Springer Monographs in Mathematics

László Fuchs

Abelian Groups
Springer Monographs in Mathematics

More information about this series at http://www.springer.com/series/3733

László Fuchs

Abelian Groups

123
László Fuchs
Mathematics Department
Tulane University
New Orleans, LA, USA

ISSN 1439-7382 ISSN 2196-9922 (electronic)

Springer Monographs in Mathematics
ISBN 978-3-319-19421-9 ISBN 978-3-319-19422-6 (eBook)
DOI 10.1007/978-3-319-19422-6

Library of Congress Control Number: 2015945018

Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.

springer.com)
Dedicated with love
to my wife Shula,
our daughter Terry, and son David
Preface

The theory of abelian groups is a branch of (abstract) algebra, which deals with
commutative groups, named after the Norwegian mathematician Niels H. Abel.
Curiously enough, it is rather independent of general group theory: its methods bear
only a slight resemblance to the non-commutative case. (However, there is a close
relationship to the theory of modules, especially over integral domains.)
This is my third book on abelian groups. My first book was published in 1958;
it was a reasonably comprehensive systematic summary of the theory at that time.
With the advent of homological methods and the explosion of the theory of torsion
groups, an extended version was needed that was published in 1970 and 1973. Even
two volumes could not claim any more a comprehensive status, but they seem to
have presented accurately the main streams in the theory. Today, after four decades
of developments and thousands of publications, it is hardly imaginable to have a
comparably complete volume. As a consequence, I had to scale down my goals. By
no means could I be content with just a broad introductory volume, so I had in mind
a monograph which goes much beyond being a mere introduction and concentrates
on most of the advanced ideas and methods of today’s research. I do hope that this
volume will provide a thorough and accurate picture of the current main trends in
abelian group theory.
The audience I have in mind with this book is anybody with a reasonable
mathematical maturity interested in algebra. My aim has been to make the material
accessible to a mathematician who would like to study abelian groups or who
is looking for particular results on abelian groups needed in his/her field of
specialization. I have made an effort to keep the exposition as self-contained as
possible, but I had to be satisfied with stating a few very important theorems without
proofs whenever the proofs would not fit in this volume. The specific prerequisites
are sound knowledge of the rudiments of abstract algebra, in particular, basic group
and ring theory. Also required is some exposure to category theory, topology, and set
theory. A few results where more advanced set theory is indispensable are included
for those who are willing to indulge in a bit more sophisticated set theory. In order to
fill the need of a reference source for experts, much additional information about the
topics discussed is included also in the “Notes” at the end of sections with reference

vii
viii Preface

to relevant publications. Many results discussed are collected from scattered articles
in the literature; some theorems appear here for the first time in a book.
The writing of such a book requires, inevitably, making tough choices on what
to include and what to leave out. I have tried to be selective so that the central ideas
stand out. My guiding principle has been to give preference to important methods
and typical results as well as to topics, which contain innovative ideas or which
I felt were particularly instructive. It is no secret that several flourishing areas of
module theory have their origin in abelian groups where the situation is often more
transparent—I have paid special attention to these results as well. As I have tried
to make the book more accessible for starters, generality is sometimes sacrificed in
favor of an easier proof. I have shied away from theorems with too technical proofs
unless the result, I believe, is theoretically or methodically extremely important. In
the selection of additional topics, the guide was my personal interest (which I view
as an indisputable privilege of authors).
In order to give a fair idea of the current main streams in abelian group theory, it
is impossible to ignore the numerous undecidable problems. Abelian group theory is
not only distinguished by its rich collection of satisfactory classification theorems,
but also for having an ample supply of interesting undecidable problems. Shelah’s
epoch-making proof of the undecidability of Whitehead’s problem in ZFC marked
the beginning of a new era in the theory of abelian groups with set-theoretical
methods assuming the leading role. The infusion of ideas from set theory created
as radical a change in the subject as Homological Algebra did a quarter of a century
before. Therefore, I had to take a more penetrating approach to set-theoretical
methods and to discuss undecidable problems as well, but I treat these fascinating
problems as interesting special cases of theoretical importance, rather than as a main
object of a systematic study. Though a short survey of set-theoretical background is
given, the reader is well advised to consult other sources.
I am a bit leisurely, primarily in the first half of the book, with the method of
presentation and the proofs in order to assist students who want to learn the subject
thoroughly. I have included a series of exercises at the end of each section. As is
customary, some of them are simple to test the reader’s comprehension, but the
majority give noteworthy extensions of the theory or sidelines to enrich the topics
studied in the same section. A student should conscientiously solve several exercises
to check how he/she commands the material. Serious attempts have been made to
provide ample examples to illustrate the theory. Good examples not only serve to
motivate the results, but also provide an explicit source of ideas for further research.
Past experience suggests that listing open problems is a good way of encouraging
young researchers to start thinking seriously on the subject. In view of this, I am
listing open problems at the end of each chapter, with a brief commentary if needed.
I have not given any serious thought to their solutions; so a lucky reader may find
quick answers to some of them.
This volume follows basically the same line of development as my previous
books on abelian groups. With the kind permission of the publishing company
Elsevier, I could include portions of my two volume book “Infinite Abelian Groups.”
The book was out of print by 1990, and at that time I was playing with the idea
Preface ix

of revising it, but decided to join forces with my friend Luigi Salce to write a
monograph on modules over non-noetherian integral domains. When I felt ready to
get involved in working on a book on abelian groups, Hurricane Katrina interfered:
it forced us to leave our home for three-and-a-half years. It took me some time to
brace myself for such a big task as writing a book. When I realized that a mere
revision would not be satisfactory, I had no choice but to start working on a new
book.
The first two chapters of this volume are introductory, and the reader who is
knowledgeable in group theory will want to skip most of the sections very quickly.
For those who are not so familiar with more advanced set-theoretical methods, it
might help to read Sect. 4 carefully before proceeding further. The text starts in
earnest in Chapter 3, which provides a thorough discussion of direct sums of cyclic
groups. Chapter 4 is devoted to divisibility and injectivity, while Chapter 5 explores
pure subgroups along with basic subgroups. In Chapter 6, we introduce algebraically
compact groups motivated by purity and topological compactness.
In the next three chapters, we embark on the fundamental homological machinery
needed for abelian groups. The insight developed here is essential for the rest of the
book. We conclude Chapter 9 with the theory of cotorsion groups.
Armed with the powerful tools of homological algebra, we proceed to the next six
chapters, which form the backbone of the theory, exploring the structure of various
classes of abelian groups. The structure theory is a vast field of central importance
that is not easy to organize into a coherent scheme. Starting with torsion groups, we
first deal with p-groups that contain no elements of infinite height. Then we go on
to explore the Ulm-Zippin theory of countable p-groups leading to the highlights
of the impressive theory of totally projective p-groups. Chapters 12 and 13 provide
a large amount of material on torsion-free groups—an area that has shown great
advancement in the past quarter of century. We explore indecomposable groups,
slender groups, and vector groups. The discussion culminates in the proof of the
undecidability of the famous Whitehead problem. Chapter 14 serves as a broad
introduction to the fascinating theory of Butler groups. In Chapter 15, the main
results on mixed groups are presented.
The final three chapters deal with endomorphism rings, automorphism groups,
and the additive groups of rings. Some ideas are introduced that interact between
abelian groups and rings.
I stress that the reader should by no means take this book as a complete survey
of the present state of affairs in abelian group theory. A number of significant and
more advanced results in several areas of the theory, as well as a wealth of important
topics (like group algebras) are left out, not to mention topics like the more advanced
theory of finite and infinite rank torsion-free groups, as well as p-groups more
general than totally projective groups. In spite of all these, I sincerely hope that
the material discussed provides a significant amount of information that will open
up new and promising vistas in our subject.
As far as the bibliography is concerned, references are not provided beyond the
list of works quoted in the text and in the “Notes.” No reference is given to the
exercises. This self-imposed restriction was necessary in view of the vast literature
x Preface

on abelian groups. The reader who is interested in going beyond the contents of this
book should use the list of references as a guide to other sources. References outside
the theory of abelian groups are usually given in an abbreviated form embedded in
the “Notes.” The system of cross references is simple: lemmas, theorems etc. are
numbered in each chapter separately. Books are cited by letters in square brackets. I
have attempted to give credit to the results, and I apologize if I have made mistakes
or omissions.
A remark is in order about notational conventions. We are using the functional
notation for maps, thus .x/, or simply x, is the image of x under the map .
Accordingly, the product  of two maps is defined as . /.x/ D . .x//. This
is not a universally adopted convention in abelian group theory, but is predominant.
The “Table of Notations” should be consulted if symbols are not clear. I have
introduced new notation or terminology only in a few places where I found the
frequently used terms clumsy or confusing, or if a name was missing.
I am grateful to my friends and colleagues Lutz Strüngmann, Luigi Salce,
Kulumani M. Rangaswamy, Claudia Metelli, Brendan Goldsmith, and Ulrich
Albrecht for their comments on portions of an earlier version of the manuscript.
I have greatly benefitted from the comments I received from them. I apologize for
the errors which remain in the text.

Metairie, Louisiana, USA László Fuchs

March 27, 2015
Contents

1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
2 Maps and Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 6
3 Fundamental Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
4 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 20
5 Families of Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
6 Categories of Abelian Groups . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
7 Linear Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 36
8 Modules .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 39
2 Direct Sums and Direct Products . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
1 Direct Sums and Direct Products . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
2 Direct Summands.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 50
3 Pull-Back and Push-Out Diagrams . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 54
4 Direct Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
5 Inverse Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
6 Direct Products vs. Direct Sums . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
7 Completeness in Linear Topologies . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 69
3 Direct Sums of Cyclic Groups .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
1 Freeness and Projectivity . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
2 Finite and Finitely Generated Groups . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 80
3 Factorization of Finite Groups .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 86
4 Linear Independence and Rank .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 91
5 Direct Sums of Cyclic Groups .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 94
6 Equivalent Presentations . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 99
7 Chains of Free Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
8 Almost Free Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 112
9 Shelah’s Singular Compactness Theorem .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 120
10 Groups with Discrete Norm.. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 122
11 Quasi-Projectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127

xi
xii Contents

4 Divisibility and Injectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131

1 Divisibility .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
2 Injective Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 134
3 Structure Theorem on Divisible Groups . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 140
4 Systems of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 143
5 Finitely Cogenerated Groups . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 145
5 Purity and Basic Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 149
1 Purity .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 149
2 Theorems on Pure Subgroups.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 155
3 Pure-Exact Sequences .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 159
4 Pure-Projectivity and Pure-Injectivity .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 163
5 Basic Subgroups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 166
6 Theorems on p-Basic Subgroups . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 173
6 Algebraically Compact Groups . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183
1 Algebraic Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183
2 Complete Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 190
3 The Structure of Algebraically Compact Groups .. . . . . . . . . . . . . . . . . . . . 195
4 Pure-Injective Hulls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 199
5 Locally Compact Groups . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
6 The Exchange Property . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 206
7 Homomorphism Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 213
1 Groups of Homomorphisms . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 213
2 Algebraically Compact Homomorphism Groups .. . . . . . . . . . . . . . . . . . . . 220
3 Small Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 225
8 Tensor and Torsion Products .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 229
1 The Tensor Product .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 229
2 The Torsion Product .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 237
3 Theorems on Tensor Products . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 242
4 Theorems on Torsion Products . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 245
5 Localization.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 251
9 Groups of Extensions and Cotorsion Groups. . . . . . . .. . . . . . . . . . . . . . . . . . . . 255
1 Group Extensions.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 255
2 Exact Sequences for Hom and Ext . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 261
3 Basic Properties of Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 266
4 Lemmas on Ext .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 270
5 The Functor Pext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 275
6 Cotorsion Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 282
7 Cotorsion vs. Torsion.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 287
8 More on Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 291
9 Cotorsion Hull and Torsion-Free Cover .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 294
Contents xiii

10 Torsion Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 299

1 Preliminaries on p-Groups . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 299
2 Fully Invariant and Large Subgroups .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 307
3 Torsion-Complete Groups . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 310
4 More on Torsion-Complete Groups . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 318
5 Pure-Complete and Quasi-Complete p-Groups . . .. . . . . . . . . . . . . . . . . . . . 321
6 Thin and Thick Groups.. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 325
7 Direct Decompositions of Separable p-Groups . . .. . . . . . . . . . . . . . . . . . . . 328
8 Valuated Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 334
9 Separable p-Groups That Are Determined by Their Socles .. . . . . . . . . 339
11 p-Groups with Elements of Infinite Height . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 343
1 The Ulm-Zippin Theory.. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 343
2 Nice Subgroups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 352
3 Simply Presented p-Groups .. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 354
4 p-Groups with Nice Systems. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 362
5 Isotypeness, Balancedness, and Balanced-Projectivity . . . . . . . . . . . . . . 364
6 Totally Projective p-Groups .. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 371
7 Subgroups of Totally Projective p-Groups . . . . . . . .. . . . . . . . . . . . . . . . . . . . 377
8 p -Purity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 385
9 The Functor p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 392
10 p!Cn -Projective p-Groups.. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 397
11 Summable p-Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 401
12 Elongations of p-Groups . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 403
12 Torsion-Free Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 409
1 Characteristic and Type: Finite Rank Groups .. . . .. . . . . . . . . . . . . . . . . . . . 409
2 Balanced Subgroups.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 417
3 Completely Decomposable Groups.. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 423
4 Indecomposable Groups . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 431
5 Pathological Direct Decompositions of Finite Rank Groups .. . . . . . . . 438
6 Direct Decompositions of Finite Rank Groups: Positive Results . . . . 446
7 Substitution Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 449
8 Finite Rank p-Local Groups . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 454
9 Quasi-Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 458
10 Near-Isomorphism.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 465
11 Dualities for Finite Rank Groups . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 470
12 More on Finite Rank Groups . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 475
13 Torsion-Free Groups of Infinite Rank . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 481
1 Direct Decompositions of Infinite Rank Groups ... . . . . . . . . . . . . . . . . . . . 481
2 Slender Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 489
3 Characterizations of Slender Groups . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 496
4 Separable Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 501
5 Vector Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 509
6 Powers of Z of Measurable Cardinalities . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 513
xiv Contents

7 Whitehead Groups If V = L . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 519

8 Whitehead Groups Under Martin’s Axiom.. . . . . . .. . . . . . . . . . . . . . . . . . . . 524
14 Butler Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 529
1 Finite Rank Butler Groups .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 529
2 Prebalanced and Decent Subgroups . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 537
3 The Torsion Extension Property . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 541
4 Countable Butler Groups.. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 546
5 B1 - and B2 -Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 548
6 Solid Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 551
7 Solid Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 557
8 Butler Groups of Uncountable Ranks . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 563
9 More on Infinite Butler Groups .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 569
15 Mixed Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 573
1 Splitting Mixed Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 573
2 Baer Groups are Free .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 579
3 Valuated Groups. Height-Matrices . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 584
4 Nice, Isotype, and Balanced Subgroups .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 590
5 Mixed Groups of Torsion-Free Rank One .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 593
6 Simply Presented Mixed Groups . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 596
7 Warfield Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 599
8 The Categories WALK and WARF . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 605
9 Projective Properties of Warfield Groups . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 608
16 Endomorphism Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 613
1 Endomorphism Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 613
2 Endomorphism Rings of p-Groups . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 622
3 Endomorphism Rings of Torsion-Free Groups . . .. . . . . . . . . . . . . . . . . . . . 629
4 Endomorphism Rings of Special Groups.. . . . . . . . .. . . . . . . . . . . . . . . . . . . . 634
5 Special Endomorphism Rings . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 636
6 Groups as Modules Over Their Endomorphism Rings . . . . . . . . . . . . . . . 643
7 Groups with Prescribed Endomorphism Rings . . .. . . . . . . . . . . . . . . . . . . . 650
17 Automorphism Groups.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 655
1 Automorphism Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 655
2 Automorphism Groups of p-Groups .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 660
3 Automorphism Groups of Torsion-Free Groups . .. . . . . . . . . . . . . . . . . . . . 664
18 Groups in Rings and in Fields . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 673
1 Additive Groups of Rings and Modules .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 673
2 Rings on Groups .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 677
3 Additive Groups of Noetherian Rings . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 682
4 Additive Groups of Artinian Rings . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 684
5 Additive Groups of Regular Rings . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 689
6 E-Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 693
Contents xv

7 Groups of Units in Commutative Rings . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 697

8 Multiplicative Groups of Fields. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 701

References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 707

Author Index.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 731

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 737

Table of Notations

Set Theory

; : subset, proper subset

[; \: set union, intersection
: cartesian product
X n Y D fx 2 X j x … Yg
jSj: cardinality of the set S
¿: empty set
fa 2 A j g: set of elements of A satisfying 
fai gi2I : the set of all ai with i 2 I
; ; ; : : : : cardinals
˛; ˇ; : : : ; ; ; : ordinals
@1 : finite, @0 : countable cardinal
2@0 : continuum
@ : th infinite cardinal, th aleph
!: first infinite ordinal or the set f0; 1; : : : ; n; : : : g
! : th initial ordinal (j! j D @ )
cf : cofinality of the ordinal 
): implication
,: equivalence, if and only if
8: for all
9: there exists
:: negation
ZFC : Zermelo-Fraenkel axioms of set theory + Axiom of Choice
CH: Continuum Hypothesis
GCH: Generalized Continuum Hypothesis
V: model of set theory
L: Gödel’s Axiom of Constructibility
V = L: Axiom L is assumed
}: Jensen’s Diamond Principle

xvii
xviii Table of Notations

MA: Martin’s Axiom

Ab: category of abelian groups
C; D; : : : : categories
V; Vp : category of valuated groups or vector spaces
G; H; : : : : G./; H./-families of subgroups

Maps

!: mapping, homomorphism
˛
A!B: map ˛ from A to B
7!: corresponds to
!: P quasi-homomorphism
1A : identity map on A
nP : multiplication by integer n
  A: restriction of map  to A
Im : image of map 
Ker : kernel of map 
Coker : cokernel of map 
; r: diagonal, codiagonal map
˚Q i : direct sum of maps i
i : direct product of maps i
[i i : union of a chain of maps
e W 0 ! A ! B ! C ! 0: exact sequence

Groups, rings

0: number 0, element 0, or subgroup f0g

N: set of positive integers
Z: group of integers
Q: group of rational numbers
R: group of real numbers
C: group of complex numbers
T Š R=Z: group of complex numbers of absolute value 1
Z.n/: cyclic group of order n
Z.p1 /: quasi-cyclic p-group
Z.p/ : group or ring of rational numbers with denominators coprime to p
Q.p/ : group or ring of rational numbers whose denominators are powers of p
Jp : group or ring of p-adic integers
Qp : group or field of p-adic numbers
H : generalized Prüfer group of length 
Table of Notations xix

N .p/: Nunke group of length  for prime p

Pˇ : Walker group of length ˇ
A; B; C; G; H; T; : : : : groups
Ap : p-component of group A
A.p/ : localization of A at prime p
E.A/: injective hull of A
Q Z-adic or p-adic completion of A
A:
O Pontryagin dual of A
A:
B: torsion-complete group with basic subgroup B
A : cotorsion hull of A
A.t/: subgroup of elements of types  t
A .t/: subgroup generated by types > t
AŒt
; A Œt
: subgroups defined in terms of t
G; H: non-commutative groups
z.S/: centralizer of set S in a group
R, S, . . . : rings
RC : additive group of ring R
U.R/: group of units of ring R
F : multiplicative group of field F
J: Jacobson radical of the ring
R M: left module M over R

Special Notation

gcd: greatest common divisor

lcm: least common multiple
rk A: rank of A
rk0 A; p-rk A: torsion-free rank, p-rank of A
rkp A: p-corank of A
fin rk A: final rank of p-group A
supp: support of a vector
bpd.A/: balanced projective dimension of A
Type.A/: type set of torsion-free finite rank group A
IT.A/, OT.A/: inner, outer type of torsion-free group A

Symbols

Š: isomorphism
: near-isomorphism
: quasi-isomorphism
ŠC : isomorphism in category C
xx Table of Notations

o.a/: order of element a

nja: integer n divides group element a
hp .a/: (transfinite) height of element a at prime p
.a/: characteristic of element a in torsion-free A
A./ D fa 2 A j .a/  g
t.a/: type of element a
B
A; B < A: B is a (proper) subgroup of A
G: normal P subgroup
B C C; Bi : subgroup generated by subgroups B; C, by subgroups Bi
hSi: subgroup generated by S
hSi : pure subgroup generated by S in a torsion-free group
jA W Bj: index of subgroup B in A
t.A/ D tA: torsion subgroup of A
nA D fna j a 2 Ag where n 2 N
AŒn
D fa 2 A j na D 0g where n 2 N
n1 B D fa 2 A j na 2 Bg for B < A
C : Z-adic closure of subgroup C in the given group
p A: set of elements in A of p-height  
A1 D \n2N nA : first Ulm subgroup of A
A : th Ulm subgroup of A
A D A =A C1 : th Ulm factor of A
u D .0 ; 1 ; : : : ; n ; : : : /: indicator, increasing sequence of ordinals and 1
A.u/: fully invariant subgroup of A defined by u
AkB: A is compatible with B
L.A/: lattice of subgroups in A
T: lattice of all types
t, s, . . . : types in torsion-free groups
t.A/: type of the torsion-free group A
`.A/: length of p-group A
f .A/ D p AŒp
=p C1 AŒp
: th UK-invariant of p-group A
f .A; G/: th Hill invariant of A relative to subgroup G
rkAt D rk.A.t/=A .t//: Baer invariant for t
j
A D fAi .i 2 I/I i g: direct (inverse) system
M, kaik k: matrix
H.a/ D kik k: height matrix of a in mixed group
A.H/ D fa 2 A j H.a/  Hg

Operations

A ˚ C: direct
Q sum of A and C
˚i2I Ai ; i2I Ai : direct sum, direct product of the Ai with i changing over I
˚ A; A./ : direct sum of  copies of A
Table of Notations xxi

Q 
Q<A; ; A : direct product of  copies of A
Qi2I Ai : -product of Ai
K Ai : K-product (K=ideal in power-set of I)
A.B/ : Boolean power of A
Hom.A; C/: group of homomorphisms from A to C
Homs .A; C/: group of small homomorphisms
Q Hom.A; C/: group of quasi-homomorphisms
HomR .A; C/: group of R-homomorphisms between R-modules R A;R C
End A: endomorphism ring (group) of A
Ends A: ring of small endomorphisms of A
Q End A: quasi-endomorphism ring of torsion-free A
Aut A: automorphism group of A
A ˝ C: tensor product of A and C
A ˝R C: tensor product of AR and R C over R
Tor.C; A/: torsion product of C and A
Ext.C; A/: group of extensions of A by C
Pext.C; A/: group of pure extensions of A by C
Bext.C; A/; PBext.C; A/: group of (pre)balanced extensions of A by C
Char A: character group of A in Pontryagin duality
Mult A: group of ring multiplications on A
TrA .G/: trace of A in G
lim; lim: direct, inverse limit
!1 
lim : first derived functor of inverse limit

Chapter 1
Fundamentals

Abstract The aim of this introductory chapter is twofold. First, to refresh the reader’s memory
about the basics before entering into the study of more serious topics. A fairly large portion of the
material can be found in standard textbooks on algebra. So, the reader may skip some or all of the
sections and turn to the appropriate places when needed. Secondly, we have to fix the fundamental
terminology to be used throughout this volume. In the 1950s, as in every rapidly developing field,
the terminology in abelian groups often varied from author to author. Slowly standardization took
place, and we will use here the widely accepted, the most familiar or the more appropriate names
for the concepts, with a couple of exceptions where new terminology is warranted.
The topics covered in this chapter include maps, diagrams, sets, categories. Most of the proofs
will be omitted as they are standard and available in textbooks.
For the comfort of the reader, especially, of those who look for references, we avoid the use of
abbreviations in the text as far as reasonable.

1 Basic Definitions

Throughout this book, ‘group’ will mean an additively written abelian group. That
is, a group is a set A satisfying the following conditions:
1. with every pair a; b 2 A, an element a C b 2 A is associated, called the sum of a
and b;
2. associative law: a C .b C c/ D .a C b/ C c for all a; b; c 2 A;
3. commutative law: a C b D b C a for all a; b 2 A;
4. there is an element 0, called zero, such that a C 0 D a for all a 2 A;
5. to each a 2 A there exists an x 2 A satisfying a C x D 0; this is x D a, the
inverse to a.
Note that a group is never empty. The associative law enables us to write a sum
of more than two summands without parentheses, and due to the commutative law,
the terms of a sum may be permuted. We write a  b to mean a C .b/, and a  b
for the inverse of a C b. The sum a C    C a (n summands) is abbreviated as na
(called a multiple of a), and a      a (n summands) as na with n 2 N (where
N denotes the set of positive integers). A sum without terms is 0, thus 0a D 0 for all
a 2 A. Notice that we do not make distinction in the notation between the integer 0
and the group element 0.
The same symbol is used for a group and for the set of its elements. The order
of a group A is the cardinal number jAj of the set of its elements. According as jAj

© Springer International Publishing Switzerland 2015 1

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_1
2 1 Fundamentals

is finite, countable, or uncountable, we call A a finite, countable or uncountable

group.
Subgroups A subset B of a group A is a subgroup if the elements of B
form a group under the same operation. A non-empty subset B of a group A is a
subgroup if and only if a; b 2 B implies a C b 2 B and b 2 B implies b 2 B, or
more simply, if and only if a; b 2 B implies a  b 2 B. If A is a finite group and
B is a subgroup, then by Lagrange’s theorem jBj is a divisor of jAj. Every subgroup
of A contains the element 0, and the subgroup f0g consisting of the element 0 alone
is called the trivial subgroup of A (there being no danger of confusion, the trivial
subgroup will be denoted by 0 instead of f0g). A subgroup of A, different from A,
is said to be a proper subgroup. We shall use the following convention: B
A
will denote that B is a subgroup of A, while B < A will mean that B is a proper
subgroup. In contrast, whenever we wish to claim that X is merely a subset (resp. a
proper subset) of A, then we shall write X  A (resp. X  A).
The set-theoretic intersection B \ C of two subgroups B; C of A is again a
subgroup. More generally, if fBi gi2I is a family of subgroups of A, for any index
set I, then their intersection B D \i2I Bi is likewise a subgroup of A. If I is the
empty set, then we agree that B D A. If the subgroups B and C satisfy B \ C D 0,
we will say that B and C are disjoint. This terminology is not consistent with
the set-theoretical meaning of the word, but it is an accepted agreement that the
group-theoretical disjointness should mean that the subgroups have nothing else in
common other than what they ought to have.
If S is a subset of A, the symbol hSi will denote the subgroup of A generated
by S, i.e. the intersection of all subgroups of A that contain S. If S D f: : : ; ai ; : : : g,
then we write hSi D h: : : ; ai ; : : : i, or simply hSi D hai ii2I . This hSi consists of all
linear combinations (i.e., sums of the form) n1 a1 C    C nk ak .ai 2 S; ni 2 Z/
where k denotes a non-negative integer. We set h¿i D 0. In case hSi D A, we say
that S is a generating system of A, and its elements are generators. If B; C are
subgroups of A, then the subgroup hB; Ci they generate consists of the elements
b C c .b 2 B; c 2 C/. We may write, therefore, hB; Ci D B C C. For a possibly
infinite collection of subgroups fBi gi2I of A, the subgroup B they generate P consists
of all finite sums bi1 C    C bik with bij 2 Bij ; we shall then write B D i2I Bi .
A group hai, generated by a single element a, is called cyclic. The order of
this cyclic group is called the order of the element a; notation: o.a/. The order
of an element can be a positive integer or the symbol 1. If a is of finite order
n, then na D 0, so hai D f0; a; : : : ; .n  1/ag, while if o.a/ D 1, then
hai D f0; ˙a; : : : ; ˙na; : : : g (all different elements).
By a superfluous subgroup of a group A is meant a subgroup G
A such that
X C G D A holds for a subgroup X
A only if X D A, i.e. the elements of G
are superfluous in any generating system. Evidently, subgroups of a superfluous
subgroup are superfluous, and the sum of two superfluous subgroups is again
superfluous.
1 Basic Definitions 3

Example 1.1.
(a) The only superfluous subgroup of the group Z of integers is the 0 subgroup. In fact, if nZ is
any non-zero subgroup and 1 < m 2 N satisfies gcd.n; m/ D 1, then mZ C nZ D Z, but
mZ ¤ Z.
(b) In a cyclic group of prime power order, all proper subgroups are superfluous.

The set of all subgroups of a group A is partially ordered under the inclusion
relation. It is a lattice, where B \ C and B C C are the lattice-operations ‘inf’
and ‘sup,’ respectively, for subgroups B; C of A. This lattice L.A/ has a minimum
and a maximum element (0 and A), and satisfies the important modular law: for
subgroups B; C; D of A,

B C .C \ D/ D .B C C/ \ D provided B
D:

In fact, the inclusion relation

being obvious, we need only verify that every
d 2 .B C C/ \ D belongs to the subgroup on the left. We can write d D b C c
with b 2 B; c 2 C. Thus d  b D c belongs to both D and C. Hence c 2 C \ D, and
d D b C c 2 B C .C \ D/.
Factor Groups If B
A and a 2 A, then aCB D faCb j b 2 Bg is called a coset
of A modulo B. Any element in a coset, also called a representative of the coset,
identifies the coset. The cosets of A mod B are pairwise disjoint, and A is their set-
theoretical union. Elements x; y 2 A are congruent (written x  y) mod B, if they
belong to the same coset mod B. A set consisting of exactly one representative from
each coset of A mod B is a complete set of representatives mod B. Its cardinality
is the index of B in A, denoted as jA W Bj: If this cardinal number is finite, we call B
of finite index in A. If A is a finite group, then jA W Bj D jAj=jBj.
The cosets of A mod B form a group AN D A=B known as the factor group
(or quotient group) of A mod B. In this group, the sum of two elements C1 ; C2
(which are cosets in A) is defined as the coset consisting of the elements fc1 C
c2 j c1 2 C1 ; c2 2 C2 g; it is uniquely determined as the coset represented by any of
its elements. The zero of AN is the coset B, and the inverse of a coset C is the coset
C D fc j c 2 Cg.
We shall frequently refer to the natural bijection between the subgroups of the
factor group A=B and the subgroups of A containing B. The elements of A contained
in elements of a subgroup CN of AN D A=B form a subgroup C of A containing B. On
the other hand, if B
C
A, then the cosets of A mod B containing only elements
from C form a subgroup CN
A=B. This correspondence C $ CN is a bijection, and
we may write CN D C=B. Note that jCj N D jC W Bj and jAN W Cj
N D jA W Cj.

The Torsion Subgroup If every element of a group A is of finite order, then A is

called a torsion (or periodic) group, while A is torsion-free if all its elements, with
the exception of 0 (which has order 1), are of infinite order. Mixed groups contain
both non-zero elements of finite order and elements of infinite order. By a primary
group or p-group is meant a group the orders of whose elements are powers of a
fixed prime p. An abelian group A has a unique p-Sylow subgroup Ap for each prime
p: Ap consists of all elements of A whose orders are powers of p.
4 1 Fundamentals

Theorem 1.2. In a group A, the set T of elements of finite order is a subgroup. T is

a torsion group and A=T is torsion-free.
Proof. Since 0 2 T, the set T is not empty. If a; b 2 T, i.e. ma D 0; nb D 0 for some
m; n 2 N, then mn.a  b/ D 0, and so a  b 2 T. Thus T is a torsion subgroup. To
show that A=T is torsion-free, let a C T be a coset of finite order, i.e. m.a C T/  T
for some m 2 N. This means that ma 2 T, so n.ma/ D 0 for some n 2 N. Thus a is
of finite order, i.e. a 2 T, and a C T D T is the zero of A=T. Consequently, A=T is
torsion-free. 
We shall call T the (maximal) torsion subgroup or the torsion part of A, and
we shall denote it by t.A/ or tA. (If we refer to the torsion subgroup of A, then we
always mean the maximal torsion subgroup.)
The following notations are typical for abelian groups; they will be used all
the time without explanation. Given a group A and an integer n 2 N, define the
subgroups:

nA D fna j a 2 Ag and AŒn

D fa 2 A j na D 0g:

Thus b 2 nA if and only if the equation nx D b has a solution for x in A, and

c 2 AŒn
exactly if o.c/jn. A fundamental concept is the pure subgroup. Purity will
be discussed in Chapter 5, here we just state the definition: a subgroup G of A is
a pure subgroup if nG D G \ nA for every n 2 N, i.e. if whenever the equation
nx D g 2 G admits a solution in A for x, then it is also solvable in G.
Ulm Subgroups The first Ulm subgroup of A is defined as

A1 D \n2N nA:

The second Ulm subgroup is A2 D .A1 /1 , etc. We shall also need the th Ulm
subgroups for ordinals ; these are defined transfinitely by the rules: A C1 D .A /1 ,
and A D \ < A if  is a limit ordinal (see Sect. 4). (We view A D A0 .) The
Ulm length of A is the smallest cardinal  such that A C1 D A (which exists by
cardinality reason). The th Ulm factor of A is the factor group

A D A =A C1 :

Given a 2 A, the largest non-negative integer n for which the equation pn x D a

is solvable for x 2 A is said to be the p-height hp .a/ of a. If this equation is solvable
for every integer n > 0, then a is of infinite p-height, hp .a/ D 1. The element 0
is of infinite height at every prime. If it is completely clear from the context which
prime p is meant, then we may simply talk of the height of a and write h.a/. (In
Chapter 11, we shall discuss transfinite heights.)
The socle s.A/ of a group A is the set (actually the subgroup) of all the elements
a 2 A such that o.a/ is a square-free integer. If s.A/ D A, A is called an elementary
group. If A is a p-group, then s.A/ D AŒp
.
1 Basic Definitions 5

A subgroup E of A is essential if its intersection with any non-zero subgroup of

A is non-zero. It is easily checked that (i) E is essential in A if and only if it contains
the socle of A and A=E is torsion; (ii) the property of being an essential subgroup is
transitive; and (iii) the intersection of two essential subgroups is again an essential
subgroup.
F Notes. Theorem 1.2 is an elementary, but fundamental result. It means that a typical group
can be thought of as being a composite of a torsion and a torsion-free group. This, however, does
not reduce the theory of mixed groups to those of these constituents, since a major issue that
remains is to find out how they are glued together to form the mixed groups. It is hard to trace the
history of Theorem 1.2.
Generalizations of ‘torsion’ exist for modules, albeit not over all rings. If we mean by a ‘torsion’
element one whose annihilator in the ring is ¤ 0, then the left Ore domains are exactly those rings
R for which in every left R-module M the torsion elements form a submodule T and M=T has
no torsion ¤ 0. There is an extensive literature on torsion theories in module categories, even in
additive categories; see, e.g., J. Golan’s book Torsion Theories (1986).

Exercises

(1) The associativity and commutativity laws can be combined into a single law:
.a C b/ C c D a C .c C b/ for all a; b; c 2 A.
(2) (a) Let B1 ; : : : ; Bk be subgroups of the group A, and let B D B1 \    \ Bk .
The index jA W Bj is not larger than the product of the indices jA W Bi j.
(b) The intersection of a finite number of subgroups of finite index is again a
subgroup of finite index.
(3) Let B; C be subgroups of A.
(a) For every a 2 A, the cosets a C B and a C .B C C/ have non-zero
intersections with the same cosets mod C.
(b) A coset mod B contains exactly jB W .B \ C/j pairwise incongruent
elements mod C.
(4) (O. Ore) The group A has a common system of representatives mod two of its
subgroups, B and C, if and only if jB W .B \ C/j D jC W .B \ C/j. [Hint: for
necessity use Exercise 3; for sufficiency, divide the cosets mod B into blocks
mod .B C C/, and define a bijective correspondence within the blocks.]
(5) (N.H. McCoy) (a) Let B; C; G be subgroups of A such that G is contained in
the set union B [ C. Then either G
B or G
C. [Hint: if b 2 .B \ G/ n C,
then c 2 C \ G implies b C c 2 B \ G; c 2 B \ G.]
(b) The same fails for the set union of three subgroups.
(6) If n D pi11    pikk is the canonical representation of the integer n > 0, then
nA D pi11 A \    \ pikk A and AŒn
D AŒpi11
˚    ˚ AŒpikk
.
(7) (Honda) If B
A and m 2 N, set m1 B D fa 2 A j ma 2 Bg. Prove that
(a) m1 B is a subgroup of A containing B; (b) m1 0 D AŒm
; (c) m1 mB D
B C AŒm
; (d) m.m1 B/ D B \ mA; (e) m1 n1 B D .mn/1 B where n 2 N.
6 1 Fundamentals

(8) A superfluous subgroup is contained in every maximal subgroup.

(9) (B.H. Neumann) Let ai C Bi .i D 1; : : : ; n/ be cosets of subgroups Bi of A
such that their set union is all of A. Then one of Bi has finite index in A. [Hint:
induct on n.]
(10) Let B be a subgroup of A. (a) If B is torsion, then B
tA. (b) If A=B is torsion-
free, then tA
B.
(11) For every n 2 N, we have n.tA/ D t.nA/.
(12) Prove the triangle inequality for the heights:

hp .a C b/  minfhp .a/; hp .b/g

for the elements a; b of any group A. Equality holds if hp .a/ ¤ hp .b/.

(13) (a) Prove the inclusion relations .B \ C/ C .B \ D/
B \ .C C D/ and
B C .C \ D/
.B C C/ \ .B C D/ for subgroups B; C; D of a group A.
(b) Find examples where proper inclusions hold.

2 Maps and Diagrams

Homomorphisms Let A and B be arbitrary groups. A homomorphism

˛
˛ W A ! B (often denoted as A ! B) is a function that associates with every
element a 2 A a unique element b 2 B, written as ˛.a/ D b (or simply as ˛a D b),
such that it preserves addition:

˛.a1 C a2 / D ˛.a1 / C ˛.a2 / for all a1 ; a2 2 A:

A is the domain and B is the codomain or range of ˛. If there is no need to name

the homomorphism, then we write simply A ! B.
A homomorphism ˛ W A ! B gives rise to two subgroups: Ker ˛
A and
Im ˛
B. Ker ˛ is the kernel of ˛: the set of all a 2 A with ˛a D 0, while
Im ˛, the image of ˛, consists of all b 2 B such that there is an a 2 A with
˛a D b. The factor group B= Im ˛ is called the cokernel of ˛; notation: Coker ˛.
If Im ˛ D B, then ˛ is surjective or epic; we also say that it is an epimorphism.
If Ker ˛ D 0, ˛ is said to be injective or monic; also, ˛ is a monomorphism.
If both Im ˛ D B and Ker ˛ D 0, then ˛ is a bijection; in this case, it is called an
isomorphism. The groups A and B are isomorphic (denoted as A Š B) if there is an
isomorphism ˛ W A ! B. Then the inverse map ˛ 1 W B ! A exists, and is again an
isomorphism. Abstractly, no distinction is made between isomorphic groups, unless
they are distinct subgroups of the same larger group under consideration. If G is a
subgroup of both A and B, and if ˛ W A ! B fixes the elements of G, then ˛ is a
homomorphism over G.
2 Maps and Diagrams 7

A homomorphism ˛ with Im ˛ D 0 is referred to as a zero homomorphism; it

will be denoted by 0. If A
B, then the map that assigns to every a 2 A itself may
be regarded as a homomorphism of A into B; it is called the injection or inclusion
map. The injection 0 ! A is the unique homomorphism of the group 0 into A. If
˛ W A ! B and C
A, then the restriction D ˛  C has the domain C and
codomain B. In this case, ˛ is viewed as an extension of , written
˛.
If ˛0 <    < ˛n < : : : is a chain of extensions of maps (i.e., each map is an
extension of its predecessor), then [n<! ˛n is their ‘union.’ This is a map from the
union of the domains of the ˛n to the common codomain.
Let ˛ W A ! B and ˇ W B ! C be homomorphisms between groups; here, the
codomain of ˛ is the same as the domain of ˇ. The composite map A ! B ! C,
called the product of ˛ and ˇ and denoted by ˇ ı ˛ or simply by ˇ˛, is a
homomorphism A ! C (notice the order of factors!). Keep in mind that ˇ˛ acts
according to the rule

.ˇ˛/a D ˇ.˛a/ for all a 2 A:

The associative law .ˇ˛/ D . ˇ/˛ holds whenever the products ˇ˛ and ˇ are
defined. ˛ is right-cancellable (i.e. ˇ˛ D ˛ always implies ˇ D ) exactly if ˛
is an epimorphism, and left-cancellable (˛ˇ D ˛ always implies ˇ D ) if and
only if it is a monomorphism. The product of two epimorphisms (monomorphisms)
is again one.
If ˛; ˇ W A ! B are homomorphisms, then their sum ˛ C ˇ is again a
homomorphism A ! B defined as

.˛ C ˇ/a D ˛a C ˇa for all a 2 A:

Under this operation, the homomorphisms from A to B form an abelian group,

denoted as Hom.A; B/ (to be studied in Chapter 7).
A homomorphism of A into itself is called an endomorphism, and an isomor-
phism with itself an automorphism. The endomorphisms of A form a ring End A,
called the endomorphism ring of A, and the automorphisms of A form a group
Aut A, called the automorphism group of A (which is rarely commutative). The
identity automorphism 1A of A is the identity both in End A and in Aut A. A subgroup
of A that is carried into itself by every endomorphism (automorphism) of A is said
to be a fully invariant (characteristic) subgroup of A.
Both the sum and the intersection of fully invariant subgroups of A are fully
invariant in A. Thus the fully invariant subgroups of A form a sublattice in the lattice
L.A/ of subgroups of A. If S is a subset of the group A, then we can talk about the
fully invariant subgroup generated by S: this is the intersection of all fully invariant
subgroups containing S, and coincides with the set of all sums of the images of
elements in hSi under endomorphisms of A.
Let ˛ W A ! B be a homomorphism, and set K D Ker ˛. The map a C K 7! ˛a
from the factor group A=K to Im ˛ is an isomorphism. Thus Im ˛ Š A=K. The map
8 1 Fundamentals

A ! A=K acting as a 7! a C K is called the natural or canonical homomorphism.

There is an important isomorphism

B=.B \ C/ Š .B C C/=C if B; C
A;

known as the first isomorphism theorem by Emmy Noether. The natural isomor-
phism is b C .B \ C/ 7! b C C for b 2 B.
If C
B
A, then there is an epimorphism A=C ! A=B acting as a C C 7!
a C B, whose kernel is B=C. This leads to the second isomorphism theorem:

A=B Š .A=C/=.B=C/ where C

B
A:

Exact Sequences A sequence of groups Ai and homomorphisms ˛i ,

˛1 ˛2 ˛k
A0 ! A1 ! : : : ! Ak .k  2/;

is called an exact sequence if Im ˛i D Ker ˛iC1 for i D 1; : : : ; k  1. In particular,

˛ ˇ
the sequence 0 ! A ! B is exact if and only if ˛ is monic, and B ! C ! 0 is
˛
exact if and only if ˇ is epic. The exactness of 0 ! A ! B ! 0 is equivalent to ˛
being an isomorphism. The most frequently used exact sequences are of the form

˛ ˇ
e W 0 ! A ! B ! C ! 0;

and are called short exact sequences; here, ˛ is an injection of A in B such that ˇ is
a surjective map with Im ˛ as kernel. Whenever convenient, we may identify A with
the subgroup Im ˛ of B, and C with the quotient group B=A; in other words, A can
be treated as a subgroup, and C as a factor group of B.
Another important exact sequence arises from an arbitrary group homomorphism
 W A ! B. This is the sequence

˛  ˇ
0 ! Ker  ! A ! B ! Coker  ! 0

where ˛ denotes the embedding, and ˇ the canonical map mod Im .

Diagrams The balance of this section is devoted to diagrams. Diagrams not
only assist us in understanding statements, but they also clarify proofs, and most
importantly, help intuition.
Roughly speaking, a diagram consists of capital letters, representing groups,
and arrows between certain pairs of capital letters, representing homomorphisms
between the corresponding groups. A diagram is commutative if we get the same
composite homomorphism no matter how we move along directed arrows on
different paths from one group to another group in the diagram. For instance, the
diagram
2 Maps and Diagrams 9

μ ν
A −−−−→ B −−−−→ C
⏐ ⏐ ⏐
⏐α ⏐β ⏐γ
  
μ ν
A −−−−→ B  −−−−→ C 

is commutative exactly if the homomorphisms ˇ and 0 ˛ of A into B0 are equal,

and the same holds for the homomorphisms  and  0 ˇ of B into C0 . Then the
equality of the homomorphisms ;  0 ˇ;  0 0 ˛ is a simple consequence. In
diagrams, the identity map is often denoted by the sign of equality, as, e.g., in
α
A −−−−→ B
 ⏐
 ⏐β
 
γ
A −−−−→ C

This diagram is commutative exactly if D ˇ˛.

The following two lemmas are dual to each other. In the proofs, the technique
with maps is instructive.
Lemma 2.1. A diagram

G
⏐
⏐η

α β
0 −−−−→ A −−−−→ B −−−−→ C

with exact row can be embedded in a commutative diagram

G
⏐
⏐η
φ 
α β
0 −−−−→ A −−−−→ B −−−−→ C

exactly if ˇ D 0. In this case,  W G ! A is uniquely determined.

Proof. If such a  exists, then  D ˛ implies ˇ D ˇ˛ D 0 D 0, thus the

stated condition is necessary. Conversely, if ˇ D 0 holds, then Im 
Ker ˇ.
By the exactness of the row, ˛ is monic and Ker ˇ D Im ˛, so Im 
Im ˛.
Hence the map  D ˛ 1  W G ! A is well defined and obviously satisfies ˛ D .
If  0 W G ! A also satisfies ˛ 0 D , then ˛ 0 D ˛ implies  0 D , since ˛ is a
monomorphism. 
In order to save space, the two diagrams may be combined to a single one:
10 1 Fundamentals

G
⏐
⏐η
φ 
α β
0 −−−−→ A −−−−→ B −−−−→ C

where the solid arrows represent given maps, while the broken arrow designates a
homomorphism to be “filled in.”
Lemma 2.2. A diagram

α β
A −−−−→ B −−−−→ C −−−−→ 0
⏐
⏐
η
ψ

with exact row can be completed by a map W C ! G so as to get a commutative

triangle exactly if ˛ D 0. Such a is unique.
Proof. If such a exists, then from  D ˇ we obtain ˛ D ˇ˛ D 0 D 0,
and the necessity is clear. Conversely, assume ˛ D 0, and define W C ! G as
follows. For c 2 C set c D b with some b 2 B satisfying ˇb D c. This is a good
definition, for if also b0 D c for b0 2 B, then b0  b 2 Ker ˇ D Im ˛
Ker , so
b0 D b. Uniqueness follows from the surjectivity of ˇ. 
Next, we prove a lemma that has several applications. In its proof, the preceding
lemmas are used without explicit reference.
Lemma 2.3. Suppose

α β
0 −−−−→ A −−−−→ B  −−−−→ C  −−−−→ 0
φ σ ψ ρ η

α β
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

is a commutative diagram with exact rows. There exists a map  W B0 ! A making

the upper triangle commute .i.e., ˛ 0 D / if and only if there is a map  W C0 ! B
making the lower triangle commute .i.e., ˇ D /:
2 Maps and Diagrams 11

Proof. If  W B0 ! A satisfies ˛ 0 D , then . ˛/˛ 0 D ˛ ˛ D 0 implies the

existence of a map W C0 ! B such that ˇ 0 D  ˛. Hence from ˇˇ 0 D ˇ 
ˇ˛ D ˇ 0 we obtain ˇ D . Conversely, if there is a map W C0 ! B such that
ˇ D , then  D  ˇ 0 satisfies ˇ D ˇ  ˇˇ 0 D ˇ 0  ˇ 0 D 0. Therefore,
there is a map  W B0 ! A with ˛ D . It satisfies ˛˛ 0 D ˛ 0  ˇ 0 ˛ 0 D ˛, that
is, ˛ 0 D , as desired. 
The next two lemmas give us an opportunity to get acquainted with a simple, but
extremely useful technique, called diagram chasing. The first lemma is a prelude
to the second one.
Lemma 2.4. Suppose we are given a commutative square as the center square in
the diagram

α
0 → Ker α −−−−→ A −−−−→ A −−−−→ Coker α → 0
⏐ ⏐
⏐
μ
⏐ν
μ  ν

β
0 → Ker β −−−−→ B −−−−→ B  −−−−→ Coker β → 0

with exact rows. Then  induces a map 0 W Ker ˛ ! Ker ˇ, and  induces a map
 0 W Coker ˛ ! Coker ˇ making the left and the right squares commute. 0 is monic
if so is , and  0 is epic if so is .
Proof. The map 0 D   Ker ˛ carries evidently Ker ˛ into B. We have to check
that it lands in Ker ˇ. So we pick an a 2 Ker ˛, and want to show that a 2 Ker ˇ,
i.e. ˇa D 0. This is indeed true, since ˇ D ˛ by the commutativity of the
central square, and ˛a D 0 by the choice of a. It is also clear that 0 is monic if 
is monic.
A kind of dual argument applies to the other half of the claim. Let x 2 Coker ˛,
i.e. x is a coset a0 C Im ˛ .a0 2 A0 /. Then a0 C Im ˇ is independent of the selection
of the representative a0 of the coset, because if a00 D a0 C ˛a for an a 2 A, then
a00 D a0 C˛a D a0 Cˇa 2 a0 CIm ˇ. It is immediate that the correspondence
a0 C Im ˛ 7! a0 C Im ˇ is a homomorphism  0 W Coker ˛ ! Coker ˇ. If  is
epic, then for every coset b0 C Im ˇ .b0 2 B0 / there exists an a0 2 A0 such that
a0 D b0 C Im ˇ, whence the surjectivity of  0 is evident. 
The Snake Lemma The next, forbiddingly looking diagram is not as formidable
as it appears at the first sight. The lemma provides us with a long exact sequence
that will have significant applications in later chapters.
12 1 Fundamentals

Lemma 2.5 (Snake Lemma or Kernel-Cokernel Sequence). Let

0 0 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
μ1 ν1 δ
0 −−−−−→ Ker α −−−−−→ Ker β −−−−−→ Ker γ −−−−−→
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
μ ν
0 −−−−−→ A −−−−−→ B −−−−−→ C −−−−−→ 0
⏐ ⏐ ⏐
⏐α ⏐ ⏐γ
 β 
μ ν
0 −−−−−→ A −−−−−→ B −−−−−→ C −−−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
δ 2 μ 2 ν
−−−−−→ Coker α −−−−−→ Coker β −−−−−→ Coker γ −−−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 0 0

be a commutative diagram, where the two middle rows and all the columns are
exact. Then there exist maps making the top and bottom rows exact, as well as a
map ı W Ker ! Coker ˛, called connecting homomorphism, making the long
sequence

ı
0 ! Ker ˛ ! Ker ˇ ! Ker ! Coker ˛ ! Coker ˇ ! Coker ! 0

exact.
Proof. To start with, note that the vertical maps from the kernels are viewed as
inclusions, while those into the cokernels as canonical epimorphisms.
First we define the connecting homomorphism ı. Pick an element c 2 Ker .
There is a b 2 B such that b D c. As c D 0, we have  0 ˇb D 0, which implies
that some a0 2 A0 satisfies 0 a0 D ˇb. Define ıc D a0 C Im ˛ 2 Coker ˛. It is urgent
to check that this is independent of the choice of b 2 B. This is easily done: another
choice b D b C x .x 2 A/ leads to a D a0 C ˛x which gives the same coset in
Coker ˛ as a0 .
Since the maps 1 ; 1 act like ; , it is easily seen that the sequence of kernels
is exact. A similar comment settles the exactness of the sequence of the cokernels.
It remains to check exactness at Ker and Coker ˛.
It is clear that the composite maps at these places are 0. Clearly, ıc D 0 if and
only if a0 D ˛a for some a 2 A. Then ˇb D 0 a0 D ˇa, thus b  a 2 Ker ˇ. As
c D b D .b  a/, the exactness at Ker follows at once. Similarly, a0 C Im ˛ 2
Ker 2 .a0 2 A0 / means that 0 a0 2 Im ˇ, so there is a b 2 B with 0 a0 D ˇb. Then
b D c 2 Ker is mapped by ı upon a0 C Im ˛. 
2 Maps and Diagrams 13

The 5- and the 3  3-Lemma Finally, we state two more lemmas that are often
useful in applications. The direct proofs are rather technical and omitted. They can
be found, e.g., in Mac Lane [M], but they can also be derived from the Snake
Lemma.
Lemma 2.6 (The 5-Lemma). Suppose

1 α 2 α 3 α 4 α
A1 −−−− → A2 −−−− → A3 −−−− → A4 −−−− → A5
⏐ ⏐ ⏐ ⏐ ⏐
⏐γ1 ⏐γ2 ⏐γ3 ⏐γ4 ⏐γ5
    
β1 β2 β3 β4
B1 −−−−→ B2 −−−−→ B3 −−−−→ B4 −−−−→ B5

is a commutative diagram with exact rows. Then

(a) if 1 is epic and 2 ; 4 are monic, then 3 is monic;
(b) if 5 is monic and 2 ; 4 are epic, then 3 is epic;
(c) if 1 is epic, 5 is monic, and 2 ; 4 are isomorphisms, then 3 is an
isomorphism. u
t
Lemma 2.7 (The 3  3-Lemma). Assume that the diagram

0 0 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
1 α 1 β
0 −−−−−→ A1 −−−−−→ B1 −−−−−→ C1 −−−−−→ 0
⏐ ⏐ ⏐
⏐
λ1 
⏐
μ1 
⏐ν
 1
2 α 2 β
0 −−−−−→ A2 −−−−−→ B2 −−−−−→ C2 −−−−−→ 0
⏐ ⏐ ⏐
⏐
λ2 
⏐
μ2 
⏐ν
 2
3 α 3β
0 −−−−−→ A3 −−−−−→ B3 −−−−−→ C3 −−−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 0 0

is commutative and has exact columns. If the first two rows or the last two rows are
exact, then all three rows are exact. t
u
F Notes. Only with the advance of category theory became the maps (homomorphisms)
between groups, and more generally, between algebraic systems, major players in algebra. Their
overall importance is manifest.
We will often draw diagrams to support proofs, though frequently, diagrams that can easily be
supplied by the reader will be skipped in order to save space. We will see that even simple minded
diagrams prove very effective in understanding claims, and can be extremely helpful in following
proofs. Readers are encouraged to draw diagrams whenever possible, and it is hoped that serious
students will fall into this useful habit.
14 1 Fundamentals

Exercises

(1) Assume ˛ W A ! B and ˇ W B ! C. Prove that

(a) Ker ˇ˛  Ker ˛, and equality holds if ˇ is a monomorphism;
(b) Im ˇ˛
Im ˇ, and equality holds if ˛ is an epimorphism.
(2) Let again ˛ W A ! B and ˇ W B ! C.
(a) If ˇ˛ is a monomorphism, then ˛ is monic, but ˇ need not be monic.
(b) If ˇ˛ is an epimorphism, then ˇ is, but ˛ is not necessarily epic.
(3) For every group A and for every positive integer m there are exact sequences
0 ! AŒm
! A ! mA ! 0 and 0 ! mA ! A ! A=mA ! 0:
(4) Suppose ˛ W A ! B and B0
B.
(a) If we write ˛ 1 B0 D fa 2 A j ˛a 2 B0 g, then we have ˛.˛ 1 B0 /
B0 .
(b) For A0
A, the following is true: A0
˛ 1 .˛A0 /.
(5) Let  2 End A and m 2 N. Then Ker m D m1 Ker .
(6) Let B and C be subgroups of A such that A D B C C, and ˇ W B ! G,
W C ! G homomorphisms into the same group G. There is an ˛ W A ! G
with ˛  B D ˇ and ˛  C D if and only if ˇ and are equal on B \ C.
(7) For every m 2 N, and for every fully invariant subgroup H of A, the subgroups
mH and HŒm
are likewise fully invariant in A.
(8) Full invariance is a transitive property: a fully invariant subgroup of a fully
invariant subgroup is fully invariant.
(9) (a) If H is a fully invariant subgroup of A, and  is an endomorphism of A, then
the correspondence a C H 7! a C H is an endomorphism of A=H.
(b) If H is fully invariant in A, and G=H is fully invariant in A=H, then G is
fully invariant in A.
(10) If Hi .i 2 I/Pare fully invariant (characteristic) subgroups of A, then so are
\i2I Hi and i2I Hi .
˛ ˇ
(11) Let 0 ! A ! B ! C ! 0 be an exact sequence, and B0
B. Then there
˛0 ˇ0
exist A0
A and C0
C such that the sequence 0 ! A0 ! B0 ! C0 ! 0
is exact where ˛ 0 D ˛  A0 and ˇ 0 D ˇ  B0 .
(12) (a) In a diagram

α
A −−−−−→ B
⏐
⏐β
φ 
γ
0 −−−−−→ C −−−−−→ D

with exact row, an arrow  W A ! C can be filled in to make the diagram

commute if and only if Im ˇ˛
Im . Such a  is unique.
(b) Formulate and prove the dual of (a).
3 Fundamental Examples 15

3 Fundamental Examples

Our next order of business is to get acquainted with the groups that we will meet all
the time in the sequel.
Cyclic Groups They have been defined above as groups that can be generated
by a single element: C D hci. A cyclic group can be finite or infinite.
The elements of an infinite cyclic group C generated by c are nc (all distinct)
with n running over the additive group Z of integers. C is isomorphic to Z, an
isomorphism is given by the correspondence nc 7! n 2 Z. Thus all infinite cyclic
groups are isomorphic. Along with c, also c can be a generator of C, but no other
element alone can generate C.
The list of elements of a cyclic group C D hci of order m is: 0; c; 2c; : : : ;
.m  1/c. Because of mc D 0, we compute in C just as with the integers mod m.
Consequently, C is isomorphic to the additive group of residue classes of the integers
mod m; this group is Z=mZ. Thus all finite cyclic groups of the same order are
isomorphic; we shall use the notation Z.m/. (In the literature, often Zm is used,
though Zp is in conflict with the notation used for the localization of Z at p.) Along
with c, every kc with gcd.k; m/ D 1 can serve as a single generator of C, but only
these.
The simple abelian groups (0 is the only proper subgroup) are of prime order;
they are cyclic, isomorphic to Z=pZ for some prime p. They can be generated by
any element ¤ 0.
Theorem 3.1. Subgroups of cyclic groups are cyclic.
Proof. Let C D hci be any cyclic group, and B a subgroup in C. If B D 0, then B is
evidently cyclic, so for the rest of the proof we may assume that B ¤ 0. Therefore,
there exists kc 2 B with kc ¤ 0. Then also kc 2 B, so B must contain multiples of
c with positive coefficients. Among such coefficients there is a minimal one, say, n.
We are going to show that B D hnci. As nc 2 B, the inclusion hnci
B is obvious.
To prove the reverse inclusion, let sc be an arbitrary element of B where s 2 Z.
Euclidean division yields s D qn C r with q; r 2 Z such that 0
r < n. Now
rc D .s  qn/c D sc  q.nc/ 2 B, whence the choice of n implies r D 0. Hence
sc D q.nc/ 2 B, proving that B
hnci. 
An infinite cyclic group C D hci has infinitely many subgroups; these are: hmci
for m D 0; 1; 2; : : : . The subgroups of a finite cyclic group C D hci of order n are
hmci for positive divisors m of n.
Cocyclic Groups A cyclic group can be characterized as a group C containing
an element c such that every homomorphism  W A ! C (for any group A) with
c 2 Im  is surjective. Dualizing this concept, we arrive at the definition of cocyclic
groups: C is cocyclic if there is an element c 2 C such that any homomorphism
 W C ! A with c … Ker  is monic. In this situation, c is called a cogenerator of
C. Since every subgroup is the kernel of a suitable homomorphism, a cogenerator
c must belong to all non-zero subgroups of C. Hence the intersection of all
16 1 Fundamentals

non-zero subgroups of a cocyclic group C is not 0; this is the unique smallest non-
zero subgroup in C, it is a simple group generated by c. Conversely, if a group has a
unique smallest subgroup ¤ 0, then the group is cocyclic, and any non-zero element
in the smallest subgroup is a cogenerator.
Example 3.2. A cyclic group C D hci of prime power order pk is cocyclic, where any element of
order p is a cogenerator. This follows from the simple fact that the only subgroups of such a C are
those in the chain

0 < hpk1 ci <    < hpci < hci:

Let p denote a prime number. The pn th complex roots of unity, with n running
over all integers > 0, form an infinite multiplicative group; in accordance with our
convention, we will switch to the additive notation. This group, called a quasi-cyclic
group or a group of type p1 (notation: Z.p1 /), can be defined as follows: it is
generated by elements c1 ; c2 ; : : : ; cn ; : : : such that

pc1 D 0; pc2 D c1 ; : : : ; pcnC1 D cn ; : : : .n 2 N/:

Here o.cn / D pn , and Z.p1 / is the union of the ascending chain of cyclic subgroups
hcn i; these are the only non-zero proper subgroups of Z.p1 /.
Theorem 3.3. A group C ¤ 0 is cocyclic if and only if it is isomorphic to Z.pk / for
some prime p and for some k 2 N [ f1g.
Proof. Let c be a cogenerator of C. Then hci is the smallest subgroup ¤ 0 of C, and
therefore c must be of prime order p. Since c lies in every non-zero subgroup of C,
C cannot contain elements of infinite order, neither elements whose order contains
a prime factor ¤ p, i.e. C is a p-group. As a basis of induction, assume that for an
integer n, C contains at most one subgroup Cn of order pn , and this is cyclic, say,
Cn D hcn i, containing all elements of orders
pn . This is evidently true for n D 1.
If A; B are subgroups of order pnC1 in C, then there are elements a 2 A n Cn and
b 2 BnCn , and their orders must be pnC1 . We may pick a; b such that pa D cn D pb.
Hence p.a  b/ D 0, so a  b D tcn for some t 2 Z. We conclude that a D b C tpb
and b D a  tpa, i.e. a 2 hbi and b 2 hai. Consequently, A D hai D hbi D B is
cyclic, and is the only subgroup of order pnC1 in C. Moreover, it must contain all
the elements of order
pnC1 . Thus C is the union of an ascending chain of cyclic
groups of orders pn , so it must be of the form C D Z.pk / for some k
1. 
The proof also shows that all quasi-cyclic groups belonging to the same prime p
are isomorphic. Since the proper subgroups of Z.p1 / are cyclic of type Z.pk /, all
of its quotient groups ¤ 0 are isomorphic to Z.p1 /.
Elementary p-Groups An elementary group is defined as a group the orders
of whose elements are square-free integers. We will see later on that such a group
decomposes into the direct sum of elementary p-groups, for different primes p.
Therefore, here we will focus our attention on elementary p-groups.
3 Fundamental Examples 17

In an elementary p-group A all the elements ¤ 0 have order p, i.e. pa D 0 for all
a 2 A. This means that the integers that are congruent mod p operate the same way
on the elements of A, so that A can be viewed as a vector space over the prime field
Z=pZ of characteristic p. Hence A admits a basis fai gi2I and every element a 2 A
can be expressed uniquely as

a D r1 ai1 C    C rn ain .ri 2 Z=pZ; ij 2 I/:

The cardinality of a basis is uniquely determined by the group: it is its dimension

as a Z=pZ-vector space. It is also determined by the order of the group.
Rational Groups Under addition, the rational numbers form a torsion-free
group, called the full rational group, denoted by Q. Like Z.p1 /, Q can also be
obtained as the union of an infinite ascending chain of cyclic groups, this time the
cyclic groups are infinite:

Z D h1i < h2Š1 i <    < hnŠ1 i < : : : :

Thus Q has a generating system fc1 D 1; c2 ; : : : ; cn ; : : : g such that

2c2 D c1 ; 3c3 D c2 ; : : : ; .n C 1/cnC1 D cn ; : : : :

It is easily seen that Q is locally cyclic in the sense that its finitely generated
subgroups are cyclic. In fact, every finite set of its elements is contained in some
hnŠ1 i; therefore, the subgroup they generate is a subgroup of a cyclic group, so
itself cyclic. Q contains numerous proper subgroups that are not finitely generated,
as the group Z.p/ of all rational numbers whose denominators are prime to the
prime p, or the group Q.p/ of rational numbers whose denominators are powers
of p. The subgroups of Q, called rational groups, are the building bricks of torsion-
free groups, and as such they are of fundamental importance in the theory of
torsion-free groups; see Chapters 12–14.
Every proper factor group Q=A of Q (i.e., A ¤ 0) is a torsion group, since
every non-zero rational number has a non-zero multiple in A. In particular, Q=Z
is isomorphic to the group of all complex roots of unity, p an isomorphism being
given by the map r C Z 7! e2ir (where r 2 Q; i D 1; and e is the base of
natural logarithm). We have Q=hri Š Q=Z for every rational number r ¤ 0, while
Q=Z.p/ Š Z.p1 /. (Z.p1 / is the p-Sylow subgroup of Q=Z.)
p-adic Integers The p-adic integers appear naturally on the scene in a variety of
ways; they play a substantial role in several branches of abelian group theory.
Let p be a prime, and Z.p/ the ring of rational numbers whose denominators are
prime to p (this is the localization of Z at p, it is a discrete valuation ring). The non-
zero ideals in Z.p/ are principal ideals generated by pk with k D 0; 1; : : : . If the set
of these ideals pk Z.p/ is declared to be a fundamental system of neighborhoods of 0,
then Z.p/ becomes a (Hausdorff) topological ring, and we may form its completion
Jp in this topology (this completion process is described in more detail for groups
18 1 Fundamentals

in Sect. 7 in Chapter 2). Jp is again a ring, called the ring of p-adic integers, whose
non-zero ideals are pk Jp with k D 0; 1; 2; : : : , and which is complete (i.e., every
Cauchy sequence is convergent) in the topology defined by its ideals.
For all practical reasons, the elements of Jp may be represented as power series
in p. Note that f0; 1; : : : ; p1g is a complete set of representatives of Z.p/ mod pZ.p/ ,
and more generally, f0; pk ; 2pk ; : : : ; .p 1/pk g is a complete set of representatives of
pk Z.p/ mod pkC1 Z.p/ . Let  2 Jp , and a0 ; : : : ; an ; : : : a sequence in Z.p/ converging
to  (dropping to a subsequence, we may accelerate convergence, and assume
an   2 pn Jp for all n 2 N). Owing to the definition of convergence, almost
all an belong to the same coset mod pZ.p/ , say, to the one represented by some
s0 2 f0; 1; : : : ; p  1g. Almost all differences an  s0 belong to the same coset of
pZ.p/ mod p2 Z.p/ , say, to the one represented by s1 p. So proceeding,  defines a
sequence s0 ; s1 p; : : : ; sn pn ; : : : , which is the same for every sequence converging
to . Accordingly, we assign to  the (formal) infinite series

 D s0 C s1 p C s2 p2 C    C sn pn C : : : .sn 2 f0; 1; : : : ; p  1g/:

Its partial sums bn D s0 C s1 p C    C sn pn .n D 0; 1; 2; : : : / form a Cauchy

sequence in Z.p/ that converges to  in Jp because   bn 2 pk Jp for n  k. From the
uniqueness of limits it follows that in this way different elements of Jp are associated
with different series, and since every infinite series s0 Cs1 pC  Csn pn C: : : defines
an element  2 Jp , we may identify the elements of Jp with the corresponding series.
Let us see how to compute in Jp . If  D r0 C r1 p C    C rn pn C : : : with
rn 2 f0; 1; : : : ; p  1g is another p-adic integer, then the sum  C  D t0 C t1 p C
   C tn pn C : : : and the product  D v0 C v1 p C    C vn pn C : : : are computed as
follows: t0 D r0 C s0  `0 p; tn D rn C sn C `n1  `n p and v0 D r0 s0  m0 p; vn D
r0 sn Cr1 sn1 C  Crn s0 Cmn1 mn p for n D 1; 2; : : : where the integers `n ; mn are
uniquely determined by the rule that all of tn ; vn are integers in the set f0; : : : ; p1g.
As to subtraction and division, note that if, e.g., s0 ¤ 0, then the negative of  is
 D .p  s0 /p C .p  s1  1/p C .p  s2  1/p2 C : : : , and the inverse  1 of 
exists if and only if s0 ¤ 0 in which case it may be computed by using the inverse
rule of multiplication.
We shall denote by Jp both the ring and the group of the p-adic integers. Qp will
denote the field of quotients of Jp (and its additive group); its elements are of the
form pk  with  2 Jp ; k 2 Z.
F Notes. It will perhaps be instructive to mention a few important applications of abelian
groups, in particular, to illustrate the groups in this section by pointing out a few applications
outside group theory.
The most widespread applications of abelian groups outside algebra are in algebraic topology
and algebraic geometry. A main step was the reinterpretation of Betti numbers as invariants of
finite groups. In algebra, the additive groups of rings and the unit groups of commutative rings
play a leading role, see Chapter 18. Elementary p-groups are the additive groups of fields of prime
characteristics. The case p D 2 plays a prominent role in computer theory as well as in coding
theory. Class groups of integral domains are abelian groups. As a matter of fact, L. Claborn [Pac.
J. Math. 18, 219–222 (1966)] proved that every abelian group occurs in this way.
3 Fundamental Examples 19

Exercises

(1) Neither Z.p1 / nor Q is finitely generated, and any finite subset of a generating
set of these groups can be dropped without spoiling the generating property.
(2) (a) Subgroups and quotient groups of locally cyclic groups are locally cyclic.
(b) The (multiplicative) group of complex roots of unity is locally cyclic.
(3) The additive group R of the real numbers is isomorphic to the multiplicative
group of the positive reals. [Hint: x 7! ex :]
(4) If B; C are subgroups of a cyclic group A, then A=B Š A=C implies B D C.
(5) In a cocyclic group all subgroups are fully invariant.
(6) An elementary group of order pr contains exactly

.pr  1/.pr1  1/    .prtC1  1/=.p  1/.p2  1/    .pt  1/

different subgroups of order pt .t

r/.
(7) Let A be a finite group. Suppose A has a set of subgroups, say B1 ; : : : ; Bn
.n > 1/, such that every non-zero element of A belongs to exactly one of the
Bi . Show that A is an elementary p-group.
(8) A subgroup M of A is called maximal if M < A and M
B < A implies
B D M.
(a) A subgroup is maximal if and only if its index is a prime number.
(b) The cyclic groups Z.pk / .k D 1; 2; : : : / as well as Jp have exactly
one maximal subgroup. Z has infinitely many maximal subgroups, while
neither Z.p1 / nor Q has any maximal subgroup.
(9) (a) The intersection of all maximal subgroups of A of the same prime index p
is equal to pA.
(b) The Frattini subgroup of A is defined as the intersection of all maximal
subgroups of A. Show that it is the intersection of the subgroups pA with
p running over all primes.
(c) Find the Frattini subgroups of Z.n/; Z.p1 /; Z; Q; Z.p/ ; and Jp .
(10) Prove the isomorphisms Q=Z.p/ Š Z.p1 /, Z.p/ =pn Z.p/ Š Z.pn /; Jp =pn Jp Š
Z.pn /.
(11) (a) Let A be an infinite group all of whose proper subgroups are finite. Then
A Š Z.p1 / for some prime p.
(b) An infinite group B satisfying B=C Š B for every proper subgroup C is
isomorphic to Z.p1 / for some prime p.
(12) (a) Find the sum, product, and quotient of the 3-adic integers  D 2 C 1  3 C
2  32 C 1  33 C 2  34 C : : : and  D 2 C 2  3 C 2  32 C 2  33 C 2  34 C : : : .
(b) Verify that 1 D .p  1/ C .p  1/p C    C .p  1/pn C : : : in Jp for
every prime p (thus  D 1 in (a)).
(13) Prove that a p-adic integer  D s0 C s1 p C    C sn pn C : : : (where sn 2
f0; : : : ; p  1g) is a unit in Jp if and only if s0 ¤ 0:
20 1 Fundamentals

4 Sets

Since the publications of Baer [6] and Kulikov [1], set theory has played a significant
role in abelian group theory. Its importance catapulted with the epoch making paper
Shelah [1], and since that it has been impossible to study abelian groups thoroughly
without a working knowledge in the theory of sets.
First a word about classes and sets. We are going to deal with both of them
(though only rarely with classes), but we avoid trying to explain what they are, as
they are most fundamental concepts. We will assume that the reader has acquired a
working knowledge about classes and sets. It is of vital importance to distinguish
between them, the rule of thumb being that a set does have cardinality, while a
proper class (i.e., a class that is not a set) is too big to be measured by any cardinal;
e.g., all the abelian groups form a proper class.
ZFC Throughout we will take for granted the ZFC axioms of Set Theory, i.e.
the axioms of Zermelo-Fraenkel along with the Axiom of Choice. They formalize
the basic set-theoretical properties that mathematicians use every day without
mentioning them explicitly. We do not list these axioms, as we do not plan to
refer to any of them explicitly, and their exact forms will never be required.
So, we just accept them with the usual understanding that, if nothing is said,
then we work in a model of ZFC. However, occasionally, we will adjoin another
set-theoretical hypothesis consistent with ZFC as an additional axiom, like the
Continuum Hypothesis (CH), the Generalized Continuum Hypothesis (GCH), or
Gödel’s Axiom of Constructibility (L). We recall that it was a major accomplishment
by P.J. Cohen to show that both CH and GCH are independent of ZFC, while K.
Gödel proved that GCH (and hence CH) holds if L is assumed.
We will frequently refer to Zorn’s lemma (known to be equivalent to the Axiom
of Choice in ZFC) as a main tool in existence proofs. In order to formulate it, recall
the definition of a partially ordered set or, in short, a poset. This is a set P equipped
with a binary relation
such that a
a (reflexivity); a
b and b
a imply a D b
(antisymmetry); a
b and b
c imply a
c (transitivity) for all a; b; c 2 P.
A subset C of P is a chain if a; b 2 C implies a
b or b
a. An element u 2 P is
an upper bound for C if c
u for all c 2 C, and P is inductive if every chain in P
has an upper bound in P. An element a 2 P is maximal in P if a
x 2 P implies
a D x. (In contrast, a maximum element b 2 P is one that satisfies the stronger
relation: x
b for all x 2 P.)
Zorn’s Lemma. A partially ordered set that is inductive contains maximal
elements.
Thus, in the applications of this lemma, we need three checks: it is a set, it is
partially ordered, and finally, it is inductive.
If S is a subset (or a subgroup) of the group A, then a subgroup B
A that is
maximal with respect to the property B \ S D ¿ (B \ S D 0) is called an S-high
subgroup. Its existence is guaranteed by Zorn’s lemma.
4 Sets 21

Ordinals and Cardinals We assume the reader is familiar with the theory of
cardinals and ordinals. The cardinality of a set X will be denoted by jXj. If  is
a cardinal,  C will stand for the first cardinal strictly larger than , and 2 for
the cardinality of the set of all subsets of a set of cardinality . We adhere to the
standard notation, and will use the symbol @ (Hebrew letter aleph) for the th
infinite cardinal. In particular, @0 stands for countable (strangely enough, @1 may
indicate finiteness), and 2@0 for the continuum. CH claims that 2@0 D @1 , and GCH
asserts that 2 D  C for each infinite cardinal . Addition and multiplication of
cardinals are trivialities:  C  D  D maxf; g if at least one of ;  is infinite.
If convenient, we will view an ordinal  as the set of ordinals less than , and
a cardinal number as an initial ordinal, i.e. an ordinal which is not equinumerous
with any smaller ordinal. The th initial ordinal is denoted by ! ; in particular, !0
(or just !) is the smallest infinite ordinal. If for an ordinal  and a cardinal , we
write  <  or  2 , we mean that the cardinality of  is less than ; in particular,
n < ! or n 2 ! is another way of saying that n is a non-negative integer. The
notation X . < / means that the sets X are indexed by ordinals  less than .
Addition and multiplication of ordinals are associative, but not commutative; we
will not need them except when decomposing an ordinal  D !˛ C n into segments
of length !; here, ˛ is an ordinal, n  0 an integer, both uniquely determined by .
Note that 2! D 2 C 2 C    C 2 C    D ! ¤ !2 D ! C !, the latter being the
second smallest limit ordinal.
An ordinal  is a successor ordinal if it is of the form  D  C 1 for some
ordinal  (called the immediate predecessor of ). If  > 0 has no immediate
predecessor, it is called a limit ordinal. Every infinite cardinal  represents a limit
ordinal. The cofinality of , cf , is the smallest cardinal  (= initial ordinal) such
that there is a subset C in  whose cardinality is  and whose supremum is ,
sup C D . A cardinal  is called regular if cf  D . Otherwise it is singular. For
every cardinal , cf  is regular. For instance, the cardinals @n for integers n  0 are
regular, while @! is a singular cardinal with cf @! D !.
A consequence of Gödel’s theorem is that it is consistent with ZFC that no regular
limit cardinal exists. On the other hand, it is not possible to prove that the hypothesis
of the existence of regular limit cardinals is consistent with ZFC.
Cubs and Stationary Sets Let  be an uncountable regular cardinal. A subset
C of  (i.e., a set of ordinals < ) is said to be unbounded if sup C D , and closed
if X  C and sup X <  imply sup X 2 C. A cub is a closed unbounded subset
of . A cub C in  is order-isomorphic to , and so it can be reindexed by using all
the ordinals < , i.e. it can be written as C D ff .˛/g˛< for an order-preserving
bijection f W  ! C. It is easily seen by using a routine back-and-forth argument:
Lemma 4.1. The intersection of two cubs in  is again a cub in .
Proof. The proof is the same as for Lemma 4.3 below, so we may skip it. 
A subset E of a regular cardinal  is said to be stationary if it intersects every
cub in . Actually, the intersection of a stationary set with a cub is again a stationary
set—this follows at once from Lemma 4.1. It is easily seen that a stationary subset
in  must have cardinality .
22 1 Fundamentals

It might be helpful to think of sets containing a cub as analogs of sets of measure

1, and stationary sets as analogs of sets of measure > 0.
Example 4.2.
(a) Cubs are obvious examples for stationary subsets.
(b) For a regular cardinal , the set E of those limit ordinals in  >  that are cofinal with  is
a stationary subset of . To show that E intersects every cub C in , let  be the supremum
of a subset of cardinality  in C. Since  is regular, cf  D , and since  C is regular,
 <  C  . Hence  2 C \ E, and E is stationary in .
Filtrations By a filtration of a set X of cardinality  (almost always uncount-
able regular) we mean a family fX˛ g˛< of subsets of X such that the following
holds:
(i) ˛
ˇ implies X˛  Xˇ (i.e., it is a well-ordered ascending chain);
(ii) jX˛ j < S
 for all ˛ < ;
(iii) Xˇ D ˛<ˇ X˛ whenever ˇ is a limit ordinal <  (we will refer to this by
sayingSthat the chain of the X˛ is continuous or smooth);
(iv) X D ˛< X˛ .
It is a routine exercise in elementary set theory to show that such filtrations
always exist. Though a set has numerous filtrations, it ought to be emphasized that,
for uncountable regular cardinals, it really does not matter which filtration is chosen,
since we have:
Lemma 4.3. If  is an uncountable regular cardinal, and if fX˛ g˛< and fY˛ g˛<
are two filtrations of the same set X of cardinality , then

E D f˛ <  j X˛ D Y˛ g

is a cub in .
Proof. It is evident that the set E is closed in . To prove that it is also unbounded,
note that for every member X˛ of the first filtration there is a member Yˇ of the
second filtration such that ˛
ˇ and X˛  Yˇ . To this Yˇ we can find an X˛1
such that ˇ
˛1 and Yˇ  X˛1 . There is a Yˇ1 with ˛1
ˇ1 and X˛1  Yˇ1 .
In this way, we obtain an increasing sequence of ordinals ˛
ˇ
˛1
ˇ1

  
˛n
ˇn
: : : (all < ) along with an increasing sequence of subsets,
X˛  Yˇ  X˛1  Yˇ1     SX˛n  Yˇn S: : : , such that the ordinal supn<! ˛n D
supn<! ˇn D satisfies X D n<! X˛n D n<! Yˇn D Y . This argument (which
is called a back-and-forth argument) establishes the existence of an index such
that ˛ < <  and 2 E. Thus E is unbounded in . 
Filters The set of subsets of a set X is called the power set of X, denoted P.X/.
As mentioned above, its cardinality is 2jXj . A filter D on a set X is a set of subsets
of X such that
(i) ¿ … D; X 2 D;
(ii) if Y 2 D and Y  Z  X; then Z 2 D; and
(iii) U; V 2 D implies U \ V 2 D.
4 Sets 23

The principal filter generated by an element x 2 X consists of all subsets of X that

contain x. If  is an infinite cardinal, we say that D is -complete if the intersection
of any <  members of D also belongs to D; if this is not the case, we say the
filter is -incomplete. An ultrafilter on X is a filter U such that for every subset
Y of X, either Y or its complement X n Y belongs to U. An ultrafilter U on X may
be interpreted as a f0; 1g-valued, finitely additive measure  on the power set P.X/
where .Y/ D 1 .Y  X/ means Y 2 U. Ultrafilters are the maximal filters on X,
and every filter is contained in an ultrafilter.
Inaccessible and Measurable Cardinals A cardinal  is said to be (strongly)
inaccessible if it is a regular uncountable cardinal such that  <  implies 2 < .
(It is thus a limit cardinal as well; in the presence of GCH, it is nothing else than a
regular limit cardinal.)
A weakly compact cardinal  is inaccessible, and has the property that, for every
-complete Boolean lattice B of subsets of a set of cardinality , every -complete
filter on B extends to a -complete ultrafilter on B. (It seems there is no simple
way to define weak compactness.) A characteristic property of weak compactness
is displayed in Lemma 4.6.
A cardinal  is called measurable if there is a non-principal @1 -complete
ultrafilter on . Equivalently, a set X of cardinality  admits a countably additive
measure  which assumes only the values 0 and 1, and which satisfies .X/ D 1 and
.x/ D 0 for all singletons fxg  X. It is straightforward to see that if a cardinal is
measurable, then so are all larger cardinals, and if there exists a measurable cardinal
at all, then there is a smallest one. Assuming V = L (see next), no measurable
cardinal exists.
Constructible Universe The constructible universe is a model of set theory
obtained from ZFC by adjoining Gödel’s Axiom of Constructibility. This axiom
does not allow us to form all subsets of an already existing set, but only those which
can be defined in terms of a selection principle. This model is denoted by L, and
if we assume the Axiom of Constructibility, then we can indicate this briefly by
writing V = L (V being the standard notation for the model we work in). Gödel
proved that V = L is consistent with ZFC.
An important consequence of Gödel’s axiom is the Diamond Principle; see R.
Jensen [Ann. Math. Logic 4, 229–308 (1972)].
Diamond Principle } . Let  be an uncountable regular cardinal, and E a
stationary subset of . Given a filtration fX˛ g˛< of a set X of cardinality , there is
a family fS˛ g˛2E of sets such that S˛  X˛ , and for any subset Y of X, the set

EY D f˛ 2 E j Y \ X˛ D S˛ g

is a stationary subset of .
24 1 Fundamentals

This is an amazing prediction principle: it says that no matter how we choose a

subset Y, Y will meet stationarily many times the preassigned subsets X˛ exactly in
the predicted subsets S˛ .
From the stated form of this principle we can derive another version of }; the
following proof is standard. ( between sets denotes cartesian product.)
Lemma 4.4. .} / Let E be a stationary subset of an uncountable regular cardinal
, and fX˛ g˛< a filtration of a set X of cardinality . Let Y be any countable set.
There is a family fg˛ g˛2E of functions

g ˛ W X˛ ! Y  X˛

such that, for any function g W X ! Y  X; the set

Eg D f˛ 2 E j g  X˛ D g˛ g

is stationary in .
Proof. Define X 0 D X  Y  X and X˛0 D X˛  Y  X˛ .˛ < /, and apply the
Diamond Principle to this filtration of X 0 to conclude the existence of subsets S˛ of
X˛  Y  X˛ .˛ 2 E/ with the property stated above. If for an ˛; S˛ is the graph of
a function X˛ ! Y  X˛ , then define g˛ to be this function. Otherwise, define g˛ to
be any function X˛ ! Y  X˛ whatsoever.
Now pick any function g W X ! Y  X, and let S denote its graph viewed as a
subset of X  Y  X. From the Diamond Principle we obtain that the set Eg of ˛’s
with g  X˛ D g˛ is stationary in , as desired. 
Another result of Jensen’s which will be needed is as follows.
Lemma 4.5 (V = L). Let  be a regular cardinal which is not weakly compact.
There exists a stationary subset E of  which consists of limit ordinals cofinal with
! such that, for every limit ordinal  < ; the set  \ E is not stationary in . u
t
In contrast, for weakly compact cardinals the following lemma holds.
Lemma 4.6. Let  be a weakly compact cardinal, and E a stationary subset of .
There is a regular cardinal  <  such that the set  \ E is stationary in . t
u
We now consider another set-theoretic hypothesis which is not a consequence of,
but is consistent with, ZFC (but inconsistent with V = L). In order to formulate it,
we require a couple of definitions.
Let P be a partially ordered set under a binary relation
. We say the elements
p; q 2 P have an upper bound if some r 2 P satisfies both p
r and q
r.
A subset D of P is directed (upwards) if every pair of elements of D has an upper
bound in D. A subset C of P is cofinal in P if for every p 2 P there is a c 2 C such
that p
c.
Martin’s Axiom (MA). Let P be a partially ordered set such that every subset
of P no two elements of which have an upper bound in P is countable .this is called
5 Families of Subgroups 25

the countable antichain condition/. For every family fCi gi2I of cofinal subsets of
P .where jIj < 2@0 / there is a directed subset D of P that intersects every Ci .
R. Solovay and S. Tenenbaum [Ann. Math. 94, 201–245 (1971)] proved that MA
is consistent with ZFC and the negation of the Continuum Hypothesis (: CH), i.e.
ZFC + MA + .@1 < 2@0 ) is consistent (provided so is ZFC).
F Notes. The above incomplete notes on set theory are intended to recap needed facts only,
not to serve as a systematic introduction. For more details, see, e.g., Eklof [5]. Let us point out for
information that in some cases Martin’s Axiom is not a strong enough hypothesis to prove a result;
a stronger version is Shelah’s Proper Forcing Axiom (not to be used in this volume). A powerful
prediction principle, called Black Box, was developed by Shelah; for an algebraic version, see
Göbel–Wallutis [1].
For convenience, we will often assume V = L even if the Diamond Principle or something
weaker would suffice.

Exercises

(1) Let A be a group of cardinality   @0 . The set of finitely generated subgroups

of A has cardinality .
(2) Let B; X be subgroups of A. There exists a subgroup C of A such that (i) B
C;
(ii) C \ X D B \ X; (iii) if C
C0
A and C0 \ X D B \ X, then C D C0 .
(3) Let H be a fully invariant subgroup of A, and S a subset of A such that
H \ S D ¿. There exists a fully invariant subgroup G of A that is maximal
with respect to the properties: (i) H
G; (ii) G \ S D ¿:
(4) Assume } for an uncountable regular , and let E be a stationary subset of .
For sets X; Y of cardinality  with filtrations fX˛ g˛< and fY˛ g˛< , there exists
a set ff˛ W X˛ ! Y˛ g˛< of functions such that for any function f W X ! Y, the
set f˛ 2 E j f  X D f˛ g is stationary in .

5 Families of Subgroups

In several occasions we will need a collection of subgroups to characterize a group

property or to prove a theorem. The simplest example for such a collection is an
ascending chain of subgroups indexed by the natural numbers, like

0 D A0 < A1 <    < An < : : :

whose union is the group A, i.e. A D [n<! An .

Transfinite Chains A more general version is a transfinite sequence or chain

0 D A0 < A1 <    < A < : : : . < /

26 1 Fundamentals

of subgroups, indexed by the ordinals less than an ordinal , which is called the
length of the chain. Here we usually assume that the chain is continuous or smooth
to mean that A D [< A holds for limit ordinals  <  (so that there is no hole
in the chain). In this case we talk about a smooth or a continuous well-ordered
ascending chain.
In the preceding section, we have introduced set filtrations. More often, we shall
deal with group filtrations. The definition is the same: for an infinite cardinal , by
a -filtration of a group A is meant a smooth chain fA g < of subgroups with
union A, subject to the condition that jA j <  for all  < . The -filtrations are
especially useful for uncountable regular cardinals . In this case, Lemma 4.3 is
applicable, i.e. the intersection of two -filtrations is again one.
Sometimes we shall need a pure -filtration, where the subgroups are assumed
to be pure subgroups.
G- and H-Families We will encounter collections of subgroups that are no
longer linearly ordered. Their significance is greatly enhanced by the fact that they
are not only useful in some proofs, but also instrumental in characterizing groups
with certain properties. Following P. Hill, we define various families of subgroups.
Let  D @ .  1/. (Recall @1 is to be interpreted as ‘finite.’) By an H./-
family of subgroups of A is meant a collection H of subgroups of A satisfying the
following conditions:
H1 . 0; A 2 H; P
H2 . H is closed under unions, i.e. Ai 2 H .i 2 I/ implies i2I Ai 2 H for any index
set I;
H3 . if C 2 H, and X is any subset of A of cardinality
, then there is a subgroup
B 2 H that contains both C and X, and is such that jB=Cj
.
It is easily checked that in the presence of H2 , it suffices to assume H3 only for
C D 0.
A G./-family G is defined similarly with H2 replaced by the following weaker
condition:
G2 . G is closed under unions of chains.
Obviously, every H./-family is a G./-family, but the converse fails in general,
see Example 5.2 below. Sometimes we will need the rank versions of these families
(for the definition of rank, see Sect. 4 in Chapter 3). H  ./- and G ./-families
are defined similarly for torsion-free groups A: in these cases the subgroups in the
families are required to be pure subgroups (Sect. 1 in Chapter 5) and in condition
H3 , ‘rank’ is to be used in place of ‘cardinality.’
Example 5.1. Let X be a generating system of A. If we let Y run over all subsets of X, then the
subgroups hYi generated by the Y form an H.@0 /-family of subgroups.
Example 5.2. This example relies on the concept of pure subgroup. We claim that every torsion-
free group A has a G.@0 /-family of pure subgroups. In fact, select a maximal independent set X in
A, and for a subset Y of X, let AY denote the smallest pure subgroup of A that contains Y. Then the
set of all AY is a G.@0 /-family. However, this is in general not an H.@0 /-family, since the sum of
pure subgroups need not be pure.
5 Families of Subgroups 27

We shall see that the existence of families of subgroups of various kinds has a
strong influence on the group structure. Actually, it works in both directions: a global
property of groups may turn out to be equivalent to having a family of a certain kind
of subgroups.
A notable special case is when the group A (of cardinality ) happens to be a
direct sum of subgroups. If A D ˚i2I Ai (jAi j < ) is a direct sum decomposition
(see Sect. 1 in Chapter 2), then the standard way of defining an H./-family of
summands in A is to consider the set of all partial summands in this decomposition:
BJ D ˚i2J Ai with J ranging over all subsets of I.
In the next proof, we use again a back-and-forth argument.
Lemma 5.3. For an uncountable regular , the intersection of at most  families of
H./- or G./-families of subgroups is again a family of the same kind.
Proof. We give a detailed proof for the intersection of two H./-families, the
general case follows the same pattern. Let H1 and H2 denote two H./-families of
subgroups of A, and let H be their intersection. Conditions H1 and H2 are evidently
satisfied for H, so we proceed to the verification of H3 . Given any B 2 H and a set
X with jXj
, we can find a B1 2 H1 such that B [ X
B1 and jB1 =Bj
,
and then a C1 2 H2 such that B1
C1 and jC1 =Bj
. Continuing, there is a
B2 2 H1 satisfying C1
B2 and jB2 =Bj
, and then again a C2 2 H2 with
B2
C2 and jC2 =Bj
, etc. We obtain an increasing sequence of subgroups
B
B1 S
C1
   S

Bn
Cn
: : : where Bn 2 H1 and Cn 2 H2 . If we
set C D n<! Bn D n<! Cn , then it is clear that C 2 H, and we also have the
required jC=Bj
. 
Summands of Families The following lemma deals with families in direct
summands.
Lemma 5.4. Assume A D B ˚ C has an H./-family A of subgroups .  @0 /.
Then the B-components B0 of those A0 2 A which decompose as A0 D B0 ˚ C0 with
C0
C form an H./-family B of subgroups in B.
Proof. We prove that the set B of the B-components of the groups A0 satisfies H3 .
Let B0 2 B and X a subset of B of cardinality
. There are A0 2 A such that
A0 D B0 ˚ C0 with C0
C, and A1 2 A such that A0 [ X  A1 and jA1 =A0 j
.
There are subgroups B1
B; C1
C such that B0
B1 ; C0
C1 , A1
B1 ˚ C1
with jB1 =B0 j
; jC1 =C0 j
: We can now find A2 2 A with B1 ˚ C1
A2 and
jA2 =A1 j
. Proceeding in a similar fashion, we obtain sequences An 2 A; Bn
B;
and Cn
C .n < !/ with the following properties:

An
Bn ˚ Cn
AnC1 ; jAnC1 =An j
:

Then A0 D [n An 2 A will satisfy A0 D .A0 \ B/ ˚ .A0 \ C/, whence A0 \ B 2 B is

immediate. To complete the proof, we note that B0 [X  A0 \B and j.A0 \B/=B0 j

jA0 =A0 j
. 
28 1 Fundamentals

If  is an infinite cardinal, and if we have a G./-family G of subgroups, then it is

easy to extract from it a -filtration. The procedure is obvious: we start with A0 D 0,
and if we have A 2 G, then for A C1 we pick any B 2 G with jB=A j
, and take
unions at limit ordinals. We have to reach the whole group for obvious cardinality
reason.
H-Family from a Chain The following ingenious lemma also holds when the
group G is torsion-free and ‘cardinality’ is replaced by ‘rank.’
Theorem 5.5 (Hill [16]). Suppose the group G is the union of a smooth chain

0 D G0 < G 1 <    < G ˛ < : : : .˛ < / (1.1)

of subgroups such that for each ˛ C 1 < ,

G˛C1 D G˛ C A˛

holds for some subgroup A˛ of cardinality

 D @ where   1. Then G
admits an H./-family C of subgroups such that every C 2 C has a smooth chain
of subgroups Cˇ with union C whose factors are isomorphic to factors in the chain
(1.1), and satisfy CˇC1 D Cˇ C A˛ for some ˛ < .
If the groups G˛ are pure in G, then the members of C can be chosen to be pure.
P
Proof. To start with, observe that hypotheses imply Gˇ D ˛<ˇ A˛ for all ˇ < .
Thus each g 2 G is contained in the sum of a finite number of A˛ . A subset S of 
will be called blocked if every ˇ 2 S satisfies
X
Gˇ \ A ˇ
A˛ :
˛2S;˛<ˇ
P
For a blocked subset S of , we will set G.S/ D ˛2S A˛ , and claim that

C D fG.S/ j S a blocked subset in g

is a desired H./-family in G. The proof is completed in several steps.

1ı Unions of blocked
S subsets of  are blocked. Let Si .i 2 I/ be blocked subsets
of , and ˇ 2 i2I Si . Then ˇ 2 Sj for some j 2 I, and Gˇ \ Aˇ is evidently
S in the sum of the A˛ for ˛ < ˇ with all ˛ 2 Sj , and a fortiori with all
contained
˛ 2 i2I Si .
2ı A subset of  of cardinality
 is contained in a blocked subset of  of cardina-
lity
. By 1ı , it suffices to verify this for a single ˇ < . We induct on ˇ.
If ˇ D 0, then the claim is true, as f0g is evidently a blocked subset of . The
subgroup Gˇ \Aˇ is of cardinality
, so it contains a generating set fai j i 2 Ig
with jIj
. The ai ’s are in Gˇ , so each is contained in a finite sum of the A˛ ’s
with ˛ < ˇ. By induction, there is a blocked subset S0   such that jS0 j
,
and all the ai are contained in G.S0 /. There is no loss of generality in assuming
that S0  ˇ, since otherwise S0 can be replaced by the blocked subset S0 \ ˇ. To
5 Families of Subgroups 29

show that S D S0 [fˇg is a blocked subset of , it suffices to check the definition

for ˇ. G.S0 / contains all the ai , hence it contains Gˇ \ Aˇ as well.
ı
3 C is an H./-familyPof subgroups inS G. Obviously,
 both ¿ and  are blocked
subsets of . Since i2I G.Si / D G i2I S i holds for blocked subsets Si  ,
C is closed under arbitrary unions. If X is a subset of G of cardinality
, then
there is a blocked subset S   such that jSj
; X  G.S/, where G.S/ has
cardinality
.
4ı G.S/ has
P chains as desired. Let S be a blocked subset of , and form the chain
Cˇ D ˛2S;˛<ˇ A˛ for all ˇ 2 S. This is a smooth chain of subgroups of C D
P P 
G.S/ with union C. As Gˇ \Aˇ
˛2S;˛<ˇ A˛
Gˇ implies ˛2S;˛<ˇ A ˛ \
Aˇ D Gˇ \ Aˇ , we clearly have
0 1,0 1
X X
CˇC1 =Cˇ D @ A˛ A @ A˛ A Š Aˇ =.Gˇ \ Aˇ / Š GˇC1 =Gˇ :
˛2S;˛ˇ ˛2S;˛<ˇ

The second part of the claim is clear.

5ı The G.S/ are pure in G if the G˛ are pure in G. Let g 2 G.S/, thus g D a˛1 C
   C a˛k with a˛j 2 A˛j , where ˛1 <    < ˛k in S. If nx D g for some n 2 Z
and x 2 G, then by purity x 2 G˛k C1 can be assumed, so let x D y C b˛k
with y 2 G˛k ; b˛k 2 A˛k . Hence
P ny D nx  nb˛k D a˛1 C P   C a˛k  nb˛k , so
a˛k  nb˛k 2 G˛k \ A˛k
˛2S;˛<˛k A˛ . We obtain ny 2 ˛2S;˛<˛k A˛ , whence
an induction on the largest index ˛k implies that we may pick y 2 G.S/; then
x 2 G.S/ as well. 
Almost Disjoint Subsets The following two lemmas will be needed in later
proofs. They deal with the existence of almost disjoint sets. Two infinite sets are
called almost disjoint if their intersection is finite. In the following lemmas, the
symbol  stands for the initial ordinal of the power 2@0 .
Lemma 5.6. There is a set fS j  2 g of countable sets S that are pairwise
almost disjoint.
Proof. For an irrational real number r let Sr be a sequence of rational numbers with
limit ˛. It is clear that Sr \ Ss is finite whenever r ¤ s. 
The following generalization of Lemma 5.6 will be needed in Sect. 6 in
Chapter 13. We state it without proof.
Lemma 5.7 (W. Sierpinski). Let  be an infinite cardinal. There exists a set C of
subsets of  such that
(i) jXj D  for each X 2 C;
(ii) if X ¤ Y in C, then jX \ Yj < ; and
(iii) jCj > . t
u
30 1 Fundamentals

For the proof of Proposition 8.7 in Chapter 14, almost disjoint sets seems
insufficient, a collection of disjoint sets is needed. We establish the existence of two
collections † and †0 (both of the power of the continuum) of subsets of a countable
set S, such that for each fixed  < , fS \ S0 j  < g is a collection of almost
disjoint sets, where S 2 †; S0 2 †0 .
Lemma 5.8 (Dugas–Thomé [1]). Given a countably infinite set S, there exist two
families,

† D fS j  < g and †0 D fS0 j  < g;

of almost disjoint subsets of S such that

S S
(i)  < S D S and  < S0 D S;
(ii) for all ;  < , the intersection S \ S0 is infinite.
Proof. For n < !, let Fn denote the set of all functions S f from the set nN D
f0; 1; : : : ; n  1g to the set f0; 1g. Evidently, the set S D n<! .Nn  Fn / is countable.
For a map  W ! ! f0; 1g, we define

S D f.Nn;   nN / j n < !g:

Then the set † D fS j  W ! ! f0; 1gg is an almost disjoint family of countable
subsets of S; its size is 2@0 .
Next we select an almost disjoint family fT j  < g of subsets of ! (see
Lemma 5.6), and define the countable sets

S0 D f.Nn; f / j n 2 T ; f 2 Fn g:

Then for  ¤  2 , the intersection

S0 \ S0 D f.Nn; f / j n 2 T ; f 2 Fn g \ f.Nn; f / j n 2 T ; f 2 Fn g D

D f.Nn; f / j n 2 T \ T ; f 2 Fn g

has to be finite, because T \T is a finite set. Hence †0 D fS0 j  < g is an almost
disjoint family of countable subsets of S. On the other hand, for any  W ! ! f0; 1g
and any  < , the intersection

S \ S0 D f.Nn;   nN / j n < !g \ f.Nn; f / j n 2 T ; f 2 Fn g D

D f.Nn;   nN / j n 2 T g

is infinite. 
6 Categories of Abelian Groups 31

F Notes. A word of warning is in order. We follow the customary definitions for the H./-
and G./-families which do not confirm with the usual practice involving a cardinal  (that would
be <  rather than our  ).
Hill prefers to use the term Axiom-3 family for a G.@0 /-family; the name is chosen to remind
that having such a family is a condition that is weaker than, but similar to, the first and the second
axioms of countability hypotheses widely used in topology.

Exercises

(1) Give a detailed proof of Lemma 5.3 for the intersection of  families.
(2) A group of cardinality  C has a -filtration.
(3) If fA g < is a -filtration of A, and C is a cub in  (an infinite cardinal), then
fA g 2C is also a -filtration of A.
(4) Let B be a subgroup in A. If B has a G./-family B of subgroups, then there is
a G./-family A in A such that B D fB \ X j X 2 Ag.
(5) Suppose ˛ W A ! B is an epimorphism.
(a) ˛A is an H./-family in B if A is an H./-family of subgroups in A.
(b) If B has a G./- (or H./-)family B of subgroups, then there is a G./-
(H./-)family A in A such that ˛A D B.
(6) Prove Lemma 5.6 as follows: each Sr (r is a positive real number) is a set of
lattice points in the first quarter of the plane. For r, we let Sr consist of the
lattice points whose distances from the line y D rx are less than 1.

6 Categories of Abelian Groups

In the theory of abelian groups it is often convenient and more suggestive to express
situations in the categorical language. In fact, categories and functors appear to be
the right unifying concepts, and the categorical point of view frequently provides
a better understanding. We survey some basic facts on categories and exhibit some
important concepts connected to them, confining ourselves to those that will be
needed in this volume. We are not borrowing sophisticated results from category
theory, we mostly use it as an appropriate language.
Categories A category C is a class of objects A; B; C; : : : and morphisms
(sometimes called arrows) ˛; ˇ; ; : : : satisfying the following axioms:
C1. With each ordered pair of objects, A; B 2 C, there is associated a set
MorphC .A; B/ (or simply Morph.A; B/) of morphisms in C such that every
morphism in C belongs to exactly one Morph.A; B/. In case ˛ 2 Morph.A; B/,
we write ˛ W A ! B and call ˛ a map from A to B. A is the domain and B the
range or codomain of ˛.
C2. Composition. With ˛ 2 Morph.A; B/ and ˇ 2 Morph.B0 ; C/ there is associated
a unique element in Morph.A; C/, called their product ˇ˛ if and only if
B D B0 .
32 1 Fundamentals

C3. Identity. For every object A, there is a morphism 1A 2 Morph.A; A/, the
identity morphism of A, such that 1A ˛ D ˛ and ˇ 1A D ˇ provided the
products are defined.
C4. Associativity. Whenever the products are defined, the composition is associa-
tive: .ˇ˛/ D . ˇ/˛.
One verifies at once that the identity 1A is uniquely determined by the object A.
Hence there is a bijection between the objects A and the identities 1A . In view of
this, categories can be (and often are) defined in terms of morphisms only.
Two objects, A; C 2 C, are said to be isomorphic if there exist morphisms, ˛ W
A ! C and W C ! A, such that ˛ D 1A and ˛ D 1C . Manifestly, isomorphism
is an equivalence relation in C.
In the category Ab of abelian groups, the objects are the abelian groups, and the
morphisms are the homomorphisms between them. In particular, Morph.A; B/ D
Hom.A; B/ (see Chapter 7), and the identities are the identity automorphisms of the
groups.
The category D is a subcategory of C if the objects of D are objects in C, and
the morphisms of D are morphisms in C. D is a full subcategory of C if it is a
subcategory with MorphD .A; B/ D MorphC .A; B/ for all A; B 2 D. The cartesian
product C D of two categories consists of the objects .C; D/ and morphisms . ; ı/
with C; 2 C, and D; ı 2 D, where the morphisms act coordinate-wise.
Functors If C and D are categories, then a covariant functor F W C ! D
assigns to each object A 2 C an object F.A/ 2 D, and to each morphism ˛ W A ! B
a morphism F.˛/ W F.A/ ! F.B/ such that
(i) if a product ˇ˛ is defined in C, then F.ˇ/F.˛/ is defined in D and F.ˇ˛/ D
F.ˇ/F.˛/;
(ii) F carries the identity 1A to the identity 1F.A/ .
Thus a covariant functor preserves domains, codomains, products, and identities.
The identity functor 1C , defined by 1C .A/ D A; 1C .˛/ D ˛ for all A; ˛ 2 C, is a
covariant functor of C into itself. A contravariant functor G W C ! D is defined
similarly by reversing arrows: it assigns an object G.A/ 2 D to every object A 2 C,
and a morphism G.˛/ W G.B/ ! G.A/ to every morphism ˛ W A ! B. It is subject
to the conditions: G.1A / D 1G.A/ ; G.ˇ˛/ D G.˛/G.ˇ/ for ˛; ˇ 2 C whenever ˇ˛
is defined. The unqualified term ‘functor’ will mean ‘covariant functor.’
Example 6.1. Let t W Ab ! T be the functor on the category of abelian groups to the subcategory
T of torsion abelian groups such that for A 2 Ab, t.A/ is the torsion subgroup of A, and for
˛ W A ! B in Ab, t.˛/ D ˛  t.A/ W t.A/ ! t.B/:
Example 6.2. For a positive integer n, let the functor Mn W Ab ! Ab assign to A 2 Ab the
subgroup nA, and to ˛ W A ! B the induced homomorphism ˛  nA W nA ! nB.
Example 6.3.
(a) Let Bn be the full subcategory of Ab consisting of the n-bounded groups for a fixed n 2 N.
F W Ab ! Bn is a functor assigning to A 2 Ab the subgroup AŒn
2 Bn , and to ˛ W A ! B the
restriction ˛  AŒn
.
6 Categories of Abelian Groups 33

(b) We get another functor G W Ab ! Bn by letting G.A/ D A=nA and G.˛/ W aCnA 7! ˛aCnB
for ˛ W A ! B.
Example 6.4. In this example, F is the full subcategory of Ab that consists of the torsion-free
groups. The functor F W Ab ! F assigns to a group A the factor group A=t.A/, and to ˛ W A ! B
the induced map F.˛/ W a C t.A/ 7! ˛a C t.B/.

Suppose C; D; E are categories, and F W C ! D; GW D ! E are functors. The

composite GF is a functor from C to E where GF.A/ D G.F.A// and GF.˛/ D
G.F.˛// for all A; ˛ 2 C. Clearly, GF is covariant if both F and G are covariant or
both are contravariant, and is contravariant if one of F; G is covariant and the other
is contravariant.
We shall have occasions to consider functors in several variables, covariant in
some of their variables, and contravariant in others. For instance, if C; D; E are
categories, then a bifunctor F W C  D ! E, covariant in C and contravariant in
D, assigns to each pair .C; D/ 2 C  D of objects an object F.C; D/ 2 E, and
to each pair ˛ W A ! C; ˇ W B ! D of morphisms ˛ 2 C; ˇ 2 D a morphism
F.˛; ˇ/ W F.A; D/ ! F.C; B/ in E such that

F. ˛; ıˇ/ D F. ; ˇ/F.˛; ı/ and F.1C ; 1D / D 1F.C;D/ (1.2)

whenever ˛; ıˇ are defined. The quintessence of these relations is made clear in

the commutativity of the diagram
F (α,1D )
F (A, D) −−−−−→ F (C, D)
⏐ ⏐
⏐
F (1A ,β)
⏐F (1 ,β)
 C
F (α,1B )
F (A, B) −−−−−→ F (C, B)

In the theory of abelian groups, one encounters almost exclusively additive

functors, i.e. functors F satisfying F.˛ C ˇ/ D F.˛/ C F.ˇ/ for all morphisms
˛; ˇ whenever ˛ C ˇ is defined. For an additive functor F, one always has F.0/ D 0
where 0 stands for the zero group or for the zero homomorphism. Also F.n˛/ D
nF.˛/ holds for every n 2 Z.
Exact Sequences One of the fundamental questions concerning functors in
abelian groups is to find out how they behave for subgroups and quotient groups.
This is most efficiently studied in terms of exact sequences. If F is a covariant func-
˛ ˇ
tor from a subcategory C of Ab into the category D, and if 0 ! A!B!C ! 0
is an exact sequence in C, then F is called left or right exact according as

F.˛/ F.ˇ/ F.˛/ F.ˇ/

0 ! F.A/!F.B/!F.C/ or F.A/!F.B/!F.C/ ! 0
34 1 Fundamentals

is exact in D; F is exact if it is both left and right exact. For a contravariant F, the
displayed sequences are replaced by
F.ˇ/ F.˛/ F.ˇ/ F.˛/
0 ! F.C/!F.B/!F.A/ and F.C/!F.B/!F.A/ ! 0;

respectively. Subfunctors of the identity (assigning a subgroup to every group) are

always left exact, while quotient functors (assigning some factor group) are right
exact.
Assume F and G are covariant functors C ! D. By a natural transformation
ˆ W F ! G is meant a function assigning to each object A 2 C a morphism
ˆA W F.A/ ! G.A/ in D in such a way that for all morphisms ˛ W A ! B in C,
the following diagram in D commutes:

F (α)
F (A) −−−−→ F (B)
⏐ ⏐
⏐
ΦA 
⏐Φ
 B
G(α)
G(A) −−−−→ G(B)

In this case, we also say that ˆA is a natural morphism from F.A/ to G.A/. If ˆA is
a bijection (isomorphism) for every A 2 C, then ˆ is called a natural equivalence
(natural isomorphism). (It is important to understand that natural isomorphism is
far more than just an isomorphism between two groups: it is an individual case of a
functorial isomorphism.)
In general, two categories, C and D, are called equivalent if there are functors
F W C ! D and G W D ! C such that FG is naturally equivalent to 1D , and GF is
naturally equivalent to 1C .
Example 6.5. The skeleton of a category C is the category S such that (i) S is a full subcategory
of C ; (ii) every object A 2 C is isomorphic to a unique object S 2 S . The embedding functor
F W S 7! S from S into C , and the assignment functor G W A 7! S define an equivalence between C
and S .
In particular, if we select an abelian group from each isomorphy class in Ab, then the full
subcategory whose objects are the selected groups is a skeleton of Ab.
Example 6.6. For a fixed group G, the assignments FG W G ˚ A 7! A and . ; ˛/ 7! ˛ for all A; ˛
in Ab and 2 End G define a functor. Furthermore, each homomorphism  W G ! H between
groups gives rise to a natural transformation ˆ W FG ! FH .

Anticipating the functor Hom that will be discussed in Chapter 7, we can explain
what we mean by adjoint functors.
Let A and C be two categories, and F W A ! C; G W C ! A functors between
them. The functors are called adjoint (more precisely, F is the left adjoint of G, and
G is the right adjoint to F) if we have

HomA .A; G.C// Š HomC .F.A/; C/ 8A 2 A; C 2 C; (1.3)

where the isomorphism is natural both in A and in C.

6 Categories of Abelian Groups 35

Example 6.7. Consider the categories Ab and Bn of abelian groups and n-bounded abelian groups,
respectively. Define F W Ab 7! Bn via A 7! A=nA (canonical map) and G W Bn 7! Ab via
G W CŒn
7! C (injection). Then F is the left adjoint of G as is demonstrated by Hom.A; CŒn
/ Š
Hom.A=nA; C/.
F Notes. Category theory is the product of the twentieth century, initiated by S. Eilenberg and
S. Mac Lane. The revolutionary new idea was to shift the emphasis from the objects (like groups,
rings, etc.) to the maps between them, and instead of focusing on one particular object, the entire
class of similar objects became the subject of study. This point of view has proved very fruitful, it
has penetrated into many branches of mathematics. The category of abelian groups has been under
strict scrutiny of category theorists.
Functorial subgroups were studied by B. Charles. Let F W Ab ! Ab be a functor such that (i)
F.A/  A, and (ii) if  W A ! B is a morphism in Ab, then F./ D   F.A/, for all groups A.
Then F.A/ is a functorial subgroup of A. Nunke calls such a functor subfunctor of the identity;
he develops a more extensive theory. The torsion subgroup and its p-components are prototypes of
functorial subgroups.

Exercises

(1) Both the torsion groups and the torsion-free groups form a subcategory in Ab.
The same holds for p-groups.
(2) A category with a single object is essentially a monoid (i.e., a semigroup with
identity) of morphisms.
(3) Give a detailed proof of our claim that the cartesian product of two categories
is again a category.
(4) Prove that the following is a category: the objects are commutative squares of
the form

A1 −−−−−→ A2
⏐ ⏐
⏐ ⏐
 

A3 −−−−−→ A4

where Ai are groups. The morphisms are quadruples .˛1 ; ˛2 ; ˛3 ; ˛4 / of group

maps (˛i W Ai ! Bi ) making all the arising squares between the objects
commutative.
(5) The composite of two natural transformations is again one. Does this composi-
tion obey the associative law?
(6) Prove that, for an integer n > 0, F W A ! A=nA with F.˛/ W a C nA 7! ˛a C nB
for ˛ W A ! B is a functor from Ab to Bn .
(7) We get a functor U W Ab ! Ab by assigning the Ulm subgroup A1 to a group
A, and the restrictions to the Ulm subgroups of the homomorphisms.
(8) The equivalence of categories is a reflexive, symmetric, and transitive relation.
36 1 Fundamentals

7 Linear Topologies

In abelian groups, topology can be introduced in various ways which are natural
in one sense or another. The importance of certain topologies will be evident from
subsequent developments, especially when completeness will be discussed.
Linear Topologies The most important topologies on groups to be considered
here are the linear topologies; this means that there is a base (fundamental system)
of neighborhoods about 0 which consists of subgroups such that all the cosets of
these subgroups form a base of open sets for the topology. A more formal definition
for linear topologies can be given as follows.
Let u be a filter in the lattice L.A/ of all subgroups of A. u defines a topology on
A, if we declare the set of subgroups U 2 u to be a base of open neighborhoods
about 0, and for every a 2 A, the cosets a C U .U 2 u/ as a base of open
neighborhoods of a. Since the intersection of two cosets, a C U and b C V, is either
vacuous or a coset mod U \ V (this subgroup also belongs to u whenever U; V 2 u/,
all open sets will be unions of cosets a C U with a 2 A; U 2 u. The continuity
of the group operations is obvious from the simple observation that x  y 2 a C U
implies .x C U/  .y C U/  a C U: Thus A is always a topological group under
the arising topology, which may be called the u-topology of A; .A; u/ will denote A
as a topological group equipped with the u-topology.
Note the following simple facts on u-topologies.
(A) The u-topology on a group A is discrete
T exactly if f0g 2 u, indiscrete if u D
fAg, and Hausdorff if and only if U2u U D 0.
(B) Open subgroups are closed. Indeed, the complement of an open subgroup B is
open as the union of the open cosets a C B .a … B/.
(C) If the u-topology of A is Hausdorff, then it makes A into a 0-dimensional
topological group. In fact, the subgroups U 2 u are by (B) both open and closed.
(D) The closure
T of a subgroup B of A in the u-topology of A is given by the formula
B D U2u .B C U/:
Some topologies satisfy the first axiom of countability, i.e. there is a countable
base of neighborhoods about 0. If fUn j n < !g is such a system of neighborhoods,
then fU0 \    \ Un j n < !g is also one; thus, in this case we may assume without
loss of generality that the sequence of the Un is decreasing. These topologies—
if Hausdorff—are exactly the metrizable linear topologies; in fact, a metric can
be introduced by setting kak D en .e denotes the base of the natural logarithm)
whenever a 2 A belongs to Un n UnC1 .
The u-topology is considered finer than the v-topology, and the v-topology
coarser than the u-topology, if v  u. Thus the discrete topology is the finest,
and the indiscrete topology is the coarsest of all topologies on any group.
A compact group is a topological group whose topology is compact.
The following special topologies are significant, especially the Z-adic topology
that will be used constantly. In Examples 7.1–7.4, every group homomorphism is
continuous.
7 Linear Topologies 37

Example 7.1. The Z-adic topology on a group A is defined by letting fnA j n 2 Ng be a base of
neighborhoods about 0. This is a u-topology, where u consists of all U  A such that A=U is a
bounded group. This topology is Hausdorff if and only if the first Ulm subgroup A1 of A vanishes.
A subgroup G of A is closed exactly if the first Ulm subgroup of A=G is 0.
Example 7.2. In the p-adic topology (for a prime p) the subgroups pk A .k < !/ are declared
to form a base of neighborhoods about 0. This is likewise a u-topology, with u consisting of all
U  A such that A=U is a bounded p-group.
Example 7.3. In order to define the Prüfer topology, we choose the filter u to consist of all U  A
such that A=U satisfies the minimum condition on subgroups (see Sect. 5 in Chapter 4). This is
always a Hausdorff topology in which all subgroups are closed.
Example 7.4. In the finite index topology, the subgroups of finite indices constitute a base of
neighborhoods of 0; equivalently, u consists of the subgroups of finite indices in A. This is coarser
than both the Z-adic and the Prüfer topologies.
Example 7.5. The following groups are usually viewed as being equipped with the interval
topology (which is not a linear topology): the multiplicative group T of complex numbers of
absolute value 1, and the additive group R of reals.

Let .A; u/ be a topological group, and B a subgroup of A. The induced topology

on B has the subgroups B\U with U 2 u as a base of neighborhoods of 0. There is an
induced topology on the factor group A=B as well, where the subgroups .B C U/=B
with U 2 u are the neighborhoods of B=B. This topology is Hausdorff if and only if
B is closed, and discrete when B is open.
We will need later the following simple lemma.
Lemma 7.6. Assume .A; u/ is a Hausdorff topological group, and B is a subgroup
of A. If mB D 0 for some 0 ¤ m 2 Z, then also mB D 0 for the closure B of B in
the u-topology of A.
Proof. For an a 2 A, a 2 B means that a 2 B C U for all U 2 u. Then ma 2 mU

U, thus ma 2 \U2u U D 0: 

Q If f.Ai ; ui / j i 2 Ig is a family of topological groups, then their cartesian product

Ai is usually equipped with the product (Tychonoff) topology; this is a linear
topology (provided all ui are linear) inQwhich a base u of the neighborhoods of 0
consists of all subgroups of the form Xi where Xi 2 ui such that Xi D Ai for
almost all i 2 I.
A homomorphism  W A ! C between two topological groups is a group
homomorphism that is at the same time continuous in the topological sense, i.e.
for every open set V about 0 2 C there is an open neighborhood U of 0 2 A such
that .U/  V. A topological isomorphism  is a group isomorphism such that
both  and  1 are continuous.
Groups and Hausdorff Topology We now prove a theorem showing that all
infinite abelian groups can be equipped with a non-discrete Hausdorff topology.
The result is of theoretical importance, no use will be made of it in the sequel.
Theorem 7.7 (Kertész–Szele [1]). Every infinite abelian group can be made into
a non-discrete Hausdorff topological group.
38 1 Fundamentals

Proof. (We need some simple facts that will be proved only later on.) The
Prüfer topology makes an infinite abelian group A into a Hausdorff topological
group. This is a discrete topology if and only if A itself satisfies the minimum
condition on subgroups. Then by Theorem 5.3 in Chapter 4, such an A, if
infinite, contains a subgroup Z.p1 / for some p. The embedding of this subgroup
in the multiplicative group of complex numbers of absolute value 1 induces a
non-discrete (non-linear) topology on Z.p1 /, and by translations one obtains a non-
discrete topology on A. 
Functorial Topologies Following Charles [3], we introduce functorial topolo-
gies. A functorial topology is defined in terms of a functor T on the category
Ab of abelian groups to the category L of (linearly) topologized abelian groups
such that T.A/ is the group A furnished with a linear topology and T./ D 
for every homomorphism  W A ! B in A. (The main point is that a functorial
topology assigns a topology to every group A such that all group homomorphisms
are continuous.) In this sense, all of our examples above (with the exception of the
last one) are functorial topologies.
A general method of obtaining a functorial linear topology is to choose an
arbitrary class X of groups and to declare, for each group A, the subgroups Ker 
taken for all  W A ! X 2 X as a subbase of neighborhoods of 0 2 A. The discrete
groups in the arising topology T.X / are exactly the subgroups of finite direct sums
of groups in X . Thus if we assume, to start with, that X is closed under taking finite
direct sums and subgroups, then different choices of X define different functorial
topologies.
It should be pointed out that there are functorial linear topologies which are not
obtainable in the indicate way by using an appropriate class X . The large subgroup
topology of a p-group (Sects. 2–3 in Chapter 10) is an example for such a functorial
topology.
F Notes. It is an interesting question as to which groups can carry certain special kind of
topology. Minimal and maximal functorial topologies, e.g., were studied by Boyer–Mader [1] and
by Fay–Walls [1].

Exercises

(1) The u-topology is finer than the v-topology on a group A if and only if the
identity map .A; u/ ! .A; v/ is continuous.
(2) In a linear topology, a subgroup is closed if and only if it is the intersection of
open subgroups. [Hint: use (D).]
(3) The sum of two closed subgroups is again closed.
(4) Prove that (a) the Prüfer topology is always Hausdorff; (b) every subgroup is
closed.
(5) Let G be a closed subgroup in a linear topology of A. If A=G satisfies the
minimum condition on subgroups, then G is open.
8 Modules 39

(6) The group Jp is compact both in the finite index and in the p-adic topologies.
(7) Let B be a closed subgroup of the topological group A. Prove that
(a) the canonical map A ! A=B is an open, continuous homomorphism
(as usual, A=B is furnished with the induced topology);
(b) A=B is discrete if and only if B is open in A;
(c) if  W A ! C is an open, continuous epimorphism between topological
groups, then the topological isomorphism A= Ker  Š C holds.
(8) The topology T.X / on a group A (see above) is Hausdorff if and only if
homomorphisms from A into groups in X separate points (i.e., for each 0 ¤
a 2 A there are X 2 X and  W A ! X with a ¤ 0/.
(9) Let u be a linear topology of the group A whose first Ulm subgroup A1 D 0.
All subgroups of A are closed in u if and only if u is finer than the finite index
topology.

8 Modules

Numerous theorems on abelian groups can be generalized, mutatis mutandis, to

unital modules over principal ideal domains, even over Dedekind domains. Some
results admit generalizations to modules over all integral domains, or possibly to
arbitrary associative rings. It is a delicate question to find the natural boundaries of
the validity of a particular theorem, i.e. to describe the largest class of rings over
which the theorem in question is still valid—discussion of problems like this is
beyond our present aim. But we cannot avoid modules completely: those over the
ring Jp of p-adic integers naturally appear in the scene in a variety of ways. Nolens
volens, we have to deal with them.
Modules Let R be an associative ring with identity 1, and M an abelian group
such that
(i) with r 2 R and a 2 M there is associated an element ra 2 M;
(ii) r.a C b/ D ra C rb for all r 2 R; a; b 2 M;
(iii) .r C s/a D ra C sa for all r; s 2 R; a 2 M:
(iv) .rs/a D r.sa/ for all r; s 2 R and a 2 M;
(v) 1a D a for all a 2 M.
In this case, M is said to be a left R-module or a left module over R. A right module
is defined similarly, with the elements of R acting on the right (the crucial difference
is in postulate (iv) which tells us which factor in a product is to be applied first). If
R is commutative, no distinction is necessary between left and right R-modules.
A submodule N of a left R-module M is a subset that is an R-module under the
operations of M, i.e. a subgroup of M such that rN
N for all r 2 R. In this case,
the factor group M=N becomes an R-module, where r.a C N/ D ra C N for cosets
a C N and for r 2 R.
40 1 Fundamentals

For two R-modules, M and N, an R-homomorphism is a group homomorphism

 W M ! N that respects multiplication by elements r 2 R, i.e. .ra/ D r.a/ for
all r 2 R and a 2 M. The meaning of R-isomorphism, etc. should be obvious.
Example 8.1.
(a) If R is a field, then an R-module is just an R-vector space. In abelian group theory, vector
spaces over the prime fields (Q and Z=pZ) are ubiquitous.
(b) If R is the ring Z of integers, then every abelian group A can be viewed as a Z-module under
the natural definition of multiplication of a 2 A by n 2 Z: na is the nth multiple of a.
Occasionally, we will deal with p-local groups. These are exactly those groups
in which the elements are uniquely divisible by every prime ¤ p. They are genuine
Z.p/ -modules, i.e., modules over the localization of Z at the prime p. The torsion
subgroup t.A/ of a p-local group A is a p-group and A=t.A/ is q-divisible torsion-
free for all primes q ¤ p.
Example 8.2.
(a) The group Jp is p-local, and so is the additive group of Z.p/ .
(b) For any group A, the tensor product A ˝ Z.p/ is a p-local group.
p-adic Modules Let us say a few words about modules over the rings Z.p/
and Jp . The following observation will be used frequently.
Lemma 8.3. Every p-group is, in a natural way, a module over the ring Jp of the
p-adic integers.
Proof. If  D s0 Cs1 pCs2 p2 C  Csn pn C   2 Jp , and if a 2 A is of order pn , then
a D .s0 Cs1 pCs2 p2 C  Csn1 pn1 /a is the natural definition. The element on the
right does not change if we use a longer partial sum for . The module properties
are pretty clear. 
Modules over Jp are called p-adic modules. For any group A, Jp ˝ A is a p-adic
module, and a 7! 1 ˝ a .a 2 A/ is the canonical map  W A ! Jp ˝ A.  is
universal for A in the sense that if M is any p-adic module and ˛ W A ! M any
homomorphism, then there is a unique Jp -map W Jp ˝ A ! M such that ˛ D .
F Notes. Though several theorems in abelian group theory can be phrased more naturally as
statements on modules over integral domains, or just over Z.p/ or Jp , we hesitate to enter unexplored
territory, and will phrase the results to abelian groups only. In this way, inevitably some flavor
is lost, but strict limitations had to be honored. The only exceptions will be cases when p-adic
modules will emerge naturally.

Exercises

(1) For any ring R, the cyclic left R-module Ra is R-isomorphic left annihilator
Ann a of a.
(2) For a submodule N of an R-module M, Ann a
Ann.a C N/ for all a 2 M.
(3) If R is an integral domain, then the elements a of an R-module M with
Ann a ¤ 0 form a submodule in M (called the torsion submodule).
(4) Every torsion group is a module over the completion Z Q of Z in the Z-adic
topology. This ring is the cartesian product of the rings Jp for all primes p.
8 Modules 41

(5) Let S ! R be a ring homomorphism preserving identities. Then every

R-module M becomes an S-module via sa D .s/a for a 2 M.

Problems to Chapter 1

PROBLEM 1.1. Characterize the lattices of fully invariant subgroups of torsion-

free groups.
PROBLEM 1.2. What are the singular submodules of A over its endomorphism
ring End A?
Chapter 2
Direct Sums and Direct Products

Abstract The concept of direct sum is of utmost importance for the theory. This is mostly due
to two facts: first, if we succeed in decomposing a group into a direct sum, then it can be studied
by investigating the summands separately, which are, in numerous cases, simpler to deal with. We
shall see that almost all structure theorems in abelian group theory involve, explicitly or implicitly,
some direct decomposition. Secondly, new groups can be constructed as direct sums of known or
previously constructed groups.
Accordingly, there are two ways of approaching direct sums: an internal and an external way.
Both will be discussed here along with their basic features. The external construction leads to the
unrestricted direct sum, called direct (or cartesian) product, which will also play a prominent role
in our future discussions. We present interesting results reflecting the fundamental differences in
the behavior of direct sums and products in the infinite case. Pull-back and push-out diagrams will
also be dealt with.
Important concepts are the direct and inverse limits that we shall use on several occasions. The
final section of this chapter discusses completions in linear topologies.
A reader who is well versed in group theory can skip much of this chapter.

1 Direct Sums and Direct Products

Internal Direct Sum Let B; C be subgroups of the group A, and assume they
satisfy
(i) B C C D A; and
(ii) B \ C D 0:
Condition (i) tells us that every element a 2 A can be written as a D b C c with
b 2 B; c 2 C, while (ii) implies that such b; c are unique. For, if a D b0 C c0 with
b0 2 B; c0 2 C, then b  b0 D c0  c 2 B \ C D 0. We will refer to b; c as the
coordinates of a (in the given direct sum decomposition of A). In this case we write
A D B ˚ C, and call A the (internal) direct sum of its subgroups B and C. (Recall
that if (ii) is satisfied, we say that B and C are disjoint.)
Let Bi .i 2 I/ be a set of subgroups in A subject to the following conditions:
P
(i) i2I Bi D A, i.e. thePsubgroups Bi combined generate A; and
(ii) for every i 2 I, Bi \ j¤i Bj D 0.

© Springer International Publishing Switzerland 2015 43

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_2
44 2 Direct Sums and Direct Products

Again, (i) means that every element a 2 A can be written as a finite sum a D
bi1 C    C bin with bij belonging to different components Bij , while (ii) states that
such an expression is unique. We then write

A D B1 ˚    ˚ Bn or A D ˚i2I Bi

according as the index set is finite or infinite. We call these direct decompositions
of the group A, and the Bi (direct) summands of A. If A D B ˚ C, C
is a complementary summand or a complement to B. A is called (directly)
indecomposable if A D B ˚ C implies that either B D 0 or C D 0.
Let a 2 A D B ˚ C, and write a D b C c with b 2 B; c 2 C. The maps

 W A ! B;  W A ! C given by  W a 7! b;  W a 7! c

are surjective maps; they can also be regarded as endomorphisms of A. They satisfy
b D b; c D 0; c D c; b D 0 as well as a C a D a, thus

 2 D ; 2 D ;  D 0 D ;  C  D  C  D 1A : (2.1)

If we mean by a projection an idempotent endomorphism, and by orthogonal

endomorphisms those with 0 products (in both orders), then (2.1) may be expressed
by saying that a direct decomposition A D B ˚ C defines a pair of orthogonal pro-
jections with sum 1A . Conversely, any pair ;  of endomorphisms satisfying (2.1)
yields a direct decomposition A D A˚A: In fact, idempotency and orthogonality
imply that any element common to A and A must be both reproduced and
annihilated by  and , so A \ A D 0, while  C  D 1A guarantees that
A C A D A.
If A is the direct sum of several subgroups, A D ˚i2I Bi , the decomposition can
also be described in terms of pairwise orthogonal projections. The ith projection
i W A ! Bi assigns to the element a D bi1 C    C bin the term bi 2 Bi (which can
very well be 0). Then we have:
(a) i j D 0 or i according as i ¤ j or i D j; P
(b) for every a 2 A, almost all of i a are 0, and i2I i a D a.
Conversely, if fi j i 2 Ig is a set of endomorphisms of A satisfying (a) and (b), then
A is the direct sum of the subgroups i A.
Some of the most useful properties of direct sums are listed as follows:
(A) If A D B ˚ C, then C Š A=B. Thus the complement of B in A is unique up to
isomorphism.
(B) If A D B ˚ C, and if G is a subgroup of A containing B, then we have
G D B ˚ .G \ C/.
(C) If a 2 A D B ˚ C, and if a D b C c .b 2 B; c 2 C/, then o.a/ D
lcm fo.b/; o.c/g provided both orders are finite.
P Otherwise, o.a/ D 1.
(D) If A D ˚i2I Bi and if Ci
Bi for each i, then i Ci D ˚i Ci .
1 Direct Sums and Direct Products 45

(E) If A D ˚i Bi , where each Bi is a direct sum Bi D ˚j Cij , then A D ˚i ˚j Cij .

This is a refinement of the given decomposition of A. Conversely, if A D
˚i ˚j Cij , then A D ˚i Bi where Bi D ˚j Cij .
˛ ˇ
(F) If in the exact sequence 0 ! B!A!C ! 0; Im ˛ is a summand of A,
then A Š B ˚ C. In this case, we say that the exact sequence is splitting.
Any map W C ! A satisfying ˇ D 1C is called a splitting map; then
A D Ker ˇ ˚ Im . Of course, there is another map: ı W A ! B with ı˛ D 1B
indicating splitting: A D Im ˛ ˚ Ker ı.
Two direct decompositions of A, A D ˚i Bi and A D ˚j Cj are called isomorphic
if there is a bijection between the two sets of components, Bi and Cj , such that
corresponding components are isomorphic.
We now prove a fundamental result.
Lemma 1.1. Let C D hci be a finite cyclic group where o.c/ D m D pr11    prkk with
different primes pi . Then C has a decomposition into a direct sum

C D hc1 i ˚    ˚ hck i .o.ci / D pri i /

with uniquely determined summands.

Proof. Define mi D mpr i
i
and ci D mi c .i D 1; : : : ; k/. Then the mi are relatively
prime, so there are si 2 Z such that s1 m1 C    C sk mk D 1. Then c D s1 m1 c C    C
sk mk c D s1 c1 C    C sk ck shows that the ci generate C. Clearly, hci i is of order pri i ,
so disjoint from hc1 ; : : : ; ci1 ; ciC1 ; : : : ; ck i which has order mi . Hence we conclude
that C D hc1 i ˚    ˚ hck i.
The uniqueness of the summands hci i (but not of the generators ci ) follows from
the fact that hci i is the only subgroup of C that contains all the elements whose
orders are powers of pi . t
u
Decomposition of Torsion Groups One of the most important applications of
direct sums is the following theorem that plays a fundamental role in abelian group
theory.
Theorem 1.2. A torsion group A is the direct sum of p-groups Ap belonging to
different primes p:

A D ˚p Ap :

The Ap are uniquely determined by A.

Proof. Given A, let Ap consist of all a 2 A whose orders are powers of the prime p.
Since 0 2 Ap , Ap is not empty. If a; b 2 A, i.e. pm a D 0 D pn b for integers m; n  0,
then pnCm .a  b/ D 0, so a  b 2 Ap , and Ap is a subgroup of A. If p1 ; : : : ; pk are
primes ¤ p, then Ap \ .Ap1 C    C Apk / D 0, since every element of Ap1 C    C Apk
is annihilated by a product of powers of p1 ; : : : ; pk . Thus the Ap generate their direct
sum in A; it must be all of A, as it is obvious in view of Lemma 1.1.
46 2 Direct Sums and Direct Products

If A D ˚p Bp is another decomposition of A into p-groups Bp with different

primes p, then by the definition of the Ap we have Bp
Ap for each p. If we had
Bp < Ap for some p, then ˚p Bp could not equal A. t
u
The subgroups Ap are called the primary components or the p-components
of A. They are, as is seen from the definition, fully invariant in A. If A is not
torsion, then the p-components Tp of its torsion part T D tA may be referred
to as the p-components of A. (In this case, however, Tp need not be a summand
of A.) Theorem 1.2 is of utmost importance as it makes it possible to reduce the
structure theory of torsion groups to primary groups.
Example 1.3. The group Q=Z is isomorphic to the multiplicative group of all complex numbers
that are nth roots of unity for some integer n > 0. It is a torsion group whose p-component is
Z.p1 / (this corresponds to the subgroup of all pk th roots of unity .k D 0; 1; 2; : : : /). Hence

Q=Z Š ˚p Z.p1 /:

Another crucial direct sum decomposition is a trivial consequence of a vector

space theorem.
Theorem 1.4. An elementary group is a direct sum of cyclic groups of prime orders.
Proof. By Theorem 1.2 only p-groups need to be considered. An elementary p-
group is a Z=pZ-vector space, and as such it is the direct sum of one-dimensional
spaces, i.e. of groups of order p. t
u
External Direct Sum While the internal direct sum serves to break a group into
smaller pieces, in case of external direct sums we glue together groups to create a
new larger group.
We start with two unrelated groups, B and C, and construct a new group A that is
the direct sum of two subgroups B0 and C0 , such that B0 Š B; C0 Š C. The set of all
pairs .b; c/ with b 2 B; c 2 C forms a group A under the rules:
(a) .b1 ; c1 / D .b2 ; c2 / if and only if b1 D b2 ; c1 D c2 ;
(b) .b1 ; c1 / C .b2 ; c2 / D .b1 C b2 ; c1 C c2 /.
Then B0 D f.b; 0/ j b 2 Bg Š B; C0 D f.0; c/ j c 2 Cg Š C under
the correspondences b 7! .b; 0/; c 7! .0; c/; they are subgroups of A such that
A D B0 ˚ C0 (internal direct sum). If we think of B; C being identified with B0 ; C0
under the indicated mappings, then we may also write A D B ˚ C, and call A the
external direct sum of B and C. (We write A Š B ˚ C to say that A is a direct sum
of two subgroups isomorphic to B and C.)
Direct Products A vector .: : : ; bi ; : : : / over the set fBi gi2I of groups has exactly
one coordinate bi from Bi , viz. in the ith position, for each i 2 I. Such a vector can
also be interpreted as a function f defined over I such that f .i/ D bi 2 Bi for every
i 2 I. Equality and addition of vectors are defined coordinate-wise (for functions,
1 Direct Sums and Direct Products 47

we would say point-wise). In this way, the set of all vectors becomes a group C,
called the direct product or the cartesian product of the groups Bi ; in notation:
Y
CD Bi :
i2I

The correspondence i W bi 7! .: : : ; 0; bi ; 0; : : : ; 0; : : : / where bi is the ith

coordinate and 0’s are everywhere else, is an isomorphism of Bi with a subgroup
B0i of C. The groups B0i .i 2 I/ generate their direct sum B in C which consists of
all vectors with finite support, where support means supp c D fi 2 I j ci ¤ 0g if
c D .: : : ; ci ; : : : / 2 C. B is the external direct sum of the Bi , B D ˚i2I Bi . Clearly,
B D C whenever I is finite.
.I/
For a group A, and for a set I, AQ D ˚i2I A will denote the direct sum of jIj
copies of A, and the symbol A D i2I A will stand for the direct product of jIj
I

copies of A. The corresponding notations A./ and A for a cardinal  should be

clear.
The external direct sums and direct products can also be described in terms of
systems of maps. The functions

B W b 7! .b; 0/; C W c 7! .0; c/; B W .b; c/ 7! b; C W .b; c/ 7! c

are called the (coordinate) injection and projection maps, respectively. They
satisfy

B B D 1B ; C C D 1C ; B C D 0 D C B ; B B C C C D 1B˚C :

For an arbitrary number of components Bi .i 2 I/, we have injections i and

projections i satisfying
i Y i
Bi ! C D Bi ! Bi
i2I

where i bi D .: : : ; 0; bi ; 0; : : : /; i .: : : ; bj ; : : : ; bi ; : : : / P
D bi satisfy the conditions:
(i) j i D 1Bi or 0 according as i D j or i ¤ j; and (ii) i2I i i D 1C (formally).
Similarly for an infinite direct sum ˚i2I Bi , in which case any given element is
annihilated by almost all i .
The following ‘universal’ properties are crucial.
Theorem 1.5. Let ˇi W Bi ! A .i 2 I/ denote arbitrary homomorphisms, and i W
Bi ! ˚i2I Bi the injection maps. There is a unique homomorphism  W ˚i2I Bi ! A
such that ˇi D i for every i.
Proof. Write b 2 ˚i2I Bi in the form b D 1 1 b C    C n n b where the i
are the projection maps of the direct sum. It is immediately checked that b D
ˇ1 1 b C    C ˇn n b 2 A defines a homomorphism  W ˚i2I Bi ! A with ˇi D i .
48 2 Direct Sums and Direct Products

If  0 is another such map, then .   0 /i D 0 for each i, so .   0 /b vanishes for

all b 2 ˚i2I Bi , i.e.  D  0 . t
u
Q
Theorem 1.6. Let ˛i W A ! Bi .i 2 I/ denote homomorphisms and Q i W i2I Bi !
Bi the projection maps. There exists a unique map W A ! i2I Bi such that
˛i D i for each i 2 I.
Q
Proof. Define .a/ D .: : : ; ˛i a; : : : / 2 i2I Bi . This is obviously a homomor-
phism satisfying ˛i D i . If also 0 has the same property, then i .  0 /a D 0
for all a 2 A, thus .  0 /a D 0. This means D 0 . t
u
A notational agreement: if ˛i W Ai ! Bi .i 2 I/ are homomorphisms, then ˚i2I ˛i
will denote the map ˚i2I Ai ! ˚i2I BQ i that carries Q the ith coordinates
Q to the ith
coordinates as given by ˛i . The map i2I ˛i W i2I Ai ! i2I Bi has similar
meaning. Q
For a group G, the diagonal map G W G ! G (arbitrary number of
components) acts as G W g 7! .: : :P ; g; : : : ; g; : : : /, and the codiagonal map
rG W ˚ G ! G as rG W .: : : ; gi ; : : : / 7! i gi .
Subdirect Products Among the subgroups of the direct product, the subdirect
products are most important. A group G is a subdirect Q product of the groups
Bi .i 2 I/ if it is a subgroup of the direct product A D i2I Bi such that i G D Bi
for all projections i W A ! Bi . This means that for every bi 2 Bi , G contains
at least one vector whose ith coordinate is exactly bi . If Ki D Ker.i  G/, then
\i2I Ki D 0. Conversely, if Ki are subgroups of a group G such that \i2I Ki D 0,
then G is a subdirect product of the factor groups G=Ki , via
Y
g 7! .: : : ; g C Ki ; : : : / 2 .G=Ki / where g 2 G:
i2I

If the index set I is finite, then we also say that we have a subdirect sum.
Lemma 1.7 (Łoś). Every group is a subdirect product of cocyclic groups.
Proof. For every non-zero a in group A, let Ka be a subgroup of A maximal without a
(argue with Zorn). Thus every subgroup of A that properly contains Ka also contains
a, i.e. the coset a C Ka is a cogenerator in A=Ka , so this factor group is cocyclic.
Since \0¤a2A Ka D 0, it follows that A is a subdirect product of the cocyclic groups
A=Ka . t
u
There are numerous subdirect products contained in a direct product of groups,
but there is no complete survey of them. The only exception is the case of subdirect
sums of two groups.
Let G be a subdirect sum of B and C. The elements b 2 B with .b; 0/ 2 G form a
subgroup B0
B and the elements c 2 C with .0; c/ 2 G form a subgroup C0
C.
It is straightforward to check that the correspondence b C B0 7! c C C0 whenever
.b; c/ 2 G is an isomorphism of B=B0 with C=C0 . Thus G consists of those .b; c/ 2
B ˚ C for which the canonical epimorphisms B ! B=B0 and C ! C=C0 map b
1 Direct Sums and Direct Products 49

and c upon corresponding cosets. The groups B0 and C0 are called the kernels of the
subdirect sum. Conversely, if we are given the groups B; C along with epimorphisms
ˇ W B ! F; W C ! F for some group F, then the elements .b; c/ 2 B ˚ C with
ˇb D c form a group G that is a subdirect sum of B and C. It is easy to verify the
isomorphisms

G=B0 Š C; G=C0 Š B; B=B0 Š G=.B0 ˚ C0 / Š C=C0 :

We mention that the subdirect sum G in the preceding paragraph may also be
ˇ
obtained as a pull-back of the maps ˇ; where B!B=B0 Š C=C0 C: See
Exercise 3 in Sect. 3.)
Q Let K be an ideal in the Boolean Q lattice of all subsets of I; then the K-product
QK<A i is the set of all vectors in i2I Ai whose supports belong to K. The -product
i2I A i consists of vectors with support < .
Ultraproducts The following construction is based on the notion of filters. Let I
be an infinite index set and F a filter on the subsets of I. TheQ filtered direct product
of groups Ai .i 2 I/ is a subgroup of the direct product A D i2I Ai consisting of all
vectors a D .: : : ; ai ; : : : / 2 A for which the null-set n.a/ D fi 2 I j ai D 0g 2 F . It
is routine
Q to check that this is in fact a (pure) subgroup of A, which we shall denote
as F i2I i . The factor group
A
Y Y YF
Ai =F D Ai = Ai
i2I
i2I i2I

Q
Qcalled the reduced product with respect to F . Thus a; b 2 i2I Ai are equal in
is
i2I Ai =F exactly if supp .a  b/ 2 F . Q
The most important special case is when F is an ultrafilter U. Then i2I Ai =U
is called the ultraproduct of the Ai . If U is a principal
Q ultrafilter,
Q i.e. it consists of
those subsets of I that contain a fixed j 2 I, then U i2I A i D i2J A i where J D Infjg.
In this case, the ultraproduct is just Aj . Therefore, only ultraproducts with respect to
non-principal ultrafilters are of real interest.
Example 1.8. The filter F that consists of all subsets of I with finite complements is non-principal,
QF
and i2I Ai contains the direct sum ˚i2I Ai .
F Notes. A noteworthy generalization of direct powers, studied by Balzerzyk [3], Eda [1],
relies on a complete Boolean lattice B with 0 as smallest and 1 as largest element. By the Boolean
power A.B/ of the group
W A is meant the set of functions f W A ! B such that (i) f .a/ ^ f .b/ D 0 if
a ¤ b in A, and (ii) a2A f .a/ D 1. The sum f C g of two functions is defined via

.f C g/.a/ D _aDxCy .f .x/ ^ g.y//

for all possible x; y 2 A satisfying x C y D a. In case B is the power-set of a set I, then the
elements f 2 A.B/ are in a bijective correspondence with the elements fN 2 AI such that f .a/ D
fi 2 I j fN .i/ D ag 2 B where a 2 A.
The primary decomposition Theorem 1.2 is of central importance in abelian group theory. Its
roots are in elementary number theory; this kind of decomposition was used by C.F. Gauss. In its
50 2 Direct Sums and Direct Products

complete, final form is due to Frobenius–Stickelberger [1]. The result generalizes straightforwardly
to torsion modules over Dedekind domains.
In contrast to Theorem 1.2, Theorem 1.4 easily generalizes to arbitrary modules: if a module
is the union of simple submodules, then it is a direct sum of simple modules (it is then called
semi-simple). Semi-simple modules may be characterized by the property that every submodule is
a direct summand.
The result on the subdirect sum of two groups is due to R. Remak; he dealt with finite,
not necessarily commutative groups. Ultraproducts have profound implications in various areas,
especially in model theory. See Eklof [1] for their structure.

Exercises

(1) Let B; C be subgroups of A, and B ˚ C their external direct sum. There is an

exact sequence 0 ! B \ C ! B ˚ C ! B C C ! 0.
(2) Determine when the direct product of infinitely many torsion groups is again a
torsion group.
˛i ˇi
(3) If 0 ! Ai !Bi !Ci ! 0 are exact sequences for i 2 I, then so are
Y Q Y Q Y
˚˛i ˚ˇi ˛i ˇi
0 ! ˚Ai ! ˚ Bi ! ˚ Ci ! 0 and 0 ! Ai ! Bi ! Ci ! 0:

(4) If G is a subdirect sum of B and C, then B C G D B ˚ C D G C C.

(5) Let B; C be subgroups of A such that B \ C D 0. If .B C C/=C is a summand
of A=C, then B is a summand of A.
(6) (a) The subdirect sum of Z.pm / and Z.pn / .0 < m
n/ with kernels Z.pmk /
and Z.pnk / is isomorphic to Z.pn / ˚ Z.pmk /.
(b) The subdirect sum of Z.p1 / and Z.p1 / with kernels Z.pm / and Z.pn / .0 <
m
n/ is isomorphic to Z.p1 / ˚ Z.pm /.
(7) A group A is called subdirectly irreducible if in any representation of A as a
subdirect product of groups Ai , one of the coordinate projections i W A ! Ai
is an isomorphism. Prove that A is subdirectly irreducible if and only if it is
cocyclic.

2 Direct Summands

Direct Summands In this section, we collect a few criteria for a subgroup to be

a summand. We start with the most frequently used criterion.
Lemma 2.1. A subgroup B of A is a summand of A if and only if A has an
idempotent endomorphism  satisfying A D B; equivalently, the injection B ! A
followed by  W A ! B is the identity 1B of B.
2 Direct Summands 51

Proof. If A D B ˚ C, then the projection  on the first summand, viewed as an

element of End A, is as desired. Conversely, if  is an idempotent endomorphism,
then A D A ˚ .1  /A. t
u
Putting it in a different way, lemma states that B is a summand of A exactly if the
identity map of B extends to an endomorphism A ! B.
If B is a summand of A, then the complementary summand is unique up to
isomorphisms (recall: it is Š A=B), but it is far from being unique as a subgroup.
The following result explains how to obtain from one complement all the other
complements.
Lemma 2.2. Let A D B ˚ C be a direct decomposition with projections ˇ; . If
also A D B ˚ C0 with projections ˇ0 ; 0 , then, for some endomorphism  of A, we
have

ˇ0 D ˇ C ˇ ; 0 D  ˇ : (2.2)

Conversely, if the maps ˇ0 ; 0 are of the form (2.2), then A D B ˚ 0 A.

Proof. If we are given the two direct decompositions, then let  D  0 . Then B

Ker , so  D ˇ C D  . If a D bCc D b0 Cc0 with b; b0 2 B; c 2 C; c0 2 C0 ,

then a D c  c0 D b0  b 2 B, thus ˇ D . Hence 0 D   D  ˇ and
ˇ0 D 1A  0 D ˇ C  0 D ˇ C ˇ :
Conversely, if ˇ0 ; 0 are obtained from ˇ; as given in (2.2) with any  2 End A,
then ˇ0 C 0 D 1A ; ˇ02 D ˇ0 ; 02 D 0 ; and ˇ0 0 D 0 ˇ0 D 0. Thus A D ˇ0 A ˚ 0 A
where ˇ0 A D ˇA D B. t
u
If ˇ is a central idempotent (commutes with all endomorphisms), then
ˇ0 D ˇ C ˇ D ˇ and 0 D . Thus the complements cannot be changed
(they are fully invariant in A).
In general, a subgroup of a direct sum does not decompose along the summands.
However, there is an important exceptional case.
Lemma 2.3. If A D B ˚ C, and if G is a fully invariant subgroup of A, then

G D .G \ B/ ˚ .G \ C/:

Proof. Let ˇ; be the projections attached to the given direct sum. By full
invariance, both ˇG and G are subgroups of G. Evidently, ˇG and G generate a
direct sum in A, and since ˇ C D 1A , we have G D ˇG ˚ G. Since ˇG
G \ B
and G
G \ C; and proper inclusion is out of question, we have ˇG D G \ B and
G D G \ C. t
u
The following is a useful lemma.
52 2 Direct Sums and Direct Products

Lemma 2.4 (Kaplansky [K]). If the factor group A=B is a direct sum: A=B D
˚i2I .Ai =B/, and if B is a direct summand in every Ai , say, Ai D B ˚ Ci , then B is a
summand of A. More precisely,

A D B ˚ .˚i2I Ci /:

Proof. It is clear that the groups B and the Ci generate A. Assume that b C c1 C
   C cn D 0 for some b 2 B and cj 2 Cj .j D 1; : : : ; n/. Passing mod B, we obtain
.c1 C B/ C    C .cn C B/ D B, whence the given direct sum forces cj 2 B for every
j. Thus cj 2 Cj \ B D 0; and hence also b D 0. Consequently, B and the Ci generate
their direct sum in A. t
u
Summands of Large Direct Sums The following theorem has several applica-
tions in the study of properties inherited by summands.
Theorem 2.5 (Kaplansky [2]). Summands of a direct sum of countable groups are
also direct sums of countable groups.
Proof. Let A D ˚i2I Ai D B ˚ C where each summand Ai is countable. Pick any
summand A1 , a generating system faj gj2J of A1 , and write aj D bj C cj .bj 2 B;
cj 2 C/. Note that each bj and each cj has but a finite number of non-zero coordinates
in the direct sum A D ˚i Ai . Collecting all the Ai that contain at least one non-
zero coordinate of some bj or cj , and then forming their direct sum, we obtain
a countable direct summand X1 of A. Next, we repeat the same process with X1
replacing A1 : select a generating system for X1 and collect all the Ai which have
non-zero coordinates of the B- and C-coordinates of the generators, to obtain a
larger countable summand X2 of A. Continuing the same way, we get a chain
X1
X2
  
Xn
: : : of countable summands of A whose union is a countable
summand AN 1 such that AN 1 D .B \ AN 1 / ˚ .C \ AN 1 /.
A smooth chain of summands S of A is defined as follows. Each S is a direct
sum of some Ai . Set S0 D 0. If S is defined for an ordinal  and S < A, then
pick an Ai not in S and let S C1 D S C AN i (where AN i is obtained by repeating
the above process for Ai using components not in S ). For limit ordinals  we set
S D [< S . It is evident that for some ordinal 
jAj we will reach S D A. It is
also clear that S C1 =S is countable, and every S is a direct sum of a subset of the Ai
such that S D .B \ S / ˚ .C \ S / for all 
. Setting B \ S C1 D .B \ S / ˚ B ,
it is clear that the B are countable and generate their direct sum in B. Since the B
together generate B, we have B D ˚ < B , as claimed. t
u
Example 2.6. Let G be any countable group, and A D ˚<!1 G where G Š G for each  . If
A D B ˚ C, then both B and C are direct sums of countable groups (not necessarily isomorphic
to G).
F Notes. Kaplansky’s Theorem 2.5 holds for countably generated modules over arbitrary
rings. It has been extended to -generated modules by C. Walker [2] for any infinite cardinal .
2 Direct Summands 53

Exercises

(1) (Grätzer) Let B be a subgroup of A, and C a B-high subgroup in A. Then

A D B ˚ C if and only if pa D b C c .a 2 A; b 2 B; c 2 C/ for a prime p
implies b 2 pB.
(2) Suppose C < B < A. Prove that
(a) if B is a summand of A, then B=C is a summand of A=C;
(b) if C is a summand of A and B=C is a summand of A=C, then B is a
summand of A.
(3) Let B be a summand of A, and let fi j i 2 Ig be the set of all projections of
A onto B. These projections form a semigroup such that i j D j .
(4) A group A has no summand isomorphic to itself if and only if one-sided units
in its endomorphism ring End A are twosided.
(5) Let  denote an endomorphism of A.
(a) If, for some n, Im nC1 D Im n , then Ker n C Im n D A.
(b) If, for some n, Ker nC1 D Ker n , then Ker n \ Im n D 0.
(6) Assume A D B ˚ C D B0 ˚ C0 , and let ˇ W A ! B; ˇ 0 W A ! B0 denote the
projections in the given decompositions. Then B Š B0 if and only if there are
; 2 End A such that  D ˇ and  D ˇ 0 .
(7) (a) (Grätzer–Schmidt) Let B be a direct summand of A. The intersections of
all the complements of B in A is the maximal fully invariant subgroup of
A that is disjoint from B. [Hint: Lemma 2.2.]
(b) A complement to a direct summand of A is unique if and only if it is fully
invariant in A.
(8) Call a subgroup G of A projection-invariant if G
G for every projection
 of A onto a summand. Prove that: (a) G is projection-invariant in A if and
only if G D G \ A for all projections ; (b) intersections of projection-
invariant subgroups are projection-invariant, and so are subgroups generated
by projection-invariant subgroups; (c) Lemma 2.2 holds for projection-
invariant G; (d) a projection-invariant summand is a fully invariant subgroup.
(9) (Kulikov) A direct decomposition A D ˚i2I Ai has a common refinement with
every direct decomposition of A if and only if every Ai is projection-invariant.
(10) (Fuchs) B < A is an absolute direct summand of A if A D B ˚ C for every
B-high subgroup C. (a) Prove that B is an absolute direct summand if and
only if it is either injective (see Chapter 4) or A=B is a torsion group whose
p-component is annihilated by pk whenever B n pB contains an element of
order pk . (b) Find all absolute direct summands of a bounded group.
(11) (Irwin–Walker) Let A D ˚i2I Ai and Bi
Ai for each i. If Ci is Bi -high in
Ai , then ˚i Ci is ˚i Bi -high in A.
(12) (Enochs) Let A be a p-group and A D B˚C D B0 ˚C0 direct decompositions
of A such that BŒp
D B0 Œp
. Then A D B ˚ C0 D B0 ˚ C: [Hint: use induction
of the order of a 2 A to show a 2 B ˚ C0 :]
54 2 Direct Sums and Direct Products

(13) (C. Walker) Generalize Theorem 2.5 to larger cardinalities : summands of

direct sums of -generated groups are of the same kind.
(14) A supplement subgroup S to some C < A is defined to be minimal with
respect to the property A D C C S. S has this property if and only if S \ C is
superfluous in A. [Hint: use the modular law in both directions.]

3 Pull-Back and Push-Out Diagrams

Pull-Backs With the aid of direct sums, we can describe two important methods
in constructing certain commutative diagrams.
Theorem 3.1. Given the homomorphisms ˛ W A ! C and ˇ W B ! C, there exists
a group G, unique up to isomorphism, along with homomorphisms W G ! A; ı W
G ! B, such that the diagram
γ
G −−−−→ A
⏐ ⏐
⏐ ⏐α
δ 
β
B −−−−→ C

is commutative, and if
γ
G −−−−→ A
⏐ ⏐
⏐ ⏐α
δ 
β
B −−−−→ C

is any commutative diagram, then there exists a unique homomorphism  W G0 ! G

such that  D 0 and ı D ı 0 .
Proof. Given ˛; ˇ, define G as the subgroup of the direct sum A˚B consisting of all
pairs .a; b/ .a 2 A; b 2 B/ such that ˛a D ˇb, and let W .a; b/ 7! a; ı W .a; b/ 7! b.
This makes the first diagram commutative.
If the second diagram is commutative, then define  W G0 ! G as g0 D
. g ; ı g / for g0 2 G0 ; here . 0 g0 ; ı 0 g0 / 2 G, since ˛ 0 D ˇı 0 . Evidently, g0 D
0 0 0 0

0 g0 and ıg0 D ı 0 g0 for every g0 2 G0 . It is easy to see that Ker D .0; Ker ˇ/ and
Ker ı D .Ker ˛; 0/. Therefore, if  0 W G0 ! G also satisfies  0 D 0 ; ı 0 D ı 0 ,
then .   0 / D 0 D ı.   0 /, and so Im.   0 /
Ker \ Ker ı D 0: Hence
   0 D 0, thus  is unique.
The uniqueness of G can be verified by considering a G0 with the same prop-
erties. Then by what has already been shown, there are unique maps  W G ! G0 ;
0 W G0 ! G with the indicated properties. Then 0  W G ! G is a unique map
3 Pull-Back and Push-Out Diagrams 55

(if applied to the case G0 D G), so it must be the identity; the same holds for
 0 W G0 ! G0 , whence the uniqueness of G is manifest. u
t
Push-Outs The group G of the preceding theorem is called the pull-back of the
maps ˛ and ˇ. Our next task is to prove the dual, where the group H will be called
the push-out of ˛ and ˇ.
Theorem 3.2. Assume that ˛ W C ! A; ˇ W C ! B are homomorphisms. There exist
a group H, unique up to isomorphism, and homomorphisms W A ! H; ı W B ! H,
such that the diagram
α
C −−−−→ A
⏐ ⏐
⏐
β
⏐γ

δ
B −−−−→ H

is commutative, and for every commutative diagram

α
C −−−−→ A
⏐ ⏐
⏐
β
⏐ 
γ
δ
B −−−−→ H 

there is a unique homomorphism W H ! H 0 satisfying D 0 and ı D ı0.

Proof. Starting with ˛; ˇ, define H as the factor group of A˚B modulo the subgroup
X D f.˛c; ˇc/ j c 2 Cg. Let W a 7! .a; 0/ C X; ı W b 7! .0; b/ C X .a 2 A; b 2 B/
be the maps induced by the injections. Then ˛c D ıˇc for every c 2 C assures the
commutativity of the first diagram.
If the second diagram is commutative, then we let W .a; b/ C X 7! 0 a C
0 0
ı b 2 H . One can readily check that this definition is independent of the chosen
representative .a; b/ of the coset, and moreover, it satisfies D 0 and ı D ı 0 .
The uniqueness follows from the simple fact that Im and Im ı generate H, and
therefore, if 0 D 0 ; 0 ı D ı 0 for some 0 W H ! H 0 , then .  0 / D 0 D
.  0 /ı implies that  0 maps the whole of H upon 0. An argument similar to
the one at the end of the proof of the preceding theorem establishes the uniqueness
of H. t
u
The following observations are of importance.
(a) If in the pull-back diagram, ˛ is monic, then so is ı; if ˛ is epic, so is ı. In
view of the uniqueness of the pull-back diagram, it is enough to prove the claim
for the group G as constructed in the proof above. That Ker ˛ D 0 implies
Ker ı D 0 is immediately seen from the proof. Furthermore, if ˛ is epic, then to
every b 2 B there is an a 2 A such that ˛a D ˇb, and so ı is also epic.
56 2 Direct Sums and Direct Products

(b) If in the push-out diagram, ˛ is monic, then so is ı; if ˛ is epic, so is ı.

Again, we need only show this for H as defined above. Now clearly Ker ı D 0
whenever Ker ˛ D 0. If ˛ is epic, then to every a 2 A there is a c 2 C with
˛c D a, and so ı maps b C ˇc upon .0; b C ˇc/ C X D .a; b/ C X. Hence ı is
epic as well.

Exercises

(1) If B D 0 in the pull-back diagram above, then G Š Ker ˛.

(2) (a) If C D 0 in the pull-back diagram, then G Š A ˚ B.
(b) If C D 0 in the push-out diagram, then H Š A ˚ B.
(3) If both ˛; ˇ are surjective in the pull-back diagram, then G is a subdirect sum
of A and B with kernels Ker ˛; Ker ˇ.
(4) The pull-back diagram above is a push-out diagram (for ; ı) exactly if the map
r.˛ ˚ ˇ/ W A ˚ B ! C is surjective.
(5) If in the diagram

A1 −−−−−→ A2 −−−−−→ A3
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

B1 −−−−−→ B2 −−−−−→ B3

each of the two squares is a pull-back, then the outer rectangle is also a pull-
back.
(6) Formulate and prove the dual of the preceding exercise for push-outs.
(7) Using the notations of the above pull-back and push-out diagrams, the
sequences 0 ! G ! A ˚ B ! C ! 0 and 0 ! C ! A ˚ B ! H ! 0 (with
the obvious maps) are exact.

4 Direct Limits

Direct Systems Let fAi .i 2 I/g be a system of groups where the index set I is
partially ordered and directed (upwards) in the sense that to each pair i; j 2 I, there
is a k 2 I such that both i
k and j
k. Suppose that for every pair i; j 2 I with
j
i
j, there is a homomorphism i W Ai ! Aj (called connecting map) subject to
the conditions:
(i) ii is the identity map of Ai for all i 2 I; and
j
(ii) if i
j
k in I, then jk i D ik .
4 Direct Limits 57

j
In this case, A D fAi .i 2 I/I i g is called a direct system. (If the index set is !,
then it suffices to specify only nnC1 for all n < !, because the other nm are then
determined by rule (ii).) By the direct or injective limit, or colimit of A is meant a
group A such that
j
(a) there are maps i W Ai ! A such that i D j i holds for all i
j;
j
(b) if G is any group, and i W Ai ! G .i 2 I/ are maps satisfying i D j i for all
i
j, then there is a unique map ˛ W A ! G such that i D ˛i for all i 2 I.
We write: A D lim Ai ; and call the maps i W Ai ! A canonical.
!i2I
j
Theorem 4.1. A direct system A D fAi .i 2 I/I i g of groups has a limit, unique
up to isomorphism.
Proof. We form the direct sum A D ˚i Ai , and consider the subgroup B
A
j
generated by the elements ai  i ai for all ai 2 Ai and for all i
j in I. Our
claim is that A=B D A is the direct limit of A.
The elements of A=B are cosets of the form ai1 C    C ain C B with aij 2 Aij . If
i 2 I is such that i1 ; : : : ; in are all
i, then this coset is ii1 ai1 C    C iin ain C B,
since ai1 C    C ain  ii1 ai1      iin ain D .ai1  ii1 ai1 / C    C .ain  iin ain / 2 B.
Thus every element in A=B can be written as ai C B for some ai 2 Ai . In particular,
B consists of all finite sums of the form b D ai1 C    C ain with aij 2 Ain for which
there is an i 2 I such that i1 ; : : : ; in
i and ii1 ai1 C    C iin ain D 0:
Consider the maps i W Ai ! A=B acting as ai 7! ai C B. They obviously satisfy
j
i D j i .i
j/. If G is any group as stated in (b), then define ˛ W A=B ! G
j
by ˛.ai C B/ D i ai . Owing to i D j i , this definition is independent of the
choice of the coset representative, and since ˛ is evidently additive, ˛ is a genuine
homomorphism. It satisfies i D ˛i for all i 2 I, as required. If ˛ 0 W A=B ! G also
satisfies i D ˛ 0 i for all i 2 I, then .˛  ˛ 0 /i D 0 for each i 2 I, thus ˛  ˛ 0 sends
every ai C B D i ai to 0, i.e. ˛ D ˛ 0 . It follows that A=B is a limit of the given direct
system, so we can write A D A=B.
To show that A is unique up to isomorphism, suppose that also A0 shares
properties (a)–(b). Then there exist unique maps ˛ W A ! A0 and ˛0 W A0 ! A as
required by (b). Also, ˛0 ˛ W A ! A and ˛˛0 W A0 ! A0 are unique, so they are the
identity maps. Consequently, A0 Š A . t
u
We now list some of the most useful properties of direct limits.
(A) A is the set union of the subgroups i Ai .i 2 I/:
j
(B) If i ai D 0 for some ai 2 Ai , then there is a j  i such that i ai D 0. Indeed,
if i ai D 0, then ai 2 B, and the proof above establishes this claim.
j
(C) If every i is a monic map, then all the i are monomorphisms. This follows
from (B).
(D) If J is a cofinal subset of I, then the system restricted to J has the same direct
limit: lim Ai Š lim Ai . In fact, if the first group is A0 =B0 , then aj C B0 7!
!J !I
aj C B is an isomorphism of A0 =B0 with A=B.
58 2 Direct Sums and Direct Products

Example 4.2. Let fAi .i 2 I/g be the collection of all subgroups of a group A where the index set
j
I is partially ordered by the rule: i  j if and only if Ai  Aj . Let i W Ai ! Aj denote the injection
j
map for i  j. Then A D fAi .i 2 I/I i g is a direct system with limit A.
Example 4.3. If we admit only finitely generated subgroups of A in the direct system A D fAi .i 2
j j
I/I i g with the injection maps i , the direct limit is still A. In the special case where A is arbitrary
torsion-free, we get A as the direct limit of finitely generated free groups.
Example 4.4. Let A D ˚j2J Cj be a direct sum. Let i range over the set I of finite subsets of J, so
that i  k in I means that i is a subset of k. If we define Ai D ˚j2i Cj for all i 2 I, and ik W Ai ! Ak
to be the obvious inclusion map, then we get a direct system whose limit is A.

Maps Between Direct Systems We consider homomorphisms between direct

limits that are induced by homomorphisms between direct systems. If A D fAi .i 2
j j
I/I i g and B D fBi .i 2 I/I i g are direct systems with the same index set I, then
by a homomorphism ˆ W A ! B we mean a set of homomorphisms ˆ D fi W
Ai ! Bi j i 2 Ig such that the diagrams

πj
Ai −−−i−→ Aj
⏐ ⏐
⏐
φi 
⏐φ
 j
ρj
Bi −−−i−→ Bj

commute for all i

j in I.
Lemma 4.5. If ˆ is a homomorphism between the direct systems A and B, then
there exists a unique morphism ˆ W A D lim Ai ! B D lim Bi making all the
! !
diagrams

i π
Ai −−−− → A∗
⏐ ⏐
⏐
φi 
⏐Φ
 ∗
ρi
Bi −−−−→ B∗

commute .i ; i denote the canonical morphisms/. ˆ is an epimorphism

.monomorphism/ if all the i are epimorphisms .monomorphisms/.
j j
Proof. Since the maps i i W Ai ! B satisfy the condition j j i D j i i D i i
for every pair i
j, the existence of a unique homomorphism ˆ W A ! B such
that i i D ˆ i for each i 2 I is guaranteed. This proves the first assertion.
If all the i are epic, then the subgroups i Bi D i i Ai cover B , so ˆ must
be epic. If all the i are monic, then pick an a 2 Ker ˆ . There is j 2 I such that
a D j aj for some aj 2 Aj . Hence j j aj D ˆ j aj D ˆ a D 0, and so by (B)
we have a k  j with jk j aj D 0. But jk j D k jk and k is monic, so jk aj D 0,
whence j aj D 0 and a D 0. t
u
4 Direct Limits 59

We now move to three direct systems: A, B as above, and a third one, C D

j
fCi .i 2 I/I i g, all with the same directed index set I. If ˆ W A ! B and ‰ W B ! C
i i
are homomorphisms between them such that the sequence 0 ! Ai !Bi !Ci ! 0
is exact for each i 2 I, then we say that the sequence

ˆ ‰
0 ! A!B!C ! 0 (2.3)

is exact. It is an important fact that direct limits of exact sequences is exact. More
precisely,
Theorem 4.6. Let A; B; C be direct systems over the same index set I, and ˆ W A !
B and ‰ W B ! C homomorphisms between them. If the sequence (2.3) is exact,
then the sequence

ˆ ‰
0 ! A D lim Ai !B D lim Bi !C D lim Ci ! 0
! ! !
i i i

of direct limits is likewise exact.

Proof. Exactness at A and C is guaranteed by Lemma 4.5, so we prove exactness
only at B . By Lemma 4.5, the diagram

φi ψi
0 −−−−→ Ai −−−−→ Bi −−−−→ Ci −−−−→ 0
⏐ ⏐ ⏐
⏐
πi 
⏐
ρi 
⏐σ
 i
∗ Φ ∗ Ψ
0 −−−−→ A∗ −−−− → B∗ −−−− → C∗ −−−−→ 0

is commutative for all i 2 I. If a 2 A , then i ai D a for some ai 2 Ai , so ‰ ˆ a D

‰ ˆ i ai D ‰ i i ai D i i i ai D 0. Next let b 2 Ker ‰ . For some bi 2 Bi ,
we have i bi D b, whence i i bi D ‰ i bi D ‰ b D 0. There exists j 2 I with
j j
i i bi D 0, thus j bj D j i bi D 0. Since the top row in the diagram is exact,
there is an aj 2 Aj with j aj D bj . Setting a D j aj , we arrive at ˆ a D ˆ j aj D
j
j j aj D j i bi D i bi D b, i.e. b 2 Im ˆ , and the bottom row is exact at B . ut

Exercises

(1) Show that lim Z.pn / D Z.p1 /, using inclusion maps.

!n
(2) Let An Š Z .n < !/ with nnC1 W An ! AnC1 multiplication by n. Prove that
lim An Š Q.
!n
j
(3) If every i is an onto map, then all the i are epimorphisms.
(4) A group is locally cyclic if and only if it is a direct limit of cyclic groups.
60 2 Direct Sums and Direct Products

j
(5) (a) Let A be the limit of the direct system A D fAi .i 2 I/I i g, and a 2 A .
There exist a j 2 I and an aj 2 Aj such that j aj D a and o.aj / D o.a/.
(b) Direct limit of torsion (torsion-free) groups is again torsion (torsion-free).
(6) If G is finitely generated, and ˛ W G ! A (notations as above), then there exist
a j 2 I and an ˛j W G ! Aj such that ˛ D j ˛j .
j j
(7) If A D fAi .i 2 I/I i g and B D fBi .i 2 I/I i g are direct systems of groups,
j j
then A ˚ B D fAi ˚ Bi .i 2 I/I i ˚ i g is likewise a direct system whose direct
limit is the direct sum of the direct limits of A and B.
(8) What is wrong with the following argument? Because of Theorem 4.6, the
sequence 0 ! Z.p1 / ! Z.p1 / ! Z.p1 / ! 0 must be exact, since it can
be obtained as the direct limit of the exact sequences 0 ! Z.pm / ! Z.p2m / !
Z.pm / ! 0 .m 2 N/.

5 Inverse Limits

Inverse Systems Inverse limits are dual to direct limits: we just reverse the
arrows.
Assume fAi j i 2 Ig is a collection of groups, indexed by a poset I, and for
each pair i; j 2 I of indices with i
j there is given a connecting homomorphism
j
i W Aj ! Ai such that
(i) ii is the identity map of Ai for all i 2 I; and
j
(ii) if i
j
k in I, then i jk D ik .
j
In this case, A D fAi .i 2 I/I i g is called an inverse system. By the inverse or
projective limit, or simply limit, of this inverse system is meant a group A such
that
(a) there are maps i W A ! Ai such that i D i j for all i
j; and
j
j
(b) if G is any group with maps i W G ! Ai .i 2 I/ subject to i D i j for i
j,

then there is a unique map  W G ! A satisfying i D i  for all i 2 I.
We write: A D lim Ai ; and call the maps i W A ! Ai canonical.
i2I
j
Theorem 5.1. An inverse system A D fAi .i 2 I/I i g of groups has a limit, unique
up to isomorphism.
Q
Proof. Consider the subgroup A in the direct product A D i2I Ai that consists of
j
all vectors a D .: : : ; ai ; : : : / whose coordinates satisfy i aj D ai for all i
j. This
is in fact a subgroup as is seen immediately. The projection maps i W a 7! ai satisfy
i D i j , so (a) holds for A .
j

To verify (b), let Q G be a group as stated in (b), and forj g 2 G define  W g 7!

.: : : ; i g; : : : / 2 i Ai . Owing to the condition i D i j , we have g 2 A .
Clearly,  W G ! A satisfies i D i  for all i 2 I. If i D i  0 holds also
5 Inverse Limits 61

for  0 W G ! A , then i .   0 / D 0 for all i, so every coordinate projection of

.   0 /G is 0, hence  D  0 .
In order to establish the uniqueness of A , we can mimic the argument at the end
of the last paragraph almost word-by-word. t
u
It is worthwhile noting the following properties of inverse limits.
j
(A) If I is directed, and if in the inverse system A D fAi .i 2 I/I i g all connecting
maps i are monomorphisms, then so are all the i . In fact, assume a 2 A is
j

such that i a D 0. Given j 2 I, there is a k 2 I with i; j

k. Then ik k a D
i a D 0, whence ik monic implies k a D 0. Therefore, j a D jk k a D 0
for all j 2 I, and so a D 0. (Exercise 5 will show that, in general, the same
fails for epimorphisms.)
(B) If I is directed, and if J is a cofinal directed subset in I, then we have
lim Ai D lim Aj :
i2I j2J Q
(C) A is the intersection of kernels of certain endomorphisms of i Ai . For, every
pair i
j in I defines an endomorphism
j
ij W .: : : ; ai ; : : : ; aj ; : : : / 7! .: : : ; ai  i aj ; : : : ; aj ; : : : /:

Comparing this with the proof of Theorem 5.1, it becomes evident that
A D \ij Ker ij :
j
(D) If all the groups in the inverse system A D fAi .i 2 I/I i g are Hausdorff topo-
j
logical groups and the connecting maps i areQ continuous homomorphisms,
then the inverse limit A is a closed subgroup of i Ai .which is equipped with
the product topology/, and the canonical maps i W A ! Ai are continuous.
Indeed, then the endomorphisms ij in (C) are continuous, so their kernels as
well as the intersection Q of the kernels are closed subgroups. A carries the
topology inherited from i Ai , so the continuity of the i is obvious.
Q
Example 5.2. Let A D ˛2J B˛ be the direct product of the groups B˛ . Let I denote the set of all
finite subsets of J, partially ordered by inclusion. For i 2 I, set Ai D ˚˛2i B˛ , and for i  j in I
j j
let i be the projection map Aj ! Ai . This gives rise to an inverse system A D fAi .i 2 I/I i g.
 
We now claim: A D lim Ai Š A. To prove this, let i W A ! Ai be the ith canonical map,
i2I
and i W A ! Ai the ith projection map. By definition, there is a unique map  W A ! A
such that i  D i . If a D 0 for some a 2 A, then i a D i a D 0 for all i 2 I, so  is
monic. If a D .: : : ; ai ; : : : ; aj ; : : : / 2 A , then write ai D b˛1 C    C b˛k with b˛` 2 B˛` if
j
i D f˛1 ; : : : ; ˛k g. If i  j, then by the choice of i , the B˛ -coordinates of ai are identical with

the corresponding coordinates of aj , so a defines a unique .: : : ; b˛ ; : : : / 2 A. A glance at the
definition of  in the proof of Theorem 5.1 shows that .: : : ; b˛ ; : : : / D a , so  is epic as well.
Example 5.3. Let Cn D hcn i be cyclic groups of order pn .n 2 N/, and define maps nnC1 W
CnC1 ! Cn induced by cnC1 7! cn . Now C D fCn .n 2 N/I nm g is an inverse system, and our
claim is that C D lim Cn Š Jp . If n W C ! Cn is the canonical map, and if we define
n2N
n W Jp ! Cn via n .1/ D cn , then by definition there is a unique map  W Jp ! C such
that n  D n for all n 2 N. Since only 0 2 Jp can belong to all Ker n , Ker  D 0 is clear.
Now let c D .b1 ; : : : ; bn ; : : : / 2 C with bn D kn cn .kn 2 Z/; by the choice of nnC1 we have
knC1  kn mod pn , so there is a p-adic integer  such that   kn mod pn for all n. We conclude
that n  D bn , and  must be epic.
62 2 Direct Sums and Direct Products

Example 5.4 (The Intersection of Subgroups is an Inverse Limit). Let fAi j i 2 Ig denote a set of
subgroups of a group A closed under finite intersections. We partially order I by reverse inclusion.
j
The groups Ai , along with the injection maps i W Aj ! Ai .i  Q j/, form an inverse system. Its
inverse limit will be \i2I Ai , because only the constant vectors in i2I Ai can belong to the inverse
limit A .
j
Maps Between Inverse Systems Assume A D fAi .i 2 I/I i g and B D
j
fBi .i 2 I/I i g are inverse systems, indexed by the same poset I. A homomorphism
ˆ W A ! B is a set fi W Ai ! Bi .i 2 I/g of homomorphisms subject to the
requirement that all diagrams of the form

πj
Aj −−−i−→ Ai
⏐ ⏐
⏐
φj 
⏐φ
 i
ρj
Bj −−−i−→ Bi

be commutative for all i

j.
Lemma 5.5. If ˆ W A ! B is a homomorphism between inverse systems, then there
exists a unique map ˆ W A D lim Ai ! B D lim Bi such that, for every i 2 I,
i2I i2I
the diagram

π
A∗ −−−−
i
→ Ai
⏐ ⏐
⏐ ⏐φ
Φ∗   i
ρi
B ∗ −−−−→ Bi

commutes .with canonical maps i ; i /. ˆ is monic, if so are the i .

Q
Proof.
Q TheQhomomorphisms i .i 2 I/ induce a homomorphism N D i i W
i Ai ! i Bi . The commutativity of the diagram before the lemma shows that
N 2 B , hence we can define ˆ W A ! B as
if a D .: : : ; ai ; : : : / 2 A , then a
N
the restriction of . With this ˆ we have i i a D i ai D i ˆ a, establishing the


commutativity of the diagram. If also ˆ0 W A ! B makes the diagram commute

for every i, then i .ˆ  ˆ0 / D 0 for every i, thus ˆ D ˆ0 .
Finally, if all the i are monic, and if ˆ a D 0 for some a 2 A , then i i a D
i ˆ a D 0 implies i a D 0 for every i, whence a D 0. t
u
For the inverse limits of exact sequences, we have a somewhat weaker result than
for direct limits.
j j
Theorem 5.6. Assume A D fAi .i 2 I/I i g; B D fBi .i 2 I/I i g; and C D
j
fCi .i 2 I/I i g are inverse systems over the same index set I. Let ˆ W A ! B and
i i
‰ W B ! C be homomorphisms. If the sequence 0 ! Ai !Bi !Ci ! 0 is exact
5 Inverse Limits 63

for every i 2 I, then the sequence

ˆ ‰
0 ! A D lim Ai !B D lim Bi !C D lim Ci (2.4)
  
i i i

of inverse limits is likewise exact.

Proof. Exactness at A follows from Lemma 5.5. From the definition of ˆ ; ‰  it is
evident that ‰  ˆ D 0. If i ; i ; i denote the canonical maps, then by Lemma 5.5
the diagram
Φ∗ Ψ∗
0 −−−−→ A∗ −−−−→ B ∗ −−−−→ C∗
⏐ ⏐ ⏐
⏐
πi 
⏐
ρi 
⏐σ
 i
φi ψi
0 −−−−→ Ai −−−−→ Bi −−−−→ Ci −−−−→ 0

is commutative for each i 2 I. In order to show the exactness of the top row at B ,
let b 2 Ker ‰  . In view of i i b D i ‰  b D 0 and the exactness of the bottom
j
row, for every i 2 I there is an ai 2 Ai satisfying i ai D i b. For j > i, i i aj D
j j j
i j aj D i j b D i b D i ai , whence i aj D ai as i is monic. We infer that
a D .: : : ; ai ; : : : ; aj ; : : : / 2 A . For this a we have i ˆ a D i i a D i ai D i b for
every i, so ˆ a D b: t
u
Exercise 5 will show that, in general, Theorem 5.6 cannot be improved by putting
! 0 at the end of the exact sequence (2.4). A noteworthy special case when the exact
sequence of inverse limits is exact is as follows.
Proposition 5.7. If in Theorem 5.6 we specialize: A D fAn .n < !/I nnC1 g; B D
fBn .n < !/I nnC1 g; C D fCn .n < !/I nnC1 g, and assume that all the maps nnC1
are epic, then the sequence of inverse limits is exact:

ˆ ‰
0 ! A D lim An !B D lim Bn !C D lim Cn ! 0:

n

n

n

Proof. Let c D .c0 ; : : : ; cn ; : : : / represent an element of C . We now construct by

induction an element b D .b0 ; : : : ; bn ; : : : / 2 B such that ‰  b D c . As 0 is
surjective, there is b0 2 B0 with 0 b0 D c0 . Suppose that, for some n < !, we
have found bi 2 Bi for all i
n such that i bi D ci and i1 i
bi D bi1 . Choose any
0 nC1 0
bnC1 2 BnC1 mapped upon cnC1 by nC1 . Then bn  n bnC1 D n an for some
an 2 An . If anC1 2 AnC1 is such that nnC1 anC1 D an (which exists by hypothesis),
then b0nC1 C nC1 anC1 2 BnC1 is our choice for the next coordinate bnC1 in b . It is
clear that then b 2 B is as desired. t
u
Derived Functor of Inv The inverse systems of abelian groups (with a fixed
index set I) and the morphisms between them form a category Inv.I/. The functor
64 2 Direct Sums and Direct Products

Inv.I/ 7! Ab assigning to an inverse system its inverse limit is left-exact. Since

its right-exactness fails in general, the inverse limit functor has a derived functor,
denoted lim1 . This is especially interesting in case the index set is ! when the
inverse system looks like
1 2 n nC1 nC2
C0 C1  : : : Cn CnC1  : : : :

Then lim1 Cn Š Coker where

n
W .: : : ; cn ; : : : / 7! .: : : ; cn  nC1 cnC1 ; : : : /:
Q Q
denotes the Eilenberg map W n<! Cn ! n<! Cn ; see Jensen [Je], as well
as Schochet [1]. The functor lim1 will be discussed later in Proposition 6.9 in

Chapter 9. We just point out here that, for an exact sequence of inverse systems
in Proposition 5.7, there is an exact sequence

0 ! lim An ! lim Bn ! lim Cn ! lim1 An ! lim1 Bn ! lim1 Cn ! 0:

     
n n n n n n

1Š 2Š nŠ .nC1/Š
Example 5.8. Consider the inverse system fZI nŠg W Z Z  : : : Z 
Z  : : : The Eilenberg map carries the vector .k0 ; : : : ; kn ; : : : / .kn 2 Z/ to the vector
.k0  k1 ; k1  2Šk2 ; : : : ; kn  .n C 1/ŠknC1 ; : : : /. This vector is divisible modulo ˚ Z (which group
is in Ker ), so Coker is divisible. An easy argument shows that the cokernel is torsion-free and
its cardinality is the continuum, hence lim1 fZI nŠg Š Q@0 .
n
pP pP pP
Example 5.9. We now consider three inverse systems: fZ; pg W Z Z Z  : : : , fZ; 1g W
1 1 1  
Z Z Z  : : : , and fZ=pn Z; g W 0 Z=pZ Z=p2 Z  : : : (with canonical maps ).
They fit into the exact sequence

0 ! fZ; pg ! fZ; 1g ! fZ=pn Z; g ! 0

of inverse systems. The lim-lim1 exact sequence (see above) yields the exact sequence 0 ! Z !
 
lim Z=pn Z ! lim1 fZ; pg ! 0, whence
 
lim1 fZ; pg Š Jp =Z Š Q@0 :

F Notes. The so-called Mittag-Leffler condition (not stated) is a most useful sufficient
criterion to guarantee that ! 0 can be put at the end of (2.4). See Jensen [Je].

Exercises
j j
(1) If A D fAi .i 2 I/I i g and B D fBi .i 2 I/I i g are inverse systems, then
j j
A ˚ B D fAi ˚ Bi .i 2 I/I i ˚ i g is again an inverse system. Its limit is the
direct sum of the limits of A and B.
6 Direct Products vs. Direct Sums 65

(2) Let Cn D hcn i be cyclic of order n, and for njm let nm W Cm ! Cn be the
homomorphism induced by cm 7! cn . Then C D fCn .n 2 N/I nm g is an
inverse systemQwhere N is partially ordered by the divisibility relation. Show
that lim C Š p Jp :
n
(3) Let An Š Z.p1 /, and let nnC1 W Z.p1 / ! Z.p1 / be the multiplication by
p. Then the inverse limit of the inverse system A D fAn .n 2 N/I nnC1 g is
isomorphic to the group of all p-adic numbers.
(4) The inverse limit of torsion-free groups is torsion-free, but the inverse limit of
torsion groups need not be torsion.
(5) Let Bn D hbn i Š Z and nm W bm 7! bn for all n
m in N. Let Cn D hcn i Š
Z.pn / and nm W cm 7! cn for n
m. Show that
(a) B D fBn .n 2 N/I nm g and C D fCn .n 2 N/I nm g are inverse systems, and
the epimorphisms n W bn ! cn .n 2 N/ define a map ˆ W B ! C.
(b) The induced homomorphism ˆ W B ! C between the inverse limits is
not epic. [Hint: Z ! Jp .]
(6) The inverse limit of splitting exact sequences need not be exact.
(7) Let A D fAn .n < !/I nnC1 g be an inverse system where the maps nnC1 are
epimorphisms, but not isomorphisms. Then the inverse limit A has cardinality
at least the continuum.

6 Direct Products vs. Direct Sums

One aspect of direct products that deserves special attention is related to their
homomorphisms. There is a remarkable contrast between homomorphisms from
a direct sum and from a direct product: those from direct sums are completely
determined by their restrictions to the components, but not much can be said about
homomorphisms from a direct product, except when either the components or the
target groups satisfy restrictive conditions. A most fascinating result is concerned
with homomorphisms of direct products into direct sums—this is the case that we
wish to explore here. What is a surprising, if not recondite, phenomenon about it is
that it works only up to the first measurable cardinal.
Before entering into the discussion,
Q a simple remark
P might be helpful on infinite
sums in direct products A D i2I Ai . Infinite sums j2J xj do make sense when the
terms are vectors xj D .: : : ; aji ; : : : / .aji 2 Ai / such
P that, for each i 2 I, only a finite
number of ith coordinates aji ¤ 0. (Actually, j2J xj is then a convergent sum in
the product topology.)
Q P
Example 6.1. Let A D n<! An be a countable product. Then x D n<! xn is a well-defined
element of A if xn D .0; : : : ; 0; ann ; an;nC1 ; : : : / .ani 2 Ai / .n zeros).

Maps from Direct Product into Direct Sum We start with a special case which
has independent interest. (‘Reduced’ means no divisible subgroup ¤ 0, and C1 D
\n2N nC denotes the first Ulm subgroup of C.)
66 2 Direct Sums and Direct Products

Q
Theorem 6.2 (Chase [1], Ivanov [5]). Let A D i<! Ai denote a countable
product of groups, and  W A ! C D ˚j2J Cj a homomorphism into the direct
sum of reduced groups Cj . Then there exist integers m > 0; k, as well as a finite
subset J0  J such that

.mBk /
.˚j2J0 Cj / C .˚j2J Cj1 /;
Q
where Bk D ki<! Ai .summand of A/.
Proof. Let j W A ! Cj denote the map  followed by the jth coordinate projection.
Assume the claim is false. Then we can find inductively an increasing sequence
1 D m0 < m1 <    < mk < : : : of integers, a sequence of elements bk 2 mk Bk , and
indices jk 2 J such that

mk jmkC1 ; jk .b` / D 0 for ` < k and jk .bk / … mkC1 Cjk

for all k < !. Indeed, if, for some k < !, we have b` ; m` and j` for all `
k at hand
as required, then jkC1 will be selected as an index not in [`k .supp .b` // such that
jkC1 .mkC1 BkC1 / 6
\n2N nCjkC1 for some proper multiple mkC1 of mk ; this can be
done, since otherwise the claim would be true. Only a finite Pnumber of bk have non-
zero coordinates in any Ai , therefore, the infinite sum a D k<! bk is a well-defined
element in A. Consider
X X X
jk a D jk . b` / D jk . b` / C jk .bk / C jk . b` /;
`<! `<k k<`<!

and observe that in the last sum the first term is 0, and the third term is contained in
mkC1 Cjk , but the second term is not. Since a has a finite support in C, this equation
can hold only for a finite number of indices k—an obvious contradiction. t
u
We state the following theorem for p-groups that involves transfinite heights; its
proof runs parallel to the preceding one.
Q
Theorem 6.3 (Zimmermann-Huisgen [1]). Let A D i<! Ai be a countable
product of p-groups, and  W A ! C D ˚j2J Cj a homomorphism into the direct
sum of reduced p-groups Cj . Given a limit ordinal , there exist an integer k < !,
an ordinal  < , and a finite subset J0  J such that
Y
.p Ai /
.˚j2J0 Cj / C .˚j2J p Cj /: t
u
ki<!

The Measurable Cardinal Phenomenon If we wish to extend the preceding

results to uncountable direct products, then we are confronted with an unusual
phenomenon. There is a natural boundary to the extension: the first measurable
cardinal. The reader who wishes to avoid the following delicate set-theoretical
6 Direct Products vs. Direct Sums 67

arguments can safely assume that there are no such cardinals in our model of ZFC,
and jump to Theorem 6.5.
Recall that a cardinal  is measurable if a set X of cardinality  admits a
countably additive measure  such that  assumes only two values: 0 and 1, and
satisfies .X/ D 1, while .fxg/ D 0 for all x 2 X. Here ‘countably additive’
P means
that if Xi .i < !/ are pairwise disjoint subsets of X, then .[i<! Xi / D i<! .Xi /.
Let f be a function B ! G where B D 2X is the Boolean lattice of all subsets
of a set X, and G ¤ 0 is a group. We will say f is a G-valued measure on X if it
satisfies the following conditions:
(i) f .fxg/ D 0 for every singleton fxg 2 B;
(ii) if V  U are subsets of X, then f .U/ D 0 implies f .V/ D 0;
(iii) if U; V are disjoint subsets of X, then f .U [ V/ D f .U/ C f .V/;
(iv) if Ui .i < !/ are pairwise disjoint subsets of X, then there is n 2 N such that
f .[i<! Ui / D f .U0 / C    C f .Un / and f .Ui / D 0 for all i > n.
We call f non-trivial if f .X/ ¤ 0. The following striking argument is due to J.
Łoś.
Lemma 6.4. If a non-trivial group-valued measure exists on the subsets of the set
X, then jXj is a measurable cardinal.
Proof. Assume f W 2X ! G is a non-trivial G-valued measure on X, G ¤ 0 any
group. We show that then there exists a non-trivial countably additive f0; 1g-valued
measure on X.
Consider all subsets U  X such that f .U/ D 0. From (i)–(iv) we conclude that
these U form a countably additive ideal I in the Boolean lattice B of all subsets of
X. It is readily checked that f induces a countably additive G-valued measure fN on
the Boolean quotient B=I: Let U N 0; : : : ; U
N i ; : : : be pairwise disjoint elements in B=I:
We can find representatives Ui  X of the U N i which are still pairwise disjoint. By
condition (iv), f .Ui / ¤ 0 can hold only for a finite set of indices i; in other words,
B=I is a finite Boolean lattice. Thus B=I has but a finite number of atoms, and on
them fN is not 0. Hence we derive a f0; 1g-valued measure 0 on B=I by selecting an
atom in B=I and define 0 .U/ N to be 1 or 0 according as U N does or does not contain
0
the selected atom. In the obvious manner,  gives rise to a f0; 1g-valued measure 
on B, showing that the set X is measurable. t
u
It is remarkable that Theorems 6.2 and 6.3 generalize to larger products provided
that the cardinality of the set of components is not measurable.
Q
Theorem 6.5 (Dugas–Zimmermann-Huisgen [1]). Let A D i2I Ai be a direct
product, and  W A ! C D ˚j2J Cj a homomorphism where the Cj are reduced
groups. If jIj is not a measurable cardinal, then there are an integer m ¤ 0 and
finite subsets I0  I; J0  J such that
Y
.m Ai /
.˚j2J0 Cj / C .˚j2J Cj1 /:
i2InI0
68 2 Direct Sums and Direct Products

Q
Proof. Consider the set S of all subsets S of I such that for the product AS D i2S Ai
the statement of the theorem holds (i.e., if I is replaced by S). Evidently, if S 2 S,
then all subsets of S also belong to S. Furthermore, S is not only closed under
finite unions (which is evident), but also under countable unions. In fact, if
Sk 2 S .k < !/ are pairwise disjoint subsets, then, for some n < ! we have
[n<k<!
Q Sk Q 2 S—this follows by applying Theorem 6.2 to the countable product
 W k<! . i2Sk Ai / ! C.
Once this has been established, in order to complete the proof it suffices to repeat
the arguments in Łoś’ theorem to conclude that if the claim fails, then I must be
measurable. t
u
Example 6.6. To show that the last theorem may indeed fail for a measurable index set I, let each
Ai denote a copy of the †-cyclic p-group
Q B D ˚k<! Z.p /, and let C D B. To define  W A ! B,
k

pick an a D .: : : ; ai ; : : : / 2 A D i2I Ai . a has only countably many different coordinates (as

elements of B), so the supports of the equal ones give rise to a countable partition of I into disjoint
subsets, exactly one of which has measure 1, and the rest have measure 0. If b 2 B is the element
for which the support is of measure 1, then we set .a/ D b. It is easy to see that this gives rise to
a well-defined homomorphism. It violates the conclusion of Theorem 6.5: m Im  is not contained
in any finite direct sum of cyclic groups in C, for any integer m > 0.

The proof of Theorem 6.5 also applies to verify:

Q
Theorem 6.7 (Zimmermann-Huisgen [1]). Let A D i2I Ai be a product of
p-groups, and  W A ! C D ˚j2J Cj a homomorphism where the Cj are reduced
p-groups. Given a limit ordinal , if jIj D  is not measurable, then there exist an
ordinal  < , as well as finite subsets I0  I; J0  J; such that
Y
.p Ai /
.˚j2J0 Cj /C.˚j2J p Cj /: t
u
i2InI0

F Notes. The peculiar behavior of homomorphisms from a countable direct product into an
infinite direct sum was noticed by Chase [1]. The same phenomenon of larger direct products was
observed by Dugas–Zimmermann-Huisgen [1] up to the first measurable cardinal (just as in the
case of slender groups). By using @1 -complete ultrafilters, Eda [1] gave a generalization to all
cardinals; see Lemma 2.13 in Chapter 13.
Ivanov [1] proves various theorems on so-called Fuchs-44 groups with respect to a class A,
which is closed under extensions, submodules, and direct products. G is such a group if for every
 W G ! ˚i2I Ai with Ai 2 A, there are m 2 N and a finite subset J I such that .mG/ 
˚i2J Ai .

Exercises
Q
(1) The group A D k2N Z.pk / has no unbounded †-cyclic p-group as an
epimorphic image.
7 Completeness in Linear Topologies 69

(2) Let A D Z where  is not measurable, and F a free group. Show that the image
of every homomorphism A ! F is finitely generated.
(3) (Keef) Let Ai .i 2 I/
Qbe an infinite set of unbounded separable p-groups. There
is no epimorphism i2I Ai ! ˚i2I Ai .

7 Completeness in Linear Topologies

Groups that are complete in some linear topology are very special. Therefore, we
examine completeness and the completion processes.
Linear Topologies Assume that a linear topology is defined on the group A in
terms of a filter u in the lattice L.A/ of subgroups of A. The subgroups U 2 u form a
base of open neighborhoods about 0; we label them by a directed index set I, so that
i
j for i; j 2 I means that Ui  Uj . Thus I as a (directed) poset is dual-isomorphic
to a subset of u (which has the natural order relation by inclusion).
By a net in A we mean a set fai gi2I of elements in A, indexed always by I. A net
is said to converge to a limit a 2 A if to every i 2 I there is a j 2 I such that

a k  a 2 Ui for all k  j:

If A is Hausdorff in the topology, then limits are unique; if, however, A fails to
be Hausdorff, then limits are determined only up to mod \i Ui . The classical proof
applies to show that a subgroup B of A is closed in the topology if and only if it
contains the limits of convergent nets whose elements belong to B.
A net fai gi2I is a Cauchy net if to any given i 2 I, there is a j 2 I such that

a k  a ` 2 Ui whenever k; `  j:

Since the Ui are subgroups, ak aj ; a` aj 2 Ui implies ak a` 2 Ui , for the Cauchy
character of a net it suffices to require that ak  aj 2 Ui for all k  j. Clearly, cofinal
subnets of a Cauchy net are again Cauchy nets, and such a subnet converges if and
only if the larger net also converges; moreover, the limits are then the same. To
facilitate discussion and to simplify notation, we shall concentrate without loss of
generality on Cauchy nets fbi gi2I which are neat in the sense that, for every i 2 I,
bk  bi 2 Ui holds for k  i (i.e., j D i can be chosen). If a neat Cauchy net fbi gi2I
converges to a limit b 2 A, then it converges neatly: bk  b 2 Ui for all k  i. In
a group whose topology satisfies the first axiom of countability, Cauchy sequences
fan j n < !g satisfying anC1  an 2 Un for all n 2 N are neat.
Topological Completeness A group A is said to be complete in a topology if
it is Hausdorff, and every (neat) Cauchy net in A has a limit in A. Observe that we
mean by complete groups only Hausdorff groups.
70 2 Direct Sums and Direct Products

Lemma 7.1. A subgroup of a complete group is closed if and only if it is complete

in the induced topology.
Proof. Let G be a subgroup in the complete group C. First assume G is closed in C,
and fgi gi2I is a Cauchy net in G (in the inherited topology). The net is Cauchy in C
too, so it converges to a limit c 2 C which must be in G, since G is closed. Thus G
is complete. Conversely, suppose G is complete in the induced topology, and c 2 C
is the limit of a Cauchy net fgi gi2I with gi 2 G. It is a Cauchy net in G as well, so
has a limit in G, which cannot be anything else than c. Thus G is closed in C. t
u
In the next result the countability hypothesis is essential.
Lemma 7.2. Let B be a closed subgroup of a complete group A that satisfies the
first axiom of countability. Then the factor group A=B is complete in the induced
topology.
Proof. Since B is closed, A=B is Hausdorff. Consider a base of neighborhoods about
0 in A such that U1      Um  : : : with \m2N Um D 0. Let fam C B j m 2 Ng
be a Cauchy sequence in A=B; without loss of generality, we assume that it is neat,
i.e. amC1  am C B  Um C B. We want to lift this Cauchy sequence to a Cauchy
sequence fcm j m 2 Ng in A. Let c1 D a1 , and assume that c1 ; : : : ; cm 2 A have
already been chosen such that ci 2 ai C B and ci  ci1 2 Ui1 for i D 2; : : : ; m.
Then amC1 cm D um Cbm for some um 2 Um ; bm 2 B, and set cmC1 D amC1 bm 2
amC1 C B to have cmC1  cm 2 Um : If limfcm g D a 2 A, then a C B is the limit of
the sequence fam C B j m 2 Ng in A=B. t
u
Recall that if fAj j j 2 Jg is a family of groups, each equipped with aQlinear
topology, say, defined by the filter uj in L.Aj /, then the direct product A D j2J Aj
is given the product (Tychonoff) topology: a subbase of neighborhoods of 0
consists of the subgroups j1 Uji where j W A ! Aj is the jth coordinate projection,
and Uji 2 uj . The product topology is again linear, and the j are continuous, open
homomorphisms. The direct sum A D ˚j2J Aj is a dense subgroup of A .
We should also mention the box topology on the direct product; this topology is
used, e.g., when the components are viewed in the Z-adic topology, and we want
to have this topology on their direct product. We now assume that the same poset I
serves to index a base of neighborhoods about 0 in each Aj . If fUji
Aj j i 2 Ig is
Q of Aj (where Uji
Ujk whenever k
i in I), then the box
a base in the topology
topology on A D j2J Aj is defined to have the subgroups
Y
Ui D Uji .i 2 I/
j2J

as a base of neighborhoods about 0. The box topology on A satisfies the first

countability hypothesis if all the Aj do. The inclusion Ui
j1 Uji for all j shows
that the box topology is finer than the product topology. Hence the projections j
are continuous in the box topology as well.
7 Completeness in Linear Topologies 71

Example 7.3. Actually,Q there are several methods of furnishing a direct product with a linear
topology. E.g., let G D j2I Aj be a product, and F a filter on the index set I. For each X 2 F we
form the subgroup
Y
VX D fg D .: : : ; aj ; : : : / 2 Aj j n.g/ 2 Xg
j2I

(where n.g/ D fj 2 I j aj D 0g denotes the null-set of g), and declare the subgroups VX .X 2 F /
as a base of neighborhoods about 0. This linear topology is Hausdorff if and only if F is a free
filter, i.e. \X2F X D ¿.
Example 7.4. Choose the filter of subsets of I with finite complements. Then the topology defined
in Example
Q 7.3 is the product topology. If Ai .i 2 I/ are non-trivial groups in the discrete topology,
then i Ai is complete in the product topology. The direct sum ˚i Ai is dense in the direct product.

Completions The rest of this section is devoted to the completion of groups in

linear topologies. There are two important completion processes: one is via Cauchy
nets, and another is by using inverse limits. We will employ the second method
which fits better to linear topologies.
Let A be a group with linear topology (not necessarily Hausdorff), and fUi j i 2 Ig
a base of neighborhoods of 0, with I a directed index set: i
j in I if and only if
Ui  Uj . Define the groups Ci D A=Ui , and for j  i in I, the homomorphisms
j j
i W Cj ! Ci via i W a C Uj 7! a C Ui . The limit of the arising inverse system
j
C D fCi .i 2 I/I i g will be denoted by A:M it is furnished with the topology inherited
Q Q
from the product topology of Ci . Thus, if i denotes the ith projection Ci ! Ci ,
then a subbase of neighborhoods of 0 in AM is given by the subgroups U M i D A\
M 1
i 0.
Evidently, A W a 7! .: : : ; a C Ui ; : : : / 2 AM is a homomorphism A ! AM which is
continuous and open, and A Ui D A A \ U M i holds for each i 2 I. It is clear that
Ker A is the intersection of all Ui .
Lemma 7.5. For every group A with a linear topology, the group AM is complete in
the induced topology, and the image of the map  W A ! AM is a dense subgroup of A.
M

Proof. Let aM D .: : : ; ai C Ui ; : : : / 2 A,M and let U M i  AM be an open set. As A ai lies

in the UM i -neighborhood of aM , A A is dense in A. M Therefore, to prove completeness,
we need only verify the convergence of Cauchy nets in A A to elements of A. M A neat
Cauchy net in A A is the image of a neat Cauchy net fbi gi2I in A. We claim that
bM D .: : : ; bi C Ui ; : : : / is the limit of fA bi gi2I . First, bM 2 A,
M since  j .bj C Uj / D
i
bj CUi D bi CUi for j  i. Secondly, the ith coordinate of A bi  bM is 0, so it belongs
to the open set U M i. t
u
Observe that the completion is always Hausdorff, and A W A ! AM is monic if and
only if A had a Hausdorff topology to start with.
Lemma 7.6. If  is a continuous homomorphism of the group A into a complete
group C, then there is a unique continuous homomorphism M W AM ! C such that
M A D .

72 2 Direct Sums and Direct Products

Proof. Let fai j i 2 Ig be a Cauchy net in A converging to the element aM 2 A. M

Continuity implies that fai j i 2 Ig is a Cauchy net in C. If c 2 C is its limit, then
the only possible way of defining a continuous M is to let M W aM 7! c. The rest of the
claim is straightforward. t
u
From this lemma it also follows that the completion AM of A is unique up to
topological isomorphism. Also, A W A ! AM is a natural map, for if  W A ! C
is a continuous homomorphism, then the diagram
φ
A −−−−→ C
⏐ ⏐
⏐
θA 
⏐θ
C
φ̆
Ă −−−−→ C̆

commutes where M is the map whose existence was established in Lemma 7.6.
Our main interest lies in the Z-adic topology, and in completions in that topology.
Therefore, if we say that ‘a group is complete,’ then we always mean completeness
in the Z-adic or p-adic topology (whichever is obvious), unless stated otherwise.
Furthermore, we shall use the special notation AQ for the completion of A in the
Z-adic topology.
In the next theorem we refer to linear compactness; see Sect. 3 in Chapter 6.
Theorem 7.7. Let A be any group.
(i) Its completion in the Z-adic .p-adic/ topology carries the Z-adic .p-adic/
topology.
(ii) Its completion in the finite index topology has a compact topology.
(iii) Its completion in the Prüfer topology carries a linearly compact topology.
Proof. (i) Let AQ D lim A=nA, or, equivalently, AQ D lim A=nŠA whenever
n2N n2N
we consider the collection of subgroups Un D nŠA .n 2 N/ as a decreasing
sequence of neighborhoods about 0. The elements in the induced U Q n have nth
coordinates 0, and it is easy to see that the conditions on the coordinates of
elements on AQ imply that all the ith coordinates in UQ n are 0 for i < n, while all
those for i > n are divisible by nŠ. This means that UQ n D nŠA:
Q
(ii) In the finite indexQtopology, the groups A=Ui are finite, so they are compact.
Thus the product i A=Ui is compact, and the inverse limit AM D lim A=Ui as
i2I
a closed subgroup is also compact.
(iii) The proof is similar to the one in (ii), using the linear compactness of A=Ui in
the Prüfer case.
Example 7.8. Let A D ˚n<! An be furnished with the topology where the subgroups
Uk D ˚kn<! An Q form a base of neighborhoods about 0. The completion of A in this topology is
the direct product n<! An .
7 Completeness in Linear Topologies 73

If the topology fails to satisfy the first axiom of countability, then completeness
may occur in an unexpected situation. This is demonstrated by the following
example where, for a limit ordinal , the p -topology of a p-group A is defined
by declaring the subgroups p A . < / as a base of neighborhoods of 0.
Example 7.9. Suppose  is a limit ordinal not cofinal with !, and let A . < / be p-groups

Q  . Then the A are discrete (and hence complete) in the p -topology.
such that A has length
Consequently, A D < A is complete in the p -topology which is now the box topology on
A (cf. Exercise 2).
Strangely enough, A D ˚< A is complete in the p -topology. To prove this, we show that A
is closed in A . Assume the contrary, i.e. there is x D .: : : ; a ; : : : / 2 A n A in the closure of A.
We can find a sequence 1 <    < n < : : : of ordinals with an ¤ 0. Let sup n D  0 <  and
0 Q
y 2 A such that x  y 2 p < A . Then x and y have equal coordinates in every A with  <  0
which contradicts the fact that x has infinitely many and y only finitely many non-zero coordinates
for  <  0 .

Z-adic Completeness Direct products of complete groups are complete in the

product topology. We wish to point out the following result on the Z-adic topology.
Lemma 7.10. A direct product is complete in the Z-adic topology if and only if
every component is complete in its Z-adic topology.
Proof. Summands Q inherit Z-adic topology and completeness. Conversely, assume
every Aj in G D j2J Aj is Z-adically complete and fgi j i 2 Ig is a neat Cauchy net
in G. Then fj gi j i 2 Ig is a neat Cauchy net in Aj , and if aj 2 Aj is the limit of this
net, then g 2 G with j g D aj is the limit of fgi j i 2 Ig. t
u
F Notes. While the completion in the Prüfer topology may be viewed as a ‘linear compact-
ification,’ completion in the finite index topology is not at all compactification. The latter process
kills the first Ulm subgroup of the group, so it is an embedding only for groups that are Hausdorff in
the finite index topology. A genuine ‘compactification’ can be accomplished by the so-called Stone
compactification. This is the process of embedding A in the group Hom.Hom.A; T/; T/, where T
denotes the circle group R=Z (the inner Hom is furnished with the discrete topology, and the outer
with the compact-open topology).

Exercises

(1) (a) The completions of the groups A and A= \ Ui are the same.
(b) A and A=A1 have the same Z-adic completion.
(2) A direct product is complete in the box topology if and only if every component
is complete.
(3) Every compact (linearly compact) group is complete in its topology.
(4) The direct product of discrete groups is Hausdorff and complete in every
u-topology where u is a free filter.
(5) The inverse limit of complete groups is complete. (Careful with the topology.)
(6) Compare the completions of a group in the finite index and in the Prüfer
topologies.
74 2 Direct Sums and Direct Products

Problems to Chapter 2

PROBLEM 2.1 (J. Dauns). Suppose A has the property that every summand B of
A has a decomposition B D B1 ˚ B2 with B1 Š B2 . Is then A Š A ˚ A?
PROBLEM 2.2. Study the Boolean powers A.B/ of a group A.
Cf. Balcerzyk [3], and especially, Eda [2].

PROBLEM 2.3. Represent a p-group A as a direct limit A Š lim AŒpn

. How
!n
does the structure of A change if the connecting monomorphisms AŒpn
! AŒpnC1

are modified?
Q Q
PROBLEM 2.4. Suppose  W A D Ai ! C D <@1 Cj is a homomorphism of
a product into an @1 -product. Can we say something about where the image must
be contained (like Theorem 6.5)?
Chapter 3
Direct Sums of Cyclic Groups

Abstract The study of important classes of abelian groups begins in this chapter. Not counting
the finite and finitely generated groups, the class of direct sums of cyclic groups is perhaps the best
understood class.
We give a fairly detailed account of free abelian groups, and discuss the presentation of groups
via generators and defining relations. Several sections are devoted to direct sums of cyclic groups
(called †-cyclic groups); these groups share most useful properties, and can easily be characterized
by cardinal invariants. We present a few criteria for such groups, and establish several remarkable
results, e.g. Kulikov’s theorem that passage to subgroups preserves †-cyclicity. We draw attention
to the method of smooth chains, which became the most important tool in the theory, and provides
basic machinery for several results to come.
We shall cover some of the aspects of almost free groups, but shall not pursue their theory
farther, due to the sophisticated set-theoretical arguments required.
In this chapter, in a number of proofs we have to use purity, so readers should be familiar with
the fundamental results on pure subgroups (in Chapter 5) before studying the second part of this
chapter.

1 Freeness and Projectivity

Free Abelian Groups By a free (abelian) group is meant a direct sum of

infinite cyclic groups. If these cyclic groups are generated by the elements xi .i 2 I/,
then the free group will be

F D ˚i2I hxi i:

The set fxi gi2I is a basis of F. The elements of F are linear combinations

g D n1 xi1 C    C nk xik .k  0/ (3.1)

with different xi and non-zero integers ni . In view of the definition of direct sums,
two such linear combinations represent the same element of F exactly if they differ
at most in the order of the terms. Addition is performed in the obvious way by
adding the coefficients of the same xi .

© Springer International Publishing Switzerland 2015 75

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_3
76 3 Direct Sums of Cyclic Groups

We can define F formally by starting with a set X D fxi gi2I of symbols, called
a free set of generators, and declaring F as the set of all formal expressions (3.1)
under the mentioned equality and addition. We say that F is the free group on the
set X.
Example 1.1. An immediate example for a free group is the multiplicative group of positive
rational numbers. The prime numbers form a free set of generators.

Needless to say, F is, up to isomorphism, uniquely determined by the cardinal

number  D jIj of the index set I. Thus we are justified to write F for the free group
with  free generators.  is also called the rank of the free group F, in symbols,
rk F D  (for the discussion of rank, see Sect. 4).
Theorem 1.2. The free groups F and F are isomorphic exactly if the cardinals 
and  are equal.
Proof. We need only verify the ‘only if’ part of the assertion. Observe that if F is
a free group with free generators xi .i 2 I/, then an element (3.1) of F belongs to
pF if and only if pjn1 ; : : : ; pjnk . Hence, if p is a prime, then F=pF is a vector space
over the prime field Z=pZ of characteristic p with basis fxi C pFgi2I , and so its
cardinality is pjIj or jIj according as I is finite or infinite. Thus jF=pFj completely
determines jIj. t
u
The Universal Property Free groups enjoy a universal property formulated in
the next theorem which is frequently used for the definition of free groups.
Theorem 1.3 (Universal Property of Free Groups). Let X be a free set of
generators of the free group F. Any function f W X ! A of X into any group A
extends uniquely to a homomorphism  W F ! A. This property characterizes free
sets of generators, and hence free groups.
Proof. Write X D fxi gi2I , and f .xi / D ai 2 A. There is only one way f can be
extended to a homomorphism  W F ! A, namely, by letting

g D .n1 xi1 C    C nk xik / D n1 ai1 C    C nk aik :

(The main point is that the uniqueness of (3.1) guarantees that  is well defined.) It
is immediate that  preserves addition.
To verify the second part, assume that a subset X of a group F has the stated
property. Let G be a free group with a free set Y D fyi gi2I of generators, where
the index set is the same as for X. By hypothesis, the correspondence f W xi 7! yi
extends to a homomorphism  W F ! G; this cannot be anything else than the
map n1 xi1 C    C nk xik 7! n1 yi1 C    C nk yik .  is injective, because the linear
combination of the yi is 0 only in the trivial case.  is obviously surjective, and so it
is an isomorphism. t
u
Mapping X onto a generating system of a given group, we arrive at the following
result which indicates that the group Z is a generator of the category Ab (‘generator’
in the sense used in category theory).
1 Freeness and Projectivity 77

Corollary 1.4. Every group with at most  generators is an epimorphic image of a

free group with  generators. t
u
Consequently, every group A can be embedded in a short exact sequence


0 ! H ! F !A ! 0;

where F is free group, and H D Ker . (We will see shortly that H is likewise free.)
This is called a free resolution of A. It is far from being unique, because both F and
 can be chosen in many ways.
If  is an infinite cardinal, then F has 2 subsets, and hence at most 2 subgroups
and factor groups. We conclude that there exist at most 2 pairwise non-isomorphic
groups of cardinality
. (We will learn in Corollary 3.8 in Chapter 11, that 2 is
the precise number.)
The next two theorems are fundamental, they are quoted most frequently.
Theorem 1.5. Suppose that B is a subgroup of a group A such that A=B is a free
group. Then B is a summand of A, i.e., A D B ˚ C for a subgroup C Š A=B.
Proof. That only free factor groups can share the stated property will follow
from Theorem 1.7. In order to show that free groups do have this property,
by Lemma 2.4 in Chapter 2, it suffices to verify the claim for A=B Š Z only, say
A=B D ha C Bi with a 2 A. The elements of A=B are the cosets n.a C B/ D
na C B .n 2 Z/ (all different). Hence A D B ˚ hai is immediate. t
u
This theorem can also be phrased by saying that an exact sequence 0 ! B !
A ! F ! 0 with a free group F is necessarily splitting.
Subgroups of Free Groups In the next theorem we study the subgroups of free
abelian groups. Recall the famous result in group theory that subgroups of (non-
commutative) free groups are again free. For abelian groups the situation is the
same. To prove this, we use a well ordering of the index set.
Theorem 1.6. Subgroups of free groups are free.
Proof. Let F be a free group on the set X, which we now assume to be well ordered,
say X D fx g < for some ordinal . Thus F D ˚ < hx i. For  < , define
F D ˚< hx i, and set G D G \ F for a subgroup G < F. Clearly, G D
G C1 \ F , so G C1 =G Š .G C1 C F /=F . The last factor group is a subgroup
of F C1 =F Š hx i; thus either G C1 D G or G C1 =G is an infinite cyclic group.
From Theorem 1.5 we conclude that G C1 D G ˚ hg i for some g 2 G C1
(which is 0 if G C1 D G ). It follows that the elements g generate the direct sum
˚ < hg i in G. This must be all of G, since G is the union of the G . < /. u t
78 3 Direct Sums of Cyclic Groups

Projectivity Call a group P projective if every diagram

P
⏐
⏐φ
ψ 
α β
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

with exact row can be completed by a suitable homomorphism W P ! B to a

commutative diagram, i.e. ˇ D . We then say:  is lifted to .
Theorem 1.7. A group is projective if and only if it is free.
Proof. Let ˇ W B ! C be a surjective map, and F a free group with a homomorphism
 W F ! C. For each xi in a free set X D fxi gi2I of generators of F, we pick
an element bi 2 B such that ˇbi D xi —this is possible, ˇ being epic. Owing
to Theorem 1.3, the correspondence xi 7! bi .i 2 I/ extends to a homomorphism
W F ! B. The maps ˇ and  are equal on the generators of F, so ˇ D , and
F is projective.
Next, let P be a projective group, and ˇ W F ! P an epimorphism, F a free
group. By definition, the identity map 1P W P ! P can be lifted to a map W P ! F,
i.e. ˇ D 1P . Thus P is a summand of F, so a free group by Theorem 1.6. The
isomorphism P Š P completes the proof. t
u
Thus ‘free’ and ‘projective’ have the same meaning for abelian groups. There-
fore, free resolutions may also be called projective resolutions.
Projective Cover The projective cover of a group A is defined as a projective
group P with a surjective map  W P ! A such that Ker  is a superfluous subgroup
of P. Projective covers are duals of injective hulls (to be discussed in Chapter 4), but
in contrast to their dual counterparts, they rarely exist.
Example 1.8. (a) The cyclic group Z.p/ has no projective cover. If it had one, Z would be a good
candidate, but then the kernel would not be superfluous.
(b) However, Z.p/ regarded as a Z.p/ -module does have a projective cover, since pZ.p/ is
superfluous in Z.p/ .

Theorem 1.9. A group has a projective cover if and only if it is free.

Proof. We show that the zero-group is the only superfluous subgroup of a free
group F. If H ¤ 0 is a subgroup in F, then there is a prime p with H 6
pF (since
\p pF D 0). Evidently, .H C pF/=pF ¤ 0 is a summand of the Z=pZ-vector space
F=pF, say, with complement G=pF for some pF < G < F. Then G C H D F where
G is a proper subgroup of F, so H cannot be superfluous. t
u
Defining Relations We shall discuss briefly the method of defining a group
in terms of generators and relations. Though this is well known from general
group theory, in the commutative case there are simplifications worthwhile to be
pointed out.
1 Freeness and Projectivity 79

Let fai gi2I be a set of generators of a group A, and  W F ! A an epimorphism

from a free group F D ˚i2I hxi i such that xi D ai for each i 2 I. Ker  consists of
those linear combinations m1 xi1 C    C mk xik 2 F with integral coefficients mi for
which m1 ai1 C    C mk aik D 0 holds in A. These equalities are called the defining
relations relative to the generating system fai gi2I .
It follows that the group A is completely determined by giving a set fai gi2I of
generators along with the set of all defining relations:

A D hai .i 2 I/j mj1 ai1 C    C mjk aik D 0 .j 2 J/i (3.2)

(since we are dealing exclusively with abelian groups, the commutativity relations
are not listed). Indeed, if (3.2) is given, then A is defined as the factor group
F=H, where F is a free group on the free set fxi gi2I of generators, and H is the
subgroup of F, generated by the elements mj1 xi1 C    C mjk xik for all j 2 J.
The relations between the given generators of A are exactly those which are listed
in (3.2), and their consequences. (The emphasis is on the non-existence of more
relations.) Equation (3.2) is said to be a presentation of A.
Example 1.10. A presentation of a free group F with free generators fxi gi2I is given as
F D hxi .i 2 I/ j ¿i (there are no relations between the generators). Of course, there are numerous
other presentations; e.g. Z D hx; y j 2x  3y D 0i.
Example 1.11. The group C D hx j nx D 0i for n 2 N is cyclic of order n.
F Notes. The material on free groups is fundamental, and will be used in the future without
explicit reference. Though in homological algebra, projectivity is predominant, in abelian group
theory freeness seems to prevail. Fortunately, for abelian groups, freeness and projectivity are
equivalent, while for modules, the projectives are exactly the direct summands of free modules.
Projective modules are rarely free; they are free over principal ideal domains (but not even over
Dedekind domains that are not PID), and over local rings (Kaplansky [2]).
Theorem 1.6 holds for modules over left principal ideal domains. Submodules of projectives
are again projective if and only if the ring is left hereditary, i.e., all left ideals are projective.
Theorem 1.2 holds over commutative rings or under the hypothesis that at least one of  and 
is infinite. There exist, however, rings R such that all free R-modules ¤ 0 with finite sets of
generators are isomorphic. It is perhaps worthwhile pointing out that every R-module is free if and
only if R is a field, and every R-module is projective exactly if R is a semi-simple artinian ring.
The property that all R-modules have projective covers characterizes the perfect rings, introduced
by H. Bass.
Hausen [6] defines a group P -projective for an infinite cardinal  if it has the projective
property with respect to all exact sequences 0 ! A ! B ! C ! 0 with jCj < . She establishes
various properties of -projective groups, e.g. P is -projective if and only if, for every subgroup
G with jP=Gj < , there is a summand H of P such that G  H and G=H is a free group.

Exercises

(1) Let F be a free group on n free generators. If n elements a1 ; : : : ; an 2 F generate

F, then this set is a basis of F.
80 3 Direct Sums of Cyclic Groups

(2) Prove the following converse of Theorem 1.5: a group F is free if it has the
property that whenever B < A and A=B Š F, then B is a summand of A.
(3) Give a presentation of Z.p1 /, and one of Q.
(4) Let A be presented by a set of generators and defining relations, and assume
that the set of generators is the union of two disjoint subsets, fbi gi2I and fcj gj2J ,
such that each of the defining relations contains only generators from the same
subset. Then A D B ˚ C, where B is generated by the bi , and C by the cj .
(5) Let A be presented by a set of generators and defining relations, and B by
a subset of these generators and defining relations. Show that letting the
generators of B correspond to themselves qua generators of A induces a
homomorphism B ! A.
(6) For every set of generators, there is a minimal set of defining relations relative
to these generators (i.e., no relation can be omitted). [Hint: Theorem 1.6.]
˛
(7) Let 0 ! A1 ! A2 !A3 ! 0 be an exact sequence, and i W Fi ! Ai .i D 1; 3/
epimorphisms where Fi are free. If W F3 ! A2 is such that ˛ D 3 , then
1 ˚ W F1 ˚ F3 ! A2 is epic, and its kernel is Ker 1 ˚ Ker 3 .
(8) Let 0 ! F1 ! F2 !    ! FnP ! 0 be an exact sequence of finitely generated
free groups. Prove the equality nkD1 .1/k rk Fk D 0.
(9) Assume fAn j n 2 Zg is a set of groups. Verify the existence of free groups
Fn .n 2 Z/ and a long sequence

˛n2 ˛n1 ˛n ˛nC1

: : : ! Fn1 ! Fn ! FnC1 ! : : :

such that ˛n1 ˛n D 0 and Ker ˛n = Im ˛n1 Š An for every n 2 Z.

2 Finite and Finitely Generated Groups

We turn our attention to groups with a finite number of generators. First, we discuss
finite groups separately. Though this is a special case of the general theory of finitely
generated groups (to be developed independently), a short, direct approach to the
theory of finite groups is not without merit.
Finite Groups We start with a simple lemma.
Lemma 2.1. Let A be a p-group that contains an element g of maximal order pk for
an integer k > 0. Then hgi is a direct summand of A.
Proof. If A is infinite, then use Zorn’s lemma to argue that there is a subgroup B of A
maximal with respect to the property B\hgi D 0. To show that A D hgi˚B equals
A, by way of contradiction assume that some a 2 A does not belong to A . Replacing
a by pi a if necessary, we may also suppose that pa 2 A , i.e. pa D mg C b for some
m 2 Z; b 2 B. By the maximality of the order of g, we have pk1 mg C pk1 b D
pk a D 0. Hence pk1 mg D 0, so m must be divisible by p, say, m D pm0 . Then
a0 D a  m0 g … A satisfies pa0 D b. By the maximal choice of B, hB; a0 i \ hgi ¤ 0,
2 Finite and Finitely Generated Groups 81

thus 0 ¤ ra0 C b0 D sg for some r; s 2 Z; b0 2 B. This can happen only if .r; p/ D 1,

since pa0 2 B. But then pa0 ; ra0 2 A implies a0 2 A , a contradiction. t
u
Fundamental Theorem on Finite Abelian Groups The first structure theorem
in the history of group theory was the famous Basis Theorem on finite abelian
groups.
Theorem 2.2 (Frobenius–Stickelberger [1]). A finite group is the direct sum of a
finite number of cyclic groups of prime power orders.
Proof. Thanks to Theorem 1.2 in Chapter 2, the proof reduces at once to p-groups.
In a finite p-group A ¤ 0, we select an element g of maximal order. By the preceding
lemma, A D hgi ˚ B for some subgroup B. Since B has a smaller order than A, a
trivial induction completes the proof. t
u
There is a uniqueness theorem attached to the preceding result. Again, it suffices
to state it for p-groups.
Theorem 2.3. Two direct decompositions of a finite p-group A into cyclic groups
are isomorphic.
Proof. In a direct decomposition of A collect the cyclic summands of equal orders
into a single summand to obtain a courser decomposition A D B1 ˚    ˚ Bk where
each Bi is 0 or a direct sum of cyclic groups of fixed order pi . Evidently, pk1 A D
pk1 Bk is the socle of Bk , it is an elementary p-group, its dimension (as a Z=pZ-
vector space) tells us the number of cyclic components in Bk . As this socle depends
only on A, the number of cyclic summands of order pk is independent of the choice
of the direct sum representation of A. In general, pi1 AŒp
D pi1 Bi Œp
˚    ˚
pi1 Bk Œp
modulo pi AŒp
D pi BiC1 Œp
˚    ˚ pi Bk Œp
is a Z=pZ-vector space Š
pi1 Bi Œp
whose dimension is equal to the number of cyclic summands (of order pi )
in Bi . The same argument shows that this dimension is independent of the choice of
the selected direct decomposition of A. t
u
Finitely Generated Groups We proceed to the discussion of finitely generated
groups. We start with a preliminary lemma.
Lemma 2.4 (Rado [1]). Assume A D ha1 ; : : : ; ak i, and n1 ; : : : ; nk are integers such
that gcdfn1 ; : : : ; nk g D 1. Then there exist elements b1 ; : : : ; bk 2 A such that

A D hb1 ; : : : ; bk i with b1 D n1 a1 C    C nk ak :

Proof. We induct on n D jn1 j C    C jnk j: If n D 1, then let b1 D ˙ai for any i, and
the claim is evident. Next let n > 1. Then at least two of the ni are different from
0, say, jn1 j  jn2 j > 0. Since either jn1 C n2 j < jn1 j or jn1  n2 j < jn1 j, we have
jn1 ˙n2 jCjn2 jC  Cjnk j < n for one of the two signs. gcdfn1 ˙n2 ; n2 ; : : : ; nk g D 1
and the induction hypothesis imply that A D ha1 ; : : : ; ak i D ha1 ; a2 a1 ; : : : ; ak i D
hb1 ; : : : ; bk i with b1 D .n1 ˙ n2 /a1 C n2 .a2 a1 / C n3 a3 C    C nk ak D n1 a1 C
   C nk ak : t
u
82 3 Direct Sums of Cyclic Groups

The main result on finitely generated groups is our next theorem which is
regarded as the first major result in the abstract structure theory of infinite abelian
groups. It plays an important role in several applications.
Theorem 2.5. The following conditions on a group A are equivalent:
(i) A is finitely generated;
(ii) A is the direct sum of a finite number of cyclic groups;
(iii) the subgroups of A satisfy the maximum condition.
Proof. (i) ) (ii) assume A is finitely generated, and a minimal generating set of A
contains k elements. Pick such a set with k generators, say, a1 ; : : : ; ak , with the
additional property that a1 has minimal order, i.e. no other set of k generators
contains an element of smaller order. If k D 1, then A D ha1 i, and we are done.
So let k > 1, and as a basis of induction, assume that B D ha2 ; : : : ; ak i is a direct
sum of cyclic groups. Thus it suffices to verify that A D ha1 i ˚ B, which will follow
if we can show that ha1 i \ B D 0.
By the choice of k, we have o.a1 / > 1. Working toward a contradiction, suppose
that ha1 i \ B ¤ 0, i.e. m1 a1 D m2 a2 C    C mk ak ¤ 0 with 0 < m1 < o.a1 /.
Let d D gcdfm1 ; : : : ; mk g; and write mi D dni . Then gcdfn1 ; : : : ; nk g D 1, and
from Lemma 2.4 we conclude that A D ha1 ; : : : ; ak i D hb1 ; : : : ; bk i with b1 D
n1 a1 C n2 a2 C    C nk ak . Here db1 D 0, thus o.b1 / < o.a1 /, contradicting the
choice of a1 . Thus ha1 i \ B D 0.
(ii) ) (iii) Let A D ha1 i ˚    ˚ hak i. If k D 1, then A is cyclic, and every non-
zero subgroup is of finite index in A. Hence the subgroups satisfy the maximum
condition. (iii) will follow by a trivial induction if we can show that A D B ˚ C has
the maximum condition on subgroups whenever both B and C share this property. If
A1
  
An
: : : is an ascending chain of subgroups in A, then A1 \ B
  

An \ B
: : : is one in B, so from some index m on, all An \ B are equal to Am \ B.
For n > m we have An =.Am \B/ D An =.An \B/ Š .An CB/=B
A=B Š C, whence
we conclude that from a certain index t > m on all factor groups At =.Am \ B/, and
hence all subgroups At , are equal.
(iii) ) (i) The set S of all finitely generated subgroups of A is not empty, so by
hypothesis (iii) A contains a maximal finitely generated subgroup G. For any a 2 A,
hG; ai is still finitely generated. Hence hG; ai D G, thus A D G, and A is finitely
generated. t
u
Let us point out two immediate consequences of Theorem 2.5. First, every finitely
generated group is the direct sum of a finite group and a finitely generated free
group (follows from (ii)). Secondly, subgroups of finitely generated groups are
again finitely generated (follows from (iii)).
The most essential part of the preceding theorem is the first implication. We give
another quick proof, reducing it to Theorem 2.2. If we can show that A=T is free
.T D t.A//, then A Š T ˚ A=T by Theorem 1.5, and we are done. Thus, it is
enough to consider A D ha1 ; : : : ; an i torsion-free. To start the induction on n, there
is nothing to prove if n D 1, since then A Š Z trivially. Let U=han i denote the
torsion subgroup of A=han i. Then A=U is torsion-free and has a smaller number of
2 Finite and Finitely Generated Groups 83

generators, so it is free. Hence A Š U ˚ A=U (again by Theorem 1.5), where U is

a finitely generated group isomorphic to a subgroup of Q, so it is cyclic.
Stacked Basis Theorem A third proof of Theorem 2.5 is based on the following
theorem which is of considerable interest in its own right (see the more general The-
orem 6.5). We say fai gi2I is a basis of A if A D ˚i2I hai i.
Theorem 2.6. If H is a subgroup of the free group F of finite rank k, then F and H
have ‘stacked bases:’

F D ha1 i ˚    ˚ hak i and H D hb1 i ˚    ˚ hbk i

such that there are non-negative integers m1 ; : : : ; mk satisfying

b i D mi a i .i D 1; : : : ; k/ and mi1 jmi .i D 2; : : : ; k/:

Proof. We select a free basis fx1 ; : : : ; xk g of F with the following extremal property:
H contains an element b1 D n1 x1 C   C nk xk with a minimal positive coefficient n1 .
In other words, for another basis of F, or for another permutation of the basis
elements, or for other elements of H, the leading positive coefficient is never less
than n1 .
The first observation is that n1 jni .i D 2; : : : ; k/. For, if ni D qi n1 C ri .qi ; ri 2
Z; 0
ri < n1 /; then we can write b1 D n1 a1 C r2 x2 C    C rk xk where fa1 D
x1 C q2 x2 C    C qk xk ; x2 ; : : : ; xk g is a new basis of F. By the special choice of
fx1 ; : : : ; xk g, we must have r2 D    D rk D 0. The same argument shows that if
b D s1 x1 C    C sk xk .si 2 Z/ is any element of H, then s1 D qn1 for some q 2 Z.
Hence b  qb1 2 hx2 i ˚    ˚ hxk i D F1 . We conclude that F has a decomposition
F D ha1 i ˚ F1 such that H D hb1 i ˚ H1 , where b1 D n1 a1 and H1
F1 . Using
induction hypothesis for the pair H1 ; F1 , we infer that F has a basis fa1 ; : : : ; ak g and
H has a basis fb1 ; : : : ; bk g such that bi D mi ai for some non-negative integers mi .
It remains to establish the divisibility relation m1 jm2 (the others will follow by
induction). Write m2 D tm1 C r with t; r 2 Z; 0
r < m1 . Then fa D a1 C
ta2 ; a2 ; : : : ; ak g is a new basis of F, in terms of which we have b1 C b2 D m1 a1 C
.tm1 C r/a2 D m1 a C ra2 2 H. The minimality of m1 D n1 implies r D 0. t
u
With the aid of Theorem 2.6, we can reprove the implication (i) ) (ii)
in Theorem 2.5. If A is generated by k elements, then A Š F=H, where F is a
free group on a set of k elements. Choosing stacked bases for F and H, as described
in Theorem 2.6, we obtain

A Š ha1 i=hm1 a1 i ˚    ˚ hak i=hmk ak i:

Consequently, A is the direct sum of cyclic groups: the i th summand is cyclic of

order mi if mi > 0, and infinite cyclic if mi D 0. The numbers mi are called
elementary divisors.
84 3 Direct Sums of Cyclic Groups

Fundamental Theorem on Finitely Generated Groups Of course, the num-

bers mi in Theorem 2.6 are not necessarily prime powers, but we can decompose
the finite summands into direct sums of cyclic groups of prime power orders. Cyclic
groups of prime power orders are indecomposable (and so are the infinite cyclic
groups), so we can claim the fundamental theorem:
Theorem 2.7 (Fundamental Theorem on Finitely Generated Abelian Groups).
A finitely generated group is the direct sum of finitely many indecomposable cyclic
groups, each of which is of prime power order or infinite cyclic. t
u
Whenever one has a direct decomposition, then the standard question is: to what
extent is the decomposition unique? This question is fully answered in the following
theorem.
Theorem 2.8. Any two direct decompositions of a finitely generated group into
indecomposable cyclic groups are isomorphic.
Proof. If A is finitely generated, then by Theorem 2.7 A D tA ˚ F where F Š
A=tA is finitely generated free. Both summands are uniquely determined by A up to
isomorphism. Theorems 2.3 and 1.2 guarantee the uniqueness of the decompositions
of the summands, whence the claim is evident. t
u
Invariants Thus in the decompositions of a finitely generated group A, the
orders of the indecomposable cyclic summands (but not the summands themselves)
are uniquely determined. These orders are referred to as the invariants of A.
For instance, the invariants of A Š Z ˚ Z ˚ Z ˚ Z.p/ ˚ Z.p2 / ˚ Z.q3 / ˚
Z.q3 / (with primes p; q) are: 1; 1; 1; p; p2 ; q3 ; q3 : We also say: A is of type
.1; 1; 1; p; p2 ; q3 ; q3 /.
Consequently, with every finitely generated group A, a finite system of symbols
1 and prime powers is associated. Not only is it uniquely determined by A, but
it also determines A up to isomorphism, i.e. two finitely generated groups are
isomorphic if and only if they have the same system of invariants (maybe in
different orders)—this fact is expressed by saying that this is a complete system
of invariants. Moreover, these invariants are independent in the sense that, for an
arbitrary choice of a finite system of symbols 1 and prime powers, there exists a
finitely generated group exactly with this system of invariants (this is obvious).
Example 2.9. Let C.m/ denote the multiplicative group of those residue classes of integers modulo
r
the integer m D pr11    pkk (canonical form) which are relatively prime to m. Its order is given by
Euler’s totient function '.m/. Elementary number theory tells us that
r
(a) C.m/ is the direct product of the groups C.pi i / for i D 1; : : : ; kI
(b) for odd primes p, C.pr / is cyclic of order '.pr / D pr  pr1 ;
(c) C.4/ is cyclic of order 2, while C.2r / .r
3/ is of type .2; 2r2 /.

Kaplansky’s Test Problems In his famous little red book [K], Kaplansky raises
the question about criteria for satisfactory structure theorems. He lists two test
problems that such theorems must pass in order to qualify ‘satisfactory.’ These are:
2 Finite and Finitely Generated Groups 85

Test Problem I. If the group G is isomorphic to a direct summand of H, and H is

isomorphic to a direct summand of G, are then G and H isomorphic?
Test Problem II. If G ˚ G Š H ˚ H, are G and H isomorphic?
Evidently, the structure theorem on finitely generated groups passes the test with
flying colors: both answers are easy ‘yes.’ However, some of the theorems that will
be discussed later on will fail one or both test problems.
F Notes. Whenever it seems instructive or interesting, we shall make historical remarks that
are intended to give a sense of the way in which the subject has developed, but are in no way a
comprehensive survey of the relevant contributions. As far as the fundamental theorem on finite
abelian groups is concerned, it is not clear how far back in time one needs to go to trace its origin.
It was F.C. Gauss who established a decomposition in number theory reminiscent to it. That time
the concept of a group was unknown, it took a long time to formulate and to prove the fundamental
theorem in the present form; see Frobenius–Stickelberger [1]. The theorem on finitely generated
groups may be credited to H.J.S. Smith [Phil. Trans. 151, 293–326 (1861)]. He reduced matrices
with integral entries to canonical form that bears his name.
This is the first time we encounter a structure theorem, so a few comments are in order. Such
a theorem (on any class of algebraic systems) is supposed to be in terms of easily recognizable
invariants, like natural numbers, cardinal or ordinal numbers, but they can be matrices with integral
entries, etc. ‘Invariants’ mean by definition that they are exactly the same for isomorphic objects.
A set of invariants is complete if we can reconstruct from it the object within the class by using
a method typical for the class (for finitely generated groups, this method consists in forming the
direct sum of cyclic groups with the given invariants as orders). Finally, independence means
that the system of invariants can be chosen arbitrarily, i.e. without additional restriction (in this
case, arbitrary prime powers and the sign 1, each with arbitrary multiplicities). The system
of invariants for finitely generated groups is most satisfactory, it has served as a prototype for
structure theorems in algebra.
The Kaplansky test problems have been discussed for various classes, mostly with negative
answers. de Groot modified Test Problem I by asking the isomorphy of G and H if G has a summand
G1 Š H and H has a summand H1 Š G such that, in addition, G=G1 Š H=H1 is also satisfied.
There are numerous generalizations of the theorems in this section. Kaplansky [J. Indian Math.
Soc. 24, 279–281 (1960)] proved that, for integral domains R, the torsion parts of finitely generated
R-modules are summands exactly if R is a Prüfer domain. There is an extensive literature on
commutative rings over which finitely generated torsion modules are †-cyclic. Unless the ring is
left noetherian, finitely generated left modules are different from finitely presented ones which are
somewhat better manageable. Finitely presented R-modules are †-cyclic if and only if R is an
elementary divisor ring, i.e., every matrix over R can be brought to a diagonal form by left and
right multiplications by unimodular matrices (Kaplansky [Trans. Amer. Math. Soc. 66, 464–491
(1949)]). In this case, Theorem 2.6 still holds true.

Exercises

(1) A group is finite if and only if its subgroups satisfy both the maximum and the
minimum conditions.
(2) A finite group A is cyclic exactly if jAŒp
j
p for every prime p.
 86 3 Direct Sums of Cyclic Groups

(3) (a) If the integer m divides the order of the finite group A, then A has both a
subgroup and a factor group of order m.
(b) (G. Frobenius) In a finite p-group, the number of subgroups of a fixed order
(dividing the order of the group) is  1 mod p.
(4) A group is isomorphic to a subgroup of the finite group A if and only if it is
isomorphic to a factor group of A.
(5) The number of non-isomorphic groups of order m D pr11    prkk (canonical form
of m) is equal to P.r1 /    P.rk /, where P.r/ stands for the number of partitions
of r into positive integers.
(6) If A; B are finite groups such that, for every integer m, they contain the same
number of elements of order m, then A Š B.
(7) A set fa1 ; : : : ; ak g of generators of a finite group is a basis if and only if
the product o.a1 /    o.ak / is minimal among the products of orders for all
generating sets.
(8) In a finitely generated group, every generating set contains a finite set of
generators.
(9) (a) The sum of all the elements of a finite group A is 0, unless A contains just
one element a of order 2, in which case the sum is equal to this a.
(b) From (a) derive Wilson’s congruence .p  1/Š  1 mod p, p a prime.
(10) Let A; B be finitely generated groups. There is a group C such that both A
and B have summands isomorphic to C, and every group that is isomorphic
to summands of both A and B is isomorphic to a summand of C.
(11) Any set of pairwise non-isomorphic finite (finitely generated) groups has
cardinality
@0 .
(12) (Cohn, Honda, E. Walker) Finitely generated groups A have the cancellation
property: A˚B Š A˚C implies B Š C, or equivalently, if G D A1 ˚B D A2 ˚C
with A1 Š A Š A2 , then B Š C. [Hint: enough for A1 D hai cyclic of order 1
or prime power pr .]
(13) If A and B are finitely generated groups, and if each is isomorphic to a subgroup
of the other, then A Š B.
(14) A surjective endomorphism of a finitely generated group is an automorphism.

3 Factorization of Finite Groups

In most cases, the fundamental theorem is instrumental in solving problems related

to finite abelian groups. However, there are notable exceptions where it seems the
fundamental theorem is totally irrelevant. One of these is Hajós’ theorem on the
‘factorization’ of finite abelian groups.
The problem goes back to a famous conjecture by H. Minkowski in 1896 on tiling
the n-dimensional Euclidean space by n-dimensional cubes. If the space is filled
gapless such that no two cubes have common interior points, then it was conjectured
that there exist cubes sharing n1-dimensional faces. The conjecture was rephrased
as an abelian group-theoretical problem, and solved in this form by G. Hajós. We
3 Factorization of Finite Groups 87

discuss briefly this celebrated result. The proof involves group rings, and therefore
at some point we have to switch to the multiplicative notation. It is reasonable to do
this right away.
Thus in this section, all groups are finite, written multiplicatively. Accordingly, 1
will denote the identity element of groups.
Direct Products of Subsets If S1 ; : : : ; Sk are non-empty subsets of a multiplica-
tive group G, then we say that G is a direct product of these subsets, in notation,

P : : : S
G D S1  P k; (3.3)

if each element g 2 G can be written uniquely as g D s1  : : :  sk with si 2 Si . This

definition is in line with the definition of direct sum of subgroups. We will call the
components Si factors of G, and (3.3) a factorization of G. We obviously have
(A) The cardinality of a factor is a divisor of the group order.
(B) Every subgroup H of G is a factor: G D H S P if S is a complete set of
representatives mod H.
(C) A factor Si can be replaced by gSi with any g 2 G. For this reason, there is no
loss of generality in assuming that each factor contains 1 2 G.
Periodic and Cyclic Subsets A subset P is called periodic and a non-unit g 2 G
a period of P if gP D P. Subgroups are trivially periodic. If g is a period, then so
are the elements ¤ 1 of hgi. In this case, P is the set union of certain cosets mod hgi.
P then either S or T is
Lemma 3.1. If G D hai is cyclic of order pn , and G D ST,
periodic.
Proof. Set S D fan1 ; : : : ; ank g; T D fam1 ; : : : ; am` g .ni ; mj  0/ and form the
polynomials S.z/ D zn1 C    C znk ; T.z/ D zm1 C    C zm` (with indeterminate z).
Hypothesis implies

S.z/T.z/  1 C z C z2 C    C zp
n 1 n
mod zp  1:

It follows that S.z/T.z/ is divisible by the pn th cyclotomic polynomial ˆn .z/ D

1 C zp C    C z.p1/p . This polynomial is known to be irreducible over Q, so
n1 n1

n1
one of the factors, say, S.z/ is divisible by ˆn .z/. Hence we conclude that ap is a
period of S. t
u
Our main concern is with factors that are cyclic subsets in the sense that they are
of the form

Œa
n D f1; a; : : : ; an1 g .2
n
o.a//

for some a 2 G. We need two preliminary lemmas.

Lemma 3.2 (Hajós [1]). A cyclic subset is periodic if and only if it is a group.
88 3 Direct Sums of Cyclic Groups

Proof. Let P D Œa
n be periodic with period g 2 G, so P D hgiS P for some S  G
where 1 2 S may be assumed. Evidently, g D at for some t 2 N, and P contains
1; a; : : : ; at1 , the powers of at , as well as their cosets mod hat i. This means hai  P,
so P is a (cyclic) group. t
u
Lemma 3.3 (Hajós [1]). A cyclic subset C can be written as a direct product of
cyclic subsets of prime orders such that C is a subgroup if and only if one of the
factor cyclic subsets is a subgroup.
Proof. Suppose C D Œc
n , and let n D p1    pk , a product of primes. It is an easy
computation to show that

P p1
p2 
Œc
n D Œc
p1 Œc P p1 pk1
pk :
P : : : Œc

If C is a subgroup, i.e. if cn D 1, then the last factor is also a subgroup. For the
converse, we show that if C D haiS P for some 1 ¤ a 2 C; S  C, then C has to be
a subgroup. In fact, a is then a period of C, and the claim follows from Lemma 3.2.
t
u
Hajós’ Theorem We can now state the main theorem.
Theorem 3.4 (Hajós [1]). If a finite group G is the direct product of cyclic subsets,

P : : : Œa
G D Œa1
n1  P k
nk ;

then one of the factors is a subgroup.

Proof. In view of Lemma 3.3, for the proof we may assume that the orders ni of
p
the factors are primes pi . Suppose Œak
pk is not a subgroup, i.e. ak k ¤ 1. Then from
ak G D G we derive that
p
P : : : Œa
Œa1
p1  P k1
pk1  ak k D Œa1
p1 
P : : : Œa
P k1
pk1 ; (3.4)
p
that is, the product on the right is periodic with period ak k . Delete as many factors
as possible until no more factor can be omitted without violating the periodicity
of the product. Let a 2 G denote a period of a shortest periodic subset P D
P : : : Œa
Œa1
p1  P h
ph :
Consider the subgroup H D ha1 ; : : : ; ah i of G. As P is a factor of G, it is also a
factor of H, thus jPj divides jHj, i.e. p1    ph j jHj. If we can show that jHj is the
product of not more than h primes, then P D H will follow. We will then have a
similar direct product decomposition for H, a group of smaller order, so observing
that the case h D 1 is trivial, an obvious induction will complete the proof.
It remains to substantiate the claim concerning the order of the subgroup H. We
interrupt the proof to verify a lemma that will do the job.
The Crucial Lemma The crux of the problem is to find a proper statement,
more general than actually needed for the proof, that will allow an induction to
3 Factorization of Finite Groups 89

complete the proof. We need the group ring ZŒG

to formulate such a lemma. ZŒG

consists of elements of the form

x D m1 g 1 C    C m` g ` .gi 2 G; mi 2 Z/ (3.5)

which we add and multiply according to the usual rules, respecting the multiplica-
tion rules in G.
In what follows we will assume that the expression of x is canonical, i.e. all the
gi are different, and all mi ¤ 0. With this in mind, we go on to define hxi as the
subgroup of G generated by the elements gi in (3.5), and denote by .x/ the number
of prime factors in the order of hxi. Finally, the symbol a will have double meaning:
for an a 2 G, it is either 1 C a C    C ap1 for a prime p, or 1  a. Thus a 2 ZŒG

and hai D hai.

Lemma 3.5 (Hajós [1]). Assume that the equation

x a1    ak D 0 (3.6)

holds in the group ring ZŒG

, where ai 2 G; x 2 ZŒG
. If no factor ai can be deleted
without violating the validity of the equation, then

.x; a1 ;    ; ak /  .x/ < k: (3.7)

Proof. We begin with the verification in case k D 1. Thus we have .x a/ D 0 with

non-zero factors, and what we wish to prove amounts to a 2 hxi. If a D 1  a, then
x D ax, which implies that there are b1 ; b2 2 G in the normal form of x such that
b1 D ab2 . Hence a 2 hxi in this case. If a D 1 C a C    C ap1 for some prime p,
then by multiplication by 1  a we get x.1  ap / D 0, whence ap 2 hxi. On the other
hand, from x.a C    C ap1 / D x we conclude that b1 D ai b2 for some b1 ; b2 2 hxi
and 1
i
p  1. Thus also ai 2 hxi, and therefore a 2 hxi.
We continue with induction on n D .a1 /C  C.ak /. If n D 1, then k D 1, and
we are done. Assuming k  2, we rewrite (3.6) in the form .x a1    aj /ajC1    ak D 0
for j < k, and apply the induction hypothesis to obtain

..x a1    aj /; ajC1 ; : : : ; ak /  .x a1    aj / < k  j .1

j < k/:

The index of the subgroup h.x a1    aj /; ajC1 ; : : : ; ak i in hx; a1 : : : ; aj ; ajC1 ;

: : : ; ak i evidently divides the index of hx a1    aj i in hx; a1 ; : : : ak i (cf. Exercise 1).
Hence, from the last inequality we get

.x; a1 ; : : : ; ak /  .x; a1 ; : : : ; aj / < k  j .1

j < k/: (3.8)

If .aj / D 1 for all j

k, then clearly .x; a1 ; : : : ; ak1 /  .x/
k  1, along
with (3.8) for j D k  1 yields (3.7). If, e.g., .ak /  2, then by multiplication by
p1
1  ak or by 1 C ak C    C ak for some prime p, we can replace the factor ak by
90 3 Direct Sums of Cyclic Groups

a0 D 1  a0 with 1
.a0 / D .ak /  1. After deleting superfluous factors ai , and
renumbering, we get

x a1 ; : : : a` a0 D 0 .0
`
k  1/

where no factor can be omitted, not even the last one. By induction hypothesis,
.x; a1 ; : : : ; a` ; a0 /  .x/
`. In case ` D 0, we have .x; a0 /  .x/ D 0, and
a0 2 hxi, thus .x; ak /  .x/
1. This, together with (3.8) for j D k  1, leads
to (3.7). If `  1, then manifestly .x; a1 ; : : : ; a` /.x/
`, hence along with (3.8)
for j D ` it yields the desired (3.7). t
u
Resuming the proof of Theorem 3.4, we rewrite (3.4) (after deleting superfluous
factors) as an equation in ZŒG
:

a1    ah  .1  a/ D 0
p 1
where ai D 1 C ai C    C ai i .i D 1; : : : ; h/. Applying Lemma 3.5 to the case
x D 1; we obtain .a1 ; : : : ; ah ; a/ D .a1 ; : : : ; ah ; a/
h, and a fortiori
.a1 ; : : : ; ah /
h. As pointed out above, this completes the proof. t
u
Example 3.6 (Hajós). Theorem 3.4 may fail if the factors are not cyclic. This is shown by the
following examples.
(a) Let G D hai  hbi  hci be a direct product of cyclic groups where a; b; c are generators of
orders 4; 4; 2, respectively. Then

P
G D f1; agf1; P
bgf1; a2 ; ab2 ; a3 b2 ; c; a2 bc; a2 b3 c; b2 cg:

(b) Let G D hai  hbi  hci where all the generators a; b; c are of order 4. Then

P
G D f1; agf1; P
bgf1; P
cgf1; a2 b; b2 c; c2 a; a2 b3 ; b2 c3 ; c2 a3 ; a2 b2 c2 g:

F Notes. The proof above is based on the original proof by Hajós [1] with essential
simplifications due to L. Rédei and T. Szele. Various modified versions of the problem have been
considered. One version requires the factors to be simulated subsets: a subset S of a group is
simulated if it is obtainable from a subgroup by replacing an element by an arbitrary group element.
There is an extensive literature on this difficult subject, most advanced papers are written recently
by A.D. Sands and S. Szabó. There are remarkable connections to tessellations.
It is hard to understand why so far no evidence of a link has been found between the
fundamental theorem on finite abelian groups and the Hajós theorem. Such a link would probably
avoid group rings, but it seems doubtful we could have found our way through without making use
of them.
A generalized, still unsolved version of Minkowski’s conjecture was formulated by O.H. Keller.
Its algebraized version says that if G D SŒa P    Œa
P 1
n 1  P k
nk with a subset S G, then one of the
elements ai i equals s1 s1
n
2 for some s1 ; s2 2 S.
4 Linear Independence and Rank 91

Exercises

All groups are finite, written multiplicatively.

(1) If A; B; C are finite index subgroups of the group G, and B
C, then ŒAC W AB

divides ŒC W B
.
(2) If in a group G, the subset P D Œa
p Œb
P q .a; b 2 G/ is periodic with different
primes p; q, then one of the factors is a subgroup.
(3) (Sands) Let G be cyclic of order 8. Find G D ST P such that none of S; T can
be replaced by a subgroup. [Hint: hai D f1; a2 gf1; P a; a4; a5 g.]
(4) (de Bruijn) A subset S of a cyclic group of order n is periodic if and only if
there is a proper divisor d of n such that S.z/ (defined above in Lemma 3.1) is
divisible by the polynomial f .z/ D .zn  1/.zd  1/1 .
(5) Assume G is a finite group of one of the types .2; 2; 2/; .2; 22 /; .2; 2; 3/; .2; 3; 3/;
.3; 32 /; .3; 3; 3/. If G D ST
P for subsets S; T, then S or T is periodic. [Hint: S
or T contains 2 or 3 elements.]
(6) (de Bruijn) Let G be an elementary 2-group with generators a1 ; a2 ; a3 ; b1 ;
b2 ; b3 . None of the factors is periodic in the factorization

G D f1; a1 a3 b1 ; a2 a3 ; a1 a2 b1 ; b2 ; a1 a2 a3 b2 ; a1 b1 b2 ; a2 a3 b1 b2 g
f1; a1; a2 ; a1 a2 ; b3 ; a3 b3 ; b1 b3 ; a3 b1 b3 g:

(7) (de Bruijn) Let G D hai be cyclic of order 72. It factorizes into two
non-periodic subsets: f1; a8; a16 ; a18 ; a26 ; a34 g and fa18 ; a54 ; a24 ; a60 ; a48 ; a12 ;
a17 ; a41 ; a65 ; a45 ; a69 ; a21 g.

4 Linear Independence and Rank

Motivated by linear independence and dimension in vector spaces, we are in search

for corresponding notions in groups.
Linear Independence Linear independence in groups can be defined in two
inequivalent ways: one permits only elements of infinite order to be in the system,
while the other makes no such restriction, and as a result, it is useful for torsion
and mixed groups as well. With that said, we proceed to introduce the more useful
version.
A set fa1 ; : : : ; ak g of non-zero elements in a group is called linearly indepen-
dent, or briefly, independent if

n1 a1 C    C nk ak D 0 .ni 2 Z/ implies n1 a1 D    D nk ak D 0: (3.9)

92 3 Direct Sums of Cyclic Groups

More explicitly, this means that ni D 0 if o.ai / D 1 and o.ai /jni if o.ai / is finite. By
definition, 0 is not allowed to be in an independent system. An infinite family L of
group elements is (linearly) independent if every finite subset of L is independent.
Thus independence is by definition a property of finite character.
Lemma 4.1. A subset L D fai gi2I .0 … L/ of a group is independent if and only if

hLi D ˚i2I hai i: (3.10)

Proof. If L is independent, then the intersection of the cyclic group hai i with the
subgroup generated by L n fai g is necessarily 0; hence, hLi is the direct sum of
the hai i for i 2 I. Conversely, if (3.10) holds, then a linear combination n1 ai1 C
   C nk aik D 0 (with different i1 ; : : : ; ik 2 I) can hold only in the trivial way:
n1 ai1 D    D nk aik D 0. t
u
An element g 2 A is said to depend on a subset L of A if there is a dependence
relation

0 ¤ ng D n1 a1 C    C nk ak .n; ni 2 Z/ (3.11)

for some elements ai 2 L. Thus g depends on L exactly if hgi \ hLi ¤ 0. A subset

K depends on L if every element of K depends on L.
Every element a in an independent system can be replaced, without violating
independence, by a non-zero multiple ma. Therefore, by replacing elements of finite
order by multiples of prime power order, from every independent system we can get
one in which each element is either of infinite or of prime power order.
An independent system M in A is maximal if there is no independent system
in A that properly contains M. Every element ¤ 0 of A depends on a maximal
independent system. By Zorn’s Lemma, every independent system is contained in a
maximal one. Moreover, if the original system contained only elements of infinite
or prime power orders, then a maximal one containing it can also be chosen to have
this property.
Lemma 4.2. An independent system is maximal if and only if it generates an
essential subgroup.
Proof. It suffices to observe that a non-zero element a 2 A depends on an
independent system M if and only if hai \ hMi ¤ 0. u
t
Rank of a Group By the rank rk.A/ of a group A is meant the cardinal number
of a maximal independent system containing only elements of infinite and prime
power orders. If we consider only independent systems with elements of infinite
order (of orders that are powers of a fixed prime p) which are maximal with respect
4 Linear Independence and Rank 93

to this property, then the cardinality of the system is called the torsion-free rank
rk0 .A/ (p-rank rkp .A/) of A. From the definitions it is evident that the equation
X
rk.A/ D rk0 .A/ C rkp .A/ (3.12)
p

holds with p running over all primes. Obviously, rk.A/ D 0 means A D 0.

At this point the natural question is: how unique are these various ranks? In order
to legitimize them, we need to show:
Theorem 4.3. The ranks rk.A/; rk0 .A/; rkp .A/ of a group A are invariants of A.
Proof. It suffices to prove that rk0 .A/ and rkp .A/ are independent of the choice of
the maximal independent system defining them.
It is routine to check that rk0 .A/ D rk.A=tA/. As a consequence, in proving the
invariance of rk0 .A/, we may assume without loss of generality that A is a torsion-
free group. Let fa1 ; : : : ; ak g and fb1 ; : : : ; b` g be two maximal independent
P systems
in A. Then there are integers m; mi ; n; nj with m; n ¤ 0 such that mai D `jD1 mij bj
P
and nbj D kiD1 nji ai . Hence

X
k X̀
mnai D nij mjh ah
hD1 jD1

where the corresponding coefficients on both sides must be equal. This means that
the product of matrices knij k  kmjh k is a scalar matrix mnEk (Ek denotes the k  k
identity matrix). This is impossible if k < `, thus k  ` must hold. For reasons
of symmetry, k D ` follows, i.e. equivalent finite independent systems contain the
same number of elements. This tells us that rk0 .A/ is well defined whenever it is
finite.
If rk0 .A/ is infinite, then we show that rk0 .A/ D jAj (A is still torsion-free). The
inequality
is obvious. To prove the converse, we choose a maximal independent
system L D fai gi2I . For every 0 ¤ g 2 A, there is n 2 N such that ng 2 hLi, and if
ng D ng0 .g0 2 A/, then g D g0 . Hence we conclude that jAj
jLj@0 D jLj:
Turning to the ranks rkp .A/, it is clear that rkp .A/ D rk.Tp / where Tp denotes
the p-component of T D tA. Hence it is enough to verify the claim for p-groups
A. Now if fai gi2I is a maximal independent system, then so is fpmi1 ai gi2I where
pmi D o.ai /: Therefore, rkp .A/ is the same as the rank of the socle s.A/. The socle
is a Z=pZ-vector space, its dimension is obviously the same as its rank as a group.
The uniqueness of the vector space dimension implies the uniqueness of rk.s.A// D
rkp .A/: t
u
94 3 Direct Sums of Cyclic Groups

There is another important cardinal invariant associated with groups. This is the
dimension of the Z=pZ-vector space A=.tA C pA/ which we shall call the p-corank
of A, and will be denoted as

rkp .A/ D dim A=.tA C pA/:

We will see later that this is the rank of the torsion-free part of p-basic subgroups
of A.
F Notes. The torsion-free rank of A is often defined as the dimension of the Q-vector space
Q ˝ A (then the uniqueness of rk0 .A/ follows from that of the vector space dimension). The rank
as we use here has been generalized to modules, called Goldie dimension.

Exercises

(1) Show that rk.Q/ D 1; rk.Q=Z/ D @0 ; and rk.Jp / D 2@0 for each prime p.
(2) Prove that rk.A/ D 1 exactly if A is isomorphic to a subgroup of Q or to a
subgroup of Z.p1 / for some prime p.
(3) Let B be a subgroup of A. Prove that: (i) rk.B/
rk.A/; (ii) rk.A/
rk.B/ C
rk.A=B/; (iii) rk0 .A/ D rk0 .B/ C rk0 .A=B/.
(4) The non-zero subgroups Bi .i 2 I/ of A generate their direct sum in A if and
only if every subset L D fbi gi2I with one bi ¤ 0 from each Bi is independent.
(5) A group of rank   @0 has 2 different subgroups.

5 Direct Sums of Cyclic Groups

The simplest kinds of infinitely generated groups are the direct sums of cyclic
groups. These groups admit a satisfactory classification as we shall see below. We
will feel fortunate if we are able to prove that certain groups under consideration are
direct sums of cyclic groups.
For brevity, a direct sum of cyclic groups will be called a †-cyclic group.
Kulikov’s Theorem A †-cyclic p-group contains no elements ¤ 0 of infinite
height. However, the absence of elements of infinite height does not ensure that a
p-group is †-cyclic. We are looking for criteria under which a p-group is †-cyclic.
Theorem 5.1 (Kulikov [1]). A p-group A is †-cyclic if and only if it is the union of
a countable ascending chain of subgroups,

A0
A1
  
An
: : : ; (3.13)

such that the heights of elements ¤ 0 in An .computed in A/ are bounded.

5 Direct Sums of Cyclic Groups 95

Proof. The stated condition is necessary: if A is a †-cyclic p-group, then in a

decomposition, collect the cyclic summands of the same order pn , for every n. If
we denote their direct sum by Bn , then the subgroups An D B1 ˚    ˚ Bn .n < !/
(with bound n  1 on the heights) satisfy the stated condition.
For the proof of sufficiency, suppose that the chain (3.13) is as stated. Since
we may adjoin the trivial subgroup to the beginning of the chain (3.13) and repeat
subgroups without violating the hypothesis, it is clear that there is no loss of
generality in assuming that n1 is a bound for the heights in An , that is, An \pn A D 0
for every n < !.
Accordingly, we consider the set of all chains 0 D C0
C1
  
Cn
: : : of
subgroups of A such that

An
Cn and Cn \ pn A D 0 for every n < !:

Define the chain of the Cn to be less than or equal to the chain of the Bn if and only
if Cn
Bn for all n < !: The set of all such chains in A is non-empty and is easily
seen to be inductive, so Zorn’s lemma applies to conclude that there exists a chain
0 D G0
SG1
  
Gn
: : : that is maximal in the sense defined. Needless to
say, A D n<! Gn .
The group Gn contains only elements of order
pn , so Gn \ pn1 A is in the socle
of Gn . Select a Z=pZ-vector space basis Ln of Gn \ pn1 A, and set L D [n<! Ln . For
every ci 2 L of h.ci / D ni choose an ai 2 A such that pni ai D ci . The claim is that
A0 D h : : : ; ai ; : : : i D ˚i hai i is equal to A.
First we show that hLi D AŒp
. Since evidently hLn i D Gn \ pn1 A, all the
elements ¤ 0 in hLn i are of height exactly n  1, so the hLn i generate their direct
sum, hLi D ˚n<! hLn i. Assume, as a basis of induction on k, that Gk Œp
D hL1 i ˚
   ˚ hLk i. Let a 2 GkC1 Œp
n Gk . By maximality, hGk ; ai \ pk A ¤ 0, thus 0 ¤
g C ra D b 2 pk A with some g 2 Gk ; r 2 Z, where r D 1 may be assumed.
Therefore, g C a 2 GkC1 \ pk A D hLkC1 i, thus a and hence GkC1 Œp
is contained in
hL1 i ˚    ˚ hLkC1 i. Consequently, hLi D AŒp
follows.
Assume that, for some n 2 N, we have proved that every element of A of order

pn belongs to A0 ; for n D 1, this was done in the preceding paragraph. Pick an

a 2 A of order pnC1 .n  1/: Then pn a 2 hLi, so we have pn a D m1 c1 C    C m` c`
with some cj 2 L; mj 2 Z. Let c1 ; : : : ; cr be of height  n, and crC1 ; : : : ; c` of height
< n. Then in the equation

pn a  m1 c1      mr cr D mrC1 crC1 C    C m` c`

the left-hand side is of height  n, while the right-hand side is contained in Gn1 ;
so both sides are 0. If we write mj cj D pn m0j aj .j
r/, then a  m01 a1      m0r ar
is of order
pn , so it is contained in A0 by induction hypothesis. Hence a 2 A0 as
well. t
u
Prüfer’s Theorems As corollaries we obtain the following two important,
frequently quoted results.
96 3 Direct Sums of Cyclic Groups

Theorem 5.2 (Prüfer [1], Baer [1]). A bounded group is †-cyclic.

Proof. If A is bounded, then it can have but a finite number of non-zero
p-components Ap . These components are also bounded, so we can apply Theo-
rem 5.1 with all subgroups in (3.13) equal Ap , to conclude that each Ap is †-cyclic.
t
u
Theorem 5.3 (Prüfer [1]). A countable p-group is †-cyclic if and only if it
contains no elements ¤ 0 of infinite height.
Proof. Only the ‘if’ part requires a verification. Suppose A is a countable p-group
without elements of infinite height. If fa0 ; : : : ; an ; : : : g is a generating set of A, then
A is the union of its finite subgroups An D ha0 ; : : : ; an i .n < !/, where the heights
of the elements are obviously bounded. The claim follows from Theorem 5.1. t
u
The following example shows that countability is an essential hypothesis in
Theorem 5.3.
Example 5.4 (Kurosh). Let A be the torsion part of the direct product of the cyclic groups
Z.p/; : : : ; Z.pn /; : : : . Then A is a p-group of the power of the continuum, without elements of
infinite height (by the way, each Z.pn / is a summand of A). Assume, by way of contradiction,
that A is †-cyclic, say, A D ˚n<! Bn where Bn is a direct sum of cyclic groups of fixed order
pn . Consider the socles Sn D ˚ni<! pi1 Bi ; they form, with increasing n, an infinite properly
descending chain such that Sn consists of those elements of AŒp
which are of heights
n  1.
Clearly,

a D .c1 ; : : : ; cn ; : : : / 2 AŒp
.cn 2 Z.pn //

is of height
n  1 if and only if c1 D    D cn1 D 0: This shows that each factor group
Sn =SnC1 .n D 1; 2; : : : / is of order p. Hence Bn Œp
Š Sn =SnC1 implies that the Bn are finite, and
so A is countable, a contradiction. An @1 -generated pure subgroup of A containing the direct sum
˚n Z.n/ yields an example of smallest cardinality.
A quicker counterexample is available if we make use of the isomorphism of basic subgroups:
no uncountable p-group with countable basic subgroup is †-cyclic.

Kulikov’s criterion can be generalized to arbitrary cardinalities as follows (we

make use of purity which will be discussed in Chapter 5).
Theorem 5.5 (Hill [13]). A p-group A is †-cyclic if it is the union of an ascending
chain (3.13) of †-cyclic pure subgroups An .n < !/.
Proof. For countable A, sufficiency is easy: list the generators in a sequence:
a1 ; a2 ; : : : ; ai ; : : : . If 0 D B0 < B1
  
Bi is a chain of finite pure subgroups
of A such that a1 ; : : : ; aj 2 Bj for all j
i, then choose a finite summand BiC1 of
an An containing both Bi and aiC1 ; such an n must exist. Then Bi as a bounded pure
subgroup is a summand of A, and A D [i<! Bi . Since BiC1 D Bi ˚ Ci for some
Ci
A, we get A D ˚i2N Ci .
The proof for the uncountable case is the exact analog of Theorem 7.5; we leave
the details to the reader. t
u
Isomorphy of Decompositions Though a group may have several decompo-
sitions into a direct sum of cyclic groups, one can establish a strong uniqueness
5 Direct Sums of Cyclic Groups 97

statement, just as in the finitely generated case. (Actually, one can prove more: the
Krull-Schmidt property holds for †-cyclic groups.)
Theorem 5.6. Any two direct decompositions of a group into direct sums of infinite
cyclic groups and cyclic groups of prime power orders are isomorphic.
Proof. First assume that A is a p-group. Collecting the cyclic summands of the same
order, we get a decomposition A D ˚n<! Bn where Bn is a direct sum of cyclic
groups of the same order pn . As in Example 5.4, we can argue that Bn Œp
Š Sn =SnC1
where Sn is the set of elements of heights  n  1 in AŒp
. The latter group is
independent of the representation of A as direct sum of cyclic p-groups, and the
dimension of Sn =SnC1 as a Z=pZ-vector space determines the number of cyclic
summands of order pn in any decomposition of A as a †-cyclic group.
In the general case, A D B˚C where B is a †-cyclic torsion group and C is a free
group. Then both B and C have unique decompositions (rk C being well defined),
so the same holds for A. t
u
Subgroups of †-Cyclic Groups It is extremely important and most useful that
the property of being †-cyclic is inherited by subgroups.
Theorem 5.7 (Kulikov [2]). Subgroups of †-cyclic groups are again †-cyclic.
Proof. First we dispose of the case when the group A is a p-group. By Theorem 5.1,
A is the union of an ascending chain A0
A1
  
An
: : : of subgroups, where
the heights of elements of An are bounded, say, kn is a bound in An . A subgroup B is
the union of the chain

A0 \ B
A1 \ B
  
An \ B
: : :

where the heights of elements of An \ B, computed in B, do not exceed kn . By virtue

of Theorem 5.1, B is †-cyclic.
Turning to the general case, let A be an arbitrary †-cyclic group, and B a
subgroup of A. Clearly, tB D B \ tA, and so

B=tB D B=.B \ tA/ Š .B C tA/=tA

A=tA;

where A=tA is a free group. By Theorem 1.6, B=tB is free, whence Theorem 1.5
implies that B D tB ˚ C for some free subgroup C of B. By what has been shown in
the preceding paragraph, tB is a direct sum of cyclic p-groups. Thus B is †-cyclic.
t
u
Corollary 5.8 (Kulikov [2]). Any two direct decompositions of a †-cyclic group
have isomorphic refinements.
Proof. In view of Theorem 5.7, each summand is †-cyclic. Replacing each
summand by a direct sum of cyclic groups of orders 1 or prime power, we arrive at
refinements that are isomorphic, as is guaranteed by Theorem 5.6. t
u
The next lemma provides information about pure subgroups in free groups.
98 3 Direct Sums of Cyclic Groups

Lemma 5.9. (a) A finite rank pure subgroup of a free group is a summand.
(b) (Erdős [1]) A pure subgroup H of a free group F contains a summand of F
whose rank is the same as the rank of H.
Proof. (a) A finite rank pure subgroup H is contained in a finitely generated
summand of the free group F. Then F=H is finitely generated and torsion-free,
so a free group. Therefore, H is a summand of F.
(b) If H is of finite rank, then it is a summand of F, and we are done. So assume that
H is of infinite rank . Let B D fb˛ j ˛ < g be a basis of F, and consider finite
subsets Bi of B such that hBi i \ H ¤ 0: Select a maximal pairwise disjoint set S
of such subsets Bi , and a non-zero hi in each hBi i \ H: Then the pure subgroup
hhi i? is a summand of hBi i, and hence K D ˚hhi i? is a summand of F, and so
of H. Write F D hSi ˚ G where G is generated by the basis elements not in any
member of S. Now G \ H ¤ 0 is impossible, because then the basis elements
b˛ occurring in a linear combination of a non-zero element in this intersection
form a finite subset disjoint from every finite subset in S—this contradicts the
maximality of S. Therefore, G \ H D 0. Manifestly, the cardinality of the set of
all basis elements b˛ occurring in members of S is the same as the cardinality
of S. Hence G \ H D 0 implies that rk K D rkhSi D rk F=G  rk H D :
t
u
F Notes. Various properties of †-cyclic groups have been investigated that are shared by
larger classes of groups. The name of Fuchs-5-group is used in the literature for a group in which
every infinite set is contained in a direct summand of the same cardinality. Trivial examples for such
groups are direct sums of countable groups. Hill [8] proved that for every uncountable cardinal 
there exist p-groups with this property that need not be direct sums of countable groups. The
existence of non-free @1 -separable torsion-free groups shows that not all torsion-free Fuchs-5-
groups are direct sums of countable subgroups.

Exercises

(1) For a group A, the following conditions are equivalent: (a) A is elementary;
(b) every subgroup of A is a summand; (c) A is torsion with trivial Frattini
subgroup; (d) A contains no proper essential subgroup.
(2) The direct product of   @0 copies of the cyclic group Z.pk / is a direct sum
of 2 copies of Z.pk /.
(3) Let A; B be †-cyclic groups.
(a) A ˚ A Š B ˚ B implies A Š B.
(b) A.@0 / Š B.@0 / fails to imply A Š B even if A; B are finitely generated.
(4) Let A be a countable direct sum of cyclic groups of order p2 , and B Š A ˚
Z.p/. The isomorphy classes of subgroups (and factor groups) of A are equal
to those of B, but A 6Š B.
6 Equivalent Presentations 99

(5) (Dlab)
(a) Let A be a bounded p-group, and S D fai gi2I a subset of A such that the
cosets ai C pA .i 2 I/ generate A=pA. Then S generates A.
(b) Every generating set of a bounded p-group contains a minimal generating
set (i.e. no generator can be omitted).
(6) (Szele) Improve on Example 5.4 by exhibiting an example of cardinality @1 .
(7) Let B D ˚k2N Z.pk /.
(a) Every countable p-group is an epimorphic image of B.
(b) Each p-group of infinite cardinality  is an epic image of B./ .
(8) A is †-cyclic if it contains a †-cyclic subgroup G such that A=G is bounded.
(9) (Dieudonné [1]) Let G be a p-group that contains a subgroup A such that G=A
is †-cyclic. Suppose that A is the union of a chain A0
A1
  
An
: : :
such that the heights of elements of An , computed in G, are bounded. Then G
is †-cyclic.
(10) Let A; G be p-groups, and assume C < A with †-cyclic A=C. If the
homomorphism  W C ! G does not decrease heights, then it extends
to a homomorphism A ! G. [Hint: if pn a 2 C, there is g 2 G with
.pn a/ D pn g.]
(11) An equational class or variety of groups is a class of groups that is closed
under isomorphism, the formations of subgroups, epic images, and direct
products. Prove that the following is a complete list of equational classes
of abelian groups:
(a) the class of all abelian groups;
(b) for every positive integer n, the class of n-bounded abelian groups.

6 Equivalent Presentations

This section is concerned with special kind of presentations. First, †-cyclic groups
will be considered.
Presentation with Stacked Basis We say that the group A has a presentation

with stacked bases if there is a short exact sequence 0 ! H ! F !A ! 0 where
F D ˚i2I hxi i is a free group and H D ˚i2I hni xi i is a free subgroup with ni  0
(see Theorem 2.6).
An obvious necessary condition for a group to be presented with stacked bases
is that it be a †-cyclic group. Kaplansky raised the question whether or not every
presentation of a †-cyclic group is with stacked bases. The affirmative answer was
given by Cohen–Gluck [1]. In our treatment we follow closely their argument.
100 3 Direct Sums of Cyclic Groups

As a first step, we reduce the proof of the theorem to the torsion case. This is one
of the rare situations when the discussion for torsion groups cannot be delegated to
p-groups.
Lemma 6.1 (Cohen–Gluck [1]). Let F be a free group and A D B ˚ C any group
with a free summand C. Given an epimorphism W F ! A, F admits a decomposition
F D F1 ˚ F2 such that .F1 / D B and F2 Š C.
Proof. Let W A ! C denote the projection along B. Then F D F1 ˚ F2 with
F1 D Ker  and F2 Š Im  D C. The inclusion B
F1 cannot be proper. u
t
Next, we reduce the proof to the countable case; this is a main step, supported
primarily by a straightforward back-and-forth argument.

Lemma 6.2 (Cohen–Gluck [1]). Suppose 0 ! H ! F !A ! 0 is an exact
sequence, and both F and A are direct sums of countable groups. Then there exist
.‘matching’/ direct decompositions

F D ˚ < F and A D ˚ < A (3.14)

for some ordinal  such that, for each  < ,

(i) F is countable; and
(ii) F D A :
Proof. There is nothing to prove if A is countable, so suppose A is uncountable.
Let F D ˚i2I Gi and A D ˚j2J Bj be decompositions with countable summands.
For any k 2 I, there is a countable subset Y0 of J such that Gk
˚j2Y0 Bj and a
countable subset X0 of I such that ˚j2Y0 Bj
.˚i2X0 Gi /: Arguing the same way
repeatedly, we obtain countable ascending chains of countable subsets X0     
Xn  : : : and Y0      Yn  : : : of I and J; respectively, satisfying

˚j2Yn Bj
.˚i2Xn Gi /
˚j2YnC1 Bj .n < !/:

If I0 and J0 denote the unions of the Xn and the Yn , respectively, then let F0 D
˚i2I0 Gi and A0 D ˚j2J0 Bj . They are clearly countably generated summands of F
and A, respectively, such that F0 D A0 .
Assume that we have already found, for some ordinal , smooth chains of subsets
I0      I      I and J0      J      J .
/ of I and J,
respectively, such that for all C1
, the sets IC1 n I and JC1 nJ are countable,
and the groups F0 D ˚i2IC1 nI Gi ; A D ˚j2JC1 nJ Bj satisfy .˚< F0 / D
˚< A . Using a back-and-forth argument, we adjoin to I and J countable subsets
U and V, respectively, such that putting I C1 D I [ U and J C1 D J [ V,
condition (ii) will be satisfied for F0 C1 D ˚i2U Gi ; A C1 D ˚j2V Bj : We repeat
this argument transfinitely until the index sets I and J are exhausted, where—as
usual—at limit ordinals we take unions of the previously selected subsets. Finally,
we get decompositions satisfying .˚< F0 / D ˚< A for all  < .
6 Equivalent Presentations 101

These are not yet decompositions we are looking for, we still have to modify
them to obtain ones satisfying (i)–(ii). Suppose that we have found F to satisfy
F D A for all  < . Consider the diagram

φ
Fσ −−−−→ ⊕ρ≤σ Aρ
⏐
⏐
ψ δ
φ
⊕ρ<σ Fρ −−−−→ ⊕ρ<σ Aρ

where ı denotes the projection with kernel A . Since the map in the bottom row is
surjective and F0 is a free group, we can find a map making the diagram commute.
Clearly, F D fx  x j x 2 F0 g is isomorphic to F0 . Furthermore, .x  x/ D
x   x D x  ıx 2 A shows that F
A . This inclusion cannot be proper,
thus F D A . As ˚< F ˚ F0 D ˚ F , we may replace F0 by F for each
 <  inductively, to obtain F D A for all  < . t
u
The Torsion Case We are now prepared to tackle the torsion case. The starting
point is a preliminary lemma (valid for arbitrary groups).
Lemma 6.3. Let F D F1 ˚ F2 be a free group, and  W F ! A D A1 ˚ A2 an
epimorphism such that A1
F1 . Then in the given direct decomposition, F2 can
be replaced by some G
F satisfying G
A2 .
Moreover, if F 0 is a summand of F2 with F 0
A2 , then G can be chosen so as
to contain F 0 .
Proof. Let W A ! A1 denote the projection with kernel A2 . The projectivity of F2
guarantees the existence of  making the square
ρ
F2 F1
⏐ ⏐
⏐
φ
⏐πφ

π
A1 ⊕ A2 −−−−→ A1

commutative. Setting G D .1  /F2 , evidently G D .  /F2 D 0. We

conclude that F1 ˚F2 D F1 ˚G, establishing the first claim. For the rest, it is enough
to observe that the map  can be chosen so as to act trivially on the summand F 0 . u
t
The following lemma is a crucial ingredient in the proof of Theorem 6.5 to
guarantee that no generator of F is left out in the successive decompositions.
Theorem 6.2 permits us to confine ourselves to countable groups.
102 3 Direct Sums of Cyclic Groups


Lemma 6.4. Let 0 ! H ! F !A ! 0 be a presentation of a †-cyclic torsion
group A, F a countable free group. If A0 is a finitely generated summand of A, then
there are direct decompositions

F D F1 ˚ F2 and H D .H \ F1 / ˚ .H \ F2 /

such that
(a) F1 is finitely generated and F1 contains A0 ; and
(b) A D F1 ˚ F2 .
Proof. Write A D A0 ˚ A00 where A00 is the complement of A0 in a direct
decomposition of A into cyclic groups of prime power orders. Apply Lemma 6.3
to this decomposition to get F D F0 ˚ G2 with G2
A00 .
Choose a summand A2
G2 of A00 , say A00 D A1 ˚A2 with finitely generated A1 .
Again by Lemma 6.3, we argue that F0
A0 ˚ A1 may be assumed. In this way,
we obtain a decomposition A D A0 ˚A1 ˚A2 ; where A0
F0 and A2
G2 . That
A1 D .A1 \ F0 / C .A1 \ G2 / should be clear. Assuming that the cyclic summands
in A are decomposed into their p-components, for any p, either F0 or G2 contains
an element of A1 of maximal p-power order; this generates a summand C of A1 . If
C is contained in F0 , then write A1 D C ˚ B1 , and with the aid of Lemma 6.3 we
can change G2 to a summand G such that G has trivial projection on A0 ˚ C D B0 ,
and at the same time replace A0 by B0 , and A1 by B1 to obtain A D B0 ˚ B1 ˚ A2 .
We continue in a similar fashion, next adjoining a summand of B1 to A2 , etc.
After a finite number of steps, we arrive at a decomposition F D F1 ˚ F2 , satisfying
(i) and (ii). t
u
The Stacked Basis Theorem Equipped with these lemmas, we are well pre-
pared for the proof of the main result. We keep the same notation.
Theorem 6.5 (Cohen–Gluck [1]). Every presentation of a †-cyclic group has
stacked bases.
Proof. In view of Lemma 6.1 and 6.2, the proof can be reduced to the case, in
which A is a countable †-cyclic torsion group. Then F can also be assumed to
be countable, say F D ˚i2N hxi i. We will be done if we reduce the problem to
the finitely generated case, because then a simple reference to Theorem 2.6 will
complete the proof.
By the preceding lemma, there is a decomposition F D F11 ˚ F12 such that F11
is finitely generated, x0 2 F11 , and H D .H \ F11 / ˚ .H \ F12 /. Next, F admits
a decomposition F D F21 ˚ F22 where F21 is finitely generated, contains F11 and
x1 , and H splits accordingly. Continuing in the same way, we obtain an ascending
chain F11
F21
: : : of summands of F, for which H \ Fn1 is a summand
of H. The union of the Fn1 must be all of F. If we define An .n < !/ via A0 D 0,
Fn1 D Fn1;1 ˚ An , and let Bn D H \ An , then F D ˚n<! An and H D ˚n<! Bn are
decompositions into finitely generated summands such that An and Bn are stacked.
The reduction to the finitely generated case has been accomplished, and the proof is
completed. t
u
6 Equivalent Presentations 103

Equivalent Presentations of Torsion-Free Groups The last theorem asserts

that every presentation of a †-cyclic group is equivalent to one with stacked bases
in the sense made precise by the following definition.
Let F, F 0 be free groups, H, H 0 subgroups such that F=H Š F 0 =H 0 . We say
that F=H and F 0 =H 0 are equivalent presentations of A Š F=H if there is an
isomorphism  W F ! F 0 carrying H onto H 0 .
In general, not much can be said about the situations when two presentations of
a group have to be equivalent. However, the case of torsion-free groups provides an
interesting, though not so easy positive example.
First of all, note that an obvious necessary condition for the equivalence of two
presentations F=H and F 0 =H 0 of A is that rk F D rk F 0 and rk H D rk H 0 : Our next
purpose will be to show that, if A is torsion-free, then the sole condition rk H D rk H 0
will be enough to ensure the equivalence of the presentations F=H and F 0 =H 0 .
We require an interesting preliminary lemma.
Lemma 6.6 (Erdős [1]). Let F be a free group, and H a pure subgroup of F. F has
a basis which is a complete set of representatives mod H if and only if jF=Hj D rk H.
Proof. If F has such a basis B D fb˛ j ˛ < g with (only) b0 contained in H, then
by the purity of H, B must be infinite, and obviously jF=Hj D jBj D rk F. For each
b˛ 2 B there is a unique bˇ 2 B such that b˛ C bˇ 2 H: If b˛ D bˇ ; then 2b˛ 2 H;
so b˛ D b0 ; by purity. The elements b˛ C bˇ .b˛ ¤ bˇ / and b0 form a basis of a
summand of F contained in H. Hence jBj
rk H, and necessity is established.
Turning to the proof of the sufficiency, suppose jF=Hj D rk H. From Lemma 5.9
we derive that H contains a summand G of F such that rk G D jF=Hj. Choose a basis
Y D fyj g of G, and extend it to a basis B D fb˛ j ˛ < g of F. Well-order B in such a
way that the elements of Y precede the rest of the basis elements in B. Each element
h 2 H can be written uniquely as a linear combination h D t1 b˛1 C  Cts b˛s .ti 2 Z/
with non-zero terms such that ˛1 <    < ˛s . To simplify our wording, we will say
that the ordinal ˛s is associated with h. If among the elements h 2 H associated
with the same ˛s there is one, say h0 , with jts j D 1, then in the basis B the element
b˛s can be replaced by h0 , without violating the basis character of the set. In doing
so for all possible ordinals ˛s inductively, Y remains unchanged, and the new basis
(which we continue denoting by B) will have the additional property that if h 2 H is
associated with ˛s , and in the expression for h the coefficient of b˛s is 1 in absolute
value, then necessarily b˛s D ˙h 2 H.
Split the basis B into two disjoint subsets, B D B0 [B00 , such that Y
B0 D B\H.
We keep B00 , but change B0 in order to obtain a basis B? of F which is a complete
set of representatives mod H, as desired.
First, observe that different elements b˛ and bˇ of B00 must belong to different
cosets mod H. Indeed, otherwise h D b˛  bˇ 2 H is associated with either b˛ or
bˇ , so either b˛ 2 H or bˇ 2 H, which is impossible, B00 being disjoint from H. Of
course, there are cosets mod H which do not intersect B00 . Since B0  H implies that
each coset mod H is represented by an element of the subgroup hB00 i, for each coset
104 3 Direct Sums of Cyclic Groups

mod H disjoint from B00 we can choose a representative xi 2 hB00 i. Thus B00 [X (with
X D fxi j i 2 Ig for some index set I) is a complete set of representatives mod H.
Next we show that jXj D jB0 j. On one hand, rk H D jF=Hj D jYj
jB0 j
rk F
implies jB0 j D rk F D jF=Hj. On the other hand, let b be the first element of B00 in
the chosen well-ordering of B. No two of the elements of the form b˛ b .b˛ 2 B00 /
belong to the same coset mod H, and none of these is congruent mod H to a bˇ 2 B00
(again, otherwise b˛  b  bˇ 2 H would be associated with either b˛ or bˇ , etc.).
Thus there are at least jB00 j many cosets of H which do not intersect B00 ; hence,
jB00 j
jXj follows. This together with jB00 j C jXj D rk F yields jXj D rk F: Hence
jB0 j D jXj, so there is a bijection between the set of elements fbi g of B0 and the set
of cosets fxi C Hg (where we have the corresponding elements carrying the same
index i). If in the basis B, bi 2 B0 will be replaced by bi C xi , then we obtain a new
basis B? of F which is at the same time a complete set of representatives mod H. u t
We are now able to verify the main result mentioned earlier.
Theorem 6.7 (Erdős [1]). Two presentations, F=H and F 0 =H 0 , of a torsion-free
group are equivalent if and only if rk H D rk H 0 :
Proof. To verify sufficiency, suppose rk H D rk H 0 ; as noted above, this implies
rk F D rk F 0 . We prove more than stated, viz. we show that every isomorphism
W F=H ! F 0 =H 0 is induced by an isomorphism  W F ! F 0 carrying H onto H 0 .
Since A is torsion-free, both H and H 0 are pure. Ignoring the trivial case, we may
suppose that rk H is infinite. We distinguish three cases.
Case I: rk H D jAj. Then the same is true for rk H 0 . In view of Lemma 6.6,
there exist a basis B of F and a basis B0 of F 0 which are complete sets of
representatives mod H and mod H 0 , respectively. The correspondence B ! B0
which maps b 2 B upon b0 2 B0 if and only if maps the coset b C H upon
b0 C H 0 extends uniquely to an isomorphism  W F ! F 0 under which H 0 is the
image of H. Thus the two presentations are equivalent.
Case II: rk H > jAj. Let G be a free group whose rank is rk H. Replace F by F ˚ G
and F 0 by F 0 ˚ G, but keep H and H 0 . Application of Case I to A ˚ G implies
the existence of an isomorphism  W F ˚ G ! F 0 ˚ G with H D H 0 inducing
. It is self-evident that F D F 0 .
Case III: rk H < jAj. There is a decomposition F D F1 ˚ F2 such that H
F1
and rk H D rk F1 < rk F2 D jAj. Thus A D F1 =H ˚ F2 , and yields a
similar decomposition A0 D F10 =H 0 ˚ F20 . Case I guarantees the existence of
an isomorphism F1 ! F10 mapping H upon H 0 ; this along with F2 ! F20
(restriction of ) yields an isomorphism  W F ! F 0 . t
u
F Notes. Hill–Megibben [4] furnished another proof of Theorem 6.5 as a corollary to a more
general result which they proved on equivalent presentations of arbitrary abelian groups. F=H and
F0 =H 0 are equivalent presentations if and only if, for each prime p,

dim.H C pF/=pF D dim.H 0 C pF0 /=pF0 :

7 Chains of Free Groups 105

For Dedekind domain, A.I. Generalov and M.V. Zheludev [St. Petersburg Math. J. 7, 619–661
(1996)] characterized equivalent presentations. No such study is available for larger classes of
domains, but several special cases have been investigated.
Various generalizations of the stacked basis theorem may be found in the literature. Let us
mention Ould-Beddi–Strüngmann [1] where homogeneous completely decomposable groups are
considered. Osofsky [1] studied a kind of dual to the stacked basis theorem. She proved that if
H is a subgroup of a free group F such that F=H is pn -bounded, then for every decomposition
F=H D ˚ Ci with cyclic groups Ci there is a decomposition F D ˚ Fi such that Ci D Fi =.H\Fi /.
Cutler–Irwin–Pfaendtner–Snabb [1] have a nice generalization of Lemma 6.6. They show that
a pure subgroup H in a †-cyclic group G contains a summand K of G such that rk0 .K/ D rk0 .H/
and rkp .K/ D rkp .H/ for each p. See Lemma 6.12 in Chapter 5, for the torsion case.

Exercises

(1) (Erdős) Let H be a subgroup of a group G such that G=H is torsion-free. There
is a generating system of G which is a complete set of representatives mod H if
and only if jHj
jG=Hj. [Hint: Lemma 6.2 with a presentation of G.]
(2) (Hill–Megibben) If A D F=H is a presentation of an infinite group such that F
is free and rk F > jAj, then there is a direct decomposition F D F1 ˚ F2 such
that rk F1 D jAj and F2
H.
(3) Let H0 <    < Hn < : : : be a countable ascending chain of summands of a
free group F.
(a) The union H D [n<! Hn need not be a summand
P of F.
(b) H contains a summand of F whose rank is n<! rk.Hn /. [Hint: H is pure
in F, and apply Exercise 1.]
(4) (Erdős) Let A D ˚i2I Ai be a direct sum of torsion-free groups. If F is a free
group and  W F ! A is an epimorphism, then there is a decomposition F D
˚i2IPFi such that  W Fi ! Ai for each i 2 I. [Hint: represent Ai D Fi0 =Hi0 such
that i2I rk.Hi0 /
rk.Ker /, and apply Lemma 6.6.]

7 Chains of Free Groups

We are looking for criteria for a group to be free, especially when the union of a
chain of free subgroups is again free. In this section and in the next one, we have to
use frequently purity to be discussed in Chapter 5.
Pontryagin’s Criterion In a few cases useful criteria for freeness can be
established. The one which is most often used works for countable torsion-free
groups.
106 3 Direct Sums of Cyclic Groups

Theorem 7.1 (Pontryagin [1]). A countable torsion-free group is free if and only if
each of its finite rank subgroups is free. Equivalently, for every n 2 N, the subgroups
of rank
n satisfy the maximum condition.
Proof. Because of Theorem 1.6, necessity is evident. For sufficiency, let A D
ha0 ; : : : ; an ; : : : i be a countable torsion-free group all of whose subgroups of finite
rank are free. Define A0 D 0; An D ha0 ; : : : ; an1 i .n 2 N/ (the purification of
ha0 ; : : : ; an1 i in A). Then rk An
n and rk AnC1
rk An C 1. Therefore, either A
is of finite rank—in which case there is nothing to prove—or there is a subsequence
Bn of the An , such that rk Bn D n, and A is the union of the strictly ascending chain
0 D B0 < B1 <    < Bn < : : : . Now BnC1 =Bn is torsion-free of rank 1 and finitely
generated, thus BnC1 =Bn Š Z. From Theorem 1.5 we obtain BnC1 D Bn ˚ hbn i for
some bn 2 A. This shows that the elements b0 ; b1 ; : : : ; bn ; : : : generate the direct
sum ˚n<! hbn i, whence A D ˚n<! hbn i is immediate.
By Theorem 1.6, the second formulation is equivalent to the first one. t
u
Corollary 7.2. Suppose 0 D G0 < G1 <    < Gn < : : : is a chain of countable
free groups such that each Gn is pure in the union G of the chain. Then G is free.
Proof. A finite rank subgroup of G is contained in some Gn , so it is free. The claim
is immediate from Theorem 7.1. t
u
If we have a chain like in Corollary 7.2 with the Gn as summands in a larger
group F, the union G need not be a summand of F.
Example 7.3. Let G be a free group that is the union of a countable chain of infinite rank
summands G0 < G1 <    < Gi < : : : . Our claim is that there exists a countable free group
F containing G such that each Gi is, but G is not a summand of F.
Let 0 ! H ! F0 ! Q ! 0 be a presentation of Q with countable free F0 . Let Hn .n < !/
be a chain of finite rank summands of the free group H with union H. Then F0 =Hn is free for all
n < !. Next, pick free groups F0 Š G0 and Fi Š Gi =Gi1 .i
1/. It is evident that

G Š H ˚ ˚i<! Fi and Gn Š Hn ˚ ˚ni<! Fi .n < !/:

Finally, we embed G in a free group F Š F0 ˚ ˚i<! Fi imitating the embedding of H in F0 and

keeping the Gi fixed. This F is as desired.

The Eklof–Shelah Criterion The following lemmas provide us with versatile

criteria for a group to be free.
Lemma 7.4. Let, for some ordinal ,

0 D A0 < A1 <    < A < : : : . < / (3.15)

S
be a smooth chain of pure subgroups of a group A such that A D  < A . If, for
each  < , the factor group A C1 =A is free, then A is free.
Proof. In view of the stated condition, A C1 D A ˚ B for each  < ; for a
suitable subgroup
S B of A C1 Theorem 1.5. If X denotes a basis of B , then the set
union X D  < X is a basis for A. t
u
7 Chains of Free Groups 107

We can now verify the Eklof–Shelah criterion which provides a necessary and
sufficient condition for freeness.
Theorem 7.5 (Eklof–Shelah). Let  be an uncountable regular cardinal, and
assume 0 D A0 < A1 <    < A < : : : . < / is a smooth chain of pure
subgroups of a group A such that

SA are free groups of cardinality < , and

(i) all the
(ii) A D  < A .
Then A is free if and only if the set

E D f <  j 9  >  such that A =A is not freeg

is not stationary in .
Proof. Suppose A is free. Consider a filtration fB g < of A with summands. The
set C of indices  of those subgroups A which appear in the filtration fB g < is
a cub in , so fA g 2C provides a filtration of A with summands. We see that A=A
is free for all  2 C, and so C does not intersect the set E. This proves that E is not
stationary.
Conversely, assume that E is not stationary. Then there is a cub C   which
does not intersect E. Evidently, fA g 2C is still a filtration of A. Relabeling, we have
a filtration fA g < where all factor groups A C1 =A are free. By Lemma 7.4, A is
free. t
u
Remark. For future applications we point out that both Lemma 7.4 and Theo-
rem 7.5 hold for p-groups A if ‘free’ is replaced throughout by ‘†-cyclic.’ The
proofs are the same with obvious changes.
The next lemma teaches us how to create from a short chain of direct sums with
large factor groups a long chain with small factor groups. (The main interest is in
the torsion-free case, but no such restriction is needed.)
Lemma 7.6. Assume

0 D G0 < G1 <    < Gn < : : :

is a chain of groups that are pure in the union G D [n<! Gn , where each Gn is a
direct sum of countable groups. Then there is a smooth chain

0 D A0 < A1 <    < A < : : : . < / (3.16)

of pure subgroups A of G such that

(i) A \ Gn is a summand of Gn , for every n < ! and  < ; and
(ii) A C1 =A . C 1 < / is countable and the union of an ascending chain of pure
subgroups, isomorphic to summands of Gn .n < !/.
108 3 Direct Sums of Cyclic Groups

Proof. We start choosing a fixed direct decomposition of each Gn into countable

summands, and define an H.@0 /-family Hn of summands of Gn to consist of all
direct sums of subsets of components in the chosen decomposition. We break the
proof into three steps.
Step 1. The collection

Gn D fA 2 Hn j A C Gi is pure in Gn for each i < ng

is a G.@0 /-family of subgroups of Gn .

All that we have to check is that Gn satisfies the countability condition for a
G.@0 /-family, since the other conditions are obvious. Let A 2 Gn , and H0 a
countable subgroup of Gn . Suppose that we already have a chain A D B0 <
B1 <    < Bm of subgroups in Hn such that
1. A C H0
B1 ;
2. BjC1 =Bj is countable for all j < m; and in addition,
3. for each j < m and for each i < n; .BjC1 C Gi /=.A C Gi / contains a
purification of .Bj C Gi /=.A C Gi / in Gn =.A C Gi /.
To find a next member BmC1 of the chain, for each i < n, let Vi  Gn
be a countable set that—along with .Bm C GSi /=.A C Gi /—generates a pure
subgroup in Gn =.A C Gi /. Thus HmC1 D i<n Vi is likewise a countable
set. Consequently, there is a BmC1 2 Hn such that Bm C HmC1  BmC1 and
BmC1 =Bm is countable. Then, for each i < n; .BmC1 C Gi /=.A C Gi / contains
the purification of .Bm C Gi /=.A C Gi / in Gn =.A C Gi /: The union B of the chain
of the Bm for all m < ! is a member of Hn ; B=A is evidently countable, and our
construction guarantees that .B C Gi /=.A C Gi / is pure in Gn =.A C Gi /. Thus
B C Gi is pure in Gn , i.e. B 2 Gn .
Step 2. The family

B D fA
G j A \ Gn 2 Gn for each n < !g

is a G.@0 /-family of subgroups in G.

Again, only the countability condition requires a proof. Since there are but
countably many indices n to deal with, a similar back-and-forth argument .!
times) suffices to ensure that for each A 2 B, there exists an A0 2 B such that
A0 =A is countable, as needed.
Step 3. At this point we know that B is a G.@0 /-family satisfying (i), so we can
extract a smooth chain (3.16) of pure subgroups with countable factor groups.
Evidently, the group A C1 =A is the union of the ascending chain of groups
ŒA C .Gn \ A C1 /
=A Š .Gn \ A C1 /=.Gn \ A / .n < !/, all summands of
Gn in the chosen direct decomposition. t
u
7 Chains of Free Groups 109

Hill’s Criterion The following result is a far-reaching generalization of Pon-

tryagin’s theorem.
Theorem 7.7 (Hill [9]). The union G of a countable ascending chain

0 D G0 < G1 <    < Gn < : : :

of pure subgroups, each of which is free, is a free group.

Proof. The given chain can be replaced by a chain of the A as stated in Lemma 7.6.
Apply Corollary 7.2 to the factor groups A C1 =A to conclude that they are free.
A simple reference to Theorem 7.5 completes the proof. t
u
It should be pointed out that this theorem fails to hold for longer chains, as is
shown by Theorem 8.6 below.
Before we go on, we would like to mention an important consequence of Hill’s
theorem. This is a special case of Shelah’s compactness theorem 9.2 for limit
ordinals cofinal with !.
Corollary 7.8 (Hill [21]). Suppose  is an infinite cardinal whose cofinality is !.
A group of cardinality  is free provided that all of its subgroups of cardinalities
<  are free.
Proof. Evidently, a group of cardinality  is the union of a countable ascending
chain of pure subgroups whose cardinalities are < . By hypothesis, each of these
is free, so the claim follows right away from Theorem 7.7. t
u
Another criterion worthwhile recording is the following. (Observe the difference
between Theorem 7.5 and Lemma 7.9.)
Lemma 7.9 (Eklof [5]). Let  be a regular cardinal, and

0 D A0 < A1 <    < A <    < A D A (3.17)

a smooth chain of free groups such that A =A is free whenever  is a successor
ordinal and  <  < . If the set

E D f <  j  limit ordinal; AC1 =A not freeg

is not stationary in , then A is a free group, and so is A=A for every successor
ordinal .
Proof. Suppose E is not stationary, i.e. there is a cub C   that does not
intersect E. Those A whose indices belong to C form a chain like (3.17); we may
assume that (3.17) is this subchain. In this chain, A C1 =A is free for all  < ,
so Theorem 7.5 implies that A is free. The second claim follows by applying the
result to A=A . t
u
110 3 Direct Sums of Cyclic Groups

When Torsion-Free has to Be Free A rather remarkable feature of free groups

was discovered by Griffith [5]. A torsion-free group containing a free subgroup with
bounded factor group is easily seen to be again free, and interestingly, the same
conclusion can be reached under much weaker conditions on the factor group. The
following theorem is a slightly modified version of Griffith’s theorem.
Theorem 7.10. Let A be a torsion-free group, and F a free subgroup of A. If A=F is
a p-group that admits an H.@0 /-family of subgroups such that all the factor groups
in H.@0 / are reduced, then A is free .and Š F/.
Proof. If A is of finite rank, then hypothesis implies A=F is a reduced p-group of
finite rank, so it is finite (cp. Theorem 5.3). As a finitely generated torsion-free
group, A is free.
Next assume F is of countable rank, and write F D ˚n<! Zn with Zn Š Z. For
n < !, set Fn D ˚in Zi and An D hFn i . Manifestly, An =Fn D An =.An \ F/ Š
.An CF/=F
A=F, which shows that An =Fn is reduced, and hence it must be a finite
p-group. Therefore, An is free of finite rank by the preceding paragraph. Hence A
is the union of a countable ascending chain fAn gn<! of pure free subgroups, and
by Corollary 7.2 we conclude that A itself is free.
Turning to the uncountable case, set F D ˚i2I Zi with Zi Š Z, and for the
p-group T D A=F, select an H.@0 /-family H as stated. As T is reduced, and A
is torsion-free, every non-zero element of A is divisible but by a finite number of
integers. We are going to define by transfinite induction a smooth chain ¿ D I0 
I1      I      I D I of subsets of I such that for all  < , we have
(a) jI C1 n I j
@0 , and
(b) .A C F/=F (which is Š A =F ) is a subgroup T 2 H, where F D ˚i2I Zi
and A D hF i .
Suppose  is an ordinal such that, for all  < , the subsets I have been selected as
required. If  is a limit ordinal, then we set I D [< I , as is forced by continuity.
In this case, (b) will be satisfied, since H is closed under unions. If  D  C 1 and
I ¤ I, pick any i 2 I n I . Note that .A C F/=F D T is a subgroup of countable
index in C1 =F where C1 D hA C Zi i C F, so there is a subgroup B1
A such that
C1 =F
B1 =F 2 H and jB1 =C1 j
@0 . There is a countable subset J1  I n I for
which hA ˚ ˚j2J1 Zj i C F contains B1 . We keep repeating this process, to define
an ascending chain of countable subsets Jn of I n I , along with subgroups Cn and
Bn of A (for n < !) such that C1
B1
C2
B2
: : : . If we set

I D I [ [n<! Jn ; F D ˚j2I Zj ; and A D hF i ;

then [n<! Cn =F D [n<! Bn =F will be a subgroup T 2 H, and (a)-(b) will be

satisfied for this . The factor group A =A is torsion-free and countable; it contains
.A C F /=A as a free subgroup such that the factor group is isomorphic to

A =.A C F / D A =ŒA \ .A C F/

Š .A C F/=.A C F/ Š T =T
7 Chains of Free Groups 111

which is a countable reduced p-group. Therefore, we can apply the countable case
to derive that A =A is a free group. Hence the chain of the A . < / has free
factor groups, thus their union A is a free group. t
u
Example 7.11. Reduced totally projective p-groups (Sect. 6 in Chapter 11) admit an H.@0 /-family
as stated in the theorem. However, no uncountable p-group with countable basic subgroups has
such a family.

The Summand Intersection Property We say that a group A has the summand
intersection property if the intersection of two summands in A is likewise a
summand of A. If the same holds for infinite intersections as well, then we refer
to it as the strong summand intersection property. Needless to say, this property
is shared by very special groups only. We are looking for free groups with this
property.
Proposition 7.12 (Kaplansky [K], Wilson [1]). All free groups have the summand
intersection property. A free group has the strong summand intersection property if
and only if it is countable.
Proof. Let F be a free group, and F D Bi ˚ Ci .i D 1; 2/ direct decompositions.
Then F=.B1 \ B2 / is isomorphic to a subgroup of the free group F=B1 ˚ F=B2 , so
is itself free. Hence B1 \ B2 is a summand of F.
If F is a countable free group, then the same argument with countable summands
leads to a countable factor group contained in a product of countable free groups
F=Bi . Theorem 8.2 below implies that this factor group is free, so the intersection
of countably many summands is a summand.
Finally, suppose jFj  @1 . Let A be a torsion-free, non-free group of cardinality
@1 , and  W F ! A a homomorphism that is a bijection between a basis fb j  < g
of F and the elements of A. Select homomorphisms i W F ! A with cyclic images
whose kernels Ci are summands of F containing K D Ker . This can be done such
that the intersection \i2I Ci D K. But K cannot be a summand, since F=K is not
free. t
u
F Notes. In this section, we have collected the most useful results on free groups. They have
fascinating features, no wonder that their theory attracted so many researchers. For criteria on the
existence of a basis, we refer to Kertész [1], Fuchs [1]. In view of the very useful chain criteria of
freeness, basis criteria are hardly used.
The summand intersection property for free groups was observed by Kaplansky [K]. More
on this property can be found in Wilson [1], Arnold–Hausen [1], Albrecht–Hausen [1]. Hausen
[9] proved that A.I/ has the summand intersection property if End A is a PID. This property was
investigated by Kamalov [1] for non-free groups, and by Chekhlov [2] for torsion groups.

Exercises

(1) A countable group is †-cyclic if and only if every finite set of its elements is
contained in a finitely generated direct summand.
112 3 Direct Sums of Cyclic Groups

(2) (a) A subset fai gi2I of a torsion-free group A is a basis of A if and only if it is a
minimal generating system such that, for every finite subset fa1 ; : : : ; an g, if
a 2 A depends on fa1 ; : : : ; an g, then a 2 ha1 ; : : : ; an i.
(b) The same with “minimal generating system” replaced by “maximal inde-
pendent subset.”
(3) In any presentation of Z@0 ; there are continuously many generators and
continuously many relations.
(4) (Danchev) In a p-group A, the p! A-high subgroups are †-cyclic if and only if
AŒp
is the union of an ascending chain Tn .n < !/ such that the finite heights
in Tn are bounded.
(5) The summand intersection property is inherited by summands.
(6) (Wilson) A torsion group has the summand intersection property if and only if
each of its p-components is either cocyclic or elementary.
(7) (Wilson, Hausen) A has the summand intersection property if and only if for
every direct decomposition A D B ˚ C, the kernel of any map B ! C is a
summand of A.

8 Almost Free Groups

Almost free groups are those (necessarily) torsion-free groups in which all
subgroups of smaller sizes are free. More precisely, for an infinite cardinal , we say
that a group A is -free if every subgroup of A whose rank is <  is free [AG]. The
problem of finding the cardinals  for which there exist -free groups that fail to be
 C -free was raised in [IAG]. As it turns out, it is an intricate problem, requiring
sophisticated machinery from set theory. It has been studied extensively, and a
significant amount of information has already been gained, but still much remains
to be done. Here we aim simply at giving a taste of the subject. The objective is to
understand how close almost free groups are to being free.
-Free Groups Since purification does not increase rank, it is clear that if A is
-free, then every subgroup of rank <  is contained in a pure free subgroup of the
same rank. Thus the collection C of pure free subgroups of rank <  is witness for
-freeness. In view of Theorem 7.7, C is closed under taking unions of countable
chains.
Example 8.1. In this new terminology, Pontryagin’s theorem 7.1 can be rephrased by saying that
a countable @0 -free group is free.

(A) If  <  are infinite cardinals, then -free implies -free. In particular, free
groups are trivially -free for every cardinal .
(B) Subgroups and direct sums of -free groups are -free.
(C) Extension of a -free group by a -free group is -free. More generally, we
have:
8 Almost Free Groups 113

(D) Let 0 D A0 <    < A <    < A D A be a smooth chain of groups such
that all the factor groups A C1 =A are -free. Then A is also -free. In fact,
let X be a pure subgroup of A with jXj < . Then in the smooth chain X \ A
. < / each factor group .X \ A C1 /=.X \ A / is torsion-free of cardinality
< , and therefore it is isomorphic to the free subgroup .A C.X\A C1 //=A

A C1 =A . An appeal to Lemma 7.4 completes the proof.

The Baer–Specker Group The next theorem is concerned with a prototype for
@1 -freeness; actually, the group is of major interest.
Theorem 8.2 (Baer [6], Specker [1]). The direct product of infinitely many infinite
cyclic groups is @1 -free, but not free.
Q
Proof. Write A D i2I hai i, where I is an infinite set, and hai i Š Z for each i.
The first step in the proof is to show that every finite subset fx1 ; : : : ; xm g  A is
contained in a finitely generated direct summand of A whose complement is a direct
product of infinite cyclic groups.
We induct on m. If m D 1 and x1 ¤ 0, then x1 D .: : : ; ni ai ; : : : / with ni 2 Z. If
there is an index j 2 I such that jnj j D 1, then the jth component haj i in the direct
product can be replaced by hx1 i, i.e. A D hx1 i ˚ Aj , where Aj is the set of elements
with vanishing jth coordinate, so it is also a product of infinite cyclic groups. If the
minimum n of the jni j with ni ¤ 0 is greater than 1, then setting ni D qi n C ri with
qi ; ri 2 Z; 0
ri < n, define y1 D .: : : ; qi ai ; : : : /; y2 D .: : : ; ri ai ; : : : / 2 A so
that x1 D ny1 C y2 . There must be an index j 2 I with jqj j D 1 and rj D 0, thus
A D hy1 i ˚ Aj , where y2 2 Aj with coefficients 0
ri < n. By induction on n,
Aj has a finitely generated summand B0 containing y2 , and so hy1 i ˚ B0 is a finitely
generated summand of A containing x1 such that it has a direct product of infinite
cyclic groups as a complement.
Assume that m > 1, and A D B ˚ C where B is finitely generated containing
fx1 ; : : : ; xm1 g, and C is a direct product of copies of Z. Setting xm D b C c
.b 2 B; c 2 C/ and embedding c in a finitely generated summand C0 of C, we
obtain a finitely generated summand B ˚ C0 of A, containing fx1 ; : : : ; xm g, again
with a complement that is a direct product of infinite cyclic groups.
The next step is to show that A is @1 -free. Let G be a countable subgroup of
A. A maximal independent set of a finite rank subgroup G0 of G is contained in a
finitely generated summand B of A, so by torsion-freeness, G0
B. Thus G0 is free,
and Theorem 7.1 implies that G is free.
It remains to prove that A itself is not free. We Q exhibit a non-free subgroup of A.
Let p be any prime, and H the subgroup of A0 D i<! hai i (a summand of A) that
consists of all vectors b D .n0 a0 ; n1 a1 ; : : : ; ni ai ; : : : / such that, for every integer
k > 0, almost all coefficients ni are divisible by pk . Manifestly, H contains the direct
sum S D ˚i<! hai i, and has cardinality 2@0 . Since each coset of H mod pH can be
represented by some element of S, H=pH cannot be uncountable. If H were free, we
would have jH=pHj D jHj; so H cannot be free. t
u
The countable direct product of infinite cyclic groups, i.e. the group Z@0 , is often
called the Baer-Specker group.
114 3 Direct Sums of Cyclic Groups

An immediate consequence is the following result that shows that in countable

groups free summands can be collected into a single summand.
Corollary 8.3 (K. Stein). A countable torsion-free group A can be decomposed as
A D F ˚ N where F is a free group, and N has no free factor group. N is uniquely
determined by A.
Proof. Define N as the intersection of the kernels of all homomorphismsQ  W A ! Z.
Then A=N is isomorphic to a countable subgroup of the direct product  Z, so it
is free in view of Theorem 8.2. N is then a summand of A, and we have A D F ˚ N
with F free. From the definition of N it is evident that N cannot have a non-trivial
map into Z. t
u
The next two examples show that there exist very large @1 -free groups A such
that Hom.A; Z/ D 0, and it may also happen that an @1 -free group is isomorphic to
the countable direct sum and to the countable direct product of itself.
Example 8.4 (G. Reid [1]). Let 
@1 be a non-measurable cardinal, and N the subgroup of
Z consisting of vectors with countable support. Then A D Z =N is @1 -free. To see this, let
bn C N .n < !/ be a list of elements in a countable subgroup F of A. Clearly, each bn has
uncountable support, and each sum bi C bj is equal to some bk modulo a countable index set. Thus
if we change the representatives bn by dropping all the indices in these countably many index sets,
then the new representatives form a subgroup F0 Š F. By Theorem 8.2, F0 is free, so A is @1 -free.
A homomorphism  W A ! Z may be viewed as a map   W Z ! Z such that   .N/ D 0.
By Theorem 2.8 in Chapter 13,   D 0, which means Hom.A; Z/ D 0.
Example 8.5. There exists an @1 -free group which is not free and isomorphic both to the direct
sum of countably many copies of itself, and to the direct product of countable many copies of itself.
See Proposition 4.9 in Chapter 13.

Strongly -Free The study of almost free groups brings a stronger version of
-freeness into the picture. Let  be a regular cardinal. A group A is said to be
strongly -free if every subgroup of cardinality <  is contained in a free subgroup
C of cardinality <  such that A=C is -free. Evidently, free groups are strongly
-free for any cardinal .
It is not obvious that if  <  are infinite cardinals, then strongly -free implies
strongly -free, but it is true. In fact, if a subgroup B of cardinality <  is contained
in a free subgroup C of cardinality <  with -free A=C, then B is contained in
a summand C0 of C of cardinality < . Because of (C), A=C0 is -free, being an
extension of the free group C=C0 by the -free group A=C.
The fine nuance between strongly -free and just plainly -free groups can be
better understood if we compare filtrations.
Lemma 8.6 (Eklof–Mekler [EM]). Let A be a group of cardinality , where  is
an uncountable regular cardinal.
(a) A is -free exactly if it has a filtration fA j  < g with free subgroups A of
cardinality < .
(b) A is strongly -free if and only if it admits a filtration fA j  < g with free
subgroups A of cardinality <  such that, for all  <  < , the factor groups
A C1 =A C1 are free.
8 Almost Free Groups 115

Proof. (a) Since every subgroup of cardinality <  is contained in some member
of a -filtration, the stated condition evidently implies the -freeness of A.
Conversely, if A is -free, then the subgroups in any -filtration of A are free.
(b) For sufficiency, it is enough to observe that the stated condition is equivalent to
that A is -free, and for every A C1 , the factor group A=A C1 is -free. To prove
the converse, assume A is strongly -free, and fa j  < g is a well-ordered
list of elements of A. We construct a filtration fA j  < g of A as desired, with
the additional property that a 2 A for all  <  < . Suppose that, for some
 < , we have a chain fA j 
g satisfying the requisite properties. Choose
for A C1 a subgroup of cardinality <  that contains both A and a such that
A=A C1 is -free. Then the factor groups A C1 =A C1 are free for all  > , and
by the -freeness of A, A C1 is free. t
u
Observe that in (b) we have not said anything about the freeness of the factor
groups A =A at limit ordinals .
Lemma 8.7. The Baer-Specker group P D Z@0 is not strongly @1 -free.
Proof. We prove that the direct sum S D Z.@0 / is not contained in any countable
subgroup G with @1 -free P=G. Anticipating theorems that we will prove later on,
the proof is quick. Corollary 1.12 in Chapter 6 asserts that P=S is algebraically
compact, thus for every intermediate pure subgroup S
G < P, the factor
group P=G is torsion-free and algebraically compact (see Lemma 8.1 in Chapter 9).
Therefore, it contains a subgroup isomorphic to either Q or Jp for some prime p,
and consequently, P=G can never be @1 -free. t
u
Uncountable Chains If we wish to consider chains of free groups of cofinality
exceeding !, then we are confronted with a more complicated situation. In order
to guarantee that the union of long chains of free groups will again be free, it is
necessary to impose restrictions on the factors in the chain. A typical result is as
follows.
Theorem 8.8 (Fuchs–Rangaswamy [4]). Suppose  is an uncountable regular
cardinal, and 0 D F0 < F1 <    < F < : : : . < / is a smooth chain of
groups such that, for every  < ,
(a) F is free of cardinality
, and
(b) F is a pure subgroup of F C1 .
(i) The union F of the chain is free provided the set

S D f <  j 9  >  such that F =F is not -freeg

is not stationary in .
(ii) If all jF j < , and S is stationary in , then F is -free, but not free.
Proof. The proof is similar to the one in Lemma 7.6, but the construction of the
G./-families becomes complicated at limit ordinals. The details are too long to be
reproduced here. t
u
116 3 Direct Sums of Cyclic Groups

Large Almost Free Non-free Groups We next prove that in the constructible
universe, there exist large -free groups that are not free.
Theorem 8.9 (Gregory [1]). Assume V D L. For every uncountable regular
cardinal  that is not weakly compact, there exists a -free group of cardinality
 which is not free.
Proof. Let E be a stationary subset of  that consists of limit ordinals cofinal with
!; see Lemma 4.5 in Chapter 1. Suppose for a moment that we have succeeded in
constructing a group F as the union of a smooth chain of subgroups F . < /
satisfying the following conditions for all  <  < :
(i) F is free of cardinality jj  @0 ;
(ii) if  2 E, then the quotient F C1 =F is not free;
(iii) if  … E, then F =F is free of cardinality jj  @0 .
Then F is of cardinality  and -free. Working toward contradiction, suppose F is
free. Then there exists a cub C   such that F =F is free for each pair  < 
in C. For such a pair of indices, the exact sequence 0 ! F C1 =F ! F =F !
F =F C1 ! 0 must split because of (iii). This means that F C1 =F must be free for
 2 E \ C, contrary to (ii).
It remains to construct a smooth chain of groups F . < / with the listed
properties. Starting with F0 free of rank @0 , we proceed to define F . > 0/ via
transfinite induction as follows. Assume that  is an ordinal <  such that the groups
F . < / have already been constructed, and they satisfy conditions (i)–(iii) up
to . To define F we distinguish three cases.
S
Case 1. If  is a limit ordinal, then we have no choice: F D < F . Since E\ is
not stationary in  Lemma 4.5 in Chapter 1, (iii) allows us to apply Theorem 7.5
to claim that F is free. Hence conditions (i)–(iii) hold for all ordinals
.
Case 2. If  D  C 1 and  … E, then we simply let F D F ˚ X where X is a
countable free group.
Case 3. The critical case is when  D  C 1 and  2 E. In view of the choice
of E, we have cf  D !, so  is the supremum of an increasing sequence of
non-limit ordinals 0 < 1 <    < n < : : : .n < !/: Consider the chain
F0 <    < Fn < : : : of free groups whose union is the free group F . We
are in the situation of Example 7.3, and so we can define F C1 such that the
Fn .n < !/ are, but F is not a summand of F C1 . With this choice, (i)–(iii)
will be satisfied by all ordinals
. t
u
For cardinals @n .n  1/, the existence of a stationary E of property Lemma 4.5
in Chapter 1 can be established without the hypothesis V = L, therefore we can state:
Corollary 8.10 (Eklof [2], Griffith [7], Hill [13]). For every integer n > 1, there
is a non-free @n -free group of cardinality @n . t
u
The †-Cyclic Case Several results proved above carry over to torsion and
mixed groups provided we can interpret freeness in an appropriate way. This can
8 Almost Free Groups 117

be done by introducing -cyclic groups meaning that every subgroup of cardinality

<  is †-cyclic.
A proof similar to Theorem 8.9 applies to verify:
Corollary 8.11 (Eklof [2]). If  is an uncountable regular cardinal that is not
weakly compact, then there exist -cyclic torsion groups of cardinality  that are
not †-cyclic.
For every n > 0, there are @n -cyclic torsion groups of cardinality @n which are
not †-cyclic.
Proof. Obvious modification to Theorem 8.9 is that ‘free’ should be replaced by
‘†-cyclic,’ ‘-free’ by ‘-cyclic,’ and purity should be assumed throughout. In place
of Example 7.3, a modified example should be referred to where a pure-projective
resolution of Z.p1 / is used. t
u
Mittag-Leffler Groups Most recently, a lot of attention has been devoted
to Mittag-Leffler modules. In the group case, a satisfactory characterization is
available. In the definition, we need tensor products: M is a Mittag-Leffler group
if for every collection fAi gi2I of groups, the natural map
Y Y
WM˝ Ai ! .M ˝ Ai /
i2I i2I

given by .x ˝ .: : : ; ai ; : : : // 7! .: : : ; x ˝ ai ; : : : / is monic .x 2 M; ai 2 Ai /.

Example 8.12. (a) Cyclic groups are Mittag-Leffler. This is trivial for Z, and follows for
M D Z.n/ from the fact that both the domain and the image of  are then isomorphic to
Q
i2I .Ai =nAi /.
The Prüfer groupQH!C1 (of length ! C 1) is not Mittag-Leffler. The natural map H!C1 ˝
(b) Q
n<! Z.p / ! n<! H!C1 ˝ Z.p / is not monic. (See the proof of Theorem 8.14.)
n n

Lemma 8.13 (M. Raynaud, L. Gruson). The class of Mittag-Leffler groups is

closed under taking pure subgroups, pure extensions and arbitrary direct products.
Proof. Starting with a pure-exact sequence 0 ! M 0 ! M ! M 00 ! 0, we form
the commutative diagram
  
0 −−−−−→ M  ⊗ Ai −−−−−→ M ⊗ Ai −−−−−→ M  ⊗ Ai −−−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐ 
φ  φ φ
  
0 −−−−−→ (M  ⊗ Ai ) −−−−−→ (M ⊗ Ai ) −−−−−→ (M  ⊗ Ai ) −−−−−→ 0

with pure-exact rows (see Corollary 3.7 in Chapter 5). Evidently, if  is monic, then
so is  0 . If both  0 and  00 are monic, then Lemma 2.6 in Chapter 1 (or a simple
diagram-chasing) shows that  has to be monic as well. Finally, for direct products,
the claim will be a simple consequence of Theorem 8.14 and Exercise 1. t
u
It is not difficult to characterize Mittag-Leffler groups.
118 3 Direct Sums of Cyclic Groups

Theorem 8.14 (Raynaud–Gruson). A group is Mittag-Leffler if and only if it is

@1 -cyclic.
Proof. We start the proof by showing that M is Mittag-Leffler if and only if each
of its countable pure subgroups is Mittag-Leffler. One direction the claim follows
from Lemma 8.13. P Q
For the converse, assume  maps njD1 .xj ˝ bj / .xj 2 M; bj 2 Ai / to 0. Then
by Lemma Q 1.12 in Chapter 8 and by the remark after it, the same sum vanishes in
.M 0 ˝ Ai / for a countable pure subgroup M 0
M containing the xj ’s. Hence
M cannot be Mittag-Leffler if its countable pure subgroups are not, but it is if its
countable pure subgroups are Mittag-Leffler.
It remains to prove that countable Mittag-Leffler groups are †-cyclic. Suppose
N is a countable p-groupQ which has nelements ¤ 0 of infinite p-heights.Q Clearly, all
non-zero elements of n2N .N ˝Z.p Q // have finite p-heights. However, n2N Z.p /
n

has summands Š Jp , so N ˝ n2N Z.p / has summands Š N ˝ Jp Š N with

n

elements of infinite heights. Thus N is not Mittag-Leffler. A countable Mittag-

Leffler group must therefore have separable p-components, so its torsion subgroup
is †-cyclic.
Next, let M be of finite torsion-free rank n > 0 such that M=tM is not finitely
generated. Then M contains a subgroup N such that N=tM Š Zn and M=N is an
infinite torsion group. First assume M=N is reduced. Then it is †-cyclic of the form
˚i<! Z.pki i / with ki 2 N, where the pi are not necessarily different primes, but each
prime may occur at most n times. We tensor the exact sequence

0 ! N Š tM ˚ Zn ! M ! ˚i<! Z.pki i / ! 0
Q
with A D j Z.pj / where pj varies over the (infinitely many) different primes in the
set of the pi . In the long exact sequence connecting Tor and ˝, the map Tor.N; A/ !
Tor.M; A/ is an isomorphism as N; M share the same torsion subgroup, thus the
induced sequence

ı
0 ! Tor.˚i<! Z.pki i /; A/!.tM ˚ Zn / ˝ A ! M ˝ A ! : : :

is exact. We calculate: Tor.˚i<! Z.pki i /; A/ Š ˚i Z.pi /, and note that this Tor is sent
by the connecting map ı into Zn ˝A. Q Therefore, M ˝A must contain an imageQ of the
divisible
Q group A= ˚ j Z.p j /. But j .M ˝ Z.p j // is reduced, thus M ˝ j Z.p j/ !
j .M ˝ Z.pj // is not monic. Such an M cannot be Mittag-Leffler.
A similar proof applies to show that M cannot be Mittag-Leffler if M=tM contains
a rank 1 pure Qsubgroup that is p-divisible for some prime p (in this case, we tensor
with A D n2N Z.p /). The conclusion is that if M is Mittag-Leffler, then the
n

finite rank pure subgroups of M=tM are free, i.e. M=tM is free if it is countable
by Theorem 7.1. Therefore, a countable Mittag-Leffler group is †-cyclic. t
u
F Notes. The Baer–Specker group has been investigated from various points of view, it is an
excellent source of ideas. We point out that, among others, Blass–Irwin have several interesting
papers on this group and its subgroups. In their paper [2], several interesting subgroups are dealt
8 Almost Free Groups 119

with. In the other paper [1], a core class for @1 -freeness is discussed: a well-defined class of non-
free @1 -free groups of cardinality @1 such that every non-free @1 -free group of cardinality @1
contains a subgroup from the class. Another interesting result is the existence of indecomposable
@1 -free groups by Palyutin [1] (under CH) which was generalized to rigid @1 -free groups of
cardinality @1 by Göbel–Shelah [2].
Eda [4] shows that a group is @1 -free if and only if it is contained in Z.B/ for some
Boolean lattice B. To illustrate the importance of @1 -freeness, we also mention several topological
connections. L. Pontryagin proved that a connected compact abelian group G is locally connected
exactly if its character group Char G is @1 -free, and J. Dixmier showed that it is arcwise connected
if and only if Ext.Char G; Z/ D 0 (which is stronger than @1 -freeness). We also point out that
for a compact connected group G, the nth homotopy group n .G/ D 0 for all n > 1, while
1 .G/ D Hom.Char G; Z/ is always @1 -free.
That @n -free groups need not be @nC1 -free was proved by Hill, Griffith, and then by Eklof.
Mekler–Shelah [2] study regular cardinals  for which -free implies strongly -free or  C -free.
Gregory [1] proved in L the most interesting Theorem 8.9. Assuming V = L, Rychkov [3] proves
that for each uncountable regular, not weakly compact cardinal , there exist p-groups A of final
rank  such that every subgroup C of cardinality <  is contained in a †-cyclic direct summand
of cardinality jCj@0 , but A itself is not †-cyclic, not even the direct sum of two subgroups of final
ranks .
Mittag-Leffler modules were introduced by M. Raynaud and L. Gruson [Invent. Math. 13, 1–89
(1971)].

Exercises

(1) (a) A direct product of @1 -free groups is @1 -free.

(b) The same may fail for larger cardinals.
(c) Derive from Theorem 8.14 that a direct product of Mittag-Leffler groups is
Mittag-Leffler.
(2) In a free group F, a subgroup G of cardinality <  for which F=G is -free is a
summand.
(3) An extension of a free group by a strongly -free group is strongly -free.
(4) Let A be a direct product of infinite cyclic groups, and B the subgroup of A
whose elements are the vectors with countable support. B is @1 -free, but not
free. Q
(5) In the Baer–Specker group A D k2N hek i, let D denotePthe Z-adic closure of
S D ˚k hek i. Prove that D consists of all vectors x D mk ek such that, for
every n 2 N, n divides almost all mk .
(6) Is it possible to define Mittag-Leffler groups by using only countable index
sets I?
(7) If M is Mittag-Leffler, then so is M=N for every finitely generated subgroup N
of M.
120 3 Direct Sums of Cyclic Groups

9 Shelah’s Singular Compactness Theorem

The question as to when -free implies  C -free turns out to be extremely compli-
cated for regular cardinals  (see Magidor–Shelah [1]). As far as singular cardinals
are concerned, the same question can be fully answered; this is shown by the next
theorem, a most powerful result.
The following lemma will be required in the proof of Theorem 9.2.
Lemma 9.1 (Eklof–Mekler [EM]). If  is a regular cardinal, then a  C -free group
is strongly -free.
Proof. By way of contradiction, assume that A is  C -free, but not strongly -free.
This means that A contains a subgroup B of cardinality <  which is not contained
in any subgroup of A of cardinality <  with -free factor group. Set C0 D B,
and let C1 be a pure subgroup of A of cardinality <  that contains C0 such that
C1 =C0 is not free. Repeat this with C1 in the role of C0 to obtain C2 , and continue
this process transfinitely up to  steps, taking unions at limit ordinals. We get a
chain C0 < C1 <    < C <S: : : . < / where none of the factor groups
C C1 =C is free. The union C D  < C has cardinality , and is not free because
of Theorem 7.5. This contradicts the  C -freeness of A. t
u
Theorem 9.2 (Shelah [1]). For a singular cardinal , a -free group of cardinality
 is free.
Proof. Suppose A is -free of cardinality . Let f j  < cf./g be a smooth
increasing sequence of cardinals > cf./ with  as supremum, and fA j  < cf./g
a smooth chain of pure subgroups of A with union A such that jA j D  . Set

P D fB < A j jBj
 and A=B is C freeg:

Since A is -free for all  < , by Lemma 9.1 it is strongly -free for all  < 
(including limit ordinals < ); thus, every subgroup of A of cardinality
 is
contained in a member of P . For all  < cf./, define subgroups Bk .k < !/ and
subsets Xk .k < !/ such that
(i) Bk 2 P . < cf./; k < !/;
(ii) Xk is a basis of Bk . < cf./; k < !/;
(iii) A < B0 < B1 <    < Bk < : : : and X0  X1      Xk  : : : for each
 < cf./;
(iv) B;k1
hBk \ XC1;k1 i for each  < cf./; 0 < k < !I
(v) for a limit ordinal  < cf./; Xk is the union of a chain of subsets Yk ./
where jYk ./j D  . < /; and Yk ./  BkC1 for all  < .
The construction is by induction on . In the first step, we define the subgroups
B0 . < cf.// recursively on . Let B00 be any member of P0 that contains A0 . If,
for some  < cf./, the B0Phave been defined for all  < , then pick B0 2 P
such that it contains A C < B0 ; this can be done in view of the cardinality
9 Shelah’s Singular Compactness Theorem 121

hypotheses. We are led to a well-ordered ascending chain B00 < B10 <    < B0 <
: : : (that need not be smooth) where B0 has cardinality  . Choose any basis X0
for B0 . For limit ordinals , represent X0 as the union of a chain of subsets Y0 ./
where Y0 ./ has cardinality  . < /:
The next step is to define Bk along with Xk after all Bj ; Xj (and Yj ./ only for
limit ordinals ) have been defined for all j < k and for all  < cf./, and Bk ; Xk ;
and Yk ./ for all  < . Choose Bk 2 P so as to satisfy (iv), and to contain all of
the following: (a) B;k1 ; (b) Bk for all  < ; and (c) the sets Y;k1 ./ for limit
ordinals  > . As B;k1 is a summand of Bk , we can select a basis Xk of Bk that
contains X;k1 . If  happens to be a limit ordinal, we choose the Yk ./ . < /
so as to satisfy (v). An easy cardinality argument convinces us that this can be done
in view of the hypothesis that  > cf./. It is obvious that conditions (i)–(v) are
satisfied. S
We claim that the subgroups B D k<! Bk . < cf.// form a smooth chain
B0 < B1 <    < B < : : : . < cf.// with free factor groups BC1 =B . Observe
that if  is a limit ordinal, then in view of
[ [ [[ [[ [
B D Bk D hXk i D hYk ./i
BkC1 D B ;
k<! k<! k<! < k<! < <

is continuous. Since (iv) implies that B is generated by B \

the chain of the B S
XC1S (where X D k<! Xk /; BC1 =B is indeed free. By Theorem 7.5, the group
A D  B is free. t
u
In Chapter 14, a more general form of the Singular Compactness Theorem will
be needed (for Butler groups); we state it here for groups without proof. This
axiomatic form is due to Eklof–Mekler [EM], generalizing W. Hodges’ version
[Algebra Universalis 12, 205–220 (1981)].
Assume F is a class of groups such that 0 2 F , and for each G 2 F , there is
given a family B.G/ of sets of subgroups of G. We say that G is ‘free’ if G 2 F
and B is a ‘basis’ of G if B 2 B.G/. The subgroups B 2 B are called ‘free’ factors
of G.
For a fixed infinite cardinal , the following properties .i/-.v/ are required for
every ‘free’ group G, and for every ‘basis’ B of G.
.i/ B is closed under unions of chains.
.ii/ If B 2 B and g 2 G, then there is a C 2 B that contains both B and g, and is
such that jCj
jBj C .
.iii/ Every B 2 B is ‘free’ (i.e., ‘free’ factors are ‘free’); and moreover, the set
fC 2 B j C
Bg D B  B is a ‘basis’ for B.
.iv/ If B is a ‘free’ factor of G, then for every ‘basis’ B0 of B, there exists a ‘basis’
B of G such that B0 D B  B.
.v/ Suppose B . < / is a smooth chain of ‘free’ subgroups of G with ‘bases’
B satisfying B  S B D B for all  <  <  (in particular,S B 2 B ).
Then the union B D  < B is a ‘free’ subgroup of G such that  < B is
a ‘basis’ of B.
122 3 Direct Sums of Cyclic Groups

Theorem 9.3. Suppose that the class F of groups satisfies conditions .i/-.v/ for
cardinal , and the cardinality  of the group G 2 F is a singular cardinal > .
G is ‘free’ if, for every cardinal  < , there is a family C of subgroups of G of
cardinality  satisfying the following conditions:
(a) C is a subclass of F ;
(b) C is closed under unions of chains of lengths
;
(c) every subset of G of cardinality
 is contained in a subgroup that belongs
to C . t
u
F Notes. Hill [13] showed that @! -free groups of cardinality @! are free, defeating the
conjecture that -free never implies  C -free. In a subsequent paper, he proved the same for
@!1 -free groups. Based on these results, Shelah conjectured and proved the general theorem on
singular cardinals. (The term ‘compact’ is designated in the sense used in logic, not as in topology:
properties of small substructures imply the same for the entire structure.)
Various generalizations of the compactness theorem are available in the literature which we do
not wish to review here. Let us point out that Hodges [loc.cit.] published an interesting proof of
the singular compactness theorem, based on Shelah’s ideas. The -Shelah game on a group A (for
a regular cardinal ) is introduced; it is played by two players. The players take turns to choose
subgroups of A of cardinalities <  to build an increasing chain fBn gn<! of subgroups. The players
know what subgroups have been chosen at previous steps. Bn is chosen by player I if n is even and
by player II if n is odd. Player II wins if for every odd integer n, Bn is a free summand of BnC2 ,
otherwise player I is the winner. The -Shelah game is determinate in the sense that one of the
players has a winning strategy. It is then shown that player I has no winning strategy, so player
II wins. Being a ‘free summand’ is used in a more general sense in order to obtain a singular
compactness result more general than our Theorem 9.3.

Exercises

(1) Let A be a p-group of singular cardinality . If all subgroups of A of cardinalities

<  are †-cyclic, then A too is †-cyclic. [Hint: the -free vector space AŒp
is
free.]
(2) Let A be a group of singular cardinality . If A is -cyclic, then it is †-cyclic.

10 Groups with Discrete Norm

Normed vector spaces play a most important role in functional analysis. In abelian
group theory, the idea of an integer-valued (more generally, a discrete) norm leads
to an interesting characterization of free groups—a result that has several important
applications.
10 Groups with Discrete Norm 123

Discrete Norm A norm on a group A is a function k : : : k W A ! R such that

(i) k a k  0 for all a 2 A; and k a k D 0 exactly if a D 0;
(ii) k a C b k
k a k C k b k for all a; b 2 A;
(iii) k ma k D jmj:k a k for each m 2 Z and a 2 A.
A norm k : : : k is called discrete if it also satisfies:
(iv) there is a real number  > 0 such that k a k   for all 0 ¤ a 2 A.
(Requirement (iv) is a priori less than demanding that the norms be always
integers.) We record the following elementary facts.
(A) A group with a norm has to be torsion-free. This follows at once from
properties (i) and (iii).
(B) If k : : : k is a .discrete/ norm, then so is rk : : : k for every positive r 2 R.
(C) Subgroups inherit the norm function. Discreteness is inherited as well.
(D) A norm k : : : k on a torsion-free group extends uniquely to a norm on its
divisible hull (for divisible hull, see Sect. 2 in Chapter 4). Needless to say,
an extended norm is never discrete.
Example 10.1. A free abelian group F admits a discrete norm. In fact, if fei gi2I is a free basis of
F, then
X X
k ni ei k D jni j .ni 2 Z/

defines a discrete norm on F. Another way of furnishing F with a discrete norm is by setting
X
k ni ei k D max jni j:

It would be futile to look for other groups as examples, because—as is shown by

the theorem below—only the free groups admit discrete norms.
The discussion starts with the finite rank case.
Lemma 10.2 (Lawrence [1], Zorzitto [1]). A finite rank torsion-free group with
discrete norm is free.
Proof. Let A be torsion-free of finite rank with a discrete norm k : : : k. By induction
on the rank, we prove that A is free.
Without loss of generality, we may assume that k a k  1 for all non-zero a 2 A,
and that there is an x0 2 A whose norm is < 3/2. Under this hypothesis on the norm,
x0 is evidently not divisible in A by any integer > 1, hence the cyclic subgroup hx0 i
must be pure in A. Therefore, if A is of rank 1, then hx0 i is all of A.
Let A be of rank n C 1, and assume the claim holds for groups of rank n  1.
Starting with x0 , pick a maximal independent set fx0 ; x1 ; : : : ; xn g in A. The factor
group A? D A=hx0 i is torsion-free of rank n. It is straightforward to check that one
can define a norm  in A? by setting

.r1 x?1 C    C rn x?n / D jr1 j  k x1 k C    C jrn j  k xn k

124 3 Direct Sums of Cyclic Groups

where the coefficients ri are rational numbers, and stars indicate cosets mod hx0 i.
Supposing A? is not free, induction hypothesis implies that A? cannot have a discrete
norm, so some coset y? D s1 x?1 C    C sn x?n .si 2 Q/ has a norm < 1=4. There is
an a 2 A such that a D s0 x0 C s1 x1 C    C sn xn for some s0 2 Q. By adding to a an
integral multiple of x0 if necessary, we can assume that js0 j
1=2. But then

k a k
js0 j  k x0 k C js1 j  k x1 k C    C jsn j  k xn k < 1=2  3=2 C 1=4 D 1;

a contradiction. Thus A , and hence A, is free. t

u
Free Groups and Discrete Norm We can now verify the main result.
Theorem 10.3 (Stepráns [1]). A group admits a discrete norm if and only if it is
free.
Proof. In view of our example above, it is enough to show that a group A with a
discrete norm k : : : k is free. We induct on the rank  of A. The preceding lemma
settles the case if  is finite, so assume that  is an infinite cardinal, and that the
claim holds for groups of rank < . If  D @0 , then finite rank subgroups are free,
so Pontryagin’s theorem 7.1 implies that A is free.
Next, let  be an uncountable regular cardinal, and 0 D A0 < A1 <    < A <
: : : . < / a smooth chain of Spure subgroups of the group A such that the A are
of cardinality < , and A D  < A . By induction hypothesis, the subgroups A
are free. Consider the set

E D f <  j 9  >  such that A =A is not freeg;

and suppose E is a stationary set in . Without loss of generality, we may assume

that  D  C 1 in the definition of E by thinning out the chain. For each  2 E,
pick elements x  (where  runs over a suitable index set) such that fx  C A g is a
maximal independent set of A C1 =A . As above, define a norm  in A C1 =A by
setting
X X
 . r  .x  C A // D jr  j  k x  k
 

where the coefficients r  are rational numbers, and of course, all sums are finite.
Since A C1 =A has cardinality <  and is not free, the norm
P  cannot be discrete.
Thus there is a coset y C A with norm < 12 , say, y D  s  x  C z for some
z in A .
For convenience, we assume that the underlying set of A consists of all ordinals
< , and A . 2 E/ is just the set of ordinals < . Then the correspondence
W  7! z is a regressive function from E into . Fodor’s theorem (Jech [J])
implies that there exist a z 2 A and a stationary subset E0 of E such that ./ D z
for all  2 E0 .
10 Groups with Discrete Norm 125

Choose different ;  2 E0 such that y ¤ y whose cosets have norm < 12 . We

then have
X X
ky  y k D k. s  x  C z/  . s x C z/k

 
X X

js  j  k x  k C js j  k x k D  .y C A / C  .y C A / < 1;
 

a contradiction. We conclude that E is not stationary, and hence Theorem 7.5 implies
that A is free.
To complete the proof for singular cardinals , it suffices to refer to Shelah’s
singular compactness theorem 9.2. t
u
Corollaries To underscore the significance of this result, we record a few
applications of this theorem.
Let A be an arbitrary group, and X an index set. The set of all functions f W X ! A
such that f assumes but a finite number of distinct values in A is a subgroup B.X; A/
of the cartesian power AX . In case A D Z, this subgroup consists of the bounded
integer-valued functions on X.
Theorem 10.4 (Specker [1], Nöbeling [1]). The group B.X; Z/ of bounded func-
tions on any set X into Z is a free abelian group.
Proof. For the application of Theorem 10.3 all that we have to note is that the group
B.X; Z/ carries a discrete norm. In fact, the norm of a function f 2 B.X; Z/ is defined
as the maximum of the absolute values of integers in the range of f . t
u
An immediate corollary is a far-reaching generalization.
Corollary 10.5 (Kaup–Kleane [1]). The group of all finite-valued functions on a
set X into any group A is a direct sum of copies of A.
Proof. In view of the last theorem, it suffices to verify the isomorphism B.X; A/ Š
A ˝ B.X; Z/. Let hY denote the characteristic function of the subset Y of X, i.e.
hY .x/ D 1 or 0 according as x 2 Y or not. Every f 2 B.X; A/ can be written as

f D a1 hY1 C    C ak hYk .ai 2 A/

for some k and disjoint subsets Y1 ; : : : ; Yk of X. If the characteristic functions hY are

viewed as elements of B.X; Z/, then f can be identified with the element a1 ˝ hY1 C
   C ak ˝ hYk of A ˝ B.X; Z/. t
u
An interesting corollary is concerned with continuous functions on a compact
space. J. de Groot considered the group C.X; Z/ of all continuous functions from a
topological space X into the discrete group of the integers Z. Of special interest is
the case in which X is a compact space. In this case, a continuous function from X
to Z is finite-valued, i.e. C.X; Z/ is a subgroup of B.X; Z/. As such it is free:
126 3 Direct Sums of Cyclic Groups

Corollary 10.6 (de Groot). The group of all continuous functions from a compact
space into the discrete group of the integers is free. t
u
Yamabe [1] considered, for groups A, bilinear, positive definite functions f W
p A ! Z. Note that such a function f defines a discrete norm as usual via k a k D
A 
f .a; a/ for a 2 A. This leads us to
Corollary 10.7. If A is a group such that there is a bilinear, positive definite
function f from A  A into the integers Z, then A has to be a free group. t
u
F Notes. This section is a typical example how a difficult question can sometimes be
rephrased to an easier one by making it more general. Specker [1] could prove only under the CH
that the group of bounded sequences of the integers is free. Nöbeling [1] succeeded in solving the
more general problem on bounded functions of integers by induction on what he called Specker
groups. Bergman [1] provided another proof by establishing an even more general theorem on
commutative torsion-free rings generated by idempotents. Finally, the powerful theorem on groups
with discrete norm was proved. It is due to Stepráns [1] who proved it after Lawrence [1], Zorzitto
[1] settled the countable case. As shown above, this result has important applications.
Hill [14] found an interesting generalization of Bergman’s version by dropping the condition of
torsion-freeness: the additive group of a commutative ring generated by idempotents is †-cyclic.

Exercises

(1) Find a discrete norm on a free group of rank @0 that is not a multiple of any of
examples in Example 10.1.
(2) Let P D ZX and B D B.X; Z/ for an infinite set X. Show that P=B is divisible.
[Hint: for a 2 P; n 2 N find c 2 P with a D nc C b with b 2 B.]
(3) (Nöbeling) Recalling that every element f 2 B.X; A/ can be written as f D
a1 hY1 C  Cak hYk .ai 2 A/ for some k and disjoint subsets Y1 ; : : : ; Yk of X, call a
subgroup S of B.X; A/ a Specker group if f 2 S implies that AhY1 ; : : : ; AhYk are
contained in S. Prove that the following conditions are equivalent for a subgroup
S of B.X; Z/:
(a) S is a Specker group;
(b) f 2 S implies hY where Y denotes the support of f ;
(c) S is a pure subgroup and a subring in ZX . [Hint: (a) , (b) , (c).]
(4) The intersection of Specker subgroups in B.X; Z/ is again a Specker group.
(5) If S is a Specker subgroup in B.X; Z/ and Y  X, then ShY is also a Specker
group.
(6) (Bergman) Let R be a commutative ring with identity whose additive group
RC is torsion-free. If R is generated as a ring by a set E of idempotents, then
RC is a free group. It can be freely generated by idempotents that are products
of elements of E. [Hint: assume E is a multiplicative semigroup, well-order its
elements and show that the elements of E which are not linear combinations of
preceding elements of E in the ordering form a basis of RC .]
11 Quasi-Projectivity 127

11 Quasi-Projectivity

Another fundamental concept in this circle of ideas is that of quasi-projectivity. It

is a natural generalization of projectivity where the projective property is required
only with respect to the group itself (so ‘self-projective’ would probably be a better
name).
Quasi-Projective Groups Thus, a group P is called quasi-projective if
for every exact sequence with P in the middle and for every homomorphism
 W P ! P=G

P
⏐
⏐φ
θ 
α β
0 −−−−→ G −−−−→ P −−−−→ P/G −−−−→ 0

there exists an endomorphism  of P making the triangle commute: ˇ D . Free

groups are quasi-projective, but not only these.
Example 11.1. (a) All cyclic groups are quasi-projective.
(b) Elementary groups are quasi-projective.

A few properties that are worth noting are as follows.

(A) Summands of quasi-projective groups are quasi-projective. If P D X ˚ Y and
G
X, then for any homomorphism  W X ! X=G, the map  ˚ 1Y W P !
X=G ˚ Y D P=G lifts to a  W P ! P, and   X is a desired endomorphism
of X.
(B) A torsion group P D ˚p Pp is quasi-projective if and only if its p-components
Pp are. Necessity follows from (A), and sufficiency is straightforward.
(C) Factor groups modulo fully invariant subgroups inherit quasi-projectivity. To
see this, let S be fully invariant in the quasi-projective group P, and ˇ W P=S ! X
an epimorphism. If  W P=S ! X is any map and W P ! P=S is the canonical
homomorphism, then by the quasi-projectivity of P, there is a map  W P ! P
such that ˇ  D  .
γ
P −−−−→ P/S
⏐
⏐φ
θ 
γ β
P −−−−→ P/S −−−−→ X

Since S is fully invariant in P,  induces a map  0 W P=S ! P=S such that

 D  0 . can be canceled in ˇ 0 D  , thus ˇ 0 D .
(D) Let G be a subgroup of a quasi-projective P such that P=G is isomorphic to a
summand A of P. Then G is a summand of P. Let ˛ W A ! P and  W P ! A be
128 3 Direct Sums of Cyclic Groups

the canonical injection and projection maps, respectively. If ˇ W A ! P=G is an

isomorphism, and W P ! P=G is the canonical map, then by quasi-projectivity
there is a  W P ! P rendering the diagram
ρ
P −−−−→ A
⏐
⏐β
θ
γ
P P/G

commutative. Define the homomorphism ı W P=G ! P as ı D ˛ˇ 1 ; then

ı D ˛ˇ 1 D ˇ˛ˇ 1 D 1P=G . This means that the exact sequence 0 !

G ! P!P=G ! 0 is splitting.
(E) Let G be a subgroup of the quasi-projective group P such that there is an
epimorphism  W G ! P. Then K D Ker  is a summand of G. Let ˛ W G=K ! P
be the isomorphism induced by . We have an injection ˇ W P ! P=K with
Im ˇ D G=K and ˇ˛ D 1G=K . By quasi-projectivity, there is a  W P ! P such
that  D ˇ where W P ! P=K denotes the canonical map. We argue that
.P/
1 .G=K/ D G. Now ı D ˛ W G=K ! G satisfies ı D 1G=K , thus

0 ! K ! G!G=K Š P ! 0 is a splitting exact sequence.
Structure of Quasi-Projective Groups A complete classification of quasi-
projective groups can be given in terms of cardinal invariants, based on the following
theorem.
Theorem 11.2 (Fuchs–Rangaswamy [2]). A group is quasi-projective if and only
if either
(i) it is a free group; or
(ii) it is a torsion group such that each of its p-components is a direct sum of cyclic
groups of fixed order pkp .
Proof. Free groups F are obviously quasi-projective, and (C) implies that the groups
F=nF are also quasi-projective for every n 2 N. By (B), the same holds for the direct
sum ˚.F=pkp F/ with different primes p. As F=pkp F is a direct sum of cyclic groups
of fixed order pkp , the sufficiency follows.
Conversely, assume P is quasi-projective. If P is torsion, then it cannot have
a summand Z.p1 /, because by (A) this summand would be quasi-projective, so
by (D) it would contain every Z.pn / as a summand—this is impossible. Thus P is
reduced. It cannot have a summand of the form C D Z.pn / ˚ Z.pm / with n > m,
since there is an epimorphism Z.pn / ! Z.pm / whose kernel is not a summand of C
(cp. (D)). Therefore, the p-components of P are bounded by some pk with no cyclic
summands of different orders. Hence (ii) holds for P if torsion.
If P is torsion-free, then let F be a free subgroup of P generated by a maximal
independent set, so that P=F is a torsion group. Let W P ! P=F denote the natural
map. We distinguish two cases according as P is of finite or infinite rank. If rk P is
finite, then for every map  W P ! P=F there is a  W P ! P with  D  , and
11 Quasi-Projectivity 129

for different  we have different . If P=F were infinite, then it had continuously
many automorphisms (see Sect. 2 in Chapter 17, Exercise 3), so there would be this
many choices for . But a finite rank P has only countably many endomorphisms.
Thus P=F must be finite, P is finitely generated, so P is finitely generated free. If
rk P is infinite, then we can find a surjective map F ! P, and (E) shows that P is
isomorphic to a summand of F, so it is free.
Finally, suppose P is mixed. (C) implies that P=tP is quasi-projective, so free:
P D tP ˚ F with F a free group. If none of the summands is 0, then there is
an epimorphism F ! C
tP; C cyclic, whose kernel is not a summand of F,
contradicting (D). Thus P cannot be mixed. t
u
F Notes. The listed properties of quasi-projectivity were borrowed from the pioneering paper
L. Wu–J.P. Jans [Ill. J. Math. 11, 439–448 (1967)], and from Fuchs–Rangaswamy [2]. The majority
of the results (e.g., Theorem 11.2 is an exception) are valid for modules as well.

Exercises

(1) Describe the complete set of cardinal invariants attached to a quasi-projective

group.
(2) P./ is quasi-projective for every cardinal  whenever P is quasi-projective.
(3) (a) Suppose P D ˚n<! Pn where the Pn are fully invariant in P. P is quasi-
projective if and only if every Pn is quasi-projective.
(b) Claim (a) may fail if the summands are not fully invariant.
(4) Fully invariant subgroups inherit quasi-projectivity.
(5) Only quasi-projective groups admit quasi-projective covers.
(6) Let G < P; P a quasi-projective group. Then j End P=Gj
j End Pj.

Problems to Chapter 3

PROBLEM 3.1. Characterize almost free groups in which the intersection of two
direct summands is again a summand.
PROBLEM 3.2. For which ordinals  do there exist (strongly) @ -free groups
that are not (strongly) @ C1 -free?
PROBLEM 3.3 (IRWIN). Is there a core class of @1 -free groups? That is, a small
collection of @1 -free groups, each of cardinality @1 , such that every @1 -free group
contains a member of this class.
Cf. Blass–Irwin [1].

PROBLEM 3.4. Let A be the free lattice-ordered group generated by the partially
ordered group G. Relate A to G as groups.
Chapter 4
Divisibility and Injectivity

Abstract Most perfect objects in the category of abelian groups are those groups in which we
can also ‘divide:’ for every element a and for every positive integer n, the equation nx D a has a
(not necessarily unique) solution for x in the group. These objects are the divisible groups which
are universal in the sense that every group can be embedded as a subgroup in a suitable divisible
group.
The divisible groups form one of the most important classes of abelian groups. In our
presentation, we focus on their most prominent properties, many of them may serve as their
characterization. Their outstanding feature is that they coincide with the injective groups, and
as such they are direct summands in every group containing them as subgroups. Moreover, they
constitute a class in which the groups admit a satisfactory characterization in terms of cardinal
invariants.
The concluding topic for this chapter is concerned with a remarkable duality between maximum
and minimum conditions on subgroups.

1 Divisibility

Since multiplication of group elements by integers makes sense, it is natural to

consider divisibility of group elements by integers. Divisibility offers a great deal of
information on how an element fits in the group.
Divisibility of Elements We shall say that the element a of the group A is
divisible by n 2 N, in symbols: nja, if the equation

nx D a .a 2 A/ (4.1)

is solvable for x in A, i.e., there exists a b 2 A such that nb D a. Evidently, (4.1) is

solvable if and only if a 2 nA.
We list some elementary consequences of the definition.
(a) If x D b is a solution to (4.1), then the coset b C AŒn
is the set of all solutions
of (4.1).
(b) If A is torsion-free, then (4.1) has at most one solution.
(c) If gcdfn; o.a/g D 1; then (4.1) is solvable. For if r; s 2 Z are such that nr C
o.a/s D 1, then x D ra satisfies nx D nra D nra C o.a/sa D a:

© Springer International Publishing Switzerland 2015 131

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_4
132 4 Divisibility and Injectivity

(d) mja and nja imply lcmfm; ngja. Indeed, if r; s satisfy mr C ns D d D gcdfm; ng;
and if b; c 2 A are such that mb D a D nc, then (lcmfm; ng/.rc C sb/ D
mnd1 .rc C sb/ D md 1 ra C nd1 sa D a.
(e) nja and njb with a; b 2 A imply nj.a ˙ b/.
(f) If A D B ˚ C is a direct sum, then nja D b C c .b 2 B; c 2 C/ if and only if
both njb in B and njc in C. The same holds for infinite direct sums and direct
products.
(g) If ˛ W A ! B is a homomorphism, then nja in A implies nj˛b in B.
(h) If p is a prime, then pk ja is equivalent to k
hp .a/.
Divisibility of Groups A group D is called divisible if

njd for all d 2 D and all 0 ¤ n 2 Z:

Thus D is divisible exactly if nD D D for every integer n ¤ 0.

Example 1.1. The groups 0; Q; Q=Z; Z.p1 /; R are divisible groups, but no non-zero cyclic
group is divisible.

A most useful criterion of divisibility is our next lemma.

Lemma 1.2. A group D is divisible if and only if every homomorphism  W Z ! D
can be extended to a homomorphism  W Q ! D.
Proof. Let  be a homomorphism of Z into the divisible group D. We think of Q
as the union of the chain h1i < h.2Š/1 i <    < h.nŠ/1 i < : : : of infinite cyclic
groups. The element .1/ D d1 2 D is divisible by 2, so there is a d2 2 D with
2d2 D d1 , and we can extend  to 2 W h.2Š/1 i ! D by letting 2 ..2Š/1 / D d2 .
If we have an extension n W h.nŠ/1 i ! D with n ..nŠ/1 / D dn , then we select a
dnC1 2 D satisfying .n C 1/dnC1 D dn , and define nC1 ...n C 1/Š/1 / D dnC1 to
extend n . At the end of this stepwise process we arrive at a desired homomorphism
 W Q ! D (whose restriction to h.nŠ/1 i equals n ).
Conversely, assume that the group D has the indicated property, and pick any
d 2 D. There is a homomorphism  W Z ! D with .1/ D d. By hypothesis,  can
be extended to a map  W Q ! D. Then the element .n1 / satisfies n.n1 / D d,
establishing the divisibility of d. t
u
A group D is said to be p-divisible (p a prime) if pk D D D for every positive
integer k. Since pk D D p    pD, it is obvious that p-divisibility is implied by pD D
D; i.e. every element of D is divisible by p. Then every element is of infinite p-height
in D.
(A) A group is divisible if and only if it is p-divisible for every prime p. Indeed,
if pD D D for every prime p and n D p1 p2    pk with primes pi , then nD D
p1 p2    pk D D p1    pk1 D D    D p1 D D D:
(B) A p-group is divisible if and only if it is p-divisible. In view of (c), for a p-group
A we always have qA D A whenever the primes p; q are different.
(C) A p-group D is divisible exactly if every element of order p is of infinite height.
Only sufficiency requires a proof. So assume every element of order p is of
1 Divisibility 133

infinite height, and let a 2 D be of order pk . We induct on k to prove that

pja. For k D 1, the claim is included in the hypothesis, so assume k > 1.
By hypothesis, pk1 a has infinite height, thus pk1 a D pk b for some b 2 D.
Since o.a  pb/
pk1 ; we have pja  pb by induction hypothesis, whence
pja, indeed.
(D) Epimorphic images of a divisible group are divisible. This is an immediate
consequence of (g) above.
(E) A direct sum .direct product/ of groups is divisible if and only if each
component is divisible. This follows at once from (f). P
(F) If Di .i 2 I/ are divisible subgroups of A, then so is their sum i2I Di . This is
evident in view of (e).
An immediate consequence is that the sum of all divisible subgroups of a group
is again divisible, so we have the following result:
Lemma 1.3. Every group A contains a maximal divisible subgroup D. D contains
all divisible subgroups of A. t
u
Groups that contain no divisible subgroups other than 0 are called reduced.
Embedding in Divisible Groups Recall that free groups are universal in the
sense that every group is an epic image of a suitable free group. The next result
shows that divisible groups have the dual universal property.
Theorem 1.4. Every group can be embedded as a subgroup in a divisible group.
Proof. The infinite cyclic group Z can be embedded in a divisible group, namely,
in Q. Hence every free group can be embedded in a direct sum of copies of Q, which
is a divisible group. Now if A is an arbitrary group, then A Š F=H for some free
group F and a subgroup H of F. If we embed F in a divisible group D, then A will
be isomorphic to the subgroup F=H of the divisible group D=H. t
u
It follows that, for every group A, there is an exact sequence

0!A!D!E!0

with D and (hence) E divisible.

F Notes. There is no need to emphasize the relevance of divisibility in the theory of abelian
groups: the reader will soon observe that a very large number of theorems rely on this concept. (In
early literature, divisible groups were called complete groups.)

Exercises

(1) The additive group of any field of characteristic 0 is divisible.

(2) The factor group Jp =Z is divisible.
134 4 Divisibility and Injectivity

(3) A group is divisible exactly if it satisfies one of the following conditions:

(a) it has no finite epimorphic image ¤ 0;
(b) it has no maximal subgroups (it coincides with its own Frattini sub-
group).
(4) If fai gi2I is a generating set (or a maximal independent set) in a group D, and
if njai in D for every i 2 I and every n 2 N, then D is divisible.
(5) A direct sum (direct product) of groups is reduced if and only if every
component is reduced.
(6) Let 0 ! A ! B ! C ! 0 be an exact sequence. If both A and C are
divisible (p-divisible), then so is B; if both are reduced, then B is also reduced.
(7) The maximal divisible subgroup of a torsion-free group coincides with the
first Ulm subgroup of the group.
(8) Let A be the direct product, and B the direct sum of the groups Bn .n < !/.
A=B is a divisible group if and only if, for every prime p, pBn D Bn holds for
almost all n.
(9) Direct limits of divisible groups are divisible.
(10) (Szélpál) Assume A is a group such that all non-zero factor groups of A are
isomorphic to A. Show that A Š Z.p/ or A Š Z.p1 / for some p.
(11) Using Hom and Ext, show that (a) D is divisible if and only if Ext.Q=Z; D/ D
0; (b) A is reduced if and only if Hom.Q; A/ D 0.

2 Injective Groups

Injective groups are dual to projective groups; they are defined by dualizing the
definition of projectivity.
Injectivity A group D is said to be injective if, for every diagram

α
0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0
⏐
⏐ξ
 η

with exact rows and a homomorphism  W B ! D, there is a homomorphism

 W A ! D making the triangle commute: ˛ D . If B is identified with its
image in A, then the injectivity of D can be interpreted as the extensibility of any
homomorphism  W B ! D to a homomorphism of any group A containing B into D.
Our next purpose is to show that the injective groups are precisely the divisible
groups.
2 Injective Groups 135

Theorem 2.1 (Baer [8]). A group is injective if and only if it is divisible.

Proof. That an injective group is divisible follows at once from Lemma 1.2. In order
to verify the converse, let D be a divisible group, and  W B ! D a homomorphism
from a subgroup B of the group A. Consider all groups G between B and A, such
that  has an extension  W G ! D. The set S of all pairs .G; / is partially ordered
by setting

.G; /
.G0 ;  0 / if and only if G
G0 and  D  0  G:

S is not empty, since .B; / 2 S, and is inductive, since

S every chain .GiS ; i / .i 2 I/
has an upper bound in S, viz. .G; / where G D i2I Gi and  D i2I i . By
Zorn’s lemma, there exists a maximal pair .G0 ; 0 / in S. We claim: G0 D A.
By way of contradiction, suppose G0 < A. If a 2 A n G0 is such that na D g 2 G0
for some n 2 N, then choose a minimal such n. By the divisibility of D, some d 2 D
satisfies nd D 0 g. It is straightforward to check that

x C ka 7! 0 x C kd .x 2 G0 ; 0
k < n/

is a genuine homomorphism of hG0 ; ai into D. If na … G0 for all n 2 N, then the

correspondence x C ka 7! 0 x C kd .x 2 G0 / is a homomorphism for any choice of
d 2 D (no restriction on k 2 Z). In either case, G0 < A contradicts the maximality
of .G0 ; 0 /. Hence 0 W A ! D is a desired extension of . t
u
Baer’s Criterion From the proof we derive the famous Baer criterion for
injectivity (whose real importance lies in the fact that its analogue holds for modules
over any ring):
Corollary 2.2. A group D is injective if and only if, for each n 2 N, every
homomorphism Zn ! D extends to a homomorphism Z ! D.
Proof. A careful analysis of the preceding proof shows that the only place where
the injectivity of D was needed was to assure the existence of a d 2 D. The same
conclusion can be reached if we extend the map Zn ! D given by n 7! 0 g to
Z ! D, and pick d as the image of 1. t
u
Another simple result is a trivial consequence of Theorem 2.1 and Corollary 2.2,
but it is important enough to record it as a corollary.
Corollary 2.3. Epic images of injective groups are injective. t
u
The Summand Property We are now able to show that injective (divisible)
subgroups are always summands.
Corollary 2.4 (Baer [8]). A divisible subgroup D of a group A is a summand, A D
D ˚ C for some subgroup C of A. Here C can be chosen so as to contain any
preassigned subgroup B of A with D \ B D 0. .Thus D-high subgroups are always
summands./
 136 4 Divisibility and Injectivity

Proof. By Theorem 2.1, the identity map 1D W D ! D extends to a homomorphism

 W A ! D. Therefore, A D D ˚ Ker . If D \ B D 0, then the same argument
implies that the map D ˚ B ! D which is the identity on D and trivial on B extends
to an  W A ! D. Then evidently, B
Ker . t
u
Given a group A, consider the subgroup generated by all divisible subgroups of A.
From Sect. 1(F) we know that D is divisible, it is the maximal divisible subgroup
of A. By Corollary 2.4, A D D ˚ C, where evidently, the summand C has to be
reduced. We thus have the first part of
Theorem 2.5. Every group A is the direct sum of a divisible group D and a reduced
group C,

A D D ˚ C:

D is a uniquely determined subgroup of A, C is unique up to isomorphism.

Proof. To verify the second claim, it is clear that if A D D ˚ C with D divisible and
C reduced, then D ought to be the unique maximal divisible subgroup of A. Hence
D is unique, and a complement is as always unique up to isomorphism. t
u
A consequence of the last theorem is that a problem on abelian groups can be
often reduced to those on divisible and reduced groups.
We now summarize as a main result:
Theorem 2.6. For a group, the following conditions are equivalent:
(i) it is divisible;
(ii) it is injective;
(iii) it is a direct summand in every group containing it.
Proof. We had proved earlier the implications (i) , (ii) ) (iii), so only (iii) ) (i)
is needed to complete the proof. Let D satisfy (iii). Theorem 1.4 shows that D
E
for a divisible group E. Then (iii) implies D is a summand of E, so (i) follows. u t
Injective Hulls Theorems 1.4 and 2.1 guarantee that every group can be
embedded in an injective group. This important fact can be considerably improved
by establishing the existence of a minimal embedding.
A minimal divisible (injective) group containing the group A is called a divisible
hull or injective hull of A, it will be denoted as E.A/. The main result on injective
hulls is as follows.
Theorem 2.7. Any injective group containing A contains an injective hull of A. The
injective hull of A is unique up to isomorphism over A.
Proof. Let E be an injective group containing A. There exists a subgroup C of E
maximal with respect to the property A \ C D 0. For each c0 2 C and prime p, there
is x 2 E such that px D c0 2 C. If x … C, then c C kx D a ¤ 0 .c 2 C; a 2 A/
for some k 2 Z prime to p. Then pc C kc0 D pa must be 0, which shows that kc0 ,
and hence c0 is divisible by p in C. That is, C is divisible. By Corollary 2.4, we can
2 Injective Groups 137

write E D C ˚ D with A
D. Clearly, D is divisible, and by the maximality of
C, D cannot have any proper summand still containing A. Therefore, D is minimal
divisible containing A, and by the choice of C, A is essential in D.
If D1 and D2 are two minimal injective groups containing A, then because of
Theorem 2.1, the identity map 1A of A extends to a homomorphism  W D1 ! D2 .
Since D1 is divisible containing A, we have A
D1
D2 . By the minimality
of D2 , the last
must be equality. Now jA D 1A implies Ker  \ A D 0 which
means that  is monic. Hence  is an isomorphism between D1 and D2 leaving A
element-wise fixed. t
u
An important consequence of the preceding proof is
Corollary 2.8. An injective group E containing A is an injective hull of A if and
only if A is an essential subgroup in E. t
u
Example 2.9.
(a) The injective hull of a torsion-free group A is its divisible hull; it can also be obtained as
Q ˝ A.
P basic subgroup B D ˚i2I hbi i, embed
(b) In order to get the injective hull of a p-group A with
each hbi i in a quasi-cyclic group Ci , and form AC i2I Ci . (Check that this group is divisible,
and A is essential in it.)
Quasi-Injectivity We now embark upon a noteworthy generalization of injec-
tivity. A group A is called quasi-injective if every homomorphism of every
subgroup into A extends to an endomorphism of A.
(A) Summands of quasi-injective groups are again quasi-injective.
(B) Powers of a quasi-injective group are quasi-injective.
(C) A torsion group A is quasi-injective if and only if all of its p-components Ap are
quasi-injective.
(D) Neither direct sums nor direct products of quasi-injective groups are necessarily
quasi-injective (this is trivial from Theorem 2.11).
Theorem 2.10 (Johnson–Wong). A group is quasi-injective if and only if it is a
fully invariant subgroup of its injective hull.
Proof. First assume A is fully invariant in an injective group E, and let W B ! A
where B
A. By injectivity,  extends to an endomorphism W E ! E which—by
full invariance—must map A into itself.
Conversely, let A be quasi-injective, and  an endomorphism of the injective hull
E of A. The subgroup B D fa 2 A j a 2 Ag is mapped by  into A, so  extends
to a map W A ! A. If .  /a 2 A for an a 2 A, then a 2 A, so a 2 B. Thus
.  /A \ A D 0, whence .  /A D 0 by the essential character of A. This
means that   A D , and so .A/
A; indeed. t
u
From full invariance it follows that if E D E1 ˚ E2 is a direct decomposition
of the injective hull of a quasi-injective group A, then A has a corresponding
decomposition: A D .A \ E1 / ˚ .A \ E2 /.
A consequence of the last lemma is that every group A admits a quasi-injective
hull: a smallest quasi-injective group containing A as a subgroup. This is simply
138 4 Divisibility and Injectivity

the fully invariant subgroup generated by A in the injective hull E.A/ of A.

This characterization of quasi-injective groups enables us to prove the following
structural result on them.
Theorem 2.11 (Kil’p [1]). A group is quasi-injective exactly if it either injective
or is a torsion group whose p-components are direct sums of isomorphic cocyclic
groups.
Proof. If A is a group as stated, then it is immediately seen that it is fully invariant
in its injective hull.
Conversely, let A be a quasi-injective group, and E its injective hull. If A contains
an element a of infinite order, then for every b 2 E there is a map  W hai ! E
with a D b, so the only fully invariant subgroup of E containing a is E itself. Thus
A D E in this case. If A is a torsion group, then its p-components Ap are likewise
quasi-injective, so fully invariant in their injective hulls Ep . The latter group is a
direct sum of copies of Z.p1 / (see Theorem 3.1), and its non-zero fully invariant
subgroups are the direct sums of copies of a fixed subgroup Z.pk / of Z.p1 /; where
k 2 N or k D 1. t
u
More on Quasi-Injectivity Quasi-injective groups have several remarkable
properties which led to various generalizations of quasi-injectivity in module
categories (for abelian groups some of them coincide with quasi-injectivity). We
mention a few interesting facts for illustration.
Proposition 2.12. A quasi-injective group has the following properties:
(i) it is a CS-group: high subgroups are summands;
(ii) it is an extending group: every subgroup is contained as an essential subgroup
in a direct summand;
(iii) a subgroup that is isomorphic to a summand is itself a summand;
(iv) if B; C are summands and B \ C D 0, then B ˚ C is also a summand.
Proof. (i) , (ii) is routine.
(ii) Let G be a subgroup of the quasi-injective group A, and let E1 denote the
injective hull of G in the injective hull E of A. Then E D E1 ˚ E2 holds for
some E2
E, and A D .A \ E1 / ˚ .A \ E2 /. Clearly, G is essential in the
first summand of A.
(iii) Let H be a subgroup, and G a summand of A with inverse isomorphisms
W G ! H; ˇ W H ! G. Now ˇ followed by the injection map G ! A
extends to an endomorphism ˛ W A ! A. If this is followed by the projection
A ! G and then by , then the composite is 1H . As this extends to A ! H,
H is a summand.
(iv) Let A D B ˚ B0 and  W A ! B0 the projection. Then B ˚ C D B ˚ C where
  C is an isomorphism. By (iii), C is a summand of A and hence of B0 . It
follows B ˚ C is a summand of A. t
u
2 Injective Groups 139

F Notes. The true significance of injectivity of groups lies not only in its extremely important
role in the theory of abelian groups, but also in the fact that it admits generalizations to modules
over any ring such that most of its relevant features carry over to the general case. The injective
property was discovered by Baer [8]. He also proved that for the injectivity of a left R-module M,
it is necessary and sufficient that every homomorphism from every left ideal L of R into M extends
to an R-homomorphism R ! M. This extensibility property, with L restricted to principal left
ideals generated by non-zero divisors, is perhaps the most convenient way to define divisible R-
modules. It is then immediately clear that an injective module is necessarily divisible. For modules
over integral domains, the coincidence of injectivity and divisibility characterizes the Dedekind
domains (see Cartan–Eilenberg [CE]). For torsion-free modules over Ore domains divisibility
always implies injectivity.
Mishina [3] calls a group A weakly injective if every endomorphism of every subgroup extends
to an endomorphism of A. Besides quasi-injective groups, only the groups of the form A D D ˚ R
have this property where D is torsion divisible and R is a rational group.
Epimorphic images of injective left R-modules are again injective if and only if R is left
hereditary (left ideals are projective); an integral domain is hereditary exactly if it is Dedekind.
Note that the semi-simple artinian rings are characterized by the property that all modules over
them are injective.
It is an easy exercise to show that over left noetherian rings every left module contains a
maximal injective submodule. This is not necessarily a uniquely determined submodule, unless the
ring is, in addition, left hereditary. E. Matlis [Pac. J. Math. 8, 511–528 (1958)] and Z. Papp [Publ.
Math. Debrecen 6, 311–327 (1959)] proved that every injective module over a left noetherian ring
R is a direct sum of directly indecomposable ones. If, in addition, R is commutative, then the
indecomposable injective R-modules are in a one-to-one correspondence with the prime ideals P
of R, namely, they are the injective hulls of R/P (for R D Z take P D .0/ or .p/); cf. Matlis
[loc. cit.]. It is remarkable that direct sums (and direct limits) of injective left modules are again
injective if and only if the ring is left noetherian.
Before the term ‘essential extension’ was generally accepted, some authors were using
‘algebraic extension’ instead. Szele [3] developed a theory of ‘algebraic’ and ‘transcendental’
extensions of groups, modeled after field theory, where ‘algebraic’ meant ‘essential,’ while
‘transcendental’ was used for ‘non-essential’ extensions. He established the analogue of algebraic
closure (injective hull) a few years before the Eckman–Schopf paper on the existence of injective
hulls was published.
Quasi-injectivity was introduced by R.E. Johnson and E.T. Wong [J. London Math. Soc. 36,
260–268 (1961)]. As far as generalizations of quasi-injectivity are concerned, interested readers
are referred to the monographs S.H. Mohamed and B.J. Müller, Continuous and Discrete Modules,
London Math. Soc. Lecture Notes 17 (1990), and N.V. Dung, D.V. Huynh, P.F. Smith, R. Wisbauer,
Extending Modules, Pitman Research Notes 313 (1994).

Exercises

(1) (Kertész) A group is divisible if and only if it is the endomorphic image of

every group containing it.
(2) If A D ˚i2I Bi , then E.A/ D ˚i2I E.Bi / where E./ stands for the injective
hull.
(3) Every automorphism of a subgroup A of an injective group D is induced by
an automorphism of D.
(4) A direct sum of copies of Z.p1 / is injective as a Jp -module as well.
140 4 Divisibility and Injectivity

(5) Let A be a torsion-free group, and E the set of all pairs .a; m/ 2 A  Z with
m ¤ 0 subject to
(a) .a; m/ D .b; n/ if and only if mb D na;
(b) .a; m/ C .b; n/ D .na C mb; mn/:
Show that E is a divisible hull of the image of A under the map a 7! .a; 1/.
(6) If C is a subgroup of the group B such that B=C is isomorphic to a subgroup
H of G, then there exists a group A containing B such that A=C Š G.
(7) Given A and integer n > 0, there exists an essential extension C of A such
that A D nC. Is C unique up to isomorphism (over A)?
(8) (a) (Charles, Khabbaz) A subgroup A of a divisible group D is the inter-
section of divisible subgroups of D if and only if, for every prime p,
AŒp
D DŒp
implies pA D A:
(b) (Bergman) Every group is the intersection of divisible subgroups in a
suitable divisible group. [Hint: push-out of two different injective hulls.]
(9) (Szele) Let B be a subgroup of A. Call an a 2 A of infinite or prime power
order algebraic over B if a D 0 or hai \ B ¤ 0. A is algebraic over B if
every a 2 A is algebraic over B.
(a) A is algebraic over B if and only if B is an essential subgroup of A.
(b) A is a maximal algebraic extension of B exactly if A D E.B/:
(c) Derive Theorem 2.1 from the existence of maximal algebraic extensions
in E.
(10) The group A Š Z.p2 / ˚ Z.p/ is a CS-group, but not quasi-injective.

3 Structure Theorem on Divisible Groups

Structure of Divisible Groups The groups Q and Z.p1 / were among our first
examples for divisible groups. The main theorem of this section will show that there
are no divisible groups other than the direct sums of copies of Q and Z.p1 / with
various primes p.
Theorem 3.1. A divisible group D is the direct sum of groups each of which is
isomorphic either to the additive group Q of rational numbers and or to a quasi-
cyclic group Z.p1 / W

D Š .˚ Q/ ˚ ˚p .˚p Z.p1 //:

The cardinal numbers ; p .for every prime p/ form a complete and independent
system of invariants for D.
Proof. The torsion part T D tD of D is divisible, so by Corollary 2.4 it is a
summand: D D T ˚ E, where E is torsion-free and divisible. The p-components
3 Structure Theorem on Divisible Groups 141

Tp of T D ˚p Tp are divisible, so it suffices to show that Tp is a direct sum of copies

of Z.p1 /, and E is a direct sum of copies of Q.
Owing to divisibility, for each a 2 Tp we can find a sequence a D
a0 ; a1 ; a2 ; : : : ; an ; : : : in Tp such that panC1 D an for n D 0; 1; : : : . Thus every
element embeds in a subgroup Š Z.p1 /. Consider the set S of subgroups Bi in
Tp that are direct sums of subgroups Š Z.p1 /, and partially order S by declaring
Bi
Bj if Bi is a summand of Bj . Use Zorn’s lemma to argue that Tp contains a
maximal B 2 S. Such a B is injective, so Tp D B ˚ C. If C ¤ 0, then it must contain
a subgroup Š Z.p1 /, contradicting the maximal choice of B. Hence Tp D B, and
Tp is a direct sum of copies of Z.p1 /. The proof for E is similar, making use of the
embeddability of every element in a subgroup Š Q.
To show that the cardinal numbers of the summands Q and Z.p1 / do not depend
on the special choice of the decompositions, it is enough to note that these cardinals
are exactly the ranks rk0 .D/ and rkp .D/, which are uniquely determined by D. They
do form a complete system of invariants for D, since if given rk0 .D/ D  and
rkp .D/ D p , we can uniquely reconstruct D as a direct sum of  copies of Q and
p copies of Z.p1 / for each prime p. Moreover, that these cardinals can be chosen
arbitrarily is obvious. t
u
Note that Corollary 2.8 implies that

rk0 .E.A// D rk0 .A/ and rkp .E.A// D rkp .A/ for every prime p:

Consequently, the structure of the divisible hull E.A/ of A is completely determined

by the ranks of A.
Example 3.2. The additive group R of the real numbers is a torsion-free divisible group of the
power of the continuum 2@0 . Hence R Š ˚ Q, where  D 2@0 .
Example 3.3. The multiplicative group of the positive real numbers is a torsion-free divisible
group of the power of the continuum 2@0 . It is isomorphic to R under the correspondence
r 7! log r:
Example 3.4. The multiplicative group R of the non-zero real numbers is the direct product of
the group in Example 3.3 and the multiplicative cyclic group h  1i Š Z.2/.
Example 3.5. The multiplicative group T of complex numbers of absolute value 1, the circle
group, is isomorphic to R=Z, the (additive) group of reals mod 1. The torsion subgroup is Š Q=Z,
so the p-components are quasi-cyclic p-groups. Therefore,

R=Z Š ˚p Z.p1 / ˚ .˚ Q/

where again  D 2@0 .

Example 3.6. The multiplicative group C of all complex numbers ¤ 0 is the direct product of
the circle group and the multiplicative group of the positive real numbers. It is isomorphic to its
subgroup: the circle group.

Cogenerators A group C is called a cogenerator of the category Ab of abelian

groups if every abelian group is contained in a suitable direct product of copies of
C, or, equivalently, every non-zero abelian group has a non-trivial homomorphism
into C.
142 4 Divisibility and Injectivity

Theorem 3.7. A group is a cogenerator of Ab if and only if it has a summand

isomorphic to Q=Z. Thus Q=Z is the minimal cogenerator of the category Ab.
Proof. Every non-trivial homomorphic image of a quasi-cyclic group is isomorphic
to the group itself, so any cogenerator C must contain Z.p1 / for each prime p.
These Z.p1 / generate their direct sum in any group. Such a direct sum is Š Q=Z,
an injective group, so a summand of C.
Conversely, any group A ¤ 0 has a non-trivial homomorphism into Q=Z. For,
0 ¤ a 2 A implies hpai < hai for some prime p, and then the cyclic group hai can
be mapped upon the cyclic subgroup hci of order p in Z.p1 / via a 7! c. This map
extends to a homomorphism of A ! Q=Z. t
u
It is an important fact that the endomorphism rings of injective groups are very
special. We do not discuss them here, because later we will learn more about these
rings. We refer to Theorem 4.3 in Chapter 16.

Exercises

(1) Find the cardinal invariants for the following groups: (a) the direct product of
 copies of Z.p1 /; (b) the direct product of  copies of Q=Z; (c) the direct
product of  copies of R=Z; here  denotes an infinite cardinal.
(2) Any two direct decompositions of a divisible group have isomorphic
refinements.
(3) If A; B are divisible groups, each containing a subgroup isomorphic to the
other, then A Š B.
(4) If A is divisible, and B is a group such that A ˚ A Š B ˚ B, then B Š A.
(5) Find minimal cogenerators for the following categories: (a) torsion-free
groups; (b) torsion groups; (c) p-groups.
(6) (Szele) A group contains no two distinct isomorphic subgroups if and only if
it is isomorphic to a subgroup of Q=Z.
(7) (Kertész) Let A be a p-group in which the heights of elements of finite heights
are bounded by an integer m > 0. Then A is the direct sum of cocyclic groups.
[Hint: pm A is divisible.]
(8) (a) (E. Walker) Any torsion-free group of infinite rank is a subdirect sum of
copies of the group Q.
(b) An unbounded p-group is a subdirect sum of quasi-cyclic groups.
(9) (a) For every infinite cardinal , there is a group U of cardinality  which
contains an isomorphic copy of every group of cardinality
.
(b) In the set of groups U with the indicated property there is one that is
isomorphic to a summand of every other one.
(10) (W.R. Scott) An infinite group A is a Jónsson group if every proper subgroup
has cardinality < jAj. Prove that A is a Jónsson group if and only if A Š
Z.p1 / for some prime p. [Hint: it is indecomposable, divisible, torsion.]
4 Systems of Equations 143

(11) A group is hopfian if its surjective endomorphisms are automorphisms; it is

cohopfian if its injective endomorphisms are automorphisms. Show that the
only hopfian-cohopfian torsion-free groups are the finite direct sums of Q. (It
is difficult to construct an infinite hopfian-cohopfian p-group.)

4 Systems of Equations

By the definition of divisible groups D, all ‘linear’ equations of the form nx D d 2 D

with positive integers n are solvable for x in D. It is natural to raise the question
of solvability of systems of linear equations in D. We are going to show that all
consistent systems of linear equations are solvable in any divisible group.
Systems of Linear Equations By a system of equations over a group A is
meant a set of equations
X
nij xj D ai .ai 2 A; i 2 I/ (4.2)
j2J

where the coefficients nij are integers such that, for any fixed i 2 I, almost all nij D 0;
here, fxj gj2J is a set of unknowns, while I; J are index sets of arbitrary cardinalities.
Equation (4.2) is a homogeneous system if ai D 0 for all i 2 I. We say that
xj D gj 2 A .j 2 J/ is a solution to (4.2) if Eq. (4.2) are satisfied in A whenever
the xj are replaced by the gj . Sometimes, it is convenient to view a solution
xj D gj 2 A .j 2 J/ as an element .: : : ; gj ; : : : / in the direct product AJ .
For the solvability of the system (4.2), a trivial necessary condition is that it
be consistent in the sense that, if a linear combination of the left sides of some
equations vanishes (i.e., the coefficients of all the unknowns are 0), then it equals
0 2 A when the corresponding right-hand sides are substituted. Following Kertész,
we give another, more versatile interpretation of the consistency and solvability of
systems of equations.
The left members of the equations in (4.2) may be thought of as elements of
the free group F on the set fxj gj2J of unknowns. Let H denote the subgroup of F
generated by the left hand sides of the equations in (4.2). It is readily checked that
the correspondence
X
nij xj 7! ai .i 2 I/ (4.3)
j2J

induces a homomorphism  W H ! A if and only if every representation of 0 as a

linear combination of the left-hand sides is mapped by  upon 0, i.e. if the system is
consistent in the sense above. Accordingly, we call (4.2) a consistent system if (4.3)
extends to a homomorphism  W H ! A.
144 4 Divisibility and Injectivity

Clearly, two consistent systems define the same pair .H; / exactly if the
equations of either system are linear combinations of the equations of the other
system, i.e. if the two systems of equations are equivalent. Thus a consistent system
(or any of its equivalent systems) may be viewed as a pair .H; /, where H is a
subgroup of the free group F on the set of unknowns, and  is a homomorphism
H ! A.
Solvability of Systems of Equations Manifestly, xj D gj 2 A .j 2 J/ is a
solution of (4.2) if and only if the correspondence

xj 7! gj .j 2 J/ (4.4)

extends to a homomorphism  W F ! A whose restriction to H is . Moreover, the

extensions  W F ! A of  W H ! A are in a bijective correspondence with the
solutions of (4.2), so we may use the notation .F; / for a solution of (4.2).
Theorem 4.1 (Gacsályi [1]). Every consistent system of equations over A is
solvable in A if and only if A is an injective group.
Proof. The necessity is evident, since a single equation nx D a 2 A with n ¤ 0 is
a consistent system. Turning to the proof of sufficiency, let .H; / be a consistent
system of equations over a divisible group A. By Theorem 2.1,  extends to a
homomorphism  W F ! A, that is, a solution exists. t
u
Consistency being a property of finite character, we conclude at once:
Corollary 4.2 (Gacsályi [1]). A system of equations over an injective group D is
solvable in D if and only if every finite subsystem has a solution in D. u
t
It is worthwhile mentioning the following characterization of summands in terms
of solvability of equations.
Proposition 4.3 (Gacsályi [1]). A subgroup B of a group A is a direct summand
exactly if every system of equations over B that is solvable in A can also be solved
in B.
Proof. If B is a summand, say, A D B ˚ C, then the B-coordinates of a solution in
A provide a solution in B.
Conversely, assume that any system over B is solvable in B whenever it has a
solution in A. For each coset u of A mod B, select a representative a.u/ 2 A, and
consider the system

xu C xv  xuCv D a.u/ C a.v/  a.u C v/ 2 B for all u; v 2 A=B:

By hypothesis, it has a solution xu D b.u/ 2 B. Then the representatives a.u/  b.u/

of the cosets u form a subgroup C of A, and A D B ˚ C. t
u
5 Finitely Cogenerated Groups 145

F Notes. The idea of considering systems of equations over a group was suggested by Szele
whose student Gacsályi developed the theory in two papers. Generalizations to modules are due to
Kertész who published several papers on the subject, starting with [Publ. Math. Debrecen 4, 79–86
(1955)].

Exercises

(1) A system of equations over a group A is consistent if and only if it is solvable

in some group containing A.
(2) A system of equations over an injective group contains maximal solvable
subsystems.
(3) Prove that, for any prime p, the equation system

x1  px2 D 1; x2  p2 x3 D 1; : : : ; xn  pn xnC1 D 1; : : :

over Z is not solvable in Z, though each of its finite subsystems is solvable.

(4) A homogeneous system .H; 0/ over an arbitrary group A admits a non-trivial
solution in A if and only if there exists a non-zero homomorphism  W F=H ! A
(notation as above). The maps  are in a bijective correspondence with the non-
trivial solutions of the system.
(5) A homogeneous system of n equations with n C 1 unknowns over any group
A ¤ 0 always has a non-trivial solution in A.

5 Finitely Cogenerated Groups

We turn our attention to a concept dual to finite generation.

Finite Cogeneration A set C of non-zero elements in a group A is called
a set of cogenerators if, every non-zero subgroup of A contains an element of
C. Equivalently, for any group G, and for any homomorphism  W A ! G,
C \ Ker  D ¿ implies that  is monic.
Example 5.1. For a set C of cogenerators, the subgroup hCi is an essential subgroup in A, and the
set of elements in an essential subgroup with 0 omitted is always a set of cogenerators.
Example 5.2. In a cocyclic group Z.pk / .k 2 N [ 1/, a generator of its minimal subgroup Z.p/
is a singleton cogenerator of this group.

A group is finitely cogenerated if it has a finite set of cogenerators. The

following theorem is an analogue of Theorem 2.5 in Chapter 3, and points out a
beautiful duality between maximum and minimum conditions.
146 4 Divisibility and Injectivity

Theorem 5.3 (Prüfer [1], Kurosh [1], Yahya [1]). For a group A, the following
conditions are equivalent:
(i) A is finitely cogenerated;
(ii) A is an essential extension of a finite group;
(iii) A is torsion of finite rank;
(iv) A is a direct sum of a finite number of cocyclic groups;
(v) the subgroups of A satisfy the minimum condition.
Proof. (i) ) (ii) By hypothesis, A has a finite set C of cogenerators. A cannot have
elements a of infinite order, for otherwise we could select a cyclic subgroup in hai
disjoint from C. Thus the elements in C are of finite order, whence hCi is finite.
A must be an essential extension of hCi, so (ii) follows.
(ii) , (iii) is straightforward.
(ii) ) (iv) Let A be an essential extension of a finite subgroup B. It follows that
A is a torsion group with a finite number of non-zero p-components, and in order
to prove (iv), we may without loss of generality assume that A is a p-group. Write
A D D ˚ F where D is divisible and F reduced. As AŒp
D BŒp
is finite, there is
a bound pm for the heights of elements in FŒp
, whence pmC1 F D 0 follows, i.e. F
is finite. Both D and F are direct sums of cocyclic groups, their number ought to be
finite, due to the finiteness of the socle.
(iv) ) (v) Observe that if A is quasi-cyclic, then it enjoys the minimum condition
on subgroups. To complete the proof, we show that if A D U ˚ V and both of U; V
have the minimum condition of subgroups, then the same holds for A. If B1 
B2      Bn  : : : is a descending chain of subgroups in A, then in the chain
B1 \ U  B2 \ U      Bn \ U  : : : there is a minimal member, say Bm \ U.
Then the chain Bn =.Bm \ U/ D Bn =.Bn \ U/ Š .Bn C U/=U
A=U Š V for n  m
also contains a minimal member, say Bt =.Bm \ U/. Then Bt is minimal in the given
chain.
Finally, (v) ) (i). The minimum condition is inherited by subgroups, so A cannot
contain elements of infinite order, neither can the socle of A be infinite. Thus the
socle contains a finite set of cogenerators. t
u
From the equivalence of (i) and (v) we infer that factor groups of finitely
cogenerated groups are again finitely cogenerated. Observe that (ii) is equivalent
to the finiteness of the socle in a torsion group.
Countable Number of Subgroups The groups in Theorem 5.3 have but
countably many subgroups. There are only few other groups with this special
property.
Proposition 5.4 (Rychkov–Fomin [1]). A group A has fewer than continuously
many subgroups if and only if it is an extension of a finitely generated group by
a finite rank divisible subgroup of Q=Z. Then the set of subgroups is countable.
Proof. It is an easy exercise to show that finitely generated groups and finite rank
subgroups of Q=Z have but a countable number of subgroups. A proof like (iv)
) (v) above shows that this remains true for their extensions.
5 Finitely Cogenerated Groups 147

For the converse, observe that a group of infinite rank has at least 2@0 subgroups,
since different subsets of a maximal independent set generate different subgroups.
Thus, if A has less than 2@0 subgroups, then rk A is finite. A maximal independent
set in such an A generates a finitely generated subgroup F. The factor group A=F
is torsion, it must also be of finite rank, so it satisfies the minimum condition on
subgroups (Theorem 5.3); if its finite part is included in F, then A=F is divisible of
finite rank. The group Z.p1 / ˚ Z.p1 / has 2@0 subgroups, so A=F < Q=Z. t
u
The Prüfer Topology As another application of Theorem 5.3, we show:
Proposition 5.5. Let S be a finite subset in the group A, and B a subgroup of A
maximal disjoint from S. Then A=B satisfies the minimum condition for subgroups.
If jSj D 1, then A=B is cocyclic.
Proof. Every subgroup of A that contains B properly must intersect S, i.e., every
non-zero subgroup of A=B contains one of the cosets s C B with s 2 S. Hence A=B
is finitely cogenerated, and a reference to Theorem 5.3 completes the proof. t
u
Recall that the Prüfer topology of a group A is defined by declaring those
subgroups U of A as a base of open neighborhoods of 0 for which A=U satisfies
the minimum condition on subgroups. The preceding proposition is nothing else
than asserting that the Prüfer topology is always Hausdorff.
F Notes. While groups with maximum condition on subgroups are quite familiar to many
mathematicians, because they have numerous applications, groups with minimum condition are
often ignored due to their limited occurrence.

Exercises

(1) Let 0 ! A ! B ! C ! 0 be an exact sequence of groups. B satisfies the

minimum condition on subgroups if and only if so do A and C.
(2) An endomorphism  of a group with minimum condition is an automorphism if
and only if Ker  D 0.
(3) If A has minimum condition on subgroups, and if A ˚ B Š A ˚ C, then B Š C.
(4) If A has the minimum condition on subgroups and the group B satisfies A˚A Š
B ˚ B, then B Š A.
(5) If A is finitely cogenerated, then a minimal set of cogenerators is contained in
the socle of A.
(6) (Kulikov) Suppose A is a direct sum of cocyclic groups.
(a) Every summand of A is likewise a direct sum of cocyclic groups.
(b) Any two direct decompositions of A have isomorphic refinements.
(7) (de Groot) Let A and B be direct sums of cocyclic groups. If each is isomorphic
to a pure subgroup of the other, then A Š B.
148 4 Divisibility and Injectivity

(8) A group satisfies the minimum condition on fully invariant subgroups exactly if
it is a direct sum of groups Q, Z.p1 / for finitely many different primes p, and
a bounded group.

Problems to Chapter 4

PROBLEM 4.1. Call A endo-divisible if every E-homomorphism L ! A from a

principal left ideal L of E D End A extends to E ! A. Which groups are endo-
divisible?
PROBLEM 4.2. Characterize the quasi-injective hull of a p-group over its endo-
morphism ring.
Chapter 5
Purity and Basic Subgroups

Abstract In this chapter, we are going to discuss a basic concept: pure subgroup. This concept
has been one of the most fertile notions in the theory since its inception in a paper by the
pioneer H. Prüfer. The relevance of purity in abelian group theory, and later in module theory, has
tremendously grown with time. While abelian groups have been major motivation for a number of
theorems in category theory, purity has served as a prototype for relative homological algebra, and
has played a significant role in model theory as well.
Pure subgroups, and their localized version: p-pure subgroups, are often used as a weakened
notion of summands. In contrast to summands, most groups admit a sufficient supply of pure
subgroups: every infinite set of elements embeds in a pure subgroup of the same cardinality. They
are instrumental in several results that furnish us with criteria for a summand.
Every group contains, for every prime p, a p-pure subgroup, called p-basic subgroup, that is (if
not zero) a direct sum of infinite cyclic groups and cyclic p-groups. Basic subgroups are unique up
to isomorphism, and store relevant information about the containing group. Basic subgroups were
introduced by Kulikov for p-groups, and occupy a center stage in the theory of these groups.

1 Purity

Pure Subgroups A subgroup G of a group A is called pure if the equation

nx D g 2 G .n 2 N/ is solvable for x in G whenever it is solvable in A. This
amounts to saying that G is pure in A if, for any g 2 G, njg in A implies njg in G.
As njg in G means g 2 nG, we see that G is pure in A if and only if

nG D G \ nA for every n 2 N: (5.1)

Thus purity means that the divisibility properties of the elements in G by integers
are the same whether computed in A or in G.
If we equip A and a pure subgroup G with their Z-adic topologies, then (5.1)
implies that the topology of G inherited from A is equal to its own Z-adic topology
(but the converse fails).
We will often need the concept of p-purity for a prime p. A subgroup G of A is
p-pure if

pk G D G \ pk A for every k 2 N; (5.2)

© Springer International Publishing Switzerland 2015 149

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_5
150 5 Purity and Basic Subgroups

or, in other words, the finite p-heights of elements in G are the same as (i.e., not less
than) in A. Again, it follows that the induced p-adic topology on G coincides with
its own p-adic topology.
We pause for a moment to clarify the connection between purity and p-purity.
Our claim is that G is pure in A if and only if it is p-pure in A for every prime p.
This is trivial one way, so assume that G is p-pure for every p. If n D pr11    prkk is
the canonical representation of n 2 N, then nG D pr11 G \    \ prkk G D .G \ pr11 A/ \
   \ .G \ prkk A/ D G \ nA. Hence G is pure in A if and only if G.p/ pure in A.p/
holds for every localization. Thus in a p-group, purity and p-purity are equivalent.
Q
Example 1.1. The direct sum A D ˚i2I Ai is always pure in the direct product AN D i2I Ai .
Example 1.2. Rational groups (i.e., subgroups of Q) and cocyclic groups contain no pure
subgroups other than 0 and themselves.

Basic Properties of Purity Next, we assemble several useful facts concerning

pure subgroups.
(A) Direct summands are pure subgroups. In particular, 0 and A itself are pure
subgroups of A.
(B) The torsion part of a mixed group and its p-components are pure subgroups.
(C) A subgroup of a divisible group is pure if and only if it is divisible .and hence
a summand/.
(D) If A=G is torsion-free, then G is pure in A. In fact, na D g 2 G .n 2 N/ with
a 2 A implies a 2 G.
(E) If A is a p-group, and if the elements of order p of a subgroup G have the same
finite heights in G as in A, then G is pure in A. We use induction on the order
to verify that if g 2 G is divisible by pk in A, then also pk jg in G. For g 2 G of
order p, this being true by hypothesis, assume that the claim holds for elements
g 2 G of orders < pn where n  2. If o.g/ D pn , and if some a 2 A satisfies
pk a D g, then by induction hypothesis there is an h 2 G such that pkC1 h D pg.
Now pk h  g is either 0 or of order p, and pk .h  a/ D pk h  g. By induction
hypothesis, pk g0 D pk h  g for some g0 2 G, whence pk .h  g0 / D g with
h  g0 2 G. Thus the height of g is not smaller in G than in A.
(F) If G is a pure subgroup of a p-group A such that GŒp
D AŒp
, then G D A.
Again, we use induction on the order to prove that every a 2 A belongs to
G. Let o.a/ D pn .n  2/, thus pn1 a 2 AŒp
D GŒp
. Owing to purity,
pn1 g D pn1 a for some g 2 G. Here o.g  a/
n  1, so by induction
g  a 2 G, so also a 2 G.
(G) In torsion-free groups, intersection of pure subgroups is again pure. As a
matter of fact, an equation nx D g has at most one solution in A; therefore,
if it is solvable in A, then its unique solution belongs to every pure subgroup
containing g.
In view of this, in a torsion-free group A, for every subset S  A, there
exists a minimal pure subgroup containing S. This is the intersection of all
pure subgroups containing S; our notation for this subgroup is hSi . It may be
1 Purity 151

called the pure subgroup generated by S. It is easy to check that hSi =hSi is
precisely the torsion subgroup in A=hSi.
(H) Purity is an inductive property: the union of a chain of pure subgroups is pure.
For, if G is the union of a chain G1
  
G
: : : of pure subgroups, and
if nx D g 2 G is solvable in A, then it is solvable in G for every index  with
g 2 G . It is a fortiori solvable in G.
The following theorem is of utmost importance, it lists the most frequently
needed properties of purity.
Theorem 1.3 (Prüfer [2]). Let B; C be subgroups of the group A such that C

B
A. We then have:
(i) if C is pure in A, then it is pure in B;
(ii) if C is pure in B and B is pure in A, then C is pure in A;
(iii) if B is pure in A, then B=C is pure in A=C;
(iv) if C is pure in A and B=C is pure in A=C, then B is pure in A.
Proof.
(i) is obvious.
(ii) Under the stated hypotheses, nC D C \ nB D C \ .B \ nA/ D .C \ B/ \ nA D
C \ nA for every n 2 N.
(iii) follows from the equalities n.B=C/ D .nB C C/=C D Œ.B \ nA/ C C
=C D
ŒB \ .nA C C/
=C D B=C \ n.A=C/ (we used the modular law).
(iv) Assuming the stated hypotheses, let na D b 2 B for some a 2 A and n 2 N.
Then n.a C C/ D b C C whence hypothesis implies that there is a b0 2 B such
that n.b0 C C/ D b C C, i.e. nb0 D b C c for a suitable c 2 C. In view of the
purity of C, from n.b0  a/ D c we get a c0 2 C satisfying nc0 D c. It only
remains to check that b0  c0 2 B and n.b0  c0 / D b. t
u
Thus (iii) and (iv) combined claim that the natural correspondence between
subgroups of A=C and subgroups of A containing the pure subgroup C preserves
purity.
Lemma 1.4. Let B be a pure subgroup of A. If B is torsion-free and A=B is torsion,
then B is a summand of A.
Proof. If T denotes the torsion subgroup of A, then B ˚ T is an essential subgroup
in A. We claim that it is all of A. For every a 2 A n T there is an integer n > 0 such
that na D b C t with b 2 B; t 2 T. For some m > 0 we have mt D 0, so mna D mb.
By purity, some b0 2 B satisfies mna D mnb0 . Then a  b0 2 T and a 2 B C T. u t
Embedding in Pure Subgroup A fundamental property of purity is stated in
the following theorem.
Theorem 1.5 (Szele). Every finite subgroup can be embedded in a countable pure
subgroup, and every subgroup of infinite cardinality in a pure subgroup of the same
power.
152 5 Purity and Basic Subgroups

Proof. Let B be a subgroup of A of cardinality . Consider all equations nx D b for

all n 2 N and all b 2 B that are solvable in x 2 A. For each such equation, we adjoin
a solution an;b 2 A to B in order to obtain a subgroup B1
A in which all these
equations are solvable; thus, B1 D hB; an;b .8 n; b/i. We repeat this precess with B1
in place of B to get a subgroup B2 in which all equations with right members in
B1 are solvable whenever they admit a solution in A. Thus proceeding, we form the
union G of the chain B
B1
  
Bm
: : : .m < !/. This G must be pure in A,
since every equation nx D g 2 G that is solvable in A is solvable in BmC1 if g 2 Bm .
As far as the size of G is concerned, it is clear that jGj
@0 , whence both parts of
our claim follow. t
u
For torsion-free groups, (G) above gives a more powerful statement: there is a
minimal pure subgroup containing a given subgroup—a pure subgroup of the same
rank. However, in torsion and mixed groups, it is unpredictable if there is a minimal
pure subgroup containing a subgroup.
Example 1.6. Let A D ˚n<! han i where o.an / D p2n , and consider the subgroup L D
˚n<! ha2n C pa2nC1 i. Suppose there is a minimal pure subgroup C containing L. Clearly,
p4nC1 a2nC1 2 LŒp
, thus p4nC1 c2nC1 D p4nC1 a2nC1 for some c2nC1 2 C, so we have
a2n Cp.a2nC1 c2nC1 / 2 C. Hill–Megibben [1] prove that any minimal pure subgroup C containing
a subgroup L must satisfy pm CŒp
 LŒp
for some m 2 N. Therefore, if 4n  1 > m, then
p4n1 a2n C p4n .a2nC1  c2nC1 / 2 pm CŒp
 LŒp
. It follows that p4n1 a2n 2 LŒp
which is clearly
impossible. Thus no such C exists.

A subgroup C of A is purifiable if there exists a pure subgroup G

A containing
C such that A has no pure subgroup containing C and properly contained in G. There
is no satisfactory characterization of purifiable subgroups. We state a relevant result
without proof.
Theorem 1.7 (Hill–Megibben [1]). A p-group A has the property that all of its
subgroups are purifiable if and only if A D B ˚ D, where B is a bounded and D is a
divisible group. t
u
It is useful to keep in mind that groups admit pure @0 -filtrations:
Proposition 1.8. For every pure subgroup G of A, there is a smooth chain G D
G0 < G1 <    < G <    < G D A of pure subgroups for some ordinal  such
that each G C1 =G is countable.
Proof. To obtain G C1 from G , choose G C1 =G as a countable pure subgroup of
A=G (Theorem 1.5). t
u
A word on purity in p-adic modules. Since p-adic modules are q-divisible for
every prime q ¤ p, purity is the same as p-purity. (p-purity in the p-adic sense is the
same as in groups.)
Neat Subgroups There exist several concepts related to purity that deserve
mentioning, besides the isotype subgroups that will be discussed in Chapters 11
and 15.
1 Purity 153

Honda [1] introduced a widely used generalization of purity. He called a

subgroup N of A a neat subgroup if, for every prime p, the solvability of the equation
px D b 2 N in A implies that it is also solvable in N. Equivalently, if

pN D N \ pA for every prime p:

In torsion-free groups neatness does not offer anything new: it coincides with purity
(Exercise 12). However, for torsion groups the situation is different.
Example 1.9. Here is a simple example of a neat subgroup that is not pure. Let A D hui ˚ hvi
where o.u/ D p3 ; o.v/ D p. The subgroup N D hpu C vi satisfies pN D N \ pA, however,
0 D p2 N < N \ p2 A D hp2 ui:

The importance of neatness lies in the following basic fact.

Lemma 1.10 (Honda [3]). High subgroups are neat.
Proof. Let B be a C-high subgroup of A, and pa D b 2 B .a 2 A/. If a … B,
then hB; ai \ C ¤ 0, so there is a non-zero c 2 C with b0 C ka D c for some
0 < k < p and b0 2 B. Then both pa and ka belong to B ˚ C, thus a D b0 C c0 for
some b0 2 B; c0 2 C. Hence b D pa D pb0 C pc0 along with B \ C D 0 implies
pa D pb0 , so pb0 D b. t
u
-Purity We mention quickly another generalization of purity that involves
higher cardinalities. Let  be an infinite cardinal. A subgroup B of A is called -pure
if B is a summand in every subgroup C of A that contains B such that jC=Bj < .
Using Theorem 2.12 below, it is an easy exercise to check that @0 -purity is the same
as purity.
Example 1.11. Let F be a free group with an epimorphism  W F ! Q. Then Ker  is @0 -pure,
but not @1 -pure in F.
Example 1.12. Let F be a free group, and H a subgroup such that F=H is an @1 -free, but not a
free group (e.g., the Baer–Specker group). Then H is @1 -pure in F, but it is not a summand.
Example 1.13. Let H be a pure subgroup of a †-cyclic p-group F. Assume that F=H is of
cardinality @1 and has no elements of infinite heights, but it is not †-cyclic. Then H is @1 -pure in
F, but not a summand. (Use Theorem 2.9.)
F Notes. Purity was introduced by Prüfer in his seminal paper [2] under the name
‘Servanzuntergruppe’ (serving subgroup); it played an important role already in his papers. The
early literature followed this terminology, until the term ‘pure’ was firmly established on the
initiative of Kaplansky who adopted this simpler term from a paper by J. Braconnier.
The question of finding a minimal pure subgroup containing a given subgroup attracted several
researchers. Benabdallah–Okuyama [1] introduce new invariants all of which vanish for purifiable
subgroups. Both purifiability and its generalizations have been studied extensively by Okuyama;
interested readers are advised to consult his papers [1], [2], where several new concepts were
introduced. Pierce [2] characterizes and classifies centers of purity in p-groups; these are defined
as subgroups C such that every C-high subgroup is pure.
The idea of -purity is credited to Gacsályi [1]. The search for cardinals  for which -purity
implies  C -purity is still on. Janakiraman–Rangaswamy [1] discuss strong purity: C is strongly
pure in A if for every c 2 C, there is a homomorphism  W A ! C such that .c/ D c.
Finitely generated and finite rank torsion-free strongly pure subgroups are summands. For a further
154 5 Purity and Basic Subgroups

generalization of purity, we refer to C. Walker [1], Nunke [5], B. Stenström [J. Algebra 8, 352–361
(1968)]. There are a number of papers in the Russian literature on so-called !-purity, introduced
by Mishina and Skornyakov. Honda’s neat subgroups generated a great deal of interest, even their
homological aspects were explored.

Exercises

(1) Neither the sum nor the intersection of direct summands need be a pure
subgroup.
(2) If G is pure in A, then nG is pure in nA for every n 2 N.
(3) If G is a pure subgroup in each member of a chain B0
  
Bi
: : : , then
it is pure in their union B D [i Bi .
(4) If B \ C and B C C are pure subgroups of A, then so are B and C.
(5) A pure subgroup that is essential in A cannot be a proper subgroup.
(6) If G is pure in A, then so is tA C G.
(7) Let G be a pure subgroup of A. Then:
(a) G1 D G \ A1 (first Ulm subgroups);
(b) .G C A1 /=A1 is pure in A=A1 ;
(c) G
A1 if and only if G is divisible.
(8) Let B be a pure subgroup of A, and S a subset of A such that B \ S D ¿. There
exists a pure subgroup C of A containing B and maximal with respect to the
property C \ S D ¿.
(9) A group is pure in every group containing it as a subgroup if and only if it is
divisible.
(10) (a) Describe the groups in which every subgroup is pure.
(b) Find the torsion groups in which every pure subgroup is a summand.
(11) Give an example of a non-pure subgroup such that its own Z-adic topology is
the same as the relative Z-adic topology.
(12) (Honda)
(a) In torsion-free groups, neatness coincides with purity.
(b) Neatness is an inductive property.
(c) If N is neat in A, and if either N or A=N is an elementary p-group, then N
is a summand of A.
(13) A group A has no neat subgroups other than 0 and A if and only if rk.A/ D 1.
(14) (a) Let E denote an injective hull of A. A subgroup B is neat in A if and only
if B D A \ D for a divisible subgroup D of E.
(b) Every subgroup C of A can be embedded in a neat subgroup of A which is
minimal neat containing C.
(15) (Rangaswamy) A subgroup G in A is the intersection of neat subgroups of A
if and only if AŒp
6
G whenever .A=G/Œp
¤ 0.
2 Theorems on Pure Subgroups 155

(16) (Irwin) Let H be an A1 -high subgroup of A. Prove that:

(a) H is pure in A. [Hint: choose n minimal with pn x D b 2 H solvable in A,
but not in H, and consider A1 \ hH; pn1 ai where pn a D b.]
(b) A=H is the divisible hull of .A1 ˚ H/=H Š A1 .
(17) (V = L) For every infinite cardinal @ that is not compact, there is an
example of an @ -pure subgroup that is not @ C1 -pure. [Hint: Theorem 8.9
in Chapter 3.]

2 Theorems on Pure Subgroups

The results of this section are fundamental, especially for p-groups.

The Summand Property We find important criteria under which a pure
subgroup becomes a summand.
Lemma 2.1 (Szele [4]). Suppose that B is a subgroup of A such that B is a direct
sum of cyclic groups of the same order pk . The following are equivalent:
(a) B is a pure .p-pure/ subgroup of A;
(b) B satisfies B \ pk A D 0;
(c) B is a direct summand of A.
Proof. (a) ) (b) If B is p-pure in A, then B \ pm A D pm B for every m 2 N. The
choice m D k yields pk B D 0. Thus (b) follows from (a).
(b) ) (c) Assuming (b), let C be maximal in A with respect to the properties
pk A
C and C \ B D 0. We show that A D B ˚ C. If there is an a 2 A not in B ˚ C,
then there is also one with pa 2 B ˚ C, so pa D b C c with b 2 B; c 2 C. Then
pk1 b C pk1 c D pk a 2 C implies pk1 b D 0. By the assumption on the structure
of B, there is a b0 2 B satisfying pb0 D b. The maximal choice of C guarantees that
the subgroup hC; a  b0 i must contain a non-zero b0 2 B, thus b0 D c0 C m.a  b0 /
for some c0 2 C; m 2 Z. Since B \ C D 0 and p.a  b0 / D c 2 C, we must have
.m; p/ D 1. But then both m.a  b0 / D b0  c0 and p.a  b0 / D c are in B ˚ C, so
a  b0 2 B ˚ C, and a 2 B ˚ C follows.
(c) ) (a) is trivial. t
u
The following two corollaries are used frequently.
Corollary 2.2 (Prüfer [2], Kulikov [1]). Every element of order p and of finite
p-height can be embedded in a cyclic summand of the group.
Proof. If a 2 A is of order p and of height k < 1, then let pk b D a for some b 2 A.
Then hbi is pure of order pkC1 , and Lemma 2.1 applies. t
u
Corollary 2.3 (Kulikov [1]). If a group contains non-zero elements of finite order,
then it has a cocylic direct summand.
156 5 Purity and Basic Subgroups

Proof. If the group contains a quasi-cyclic group Z.p1 / for some prime p, then this
is a direct summand. If it does not contain any quasi-cyclic subgroup, but it contains
elements of order p, then it must also contain one of finite height; cf. (E). The claim
follows from Corollary 2.2. t
u
Another corollary worthwhile recording is as follows.
Corollary 2.4 (Kulikov [1]). A directly indecomposable group is either cocyclic
or torsion-free.
Proof. By Corollary 2.3 there are no indecomposable mixed groups, and the only
indecomposable torsion groups are the cocyclic groups. t
u
Bounded Pure Subgroups We now proceed to show that certain pure sub-
groups are necessarily summands.
Theorem 2.5 (Prüfer [2], Kulikov [1]). A bounded pure subgroup is a direct
summand.
Proof. If B is a bounded group, then Lemma 2.1 allows us to decompose B D B1 ˚C
where B1 is a direct sum of cyclic groups of the same order pk , and the bound for C
is less than the bound for B. If B is pure in A, then so is B1 , and Lemma 2.1 implies
that A D B1 ˚ A1 for some A1
A. Then B D B1 ˚ C1 with C1 D B \ A1 Š C.
Here C1 is pure in A1 , and by induction, C1 is a summand of A1 , and hence B is
one of A. t
u
Corollary 2.6 (Khabbaz [2]). pn A-high subgroups .n 2 N/ of a group A are
summands.
Proof. We show that a pn A-high subgroup B is a bounded pure subgroup. Since
pn B
B \ pn A D 0, B is a bounded p-group. By induction, we prove that B \ pk A

pk B for integers k with 0

k
n. This being trivially true for k D 0, assume it is
true for some k where 0
k < n. Let b D pkC1 a ¤ 0 .b 2 B; a 2 A/. If pk a 2 B,
then by induction hypothesis pk a D pk b0 for some b0 2 B, and then b D pkC1 b0 . If
pk a … B, then hB; pk ai contains a non-zero element b0 C pk a 2 pn A, where b0 2 B
is also in pk A, so b0 2 pk B as well. Now b0 D pn a0  pk a for an a0 2 A, and
hence b D pkC1 a D pnC1 a0  pb0 ; where the term pnC1 a0 has to vanish because of
B \ pn A D 0. Thus b D pb0 2 pkC1 B: t
u
Corollary 2.7 (Erdélyi [1]). A p-subgroup embeds in a bounded summand if and
only if the heights of its elements .computed in the containing group/ are bounded.
Proof. The necessity being obvious, suppose that pm is an upper bound for the
heights in the subgroup B of A. There is a pmC1 A-high subgroup C such that B
C.
Invoke the preceding corollary to conclude that C is a summand of A. t
u
Cosets of Pure Subgroups Here is another characterization of purity.
Lemma 2.8 (Prüfer [1]). A subgroup B of A is pure if and only if every coset of A
modulo B can be represented by an element that has the same order as the coset.
2 Theorems on Pure Subgroups 157

Proof. Let B be pure in A, and a 2 A n B. If the order of the coset a C B is infinite,

then every element in the coset has infinite order. If the order of a C B is n < 1,
then na 2 B, so purity implies that nb D na for some b 2 B. Then a  b.2 a C B/
is of order
n, so of order n.
Conversely, if the stated condition holds, and if na D b 2 B for some a 2 A, then
choose b0 2 B such that o.ab0 / D o.aCB/. Then n.ab0 / D 0, and nb0 D na D b
establishes the purity of B in A. t
u
We arrive at another important result:
Theorem 2.9 (Kulikov [1]). If B is a pure subgroup of A such that A=B is a direct
sum of cyclic groups, then B is a summand of A.
Proof. In view of Lemma 2.4 in Chapter 2, it is enough to deal with the case of
cyclic A=B. By Lemma 2.8, we can select a representative a 2 A in a generating
coset a C B such that o.a/ D o.a C B/. Then hB; ai D A and B \ hai D 0, so that
A D B ˚ hai. t
u
In particular, a pure subgroup of finite index is a summand.
Theorem 2.10. For a subgroup B of a group A the following conditions are
equivalent:
(i) B is pure in A;
(ii) B=nB is a direct summand of A=nB for every n 2 N;
(iii) B is a direct summand of n1 B.
Proof. (i) , (ii) If (i) holds, then from Theorem 1.3(iii) we infer that B=nB is pure
in A=nB. Hence (ii) follows at once from Theorem 2.5. Conversely, suppose (ii),
and let na D b 2 B with a 2 A; n 2 N. There is a subgroup C
A such that
A=nB D B=nB ˚ C=nB, so a D b0 C c with b0 2 B; c 2 C. Then b D na D nb0 C nc
implies nc D b  nb0 2 C \ B D nB, whence b 2 nB follows.
(i) ,(iii) If B is pure in A, then the boundedness of n1 B=B implies (iii) because
of Theorem 5.2 in Chapter 3 and Theorem 2.9. Conversely, assume again na D
b 2 B with a 2 A; n 2 N. By hypothesis, n1 B D B ˚ C for some C
A, where
obviously nC
B \ C D 0. Since a 2 n1 B, we have a D b0 C c with b0 2 B; c 2 C.
But then nb0 D nb0 C nc D na D b, establishing purity. t
u
By making use of Theorem 2.9, we can verify the analog of Theorem 7.5 in
Chapter 3 for †-cyclic torsion groups.
Theorem 2.11. Suppose 0 D A0 < A1 <    < A < : : : . < / with an
uncountable regular cardinal  is a smooth chain of pure subgroups of a p-group A
such that
 are †-cyclic of cardinality < , and
(i) the AS
(ii) A D  < A .
158 5 Purity and Basic Subgroups

Then A is †-cyclic if and only if the set

E D f <  j 9  >  such that A =A is not †-cyclicg

is not stationary in .
Proof. See the proof of Theorem 7.5 in Chapter 3. t
u
Solvability of Finite Systems Pure subgroups B of A were defined in terms of
the solvability of single equations nx D b 2 B with one unknown. Proposition 4.3
in Chapter 4 shows that the same may fail for systems of equations with infinitely
many unknowns. However, if we restrict ourselves to equations with a finite number
of unknowns only, then we get the following result.
Theorem 2.12 (Prüfer [1]). Let

ni1 x1 C    C nim xm D bi .bi 2 B; nij 2 Z; i 2 I/

be a system of equations with a finite number of unknowns, x1 ; : : : ; xm , over a pure

subgroup B of A. If it has a solution in A, then it is solvable also in B.
Proof. Let xj D aj .j D 1; : : : ; m/ be a solution in A. hB; a1 ; : : : ; am i=B is †-cyclic,
so owing to Theorem 2.10, B is a summand of C D hB; a1 ; : : : ; am i, say, C D B˚C0 .
The B-coordinates of the aj provide a solution in B. t
u
F Notes. Theorems 2.5 and 2.9 are most essential results on pure subgroups, they are
indispensable and will be used throughout without reference to them. Generalization of purity to
modules over arbitrary (associative) rings was given by P. Cohn, using property in Theorem 2.12
with a finite number of equations. In the theory of modules over integral domains, a weaker version,
the so-called ‘relative divisibility’ turned out to be most useful; this concept is due to Warfield [Pac.
J. Math. 28, 699–719 (1969)]. This adheres to the above definition of purity with ring elements
replacing integers.

Exercises

(1) If a reduced p-group contains elements of arbitrarily large orders, then it also
has cyclic summands of arbitrarily large orders.
(2) Let A be a bounded p-group, and B a subgroup of AŒp
. A has a direct summand
C such that CŒp
D B.
(3) If G is a pure subgroup of A D B ˚ C such that G \ C is essential both in G
and in C, then A D B ˚ G.
(4) Call a group A pure-simple if it contains no pure subgroups other than 0 and
A. Prove that A is pure-simple if and only if it isomorphic to a subgroup of Q
or Z.p1 / for some prime p.
(5) A group satisfies the maximum [minimum] condition on pure subgroups if
and only if it is of finite rank.
3 Pure-Exact Sequences 159

(6) (Cutler, E. Walker) If the groups A and G satisfy nA Š nG for some n 2 N,

then there exist groups A0 and G0 such that A ˚ A0 Š G ˚ G0 with nA0 D 0 D
nG0 . [Hint: find maximal n-bounded subgroups.]
(7) If nA D ˚i2I Ci is a direct decomposition, then there exist subgroups Bi such
that A D ˚i2I Bi with nBi D Ci .
(8) If B is a pure subgroup of A, then .A=B/Œn
Š AŒn
=BŒn
for every n 2 N.
(9) (Kulikov) Suppose a 2 A is an element of smallest finite order in the coset
a C pA. Then hai is a summand of A.
(10) (Mader) Let A be an infinite p-group such that jA=pn Aj < jAj for some n 2 N.
Then A contains a pn -bounded summand of cardinality jAj.
(11) The closure C of a pure subgroup C of A (in the Z-adic topology) is pure if
and only if .A=C/1 is a divisible group.
(12) Let B ˚ C be a pure subgroup in a reduced torsion-free group A. Then the
closures B ; C in the Z-adic topology of A are still disjoint, and B ˚ C is
pure in A.
(13) (Göbel–Goldsmith) In every group ¤ 0, the set of all proper pure subgroups
contains a maximal member. [Hint: argue separately for torsion-free groups.]
(14) A group is called absolutely pure if it is a pure subgroup in any group in
which it is contained as a subgroup. Show that D is absolutely pure if and only
if it is divisible.
(15) A subgroup B is -pure in A if and only if every system of equations over B
with less than  unknowns is solvable in B whenever it admits a solution in A.
[Hint: Theorem 2.12.]
(16) A subgroup G of a group A can be embedded in an @ -pure subgroup of
cardinality
jGj@ , where  D   1 or  according as  is a successor
or a limit ordinal. [Hint: argue as in Theorem 1.5 and preceding exercise.]

3 Pure-Exact Sequences

Pure-Exactness A short exact sequence

˛ ˇ
0 ! A!B!C ! 0 (5.3)

is said to be pure-exact if Im ˛ is a pure subgroup of B. It is p-pure-exact if Im ˛

is p-pure in B.
To simplify notation, in the next theorem we shall use the same letter for
homomorphisms and for maps that they induce.
Theorem 3.1. An exact sequence (5.3) is pure-exact if and only if it satisfies one
.and hence all/ of the following equivalent conditions:
˛ ˇ
(a) 0 ! nA!nB!nC ! 0 is exact for every n 2 N;
˛ ˇ
(b) 0 ! A=nA!B=nB!C=nC ! 0 is exact for every n;
160 5 Purity and Basic Subgroups

˛ ˇ
(c) 0 ! AŒn
!BŒn
!CŒn
! 0 is exact for every n;
˛ ˇ
(d) 0 ! A=AŒn
!B=BŒn
!C=CŒn
! 0 is exact for every n.
Moreover, the sequences (b) and (c) are splitting exact.
Proof. We prove only (a) and (c), because then (b) and (d) will follow from the
3  3-lemma. That the compositions of ˛ and ˇ are 0 throughout is evident.
(a) It is clear that, for all n, ˛ is monic if and only if it is monic in (5.3). Ker ˇ is
˛A \ nB which is ˛.nA/ if and only if (5.3) is exact at B. Finally, for all n, ˇ is
epic exactly it is epic in (5.3).
(c) Again, for all n, ˛ is monic if and only if it is monic in (5.3), and ˇ is epic for
every n exactly if every element of order n is the image of an element of order
n in B. Ker ˇ is equal to ˛.nA/ if and only if (5.3) is exact at B.
The last claim follows straightforwardly by showing that (b) and (c) are pure-
exact and the groups are bounded. t
u
The next theorem characterizes pure-exact sequences in terms of their injective
and projective properties.
Theorem 3.2. An exact sequence (5.3) is pure-exact if and only if the finite cyclic
groups have the injective property relative to it if and only if the finite cyclic groups
have the projective property relative to it.
Proof. We prove the claim only for injectivity, a dual proof applies to projectivity.
Let  W A ! H be a homomorphism into a cyclic group H; without loss of generality
we may assume  is epic. The existence of a W B ! H with ˛ D  is equivalent
to the extensibility of the isomorphism A= Ker  Š H to a map B ! H, i.e. to
the fact that ˛.A= Ker / is a summand of B=˛ Ker . An appeal to Theorem 3.1
completes the proof. t
u
Direct Limits and Purity We turn our attention to the behavior of pure-exact
sequences towards direct limits. Interestingly, direct limits of pure-exact sequences
are again pure-exact.
j j
Theorem 3.3. Let A D fAi .i 2 I/I i g, B D fBi .i 2 I/I i g and C D
j
fCi .i 2 I/I i g be direct systems of groups, and let ˆ W A ! B; ‰ W B ! C
be homomorphisms between them such that, for every i 2 I, the sequence 0 !
i i
Ai !Bi !Ci ! 0 is pure-exact. Then the sequence

ˆ ‰
0 ! A !B !C ! 0 (5.4)

of direct limits is likewise pure-exact.

Proof. In view of Theorem 4.6 in Chapter 2, we need only check the purity of
ˆ .A / in B . Let nb D ˆ a for a 2 A ; b 2 B , and for some n 2 N. Then there
are ai 2 Ai ; bi 2 Bi for some i 2 I such that i ai D a; i bi D b for the canonical
maps i W Ai ! A ; i W Bi ! B . Because of the commutativity of the diagram
3 Pure-Exact Sequences 161

in Theorem 4.6 in Chapter 2, we have i nbi D ˆ i ai D i i ai , whence we get

j
i .nbi  i ai / D 0; and so i .nbi  i ai / D 0 for some index j  i. Therefore,
j j j j j
ni bi D i i ai D j i ai , and since i bi 2 Bj , i ai 2 Aj , the pure-exactness for
index j implies that j na0j D j i ai for some a0j 2 Aj . Apply j , and observe that
j

j j D ˆ j to obtain nˆ a0 D ˆ a with a0 D j a0j 2 A . t

u
Next we point out a most interesting connection of purity with splitting exact
sequences; this is another convincing evidence that purity is a very natural concept.
Theorem 3.4. A sequence is pure-exact if and only if it is the direct limit of splitting
exact sequences.
Proof. That the direct limit of splitting exact sequences is pure-exact follows at once
˛ ˇ
from the preceding theorem. To verify the converse, assume 0 ! A!B!C !
0 is a pure-exact sequence, and fCi .i 2 I/g is the family of finitely generated
j
subgroups of C; here, I is a poset where i
j means Ci
Cj . If i W Ci ! Cj is the
injection map for i
j, then we may view C as the direct limit of the direct system
j
C D fCi .i 2 I/I i g. It follows that B is the direct limit of the direct system B D
fBi D ˇ 1 Ci .i 2 I/I i g where i stands for the injection map Bi ! Bj . Finally, we
j j
j
let A D fAi D A .i 2 I/I i D 1A g. Then ˛ and ˇ induce homomorphisms A ! B
˛ ˇ
and B ! C such that all the sequences 0 ! Ai !Bi !Ci ! 0 are pure-exact;
they are actually splitting because of Theorem 2.9. It is obvious that the direct limit
of these splitting exact sequences is the exact sequence we started with. t
u
The last theorem can be applied to derive a useful corollary.
Corollary 3.5 (Yahya [1]). Suppose F is a covariant additive functor Ab ! Ab
that commutes with direct limits. Then F carries a short pure-exact sequence into a
pure-exact sequence.
Proof. In view of Theorem 3.4, a pure-exact sequence 0 ! A ! B ! C ! 0 can
be represented as the direct limit of splitting exact sequences. Applying F to these
sequences, we get a direct system of splitting exact sequences whose limit will be
the pure-exact sequence 0 ! F.A/ ! F.B/ ! F.C/ ! 0. t
u
Tensor Products and Purity We hesitate to include in this section results on
tensor product (and on Hom), because we have not as yet defined these functors.
But these results are essential for purity, and we feel it is reasonable to discuss them
here. First, we rephrase part of Theorem 3.1 in terms of functors.
Corollary 3.6. The exact sequence (5.3) is pure-exact if and only if one .and hence
both/ of the following sequences is exact for every n:
˛ ˇ
(a) 0 ! Z.n/ ˝ A!Z.n/ ˝ B!Z.n/ ˝ C ! 0;
˛ ˇ
(b) 0 ! Hom.Z.n/; A/! Hom.Z.n/; B/! Hom.Z.n/; C/ ! 0:
162 5 Purity and Basic Subgroups

Proof. The proof is immediate, all that we have to observe is that both Z.n/ ˝ G Š
G=nG and Hom.Z.n/; G/ Š GŒn
are natural isomorphisms for every group G; see
Sect. 1(B) in Chapter 8 and Example 1.2 in Chapter 7. t
u
The next corollary is an important result on purity, it is intimately related to the
foregoing discussions.
Corollary 3.7. An exact sequence (5.3) is pure-exact if and only if, for every group
G, the induced sequence

1G ˝˛ 1G ˝ˇ
0 ! G ˝ A ! G ˝ B ! G ˝ C ! 0

is exact. It is even pure-exact, if so is (5.3).

Proof. Suppose (5.3) is an exact sequence. The exactness of the tensored sequence
holds for finitely generated groups G, as is shown in Corollary 3.6(a). By The-
orem 3.3, exactness is preserved under taking direct limits. Since tensor product
commutes with direct limits, the claim also holds for direct limits of finitely
generated groups, so also for any G. The converse is a consequence of Corollary 3.6
if we choose G as finite cyclic groups. Purity follows from Corollary 3.5 at
once. t
u
F Notes. The results above show that purity has several remarkable characterizations.
Several mathematicians noticed almost simultaneously that pure-exact sequences are exactly the
direct limits of splitting exact sequences, this being true also in the module-theoretic version.
Interestingly, for the generalization of purity to modules, the results of this section continue to
hold mutatis mutandis. For integral domains, see, e.g., Fuchs–Salce, Modules over non-Noetherian
Domains (2001).
In general, pure submodules of injective left R-modules need not be injective; they are always
injective exactly if R is left noetherian. The von Neumann regular rings can be characterized as
rings over which all exact sequences are pure-exact.

Exercises

(1) Show that (5.3) is pure-exact if and only if the induced sequence

0 ! Hom.A; Z.n// ! Hom.B; Z.n// ! Hom.C; Z.n// ! 0

is exact for every n 2 N. [Hint: Hom.G; Z.n//.]

(2) If C is a pure subgroup of the p-group A, then AŒp
=CŒp
Š .A=C/Œp
.
(3) If (5.3) is a pure-exact sequence, then the sequence 0 ! tA ! tB ! tC ! 0
of torsion subgroups is exact.
(4) Suppose (5.3) is pure-exact. The sequences 0 ! A1 ! B1 ! C1 ! 0 of first
Ulm subgroups and 0 ! A=A1 ! B=B1 ! C=C1 ! 0 of first Ulm factors
need not be exact; but if one is exact, then so is the other.
4 Pure-Projectivity and Pure-Injectivity 163

(5) Let E denote a class of exact sequences.

(a) The class of groups that enjoy the injective property with respect to every
member of E is closed under taking direct products and summands.
(b) The same for ‘projective’ in place of ‘injective,’ but change ‘direct product’
to ‘direct sum.’

4 Pure-Projectivity and Pure-Injectivity

In Theorem 3.2 the pure-exact sequences were characterized by properties that the
finite cyclic groups have the projective as well as the injective property relative to
them. Our next goal is to find all groups that have the projective, resp. the injective
property relative to all pure-exact sequences.
Pure-Projective Groups A group P is called pure-projective if it enjoys the
projective property relative to the class of pure-exact sequences; i.e., if every
diagram

P
⏐
ψ ⏐φ

α β
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

with pure-exact row can be completed by a map W P ! B such that ˇ D .

Example 4.1. All cyclic groups are pure-projective: Z is because it is projective, and all finite
cyclic groups because of Theorem 3.2. Hence all †-cyclic groups are pure-projective.

In order to find all pure-projective groups, we prove a lemma (which can be

interpreted as the existence theorem on pure-projective resolutions; we say: there
are enough pure-projectives).
Lemma 4.2. Every group A can be embedded in a pure-exact sequence
˛
0 ! B ! P!A ! 0 (5.5)

where P is †-cyclic .hence B as well/.

Proof. For every a 2 A, let hca i Š hai be a cyclic group, and define P D ˚a2A hca i.
Let ˛ W P ! A act via ˛ W ca 7! a for all a 2 A. This is a well-defined epimorphism,
and if we set B D Ker ˛, then we get the exact sequence (5.5). B is pure in P, since
by construction, every coset mod B is represented by an element of the same order
(cf. Lemma 2.8). Ker ˛ D B is also †-cyclic, since it is a subgroup in a †-cyclic
group (Theorem 5.7 in Chapter 3). t
u
164 5 Purity and Basic Subgroups

Theorem 4.3 (Maranda [1]). A group is pure-projective if and only if it is

†-cyclic.
Proof. Theorem 3.2 implies that †-cyclic groups are pure-projective.
Conversely, assume that A is a pure-projective group. By Lemma 4.2, there exists
a pure-exact sequence (5.5) with †-cyclic P. By pure-projectivity, there is a map
˛ W A ! P such that ˛ D 1A . This means, A is isomorphic to a summand of P, and
hence it is †-cyclic. t
u
Pure-Injective Groups Turning to the dual concept, a group H is defined to be
pure-injective if it has the injective property relative to all pure-exact sequences. In
other words, any diagram

α β
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
⏐
⏐
φ
ψ

with pure-exact row can be completed with a map W B ! H such that ˛ D .

As we shall see, the theory of pure-injective groups can be incorporated in the theory
of algebraically compact groups, so here we restrict ourselves to a few elementary
results. For more information, we refer to Sect. 4 in Chapter 6.
Example 4.4. Injective groups are trivially pure-injective, and so are all the cocyclic groups (cf.
Theorem 3.2).
Example 4.5. The group Jp is pure-injective. This will follow from Theorem 1.2 in Chapter 6 as
Jp is a compact group.

The next lemma provides pure-injective resolutions (so there are enough pure-
injectives).
Lemma 4.6 (Łoś [1]). Every group can be embedded as a pure subgroup in a direct
product of cocyclic groups.
Proof.QLet fHi .i 2 I/g be the set of all cocyclic factor groups of the group A, and set
H D i2I Hi . The canonical maps i W A ! Hi induce a homomorphism  W A ! H
which must be an embedding, since every non-zero a 2 A is excluded from the
kernel of some i (see Proposition 5.5 in Chapter 4). To verify the purity of Im 
in H, we show that if a 2 A is such that a … pn A, then also a … pn H. Let C be
a subgroup of A maximal with respect to the properties pn A
C and a … C. Then
by Proposition 5.5 in Chapter 4 A=C is cocyclic, and since it is bounded, it must be
cyclic of order pk for some k
n. Thus A=C D Hi for some i, where i a is of height
k  1, so that a 2 pn H is impossible. t
u
4 Pure-Projectivity and Pure-Injectivity 165

Theorem 4.7.
(a) A group is pure-injective if and only if it is a summand of a direct product of
cocyclic groups.
(b) Every group embeds as a pure subgroup in a pure-injective group.
Proof.
(a) A straightforward argument shows that any injective property is preserved by
taking direct products or passing to summand, so for the ‘if’ part it is enough
to show that cocyclic groups are pure-injective. This is obvious for quasi-cyclic
groups (they are injective), and this also holds for finite cyclic groups as is clear
from Theorem 3.2.
On the other hand, if A is pure-injective, then by Lemma 4.6 A embeds as a
pure subgroup in a direct product H of cocyclic groups. By pure-injectivity, the
identity map 1A factors through H ! A, which shows that A is a summand of H.
(b) Combine Lemma 4.6 and (a). t
u
Pure-essential extensions and pure-injective hulls will be discussed in Sect. 4 in
Chapter 6.
F Notes. Pure-projectivity and pure-injectivity were studied by Maranda [1]. Over any ring
(commutative or not), the pure-projective left modules are the summands of direct sums of finitely
presented left modules. Hence it is easy to conclude that all cyclic left modules are pure-projective
if and only if the ring is left noetherian. There is an extensive literature on pure-injective modules,
they play an important role also in model theory.
Quasi-pure-projectivity and -injectivity have also been discussed in several publications. (In
the definition of the latter imitate pure-injectivity with H D B above.) See, e.g., Reid [4], and the
literature cited there. Cf. also Chekhlov [2], and the survey Chekhlov–Krylov [1].

Exercises

(1) Give an example for a pure-injective group that is not a product of cocyclic
groups (only a summand of such a group). [Hint: torsion-free.]
(2) Define p-pure-projectivity and p-pure-injectivity, and show that there are
enough p-pure-projective and p-pure-injective groups.
(3) There are enough neat-projective and neat-injective groups.
(4) If G and H are pure-injective groups such that each of them is isomorphic to a
pure subgroup of the other, then G Š H. [Hint: G Š H ˚ A; H Š G ˚ B implies
that G contains a pure subgroup Š A.@0/ ˚ B.@0 / .]
(5) (C. Walker) Let  denote a non-limit cardinal. A group is -pure-projective if
and only if it is (a summand of) a direct sum of groups of cardinalities < .
(6) A homomorphism ˛ W A ! B is called pure if both Ker ˛ is pure in A and
Im ˛ is pure in B. Groups with pure homomorphisms as morphisms form a
subcategory of Ab which has enough projectives and injectives.
166 5 Purity and Basic Subgroups

5 Basic Subgroups

p-Independence Let A be an arbitrary group, and p any prime which will be kept
fixed in this section. A set fai gi2I of non-zero elements of A is called p-independent
if, for every finite subset fa1 ; : : : ; ak g and for every positive integer r,

n1 a1 C    C nk ak 2 pr A .ni ai ¤ 0; ni 2 Z/ implies pr jni .i D 1; : : : ; k/:

Thus, by definition, p-independence is of finite character, thus every p-independent

system can be extended to a maximal one. These maximal systems are of special
interest.
(A) A p-independent system is an independent set. For, if fa1 ; : : : ; ak g is
p-independent, and if n1 a1 C    C nk ak D 0 with ni ai ¤ 0, then definition
guarantees that pr jni holds for all r > 0, which is impossible. Thus the
subgroup generated by a p-independent set fai gi2I is ˚i2I hai i.
(B) A subgroup generated by a p-independent system is p-pure. Let C be the
subgroup generated by the p-independent system fai gi2I . Assume c 2 C \pr A,
i.e. c D n1 a1 C    C nk ak 2 pr A with ni ai ¤ 0; ni 2 Z. By p-independence,
we have ni D pr mi for suitable mi 2 Z. This shows that c D pr .m1 a1 C    C
mk ak / 2 pr C.
(C) If a 2 A belongs to a p-independent system, then o.a/ is either 1 or p` for
some `  1. For, if o.a/ D m and p` is the highest power of p that divides m,
then p` a 2 pr A for every r > 0. Therefore, pr jp` for all r, unless p` a D 0.
(D) If an independent set containing only elements of infinite and p-power orders
generates a p-pure subgroup, then it is p-independent. Let fai gi2I be such an
independent set in A, and C D h : : : ; ai ; : : : i D ˚i2I hai i. If n1 a1 C    C
nk ak 2 pr A with ni ai ¤ 0, then by p-purity we have n1 a1 C    C nk ak D
pr .m1 a1 C    C mk ak / for some mi 2 Z. On account of independence, we
infer that ni ai D pr mi ai for i D 1; : : : ; k. Hence the hypothesis on the orders
of the elements ai implies pr jni .
p-Basic Subgroups By a p-basic subgroup B of A we mean a subgroup of A
that satisfies the following three conditions:
(i) B is a direct sum of cyclic p-groups and infinite cyclic groups;
(ii) B is p-pure in A;
(iii) A=B is p-divisible.
Thus B has a basis which we will call a p-basis of A.
Evidently, A is a p-basic subgroup of itself if and only if (i) holds for A, and 0 is a
p-basic subgroup of A if and only if A is p-divisible. If A is equipped with the p-adic
topology, then conditions (i)–(iii) imply that B is Hausdorff in its p-adic topology
that is the same as the p-adic topology inherited from A, and furthermore, B is dense
in A.
5 Basic Subgroups 167

Lemma 5.1. A set of group elements is a p-basis if and only if it is a maximal

p-independent system.
Proof. Suppose fai gi2I is a p-basis in the group A. Its p-independence follows from
(i)–(ii) and (D). (iii) implies that for 0 ¤ g 2 A there is a relation of the form
g C n1 a1 C    C nk ak 2 pA with ni 2 Z. This shows that adjoining g, the given set
would loose its p-independence.
Conversely, let fai gi2I denote a maximal p-independent system in A, and B the
subgroup it generates. By (A), this system is independent, so (i) follows at once
from (C). Since (B) implies (ii), only (iii) remains to be verified.
Let g 2 A n B. By maximality, there is a dependence relation n0 g C n1 a1 C    C
nk ak 2 pr A with ni ai ¤ 0 .ni 2 Z/ such that pr does not divide n0 . If o.g/ D p,
then necessarily gcdfn0; pg D 1, and g is divisible by p mod B. We induct on t to
show that if o.g/ D pt , then g is divisible by p mod B. Let ps denote the highest
power of p that divides n0 . Evidently, s < r and ni D ps mi for some mi 2 Z and
for all i. Thus ps .m0 g C m1 a1 C    C mk ak / D pr a for some a 2 A. The element
g0 D m0 g  prs a C m1 a1 C    C mk ak has order dividing ps .< pt /, so the induction
hypothesis implies that g0 is divisible by p mod B. As r  s  1, the same must be
true for m0 g, and so for g. For a g of infinite order, it suffices to point out that o.g0 /
is a power of p, so g0 , and hence g, is divisible by p mod B. t
u
Theorem 5.2 (Kulikov [2], Fuchs [11]). Every group contains p-basic subgroups,
for every prime p.
Proof. Maximal p-independent systems exist in every group. By Lemma 5.1, each
of them generates a p-basic subgroup. t
u
Example 5.3. In a divisible group, 0 is the only p-basic subgroup, for every p.
Example 5.4. The cyclic subgroup generated by 1 in Jp is a p-basic subgroup of Jp . For primes
q ¤ p, the q-basic subgroups are 0.
Example 5.5. Let A D ˚1 nD1 han i where han i Š Z.p /. Then A is a p-basic subgroup of itself.
n

But A is not its only p-basic subgroup; e.g., B D ˚1nD1 han  panC1 i is another p-basic subgroup,
properly contained in A.
Example 5.6 (Kulikov [2]). Let B D ˚1 nD1 Bn where Bn is a direct sum of cyclic groups Š Z.pn /.
Q1
Then B is a p-basic subgroup of the torsion part B of the direct product A D nD1 Bn . Observe
that B consists of all vectors c D .b1 ; : : : ; bn ; : : : / with bn 2 Bn such that there is an integer k
with pk bn D 0 for every n, while B contains only the vectors with almost all bn D 0. Conditions
(i) and (ii) in the definition of basic subgroups are clearly satisfied. The same is true for (iii), as
c  .b1 C    C bk / D .0; : : : ; 0; bkC1 ; bkC2 ; : : : / must be divisible by p in view of pk bn D 0 and
o.bn / D pn .
Example 5.7. Consider the Prüfer group H!C1 . This group is generated by the elements
an .n < !/ subject to the defining relations pa0 D 0; pn an D a0 .n
1/. B D ˚n2N han  panC1 i
is a basic subgroup, and H!C1 =B Š Z.p1 /:
Example 5.8. In the Baer–Specker group A D Z@0 , the subgroup B of bounded vectors is a free
pure subgroup (Theorem 10.4 in Chapter 3). Moreover, A=B is divisible, because for every a 2 A,
for each prime p there is a b 2 B such that pja  b. In fact, such a b can be chosen to have all
coordinates from the set f0; 1; : : : ; p  1g. Consequently, B is a p-basic subgroup of A for every
prime p.
168 5 Purity and Basic Subgroups

Let B be a p-basic subgroup of A. We collect the cyclic summands of the same

order in a direct decomposition of B, and form their direct sums to obtain

B D B0 ˚ B1 ˚    ˚ Bn ˚ : : : (5.6)

where B0 denotes the direct sum of the infinite cyclic summands, while Bn for n  1
is the direct sum of the cyclic summands of order pn . Clearly, B1 ˚    ˚ Bn is a
summand in B, and hence it is pure in A, so it is a bounded summand of A, for
every n:

A D B1 ˚    ˚ Bn ˚ An :

We are going to show that An can be chosen as

An D Bn C pn A where Bn D B0 ˚ BnC1 ˚ BnC2 ˚ : : : (5.7)

so that An D BnC1 ˚ AnC1 for all n 2 N. In fact, (iii) above guarantees that, for any
given n, each a 2 A is of the form a D b C pn c .b 2 B; c 2 A/, thus B1 ; : : : ; Bn ; Bn
and pn A together generate A. If a 2 Bn C pn A is also contained in B1 ˚   ˚ Bn , then
from a D b C pn c with b 2 Bn ; c 2 A we conclude that pn c 2 B. But B1 ˚    ˚ Bn
is pn -bounded, thus pn c 2 Bn . Consequently, a 2 Bn \ .B1 ˚    ˚ Bn / D 0, and An
can be chosen as stated. Thus An D BnC1 ˚ AnC1 , in fact.
Theorem 5.9 (Baer [1], Boyer [1]). For a subgroup B in (5.6) of A to be a p-basic
subgroup it is necessary and sufficient that
(a) B0 is pure in A; and
(b) for every n < !, A D B1 ˚    ˚ Bn ˚ .Bn C pn A/ with Bn in (5.7).
Proof. It remains to verify sufficiency. Let B D ˚n<! Bn satisfy (a)-(b). Then B is
obviously †-cyclic and p-pure in A. To prove that A=B is p-divisible, write a 2 A
in the form a D b1 C    C bn C c C pk g with bi 2 Bi ; c 2 Bk ; g 2 A. Then
a C B D pk g C B D pk .g C B/, whence pk ja C B. Thus A=B is p-divisible. t
u
p-Basic in Torsion-Free Groups Let us have a closer look at two important
special cases: when A is torsion-free and when A is a p-group.
If A is torsion-free, then its p-basic subgroups are free groups. Even if A is not
free, it may very well happen that a free subgroup B is p-basic for every prime p.
This is the case when A=B is torsion-free divisible. An example is the Baer–Specker
group; see Example 5.8.
By the way, it is rather easy to construct a p-basic subgroup B in a torsion-free
group A. Select a basis ai C pA .ai 2 A; i 2 I/ in the vector space A=pA. We
claim that B D ˚i2I hai i is a p-basic subgroup of A. To see that the ai are p-pure-
independent, suppose that n1 a1 C    C nk ak 2 pr A for some r > 0 where ni 2 Z.
Since divisibility by p is unique in A, we may assume that not all of the ni are
divisible by p. Then n1 a1 C    C nk ak 2 pA, and omitting the terms ni ai with pjni
we would get a non-trivial dependence relation for the ai C pA. The maximality
5 Basic Subgroups 169

of the p-pure-independent set fai gi2I follows from the isomorphism B=pB Š A=pA
induced by the embedding B ! A (cp. Sect. 6(B)).
Basic Subgroups in p-Groups p-basic subgroups play a significantly more
important role in torsion groups than in torsion-free or mixed groups. Of course,
in the torsion case, the focus is on p-groups. If A is a p-group, then its only non-
trivial q-basic subgroup for a prime q is when q D p. In this case, we simply say
‘basic subgroup’ to mean ‘p-basic subgroup.’
In order to better understand this concept, we mention here two different ways
of obtaining basic subgroups in p-groups. One is based on using maximal bounded
summands, and the other relies on the socle.
Proposition 5.10 (Szele [5]). A †-cyclic subgroup B D ˚1 nD1 Bn of a p-group A,
where Bn is a direct sum of cyclic groups of order pn , is a basic subgroup if and only
if, for every n > 0, B1 ˚    ˚ Bn is a maximal pn -bounded summand of A, i.e. a
pn A-high subgroup of A.
Proof. If B is basic in A, then (using the above notation) An D Bn C pn A contains no
elements of order p and of height < n, hence An has no cyclic summands of order

pn , i.e. B1 ˚    ˚ Bn is a maximal pn -bounded summand of A. Conversely, if B

satisfies the stated condition, then (i) and (ii) in the definition of p-basic subgroups
are obvious. If A=B were not divisible, then by Corollary 2.3 it would have a cyclic
summand C=B D hc C Bi Š Z.pm / for some m > 0. By Theorem 2.9, C D B ˚ hci,
and clearly, C is pure in A. But then B1 ˚    ˚ Bm ˚ hci would be a larger pm -
bounded summand of A. The last claim is evident in view of Corollary 2.6. t
u
Proposition 5.11 (Charles [2]). Let B D ˚1 nD1 Bn be a subgroup of a p-group
A, where Bn denotes a direct sum of cyclic groups of order pn . A necessary and
sufficient condition for B to be a basic subgroup of A is that we have

pn AŒp
D BnC1 Œp
˚ pnC1 AŒp
for every n 2 N: (5.8)

Proof. Necessity is a consequence of the equality An D BnC1 ˚ AnC1 (notation as

above) once we observe that An Œp
D pn AŒp
; this equality follows by examining the
socles of the Bi and Bn . To prove the converse, suppose (5.8) for all n 2 N. Then
AŒp
D B1 Œp
˚    ˚ BnC1 Œp
˚ pnC1 AŒp
, and it is clear that a pnC1 -bounded pure
subgroup larger than B1 ˚    ˚ BnC1 would intersect pnC1 AŒp
, so the claim follows
from Proposition 5.10. t
u
A basic subgroup isomorphic to B Š ˚1
nD1 Z.p / is often called the standard
n

basic subgroup.
Subgroups of Basic Subgroups The next theorem characterizes the subgroups
of basic subgroups. The analogy with Kulikov’s Theorem 5.1 in Chapter 3 is
apparent.
170 5 Purity and Basic Subgroups

Theorem 5.12 (Kovács [1]). A subgroup C of a p-group A is contained in a basic

subgroup of A if and only if C is the union of an ascending chain

0 D C0
C1
  
Cn
: : :

of subgroups such that the heights of elements in Cn .computed in A/ are bounded

for every n > 0.
Proof. If the subgroup C is contained in a basic subgroup B D ˚1 nD1 Bn , where
Bn Š ˚ Z.pn /, then the subgroups Cn D C \ .B1 ˚    ˚ Bn / satisfy the stated
condition.
To prove sufficiency, we may assume without loss of generality that Cn \pn A D 0
(we may adjoin or delete members in the chain). Consider ascending chains 0 D
G0
G1
  
Gn
: : : of subgroups of A subject to the following conditions:

Cn
Gn and Gn \ pn A D 0 8 n 2 N:

We introduce a partial order in the set of all such chains in the obvious manner: a
chain of the Gn is
than the chain of the G0n if Gn
G0n for each n 2 N. Then Zorn’s
lemma ensures the existence of a maximal chain which we will denote (without
danger of confusion) again by Gn .
We claim that the union B D [1 nD1 Gn for a maximal chain is a basic subgroup
of A. The main step in the proof is to show that Gn is pn A-high in A. If we
show this, then we can finish the proof quickly by referring to Khabbaz’s theorem
(Corollary 2.6) that Gn is a summand of A, clearly a maximal pn -bounded one, so
Szele’s theorem (Proposition 5.10) applies.
By way of contradiction, suppose that there is an a 2 A such that, for some
n > 0, a … Gn ; pa 2 Gn satisfies hGn ; ai \ pn A D 0: By the maximal choice of the
chain, there is m  n C 1 with hGm ; ai \ pm A ¤ 0; and it is safe to assume that
m D n C 1. Thus, gnC1 C a D pnC1 c ¤ 0 for some gnC1 2 GnC1 ; c 2 A. Hence
pgnC1 C pa D pnC2 c 2 GnC1 \ pnC1 A D 0 implies pgnC1 D pa 2 Gn . Now
gnC1 … Gn , for otherwise gnC1 C a D pnC1 c 2 hGn ; ai \ pn A D 0, a contradiction.
Then Gn being maximal in GnC1 with respect to being disjoint from pn A, we argue
that there is a gn 2 Gn such that gnC1 C gn D pn d ¤ 0 for some d 2 A. But then
agn D pnC1 cpn d 2 hGn ; ai\pn A D 0 implies a D gn 2 Gn , a contradiction. u t
Q
Corollary 5.13. Let A be the torsion part of the direct product i2I Ai of p-groups
Ai . Suppose Bi D ˚1 nD1 Bin is a basic subgroup of Ai where Bin Š ˚ Z.p / for each
n

n 2 N, and for each i 2 I. Then

M1 Y
BD Bn with Bn D Bin
nD1 i2I

is a basic subgroup of A.
Proof. For every n > 0, we have Ai D Bi1 ˚    ˚ Bin ˚ Ain where the last summand
has no cyclic summand of order
pn . Hence we obtain A D B1 ˚  ˚Bn ˚An with
5 Basic Subgroups 171

Q
An as the torsion part of i2I Ain . It is clear that An cannot have any cyclic summand
of order
pn . An appeal to Proposition 5.10 concludes the proof. t
u
Quasi-Basis The existence of basic subgroups makes it possible to find a simple
generating set in p-groups (that will be used later on).
Let B denote a basic subgroup in a p-group A. We write

B D ˚i2I hai i and A=B D ˚j2J Cj where Cj Š Z.p1 /:

If Cj is generated by the cosets cjn .n 2 N/ such that pcj1 D 0; pcjnC1 D cjn .n 2 N/,
then by the purity of B in A, we can pick a representative cjn 2 A of each cjn such
that o.cjn / D o.cjn /. The cjn satisfy the relations

pcj1 D 0; pcjnC1 D cjn  bjn .n 2 N/

where bjn 2 B is necessarily of order

pn . We now consider the set fai .i 2 I/,
cjn .j 2 J; n 2 N/g, called a quasi-basis of A. It satisfies:
Proposition 5.14 (Fuchs [2]). If fai ; cjn g is a quasi-basis of the p-group A, then
every x 2 A can be written in the form

x D s1 ai1 C    C sk aik C t1 cj1 n1 C    C tm cjm nm

with distinct indices, where si ; tj 2 Z and every tj is prime to p. This form is unique
for the given quasi-basis in the sense that the terms sai and tcjn are determined by x.
Proof. Given x 2 A, we first write the coset x D x C B as

x D t1 cj1 n1 C    C tm cjm nm with gcdftj ; pg D 1:

Then x.t1 cj1 n1 C  Ctm cjm nm / 2 B is equal to a linear combination s1 ai1 C  Csk aik .
That all this is done with uniquely determined terms is obvious. t
u
A frequently quoted corollary is the following (cp. also Theorem 3.2 in
Chapter 10).
Corollary 5.15 (Kulikov [3]). A basic subgroup B of a reduced p-group A satisfies

jAj
jBj@0 :

Proof. Using the notation we developed for quasi-basis, we observe that if for
indices j ¤ k, the equality bjn D bkn were true for all n 2 N, then the differences
cjn ckn .n 2 N/ would generate a divisible subgroup in A. Therefore, the cardinality
of a reduced A cannot exceed the cardinality of the set of sequences fbjn gn2N which
is jBj@0 : t
u
172 5 Purity and Basic Subgroups

F Notes. Prüfer [1] proves that every countable p-group contains a subgroup satisfying
conditions (i)–(iii) listed for basic subgroups. Also, Baer used a basic subgroup in a very special
situation, but failed to recognize its potential. It was Kulikov [2] who developed the theory of basic
subgroups for p-groups of arbitrary cardinalities. Szele should be given the credit for recognizing
their utmost importance and popularizing them. It is hard to think of the theory of p-groups without
basic subgroups.
The material of this section is based on two papers: the pioneering article by Kulikov [2] where
basic subgroups of arbitrary p-groups were introduced, and a wealth of details about them was
published, and a paper by the author Fuchs [11] where the p-basic subgroups were defined for
every group. p-basic subgroups turn out to enjoy several properties of basic subgroups in p-groups
without any need for localizing the groups. A major difference is that Szele’s Theorem 6.10 fails in
general. Kulikov also developed a theory of basic submodules over the localization Z.p/ of Z at a
prime p and over Jp (which actually covers all discrete valuation domains). Further generalization
(to arbitrary valuation domains) we refer to Fuchs–Salce, Modules over non-Noetherian Domains
(Amer. Math. Soc., 2001).
Global basic subgroups (without reference to any prime) in torsion-free groups were investi-
gated in several papers by Blass–Irwin. B is basic in the torsion-free group A if it is a free pure
subgroup, and A=B is divisible. Of course, it is too much to demand from a torsion-free group
to contain such a subgroup, but if it does contain one, then it may contain many, all of them
isomorphic. Dugas–Irwin [2] prove that a strongly @1 -free group of cardinality @1 has such a basic
subgroup.
Mutzbauer–Toubassi [1] construct a quasi-basis that has additional properties.

Exercises

(1) If Bi is a p-basic subgroup in Ai for i 2 I, then ˚i Bi is p-basic in ˚i Ai .

(2) If B is p-basic in A, then for every integer m, mB is p-basic in mA.
(3) A subgroup C of A is p-pure in A if and only if a p-basic subgroup of C is
p-pure in A.
(4) Let Cn .n < !/ be pure cyclic subgroups of different orders in the p-group A.
The subgroup they generate is pure in A and is their direct sum.
(5) Let A be a reduced p-group.
(a) pn AŒp
is an essential subgroup of An D Bn C pn A in (5.6).
(b) The subgroup An is an absolute direct summand in A.
(6) (a) Let A be a p-group without elements of infinite height. A countable
subgroup of A can be embedded in a basic subgroup of A.
(b) This fails in general for uncountable subgroups.
(7) (E. Walker) Generalize Corollary 5.15 as follows: If A is a reduced group, and
if B is dense in A, then jAj
jBj@0 .
(8) (Irwin) If A is a p-group with elements of infinite height, then every A1 -high
subgroup contains a basic subgroup of A.
(9) (Kaplansky, Bourbaki) Call fa1 ; : : : ; ak g pure-independent if mx D n1 a1 C
   C nk ak with x 2 A; m; ni 2 Z implies mjni ai for i D 1; : : : ; k. An infinite
set is pure-independent if every finite subset has this property.
6 Theorems on p-Basic Subgroups 173

(a) A set is pure-independent if and only if it is independent and generates a

pure subgroup.
(b) A pure-independent set is contained in a maximal one.
(c) A pure-independent set S is maximal if and only if A=hSi is divisible.
(d) In a p-group A, a maximal pure-independent set is a p-basis.
(10) Define neat-independence, and analyze its properties.

6 Theorems on p-Basic Subgroups

In this section we prove a number of useful results on p-basic subgroups. We start

with elementary properties most of which have been established by Kulikov for
p-groups; they immediately extend to the general case.
Throughout the first part of this section, A will denote an arbitrary group, and B
a p-basic subgroup of A, p any prime.
Results on Basic Subgroups
(A) A D B C pn A for every n 2 N. This is an immediate consequence of the p-
divisibility of A=B.
(B) For every integer n > 0, there is a natural isomorphism B=pn B Š A=pn A. By
(A) and the first isomorphism theorem, we have A=pn A D .B C pn A/=pn A Š
B=.B \ pn A/ D B=pn B. We have used the equality B \ pn A D pn B that holds
because of the p-purity of B in A.
(C) For every integer n  0, pn A=pn B Š A=B is a natural isomorphism. The proof
is similar to (B).
(D) Let C be a p-pure subgroup of A. A p-basic subgroup B of C is a summand of a
suitable p-basic subgroup of A. All that we have to do is to extend a p-basis of
C to one of A.
(E) Transitivity: a p-basic subgroup C of a p-basic subgroup B of A is p-basic in A.
The only non-trivial part is that A=C is p-divisible, but this follows at once from
the p-divisibility of B=C and A=B.
(F) Let A0 D A=A1 be the 0th Ulm factor of the p-group A. Then the image B0 of
a basic subgroup B of A under the canonical homomorphism  W A ! A0 is
a basic subgroup of A0 . Moreover,   B is an isomorphism B ! B0 . Since
B \ A1 D 0 is obvious,   B is an isomorphism, so B0 is †-cyclic. B pure in
A implies B pure in A (Theorem 1.3). Finally, A=B is p-divisible, since
A=B D .A=A1 /=Œ.B C A1 /=A1
Š A=.B C A1 / is an epic image of the p-
divisible group A=B.
(G) Let B be a p-basic subgroup of a group A that has no p-divisible subgroup ¤ 0.
If ;  are endomorphisms of A such that   B D   B, then  D . The
kernel of    contains B, hence Im.  / is an epic image of A=B, and as
such it is p-divisible, so 0.
174 5 Purity and Basic Subgroups

(H) If C0 <    < Cn < : : : is an ascending chain of basic subgroups of a p-group,

then the union C D [n<! Cn is also a basic subgroup. Everything is easy to
check, except that C is also †-cyclic. But this follows at once from Theorem 5.5
in Chapter 3.
(J) Our results carry over to Z.p/ - and Jp -modules. The only difference is that in the
definition of basic subgroups instead of Z, Z.p/ , resp. Jp is to be used.
Basic Subgroups are Isomorphic A most relevant question is to what extent
p-basic subgroups are unique. We will show that, though they are, in general, not
unique as subgroups, they are still all isomorphic.
First of all we point out that in general a group contains infinitely many different
p-basic subgroups.
Lemma 6.1. Let A D ˚1 nD1 han i, where either o.an / D p with k1 <    < kn <
kn

: : : or o.an / D 1 for every n. Then A contains a proper p-basic subgroup.

Proof. Define bn D an  pknC1 kn anC1 for n D 1; 2; : : : , where we can use any
increasing sequence for the kn in case the an are of infinite order. A straightforward
calculation shows that fbn gn2N is a p-independent system, and the subgroup B D
˚1nD1 hbn i they generate does not contain a1 . Since A=B is p-divisible (generated by
the cosets an C B for all n), B is p-basic in A. t
u
Thus, if A is a p-group with an unbounded basic subgroup B, then Lemma 6.1
allows us to get a proper basic subgroup B0 of B, which will be also basic in A; see
(E). In this way, we can form an infinite descending chain B > B0 > B00 > : : : of
basic subgroups. However, we have:
Theorem 6.2 (Kulikov [2], Fuchs [11]). For every prime p, all p-basic subgroups
of a group are isomorphic.
Proof. Suppose B D ˚1 nD0 Bn (where B0 Š ˚ Z; Bn Š ˚ Z.p /) is p-basic in A.
n

The number of cyclic summands in Bn .n > 0/ is equal to the number of cyclic

summands of order pn in B=pk B for every k > n. Since B=pn B Š A=pn A by (B),
this number is determined by A, i.e. is independent of the choice of the p-basic
subgroup B.
It remains to prove that also B0 is unique up to isomorphism. First we show that
pB0 D B0 \ .T C pA/ where T D tA. The inclusion
being obvious, let b0 2
B0 \ .T C pA/, thus b0 D c C pa with c 2 T; a 2 A. By changing a, we may assume
o.c/ D pr for some pr ; then pr b0 D prC1 a 2 B0 , so pr b0 D prC1 b for some b 2 B0 .
By the torsion-freeness of B0 , we obtain b0 D pb 2 B0 , proving the reverse inclusion
. We now infer B0 =pB0 D B0 =ŒB0 \ .T C pA/
Š .B0 C T C pA/=.T C pA/ D
A=.T C pA/ where we have used the inclusion relation B0 C T C pA  B C pA D A.
Consequently, the rank of B0 which is obviously equal to the rank of B0 =pB0 does
not depend on the choice of B. t
u
A consequence of this theorem is that the p-basic subgroups provide most
valuable invariants for the group. More precisely, if B D ˚1 nD0 Bn with
B0 Š ˚0 Z; Bn Š ˚n Z.pn / is a p-basic subgroup of A, then the sequence
6 Theorems on p-Basic Subgroups 175

0 ; 1 ; : : : ; n ; : : : of cardinal numbers yields a system of invariants for A. For every

prime p, A has such a collection. However, in general, the set of these invariants
(even if taken for all p) is far from being a complete system of invariants (but—as
we shall see—they completely determine the pure-injective hull of A).
If A happens to be a p-group, then 0 D 0, and only the invariants for the same
prime p are of interest. In this case, the invariant n .n  1/ is the n  1st UK-
invariant fn1 .A/ D pn1 AŒp
=pn AŒp
of A. To substantiate this claim, note that n D
dim Bn Œp
and (in the notation of Sect. 5) pn1 AŒp
D An1 Œp
D Bn Œp
˚ An Œp
D
Bn Œp
˚ pn AŒp
:
In which groups is a p-basic subgroup unique? We answer this question for p-
groups, and delegate the general problem to the exercises.
Lemma 6.3 (Kulikov [2]). A p-group has a unique basic subgroup if and only if it
either bounded or divisible.
Proof. If a p-group A is bounded, then the only basic subgroup of A is itself, while
if A is divisible, then its only basic subgroup is 0. On the other hand, if A has
an unbounded basic subgroup, then—as was pointed out above—it has an infinite
descending chain of basic submodules. The only remaining case is when A is a direct
sum of a bounded group B and a divisible group D, both non-zero. In this case, a
cyclic summand hbi of B can be replaced by hb C di with d 2 DŒp
to obtain a basic
subgroup different from B. t
u
Basic Subgroups in p-Adic Modules We wish to state without repetition of
proofs that the results in this section above carry over to p-adic modules. (As
a matter of fact, they hold for modules over discrete valuation domains.) In the
definition of basic subgroup, Z is usually replaced by Z.p/ or Jp .
Basic Subgroups in p-Groups In the rest of this section, we confine our
attention to p-groups.
Let A denote an unbounded reduced p-group. For any basic subgroup B of A,
the factor group A=B is divisible, so a direct sum of quasi-cyclic groups Z.p1 /, the
number rk.A=B/ of such summands is an invariant of A=B. However, this number is
not an invariant for A.
The set of cardinal numbers rk.A=B/, taken for all basic subgroups B of A,
contains a minimal cardinal, and we may select a basic subgroup Bu for which
rk.A=Bu / is minimal. Such a Bu will be called an upper basic subgroup of
A. If rk.A=Bu / is infinite, then Bu contains infinitely many different upper basic
subgroups.
The following is a key example (already mentioned in Example 5.7). It is the
Prüfer group of length ! C 1, the simplest reduced p-group with elements of infinite
height. Here we use it to show that rk.A=Bu / D 1 (and hence rk.A=Bu / D k for any
k 2 N) is a possibility for reduced p-groups A.
Example 6.4 (Prüfer [1]). Let A D H!C1 D ha0 ; a1 ; : : : ; an ; : : : i where the generators are subject
to the defining relations

pa0 D 0; pa1 D a0 p2 a2 D a0 ; : : : ; pn an D a0 ; : : : :
176 5 Purity and Basic Subgroups

Manifestly, o.an /  pnC1 for each n. If C Š Z.p1 / is defined as in Sect. 3 in Chapter 1, then the
correspondence an 7! cnC1 .n  !/ gives rise to an epimorphism A ! C (relations in A hold for
the images in C), showing that o.an /
pnC1 . Since a0 ¤ 0 is of infinite height, A is not †-cyclic
(but A=ha0 i is †-cyclic: the cosets of an .n
1/ form a basis). a 2 A may be written in the form
a D t0 a0 C t1 a1 C    C tn an with 0  t0 < p and 0  ti < pi .i D 1; : : : ; n/ for some n. In case
tn an ¤ 0, the correspondence a0 ; a1 ; : : : ; an1 7! 0 and an 7! c1 extends to a homomorphism
 W A ! C such that a ¤ 0; this shows that a D 0 if and only if t0 a0 D t1 a1 D    D tn an D 0.
It is now routine to show that A1 D ha0 i and fa1  pa2 ; : : : ; an  panC1 ; : : : g is a p-basis of A.
The subgroup B generated by this p-basis is an upper basic subgroup, since clearly A=B Š Z.p1 /.
Thus rk.A=B/ D 1.
The direct sum of k copies of H!C1 yields an example where rk.A=Bu / D k.

Final Rank Szele [5] defines the final rank of a p-group A as the infimum of
the cardinals rk.pn A/ with n < !:

fin rk.A/ D inf rk.pn A/:

n<!

Often it is inconvenient to work with p-groups whose ranks are larger than their
final ranks. There is a remedy:
Lemma 6.5 (Szele [5]). A reduced p-group A decomposes as A D C ˚ A0 where C
is bounded and rk A0 D fin rk A0 D fin rk A.
Proof. We use the same notations as above. If rk.pm A/ is equal to the final rank of
A, then decompose A D B1 ˚    ˚ Bm ˚ Am where C D B1 ˚    ˚ Bm is a maximal
pm -bounded summand, and pm C D 0. Then pm A D pm Am , and A D C ˚ Am is a
desired decomposition. t
u
Lower Basic Subgroups Resuming the question of factor groups A=B modulo
basic subgroups B, it is clear that there is an upper bound for the ranks rk.A=B/: they
cannot exceed rk A. Hence there is a least upper bound for the ranks rk.A=B/; this
is easy to describe. Property (C) above guarantees that rk.A=B/ D rk.pn A=pn B/

rk.pn A/, so that always rk.A=B/

fin rk.A/. Call a basic subgroup B` lower if it
satisfies rk.A=B` / D fin rk.A/. Its existence is established in:
Proposition 6.6 (Fuchs [2]). Every basic subgroup of a p-group contains a lower
basic subgroup.
Proof. Let A be a reduced p-group. If fin rk.A/ is finite, then fin rk.B/ is also finite,
in which case it must be 0 as B is †-cyclic. Then B is bounded, it is the unique basic
subgroup, so it is a lower basic.
Next assume  D fin rk.A/ is infinite. If a basic subgroup B is not lower, then
rk.A=B/ D rk.pn A=pn B/ < 
rk.pn A/ implies rk.pn B/ D rk.pn A/. This equality
means that the cardinality of the cyclic summands hai i of order > pn in B is at least
, for every integer n. If this is the case, then we may decompose B into a direct
sum of groups Cj , where each Cj is an unbounded †-cyclic group, and j runs over
6 Theorems on p-Basic Subgroups 177

an index set J of cardinality . We quote Lemma 6.1 to argue that each Cj contains
properly a basic subgroup Bj < Cj . The direct sum B0 D ˚j2J Bj is evidently a
0
basic subgroup of B, and Phence of A. This B is lower, because by construction
0 0
rk.A=B /  rk.B=B / D j2J rk.Cj =Bj /  . t
u
Another property of basic subgroups is featured in the next lemma.
Lemma 6.7. Let A be a reduced p-group, and B a basic subgroup of a subgroup G
of A. If B
p A for some ordinal , then also G
p A.
Proof. A=B is viewed as an extension of p A=B by A=p A. The latter group
is reduced, so every divisible subgroup of A=B, in particular G=B, ought to be
contained in p A=B. u
t
Szele’s Theorem The next result provides a good illustration of how isomor-
phism properties between a p-group and its basic subgroup are applied in practice.
Lemma 6.8 (Fuchs [3]). If the group G is an epimorphic image of the p-group A,
then every basic subgroup of G is an epimorphic image of every basic subgroup
of A.
Proof. The restriction of an epimorphism  W A ! G to pn A is an epimorphism
n W pn A ! pn G which induces the epic map N n W pn A=pnC1A ! pn G=pnC1 G.
Owing to (B), pn A=pnC1A Š pn B=pnC1 B and pn G=pnC1 G Š pn C=pnC1 C are natural
isomorphisms (B; C denote the basic subgroups). Take into account that pn B=pnC1 B
is a direct sum of groups of order p, where the number of summands equals the
number of summands in B that are of order  pnC1 , and the same holds for C.
The existence of an epimorphism pn B=pnC1 B ! pn C=pnC1 C implies an inequality
between the cardinalities of sets of components of order  pnC1 in B and in C,
for every n. Considering that both B and C are †-cyclic groups, this suffices to
guarantee the existence of a desired epimorphism (argument with cardinalities is
needed). t
u
Incidentally, this theorem leads us to a very interesting corollary about †-cyclic
groups that are epic images of a p-group.
Corollary 6.9 (Szele [5]). If a †-cyclic group is an epimorphic image of a p-group
A, then it is an epic image of every basic subgroup of A.
Proof. Apply Lemma 6.8 with G D C. t
u
While we are still on this theme, let us reveal another piece of information on
basic subgroups of p-groups: a most surprising feature. (We reproduce Szele’s proof,
a shorter proof can be given by using large subgroups (Sect. 2 in Chapter 10.)
Theorem 6.10 (Szele [5]). In a p-group, basic subgroups are endomorphic images.
178 5 Purity and Basic Subgroups

Proof. Lemma 6.5 above allows us to reduce the proof to the case when
fin rk A D rk A D . Let D denote the set of cyclic summands hbi i of a basic
subgroup B in an arbitrary, but fixed decomposition. We define a function f W D ! D
subject to the following conditions:
(i) f is one-to-one; and
(ii) if f W hbi i 7! hbj i and o.bi / D pki ; o.bj / D pkj , then kj  2ki .
By the choice of , for every k 2 N, D contains  summands of order  pk , so such
a function f does exist. Once we have such an f , define a surjective endomorphism
 of B as follows:
(a) bj D bi whenever f W hbi i 7! hbj i;
(b) bj D 0 if hbj i … Im f .
We now extend  to a homomorphism  W A ! B. If a 2 A is of order pk ,
then decompose A D B1 ˚    ˚ Bn ˚ An with n  2k. Write a D b C c with
b 2 B1 ˚    ˚ Bn ; c 2 An , and define a D b. This definition is independent of
the choice of n (provided n  2k), for the B` -coordinate of a in the decomposition
A D B1 ˚    ˚ B` ˚ A` is divisible by p`k and therefore it is carried into 0 by 
whenever `  2k. That  preserves sums is evident, so it is an endomorphism of A
with image B. t
u
Example 6.11. LetQBn D hbn i Š Z.pn / and B D ˚1 nD1 Bn . Define A as the torsion subgroup of
1
the direct product nD1 Bn . With the notation of the preceding proof, we let f W hbn i 7! hb2n i,
so the map  W B ! B is defined by setting .b2n / P D bn and .b2n1 / D 0 for all n. If a D
1
.a1 ; : : : ; an ; : : : / .an 2 Bn / is of order pr , then a D nD1 .an / which is actually a finite sum,
because .an / D 0 for all n > 2r.

†-Cyclic Summands Contained in Basic Subgroups We continue this section

with an interesting feature of basic subgroups. First, a preliminary lemma.
Lemma 6.12 (Cutler–Irwin [1]). Let G be a pure subgroup of final rank 
contained in the †-cyclic p-group F. Then G contains a summand of F that has
final rank .
Proof. Decompose the socle FŒp
D ˚n<! Vn where all the non-zero elements of Vn
are of height n, and set Vn D ˚i2In hvni i. For each n < !, we select a maximal set
of disjoint finite subsets Snj  fvni j i 2 In g such that .hSnj i C pnC1 F/ \ G ¤ 0.
For each Snj , pick snj 2 G which is a linear combination of elements in Snj plus an
element of greater height. If gnj 2 G satisfies pn gnj D snj , then it is clear that the set
of all such gnj for a fixed n generates a summand Gn of F contained in G. These Gn
are clearly independent, so H D ˚n<! Gn
G is a summand of F.
In order to prove that fin rk H D , observe that FŒp
D S ˚ T where S is
generated by the set union of all selected Snj , and T is generated by all vni not in S.
Evidently, T \ G D 0 by the maximal choice of the set of the Snj . Hence the final
rank of H cannot be less than . t
u
Theorem 6.13 (Cutler–Irwin [1]). If a p-group A contains a †-cyclic summand of
final rank , then every basic subgroup of A contains such a summand of A.
6 Theorems on p-Basic Subgroups 179

Proof. Assume A D F ˚ G where F is a †-cyclic p-group of final rank , and

B D ˚n2N Bn is a basic subgroup of A; here, Bn D ˚ Z.pn /. Observe that the
subgroup .B1 ˚    ˚ Bn / C G is a summand of A for each n, and therefore their
union B C G is pure in A. Hence .B C G/=G is a pure subgroup of the †-cyclic
group A=G; moreover, it must be a basic subgroup, since the density of B in A is
preserved mod G. This means that .B C G/=G Š A=G by the uniqueness of basic
subgroups up to isomorphism. By Lemma 6.12, there is a summand .H C G/=G of
A=G contained in .B C G/=G of final rank . The generators of the cyclic summands
in H may be chosen to belong to B. t
u
Starred p-Groups Not much, but still something can be said about reduced p-
groups whose basic subgroups have the same cardinality as the group; these groups
are called starred (Khabbaz [1]). It is pretty clear that this property is preserved by
arbitrary direct sums, but not necessarily by summands.
A kind of converse to Theorem 6.13 holds for starred p-groups.
Theorem 6.14 (Khabbaz [1]). Suppose that A is a starred p-group. Then A has a
†-cyclic summand of cardinality jAj.
Proof. First assume A1 D 0. The countable case being trivial (since then A itself is
†-cyclic), let  D jAj D jBj > @0 where B D ˚n Bn with Bn D ˚n Z.pn /. There
is no loss of generality in assuming that A is a subgroup of its torsion-completion B
(Sect. 3 in Chapter 10), so every a 2 AnB may be viewed as a vector .b1 ; : : : ; bn ; : : : /
with infinitely many non-zero coordinates. If n D  for some n, there is nothing to
prove: Bn is a sought-after summand.
P Thus, for the rest of the proof we may assume
that n <  for each n and n n D , i.e.  is a limit cardinal.
A moment reflection shows that there P exists a sequence n0 <    < ni < : : :
satisfying n0 <    < ni < : : : with i ni D . Let X be a set of representatives
of the cosets mod B; evidently jXj
. Let X D [i<! Xi be a partition of X such
that jXi j
ni . Each x 2 X is represented by an infinite vector with nth coordinate in
Bn , and it is clear that no harm is done if we change a finite number of coordinates.
Taking advantage of this freedom, we may assume that all the jth coordinates of
every x 2 Xi are 0 whenever j
ni . Since the i ’s are increasing, it follows that each
BniC1 has a summand B0iC1 of cardinality i that contains all the BniC1 -coordinates
of the x 2 X, so its complement Ci in BniC1 has cardinality iC1 . Then C D ˚i Ci
is a †-cyclic summand of A of cardinality , its complement is generated by the B0i
along with X and the Bn with n ¤ ni .
If A1 ¤ 0, then let  W A ! A=A1 be the canonical map. What has been proved
can be applied to .A/ with basic subgroup .B/ Š B to find a decomposition
.A/ D C0 ˚ G=A1 where C0 is †-cyclic of cardinality jAj. Then G is pure in A and
A=G Š C0 is †-cyclic, so G is a summand of A whose complement is a summand
of the desired kind. t
u
F Notes. The role played by basic subgroups in p-groups is absolutely fundamental; their
various characterizations enable us to view them from different angles. The existence of lower and
upper basic subgroups shows that it is impossible to find a standard location for basic subgroups.
Basic subgroups are crucial in the study of torsion-complete and algebraically compact groups.
180 5 Purity and Basic Subgroups

So far no relevant application has been found for Szele’s theorem which is a rather unexpected
result on basic subgroups. But it was instrumental in leading Pierce to the definition of small
homomorphisms. Keef [3] has an interesting, more general version Q of Theorem 6.10: for any
sequence of p-groups Gn .n < !/, there is an epimorphism  W t. n<! Gn / ! ˚n<! Gn .

Exercises

(1) Let B denote a p-basic subgroup of A. For every n  0, we have

(a) .A=B/Œp
Š AŒp
=BŒp
;
(b) .pn A=pn B/Œp
Š pn AŒp
=pn BŒp
;
(c) pn AŒp
D pn BŒp
C pnC1 AŒp
;
(d) pnC1 BŒp
D pn BŒp
\ pnC1 AŒp
;
(e) pn BŒp
=pnC1 BŒp
Š pn AŒp
=pnC1 AŒp
.
(2) If B is a p-basic subgroup of A, then for all integers n; k > 0, we have pn A D
pn B C pnCk A and pn B=pnCk B Š pn A=pnCk A.
(3) Let C be a p-pure subgroup of A, and fai gi2I a p-basis of C. Furthermore, let
fbj C Cgj2J be a p-basis of A=C such that o.bj / D o.bj C C/ for each j 2 J.
Then fai ; bj gi2I;j2J is a p-basis of A.
(4) Let 0 ! A ! G ! C ! 0 be a p-pure-exact sequence, and BA ; BC p-basic
subgroups of A; C. There is a p-basic subgroup BG of G such that with the
induced mappings the sequence 0 ! BA ! BG ! BC ! 0 is splitting exact.
[Hint: Exercise 3.]
(5) In a p-group A, the intersection of all basic subgroups is either A or 0.
(6) (Khabbaz–Walker, Hill) If the p-group A has infinitely many basic subgroups,
then it has jAjjBj different basic subgroups.
(7) Let  be a cardinal satisfying rk.A=Bu /

rk.A=B` / for an upper and a
lower basic subgroup Bu ; B` of a p-group A. There exists a basic subgroup B
in A with rk.A=B/ D . Q
(8) Suppose A is the torsion part of 1 nD1 han i where o.an / D p . Then every
n

basic subgroup of A is both upper and lower.

(9) A p-group A with A1 D 0 has a decomposition A D C ˚ A0 where C is
†-cyclic, and every basic subgroup of A0 is both upper and lower.
(10) Let B be a basic subgroup of the p-group A, and G fully invariant in A. Then
G \ B is basic in G.
(11) (Irwin–Keef) If the p-group A D G ˚ H has a †-cyclic summand of final rank
@0 , then either G or H has such a summand. [Hint: the direct sum of their basic
subgroups has such a summand; use isomorphic refinement.]
(12) (Nunke) Let A be a reduced p-group. A basic subgroup B of A satisfies jBj D
jAj if and only if jCj D jAj for every dense subgroup C of A.
(13) Let A be a reduced starred p-group. Every subgroup of A is an endomorphic
image of A. [Hint: Theorem 6.10.]
6 Theorems on p-Basic Subgroups 181

(14) (Szele) Every p-group of cardinality

 (infinite cardinal) is an epic image
of the p-group A if and only if 
fin rk.B/ holds for its basic subgroup B.

Problems to Chapter 5

PROBLEM 5.1. For which cardinals  is there a universal p-group for purity? We
mean a p-group U of cardinality of  such that every p-group of cardinality

embeds in U as a pure subgroup. The same question for torsion-free groups.
PROBLEM 5.2. For which ordinals  are there @ -pure subgroups that fail to be
@ C1 -pure?
The answer is affirmative for isolated ordinals  up to the first inaccessible cardinal.

PROBLEM 5.3. What are the special features of the subgroups of A that are pure
End A-modules? (Purity in the module sense à la P. Cohn.)
Cf. Turmanov [1].

PROBLEM 5.4. Describe the pure-injective hull of a group over its own endo-
morphism ring.
See Vinsonhaler–Wickless [2] for injectivity.

PROBLEM 5.5. Study the pure-projective and pure-injective dimensions over the
endomorphism ring.
Chapter 6
Algebraically Compact Groups

Abstract In the preceding chapter we have encountered groups that were summands in every
group containing them as pure subgroups: the pure-injective groups. In this chapter, we collect a
large amount of additional information about these groups. Interestingly, these are precisely the
summands of groups admitting a compact group topology, and the reduced ones are nothing else
than the groups complete in the Z-adic topology. From Sect. 4 in Chapter 5 we know that every
group can be embedded as a pure subgroup in a pure-injective (i.e., in an algebraically compact)
group, and here we show that the significance of this embedding is enhanced by the fact that
minimal embeddings exist and are unique up to isomorphism. Thus the theory of algebraically
compact groups runs, in many respects, parallel to the theory of injective groups, a fact that was
first pointed out by Maranda [1].
The theory of algebraically compact groups is quite satisfactory: these groups admit complete
characterization by cardinal invariants. We shall often meet algebraically compact groups in
subsequent discussions.
We close this chapter with the discussion of the exchange property. This is a remarkable, but
rather rare phenomenon. Groups with this property show the best behavior as summands.

1 Algebraic Compactness

A group is called algebraically compact if it is (algebraically) a summand of a

group that admits a compact group topology. Evidently, this property is inherited by
summands and direct products.
Example 1.1. As the circle group T is compact, its summand Q=Z is algebraically compact. Direct
products of copies of a finite group are compact groups, thus their summands: the bounded groups
are also algebraically compact.

Characterizations of Algebraic Compactness It is convenient to start the

discussion with the main characterizations of algebraic compactness. The
equivalence of (b) with (e) has already been observed in Theorem 4.7 in Chapter 5.
Theorem 1.2 (Balcerzyk [1], Łoś [1], Maranda [1]). The following conditions on
a group A are equivalent:
(a) A is a summand in every group which contains it as a pure subgroup;
(b) A is a summand of a direct product of cocyclic groups;
(c) A is algebraically compact;

© Springer International Publishing Switzerland 2015 183

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_6
184 6 Algebraically Compact Groups

(d) a system of equations over A is solvable in A provided that each of its finite
subsystems has a solution in A;
(e) A is pure-injective.
Proof. (a) ) (b) By Lemma 4.6 in Chapter 5, A can be embedded as a pure
subgroup in a direct product of cocyclic groups. Hence it is clear that (a)
implies (b).
(b) ) (c) Since the direct product of compact groups is compact in the product
topology, and since the property of being a summand is transitive, (c) will be
established as soon as we can show that cocyclic groups share property (c).
But this is evident, since the cyclic groups Z.pk / .k < 1/ are compact in the
discrete topology, while the quasi-cyclic groups Z.p1 / are summands of the
circle group T D R=Z (the reals mod 1) which is a well-known compact group.
(c) ) (d) Consider the system
X
nij xj D ai 2 A .i 2 I/ (6.1)
j2J

of linear equations where nij 2 Z such that, for a fixed i, almost all the
coefficients nij are 0, and suppose that every finite subsystem is solvable in A. By
hypothesis (c), for some group B, the group C D A ˚ B admits a compact group
topology. We may equally well view our system (6.1) to be over C, and finitely
solvable in C. A solution of the ith equation can be regarded as an element
.: : : ; cj ; : : : / in the cartesian power CJ such that xj D cj 2 C .j 2 J/ satisfies the
ith equation. The set of all solutions to the ith equation is thus a subset Xi of the
compact space CJ ; moreover, it is a closed subset, since it is defined in terms
of an equation. The hypothesis that every finite subsystem of (6.1) is solvable
in C amounts to the condition that every finite set of the Xi .i 2 I/ has a non-
empty intersection. By the compactness of CJ , the intersection \i2I Xi is not
vacuous. This means that the entire system (6.1) admits a solution in C. The
A-coordinates of a solution yield a solution of (6.1) in A.
(d) ) (e) Let B denote a pure subgroup of C, and  W B ! A. Let fcj gj2J be a
generating system of C mod B, and
X
nij cj D bi 2 B .nij 2 Z; i 2 I/
j2J

the list of all the relations between these cj and elements of B. We pass to the
corresponding system
X
nij xj D bi 2 A .i 2 I/ (6.2)
j2J

of equations. A finite subsystem of (6.2) contains but a finite number of

unknowns xj1 ; : : : ; xjk explicitly. By purity, B is a direct summand of B0 D
hB; cj1 ; : : : ; cjk i, B0 D B ˚ C0 (cf. Theorem 2.9 in Chapter 5), and the images
1 Algebraic Compactness 185

of the B-coordinates of cj1 ; : : : ; cjk under  yield a solution in A. Consequently,

(6.2) satisfies the hypotheses of (d), so we can infer that there is a solution
xj D aj 2 A .j 2 J/ of the entire system (6.2). The correspondence cj 7! aj
gives rise to an extension of  to a morphism C ! A, establishing (e).
(e) ) (a) Let A be a pure-injective group contained as a pure subgroup in a group
G. Owing to pure-injectivity, the identity map 1A W A ! A can be factored as
A ! G ! A. Thus A is a direct summand of G. t
u
The reduction of conditions (a), (d), and (e) to the countable case will be useful
in the sequel.
Corollary 1.3. We have (a) , (a0 ), (d) , (d0 ), and (e) , (e0 ), where
(a0 ) A is a summand in every group G in which it is a pure subgroup with G=A
countable;
(d0 ) every countable system of equations over A is solvable in A provided that each
of its finite subsystems has a solution in A;
(e0 ) A is injective with respect to pure-exact sequences 0 ! B ! C ! C=B ! 0
with countable C=B.
Proof. We give a proof for (e0 / ) (e), and leave the rest to the reader (Exercise 4).
In order to extend a map  W B ! A to C ! A if C=B is uncountable, we take a pure
@0 -filtration B=B < C1 =B <    < C =B < : : : of C=B (with subgroups C
C).
Since C is pure in C C1 of countable index, once we have a map  W C ! A, this
can be extended to the next level. As always, at limit ordinals we take the union of
maps already defined. t
u
For a reduced group, condition Theorem 1.2 (b) can be refined:
Corollary 1.4. A reduced algebraically compact group is a summand of a direct
product of cyclic p-groups. Its first Ulm subgroup is 0.
Proof. If A is reduced algebraically compact, then for some group B, A˚B D D˚E
where D is the direct product of quasi-cyclic groups, while E is a direct product of
finite cyclic groups. D is the divisible part, so fully invariant; thus, D D .A \ D/ ˚
.B\D/ where the first summand must vanish. Thus D
B and A˚.B=D/ Š E. u t
From Theorem 1.2(b) we derive
Q
Corollary 1.5. Reduced algebraically compact groups are Z-modules.
Q
Proof. Take into account that group summands of reduced Z-modules are also
Q
Z-module summands. t
u
Example 1.6. The group Jp of the p-adic integers is compact in the p-adic topology (so it is
Q Jp is the
algebraically compact). Since inverse limit of the cyclic groups Z.pk / for k D 1; 2; : : : , it
is contained in the group k<! Z.p / which is compact in the product topology. The inverse limit
k

is a closed subgroup in the product, so the compactness of Jp is immediate.

Example 1.7 (Fuchs [IAG]). The additive group of an injective module M over any ring R
is algebraically compact. Indeed, if D is the divisible hull of M as an abelian group, then M
186 6 Algebraically Compact Groups

is a submodule and hence a summand of the injective R-module HomZ .R; D/ which is by
Theorem 2.11 in Chapter 7 algebraically compact.
Example 1.8. The group Z is not algebraically compact. Perhaps the easiest way to see it is to
Q
point out that Z is not a Z-module.

For the sake of easy reference we formulate the following immediate conse-
quence of (b) or (c) in Theorem 1.2.
Corollary 1.9. A direct product of groups is algebraically compact if and only if
every component is algebraically compact. u
t
Combining Lemma 4.6 in Chapter 5 with the theorem above, we conclude:
Corollary 1.10. Every group can be embedded as a pure subgroup in an alge-
braically compact group. t
u
We will see in Theorem 4.2 below that there is a minimal such algebraically
compact group.
Algebraically Compact Factor Groups Next, we will show that algebraically
compact groups are abundant among quotient groups of direct products.
Let Gi .i 2 I/ be a family of groups with an arbitrary index set I, and let K denote
an ideal in the Boolean lattice 2I of all subsets of I. K will stand for the -ideal
generated by K, i.e. it consists of countable unions of subsets in K. (It is easy to
see that these are exactly the countable unions of pairwise disjoint subsets from K.)
Our concern is the relation between K- and K -products (for definition, see Sect. 1
in Chapter 2).
Theorem 1.11 (Fuchs [13]). If K and K are defined as above, then the factor
group
Y Y
AD Gi = Gi (6.3)
K K

is algebraically compact for any collection of groups Gi .i 2 I/.

Proof. Ignoring a trivial case, assume K ¤ K . In view of Theorem 1.2 and
Corollary 1.3, it suffices to establish a solution in A for the system
X1
njk xk D aN j 2 A .njk 2 Z; j < !/ (6.4)
kD1

with a countable number of unknowns and equations, under the Q hypothesis that
every finite subsystem
Q of (6.4) admits a solution in A. Here a j 2 K Gi , and bars
denote cosets mod K Gi .
Let xk D cN m
k 2 A be a solution of the system consisting of the first m equations,
with the understanding that the value 0 is assigned to the unknowns
Q not occurring
explicitly in these equations. The representatives aj ; cmk 2 K? Gi in the cosets
1 Algebraic Compactness 187

aN j ; cN m N
k ; respectively, are such that 0 is always chosen as a representative
S of the coset 0.
All the supports s.aj / belong to K? , and so does their union Y D j<! s.aj /.
We now move to the system
X1 Y
njk xk D aj 2 Gi .j < !/: (6.5)
kD1
K?

Plugging xk D cm P with indices j

m, the set Ym of indices i 2 I
k in the equations
k and aj differ for j D 0; 1; : : : ; m certainly
for which the ith coordinates of k njk cm
belongs to K. There is no difficulty in constructing subsets Xm .m < !/ so as to
satisfy

X0 D Y0 ; X1  Y1 n Y0 ; : : : ; Xm  Ym n .Y0 [    [ Ym1 /; : : : ;
S
subject to the condition m<! Xm D Y [ Y0 [    [ Ym [ : : : :
Now the proof can easily be completed. Set bk D .: : : ; cki ; : : : /, where cki is
?
P of ck if i 2 Xm , and 0 otherwise. Clearly, bk 2 K . As the ith
m
the ith coordinate
coordinate of k njk bk agrees with the ith coordinate of aj for all i 2 Xm .j
m/; it
follows that xk D bN k 2 A .k 2 N/ is a solution of the system (6.4). t
u
There is a special case that is particularly worth mentioning. This is when the
index set I is countable, and K is the ideal consisting of all finite subsets of I. Then
K D I, and we get a most interesting corollary.
Corollary 1.12 (Balcerzyk [2], Hulanicki [3]). For any countable family of
groups Gn .n D 1; 2; : : : / the factor group
1
Y 1
M
Gn = Gn
nD1 nD1

is algebraically compact. t
u
On the basis of this result it would be tempting to conjecture that direct
products modulo direct sums are always algebraically compact. It turns out that
Corollary 1.12 is true only for countable index sets (see Notes).
A proof similar to the one in Theorem 1.11 yields:
Theorem 1.13. If fAi j i 2 Ig is a set of Q
groups, and F is an !1 -complete ultrafilter
on the index set I, then the ultraproduct i2I Ai =F is algebraically compact. t
u
Large Products Mod Direct Sum For uncountable index sets I, the factor groups
Q
i2I Ai = ˚i2I Ai are only exceptionally algebraically compact. As observed by
Göbel–Rychkov–Wald [1], a weaker conclusion may be drawn which we prove
in the next theorem (in an equivalent form). We start with the observation that all
torsion epic images of an algebraically compact group are of the form B ˚ D, with
bounded B and divisible D. The easiest way to prove this is to refer to Corollary 3.5.
188 6 Algebraically Compact Groups

Theorem 1.14. For a non-measurableQ index set I and any groups Ai , every reduced
torsion epic image of the factor group i2I Ai = ˚i2I Ai is bounded.
Q
Proof. Let  W A D i2I Ai = ˚i2I Ai ! T; where T is a reduced torsion group.
Im  D T may be assumed. We consider  as a homomorphism of the product
that vanishes on ˚Ai . If Im  is unbounded, then the image of  followed by an
epimorphism of T onto its basic subgroup B Theorem 6.10 in Chapter 5 is still
unbounded. In this situation, in the application of Theorem 6.5 in Chapter 2 we
can ignore I0 and J0 , and conclude that there is an m 2 N such that m Im. / is
contained in the Ulm subgroup B1 D 0. Hence Im. / is bounded, and so is Im .
t
u
@1 -Algebraic Compactness It seems reasonable to consider generalizations
of algebraic compactness for higher cardinalities. @1 -algebraic compactness was
studied by Megibben [6]. He proved that this concept does not yield anything new.
However, for higher cardinalities the situation is different, though it does not seem
to be too interesting. Here we wish to discuss only the @1 case. In the proof, we need
a couple of results to be proved later, in particular, the characterization of algebraic
compactness from Proposition 5.8 in Chapter 9.
A group M is said to be @1 -algebraically compact if it has the injective property
relative to all @1 -pure-exact sequences. Equivalently, it is a summand in every group
in which it is contained as an @1 -pure subgroup. All algebraically compact groups
are evidently @1 -algebraically compact.
In the proof of the next theorem, the following simple fact is needed from set
theory. For an infiniteP cardinal  there is a cardinal  >  such that  <  @0 .D 2 /.
(For instance,  D i<! i is such a cardinal if 0 D  and iC1 D 2 for all i; see
i

note after Proposition 7.10 in Chapter 10.)

Theorem 1.15 (Megibben [6]). A group is @1 -algebraically compact if and only if
it is algebraically compact.
Proof. By Proposition 5.8 in Chapter 9, for the algebraic compactness of a group G,
it suffices to show that Ext.Q; G/ D 0 and Pext.Q=Z; G/ D 0 hold for G. We start
the proof by showing that an @1 -algebraically compact group G satisfies
(i) Ext.F; G/ D 0 for all @1 -free groups F; and
(ii) Pext.T; G/ D 0 for all separable torsion groups T.
To prove (i), observe that if G is @1 -algebraically compact, and if 0 ! G ! A !
F ! 0 is an exact sequence with @1 -free F, then the sequence has to split, being
@1 -pure-exact. The argument for (ii) is entirely similar with a pure-exact sequence.
Now suppose G is @1 -algebraically compact. Choose a cardinal  > jGj such
that  <  @0 . Let F be the Z-adic closure of the free group F0 D Z./ in the direct
product Z . F is @1 -free and has cardinality  @0 . Thus F=F0 Š ˚ @0 Q. We obtain
the induced exact sequence

G Š Hom.F0 ; G/ ! Ext.F=F0 ; G/ ! Ext.F; G/ D 0:

1 Algebraic Compactness 189

Q
Here jG j D 2 , while j Ext.F=F0 ; G/j D  @0 j Ext.Q; G/j is either 0 or is at least
@
2 0 . The latter alternative is impossible, proving Ext.Q; G/ D 0.
The proof for Pext.Q=Z; G/ D 0 is exactly the same, using embedding in torsion
completion. t
u
F Notes. The problem of describing the algebraic structure of compact abelian groups led
Kaplansky [K] to the discovery of algebraically compact groups as summands (in the algebraic
sense) of compact groups. A different line of investigation started with Łoś [1, 2], who considered
groups that were summands in every group containing them as pure subgroups. Balcerzyk [1]
noticed that the classes discussed by Kaplansky and Łoś were identical. The study of pure-
injectivity as an analogue of injectivity was initiated by Maranda [1].
Łoś [1] considers abstract ‘limits’ (not in the topological sense), called ! -limits, and shows
that a group admits such a limit if and only if it is algebraically compact.
The theory of pure-injective modules over general rings was developed by Warfield [Pac. J.
Math. 28, 699–720 (1969)] based on P. Cohn’s definition of purity. Lot of information is available
about the structure of pure-injective modules, especially over particular rings.
Numerous papers deal with factor groups of direct products modulo direct sums. Gerstner [1]

shows that ZQ =Z./ is never algebraically compact if  is uncountable. Rychkov [1] proves that the
factor group i2I Gi = ˚i2I Gi is algebraically compact if and only if, with the exception of at most
countably many indices i, all the Gi are algebraically compact. Dugas–Göbel [1] studies such factor
groups more generally, using filters of the index set. Franzen [1] discusses when the factor group
of a filter-product modulo another filter-product is algebraically Q compact. Göbel–Rychkov–Wald
[1] show that if I has non-measurable cardinality, then A D i2I Ai = ˚i2I Ai is always a Fuchs-
Q./ Q./
44-group (see Notes to Sect. 6 in Chapter 2). Wald [1] studies the quotients Ai = Ai for
cardinals  >  (only vectors with supports <  (resp. < ) are taken).

Exercises

(1) (a) A1 is the divisible part of an algebraically compact group A.

(b) A group is algebraically compact if and only if it is the direct sum of an
injective group and a reduced algebraically compact group.
(2) (Sa̧siada) A group A is algebraically compact if it is a summand in every
group G in which it is a pure subgroup such that G=A Š Q or Z.p1 /.
(3) (a) A pure subgroup C of an algebraically compact group A is algebraically
compact provided A=C is reduced.
(b) A group A need not be algebraically compact even if both the subgroup
C and the factor group A=C are algebraically compact.
(4) Complete the proof of Corollary
Q 1.3.
(5) The obvious map A ! n2N A=nA ˚ E is an embedding of A as a pure
subgroup in an algebraically compact group; here, E stands for the injective
hull of A.
(6) (a) Show that A is isomorphic to a pure subgroup of the algebraically
compact group A@0 =A.@0/ . [Hint: consider the diagonal in A@0 .]
(b) Every group A can be embedded as a pure subgroup in an algebraically
compact group of cardinality
jAj@0 .
190 6 Algebraically Compact Groups

(7) Prove that Z@0 =Z.@0 / Š .Q ˚ Z/ Q @0 .

(8) Let Gn .n D 1; 2; : : : / be reduced groups such that thereQis no integer m > 0
satisfying mGn D 0 for almost all n. The factor group n Gn = ˚n Gn is not
reduced.
(9) (Baumslag–Blackburn)
Q Let Gn .n D 1; 2; : : : / be reduced groups. ˚n Gn is a
summand of n Gn if and only if mGn D 0 holds for some m > 0 and for
almost all n. [Hint: for necessity
Q see preceding exercise.]
(10) Let H denote the subgroup of i2I Gi consisting Q of elements with countable
support. Prove that H= ˚i Gi is a summand of i Gi = ˚i Gi .
(11) (a) If A is pure-injective, then for every cardinal , the codiagonal homo-
morphism r W A./ ! A extends to a homomorphism: A ! A.
(b) If A has this extension property for some infinite cardinal  > jAj, then
it is pure-injective.
(12) A group G is @1 -algebraically compact if and only if every system of
equations over G is solvable in G whenever every countable subsystem has a
solution in G.

2 Complete Groups

In this section we examine more closely the groups that are complete in their Z-adic
or p-adic topologies. This will not be a new class: we shall see that these groups
coincide with the reduced algebraically compact groups. (Recall that according to
our agreement, only Hausdorff groups will be called complete.)
Example 2.1. The groups that are discrete in their Z-adic topologies, i.e. the bounded groups, are
evidently complete.
Example 2.2. The group Jp of p-adic integers is (reduced and compact, and hence) complete in its
p-adic topology.

To simplify terminology, we agree that in the absence of firm indication to the

contrary, complete group means a group that is complete in its Z-adic topology (in
p-groups, this is identical to the p-adic topology).
Properties of Complete Groups We start with a simple observation.
Lemma 2.3. A group A that is complete in its p-adic topology is a Jp -module, and
is complete also in the Z-adic topology. It satisfies qA D A for all primes q ¤ p.
Proof. Recall that such an A is in a natural way a Jp -module: if a 2 A and  D
s0 C s1 p C    C sn pn C    2 Jp .0
sn < p/, then

s0 a; .s0 C s1 p/a; : : : ; .s0 C s1 p C    C sn pn /a; : : :

2 Complete Groups 191

is a Cauchy sequence in A, hence it converges to a limit in A which we define as a.

qJp D Jp implies qA D A for all primes q ¤ p. As a consequence, the p-adic and
the Z-adic topologies are identical on A. t
u
In a similar fashion, we can prove that every group that is complete in the Z-adic
Q
topology is a Z-module.
Recall that groups that are Jp -modules were called p-adic groups. p-adic groups
have basic subgroups, unique up to isomorphism.
Theorem 2.4. (i) A complete p-adic group is the completion of any of its basic
subgroups.
(ii) A †-cyclic p-group is basic in its completion.
(iii) Two complete p-adic groups are isomorphic if and only if their basic sub-
groups are isomorphic.
Proof. (i) This follows from the fact that a basic subgroup is dense and the
induced topology on it is the same as its p-adic topology.
(ii) is clear from (i).
(iii) One way the claim is trivial. Conversely, a complete p-adic group is deter-
mined by a basic subgroup. t
u
Completeness and Algebraic Compactness The principal result on complete
groups is the following theorem, which is essentially due to Kaplansky [K].
Theorem 2.5. A group is complete in its Z-adic topology if and only if it is a
reduced algebraically compact group.
Proof. Assume A is reduced and algebraically compact. Owing to Corollary 1.4, A is
a summand of a direct product of cyclic groups Z.pk /. Each component is complete
in its Z-adic topology, so the same holds for the summands of their direct product
(see Lemma 7.10 in Chapter 2).
Conversely, suppose that A is complete in its Z-adic topology and is pure in the
group G. If G1 D 0, then we expand G to its completion G; Q this contains G, and
0
hence also A as a pure subgroup. A basic subgroup B of A is a summand of a basic
subgroup B D B0 ˚ B00 of G,Q whence GQ D BQ 0 ˚ BQ 00 D A ˚ BQ 00 . Thus A is a summand
of G, too. If G ¤ 0, then factoring out G1 , the image of A remains pure in G=G1 ,
1

so .A C G1 /=G1 Š A is a summand of G=G1 . As A \ G1 D 0, A is a summand of

G. Consequently, A is algebraically compact. t
u
Since the Z-adic topology on A induces a topology on a subgroup B of A that
is coarser than the Z-adic topology of B, we combine the preceding theorem with
Lemma 7.2 in Chapter 2 to derive at once:
Corollary 2.6. Let A be a reduced algebraically compact group, and B a subgroup
such that .A=B/1 D 0. Then both B and A=B are algebraically compact. t
u
Another noteworthy observation is the following.
Corollary 2.7. If  is an endomorphism of a complete group A, then both Ker 
and Im  are complete groups.
192 6 Algebraically Compact Groups

Proof. Apply the preceding corollary to B D Ker , and note that A= Ker  is
complete by virtue of Lemma 7.2 in Chapter 2. t
u
Lemma 2.8 (Kaplansky [K]). Let G be a pure subgroup of a complete group C.
Then the Z-adic closure of G in C is a summand of C. In particular, a pure closed
subgroup is a summand.
Proof. A basic subgroup B0 of G extends to a basic subgroup B D B0 ˚ B00 of C.
Then C D BQ 0 ˚ BQ 00 by Theorem 2.4, where G
BQ 0 . Thus BQ 0 is the closure of G in C.
t
u
Completions We turn our attention to the completion process. In the next result
we expand Theorem 7.7(i) in Chapter 2.
Theorem 2.9. For any group A, the inverse limit

AQ D lim.A=nA/

with the connecting maps a C knA 7! a C nA .a 2 AI n; k 2 N/ is a complete group.
The canonical map

A W a 7! .: : : ; a C nA; : : : / 2 AQ

has A1 for its kernel, and an isomorphic copy of A=A1 for its image. A .A/ is pure
Q and the factor group A=
in A, Q A .A/ is divisible.
Q
Proof. Since the factor groups A=nA are bounded, and hence complete, .A=nA/
is complete, and its pure closed subgroup (the inverse limit) is complete by
Lemma 2.8. Clearly, a D 0 amounts to a 2 nA for every n, whence Ker  D A1
and Im  Š A=A1 . If bQ D .: : : ; bn C nA; : : : / .bn 2 A/ satisfies mbQ D a for some
a 2 A and m 2 N, then mbn  a 2 nA for every n, in particular, for n D m, whence
a 2 mA, and the purity of A in AQ follows. To prove that A=AQ is divisible, we show
Q
that every b D .: : : ; bn C nA; : : : / .bn 2 A/ is divisible by every m 2 N mod A.
Since mjbm  bkm for each k 2 N, we have mjbQ  bm , in fact. t
u
An immediate consequence of the preceding theorem is the inequality

Q
jAj@0 ;
jAj (6.6)
Q
which is evident in view of AQ
n A=nA.
The group AQ is called the Z-adic completion of A. The map A W A ! AQ is
natural in the categorical sense, so the correspondence A 7! AQ is functorial. Indeed,
we have
Q
Proposition 2.10. Every homomorphism ˛ W A ! B induces a unique Z-
Q Q
homomorphism ˛Q W A ! B making the diagram
2 Complete Groups 193

α
A −−−−→ B
⏐ ⏐
μA
⏐ ⏐μB

A B

commute.
Proof. The proof is given in the comments after Lemma 7.6 in Chapter 2. t
u
We hasten to point out that the completion functor: A 7! AQ is a pure-exact
functor in the sense that it carries short pure-exact sequences into pure-exact
sequences. Moreover, a stronger result holds:
˛ ˇ
Theorem 2.11. If 0 ! A!B!C ! 0 is a pure-exact sequence, then the
induced sequence

˛Q Q̌
0 ! AQ !BQ !CQ ! 0 (6.7)

of completions is splitting exact.

Proof. If the given sequence is pure-exact, then the induced sequence 0 ! A=nA !
B=nB ! C=nC ! 0 is exact for every n; see Theorem 3.1 in Chapter 5. The
completion functor (as inverse limit) is left-exact (Theorem 5.6 in Chapter 2), so for
the exactness of (2) it suffices to show that ˇQ is a surjective map. By Lemma 7.2 in
Chapter 2, Im ˇQ is complete and contains ˇB as a dense subgroup, so necessarily
Im ˇQ D C.Q What remains to be proved is only the purity of Im ˛Q in B. Q The map
1 1 1
a C A 7! ˛a C B carries A=A onto a pure subgroup of B=B , which along with the
purity of B .B/ in BQ shows that B .˛A/ is pure in B.Q In view of B ˛ D ˛Q A and
Q
the divisibility of ˛Q A=˛ Q Q The algebraic
Q A .A/, we infer that ˛Q A must be pure in B.
compactness of AQ implies the splitting. t
u
Q a p-basic subgroup of A
Corollary 2.12. Under the canonical map A W A ! A,
Q
maps upon a p-basic subgroup of A.
Proof. Since Ker  D A1 , Sect. 6(F) in Chapter 5 implies that  maps p-basic
subgroups of A isomorphically upon p-basic subgroups of A. Owing to the
Q
divisibility of A=A Q the claim follows.
and the purity of A in A, t
u
Example 2.13. TheQp-adic (as well as the Z-adic) completion of Z.p/ is Jp , and the Z-adic
completion of Z is p Jp with p running over all primes.
Q
Example 2.14. The Z-adic completion of ˚p Z.p/ is p Z.p/.
Example 2.15. Let B D ˚n Bn where Bn Š ˚ Z.pn / with an arbitrary number of components.
The
Q p-adic (Z-adic) completion of B is C, where C=B denotes the divisible part of the factor group
. n Bn /=B.
194 6 Algebraically Compact Groups

Direct Decompositions of Complete Groups We focus our attention on infinite

direct decompositions of complete groups. We start with a lemma that has its own
interest.
Lemma 2.16. Assume that the complete group C is contained in the direct sum
A D ˚i2I Ai of groups with A1i D 0 for every i. Then there is an integer m > 0 such
that mC is contained in the direct sum of a finite number of the Ai .
Proof. If the conclusion fails, then there exist an increasing sequence of integers,
m1 <    < mj < : : : with mj jmjC1 ; and groups Hj , each being a direct sum of
finitely many Ai , such that the Hj generate their direct sum in A and

j1 j
mj C \ ˚kD1 Hk < mj C \ ˚kD1 Hk .j D 1; 2; : : : /:

Pick an element cj in the right, which is not contained in the left side. Evidently,
cj1 has 0, while cj has a non-zero coordinate in Hj . Thus the Cauchy sequence
c1 ; : : : ; cj ;    2 C cannot have a limit in A. t
u
We can now state:
Theorem 2.17. If C D ˚i2I Ci is a direct decomposition of a complete group, then
all the groups Ci are complete, and there exists an integer m > 0 such that mCi D 0
for almost all i.
Proof. The first claim is evident, while the second one is an immediate consequence
of Lemma 2.16. t
u
F Notes. The theory of complete groups is due to Kaplansky [K]; he gives credit to I.
Fleischer for the torsion-free case. That complete groups can be characterized by their basic
subgroups is a consequence of Theorem 2.4, and will also follow from the next section where
this question is settled for algebraically compact groups.
Groups complete in the Prüfer topology are extremely important: they are the linearly compact
groups. They will be discussed in the next section.

Exercises

(1) Let m > 0 be an integer. A is complete if and only if mA is complete.

(2) The direct product of groups Ci is complete if and only if each Ci is complete.
(3) (a) If A D A1 ˚    ˚ An , then AQ D AQ 1 ˚    ˚ AQ n .
(b) This fails in general for infinite direct sums.
Q
(4) p Z.p/ cannot be written as an infinite direct sum of non-zero groups.
(5) A complete torsion-free group may have decompositions into the direct sum of
any finite number of summands, but never into the direct sum of infinitely many
non-zero groups.
3 The Structure of Algebraically Compact Groups 195

(6) Let B be a p-basic subgroup of a group A that is complete in its p-adic topology.
Every homomorphism  W B ! C extends uniquely to a homomorphism   W
A ! C provided that C is complete in the p-adic topology.
(7) (a) The inverse limit of reduced algebraically compact groups is again alge-
braically compact.
(b) The same is not true in general for non-reduced algebraically compact
groups.

3 The Structure of Algebraically Compact Groups

Having got acquainted with several remarkable properties of algebraically compact

groups, most of which may even serve as a characterization, it is time to have a closer
look at their structure. The main theorem will tell us that they have a satisfactory
structure theorem: they admit a complete and independent set of cardinal invariants.
p-adic Algebraically Compact Groups Since an algebraically compact group
is a direct sum of an injective group (which can be characterized by invariants
(Theorem 3.1 in Chapter 4)) and a reduced algebraically compact group, we may
restrict our considerations to the reduced case. We state a lemma that is essentially
the same as Theorem 2.4.
Lemma 3.1. Let Ap be complete in its p-adic topology, and let Bp denote a p-basic
subgroup of Ap . Then Ap D BQ p , and the correspondence Ap $ Bp is a bijection
between groups that are complete in the p-adic topology and groups that can be
p-basic subgroups. t
u
The Structure Theorem We are prepared for the proof of the structure theorem
on algebraically compact groups.
Theorem 3.2 (Kaplansky [K]). A reduced algebraically compact group A is of the
form
Y
AD Ap ; (6.8)
p

where for each prime p, Ap is a uniquely determined p-adic algebraically compact

group. The invariants of the p-basic subgroups of A serve as a complete and
independent set of invariants of A.
Proof. Let A be reduced and algebraically compact. Then, for a suitable B, A ˚ B D
C is a direct product of cyclic p-groups (see Corollary 1.4). Collect the components
Z.pk / belonging to the same prime p, and form their direct product Cp . The Cp are
fully Q
invariant in
Q C, henceQ Cp D Ap ˚ Bp with 0Ap D A \ Cp ; Bp D B \ CpQ ; thus
C D p Cp D p Ap ˚ p Bp . Q The closure of A D ˚p Ap in C must contain p Ap
because of the divisibility of . p Ap /=A0 . A is closed in C, whence the inclusion
196 6 Algebraically Compact Groups

Q Q
p ApQ
A is Q immediate. Analogously,
Q weQhave p Bp
B, and consequently,
C D p Ap ˚ p Bp implies p Ap D A, p Bp D B. As a direct summand of a
complete group, Ap must be complete in its Z-adic topology, which is now identical
with its p-adic topology.
Finally, the uniqueness of the components Ap in (6.8) follows at once from
the equality Ap D \q¤p qk A .k D 1; 2; : : : /, where q denotes primes. This is a
consequence of the relations qAp D Ap and \k qk Aq D 0. t
u
The group Ap in the preceding theorem is called the p-adic component of the
algebraically compact group A. As is shown in Lemma 2.3, it is a Jp -module.
Let us emphasize that owing to the theorem, groups Ap that are complete in
their p-adic topologies can be totally characterized by the invariants of their p-basic
subgroups Bp , i.e. by the cardinal numbers 0 and n .n D 1; 2; : : : / of the sets of
components Š Z, resp. of Z.pn / in a direct decomposition of Bp . This countable set
of invariants is complete: they determine Ap up to isomorphism, and, in addition,
this set is independent: it can be chosen arbitrarily. More explicitly,

Ap Š BQ p Š p-adic completion of ˚0 Z ˚ .˚1

nD1 ˚n Z.p //:
n

Q
Example 3.3. (a) We prove that Z Q Š p Jp . It is a reduced torsion-free algebraically compact
Q Z
group, so by the structure theorem it is the product of copies of Jp . Since Z=p Q Š Z=pZ
(inverse limit), each Jp occurs exactly once in the product.
(b) Let  denote an infinite cardinal. Then Jp is isomorphic to the p-adic completion of a direct
sum of 2 copies of Jp . For, it is a torsion-free p-adic algebraically compact group whose
p-basic subgroups have the same rank as Jp =pJp Š Z.p/ D ˚2 Z.p/.
Example 3.4. Let n .n D 1; 2; : : : / be arbitrary
Q1 cardinals. Then the p-adic completion of A D
˚1nD1 ˚n Z.p / is the subgroup of C D
n
nD1 Œ˚n Z.p /
which contains A and corresponds to
n

the divisible part of C=A.

Corollaries to the Structure Theorem The structure theorem allows us to derive

the following useful corollaries.
Corollary 3.5. A reduced torsion group is algebraically compact if and only if it is
bounded.
Proof. Bounded groups are algebraically compact. It is evident that, in a decompo-
sition (6.8) of a reduced torsion group A, only a finite number of Ap ¤ 0 may occur.
If the basic subgroup Bp is unbounded, then BQ p would contain elements of infinite
order. Thus the p-basic subgroups of our A must be bounded. Then Ap D Bp , and
the claim follows. t
u
Corollary 3.6 (Kaplansky [K]).
(i) Every reduced algebraically compact group ¤ 0 contains a summand isomor-
phic to Z.pk / .k D 1; 2; : : : / or to Jp , for some prime p.
(ii) The indecomposable algebraically compact groups are the following:
Z.pk / .k
1/ and Jp for all primes p, and Q.
3 The Structure of Algebraically Compact Groups 197

Proof. (i) If A is algebraically compact and reduced, and if p is a prime with

Ap ¤ 0, then a p-basic subgroup Bp of Ap contains either a finite cyclic
summand Z.pk / or a summand Š Z. It follows that BQ p contains a pure subgroup
Š Z.pk / or Š Jp . These are summands, as they are algebraically compact.
(ii) This is immediate from (i). t
u
Q
Example 3.7. The group P D n2N han i (where o.an / D pn ) is algebraically compact, being a
product of algebraically compact groups. The completion B Q of the pure subgroup B D ˚n2N han i
is a summand of P such that P=B Q is torsion-free, reduced, and algebraically compact. As P=pP Š
˚2@0 Z.p/, while B=pB Š B Q =pB
Q is countable, it follows that P=B
Q is the p-adic completion of a free
group of rank 2@0 (or of ˚2@0 Jp ).
By the way, it is easy to find a direct summand Š Jp in P. E.g., the vector x D .a1 ; : : : ; an ; : : : /
generates a Jp -submodule Jp x Š Jp whose purity (as a group or as a module) is obvious. Hence
Jp x is a summand of P.

Linear Compactness We now consider briefly linearly compact groups. A lin-

early compact group is a group A with a linear topology such that if a collection
aj C Aj .j 2 J/ of cosets modulo closed subgroups Aj has the finite intersection
property (i.e., any finite number of them have a non-void intersection), then the
intersection of all of them is not empty.
We make the following observations that are derived from the theory of
topological groups (some follow from results proved here).
(a) A subgroup of a linearly compact group is linearly compact in the induced
topology if and only if it is closed.
(b) The image of a linearly compact group under a continuous homomorphism is
again linearly compact.
(c) Direct products of linearly compact groups are linearly compact.
(d) Inverse limits of linearly compact groups are linearly compact.
(e) A reduced linearly compact group is complete in its topology.
(f) A group which is compact in a linear topology is linearly compact.
Groups that are linearly compact in the discrete topology admit a complete
classification.
Lemma 3.8 (Leptin, Schöneborn). A group is linearly compact in the discrete
topology if and only if it satisfies the minimum condition on subgroups.
Proof. Assume A is linearly compact in the discrete topology; then so is every
subgroup of A. A does not contain elements of infinite order, for Z is not linearly
compact: if  D s0 Cs1 pCs2 p2 C: : : is a non-rational p-adic integer, then the closed
cosets .s0 Cs1 p C  Csn1 pn1 /Cpn Z .n 2 N/ have no elements in common. Thus
A is torsion. Choose a basis fbj gj2J for its socle S with each bj of prime order, and let
Bk be the subgroup generated by all bj with j ¤ k: The cosets bk C Bk .k 2 J/ have
the finite intersection property, but the intersection of all of them is empty unless
J is a finite set. Hence S is finite which is equivalent to the minimum condition on
subgroups (see Theorem 5.3 in Chapter 4).
Conversely, if A satisfies the minimum condition, and if the cosets ai C Ai .i 2 I/
have the finite intersection property, then there is a minimal finite intersection
198 6 Algebraically Compact Groups

Ai1 \    \ Ain . It follows that the intersection of all ai CAi is equal to the intersection
of the cosets corresponding to this minimal intersection, and so it is not empty. u t
It follows that if A is any linearly compact group, and if U is an open subgroup
of A, then A=U satisfies the minimum condition on subgroups. In fact, A=U is then
linearly compact in the discrete topology.
Theorem 3.9. A group is linearly compact if and only if it is an inverse limit of
groups with minimum condition on subgroups.
Proof. The ‘if’ part is an immediate consequence of Lemma 3.8 and (d). To prove
the ‘only if’ part, assume A linearly compact, and let the subgroups Ui .i 2 I/
form a fundamental system of neighborhoods about 0. Without loss of generality,
the system of the Ui may be assumed to be closed under finite intersections. A=Ui is
discrete and linearly compact, so it satisfies the minimum condition, for every i 2 I.
j
The groups A=Ui with the natural maps i W a C Ui 7! a C Uj (for Ui
Uj ) form
an inverse system whose inverse limit must be A, since the image of A under the
natural map is a dense subgroup of the inverse limit, and by (a), the image must be
the whole inverse limit. t
u
Finally we verify:
Corollary 3.10. Linearly compact groups are algebraically compact.
Proof. We show that a linearly compact Pgroup A satisfies Theorem 1.2(d). A solu-
tion of the ith equation in the system j2J nij xj D ai 2 A .i 2 I/ is a coset of a
closed subgroup Ci of AJ . The system is finitely solvable if these cosets have the
finite intersection property. Linear compactness implies that the system is solvable.
t
u
Example 3.11. Jp is linearly compact in the p-adic topology. Jp is linearly compact in the finite
index topology, for every cardinal .
Example 3.12. The additive group Q 1
p of the p-adic numbers is the inverse limit of groups Z.p /;
it is linearly compact in the Prüfer topology.
F Notes. Kaplansky [K] gives a full description of algebraically compact groups. The theory
of algebraically compact modules contains several remarkable generalizations of the group case
and much more. See, e.g., Zimmermann-Huisgen–Zimmermann [1], Fuchs–Salce’s book Modules
over non-Noetherian Domains, 2001.
Eklof–Mekler [1] investigates the structure of ultraproducts of abelian groups and proves inter
alia that over an !-incomplete ultrafilter they are algebraically compact.

Exercises

(1) A countable algebraically compact group is the direct sum of a divisible group
and a bounded group.
4 Pure-Injective Hulls 199

(2) The group R of reals admits infinitely many different linearly compact
topologies.
(3) In an algebraically compact group, a finite set of elements is contained in a
summand that is a finite direct sum of Q, cocyclic groups, and copies of Jp for
various primes p. Q
(4) Determine the invariants of the algebraically compact group Π1 n 
nD1 Z.p /

where  is an infinite cardinal.

(5) (Leptin) A group with a linear topology is linearly compact if and only if it
is complete in this topology, and the factor groups modulo open subgroups
satisfy the minimum condition. [Hint: (e).]
(6) If A; B are algebraically compact groups such that A˚A Š B˚B, then A Š B.
(7) If A and B are algebraically compact groups, each isomorphic to a pure
subgroup of the other, then A Š B. [Hint: Theorem 3.2.]
(8) Jp has the cancellation property: if A; B are groups such that A ˚ Jp Š B ˚ Jp ,
then A Š B. [Hint: reduce the proof to reduced
Q torsion-free groups.]
(9) (Balcerzyk) Prove that ZN =Z.N/ Š D ˚ p Ap , where D is a Q-vector space
of dimension 2@0 , and Ap is the p-adic completion of ˚2@0 Jp . [Hint: the group
ZN =.pZN C Z.N/ / has cardinality 2@0 .]
(10) Are quasi-injective groups algebraically compact?
(11) The direct sum ˚@0 Z.pn / for any fixed n 2 N is algebraically compact, but it
cannot be linearly compact under any linear topology.

4 Pure-Injective Hulls

The striking analogy between injective and pure-injective groups can be pushed
further by pointing out the analogue of the injective hull.
Pure-Essential Extensions We first introduce the following notation. For a pure
subgroup G of A, K.G; A/ will denote the set of all subgroups H
A such that
(i) G \ H D 0; and
(ii) .G C H/=H is pure in A=H.
Since (i) implies G C H D G ˚ H, condition (ii) amounts to the fact that if
nx D g C h .n 2 N/ with g 2 G; h 2 H is solvable for x in A, then g is divisible by
n in G. Hence the set K.G; A/ is closed under taking subgroups. The purity of G in
A assures that 0 2 K.G; A/, so K.G; A/ is never empty.
Moreover, the set K.G; A/ is inductive. To prove this, let Hi .i 2 I/ be a chain of
subgroups in K.G; A/, and H their union. H obviously satisfies (i). Suppose nx D
g C h .g 2 G; h 2 H/ is solvable in A. For some i 2 I, h 2 Hi 2 K.G; A/ whence
g 2 nG, and so H 2 K.G; A/.
Following Maranda [1], we call a group A a pure-essential extension of its pure
subgroup G, and G a pure-essential subgroup of A, if K.G; A/ D f0g. A is a
maximal pure-essential extension of G if it is a pure-essential extension, but no
200 6 Algebraically Compact Groups

group properly containing A is a pure-essential extension of G. It is easily checked

that if Ai .i 2 I/ is a chain of pure-essential extensions of G, then A D [i2I Ai is also
a pure-essential extension of G.
Lemma 4.1. Suppose C is a pure-essential extension of G, and A an algebraically
compact group containing G as a pure subgroup. Then the identity map 1G extends
to an embedding  W C ! A.
Proof. Due to the pure-injectivity of A, the existence of an extension  W C ! A of
1G is evident. Since G is pure in C, we have Ker  2 K.G; C/. By pure-essentiality,
K.G; C/ D f0g, establishing our claim. t
u
Pure-Injective Hulls A is a pure-injective hull of G if it is a minimal pure-
injective group containing G as a pure subgroup. To prove the existence of such
a hull, we need a few important facts on pure-essential extensions.
Theorem 4.2 (Maranda [1]).
(i) Every pure-essential extension of a group G is contained in a maximal pure-
essential extension of G.
(ii) A is a maximal pure-essential extension of G if and only if it is a pure-injective
hull of G.
(iii) Every group G has a pure-injective hull, unique up to isomorphism over G.
Every pure-injective group containing G as a pure subgroup contains a pure-
injective hull of G.
Proof. (i) There exists a pure-injective group H containing G as a pure sub-
group Theorem 4.7 in Chapter 5. From Lemma 4.1 we derive that every
pure-essential extension of G must have cardinality
jHj. Therefore, the non-
isomorphic pure-essential extensions of G form a set, so it only remains to
appeal to Zorn’s lemma to get a maximal member containing any given one.
(ii) Let A be a maximal pure-essential extension of G, and C a pure-injective
group containing G as a pure subgroup. Select a maximal member M in the set
K.G; C/. Passing mod M, we obtain G Š .G C M/=M
.A C M/=M
C=M.
By the maximal choice of M, .G C M/=M is pure-essential in C=M. As G is
pure-essential in A, we must have A \ M D 0, so A Š .A C M/=M. Therefore,
.A C M/=M D C=M by the maximality of A, whence C D A ˚ M, so A
is pure-injective. To see that it is minimal pure-injective, assume that A0 is a
pure-injective group with G
A0
A. By Lemma 4.1, the identity 1G extends
to a monomorphism  W A ! A0 . Then A is also a maximal pure-essential
extension of G, and since A
A, A D A is the only possibility. Hence
A0 D A is indeed minimal.
With the same notations, there is a map  W A ! C which is the identity
on G; it has to be monic on A (see Lemma 4.1). By the preceding paragraph,
A, and hence A, is pure-injective, so A D C provided C is minimal. This
means that C is a maximal pure-essential extension of G.
(iii) The statements (i) and (ii) guarantee the existence of a pure-injective hull A
for every group G. If A0 is any pure-injective group containing G as a pure
4 Pure-Injective Hulls 201

subgroup, then the identity map 1G can be extended to a monomorphism  W

A ! A0 . If A0 happens to be also minimal, then necessarily A D A0 . Thus 
is then an isomorphism. t
u
We next prove a useful criterion for the pure-injective hull.
Lemma 4.3. A pure-injective group A containing G as a pure subgroup is the pure-
injective hull of G exactly if
(a) the maximal divisible subgroup of A is the injective hull of G1 ; and
(b) the factor group A=G is divisible.
Proof. First, assume A is a pure-injective hull of G. Write A D D ˚ C with D
divisible and C complete. In view of the purity of G in A,

G1 D \n nG D \n .G \ nA/ D G \ A1 D G \ D:

Since every non-zero summand of D must intersect G, D has to be an injective

hull of G1 . Define E
A such that E=G is the first Ulm subgroup of A=G. Then
E D D ˚ .C \ E/, and C=.C \ E/ Š .C C E/=E D A=E Š .A=G/=.E=G/ has
trivial Ulm subgroup. Thus C \ E is closed in C, so it is complete. Consequently, E
is pure-injective containing G, so E D A by minimality. Hence A=G coincides with
its own Ulm subgroup, which means that it is divisible.
To prove sufficiency, assume (a) and (b) hold for the pure-injective A containing
G as a pure subgroup. There is a pure-injective hull E of G contained in A. From the
first part of the proof it follows that E=G is divisible, so E is pure in A, and hence it
is a summand. It cannot be a proper summand, so A D E. t
u
It is now easy to describe how to obtain the pure-injective hull of a group.
Theorem 4.4. The pure-injective hull of a group G is isomorphic to the direct sum
Q of G.
of the injective hull of G1 and the completion G
Proof. Let D denote the injective hull of G1 , and let  W G ! D be an extension
of the inclusion map G1 ! D. If  W G ! G Q stands for the canonical map, then
consider the map W G ! D ˚ G Q which is the composite of the diagonal map
G ! G ˚ G followed by  ˚ . The purity of G in G Q guarantees that Im is pure
in D ˚ G.Q From Lemma 4.3 the assertion follows at once. t
u
The last theorem makes it possible to determine the complete system of invariants
for the pure-injective hull of a group G in terms of certain invariants of G;
cf. Theorem 3.1 in Chapter 4 and Theorem 3.2.
Pure-Essential is Not Transitive The analogy of ‘essential’ with ‘pure-essential’
subgroups breaks down when transitivity is considered.
Lemma 4.5 (Fuchs–Salce–Zanardo [1]).
(i) The property of being a “pure-essential extension” is not transitive.
(ii) However, it is transitive for p-local groups.
202 6 Algebraically Compact Groups

Proof. (i) Let B D ˚n2N Z.pn /, a †-cyclic group. Since B=B Q contains copies of
Q, it is easy to find a subgroup C of BQ such that C=B Š Z.q/ where q ¤ p is
a prime. Manifestly, B is pure-essential in C, since B is not a summand in C,
and C=B is torsion-free of rank 1. By Theorem 4.4, C is pure-essential in its
Z-adic completion which is CQ D BQ ˚ Jq (see Theorem 2.11). However, B is not
Q since it is contained in its first summand.
pure-essential in C,
(ii) The arguments in the proof of Lemma 4.3 can be applied to show that a pure
subgroup G of a p-local group A is pure-essential if and only if (a) G1 is essential
in A1 , and (b) A=G is p-divisible. Since properties (a) and (b) are transitive, the
claim follows. t
u
F Notes. Every module over any ring admits a pure-injective hull, unique up to isomorphism.
However, pure-injective hulls cannot be obtained in general via completion. There is an extensive
theory of pure-injectivity that started to develop following the pattern of groups. Pure-injective
modules admit indecomposable summands, so that their structure is considerably simpler to study.
Pure-injective hulls may compete in significance with injective hulls. They play important roles
in the structure theory of modules over arbitrary rings, not to mention their relevance in model
theory. As illustration, let us refer to the interesting result by G. Sabbagh [C.R. Acad. Sci. Paris,
271, A909–A912 (1970)] which states that for any ring R, an R-module is elementarily equivalent
to its pure-injective hull. Thus the elementary theory of modules is reduced at once to the case of
pure-injective modules.

Exercises

(1) If A is a pure-essential extension of G, and if B is a pure subgroup of A

containing G, then B is a pure-essential extension of G.
(2) Prove that if G is pure in A, H
A1 , and H \ G D 0, then H 2 K.G; A/:
(3) Let A be a p-group and B its basic subgroup.
(a) K.B; A/ is the set of all subgroups of A1 .
(b) B is pure-essential in A if and only if A1 D 0.
(4) If C is a pure-essential extension of G, and C=G is not divisible, then there is a
pure-essential extension C0 of G such that C < C0 and C0 =A is divisible.
(5) Show that if Ai .i 2 I/ is a chain of pure-essential extensions of G, then A D
[i2I Ai is also a pure-essential extension of G.
(6) Let A be a pure-injective group containing G as a pure subgroup. A maximal
pure-essential extension of G in A is a summand of A.
(7) Give anQ example where An .n < !/ are the pure-injective
Q hulls of the groups
Gn , but n An is not the pure-injective hull of n Gn .
(8) Using invariants of a group, determine a complete system of invariants for its
pure-injective hull. [Hint: ranks.]
(9) Find conditions on the invariants of a pure-injective group A such that A has no
proper summands isomorphic to A.
5 Locally Compact Groups 203

5 Locally Compact Groups

Having discussed compact and linearly compact groups, it is worth while making a
few comments on the locally compact case. Besides the Pontryagin duality theory
we rely heavily on homological machinery, so a reader not familiar with the material
of Chapters 7–9 is advised to read this section after studying these chapters.
Locally Compact Extensions It is well known that a locally compact (abelian)
group is the direct sum of Rn (for some integer n  0) and a group G that contains an
open compact subgroup C. Thus G fits into an exact sequence 0 ! C ! G ! A !
0 where C is compact and A is discrete. Conversely, any extension of a compact C
by a discrete A yields a well-defined locally compact group G, as the only topology
compatible with the given topologies of C and A is the one in which a base of
neighborhoods of 0 in C is also a base of neighborhoods of 0 in G. Consequently,
if C is compact and A is discrete, then the elements of Ext.A; C/ may be viewed as
the locally compact extensions of C by A (the equivalence of extensions is the same
whether C is discrete or compact).
Recall that the Pontryagin dual (or character group) Char G of a locally
compact group G, i.e., the group of all continuous homomorphisms of G into the
compact group T D R=Z, equipped with the compact-open topology, is again
locally compact. We write
O D Char G D Hom.G; T/;
G

where Hom now stands for continuous homomorphisms. Importantly, if G is

O is discrete, and vice versa. The Pontryagin Duality asserts that
compact, then G
the map
OO
G W G ! G; acting as G .x/./ D .x/ O
.x 2 G;  2 G/

implements a canonical algebraic and topological isomorphism between G and its

OO
second dual G.
Hom, Ext with Compact Second Argument In the sequel, the letters M; N will
O NO are discrete groups. The group
stand for compact groups, so their duals M;
O
Hom.M; N/ carries a compact topology: the compact-open topology makes it into a
compact group.
Lemma 5.1. For compact groups M and N, there are natural .topological/ isomor-
phisms of compact groups

O N/ ! Hom.N;
MN W Hom.M; O M/;

O N/ ! Hom.M
MN W Hom.M; O ˝ N;
O T/ D .M
O ˝ N/O:
O
204 6 Algebraically Compact Groups

Proof. It is enough to consider a sequence of natural isomorphisms:

O N/ Š Hom.M;
Hom.M; O Hom.N;
O T// Š Hom.M
O ˝ N;
O T/

Š Hom.NO ˝ M;
O T/ Š Hom.N;
O HomZ .M;
O T// Š Hom.N;
O M/;

where we have made repeated use of Theorem 1.13 in Chapter 8. t

u
If the isomorphism Theorem 3.8 in Chapter 9 is used with the following cast of
O B D NO are discrete groups, and C D T, then we get
characters: A D M;

O Hom.N;
Ext.M; O T// Š Hom.Tor.M;
O N/;
O T/:

Consequently,
Lemma 5.2. For compact M and N, there is a natural isomorphism

O N/ Š .Tor.M;
Ext.M; O N//O
O

O N/ into a compact group.

that makes Ext.M; t
u
Since Tor is symmetric in its arguments, Lemma 5.2 shows that the switch
O M/
involution in Tor.N; O induces a natural isomorphism between the compact groups
O O
Ext.M; N/ and Ext.N; M/. This isomorphism can easily be described explicitly:
Theorem 5.3. For compact groups M; N, there is a natural topological isomor-
phism

O N/ Š Ext.N;
Ext.M; O M/

O ! 0 to its Pontryagin dual 0 ! M !

that carries the extension 0 ! N ! G ! M
GO ! NO ! 0.

Proof. As G is locally compact, Pontryagin duality yields a natural bijection MN

between the two Exts whose inverse is the map NM . This bijection respects the
group operations, since the Baer sum is defined by using diagonal and codiagonal
maps (which are dual to each other). t
u
Duality of Long Exact Sequences The purpose of including this section in this
volume was also to point out that the classical induced long exact sequences The-
orem 2.4 in Chapter 8 and Theorem 2.3 in Chapter 9 become intertwined once
topology gets involved. In fact, we can verify a surprising duality between these
long sequences.
5 Locally Compact Groups 205

Theorem 5.4. Let M be compact, A; B; C discrete, and the sequence 0 ! A !

B ! C ! 0 exact. The Pontryagin dual of the exact sequence

O ! Tor.B; M/
0 ! Tor.A; M/ O ! Tor.C; M/
O !

O !B˝M
!A˝M O !C˝M
O !0

of discrete groups is the exact sequence

0 ! Hom.C; M/ ! Hom.B; M/ ! Hom.A; M/ !

! Ext.C; M/ ! Ext.B; M/ ! Ext.A; M/ ! 0

of compact groups .where the maps are continuous/.

Proof. Application of the exact functor Hom.; T/ to the first long exact sequence
yields the second one, after making use of Lemma 5.1 and 5.2. The continuity of
the maps is straightforward. t
u
Autodual Extensions As an application, consider the special case M D N. Then
MM is an involution of the compact group Ext.M; O M/. Of particular interest are
those extensions which are autodual in the sense that MM carries them into equiv-
alent extensions, and those which antidual, i.e. MM acts on them as multiplication
by 1. It is readily checked that both the autodual and the antidual extensions in
O M/ form a subgroup Ext.M;
Ext.M; O M/C and Ext.M; O M/ , respectively.

Proposition 5.5. Suppose M is compact such that multiplication by 2 is an

O M/. Then there is a direct decomposition
automorphism of Ext.M;

O M/ D Ext.M;
Ext.M; O M/C ˚ Ext.M;
O M/

where the summands are closed subgroups.

Proof. All that we have to point out is that, for every element e of Ext.M; O M/, e
1 1
is the Baer sum of 2 .e C eO/ and 2 .e  eO /, which are evidently auto- and antidual
extensions, respectively. t
u
Corollary 5.6 (Fuchs–Hofmann [1]). All the extensions of a compact group M by
O are autodual if and only if the torsion part of M
its dual M O is locally cyclic.

Proof. By Lemma 5.2, the condition can be rephrased as saying that the involution
O M/
of Tor.M; O corresponding to the switch of the arguments is the identity. Since if
O
M D A ˚ B, then

O M/
Tor.M; O Š Tor.A; A/ ˚ Tor.B; A/ ˚ Tor.A; B/ ˚ Tor.B; B/;

and the involution in question would switch the two middle summands, the condition
Tor.A; B/ D 0 is necessary. This means that, for each prime p, one of the
206 6 Algebraically Compact Groups

O is locally cyclic, i.e.

p-components Ap ; Bp must be 0, i.e. the torsion part of M
O
tM
Q=Z. As Tor vanishes for torsion-free groups, it is pretty clear that this
condition is sufficient. t
u
Example 5.7 (Samelson). Consider the extension 0 ! Jp ! Q 1
p ! Z.p / ! 0 where Jp
carries the obvious compact topology. This extension is self-dual, and so are all extensions of Jp
by Z.p1 /.
F Notes. More on the homological aspects of topological groups may be found in K.H.
Hofmann–S.A. Morris, The Structure of Compact Groups (de Gruyter, 1998), and on autoduality
in Fuchs–Hofmann [1].
Loth [3] investigates how group properties are reflected in the Pontryagin duals.

Exercises

(1) If 0 ! L ! M ! N ! 0 is an exact sequence of compact groups, then for any

discrete A, the sequence

0 ! Hom.A; L/ ! Hom.A; M/ ! Hom.A; N/ !

! Ext.A; L/ ! Ext.A; M/ ! Ext.A; N/ ! 0

of compact groups is exact with continuous maps.

(2) Find the Pontryagin dual of the long exact sequence in the preceding exercise.
(3) An extension equivalent to an autodual extension is autodual as well.
(4) Describe the auto- and antidual extensions in Ext.Z.p/ ˚ Z.p1 /; Z.p/ ˚ Jp / if
p ¤ 2.

6 The Exchange Property

We end this chapter with a brief study of the Exchange Property, a property that has
come to prominence in the literature. It is closely connected with the problem of
uniqueness of direct decompositions, with the Krull-Schmidt theorem, as we shall
see below.
Exchange Property A group G is said to have the (finite) exchange property if,
for any direct decomposition

A D G ˚ H D ˚i2I Ai

of a group A, where Ai ; H are arbitrary groups, and I is a (finite) index set, there
always exist subgroups Bi of Ai for i 2 I such that
 
A D G ˚ ˚i2I Bi :
6 The Exchange Property 207

Evidently, Bi must be a summand of Ai : it is a summand of A contained in Ai . We

shall see that some special groups, like injective, quasi-injective, and pure-injective
groups, enjoy the exchange property.
The next lemma will be needed in the following proofs (it holds also for
modules).
Lemma 6.1 (Crawley–Jónsson [1]). It suffices to check the exchange property for
G in case the Ai are isomorphic to subgroups of G.
Proof. Let A D G ˚ H D ˚ Ai with arbitrary Ai . Set Hi D H \ Ai , and let overbars
denote images under the canonical map  W A ! A=H 0 where H 0 D ˚ Hi : Then
AN D G
N ˚H N D ˚ AN i , where

AN i Š Ai =.Ai \ H 0 / D Ai =.Ai \ H/ Š .Ai C H/=H

A=H Š G:

By hypothesis, there exist BN i

AN i satisfying AN D G
N ˚ .˚ BN i /. Setting
1 N
Bi D  Bi \ Ai ; we clearly have A D G ˚ .˚ Bi /. t
u
Groups with the Exchange Property It seems sensible to start the study with
indecomposable groups. (It is easily checked that, for indecomposable groups, the
finite and the general exchange properties are equivalent.) Observe that a group with
local endomorphism ring is necessarily indecomposable.
Theorem 6.2 (Warfield [2]). An indecomposable group G has the finite exchange
property if and only if its endomorphism ring End G is local.
Proof. Suppose End G is local, and A D G ˚ H D B ˚ C. Let ; ˇ; denote the
projections of A onto its summands G; B and C, respectively. ˇ C D 1A implies
ˇ C   D  where the terms will be viewed as endomorphisms of G; in
particular,  acts as the identity on G. End G local implies that either ˇ or   is
an automorphism of G; say, the first alternative holds. Setting K D Im ˇ, it is clear
that both  K W K ! G and ˇ G W G ! K are isomorphisms. As K D ˇG D G,
we have A D K ˚ H. Now K
B implies B D K ˚ B0 for some B0
B, thus
A D G ˚ H D K ˚ B0 ˚ C. Taking into account that ˇ induces an isomorphism of
G with K, we conclude that A D G ˚ B0 ˚ C, i.e. G enjoys the exchange property.
To prove the converse, assume End G is not local. Thus there are endomorphisms
˛ and ˇ of G, which are not automorphisms, such that ˛  ˇ D 1G . Then A D
G1 ˚ G2 .Gi Š G/ also decomposes as A D H1 ˚ H2 where H1 D f.g; g/ j g 2 Gg
and H2 D f.ˇg; ˛g/ j g 2 Gg are indecomposable. However, neither A D G1 ˚ H1
nor A D G1 ˚ H2 , so G does not have the exchange property. t
u
Several important groups have the exchange property. For instance,
Theorem 6.3 (Warfield, Fuchs). Injective groups, and more generally, quasi-
injective groups share the exchange property.
Proof. We prove the claim for a quasi-injective G. Suppose A D G ˚ H D ˚i2I Ai
where each Ai is isomorphic to a subgroup of G. Pick a subgroup B of A maximal
208 6 Algebraically Compact Groups

with respect to the properties: (i) B D ˚i2I Bi with Bi

Ai , and (ii) G \ B D 0. We
claim A D G ˚ B.
The natural homomorphism  W A ! A=B maps G isomorphically into A, so G
is a quasi-injective subgroup in A=B D ˚i .Ai CB/=B: Owing to the maximal choice
of B, G\.Ai CB/=B is an essential subgroup in .Ai CB/=B for each index i, whence
we conclude that G is essential in A=B: As Ai
A=B, where Ai is isomorphic to
a subgroup of G, and G is fully invariant in its injective hull (that contains A=B),
we can argue that necessarily Ai
G: Therefore, A
G, whence G D A=B,
and A D G ˚ B follows. t
u
Our next task is to establish a sufficient condition on a group to have the exchange
property, this will be needed in the applications. The theorems above suggest strong
dependence on the endomorphism ring. This is confirmed by the next theorem.
An exchange ring is a (not necessarily commutative) ring E such that the left
module E E has the finite exchange property.
Theorem 6.4 (Warfield). A group G has the finite exchange property if and only if
E D End G is an exchange ring.
Proof. In the proof we use the ad hoc notations X _ D Hom.X; G/ for a group X,
and Y ^ D HomE .Y; G/ for a left E-module Y (where G is viewed as a left E-
module). Thus X _ is a left E-module, and Y ^ is a group. In this notation, G_ D E E
and E^ D G. Consequently, the canonical maps G ! G_^ and E ! E^_ are
isomorphisms.
Assuming G has the exchange property, let E M D E ˚ N D L1 ˚ L2 with a left
E-module N and left ideals Li of E. Hence M ^ D G ˚ N ^ D L^ ^
1 ˚ L2 are group
decompositions. Hypothesis implies the existence of summands Ci
L^ i such that
M ^ D G ˚ C1 ˚ C2 : Thus, M ^_ D E ˚ C1_ ˚ C2_ : In view of the isomorphism
E ! E^^ , the canonical maps Li ! Li ^^ are monic, so the map M ! M ^^
is likewise monic. It remains to set Bi D Li \ Ci^ to obtain M D E ˚ B1 ˚ B2 ,
establishing the exchange property of E.
Since the group G and the ring E play symmetric roles in the last argument, the
proof is complete. t
u
Exchange Rings We now investigate properties of endomorphism rings that imply
the exchange property. We need a definition: for an ideal H of the ring R, we say that
idempotents lift modulo H if every idempotent coset in R/H contains an idempotent
element of R.
Theorem 6.5. A ring E is an exchange ring if it is von Neumann regular modulo its
Jacobson radical J, and idempotents lift mod J.
Proof. Assume E is a ring as stated. First we prove that if a C b D 1 .a; b 2 E/,
then there exists an idempotent e 2 Ea such that 1  e 2 Eb. We start with the
special case when J D 0. By regularity, if given a 2 E, there exists a c 2 E such
that aca D a. Then f D ca is an idempotent, and so is e D f C .1  f /a 2 Ea. A
simple calculation leads to 1  e D .1  f /.1  a/ D .1  f /b 2 Eb.
6 The Exchange Property 209

In the general case, let overbars denote cosets mod J. By the preceding paragraph,
for a 2 E, there exists an x 2 E such that xN 2 D xN 2 EN N a and 1N  xN 2 EN b.
N We may
assume, without loss of generality, that x 2 Ea. Hypothesis guarantees that we can
find an idempotent f 2 xN . Then u D 1  f C x is a unit in E (as x  f 2 J), so u1 2 E
exists. Setting g D u1 fu D u1 fx 2 Ex
Ea, and observing that gN D fN D aN , we
have g2 D g and .1  g/  y.1  a/ 2 J for some y 2 E. Consequently, g C y.1  a/ is
a unit in E, so it has an inverse v 2 E. It remains to define e D g .1 g/vg 2 Eg

Ea, and to check that e2 D e and 1  e D .1  g/.1  vg/ D .1  g/vy.1  a/ 2 Eb.

Suppose E is the endomorphism ring of the group A. Let A D G ˚ H D A1 ˚ A2 ,
and denote the projections onto G; A1 ; A2 by ; ˛1 ; ˛2 , respectively. Evidently, E
is the endomorphism ring of G, and  ˛1  C  ˛2  D . By what has been proved
above, there exist ˇi 2 E .i D 1; 2/ such that i D ˇi ˛i  are orthogonal
idempotent endomorphisms of G with 1 C 2 D . Since i is an idempotent,
ˇi may be chosen so as to satisfy i ˇi D ˇi . Note that

ˇi D ˇi D ˇi ;  i D i D i ; ˇi ˛i ˇi D ˇi :

The maps i D ˛i ˇi ˛i .i D 1; 2/ are orthogonal idempotents satisfying i i D

˛i ˇi ˛i ˇi ˛i  D ˛i ˇi ˛i  D i . Setting Bi D Ai \ Ker i , it is easily seen that
Bi D ImŒ.1  i /˛i
D Im.˛i  i /. We now claim that A D G ˚ B1 ˚ B2 .
First, if a 2 G \ .B1 ˚ B2 /, then from 1 a D 0 D 2 a we obtain that a D
1 aC 2 a D ˇ1 ˛1 aCˇ2 ˛2 a D .ˇ1 ˛1 ˇ1 /˛1 aC.ˇ2 ˛2 ˇ2 /˛2 a D ˇ1 1 aCˇ2 2 a D 0:
Evidently, A D Im 1 C Im 2 C Im.˛1  1 / C Im.˛2  2 /, hence it only remains
to prove that Im i
G C B1 C B2 . We have ˛i ˇi i D ˛i ˇi ˛i ˇi ˛i D ˛i ˇi ˛i D ˛i i ,
and for i ¤ j, also j ˇi D j .ˇi / D .j /. i ˇi / D .j j /. i ˇi /: It follows that
Im.i  ˇi i / is annihilated by both i and ˛i , so it is contained in Bj .j ¤ i/.
Consequently, Im i
Im ˇi i C Im.i  ˇi i /
G C Bj . t
u
For the proof of Theorem 6.7 below, we require the following lemma that is a
special case of a result by Zimmermann-Huisgen–Zimmermann [1].
Lemma 6.6. Let the ring E be algebraically compact as a left module over itself.
N D E=J is von Neumann regular, and its idempotents lift to E.
Then E
Proof. Suppose LN is a finitely generated left ideal of E. N Clearly, there is a finitely
generated left ideal L of E that maps upon LN via the canonical homomorphism  W
E ! E. N First we verify the existence of a finitely generated left ideal U minimal
with respect to the property L + U = E.
Let W be a left ideal of E satisfying L+W=E, and fUi j i 2 Ig a descending
chain of finitely generated left T ideals of E, contained in W, such that L + Ui D E
for every i 2 I. As U D i2I Ui is again a left ideal, it remains only to show
that L + U = E (by minimality, U is then even singly generated, since 1 2 E). Let
fa1 ; : : : ; an g and fui1 ; : : : ; uimi g be sets of generators of L and Ui , respectively. The
condition L + Ui D E translates into the solvability of the system of equations

x1 a1 C    C xn an C yi1 ui1 C    C yimi uimi D 1 .i 2 I/

210 6 Algebraically Compact Groups

in E for the unknowns xj ; yik . The chain property of the set of the Ui guarantees that
for every finite subset J of I, the system consisting of equations with indices in J
admits a simultaneous solution. Algebraic compactness implies the existence of a
global solution xj D ej 2 E and yik D cik 2 E. Then U is the principal left ideal
generated by u D 1  .e1 a1 C    C en an /.
We claim that for this U, the intersection L \ U is superfluous in E (i.e., it can
be omitted from every generating system of E). First we show that it is superfluous
in U. Suppose that U D .L \ U/ C V for some left ideal V
U. Then E D
L C .L \ U/ C V D L C V implies V D U by the minimality of U. Next, let
E D .L\U/CA for a left ideal A
E. By the modular law, U D .L\U/C.A\U/,
whence U D A \ U follows as L \ U is superfluous in U. But then U
A, thus
E D .L \ U/ C U C A D U C A D A, and L \ U is superfluous in E. Hence
L \ U
J, and passing mod J, we obtain that the sum LN C U N D EN has to be direct.
A ring in which every finitely generated left ideal is a summand is known to be von
Neumann regular, so EN is von Neumann regular.
It remains to verify that idempotents lift modulo J. For an idempotent coset xCJ,
consider the set S of all cosets f C A of ideals A of E such that f C A
x C J (so
A
J) and f 2 2 f C A. The set S is not empty, and a proof like the one above (for
the existence of a minimal U) will guarantee that it contains a minimal member, say,
e C B .B
J/. To complete the proof, we show that e is an idempotent.
Since .1  2e/2 D 1 C 4.e2  e/ 2 1 C J is invertible, so is 1  2e. Hence
y D .e2  e/.1  2e/1 2 B. Set f D e C y 2 x C J and C D y2 E
B
J. Then
.e C y/2  .e C y/ D e2  e C .2e  1/y C y2 D y2 shows that we have f C C 2 S. As
f C C
e C B, necessarily B D C. Hence y 2 C implies y D y2 r for some r 2 E,
and from 1  yr 2 1 C J we obtain y D 0. Consequently, e D f is idempotent. u t
We are now prepared for the proof of the following important theorem.
Theorem 6.7 (Warfield [3]). Algebraically compact groups have the finite
exchange property.
Proof. Combining Theorem 6.5, Lemma 6.6 with Theorem 6.4, it follows that
a group enjoys the exchange property whenever its endomorphism ring is a left
algebraically compact ring. As left and right exchange rings are identical (this
follows from Theorem 6.4, since the opposite ring Eopp is the endomorphism ring
of E E). An appeal to Theorem 4.2 in Chapter 16 completes the proof. t
u
The next theorem is the main result on direct sums of groups with local
endomorphism rings. We state it here without proof.
Theorem 6.8 (Azumaya, Crawley–Jónsson [1]). Assume
M
AD Ai
i2I

where each Ai is a countably generated group with local endomorphism ring. Then
every direct decomposition of A can be refined to a direct decomposition isomorphic
6 The Exchange Property 211

to the given one. In particular, every summand C of A satisfies C Š ˚i2J Ai for a

suitable subset J of I. u
t
A class C of groups is said to have the Krull-Schmidt property if every group in
C is a direct sum of indecomposable members of C, and such a direct decomposition
is unique up to isomorphism. Hence Theorem 6.8 can be rephrased by stating that
direct sums of countably generated groups with local endomorphism rings enjoy the
Krull-Schmidt property.
F Notes. The exchange property for algebraic systems was introduced by Crawley–Jónsson
[1], it became a well-researched subject. It is still an open problem whether or not the finite
exchange property implies the unrestricted exchange property (it is true for indecomposable
modules). The proofs in the text are modeled after arguments by W.K. Nicholson [Trans. Amer.
Math. Soc. 229, 269–278 (1977)]. The exchange property for injective modules was proved
by Warfield [Pacific J. Math. 31, 263–276 (1969)] and for quasi-injectives by Fuchs [Annali
Scuola Norm. Pisa 23, 541–546 (1969)]. Monk [Proc. Amer. Math. Soc. 35, 349–353 (1972)]
characterizes exchange rings as rings R such that for all a 2 R there exist b; c 2 R with
bab D b; c.1  a/.1  ba/ D 1  ba. For Theorems 6.4 and 6.7, see Warfield [Math. Ann.
199, 31–36 (1972)], and for Theorem 6.8, Azumaya [Nagoya Math. J. 1, 117–124 (1950)].
There are several results facilitating the proofs that certain groups (or modules) enjoy the
exchange property. For instance, for the finite exchange property it suffices to check for an index
set of cardinality 2 (Exercise 1). Furthermore, in Lemma 6.1 it is enough to assume that all the
groups Ai are isomorphic to G (Zimmermann-Huisgen–Zimmermann [1]).
Ivanov [4] discusses an ‘almost exchange property’ of groups G. His definition reads as follows:
if A D G ˚ H D ˚i2I Ai , then there are a partition of I into finite subsets Ij and subgroups
Cj  ˚i2Ij Ai such that A D G ˚ .˚j Cj /. Closure under finite sums is established.

Exercises

(1) For the finite exchange property it suffices to check for 2 summands.
(2) (a) A finite direct sum has the finite exchange property exactly if all summands
enjoy this property.
(b) In general, the exchange property is not preserved when taking infinite
direct sums. [Hint: a direct sum of unbounded cyclic p-groups.]
(3) (a) Finite groups have the exchange property.
(b) Z does not have the exchange property.
(4) If A D B ˚ C D G ˚ H and B is an unbounded †-cyclic p-group, then either G
or H has an unbounded †-cyclic summand. [Hint: keep using Exercise 3.]
(5) Suppose that G D A1 ˚    ˚ An where all End Ai are local. Use the exchange
property to show:
(a) a decomposition of G into indecomposable summands is unique up to
isomorphism;
(b) a summand of G is isomorphic to a partial direct sum of the given
decomposition.
212 6 Algebraically Compact Groups

Problems to Chapter 6

PROBLEM 6.1. Which reduced algebraically compact groups carry an injective

module structure over some ring with 1?
PROBLEM 6.2. Which algebraically compact rings can be endomorphism rings
of groups?
Such rings are not cotorsion-free, so the results in Sect. 7 in Chapter 16 are not applicable.

PROBLEM 6.3. Does the finite exchange property imply the general exchange
property?
This is a well-known difficult open problem.
Chapter 7
Homomorphism Groups

Abstract The fact that the homomorphisms of a group into another group form an abelian group
has proved extraordinarily profound not only in abelian group theory, but also in Homological
Algebra where the functor Hom is one of the cornerstones of the theory. Our first aim is to find
relevant properties of Hom both as a bifunctor and as a group.
It is rather surprising that in some significant cases Hom.A; C/ is algebraically compact; for
instance, when A is a torsion group, or when C is algebraically compact. In the special situation
when C is the additive group T of the reals mod 1, in which case Hom.A; T/, furnished with
the compact-open topology, will be the character group of A, our description leads to a complete
characterization of compact abelian groups by cardinal invariants. An analogous result deals with
the linearly compact abelian groups.
The final section discusses special types of homomorphisms that play important roles in the
theory of torsion groups.

1 Groups of Homomorphisms

Homomorphism Groups We have already noticed earlier that, if ˛ and ˇ are

homomorphisms of A into C, then their sum ˛ C ˇ, defined as

.˛ C ˇ/a D ˛a C ˇa .a 2 A/;

is again a homomorphism A ! C. It is now routine to check that the homomor-

phisms of A into C form an abelian group under addition. This group is called the
homomorphism group of A into C and is denoted by Hom.A; C/. The zero in this
group is the trivial homomorphism mapping A to 0 2 C, and the inverse ˛ of
˛ W A ! C maps a 2 A upon .˛a/ 2 C.
If A D C, then the elements of Hom.A; A/ are the endomorphisms of A, and
the group Hom.A; A/ D End A is called the endomorphism group of A. This
group carries a ring structure where the product ˛ˇ of ˛; ˇ 2 End A is defined
by .˛ˇ/a D ˛.ˇa/ for a 2 A (observe the order of maps). The ring identity is the
identity automorphism of A.
Next we list some simple facts on homomorphism groups.
(A) There are two important necessary conditions to satisfy when we are looking
for homomorphisms ˛ W A ! C: one is that if a 2 A is annihilated by n 2 N,
then also n.˛a/ D 0 2 C, and the other is that we must have hp .˛a/  hp .a/.

© Springer International Publishing Switzerland 2015 213

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_7
214 7 Homomorphism Groups

(B) Hom.A; C/ D 0 in the following cases: (i) A is torsion and C is torsion-free;

(ii) A is a p-group and C is a q-group, for primes p ¤ q; (iii) A is divisible and
C is reduced.
(C) If CŒn
D 0 for some n 2 N, then Hom.A; C/Œn
D 0 for every group A. Indeed,
if ˛ W A ! C and n˛ D 0, then for a 2 A we have n.˛a/ D .n˛/a D 0 whence
CŒn
D 0 implies ˛a D 0, i.e. ˛ D 0.
(D) Hom.A; C/ is torsion-free whenever C is torsion-free.
(E) If C is torsion-free and divisible, then so is Hom.A; C/ for every A. In order
to show that Hom.A; C/ is now also divisible, pick an ˛ 2 Hom.A; C/ and
an n 2 N. For a 2 A, there exists a unique c 2 C with nc D ˛a, and thus
we may define a map ˇ W A ! C via ˇa D c. It follows readily that ˇ is a
homomorphism A ! C satisfying nˇ D ˛.
(F) If nA D A for some n 2 N, then Hom.A; C/Œn
D 0. Indeed, let ˛ 2 Hom.A; C/
with n˛ D 0. Write a 2 A as a D nb for some b 2 A. Then ˛a D ˛.nb/ D
.n˛/b D 0 shows that ˛ D 0.
(G) If A is divisible, then Hom.A; C/ is torsion-free.
(H) If A is torsion-free and divisible, then the same holds for Hom.A; C/, for any
C. The proof is similar to the one in (E).
Example 1.1. If A D Z, then every ˛ W Z ! C is completely determined by ˛.1/ D c 2 C.
Moreover, evidently, for every c 2 C there is a homomorphism W Z ! C such that .1/ D c:
Since ˛.1/ D c1 and ˇ.1/ D c2 imply .˛ C ˇ/.1/ D c1 C c2 , the correspondence 7! c given
by .1/ D c is a natural isomorphism between Hom.Z; C/ and C,
Hom.Z; C/ Š C for all groups C:

Example 1.2. If A D Z.m/ with m 2 N, then again, every homomorphism ˛ W Z.m/ ! C is

determined by the image ˛.1N / D c of the coset 1N D 1 C mZ, but here mc D 0 must hold, i.e.
c 2 CŒm
. Conversely, each such c gives rise to a homomorphism W 1N 7! c, and as in the preceding
example, the correspondence 7! c given by .1N / D c is a natural isomorphism

Hom.Z.m/; C/ Š CŒm
for all groups C:

Example 1.3. From the preceding example we obtain

Hom.Z.pk /; Z.pn // Š Z.p` / where ` D minfk; ng:

Example 1.4. Next, let C be quasi-cyclic, say, C D hc1 ; : : : ; cn ; : : : i with the defining relations
pc1 D 0; pcnC1 D cn .n
1/. If  is an endomorphism of C, then write cn D kn cn with an integer
kn .0  kn < pn / for every n. Now kn cn D cn D .pcnC1 / D pcnC1 D pknC1 cnC1 D knC1 cn
implies kn  knC1 mod pn . This means that the sequence of the kn is a Cauchy sequence in Jp , so it
has a limit, say  2 Jp is the limit. The correspondence  7!  between the endomorphisms  of C
and the p-adic integers  is evidently additive. If the endomorphisms 1 and 2 define the same ,
then 1 2 maps every cn to 0, i.e. 1 D 2 . On the other hand, if  D s0 Cs1 pC  Csn pn C: : :
is any p-adic integer, then the correspondence cn 7! .s0 C s1 p C    C sn pn /cn for all n (which
we may write simply as cn ) extends uniquely to an endomorphism  of C such that  7! . We
conclude:

End Z.p1 / Š Jp :

Example 1.5. Consider Q.p/ , the group of rational numbers with powers of p as denominators,
and C D Z.p1 /. Suppose Q.p/ D h1; p1 ; : : : ; pn ; : : : i and C as in the preceding example
1 Groups of Homomorphisms 215

(and cn D 0 for n  0). A p-adic number  D pk  (with a p-adic unit  and k 2 Z) induces a
homomorphism  W Q.p/ ! Z.p1 / such that pn 7! cnk for all n. As in the preceding example,
we can convince ourselves that different p-adic numbers  give rise to different homomorphisms,
and every homomorphism  W Q.p/ ! Z.p1 / arises in this way. Consequently, Hom.Q.p/ ; Z.p1 //
is isomorphic to the additive group of all p-adic numbers, i.e., we have

Hom.Q.p/ ; Z.p1 // Š ˚ Q with  D 2@0 :

Example 1.6. If A D C D Jp , then it is evident that multiplication by a fixed p-adic integer 

is an endomorphism of Jp (which we denote by ), P and different p-adic integers yield different
endomorphisms of Jp , since they map 1 2 Jp differently. Let  2 End Jp such that .1/ D . Then
 and P are identical on Z  Jp , so Z  Ker.  /.
P But Jp =Z is divisible, while Jp is reduced, so
Jp = Ker.  /
P D 0. It follows that  D ,
P and we have

End Jp Š Jp :

Hom and Direct Sums and Products Our next concern is the behavior of Hom
towards direct sums and direct products. The following theorem is fundamental.
Theorem 1.7. For an arbitrary index set I, there are natural isomorphisms
Y
Hom.˚i2I Ai ; C/ Š Hom.Ai ; C/ (7.1)
i2I

and
Y Y
Hom.A; Ci / Š Hom.A; Ci /: (7.2)
i2I i2I

Proof. In order to prove (7.1), let i W Ai ! ˚ Ai and i W ˚ Ai ! Ai denote the

injection and the projection maps, respectively. We map the left side of (7.1) to the
right side by sending ˛ W ˚ Ai ! C to .: : : ; ˛i ; : : : / where ˛i W Ai ! C. This
is evidently a homomorphism  from the left to the right Q side. It is clear that 
maps ˛ to 0 only if ˛ D 0. Since every .: : : ; ˛i ; : : : / 2 Hom.Ai ; C/ defines an
˛ 2 Hom.˚ Ai ; C/ via ˛ D ˚.˛i i /,  is epic
Q as well. Q
For the proof of (7.2), let i W Ci ! Ci and i W Ci ! C Qi denote the
injection and the projection maps, respectively. Every ˇ 2 Hom.A; Ci / defines
a homomorphism i ˇ 2 Hom.A; Ci / for each i. As in the preceding paragraph, we
conclude that the correspondence ˇ 7! .: : : ; i ˇ; : : : / is an isomorphism of the left-
hand side of (7.2) with its right-hand side. t
u
We can now derive the following corollary.
Corollary 1.8. Assume A is a torsion group with p-components Ap , and C is a group
with p-components Cp . Then
Y
Hom.A; C/ Š Hom.Ap ; Cp /:
p
216 7 Homomorphism Groups

Proof. Apply (7.1) and observe that Hom.Ap ; C/ D Hom.Ap ; Cp /: t

u
Example 1.9. For any group A,
Y
Hom.A; Q/ Š Q:
rk0 .A/

Because of (E), the description of Hom.A; Q/ becomes a simple calculation in cardinal arithmetics.
If F is a free subgroup of A generated by a maximal independent system of elements of infinite
order only, then every  W F ! Q extends uniquely to a map ˛ W A ! Q. This amounts to saying
that Hom.F; Q/ Š Hom.A; Q/ naturally. The former Hom is evaluated by using (7.1).

Hom As Bifunctor The correct way of viewing Hom is as a functor

AbAb!Ab associating the group Hom.A; C/ with the ordered pair .A; C/ 2
Ab  Ab. In the balance of this section we investigate the functorial behavior of
Hom.
Let ˛ W A0 ! A and W C ! C0 be fixed homomorphisms. An  2 Hom.A; C/
˛ 
defines a homomorphism A0 ! C0 as the composite A0 !A!C!C0 . The
correspondence  7! ˛ is a homomorphism

Hom.˛; / W Hom.A; C/ ! Hom.A0 ; C0 /;

called the homomorphism induced by ˛ and . Clearly, Hom.1A ; 1C / D 1Hom.A;C/ .

˛0 ˛ 0
Furthermore, if A00 !A0 !A and C!C0 !C00 , then

Hom.˛˛ 0 ; 0 / D Hom.˛ 0 ; 0 / Hom.˛; /:

Evidently, Hom.˛; / is additive in both arguments. Therefore, we can conclude:

Theorem 1.10. Hom is an additive bifunctor Ab  Ab ! Ab, contravariant in the
first and covariant in the second argument. t
u
It is often convenient to use abbreviated notations (provided there is no danger
of confusion):

˛  D Hom.˛; 1C / and  D Hom.1A ; /:

The following result describes the behavior of Hom towards direct and inverse
limits.
Theorem 1.11 (Cartan–Eilenberg [CE]). Assume

A D fAi .i 2 I/I ik g and C D fCj .j 2 J/I j` g

are a direct and an inverse system of groups, respectively, and let A D lim Ai ; C D
!
lim Cj with canonical maps i W Ai ! A and j W C ! Cj . Then


H D fHom.Ai ; Cj / ..i; j/ 2 I  JI Hom.ik ; j` /g

1 Groups of Homomorphisms 217

is an inverse system of groups whose inverse limit is Hom.A; C/ with Hom.i ; j /

as canonical maps.
Proof. It is straightforward to check that H is an inverse system; let H denote its
inverse limit. From the required commutativity of the triangles we can conclude
that there exists a unique map  rendering all triangles

ξ
Hom(A, C) H

Hom(πi, ρj) ξij

Hom(Ai, Cj)

commutative, where the ij are the canonical maps. To show that  is monic, let
 2 Ker . Then ij  D 0; that is, j i D Hom.i ; j / D 0 for all i; j. Thus the
map i W Ai ! C is 0, because all of its jth coordinates are 0, and since [i i Ai D A;
we have  D 0. Q
Any  2 H is of the form  D .: : : ; ij ; : : : / 2 Hom.Ai ; Cj / where the
coordinates ij satisfy the requisite postulates. Define QW A ! C as follows: if
a D i ai , then for this i set a D .: : : ; ij ai ; : : : / 2 Cj . It is straightforward
to verify the independence of a of the choice of i as well as the homomorphism
property of . Considering that ij  D ij and ij  D j i D ij , we must have
 D , showing that  is epic. Thus  is an isomorphism. t
u
Hom and Exact Sequences Next we prove a most frequently used application
of the Hom functor.
˛ ˇ
Theorem 1.12. If 0 ! A!B!C ! 0 is an exact sequence, then so are the
induced sequences

˛ ˇ
0 ! Hom.G; A/! Hom.G; B/! Hom.G; C/ (7.3)

and
ˇ ˛
0 ! Hom.C; G/! Hom.B; G/! Hom.A; G/ (7.4)

for every group G. Equation (7.3) can always be completed to an exact sequence
with ! 0 if G is a free group, and (7.4) if G is a divisible group.
Proof. Let  W G ! A. If ˛ D 0, then ˛ monic implies  D 0, so ˛ is also
monic. Furthermore, ˇ˛ D 0 shows that ˇ ˛ D 0 as well. If  W G ! B is such
that ˇ D 0, then Im 
Ker ˇ D Im ˛, so there is a  W G ! A with  D ˛.
Thus (7.3) is exact. If we continue with ! 0, then exactness at Hom.G; C/ would
mean that for every  W G ! C there is a  W G ! B such that ˇ D , this holds for
free groups G, due to their projective property.
Next let  W C ! G. If ˇ D 0, then  D 0, since ˇ is epic; thus ˇ  is monic.
From ˇ˛ D 0 we obtain ˛  ˇ  D 0. Assume  W B ! G satisfies ˛ D 0. This
218 7 Homomorphism Groups

means that Ker ˇ D Im ˛

Ker , so there is a  W C ! G with  D ˇ. Thus (7.4)
is exact. If G is injective, then for every  W A ! G there is a  W B ! G such that
 D ˛, so ! 0 can be added to (7.4). t
u
More can be said if we start with a pure-exact sequence.
Proposition 1.13 (Fuchs [11]). If the sequence given in the preceding theorem is
pure-exact .p-pure-exact/, then so are (7.3) and (7.4).
Proof. First, let  W G ! B;  W G ! A satisfy n D ˛ .n 2 N/. Thus n
maps G into ˛A, and so Im 
n1 ˛A. By Theorem 2.10 in Chapter 5, n1 ˛A D
˛A ˚ X where nX D 0. If  denotes the projection onto the first summand, then
 D ˛ 1  W G ! A satisfies n D , establishing the first claim.
Next, assume n D ˇ holds for  W B ! G;  W C ! G and n 2 N. Then
.n˛/ D ˇ˛ D 0 shows that n˛A
Ker . Owing to Theorem 2.10 in Chapter 5,
there is a direct decomposition B=.n˛A/ D ˛A=.n˛A/˚B0 =.n˛A/ for some B0
B.
Define  as the composite map B ! B=.n˛A/ ! B0 =.n˛A/ ! G, where the second
map is the canonical projection, while the third is induced by . Clearly, n D n
and ˛A
Ker . Because of this inclusion, there is a homomorphism  W C ! G
such that  D ˇ. Hence n.ˇ/ D n D n D ˇ, completing the proof. t
u
Small Groups A group G is said to be small if there is a natural isomorphism

Hom.G; ˚i2I Ci / Š ˚i2I Hom.G; Ci /

for every set of groups Ci . Equivalently, the image of every homomorphism of G

into an infinite direct sum is already contained in the direct sum of a finite number
of summands. If the groups Ci in the definition are restricted to a class C of groups
(e.g., to torsion-free groups), then G is called C-small. In particular, if C is the direct
sum of copies of G itself, then G is self-small.
Example 1.14. (a) Finitely generated groups are small, while finite rank torsion-free groups are
F -small, where F denotes the class of torsion-free groups.
(b) The quasi-cyclic group Z.p1 / is not small: it has a homomorphic image in ˚@0 Z.p1 / that
has non-zero projection in every summand (elements of order pn have non-zero coordinates
in the first n summands).

(a) Epic images of small groups are small.

(b) A finite direct sum of groups is small if and only if each component is small.
(c) A group is small if and only if it is finitely generated. If G is small, then its
copy in an injective group must be contained in the direct sum of finitely many
summands, whence we infer that G is of finite rank. The torsion subgroup
of G cannot have infinitely many non-zero p-components, nor a quasi-cyclic
summand (see Example 1.14b), so it must be finite. G=tG is a small torsion-free
group, hence all of its torsion homomorphic images have to be small, so finite.
Hence G=tG is a finite extension of a finitely generated free subgroup, so itself
finitely generated.
(d) A torsion-free group is small in the category of torsion-free groups if and only
if it is of finite rank.
1 Groups of Homomorphisms 219

F Notes. The group structure of Hom has been the main topic of numerous investigations.
It is impossible to survey them without the extensive knowledge of the material in later chapters.
Perhaps the most important results are due to Pierce [1] that give a very precise description of Hom
in case of p-groups, making use of Theorem 2.1. No comparable study is expected for torsion-free
groups.
One question which we would like to point out here is concerned with the problem as to what
extent the functor Hom.A; / determines the group A. That A is by no means determined by this
functor was proved by Hill [12] for p-groups and by Sebel’din [2] for torsion-free groups. The
counterexamples are: 1) A D ˚2@0 B and A0 D ˚2@0 B where B D ˚@0 ˚n<! Z.pn / (B is the
torsion-completion of B, see Sect. 3 in Chapter 10); and 2) A D ˚@0 Z ˚ Q and A0 D A ˚ Q: Then
Hom.A; G/ Š Hom.A0 ; G/ holds for all G. Albrecht [7] deals with this question for p-groups and
cotorsion groups.
An important generalization of homomorphism groups is concerned with groups with distin-
guished subgroups. The objects of the category Abn are A D fAI Ai .i < n/g where A 2 Ab, and
A0 ; : : : ; An1 are fixed subgroups of A. If C D fCI Ci .i < n/g is another object in this category,
then  W A ! C is a morphism if  2 Hom.A; C/ such that .Ai /  Ci for all i < n. Results
on homomorphism groups in such categories are instrumental in several questions concerning
ordinary homomorphism groups.

Exercises

(1) Show that Hom.A; C/ is isomorphic to a subgroup of CA .

(2) We have Hom.A; C/ Š Hom.C; A/ and Hom.A; Q=Z/ Š A if both A and C
are finite groups.
(3) If A is torsion-free and C is divisible, then Hom.A; C/ is divisible.
(4) Prove that Hom.A; Z.m// Š Hom.A=mA; Z.m// for all m 2 N.
(5) If C is torsion-free, then Hom.Q; C/ is isomorphic to the maximal divisible
subgroup of C.
(6) If the sequence 0 ! A ! B ! C ! 0 is pure-exact, then (7.3) can be
completed with ! 0 if G is †-cyclic, and (7.4) can so be completed if G is
pure-injective.
(7) (a) If A is a torsion group, then the set union [ Im ˛, taken for all ˛ 2
Hom.A; C/, is a subgroup of C.
(b) The same is not necessarily true if A is torsion-free. [Hint: A of rank 2
with End A Š Z, and C D A ˚ A:]
(8) Prove End Jp Š Jp via the isomorphism End Jp Š lim Hom.Jp ; Z.pn //:
n
(9) Describe the structures of End.˚ Q/ and End.˚ Jp / for a cardinal .
(10) If either A or C is a p-group, then Hom.A; C/ is a Jp -module.
(11) If ˛ 2 Aut A, 2 Aut C, then Hom.˛; / is an automorphism of Hom.A; C/.
(12) (Gerdt) G is small if and only if G
˚i2I Ci implies G
˚i2J Ci for some
finite subset J  I. (Thus it suffices to consider monomorphisms.)
(13) (Gerdt) If D is the class of divisible groups, then D-small groups are small.
220 7 Homomorphism Groups

2 Algebraically Compact Homomorphism Groups

Having considered elementary properties of Hom as well as the exact sequences

involving Homs, we turn our attention to special situations when Hom.A; C/ is of
great interest. We concentrate on cases in which Hom is algebraically compact.
Hom for Torsion Groups We start with the remarkable fact that if A is a
torsion group, then Hom.A; C/ has to be algebraically compact, and hence it can
be characterized by invariants describable in terms of the invariants of A and C.
Theorem 2.1 (Harrison [2], Fuchs [11]). If A is a torsion group, then Hom.A; C/
is a reduced algebraically compact group, for any C.
Proof. It suffices to prove that if A is a p-group, then H D Hom.A; C/ is complete
in its p-adic topology. To show that H is Hausdorff, suppose  2 H is divisible by
every power of p. If a 2 A is of order pk , and if  2 H satisfies pk  D , then
a D pk a D pk a D 0 shows that  D 0. Next, let 1 ; : : : ; n ; : : : be a Cauchy
sequence in the p-adic topology of H; dropping to a subsequence if necessary, we
may assume it is neat: nC1  n 2 pn H for each n; i.e. nC1  n D pn n for some
n 2 H. Let

 D 1 C .2  1 / C    C .nC1  n / C : : : :

This is a well-defined map A ! C, since for a 2 A of order pk , we have

.nC1  n /a D 0 for all n  k, so that the image a D 1 a C .2  1 /a C
   C .k  k1 /a is well defined. Furthermore,

  n D .nC1  n / C .nC2  nC1 / C    D pn .n C pnC1 C : : : /;

where the infinite sum in the parentheses belongs to H. Thus   n 2 pn H, and  is

the limit of the given Cauchy sequence. Consequently, H is complete. t
u
We give a second, shorter proof based on Theorem 1.11. As a torsion group, A
is the direct limit of its finite subgroups Ai . By Theorem 1.11, Hom.A; C/ is then
the inverse limit of the groups Hom.Ai ; C/ which are bounded in view of Sect. 1(F).
Hence Hom.A; C/ is the inverse limit of complete groups, and the assertion follows
from Sect. 2, Exercise 7 in Chapter 6.
The invariants of Hom.A; C/ for p-groups A can be computed, but the computa-
tion is very technical and lengthy, so we just refer the interested reader to Pierce [1];
see also Fuchs [IAG].
Hom for Compact Groups A most interesting case is when the Hom is a
compact group. Next, we take a look at this situation.
We want to give an algebraic characterization of those groups that can carry
a compact group topology. In Sect. 5 in Chapter 6 we have introduced the group
Char G D G O as the group of all continuous homomorphisms of the topological
group G into the circle group T D R=Z, and observed that G O is compact if and
2 Algebraically Compact Homomorphism Groups 221

only if G is discrete. Actually, this is all that we need from the Pontryagin duality to
describe the structure of compact groups:
Proposition 2.2. A group A can carry a compact group topology if and only if it is
of the form

A D Hom.G; T/

for some group G. t

u
Here Hom can be viewed in the algebraic or in the topological sense.
Character Groups Accordingly, our problem became purely algebraic: to
classify the groups of the form Hom.G; T/. Algebraically, T is nothing else than
the direct product of quasi-cyclic groups, one for each prime p. Hence
Y
Char G Š Hom.G; Z.p1 //:
p

Consequently, it suffices to deal with Hom.G; Z.p1 // only.

In describing the structure of this Hom, crucial role is played by the p-basic
subgroups of G. So let us fix a p-basic subgroup B of G, and write

B D ˚1
nD0 Bn where B0 D ˚0 Z; Bn D ˚n Z.pn / for n  1:

Here n .n  0/ are cardinal numbers, uniquely determined by G. The p-component

of G=B is of the form ˚ Z.p1 /; the fact that the cardinal  depends on the choice of
B is not relevant (as we shall see below), but it can be made unique by choosing e.g.
a lower basic subgroup in the p-component of G. Finally, we let  D rk0 .G=B/. A
full characterization of Hom.G; Z.p1 // may be given with the aid of these cardinal
numbers.
Theorem 2.3 (Fuchs [10]). Using the above notation, for any group G we have

Y 1 Y
Y Y Y
Hom.G; Z.p1 // Š Z.p1 / ˚ Z.pn / ˚ Jp ˚ Q: (7.5)
0 nD1 n  @0

Proof. The p-pure exact sequence 0 ! B ! G ! G=B ! 0 induces the p-pure-

exact sequence

0 ! Hom.G=B; Z.p1 // ! Hom.G; Z.p1 // ! Hom.B; Z.p1 // ! 0:

Now Theorem 1.7 shows that

1
Y Y 1 Y
Y
Hom.B; Z.p1 // D Hom.Bn ; Z.p1 // Š Z.p1 / ˚ Z.pn /:
nD0 0 nD1 n
222 7 Homomorphism Groups

If we write G=B D ˚ Z.p1 / Q ˚ H with zero p-component

Q for H, then because
of Hom.˚ Z.p1 /; Z.p1 // Š  End.Z.p1 // Š  Jp , it remains to evaluate
Hom.H; Z.p1 //. The p-pure subgroup L in H generated by a maximal independent
set of elements of infinite order is torsion-free and p-divisible, and H=L is a torsion
group with zero p-component, so L D ˚ Q.p/ . (This group L is not unique, not
even its cardinality is well defined, but this does not influence the outcome.) The
exactness of 0 ! L ! H ! H=L ! 0 implies that of 0 D Hom.H=L; Z.p1 // !
Hom.H; Z.p1 // ! Hom.L; Z.p1 // ! 0, thus we obtain
Y YY
Hom.H; Z.p1 // Š Hom.L; Z.p1 // Š Hom.Q.p/ ; Z.p1 // D . Q/;
  @0

where we have used Example 1.5. We observe that Hom.G=B; Z.p1 // is alge-
braically compact, so its purity in Hom.G; Z.p1 // implies that it is a summand.
This completes the proof. t
u
If we determine the cardinal numbers 0 ; n ; ;  for all primes, then Char G will
be the direct product of groups (7.5) with p ranging over all primes. The group on
the right side of (7.5) does not depend on the choice of , since the second summand
always has a summand that is the product of  0 copies of Jp where  0 D fin rk tB, so
(7.5) has always the product of fin rk tG copies of Jp . A similar comment applies to
the choice of  (see also Theorem 2.6 below).
Observe that the first and the fourth summands in (7.5) come from elements of
infinite order, while the two middle summands from the torsion subgroup of G.
Hence:
Corollary 2.4. Char G is reduced if and only if G is a torsion group, and is divisible
if and only if G is torsion-free. t
u
Since groups G can be found with arbitrarily chosen cardinals n and , for every
prime p, we can conclude:
Corollary 2.5 (Hulanicki [1], Harrison [1]). A reduced group is the character
group of some .torsion/ group exactly if it is the direct product of finite cyclic groups
and groups Jp for .distinct or equal/ primes p. t
u
For divisible groups, a simple inequality must be satisfied.
Theorem 2.6 (Hulanicki [1], Harrison [1]). A divisible group ¤ 0 is the charac-
ter group of some .torsion-free/ group if and only if it is of the form
YY Y
Z.p1 / ˚ Q where   @0 :
p p 

Proof. If G is a torsion-free group, then its rank is, in the above notation, 0 .p/ C
.p/ (the dependence on p must be indicated, but the sum is the same for every
2 Algebraically Compact Homomorphism Groups 223

prime p). This shows that Char G will have the stated form withQ p D 0 .p/, unless
0 .p/ D 0 for every prime p. In this case, the direct sum with @0 Q does not change
the isomorphy class of the first direct product.
Conversely, given a divisible groupP of the stated form, it is an easy exercise to
check that  may be replaced by C p p . This says, in short, that   p may be
assumed. Define G as a direct sum of rational groups Gi such that, for every prime
p, p of them satisfy pGi ¤ Gi and  of them satisfy pGi D Gi . Then Char G will
be as desired. t
u
Example 2.7. (a) For discrete groups Z.p1 /; Q we have Char Z.p1 / Š Jp and Char Q Š R.
(b) For discrete Jp , Char Jp Š Z.p1 / ˚ ˚2@0 Q.

Corollary 2.8 (Kakutani). The character group of a group of infinite cardinality

 is of the power 2 .
Proof. The cardinality of an infinite group G is the sum of the cardinalities used
in Theorem 2.3, taken for all primes p. (7.5) implies that then the group Char G
must have cardinality 2 . t
u
The theorems above are convincing evidence that the algebraic structure of
compact groups is extremely special. The cardinality of the set of all non-isomorphic

groups of infinite cardinality
2 is 22 , but the number of those that can carry a
compact group structure is minuscule. For instance, if  D @˛ with j˛j
@0 , then
there are only countably many, pairwise (algebraically) non-isomorphic groups of
cardinality  that can be compact topological groups, provided we assume GCH.
Indeed, then the cardinal invariants in Theorems 2.3 and 2.6 can be chosen not more
than countably many ways, and they are unique due to GCH.
Example 2.9. This is an example of a group that can carry one and only one compact group
topology: Jp for any cardinal  (it is compact in the finite index topology). In fact, Theorem 2.3
shows that the only discrete group whose character group is Š Jp is the group ˚ Z.p1 / where 
is unique if GCH holds (note that jJp j D 2 ).

In contrast, some groups may be furnished with as many distinct compact

topologies as possible, as is shown by the following theorem:
Theorem 2.10 (Fuchs [10]). For any infinite cardinal , there exist 2 non-
isomorphic compact topological groups of power 2 that are algebraically all
isomorphic.
Proof. In the proof we refer to Corollary 3.8 in Chapter 11 that asserts the
existence of 2 non-isomorphic p-groups of cardinality : they can be chosen with
isomorphic basic subgroups ˚1 nD1 ˚ Z.p /, and they have the same final rank .
n

By virtue
Q of TheoremQ 2.3, their character groups are algebraically isomorphic to
1 Q
nD1  Z.p n
/ ˚  Jp ; however, by the Pontryagin duality theory, they are not
isomorphic as topological groups. t
u
While we are still on the subject of compactness, it is worthwhile pointing out
that Hom preserves (algebraic) compactness in the second argument.
224 7 Homomorphism Groups

Theorem 2.11. If A is .algebraically/ compact, then Hom.G; A/ is .algebraically/

compact for every group G.
Proof. Hom.G; A/ is isomorphic to a subgroup of the group AG of all functions from
G to A. If A is compact, then AG is a compact group in which Hom is a closed subset.
Hence Hom.G; A/ is a compact group. (For the proof of algebraic compactness, use
the summand property.) t
u
Linearly Compact Groups The structure of linearly compact groups (see
Sect. 3 in Chapter 6) is similar to the compact case, though there are some notable
differences. First and foremost is that the Kaplansky duality replaces the Pontryagin
duality.
In Kaplansky’s theory, the duality is established between the category of
linearly compact and the category of discrete p-adic modules, for a fixed prime
p. Thus only those abelian groups are participating in the duality that are also
Jp -modules. The characters are continuous homomorphisms into the discrete p-
adic module Z.p1 /. If M is a discrete p-adic module, then its character module
HomJp .M; Z.p1 // is a linearly compact Jp -module, furnished with the compact-
open topology. On the other hand, if M is a linearly compact Jp -module, then the
continuous homomorphisms of M into Z.p1 / yield a discrete Jp -module.
Example 2.12. The Kaplansky dual of the discrete group Z.p1 / is the linearly compact group Jp ,
and vice versa.

Consequently, the linearly compact p-adic modules are, from the pure algebraic
point of view, nothing else than the groups HomJp .M; Z.p1 // where M ranges over
the class of discrete p-adic modules. Hence, from the proof above on the character
groups one can derive:
Theorem 2.13 (Fuchs [15]). A group admits a linearly compact topology if and
only if it is the direct product of groups of the following types:
(a) Cocyclic groups: Z.pn /; Z.p1 / for any prime p and n 2 N;
(b) The additive group Jp of the p-adic integers and the additive group of the field
Qp of the p-adic numbers, for each prime p. t
u
F Notes. The character group of a discrete left module (over any ring) is a right module that is
compact in the compact-open topology. Theorem 2.11 also extends to modules. It was S. Lefschetz
who introduced linearly compact vector spaces, and later the theory was extended to modules.
Linearly compact modules have an extensive theory.

Exercises

(1) Prove that Hom.Q; Q=Z/ Š Q@0 .

(2) (a) If A is algebraically compact, and if H is a pure subgroup of a group G, then
Hom.G; A/ Š Hom.H; A/ ˚ Hom.G=H; A/.
(b) Show that then Char G Š Char H˚ Char G=H (isomorphism in the
algebraic sense only).
 3 Small Homomorphisms 225

(3) The additive group R of the reals can be furnished with infinitely many distinct
topologies, each yielding non-isomorphic compact groups. [Hint: Char .˚n Q/.]
(4) The group A D ZN =Z.N/ is the character group of ˚@0 .Q ˚ Q=Z/. [Hint:
Exercise 9 in Sect. 3 in Chapter 6.]
(5) If C is a complete group, then Hom.A; C/ is the inverse limit of bounded groups.
(6) (a) Assume A is equipped with a non-discrete compact topology. Then it
has a subgroup (algebraically) isomorphic either to 1) Jp for some p,
or to 2) an infinite direct product of cyclic groups of prime orders.
[Hint: Theorem 2.3.]
(b) A group admits a non-discrete locally compact topology if and only if it has
a subgroup of kind 1) or 2).
(7) (Faltings) Let A be a p-group. Then t.Hom.A; Z.p1 // Š A if and only if A is
torsion-complete with finite UK-invariants.
(8) Calculate the invariants of the algebraically compact group Hom.A; C/ in case
A is †-cyclic and C D ˚ Z.p1 /.
(9) Find the invariants of Hom.A; C/ if A is torsion-free and C D ˚ Z.p1 /. [Hint:
take p-basic in A.]
(10) Give a detailed proof of Theorem 2.13 for linearly compact groups.

3 Small Homomorphisms

This section should be read after getting familiar with the basic material from
Chapter 10; in particular, with large subgroups to be discussed in Sect. 2 there.
Small Homomorphisms Let A; C be p-groups. Following Pierce [1], we call a
homomorphism  W A ! C small if Ker  contains a large subgroup of A. In other
words, the Pierce condition (Sect. 2 in Chapter 10) must be satisfied: given k > 0,
there exists n > 0 such that
pn AŒpk

Ker :

Example 3.1. The map  in the proof of Szele’s theorem 6.10 in Chapter 6 is a small endomor-
phism of the p-group A; its image is a basic subgroup.

(A) Elements of infinite height belong to the kernel of every small homomorphism,
since the first Ulm subgroup is contained in each large subgroup (see Sect. 2(D)
in Chapter 10).
(B) The small homomorphisms  W A ! C form a subgroup of the group
Hom.A; C/. Observe that if pn1 AŒpk

Ker 1 and pn2 AŒpk

Ker 2 , then
pn AŒpk

Ker.1 C 2 / holds with n D maxfn1 ; n2 g. The group of small
homomorphisms will be denoted by Homs .A; C/.
(C) Homs .A; C/ D Hom.A; C/ whenever either A or C is bounded. The latter holds
as pm A is a large subgroup of A for each m > 0.
226 7 Homomorphism Groups

(D) The factor group Hom.A; C/= Homs .A; C/ is torsion-free. This follows from the
fact that if pm  is a small homomorphism for some  W A ! C and m > 0, then
 must be small, as well.
(E) If  W A ! C is a small homomorphism, then B C Ker  D A for any basic
subgroup B of A. This is a consequence of the definition of large subgroups.

Lemma 3.2 (Pierce [1]). Let B be a basic subgroup of the p-group A. There is a
natural isomorphism

Homs .A; C/ ! Homs .B; C/

given by the restriction map:  7!   B where  W A ! C.

Proof. If  W A ! C is a small homomorphism, then (E) shows that, for every a 2 A,
the image a is the same as b if a  b mod Ker  (b 2 B). Hence it is clear that
  B is different for distinct ’s.
Conversely, we show that if we are given a small homomorphism W B ! C,
then we can extend it to a small  W A ! C. By definition, there is a large subgroup
B.u/ of B contained in Ker . Then L D A.u/ is a large subgroup of A such that
L \ B D B.u/ by the purity of B in A. Since A=L D .L C B/=L Š B=.L \ B/ and
there is a natural homomorphism B=.L \ B/ ! B= Ker , we have the composite
map  W A ! A=L ! B= Ker , which is evidently small and coincides with
on B. t
u
Homs As a Summand Perhaps more interesting is that Homs .A; C/ is a
summand of Hom.A; C/. This is demonstrated by the next theorem.
Theorem 3.3 (Pierce [1]). For p-groups A; C, Homs .A; C/ is a direct summand of
Hom.A; C/, complete in its p-adic topology. We have

Hom.A; C/ Š FQ ˚ Homs .A; C/

where FQ is the p-adic completion of a free group F.

Proof. To show that Homs .A; C/ is complete, let 1 ; : : : ; i ; : : : be a neat Cauchy
sequence in Homs .A; C/. It is Cauchy also in Hom.A; C/, thus, by the completeness
of this Hom, it has a limit in Hom.A; C/, which must be D 1 C .2  1 / C
   C .iC1  i / C : : : . It remains to show that is small. By the Cauchy property,
iC1  i D pi i for some i 2 Homs .A; C/. Pick a k 2 N and let pni AŒpk

Ker i for suitable ni 2 N, for each i. Since AŒpk

Ker pk i
Ker pi i whenever
i  k, if we choose n D maxfn0 ; n1 ; : : : ; nk ; kg, then pn AŒpk

Ker.iC1  i / for
all i, showing that is a small homomorphism. This proves that Homs .A; C/ is a
complete group.
Since by (D) Hom.A; C/= Homs .A; C/ is torsion-free, Homs .A; C/ is by algebraic
compactness a summand of Hom.A; C/. A complementary summand is p-adically
complete (as a summand of Hom.A; C/) and torsion-free, so it must be the p-adic
3 Small Homomorphisms 227

completion of a free group. (The rank of F can be computed as  where  ./

denotes the final rank of the basic subgroup of A (resp. C/ if these are infinite.) u
t
The special case A D C leads to the subgroup Ends .A/ of End A consisting
of the small endomorphisms of A. It is a two-sided ideal: that  is small for all
 2 End A whenever  2 Ends A is pretty obvious. That the same holds for  too
follows easily, see, e.g., Exercise 2. Thus End.A/=Ends .A/ is a torsion-free ring on
an algebraically compact group.
Exact Sequence for Homs We now prove an analogue of Theorem 1.12 for
Homs .
˛ ˇ
Proposition 3.4 (Pierce [1]). Let 0 ! A!B!C ! 0 be an exact sequence of
p-groups. Then the induced sequence

˛ ˇ
0 ! Homs .G; A/! Homs .G; B/! Homs .G; C/

is likewise exact for every group G. If the first sequence is pure-exact, then so is the
induced sequence even if we append ! 0 to the end.
Proof. For the first part, the only non-obvious claim is Im ˛  Ker ˇ . If  2
Homs .G; B/ satisfies ˇ D 0, then ˇ.G/ D 0, so G
Im ˛. Hence there is a
 W G ! A such that ˛ D . Since evidently Ker  D Ker  (˛ being monic), we
have  2 Homs .G; A/, as desired.
Assume the given sequence is pure-exact, and  2 Homs .G; C/. If H denotes a
basic subgroup of G, then by Theorem 4.3 in Chapter 5 there is a homomorphism
W H ! B such that ˇ D   H. It is readily checked that the map  W G D
H C Ker  ! B is well defined if we apply to H and send Ker  to 0. Furthermore,
 is small, and ˇ ./ D , so ˇ is surjective. To verify purity, let  2 Homs .G; B/
satisfy pk  2 ˛.Homs .G; A//. Then by Proposition 1.13 there is a W G ! A
satisfying ˛.pk / D pk . From the equality of the kernels we derive that pk is a
small homomorphism. Hence is small as well, and the proof is complete. t
u
Note that we do not claim that the sequence for Homs .; G/ is exact. As Pierce
[1] points out, in contrast to Proposition 3.4, this is not true in general.
F Notes. Megibben [2] shows that an unbounded torsion-complete p-group has a non-small
homomorphism into a separable p-group C if and only if C has an unbounded torsion-complete
subgroup. A result by Monk [2] states that the finite direct decompositions of End A= Ends A are
induced by those of End A, so that they correspond to certain direct decompositions of A.
The concept of small homomorphism has been extended to the torsion-free and mixed cases by
Corner–Göbel [1]. The general version, called inessential homomorphism, is based on the ideal
Ines A of End A; this is the set of all kinds of endomorphisms that are always present in groups
(like those with finite rank images). Interested readers are advised to consult this interesting paper.
228 7 Homomorphism Groups

Exercises

(1) (Pierce) Every small homomorphism  W pk A ! pk C can be extended to a small

homomorphism A ! C.
(2) (Pierce) (a)  W A ! C is small if and only if for every k > 0 there is an n > 0
such that o.a/
pk and h.a/  n imply .a/ D 0.
(b)  is small if and only if, for every k > 0, there is an n > 0 such that
pn a ¤ 0 .a 2 A/ implies o.a/
o.pk a/.
(3) (Pierce) Prove that Homs .A1 ˚ A2 ; C/ Š Homs .A1 ; C/ ˚ Homs .A2 ; C/ and
Homs .A; C1 ˚ C2 / Š Homs .A; C1 / ˚ Homs .A; C2 /.
(4) (Pierce) If G is a pure subgroup of the p-group A, then every small homomor-
phism  W G ! C extends to a small homomorphism A ! C.
(5) (Pierce) Let A; C be arbitrary p-groups. Homs .A; C/ D Hom.A; C/ if either (a)
A has bounded basic subgroups and C is reduced, or (b) C is bounded.
(6) The composite of two small homomorphisms is small.
(7) Suppose A; C are p-groups.  W A ! C is small exactly if .A1 / D 0 and the
induced map A=A1 ! C is small.
(8) Let A be a p-group with unbounded basic subgroup B. Prove Szele’s
theorem 6.10 in Chapter 5 by first mapping B onto itself by a small
endomorphism, and then applying Lemma 3.2.
(9) Let A be a p-group and B a basic subgroup of A. Then, for every p-group C,
the torsion subgroups of Hom.A; C/ and Hom.B; C/ are isomorphic. [Hint: the
elements in these Homs are small homomorphisms.]

Problems to Chapter 7

PROBLEM 7.1. Can the groups Hom.M; Z/ be characterized for monotone

subgroups M of Z@0 ?
See Sect. 2 in Chapter 13 for monotone subgroups.

PROBLEM 7.2. Call  2 Hom.A; B/ a right universal homomorphism for A ! B

if every 2 Hom.A; B/ factors uniquely as D  with  2 End B. It is left
universal if D  with unique  2 End A. Study the cases when uniqueness is
not required, so Hom is singly generated (on the right or on the left) over End.
Right universal homomorphisms, called localizations, were discussed by Dugas [3] for torsion-
free groups. Left universal homomorphisms were completely described by Chachólski–Farjoun–
Göbel–Segev [1] for divisible groups B under the name of cellular cover, and for arbitrary abelian
groups by Fuchs–Göbel [2]. See also Dugas [4].
Chapter 8
Tensor and Torsion Products

Abstract The tensor product of groups is one of the most important concepts and
indispensable tools in the theory of abelian groups. They compete in importance with
homomorphism groups, but their features are totally different.
Tensor products can be introduced in various ways. We define them via generators and defining
relations, and then we show that they have the universal property for bilinear maps. Tensoring is
a bifunctor that is right exact in both arguments. The exact sequence of tensor products is a most
useful asset, both as a tool in proofs and as a device in discovering new facts. Exactness on the left
can be restored by introducing the functor Tor, the torsion product, that is of independent interest.
If one of the groups is a torsion group, then the tensor product can be completely described by
invariants. In particular, the tensor product of two torsion groups is always a direct sum of cyclic
groups. The torsion product behaves differently, it raises more challenging problems. The tensor
product of torsion-free groups is a difficult subject.
Various facts concerning groups that were proved originally in an ad hoc fashion may be
verified more clearly, and perhaps more elegantly, by using homological methods, in particular,
the long exact sequences connecting the tensor and torsion products (as well as Hom and Ext).

1 The Tensor Product

Bilinear Functions and the Tensor Product Suppose A and C are arbitrary
groups, and g is a function defined on the set A  C with values in a group G,
g W A  C ! G. We say that g is a bilinear function if it satisfies

g.a1 C a2 ; c/ D g.a1 ; c/ C g.a2 ; c/ (8.1)

and

g.a; c1 C c2 / D g.a; c1 / C g.a; c2 / (8.2)

for all a; a1 ; a2 2 A; c; c1 ; c2 2 C. It follows at once that a bilinear function obeys

the following simple rules: g.a; 0/ D 0 D g.0; c/; g.na; c/ D ng.a; c/ D g.a; nc/
for all a 2 A; c 2 C, and n 2 Z.
We are going to define a group (that will be denoted by A ˝ C) as well as a
bilinear function e W A  C ! A ˝ C, such that e is the most general bilinear function
in the sense that any bilinear function g W A  C ! G for any group G factors

© Springer International Publishing Switzerland 2015 229

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_8
230 8 Tensor and Torsion Products

uniquely through it: there is a unique group homomorphism  W A ˝ C ! G such

that g D e. This is the universal property of the tensor product.
Define X as the free group on the set A  C; i.e. the free generators of X are the
pairs .a; c/ with a 2 A; c 2 C. Let Y denote the subgroup of X generated by the
elements of the form

.a1 C a2 ; c/  .a1 ; c/  .a2 ; c/ and .a; c1 C c2 /  .a; c1 /  .a; c2 / (8.3)

for all a; a1 ; a2 2 A; c; c1 ; c2 2 C. The tensor product of A and C is defined as

A ˝ C D X=Y:

If we write the coset .a; c/ CPY as a ˝ c, then a typical element of A ˝ C will be

written as a finite sum u D i ki .ai ˝ ci / with ai 2 A; ci 2 C; ki 2 Z, where the
elements are subject to the rules

.a1 C a2 / ˝ c D a1 ˝ c C a2 ˝ c and a ˝ .c1 C c2 / D a ˝ c1 C a ˝ c2 : (8.4)

We also observe that na ˝ c D n.a ˝ c/ D a ˝ nc for all a 2 A; c 2 C; n 2 Z.

As a consequence,
P the elements of A ˝ C can be written in a simplified form as
u D i .ai ˝ ci / where ai 2 A; ci 2 C. However, it should be emphasized that
ordinarily u has many expressions of this form.
The notation a ˝ c is in principle ambiguous: it has a different meaning when
it is considered as an element of A ˝ C or as an element of A0 ˝ C0 for subgroups
A0
A; C0
C containing a and c, respectively (e.g., it can be 0 in A ˝ C, but not
in A0 ˝ C0 ; see (H)). Therefore, one always has to specify to which tensor product
the element a ˝ c belongs, unless its membership is clear from the context.
From the definition it is clear that e W .a; c/ 7! a ˝ c is a bilinear map of A  C
into A ˝ C. If g W A  C ! G is an arbitrary bilinear map, then the correspondence
a ˝ c 7! g.a; c/ extends to a homomorphism  W A ˝ C ! G, since the generators
of A ˝ C are subject only to the relations (8.3) (and to their consequences) that hold
for the respective images in G. Furthermore, it is obvious that no other map  can
satisfy g D e, i.e. given g, there is a unique  making the triangle

A×C
e  g
φ
A⊗C G

commute. We can now state:

Theorem 1.1. Given a pair A; C of groups, there exist a group A˝C and a bilinear
function

eW A  C ! A ˝ C
1 The Tensor Product 231

with the following property: if g W A  C ! G is a bilinear function into any group

G, then there is a unique homomorphism  W A ˝ C ! G making the above triangle
commute. This property determines A ˝ C up to isomorphism.
Proof. It remains to prove the last statement. Assume there are a group B, and a
bilinear function f W A  C ! B with the same properties as A ˝ C and e. By the
first part of the theorem, there is a unique homomorphism  W A ˝ C ! B with
e D f , and by the hypothesis on B, there is a unique W B ! A ˝ C with f D e.
Hence e D e; f D  f , and in view of the uniqueness property, we conclude
that  D 1A˝C and  D 1B , so A ˝ C Š B, as claimed. t
u
As mentioned above, A ˝ C is called the tensor product of A and C, and we will
refer to the map e W A  C ! A ˝ C as the tensor map. From the symmetry of
the roles of A and C in the definition it follows right away that the correspondence
a ˝ c 7! c ˝ a induces a natural isomorphism

A ˝ C Š C ˝ A:

It is easy to check, for arbitrary groups A; B; C; the associative law

.A ˝ B/ ˝ C Š A ˝ .B ˝ C/

either by the universal property or by the equality of generators.

Basic Properties Before stating some relevant properties of tensor products, we
record a simple lemma that is instrumental in exploring tensor products.
Lemma 1.2. Let a 2 A; c 2 C.
(a) If m; n 2 Z such that ma D 0 D nc, then d.a ˝ c/ D 0 where d D gcdfm; ng;
(b) if mja and mc D 0, then a ˝ c D 0;
(c) if mja and njc, then mnj.a ˝ c/.
Proof. (a) If s; t 2 Z such that sm C tn D d, then d.a ˝ c/ D sma ˝ c C tna ˝ c D
sma ˝ c C a ˝ tnc D 0.
(b) If a D ma0 .a0 2 A/, then a ˝ c D ma0 ˝ c D a0 ˝ mc D 0.
(c) If also c D nc0 .c0 2 C/, then a ˝ c D mn.a0 ˝ c0 /. t
u
Here is a short list of the most useful properties of tensor products. (Note how
important an ingredient the universal property is in the proofs of (A)-(B).) The
proofs of (C)-(H) are straightforward, and will be left to the reader.
(A) There is a natural isomorphism Z ˝ C Š C for every group C.P
The elements of Z ˝ C may be
P Pbrought to the simpler form i .ni ˝ ci / D
i .1 ˝ n i c i / D 1 ˝ c for c D i ni ci 2 C. Thus the map  W c 7! 1 ˝ c is an
epimorphism C ! Z ˝ C. Clearly, .m; c/ 7! mc is a bilinear map Z  C ! C,
so the universal property implies that there is a unique W Z ˝ C ! C acting
as .1 ˝ c/ D c. Consequently,  and are inverse to each other.
(B) There is a natural isomorphism Z.m/ ˝ C Š C=mC for every integer m > 0
and group C.
232 8 Tensor and Torsion Products

Here again,  W c 7! 1N ˝ c is an epimorphism C ! Z.m/ ˝ C where 1N D

1 C mZ. We have mC
Ker , since 1 ˝ mc D m ˝ c D 0 ˝ c D 0. Now
.n; c/ 7! nc C mC is a bilinear map Z.m/  C ! C=mC, so by the universal
property, there is a homomorphism W Z.m/ ˝ C ! C=mC such that  is
the canonical map C ! C=mC. Thus Ker  D mC.
(C) The heights satisfy hp .a ˝ c/  hp .a/ C hp .c/ .a 2 A; c 2 C/. Thus if either A
or C is p-divisible (resp. divisible), then so is A ˝ C.
(D) If hp .a/  ! for a 2 A and C is a p-group, then a ˝ c D 0 for every c 2 C.
Thus if A is p-divisible and C is a p-group, then A ˝ C D 0.
(E) If a 2 mA and c 2 CŒm
for some m 2 Z, then a ˝ c D 0.
(F) If either A or C is a p-group .torsion group/, then so is A ˝ C.
(G) If A is a p-group and C is a q-group for different primes p; q, then A ˝ C D 0.
Example 1.3. We have the following natural isomorphisms: Z ˝ Z.n/ Š Z.n/, Z.pn / ˝ Z.pm / Š
Z.pk / with k D minfm; ng, and Z.n/ ˝ Z.m/ Š Z.d/ where d D gcdfm; ng.
Example 1.4. Suppose A is a rational group. Then everyP x 2 A ˝ C can be written in the form
n
x D a˝c with some a 2 A; c 2 C. In fact, as always, x D iD1 .ai ˝ci / holds with ai 2 A; ci 2 C,
but in the present case A is locally cyclic, i.e. there exists an a 2 A such that each ai is an integral
multiple of a, say ai D mi a .mi 2 Z/. Thus
X X X
xD .mi a ˝ ci / D .a ˝ mi ci / D a ˝ . mi ci /
P
where mi ci D c 2 C. This form is not unique: if a D ma0 , then also x D a0 ˝ mc.

(H) It is very important to keep in mind that if B is a subgroup of A, then B ˝ C

need not be a subgroup of A ˝ C. For instance, Z < Q, but Z ˝ Z.p/ Š Z.p/ is
not a subgroup of Q ˝ Z.p/ D 0; or, Z.p/ < Z.p1 /, but Z.p/ ˝ Z.p/ Š Z.p/
is not a subgroup of Z.p1 / ˝ Z.p/ D 0. (We will see later that this cannot
happen if the containment is pure.)

Tensor Product As Bifunctor Let ˛ W A ! A0 ; W C ! C0 be homomorphisms.

Clearly, .a; c/ 7! ˛a˝ c is a bilinear map AC ! A0 ˝C0 , hence there is a unique
homomorphism  W A ˝ C ! A0 ˝ C0 such that .a ˝ c/ D ˛a ˝ c. This map will
be denoted as  D ˛ ˝ , thus

.˛ ˝ /.a ˝ c/ D ˛a ˝ c .a ˝ c 2 A ˝ C; ˛a ˝ c 2 A0 ˝ C0 /:

To avoid complicated notations, we will often write simply

˛ D ˛ ˝ 1C ;  D 1 A ˝

whenever there is no danger of confusion. We also observe:

.˛ ˝ /.˛ 0 ˝ 0 / D ˛˛ 0 ˝ 0 ;

.˛1 C ˛2 / ˝ D .˛1 ˝ / C .˛2 ˝ /; ˛ ˝ . 1 C 2 / D .˛ ˝ 1 / C .˛ ˝ 2 /

1 The Tensor Product 233

for matching homomorphisms ˛; ˛1 ; ˛2 ; ; 1 ; 2 . Hence we can state:

Theorem 1.5 (Cartan–Eilenberg [CE]). The tensor product is an additive bifunc-
tor Ab  Ab ! Ab, covariant in both variables. t
u
In the next lemma we prove an important property: tensor products commute
with direct limits, and, in particular, with direct sums.
Lemma 1.6. The tensor product commutes with direct limits.
j
Proof. Let A D fAi .i 2 I/I i g be a direct system of groups with direct limit A and
canonical maps i W Ai ! A. Then
j
C D fAi ˝ C .i 2 I/I i ˝ 1C g

is also a direct system; let H denote its direct limit, and i W Ai ˝ C ! H the
canonical maps. The homomorphisms i ˝ 1C W Ai ˝ C ! A ˝ C satisfy i ˝ 1C D
j
.j ˝ 1C /.i ˝ 1C / for i
j. By Theorem 1.1, there is a unique  W H ! A ˝ C such
that i D i ˝ 1C for each i. Our claim is that  is an isomorphism. Given a 2 A,
write a D i ai for some ai 2 Ai , and observe that g W .a; c/ ! i .ai ˝ c/ is a well
defined bilinear map A  C ! H (i.e., it does not depend on the choice of i and ai ).
Therefore, there is a homomorphism  W A ˝ C ! H satisfying .a ˝ c/ D g.a; c/.
Now a ˝ c D .i ˝ 1C /.ai ˝ c/ D i .ai ˝ c/ D g.a; c/ shows that  and  are
inverse on the generators of A ˝ C and H. t
u
Corollary 1.7. (i) For all groups Ai ; Cj we have

.˚i2I Ai / ˝ .˚j2J Cj / Š ˚i2I ˚j2J .Ai ˝ Cj /

(ii) In case A D ˚p Ap is a torsion group with p-components Ap , then A ˝ C D

˚p .Ap ˝ C/ for every C. If C is also a torsion group, C D ˚p Cp , then
A ˝ C D ˚p .Ap ˝ Cp /. t
u
Tensor Product and Exact Sequences We have come to study the behavior of
the tensor product towards short exact sequences. In view of the symmetry, we may
confine ourselves to tensoring from one side only. The next theorem shows that ˝
is a right exact functor.
˛ ˇ
Theorem 1.8 (Cartan–Eilenberg [CE]). If 0 ! A!B!C ! 0 is an exact
sequence of groups, then so is the sequence

˛ ˇ
A ˝ G!B ˝ G!C ˝ G ! 0

for every group G.

Proof. Since ˇ˛ D 0, we have ˇ ˛ D .ˇ˛/ D 0. Thus we need only prove
that the homomorphism  W H D .B ˝ G/= Im ˛ ! C ˝ G induced by ˇ is
an isomorphism. Given c 2 C, choose a b 2 B such that ˇb D c. The mapping
234 8 Tensor and Torsion Products

.c; g/ 7! .b ˝ g/ C Im ˛ 2 H is clearly well-defined and bilinear, so there is a

homomorphism W C ˝ G ! H such that .c ˝ g/ D .b ˝ g/ C Im ˛ . It is now
clear that  and are inverse to each other on the generators. t
u
Example 1.9. To show that in Theorem 1.8 the tensored sequence cannot be extended with 0 !
on the left, consider the inclusion map Z.p/ ! Z.p1 /. If it is tensored with Z.p/, then the induced
map Z.p/ ˝ Z.p/ D Z.p/ ! Z.p1 / ˝ Z.p/ D 0 is no longer monic. (See also (H).)
Example 1.10. We show that A D Jp ˝Jp contains an uncountable divisible torsion-free subgroup.
@
The exact sequence 0 ! Z.p/ ! Jp ! Q.2 0 / ! 0 induces the exact sequence 0 ! Z.p/ ˝ Jp !
.2@0 /
Jp ˝ Jp ! Q ˝ Jp ! 0 (Theorem 3.5 will show that tensoring with a torsion-free group
preserves exactness). From the exact sequence 0 ! Z ! Z.p/ ! ˚q¤p Z.q1 / ! 0 it follows
@
that Z.p/ ˝ Jp Š Jp , so algebraically compact, while Q.2 0 / ˝ Jp is a large torsion-free divisible
@
group. Hence the tensored sequence splits, and the claim follows: Jp ˝ Jp Š Jp ˚ Q.2 0 / .

A repeated applications of the last theorem shows that if ˇ W B ! C and ˇ 0 W

B ! C0 are epimorphisms, then so is ˇ ˝ ˇ 0 W B ˝ B0 ! C ˝ C0 . With a somewhat
0

greater effort one can prove a little bit more:

˛ ˇ ˛0 ˇ0
Corollary 1.11. Let A!B!C ! 0 and A0 !B0 !C0 !0 be exact sequences.
Then the sequence

 ˇ˝ˇ 0
.A ˝ B0 / ˚ .B ˝ A0 /!B ˝ B0 !C ˝ C0 ! 0

is exact where  D rŒ.˛ ˝ 1B0 / ˚ .1B ˝ ˛ 0 /

.
Proof. Consider the following commutative diagram with exact rows and columns:

B ⊗ A −−−−→ C ⊗ A
⏐ ⏐
⏐ ⏐
1B ⊗α  (1C ⊗α )
α⊗1  β⊗1 
A ⊗ B  −−−−B→ B ⊗ B  −−−−B→ C ⊗ B  −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
1A ⊗β   1B ⊗β   1C ⊗β 
α⊗1  β⊗1 
A ⊗ C  −−−−C→ B ⊗ C  −−−−C→ C ⊗ C  −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
0 0 0

Hence it is clear that the map ˇ ˝ ˇ 0 D .1C ˝ ˇ 0 /.ˇ ˝ 1B0 / is surjective, and it is
easy to verify by simple diagram chasing that its kernel is the union of Ker.ˇ ˝ 1B0 /
and Ker.1B ˝ ˇ 0 /. t
u
Vanishing
P Elements in Tensor Products An immediate corollary to Lemma 1.6
is that if niD1 .ai ˝ ci / vanishes in the tensor product A ˝ C, then there are
1 The Tensor Product 235

finitely
Pn generated subgroups A0
A; C0
C with ai 2 A0 ; ci 2 C0 such that
0 0
iD1 .ai ˝ ci / D 0 as an element of A ˝ C . Actually, we can obtain more detailed
information about what forces an element of a tensor product to become 0.
Lemma 1.12. In the tensor product A ˝ C of the groups A; C, a relation

X
n
.ai ˝ ci / D 0 .ai 2 A; ci 2 C/
iD1

holds if and only if there exist elements bj 2 C .j D 1; : : : ; m/ as well as integers

rij 2 Z .i D 1; : : : ; nI j D 1; : : : ; m/ such that

X
m X
n
ci D rij bj for all i and rij ai D 0 for all j:
jD1 iD1

Proof. Sufficiency follows from a simple computation:

X
n X
n X
m X
m X
n
.ai ˝ ci / D Œai ˝ . rij bj /
D Œ. rij ai / ˝ bj
D 0:
iD1 iD1 jD1 jD1 iD1

P
For the converse, assume niD1 .ai ˝ ci / D 0. Let F D ˚k2I hxk i be a free group,
and W F ! A an epimorphism such that xi 7! ai for i D 1; : P : : ; n (we assume that
these indices are in I). Hypothesis implies that the element niD1 .xi ˝ ci / in the
second term of the exact sequence Ker  ˝ C ! F ˝ C ! A ˝ C ! 0 belongs to
the image
P of the first map.
P Hence there exist elements P yj 2 Ker  and bj 2 C such
that niD1 .xi ˝ ci / D m jD1 .y j ˝ b j /: Writing y j D k rkj xk .rkj 2 Z/, we get

X
n XX
m XX
m
.xi ˝ ci / D .xk rkj ˝ bj / D .xk ˝ rkj bj /:
iD1 k jD1 k jD1

Pm Pn
Since the xi are free generators, hence we obtain ci D jD1 rij bj and iD1 rij ai D
yj D 0: t
u
For later application we point out that the elements bj in the proof may be chosen
in any preassigned pure subgroup C0 of C that contains the ci . InPfact, it will follow
from Theorem 3.1 that A ˝ C0 ! A ˝ C is a monic map, so if .ai ˝ ci / vanishes
in A ˝ C, then it must be 0 in A ˝ C0 .
A Natural Isomorphism A most important natural isomorphism connecting
Hom with the tensor product is the content of the following theorem.
Theorem 1.13. For any three groups A; B; C, there is a natural isomorphism

 W Hom.A ˝ B; C/ Š Hom.A; Hom.B; C//;

236 8 Tensor and Torsion Products

acting as follows: if  W A ˝ B ! C, then

Œ./a
.b/ D .a ˝ b/ for all a 2 A; b 2 B:

Proof. Clearly, ./ assigns to a 2 A a map ./a W B ! C that sends b 2 B to

.a ˝ b/ 2 C: The homomorphism properties of ./a are rather obvious. It is also
clear that the correspondence  W  7! ./a is a homomorphism, so it remains to
find an inverse to . Pick a  W A ! Hom.B; C/, and observe that .a; b/ 7! .a/.b/
is a bilinear function g W A  B ! C: Hence there exists a unique  W A ˝ B ! C
such that .a ˝ b/ D .a/.b/: The mapping W  7!  is readily seen to be inverse
to . t
u
F Notes. Tensor products for groups were introduced by H. Whitney. It is one of the
fundamental functors in Homological Algebra and in Category Theory. Tensor products A ˝R C
for modules over a ring R are defined when A is a right and C is a left R-module, and in
forming Y, the above generators are complemented with those of the form .ar; c/  .a; rc/ for
all a 2 A; c 2 C; r 2 R. Theorem 1.13 is an important result: it claims that the tensor product and
the Hom are adjoint functors.

Exercises

(1) The bilinear functions A  C ! G form a group under addition. This group
is Š Hom.A ˝ C; G/:
(2) There is a (non-natural) isomorphism A ˝ C Š Hom.A; C/ for finite groups
A; C.
(3) Characterize the groups C that satisfy in turn the following conditions:
(a) A ˝ C Š A for every group A;
(b) A ˝ C Š A=mA with a fixed integer m, for every A;
(c) A ˝ C is a divisible group for every A;
(d) A ˝ C D 0 for every torsion group A;
(e) A ˝ C D 0 for every p-group A;
(f) A ˝ C is a torsion group for every A.
(4) Describe the structure of the following tensor products: (a) Jp ˝QI (b) Jp ˝Jq
for primes p ¤ q; (c) Jp ˝ C for a torsion group C.
(5) (a) If a 2 A; c 2 C are elements of infinite order, then the same holds for
a ˝ c in A ˝ C.
(b) Let fai gi2I and fcj gj2J be maximal independent sets in the torsion-free
groups A and C, respectively. Show that fai ˝ cj gi2I;j2J is a maximal
independent set in A ˝ C.
(c) We have rk0 .A ˝ C/ D rk0 .A/  rk0 .C/ for any groups A; C.
2 The Torsion Product 237

(6) (a) If A is torsion-free, then Q ˝ A is naturally isomorphic to the divisible

hull of A; a natural embedding being given by a 7! 1 ˝ a.
(b) Claim (a) is false if A has elements ¤ 0 of finite order.
(7) Assume that the element a 2 A is such that a ˝ c D 0 in A ˝ C for all torsion
groups C and for all c 2 C. Prove that a 2 A1 .
(8) Define multilinear functions g W A1      An ! G, the corresponding tensor
product A1 ˝    ˝ An , and the tensor map e for multilinear functions. Prove
the universal property and the associativity relations

.A1 ˝ A2 / ˝ A3 Š A1 ˝ A2 ˝ A3 Š A1 ˝ .A2 ˝ A3 /:

(9) There is a natural homomorphism

Y Y
A˝ Ci ! .A ˝ Ci /
i i

which in general fails to be an isomorphism. [Hint: A D Q; Ci unbounded

torsion groups. See Mittag-Leffler groups in Sect. 8 in Chapter 3.]
(10) If ˛ 2 End A, 2 End C, then ˛ ˝ 1C and 1A ˝ are commuting
endomorphisms of A ˝ C.
(11) (a) Assume A and C are torsion-free groups, and pt ja ˝ c holds for some
a 2 A; c 2 C, prime p, and integer t > 0. Show that there are integers
r; s  0 such that pr ja; psjc with r C s D t.
(b) Using the notion of type of torsion-free groups to be introduced in Sect. 1
in Chapter 12, find the type of A ˝ C if A; C are rank one torsion-free
groups.
(12) (Corner) (a) If A ˝ C is p-divisible, then one of A; C is p-divisible. [Hint:
argue with p-basic subgroups.]
(b) Let A be a torsion group. A ˝ C D 0 if and only if (i) pA ¤ A implies
p.tC/ D tC, and (ii) Ap ¤ 0 implies p.C=tC/ D C=tC.
(13) (Head) Let B be a subgroup of a p-group A, and C a reduced group. The
canonical map B ˝ C ! A ˝ C is monic exactly if B \ pn A D pn B holds
whenever C has a cyclic summand of order pn .

2 The Torsion Product

In Theorem 1.8 it was proved that the tensor product is right exact, and Example 1.9
shows that it is not left exact. Now the question arises whether or not it is possible to
salvage some sort of exactness by introducing another functor. It is our next goal to
define a bifunctor that leads to a long exact sequence continuing the tensor sequence
to the left.
Torsion Products The torsion product Tor.A; C/ of the groups A; C is defined
as the group generated by the triples .a; m; c/ with a 2 A; c 2 C; m 2 N such that
ma D 0 D mc, subject to the following relations:
238 8 Tensor and Torsion Products

.a1 C a2 ; m; c/ D .a1 ; m; c/ C .a2 ; m; c/ where ma1 D ma2 D mc D 0;

.a; m; c1 C c2 / D .a; m; c1 / C .a; m; c2 / where ma D mc1 D mc2 D 0;
.a; mn; c/ D .na; m; c/ where mna D mc D 0;
.a; mn; c/ D .a; m; nc/ where ma D mnc D 0

for all a; a1 ; a2 2 A; c; c1 ; c2 2 C; m; n 2 N. (Thus the relations are assumed to hold

whenever the right-hand sides are defined.) In view of the apparent symmetry, we
have

Tor.A; C/ Š Tor.C; A/ for all A; C;

this being a natural isomorphism via .a; m; c/ $ .c; m; a/.

The first two relations above imply .0; m; c/ D 0 D .a; m; 0/; furthermore,
.na; m; c/ D n.a; m; c/ D .a; m; nc/ for each n 2 Z whenever theP symbols are
meaningful. Thus every element x of Tor.A; C/ is a finite sum x D i .ai ; mi ; ci /
with mi ai D 0 D mi ci .ai 2 A; ci 2 C; mi 2 N/, so there exist finite subgroups
A0
A; C0
C such that x 2 Tor.A0 ; C0 /
Tor.A; C/. Note that m.a; m; c/ D 0, so
Tor is always a torsion group (as its name indicates).
We list a number of elementary properties of Tor.
(A) There is a natural isomorphism Tor.A; C/ Š Tor.t.A/; t.C//. This follows from
the fact that in the definition of Tor, only elements of finite order are involved.
In particular, Tor.A; C/ D 0 if either A or C is torsion-free.
(B) If nA D 0 for some n 2 Z, then n Tor.A; C/ D 0 for every C. Clearly, all the
generators of Tor are now annihilated by n.
(C) Tor.A; C/ D 0 if A is p-group and C is a q-group for different primes p; q.
(D) Multiplication by an integer n on A or on C induces multiplication by the same
n on Tor.A; C/. This is an immediate consequence of the additivity of Tor.
(E) If C D hci is a cyclic group of order n, then there is a natural isomorphism
Tor.A; C/ Š AŒn
. Now, every x 2 Tor.A; C/ is annihilated by n, so x can be
written in the form x D .a; n; c/ with na D 0. In this special case, the elements
of Tor have unique forms, since .a; n; c/ D 0 follows from the relations only if
a D 0. Hence Tor.A; C/ Š AŒn
.
(F) We have Tor.Z.pk /; Z.p` // Š Z.pminfk;`g / and Tor.Z.pk /; Z.q` // D 0 if p; q
are distinct primes. This follows from (E) via a straightforward calculation.
Tor As Bifunctor In order to examine the behavior of Tor as a bifunctor
Ab  Ab ! Ab, choose homomorphisms ˛ W A ! A0 ; W C ! C0 . Evidently,
if .a; m; c/ is a generator of Tor.A; C/, then .˛a; m; c/ will be a generator of
Tor.A0 ; C0 /. Moreover, the function .a; m; c/ 7! .˛a; m; c/ between the generators
extends uniquely to a group homomorphism

Tor.˛; / W Tor.A; C/ ! Tor.A0 ; C0 /:

2 The Torsion Product 239

It is pretty obvious that Tor.˛; / Tor.˛ 0 ; 0 / D Tor.˛˛ 0 ; 0 / for matching

homomorphisms, as well as Tor.1A ; 1C / D 1Tor.A;C/ . Hence we conclude:
Theorem 2.1 (Cartan–Eilenberg [CE]). The torsion product is an additive
bifunctor Ab  Ab ! Ab, covariant in both variables. t
u
Next we prove the important result:
Theorem 2.2. Tor commutes with direct limits .and so also with direct sums/:

Tor.lim Ai ; C/ D lim Tor.Ai ; C/:

! !
i i

j
Proof. Let A D fAi .i 2 I/I i g be a direct system of groups with direct
limit A, and canonical maps i W Ai ! A. We get the direct system C D
j
fTor.Ai ; C/ .i 2 I/I Tor.i ; 1C /g; let T denote its direct limit, and i the canonical
maps Tor.Ai ; C/ ! T. The homomorphisms Tor.i ; 1C / W Tor.Ai ; C/ ! Tor.A; C/
j
satisfy Tor.i ; 1C / D Tor.j ; 1C / Tor.i ; 1C / for i
j. T being a direct limit, there
is a unique  W T ! Tor.A; C/ such that i D Tor.i ; 1C /. To show that  is an
isomorphism, let .a; m; c/ 2 Tor.A; C/ where of course ma D 0 D mc. We can write
a D i ai for some ai 2 Ai with mai D 0, whence Tor.i ; 1C / W .ai ; m; c/ 7! .a; m; c/
shows that  is a surjective map. If x 2 Ker , then there is an index i 2 I with
i y D x for some y 2 Tor.Ai ; C/. Now Tor.i ; 1C /y D i y D x D 0 implies
j
the existence of a j  i such that Tor.i ; 1C /y D 0. Apply j , and notice that
j
j Tor.i ; 1C / D i to conclude that x D i y D 0: This proves that  is an
isomorphism. t
u
An immediate consequence of the preceding theorem, combined with (A), is that
if A; C are any groups, then Tor.A; C/ Š ˚p Tor.Ap ; Cp / holds for the p-components
Ap and Cp .
(G) For any three groups A; B; C, there is a natural isomorphism

Tor.A; Tor.B; C// Š Tor.Tor.A; B/; C/:

We can represent both sides as direct limits of the respective finitely generated
subgroups, since Tor commutes with direct limits. Consequently, it suffices to verify
the isomorphism for finitely generated groups. These are †-cyclic groups, Tor
commutes with direct sums, so a further reduction to cyclic groups is immediate. If
one of the groups is infinite cyclic or if not all of A; B; C belong to the same prime,
then both sides are 0. If A; B; C are cyclic of orders pk ; p` ; pm , respectively, then a
quick calculation (using (E)) shows that both sides are isomorphic to Z.pn / with
n D minfk; `; mg. Hence the stated isomorphism is evident. It is easy to check that
the isomorphism is natural for cyclic groups, and direct limits preserve the naturality
of maps.
The following natural isomorphism makes it possible to identify the torsion part
of a group via Tor.
240 8 Tensor and Torsion Products

Theorem 2.3. For every group C, we have

Tor.Q=Z; C/ Š t.C/:

Proof. We start observing that Q=Z is the direct limit of the direct system A D
fZ.m/I mn g where mn W Z.m/ ! Z.n/ is the natural injection for mjn, acting as
1 C mZ 7! nm1 C nZ. From Theorem 2.2 we conclude that Tor.Q=Z; C/ is the
direct limit of the direct system C D fTor.Z.m/; C/I Tor.mn ; 1C /g. In view of (E),
we can replace Tor.Z.m/; C/ by CŒm
, and Tor.mn ; 1C / becomes the inclusion map
CŒm
! CŒn
, as is easily verified. Thus the direct limit will be the union of the
CŒm
for all m > 0, which is exactly t.C/. t
u
Tor and the Tensor Product Let
˛ ˇ
0 ! A!B!C ! 0 (8.5)

be an exact sequence. Define the connecting homomorphism

ı W Tor.G; C/ ! G ˝ A

as follows: if .g; m; c/ 2 Tor.G; C/ and if ˇb D c .b 2 B/, then let ı W .g; m; c/ 7!

g ˝ a where ˛a D mb .a 2 A/. Thus ı is defined on the set of generators, and in
view of the relations in Tor and in ˝, it extends to all of Tor.G; C/. Importantly,
the definition is independent of the choice of b; a, because if b0 ; a0 are other choices,
then b0 D b C ˛x .x 2 A/; a0 D a C mx, and so g ˝ a0 D g ˝ a C g ˝ mx where
g ˝ mx D mg ˝ x D 0.
The exactness of the sequence in the following theorem is one of the most
important results.
Theorem 2.4 (Cartan–Eilenberg [CE]). If .8:5/ is exact, then so is for any group
G the sequence

˛ ˇ ı
0 ! Tor.G; A/! Tor.G; B/! Tor.G; C/!
1G ˝˛ 1G ˝ˇ
! G ˝ A ! G ˝ B ! G ˝ C ! 0:

Proof. Since both tensor and torsion products commute with direct limits, and since
direct limits of exact sequences are exact, it will be enough to verify exactness for
G Š Z and G Š Z.m/ .m 2 N/ only. The first case is trivial, since it just leaves the
given exact sequence (8.5) unchanged (all Tors are 0). If G D hgi Š Z.m/, then the
elements of the Tors are of the form .g; m; x/ with mx D 0, while those in the tensor
products are of the form g ˝ x with x taken mod mX (here X D A; B; C). That
the composite of any two maps is 0 is clear except at places where ı is involved.
Then ıˇ .g; m; b/ D ı.g; m; ˇb/ D g ˝ a D 0, because ˛a D mb D 0, and
.1G ˝ ˛/ı.g; m; c/ D .1G ˝ ˛/.g ˝ a/ D .1G ˝ ˛a/ D 0 as ˛a 2 mB.
2 The Torsion Product 241

It remains to prove that the kernels are included in previous images for the first
three maps. If ˛ .g; m; a/ D .g; m; ˛a/ D 0, then ˛a D 0 and a D 0, thus ˛
is monic. If ˇ.g; m; b/ D 0, then similarly ˇb D 0, thus some a 2 A satisfies
˛a D b with ma D 0. Hence ˛ .g; m; a/ D .g; m; b/, and the exactness at the
second Tor follows. Next, suppose ı.g; m; c/ D 0 where mc D 0 .c 2 C/. Then
with the above notation g ˝ a D 0, so a 2 mA, i.e. a D ma0 for some a0 2 A. Thus
˛ma0 D ˛a D mb, i.e. m.b  ˛a0 / D 0. Hence .a; m; ˇ.b  ˛a0 // 2 Tor.G; B/ maps
upon .g; m; c/. Finally, if .1A ˝ ˛/.g ˝ a/ D 0, then ˛a D mb for some b 2 B, and
for this b we have ı.g; m; ˇb/ D g ˝ a. This implies exactness at G ˝ A, completing
the proof. t
u
By the way, if (8.5) is chosen as a free resolution of C, say, 0 ! H ! F ! C !
0 with free group F, then Tor.G; C/ can be defined as the kernel of the induced map
G ˝ H ! G ˝ F. This interpretation leads to an alternative set of generators and
defining relations for Tor.G; C/.
F Notes. The torsion product Tor.A; C/ is defined over any ring R if A is a right and C is a
left module. Theorem 2.4 carries over, but the left 0 has to be removed and the sequence extended
n .; /.
to the left by terms of higher functors TorR
S.E. Dickson [Trans. Amer. Math. Soc. 121, 223–235 (1966)] defines torsion theory for
abelian categories A as a pair .T ; F / of classes in A such that T ; F are maximal with respect
to the property HomA .T; F/ D 0 for all T 2 T ; F 2 F . T is closed under homomorphic images,
direct sums and extensions, while F is closed under taking kernels, direct products and extensions.
An example for a torsion theory in Ab is (besides the obvious one) when T is the class of divisible,
and F is the class of reduced groups.
A more general concept is the radical. A functor R W Ab ! Ab defines a radical in Ab if
R.A/  A and R.A=R.A// D 0 for all A 2 Ab. E.g. p (for any ordinal  ) is a radical functor in the
category of torsion groups. More generally, if X is any class of groups, then RX .A/ D \ Ker 
with  2 Hom.A; X/ .8 X 2 X / is a radical functor. An idempotent radical defines a torsion
theory.

Exercises

(1) Prove that .a; m1 C m2 ; c/ D .a; m1 ; c/ C .a; m2 ; c/ holds for the generators of
Tor whenever the right hand side makes sense.
(2) For finite groups A; C, Tor.A; C/ Š A ˝ C is a (non-natural) isomorphism.
(3) Assume .a; pn ; c1 / D .a; pn ; c2 / in Tor.A; C/ where A D hai Š Z.pn / and
o.c1 / D o.c2 / D pn . Prove that c1 D c2 .
(4) (a) A is torsion-free if and only if Tor.A; C/ D 0 for every group C.
(b) Tor.A; A/ D 0 implies that A is torsion-free.
(5) If A is torsion-free, then A ˝ Tor.B; C/ Š Tor.A ˝ B; C/: [Hint: direct limit.]
(6) For all A; C, and m 2 Z, one has Tor.A ˝ Z.m/; C/ Š A ˝ Tor.Z.m/; C/ and
Tor.Z.m/; A/ ˝ Tor.Z.m/; C/ Š Tor.Tor.Z.m/; A/; C/:
(7) If A satisfies Tor.A; C/ Š t.C/ for every group C, then A Š Q=Z ˚ G with G
torsion-free.
242 8 Tensor and Torsion Products

(8) Formulate and prove the dual of Corollary 1.11 for Tor.
(9) Give counterexamples to show that Tor Q does not commute with (a) direct
products; (b) inverse limits. [Hint: A D p Z.p1 /.]

3 Theorems on Tensor Products

We wonder under what conditions the exactness of a short exact sequence is

preserved when tensoring it with a group G or when forming the torsion product
with G. We offer two independent conditions, one relates the sequence itself (pure-
exactness), while the other specifies the groups G (torsion-freeness).
Tensor Product and Pure-Exact Sequences We verify a frequently used
property of pure-exact sequences.
Theorem 3.1 (Harrison [1], Fuchs [11]). If 0 ! A ! B ! C ! 0 is a pure-
exact sequence, then so are the sequences

0!A˝G!B˝G!C˝G!0 (8.6)

and

0 ! Tor.A; G/ ! Tor.B; G/ ! Tor.C; G/ ! 0 (8.7)

for every group G.

Proof. Pure-exactness for both sequences follows from Theorem 3.1 or Corol-
lary 3.6 in Chapter 5, since the claim is true for cyclic groups, and both tensor
and torsion products commute with direct limits. t
u
An important consequence of this theorem is that if A0 ; C0 are pure subgroups in
A; C, respectively, then A0 ˝C0 may be regarded as a (pure) subgroup of A˝C under
the natural map. In other words, a0 ˝ c0 .a0 2 A0 ; c0 2 C0 / is the same if viewed as
an element of A0 ˝ C0 or as one of A ˝ C. Needless to say, the same conclusion can
be stated for p-purity.
Lemma 3.2 (Fuchs [11]). If B is a p-basic subgroup of A, and C is a p-group, then
there is a natural isomorphism

A ˝ C Š B ˝ C:

Proof. We start with the p-pure-exact sequence 0 ! B ! A ! A=B ! 0, and ten-

sor it with C to obtain the exact sequence 0 ! B ˝ C ! A ˝ C ! .A=B/ ˝ C ! 0.
Since A=B is p-divisible and C is a p-group, their tensor product vanishes. The
exactness of 0 ! B ˝ C ! A ˝ C ! 0 amounts to the claim in the lemma. t
u
3 Theorems on Tensor Products 243

Tensor Products of Torsion Groups The last result enables us to determine

explicitly the tensor product of two groups if one of them is a torsion group. Let
Bp denote a p-basic subgroup of A, and Cp the p-component of the torsion group C,
then

A ˝ C Š ˚p .A ˝ Cp / Š ˚p .Bp ˝ Cp /: (8.8)

The last tensor products are easily computed as the groups Bp are †-cyclic. Note
that these isomorphisms also show that, for a torsion group C, we have A ˝ C Š
.t.A/ ˝ C/ ˚ .A=t.A/ ˝ C/: More generally,
Lemma 3.3 (Harrison [2]). If B is a pure subgroup of A, and if C is a torsion
group, then

A ˝ C Š ŒB ˝ C
˚ Œ.A=B/ ˝ C
:

Proof. Referring again to the direct limit property, the proof can be simplified to the
special case C Š Z.pk / of cocylic groups. If k is an integer, then we know that the
pure-exact sequence 0 ! B ! A ! A=B ! 0 splits when tensored with a finite
cyclic group. If k D 1, then (8.6) along with the injectivity of B ˝ Z.p1 / implies
the claim. t
u
If both A and C are torsion groups, then their tensor product is very simple to
describe:
Theorem 3.4 (Harrison [1], Fuchs [6]). The tensor product of torsion groups is a
direct sum of finite cyclic groups.
Proof. If both A and C are torsion groups, then Lemma 3.2 permits us to replace
Cp in (8.8) by its basic subgroup, so Bp ˝ Cp will be a direct sum of finite cyclic
groups. t
u
Tensor Products and Torsion-Free Groups The other case when exactness is
preserved under tensor products is recorded in the following theorem.
Theorem 3.5 (Dieudonné). Suppose 0 ! A ! B ! C ! 0 is an exact sequence.
If G is a torsion-free group, then the induced sequence .8:6/ is also exact.
Hence the tensor product of torsion-free groups is again torsion-free.
Proof. As Tor.G; / D 0 for torsion-free G, the claim follows at once from Theo-
rem 2.4. t
u
Very little is known about the structure of the tensor product of torsion-
free groups, except when the groups are completely decomposable. In this case,
complete information is available about the tensor product, based on the tensor
product of two rank one torsion-free groups; see Chapter 12. A very modest result
is related to p-basic subgroups.
244 8 Tensor and Torsion Products

Lemma 3.6. Let A and C be torsion-free groups with p-basic subgroups B and D,
respectively. Then A ˝ C is torsion-free, and its p-basic subgroups are isomorphic
to B ˝ D.
Proof. From the p-pure-exact sequence 0 ! B ! A ! A=B ! 0 we deduce the
p-pure-exact sequence 0 ! B ˝ C ! A ˝ C ! .A=B/ ˝ C ! 0. As .A=B/ ˝ C
is p-divisible and torsion-free, the p-basic subgroups of B ˝ C are also p-basic in
A ˝ C. A repetition of this argument leads to the conclusion that B ˝ D is p-basic
in A ˝ C. t
u
As far as mixed groups are concerned, we are able to get some information about
their torsion and torsion-free parts.
Theorem 3.7. For any groups A; C, there are isomorphisms

t.A ˝ C/ Š Œt.A/ ˝ t.C/

˚ Œt.A/ ˝ C=t.C/
˚ ŒA=t.A/ ˝ t.C/
;
.A ˝ C/=t.A ˝ C/ Š A=t.A/ ˝ C=t.C/:

Proof. It is easily checked that the kernel of the natural epimorphism A ˝ C !

A=t.A/ ˝ C=t.C/ is generated by A ˝ t.C/ and t.A/ ˝ C (which are subgroups of
A ˝ C; cf. Theorem 3.3). Evidently, A ˝ t.C/ Š Œt.A/ ˝ t.C/
˚ ŒA=t.A/ ˝ t.C/
,
and t.A/ ˝ C Š Œt.A/ ˝ t.C/
˚ Œt.A/ ˝ C=t.C/
are torsion groups. t
u
F Notes. Theorem 3.5 sounds trivial, but one should not forget that it was published before
the powerful homological machinery existed. It is a special case of a general theorem on modules
that states that exactness is preserved by tensoring a short exact sequence with a flat module. By the
way, flat modules are defined by the property that Tor vanishes identically for them. (Flat abelian
group is the same as torsion-free group.)
Tensor products need not be reduced even if the components are reduced as is shown
by Example 1.10. A. Fomin [1] studied the tensor powers of torsion-free groups A of rank n and 
p-corank k. He found that A˝  ˝A .k C1 factors) contains a p-divisible subgroup of rank kC1 n
.

Exercises

(1) The following are equivalent:

(a) The sequence 0 ! A ! B ! C ! 0 is pure-exact;
(a) 0 ! A ˝ G ! B ˝ G ! C ˝ G ! 0 is exact for every G;
(b) 0 ! Tor.A; G/ ! Tor.B; G/ ! Tor.C; G/ ! 0 is exact for each G.
(2) (a) (Harrison) If 0 ! A ! B ! C ! 0 is a pure-exact sequence, then
0 ! A ˝ G ! B ˝ G ! C ˝ G ! 0 is splitting exact for every torsion G.
(b) If a group G has property (a) for every pure-exact sequence, then it must be
torsion. [Hint: B free and C D Q.]
4 Theorems on Torsion Products 245

(3) (a) If A ˝ C contains a copy of Z.p1 /, then either A or C has a subgroup

Š Z.p1 /.
(b) Examine when A ˝ C Š Z.p1 /.
(4) (a) Prove that A ˝ C Š Z implies A Š C Š Z.
(b) If A ˝ C is a non-trivial free group, then both A and C are free.
(5) If A is torsion-free, then AQ Š Hom.Q=Z; A ˝ Q=Z/ (Z-adic completion).
(6) If the sequence 0 ! A ! B ! C ! 0 is pure-exact, then the same holds
for the sequence 0 ! tA ! tB ! tC ! 0 of torsion subgroups. [Hint:
Tor.Z.p1 /; /.]

4 Theorems on Torsion Products

While tensoring with a p-group implements drastic structural simplification, this is

not the case for the torsion product, though we may notice some smoothing effect.
In this section, we explore some interesting features of Tor.
In order to get more information about Tor, we want to take advantage of its
left exactness. The first and foremost fact to be observed is that if A0
A and
C0
C, then Tor.A0 ; C0 /
Tor.A; C/. This is a powerful property that will be used
throughout without mentioning it explicitly.
Elementary Facts Our first concern is how Tor behaves towards multiplication
by integers.
Lemma 4.1 (Nunke [4]). For every integer n, we have

n Tor.A; C/ D Tor.nA; nC/:

Proof. Starting with the exact sequence 0 ! nA ! A ! A=nA ! 0 and the same
for C, we obtain the exact sequence

0 ! Tor.nA; nC/ ! Tor.A; C/ ! Tor.A=nA; C/ ˚ Tor.A; C=nC/;

using the commutative diagram dual to Corollary 1.11. Hence we conclude that
n Tor.A; C/
Tor.nA; nC/, since the direct sum in the displayed formula is
annihilated by n. To prove the converse inclusion, pick a generator x D .na; m; nc/ 2
Tor.nA; nC/ where mna D 0 D mnc .a 2 A; c 2 C/. Here x D .a; nm; nc/ D
n.a; nm; c/ 2 nTor.A; C/. t
u
Interestingly, in the last lemma the integer n can be replaced by p for any
ordinal .
Lemma 4.2 (Nunke [4]). For all ordinals  and p-groups A; C, we have

p Tor.A; C/ D Tor.p A; p C/:

246 8 Tensor and Torsion Products

Thus the length of the p-group Tor.A; C/ can exceed neither the length of A nor the
length of C.
Proof. First we observe that, by the left exactness of Tor, Tor.p A; p C/ may be
regarded as a subgroup of Tor.A; C/. To prove the claim by induction, note that the
case  D 0 is trivial, and the step from  to  C1 follows from the preceding lemma
along with the induction hypothesis:

p  p Tor.A; C/ D p Tor.p A; p C/ D Tor.p  p A; p  p C/:

If  is a limit ordinal, then

p Tor.A; C/ D \< p Tor.A; C/ D \< Tor.p A; p C/  Tor.p A; p C/:

The reverse inclusion follows in the same way as in the proof of Lemma 4.1. t
u
Lemma 4.3 (Nunke [4]).
(i) For all groups A; C, and integers n > 0,

Tor.A; C/Œn
Š Tor.AŒn
; CŒn
/:

(ii) The ranks satisfy rkp .Tor.A; C// D rkp .A/  rkp .C/:
Proof. (i) Multiplication by n yields the exact sequence 0 ! AŒn
!
nP
A!nA ! 0, whence we derive the exact sequence 0 ! Tor.AŒn
; C/ !
nP
Tor.A; C/! Tor.nA; C/. Thus Tor.A; C/Œn
D Ker nP D Tor.AŒn
; C/:
Applying the same argument to C, we obtain the desired isomorphism.
(ii) Since the rank of a p-group is the dimension of its socle as a Z=pZ-vector space,
and since Tor commutes with direct sums, the equality follows from the proved
isomorphism, applied for n D p. t
u
A consequence of this lemma is that if A; C are non-zero p-groups, then Tor.A; C/
is never 0.
UK-Invariants of Tor The following result will enable us to determine the UK-
invariants of Tor for p-groups in terms of their invariants.
Theorem 4.4 (Nunke [4]). If A; C are p-groups, then the th UK-invariant of their
torsion product is

f .Tor.A; C// D f .A/ f .C/ C f .A/ rk.p C1 C/ C rk.p C1 A/ f .C/:

4 Theorems on Torsion Products 247

Proof. Write p AŒp

D p C1 AŒp
˚ G and p CŒp
D p C1 CŒp
˚ H where
rk.G/ D f .A/ and rk.H/ D f .C/. Combining Lemmas 4.2 and 4.3 we obtain
.p Tor.A; C//Œp
D Tor.p AŒp
; p CŒp
/ whence

.p Tor.A; C//Œp

D Tor.p C1 AŒp
; p C1 CŒp
/ ˚ Tor.G; p C1 CŒp
/
˚ Tor.p C1 AŒp
; H/ ˚ Tor.G; H/Œp
:

The first summand on the right is .p C1 Tor.A; C//Œp

, so the th UK-invariant of
Tor will be the sum of the ranks of the remaining three summands. The claim
follows from the observation that the rank of the last summand is rk.G/ rk.H/ D
f .A/f .C/. t
u
Anticipating results from Chapter 11 that direct sums of countable p-groups are
completely determined by their UK-invariants, we can state:
Corollary 4.5. If the p-groups A; C are direct sums of countable groups, then so is
their torsion product. In this case, Tor.A; C/ can be completely characterized by the
UK-invariants of A and C.
Proof. As Tor commutes with direct sums, the first claim is evident. The rest follows
from our discussion above. t
u
Tor and Intersection of Subgroups Additional relevant properties are listed in
the following two lemmas.
Lemma 4.6 (Nunke [4]). Suppose A0 ; A00 are subgroups of A and C0 ; C00 are
subgroups of C. Then
(i) Tor.A0 \ A00 ; C/ D Tor.A0 ; C/ \ Tor.A00 ; C/I
(ii) Tor.A0 ; C0 / D Tor.A; C0 / \ Tor.A0 ; C/I
(iii) Tor.A0 \ A00 ; C0 \ C00 / D Tor.A0 ; C0 / \ Tor.A00 ; C00 /:
Proof. (i) From the commutative diagram

0 −−−−−→ A ∩ A −−−−−→ A −−−−−→ A /(A ∩ A ) −−−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ A −−−−−→ A −−−−−→ A/A −−−−−→ 0

with exact rows and monic vertical maps we get the commutative diagram

0 −−−−−→ Tor(A ∩ A , C) −−−−−→ Tor(A , C) −−−−−→ Tor(A /(A ∩ A ), C)

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ Tor(A , C) −−−−−→ Tor(A, C) −−−−−→ Tor(A/A , C)

248 8 Tensor and Torsion Products

whose rows are exact and the vertical maps are monic. Diagram chasing shows
that Tor.A0 \ A00 ; C/ must be the stated intersection, thus (i) holds.
(ii) follows in the same way from the commutative diagram

0 −−−−−→ Tor(A , C  ) −−−−−→ Tor(A, C  ) −−−−−→ Tor(A/A , C  )

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ Tor(A , C) −−−−−→ Tor(A, C) −−−−−→ Tor(A/A , C)

with exact rows and monic vertical maps.

(iii) Using (i) and (ii), we calculate

Tor.A0 ; C0 / \ Tor.A00 ; C00 / D Tor.A; C0 / \ Tor.A0 ; C/ \ Tor.A; C00 / \ Tor.A00 ; C/

D Tor.A0 \ A00 ; C/ \ Tor.A; C0 \ C00 /
D Tor.A0 \ A00 ; C0 \ C00 /;

as desired.
t
u
Lemma 4.7 (Nunke [4]).
(i) For every x 2 Tor.A; C/, there exist unique finite subgroups A0
A; C0
C
with x 2 Tor.A0 ; C0 / such that if x 2 Tor.A00 ; C00 / .A00
A; C00
C/, then
A0
A00 and C0
C00 .
(ii) For every infinite subgroup G < Tor.A; C/, there exist subgroups A0
A; C0

C such that G
Tor.A0 ; C0 / and jA0 j C jC0 j
jGj.
Proof. (i) Consider pairs A00 ; C00 of finite subgroups with A00
A; C00
C satisfying
x 2 Tor.A00 ; C00 /. From Lemma 4.6(iii) we conclude that the sets of such A00 and C00
are closed under intersection, so there is a unique minimal pair A0 ; C0 . (ii) follows
from (i). t
u
We can say something definitive about the basic subgroups of Tor, also for those
of the subgroups of Tor.
Lemma 4.8 (Nunke [4]). Assume A; C are reduced p-groups. Every subgroup G of
Tor.A; C/ is starred, i.e. its basic subgroups are of the same cardinality as G itself.
Proof. (Keef) Let B denote a basic subgroup of G. There is nothing to prove if B is
finite, so assume it is infinite. By Lemma 4.7, there are subgroups A0
A; C0
C
of cardinality
jBj such that B
Tor.A0 ; C0 /. There are maps ˛ W Tor.A; C/ !
Tor.A=A0 ; C/ and W Tor.A; C/ ! Tor.A; C=C0 / with kernels Tor.A0 ; C/ and
Tor.A; C0 /, respectively, so that the kernel of the map ˛ ˚ W Tor.A; C/ !
Tor.A=A0 ; C/ ˚ Tor.A; C=C0 / is equal to Tor.A0 ; C/ \ Tor.A; C0 / D Tor.A0 ; C0 /:
Since B is mapped by ˛ ˚ onto 0 in a reduced group, the image of G under ˛ ˚
4 Theorems on Torsion Products 249

must also be 0 (see Lemma 6.7 in Chapter 5). This means that G
Tor.A0 ; C0 /, thus
jGj
j Tor.A0 ; C0 /j D jBj: t
u
Dependence on the Underlying Set We now show that, interestingly, the
structure of Tor might depend on the set-theoretical model in which it is considered.
(For the torsion-completion B, see Sect. 3 in Chapter 10.)
Theorem 4.9 (Keef [2]). Let B be an unbounded countable †-cyclic p-group. Then
Tor.B; B/ is †-cyclic if and only if CH holds.
Proof. Assume CH, so that we can create a filtration of the torsion-complete
p-group B (whose cardinality is 2@0 ) with countable pure (†-cyclic) subgroups
C . < !1 /. Consider the pure-exact sequences

0 ! Tor.C ; C / ! Tor.C ; C C1 / ! Tor.C ; C C1 =C / ! 0;

0 ! Tor.C ; C C1 / ! Tor.C C1 ; C C1 / ! Tor.C C1 =C ; C C1 / ! 0:

These sequences split, as the last Tors are †-cyclic, so Tor.C C1 ; C C1 / Š

Tor.C ; C / ˚ Tor.C ; C C1 =C / ˚ Tor.C C1 =C ; C C1 /: Therefore, in the smooth
chain of the Tor.C ; C /, every group is a summand of the next one with †-cyclic
complementary summand, and since evidently Tor.B; B/ D [ <!1 Tor.C ; C /, that
Tor.B; B/ is †-cyclic follows at once from Theorem 2.11 in Chapter 5.
Conversely, by way of contradiction, suppose T D Tor.B; B/ is †-cyclic, and
@1 < 2@0 . Using a fixed direct decomposition of T into cyclic summands, a
standard back-and-forth argument leads us to a pure subgroup C such that B <
C < B; jCj D @1 , and Tor.C; C/ is a summand of T. Tor.C; C/ is obviously
a summand of Tor.B; C/ as well, hence the exact sequence 0 ! Tor.C; C/ !
Tor.B; C/ ! Tor.B=C; C/ ! 0 is splitting. Consequently, the last Tor is †-cyclic.
But B=C is divisible of cardinality 2@0 , hence Tor.B=C; C/ Š ˚ C ¤ 0, and C is †-
cyclic. But C cannot be †-cyclic, since it is uncountable and its basic subgroups are
countable. t
u
When Tor is -Cyclic Call a p-group -cyclic if every subgroup which has
cardinality <  is †-cyclic. The main interest lies of course in groups of cardinality
  that are -cyclic.
Example 4.10. It is clear that for every uncountable cardinal , -cyclic p-groups are separable
(in the sense of Sect. 1 in Chapter 10), and in turn, every separable p-group is @1 -cyclic by Prüfer’s
theorem.

Lemma 4.11 (Nunke [4]). If the p-groups A and C are -cyclic, then Tor.A; C/ is
 C -cyclic.
Proof. It suffices to prove that T D Tor.A; C/ is †-cyclic provided jAj D jCj D .
Choose smooth chains 0 D A0 <    < A < : : : and 0 D C0 <    < C <
: : : . < / of pure subgroups with unions A and C, respectively, such that the links
250 8 Tensor and Torsion Products

have cardinality < , so they are †-cyclic by hypothesis. Then T is the union of the
smooth chain T D Tor.A ; C / . < / of pure subgroups. For each , there is a
pure-exact sequence

0 ! T ! T C1 ! Tor.A C1 =A ; C C1 / ˚ Tor.A C1 ; C C1 =C /

(the dual of Corollary 1.11). The last direct sum of the Tors is †-cyclic, since C C1
and A C1 are both †-cyclic, thus T C1 D T ˚ H for some †-cyclic H . Hence
we conclude that T D ˚ < H is †-cyclic. t
u
In particular, if A; C are separable p-groups (thus @1 -cyclic), then Tor.A; C/ is
@2 -cyclic. A simple induction shows that if A1 ; : : : ; A2n are separable p-groups, then
their torsion product Tor.A1 ; : : : ; A2n / is @nC1 -cyclic.
F Notes. More on Tor can be found in Nunke [4]. For iterated torsion products, see Keef
[2]. Keef [9] gives a complete answer to Nunke’s question as to when Tor.A; C/ is †-cyclic.
He introduces new invariants whose values are collections of finite sets of uncountable regular
cardinals.
The problem has been raised about the relation of reduced p-groups A; C if they satisfy
Tor.A; G/ Š Tor.C; G/ for all reduced groups G. Hill [12] has several positive results, and Cutler
[1] provides examples showing that A; C need not be isomorphic. Keef [7] shows that reduced p-
groups A; C satisfy Tor.A; G/ Š Tor.C; G/ for all reduced p-groups G if and only if they have the
same UK-invariants and A ˚ X Š C ˚ Y for some †-cyclic groups X; Y. For an excellent survey
of Tor, see Keef [8].

Exercises

(1) For all A; C, we have j Tor.A; C/j

jAj  jCj.
(2) If A; C are p-groups of length A ; C , then the length of Tor.A; C/ is exactly
minfA ; C g.
(3) The divisible part of Tor.A; C/ is the torsion product of the maximal divisible
subgroups of A and C.
(4) (Richman) If A; C are p-groups, and u D .a; pk ; c/ 2 Tor.A; C/ is a generator,
then hp .u/ D minfhp .a/; hp .c/g.
(5) (Nunke) Let a 2 A; c 2 C be of the same order pm , and A0
A; C0
C
such that the generator .a; pm ; c/ of Tor.A; C/ belongs to Tor.A0 ; C0 /. Then also
a 2 A0 ; c 2 C 0 .
(6) The torsion product T D Tor.A; C/ of two reduced p-groups has a †-cyclic
summand of the same cardinality as T. (We could be more specific and claim a
summand isomorphic to a basic subgroup of T.) [Hint: Lemma 4.8 and Lemma
6.12 in Chapter 5.]
(7) (Nunke) If A; C are p-groups such that Tor.A; C/ is, but C is not †-cyclic, then
A must be separable.
5 Localization 251

(8) (Keef) If A0
A; C0
C, then Tor.A0 ; C0 / is isotype (Sect. 5 in Chapter 11) in
the direct sum Tor.A; C0 / ˚ Tor.A0 ; C/. [Hint: Lemmas 4.2 and 4.6.]

5 Localization

Localization at Primes Let S denote a set of primes. The localization of Z at

S is defined as the set of rational numbers rs such that all the prime divisors of the
denominator s belong to S. It is easy to check that these rational numbers form a
subring of Q; it is denoted by S1 Z. By far the most important special case is when
S consists of all primes with the exception of a single prime p, in which case the
localization is denoted as Z.p/ , and is called the localization at p.
We are particularly interested in the localization of groups at a prime p. The
localization of group A at p is defined by the formula

A.p/ D Z.p/ ˝ A:

Thus A.p/ is a Z.p/ -module. The localization map p .A/ W A ! A.p/ acting as p .A/ W
a 7! 1 ˝ a is a natural homomorphism. (We often simplify notation and write p if
there is no danger of confusion.) p has the universal property that if W A ! M
is a group homomorphism into a Z.p/ -module M, then there exists a unique Z.p/ -
homomorphism 0 W A.p/ ! M such that D 0 p . It is routine to check that
Ker p is the direct sum of the q-components of the torsion subgroup of A for all
primes q ¤ p.
If p .A/ is an isomorphism, we say that A is p-local. A p-local group is
nothing else than a group with unique q-divisibility for every prime q ¤ p. Group
homomorphisms between p-local groups are automatically Z.p/ -homomorphisms.
(A) The localization at p is functorial: if  W A ! C is a homomorphism, then
there exists a unique Z.p/ -homomorphism p W A.p/ ! C.p/ such that p .C/ D
p p .A/. The map p carries 1 ˝ a .a 2 A/ to 1 ˝ a.
(B) Localization is an exact functor Ab ! Mod-Z.p/ from the category of abelian
groups into the category of Z.p/ -modules. More explicitly, this means that if
˛ ˇ
0 ! B!A!C ! 0 is an exact sequence, then so is the sequence 0 !
˛p ˇp
B.p/ !A.p/ !C.p/ ! 0. In fact, tensoring by the torsion-free Z.p/ preserves
exactness. In particular, we have

A.p/ =B.p/ Š .A=B/.p/ :

(C) If A is a torsion group, then its localization at p is just its p-component Ap . In

our notation, A.p/ D Ap . Thus localization at p leaves a p-group unchanged.
252 8 Tensor and Torsion Products

(D) Localization commutes with arbitrary direct sums:

.˚i2I Ai /.p/ D ˚ı2I .Ai /.p/ :

(The same fails for direct products.)

(E) For every group A, there is a canonical embedding
Y
W A! A.p/ ;
p

acting as .x/ D .: : : ; p .x/; : : : /. This is clear, since the intersection of the

kernels of the p for all p is 0. Consequently, A D 0 if and only if A.p/ D 0 for
all primes p.
(F) A homomorphism  W A ! C is injective .respectively, surjective/ if and only
if p W A.p/ ! C.p/ is injective .respectively, surjective/ for all primes p. If 
is monic, then p is also monic by (B). Conversely, the sequence 0 ! K !

A!C is exact (K D Ker ), so the same is true for the localization sequence
p
0 ! K.p/ ! A.p/ !C.p/ . By hypothesis, K.p/ D 0 for all p, so K D 0 by (E).
A similar proof applies to the epic case.
Thus  W A ! C is an isomorphism if and only if p W A.p/ ! C.p/ is an
isomorphism for every prime p. However, it should be kept in mind that A Š C
need not be true even if A.p/ Š C.p/ for all p (Exercise 4).
Localization of Torsion-Free Groups If G is torsion-free, then the canonical
map p .G/ is monic for each p. Thus p .G/ embeds G as a subgroup in G.p/ which
in turn is embedded in the divisible hull of G. These embeddings are natural which
makes it possible to form intersections of distinct localizations.
Lemma 5.1. For any torsion-free group G,

G D \p G.p/

where the intersection is taken in the injective hull of G.

Proof. This is a special case of a general theorem on localization over commutative
rings, but it is easy to give a quick proof for groups. Since G
G.p/ for all primes
p, it is enough to prove that every x 2 \p G.p/ belongs to G. For a p, there is n 2 N,
coprime to p, such that nx 2 G. Also, for every prime divisor pj of n we can find an
nj 2 N coprime to pj with nj x 2 G. If t; tj 2 Z satisfy tn C t1 n1 C    C t` n` D 1,
then x D tnx C t1 n1 x C    C t` n` x 2 G. t
u
F Notes. There is an increasing interest in localizations. They can be defined as idempotent
functors L W Ab ! Ab along with the natural transformation 1Ab ! L.
5 Localization 253

Exercises

(1) If F is a free group, then F.p/ is a free Z.p/ -module.

(2) Localization commutes with finite intersections: .G \ H/.p/ D G.p/ \ H.p/ .
(3) Find a counterexample to show that in general direct products do not commute
with localization.
(4) A and C need not be isomorphic even if A.p/ Š C.p/ for every prime p. [Hint: Z
and the group of rational numbers with square-free denominators.]
(5) Hom.A; C/.p/ Š Hom.A.p/ ; C.p/ / need not be true, not even for torsion-free
A; C. [Hint: A D Z.p/ ; C D Z.]
(6) (Hasse principle for abelian groups) Let B be a subgroup of A with A=B torsion,
and let  W B ! C be a homomorphism. Suppose that, for each prime p, the
localization map p W B.p/ ! C.p/ has an extension p W A.p/ ! C.p/ . Then there
exists a homomorphism   W A ! C that extends  and satisfies p D p for
each prime p.

Problems to Chapter 8

PROBLEM 8.1. Find the groups A with Tor.A; A/ Š A.

PROBLEM 8.2. Determine the Ulm factors of Tor.A; C/ for p-groups A; C.
PROBLEM 8.3. When can a reduced p-group A be written as Tor.B; C/ for
reduced p-groups B; C?
PROBLEM 8.4. When is a rank 4 torsion-free group the tensor product of two
rank 2 torsion-free groups? More generally, when is a torsion-free group a non-
trivial tensor product?
Lausch [1] has results on the rank 4 case.
Chapter 9
Groups of Extensions and Cotorsion Groups

Abstract The extension problem for abelian groups (as a special case of the general
group-theoretical question formulated by O. Schreier) consists in constructing a group from a
normal subgroup and the corresponding factor group. The classical way of discussing extensions
is via factor sets which we follow in our presentation (simplified for the abelian case). Then we
introduce Baer’s group Ext, an extremely important device, and discuss its fundamental properties.
The intimate relationship between Hom and Ext has been pointed out by Eilenberg–MacLane [1];
this led to the interpretation of Ext as a derived functor of Hom and has been exploited extensively
in Homological Algebra. Another important functor is Pext, the group of pure extensions, which
appears unexpectedly as the first Ulm subgroup of Ext.
The investigation of the group structure of Ext leads to the concept of cotorsion group, a
generalization of algebraic compactness. We give special prominence to cotorsion groups that
occur not only as Ext, but also in several other forms.

1 Group Extensions

The Extension Problem Given two groups, A and C, the extension problem
consists in finding all groups B such that B contains a subgroup A0 isomorphic to A
and B=A0 Š C. This situation can be expressed in terms of the exact sequence
 
e W 0 ! A!B!C ! 0; (9.1)

where  stands for the inclusion map, and  is an epimorphism with kernel A. In
this case, we say that B is an extension of A by C.
It is our next aim to survey all extensions for fixed A and C. This can be done
in different ways. We first describe extensions via factor sets (in a pedestrian way),
and then we discuss them by using short exact sequences (which is a more attractive
and more powerful method).
Let a; b; : : : denote elements of A, and u; v; w; : : : those of C. Assuming that B
is an extension of A by C, we pick a transversal; this is a function g W C ! B
that assigns an element of B in the coset corresponding to u, i.e. g.u/ 2  1 u. For
convenience, it is always assumed that g.0/ D 0. Once the function g is selected,
every b 2 B has a unique form b D g.u/ C a with u 2 C; a 2 A. Since g.u/ C g.v/
and g.u C v/ belong to the same coset mod A, there is an f .u; v/ 2 A such that

g.u/ C g.v/ D g.u C v/ C f .u; v/:

© Springer International Publishing Switzerland 2015 255

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_9
256 9 Groups of Extensions and Cotorsion Groups

Thus we have a function f W C C ! A, uniquely determined by the extension B and

by the choice of representatives g.u/. The commutative and associative laws imply
that f satisfies

f .u; v/ D f .v; u/; f .u; v/ C f .u C v; w/ D f .u; v C w/ C f .v; w/ (9.2)

for all u; v; w 2 C. A function f satisfying these two equations is called a factor

set on C to A. Evidently, a splitting map for  (if it exists) is also a transversal for
which the factor set is identically 0.
Assume, conversely, that we are given two groups, A and C, along with a factor
set ff .u; v/gu;v2C on C to A. We will construct a group B as the set of all pairs
.u; a/ 2 C  A with the operation

.u; a/ C .v; b/ D .u C v; a C b C f .u; v//:

The commutative and associative laws in B are consequences of conditions (9.2),

while .0; 0/ is the zero in B, and .u; a  f .u; a// is the inverse to .u; a/.
Manifestly, the mappings  W a 7! .0; a/;  W .u; a/ 7! u will make the sequence
(9.1) exact. Thus B is an extension of A by C, where the choice g.u/ D .u; 0/
corresponds to the factor set f :
The relation between extensions and factor sets being clarified, it is now obvious
that the extension problem amounts to finding all factor sets. The direct sum C ˚ A
is always a solution to the extension problem; it is called the splitting extension. If
g.u/ D .u; 0/ 2 C ˚ A is chosen, then f .u; v/ is identically 0.
Equivalence of Extensions While a factor set determines the extension
uniquely, the converse is not true: a factor set depends also on the representatives
selected. In order to remedy the problem with this ambiguity, an equivalence relation
is introduced for factor sets. We shall call the factor sets f1 and f2 equivalent if they
are coming from the same extension by using another set of representatives. This
means that the new representatives will be g.u/ C h.u/ with some h.u/ 2 A, and
accordingly, the new factor set will look like f .u; v/ C h.u/ C h.v/  h.u C v/.
A factor set f 0 .u; v/ of the form f 0 .u; v/ D h.u/ C h.v/  h.u C v/ with any function
h W C ! A is called a transformation set or coboundary; we always choose
h.0/ D 0. Consequently, by definition, two factor sets are equivalent if and only if
their difference is a coboundary.
One should be aware of the fact that two equivalent, but different, factor sets
do not necessarily define identical extensions. However, the two extensions, B and
B0 , are essentially the same in the sense that the correspondence ˇ W .u; a/ 7!
.u; a C h.u//0 is not only an isomorphism ˇ W B ! B0 , but it is a special one
that induces the identity maps both on A and on C. This is tantamount to saying that
the diagram
1 Group Extensions 257

μ ν
−−−−→ A −−−−→ B −−−−→ C −−−−→ 0
 ⏐ 
 ⏐ 
 β 
μ ν
−−−−→ A −−−−→ B  −−−−→ C −−−−→ 0

is commutative. In this case, the extensions e and e0 themselves are called equiva-
lent. Thus, there is a bijective correspondence between the equivalence classes of
extensions of A by C and the equivalence classes of factor sets f W C  C ! A.
In this correspondence, the splitting extensions form an equivalence class, and the
corresponding equivalence class of factor sets consists of factor sets of the form
h.u/ C h.v/  h.u C v/ for arbitrary functions h W C ! A (with h.0/ D 0 as agreed).
The Group Ext We have come to a leading idea: instead of being satisfied with
a survey of the collection of all extensions, one tries to furnish this set with a proper
algebraic structure which would provide a more powerful tool in the exploration.
This was done by R. Baer who introduced a group structure, creating a fascinating
theory.
If f ; f 0 W C  C ! A are factor sets, then their sum f C f 0 defined as

.f C f 0 /.u; v/ D f .u; v/ C f 0 .u; v/

is again a factor set. Under this composition, the factor sets form a group, denoted
Z.C; A/. The coboundaries form a subgroup B.C; A/, and what has been concluded
above can be rephrased by saying that there is a bijective correspondence between
the equivalence classes of extensions of A by C and the elements of the factor group
Z.C; A/=B.C; A/. This factor group is generally called the group of extensions of
A by C:

Ext.C; A/ D Z.C; A/=B.C; A/:

Having defined Ext in terms of factor sets, we now proceed to another interpre-
tation: via short exact sequences (9.1). If we think of extensions of A by C as such
sequences, then it seems reasonable to create and to study first a category whose
objects are short exact sequences. In doing so, the first order of business is to define
the morphisms between two exact sequences, e and e0 . The right definition is pretty
clear: it is a triple .˛; ˇ; / of group homomorphisms rendering the diagram

μ ν
−−−−→ A −−−−→ B −−−−→ C −−−−→ 0
⏐ ⏐ ⏐
⏐
α
⏐
β
⏐
γ

μ ν
−−−−→ A −−−−→ B  −−−−→ C  −−−−→ 0
258 9 Groups of Extensions and Cotorsion Groups

commutative. There is no difficulty in showing that in this way we have indeed

defined a category which will be denoted E.
In accordance with our definition of equivalent extensions above, we will say that
the exact sequences e and e0 (qua extensions) are equivalent (notation: e  e0 ) if
A D A0 ; C D C0 and there is a morphism .1A ; ˇ; 1C / from e to e0 . That ˇ is then an
isomorphism follows at once from Lemma 2.6 in Chapter 1.
The beauty of treating extensions as short exact sequences lies in the fact that we
can work with commutative diagrams, making most proofs more transparent. But
first we must learn the basic facts about the category E.
To start with, we concentrate on extensions of a fixed group A. If W C0 ! C is a
homomorphism, then there is a pull-back square

ν
B  −−−−→ C 
⏐ ⏐
⏐
β
⏐
γ

μ ν
0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

From Sect. 3(a) in Chapter 2 we know that  0 is an epimorphism (since so is ), and
the pull-back property shows that Ker  0 Š Ker  Š A. Hence there exists a monic
map 0 W A ! B0 acting as 0 W a 7! .a; 0/ 2 B0 .
B ˚ C0 /, so that the diagram

μ ν
−−−−→ A −−−−→ B  −−−−→ C  −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
 β γ

μ ν
−−−−→ A −−−−→ B −−−−→ C −−−−→ 0

has exact rows and commutative squares. The top row is an extension of A by C0
which we have denoted by e to indicate its origin from e via . Note that  D
.1A ; ˇ; / is a morphism e ! e in the category E.
Now suppose we have a similar commutative diagram

μ ν 
: 0 −−−−→ A −−−−→ B  −−−−→ C  −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
 β  γ

μ ν
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

with exact rows, for another group B00 . By the pull-back property there is a unique
 W B00 ! B0 such that  0  D  00 and ˇ D ˇ 0 . Since the maps 00 ; 0 W A ! B0
are such that ˇ.00 / D ˇ 0 00 D  D ˇ0 and  0 .00 / D  00 00 D 0 D  0 0 ,
the uniqueness assertion on pull-backs implies 00 D 0 . Hence .1A ; ; 1C0 / is an
1 Group Extensions 259

E-morphism e00 ! e , and so e00  e . This shows that e is unique up to

equivalence. Once this has been established, we can then assert that e1C  e and
0
e. 0 /  .e / 0 for C00 !C0 !C. If we view Ext.; A/ for a fixed A as a functor
Ab ! Ab, then its contravariant character is evident.
Next we switch the roles: we keep C fixed and let A vary. Given a map ˛ W A ! A0 ,
let B0 be defined by the push-out square (on the left)

μ ν
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
⏐ ⏐ 
⏐
α
⏐
β


μ ν
α : 0 −−−−→ A −−−−→ B  −−−−→ C −−−−→ 0

where we have already completed the diagram to have an exact sequence at the
bottom row. This could be done, since 0 is a monomorphism, and if B0 is defined
as a factor group of A0 ˚ B, then  0 ..a0 ; b/ C H/ D b makes the diagram commute.
The bottom row is an extension of A0 by C which was denoted as ˛e. As before, we
can show (using dual arguments) that ˛e is unique up to equivalence, furthermore,
˛ ˛0
1A e  e and .˛ 0 ˛/e  ˛ 0 .˛e/ for A!A0 !A00 . Thus, for a fixed C, Ext.C; / is a
covariant functor Ab ! Ab.
Combining the maps ˛ W A ! A0 and W C0 ! C, we obtain the important
associative law: ˛.e /  .˛e/ . Indeed, by making use of the pull-back property of
.˛e/ , it is easy to verify the existence of a morphism .˛; ˇ 0 ; 1C / W e ! .˛e/ and
the commutativity of the square

(1,β  ,γ)
γ −−−−−→
⏐ ⏐
⏐ ⏐(α,β,1)
(α,β  ,1) 
(1,β,γ)
(α )γ −−−−→ α

The Baer Sum of Extensions It was shown above that the extensions (more
precisely, the equivalence classes of extensions) of A by C form a group. We wonder
how to describe the group operation in the language of short exact sequences. We
will use the diagonal map G W G ! G ˚ G (where g 7! .g; g/), as well as the
codiagonal map rG W G ˚ G ! G (where .g1 ; g2 / 7! g1 C g2 ). If we understand by
˛i ˇi
the direct sum of extensions ei W 0 ! Ai !Bi !Ci ! 0 .i D 1; 2/ the extension

˛1 ˚˛2 ˇ1 ˚ˇ2
e1 ˚ e2 W 0 ! A1 ˚ A2 ! B1 ˚ B2 ! C1 ˚ C2 ! 0;

then we can state:

260 9 Groups of Extensions and Cotorsion Groups

Proposition 1.1 (Baer). The sum of the extensions e1 and e2 of A by C is the

extension

e1 C e2 D rA .e1 ˚ e2 /C :

Proof. What we have to establish is that if fi W C  C ! A is a factor set

belonging to ei .i D 1; 2/, then f1 C f2 belongs to rA .e1 ˚ e2 /C : It is clear that
.f1 .c1 ; c2 /; f2 .c01 ; c02 // with ci ; c0i 2 C is a factor set belonging to the direct sum e1 ˚e2 ,
and .f1 .c1 ; c2 /; f2 .c1 ; c2 // is one corresponding to .e1 ˚ e2 /C : An application of rA
yields the factor set f1 .c1 ; c2 / C f2 .c1 ; c2 /. t
u
It is now easy to conclude that for the homomorphisms ˛i W A ! A0 and
i W C0 ! C the following equivalences hold for the extensions e; e1 ; e2 of A by C:

˛.e1 C e2 /  ˛e1 C ˛e2 ; .e1 C e2 /  e1 C e2 ; (9.3)

.˛1 C ˛2 /e  ˛1 e C ˛2 e; e. 1 C 2 /  e 1 C e 2 : (9.4)

The equivalences (9.3) express the fact that ˛ W e 7! ˛e and  W e 7! e are group
homomorphisms

˛ W Ext.C; A/ ! Ext.C; A0 /;  W Ext.C; A/ ! Ext.C0 ; A/;

while (9.4) asserts that .˛1 C ˛2 / D .˛1 / C .˛2 / and . 1 C 2 / D . 1 / C . 2 / ,

i.e. the correspondence ExtW C  A ! Ext.C; A/ with  ˛ 7!  ˛ D ˛  is an
additive bifunctor Ab  Ab ! Ab. This fact is important enough to be recorded as
a theorem:
Theorem 1.2 (Eilenberg–MacLane [1]). Ext is an additive bifunctor Ab  Ab !
Ab, contravariant in the first, and covariant in the second variable. t
u
In order to be consistent with the functorial notation for homomorphisms, we
shall also use the notation Ext. ; ˛/ for  ˛ D ˛  , thus Ext. ; ˛/ W Ext.C; A/ !
Ext.C0 ; A0 / acting as e 7! ˛e . In the sequel, we will often deal with diagrams of
extensions, emphasizing that in all of these diagrams extensions can be replaced by
equivalent extensions without causing any harm to commutativity or exactness.
A useful observation: if C is a p-group, then there are natural isomorphisms

Ext.C; A/ Š Ext.C; Z.p/ ˝ A/ and Ext.A; C/ Š Ext.Z.p/ ˝ A; C/:

In both cases the right-hand sides can also be interpreted as ExtZ.p/ .

F Notes. If we wish to develop extensions solely qua short exact sequences, then the Baer sum
would serve as the definition of the sum of two extensions, and then we have to verify: (1) e1 C e2
is indeed an extension of A by C which stays in the same equivalence class if e1 ; e2 are replaced
by equivalent extensions, and (2) the equivalence classes form a group under this operation. For
details of this approach, we refer to MacLane [M]. A third method of introducing Ext is as the
2 Exact Sequences for Hom and Ext 261

derived functor of Hom, this is a popular way of defining Ext in Homological Algebra, where also
the higher Extn functors are dealt with (for n > 1, these are trivial for abelian groups).
It was O. Schreier (1926) who started a more systematic investigation of the extensions of
non-commutative groups (in a multiplicative notation): given the groups A; C, new groups G
have to be constructed such that G contains a normal subgroup N isomorphic to A satisfying
G=N Š C. Besides factor sets, he used also automorphisms of A (these are not needed in the
commutative case) to describe extensions. Extensions without factor sets were discussed by Baer
(1935). Topological applications were discovered by S. Eilenberg and S. Mac Lane (1942).
A note on the terminology is in order. A few authors apply a reverse terminology: if B=A Š C,
then they say that B is an extension of C by A. It seems more natural to extend on the top than in
the bottom.

Exercises

(1) A factor set on any group C to a divisible group D is a transformation set. [Hint:
set up a system of equations.]
(2) Let B be a basic subgroup of the reduced p-group A. For any group C, any factor
set f W C  C ! A is equivalent to a factor set g W C  C ! B.
3 ˇ 3
(3) The exact sequences 0 ! Z !Z!Z.3/ ! 0 and 0 ! Z !Z!Z.3/ !
0 represent different elements in Ext.Z.3/; Z/ if ˇ W 1 7! 1 C 3Z; W 1 7!
2 C 3Z.
(4) There are p inequivalent extensions of Z.p/ by Z.p/, but only two non-
isomorphic ones.
(5) In terms of factor sets f , the extension e corresponds to the composite map
 f f ˛
C0  C0 !C  C!A, and ˛e to C  C!A!A0 (notation as in the text).
(6) If .˛; ˇ; / W e ! e0 is a morphism in the category E, then ˛e  e0 .
(7) (a) Let ˛ W A ! A0 be an epimorphism. Then ˛e with e as in (9.1) is equivalent
to the extension 0 ! A= Ker ˛ ! B=.Ker ˛/ ! C ! 0 (obvious maps).
(b) Let W C0 ! C be a monic map. Then e is equivalent to 0 ! A !
 1 .Im / ! Im ! 0.
(8) (a) If an extension is represented by a pure-exact sequence, then the same holds
for all equivalent extensions.
(b) Show that the Baer sum of two extensions represented by pure-exact
sequences is again pure-exact.
(9) (a) Let ˛ be an automorphism of A. When is e equivalent to ˛e?
(b) When does e  e hold for some 2 Aut C?

2 Exact Sequences for Hom and Ext

The functorial aspects of Ext having been settled, we move on to gather more
information about its formal properties, in particular, its behavior towards exact
sequences. What might come as a surprise is that it is intimately connected with
262 9 Groups of Extensions and Cotorsion Groups

the functor Hom: Ext emerges as an “error term” for the failure of exactness of the
Hom sequence in Sect. 1 in Chapter 7.
Preparatory Lemmas We need to do a little technical work before we prove
our main result in this section. We start with an easy, but very useful observation.
(Compare with Lemmas 2.1 and 2.2 in Chapter 1.)
Lemma 2.1. Suppose we are given the diagram

G
⏐
ξ ⏐
η

α β
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
⏐
⏐
χ
ζ

with exact row. Then

(i) there exists a map  W G ! B such that ˇ D  if and only if e splits;
(ii) there is a map  W B ! H such that ˛ D  if and only if e splits.
Proof. If there is such a , then the diagram

0 −−−−→ A −−−−→ A ⊕ G −−−−→ G −−−−→ 0

 ⏐ ⏐
 ⏐∇(α⊕ξ) ⏐η
  
α β
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

(with the obvious maps in the top row) commutes, hence the top row is  e.
Conversely, if e W 0 ! A ! B0 ! G ! 0 splits, then a splitting map followed
by B0 ! B yields a map  W G ! B with the desired property. For  the proof is
dual. t
u
The following lemma unveils the basic relationship between Hom and Ext.
Actually, it is the jumping board to Theorem 2.3; it deals with a special case needed
in the proof of Theorem 2.3.
Lemma 2.2. Let 0 ! H ! F ! C ! 0 be a free resolution of C and 0 ! A !
D ! N ! 0 an injective resolution of A.
(i) There exists a natural map ı  W Hom.H; A/ ! Ext.C; A/ such that following
the sequence is exact:

ı
0 ! Hom.C; A/ ! Hom.F; A/ ! Hom.H; A/! Ext.C; A/ ! 0: (9.5)
2 Exact Sequences for Hom and Ext 263

(ii) There is a natural map ı W Hom.C; N/ ! Ext.C; A/ that makes the following
sequence exact:

ı
0 ! Hom.C; A/ ! Hom.C; D/ ! Hom.C; N/! Ext.C; A/ ! 0: (9.6)

Proof. Starting from the middle row, consider the following diagram with exact
rows and commutative squares:

μ
1: 0 −−−−→ H −−−−→ F −−−−→ C −−−−→ 0
⏐ ⏐ 
⏐
χ ζ ⏐φ 
 
α β
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
 ⏐ ⏐
 ⏐ ξ ⏐η
 ψ 
ν
2: 0 −−−−→ A −−−−→ D −−−−→ N −−−−→ 0,

where  exists, because F is free, and exists, because D is divisible; then  and 
are automatic. Thus e  e1 and e  e2 , showing that the maps in (9.5) and (9.6)
between the Hom and Ext are surjective. To prove exactness at the last Homs, we
refer to the preceding lemma to argue that e  e1 is splitting if and only if there is a
map  W F ! A such that  D , and similarly, e  e2  is splitting exactly if there
is a map  W C ! D such that  D . This amounts to saying that the kernels are
the images of the maps Hom.F; A/ ! Hom.H; A/ and Hom.C; D/ ! Hom.C; N/,
respectively. We leave it as an exercise to the reader to show that the maps ı  ; ı
(called connecting maps) are natural. t
u
The Hom-Ext Exact Sequences We now have all the ingredients to verify the
extremely important long exact sequences.
˛ ˇ
Theorem 2.3 (Cartan–Eilenberg [CE]). If 0 ! A!B!C ! 0 is an exact
sequence, then so are the sequences

0 ! Hom.C; G/ ! Hom.B; G/ ! Hom.A; G/ !

ı ˇ ˛
! Ext.C; G/! Ext.B; G/! Ext.A; G/ ! 0 (9.7)

and

0 ! Hom.G; A/ ! Hom.G; B/ ! Hom.G; C/ !

ı ˛ ˇ
! Ext.G; A/! Ext.G; B/! Ext.G; C/ ! 0 (9.8)

for every group G. All the maps are natural.

264 9 Groups of Extensions and Cotorsion Groups

Proof. The proof relies on the Snake Lemma 2.5 in Chapter 1. We first choose free
resolutions 0 ! H1 ! F1 ! A ! 0 and 0 ! H2 ! F2 ! C ! 0, and note
that there is a free resolution 0 ! H1 ˚ H2 ! F1 ˚ F2 ! B ! 0 (see Sect. 1,
Exercise 7 in Chapter 3). These are used in the commutative diagram

1 ρ σ
1
0 −−−−−→ Hom(C, G) −−−−−→ Hom(B, G) −−−−−→ Hom(A, G)
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
μ ν
0 −−−−−→ Hom(F2 , G) −−−−−→ Hom(F1 ⊕ F2 , G) −−−−−→ Hom(F1 , G) −−−−−→ 0
⏐ ⏐ ⏐
⏐α ⏐ ⏐γ
 β 
μ ν
0 −−−−−→ Hom(H2 , G) −−−−−→ Hom(H1 ⊕ H2 , G) −−−−−→ Hom(H1 , A) −−−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
ρ
2 σ
2
Ext(C, G) −−−−−→ Ext(B, G) −−−−−→ Ext(A, G) −−−−−→ 0

where the middle rows are exact, because Fi ; Hi are free groups, while the columns
(bordered with 0’s, not shown) are exact, thanks to Lemma 2.2. By Lemma 2.5 in
Chapter 1, (9.7) follows from the diagram. The proof of (9.8) is similar. t
u
Purity in the Exact Sequence on Ext We complement Theorem 2.3 with the
following result where the starting short exact sequence is pure-exact.
˛ ˇ
Lemma 2.4. Suppose 0 ! A!B!C ! 0 is a pure-exact sequence. Then, for
every group G, the induced homomorphisms

ˇ  W Ext.C; G/ ! Ext.B; G/ and ˛ W Ext.G; A/ ! Ext.G; B/

map onto pure subgroups.

Proof. We give a detailed proof for the first part (using Theorem 5.2 below), and
leave the rest to the reader as an exercise. In the following commutative diagram,
we start from the bottom row and then form the preceding rows by using the maps
in the right column as indicated:

βσρ : 0 −−−−−→ G −−−−−→ K  −−−−−→ C[n] −−−−−→ 0

 ⏐ ⏐
 ⏐ ⏐ρ
  

βσ : 0 −−−−−→ G −−−−−→ K −−−−−→ B[n] −−−−−→ 0

 ⏐ ⏐
 ⏐ ⏐σ
  

β : 0 −−−−−→ G −−−−−→ H  −−−−−→ B −−−−−→ 0 (∈ Im β ∗ )

 ⏐ ⏐
 ⏐ ⏐β
  

: 0 −−−−−→ G −−−−−→ H −−−−−→ C −−−−−→ 0

2 Exact Sequences for Hom and Ext 265

Here the rows are exact,  denotes the injection map, and  exists, due to the purity
of the given exact sequence. The composite map ˇ W CŒn
! C will be the natural
injection. If the last but one row belongs to n Ext.B; G/, then Theorem 5.2 below
shows that the second row splits, and hence so does the top row. This means that the
bottom exact sequence belongs to n Ext.C; G/, so the last but one row is in n Im ˇ  .
This holds for every n 2 N, thus Im ˇ  is pure in Ext.B; G/. t
u
Ext and Direct Sums, Products We have to find out how Ext behaves towards
direct sums and products. In view of Theorem 2.3, it should not come as a surprise
that Ext imitates Hom in this respect.
Theorem 2.5. For all groups A; Ai ; C; Ci , there exist natural isomorphisms
Y
Ext.˚i2I Ci ; A/ Š Ext.Ci ; A/; (9.9)
i2I

!
Y Y
Ext C; Ai Š Ext.C; Ai /: (9.10)
i2I i2I

Proof. We prove (9.9), the proof of (9.10) runs dually. We start with free resolutions
of the Ci , 0 ! Hi ! Fi ! Ci ! 0 with Fi free, to obtain the exact sequences
Hom.Fi ; A/ ! Hom.Hi ; A/ ! Ext.Ci ; A/ ! Ext.Fi ; A/ D 0. The exact sequence
0 ! ˚i Hi ! ˚i Fi ! ˚i Ci ! 0 induces the top exact sequence in the
commutative diagram

Hom(⊕i Fi , A) −−−−−→ Hom(⊕i Hi , A) −−−−−→ Ext(⊕i Ci , A) −−−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
  
i Hom(Fi , A) −−−−−→ i Hom(Hi , A) −−−−−→ i Ext(Ci , A) −−−−−→ 0

We know from Theorem 1.7 in Chapter 7 that the first two vertical maps represent
natural isomorphisms, whence we can conclude that there is a natural isomorphism
between the two Exts in the diagram. t
u
F Notes. The long exact sequence for Hom-Ext generalizes for modules over arbitrary rings.
However, the 0 at the right end is then replaced by sequences of higher Exts.
The functor Ext does not behave in the same way towards direct and inverse limits as Hom. If
j j
fCi .i 2 I/ji g is a direct system of groups with direct limit C, then fExt.Ci ; A/.i 2 I/j Ext.i ; 1A /g
is an inverse system. The most we can say in general is that there is a natural homomorphism
 W Ext.C; A/ ! lim Ext.Ci ; A/. For the special case when the direct system is indexed by the

natural numbers, see Lemma 5.9. In general, the so-called Mittag-Leffler condition guarantees the
surjectivity of . Q Q
Göbel–Prelle [1] ask for groups G with the properties like Ext. Ai ; G/ Š i Ext.Ai ; G/ for
all choices of the Ai , and show that “only G Š Q” is the answer.
266 9 Groups of Extensions and Cotorsion Groups

Exercises

(1) The group G is divisible if and only if, for every monomorphism ˛ W A ! B, the
induced map ˛  W Ext.B; G/ ! Ext.A; G/ is epic.
(2) G is free if and only if, for every epimorphism ˇ W B ! C, the induced map
ˇ W Ext.G; B/ ! Ext.G; C/ is epic.
(3) Let ˛ W A ! B be a monomorphism such that ˛ W Ext.G; A/ ! Ext.G; B/ is
monic for every G. Prove that Im ˛ is a summand of B.
(4) Let ˇ W B ! C be an epimorphism such that ˇ  W Ext.C; G/ ! Ext.B; G/ is
monic for every G. Then Ker ˇ is a summand of B.
(5) Suppose A satisfies Hom.A; Z/ D 0. Then rkp Ext.A; Z/ D rkp Ext.A=pA; Z/.
(6) Prove that Ext.Q; Z/ Š Q@0 . [Hint: 0 ! Z ! Q ! ˚p Z.p1 / ! 0.]
(7) Verify the isomorphism (for any p) Ext.Z.p1 /; Jp / Š Jp . [Hint: 0 ! Jp !
Qp ! Z.p1 / ! 0.]
(8) Assume B is a basic subgroup of the p-group A. For any group G, the groups
Ext.G; A/ and Ext.G; B/ are epimorphic images of each other. [Hint: Theorem
6.10 in Chapter 5.]

3 Basic Properties of Ext

In this section our objective is to record a number of elementary and some advanced
properties of the group of extensions. We shall make extensive use of the long exact
sequences stated in Theorem 2.3 without referring to them explicitly.
Induced Endomorphisms If e W 0 ! A ! B ! C ! 0 is an exact sequence,
and ˛ 2 End A; 2 End C, then both ˛e and e are again extensions of A by C. It is
a routine calculation to verify that the correspondences ˛ W e 7! ˛e and  W e 7! e
are endomorphisms of Ext.C; A/; they will be called the induced endomorphisms
of Ext. Their actions commute, since ˛  D  ˛ . Consequently, Ext.C; A/ is a
bimodule over the endomorphism rings End A and End C.
Our next lemma points out a most useful fact.
Lemma 3.1. Multiplication by an integer n on A or on C induces multiplication
by the same n on Ext.C; A/. The same holds for multiplication by a p-adic integer
.provided it is defined on A or on C/.
Proof. If ˛i 2 End A, then .˛1 C  C˛n / D .˛1 / C  C.˛n / . Since 1A obviously
acts as the identity on Ext, it is manifest that nP 2 EndA induces multiplication by n
on Ext.C; A/. The proof for C is analogous.
A p-adic integer  is the limit of a sequence ni 2 Z .i < !/ in the p-adic
topology, say, pi j  ni for i < !. By what has been proved, multiplication by 
on Ext.C; A/ is the limit of multiplications by ni , and hence it can be identified with
the multiplication by . t
u
3 Basic Properties of Ext 267

Elementary Facts on Ext Next, we collect several elementary results on Ext;

they are useful in computing Ext. We start with two rather trivial remarks which
reveal important features of Ext.
(A) A group C satisfies Ext.C; A/ D 0 for all A exactly if it is free. Thus, we claim
that every extension by C splits if and only if C is free. The “if” part is the same
as Theorem 1.5 in Chapter 3, while the “only” if part follows from Theorem
1.7 in Chapter 3.
(B) A group A satisfies Ext.C; A/ D 0 for all C if and only if it is divisible. This is
a restatement of the equivalence (i) , (iii) in Theorem 2.6 in Chapter 4.
(C) If, for some m 2 N, either mA D 0 or mC D 0, then also m Ext.C; A/ D 0.
This is an obvious consequence of Lemma 3.1.
(D) If mA D A for some integer m 2 N, then also m Ext.C; A/ D Ext.C; A/:
P
m
This follows from the exactness of Ext.C; A/! Ext.C; A/ ! 0 which is a
P
m
consequence of the exactness of A!A ! 0.
(E) Ext.Q; A/ is torsion-free divisible for each group A. This is obvious if we
combine (C) and (D).
(F) CŒm
D 0 implies m Ext.C; A/ D Ext.C; A/ for every m 2 N. In particular,
Ext.C; A/ is divisible whenever C is torsion-free. Hypothesis guarantees that
P
m
the sequence 0 ! C!C is exact, whence the exactness of the induced
P
m
Ext.C; A/! Ext.C; A/ ! 0 follows.
(G) For all groups A, and for every integer m > 0,

Ext.Z.m/; A/ Š A=mA:

P
m
From the exact sequence 0 ! Z!Z ! Z.m/ ! 0 we obtain the exact
sequence

P
m
Hom.Z; A/ Š A! Hom.Z; A/ Š A ! Ext.Z.m/; A/ ! Ext.Z; A/ D 0:

A moment’s reflection shows that the image of the first Hom in the second one
is mA, whence the stated isomorphism is evident.
Example 3.2. If m; n are relatively prime integers satisfying mA D 0 and nC D 0, then
Ext.C; A/ D 0. This is an immediate consequence of (C).
Example 3.3. Using (G), a simple calculation with the integers m; n shows that

Ext.Z.m/; Z.n// Š Z.d/;

where d D gcdfm; ng. Thus Ext.Z.m/; Z.n// D 0 whenever m; n are relatively prime.

(H) For any group C and integer m,

Ext.C; Z.m// Š Ext.CŒm

; Z.m//:
268 9 Groups of Extensions and Cotorsion Groups

P
m
The exact sequence 0 ! CŒm
! C!mC ! 0 induces the exact sequence
P
m
Ext.mC; Z.m//!Ext.C; Z.m// ! Ext.CŒm
; Z.m// ! 0. The image of
the first map must be 0 owing to Lemma 3.1 and (F), whence the claim is
immediate.
(I) If A is p-divisible and C is a p-group, then Ext.C; A/ D 0. If D is the divisible
hull of A, then D=A is torsion divisible with zero p-component. This means
that Hom.C; D=A/ D 0. The exactness of the sequence Hom.C; D=A/ !
Ext.C; A/ ! Ext.C; D/ D 0 establishes the assertion.
(J) An automorphism ˛ 2 Aut A .or 2 Aut C/ induces an automorphism of
Ext.C; A/. Using the inverse ˛N of ˛, observe that the induced endomorphisms
of Ext.C; A/ will satisfy ˛ ˛N  D ˛N  ˛ D 1: The same argument applies to  .
Example 3.4. Let R denote the group of the rational numbers with square-free denominators, and
T the direct sum ˚p Z.p/ with p running over the prime numbers. The exact sequence 0 ! Z !
R ! T ! 0 induces the exact sequence

0 ! Hom.T; T/ ! Hom.R; T/ ! Hom.Z; T/ Š T

! Ext.T; T/ ! Ext.R; T/ ! 0:

It is easily checked thatQthe map between the first Homs

Q Q is an isomorphism, and Ext.T; T/ Š
@0
p Ext.Z.p/; Z.p// Š p Z.p/. Thus Ext.R; T/ Š . p Z.p//=T Š Q .

Isomorphisms Involving Ext For later reference, it is necessary to point out:

Theorem 3.5 (Eilenberg–MacLane [1]). Let A be a torsion-free group, and D its
injective hull. If C is torsion, then

Ext.C; A/ Š Hom.C; D=A/:

Thus Ext.C; A/ is reduced algebraically compact whenever C is torsion and A is

torsion-free.
Proof. Our starting point is the exact sequence 0 ! A ! D ! D=A ! 0. In
the induced exact sequence 0 D Hom.C; D/ ! Hom.C; D=A/ ! Ext.C; A/ !
Ext.C; D/ D 0, the first Hom vanishes because C is torsion and D is torsion-free.
Hence the stated isomorphism follows at once. The second claim is a consequence
of Theorem 2.1 in Chapter 7. t
u
The choice A D Z leads to an interesting corollary on character groups:
Corollary 3.6. For a torsion group C, we have the natural isomorphism

Ext.C; Z/ Š Char C:

Proof. Just note that, for C torsion, Hom.C; R=Z/ D Hom.C; Q=Z/. t
u
3 Basic Properties of Ext 269

Example 3.7.
(a) Ext.Z.p1 /; Z/ D Char Z.p1 / D JQ
p for every prime p.
(b) Ext.Q=Z; Z/ D Char Q=Z D Z Q D p Jp .
Finally, we state without proof two important isomorphisms (that are derived
from the Künneth relations).
Theorem 3.8 (Cartan–Eilenberg [CE]). For all groups A; B; C we have

Ext.A; Ext.B; C// Š Ext.Tor.A; B/; C/: (9.11)

In case C is injective, also

Ext.A; Hom.B; C// Š Hom.Tor.A; B/; C/: (9.12)

t
u
F Notes. There is an extensive literature on the structure of Ext; for instance, what its torsion-
free or p-rank is if it is divisible, or when it can be isomorphic to Q. Interestingly, some questions
turn out undecidable in ZFC. But, tempting as it is, the discussion of these results would take us too
far afield from our principal aim here. However, let us point out that there is a full characterization
of the divisible group Ext.G; Z/ for torsion-free groups G, assuming that V = L and there is no
weakly compact cardinal. See Shelah–Strüngmann [2], and the literature quoted there.
Eklof–Huber [1] investigate when Ext vanishes, by using a new invariant, called -invariant,
which is an equivalence class in filtrations.
It is natural to ask to what extent the functor Ext.C; / or Ext.; C/ determines the group C.
For countable p-groups A; B of finite Ulm length, Moskalenko [2] shows (under CH) that they are
isomorphic whenever Ext.A; C/ Š Ext.B; C/ holds for all groups C. For finite rank torsion-free
groups, see Notes to Sect. 1 in Chapter 12.

Exercises

(1) For finite groups A; C, there is a (non-natural) isomorphism Ext.C; A/ Š

Hom.C; A/.
(2) Ext.C; A/ D 0 if A is a p-group and C is a q-group for different primes p; q.
(3) For any prime p, Ext.Z.p1 /; A/ D 0 is equivalent to pA D A.
(4) (a) Ext.C; A/=m Ext.C; A/ Š Ext.C; A=mA/ for all groups A; C .m 2 N/.
(b) Verify the following converse of (D): If, for some group A and for some
m 2 N, the equality m Ext.C; A/ D Ext.C; A/ holds for every group C,
then mA D A.
(c) Suppose the group A and the integer m 2 N have the property that
m Ext.C; A/ D 0 for every group C. Then mA is divisible.
(5) Assume that mA D A and AŒm
D 0; then also m Ext.C; A/ D Ext.C; A/ and
Ext.C; A/Œm
D 0.
(6) (a) If mC D C and CŒm
D 0 for some m 2 N, then m Ext.C; A/ D Ext.C; A/
and Ext.C; A/Œm
D 0 as well. Conclude that Ext.C; A/ is torsion-free
divisible if so is C.
270 9 Groups of Extensions and Cotorsion Groups

(b) Prove the following converse of (F): if C and m 2 N satisfy m Ext.C; A/ D

Ext.C; A/ for all groups A, then CŒm
D 0.
(c) If C is such that Ext.C; A/ is torsion-free divisible for every group A, then
C is also torsion-free divisible.
(7) (Baer) If the groups A; C satisfy p Ext.C; A/ D Ext.C; A/ for some prime p,
then either CŒp
D 0 or pA D A.
(8) If A is torsion-free, then Ext.C; A/ Š Ext.t.C/; A/ ˚ Ext.C=t.C/; A/.
(9) Prove the isomorphism Ext.Z.p/ ; Z/ Š Z.p1 / ˚ Q@0 . [Hint: 0 ! Z !
Z.p/ ! ˚q¤p Z.q1 / ! 0, apply Hom.; Z/.]
(10) Verify Ext.Q.p/ ; Z/ Š Z.p1 / ˚ Q@0 . [Hint: 0 ! Z ! Q.p/ ! Z.p1 / ! 0.]
1 2@0  1
(11) Prove Ext.Jp ; Z/ Š Z.p Q / ˚ Q . [Hint: 0 ! Jp ! Qp ! Z.p / ! 0.]
(12) Using the fact that p Z.p/ is a reduced group with ˚p Z.p/ as torsion
subgroup, show that Ext.Q; ˚p Z.p// ¤ 0.
(13) (Nunke) Prove that Hom.C; Z/ D 0 D Ext.C; Z/ implies C D 0. [Hint:
Ext.t.C/; Z/ D 0 and Ext.C=mC; Z/ D 0, then Exercise 6.]
(14) (Nunke) Every torsion-free divisible group D is representable as an Ext. [Hint:
try Ext.Q; T/ with torsion T.]
(15) For a torsion-free A, Ext.A; Z/ is torsion-free if and only if A has the projective
n
property with respect to the exact sequences 0 ! Z!Z ! Z=nZ ! 0 for
all n 2 N.
(16) (Hill–Megibben) Let A be a p-group. An exact sequence 0 ! A ! G !
Z.p1 / ! 0 represents an extension of finite order in Ext.Z.p1 /; A/ if and
only if G is not reduced.

4 Lemmas on Ext

We will need several important lemmas involving Ext that are most useful in the
theory. They relate Ext to ascending chains.
Eklof’s Lemmas We start with two lemmas due to P. Eklof. The first provides
a sufficient condition for an Ext to vanish (with no restriction on the sizes), while
the second sharpens this to a necessary and sufficient condition in the Constructible
Universe.
Lemma 4.1 (Eklof’s Lemma). Let A be any group, and

0 D C0 < C1 <    < C <    . < / (9.13)

4 Lemmas on Ext 271

.where  is any infinite cardinal/ a smooth chain of subgroups of a group C such

that
S
(a) < C D CI
(b) for each  < , Ext.CC1 =C ; A/ D 0.
Then Ext.C; A/ D 0.

Proof. Let 0 ! A ! G!C ! 0 be an extension of A by C, and let G denote
the subgroup of G that is the complete preimage of C under . Then G is the union
of the smooth chain A D G0 < G1 <    < G <    . We are going to define by
transfinite induction a smooth chain 0 D B0 < B1 <    < B <    . < / of
subgroups of G such that G D A ˚ B where B Š C .
The starting point  D 0 is obvious, and so is the case when  is a limit ordinal,
since then we have no choice other than taking unions. So suppose that, for some
ordinal  < , G and B have been defined in the desired way for all 
.
By induction, we may assume that Ext.C ; A/ D 0. Now GC1 =B can be viewed
as an extension of A by GC1 =G Š CC1 =C , so condition (b) guarantees that
.A C B /=B Š A is a summand, i.e. GC1 =B D .A C B /=B ˚ BC1 =B for
some subgroup BC1
GC1 . Clearly, GC1 D A ˚ BC1 . Once we have the
complete chain of the B for  < , we can argue that the union B D [< B
satisfies G D A ˚ B, i.e. the given exact sequence splits. t
u
Much greater effort is needed if we wish to prove something similar with
a necessary and sufficient condition. This can be done in Gödel’s Constructible
Universe, i.e. under the hypothesis V = L. We will need an easy observation which
we display as a lemma for easy reference.
In the proofs of the next lemmas, it will be convenient to consider an extension
of A by C (or by a subgroup C0 < C) to be a group built on the set C  A (on C0  A).
This can be done without loss of generality.
Lemma 4.2. Suppose  W C0 ! C is an inclusion map, and A is a group such that

Ext.C; A/ D 0 and Ext.C=C0 ; A/ ¤ 0:

Given splitting exact sequences with maps ; 0 in (9.14), there exists a homomor-
phism  W G0 ! G making the diagram

ρ
0 −−−−→ A −−−−→ G −−−−→ C  −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐φ
 χ  (9.14)
ρ
0 −−−−→ A −−−−→ G −−−−→ C −−−−→ 0

commute such that there is no splitting map  W C ! G with  0 D ; for any
splitting map  0 W C0 ! G0 of 0 .
272 9 Groups of Extensions and Cotorsion Groups

Proof. By hypothesis, in the exact sequence Hom.C; A/ ! Hom.C0 ; A/ !

Ext.C=C0 ; A/ ! Ext.C; A/ D 0 the map between the two Homs is not epic. If
 W C0 ! A is a map that does not extend to any C ! A, then define  to map
.a; c0 / 2 G0 D A  C0 onto .a C c0 ; c0 / 2 G D A  C. It is readily checked that,
by the choice of , no splitting map  W C ! A ˚ C with  0 D  may exist. u t
Keep in mind that the main point is that, under the stated hypotheses, the map
 can be chosen such that the two splitting sequences have no matching splitting
maps.
Lemma 4.3 (Eklof). Assuming V D L, let  be an uncountable regular cardinal,
A any group of cardinality
, and (9.13) a -filtration of a group C of cardinality
 such that

Ext.C ; A/ D 0 for all  < :

Then Ext.C; A/ D 0 if and only if the set

E D f <  j 9  >  such that Ext.C =C ; A/ ¤ 0g

is not stationary in .
Proof. If E is not stationary in , then choose a cub X that does not intersect E.
Keeping only the subgroups C with  2 X, an appeal to Lemma 4.1 shows that the
hypotheses are satisfied for the modified chain, and therefore Ext.C; A/ D 0.
Conversely, suppose E is stationary in . We can clearly drop those indices that
are not needed, and still have a stationary subset of , so there is no loss of generality
in assuming  D  C 1 in the definition of E. Let fA g< be a filtration of A
(repetitions are allowed). In view of the Diamond Principle } (it holds because
V D L), there exists a family fg g2E of functions g W C ! C  A . 2 E/ such
that, for every function gW C ! C  A; the set f 2 E j g  C D g g is stationary
in .
With the aid these g , we are going to construct a non-split exact sequence e W

0 ! A ! G!C ! 0 as the direct limit of splitting exact sequences eW 0 ! A !

G !C ! 0 . < / such that whenever  < ; there is a commutative diagram
ρμ
μ: 0 −−−−→ A −−−−→ Gμ −−−−→ Cμ −−−−→ 0
 ⏐ ⏐
 ν⏐ ⏐
 γμ  incl (9.15)
ρν
ν: 0 −−−−→ A −−−−→ Gν −−−−→ Cν −−−−→ 0

where the right vertical map is the inclusion map. Let  < , and assume that e
has been defined for every  <  such that required diagrams are commutative. We
distinguish three cases. (Now G is assumed to be built on the set C  A, so the 
are also inclusion maps.)
4 Lemmas on Ext 273

Case 1. If  is a limit ordinal, then we let e be the direct limit of the exact sequences
e for  <  with connecting maps .1A ;  ; incl/. This is a splitting sequence,
since by hypothesis C satisfies Ext.C ; A/ D 0.
Case 2. Let  D ı C 1: If ı … E or if gı is not a splitting map for ı ; then let

e W 0 ! A ! G !C ! 0 be a splitting extension with any choice of
ı W Gı ! G such that (9.15) commutes (with  D ı).
Case 3. The remaining case is when  D ı C 1, ı 2 E, and the selected gı W Cı !
Cı  Aı  C  A is a splitting map for ı . From Lemma 4.2 we conclude
that there are a map ı W Gı ! G and a splitting extension e W 0 ! A !

G !C ! 0 with a commutative diagram (9.15) such that there exists no
splitting map  W C ! G for  extending gı .

We now define e W 0 ! A ! G!C ! 0 as the direct limit of the splitting exact

sequences e W 0 ! A ! G !C ! 0 for  <  where G D CA D [ <  .C  A/
as sets. By way of contradiction, assume there is a splitting homomorphism
gW C ! G for : Note that we must then have g.C /
G for every  < : By
the choice of the functions g ; there is a ı 2 E (actually, stationarily many of them)
such that g  Cı D gı : This means that eıC1 has been constructed according to
Case 3 above. Since g  CıC1 is both a splitting map for eıC1 and an extension
of g  Cı ; we have reached a contradiction to the existence of a splitting g. Thus
Ext.C; A/ ¤ 0, indeed. t
u
Ext When Both Arguments Are Chains In the following lemma we turn to the
case when both arguments in Ext are unions of chains. This is a version of a lemma
by Eklof–Fuchs; it requires a different approach.
Lemma 4.4. Let  be an uncountable regular cardinal, and B . < / arbitrary
groups. We set A D ˚< B and A D ˚< B . < /. Furthermore, let again
(9.13) be a -filtration of a group C of cardinality . Suppose that

Ext.C ; A / D 0 for all  < :

If the set

E D f <  j 9  >  such that Ext.C =C ; A =A / ¤ 0g

is stationary in , then Ext.C; A/ ¤ 0.

Proof. As noted above, there is no loss of generality in assuming  D  C 1 in
the application of E. Observing that Ext.C ; B / D 0 if   , the exact sequence
0 ! C ! CC1 ! CC1 =C ! 0 induces the exact sequence

Hom.CC1 ; B / ! Hom.C ; B / ! Ext.CC1 =C ; B / ! 0:

We now define homomorphisms  W C ! B . < /. If  2 E; then by

hypothesis the last Ext ¤ 0, and we can choose a  that has no extension to
274 9 Groups of Extensions and Cotorsion Groups

CC1 ! B . If  … E, then select  D 0. Define a direct system of splitting short

exact sequences e . < / with commutative diagrams

φν
ν: 0 −−−−→ Aν −−−−→ Aν ⊕ Cν −−−−→ Cν −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
 γν  
φν+1
ν+1 : 0 −−−−→ Aν+1 −−−−→ Aν+1 ⊕ Cν+1 −−−−→ Cν+1 −−−−→ 0

where the extremal vertical arrows are the injection maps,  ; C1 are the obvious
projections, while  is defined as follows:

 W .a ; c / ! .a C  c ; c /;

where a 2 A ; c 2 C . At limit ordinals , we form the direct limit of the system

fe j  < g which will clearly be an extension of A by C . Therefore, in view of
the hypothesis Ext.C ; A / D 0, it will be equivalent to, and thus identifiable with,
the splitting sequence e .

Suppose e W 0 ! A ! G!C ! 0 is the limit of the direct system fe j  < g
with the indicated maps, and let  W A ˚ C ! G denote the map induced by
the canonical morphisms e ! e. Thus C1  D  for all  < . By way of
contradiction, assume that e splits, and let W C ! G be a splitting map for ; thus,
we can write G D A ˚ C. jC j <  implies that maps C into the direct sum of
C and a set of less than  of the B ’s. Routine arguments lead us to the conclusion
that the set

S D f <  j C
 .A ˚ C /g

is a cub in . Evidently, for each  2 S, induces a splitting map  W C ! A ˚C

for  such that   D  C .
We will denote by  the composition of  with the projection onto A , and by
 the map  followed by the projection onto A. It is clear that for all c 2 C ;
  c D  . c ; c / D . c   c ; c / is equal to c D .0; c /, whence
 D  follows. Furthermore,   c also equals

C1   c DC1  . c ; c / D C1 . c C  c ; c /

D. c C  c  C1 c ; c /;

thus C1 c D  c C  c 2 A ˚ B D AC1 : Consequently, C1 followed

by the projection of AC1 onto B carries c 2 C onto  c ; therefore it yields a
homomorphism CC1 ! B extending  . This shows that  … E, i.e., E is not
stationary in . t
u
5 The Functor Pext 275

F Notes. Lemma 4.4 was published by Eklof–Fuchs [Annali Mat. Pura Appl. 90, 363–374
(1988)] for the special case (on valuation domain) needed there.
Let us call attention to the basic difference between Lemmas 4.3 and 4.4. In the former lemma,
the second argument of Ext was kept fixed, while in the latter lemma the second argument could
grow as large as necessary to match the size of the group in the first argument. This explains why
we needed the Diamond Principle in one case, but not in the other, this difference will be more
apparent in Sect. 7 in Chapter 13.

Exercises

(1) Suppose that A satisfies Ext.X; A/ D 0 for all rank 1 torsion-free groups X.
Then Ext.C; A/ D 0 for all torsion-free groups C.
(2) If Ext.Z.p/; A/ D 0 holds for A, then Ext.C; A/ D 0 for all p-groups C.
(3) If A satisfies Ext.Z.p/; A/ D 0 for all primes p, then A is divisible.
(4) Derive Pontryagin’s Theorem 7.1 in Chapter 3 from Lemma 4.1.
(5) (Eklof–Huber) If C is a torsion-free group of countable rank such that
Ext.B; A/ D 0 holds for all finite rank subgroups B < C, then also
Ext.C; A/ D 0 for any A.

5 The Functor Pext

That the extensions of A by C in which A is a pure subgroup (we will call them pure-
extensions) play a distinguished role does not seem to be obvious at the outset. The
truly significant thing here is that these extensions form a subgroup of Ext which
can be identified as the first Ulm subgroup of Ext.
Preliminary Lemma We recall that if ˛ W A ! A and W C ! C are
endomorphisms, then there are induced endomorphisms ˛ and  of Ext.C; A/.
We begin with investigating the actions of ˛ and  in Ext.C; A/. In the proof of
the next lemma, conveniently we may view A as a subgroup of B.
Lemma 5.1 (Baer [4]). Given the exact sequence

ˇ
eW 0 ! A ! B ! C ! 0 (9.16)

and the endomorphisms ˛ W A ! A and W C ! C, we have:

(i) e 2 Im ˛ if and only if A=˛A is a summand of B=˛A;
(ii) e 2 Im  if and only if A is a summand of ˇ 1 Ker ;
(iii) if Ker D 0, then e 2 Ker  if and only if ˇ 1 Im Š A ˚ Im .
276 9 Groups of Extensions and Cotorsion Groups

Proof.
(i) (Tellman [1]) The sequence Ext.C; ˛A/ ! Ext.C; A/ ! Ext.C; A=˛A/ ! 0 is
˛
obviously exact. Since the induced map Ext.C; A/ ! Ext.C; ˛A/ is surjective,
the sequence
˛
Ext.C; A/ ! Ext.C; A/ ! Ext.C; A=˛A/ ! 0

is also exact. It shows that e 2 Im ˛ exactly if e is mapped upon the splitting

extension of A=˛A by C. This is tantamount to saying that A=˛A is a summand
of B=˛A.
(ii) An argument similar to the one in (i) yields the exact sequence


Ext.C; A/ ! Ext.C; A/ ! Ext.Ker ; A/ ! 0:

This shows that the extension e satisfies e 2 Im  exactly if it is mapped upon

the splitting extension of A by Ker .
(iii) It is clear from Lemma 2.1 that e 2 Ker  if and only if the top row in the
pull-back diagram

γ : 0 −−−−→ A −−−−→ B  −−−−→ C −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐γ
  
β
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

is splitting. As is monic, this means that the sequence e W 0 ! A !

ˇ 1 Im ! Im ! 0 is splitting. u
t
Pext as First Ulm Subgroup of Ext Our main goal in proving the preceding
lemma was to apply it to the case when the endomorphisms are multiplications by
some non-zero integer n. From Lemma 3.1 we know that then the induced endomor-
phisms on Ext are likewise multiplications by the same n. Consequently, Lemma
5.1(i) tells us that (9.16) represents an element of n Ext.C; A/ if and only if A=nA is
a summand of B=nA. Observing that e 2 n Ext.C; A/ implies e 2 m Ext.C; A/ for all
mjn, we are now in a position to derive:
Theorem 5.2 (Nunke [1], Fuchs [AG]). The exact sequence (9.16) represents
(i) an element of n Ext.C; A/ if and only if mA D A \ mB for all mjn;
(ii) an element of the first Ulm subgroup Ext.C; A/1 if and only if it is pure-exact.
In other words,

Pext.C; A/ D Ext.C; A/1 ;

where Pext.C; A/ denotes the set of pure-extensions of A by C. t

u
5 The Functor Pext 277

Thus Pext.C; A/ is not only a subset, it is also a subgroup of Ext.C; A/, called the
group of pure-extensions.
˛ ˇ
If e W 0 ! A!B!C ! 0 is a pure-extension of A by C, and if  W
A ! A0 ;  W C0 ! C are homomorphisms, then the extensions e and e are
again pure-extensions, this is easy to check directly, but also follows immediately
from Theorem 5.2 by observing that the map Ext.; / W Ext.C; A/ ! Ext.C0 ; A0 /
carries Ulm subgroups into Ulm subgroups. We can thus claim that Pext is an
additive bifunctor Ab  Ab ! Ab, contravariant in the first, and covariant in
the second variable.
From Theorem 4.3 in Chapter 5 we derive at once:
Corollary 5.3. A group C satisfies Pext.C; A/ D 0 for all groups A if and only if it
is †-cyclic. t
u
Exact Sequences for Pext Having had a first glimpse of Pext, it looks interest-
ing enough to pursue its properties. To reveal its behavior towards exact sequences,
we prove a couple of results. The next two lemmas are more general than we need
them right now, but we will find their generality useful later on.
˛ ˇ
Lemma 5.4. Suppose 0 ! A!B!C ! 0 is an exact sequence such that ˛A is
contained in the th Ulm subgroup B of B. Then ˇ maps B onto C .
Similarly, if ˛A
p B, then ˇ.p B/ D p C:
Proof. It is pretty clear that ˇ.B /
C , so only the surjectivity requires a proof.
Since Ker ˇ
B , the transfinite heights of the elements of B not in B are the
same in every coset mod Ker ˇ. As the height of an element in an epic image is
the supremum of the heights of its preimages, no element can belong to C without
being contained in ˇ.B /.
The proof for p B is the same. t
u
˛ ˇ
Lemma 5.5. If e W 0 ! A!B!C ! 0 represents an element in the th
Ulm subgroup Ext.C; A/ of Ext.C; A/, then for every group G, the connecting
homomorphisms ı  and ı act as

ı  W Hom.A; G/ ! Ext.C; G/ ; ı W Hom.G; C/ ! Ext.G; A/ :

In particular, if e is pure-exact, then Im ı  and Im ı are contained in Pext.C; G/

and Pext.G; A/, respectively.
Similarly, if e 2 p Ext.C; A/, then Im ı  2 p Ext.C; G/ and Im ı 2

p Ext.G; A/.
Proof. An  2 Hom.A; G/ induces a map  W Ext.C; A/ ! Ext.C; G/ that carries
the top row into the bottom row in the commutative diagram
278 9 Groups of Extensions and Cotorsion Groups

α β
: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
⏐ ⏐ 
⏐
η
⏐ 
 
β
η∗ : 0 −−−−→ G −−−−→ B  −−−−→ C −−−−→ 0

Homomorphisms map Ulm subgroups into Ulm subgroups, so necessarily  e 2

Ext.C; G/ . But the bottom row is exactly the image of  under ı  , whence the
claim is immediate. The dual proof establishes the statement involving ı , while the
last claim is a simple corollary, as Lemma 5.4 is available. t
u
Now we can fit together the pieces of information we obtained in these lemmas
to prove that Pext admits the same kind of long exact sequences as Ext does.
˛ ˇ
Theorem 5.6 (Harrison [1]). Let 0 ! A!B!C ! 0 be a pure-exact
sequence. Then, for every group G, the following induced sequences are exact:

0 ! Hom.C; G/ ! Hom.B; G/ ! Hom.A; G/ !

ı ˇ ˛
! Pext.C; G/! Pext.B; G/! Pext.A; G/ ! 0 (9.17)

and

0 ! Hom.G; A/ ! Hom.G; B/ ! Hom.G; C/ !

ı ˛ ˇ
! Pext.G; A/! Pext.G; B/! Pext.G; C/ ! 0: (9.18)

All the maps are natural.

Proof. In view of Lemma 5.5 and the functorial behavior of Pext, it is evident that
(9.17) and (9.18) make sense (i.e., the images of the indicated maps are correct),
only their exactness has to be established. We do not have to worry about exactness
at the Homs (see Theorem 2.3), so let us concentrate on the Pext part. Theorem 2.3
also assures the exactness of the sequence

ı ˇ ˛
Hom.A; G/! Ext.C; G/! Ext.B; G/! Ext.A; G/ ! 0:

By Lemma 5.5, Im ı 
Ext.C; G/1 , so from Theorem 5.2 we infer that ˇ 
maps Ext.C; G/1 D Pext.C; G/ upon .Im ˇ  /1
Ext.B; G/1 , thus ˛  for Pext
is surjective. But then .Im ˇ  /1
Ext.B; G/1 cannot be a proper inclusion. This
establishes the exactness of (9.17). The proof of (9.18) is similar. t
u
We observe that Theorem 5.6 also holds if we start with a p-pure-exact sequence,
and assume that G is a p-adic group.
5 The Functor Pext 279

Pext and Algebraic Compactness Pext turns out to be a most versatile device
to test algebraic compactness. First of all, it is clear that the definition of algebraic
compactness can be rephrased by saying that A is algebraically compact if and only
if Pext.C; A/ D 0 for all groups C. A reduction to certain C yields an especially
convenient criterion (Proposition 5.8). But first a comment that will be used in the
next proof.
Assume A is reduced and algebraically compact. The exact sequence 0 ! Z !
Q ! Q=Z ! 0 induces the exact sequence 0 ! Hom.Z; A/ ! Ext.Q=Z; A/ !
Ext.Q; A/ D 0, where the last Ext vanishes by the algebraic compactness of
A. Hence we obtain the natural isomorphism Ext.Q=Z; A/ Š A for a reduced
algebraically compact (= complete) A.
Proposition 5.7. Let A be a group such that A1 D 0. Then

0 ! Pext.Q=Z; A/ ! Ext.Q=Z; A/ ! AQ ! 0

is an exact sequence where AQ denotes the Z-adic completion of A. There is a natural

isomorphism

Q
Pext.Q=Z; A/ Š Hom.Q=Z; A=A/:

Proof. If A1 D 0, then A is a pure subgroup of its completion, and we can use the
pure-exact sequence 0 ! A ! AQ ! A=A Q ! 0 to derive two exact sequences

Q
0 ! Hom.Q=Z; A=A/ Q Š AQ ! 0
! Ext.Q=Z; A/ ! Ext.Q=Z; A/

and

Q
0 ! Hom.Q=Z; A=A/ Q D 0;
! Pext.Q=Z; A/ ! Pext.Q=Z; A/

Q
where we have used the divisibility of A=A to get 0 at the right end of the first
sequence. Hence the claims are evident. t
u
Observe that Proposition 5.7 also tells us the interesting fact that the completion
of a group A with trivial Ulm subgroup can also be obtained by forming the initial
Ulm factor of Ext.Q=Z; A/.
Proposition 5.8. A group A is algebraically compact if and only if it satisfies

Ext.Q; A/ D 0 and Pext.Q=Z; A/ D 0:

Proof. For the proof of sufficiency, we assume that A is reduced and the indicated
Ext and Pext vanish. Let D be the divisible hull of A and D0 the divisible hull of A1
in D. Then A C D0 contains A as a pure subgroup with .A C D0 /=A torsion divisible.
The assumption Pext.Q=Z; A/ D 0 implies Pext.E; A/ D 0 for all torsion divisible
E, so A is a summand in ACD0 . This can happen only if D0 D 0, and A1 D 0. Thus A
280 9 Groups of Extensions and Cotorsion Groups

is a pure subgroup in AQ with divisible A=A.

Q Q
Now A=A cannot contain any copy of
Q, since Ext.E; A/ D 0 for all torsion-free divisible E, and neither can it contain
any torsion divisible subgroup because of Hom.Q=Z; A=A/Q Š Pext.Q=Z; A/ D 0.
Q
Consequently, A=A D 0, establishing the claim. t
u
More on Pext In general, neither Ext nor Pext needs to convert a direct limit to
an inverse limit of Exts or Pexts, but in the countable case something definite can be
stated.
Lemma 5.9 (Nunke [1]). Let Cn .n < !/ be a direct system with connecting maps
n W Cn ! CnC1 , and limit C. Then the natural map

W Ext.C; A/ ! lim Ext.Cn ; A/ (9.19)


n<!

is surjective.
Proof. The direct system of the Cn gives rise to an inverse system with connecting
maps n W Ext.CnC1 ; A/ ! Ext.Cn ; A/. Since the natural maps n W Cn ! C induce

n W Ext.C; A/ ! Ext.Cn ; A/ which satisfy n n D nC1 , is well defined. To
prove its surjectivity, let e be an element in the inverse limit, say, represented by a
sequence of extensions en 2 Ext.Cn ; A/ .n < !/ fitting in the commutative diagrams

n: 0 −−−−→ A −−−−→ Bn −−−−→ Cn −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐φ
   n

n+1 : 0 −−−−→ A −−−−→ Bn+1 −−−−→ Cn+1 −−−−→ 0

It is clear that then the en form a direct system of exact sequences whose limit
e belongs to Ext.C; A/. A straightforward check convinces us that .e / must be
equal to e. t
u
In the following statement we keep the notations of the preceding lemma.
Proposition 5.10. Let C be a countable group which is the union of the chain
Cn .n < !/ of finitely generated subgroups. Then
(i) Ker D Pext.C; A/.
(ii) (Jensen [Je]) Pext.C; A/ Š lim1 Hom.Cn ; A/.

Proof.
(i) The inclusions n W Cn ! C induce epimorphisms n W Ext.C; A/
! Ext.Cn ; A/, and evidently, Ker is contained in the intersection of the
Ker n . As an extension 0 ! A ! B ! C ! 0 belongs to Pext.C; A/ if
and only if it is carried to 0 by all n , the claim is evident.
5 The Functor Pext 281

(ii) (Schochet [1]) We observe that C fits in the pure-exact sequence 0 !

˚n<! Cn ! ˚n<! Cn ! C ! 0 where is the Eilenberg map. By making use
of Theorem 5.6, hence we deduce the exact sequence

Hom.˚n<! Cn ; A/ ! Hom.˚n<! Cn ; A/ ! Pext.C; A/ ! 0:

These Homs can be rewritten as direct products, so that the cokernel of

the Eilenberg map between them is lim1 Hom.Cn ; A/ (see, e.g., Sect. 5 in

Chapter 2). t
u
F Notes. It is remarkable that the pure-extensions form the first Ulm subgroup of Ext. This
fact is the source of several generalizations of purity for groups, see e.g. Sect. 8 in Chapter 11. For
modules over integral domains, in general, the first Ulm submodule of Ext fails to be the collection
of pure-extensions.
Mekler [1] proves that it is consistent with GCH that there exist non-†-cyclic p-groups C such
that Ext.C; A/ D 0 for all countable groups A. Compare this with Corollary 5.3. For more on
Ext.C; A/, see Harrison [3].

Exercises

(1) Prove that an extension equivalent to a pure-extension is also a pure-extension.

(2) If C is torsion-free, then Pext.C; A/ D Ext.C; A/ for every A. Conclude that
Ext.C; A/ is divisible provided C is torsion-free.
(3) (Irwin–Walker–Walker) An extension e W 0 ! A ! B ! C ! 0 belongs
to the divisible subgroup of Ext.C; A/ if and only if the induced sequence
0 ! tA ! tB ! tC ! 0 is splitting exact.
(4) e W 0 ! A ! B ! C ! 0 is in the Frattini subgroup of Ext.C; A/ exactly if it
is neat-exact.
(5) Prove Ext.C; A=pA/ Š Ext.CŒp
; A=pA/ Š Ext.CŒp
; A/ for all A; C.
(6) Verify the natural isomorphisms
!
Y Y Y
Pext.˚i2I Ci ; A/ Š Pext.Ci ; A/; Pext C; Ai Š Pext.C; Ai /:
i2I i2I i2I

(7) If C is torsion, then Pext.C; A/ Š Pext.C; tA/: [Hint: Lemma 1.4 in Chapter
5).]
(8) If A is a †-cyclic p-group and C is a divisible p-group, then those extensions
of A by C in which A is a basic subgroup form a subgroup in Ext.C; A/.
Q
(9) If A1 D 0, then Ext.Q; A/ Š Hom.Q; A=A/:
(10) If 0 ! A ! B ! C ! 0 is a pure-exact sequence and F is †-cyclic, then the
sequence 0 ! Ext.F; A/ ! Ext.F; B/ ! Ext.F; C/ ! 0 is splitting exact.
(11) (Schoeman) A group C has the property that Ext.C; A/ is algebraically
compact for all groups A if and only if tC is †-cyclic.
282 9 Groups of Extensions and Cotorsion Groups

6 Cotorsion Groups

In this section we get acquainted with a remarkable class of groups which fits
perfectly into the structure theory of the groups of extensions. They were introduced
independently and almost simultaneously by Harrison [1], Nunke [1], and Fuchs
[11]. The name was coined by Harrison who pointed out their dual behavior to
torsion groups (Theorem 7.4).
Cotorsion Groups Here is the definition: a group G is called cotorsion if it
satisfies

Ext.A; G/ D 0 for all torsion-free groups A:

Visibly, cotorsion generalizes the concept of algebraic compactness.

Every torsion-free group A can be embedded in a direct sum of copies
Q of Q. The
inclusion map A ! ˚ Q implies that the sequence Ext.˚ Q; G/ D Ext.Q; G/ !
Ext.A; G/ ! 0 is exact. Consequently, the single equality

Ext.Q; G/ D 0

suffices to guarantee that the group G is cotorsion. This is a handy criterion for
cotorsionness.
Example 6.1. From Theorem 3.8 we can deduce that every Ext is cotorsion. Indeed, we have
Ext.Q; Ext.C; A// Š Ext.Tor.Q; C/; A/ D 0, since the Tor vanishes. (We will give another proof
in Theorem 6.5 that is independent of Theorem 3.8 whose proof has been omitted.)

As usual, we start with a list of elementary consequences of the definition. The

immediate goal is to show that the class of cotorsion groups is closed under direct
products, extensions and epic images.
Q
(A) A direct product i2I Gi is cotorsion if and only if every summand Gi is
cotorsion.
Q This isQa straightforward consequence of the natural isomorphism
Ext.Q; i Gi / Š i Ext.Q; Gi /.
(B) A group G is cotorsion if a subgroup H and the factor group G=H are
cotorsion. The exact sequence 0 ! H ! G ! G=H ! 0 implies the
exactness of Ext.Q; H/ ! Ext.Q; G/ ! Ext.Q; G=H/.
(C) Epimorphic images of cotorsion groups are cotorsion. If G is cotorsion, and
H is an epic image of G, then the sequence Ext.Q; G/ ! Ext.Q; H/ ! 0 is
exact, whence the claim is evident.
(D) Assume G is reduced and cotorsion. A subgroup H of G is cotor-
sion exactly if the factor group G=H is reduced. The exact sequence
0 ! H ! G ! G=H ! 0 leads to the exact sequence

0 D Hom.Q; G/ ! Hom.Q; G=H/ ! Ext.Q; H/ ! Ext.Q; G/ D 0:

6 Cotorsion Groups 283

Thus the middle terms are isomorphic; this shows that Ext.Q; H/ D 0 if and
only if Hom.Q; G=H/ D 0.
(E) For every endomorphism  of a reduced cotorsion group G, both Ker  and
Im  are cotorsion. The claim follows at once from (D) and (C), respectively.
(F) Inverse limits of reduced cotorsion groups are again reduced cotorsion.QThe
inverse limit G of reduced groups Gi is a subgroup in the direct product i Gi
which is, by (A), cotorsion (and reduced)
Q if so are the Gi . Because of (D), it
remains to show that the factor group i Gi =G is reduced.Q Recall that G 
was
the intersection of kernels of certain
Q endomorphisms j of i Gi (see Sect. 5(C)
in Chapter 2), and consequently, i Gi =G is a subdirect product of the groups
Im j . These groups are reduced, and hence the claim is immediate.
Characterization of Cotorsion Groups The first evidence of the intimate
relation between Ext and cotorsion groups is our next theorem.
Theorem 6.2. A reduced group G is cotorsion if and only if there is an isomorphism

Ext.Q=Z; G/ Š G:

Q
A reduced cotorsion group is a Z-module.
Proof. If this isomorphism holds, G is cotorsion by Example 6.1 (it will follow
from Theorem 6.5 too). For the converse, we start off with the exact sequence
0 ! Z ! Q ! Q=Z ! 0 which induces the exact sequence

0 D Hom.Q; G/ ! Hom.Z; G/ Š G ! Ext.Q=Z; G/ ! Ext.Q; G/ ! 0:

The connecting homomorphism between Hom and Ext is an isomorphism if and

only if the last Ext vanishes. This is the case if and only if G is cotorsion. As Q=Z
Q
is a Z-module, so is Ext.Q=Z; G/ by the induced endomorphisms. t
u
Note that the proof shows that the isomorphism in the last theorem may be
regarded to be natural for reduced cotorsion groups G.
Corollary 6.3. A reduced cotorsion group is algebraically compact exactly if it is
Hausdorff in the Z-adic topology.
Proof. A reduced cotorsion group G satisfies Ext.Q=Z; G/ Š G, therefore,
Pext.Q=Z; G/ Š G1 : G is Hausdorff if and only if G1 D 0, and the claim follows
from Proposition 5.8. t
u
Structure of Cotorsion Groups The last theorem has an important conse-
quence: the structure theory of cotorsion groups can be reduced to the p-adic case.
Indeed, cotorsion groups enjoy the following canonical decomposition.
Theorem 6.4. A reduced cotorsion group G is a direct product,
Y
GD Gp ;
p
284 9 Groups of Extensions and Cotorsion Groups

where, for each prime p, Gp is a reduced cotorsion group which is a module over the
ring Jp of p-adic integers. The Gp are uniquely determined fully invariant subgroups.
1
Proof. In view of the isomorphism
1
Q Q=Z Š ˚p Z.p /, the 1 preceding theorem
implies G Š Ext.˚p Z.p /; G/ D p Gp where Gp D Ext.Z.p /; G/ is cotorsion.
Gp is the intersection of all qn G for primes q ¤ p and all n < !, so it is fully
invariant in G. t
u
More on Ext The best way to explain the significance of cotorsion groups is
to present a few illustrative applications. The first and foremost one is (i) in the
following theorem that was the source for inspiration of the concept “cotorsion”
(the name is a different story).
Theorem 6.5.
(i) The group Ext.C; A/ is cotorsion for all groups A and C.
(ii) If C is torsion, then Ext.C; A/ is reduced cotorsion for all A.
Proof. From the injective resolution 0 ! A ! E ! D ! 0 of A (where E; D
are injective) we derive the exact sequence 0 ! Hom.C; A/ ! Hom.C; E/ !
Hom.C; D/ ! Ext.C; A/ ! 0.
(i) As D is injective, Hom.C; D/ is algebraically compact (Theorem 2.11 in
Chapter 7), so by (C), its epic image Ext.C; A/ is cotorsion.
(ii) If C is a p-group, then the Homs are reduced and algebraically compact (Theo-
rem 2.1 in Chapter 7), so the image X of the monic map between the two first
Homs is cotorsion. In view of the exact sequence 0 ! X ! Hom.C; D/ !
Ext.C; A/ ! 0, property (D) implies that Ext.C; A/ is reduced. If C is torsion,
then Ext.C; A/ is a product of reduced cotorsion groups, so is itself of the same
kind. t
u
We offer one more result on Ext showing that there are cases in which Ext is
not only cotorsion, but even algebraically compact. Recall that by Corollary 3.6,
Ext.C; Z/ is compact if C is torsion.
Proposition 6.6.
(i) For every C; Ext.C; Z/ is algebraically compact.
(ii) Ext.C; A/ is algebraically compact if so is A.
Proof.
(i) The exact sequence 0 ! Z ! R ! R=Z ! 0 induces the exact
sequence Hom.C; R/ ! Hom.C; R=Z/ ! Ext.C; Z/ ! Ext.C; R/ D 0:
Here Hom.C; R/ is torsion-free divisible, so its image in Hom.C; R=Z/ is a
summand. Therefore, Ext.C; Z/ is isomorphic to a summand of the compact
group Hom.C; R=Z/.
(ii) By Theorem 1.2 in Chapter 6, an algebraically compact A is a summand of a
direct product of cocylic groups. Hence Ext.C; A/ is a summand of a product of
groups Ext.C; Z.pk // for various primes p and various k 2 N [ f1g. These
6 Cotorsion Groups 285

groups are bounded and 0, respectively. Thus Ext.C; A/ is a summand of a

product of bounded groups, so algebraically compact. u
t
Let us stop for a moment to record a useful consequence of the proof of
Theorem 6.5.
Corollary 6.7. Every reduced group G can be embedded in the cotorsion group

G D Ext.Q=Z; G/

such that G =G is torsion-free divisible.

Proof. Consider the exact sequence 0 ! G Š Hom.Z; G/ ! Ext.Q=Z; G/ !
Ext.Q; G/ ! 0. We know from Theorem 6.5 that G D Ext.Q=Z; G/ is cotorsion,
and from Sect. 3(E) that Ext.Q; / is torsion-free and divisible. t
u
Occurrence of Cotorsion Groups In addition to Theorem 6.5, the following
two propositions afford classes of natural examples of cotorsion groups.
Proposition 6.8 (De Marco–Orsatti [1]). Let G be a topological group in a
metrizable linear topology, and Ǧ its completion in this topology. Then Ǧ=G is a
cotorsion group.
Proof. Assuming G non-discrete, let G D U1 > U2 >    > Un >    be a
base of neighborhoods
Q for 0 consisting of subgroups with \n Un D 0. Define the
map
P W U n ! Ǧ by P vector .u1 ; : : : ; un ; : : : / .un 2 Un / to
sending the infinite
un 2 Ǧ (note that the infinite series un converges in Ǧ, since Q the sequence
of the un converges to 0). Manifestly, maps the finite vectors in Un into G.
Furthermore, Im DǦ, since every element of Ǧ is the limitQof a Cauchy sequence
in G. Hence Ǧ=G is an epic image of the factor group Un = ˚ Un which is
by Corollary 1.12 in Chapter 6 an algebraically compact group. Therefore, Ǧ=G
is cotorsion. t
u
As a final illustration we mention the first derived functor lim1 of the inverse

limit functor (see Sect. 5 in Chapter 2).
Proposition 6.9 (Huber–Warfield [1]). If fAn .n < !/g is a countable inverse
system of groups, indexed by the natural numbers, then the derived functor yields a
cotorsion group lim1 An .

Proof. Let ˛n W AnC1 ! An denote the connecting homomorphisms in the inverse
system. The derived functor lim1 An is isomorphic to the cokernel of the Eilenberg
Q Q 
map ı W An ! An given by

ı W .a0 ; a1 ; : : : ; an ; : : : / 7! .a0  ˛0 .a1 /; a1  ˛1 .a2 /; : : : ; an  ˛n .anC1 /; : : : /:

It is clear that Im ı contains the direct sum of the An , and therefore lim1 An is an
Q 
epimorphic image of the factor group An = ˚ An . As above, it follows that lim1 An

is cotorsion. t
u
286 9 Groups of Extensions and Cotorsion Groups

When Cotorsion Is a Direct Sum We conclude this section with an interesting

property of cotorsion groups that is related to their direct decompositions.
Proposition 6.10. Let G D ˚i2I Ci be a direct decomposition of a reduced
cotorsion group. Then there is an integer m > 0 such that

mCi D 0 for almost all i 2 I:

Proof. We anticipate Proposition 8.3 which states that the Ulm factors of cotor-
sion groups are algebraically compact. If we factor out the first Ulm subgroups
throughout, then we get a direct decomposition G=G1 D ˚i2I Ci =Ci1 of the reduced
algebraically compact group G=G1 . By Theorem 2.17 in Chapter 6, there is an
integer m > 0 such that almost all summands Ci =Ci1 are m-bounded. If Ci =Ci1 is
bounded, then Ci1 ought to be 0. It follows that almost all of the summands Ci are
bounded by m. t
u
F Notes. Huber–Warfield [1] also proved that every cotorsion group can be obtained in the
way described in Proposition 6.9, and if we allow systems of length !1 as well, then every group
is obtainable as lim1 .

Cotorsion modules over integral domains were studied extensively by E. Matlis [Memoirs
Amer. Math. Soc. 49 (1964)]. Other generalizations were given by E. Enochs and R. Warfield,
Jr. The theory of cotorsion modules is flourishing, since Salce [2] initiated the theory of cotorsion
pairs. Cf. the monograph Approximations and Endomorphism Algebras of Modules (2006) by R.
Göbel and J. Trlifaj. The cotorsion theories for abelian groups were described by Göbel–Shelah–
Wallutis [1].

Exercises

(1) A countable cotorsion group is a direct sum of a divisible and a bounded group.
[Hint: consider G=G1 .]
(2) Let G be a cotorsion group, and D the divisible hull of G1 . Then G C D is the
pure-injective hull of G.
(3) A is torsion-free if Ext.A; G/ D 0 for all (reduced) cotorsion groups G.
(4) For a cotorsion G, there is a natural isomorphism Ext.C; G/ Š Ext.tC; G/ for
every C.
(5) Suppose 0 ! A ! B ! C ! 0 is pure-exact, and G is algebraically compact.
Argue that the induced sequence 0 ! Ext.C; G/ ! Ext.B; G/ ! Ext.A; G/ !
0 is splitting exact. [Hint: Lemma 5.5, Proposition 6.6(ii).]
(6) If G1 D 0, then .G /1 Š Hom.Q=Z; G=G/ Q Q
and G =.G /1 Š G.
(7) For all A; C, Ext.C; A/ admits a decomposition Ext.C; A/ Š D ˚ Ext.tC; A/
where the first summand is an epimorphic image of the divisible group
Ext.C=tC; A/, and the second summand is a reduced group. [Hint: Theorem
6.5(ii).]
7 Cotorsion vs. Torsion 287

Q
(8) (Wald) If Gi .i 2 I/ are cotorsion groups, then so is the subgroup of i2I Gi
that consists of the vectors whose supports are countable. [Hint: consider
Ext.Z.p1 /; /.]
(9) (Wald) G is cotorsion if every countable subgroup of G is contained in a
cotorsion subgroup of G. [Hint: Ext.Q; G/ D 0.]

7 Cotorsion vs. Torsion

The material of this section constitutes an important part of our study of cotorsion
groups. The highlight is a decomposition of reduced cotorsion groups into two
summands, mimicking the decompositions of torsion groups into reduced and
divisible parts. The relationship to torsion groups is much deeper than one would
expect, torsion and cotorsion are related to an amazing extent: the two summands
belong to categories which are equivalent to the subcategories of reduced torsion and
divisible torsion groups, respectively. These category equivalences were discovered
by Harrison [1].
Adjusted Cotorsion Groups To begin with, we state a relevant definition.
A cotorsion group that is reduced and admits no non-zero torsion-free summands is
called adjusted. (This simply says that it is reduced and no Jp is a summand.)
Lemma 7.1 (Harrison [1]). Let T be a reduced torsion group. Then Ext.Q=Z; T/
is an adjusted cotorsion group whose torsion subgroup is isomorphic to T, and
whose factor group modulo T is .torsion-free/ divisible.
Proof. From the standard exact sequence 0 ! Z ! Q ! Q=Z ! 0 we derive the
exact sequence

0 ! Hom.Z; T/ Š T ! Ext.Q=Z; T/ ! Ext.Q; T/ ! Ext.Z; T/ D 0:

Thus the claim on the factor group will follow as soon as we can show that
Ext.Q; T/ is torsion-free and divisible. But this is obvious from Sect. 3(E). What
we have proved so far implies that any torsion-free summand of Ext.Q=Z; T/ must
be divisible. Because of Theorem 6.5(ii), it has to be 0. t
u
Lemma 7.2. If T denotes the torsion subgroup of a group G, then

Ext.Q=Z; G/ Š Ext.Q=Z; T/ ˚ Ext.Q=Z; G=T/:

Proof. From the exact sequence 0 ! T ! G ! G=T ! 0 we obtain the exact

sequence

0 ! Ext.Q=Z; T/ ! Ext.Q=Z; G/ ! Ext.Q=Z; G=T/ ! 0:

288 9 Groups of Extensions and Cotorsion Groups

In view of Theorem 3.5 we know that the last Ext is Š Hom.Q=Z; D/ where D is the
cokernel of G=T in its divisible hull. Thus this Ext is the direct product of groups
Hom.Z.p1 /; D/ with p ranging over various primes, and Sect. 1(G) in Chapter 7
shows that these Homs are torsion-free. Since every Ext is cotorsion, the displayed
exact sequence must split. t
u
Canonical Decomposition of Cotorsion Groups We do not need any more
preparation for the proof of the basic decomposition theorem of reduced cotorsion
groups mentioned above.
Theorem 7.3 (Harrison [1]). Let G denote a reduced cotorsion group, and T its
torsion subgroup. There is a decomposition

G Š Ext.Q=Z; T/ ˚ Ext.Q=Z; G=T/;

where the first summand is an adjusted cotorsion group, and the second summand
is a torsion-free algebraically compact group. This direct decomposition of G into
an adjusted cotorsion and a torsion-free algebraically compact group is unique up
to isomorphism.
Proof. Combining Theorem 6.2 and Lemma 7.2, we obtain the stated decom-
position. The first summand is, by Lemma 7.1, an adjusted cotorsion group
(the cotorsion hull of T, Sect. 9), while the second summand is torsion-free and
algebraically compact as a consequence of Theorem 3.5. The uniqueness of the
decomposition follows from the fact that if G D C ˚ A is a decomposition with C
adjusted, and A torsion-free algebraically compact, then T
C and T  =T divisible
imply T 
C. Hence T  D Ext.Q=Z; T/ is a summand of C, but it cannot be a
proper one, because a complement would be torsion-free divisible. t
u
Accordingly, every cotorsion group G decomposes, uniquely up to isomorphism,
as G D D ˚ C ˚ A where D is divisible, C is adjusted cotorsion, and A is
reduced torsion-free algebraically compact. The summands D and A can fully be
characterized by cardinal invariants, while the classification of adjusted cotorsion
groups remains open, since it is equivalent to the unsolved classification problem
of torsion groups; this equivalence will be more transparent in the light of Theorem
7.4 below.
Two Category Equivalences We are now entering the discussion of two cate-
gory equivalences between certain subcategories of cotorsion and torsion groups.
Consider the category T of reduced torsion groups and the category C of adjusted
cotorsion groups; we regard them as full subcategories of Ab. Observe that if  W
T ! T 0 is a map between two reduced torsion groups, then there is an induced map
  W Ext.Q=Z; T/ ! Ext.Q=Z; T 0 / which is canonical, so that it is an isomorphism
provided so is .
We have all the tools in our arsenal to verify the following remarkable dual roles
of the functors Ext and Tor.
7 Cotorsion vs. Torsion 289

Theorem 7.4 (Harrison [1]). The category T of reduced torsion groups and the
category C of adjusted cotorsion groups are equivalent categories. The equivalence
is provided by the covariant functors:

Ext.Q=Z; / W T ! C and Tor.Q=Z; / W C ! T :

Proof. If T 2 T , then Ext.Q=Z; T/ 2 C as stated in Lemma 7.1. The same lemma

implies Tor.Q=Z; Ext.Q=Z; T// Š T, remembering that the functor Tor.Q=Z; /
picks up the torsion subgroup (Theorem 2.3 in Chapter 8). On the other hand, if
C 2 C and T D tC, then Tor.Q=Z; C/ Š T, and C0 D Ext.Q=Z; T/ is a reduced
adjusted cotorsion group with torsion subgroup Š T. But it follows from Lemma
7.1 straightforwardly that if both C and C0 are reduced adjusted cotorsion groups
with isomorphic torsion subgroups, then they are themselves isomorphic. Therefore,
Ext.Q=Z; Tor.Q=Z; C// Š C. t
u
Example 7.5. Let T be torsion and Hausdorff in the Z-adic topology (i.e., T 1 D 0). Then T  D
Ext.Q=Z; T/ is reduced, an extension of T by the divisible torsion-free group Ext.Q; T/. Its first
Ulm subgroup is the algebraically compact Pext.Q=Z; T/ Š Hom.Q=Z; TQ =T/ modulo which T 
Q
is isomorphic to the completion T.

The other category equivalence is between the category D of divisible torsion

groups and the category F of reduced torsion-free algebraically compact groups.
Both are viewed as full subcategories of Ab. Since the reduced algebraically
compact groups are exactly the complete groups, we will switch to the shorter name
in formulating the following theorem.
Theorem 7.6 (Harrison [1]). The category D of divisible torsion groups is equiv-
alent to the category F of complete torsion-free groups. The equivalence is given by
the covariant functors

Hom.Q=Z; / W D ! F and Q=Z ˝  W F ! D:

Proof. Let C denote a complete torsion-free group, and E its divisible hull. In the
arising exact sequence 0 ! C ! E ! D ! 0, the group D is divisible torsion.
The long exact sequences for Tor-tensor and for Hom-Ext yield

0 ! Tor.Q=Z; D/ Š D ! Q=Z ˝ C ! 0

and

0 ! Hom.Q=Z; D/ ! Ext.Q=Z; C/ Š C ! 0;

respectively. We have taken into account that Tor.Q=Z; / selects the torsion
subgroup, and Hom.Q=Z; / ignores torsion-free groups. The last isomorphism is
supported by Theorem 6.2. Thus there are natural isomorphisms D Š Q=Z ˝ C and
C Š Hom.Q=Z; D/, which proves that the given functors induce an equivalence
290 9 Groups of Extensions and Cotorsion Groups

on the objects. They act in the same way on the morphisms, as witnessed by the
commutative diagrams

δ
D −−−−→ D
⏐ ⏐
⏐
Hom(Q/Z,∗)
⏐Hom(Q/Z,∗)

Hom(Q/Z,δ)
C −−−−−−−−→ C 

γ
C −−−−→ C
⏐ ⏐
⏐
Tor(Q/Z,∗)
⏐Tor(Q/Z,∗)

Tor(Q/Z,γ)
D −−−−−−−→ D
t
u

An especially convenient and handy way to capture the correspondence D $ C

is via the exact sequence 0 ! C ! E ! D ! 0, where, we repeat, E is the
divisible hull of C, a direct sum of copies of Q.
Example 7.7. If D D Z.p1 /, then C D Hom.Q=Z; Z.p1 // Š Jp . There is an exact sequence
0 ! Jp ! ˚Q ! Z.p1 / ! 0. If D D Q=Z, then C D Hom.Q=Z; Q=Z/ Š Z, Q and the
Q
sequence 0 ! Z ! ˚Q ! Q=Z ! 0 is exact.
F Notes. The dual behavior of Ext-Tor as well as Hom-tensor is most interesting and most
important; there are other similar, less relevant, examples of duality in Homological Algebra.
While the generalization of Theorem 7.6 has widespread applications in the theory of modules
over integral domains (this is the Matlis category equivalence), so far Theorem 7.4 remains an
isolated result, though it easily generalizes to modules over integral domains.

Exercises

(1) A non-zero cotorsion group has a summand isomorphic to one of the following
groups: Q, Z.pk / .k
1/, Jp , for some prime p.
(2) Show that T \ C is the class of bounded groups.
(3) Let D D Z.p1 /.@0 / . Find Hom.Q=Z; D/.
(4) If G is adjusted cotorsion, then jGj
jtGj@0 .
(5) In the category equivalence of Theorem 7.6, what subcategory of F corresponds
to the category of divisible p-groups?
(6) If G; H are adjusted cotorsion groups, then Hom.G; H/ Š Hom.tG; tH/
naturally.
(7) End G Š End.tG/ for an adjusted cotorsion G.
8 More on Ext 291

8 More on Ext

Cotorsion Groups and Algebraic Compactness We continue our analysis of

cotorsion groups. First, let us mention a few more results on the relationship between
cotorsion and algebraically compact groups. As we shall see later on, they can be
quite useful.
Lemma 8.1.
(i) A reduced cotorsion group is algebraically compact if and only if its first Ulm
subgroup vanishes.
(ii) A torsion-free group is cotorsion if and only if it is algebraically compact.
Proof.
(i) This is basically equivalent to Corollary 6.3.
(ii) It is enough to prove this for a reduced group G. If G is torsion-free and reduced,
then G1 D 0. The claim follows from (i). t
u
Proposition 8.2. Every reduced cotorsion group is a quotient of a torsion-free
algebraically compact group modulo an algebraically compact subgroup.
Proof. Let 0 ! G ! E ! D ! 0 be an injective resolution of the reduced
cotorsion group G. We obtain the induced exact sequence

0 ! Hom.Q=Z; E/ ! Hom.Q=Z; D/ ! Ext.Q=Z; G/ ! Ext.Q=Z; E/ D 0:

The Homs are torsion-free and algebraically compact, while the first Ext is Š G, as
stated in Theorem 6.2. t
u
The Ulm Factors of Cotorsion Groups The last proposition confirms that
cotorsion groups can be derived from algebraically compact groups as epic images.
Miraculously, algebraically compact groups are implanted in a natural fashion in
cotorsion groups:
Proposition 8.3 (Fuchs [11]). Ulm subgroups of cotorsion groups are cotorsion,
and Ulm factors of cotorsion groups are algebraically compact.
Proof. If G is the th Ulm subgroup of the cotorsion group G, then G=G is a
reduced group. Hence the first statement follows from Sect. 6(D). Then also the th
Ulm factor G =G C1 is cotorsion, as it is guaranteed by Sect. 6(C). Its first Ulm
subgroup vanishes, so we can finish the proof by invoking Lemma 8.1(i). t
u
We note, in passing, that this result can be complemented by claiming that, in
case G has finite Ulm length, a converse is also true: if the Ulm factors of G are
algebraically compact, then G is cotorsion. (This is a consequence of Sect. 6(B).)
Corollary 8.4 (Harrison [1]). A torsion group A is cotorsion if and only if it is of
the form A D B ˚ D, where B is bounded and D is divisible.
292 9 Groups of Extensions and Cotorsion Groups

Proof. Sufficiency being obvious, assume A is torsion and cotorsion. By Proposition

8.3 its initial Ulm factor is algebraically compact. By Corollary 3.5 in Chapter 6,
a torsion group that is reduced and algebraically compact is bounded. Hence the
reduced part of A is bounded. t
u
Our next result relates the Ulm length of the torsion group T to the corresponding
adjusted cotorsion group T  D G. Specifically, we prove:
Proposition 8.5. Let T be a reduced torsion group of Ulm length . Then the
cotorsion group G D Ext.Q=Z; T/ has Ulm length  or  C 1. The subgroup G is
torsion-free, and there is a natural isomorphism

G Š Hom.Q=Z; G=.G C T//:

Proof. Since T < G, it is clear that the Ulm length of G cannot be less than . We
also have G \ T D T  D 0, so G is torsion-free cotorsion. By Lemma 8.1(ii), G
is algebraically compact, so its first Ulm subgroup vanishes, G C1 D 0.
The sequence 0 ! T ! G=G ! G=.G C T/ ! 0 is exact, and the last factor
group is divisible as an epic image of G=T. Hence Theorem 2.3 leads us to the exact
sequence

0 ! Hom.Q=Z; G=.G C T// ! Ext.Q=Z; T/ Š G

! Ext.Q=Z; G=G / Š G=G ! 0;

where the last isomorphism holds because G=G is reduced cotorsion. The map
between the two Exts is natural, so its kernel must be G . t
u
Groups That Can be Ext The groups that appear in the form Ext.C; A/ when
the arguments belong to certain subclasses have been under close scrutiny by
abelian group theorists. There are numerous publications about this subject, not only
structural results in ZFC, but also under additional set-theoretical hypotheses, not
to mention several consistency theorems. These are deeper results whose proofs
demand more substantial machinery, so we cannot discuss them here. We only
include a result with a relatively easy proof.
Theorem 8.6 (Jensen [Je]). If A is torsion-free of countable rank, then Ext.A; Z/
is either 0 or is a direct sum of the following summands:
(1) a direct sum of continuously many copies of Q, and,
(2) for each prime p, a direct sum of finitely or continuously many copies of Z.p1 /.
Proof. By Corollary 8.3 in Chapter 3, A is a direct sum of a free group and a torsion-
free group which has no non-trivial homomorphism into Z. Evidently, it suffices to
consider the second summand only, i.e. we may assume Hom.A; Z/ D 0: In this
pP
case, for a prime p, the exact sequence 0 ! A!A ! A=pA ! 0 induces the
pP
exact sequence 0 ! Ext.A=pA; Z/ ! Ext.A; Z/! Ext.A; Z/ ! 0; where in view
of Sect. 3(F), Ext.A; Z/ is a divisible group. The image of the first Ext in the second
8 More on Ext 293

Ext is exactly the p-socle of Ext.A; Z/, thus the number of summands Z.p1 / in
Ext.A; Z/ is given by the dimension of the elementary p-group Ext.A=pA; Z/ as a
Z=pZ-vector space. This is equal to the dimension of A=pA if this is finite, otherwise
it is of the power of the continuum if A=pA is countably infinite.
If A ¤ 0 and Hom.A; Z/ D 0, then A contains a finite rank pure subgroup B
which is not free (Theorem 7.1 in Chapter 3), i.e. B contains a finitely generated
free subgroup F such that B=F is infinite. From the exact sequence 0 ! F !
B ! B=F ! 0 we derive the exact sequence Hom.F; Z/ ! Ext.B=F; Z/ !
Ext.B; Z/ ! 0: Here Hom is finitely generated free and the torsion-free rank of the
first Ext is of the power of the continuum (see, e.g., Sect. 7(e) in Chapter 13), so
the last Ext must have the same torsion-free rank. As Ext.B; Z/ is an epic image of
Ext.A; Z/, the proof is complete. t
u
F Notes. For more detailed information about the Ulm subgroups and Ulm factors of the
groups Ext.Q=Z; T/, we refer to Harrison [3].
The groups Ext.G; Z/ have been studied by several authors. The study splits into the cases
according as G is torsion or torsion-free. Ext.G; Z/ is isomorphic to the character group of G if
G is torsion (so it is a compact group). For countable torsion-free G, see Theorem 8.6. For larger
torsion-free G, see the Notes to Sect. 3. Additional information: Hiller–Huber–Shelah [1] show in
L that Hom.G; Z/ D 0 implies that Ext.G; Z/ admits a compact topology.
Schultz [5] calls G a splitter if Ext.G; G/ D 0 (e.g., torsion-free algebraically compact groups).
His study includes groups whose infinite direct sums or products are also splitters. See Göbel–
Shelah [3] for more on splitters, using new ideas and methods.

Exercises

(1) For a reduced p-group T, these conditions are equivalent:

(a) Ext.Z.p1 /; T/ is algebraically compact;
(b) Ext.Z.p1 /; T/ Š T;Q
(c) T is torsion-complete.
(2) Find non-isomorphic reduced cotorsion groups with 2 Ulm factors such that the
corresponding Ulm factors are isomorphic. [Hint: B the standard basic, T pure
in torsion-completion B such that jB W Tj D jT W Bj D 2@0 ; compare the groups
Ext.Z.p1 /; B/ and Ext.Z.p1 /; T/.]
(3) (a) Let G denote a reduced cotorsion group. There is a cotorsion group A such
that A1 Š G.
(b) (Kulikov) Every cotorsion group G can be realized as G Š Pext.C; A/ for
suitable A; C.
(4) Let A be a reduced torsion-free algebraically compact group, and C an
algebraically compact subgroup. Show that there is a transfinite well-ordered
descending chain of algebraically compact subgroups A from A down to B
such that all the factors A =A C1 are algebraically compact. (At limit ordinals,
intersections are taken.) [Hint: A=C is cotorsion.]
(5) (Hiller–Huber–Shelah) Let A be torsion-free such that Hom.A; Z/ D 0. Then
the rank of the p-component of Ext.A; Z/ is either finite or of the form 2 for an
infinite cardinal .
294 9 Groups of Extensions and Cotorsion Groups

(6) For every totally disconnected compact abelian group A, there is a torsion group
T such that Ext.T; Z/ Š A. [Hint: Corollary 3.6.]

9 Cotorsion Hull and Torsion-Free Cover

Algebraically compact groups can be characterized as pure-injective groups. This

brings up the obvious question: Do cotorsion groups display any injective behavior?
This is the first question which we wish to address in this section.
The Relative Injectivity To begin with, we introduce a new kind of exact
sequence. Call an exact sequence e W 0 ! A ! B ! C ! 0 torsion-splitting
if the sequence e W 0 ! A ! B0 ! tC ! 0 (with B0
B) splits for the injection
map  W tC ! C.
(A) If C is torsion-free, or if the sequence is already splitting, then it is trivially
torsion-splitting.
(B) e is torsion-splitting if and only if torsion groups have the projective property
with respect to e.
(C) e is torsion-splitting exactly if the induced sequence 0 ! tA ! tB ! tC ! 0
of torsion subgroups is splitting exact.
The following proposition gives only a vague idea what torsion-splitting exact
sequences are like, but it helps identifying them within Ext.
Proposition 9.1. An exact sequence e W 0 ! A ! B ! C ! 0 is torsion-splitting
if and only if it represents an element in the divisible subgroup of Ext.C; A/.
Proof. We use the exact sequence 0 ! T ! C ! C=T ! 0 (where T D tC)
to derive the exact sequence Ext.C=T; A/ ! Ext.C; A/ ! Ext.T; A/ ! 0.
The extension e 2 Ext.C; A/ is torsion-splitting if and only if it maps upon
0 2 Ext.T; A/, i.e. it comes from the first Ext which is divisible (C=T is torsion-free).
As Ext.T; A/ is reduced, e is torsion-splitting exactly if it belongs to the divisible
subgroup of Ext.C; A/. t
u
It follows, in particular, that the torsion-splitting extensions of A by C form a
subgroup in Ext.C; A/.
We can now answer in the affirmative the question posed above:
Theorem 9.2. A group is cotorsion if and only if it has the injective property rel-
ative to torsion-splitting exact sequences; in particular, relative to exact sequences
0 ! A ! B ! C ! 0 with torsion-free C.
Proof. The most relevant part of this claim is a weaker form of the “only if” part: if
 W A ! G is a homomorphism of a group A into a cotorsion group G, and if A is a
subgroup of H with torsion-free quotient H=A, then  extends to a map W H ! G.
This is immediately seen from the exact sequence Hom.H; G/ ! Hom.A; G/ !
Ext.H=A; G/ D 0.
9 Cotorsion Hull and Torsion-Free Cover 295

We continue assuming G cotorsion. If 0 ! A ! B ! C ! 0 is a torsion-

splitting exact sequence and T D tC, then evidently, every map  W A ! G extends
to a map  0 W B0 ! G where B0
B; B0 =A Š T. As B=B0 is torsion-free, by what
has been proved in the preceding paragraph,  0 extends to some W B ! G.
Conversely, suppose G has the injective property relative to all torsion-splitting
exact sequences. Without loss of generality, G may be assumed reduced. By Corol-
lary 6.7, there is an exact sequence 0 ! G ! G ! D ! 0 with G cotorsion and
D torsion-free divisible. Since G has the injective property with respect to this exact
sequence, the sequence splits and G is a summand of the cotorsion group G (hence
G D G ). t
u
Thus torsion groups are projective, and cotorsion groups are injective objects for
the torsion-splitting exact sequences.
Cotorsion Hull The embedding A ! A of a reduced group A in a reduced
cotorsion group deserves special attention. A is actually the cotorsion hull of A in
the following sense: A is the minimal cotorsion group containing A with torsion-
free quotient. In fact, we have
Proposition 9.3.
(i) Let A be a reduced group and A D Ext.Q=Z; A/: Any homomorphism  W A!G
into a reduced cotorsion group G extends uniquely to   W A ! G.
(ii) The correspondence A 7! A is functorial: every homomorphism ˛ W A ! B has
a unique extension ˛  W A ! B making the following square commutative:

α
A −−−−→ B
⏐ ⏐
⏐ ⏐
 
α•
A• −−−−→ B •

Proof.

(i) The exact sequence 0 ! A!A ! A =A ! 0 leads to the exact sequence


0 D Hom.A =A; G/ ! Hom.A ; G/!Hom.A; G/ ! Ext.A =A; G/ D 0:

We infer that  is an isomorphism, proving (i).

(ii) follows in the same way, using B in place of G. t
u
Proposition 9.4. A torsion-splitting exact sequence e W 0 ! A ! B ! C ! 0 of
reduced groups induces a splitting exact sequence

e W 0 ! A ! B ! C ! 0:
296 9 Groups of Extensions and Cotorsion Groups

Proof. Applying the Hom-Ext exact sequence with Q=Z to e, the exactness of e is
immediate. Its splitting will follow once we can establish that it is torsion-splitting.
Since tC D tC, from the commutative diagram

: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
•
: 0 −−−−→ A• −−−−→ B • −−−−→ C • −−−−→ 0

it is routine to derive the projective property of torsion groups relative to the exact
sequence e . t
u
The cotorsion hull of a mixed group can be computed easily from those of
the torsion and torsion-free parts. In fact, from Lemma 7.2 it follows that for any
reduced group A, we have
A Š .tA/ ˚ .A=tA/ :

Q B D ˚n Bn with Bn D ˚ Z.p / for each n 2 N. Then B is Q
n
Example 9.5. Let equal to the
subgroup G  n Bn such that G=B is the largest torsion-free divisible subgroup in . n Bn /=B.
Q.
Example 9.6. For a reduced torsion-free group A, A Š A

Torsion-Free Cover The class of cotorsion groups (and especially their general-
izations) has attracted much attention in recent years, because they form a so-called
cotorsion pair along with the class of torsion-free groups. This means that G is
cotorsion if and only if Ext.A; G/ D 0 for every torsion-free A, and vice versa, A
is torsion-free if and only if Ext.A; G/ D 0 holds for every cotorsion G. There is
a very rich, fast growing literature available on this subject for modules. One of
the objects of research is the existence of covers and envelopes (or hulls) in the
respective classes; in our case: torsion-free covers and cotorsion hulls. The latter
has been settled above, so let us move to the question of torsion-free covers.
Needless to say, every group is an epic image of a torsion-free group, for instance,
of a free group. We wonder if there is a minimal one among such torsion-free groups.
Better yet, if there is a unique minimal one. Enochs [2] succeeded in showing that
there is always a minimal one that is even unique up to isomorphism. To make this
fact precise, we have to clarify what “minimality” should mean.
Let  W F ! A be a homomorphism of the torsion-free group F into A.  is called
a torsion-free cover of A if
(i) for every torsion-free group G and homomorphism W G ! A there is a map
 W G ! F such that  D , and
(ii) Ker  contains no non-zero pure subgroup of F.
Condition (i) says, in more prosaic terms, that maps from torsion-free groups into
A must factor through . Hence it follows that the map  has to be surjective (since
9 Cotorsion Hull and Torsion-Free Cover 297

there is a free group mapping onto A). Trivially, a torsion-free group is a torsion-free
cover of itself.
We do not wish to enter into a full-fledged study of torsion-free covers, since this
would take us too far afield. But not too much effort is needed in establishing their
existence.
Lemma 9.7 (Enochs [2]). Every group A admits a torsion-free cover  W F ! A,
and Ker  is a reduced algebraically compact group.
Proof. Let D denote the injective hull of the torsion subgroup T D tA. By the
Harrison category equivalence, there is an exact sequence 0 ! C ! E ! D ! 0
where C is a complete torsion-free group, and E is an injective torsion-free group
(a direct sum of copies of Q). Cutting down from D to T, we get an exact sequence
0 ! C ! B ! T ! 0. Here C cannot contain any pure subgroup of B, because
every rank 1 pure subgroup of B maps upon a non-zero subgroup of T (T is essential
in D).
We claim that there exists an exact sequence in the middle row, unique up to
equivalence, making the diagram

C C = Ker φ
⏐ ⏐
⏐ ⏐
 
0 −−−−→ B −−−−→ F −−−−→ H −−−−→ 0
⏐ ⏐ 
⏐ ⏐ 
 φ 
0 −−−−→ T −−−−→ A −−−−→ H −−−−→ 0

with exact rows commute where H D A=T. Indeed, there is a natural bijection
between the extensions of T by H and those of B by H, as it is clear from the exact
sequence Ext.H; C/ ! Ext.H; B/ ! Ext.H; T/ ! 0, where the first Ext vanishes
(H is torsion-free and C is algebraically compact). We claim that  W F ! A is
a torsion-free cover of A. If W G ! A with a torsion-free group G, then the
exactness of Hom.G; F/ ! Hom.G; A/ ! Ext.G; C/ D 0 implies the existence of
an  W G ! F as required by (i). Furthermore, (ii) is also satisfied, since C does not
contain any pure subgroup of B, and B is pure in F. t
u
Example 9.8. The group Jp is the torsion-free cover of Z.pk / for every integer k > 0, where
 W Jp ! Z.pk / with Ker  D pk Jp . Condition (ii) in the definition of torsion-free cover is
obviously satisfied. To check (i), let W G ! Z.pk / with torsion-free G, and pick g 2 G such that
h gi D Im . Then there is u 2 Jp such that u D g. Since hui is p-pure in Jp , the correspondence
g 7! u extends to a desired map  W G ! Jp in view of the algebraic compactness of Jp .
Example 9.9. The torsion-free cover of Z.p1 / is the additive group Q p of the field of p-adic
numbers (i.e., the quotient field of Jp ). This follows from the isomorphism Hom.Q; Z.p1 // Š Q p.
298 9 Groups of Extensions and Cotorsion Groups

F Notes. Cotorsion hulls of groups have been investigated by several authors. For the
cotorsion hulls of separable p-groups, see Moskalenko [1], and Kemoklidze [1], where also their
full transitivity has been studied.
Torsion-free covers over arbitrary integral domains are unique up to isomorphism. For a proof,
we refer to Enochs–Jenda, Relative Homological Algebra (2000), Theorem 4.2.1. A torsion-free
F is called a torsion-free precover of A if it satisfies condition (i) in the definition of torsion-free
cover. It is an important fact that every precover has a summand isomorphic to the cover.

Exercises

(1) If C is a cotorsion group, then Hom.A; C/ is cotorsion for every A. [Hint: start
with tA; A=tA.]
(2) Verify the inequality jA j
jAj@0 .
(3) A group has the projective property relative to all torsion-splitting-exact
sequences if and only if it is the direct sum of a free group and a torsion group.
(4) If A
G and G is a reduced cotorsion group with G=A torsion-free divisible,
then G Š A .
(5) (Rangaswamy) (a) For a group G, Ext.C; G/ is reduced for all groups C if and
only if G is cotorsion.
(b) For a fixed C, Ext.C; G/ is reduced for all groups G if and only if C is a
direct sum of a torsion and a free group.
(6) Show that Z ! Z.p/ cannot be a torsion-free cover. [Hint: rational groups
! Z.p/.]
(7) The torsion-free cover of a divisible group D is given by Hom.Q; D/ !
Hom.Z; D/ D D:

Problems to Chapter 9

PROBLEM 9.1. Which cotorsion groups cannot be represented as Ext.C; A/ for

any A; C?
PROBLEM 9.2. Is it possible to give an upper bound for the Ulm length of
Ext.C; A/ in terms of the Ulm lengths of A and C?
PROBLEM 9.3. Characterize p-groups A such that p Ext.A; Z/ D 0.
PROBLEM 9.4. Study the cotorsion hulls of groups over their endomorphism
rings.
PROBLEM 9.5. Relate a torsion group to its torsion-free cover.
Chapter 10
Torsion Groups

Abstract We are now prepared to plunge into an in-depth study of the major classes of abelian
groups. Divisible groups have been fully characterized, so we can concentrate on reduced groups.
Our discussion begins with the theory of torsion groups. Since a torsion group is a direct sum of
uniquely determined p-groups, it is clear that the study of torsion groups reduces immediately
to p-groups. This chapter is primarily concerned with p-groups without elements of infinite
heights (called separable p-groups), while the next chapter will concentrate on p-groups containing
elements of infinite heights.
Separable p-groups are distinguished by the property that every finite set of elements is
contained in a finite summand. This proves to be a very powerful property. However, as it turns out,
it does not simplify the group structure to the extent one hopes for: though the full potential of this
condition has not been realized, it looks probable (if not certain) that a reasonable classification in
terms of the available invariants is impossible. Every separable p-group is a pure subgroup between
its basic subgroup B and the largest separable p-group B with the same basic subgroup. Much of
the interest in these torsion-complete p-groups B comes from their numerous remarkable algebraic
and topological features, one of which is that they admit complete systems of invariants.
There is a large body of work available on separable p-groups, and a fair amount of material
will be covered on such groups that are distinguished by interesting properties, mostly those shared
by torsion-complete groups.

1 Preliminaries on p-Groups

Before embarking on an in-depth study of p-groups, we have to get familiar with

the basic tools.
Transfinite Heights and Indicators We start with transfinite heights that
provide a lot of information about how an element sits inside the group. Let A be a
p-group. Recall that for an ordinal , the subgroup p A was defined recursively:
pA D fpa j a 2 Ag, p C1 A D p.p A/, and p A D \< p A if  is a limit
ordinal. For cardinality reason, there is a smallest ordinal  with p C1 A D p A;
this subgroup must be the divisible part of A which we may denote as p1 A. We
form the descending chain

A > pA >    > pn A >    > p A >    > p A D p1 A:

The ordinal  is called the length of A.

© Springer International Publishing Switzerland 2015 299

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_10
300 10 Torsion Groups

(A) If C is a subgroup of the p-group A such that pn A

C
A for some n 2 N,
and if A has infinite length, then C has the same length.
(B) If A has length  and S
AŒpn
, then A=S has length
 C n. In fact, this is
clear from p .A=S/
AŒpn
=S.
The transfinite height h.a/ (henceforth called briefly the height) of an element
0 ¤ a 2 A is defined to be the ordinal  if a 2 p A n p C1 A, and we set h.a/ D 1
for elements a 2 p1 A. We write hp .a/ if the prime p is not obvious from the context.
More information is contained in the indicator. Let a 2 A be an element of order
pn . With a we associate the strictly increasing sequence

u.a/ D .h.a/; h.pa/; : : : ; h.pn1 a/; h.pn a/ D 1/ (10.1)

of ordinals with the symbol 1 at the end. u.a/ is called the indicator of a in A. It is
sometimes reasonable to continue the indicator with symbols 1; if we do so, then it
is easier to define the partial order for the characteristics: u.a/
u.b/ to mean that
h.pn a/
h.pn b/ for all n 2 N.
If h.pi a/C1 < h.piC1 a/, then we say there is a gap between h.pi a/ and h.piC1 a/,
or simply a gap at h.pi a/. If o.a/ D pn in a reduced p-group, there is certainly a gap
between h.pn1 a/ and h.pn a/ D 1.
Lemma 1.1 (Kaplansky [K]). Let A be a reduced p-group of length , and 0 <
1 <    < n1 <  a strictly increasing sequence of ordinals. Then u D
.0 ; 1 ; : : : ; n1 ; n D 1/ is the indicator of some a 2 A if and only if it satisfies
the gap condition:
. / If there is gap between i and iC1 , then the i th UK-invariant fi .A/ of A is
¤ 0.
Proof. Assume that in u a gap occurs at h.pi a/ for some a 2 A. This means that there
is a b 2 A such that piC1 a D pb with h.b/ > h.pi a/ D i . Now c D pi ab is of order
p and of height minfh.pi a/; h.b/g D i . Hence the group pi AŒp
=pi C1 AŒp
¤ 0, so
its dimension fi .A/ ¤ 0.
Conversely, let u D .0 ; 1 ; : : : ; n1 ; n D 1/ .n1 ¤ 1/ satisfy . /. Then
there is an an1 2 AŒp
of height n1 . We distinguish two cases. If there is no gap
between n2 and n1 , then pick an an2 of height n2 such that pan2 D an1 .
If there is a gap between them, then . / ensures the existence of a b 2 AŒp
of height
n2 , and there is a c 2 A of height > n2 with pc D an1 . Now an2 D b C c
is of height n2 and satisfies pan2 D an1 . Proceeding in the same way, we get
successively elements an1 ; an2 ; : : : ; a0 such that pai D aiC1 and h.ai / D i .i D
0; : : : ; n  1/. By construction, u.a0 / D .0 ; 1 ; : : : ; n1 ; n D 1/. t
u
Subsocles A subgroup S of the socle AŒp
of A is called a subsocle of A. A
subsocle is said to support the subgroup C of A if CŒp
D S. A subsocle S carries
the topology inherited from the p-adic topology of A, so it will make sense to talk of
a closed and of a dense subsocle, meaning closed resp. dense in AŒp
. S is discrete
in case S \ pm A D 0 for some m 2 N, i.e. the heights of the non-zero elements of S
are bounded by m  1.
1 Preliminaries on p-Groups 301

We make an important observation about pure subgroups whose socles support

summands.
Lemma 1.2 (Irwin–Walker [1], Enochs [1]). A pure subgroup supported by the
socle of a summand is itself a summand.
Proof. Suppose A D B ˚ C is a p-group, and G is a pure subgroup of A such that
GŒp
D BŒp
. Clearly, G \ C D 0 and G C C contains AŒp
. Given a 2 AŒp
,
write a D g C c with g 2 GŒp
; c 2 CŒp
: Then the height of a in G C C is
 minfh.g/; h.c/g D hA .a/. Therefore, G C C is pure in A. For a pure, essential
subgroup, we have G ˚ C D A (see Sect. 1(F) in Chapter 5). t
u
Pure subgroups supported by dense subsocles are of special interest.
Theorem 1.3 (Hill–Megibben [2]). Let S be a dense subsocle of the p-group A. If
C is a subgroup of A maximal with respect to the property CŒp
D S, then C is pure
and dense in A.
Proof. The existence of subgroups C of the stated kind follows at once from Zorn’s
lemma. We use induction on n to prove C \ pn A
pn C. For n D 1, let pa D c 2 C
with a 2 A. If a … C, then by the maximal choice of C, there is b 2 hC; ai \ AŒp

such that b … S. We can write b D c0 C ka for some c0 2 C and integer

k .1
k < p/ which may be assumed to be 1 without loss of generality. Then
pc0 D p.a  b/ D pa D c, and the case n D 1 is done.
Now assume that C \ pn A
pn C holds for some n  1, and let a 2 A
satisfy pnC1 a 2 C. By what has been shown, pnC1 a D pc for some c 2 C. Since
pn a  c 2 AŒp
, the density of S in AŒp
guarantees the existence of a d 2 S such that
pn a  c  d 2 pn A. Then c C d 2 C \ pn A, so by induction hypothesis, some c1 2 C
satisfies pn c1 D c C d. Hence pnC1 c1 D pc D pnC1 a, and the purity of C follows.
We refer to Lemma 2.8 in Chapter 5 to argue that the cosets mod C that are
of order p can be represented by elements of AŒp
. Owing to the density of S, the
elements of order p in A=C are of infinite height in A=C. Hence A=C is divisible (cf.
Sect. 1(C) in Chapter 4), and C is dense in A. t
u
Indicators with Integers In case the indicator contains only finite ordinals, an
important conclusion may be drawn. Of course, no infinite ordinal may occur if the
p-group is separable, i.e. every non-zero element is of finite height. An important
technical lemma:
Lemma 1.4 (Baer [5]). Let A be a p-group, and

u.a/ D .r0 ; r1 ; : : : ; rn1 ; rn D 1/

302 10 Torsion Groups

the indicator of a 2 A, where rj .j < n/ are integers. Let rn1 ; : : : ; rns D rn denote the
places before which gaps occur. Define the integers k1 ; : : : ; ks by the rules:

k1 D r0 ; k2 D rn1  n1 ; : : : ; ks D rns1  ns1 :

Then 0 < n1 <    < ns and 0

k1 <    < ks , and there exist elements c1 ; : : : ; cs 2
A such that
(a) o.ci / D pni Cki for i D 1; : : : ; s;
(b) C D hc1 i ˚    ˚ hcs i is a summand of A;
(c) a D pk1 c1 C    C pks cs .
Proof. We induct on o.a/. If this is p, then u.a/ D .r0 ; 1/ and everything follows
from Corollary 2.2 in Chapter 5. Thus let o.a/ D pn .n > 1/, and assume the
claim true for elements of orders
pn1 . The inequalities for the ni and ki are
obvious. Consider the element pns1 a which is of height rn1 , and choose a cs 2 A
with prn1 cs D pns1 a. Then hcs i is pure in A and is of order pns Cks . The element
a0 D a  pks cs is of order pn1
pn1 and—as is readily verified—its indicator is
u.a0 / D .r0 ; r1 ; : : : ; rns1 1 ; 1/. Invoking induction hypothesis, we argue that for
a0 there are elements c1 ; : : : ; cs1 2 A of the desired kind. Note that hcs i intersects
C0 D hc1 i ˚    ˚ hcs1 i trivially, since pns Cks 1 cs is of order p and of height
ns C ks  1, while C0 does not contain any such element. We now set C D C0 ˚ hcs i.
C must be pure in A, since elements in its socle have the same height in C as they
have in A (Sect. 1(E) in Chapter 5). A finite pure subgroup is a summand. t
u
We can now derive an important corollary to the preceding lemma.
Corollary 1.5. Let A be a separable p-group. There is an endomorphism of A
mapping a 2 A to b 2 A if and only if u.a/
u.b/.
Proof. Since endomorphisms never decrease heights, the necessity is obvious. To
prove sufficiency, assume u.a/
u.b/ for a; b 2 A. By the preceding lemma, we can
embed a and b in direct summands C D hc1 i˚  ˚hcs i and D D hd1 i˚  ˚hdt i of
A, respectively, subject to the stated conditions. It suffices to exhibit a map  W C !
D with a D b to complete the proof. Write a D pk1 c1 C    C pks cs as in condition
(c) and similarly, b D p`1 d1 C    C p`t dt where o.dj / D pmj C`j ; 0 < m1 <    <
mt ; 0
`1 <    < `t . Now 1 D h.pns a/
h.pns b/ shows that pns b D 0, thus
ns  mt , and therefore, we may map cs upon p`j ks dj C    C p`t ks dt where j is the
smallest index with ns  mj (so that the image will be of order
o.cs / D pns Cks ).
The elements a0 D a  pks cs and b0 D b  p`j dj      p`t dt satisfy u.a0 /
u.b0 /,
and by induction hypothesis C D hc1 i ˚    ˚ hcs1 i has a homomorphism into
D D hd1 i ˚    ˚ hdj1 i mapping a0 upon b0 . This extends to a homomorphism
 W C ! D with  acting on cs as indicated. t
u
Kaplansky [K] calls a reduced p-group A fully transitive if it has the property
stated in Corollary 1.5. A is transitive if the same holds for automorphisms with
inequality replaced by equality. Corollary 1.5 asserts the full transitivity of p-groups
with no non-zero elements of infinite height.
1 Preliminaries on p-Groups 303

Arbitrarily Large Lengths An obvious question concerning p-groups is

whether or not there exist p-groups of arbitrarily large length. The answer is
yes, but it is not obviously so. The first existence proof was given independently
and almost simultaneously by Kulikov [3] and the author [2]; the result follows
immediately from Theorem 1.9 below. With the advent of the theory of totally
projective p-groups, it is easy to construct examples of p-groups of arbitrarily large
length.
Theorem 1.6. There exist reduced p-groups of arbitrarily large length.
Proof. Starting with the group H0 D 0, we shall construct a p-group H , for every
ordinal , such that
(a) H is of length ;
(b) p H C1 is cyclic of order p and H C1 =p H C1 Š H ;
(c) H Š ˚< H if  is a limit ordinal;
(d) all the UK-invariants of H are
jj.
The construction is straightforward, the stated conditions indicate how to pro-
ceed. Since H0 D 0, from (b) we conclude by induction that, for an integer n > 0,
Hn is cyclic of order pn . Then (c) implies that H! D ˚n<! Z.pn /.
If  is a limit ordinal, (c) tells us how to form H . For a non-limit ordinal, write
 D  C n with limit ordinal  and integer n > 0. Two cases will be distinguished
according as n D 1 or n  2.
In case n  2, we may assume that HCn1 =p HCn1 is a cyclic group C of
order pn1 . Let B be a cyclic group of order pn and W B ! C an epimorphism. As
the induced map  W Ext.H ; B/ ! Ext.H ; C/ is surjective, there exists a group
HCn making the diagram

0 −−−−→ B −−−−→ Hμ+n −−−−→ Hμ −−−−→ 0

⏐ ⏐ 
⏐
γ
⏐ 
δ 
0 −−−−→ C −−−−→ Hμ+n−1 −−−−→ Hμ −−−−→ 0

commute and its first row exact. It is readily seen that HCn D H satisfies
conditions (a), (b), and (d).
If n D 1, then a different kind of construction is needed. The trick is to view H
as the direct sum of HC1 =p HC1 for  <  where all the subgroups p HC1 are
cyclic of order p. Using the codiagonal map r W ˚< p HC1 ! C Š Z.p/, we
define H via the push-out diagram
304 10 Torsion Groups

0 −−−−→ ⊕ρ<μ pρ Hρ+1 −−−−→ ⊕ρ<μ Hρ+1 −−−−→ Hμ −−−−→ 0

⏐ ⏐ 
⏐ ⏐ 
∇  
0 −−−−→ C −−−−→ Hμ+1 −−−−→ Hμ −−−−→ 0

where C
p HC1 , since r is epic on each summand. The reverse inclusion is a
consequence of p H D 0, so HC1 D H satisfies (a)–(b). That (d) also holds in
all cases follows rather easily from the definition by transfinite induction. t
u
We recognize H!C1 as an old acquaintance of ours. In fact, it is identical to the
Prüfer group constructed in Example 5.7 in Chapter 5. This group is a simplest
example of a reduced p-group with non-zero elements of infinite height. Because
of this resemblance, the group H defined above was named by Nunke [5] a
generalized Prüfer group of length .
The groups H not only provide us with p-groups of arbitrary length, but, more
importantly, they will serve as building blocks of the significant class of totally
projective p-groups, to be discussed in Sects. 3–6 in Chapter 11.
As an additional information on these H , let us point out:
Corollary 1.7. The Ulm factors of H are †-cyclic groups.
Proof. We argue via transfinite induction. For 
!, H is just a direct sum of
cyclic p-groups. If  D  C n with limit ordinal  and n 2 N, then H =p H has
the stated property by induction hypothesis, while the last Ulm subgroup of H is
cyclic of order pn . On the other hand, if  is a limit ordinal, then the construction
above shows that the Ulm factors of H will be the direct sums of the Ulm factors
of the H . < /, whence the claim is evident. t
u
Every Group Can Be p A Another obvious question related to the existence of
p-groups is answered by the following result that is proved in a more general setting
(for arbitrary G).
Proposition 1.8. Given a prime p, an ordinal , and a group G, there exists a group
A such that p A Š G.
Proof. First we construct a mixed group M to fit into the exact sequence 0 !
hx i Š Z ! M ! H ! 0 such that p M D hx i (thus hp .x / D ). This can
be done in the same way as in the proof of Theorem 1.6, using the existing H ’s,
with an infinite cyclic group at the very end (in place of a cyclic group of order p).
(Cf. Nunke groups in Sect. 1 in Chapter 15.)
Let B D ˚i2I hbi i be a p-basic subgroup of G. For each i 2 I, select a copy Ci
of M , with ci a generator of the cyclic group p Ci . Define A as the push-out in the
diagram
1 Preliminaries on p-Groups 305

0 −−−−→ ⊕i∈I ci  −−−−→ ⊕i∈I Ci −−−−→ ⊕i∈I Hσ −−−−→ 0

⏐ ⏐ 
⏐
γ
⏐α 
 
0 −−−−→ G −−−−→ A −−−−→ ⊕i∈I Hσ −−−−→ 0

where is defined by the correspondence ci 7! bi . It remains to prove that p A D G.

The inclusion p A
G is a consequence of p H D 0. Clearly, the generators bi of
the basic subgroup B of G will have height . The rest follows from Lemma 6.7 in
Chapter 5. t
u
Conditions on the Ulm Sequence For completeness’ sake, we state a general
necessary and sufficient condition on the Ulm sequence of p-groups. The theorem
answers a question posed by A.G. Kurosh. The proofs are too long for inclusion,
and since we are not going to need the result, we just show the necessity part, and
for the rest we refer to the original articles.
We recall that, for a p-group A, the ˛th Ulm subgroup A˛ is defined inductively
as A1 D \n<! pn A; A˛C1 D .A˛ /1 , and A D \˛< A˛ if is a limit ordinal. The
factor group A˛ D A˛ =A˛C1 is the ˛th Ulm factor, and the sequence

A0 ; A1 ; : : : ; A˛ ; : : : .˛ < / (10.2)

is the Ulm sequence of A. B˛ will denote a basic subgroup of A˛ .

Theorem 1.9 (Kulikov [3], Fuchs [2]). Let (10.2) be a well-ordered sequence of
p-groups, and  an infinite ordinal. There exists a reduced p-group A of cardinality
 with Ulm sequence (10.2) if and only if the following conditions are satisfied:
(a) P Q ˛ < ;
A˛ is separable for each
(b) 0˛< jA˛ j

0n<minf!;g jAn j;
rk B˛C1
fin rk A˛ for every ˛ C 1 < ;
(c) P
(d) ˇ<˛< jA˛ j
jAˇ j@0 for 0
ˇ < .
Proof. We establish necessity only. (a) is obvious, and so is (b) for finite ordinals .
For infinite ordinals, the first inequality in (b) follows from A D [˛< .A˛ n A˛C1 /
and jA˛ j
jA˛ nA˛C1 j. If jAm j D minn<! jAn j, then (cf. Corollary 5.15 in Chapter 5)
Y
jAj D jA0 jjA1 j    jAm1 jjAm j
jA0 jjA1 j    jAm1 jjAm j@0
jAn j:
0n<!

(d) follows from (b) applied to Aˇ and from Corollary 5.15 in Chapter 5:
P ˇ @0
ˇ<˛< jA˛ j
jA j
jAˇ j : To verify (c), note that for each n 2 N, there
˛
is an xi 2 A such that p xi D bi for every bi in a basis of B˛C1 . These xi are
n

independent mod A˛C1 , for otherwise we have r1 x1 C  Crk xk D c 2 B˛C1 , whence

r1 b1 C    C rk bk D pn c 2 B˛C1 . By the purity of B˛C1 , pn jri for i D 1; : : : ; k, which
306 10 Torsion Groups

is tantamount to the independence of the xi in A˛ . It follows that A˛ D A˛ =A˛C1

contains an independent set of elements of pn whose cardinality is not less than
rk B˛C1 . t
u
Notation In order to simplify notation, we shall write p AŒp
for the accurate
.p A/Œp
to denote the socle of p A.


F Notes. Transitivity and full transitivity have been investigated by several papers. Corner [7]
gives examples establishing the independence of these notions. Files–Goldsmith [1] prove that A
is fully transitive if and only if A ˚ A is transitive. Separable p-groups are both transitive and fully
transitive, and so are the totally projective p-groups (Sect. 6 in Chapter 11).
It is not an easy task to find a p-group that fails to contain proper subgroups isomorphic to the
entire group. Stringall [1] exhibits such an example with the standard basic subgroup.

Exercises

(1) Prove that p .p A/ D pC A for any ordinals ;  and any p-group A. [Hint:
induction.]
(2) Let A be a finite group of type .pk1 ; : : : ; pkn / with k1 <    < kn . Describe all
possible indicators of elements in A.
(3) The indicators of elements in a separable p-group form a distributive lattice
under the partial order defined above.
(4) A pure subgroup is dense in A exactly if its socle is dense in AŒp
.
(5) (Hill–Megibben) A neat subgroup supported by a dense subsocle is a pure
subgroup.
(6) In a †-cyclic p-group, (a) every subsocle supports a pure subgroup, and (b)
pure subgroups with the same support are isomorphic.
(7) (Hill) Let B0 D ˚n ha2n1 i and B00 D ˚n ha2n i with o.ak / D p2k . Define
0 00
C D ˚n ha2n1 C pa2n i. In the torsion-complete group B ˚ B (Sect. 3), the
00
pure subgroups G D B0 ˚ B and H D C C B00 have the same socle, but
they are not isomorphic. [Hint: under an isomorphism G ! H, G must have
elements carried outside H.]
(8) (Charles) Let A be a p-group, and a 2 A with hai \ A1 D 0. Then a can
be embedded in a minimal pure subgroup of A, and any two minimal pure
subgroups containing a are isomorphic over hai.
(9) Prove Corollary 1.5 for automorphisms, replacing inequality by equality.
(10) (Megibben) Check that the following group is not fully transitive: A D G˚H
where G1 Š H 1 Š Z.p/, G=G1 is torsion-complete and H=H 1 is †-cyclic.
[Hint: G1 is fully invariant in A, because every homomorphism G=G1 !
H=H 1 is small.]
(11) (Kulikov) The set of non-isomorphic p-groups of cardinality
 has
cardinality 2 . [Hint: Theorem 1.9; choose each Ulm factor A such that its
invariants are non-zero only for pk where k is either only odd or only even.]
2 Fully Invariant and Large Subgroups 307

2 Fully Invariant and Large Subgroups

Fully Invariant Subgroups Recall that a subgroup G of a group A is called

fully invariant (resp. characteristic) if every endomorphism (automorphism) of A
carries G into itself. Trivially, fully invariant subgroups are characteristic, but the
converse is not true in general, as is demonstrated by examples of suitable 2-groups
and torsion-free groups.
Example 2.1 (Kaplansky [K]). Let A D ha1 i ˚ ha2 i ˚ ha3 i where o.ai / D 2i . Consider the
subgroup G generated by all g 2 A such that o.g/ D 4; h.g/ D 0; h.2g/ D 2; these elements
are: a1 ˙ 2a3 and a1 C 2a2 ˙ 2a3 . Every automorphism carries generators into generators, so G is
characteristic. However, it is not fully invariant, since a1 … G, but the projection A ! ha1 i maps
G onto ha1 i. (For torsion-free groups, see Sect. 1, Exercise 9 in Chapter 17.)

No complete description of fully invariant subgroups of p-groups is known so

far, a satisfactory characterization is available only in special cases; see Baer [5],
Shiffman [1], Kaplansky [K]. These particular cases include the most important
subclasses of p-groups, like the separable and the totally projective p-groups.
In order to describe the fully invariant subgroups of fully transitive p-groups A,
let

u D .0 ; 1 ; : : : ; n ; : : : /

be a strictly increasing sequence of ordinals, followed possibly by a sequence of

symbols 1. The meaning of u
v should be clear. We associate with u the
subgroup

A.u/ D A.0 ; 1 ; : : : ; n ; : : : / D fa 2 A j hp .pn a/  n .n < !/g: (10.3)

This is evidently a fully invariant subgroup. We note that

(a) A.u/ D \n2N pn .pn A/;
(b) if A D ˚i2I Ai , then A.u/ D ˚i2I Ai .u/;
(c) every homomorphism  W A ! C maps A.u/ into C.u/.

As before, we say that u has a gap at n if n C 1 < nC1 ; and that u satisfies the
gap condition for A if u having a gap at n is equivalent to A having an element of
order p and of height n (in other words, the n th UK-invariant fn .A/ ¤ 0).

(d) If both u and v satisfy the gap condition for A, then does u ^ v too. For, if
u ^ v D .0 ; : : : ; n ; : : : / (point-wise infimum) has a gap at n , then either u or
v has n at the nth place, and the same must have a gap at this place. Thus the
n th UK-invariant fn .A/ ¤ 0.
Theorem 2.2 (Kaplansky [K]). Suppose A is a fully transitive p-group. A sub-
group G of A is fully invariant if and only if it can be written in the form G D A.u/
where u satisfies the gap condition. u is uniquely determined by G.
308 10 Torsion Groups

Proof. We have noticed above that every subgroup of the form (10.3) is fully
invariant. Assume, conversely, that G is a fully invariant subgroup, and define
n as the minimum of the heights h.pn g/ with g running over G. The sequence
0 ; 1 ; : : : ; n ; : : : is obviously strictly increasing (except when it reaches 1). To
verify the gap condition, suppose k C 1 < kC1 for some k. Surely, there exists an
x 2 G with h.pk x/ D k , and by definition h.pkC1 x/  kC1 ; thus, this x has a gap
at k in its indicator. By Lemma 1.1, A must contain an element of order p and of
height k .
The inclusion G
A.0 ; 1 ; : : : ; n ; : : : / is obvious. We now verify the existence
of a g 2 G such that h.pi g/ D i for i D 0; 1; : : : ; n  1: If there is no gap
in the sequence 0 ; 1 ; : : : ; n1 , and if g 2 G satisfies h.pn1 g/ D n1 , then
this g is already as desired. If there is a gap in this sequence, and if the first gap
appears between j1 and j , then there is a gj 2 G such that h.pi gj / D i for
i D 0; 1; : : : ; j  1: If the second gap lies between k1 and k .j < k/, then some
g0 2 G exists with h.pi g0 / D i for i D j; : : : ; k  1: By Lemma 1.1, A contains a
gk such that h.pi gk /  maxfh.pi g0 /; i C 1g for i D 0; 1; : : : ; j  1 and h.pi gk / D
h.pi g0 / for i  j. Because of full invariance, and hence full transitivity, as well as
Corollary 1.5, gk 2 G. Thus proceeding, we construct elements gj ; gk ; : : : ; g` 2 G
for the gaps in 0 ; 1 ; : : : ; n1 , and at the end, g D gj C gk C    C g` will
satisfy h.pi g/ D i for i D 0; 1; : : : ; n  1: Thus h.g/
h.a/ for every a 2
A.0 ; 1 ; : : : ; n ; : : : / of order
o.g/. Full transitivity shows that a 2 G, i.e. G
is of the form (10.3).
If .0 ; 1 ; : : : ; n ; : : : / and .00 ; 10 ; : : : ; n0 ; : : : / are different sequences, both
satisfying the gap condition, then let n be the first index with n ¤ n0 , say, n < n0 .
There exists an a 2 A with h.pi a/ D i for i D 0; 1; : : : ; n: This a belongs to
A.0 ; 1 ; : : : ; n ; : : : /, but not to A.00 ; 10 ; : : : ; n0 ; : : : /. t
u
Large Subgroups There is a kind of fully invariant subgroup in p-groups which
is of particular interest. It was called by Pierce [1] a large subgroup. It is defined as
a fully invariant subgroup G satisfying G C B D A for every basic subgroup B of A.
We list some relevant properties.
(A) 0 is a large subgroup if and only if A is bounded.
(B) In a bounded group, all fully invariant subgroups are large.
(C) If G is a large subgroup of A, then so is pn G for every n. Indeed, pn G is fully
invariant and satisfies pn G C B D pn G C pn B C B D pn A C B D A provided
G C B D A.
(D) If G is a large subgroup of A, then A=G is †-cyclic. In order to prove this,
observe that G C B D A implies A=G Š B=.G \ B/. Write B D ˚i2I hbi i,
and notice that the hbi i are summands of A, so the projection of G on hbi i is
G \ hbi i. Hence G \ B D ˚i2I .G \ hbi i/, and B=.G \ B/ is the direct sum of
the cyclic groups hbi i=.G \ hbi i/.
(E) A1 is contained in every large subgroup of A. This follows from (D), since if
A=G is †-cyclic, then necessarily A1
G.
Our next purpose is to single out the large subgroups from among the fully
invariant subgroups. We will say that a subgroup G of a p-group A satisfies the
2 Fully Invariant and Large Subgroups 309

Pierce condition if for every k 2 N there is an n 2 N such that every a 2 A with

o.a/
pk and h.a/  n is contained in G, i.e.

pn AŒpk

G:

Since (B) completely characterizes large subgroups in bounded groups, our focus
is on unbounded groups.
Theorem 2.3 (Pierce [1]). Let A be an unbounded, reduced p-group. For a fully
invariant subgroup G of A, these are equivalent:
(i) G is a large subgroup of AI
(ii) G D A.r0 ; r1 ; : : : ; rn ; : : : / with a strictly increasing sequence of non-negative
integers rn and symbols 1 .satisfying the gap condition/I
(iii) the Pierce condition holds for G.
Proof.
(i) ) (ii) By Theorem 2.2, G D A.0 ; 1 ; : : : ; n ; : : : / for suitable n , and in view
of (E), we have n < ! for all n 2 N for which n ¤ 1.
(ii) ) (iii) Given k 2 N, let n D rk1  k C 1. Then none of rk2  k C 2,
rk3  k C 3; : : : ; r0 exceeds n. Therefore, if a 2 A is such that o.a/
pk and
h.a/  n, then h.pi a/  n C i  ri for i D 0; 1; : : : ; k  1, and h.pi a/ D 1 for
i  k. Thus u.a/  .r0 ; r1 ; : : : ; rn ; : : : /, and a 2 G.
(iii) ) (i) We have to prove G C B D A for any basic subgroup B of A. Pick a 2 A
with o.a/ D pk , and choose n 2 N according to the Pierce condition. By the
divisibility of A=B, we have a D b C pn c for some b 2 B; c 2 A. We may
assume o.b/
pk , because 0 D pk a D pk b C pnCk c implies pk b D pnCk b0 for
some b0 2 B, so also a D .b  pn b0 / C pn .b0 C c/ where o.b  pn b0 /
pk .
But then o.pn c/
pk as well, and therefore (iii) implies pn c 2 G. This proves
G C B D A. t
u
F Notes. It was Kaplansky [K] who characterized the fully invariant subgroups in fully
transitive p-groups. The theory of large subgroups is due to Pierce [1]. Numerous papers deal
with the relation of the structure of the group to the structure of its large subgroups; e.g. large
subgroups are totally projective if and only if the group itself is totally projective.
Anyone in search for more cardinal invariants of p-groups might try to test the vector spaces
A.u/=A.v/ in case u D .0 ; 1 ; : : : ; n ; : : : / and v D .0 ; 1 ; : : : ; n ; : : : / satisfy n  n < nC1
for every n < !. However, in this way no new invariants are gained, since the dimension of the
vector space in question turns out to be just the sum of dimensions of pn AŒp
=pn AŒp
for n < !.

Exercises

(1) Find the maximal length of chains of fully invariant subgroups in a bounded
p-group.
(2) If Li is a large subgroup of the p-group Ai for i 2 I, then ˚i2I Li is large in
˚i2I Ai .
310 10 Torsion Groups

(3) Let A be a separable p-group, and a 2 A. What is the fully invariant subgroup
of A generated by a?
(4) If G is a large subgroup in the p-group A, then there is an n < ! such that
GŒp
D pn AŒp
.
(5) If A is a reduced p-group and has a large subgroup that is †-cyclic, then A
itself must be †-cyclic. [Hint: preceding exercise.]
(6) For a pure subgroup C and for a large subgroup A.u/ of a p-group A, we
always have C.u/ D C \ A.u/.
(7) (a) Give an example when A.u/ D A.v/, but u ¤ v (no gap condition is
required).
(b) If A.u/ D A.v/ and if u satisfies the gap condition, then u
v.
(8) Let A be a fully transitive p-group.
(a) For a family Gj D A.uj / .j 2 J/ of fully invariant subgroups write their
union and intersection in the form (10.3).
(b) The fully invariant subgroups of A form a complete distributive sublattice
K in L.A/.
(c) The large subgroups form a filter in K.

(9) What is the intersection of all large subgroups of a p-group?

(10) The lattices of large subgroups of the p-groups A and C are isomorphic if the
two groups have isomorphic basic subgroups. [Hint: gaps.]
(11) (Pierce) Any single basic subgroup B ¤ A is a test group for a fully invariant
subgroup G to be large: if A D G C B for the selected B, then A D G C B0
holds for all basic subgroups B0 .
(12) (Pierce) Let G be a large subgroup in a separable p-group A. If B is a basic
subgroup of A, then G \ B is a basic subgroup of G. [Hint: for purity show
B \ pG D p.B \ G/ and use (C).]
(13) (Pierce) A large subgroup of a large subgroup is a large subgroup. [Hint:
preceding exercise.]
(14) For p-groups A; C, a homomorphism  W A ! C is small if and only if Ker 
satisfies the Pierce condition if and only if Ker  contains a large subgroup
of A.
(15) Is it true that in a reduced p-group, for every element a ¤ 0 there exists a
fully invariant subgroup G maximal with respect to the property a … G?

3 Torsion-Complete Groups

Our study of p-groups continues with p-groups without elements ¤ 0 of infinite

height. This is the case when the first Ulm subgroup A1 D 0, i.e. A is Hausdorff in
the p-adic (equivalently, in the Z-adic) topology. The closure X  of a subgroup X
of A in the p-adic topology is obtained by taking the first Ulm subgroup X  =X D
.A=X/1 in A=X.
3 Torsion-Complete Groups 311

Separability We may refer to a separable p-group as a group in which every

finite subset fa1 ; : : : ; an g can be embedded in a finite direct summand. In fact, it is
straightforward to check that this definition is equivalent to the one used in Sect. 1:
the non-trivial part follows from Lemma 1.4 by using induction.
Torsion-Complete Groups Undoubtedly, the most significant class of separa-
ble p-groups, besides the class of direct sums of cyclic p-groups, is the class of
torsion-complete p-groups. Their study began with the work of Kulikov [2] (under
the name of closed p-groups), and was brought to maturity by several authors in
the 1960s. These groups have profound implications in the theory of p-groups.
Fortunately, they admit a satisfactory classification by cardinal invariants.
In this section and in the next one, we assemble various results on torsion-
complete p-groups. We will adhere to the following notation: Bn will denote a direct
sum of cyclic groups of order pn , Bn D ˚n Z.pn / for an unspecified cardinal n ,
and B will denote their direct sum ˚n2N Bn .
By a torsion-complete p-group is meant the torsion subgroup of the p-adic
completion BQ of a direct sum B of cyclic p-groups. tBQ is uniquely determined by
B, so we may denote it by B—this notation will be standard in this volume. We will
see in due course that a torsion-complete p-group can be written in the form B for
any of its basic subgroups B.
We also note that in the category of torsion groups, the categorical product is
the torsion subgroup of the direct product. Therefore, torsion-complete p-groups
are nothing else than categorical products of bounded p-groups. It is no wonder that
they share several propertiesQwith bounded groups.
Since BQ is a subgroup of n Bn , the same holds for B. Hence an element g 2 B
can be written uniquely in the form

g D .b1 ; b2 ; : : : ; bn ; : : : / with bn 2 Bn : (10.4)

A vector g in (10.4) will belong to B if and only if the orders of its coordinates bn are
bounded. The elements of B are represented by vectors with finite supports. In view
of the structures of the Bn , for any sequence fbn gn2N with bn 2 Bn of orders bounded
by pm , we have h.bn /  n  m for n  m; thus, h.g/ D minfh.b1 /; : : : ; h.bm1 /g
whenever bi ¤ 0 for at least one of i D 1; : : : ; m  1.
Example 3.1. All bounded p-groups are torsion-complete. The simplest unbounded example for a
torsion-complete group is when B is the standard basic subgroup, i.e. Bn D han i Š Z.pn / for each
n. The elements of this B are the vectors b D .k1 a1 ; : : : ; kn an ; : : : /, where 0  kn < pn , and there
is an integer m such that pm kn an D 0 for all n.

We record a few important facts on torsion-complete p-groups.

(A) B D B if and only if B is bounded. We observe that if B is unbounded, then B
will contain vectors (10.4) with infinite support.
(B) For two †-cyclic p-groups, B and B0 , we have B ˚ B0 D B ˚ B0 .
312 10 Torsion Groups

(C) B is a basic subgroup of B. In fact, as we saw in Theorem 2.4 in Chapter 6,

B is a basic subgroup in B,Q so it is also a basic subgroup in its torsion
subgroup B.
Theorem 3.2. Let B be a †-cyclic p-group of infinite cardinality . Then jBj
 @0 .
Equality holds if also fin rk B D .
Proof. The stated inequality is an immediate consequence
Q ofPthe representation
of elements of B in the form (10.4): jBj D 
n2N n
. n2N n /@0 : If also
fin rk B D , then we distinguish two cases according as the set of cardinalities
n D jBn j .n 2 N/ contains only finitely many or infinitely many n equal to .
In the second alternative the claim is obvious. In the first case, for the proof we
may ignore any Bn of cardinality , and assume that n < P  for all n. Necessarily,
infinitely many of the n are different, andQ we must have n n D  D supn n .
Jech [J] tells us that in these circumstances n n D  @0 . t
u
We are now in a position to derive the following theorem that solves at once the
structure problem for torsion-complete p-groups. It shows that they are completely
characterized by their UK-invariants.
Theorem 3.3 (Kulikov [2]). Two torsion-complete p-groups are isomorphic if and
only if their basic subgroups are isomorphic. Thus the sequence 1 ; : : : ; n ; : : : of
cardinal invariants of B D ˚n<! Bn .where Bn D ˚n Z.pn // is a complete and
independent set of invariants for B.
Proof. The ‘only if’ part is a trivial consequence of (C) and the uniqueness (up to
isomorphism) of basic subgroups. t
u
Torsion-Completion We wish to clarify the relation between arbitrary p-groups
A and torsion-complete p-groups. Let B D ˚n Bn (with the adopted notation) be a
basic subgroup of A. As in Sect. 5 in Chapter 5, we have a sequence A D B1 ˚
   ˚ Bn ˚ An .n 2 N/ of direct decompositions such that the .n C 1/st is obtained
from its predecessor by separating the summand BnC1 from An . In this way, we
can associate with each a 2 A a sequence b1 ; : : : ; bn ; : : : with bn 2 Bn such that
a D b1 C    C bn C an for some an 2 An . This gives rise to a correspondence

 W a 7! .b1 ; : : : ; bn ; : : : / .bn 2 Bn / (10.5)

Q
which is obviously a homomorphism of A into n Bn . As the order of a is a bound
for the orders of the bn ,  maps A into B. It is clear that  acts isomorphically on B,
mapping its elements to vectors with finite supports. As B is a separable group, we
have A1
Ker . On the other hand, if a D 0 for some a 2 A of order pk , then
a D an 2 An for every n, thus h.a/ D h.an /  n C 1  k for every n; this means that
h.a/  ! and a 2 A1 . Consequently, Ker  D A1 .
Theorem 3.4. Let A be a p-group, and B a basic subgroup of A. Then  in (10.5)
is a homomorphism of A onto a pure subgroup of B. The kernel of  is the first Ulm
subgroup A1 , and B=A is divisible.
3 Torsion-Complete Groups 313

In particular, a separable p-group A with basic subgroup B is isomorphic to a

pure .dense/ subgroup of B containing B.
Proof. It remains to check the purity of Im  in B. By Sect. 6(F) in Chapter 5, B is
a basic subgroup in A, so A=B is a divisible group. Manifestly, B is basic in B,
whence the purity of A in B is immediate. t
u
It is appropriate to regard the torsion-complete group B obtained from the p-
group A in the preceding theorem as the torsion-completion of A. It is unique up to
isomorphism.
As a consequence, a separable p-group may be thought of (and treated as) a
pure and dense subgroup of a torsion-complete group. This point of view gives us
considerably more leverage than regarding it as a subgroup of a divisible group. To
illustrate this, we interrupt momentarily the discussion of our main topic, and prove
the following slight generalization of Prüfer’s Theorem 5.3 in Chapter 3.
Theorem 3.5. Let A be a separable p-group, and B a pure †-cyclic subgroup of A.
If A=B is countable, then A is †-cyclic.
Proof. As B can be expanded to be a basic subgroup, we may assume B is a
basic subgroup of A. A is treated as a pure subgroup of B containing B. Let
a1 ; : : : ; am ; : : : be a countable set that, together with B, generates A. We can write
am D .bm1 ; : : : ; bmn ; : : : / 2 B with bmn 2 Bn . Each Bn is a direct sum of cyclic
groups of order pn , so there is a decomposition Bn D B0n ˚ B00n such that bmn 2 B0n
for all m, and B0n is countable. Setting B0 D ˚n B0n and B00 D ˚n B00n , we obtain
B D B0 ˚B00 and A D A0 ˚B00 where A0 D hB0 ; a1 ; : : : ; am ; : : : i. Here A0 is countable
and separable, so by Theorem 5.3 in Chapter 3 it is †-cyclic. t
u
Theorems on Torsion-Complete Groups We return to torsion-complete
groups, and continue with various algebraic characterizations. The main algebraic
features are encapsulated in the following theorem.
Theorem 3.6. For a reduced p-group A, the following conditions are equivalent.
(i) A is torsion-complete;
(ii) A is the torsion subgroup of a reduced algebraically compact group;
(iii) A is pure-injective in the category of p-groups .i.e., it has the injective property
relative to pure-exact sequences of p-groups/;
(iv) A is a direct summand in every p-group in which it is contained as a pure
subgroup.
Proof.
(i) ) (ii) This implication is obvious, since B is by definition the torsion subgroup
in the algebraically compact (= complete) group B. Q
(ii) ) (iii) Next, suppose (ii), i.e. A is the torsion part of an algebraically compact
˛ ˇ
group C. Let 0 ! G!H !K ! 0 be a pure-exact sequence of p-groups,
and  W G ! A a homomorphism. If  is viewed as a map G ! C, then by
314 10 Torsion Groups

pure-injectivity there is a  W H ! C such that  D ˛. As H is torsion,  must

land in A, so (iii) follows.
(iii) ) (iv) If (iii) holds for A, and if A is a pure subgroup in a p-group G, then the
identity map 1A extends to a map G ! A, establishing the summand property
of A. Thus (iv) is immediate.
(iv) ) (i) Finally, assume (iv). If A1 ¤ 0, then A would be pure, but not a
summand, in the torsion part G of the pure-injective hull of A (recall: G=A
is divisible and every divisible subgroup ¤ 0 of G intersects A non-trivially).
Hence A1 D 0, and so by Theorem 3.4 A may be thought of as a pure subgroup
of B where B denotes a basic subgroup of A. By hypothesis (iv), we have
B Š A ˚ B=A, where the second summand must vanish, since it is divisible,
and B is reduced. Hence (i) follows. t
u
We go on to point out that condition (iv) can be rephrased by stating that
Pext.T; A/ D 0 for all torsion groups T. As a matter of fact, this criterion can
considerably be improved: Z.p1 / is a kind of ‘test’ group for torsion-completeness.
Corollary 3.7. A p-group A is torsion-complete if and only if

Pext.Z.p1 /; A/ D 0I

equivalently, Ext.Z.p1 /; A/ is algebraically compact.

Proof. Start with a pure-exact sequence 0 ! C ! T ! ˚ Z.p1 / ! 0 where C is
a basic subgroup of an arbitrarily chosen p-group T. From Theorem 5.6 in Chapter 9
we derive the exact sequence
Y
Pext.˚ Z.p1 /; A/ D Pext.Z.p1 /; A/ ! Pext.T; A/ ! Pext.C; A/ D 0

whence the implication Pext.Z.p1 /; A/ D 0 ) Pext.T; A/ D 0 is evident.

Pext is the first Ulm subgroup of Ext, so the stated condition is equivalent to
saying that the first Ulm subgroup of Ext.Z.p1 /; A/ vanishes which means the
algebraic compactness of this Ext (Corollary 6.3 in Chapter 9). t
u
(D) We note that if A D B for a basic subgroup B of A, then also A D B0 for
any basic subgroup B0 of A. In fact, if B0 .
A/ is a basic subgroup, then A is
pure in B0 , thus the torsion-complete A is a summand of B0 . The complement
is divisible, so it must be 0.
(E) It should be pointed out that, for a †-cyclic p-group B, B is the largest
separable p-group whose basic subgroup is B. In fact, every separable p-group
with B as basic subgroup is embeddable in B.
(F) The natural map Hom.B; B/ ! Hom.B; B/ is an isomorphism. This follows
from the pure-exact sequence 0 ! B ! B ! ˚ Z.p1 / ! 0, applying
Theorem 5.6 in Chapter 9 and taking Corollary 3.7 into account.

Subgroups of Torsion-Complete Groups Next, we want to find out when

subgroups of torsion-complete p-groups are again torsion-complete.
3 Torsion-Complete Groups 315

Corollary 3.8 (Irwin–O’Neill [1]). A subgroup G of a torsion-complete p-group A

is torsion-complete if the factor group A=G is reduced.
Proof. If A=G is reduced, then the exact sequence 0 ! G ! A ! A=G ! 0
induces the exact sequence 0 D Hom.Z.p1 /; A=G/ ! Ext.Z.p1 /; G/ !
Ext.Z.p1 /; A/: Thus Ext.Z.p1 /; G/ can be regarded as a subgroup of
Ext.Z.p1 /; A/: If the first Ulm subgroup (i.e., Pext) of the latter group vanishes,
then the same holds for the former group, i.e. Pext.Z.p1 /; G/ D 0. By
Corollary 3.7, the proof is complete. t
u
More can be said if the subgroup is pure.
Corollary 3.9. For any pure subgroup G of a torsion-complete p-group A, the
factor group A=G is a direct sum of a divisible group and a torsion-complete group.
Thus the closure of a pure subgroup is a summand of A.
Proof. Because of Theorem 5.6 in Chapter 9, the pure-exactness of 0 ! G ! A !
A=G ! 0 implies that the induced map Pext.Z.p1 /; A/ ! Pext.Z.p1 /; A=G/ is
onto. By Corollary 3.7, the first Pext vanishes, and therefore so does the second. It
follows that the reduced part of A=G is torsion-complete. Consequently, the factor
group G =G is divisible for the closure G of G in A, so G is also pure in A. Thus
A=G is torsion-complete, hence G is torsion-complete by Corollary 3.8. It is pure
in A, so a summand. t
u
We draw particular attention to the next two theorems. The first result provides
a relevant information about the isomorphy of pure dense subgroups in torsion-
complete groups.
Theorem 3.10 (Leptin [1]). Two pure and dense subgroups, G and H, of a torsion-
complete p-group A are isomorphic if and only if A has an automorphism carrying
G onto H. Moreover, every isomorphism between G and H can be extended uniquely
to an automorphism of A.
Proof. We show that any isomorphism  W G ! H is induced by a unique
automorphism ˛ of A. Viewing  as a map G ! A, from the pure-exact sequence
0 ! G ! A ! A=G ! 0 we conclude that there is an ˛ W A ! A extending
G ! A; cf. Theorem 3.6(iii). Changing the roles of G and H, we obtain a map
ˇ W A ! A which extends  1 W H ! A. Now both ˇ˛ and ˛ˇ are endomorphisms
of A: the former is the identity on G, the latter is the identity on H, thus both are
identities on the respective basic subgroups. From our hypothesis it follows that both
G and H contain basic subgroups of A, thus ˇ˛ and ˛ˇ are identities on these basic
subgroups. Section 6(G) in Chapter 5 implies that ˇ˛ D 1A D ˛ˇ; thus ˛ 2 Aut A.
The uniqueness of ˛ is a consequence of the same (G), since any two extensions of
 induce the same map on a basic subgroup. t
u
An immediate consequence of Theorem 3.10 is that an isomorphism between
basic subgroups of a torsion-complete group A extends uniquely to an automor-
phism of A. Actually, this property characterizes the torsion-complete groups, as is
shown by
316 10 Torsion Groups

Theorem 3.11 (Leptin [1], Enochs [3]). A reduced p-group A is torsion-complete

if and only if every isomorphism between basic subgroups extends to an automor-
phism of A.
Proof. In order to verify the ‘if’ part, suppose A has the stated property. If A is
bounded, then it has only one basic subgroup, viz. itself, and there is nothing to
prove. So assume A is unbounded, and let B be a basic subgroup of A. As B is
unbounded, B < B. Let B0 be a subgroup of B such that B < B0 < B and B0 =B
is countable divisible. Because of Theorem 3.5, B0 is again a basic subgroup of B,
so by Theorem 3.10 there is an automorphism ˇ of B such that ˇB0 D B. Then
ˇB.< B/ is also basic in B, and in A as well, so by hypothesis, there is an ˛ 2 Aut A
such that ˛  B D ˇ  B.
Define a map  W B ! A as follows. Put b D b for every b 2 B. If x 2 BnB, then
it is included in some B0 , and we set x D ˛ 1 ˇx. This definition is unambiguous:
if B00 is another basic subgroup of B containing B0 , and if ˇ 0 2 Aut B with ˇ 0 B00 D B
and ˛ 0 2 Aut A with ˛ 0  B D ˇ 0  B, then ˇ 0 ˇ 1 B D ˇ 0 B0
ˇ 0 B00 D B and ˛ 0 ˛ 1
agrees with ˇ 0 ˇ 1 on ˇB D ˛B, and hence on B. We conclude that ˛ 0 ˛ 1 D ˇ 0 ˇ 1 ,
showing that ˛ 1 ˇ D ˛ 01 ˇ 0 , so that  is well defined. If x 2 Ker , then ˛ 1 ˇx D
0, ˇx D 0; and x D 0, that is,  is monic.
Therefore, B is a subgroup of A such that B D B. This means that B is pure
in A, and so Theorem 3.6(iv) implies that it is a direct summand of A. But A is
reduced and A=B is divisible, thus necessarily B D A. This proves that  is an
isomorphism, and A is torsion-complete. t
u
F Notes. Only a few classes of separable p-groups are well explored. In the countable case,
by Prüfer’s theorem, they are just †-cyclic groups, so the focus should be on the uncountable case.
No general structure theorem is available, and none is expected, but much is known about a few
classes with special properties (torsion-complete, quasi-complete, summable, etc. p-groups).
The theory of torsion-complete p-groups was developed by Kulikov [2], proving many of the
essential results. He also proved that any two direct decompositions of a torsion-complete group
have isomorphic refinements. Observe that the cardinality of any torsion-complete p-group A is
‘essentially’ of the form @0 for some . What we mean is that if A D G ˚ H with bounded H,
and  D jGj is minimal, then  D @0 .
There is a large body of work on torsion-complete groups, and numerous characterizations
exist for torsion-completeness. Waller [1] considers generalized torsion-completeness to p-groups
of countable limit length . Torsion-completion in the p -topology retains some of the pleasant
features of torsion-complete groups.
In retrospect, it is no surprise that torsion-completeness attracted so much attention, even before
the theory of algebraically compact groups was developed. The similarity of the two theories is
understandable, but, as far as importance is concerned, algebraic compactness is overwhelming.
It is instructive to study generalizations of separability involving higher cardinals. A p-group is
called -separable, if every subgroup of cardinality <  embeds in a †-cyclic summand of size
< . Megibben [8] investigates @1 -separable p-groups, and shows that their summands inherit this
property; he also proves several results that require additional set-theoretical hypotheses.
3 Torsion-Complete Groups 317

Exercises

(1) The torsion part of a direct product of p-groups is torsion-complete if and

only if each summand is torsion-complete. Q
(2) Using the notation of the text, prove that . n Bn /=B is a torsion-free
algebraically compact group which is not divisible unless 0.
(3) Large subgroups in torsion-complete groups are likewise torsion-complete.
(4) Using the notation in the text, an element g D .b1 ; : : : ; bn ; : : : / 2 B of order
pn generates a cyclic summand of B if and only if o.bn / D pn .
(5) (a) Let A be a separable p-group, and B an upper basic subgroup of A. If
B ¤ A, then jA=Bj  @1 , and there is a decomposition A D A0 ˚ A00 such
that A0 is †-cyclic and jA00 j D jA=Bj. [Hint: argue as in Theorem 3.5.]
(b) Every separable p-group A can be written as A D A0 ˚ A00 such that A0 is
†-cyclic, and every basic subgroup of A00 is both upper and lower.
(6) In a torsion-complete p-group, all basic subgroups are lower as well as upper.
[Hint: Theorem 3.11.]
(7) (Kemoklidze) If B is a basic subgroup of the p-group A, and if every
endomorphism of B extends to A, then either A D B or A D B.
(8) Let B be a torsion-complete p-group with fin rk B D . There is a direct
decomposition B D G ˚ H, where H is bounded and G is torsion-complete
of cardinality  @0 .
(9) Kernels (but not all images) of endomorphisms of torsion-complete groups
are again torsion-complete.
(10) (Faltings) A p-group A is isomorphic to the torsion part of Hom.A; Z.p1 //
if and only if it is torsion-complete with finite UK-invariants.
(11) Let A be a separable p-group with basic subgroup B.
A
(a) If B=A Š .Z.p1 //./ , then Pext.Z.p1 /; A/ Š .Jp /./ .
A
(b) The Ulm factors of Ext.Z.p1 /; A/ are: BQ and .Jp/./ .
(12) (Richman) Let A be a separable p-group, and G a bounded subgroup. A is
torsion-complete if and only if so is A=G. [Hint: Corollary 3.8.]
(13) Every height-preserving automorphism of the socle of B can be extended to
an automorphism of B.
(14) (Megibben) Let A be an unbounded torsion-complete p-group, and C an
arbitrary separable p-group. There exists a homomorphism of A into C which
is not small exactly if C contains an unbounded torsion-complete subgroup.
[Hint: exhibit an unbounded torsion-complete group in the image of a non-
small homomorphism.]
(15) Let T be a pure dense subgroup in the torsion-complete p-group B containing
the basic subgroup B such that rk.B=T/ D n 2 N. Then
(a) Pext.Z.p1 /; T/ Š ˚n Jp ; and
(b) if S is a separable p-group containing T as a pure dense subgroup with
rk.S=T/ D n, then S Š B.
318 10 Torsion Groups

4 More on Torsion-Complete Groups

With regard to the question of direct decompositions of torsion-complete groups,

we have definitive results.
Finite Direct Decompositions of Torsion-Complete Groups An unbounded
torsion-complete p-group has numerous direct decompositions into a finite number
of unbounded summands.
Proposition 4.1. Let B be a torsion-complete p-group with basic subgroup B. For
every direct decomposition B D C1 ˚    ˚ Ck with a finite number of summands,
there is a direct decomposition

B D C1 ˚    ˚ Ck

where bars denote p-adic closures. Also, Ci Š C i for i D 1; : : : ; k.

Proof. We represent the elements of B as in (10.4) of Sect. 3. Every a 2 B can be
written uniquely as a D a1 C    C ak where ai D .ci1 ; : : : ; cin ; : : : / and cin 2 Ci
for all n. Evidently, ai 2 Ci for i D 1; : : : ; k. The stated equality follows from this
representation. t
u
Infinite Direct Decompositions of Torsion-Complete Groups The behavior of
torsion-complete groups in infinite direct decompositions is quite unusual. We begin
with a preliminary lemma.
Lemma 4.2 (Enochs [1]). If a torsion-complete p-group A is contained in a direct
sum C D ˚i2I Ci of separable p-groups, then there exist an integer m and a finite
number of summands such that

pm AŒp

Ci1 ˚    ˚ Cik :

Proof. The proof is similar to the one in Lemma 2.16 in Chapter 6, with the
understanding that mj should mean pj , and the aj are selected in the socles. t
u
Further specialization A D C yields the following result that provides a pretty
complete information about the kind of direct decompositions torsion-complete
p-groups might have.
Theorem 4.3 (Kulikov [2]). If a torsion-complete p-group A is a direct sum A D
˚i2I Ci , then the Ci are torsion-complete, and for a sufficiently large integer m,
pm Ci D 0 for almost all i.
Proof. The torsion-completeness of the summands is evident, while the existence
of m as stated follows from the preceding lemma. t
u
Exchange Property for Torsion-Completeness The p-groups that are torsion-
complete belong to the exclusive club of groups that enjoy the exchange property.
This is proved like Theorem 6.7 in Chapter 6, the reference to Theorem 4.2 in
Chapter 16 includes torsion-complete groups as well.
4 More on Torsion-Complete Groups 319

Theorem 4.4 (Crawley–Jónsson [1]). Torsion-complete p-groups have the

exchange property. t
u
Large Subgroup Topology The balance of this section is devoted to the
topological aspects of torsion-completeness, in particular, to the large subgroup
topology.
Let A be a separable p-group, and fUj gj2J the family of its large subgroups which
we now declare as a subbase of open neighborhoods of 0. This is a linear topology
which we will denote by w. Thus the open subgroups in w are those that contain a
large subgroup, i.e. which satisfy the Pierce condition.

(a) Separable p-groups are Hausdorff in the large subgroup topology w. This is
obvious, since already the intersection of the large subgroups pn A .n 2 N/ is
the first Ulm subgroup A1 .
(b) The topology w is finer than the p-adic topology. This follows from the fact that
pn A is a large subgroup for each n < !.
(c) For a pure subgroup C of a separable p-group A, the large subgroup topology
is the same as the topology inherited from the large subgroup topology of A. In
fact, purity guarantees that C.u/ D A.u/ \ C.
(d) Every homomorphism  W A ! C between separable p-groups is continuous in
the large subgroup topology. Let c 2 C be the image of a 2 A under , and cCV
an open neighborhood of c. We may assume that V is a large subgroup of C, say,
defined by the sequence frn gn2N (cp. Theorem 2.3). If U is the large subgroup of
A defined by the very same sequence of integers, then clearly .aCU/  cCV.
(e) In a separable p-group, a bounded sequence is Cauchy in the p-adic topology
if and only if it is Cauchy in the large subgroup topology. The ‘if’ part being
obvious, let fai gi<! be a pk -bounded Cauchy sequence in the p-adic topology,
and U D A.r0 ; : : : ; rn ; : : : / a neighborhood of 0 in the large subgroup topology.
There is an index m such that ai  aj 2 prk k A for all i; j  m. Then the indicator
of ai  aj is  .rk  k; rk  k C 1; : : : ; rk ; 1 : : : /. As rk  k C `  r` .`
k/,
we have ai  aj 2 U for i; j  m.
(f) A Cauchy net in the large subgroup topology is bounded. By way of con-
tradiction, assume that there is an unbounded Cauchy net in w. This is also
Cauchy in the p-adic topology, so it contains an unbounded cofinal subsequence
a0 ; : : : ; ai ; : : : . We may assume that the orders o.ai / D pni increase with
i, and so do the heights hi of the differences ci D ai  ai1 . Define uD
.r0 ; r1 ; : : : ; rn ; : : : / such that the ri form an increasing sequence of positive
integers satisfying the inequalities: rni 1 > hi C ni for each i. Then the indicator
of ci is .hi ; hi C 1; : : : ; hi C ni  1; 1; : : : /, thus ci … A(u) for all i, contradicting
the Cauchy property of the subsequence.
Charles [3] defines the inductive topology on a p-group A by declaring those
subgroups G of A to form a subbase of neighborhoods about 0 which satisfy: for
each k 2 N,

GŒpk
D pn A \ AŒpk
for some n 2 N:
320 10 Torsion Groups

(g) The large subgroup topology on p-groups coincides with the inductive topology.
One way it is obvious: G satisfies the Pierce condition, and so it contains a large
subgroup of A. For the converse, observe that a group G satisfying the above
conditions is fully invariant in A: it is the union of the fully invariant subgroups
pn AŒpk
. In addition it satisfies the Pierce condition, so it is a large subgroup.
The principal result on the relation of torsion-completeness and the large
subgroup topology is as follows.
Theorem 4.5 (Kulikov [2], Charles [3], Cutler–Stringall [1]). For a separable
p-group A, the following are equivalent:
(i) A is torsion-complete;
(ii) every bounded Cauchy sequence in the p-adic topology of A is convergent;
(iii) A is complete in the large subgroup topology.
Proof. To fix the notation, as usual let B D ˚n2N Bn with Bn D ˚ Z.pn / be a basic
subgroup of A. Then a 2 A can be identified with a vector .b1 ; : : : ; bn ; : : : / 2 B
.bn 2 Bn /, where pm bn D 0 for every n if o.a/ D pm .
(i) ) (ii) Assuming A D B, let ak D .bk1 ; : : : ; bkn ; : : : / 2 B .k D 1; 2; : : : /
be a bounded Cauchy sequence in the p-adic topology where bkn 2 Bn . For
convenience, suppose the sequence is neat, so we have

akC`  ak D .bkC`;1  bk1 ; : : : ; bkC`;n  bkn ; : : : / 2 pk B

for all k; ` 2 N. This implies that bkC`;1 D bk1 ; : : : ; bkC`;k D bkk , showing that
the first k coordinates of ak ; akC1 ; : : : are the same, and in addition, bkC`;n 
bk;n 2 pk Bn for each n 2 N. Define b D .b11 ; b22 ; : : : ; bnn ; : : : / which belongs
to B in view of the boundedness of the sequence fak gk<! . This b is the limit of
the ak , because obviously

b  ak D .0; : : : ; 0; bkC1;kC1  bk;kC1 ; bkC2;kC2  bk;kC2 ; : : : / 2 pk B:

(ii) ) (iii) Just observe that (e) implies that a Cauchy net in the large subgroup
topology is a bounded Cauchy net in the p-adic topology.
(iii) ) (i) Let b D .b1 ; : : : ; bn ; : : : / .bn 2 Bn / be an element, say, of order pm , in
the torsion-complete group B  A. Consider the sequence an D b1 C  Cbn 2 A
for n 2 N. This is a Cauchy sequence in the large subgroup topology, since it is
bounded and anCk  an D bnC1 C    C bnCk 2 pnm A for all k  1. It ought
to have a limit in A, which cannot be anything else than b. Thus B
A, and (i)
follows. t
u
The preceding theorem says in effect that torsion-complete groups are exact
analogues of complete groups: the large subgroup topology takes over the role of
the p-adic topology.
We wind up this section with the following interesting corollary.
5 Pure-Complete and Quasi-Complete p-Groups 321

Corollary 4.6. The completion of a separable p-group in its large subgroup

topology is its torsion-completion.
Proof. Let A be a pure subgroup of the torsion-complete p-group B, and G the
completion of A in its large subgroup topology. The identity map of A extends to
a continuous map ı W G ! B which has to be surjective, because A is dense in B.
There is a similar map W B ! G, and the composite ı W B ! B must be the
identity. t
u
F Notes. Leptin [1] gave additional characterizations for torsion-completeness. In addition to
Theorem 3.10, he proved that if a basic subgroup B of a separable p-group A has the property that
every automorphism of B extends to an automorphism of A, then either A D B or A D B.
Crawley [1] solves one of the Kaplansky test problems for p-groups in the negative, by pointing
out that it is possible to have p-groups A; C satisfying A Š C ˚ C ˚ A, but A 6Š C ˚ A:
In the proceedings of the 1967 Montpellier conference, both Charles [3] and Cutler–Stringall
[1] pointed out the completeness of torsion-complete groups, using different approaches: the
former used the inductive topology arising from the topologies of the subgroups AŒpk
, induced
by the p-adic topology, while the latter used the large subgroup topology.

Exercises

(1) Every subgroup C of a torsion-complete p-group A can be embedded in a

summand G of A such that jGj
jCj@0 . [Hint: embed in pure subgroup.]
(2) A torsion-complete p-group that is contained in a †-cyclic p-group must be
bounded.
(3) A torsion-complete p-group A fails to have a proper pure subgroup Š A if and
only if its UK-invariants are finite.
(4) If C is a torsion-complete subgroup of a separable p-group A such that A=C is
bounded, then A is also torsion-complete. [Hint: apply Ext.Z.p1 /; /.]
(5) (Koyama–Irwin) A separable p-group A is torsion-complete if and only if
every decomposition B D C1 ˚ C2 of its basic subgroup B induces a direct
decomposition A D C1 ˚ C2 .
(6) (Irwin–O’Neill) Suppose A D B ˚ C is a separable p-group that contains an
unbounded torsion-complete subgroup. Then either B or C has an unbounded
torsion-complete subgroup.

5 Pure-Complete and Quasi-Complete p-Groups

It does not seem unreasonable to anticipate that torsion-complete p-groups have

many more properties of interest. Some of these are shared by wider classes of
groups as well. In this section, we discuss briefly two such special properties. Both
are related to the existence of certain pure subgroups, motivated by the observations
322 10 Torsion Groups

that in torsion-complete p-groups every subsocle supports a pure subgroup, and

closures of pure subgroups in the p-adic topology preserve purity.
Pure-Completeness A p-group A is called pure-complete if every subsocle of
A supports a pure subgroup of A. Because of Theorem 1.3, this property is shared
by all dense subsocles.
Since in a pure-complete p-group, also the subsocle consisting of all the elements
of infinite heights must support a pure subgroup (which cannot be anything else than
a divisible subgroup), we conclude that every such group decomposes into the direct
sum of a separable and a divisible group.
Example 5.1. Let B be the direct sum of cyclic groups of fixed order pk . Then every subsocle of
B supports a direct summand of B, so B is pure-complete. More generally, a torsion group with
bounded p-components is pure-complete.
Example 5.2. Divisible p-groups are pure-complete.

It is rather obvious that summands of pure-complete groups are again pure-

complete. However, the direct sum of two pure-complete p-groups need not be
pure-complete as is shown by the following example.
Example 5.3 (Hill–Megibben [2]). Suppose A and C are non-isomorphic p-groups, and there is a
height-preserving isomorphism  W AŒp
! CŒp
between their socles (such p-groups do exist, see
e.g. Theorem 9.2). Assuming A is quasi-complete (see below), it follows from the characterization
to be stated in the Notes below that so is C, and hence by Corollary 5.6 infra they are pure-
complete. However, the subsocle S D f.x; x/ j x 2 AŒp
g of the direct sum A ˚ C does not
support any pure subgroup. In fact, a pure subgroup supported by S ought to be a subdirect sum of
A and C with 0 kernels. But such a subdirect sum could exist only if A and C were isomorphic.

Lemma 5.4 (Hill–Megibben [3]).

(i) †-cyclic p-groups are pure-complete.
(ii) The direct sum of a countable number of torsion-complete p-groups is pure-
complete.
Proof. (i) This will be a simple consequence of (ii), since we can write a †-cyclic
p-group as A D ˚n2N Bn where Bn D ˚ Z.pn / is torsion-complete.
(ii) First we show that a torsion-complete p-group A is pure-complete. Actually,
we prove somewhat more: if S is a subsocle of A, and H is a pure subgroup
of A with HŒp

S, then A contains a pure subgroup G such that H
G and
GŒp
D S. Let B0 be a basic subgroup of H, and B00 a maximal †-cyclic pure
subgroup of A that contains B0 and satisfies B00 Œp

S. Finally, let B be a basic
subgroup of A that contains B00 as a summand, say, B D B00 ˚ C. We then have
00 00
A D B ˚ C. Since B00 Œp
must be dense in S, we have S
B Œp
. An appeal to
Theorem 1.3 establishes the existence of a G with the desired properties.
Now suppose A D ˚1 nD1 An , where the An are torsion-complete p-groups. Let
S denote a subsocle of A. Then Sn D S \ .A1 ˚    ˚ An / is a subsocle of the
torsion-complete p-group A1 ˚  ˚An , so it supports a pure subgroup Gn . What
we have proved in the preceding paragraph shows that the pure subgroups Gn
can be selected so as to form a chain G1
  
Gn
: : : . It is then clear that
G D [n Gn is pure in A and is supported by S. t
u
5 Pure-Complete and Quasi-Complete p-Groups 323

Quasi-Complete Groups Let A be a reduced p-group. It is called quasi-

complete (Head[1]) if the closure G (in the p-adic topology) of every pure
subgroup G is again pure in A. Since G =G can be identified as the first Ulm
subgroup of A=G, the quasi-completeness of A means that, for any pure subgroup G
of A, the first Ulm subgroup .A=G/1 of A=G is pure in A=G, i.e. divisible.
We list a few trivialities for quasi-complete groups.
(A) Reduced quasi-complete p-groups are separable. This follows from 0 D A1 .
(B) Torsion-complete groups are quasi-complete. This is a simple consequence of
Corollary 3.9.
(C) Closed pure subgroups in a quasi-complete group are quasi-complete.
(D) If G is a closed pure subgroup of a quasi-complete p-group A, then A=G is
likewise quasi-complete. For, if H=G is a pure subgroup of A=G, then H is pure
in A, so Œ.A=G/=.H=G/
1 Š .A=H/1 must be divisible.

Next we show that the property that we verified for torsion-complete groups in
the proof of Lemma 5.4 actually characterizes quasi-completeness.
Proposition 5.5 (Irwin–Richman–Walker [1], Koyama [1]). A reduced p-group
A is quasi-complete if and only if, for every pure subgroup H of A, and for every
subsocle S with HŒp

S, there exists a pure subgroup G of A such that H
G and
GŒp
D S.
Proof. Let A be quasi-complete, H a pure subgroup of A such HŒp

S
AŒp
; and
G a pure subgroup maximal with respect to the properties H
G and GŒp

S. To
prove GŒp
D S, by way of contradiction assume that there is an x 2 S n G. If the
coset x C G has finite height k in A=G, then write x C G D pk y C G with y 2 A.
Now hy C Gi is a summand of A=G, and G ˚ hyi is a pure subgroup supported
by GŒp
˚ hxi
S, a contradiction. If x C G has infinite height in A=G, then by
quasi-completeness it is contained in a subgroup G0 =G Š Z.p1 /. Clearly, G0 is
pure in A and is supported by GŒp
˚ hxi
S, again a contradiction.
Conversely, let A have the stated property. First choosing S D A1 Œp
, we obtain
1
A divisible, and hence 0. Next, for any pure subgroup H, choose G pure containing
H with socle S D H  Œp
. Then G=H is pure in A=H with socle .A=H/1 Œp
.
Consequently, G=H is divisible and equal to H  =H, so A is quasi-complete. t
u
Hence it follows at once:
Corollary 5.6. Quasi-complete p-groups are pure-complete. t
u
We isolate a preparatory result in the next lemma that will be needed in later
proofs.
Lemma 5.7 (Hill–Megibben [3]). Suppose A is a quasi-complete p-group, and B0
is an unbounded summand of a basic subgroup B of A. Then B D A C B0 .
Proof. Let B D B0 ˚ B00 , and  W B ! B00 the projection map. Then .A/ is pure
in B00 , since B00
.A/ and Coker.  A/ is divisible. We want to prove that
Coker.  A/ D 0 by showing that every b 2 B00 Œp
is in .A/. Let C denote a basic
324 10 Torsion Groups

subgroup of B0 with B0 =C Š Z.p1 /, and u 2 B0 n C an element of order p2 . One can

easily construct a pure †-cyclic subgroup X < A with XŒp
D CŒp
and b C u 2 X  .
By quasi-completeness, the closure A \ X  of X in A is pure in A, so pu 2 A \ X 
implies that pa D pu for some a 2 A \ X  . Hence b D .b C u  a/  .u  a/ 2
X  Œp
C AŒp

B0 C A, completing the proof. t
u
Another way of looking at quasi-completeness relies on the property stated in
Corollary 3.9 for torsion-complete groups.
Theorem 5.8 (Hill–Megibben [2]). A separable p-group A is quasi-complete if
and only if A=G is a direct sum of a divisible and a torsion-complete group,
whenever G is any unbounded pure subgroup of A.
Proof. For sufficiency, note that the stated condition implies that the closure of G
will also be pure in A. For the proof of necessity, assume A is quasi-complete and
contained in B, where B is a basic subgroup of A. The p-adic closure H (in B) of
an unbounded pure subgroup G is by Corollary 3.9 a summand of B, hence torsion-
complete. Since A contains a basic subgroup of G, and hence of H, from Lemma 5.7
we conclude that A C H D B. In view of quasi-completeness, .A \ H/=G Š .A=G/1
is divisible. Therefore, A=G is isomorphic to a direct sum of a divisible group and
A=.A \ H/ Š .A C H/=H D B=H, a summand of B. t
u
It follows from Lemma 5.7 that if B D B0 ˚ B00 is a decomposition of a basic
subgroup of a quasi-complete p-group A, then

.A \ B0 / ˚ .A \ B00 /
A
B0 ˚ B00

and A is a subdirect sum of B0 and B00 with kernels A \ B0 and A \ B00 . This remark
is now used to prove that quasi-complete p-groups that are not torsion-complete are
few and far between.
Theorem 5.9 (Hill–Megibben [2]). A quasi-complete p-group of final rank > 2@0
is torsion-complete.
Proof. Let A be quasi-complete of final rank > 2@0 , and B D B0 ˚ B00 a
decomposition of its basic subgroup such that B0 is countable and unbounded. In
view of the mentioned subdirect sum representation, we have B00 =.A \ B00 / Š
B0 =.A \ B0 / where the last quotient has cardinality
jB0 j
2@0 . Let S denote a
complete set of representatives of B00 modulo A \ B00 . Clearly, S must be contained
in the completion of a summand C of B00 which has cardinality not exceeding 2@0 .
Thus B00 D C ˚ C0 for some C0 < B00 of final rank > 2@0 where C0
A. By
Lemma 5.7, B D A C C0
A, and hence A is torsion-complete. t
u
Thus, if we are asking for an example of a quasi-complete p-group that is not
torsion-complete, then we should not look beyond the continuum. Such an example
(of cardinality exactly 2@0 ) was supplied by Hill–Megibben [2].
F Notes. Lady [1] proves that the torsion subgroup of a direct product of
†-cyclic p-groups is pure-complete.
6 Thin and Thick Groups 325

It was Head [1] who first studied quasi-completeness. Let us mention the following useful
characterization of quasi-completeness. Let A be a separable p-group embedded in the torsion-
completion B of its basic subgroup B. Hill–Megibben [2] prove that A is quasi-complete if and
only if BŒp
D AŒp
C S holds for every subsocle S of A where the heights of elements of S are
unbounded (S is calculated in B)—actually, this follows from the proof of Lemma 5.7. Another
interesting result is concerned with unbounded torsion-complete p-groups T of cardinality 2@0 .
Given any countable subgroup A of T, there exists a proper quasi-complete pure subgroup G < T
such that G \ A D 0 and T=G is divisible.

Exercises

(1) Define a pure-complete torsion group in the same way as it was done for p-
groups, and show that a torsion group is pure-complete if and only if all of its
p-components are pure-complete.
(2) (Hill–Megibben) If A is pure-complete, and C is †-cyclic, then A ˚ C is also
pure-complete.
(3) (Hill–Megibben) If A is a pure-complete, and C is a torsion-complete p-group
with finite UK-invariants, then A ˚ C is again pure-complete.
(4) The direct sum of two quasi-complete p-groups need not be quasi-complete.
[Hint: quasi-complete plus large torsion-complete.]
(5) (Hill–Megibben) Show that in any direct decomposition of a quasi-complete p-
group that is not torsion-complete, only one of the summands can be unbounded
(i.e., the group is essentially indecomposable). [Hint: Theorem 5.8.]
(6) (Benabdallah–Irwin) Call a subgroup G of a p-group A almost dense if, for
every pure subgroup C of A containing G, the factor group A=C is divisible.
Show that G is almost dense in A exactly if pn AŒp

G C pnC1 A holds for all
n < !.

6 Thin and Thick Groups

Small homomorphisms have been discussed in Sect. 3 in Chapter 7. Several features

make them especially interesting for study. Here we discuss two different types
of p-groups defined in terms of small homomorphisms. Their relation to torsion-
completeness will be evident.
Thin Groups A p-group C is said to be thin (Richman [1]) if every homo-
morphism  W B ! C from the standard torsion-complete p-group B into C is small
(recall: B Š ˚n2N Z.pn /). (In the category of p-groups, thin groups play similar role
as slender groups in the category of torsion-free groups; see Sect. 2 in Chapter 13.)
It is easily checked:

(a) Subgroups of thin groups are thin.

(b) Extension of a thin group by a thin group is again thin.
326 10 Torsion Groups

(c) Homomorphisms of any unbounded torsion-complete p-group into a thin group

are small. This will follow immediately from the more general Lemma 6.4 that
is needed in the proof of the subsequent theorem.
Lemma 6.1 (Richman [1]). Direct sums of thin groups are thin.
Proof. The proof follows the arguments in Lemma 2.16 in Chapter 6 and in
Lemma 4.2 with obvious modifications. u
t
Example 6.2 (Megibben [5]). †-cyclic groups are thin. This is an immediate consequence of
Lemma 6.1, observing that cyclic p-groups are trivially thin.
Example 6.3. The group Z.p1 / is not thin. The standard torsion-complete p-group B evidently
has a map  onto Z.p1 / such that B  Ker . Such a Ker  cannot contain a large subgroup of B,
since B C Ker  ¤ B.

Lemma 6.4 (Keef [6]). Let Ai .i < !/ be p-groups. If

Y
W t. Ai / ! C
i<!

is a homomorphism into a thin group C, then for each k 2 N there is n 2 N such

that
Y
. pn Ai Œpk
/ D 0:
i>n

Proof.
Q If this were not true, then for some k we could find elements an 2
i>n p n
Ai Œpk
for all nQ< ! such that Q.an / ¤ 0. Let hbn i be a cyclic group of
order p , and  W t. n<! hbn i/ ! t. Q
kCn
i<! Ai / a homomorphism mapping p bn
n

upon an . Then the composite map  W t. n<! hbn i/ ! C would not be small (its
kernel fails the Pierce condition), contrary to our hypothesis on C. t
u
The condition of non-measurability appears in the next result: another evidence
of its intimate relation to direct products.
Q
Theorem 6.5 (Keef [6]). Let i W Ai ! A D t. i2I Ai / denote the ith injection
map, where the Ai are p-groups, and I is a non-measurable index set. A p-group C
is thin if and only if it satisfies:
. / a homomorphism  W A ! C is small exactly if i is small for every i 2 I.
Proof. First, suppose C satisfies . /, and  W B ! C is a homomorphism, where B
is the torsion-completion of the standard basic subgroup B. Since the induced maps
Bn ! C are small homomorphisms, hypothesis implies that so is . Consequently,
C is thin.
Conversely, assume that C is thin, and let  W A ! C be such that i is
small for every i 2 I, but  is not small. Then Q in view of the definition of small
homomorphisms, there is a k 2 N such that . Q i2I p Ai Œp
/ ¤ 0 for all n 2 N.
n k

Consider the set S of all subsets J  I such that . i2J p Ai Œpk

/ ¤ 0 for all n 2 N.
n

As the i are small, none of J 2 S is finite. From Lemma 6.4 it follows that if
6 Thin and Thick Groups 327

J1 ; : : : ; Jj ; : : : are pairwise disjoint subsets in S, then [j>m Jj … S must hold for

some m 2 N. The proof of Lemma 6.4 in Chapter 2 applied to  (in place of f )
convinces us that I is measurable, a contradiction. t
u
Example 6.6. To show that the last theorem may fail for measurable cardinals, let I be a
measurable index set, and Ai copies of theQsame countable, unbounded †-cyclic p-group X, and
also C D X. In order to define  W A D t. i2I Ai / ! X, choose a vector a D .: : : ; ai ; : : : / 2 A,
and regard the coordinates as elements of X. There are only countably many different coordinates,
so the supports of the equal ones give rise to a countable partition of I into disjoint subsets, exactly
one of which has measure 1, and the rest has measure 0. If the support of x 2 X is the one of
measure 1, then set .a/ D x. This is a well-defined homomorphism; it is not small, though i is
small for each i 2 I.

Thick Groups A p-group G is called thick (Megibben [2]) if every homomor-

phism  W G ! C into a †-cyclic group C is small.
Example 6.7. Torsion-complete p-groups are thick. This is a consequence of Example 6.2.
Example 6.8 (Megibben [2]). A pure dense subgroup G of a torsion-complete p-group B is thick
whenever jB=Gj  @0 . Indeed, let  W G ! C be a homomorphism, where C is †-cyclic. Then
A D Im  is likewise †-cyclic. Now,  extends uniquely to  W B ! A, where A denotes the
torsion-completion of A. As Im =A is countable, by Theorem 3.5 Im  is also †-cyclic. Thus  is
small, and so is  D   G.

In the following theorem, which is a dual to Theorem 6.5, i denotes the ith
coordinate projection.
Theorem 6.9 (Rychkov–Thomé [1], Keef [6]). A p-group G is thick if and only if
the following holds:
. / for every set fAi j i 2 Ig of separable p-groups, a homomorphism  W G !
˚i2I Ai is small exactly if i  is small for every i 2 I.
Proof. If G has property . / and the Ai are cyclic p-groups, then  small
implies G is thick. For the converse, assume  W G ! ˚i2I Ai is a non-small
homomorphism, though all of i  are small. Thus for each k < ! there is n 2 N
such that .pn GŒpk
/ ¤ 0. Because the homomorphisms i  are small, we can
find inductively integers nj , elements xj 2 pnj GŒpk
, and different indices ij 2 I
such that ij .xj / ¤ 0. As Aij is separable, there is a map j W Aij ! Bi , to
a cyclic p-group, such that i .ij .xj // ¤ 0. The j ’s yield the composite map
W ˚i2I Ai ! ˚j<! Aij ! ˚j<! Bi such that .xj / ¤ 0, showing that  is not
small, so G is not thick. t
u
Theorem 6.10 (Keef [6]). If the p-groups Ai .i 2 I/ are thick, then so is the torsion
subgroup of their direct product.
Proof. If I is finite, then the assertion follows from the fact that the direct sum of
large subgroups
Q of the summands is large in the direct sum. Suppose I is infinite,
and  W t. i2I Ai / ! C where C is †-cyclic. By the analogue of Theorem Q 6.7 in
Chapter 2, there are a finite subset J  I and an integer m such that pm t. i2InJ Ai / ¤
0 is mapped by  into a finite summand of C. If this summand is annihilated by pk ,
328 10 Torsion Groups

Q  contains, along with a direct sumQof large subgroups of Ai .i 2 J/, also

then Ker
pmCk t. i2I Ai /. Therefore,  is small, and t. i2I Ai / is thick. t
u
F Notes. Megibben [2] shows that a separable p-group is thin exactly if it does not contain an
unbounded torsion-complete p-group. Generalizations of thin and thick groups have been discussed
by Rychkov–Thomé [1].
Keef [3] proves an interesting result on the torsion subgroup of a countable product of p-groups
(generalizing Theorem 6.10 in Chapter 5): for any collection fAn gn2N of p-groups, there exists an
epimorphism
!
Y
W t An ! ˚n2N An :
n2N

Exercises

(1) Every homomorphism B ! B is small (B is a †-cyclic p-group).

(2) (Keef)
(a) Given a p-group A, let S.A/ denote the class of p-groups C such that all
homomorphisms  W A ! C are small. Prove that S.A/ is closed under
taking subgroups.
(b) If A is unbounded and reduced, then every group in S.A/ is thin.
(3) In the definition of thick p-groups, can the †-cyclic C be restricted to the
standard B D ˚n2N Z.pn /?
(4) A p-group is thick if and only if its initial Ulm factor is thick.
(5) (Keef) Let A; C be reduced unbounded p-groups. There exists a non-small
homomorphism A ! C if either A is not thick or C is not thin.

7 Direct Decompositions of Separable p-Groups

Every non-trivial separable p-group has an indecomposable summand: a cyclic

group, but the group itself need not be a direct sum of indecomposable groups.
The obvious question is: what kind of decompositions a separable p-group might
have? We would like to use the knowledge that we have gained on torsion groups
so far to draw attention to some not so common direct decompositions. We will
mention a few results of this sort. We point out right away that the presence of
cyclic summands rules out the existence of superdecomposable p-groups.
Direct Decompositions We know that a separable p-group has an ample supply
of finite summands. These are of course †-cyclic groups. How about unbounded †-
cyclic summands? We refer to Khabbaz’ Theorem 6.14 in Chapter 5 showing that
in case the basic subgroups have the same cardinality as the p-group A, then A has
†-cyclic summands as large as its cardinality permits.
7 Direct Decompositions of Separable p-Groups 329

@1 -Separability Almost free groups have analogues in the torsion case:

p-groups in which every subgroup of smaller cardinality is †-cyclic, see
Corollary 8.11 in Chapter 3. A stronger version requires that every subgroup of
smaller cardinality can be embedded in a †-cyclic summand.
Theorem 7.1 (Hill [8], Crawley–Megibben [1]). There exists a separable p-group
of cardinality @1 , not a †-cyclic group, such that every countable subgroup embeds
in a †-cyclic direct summand.
Proof. Let B D ˚n Bn ; Bn Š ˚@0 Z.pn /, and C a basic subgroup of B such that
B=C Š Z.p1 /. We construct a p-group A of cardinality @1 as the union of an
ascending chain of pure subgroups, all isomorphic to B.
We start with A0 D 0. Suppose that for some ordinal ˇ < !1 , the countable
†-cyclic groups A˛ have been defined for all ˛ < ˇ to form a smooth chain,
S being pure in its successors. If ˇ is a limit ordinal, then we have to set Aˇ D
each
˛<ˇ A˛ . As a countable separable p-group, Aˇ will then be †-cyclic (Theorem 5.5
in Chapter 3). If ˇ D C 1 and is a successor ordinal, then define Aˇ D A ˚ F
where F Š B. Finally, if ˇ D C 1 and is a limit ordinal, then choose Aˇ Š B
with A embedded in Aˇ as C isSembedded in B. Then all the groups A˛ are †-cyclic
and pure in their union A D ˛<!1 A˛ . Moreover, each A˛ with non-limit ˛ is a
direct summand of every Aˇ with ˇ > ˛, where the complements form a chain, so
it is a summand of A as well. Since A has cardinality @1 , and every countable subset
of A is contained in some A˛ , it is clear that A has the required embedding property.
However, A cannot be †-cyclic, since it violates the criterion of Theorem 7.5 in
Chapter 3 (see the Remark after it): A˛C1 =A˛ Š Z.p1 / is not †-cyclic for every
limit ordinal ˛ < !1 , and the set of limit ordinals < !1 is stationary in !1 . t
u
Summands of Various Properties Features that we want to consider next are
concerned with the size of direct decompositions: when one of the summands ought
to be small or large in size.
A p-group A is said to be essentially indecomposable if it is incapable of
decomposition into two unbounded summands; i.e. if A D B ˚ C, then either B
or C is bounded. Bounded p-groups share this property, but no unbounded torsion-
complete group has it (see Proposition 4.1). The next example exhibits an essentially
indecomposable p-group of the cardinality of the continuum (for another example
see Sect. 5, Exercise 5).
Example 7.2. Pierce [1] shows that the torsion-complete group B (where B denotes the standard
basic subgroup) contains a pure subgroup G > B of cardinality 2@0 such that End G Š Jp ˚Ends G.
Idempotents in Ends G yield bounded summands, therefore only idempotents mod Ends G could
produce two unbounded summands. But Jp has only a single idempotent ¤ 0, so such a G must be
essentially indecomposable.

From the existence Theorem 7.3 in Chapter 16 one can derive the existence of
arbitrarily large essentially indecomposable p-groups, even a large family of such
groups with the property that all homomorphisms between different members are
small.
330 10 Torsion Groups

A class more general than essentially indecomposable groups was studied by

Cutler–Irwin [1]. They call a p-group A essentially finitely indecomposable, or
e.f.i. for short, if for every direct decomposition of A there is an integer n > 0
such that almost all summands are pn -bounded. An equivalent condition is: A has
no unbounded †-cyclic subgroup as a summand.
Example 7.3. Thick groups are e.f.i. To verify this, assume A is a p-group that is not e.f.i., so it
has an unbounded †-cyclic summand B such that A D B ˚ C with unbounded C. The projection
A ! B is not a small homomorphism, A is not thick.

Lemma 7.4 (Cutler–Irwin [1]). A reduced p-group is e.f.i. if and only if its initial
Ulm factor is e.f.i.
Proof. If A has an unbounded †-cyclic group as a summand, then this summand
remains summand mod the first Ulm subgroup A1 . Conversely, if A=A1 has an
unbounded †-cyclic summand, then by Theorem 6.13 in Chapter 5 every basic
subgroup of A=A1 has such a summand. Basic subgroups of A=A1 come from those
of A, so also A has an unbounded †-cyclic summand. t
u
Quasi-Indecomposable Groups A group A is called quasi-indecomposable if,
for every cardinal  < jAj, in every direct decomposition of A (into any number of
summands) there is always a summand of cardinality > . It is obvious that

(a) a reduced countable p-group is quasi-indecomposable exactly if it is

unbounded;
(b) if jAj is a successor cardinal, then quasi-indecomposability means that in every
direct decomposition one of the summands has the same cardinality as A.
Example 7.5. The torsion-complete group B with the standard basic subgroup B is quasi-
indecomposable of cardinality 2@0 .

Theorem 7.6 (Kulikov [1, 2]). For every infinite cardinal , there exist quasi-
indecomposable separable p-groups of cardinality .
Proof. Ignoring the countable case settled by (a), we distinguish three mutually
exclusive cases.
Case 1: There is an infinite cardinal  satisfying  < 
 @0 . Define A as a pure
dense subgroup of cardinality  in the torsion-complete group B with basic
subgroup B D ˚1 nD1 Bn where Bn D ˚ Z.p /. Suppose A D ˚i2I Ci with
n

Ci ¤ 0. Here necessarily jIj

jBj D , because every summand must have
a non-zero basic subgroup. Therefore, if  < , then jCi j
 for all i 2 I is
impossible, thus A is quasi-indecomposable.
Case 2:  D @ C1 is a successor cardinal, and  <  implies @0 < . Then
@@ 0 D @ , and by the Hausdorff formula (Jech [J]), @@ C1
0
D @ C1 also holds,
@0
i.e.  D . Define A D B with B D ˚n Bn ; Bn D ˚ Z.pn /. Manifestly, A
has cardinality and final rank . If A D ˚i2I Ci , then by Theorem 4.3 there is
an integer m such that almost all pm Ci D 0. At least one of the unbounded Ci
must have final rank .
7 Direct Decompositions of Separable p-Groups 331

P
Case 3:  is a limit cardinal such that  <  implies @0 < . Write  D  
with an increasing sequence of cardinals  <  . < cf /, and for each 
construct a torsion-complete group A D B (as in Case 1) with the cardinal
 . Then A D ˚ <cf  A will have cardinality and final rank . Suppose A D
˚i2I Ci . By Theorem 4.3, for each ; there is an integer m such that pm A Œp

is contained in the direct sum of a finite number of the Ci , and it is clear that
one of these must have cardinality   . Thus for every  <  there is a Ci of
cardinality > , so A is quasi-indecomposable. t
u
-Indecomposable Groups Another property we wish to look at relates to the
number of summands. A group A is -indecomposable if it cannot be written as a
direct sum of  non-zero groups. Of course, in this case, only groups of cardinality
  are of interest.
Example 7.7. The group in Case 1 of Theorem 7.6 above is -indecomposable (of cardinality
> ). In fact, every direct decomposition of A induces a direct decomposition of a basic subgroup
of A. Every basic subgroup is of cardinality , so the group cannot be a direct sum of more than 
non-zero summands.

In view of Theorem 5.3 in Chapter 3, no countably infinite, @0 -indecomposable

separable p-group may exist. However, for larger cardinalities we have a simple
criterion for the existence:
Theorem 7.8 (Szele [7], Fuchs [4], Khabbaz [1]). For an uncountable infinite
cardinal , there exist -indecomposable separable p-groups of cardinality  
if and only if either
(i) there is a cardinal  such that  < 
 @0 ; or
(ii)  is a limit cardinal of cofinality !.
(i) is a necessary and sufficient condition for the existence of groups of the stated
kind to be of cardinality equal to .
Proof. First assume there is a -indecomposable p-group A of cardinality . Let
B D ˚n Bn be a basic subgroup of A, where P Bn D ˚n Z.p /. As Bn is@ a summand,
n

we must have n <  for each n. Then n n D jBj

jAj
jBj 0 . If the first
inequality is strict, we get case (i). In the remaining cases  <P implies @0 < ,
and either  D  C with  @0 D , or  is a limit cardinal with n n D , in which
case cf  D ! is a must.
If  satisfies (i), then let A be defined as in Case 1 above in Theorem 7.6.
Example 7.7 establishes the existence of such an A.
To rule out the case when  D  C is a successor cardinal and  @0 D , note that
for such a  the basic subgroup B of any separable p-group A of cardinality   must
also be of cardinality  . If jBj D ; then also jAj D ./@0 D  (by Hausdorff
formula), and Theorem 6.14 in Chapter 5 implies that A has a †-cyclic summand of
cardinality . If jBj > ; then B, and hence also A, must have a bounded summand
of size . Such an A cannot be -indecomposable. P
If  is like in (ii), then let 0 <    < n < : : : be cardinals with n n D ;
we may assume .n /@0 D n for each n. Let A D B be the torsion-complete group
332 10 Torsion Groups

with basic subgroup B D ˚n Bn , where Bn D ˚n Z.pn /. Then rk B D fin rk B D ,

thus jAj D @0 —as it is clear from Theorem 3.2; here, @0 >  by Jech [J]. Now
if A D ˚i2I Ci , then by Theorem 4.3 there is an integer m such that almost all of
pm Ci vanish. Obviously, the cardinality of a pm -bounded summand cannot exceed
m < , so the index set I has cardinality < , and A is -indecomposable.
To prove the second claim, it only remains to show that (ii) is not an option if we
wish the group to have cardinality equal to , and (i) fails. If  is a limit cardinal
such that  <  implies @0 < , then in any separable p-group of cardinality , the
basic subgroups ought to have cardinality . We invoke Theorem 6.14 in Chapter 5
to argue that then the group has a †-cyclic summand of cardinality . t
u
Direct Product vs. Direct Sum The following theorem records a remarkable
feature of torsion groups. Interestingly, but not surprisingly, the natural boundary
for the result is the first measurable cardinal. To simplify notation, we confine our
discussion to p-groups.
Theorem 7.9 (Zimmermann-Huisgen [1]). Let Ai .i 2 I/ be reduced p-groups,
and I a non-measurable index set. In any infinite direct decomposition of the direct
product
Y
AD Ai D ˚j2J Cj ;
i2I

no more than a finite number of the Cj ’s are torsion-free.

Proof. There is a smallest ordinal  such that p Ai D 0 for all i 2 I. Then also
p Cj D 0 for all j 2 J, so there is a smallest  such that p Cj is torsion for almost
all j 2 J. We will be done if we show that  D 0.
By way of contradiction, assume  > 0. Theorem 6.7 in Chapter 2 guarantees
the existence of an ordinal  <  as well as finite subsets I0  I; J0  J such that

p BI0
.˚j2J0 Cj / C .˚j2J p Cj /;
Q 
where BI0 D i2InI0 Ai . No harm is done if we replace ˚j2J0 Cj by ˚j2J0 p Cj
  
in the last containment relation. Passing mod p A, we obtain p BI0 =p BI0 !
˚j2J0 .p Cj =p Cj /; a monomorphism. Both sides may be viewed as subgroups
of p A=p A, therefore there is an epimorphism between the cokernels:
Q    
i2I0 .p Ai =p Ai / ! ˚j2JnJ0 .p Cj =p Cj /. Here the finite product is a p-group,
so the same has to be true for the direct sum. This means that  is also an ordinal
such that almost all of p Cj are torsion. Thus  cannot be minimal unless  D 0. ut
Number of Separable p-Groups The question as to the cardinality of the set
of non-isomorphic separable p-groups of cardinality  arises naturally. Shelah [2]
proved that the cardinality is 2 for every infinite cardinal , using a more general
result of his on the number of non-isomorphic models. Here we cannot give details
of his proof; instead, we offer a weaker theorem showing only that there exist
arbitrarily large cardinals with this property.
7 Direct Decompositions of Separable p-Groups 333

Proposition 7.10. Let  be an infinite cardinal such that  @0 D 2 . Then there


exist 22 non-isomorphic separable p-groups of cardinality 2 . Moreover, they can
be chosen so as to have isomorphic basic subgroups.
Proof. The claim is a bonus of Theorem 9.2 below. t
u
Regarding Proposition 7.10, we point out that there are arbitrarily large cardinals
 satisfying the hypothesis of this proposition (Griffith
P [6]). In fact, starting with
any infinite cardinal 0 , set nC1 D 2n and  D n<! n . Then
Y Y P
 @0  n D 2n D 2 n
D 2
n<! n<!

along with the reverse inequality implies  @0 D 2 .

F Notes. Monk [2] proves that a p-group A is essentially indecomposable if and only if the
factor ring End A= Ends A has no idempotents other than 0 and 1. An interesting variant is Irwin’s
essentially finitely indecomposable p-groups defined above. The class of these groups is closed
under finite direct sums and summands.
Rychkov [3] considers p-groups A of cardinality  > @0 and of final rank  that are not †-
cyclic, but every subgroup of cardinality <  embeds in a †-cyclic summand of cardinality < .
Assuming V = L, he establishes the existence of such an A for every uncountable, regular, not
weakly compact cardinal . A has the additional property that it cannot be decomposed into the
direct sum of two summands of the same final rank.
Shelah [2] establishes the existence of families of separable p-groups like in Proposition 7.10
such that every homomorphism between two different groups is small.

Exercises

(1) A torsion-complete p-group A with fin rk A D jAj is quasi-indecomposable.

(2) An unbounded quasi-complete p-group with a countable basic subgroup is
quasi-indecomposable.
(3) In a separable p-group A, the summand C constructed for a 2 A in Lemma 1.4
is a minimal summand containing a.
(4) (Soifer, Göbel–Ziegler) A group is called almost -decomposable for a
cardinal  if, for every cardinal  < , it admits a direct decomposition
with  non-zero summands, but has no decomposition into  such summands.
There exist almost -decomposable p-groups of cardinality  for any infinite 
satisfying C D 
@0 for some cardinal . [Hint: suitable B.]
(5) The direct sum of a finite number of e.f.i. groups is e.f.i.
(6) There exists no @0 -indecomposable reduced p-group of cardinality @0 .
334 10 Torsion Groups

8 Valuated Vector Spaces

Though in a p-group A, the socle AŒp

plays a discernible role, the information about
the group itself provided by its structure is, however, next to nothing: it is a vector
space over Z=pZ, and in vector spaces all subspaces of equal dimensions are alike.
The situation changes drastically, if the elements of the socle are viewed along with
their heights that they acquired from the group. We take this as our inspiration to
embark into the study of vector spaces in which the vectors are equipped with a
‘value.’ In order to understand what such valuated vector spaces can offer, we have
to have a closer look at them.
Valuation of Vector Spaces With applications to reduced p-groups in mind,
we will restrict our discussions to vector spaces over the prime field Z=pZ of
characteristic p. Such vector spaces are easy to handle, since their one-dimensional
subspaces are finite. To simplify discussion, the values will be taken from the class
of ordinals with a symbol 1 adjoined (which is to be viewed larger than any
ordinal); this class will be denoted in this section by .
To set the stage, let V denote a vector space over the prime field Z=pZ of
characteristic p. By a valuation of V is meant a function v of V into  such that, for
all x; y 2 V,
(i) v.x/ D 1 if and only if x D 0;
(ii) v.kx/ D v.x/ if gcdfk; pg D 1;
(iii) v.x C y/  minfv.x/; v.y/g.
Note that (ii) implies that in a one-dimensional vector subspace all non-zero vectors
have the same value. It is a good idea to keep in mind an immediate consequence of
(iii):

v.x C y/ D minfv.x/; v.y/g whenever v.x/ ¤ v.y/:

In fact, this follows from combining (iii) with v.y/  minfv.x/; v.xCy/g if v.x/ >
v.y/.
The pair .VI v/ consisting of a vector space V and its valuation v is called a
valuated vector space. Its support is defined as

supp V D fv.a/ j 0 ¤ a 2 Vg  :

V is homogeneous if supp V is a singleton f g, and is finite-valued if supp V is a

finite set.
A morphism between valuated vector spaces .VI v/ and .WI w/ is a linear
transformation  W V ! W (i.e., a group homomorphism) that does not decrease
values, i.e. w.x/  v.x/ for all x 2 V.
8 Valuated Vector Spaces 335

Category V The valuated vector spaces .VI v/ with these morphisms form
a category V. Vector space isomorphisms that preserve values are the category
isomorphisms; these will be referred to as isometries.
Example 8.1. A one-dimensional valuated vector space .VI v/ is isometric to Z.p/ where every
element ¤ 0 has the same value . If g 2 V is a generator, then for a valuated vector space .WI w/
there is non-zero morphism  W V ! W if and only if there is an x 2 W with  w.x/, in which
case g D x gives rise to a morphism.
Example 8.2. For any reduced p-group A, the socle AŒp
of A becomes a valuated vector space if
the values are the heights measured in A. If  W A ! B is a homomorphism of A into a reduced
p-group B, then AŒp
is mapped into BŒp
such that the map does not decrease heights.

A subspace U of V carries the induced valuation u D v  U, and .UI u/ becomes

a subobject of .VI v/ in the category V. (Usually it is harmless to denote the induced
valuation in U also by the same symbol v.) The quotient space V=U can be equipped
with the valuation

v ? .x C U/ D sup v.x C a/ .x 2 V/
a2U

making it into a valuated vector space .V=UI v ? /, and the natural map V ! V=U
into a V-morphism. It follows that the category V admits kernels and cokernels, so
that we can talk about exact sequences of valuated vector spaces:

0 ! .UI u/ ! .VI v/ ! .WI w/ ! 0:

More explicitly, this means that .UI u/ is a subobject of .VI v/ such that the vector
space V=U with the induced valuation is isometric to .WI w/. Moreover, we have:
Proposition 8.3. The category of -valued Z=pZ-vector spaces V is an additive
category with kernels and cokernels. It admits products and coproducts.
Proof. The first claim is obvious in view of what has been said above. The product
of the valuated
Q vector spaces .Vi I vi / with i 2 I is easily seen to be the cartesian
product Vi ; i.e. it consists of all vectors x? D .: : : ; xi ; : : : / with xi 2 Vi where the
valuation is given viaQv.x? / D infi2I vi .xi /. We shall denote this categorical product
by the usual symbol i Vi . `
It is readily checked that the coproduct Vi of the .Vi I vi / is the vector space
direct sum ˚ Vi with the valuation defined as for the product. t
u
` p-groups, direct decompositions A D ˚i2I Ai are reflected in the socles
Example 8.4. For reduced
as coproducts AŒp
D i2I Ai Œp
, and vice versa.
Example 8.5. Let B D ˚1 nD1 Bn where Bn is a direct sum of cyclic groups of the same
n
`1order p .
For each n, Bn Œp
is a homogeneous vector space with constant value n1, and BŒp
D nD1 Bn Œp
:
Of course, each Bn Œp
is a coproduct of one-dimensional vector spaces.

The following lemma is elementary, but most essential.

Lemma 8.6. Let .AI v/ be a subspace of the valuated vector space .VI v/ such that
there is a 2  satisfying v.a/  for all a 2 A, and v.x/
for all x … A. Then
336 10 Torsion Groups

`
V D A B for some subobject B of V which can be chosen so as to contain any
given subspace C of V disjoint from A.
Proof. Let C be any subspace of V disjoint from A. In view of Zorn’s lemma, there
is a subspace B of V which is maximal with respect to the properties: C
B and
A \ B D 0. Then evidently, V D A ˚ B as vector spaces. From the stated conditions
on the values it follows at once`that v.a C b/ D minfv.a/; v.b/g for all a 2 A and
b 2 B. This means that V D A B, indeed. t
u
This leads to most useful decompositions of .VI v/. For 2  set

V D fa 2 V j v.a/  g and V D fa 2 V j v.a/ > g:

These are fully invariant subspaces in the sense that every endomorphism of .VI v/
(endomorphisms are V-maps!) carries them into themselves. From the preceding
lemma we conclude that there are subspaces A. /
A0 . / of V such that
a a
V D A. / V and V D A0 . / V :
`
Hence there is a subspace B. / of V such that V D B. / V : Obviously, B. / is
a homogeneous subspace isometric to the vector space V =V with constant value
. Its dimension i .V/ D dim V =V is uniquely determined by V, it may be called
the th Ulm-Kaplansky invariant, briefly the th UK-invariant, of V.
Lemma 8.7. Let A be a homogeneous subspace
` of the valuated
` vector space V, and
C a subspace of V such that A C C D A C. Then V D A B for some B  C.
Proof. If 2  is the common value of the`
non-zero elements of A, then we pass
mod V to conclude that A C ` C D A C where stars indicate cosets. By
Lemma 8.6, we have V  D A ` B for some B  C . The complete inverse

image B of B in V satisfies V D A B. t
u
Free Valuated Vector Spaces A valuated vector space F is called free if it is
the coproduct
` of one-dimensional valuated vector spaces. Thus F is of the form
F D i2I Vi where .Vi I vi / are one-dimensional valuated vector spaces, each with
constant value i . If xi 2 Vi is a basis element, then every function f W X ! W from
the basis X D fxi gi2I extends (uniquely) to a category morphism of F into a valuated
vector space .WI w/ if and only if i
w.f .xi // for all i 2 I. This property can be
used as a definition of free valuated vector spaces, it is an exact analogue of what
we have learned about free groups; of course, the stated inequality is a must. Note
that:
(A) Every homogeneous valuated vector space is free. In fact, such a space is just
a vector space with constant valuation.
(B) Two free valuated vector spaces are isometric if and only if, for every ordinal ,
the cardinalities of the free generators of value are the same for both spaces.
8 Valuated Vector Spaces 337

(C) Every valuated vector space is a V-homomorphic image of a free valuated

vector space.
(D) A finite-valued subspace A of a valuated vector space V is a free summand of
V. For, if supp A consists
` of the values 1 <    < k , then by Lemma
` 8.7 we
can write V D A k V1 for some subspace V1 such that A D A k .A \ V1 /
(A k denotes the k -homogeneous summand of A). The conclusion follows by
a straightforward induction.
Proposition 8.8 (R. Brown). Countable dimensional valuated vector spaces are
free.
Proof. In fact, if dim V is countable, then V is the union of an ascending chain
0 D V0 < V1 <    < Vn < : : : of subspaces where vectors in V`
n have n different
values. Therefore, by (D) there are`decompositions VnC1 D Un Vn for suitable
subspaces Un .n < !/. Hence V D n<! Un follows at once. t
u
Example 8.9. The socle of a countable p-group is a free valuated vector space under the height
valuation. This is clear from Proposition 8.8.

From Theorem 2.5 in Chapter 2 and Proposition 8.8 follows at once:

Theorem 8.10. Summands of a free valuated vector space are free. t
u
We do not intend to pursue the exploration of valuated vector spaces, since we
do not wish to make extensive use of them, but we cannot leave the subject without
mentioning a few interesting facts. The valuation defines a non-archimedean metric,
called ultra-metric on the valuated vector space .VI v/. The distance of a; b 2 V
is defined as d.a; b/ D v.a  b/, so it is an ordinal. The basic properties of this
distance function are not the same as for the real-valued distances:
(i) d.a; b/ D 1 is equivalent to a D b;
(ii) d.b; a/ D d.a; b/; and
(iii) d.a; c/  minfd.a; b/; d.b; c/g for all a; b; c 2 V.
The last inequality is a strengthened version of the triangle inequality. A ball
B.a; / with center a 2 V and radius 2  is defined as

B.a; / D fx 2 V j v.x  a/  g D a C V :

Note that b 2 B.a; / implies B.a; / D B.b; /, so every element in the ball can
be viewed as a center. Moreover, if two balls intersect, then one is contained in the
other. The ultra-metric defines a topology where convergence, density, etc. make
good sense.
The most important case is when the values are < !. The reader should observe
the close analogy of the dense free subspace in the next lemma to basic subgroups.
Lemma 8.11. Let V be a valuated vector space with values < !. It contains dense
free valuated subspaces; these are all isometric.
338 10 Torsion Groups

`
Proof. Define B.n/ via V n D B.n/ Vn for each n < !. Thus B.n/ is homo-
geneous,
` so free with values n. We claim that the free valuated vector space
B D n<! B.n/ is dense in V. Indeed, if a 2 V n B, then for each n < ! it has
a decomposition a D b0 C    C bn C cn with bi 2 B.i/ for all i, and cn 2 Vn ; thus,
v.a  b0      bn / > n. The UK-invariants of B are the same as those of V, whence
the claim on isometry becomes evident. t
u
F Notes. The idea of using the elements of the socle equipped with their heights can be
traced back to Charles’ proof of Kulikov’s Theorem 5.1 in Chapter 3. Hill [3] used the metric
defined by the height in the socle. A systematic study of valuated vector spaces was initiated in
the author’s paper [J. Algebra 35, 23–38 (1975)], using a more general setting. Every valuated
vector space embeds in a complete one. We point out that subspaces of free valuated vector spaces
need not be free, but summands are always free. Generalization to valuated groups turns out more
important, it has an extensive literature (for a brief discussion of valuated groups, we refer to Sect. 3
in Chapter 15).

Exercises

(1) If A and B are subspaces of the valuated `vector space V such that a 7! a C B is
an isometry from A to V=B; then V D A B.
(2) Show that B is isometric to
`V=A if A and B are subspaces of the valuated vector
space V such that V D A B.
(3) (a) A valuated vector space with countable support is free if and only if it is the
union of a countable ascending chain of finite-valued subspaces.
(b) A subspace of a free valuated vector space with countable support is free.
(c) Give an example of a countably valued vector space that is not free. [Hint:
rephrase the claims for socles of p-groups.]
(4) For every valuated vector space V with values < !, there exists a separable
p-group A`such that AŒp
with height valuation is isometric to V. `
(5) If V D A B and if C is a subspace of V containing A, then C D A .B \ C/.
(6) Prove the claim that two balls in a valuated vector space can intersect only if
one is contained in the other.
(7) Call a 2 V orthogonal to a subspace U of the valuated vector space .V; v/ (in
notation: a ? U) if v.a C u/
v.a/ for all u 2 U. We say a subspace W of V
is orthogonal to U (and write W ? U) if a ? U holds for all a 2 W. Prove that
(a) W ? U implies U ? W. `
(b) W ? U if and only if W C U D W U.
9 Separable p-Groups That Are Determined by Their Socles 339

9 Separable p-Groups That Are Determined by Their Socles

The results in this section will elucidate the importance of viewing the socle of a
p-group as a vector space furnished with the height-valuation. In this section, the
socles will be assumed to carry this valuation.
It is an immediate question as to what extent a p-group is determined by its
socle as a valuated vector space. Further questions include characterization of the
socles of p-groups in some well-defined class C or cases when the isometry of socles
implies the isomorphy of the p-groups within the class C. We deal briefly with these
questions. We should probably stress at the outset that the valuated vector space
point of view has many more applications than those treated here.
Subgroups Supported by Dense Subsocles If we are to talk about these
questions, we should probably start with showing the opposite situation: it can
very well happen that as many non-isomorphic separable p-groups as possible have
isometric socles. Before stating the relevant theorem, we prove a lemma.
Lemma 9.1 (Hill–Megibben [3]). Let A be a reduced p-group. A proper dense
subsocle S of A supports 2 different pure and dense subgroups, where  D
jpAŒp
\ Sj.
Proof. Let fci gi2I denote a basis of the vector space pAŒp
\ S, and choose ai 2
A .i 2 I/ such that pai D ci . By hypothesis, there is a u 2 AŒp
n S. For a subset
J of I, let XJ be the set of the elements gi where gi D ai if i 2 J and gi D ai C u
otherwise. S is still the socle of the subgroup generated by XJ , so by Theorem 1.3
for each J, XJ is contained in a pure and dense subgroup CJ of A, supported by S.
If K is a different subset of I, then CJ ¤ CK , since no CJ may contain both ai and
ai C u. As jIj D , we have 2 different subgroups CJ . t
u
Regarding the cardinal hypothesis in the following result, let us note that there
exist arbitrarily large cardinals  satisfying  @0 D 2 . (See the comment following
Proposition 7.10.)
Theorem 9.2 (Hill–Megibben [3]). Let  be a cardinal satisfying  @0 D 2 , and
A a separable p-group of cardinality 2 whose basic subgroups have rank and final

rank equal to . Then a proper dense subsocle S of cardinality 2 supports 22 non-
isomorphic pure .and dense/ subgroups of A.
Proof. Let B be the torsion-complete group whose basic subgroup B has rank and
final rank . By Theorem 3.2, we have jBj D 2 . Applying Lemma 9.1 to the case

 D 2 , we conclude that there are 22 different pure subgroups supported by a
proper dense subsocle S. Owing to Theorem 3.10, isomorphic dense subgroups are
carried into each other by an automorphism of B. We complete the proof by showing

that B has less than 22 automorphisms.
Each automorphism of B is entirely determined by its action on B. As any
generator of B is mapped by an automorphism upon one of  different elements,

the cardinality of Aut B is at most  D .2 / D 2 < 22 . t
u
340 10 Torsion Groups

When Socles Determine the Group We now turn to the discussion of positive
results. Our intention is to show that the †-cyclic p-groups, the direct sums
of countable p-groups, and the direct sums of torsion-complete p-groups are
determined by their socles as valuated vector spaces, within the respective classes
of groups.
Theorem 9.3.
(i) A p-group is †-cyclic if and only if its socle is a free valuated vector space with
values < !.
(ii) Groups with isometric free socles of values < ! are isomorphic.
(iii) Every free valuated vector space over Z=pZ with values < ! is isometric to
the socle of a †-cyclic p-group.
Proof. (i) If A is †-cyclic, then AŒp
is the coproduct of the socles of the cyclic
summands. These are one-dimensional spaces, so AŒp
is free as a valuated vector
space. Conversely, a free valuated vector space is the coproduct of one-dimensional
spaces each of which supports a cyclic group which is of order pnC1 if n is the
common value of the non-zero vectors. Hence also (ii) is evident.
(iii) is straightforward. t
u
For the class of countable p-groups, the analogues of (i)–(ii) hold (see Theo-
rem 1.6 in Chapter 11, while the analogue of (iii) holds true only under additional
conditions on the invariants; see Exercise 2.
Theorem 9.4.
(i) If A is a direct sum of countable p-groups, then its socle AŒp
is a free valuated
vector space with values < !1 .
(ii) Direct sums of countable p-groups are isomorphic if and only if their socles
are isometric.
Proof. (i) It suffices to prove the claim for countable p-groups A. Then dim AŒp
is
countable, so Proposition 8.8 implies that AŒp
is free as a valuated vector space.
(ii) This is equivalent to Kolettis’ theorem (Corollary 6.7 in Chapter 11). t
u
As expected, also the torsion-complete groups are determined by their socles.
However, it is not so obvious that the same holds for their arbitrarily large direct
sums.
Theorem 9.5 (Hill [5]). Two direct sums of torsion-complete p-groups are isomor-
phic if and only if they have isometric socles.
Proof. To begin with, we show that two torsion-complete p-groups with isometric
socles are isomorphic. The ‘basic’ subspaces (cp. Lemma 8.11) in the socles are
isometric, so the basic subgroups of the torsion-complete groups are isomorphic,
and hence the groups themselves are isomorphic.
9 Separable p-Groups That Are Determined by Their Socles 341

It suffices to prove the ‘if’ part of the statement. Our first step is to show that two
separable p-groups A; C with isometric socles can be embedded isomorphically as
pure subgroups in a torsion-complete p-group such that they share the same socle.
An isometry  W AŒp
! CŒp
carries the socle A0 Œp
of a basic subgroup A0 of A onto
the socle C0 Œp
of a basic subgroup C0 of C. As  is height-preserving, it extends to
an isomorphism  0 W A0 ! C0 which in turn extends to an isomorphism N W A ! C
between the torsion-completions of A and C. Then .A/ N and C have the same socle
in C.
Thus it is harmless to assume that the p-groups A D ˚i2I Ai and C D ˚j2J Cj
(where Ai ; Cj are torsion-complete) are pure subgroups with the same socle in a
torsion-complete group B. Under this hypothesis, we prove that for any summand
Ai , C admits a summand Di such that Di Œp
D Ai Œp
: From Lemma 4.2 we argue
that there is an integer m such that pm Ai Œp
is contained in a finite direct sum
Cj1 ˚    ˚ Cjk . This direct sum is torsion-complete, so by Lemma 5.4 it contains a
pure subgroup U supported by pm Ai Œp
. If we write Ai Œp
D X ˚ pm Ai Œp
, then it is
clear that X (of bounded heights) supports a summand V of C, thus there is a pure
subgroup Di D U ˚ V
C as stated. Now Ai , as a pure torsion-complete subgroup,
is a summand of B. A pure subgroup supported by the support of a summand is itself
a summand (Lemma 1.2), so Di is a summand of B, and hence of C. It is evidently
Š Ai .
Clearly, the subgroups Di generate their direct sum in C such that D D ˚i2I Di
has the same socle as AŒp
D CŒp
: The proof will be complete if we can show that
D is pure in C. Let x 2 .Di1 ˚    ˚ Dik /Œp
be of height h in C. It has the same
height in A, so all of its coordinates in the Ai have height  h. x has exactly the
same coordinates in the Di with the same heights. Thus the elements in DŒp
cannot
have smaller heights in D than in C, and so D is pure in C. t
u
Other classes of p-groups in which only isomorphic groups can have isometric
socles are the class of simply presented and the class of p!C1 -projective p-groups.
These groups will be discussed later in Sects. 3 and 10 in Chapter 11, respectively.
F Notes. By making use of the Diamond Principle, both Cutler [2], Dugas–Vergohsen [1]
proved that the only separable p-groups of cardinality  @1 that are determined by their socles are
the †-cyclic and the torsion-complete groups. Mekler–Shelah [5] established the independence
of this question in ZFC. The Keef class Kp of p-groups is the smallest class that contains the
cyclic p-groups, and is closed under taking summands, direct sums, and torsion subgroups of non-
measurable direct products. Keef [4] shows that the groups in Kp are characterized by their socles
within this class (generalizing Theorem 9.5).
As we have already mentioned above, subspaces of free valuated vector spaces need not be
free. The valuated socles of totally projective p-groups are never free once the lengths of the groups
exceed !1 , but are always embeddable in free valuated vector spaces. Valuated vector spaces that
support totally projective p-groups were characterized in the author’s paper [Symposia Math. 23,
47–62 (1979)].
There are interesting results on direct sums of torsion-complete groups. One of those is that
summands are again such direct sums. Another theorem states that any two decompositions have
isomorphic refinements. See Irwin–Richman–Walker [1], Hill [4], Enochs [1]. Ivanov [2] has an
extensive study of countable direct sums of algebraically compact and torsion-complete groups,
including cancellation properties, solutions to Kaplansky’s test problems, etc.
342 10 Torsion Groups

Hill [1] gave the first example of non-isomorphic pure subgroups supported by the same socle.
In his paper [18], he examines to what extent the socle can determine the group. Cutler [4] raises the
question as to the existence of non-isomorphic pure and dense subgroups with the same pn -socle
in a separable p-group, and shows that it is undecidable in ZFC.

Exercises

(1) (Hill) Show that the decompositions ˚i Ai D ˚j Cj where Ai ; Cj are torsion-

complete p-groups have common refinement. [Hint: argue with the exchange
property.]
(2) Find necessary and sufficient conditions on the UK-invariants of a free valuated
vector space V (with values < !1 ) to exist a countable p-group with socle
isometric to V. [Hint: Theorem 1.9 in Chapter 11.]

Problems to Chapter 10

PROBLEM 10.1. Let fC j  < g be a smooth increasing chain of p-groups

such that C is pure in C C1 , and C C1 =C is a thin group for all  < . Is the
union C D [ < C a thin group?
PROBLEM 10.2. Let fC j  < g be a smooth increasing chain of p-groups
such that C is pure in C C1 , and C C1 =C is a thick group for all  < . Is the
union C D [ < C a thick group?
PROBLEM 10.3. Call the p-groups A; C almost disjoint if only the †-cyclic
groups can have isomorphic copies in both of them. Is there any bound for the
cardinality of sets of non-isomorphic pairwise almost disjoint p-groups?
This problem for torsion-free groups was discussed by Eklof–Mekler–Shelah [1].

PROBLEM 10.4. Characterize the socle AŒp

as a valuated vector space.
Much has been done by Dugas–Vergohsen [1] under the hypothesis V = L.

PROBLEM 10.5. When do the “slices” AŒpnC1

=AŒpn
as valuated vector spaces
determine a separable p-group A?
Chapter 11
p-Groups with Elements of Infinite Height

Abstract We continue our study of torsion groups concentrating on p-groups (with

unspecified prime p) in the general case when the groups contain elements of infinite height.
Matters are more subtle here as one has to deal with transfinite heights that are the central concept
both in the search for invariants and in the proofs.
The focus of the structure theory is on p-groups that can be described by their UK-invariants.
Accordingly, this chapter is primarily devoted to countable p-groups and their generalizations: the
totally projective p-groups. The theory is perhaps the most interesting and highly satisfactory clas-
sification of a fairly large class of p-groups in terms of well-ordered sequences of cardinal numbers
(provided by their UK-invariants). The four main approaches to the theory of totally projective p-
groups (simple presentation, total projectivity, nice systems, and balanced-projectivity) underline
the extreme importance of these groups; this theory is unparalleled in beauty and richness in abelian
group theory.
Once the equivalence of the four main characterizations is established, there remain still some
intriguing questions to be answered. For instance, which well-ordered sequences of cardinals may
be the UK-invariants of a totally projective p-group? or, which is the largest class of p-groups that
includes the generalized Prüfer groups, is closed under direct sums and summands, and whose
members are distinguishable via their UK-invariants?
Needless to say, there have been various attempts to extend the well-rounded theory of totally
projective p-groups, and various generalizations have been considered in the literature. So far these
theories have produced only less remarkable results. Though several innovative techniques have
been discovered, it seems that so far they have fallen short of true significance.
The final sections of this chapter deal with questions that are spin-offs of the theory of totally
projective p-groups, and offer a glimpse into classes depending on ordinal numbers.

1 The Ulm-Zippin Theory

The main objective of this section is to present the celebrated Ulm-Zippin theory of
countable p-groups. As before, p denotes an arbitrary, but fixed prime number, and
for an element a 2 A, hA .a/ (or h.a/) will mean its transfinite height at p.
Hill Invariants Let A be a p-group, and G a subgroup of A. An element a 2 AnG
is said to be proper with respect to G if the heights satisfy

h.a/  h.x/ for all x 2 a C G:

© Springer International Publishing Switzerland 2015 343

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_11
344 11 p-Groups with Elements of Infinite Height

In other words, a has the maximal height in its coset mod G. Thus an element a 2 A
of height  is proper with respect to G if and only if a … p C1 A C G. In this case,

hA .a/ D hA=G .a C G/:

Of course such an element a need not exist in a coset, though it can be shown easily
that if the height of a coset a C G in A=G is a non-limit ordinal, then the coset does
contain an element of this height. Note that if a is proper with respect to G, then we
have

h.a C g/ D minfh.a/; h.g/g for all g 2 G:

Evidently, if G is finite, then every coset mod G contains elements proper with
respect to G. The same holds if G D p A for an ordinal .
Let now A be a reduced p-group, and G a subgroup of A. For an ordinal , we
introduce the notation

G./ D .p C1 A C G/ \ p AŒp

I

this is a subgroup between p AŒp

and p C1 AŒp
. From our remark above it follows
that an element a 2 A of order p and of height  belongs to G./ if and only if it is
not proper with respect to G. Consequently, representatives of the non-zero cosets
in the factor group

f .A; G/ D p AŒp
=G./

are exactly the elements of A that are of order p, of height , and proper with respect
to G. This Z=pZ-vector space is called the th UK-invariant of A relative to G
or the th Hill invariant of A relative to G. Sometimes we understand by this
invariant the dimension of the vector space. We can allow this ambiguity, since from
the context it will always be clear what we mean: vector space or its dimension.
Evidently, f .A; G/
f .A/ and f .A; 0/ D f .A/: Here f .A/ D p AŒp
=p C1 AŒp

is the Ulm-Kaplansky- or UK-invariant of A.

Example 1.1. Let A D Z.pnC1 / and G D pA. Then G.k/ D pk AŒp
D AŒp
for k D 0; : : : ; n, and
G.k/ D 0 for all k > n. Thus the Hill invariants f .A; G/ vanish for all  .
Example 1.2. Let A be a reduced p-group, and B its basic subgroup. Then B.n/ D .pnC1 A C
B/ \ pn AŒp
D A \ pn AŒp
D pn AŒp
, thus fn .A; B/ D 0 for all n < !. If 
!, then B. / D
.pC1 A C B/ \ p AŒp
D pC1 AŒp
, thus f .A; B/ D .pC1 AŒp
/=.p AŒp
/ D f .A/:
Example 1.3. Let A be a reduced p-group, and G D p A for some ordinal . Then G. / D
.pC1 A C p A/ \ p AŒp
D pC1 AŒp
or p AŒp
according as  <  or 
. Thus f .A; p A/
equals f .A/ if  <  and is 0 otherwise.

In order to illustrate the role of Hill invariants, we prove a simple lemma.

1 The Ulm-Zippin Theory 345

Lemma 1.4. Assume A; C are p-groups, G is a subgroup of A, and H is a subgroup

of C such that f .A; G/ D f .C; H/ for some . There exists an a 2 AŒp
of height 
and proper with respect to G if and only if there is a c 2 CŒp
of height  and proper
with respect to H.
Proof. Observe that x 2 G./ means that x 2 AŒp
is of height , and not proper
with respect to G, since—we repeat—x 2 p C1 A C G is another way of saying that
the coset x C G has a representative in p C1 A. t
u
Let A and C be reduced p-groups, and G
A; H
C subgroups. An
isomorphism  W G ! H is called height-preserving if

hC .g/ D hA .g/ for every g 2 G:

It is important to keep in mind that heights are always computed in the large groups.
Manifestly, the restriction of an isomorphism of groups A ! C to subgroups
G ! H is always height-preserving.
Kaplansky–Mackey Lemma The next lemma is crucial in extending isomor-
phisms between subgroups to larger subgroups. It is the most essential ingredient
in the proof of Ulm’s theorem below. The groups A; C in Lemma 1.5 need not be
p-groups, but the Hill invariants are computed for the chosen prime p.
Lemma 1.5 (Kaplansky–Mackey [1]). Let A and C be reduced groups, G a
subgroup of A, and H a subgroup of C such that A=G and C=H are p-groups.
Suppose  W G ! H is a p-height-preserving isomorphism; furthermore, for all
, f .A; G/
f .C; H/ and

˛ W p AŒp
=G./ ! p CŒp
=H./ (11.1)

are monomorphisms. If a 2 A is proper with respect to G and pa 2 G, then  can

be extended to a height-preserving isomorphism

  W hG; ai ! hH; ci

for a suitable c 2 C such that

˛ .hG; ai./=G.// D hH; ci./=H./ for all :

Proof. Assume that a with h.a/ D  has been chosen in its coset mod G, in addition
to being proper with respect to G, to satisfy also h.pa/ >  C 1 whenever possible.
We distinguish two cases according as such a choice is possible or not.
Case I : h.pa/ >  C 1. In this case pa D pb for some b 2 A of height   C 1.
Then h.a  b/ D , and a  b is proper with respect to G, for otherwise there
would exist a g 2 G with h.a  b C g/ > , leading to the contradiction
h.a C g/ > . By an earlier remark, a  b … G./, thus there is a u 2 p CŒp

346 11 p-Groups with Elements of Infinite Height

such that ˛ maps a  b C G./ to u C H./. This u is evidently proper with

respect to H. Since  is height-preserving, there is a d 2 C of height   C 1
such that pd D .pa/ 2 H. Now c D d C u is proper with respect to H and
satisfies h.c/ D ; pc D .pa/ 2 H.
Case II : h.pa/ D  C 1: Pick a c 2 C of height  such that .pa/ D pc: To
see that c 2 H is contradictory, assume that .g/ D c for some g 2 G. Then
pg D pa, and p.a  g/ D 0. As a was proper with respect to G, we obtain
h.a  g/ D minfh.a/; h.g/g D , showing that a0 D a  g is likewise proper
with respect to G. But h.pa0 / D 1 >  C 1 contradicts the choice of a. Thus
c … H, and we now show that it is proper with respect to H. If we had an
x 2 H with h.c C x/ > , then necessarily h.x/ D , whence we conclude that
 < h.c C x/ < h.pc C px/ D h.pa C p 1 x/ implies a0 D a C  1 x satisfies
h.a0 / D minfh.a/; h.x/g D  and h.pa0 / >  C 1, contrary to the choice of a.
Thus c … H is proper with respect to H.
In both cases, with the chosen c 2 C, we will extend  to   by letting c
correspond to a. That we get an isomorphism   W hG; ai ! hH; ci is clear from
the selection of c. To show that it is height-preserving, observe that, for all g 2 G,

h.a C g/ D minfh.a/; h.g/g D minfh.c/; h.g/g D h.c C g/:

To verify the claim on the action of   , take into consideration that if  < , then
h.a/   C 1 implies hG; ai./ > G./, while if  > , then the same inequality is
the consequence of a being proper with respect to G. Similar inequalities hold for H
and c. If  D , then in Case I, hG; ai./ D ha  bi ˚ G./ and hH; ci./ D hui ˚
H./, and the choice of u guarantees that ˛ induces an isomorphism as desired. In
Case II, no g 2 G and a0 2 A with h.a0 /   C 1 may satisfy p.a  g C a0 / D 0, so
that again hG; ai./ D G./, and similarly hH; ci./ D H./. t
u
Whenever the monomorphisms ˛ are not preassigned, in Case I any u 2
p CŒp
=H./ may be chosen.
A corollary to the preceding proof is that the th Hill invariants of A relative
to G and relative to hG; ai are the same except possibly when  D  in which case
f .A; hG; ai/C1 D f .A; G/ may hold. But certainly we have (in the notation above)

f .A; hG; ai/ D f .C; hH; ci/:

Ulm’s Theorem Equipped with this lemma, we are prepared for the proof of the
main result.
Theorem 1.6 (Ulm [1]). For the countable reduced p-groups, A and C, the follow-
ing conditions are equivalent:
(i) A and C have the same Ulm length , and for each  < , the Ulm factors A
and C are isomorphic;
(ii) A and C have the same UK-invariants: f .A/ D f .C/ for all ;
(iii) A and C are isomorphic.
1 The Ulm-Zippin Theory 347

Proof. The implication (i) )(ii) is a simple consequence of the fact that the
number of cyclic summands of order pn in the decomposition of A is equal to the
.! C n  1/st UK-invariant of A, while (iii) ) (i) is trivial. It remains to show that
(ii) ) (iii).
Assuming (ii), for each  we fix an isomorphism ˛ W p AŒp
=p C1 AŒp
!
p CŒp
=p C1 CŒp
. The elements of A; C can be arranged in countable sequences:


A D fa0 ; : : : ; an ; : : :g; C D fc0 ; : : : ; cn ; : : :g; we may even assume panC1 2

ha0 ; : : : ; an i, pcnC1 2 hc0 ; : : : ; cn i for all n. Suppose we have constructed chains
of finite subgroups 0 D G0
G1
  
Gn of A and 0 D H0
H1
  
Hn of
C, as well as height-preserving isomorphisms i W Gi ! Hi such that n  Gi D i
for i D 0; 1; : : : ; n. We may assume that ˛ induces isomorphisms p AŒp
=Gi ./ !
p CŒp
=Hi ./ for i
n. If n is odd, select the first ak in the sequence that does
not belong to Gn . We wish to extend n to a height-preserving isomorphism nC1
of GnC1 D hGn ; ak i with a subgroup HnC1 of C, containing Hn . This can be done
by using Lemma 1.5. If n is even, then we select the first c` 2 C with c` … Hn ,
and extend n1 to an isomorphism of HnC1 D hHn ; c` i with a subgroup GnC1 in A.
After each step, (11.1) will be satisfied, as is clear from Lemma 1.5. It is evident
that  D [n<! n will establish the isomorphy of A and C. t
u
Before moving on to the existence question, we record an important corollary
to the preceding theorem. Though it will follow from the more general Lemma 4.5
below, we give a short proof.
Corollary 1.7 (Zippin [1]). If A and C are countable p-groups with the same UK-
invariants, then every isomorphism between p A and p C .for any fixed ordinal /
extends to an isomorphism A ! C.
Proof. The proof is essentially the same as for the preceding theorem, just we have
to start with a prescribed isomorphism W p A ! p C. The only UK-invariants
needed in the proof are those of indices  < . As cosets of finite extensions of
p A always contain elements proper with respect to them, a repetition of the proof
in Theorem 1.6 establishes our claim. t
u
Example 1.8. The Prüfer group H!C1 defined in Sect. 1 in Chapter 10 has the UK-invariants
f .H!C1 / D 1 for  D 0; 1; : : : ; !: All countable p-groups with the same sequence of invariants
are isomorphic to H!C1 .

Zippin’s Theorem Once we have a classification theorem, we also need a result

saying what values of the invariants can occur. In order to find out what sequences
f0 .A/; : : : ; f .A/; : : : of cardinals may serve as UK-invariants in Ulm’s theorem, we
turn our attention to Zippin’s main theorem. (For generalization, see Theorem 3.7.)
Theorem 1.9 (Zippin [1]). There exists a countable reduced p-group A of length 
and with UK-invariants n0 ; : : : ; ni ; : : : ; n! ; : : : ; n ; : : : . < / if and only if
(a)  D !˛ C k is a countable ordinal where k  0 is an integer;
(b) each n is a non-negative integer or @0 such that for each  < , infinitely many
of the cardinals n ; : : : ; nCi ; : : : .i < !/ are non-zero.
348 11 p-Groups with Elements of Infinite Height

Proof. The necessity of the conditions is immediate, the sole non-trivial argument
relies on the fact that only the last Ulm factor can be bounded.
For sufficiency, we first develop a sequence A .
˛/ of †-cyclic p-groups
whose UK-invariants are n!Cj for j < ! (for j
k when  D ˛). By (b), the A are
all unbounded with the possible exception of the last one if such exists. The theorem
will be proved if we succeed in defining a p-group whose Ulm factors are A . We
use transfinite induction on ˛. If ˛ D 0, or if ˛ D 1; k D 0, then the sequence has
only one Ulm factor, and A D A0 is as desired. Hence we assume that either ˛ D 1
and k > 0, or ˛  2, and the existence has been established for shorter sequences.
The proof distinguishes several cases.
Case I. ˛  1 exists and A˛ is a cyclic group, say, of order p` . Let C denote a
countable p-group with Ulm sequence A . < ˛  1/; A0˛1 where A0˛1 D
˚i<! hci i with o.ci / D p`C`i provided A˛1 D ˚i<! hbi i with o.bi / D p`i .
Define A as the quotient of C modulo the subgroup generated by all p`i ci 
p`j cj .i; j < !/. If we denote by a the coset of p`i ci , then it is clear that A=hai is
a countable p-group with Ulm sequence A .
˛  1/. Since the exponents
`i are unbounded, a must have a height not smaller than the heights of the ci ,
so it belongs to the ˛th Ulm subgroup of A, which is hai. That hai has order p`
can be verified by taking a copy of Z.p1 /, mapping each ci upon an element
of the same order, watching that all the p`i ci should have the same image, and
then extending this map to all of C stepwise starting with A˛1 . It only remains
to point out that such a map factors through A.
Case II. ˛  1 exists and A˛ is a †-cyclic group. We reduce the construction to
Case I. We decompose each A . < ˛  1/ into a direct sum of a countable
number of unbounded summands, and for each cyclic summand of A˛ we apply
the above construction with one of the summands. Finally, we take the direct
sum of the groups constructed.
Case III. ˛ is a limit ordinal and k D 0. We decompose each A . < ˛/ into a
countable direct sum of unbounded groups: A D ˚<˛ G . We appeal to
the induction hypothesis to conclude the existence of countable p-groups G of
length  . < ˛/ with Ulm sequence G0 ; G1 ; : : : ; G . Then A D ˚<˛ G
is a countable p-group with the prescribed Ulm sequence.
Case IV. ˛ is a limit ordinal and A˛ is a cyclic group of order p` . We decompose
each A D hb i ˚ A0 where the cyclic summands hb i are selected such that
for each  < ˛ and for each m 2 N, there is a  < ˛ with o.b / D p` > pm .
By induction hypothesis, there is a countable p-group C with Ulm factors C D
hc i ˚ A0 . < ˛/ where o.c / D p`C` . Define A as the factor group of C
modulo the subgroup generated by all p` c  p` cm .;  < ˛/. In the same
way as in Case I, it follows that A will have the prescribed Ulm sequence.
Case V. ˛ is a limit ordinal and A˛ ¤ 0 is a †-cyclic group. This case can be
reduced to Case IV by imitating the method used in Case II. t
u
Example 1.10. Let G0 ; G1 be unbounded countable †-cyclic p-groups, and write G0 D ˚i<! hai i,
G1 D ˚i<! hgi i. We define a group A whose Ulm factors are isomorphic to the given groups as
follows. Let B0 D ˚i<! hbi i be a basic subgroup of G0 such that G0 =B0 is a countable direct sum:
G0 =B0 D ˚i<! Ci with Ci Š Z.p1 /. Let ci 2 G0 be a representative of a generator of the socle
1 The Ulm-Zippin Theory 349

of Ci . The group A will be defined as the group generated by fai ; gi gi<! subject to the existing
relations for the bi , elements of Ci , and for the gi (e.g., of the form pm gi D 0), to the relations
between the ci and the bi , and in addition to pci D gi for all i < !. Then the elements gi will have
infinite heights in A. To complete the argument, we still need to ascertain that the new relations do
not force any collapse in A. We need not worry about the elements generated by the ci , because
none of them becomes 0, not even modulo the subgroup generated by the gi . If a 2 A contains
generators gi , then define a map  W A ! Z.p1 / by mapping one of these generators to a non-zero
element, the rest of the gi to 0, and extend this map to all of A. Then .a/ ¤ 0, so a ¤ 0.

Theorems 5.3 in Chapter 3, 1.6, 1.9 of Prüfer, Ulm, and Zippin yield a complete
classification of countable reduced p-groups. We regard this theory as the greatest
achievement in abelian group theory in the first half of the twentieth century.
A few comments may be inserted here about this theory. If A is a countable
reduced p-group of length  D ˛! C k, then we can arrange its UK-invariants in a
convenient matrix form (to which we will refer as the UK-matrix of A):
0 1
f0 .A/ f1 .A/ : : : fn .A/ : : :
B f .A/ f C
B ! !C1 .A/ : : : f!Cn .A/ : : :C
B C
U.A/ D B : : : : : : : : : : : : : : :C D kf .A/k
B C
@f! .A/ f!C1 .A/ : : : f!Cn .A/ : : :A
::: ::: ::: ::: :::

where the rows are indexed by ordinals  < ˛ (or 

˛ if k > 0) and the columns
by non-negative integers. The cardinal numbers in the th row are the invariants of
the th Ulm factor of A. The matrix U.A/ satisfies:
(i) it is an ˛  !-matrix with a countable ordinal ˛;
(ii) the entries are non-negative integers or @0 ;
(iii) every row (with the possible exception of the very last one) contains infinitely
many non-zero entries.
Ulm’s theorem can be interpreted as saying that two reduced countable p-groups
are isomorphic, A Š C, if and only if the corresponding matrices are equal: U.A/ D
U.C/, while Zippin’s theorem says that every matrix with properties (i)–(iii) is the
UK-matrix of some reduced countable p-group.
Corollaries We conclude this section with two corollaries.
Proposition 1.11. A countably infinite reduced p-group decomposes into the direct
sum of infinitely many non-trivial groups.
Proof. Let A be a group as stated, and let U.A/ be its UK-matrix. It is an easy
exercise in cardinal arithmetics to decompose U.A/ into the sum of countably many
matrices satisfying (i)–(iii). By Zippin’s theorem, each of these matrices defines a
group An .n < !/, and Ulm’s theorem assures that A Š ˚n<! An . t
u
Proposition 1.12 (Baer [1]). A countable reduced p-group A has the property that
any two of its direct decompositions have isomorphic refinements if and only if
A1 D 0, i.e. A is †-cyclic.
350 11 p-Groups with Elements of Infinite Height

Proof. If A1 D 0, i.e. if A is a direct sum of cyclic p-groups, then evidently, its

direct decompositions have isomorphic refinements: we decompose each summand
into the direct sum of cyclic p-groups.
Conversely, assume that A1 ¤ 0. We decompose the UK-matrix U.A/ into
the sum of two matrices U.A/ D M1 C M2 such that they satisfy conditions
(i)–(iii) and have never non-zero entries at the same location. Next, define matrices
M01 ; M02 by switching the first rows of M1 and M2 , while keeping the rest of the
rows intact. There are countable reduced p-groups A1 ; A2 ; C1 ; C2 with UK-matrices
M1 ; M1 ; M01 ; M02 , respectively, and by Ulm’s theorem we have A Š A1 ˚ A2 Š
C1 ˚ C2 . However, no summand of Ai that contains elements ¤ 0 of infinite height
can be isomorphic to such a summand of Ci for i D 1; 2. t
u
It is no longer true for p-groups of higher cardinalities that they are isomorphic
whenever their corresponding Ulm factors are isomorphic.
Example 1.13 (Kulikov [2]). Let G0 D B be the torsion-complete p-group with the standard basic
subgroup B D ˚1 nD1 hbn i .o.bn / D p /, and G1 the direct sum of 2
n @0
cyclic groups hgi i .i 2 I/ of
order p. Two groups, A and C, will be defined with G0 ; G1 as Ulm sequence such that jAŒp
=A1 j D
@0 and jCŒp
=C1 j D 2@0 . Let G0 =B D ˚i2I Di .Di Š Z.p1 //, and select a canonical generating
0
set din 2 Di of order pn .n 2 N/ for each i. Now A is generated by all of b0n ; din ; g0i subject to the
0
same relations as the corresponding bn ; din ; gi except that pdi1 D 0 is replaced by pdi1 D g0i for
every i 2 I. Furthermore, for C, we write I D I1 [ I2 with disjoint subsets, both of cardinality
2@0 , and keep the same generating set and relations including pdi1 D 0 but only for i 2 I1 , and put
pdi1 D g0i for i 2 I2 . The groups A and C cannot be isomorphic, because jAŒp
=A1 j D jBŒp
j D @0 ,
while jCŒp
=C1 j
jI1 j D 2@0 .
Example 1.14 (Richman [3]). Let B and C be torsion-complete groups with basic subgroups B D
˚n2N Bn (Bn Š Z.pn /) and C D ˚n2N Cn (Cn Š Z.p2n /), respectively. Then G D B=BŒp
and
H D C ˚ C=CŒp
are extensions of an elementary p-group by a torsion-complete group. They
have the same UK-invariants: the nth invariant is 1 for all n < !, and the !th invariant is 2@0 .
However, G and H are not isomorphic: GŒp
=G1 is countable, while HŒp
=H 1 has the power of the
continuum.

For larger cardinalities, we have the following theorem which can be derived
from Theorem 1.6 with only set-theoretical argument. It will be restated as a
Theorem 6.7, so we skip its proof now.
Theorem 1.15 (Kolettis [1]). If the p-groups A and C are direct sums of countable
groups, then they are isomorphic if and only if their corresponding UK-invariants
are equal. t
u
F Notes. Prüfer’s group H!C1 (defined in 1923) was the first example for a reduced p-group
with non-zero elements of infinite height. A decade later, in 1933 Ulm [1] used ingenious matrix-
theoretical arguments to prove that the Ulm factors of a countable reduced p-group determine the
group up to isomorphism. Ulm translated the problem from p-groups into infinite-dimensional
linear algebra, and applied (his thesis supervisor) Toeplitz’s theory to obtain a kind of normal
form. Subsequently in 1935, Zippin [1] published a group-theoretical proof, and at the same
time established the important existence Theorem 1.9. Surprisingly, no fruit has been born out
of this remarkable theory for a quarter of a century, and even applications were very scarce. In
1960 the situation changed with Kolettis [1] where Ulm’s theory was extended to direct sums of
countable p-groups. The elegant proof by Kaplansky–Mackey [1] opened the door for substantial
generalizations of Ulm’s theorem. Hill [2] and Richman–Walker [2] show that Kolettis’ theorem
can be derived from Ulm’s just by pure set-theoretical argument.
1 The Ulm-Zippin Theory 351

There are numerous results in the literature involving countable p-groups, and their direct sums.
We will see some later on. In these cases, luckily, the question of isomorphy boils down to the
equality of the UK-invariants.
Rogers [1] notes that—ignoring addition—a p-group A becomes a tree under multiplication by
p: the nodes are the elements of A, and two nodes a; b 2 A are connected by an oriented edge if
a D pb. She defines UK-invariants, and derives Ulm’s theorem. This idea goes back to Crawley–
Hales [1, 2], and is explored more fully in the paper Hunter–Richman–Walker [3].
Crawley [3] points out that p-groups with a finite number of Ulm factors are determined
by the UK-invariants even if they are uncountable, but still †-cyclic (this is a consequence of
Theorem 12.2). However, Lemma 12.3 will show that this is no longer true if the Ulm factors
are not †-cyclic. In another paper [2] he shows that a reduced countable torsion group has the
cancellation (substitution) property if and only if all of its UK-invariants are finite for all primes.
In the same vein, Göbel–May [1] describe when a direct sum of countable p-groups is cancellable.
Szmielew [1] initiated a model-theoretical study of abelian groups. Barwise–Eklof [1] con-
sidered the classification problem of p-groups in the infinitary language L1! , and show that
the (modified) UK-invariants determine an arbitrary p-group up to L1! -equivalence. This is a
genuine generalization of Ulm’s theorem, since for countable groups, L1! -equivalence implies
isomorphism.

Exercises

(1) If G
H are subgroups of a p-group A, then f .A; G/  f .A; H/ for all .
(2) Let A D B ˚ C be a p-group. Then f .A; B/ D f .C/ for each ordinal .
(3) If  is a countable ordinal and a 2 p A, then a embeds in a countable subgroup
C of A such that a 2 p C.
(4) (a) For each countable ordinal , the set of non-isomorphic countable p-
groups of length  has the power of the continuum.
(b) Is the same true for the set of all non-isomorphic countable p-groups?
(5) Let A; C be countable p-groups. State a necessary and sufficient condition
on their UK-invariants for A having (a) a subgroup isomorphic to C; (b) a
summand isomorphic to C.
(6) (a) For every countable ordinal , there exists a countable reduced p-group
M./ of length  such that every countable reduced p-group of length

is isomorphic to a summand of M./.
(b) There is a reduced p-group M of length !1 and of cardinality @1 such that
every countable reduced p-group is isomorphic to a summand of M.
(7) (Kaplansky) (a) If A is a countable p-group, and if C satisfies A ˚ A Š C ˚ C;
then A Š C.
(b) For a countable p-group A with finite UK-invariants, A˚B Š A˚C implies
B Š C, where B; C are countable p-groups.
(c) The claim in (b) may fail if A has infinite UK-invariants.
(8) Find precise conditions, in terms of the UK-invariants, for a countable reduced
p-group A to satisfy A ˚ A Š A. Then also A.@0/ Š A.
(9) (Irwin–Walker) Let H be a p A-high subgroup of the p-group A, for some
ordinal . Show that f .H/ D f .A/ for  < .
352 11 p-Groups with Elements of Infinite Height

(10) Let A be a reduced countable p-group, and C a summand of A. Suppose that

p A D p C ˚ X holds for some ordinal  and a subgroup X. Then A D C ˚ B
for some B with p B D X.
(11) Let A be a p-group of length   !, and C Š Z.pn / a cyclic group. There exists
a group G such that p G Š C and G=p G Š A. [Hint: distinguish according
as the last Ulm factor of A is bounded or not.]

2 Nice Subgroups

The insight one obtains from the Ulm-Zippin theory guides the in-depth study of p-
groups of higher cardinalities. It was Hill who isolated the concept of nice subgroup
in an effort to find conditions under which Ulm’s theorem extends to a larger class
of p-groups. A close analysis of the Kaplansky–Mackey proof of Ulm’s theorem
(see Lemma 1.5) led him to the discovery of this significant type of subgroup which
embodies the property of finite subgroups relevant to the proof. We shall see that
the abundance of such subgroups in a p-group will make it possible to prove an Ulm
type theorem for a class of uncountable p-groups.
Nice Subgroups A subgroup N of a p-group A is called nice if every coset of
A mod N contains an element proper with respect to N. In other words, every coset
a C N .a 2 A n N/ contains an a C x .x 2 N/ such that hA .a C x/ D hA=N .a C N/.
We point out right away that in order to check the niceness of a subgroup N,
we need to look only at cosets of limit heights in A=N. In fact, first of all, every
element in a coset of height 0 has 0 height. Applying transfinite induction, suppose
that we know that cosets of heights
 contain elements proper with respect to
N, and hA=N .a C N/ D  C 1. Then pb C N D a C N for some b 2 A with
hA=N .b C N/ D hA .b/ D  (by hypothesis). Thus hA .pb/   C 1, and since strict
inequality is out of question, we see that pb is proper with respect to N. The case of
limit ordinals is included in the hypothesis.
The next lemma offers a most important characterization of niceness.
Lemma 2.1 (Hill [11]). A subgroup N of a p-group A is nice exactly if

p .A=N/ D .p A C N/=N for every .limit/ ordinal  and 1: (11.2)

Proof. Observe that p .A=N/ is the set of cosets whose heights in A=N are  ,
while .p A C N/=N is the image of p A under the canonical map A ! A=N. Hence
the inclusion  holds for every subgroup N of A. Now N is nice in A if and only if
every coset of height  can be represented by an element of A of height . Thus the
reverse inclusion holds for every  exactly if N is nice in A. (The reduction to limit
ordinals follows from the remarks before the lemma.) t
u
2 Nice Subgroups 353

We can rephrase the last lemma by saying that in the exact sequence 0 ! N !
A ! C ! 0 the subgroup N is nice if and only if the induced sequence p N !
p A ! p C ! 0 is exact for every  (the emphasis being on the exactness at p C).
Example 2.2.
(a) Finite subgroups are trivially nice. More generally, finite extensions of a nice subgroup are
nice. This follows from the simple observation that a finite extension cannot create new cosets
of limit heights.
(b) If A=N is a separable p-group, then N is nice in A.
Example 2.3. For every ordinal  , the subgroup p A is nice in the p-group A. Recall that heights
<  are preserved passing mod p A.

Properties of Nice Subgroups Let us acquaint ourselves with some elementary

properties of niceness.
(A) Direct summands are nice subgroups.
(B) Subgroups closed in the p-adic topology are nice. (In this case, the cosets have
finite heights only.) In particular, in a torsion-complete group, a pure subgroup
is nice if and only if it is a summand.
(C) The subgroup ˚i2I Ni is nice in ˚i2I Ai exactly if Ni is nice in Ai for each i 2 I.
(D) A nice subgroup N of A need not be nice in a subgroup B that lies between N
and A. This can be seen in a divisible group A where all subgroups are nice.
(E) The property of being nice is not transitive, in general (see Exercise 7), but the
following special case is true: a subgroup N of p A is nice in A if and only if
it is nice in p A. This is a consequence of the relation hA .a/ D  C hp A .a/
.a 2 p A/. Moreover, if N is nice in A, then N \ p A is nice in p A for every .
Visibly, (D)–(E) show that niceness does not behave like purity. However, the
following two useful properties are reminiscent of purity.
Lemma 2.4 (Hill [11]). Let A be a p-group, and M; N subgroups with N
M.
Then the following hold:
(i) if M is nice in A, then M=N is nice in A=N;
(ii) if N is nice in A and M=N is nice in A=N, then M is nice in A.
Proof.
(i) By our remark above, it suffices to check niceness at limit ordinals. We use
induction on the height. Let hA=N .a C M=N/ D  be a limit ordinal, and assume
that all cosets mod M=N of heights <  contain cosets mod N proper with
respect to M=N. This means that for every ordinal  < , there exists an x 2 M
such that hA=N .a C x C N/ > . Therefore, hA=M .a C M/  , and since
the reverse inequality is trivially true, hA=M .a C M/ D . By hypothesis, M
is nice in A, so there is an x 2 M such that hA .a C x/ D . It follows that
hA=N .a C x C N/ D , since strict inequality > is impossible.
(ii) By the hypotheses of (ii), hA=M .a C M/   implies the existence of a
y 2 M satisfying hA=N .a C y C N/ D , and of a z 2 N satisfying
hA .a C y C z/ D . t
u
354 11 p-Groups with Elements of Infinite Height

Consequently, if N is a nice subgroup of A, then under the natural correspondence

between subgroups of A containing N and subgroups of A=N, nice subgroups
correspond to nice subgroups.
F Notes. Niceness is a crucial concept in the theory. We shall see that nice subgroups play
crucial role also in mixed groups; see Chapter 15.

Exercises

(1) If M; N are subgroups of a p-group A, and if a 2 A is proper with respect to

both of them, then a is proper with respect to M \ N as well.
(2) A subgroup N is nice in the p-group A if and only if the equality N C p A D
\< .N C p A/ holds for all limit ordinals .
(3) If N is nice in A, then so is N C p A for every .
(4) (Hill) Let A be a p-group. For a subgroup N of A to be nice it is necessary and
sufficient that N \ p A is nice in p A and .N C p A/=p A/ is nice in A=p A for
all ordinals .
(5) A p-group has the property that all of its subgroups are nice if and only if it is a
direct sum of a bounded group and a divisible group.
(6) The union of an ascending chain of nice subgroups need not be nice.
(7) Let B denote an unbounded †-cyclic p-group, and B its torsion-completion.
Show that (a) BŒp
is nice in BŒp
(it is a summand); (b) BŒp
is nice in B; but (c)
BŒp
is not nice in B.
(8) Give an example where N < M < A, M=N is nice in A=N, but M fails to be
nice in A.
(9) The extensions of a p-group A by a p-group C in which A is a nice subgroup
form a subgroup in Ext.C; A/. [Hint: Baer sum.]

3 Simply Presented p-Groups

We now embark on an ambitious generalization of Ulm’s theorem: this section

and the three following ones are devoted to the theory of p-groups that can be
characterized by their UK-invariants. This theory is undoubtedly one of the major
achievements in the theory of p-groups of arbitrary cardinality; it provides the most
penetrating results known today on p-groups. We develop the theory on parallel
lines:
1. simply presented p-groups;
2. p-groups with nice systems (or nice composition series) of subgroups;
3. balanced-projective p-groups; and
4. totally projective p-groups.
Simple Presentation In this section, we deal with p-groups that have a special
kind of presentation. As a motivation, let us recall that a †-cyclic p-group can be
3 Simply Presented p-Groups 355

presented by a set of symbols x as the set of generators (corresponding to a basis),

with defining relations of the form pn x D 0 for certain integers n > 0; i.e., all the
relations involve exactly one generator. In contrast, no particular property is needed
for a group A to be presented with defining relations containing no more than three
generators. Indeed, if we choose the set of elements of A as generating system, then
the existing equalities like a C b D c for a; b; c 2 A provide relations of the desired
kind. Our present aim is to investigate those p-groups that are in between, i.e. which
can be presented by a set of generators such that all the defining relations contain at
most two generators.
Thus given a set X of generators, we now assume that every defining relation is
of the form nx D 0 or nx D my where x; y 2 X and m; n 2 N. Since we are dealing
with p-groups, we can get rid of integers prime to p, so that we may assume that
every defining relation is of the form

pn x D 0 or pn x D pm y (11.3)

where x; y 2 X. We will call a p-group A defined in this way a simply presented p-

group, and write A D hXI „i where „ stands for the collection of defining relations.
These groups, under the name of T-groups, were introduced by Crawley–Hales [1].
Faithful Simple Presentation We will need more control on the elements, and
therefore we simplify the presentation (at the expense of enlarging the generating
set). First, we replace pn x D 0 by the relations

px D x1 ; px1 D x2 ; : : : ; pxn1 D 0

after adjoining x1 ; : : : ; xn1 to X as additional generators. The same can be done for
the relations of the form pn x D pm y. As a result, we may assume without loss of
generality that every relation in „ is of the form px D 0 or px D y (x; y 2 X).
Next, in order to avoid the nuisance of working with generators that become 0
in A, and different generators that become equal in A, first we delete those x 2 X
that are 0 in A, and replace them by 0 in the relations. By doing so, we give the
same name in the relations to generators x ¤ y in X that become equal in A. To
summarize, we have now a presentation A D hXI „i such that
(i) each defining relation is of the form px D 0 or px D y with x; y 2 X;
(ii) if x 2 X, then x ¤ 0 in A;
(iii) if x; y are different elements in X, then x ¤ y in A.
A presentation satisfying (i)–(iii) will be called faithful. Thus if the simple
presentation of a p-group is faithful, then every generator occurs in exactly one
relation with coefficient p, but may occur in numerous relations with coefficient 1.
In this section we will assume (unless stated otherwise) that all the presentations
considered are faithful.
356 11 p-Groups with Elements of Infinite Height

Example 3.1.
(a) Cyclic groups of order pn are simply presented: Z.pn / D hx1 ; x2 ; : : : ; xn I px1 D 0; px2 D
x1 ; : : : ; pxn D xn1 i (faithful presentation).
(b) Quasi-cyclic groups are simply presented: Z.p1 / D hx1 ; : : : ; xn ; : : : I px1 D 0; px2 D
x1 ; : : : ; pxn D xn1 ; : : : i (faithful presentation).
(c) The Prüfer group H!C1 is simply presented: H!C1 D hx0 ; x1 ; : : : ; xn ; : : : I px0 D 0; px1 D
x0 ; : : : ; pn xn D x0 ; : : : i (this is not a faithful presentation).
Example 3.2. A clever method of constructing simply presented groups was suggested by Walker
[4]. For an ordinal ˇ, we define a simply presented p-group Pˇ of length ˇ C 1 as follows. The
generators are finite “strings” .ˇˇ1    ˇn / for n
0 where ˇ > ˇ1 > : : : > ˇn is a strictly
decreasing sequence of ordinals. The defining relations are:

p.ˇˇ1 : : : ˇn / D .ˇˇ1 : : : ˇn1 / for n

1; and p.ˇ/ D 0:

It is obvious that, for every ordinal ˇ, Pˇ is simply presented, and by induction it is not difficult to
verify that its length is ˇ C 1.

A faithful simple presentation gives rise to a partial order in the set X of

generators. For x; y 2 X we define

y<x provided pn x D y for some n > 0:

This is a partial order that satisfies the minimum condition. The minimal elements
are precisely those z 2 X for which pz D 0 is a relation in „. Note that the set
Xz D fx 2 X j x  zg is a tree with nodes corresponding to the generators in Xz , and
edges representing the relations between these generators in „.
Properties of Simple Presented Groups Divisible groups are simply pre-
sented, but the main interest lies in the reduced case. To elicit properties of reduced
simply presented p-groups, we now record a few basic facts on them. (All groups
are reduced, and simple presentations are faithful.)
(A) Direct sums of simply presented groups are simply presented. This is obvious.
(B) If M is the set of minimal elements of X in the partial ordering, then X D
P z2M Xz .disjoint union/ and
[

A D ˚z2M hXz i:

This is easy to see, since every x 2 X belongs to some Xz , and Xz ; Xy are disjoint
sets provided y ¤ z in M.
(C) A simply presented p-group of limit length is the direct sum of simply presented
groups of smaller lengths. This can be derived at once from (B) where the
subgroups hXz i with z 2 M are of smaller lengths.
(D) Every non-zero element a of a simply presented p-group can be written uniquely
in the form

a D r1 x1 C    C rk xk .k  1/ (11.4)
3 Simply Presented p-Groups 357

where x1 ; : : : ; xk are distinct elements of X and 0 < ri < p for i D 1; : : : ; k.

In view of (i)–(iii), the existence of such a representation is evident. To verify
uniqueness, suppose a D r1 x1 C    C rk xk D s1 x1 C    C sk xk holds for
distinct x1 ; : : : ; xk 2 X and for integers 0
ri ; si < p. Let x1 be a maximal one
among x1 ; : : : ; xk in the natural partial order of X. There is a homomorphism
 of hx1 ; : : : ; xk i, and hence of A, into Z.p1 / mapping x1 upon an element of
order p and x2 ; : : : ; xn to 0. Now a D r1 .x1 / D s1 .x1 / entails r1 D s1 , and
the rest follows by induction.
(E) If (11.4) is the unique representation of a ¤ 0 in a simply presented p-group
A, then a 2 p A if and only if xi 2 p A for i D 1; : : : ; k. Proof of necessity by
induction on . The assertion being trivially true for  D 0, suppose that a D pb
where b D s1 y1 C    C s` y` 2 p A for some  and distinct y1 ; : : : ; y` 2 X; here,
0 < sj < p for j D 1; : : : ; `. By induction hypothesis yj 2 p A for all j. Hence
a D s1 .py1 / C    C s` .py` /, and the set fx1 ; : : : ; xk g is coming from the set of
non-vanishing pyj . Hence necessarily xi 2 p C1 A, and we are done.
In particular, h.a/ D minfh.x1 /; : : : ; h.xk /g if a is as in (11.4).
(F) If A is simply presented, then so are A=p A and p A for every . Hence also the
Ulm factors are simply presented. This follows from (E).
(G) The Ulm factors of a simply presented p-group are †-cyclic groups. First
assume A is separable. Then the minimal elements in a faithful generating
system are of finite height, so from (E) it follows that the heights of the elements
in each hXz i in the decomposition (B) are bounded. Therefore, the groups hXz i
are †-cyclic. In the general case, by (F) the Ulm factors are simply presented,
they are also separable, so also †-cyclic.
(H) An infinite simply presented p-group A satisfies
X
jAj D jA0 j D fn .A/;
n<!

where A0 D A=A1 is the initial Ulm factor. As the Ulm factors are †-cyclic
groups, the second equality is evident. To prove the first, we induct on the length
 of the Ulm sequence. If A ¤ 0, but AC1 D 0, then Theorem 1.9(c) in
Chapter 10 implies jA=A j  jA j. As A=A is also simply presented and has the
same initial Ulm factor, by induction we obtain jAj D jA=A jjA j D jA=A j D
jA0 j: If the first ordinal  with A D 0 is a limit ordinal, then A D ˚ Ai with
summands for which the induction hypothesis applies. Hence the claim is clear.
A consequence of (H) is that the ranks of the Ulm factors of simply presented
groups never increase with the indices: rk A˛  rk Aˇ if ˛ < ˇ.
The following lemma is crucial.
Lemma 3.3. The generalized Prüfer groups H are simply presented.
Proof. From the definition of H it is manifest that if H is simply presented, then
so is H C1 . Invoking (C) and (A) at limit ordinals, the claim is established. t
u
358 11 p-Groups with Elements of Infinite Height

We will refer to the next result in the proof of the main Theorem 3.6.
Lemma 3.4. Let A D hXI „i be a simply presented p-group, and Y a subset of X.
The subgroup N D hYi is a nice subgroup of A.
Proof. Let a 2 A n N, and write a D r1 x1 C    C rk xk C s1 y1 C    C s` y` as
in (11.4) where xi and yj are distinct elements in X n Y and in Y, respectively; here
ri ; sj are positive integers < p. We claim that b D r1 x1 C    C rk xk 2 a C N is proper
with respect to N. Pick any c D t1 y01 C    C tm y0m 2 N, and calculate the height of
the element b C c 2 a C N. Evidently, we have h.b C c/
mini;j fh.xi /; h.y0j /g

mini fh.xi /g D h.b/. t

u
From the preceding proof it is clear that an x 2 X satisfying px 2 hYi is proper
with respect to hYi.
Crawley–Hales Theorem We now have all the ingredients to establish the main
structure theorem. This remarkable result does no less than characterizes a large
class of p-groups in terms of cardinal invariants.
Theorem 3.5 (Crawley–Hales [1]). Two simply presented p-groups are isomor-
phic if and only if their corresponding UK-invariants are identical.
Some advantage is to be gained by proving this result in the following strength-
ened form, in order to be applicable to mixed groups as well.
Theorem 3.6 (Hill [11], Walker [3]). For a prime p, let G; H be nice subgroups
of A; C, respectively, such that A=G D hXI †X i and C=H D hYI †Y i are simply
presented p-groups of length
. If
(i) there is a p-height-preserving isomorphism  W G ! H; and
(ii) the Hill invariants for the prime p are equal:

f .A; G/ D f .C; H/ for every 

;

then  extends to an isomorphism   W A ! C.

If only an inequality f .A; G/
f .C; H/ holds for every , then there is an
extension   which is a homomorphism.
Proof. Since certain restraints are needed to control the step-wise extension of ,
we select, for each , an arbitrary, but fixed isomorphism

˛ W p AŒp
=G./ ! p CŒp
=H./:

Consider the set of all pairs .G ;  / subject to the following conditions:
(a) G D hG; X i (for some subset X  X) is a nice subgroup of A;
(b)  is a height-preserving isomorphism of G with a nice subgroup H D
hH; Y i, where Y  Y;
(c)   G D ;
(d) for each , ˛ induces an isomorphism G ./=G./ ! H ./=H./.
3 Simply Presented p-Groups 359

The set of pairs .G ;  / is not empty, and is partially ordered in the obvious way:
.G ;  /
.G ;  / exactly if G
G and  D   G . Application of
Zorn’s lemma yields a maximal pair .G ;   /, where   is a height-preserving
isomorphism of G D hG; X  i with H  D hH; Y  i for some X   X and Y   Y.
Note that (d) guarantees that f .A; G / D f .C; H  / for every . We claim that
X  D X and Y  D Y.
Suppose the contrary, i.e. some x 2 X is missing from X  . Then there is also such
an x with px 2 G . By making use of Lemma 1.5, we can extend   to a height-
preserving isomorphism 0 W G0 D hG ; xi ! H0
C (where H 
H0 ) such that
(d) is preserved. There are a finite number of generators y1 ; : : : ; ym 2 Y such that
H0
H1 D hH; y1 ; : : : ym i. Now we invert the procedure: by the niceness of H0 ,
Lemma 1.5 is applicable to H0 , 01 , and to the generators y1 ; : : : ; ym to obtain a
finite extension G1 of G0 with a height-preserving isomorphism 1 W G1 ! H1 such
that 1  G0 D 0 and ˛ .G1 ./=G.// D H1 ./=H./. If G1 ¤ A, then we
select x1 ; : : : ; xn 2 X such that G1
hG0 ; x1 ; : : : ; xn i D G2 . We repeat this process,
alternately in C and A, to find a countable ascending chain G < G0
G1
: : :
of subgroups of A along with height-preserving isomorphisms k W Gk ! Hk
C

satisfying the requisite conditions, in particular, kC1  Gk D k for k < !. Setting
G D [k<! Gk and  D [k<! k , we get a pair .G ;   / that also belongs to
   

the set of pairs under consideration. This contradicts the maximal choice of G , so
we must have G D A.
To show that the simply presented group H  is all of C, suppose y 2 Y of height
 is missing from H  . As py D u 2 Y \ H  can be assumed, there is v 2 Y \ H 
such that pv D u and h.v/  . By construction, H  ./ D 0, which amounts to
p CŒp

p C1 C C H  , so we can write y  v D z C w with z 2 C, h.z/   C 1 and
w 2 H  . We must have h.w/  , thus if we write z D y  v  w in the canonical
form (11.4), using the generators Y, then y will have coefficient 1, so the right-hand
side cannot equal z.
The final claim follows from our argument. The proof of Theorem 3.6 is
complete, and Theorem 3.5 is the special case G D H D 0. t
u
UK-Invariants of Simply Presented p-Groups The structure Theorem 3.5 will
be much more satisfactory if we can answer the question as to which well-ordered
sequences

 0 ; 1 ; : : : ;  ; : : : . < /

of cardinals can be realized as UK-invariants of simply presented p-groups. We

note right away that in view of the existence of generalized Prüfer groups, there is
no restriction on the length  of the sequences. In order to state the precise condition
we need a new definition.
We call a function  from the set of ordinals  <  to the cardinals -
admissible if it satisfies
 D supf C 1 j P
(i) P  ./ ¤ 0gI and
(ii)  C!  ./
n<!  . C n/ for all  with  C ! < .
360 11 p-Groups with Elements of Infinite Height

Theorem 3.7 (Crawley–Hales [1], Hill [11]). Let  be a function from the
ordinals  <  to a set of cardinals. There exists a simply presented p-group A
of length  such that f .A/ D  ./ for all  <  if and only if the function  is
-admissible.
Proof. For necessity, we show that  satisfies (i)–(ii) if A is simply presented of
length  and  ./ D f .A/. (i) is obvious. (ii) is nothing else than Theorem 1.9(c)
in Chapter 3 combined with (H), observing that the UK-invariants of the ˛th Ulm
factor A˛ are fˇ .A/ where !˛
ˇ < !.˛ C 1/, so rk A˛ is equal to the sum of these
cardinals.
To prove sufficiency, let  be a -admissible function; by adjoining 0’s if needed
we may assume that  D ! is a limit ordinal. First of all, note that the future
Ulm factors A0˛ can be constructed as †-cyclic groups with invariants  .ˇ/ where
!˛
ˇ < !.˛ C 1/. Hence the case 
! is settled. Write  D 0 C n where 0
is a limit ordinal and n < !. We distinguish three cases.
Case I: n  2. The first step is to choose a basis fbi j i 2 Ig in A0 . By induction
hypothesis, there is a simply presented G whose Ulm factors are G˛ Š A0˛ for
˛
  2, while G1 is A01 , but modified as follows. Decompose A01 D
˚i2I Hi into unbounded summands; (ii) ensures that this can be done with I
as index set. If x 2 Hi is a basis element, then we change it to be x0 of order
o.x/o.bi /, to get G1 as modified A01 . Finally, A is defined as the factor group
of G by identifying all the o.x/x0 for the same i, for every i. It is easily seen that
A is simply presented, and A˛ Š A0˛ for each ˛
.
Case II: n DP0, i.e.  is a limit ordinal. We will show below:  can be written
as a sum i of i -admissible functions of lengths i < . If Ai is a simply
presented p-group with the i as the function for the sequence of UK-invariants,
then A D ˚i Ai will have the prescribed function  .
The proof of the claim on -admissible functions for limit ordinals
P  requires
a delicate set-theoretical argument. In the set of cardinals  D  <  ./
for all  < , let  denote the smallest with the smallest index. Set J D
fˇ j
ˇ < g, and consider the cardinals
X
 D min  . C n/
k<!
kn<!

for all limit ordinals  < . It is an elementary exercise in cardinal arithmetic

to show Pthat the restrictionP of  to the interval Œ;  C !/ decomposes as
 D ˇ2J ˇ such that kn<! ˇ . C n/   for all k < !. (This is
essentially the same argument as the one needed to decompose a †-cyclic p-
group of final rank  into a direct sum of  summands, each of final rank .) We
extendP ˇ by putting ˇ ./ D 0 whenever P <  or   PC !. Next, define

Pˇ D  ˇ for ˇ 2 J: Then we have  D ˇ2J ˇ and n<! ˇ . C n/ 

 C!  ./ for every  and for every ˇ. Finally, let
3 Simply Presented p-Groups 361

8
ˆ
<ˇ ./
ˆ if  < ˇ;
P
ˇ ./ D  ./ if  D ˇ;
ˆ ˇ
:̂0 if  > ˇ:
P
Then ˇ has length ˇ C 1 < , and ˇ2J ˇ D . In view of the inequalities
X X X X
ˇ . C n/ D ˇ . C n/  ./  ˇ ./
n<! n<! 
 Cn 
 Cn

for every  with  C ! < ˇ C 1, we infer that ˇ is .ˇ C 1/-admissible.

Case III: n D 1: Before entering into the discussion of this case, it should be
pointed out that a corollary to the proof in Case II is that the constructed simply
presented group (we call it now G) decomposes into the direct sum of groups
Gi such that for every ordinal  <  as many as jJj of them have length > !.
By construction, this G has the Ulm sequence A0˛ .˛ < /. Again choosing a
basis fbi j i 2 Ig in A0 , we observe that (ii) implies jIj
jJj. As a result, the
proof reduces to the case when the last Ulm factor is a cyclic group hbi i. It is
not difficult to see that there is no loss of generality in assuming that each Gi
has a last Ulm factor that is a cyclic group. Then the method of Case I can be
applied, and the direct sum of the constructed groups will be as desired. t
u
Finally, we estimate the cardinality of the set of non-isomorphic simply presented
groups of a given cardinality.
Corollary 3.8. For every infinite cardinal , there exist 2 non-isomorphic simply
presented p-groups of cardinality .
Proof. We apply Theorem 3.7 with  D , and refer to its proof where the Ulm
factors were chosen as unbounded †-cyclic groups. If we let each  ./ be  or 0
(or either only even or only odd invariants), then we have quite a freedom to satisfy
the required condition for admissibility, so that we will have 2 different choices.
Distinct choices mean non-isomorphic simply presented groups. (All constructed
groups will have isomorphic basic subgroups if we fix the initial Ulm factor.) t
u
F Notes. Simple presentation is not the historical route to the theory of p-groups that can be
classified by their UK-invariants, but—as Hill points out—it is considered by many as the most
aesthetic approach. The study of simply presented p-groups was initiated by Crawley–Hales [1];
they proved most of the main results on them.

Exercises

(1) Give a faithful simple presentation of Q.

(2) A separable p-group is simply presented if and only if it is †-cyclic.
362 11 p-Groups with Elements of Infinite Height

(3) If A is a p-group such that both p A and A=p A are simply presented, then so
is A.
(4) If A is simply presented and reduced, then fin rk A D fin rk B for any basic
subgroup B of A.
(5) A reduced simply presented p-group is fully transitive. [Hint: Theorem 3.6.]
(6) If A is a simply presented p-group, and C satisfies A ˚ A Š C ˚ C, then A Š C.
(7) Two simply presented p-groups are isomorphic if each of them is isomorphic to
an isotype subgroup of the other (for isotypeness, see Sect. 5). [Hint: the UK-
invariants of an isotype subgroup cannot be larger than those of the group.]
(8) Let Pˇ denote the group defined in Example 3.2. Prove that if ˇ D C ı, then
p Pˇ D Pı .

4 p-Groups with Nice Systems

Our main concern here is to make good our claim that p-groups with an abundance
of nice subgroups can be classified by their UK-invariants. In this section, we are
making a big step towards this goal.
Nice Composition Chains Let A be a p-group and

0 D N0 < N1 <    < N <    < N D A (11.5)

a smooth chain of subgroups in A such that

(a) each N is a nice subgroup of A; and
(b) NC1 =N Š Z.p/ for every  < .
Such a chain will be called a nice composition chain for A.
Nice Systems There is a stronger version of nice composition chain. By a nice
system is meant an H.@0 /-family N of nice subgroups (Hill [11]). It is an easy
exercise to extract a nice composition chain from a nice system: just select from
a nice system a smooth chain with countable factor groups, and then refine the
chain by making use of Example 2.2. Surprisingly, the converse is also true: every
group with a nice composition chain admits a nice system—this will follow from
Theorem 5.9 below.
Example 4.1.
(a) Every countable p-group admits a nice system: the set of all finite subgroups is such a system.
Recall that finite subgroups are always nice.
(b) Divisible groups have nice systems, since all of their subgroups are nice.
(c) Simply presented p-groups have a nice system: the set of subgroups generated by various
subsets of the generating set.

Lemma 4.2 (Hill [11]). The class of groups that admit nice systems is closed under
direct sums and summands.
4 p-Groups with Nice Systems 363

Proof. Let A D ˚i2I Ai where each Ai has a nice system Ni . Define N to consist
of all subgroups N D ˚i2I Ni with Ni 2 Ni .i 2 I/. Then each N is nice in A, and
it is straightforward to see that N is closed under taking sums. Finally, a countable
subset X of A has non-zero projection Xi only for i in a countable subset J  I.
For each i 2 J, there is an Mi 2 Ni such that Ni [ Xi  Mi and jMi =Ni j
@0 . Then
M 2 N obtained from N by replacing the Ni by Mi for i 2 J is as desired.
For summands, we appeal to Lemma 5.4 in Chapter 1 which tells us that a
summand B of a group A with an H.@0 /-family N also admits an H.@0 /-family
M of subgroups that are summands of groups in N . Summands are nice subgroups,
and summands of groups in N remain nice in B, so groups in M ought to be nice
in B. t
u
The following is an important observation, a restatement of Example 3.1(c).
Proposition 4.3 (Crawley–Hales [1], Hill [7]). The generalized Prüfer groups H ,
and more generally, simply presented p-groups admit nice systems.
Proof. From the construction of the generalized Prüfer groups it is clear that if H
has a system N of nice subgroups, then the preimages of the subgroups in N under
the natural epimorphism H C1 ! H yield a nice system in H C1 . The limit case is
obvious. The rest is a trivial corollary to Lemma 4.2. t
u
The next two results provide more information about the groups under
consideration.
Lemma 4.4. If A is a p-group such that both its nice subgroup N and the factor
group A=N admit nice composition chains, then so does A.
Proof. It suffices to refer to the simple fact that a nice composition chain in A=N
lifts to a nice composition chain between N and A. t
u
Groups with Nice Composition Chains In the next few results we explore the
implications of the hypothesis of having a nice composition chain. The following
lemma was proved by Hill [11] for nice systems, and later by Fuchs [IAG] for nice
composition chains.
Lemma 4.5. Suppose A; C are reduced p-groups, and  is a height-preserving
isomorphism between a nice subgroup G of A and a subgroup H of C. Suppose
that
(i) the factor group A=G has a nice composition chain, and
(ii) the Hill invariants satisfy f .A; G/
f .C; H/ for each ordinal .
Then  extends to a height-preserving isomorphism of A with a subgroup of C.
Proof. For each , we pick a monic map ˛ W p AŒp
=G./ ! p CŒp
=H./. We


also select a nice composition chain G D G0 < G1 <    < G <    < G D A,
and consider the set of all pairs .G ;  / such that
364 11 p-Groups with Elements of Infinite Height

(a)  is a height-preserving isomorphism of G with a subgroup H of C;

(b)   G D , and  is the restriction of  0 to G if  <  0 ;
(c) for each , ˛ induces an isomorphism G ./=G./ ! H ./=H./.
Evidently, there is a maximal pair .G ;  / in this set. (c) ensures that f .A; G /

f .C; H / for each . If  < , then owing to (a), we are in the situation of
Lemma 1.5, so  W G ! H can be extended to a height-preserving isomorphism
of GC1 with a subgroup of C such that (c) is not violated. This is a contradiction to
the choice of , so  D , and N D A, proving the assertion. t
u
We derive the following immediate corollary.
Corollary 4.6. The preceding lemma continues to hold if condition (ii) is dropped,
 is assumed not to decrease heights, and the extension is required to be a non-
height-decreasing homomorphism only.
Proof. We apply Lemma 4.5 to the groups A and A ˚ C with the nice subgroups
G and G ˚ H. The UK-invariants satisfy condition (ii), and the homomorphism
 W G ! G ˚ H is height-preserving. Hence Lemma 4.5 guarantees that there is a
height-preserving monomorphism A ! A ˚ C. The projection to C is a map as
desired. t
u
F Notes. Groups with nice systems were studied by Hill [7]; he used the terminology “Axiom-
3 groups” (referring to their H.@0 /-family property). He proved the generalized Ulm theorem for
these groups—a major result. That it suffices to assume a priori the existence of nice composition
chains was observed in [IAG]. It turns out that this is not a weaker hypothesis after all; actually, we
do not need a deep result like Theorem 5.9 to show that a group with a nice composition chain has
a nice system: Hill has an ingenious direct proof in [16], based on his Theorem 5.5 in Chapter 1.

Exercises

(1) Let A be a p-group, and n > 0 an integer. pn A has a nice system if and only if A
has one.
(2) An unbounded torsion-complete p-group has no nice composition chain. [Hint:
it is enough to prove if basic subgroup is countable; in a nice composition chain
from a countable index on the cokernels are not nice subgroups.]
(3) The torsion subgroup of a direct product of infinitely many unbounded reduced
p-groups with nice composition chains never admits such a chain. [Hint: hint in
preceding exercise.]
(4) Let C be a p-group. A reduced p-group A with a nice composition chain is
isomorphic to an isotype subgroup (see next section) of C exactly if f .A/

f .C/ for all :

5 Isotypeness, Balancedness, and Balanced-Projectivity

We shall now turn our attention to subgroups that are of immense help in developing
the theory of totally projective p-groups. We start with an important device that
5 Isotypeness, Balancedness, and Balanced-Projectivity 365

refines the concept of purity in exactly the same way as transfinite heights improve
on ordinary heights.
Isotype Subgroups A subgroup G of a p-group A is called isotype (Kulikov
[3]) if

p G D G \ p A for every ordinal I

in other words, transfinite heights of elements in G are the same whether computed
in G or in A. Equivalently, the exact sequence 0 ! G ! A ! A=G ! 0 implies
the exactness of

0 ! p G ! p A ! p .A=G/:

Let us learn some basic facts about isotype subgroups.

(a) An isotype subgroup is isotype in every isotype subgroup that contains it.
(b) The property of being isotype is transitive.
(c) Isotypeness is an inductive property.
(d) G is isotype in A if and only if p GŒp
D G\p AŒp
for every ; i.e. the elements
in the socle of G have the same heights in G as in A. This can be verified in the
same way as it was done in Sect. 1(F) in Chapter 5 for finite heights.
(e) If G is isotype in A, then for every ordinal , G=p G is isotype in A=p G.
Moreover, if G=p G is viewed as a subgroup of A=p A in the natural way, then
we can claim that G=p G is isotype in A=p A. These follow at once from the
preservation of heights <  mod p A.
We do not intend to pursue the exploration of isotypeness much further, but we
wish to point out an interesting application of this concept. (For a stronger result,
see Theorem 8.3 below.)
Lemma 5.1 (Irwin–Walker [2]). p A-high subgroups of a p-group A are isotype,
for any ordinal .
Proof. To prove this claim, let G be p -high in A. Transfinite induction is used to
establish the inclusion relation G \ p A
p G for all 
. This inclusion holds
trivially for  D 0, and is true for a limit ordinal provided that it holds for all smaller
ordinals. Thus, it remains to verify that it holds for C1, assuming it holds for . Let
g 2 G \ pC1 A; thus g D pa for some a 2 p A. If a 2 G, then a 2 G \ p A D p G,
and g 2 pC1 G. If a … G, then ha; Gi \ p A ¤ 0, say, na C g0 D a0 ¤ 0 with
g0 2 G; a0 2 p A; n 2 Z. As pa 2 G, n must be coprime to p, so n D 1 may be
assumed. From pa C pg0 D pa0 2 G \ p A D 0 we obtain pa0 D 0 and pg0 D g.
Now  <  implies g0 D a0  a 2 G \ p A D p G, whence g D pg0 2 pC1 G is
immediate. t
u
Balanced Subgroups We now combine the concepts of nice and isotype
subgroups to create a major concept, called balanced subgroup, which plays a more
important role in the theory. Balancedness crystallizes an idea that is common to
366 11 p-Groups with Elements of Infinite Height

tractable classes of torsion, torsion-free and mixed groups. Its real significance can
be better observed if we treat the torsion and torsion-free cases separately, before
embarking into the general discussion needed for mixed groups (see Chapter 15).
A subgroup B of a p-group A is called balanced if it is both nice and isotype.
Equivalently, an exact sequence 0 ! B ! A ! C ! 0 is balanced-exact if and
only if the induced sequence

0 ! p B ! p A ! p C ! 0 (11.6)

is exact for every ordinal . (An interesting observation that follows from what has
been said above is that the exactness of (11.3) at a non-limit ordinal  is responsible
for the isotypeness of B in A, and at limit ordinals for the niceness of B in A.) Thus,
for the last balanced-exact sequence, the commutative diagram

0 0 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ pσ B −−−−−→ pσ A −−−−−→ pσ C −−−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ B −−−−−→ A −−−−−→ C −−−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 −−−−−→ B/pσ B −−−−−→ A/pσ A −−−−−→ C/pσ C −−−−−→ 0

⏐ ⏐ ⏐
⏐ ⏐ ⏐
  

0 0 0

has exact columns and exact first and second rows. By virtue of the 33-Lemma 2.7
in Chapter 1, we conclude that even the third row is exact:

0 ! B=p B ! A=p A ! C=p C ! 0: (11.7)

Conversely, the exactness of (11.7) for all  is equivalent to the balancedness of

0 ! B ! A ! C ! 0.
From what has been said it follows readily that if B is balanced in A, then p B is
balanced in p A, and B=p B is balanced in A=p A for all .
That there is an abundance of balanced subgroups which are not direct summands
will be apparent from the existence of balanced-projective resolutions; see below.
The next examples are of special interest.
Example 5.2. Let B be an unbounded torsion-complete p-group, and 0 ! G ! C ! B ! 0 a
pure-projective resolution of B; thus, C; G are †-cyclic p-groups. Then G is closed in the p-adic
topology of C, so it is nice in C. As G is pure with no elements of infinite height, it is isotype in
C. Thus G is balanced in C. Since C is, but B is not a †-cyclic group, it is evident that G is not a
summand.
5 Isotypeness, Balancedness, and Balanced-Projectivity 367

Example 5.3. Consider the group C in Example 1.13. For every c 2 CŒp
, select a homomorphism
c of a copy Ac of the Prüfer group H!C1 into C such that some ac 2 Ac of height h.c/ maps upon
c. In this way, we get an epimorphism

 D r.˚c / W A D ˚c2CŒp
Ac ! C

whose kernel will be balanced in A (this will follow e.g. from Proposition 5.5 below). However,
Ker  cannot be a summand, since C is not a direct sum of countable groups.

Properties of Balancedness Some needed facts about the behavior of balanced-

ness are presented in the next lemma. The resemblance to purity is apparent.
Lemma 5.4. Let A be a p-group, and C < B subgroups in A.
(i) If C is balanced in A, then it is balanced in B.
(ii) If C is balanced in B, and B is balanced in A, then C is balanced in A.
(iii) If B is balanced in A, then B=C is balanced in A=C.
(iv) If C is balanced in A, and B=C is balanced in A=C, then B is balanced in A.
Proof.
(i) We only need to prove that C is nice in B. Let hB=C .b C C/ D  .b 2 B/ for
a limit ordinal , and assume, by induction, that cosets of B mod C of heights
<  contain elements proper with respect to C. Thus, for every  < , there is
a c with b C c 2 p B. As C is balanced in A, b C c 2 p A for some c 2 C.
Then c  c 2 p A \ C D p C whence b C c D .b C c / C .c  c / 2 p B for
all  < , and hB .b C c/ D .
(iii) In view of Lemma 2.4(i), it suffices to ascertain that B=C is isotype in A=C. If
b C C 2 p .A=C/ for b 2 B, then some a 2 p A and c 2 C satisfy b C c D
a 2 B \ p A D p B. Hence b C C 2 p B C C
p .B=C/; and (iii) follows.
(ii) Because of the transitivity of isotypeness, it is enough to show that C is nice
in A. Let hA=C .a C C/ D  .a 2 A/. If a 2 B, then (iii) implies that a C C has
height  in B=C, thus h.aCc/ D  for some c 2 C. If a … B, then h.aCB/  
and h.a C b/ D h.a C B/ for some b 2 B. Visibly, h.a  b C a C C/  ,
whence h.b C c/   and h.a C c/ D h.a C b  b C c/   for some c 2 C.
Consequently, h.a C c/ D , and (ii) is done.
(iv) In view of Lemma 2.4(ii), only isotypeness needs to be checked. Suppose
b 2 B \ p A. Then b C C 2 B=C \ p .A=C/ D p .B=C/ implies that
b  b1 2 C holds for some b1 2 p B. Therefore, b  b1 2 C \ p A D p C, and
b 2 p B. t
u
We draw attention to the following proposition that records most essential
characterizations of balancedness.

Proposition 5.5. For an exact sequence 0 ! B ! A!C ! 0 of p-groups, the
following are equivalent:
.i/ B is balanced in A;
.ii/ 0 ! p B ! p A ! p C ! 0 is exact for every ordinal ;
368 11 p-Groups with Elements of Infinite Height

.iii/ 0 ! B.u/ ! A.u/ ! C.u/ ! 0 is exact for all increasing sequences u D

.0 ; 1 ; : : : ; n ; : : :/ of ordinals and 1;
.iv/ .p AŒp
/ D p CŒp
holds for each .
Proof. The equivalence of .i/ and .ii/ has already been established above.
To prove .ii/ ) .iii/; note that the map B.u/ ! A.u/ is monic, since hB .pn b/ D
hA .pn b/ for all b 2 B and n  0. It is also clear that A.u/
C.u/. In order to
prove that this is not a proper inclusion, we use induction on the order of c 2 C.u/,
and show that p CŒpk

.p AŒpk
/ for k  1. For k D 1, pick c 2 p CŒp
, and
argue that there is an a 2 p A with a D c. Now pa 2 p C1 A \ B D p C1 B
implies that pb D pa for some b 2 p B. Since ˛ maps a  b 2 p AŒp
upon c,
the reverse inclusion follows. Assume the assertion is true for k  1 and for all .
If c 2 p CŒpk
.k > 1/, then there is x 2 A such that .x/ D c, and by induction
hypothesis, some y 2 p C1 AŒpk1
satisfies .y/ D pc 2 p C1 CŒpk1
. Choose
an a0 2 p AŒpk
with pa0 D y, and note that .x  a0 / D c  .a0 / 2 p CŒp
.
Consequently, c  .a0 / D .a1 / for some a1 2 p AŒp
, and c D .a0 C a1 / with
a0 C a1 2 p AŒpk
. Hence .iii/ is exact at C.u/.
.iv/ is a special case of .iii/.
For the implication .iv/ ) .ii/; it remains to check that B\p A D p B for all .
This being trivially true for  D 0, suppose it holds for . To verify it for  C1, pick
a b 2 B \ p C1 A and an a0 2 p A with pa0 D b. As .a0 / 2 p CŒp
, we can find an
a1 2 p AŒp
such that .a1 / D .a0 /. Then a D a0  a1 2 B \ p A D p B, whence
b D pa 2 p C1 B follows. The induction step for limit ordinals is automatic. t
u
The Extension Lemma We have come to a crucial lemma about extending
maps that do not decrease heights. The idea is borrowed from Hill [11].
Lemma 5.6. Given a commutative diagram

0 −−−−→ N −−−−→ G
⏐ ⏐
⏐
ψ
⏐φ

α
0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0

with exact rows, suppose that

(i) B is a balanced subgroup of A;
(ii) does not decrease heights .heights in N are computed in G/;
(iii) there is a g 2 G proper with respect to N satisfying pg 2 N.
Then can be extended to a map

W hN; gi ! A


which does not decrease heights either, and satisfies ˛ .g/ D .g/.
5 Isotypeness, Balancedness, and Balanced-Projectivity 369

Proof. Using the notation h.g/ D , (i) implies that some a 2 p A satisfies ˛a D
g. Visibly, pa  .pg/ 2 B \ p C1 A D p C1 B, whence the existence of a b 2 p B
with pb D pa  .pg/ is immediate. To define  , it suffices to specify   N D
and  g D a  b 2 A. It is readily seen that this is a homomorphism hN; gi ! A
such that ˛  g D ˛a D g. Obviously, h.ab/   D h.g/. In order to check that

does not decrease heights, it is enough to show that h.g C x/
h.a  b C x/
for every x 2 N. From (iii) we infer that, on one hand, h.g C x/ D minfh.g/; h.x/g,
and, on the other hand, h.a  b C x/  minfh.a  b/; h. x/g  minf; h.x/g. This
completes the proof. t
u
Balanced-Projective p-Groups Fundamental to our treatment is the concept
of balanced-projectivity. A p-group G will be called balanced-projective if it has
the projective property with respect to balanced-exact sequences of p-groups. Our
next aim is to characterize these p-groups, and at the same time to obtain relevant
information about groups with nice composition chains. The following two results
are preludes to Theorem 5.9 below. (Recall the H denotes the th generalized
Prüfer group.)
Lemma 5.7 (Nunke [5]). Suppose A is a p-group, and a 2 A is of order
pn and
of height  . Then there exists a homomorphism
 W H Cn ! A such that z D a

where hzi D p H Cn is cyclic of order pn .

Proof. The correspondence z 7! a gives rise to a homomorphism  W p H Cn ! hai
that does not decrease heights, computed in H Cn and A, respectively. It remains to
refer to Corollary 4.6 to complete the proof. t
u
Proposition 5.8. Let A be a reduced p-group of length . There exist a group G that
is a direct sum of generalized Prüfer groups of lengths
, and a balanced-exact
sequence
0 ! K ! G ! A ! 0: (11.8)

Proof. For a 2 A of order pn and of height , we select a copy Ga of H Cn and

a map a W Ga ! A as described in Lemma 5.7. Define the group G D ˚a2A Ga ,
and the map  D r.˚a / W G ! A. It is evident that .p HŒp
/ D p AŒp
for
every 
, so that by virtue of Proposition 5.5 we can claim with confidence that
Ker  D K is balanced in G. t
u
The next theorem will show that the generalized Prüfer groups are balanced-
projective. Consequently, (11.8) may be viewed as a balanced-projective resolu-
tion of A.
We now have all the tools for the core theorem that completely describes the
balanced-projective p-groups. Main portions of this result are due to Hill [7].
Theorem 5.9. For a reduced p-group A, the following conditions are equivalent:
370 11 p-Groups with Elements of Infinite Height

.a/ A has a nice system;

.b/ A admits a nice composition chain;
.c/ A is balanced-projective;
.d/ A is a summand of a direct sum of generalized Prüfer groups.
Proof. The implication .a/ ) .b/ is trivial.
.b/ ) .c/ We start with a balanced-exact sequence in the bottom row in the
following diagram, and with a map  W A ! C:

A
⏐
ψ ⏐φ

α
0 −−−−→ B −−−−→ G −−−−→ C −−−−→ 0

We select a nice composition chain 0 D N0 < N1 <    < N <    < N D A,

and define the map W A ! G step-wise, by transfinite recursion for  W N ! G:
at limit ordinals we take unions, while passing from N to NC1 we simply refer to
Lemma 5.6. Then ˛ D , proving .c/.
.c/ ) .d/ Owing to Proposition 5.8, there exists a balanced-exact sequence
0 ! K ! G ! A ! 0 where G is a direct sum of generalized Prüfer groups. .c/
implies that the sequence splits, whence .d/ is evident.
.d/ ) .a/ This implication is an immediate consequence of Proposition 4.3. ut
F Notes. Griffith [7] discusses a subfunctor of Ext that is actually Bext (subgroup of balanced
extensions) for torsion groups. Balancedness was studied under different names separately for
p-groups and torsion-free groups, until Hunter [1] developed a unified theory, including mixed
groups. Balanced-projectivity was also studied by Warfield [7].
The problem of balanced-injective p-groups was settled by Griffith [7]: there is nothing new,
they are just the torsion-complete p-groups. To prove this in one direction, observe that maps into
a torsion-complete p-group kill elements of infinite heights, so balancedness reduces to purity.

Exercises

(1) G is isotype in A if and only if p G is pure in p A for every .

(2) A pure subgroup G of A is isotype whenever A=G is separable.
(3) (Kulikov) A subgroup C of A is isotype if and only if, for an ordinal , p C is
isotype in p A, and C=p C is isotype in A=p A.
(4) If G is isotype in A, then f .G/
f .A/ for every .
(5) (Irwin–Walker) In a p-group A, every subgroup with elements of heights

 can be embedded in an isotype subgroup of cardinality

jj@0 whose
elements are again of heights
.
(6) Let A be generated by symbols fan ; bn gn<! subject to the relations pa0 D
0; pb0 D a0 ; pn an D a0 ; pn bn D b0 for all n  1. Then G D ha0 ; : : : ; an ; : : : i
is pure, but not isotype in A.
6 Totally Projective p-Groups 371

(7) A balanced subgroup of a countable p-group is a summand. [Hint: balanced-

projectivity.]
(8) Prove Lemma 5.7 by induction on . [Hint: it is trivial for finite , and easy
for limit ; for successor ordinals use y 2 p H Cn1 with py D z and b 2 A
with pb D a; h.b/    1, and extend  to hyi ! hbi.]
(9) (Warfield) A short exact sequence of p-groups relative to which the general-
ized Prüfer groups have the projective property is balanced-exact.
(10) (E. Walker) In the next section we show that the balanced-projective p-groups
are exactly the simply presented ones. Using this, verify the following method
of getting a balanced-projective resolution 0 ! K ! G ! A ! 0 of a p-
group A. G is generated by xa , one generator for each a 2 A, subject to the
defining relations x0 D 0, and pxa D xb if and only if pa D b holds in A. The
correspondence xa 7! a induces an epimorphism  W G ! A, and K D Ker 
is balanced in G.
(11) If G is a balanced subgroup of A, then Tor.G; X/ is balanced in Tor.A; X/ for
every group X. [Hint: Lemma 4.2 and Theorem 3.1 in Chapter 8.]

6 Totally Projective p-Groups

A most important type of p-group was discovered by Nunke [3] via homological
considerations. Later, it turned out that the class of p-groups with nice systems
introduced by Hill coincides with this class.
Total Projectivity Recall that †-cyclic groups can be characterized as groups A
satisfying Pext.A; C/ D 0 for all groups C. If we restrict our attention to p-groups,
and take into consideration that Pext is just the first Ulm subgroup of Ext, then we
can claim that a p-group A is †-cyclic exactly if p! Ext.A; C/ D 0 for all groups C.
Nunke defined classes of p-groups A using this observation by replacing ! by an
arbitrary ordinal.
Let  stand for an ordinal. A p-group A is called p -projective if

p Ext.A; C/ D 0 for all groups C;

and totally projective if

p Ext.A=p A; C/ D 0 for all ordinals  and all groups C: (11.9)

Accordingly, A is totally projective if and only if the factor group A=p A is p -

projective for each .
Observe that if  < , then p -projective implies p -projective.
Example 6.1.
(a) A pn -bounded group is pn -projective, and an unbounded †-cyclic p-group is p! -projective.
(b) A pn -bounded group is totally projective, and so is any †-cyclic p-group.
372 11 p-Groups with Elements of Infinite Height

(c) The Prüfer group H!C1 is both p!C1 -projective and totally projective. This will follow from
Lemmas 6.3 and 6.4 infra.

We will need the following result on Ext.

Proposition 6.2 (Irwin–Walker–Walker [1], Nunke [5]). If the subgroup B of A
satisfies B
p A for some ordinal , then for every group G, the exact sequence
ˇ ˛
0 ! B ! A ! C ! 0 induces the exact sequence

ˇ ˛
Hom.G; C/ ! Ext.G; B/ ! p Ext.G; A/ ! p Ext.G; C/ ! 0:

Proof. We refer to Theorem 2.3 in Chapter 9, and show that dropping to subgroups
is legitimate. Consider the Nunke group N of length  for the prime p (Sect. 1
in Chapter 15) with the exact sequence 0 ! Z ! N ! H ! 0 where
p N Š Z. There is a natural homomorphism ıA W A Š Hom.Z; A/ !
Ext.H ; A/ whose kernel is Ker ıA D Im.Hom.N˛ ; A/ ! A/ D p A. ıA
induces the map ıA W Ext.G; A/ ! Ext.G; Ext.H ; A//. We also have ıExt.G;A/ W
Ext.G; A/ ! Ext.H ; Ext.G; A//, where the last Ext is naturally isomorphic to
Ext.G; Ext.H ; A// (see Theorem 3.8 in Chapter 9). Owing to Lemma 5.5 in
Chapter 9, Im ıA
p Ext.H ; A/ D 0, thus ıA ˇ D 0, and so ıA ˇ  D 0 D
ıExt.G;A/ ˇ  . Hence Im ˇ 
Ker ıExt.G;A/ D p Ext.G; A/: That the displayed
sequence is exact at the last two Exts is straightforward. t
u
A few simple remarks are inserted here.
(A) The class of p -projective p-groups .for a fixed / as well as the class of totally
projective p-groups is closed under direct sums and summands.
(B) A p -projective group A has length
. To prove this, consider for any C the
following commutative diagram with exact rows:

α
Hom(A/pσ A, C/pσ C) −−−−−→ Hom(A, C/pσ C) −−−−−→ Hom(pσ A, C/pσ C)
⏐ ⏐ ⏐
⏐ ⏐γ ⏐
  
β
Ext(A/pσ A, pσ C) −−−−−→ Ext(A, pσ C) −−−−−→ Ext(pσ A, pσ C) −−−−−→ 0
⏐
⏐


pσ Ext(A, C) = 0

where is surjective (Proposition 6.2), and ˛ D 0 (p A is in the kernel of every

map A ! C=p C). We conclude that ˇ D 0, whence Ext.p A; p C/ D 0. This
implies p A D 0, for otherwise we could choose C, e.g. such that p C Š Z.p/
to have the last Ext ¤ 0.
In particular, a pn -projective group is pn -bounded .n 2 N/.
(C) It can very well happen that a p C1 -projective p-group has length  even if it
is not p -projective. See Sect. 9.
6 Totally Projective p-Groups 373

Lemma 6.3 (Nunke [3], Irwin–Walker–Walker [1]). If S is a subsocle of a p-

group A such that A=S is p -projective, then A is p C1 -projective.
Proof. The claim is obvious if  < !, so assume   !. We have an exact sequence
ı
Hom.S; /! Ext.A=S; /! Ext.A; / ! Ext.S; / ! 0. Here p Ext.S; / D 0
implies pExt.A; /
Im ı, so it suffices to verify p C1 Im ı D 0, because then
p C1 Ext.A; / D 0 (observe that 1 C  D  as   !).
More generally, we show that if S is a subsocle of a p-group E satisfying p E D 0,
then p C1 .E=S/ D 0. But this is clear from p .E=S/
EŒp
=S. (See Sect. 1(B) in
Chapter 10.) t
u
The same argument applies to establish the corresponding result on total
projectivity.
Lemma 6.4 (Nunke [3]). If A is a p-group satisfying p C1 A D 0 such that A=p A
is totally projective, then A is likewise totally projective. t
u
An immediate consequence of this lemma is that for the total projectivity of a
p-group A, it suffices to check that p Ext.A=p A; C/ D 0 holds for limit ordinals .
Indeed, it is evident from Lemma 6.4 that if (11.9) holds for , then it also holds for
 C 1.
A Characterization of Total Projectivity We now relate nice composition
chains to total projectivity.
Theorem 6.5 (Hill [11]). A p-group is totally projective if and only if it satisfies
one .and hence all/ of the equivalent conditions of Theorem 5.9.
Proof. The first task is to show that a direct summand of a direct sum of generalized
Prüfer groups is totally projective. Because of (A), it suffices to establish the total
projectivity of H for all . As expected, a straightforward induction on  will do
the job. Indeed, passage from  to  C 1 is trivial by virtue of Lemma 6.4, while
for limit ordinals  it is a consequence of being the direct sum of groups H with
 < .
Conversely, suppose A is totally projective. We induct on the length  of A. If
 D n 2 N, then A is a bounded group, so †-cyclic and balanced-projective. If A
is totally projective of length  C 1, then A=p A is totally projective of length ,
consequently, by induction hypothesis, it has a nice composition chain. So does the
elementary p-group p A that is nice in A, thus also A has a nice composition chain.
Finally, assume that the length  of A is a limit ordinal. We show that any
balanced-exact sequence e W 0 ! B ! G ! A ! 0 is splitting. From Sect. 5
we know that the sequence e W 0 ! B=p B ! G=p G ! A=p A ! 0 is exact
for all  < , and the large diagram commutes. Starting from the bottom row, we
construct the following diagram:
374 11 p-Groups with Elements of Infinite Height

α β
: 0 −−−−−→ B −−−−−→ G −−−−−→ A −−−−−→ 0


μ ν 
γ δ
ρφ : 0 −−−−−→ B/pρ B −−−−−→ H −−−−−→ A −−−−−→ 0
 ⏐ ⏐
 ⏐ ⏐φ
 ψ 
 ξ
ρ : 0 −−−−−→ B/pρ B −−−−−→ G/pρ G −−−−−→ A/pρ A −−−−−→ 0,

We get the middle row as a pull-back. The canonical map  W G ! G=p G

followed by  equals ˇ, so the pull-back property ensures the existence of a
unique  W G ! H such that  D  and ı D ˇ. If  W B ! B=p B denotes
the canonical map, then for the commutativity of the diagram it only remains to
check that  D ˛. Now  D  D ˛ D ˛ and ı  D 0 D ˇ˛ D ı˛
show that  and ˛ are maps B ! H which become equal if followed by or ı.
Since H was a pull-back,  D ˛ follows.
By a remark in Sect. 5, after (11.7), B=p B is balanced in G=p G, and
therefore e must split (by induction hypothesis, A=p A is balanced-projective).
It follows that e  also splits, and hence in view of the exact sequence
 
Ext.A; p B/ !Ext.A; B/!Ext.A; B=p B/ we conclude that e 2 Im  where
 W p B ! B is the inclusion map. By Proposition 6.2, Im 
p Ext.A; B/, and
therefore e 2 \< p Ext.A; B/ D p Ext.A; B/. The last group is 0, if A is totally
projective, i.e. e is splitting. t
u
Before continuing, let us summarize briefly the various characterizations of the
class of p-groups we are studying. The theorems we have proved so far already show
that for a reduced p-group A the following are equivalent:
1. A is simply presented;
2. A has a nice system (or a nice composition chain);
3. A is a summand of a direct sum of generalized Prüfer groups;
4. A is balanced-projective;
5. A is totally projective.
And more importantly, these groups are characterized by their UK-invariants.
Though these groups can be referred to by any of the listed properties, we
more often use the term ‘totally projective p-groups,’ because chronologically
this property was first introduced. However, total projectivity is perhaps the least
practical criterion to recognize groups in this class.
Consequences of the Main Result Capitalizing on Theorem 6.5, we can get
valuable information about how totally projective p-groups are related to direct sums
of countable groups.
Theorem 6.6 (Nunke [5]). A reduced p-group is a direct sum of countable groups
if and only if it is totally projective of length
!1 .
6 Totally Projective p-Groups 375

Proof. Combining Example 4.1, Theorems 5.9 and 6.5, it follows that a direct sum
of countable p-groups is totally projective; evidently, its length cannot exceed !1 .
Conversely, a totally projective p-group A of length
!1 is a summand of a direct
sum of generalized Prüfer groups H of lengths  < !1 . These H are countable,
and an appeal to Kaplansky’s Theorem 2.5 in Chapter 2 completes the proof. t
u
Corollary 6.7 (Kolettis [1]). Direct sums of countable p-groups can be classified
by their UK-invariants. t
u
If we concentrate on the entire class of totally projective p-groups rather than
on the individual groups, then we can establish two remarkable characterizations:
one as the smallest, and the other as the largest class satisfying certain conditions.
Accordingly, we now prove two theorems. In the next result, an elongation of a
p-group B by a p-group C means an extension A of B by C such that B D p A for
some ordinal  (which is equal to the length of C).
Theorem 6.8 (Parker–Walker [1]). The class of totally projective p-groups is
the smallest class C of groups that is closed under isomorphism, and shares the
following properties:
(a) Z.p/ 2 C;
(b) C is closed under the formations of arbitrary direct sums and direct summands;
(c) C is closed under elongations.
Proof. From (a)–(c) it follows that cyclic and †-cyclic p-groups are in C. Then by
(c) the generalized Prüfer groups of length ! C n and !2 also belong to this class.
A straightforward induction convinces us that all generalized Prüfer groups H
belong to C, and then (b) guarantees that all totally projective p-groups are members
of C. Since the class of totally projective p-groups shares the listed properties, the
assertion follows. t
u
The second theorem we hinted at above asserts that the class of totally projective
p-groups is the natural boundary for the extension of Ulm’s theorem.
Theorem 6.9. The class of totally projective p-groups is the largest class B of
p-groups that
(a) contains the generalized Prüfer p-groups,
(b) is closed under direct sums and summands, and
(c) non-isomorphic groups in B are distinguishable by their UK-invariants.
Proof. We need to show that only totally projective groups can belong to B. Let A
be a member of B, say, of length . Consider the direct sum H D ˚ ˚  H of
generalized Prüfer groups where  D jAj@0 . By (a)–(b), H 2 B, and it is clear that
f .H/ D  for all 
. By (b), H ˚ A 2 B. Evidently, f .H ˚ A/ D f .H/ for all
 < , thus (c) implies H ˚ A Š H. This amounts to saying that A is a summand of
a direct sum of generalized Prüfer groups, so it is totally projective. t
u
376 11 p-Groups with Elements of Infinite Height

Transitivity and Full Transitivity We plan to show that totally projective p-

groups A are transitive (fully transitive) in the sense that if a; b 2 A satisfy u.a/ D
u.b/ (u.a/
u.b/), then there exists an ˛ 2 Aut A (˛ 2 End A) such that ˛.a/ D b.
Theorem 6.10. Totally projective p-groups are both transitive and fully transitive.
Proof. Suppose A is totally projective. If a; b 2 A satisfy u.a/ D u.b/, then there
exists a height-preserving isomorphism  W hai ! hbi; these subgroups are finite,
and hence nice. Furthermore, we have A=hai Š A=hbi, both totally projective.
Indeed, these factor groups have the same UK-invariants: this is obvious if the
only gap in u.a/ is at 1, and it follows by induction for a finite number of gaps.
Theorem 3.6 implies the existence of ˛ 2 Aut A such that ˛.a/ D b.
In case u.a/
u.b/, we apply what has been proved for the group A ˚ A0 (where
A Š A) and to the elements .a; 0/ and .b; a0 /. After application of ˛ 2 Aut.A ˚ A0 /,
0

we project to the first summand. t

u
F Notes. Totally projective p-groups were introduced by R. Nunke in order to give a
homological description to direct sums of countable p-groups. He proved that a totally projective
p-group is a direct sum of countable groups provided its length is  !1 . In view of Kolettis’
[1] remarkable result that direct sums of countable p-groups are classifiable by UK-invariants,
E. Walker conjectured that such a classification extends to totally projective p-groups. A good
progress was made toward the proof of this conjecture when it was confirmed for groups of lengths
 !1 by Parker–Walker [1]. The credit of extending Ulm’s theorem to arbitrary p-groups with nice
systems belongs to Hill. Walker [3] simplified the proof, and pointed out that the extension lemma
may start from any group, so it can be applied to mixed groups as well.
There is a large literature on various aspects of total projectivity. For example, Hill [15] shows
that the set union of a countable number of totally projective p-groups is again totally projective.
Göbel–May [4] give conditions on a direct sum of countable groups when it is cancellable on both
sides of a direct sum. The compact groups that are Pontryagin duals of totally projective p-groups
were described by Kiefer [1].
It is interesting that in several cases, for a p-group A, the group V.A/=A is totally projective.
Here V.A/ denotes the (multiplicative) group of normalized units in the group algebra F.A/ over a
field F.
Since the classification of all p-groups seems to be an unrealistic goal at this time, one has
to be content with aiming at classifying significant subclasses. There are a couple of successful
attempts using additional types of invariants describing classes of p-groups larger than that of
totally projective p-groups whose members admit complete sets of invariants. Space forbids our
presenting any of these generalizations, but we will call attention to A- and S-groups in Notes to
Sect. 7.

Exercises

(1) Show that in the definition of total projectivity, C may be restricted to p-groups.
[Hint: check the proofs.]
(2) (Irwin–Walker–Walker) If A is p -projective, then for every group C,
Ext.A; C/ Š Ext.A; C=p C/.
7 Subgroups of Totally Projective p-Groups 377

(3) If A is a totally projective p-group, and C is a subgroup satisfying pn A

C
A
for some n 2 N, then C is also totally projective. [Hint: p! C D p! A and C=p! C
is †-cyclic.]
(4) (Nunke, Hill–Megibben) Let  denote a countable ordinal. If A is a p-group
such that both p A and A=p A are direct sums of countable groups, then A is
also a direct sum of countable groups.
(5) (Nunke) Let A be a p-group whose nth Ulm subgroup vanishes for some
integer n. If all the Ulm factors of A are †-cyclic groups, then A is a direct
sum of countable groups.
(6) (Nunke) If A is a p -projective p-group, then so is Tor.A; C/ for any C. [Hint:
trivial proof by Ext.A; Ext.B; G// Š Ext.Tor.A; C/; G/.]
(7) A p-group is totally projective if and only if it is a summand of a direct sum of
copies of groups Pˇ for various ˇ (see Example 3.2).
(8) (Irwin–Walker–Walker [1]) A subgroup B of a p-group A is contained in
the Ulm subgroup A1 if and only if the sequence 0 ! Hom.G; B/ !
Hom.G; A/ ! Hom.G; A=B/ ! Ext.G; B/ ! Pext.G; A/ ! Pext.G; A=B/ !
0 is exact for all groups G.

7 Subgroups of Totally Projective p-Groups

When is a subgroup of a totally projective group again totally projective? We have

a better chance for obtaining a useful answer to this question if we concentrate on
special subgroups, like fully invariant or isotype. But first let us see an example
demonstrating that even subgroups of direct sums of countable groups need not be
such direct sums.
Example 7.1 (Nunke [5]). Let B be an unbounded countable †-cyclic p-group, and H!C1 the
Prüfer group of length ! C 1. The group A D Tor.B; H!C1 / is contained in Tor.D; H!C1 / (where
D D ˚Z.p1 / is the divisible hull of B), so A is contained in a direct sum of countable groups
H!C1 . By Lemma 4.2 in Chapter 8, A has length  !, therefore, if A is a direct sum of countable
groups, then it must be †-cyclic. But A cannot be †-cyclic, because the pure-exact sequence
0 ! B ! B ! ˚Z.p1 / ! 0 induces the pure-exact sequence 0 ! Tor.B; H!C1 / !
Tor.B; H!C1 / ! Tor.˚Z.p1 /; H!C1 / ! 0, where the first Ulm subgroup of the last Tor has
cardinality 2@0 , and such a group cannot be the image of a †-cyclic p-group A modulo a countable
subgroup Tor.B; H!C1 /.

Fully Invariant in Totally Projective To begin with, we point out that every
fully invariant subgroup of a totally projective p-group A is of the form A.u/
for some u = .0 ; 1 ; : : : ; n ; : : :/ with an increasing sequence of ordinals and
symbols 1. In fact, totally projective groups are fully transitive by Theorem 6.10,
and the rest follows from Theorem 2.2 in Chapter 10.
Theorem 7.2 (Fuchs–E. Walker, Linton [1]). Fully invariant subgroups of totally
projective p-groups are likewise totally projective.
378 11 p-Groups with Elements of Infinite Height

Proof. Since full invariance commutes with arbitrary direct sums, it suffices to
prove the claim for the generalized Prüfer groups of the form H C1 . Consider
H C1 .u/ where u = .0 ; 1 ; : : : ; n ; : : :/. If  C 1 < n for some n, then H C1 .u/ is
bounded, so it is trivially totally projective. If n <  C 1 for all n 2 N, then we
induct on . Evidently, we have

Œpn .pn H C1 /
=p H C1 D pn Œpn .H C1 =p H C1 /
Š pn .pn H /

(naturally), and H C1 .u/ D \n2N pn .pn H C1 /. Consequently, we have

H C1 .u/=p H C1 Š H .u/. Thus H C1 .u/ is totally projective as an extension
of the cyclic group p H C1 by the totally projective H .u/. t
u
Separative Subgroups The following concept will be needed; it was introduced
by Hill [17]. It generalizes the notion of nice subgroups.
A subgroup B of a p-group A is called separative (or, separable in the sense of
Hill) if, for each a 2 A, there is a countable subset fbn j n < !g in B such that

supfhA .a C x/ j x 2 Bg D supfhA .a C bn / j n < !g:

In other words, the set fhA .a C x/ j x 2 Bg of ordinals either contains a largest

element or is cofinal with !. By a separative chain from A to G is meant a smooth
chain A D A0 <    < A˛ <    < A D G (for some ordinal ) of separative
subgroups of G, such that, for each ˛ < , the factor group A˛C1 =A˛ is countable.
Let us point out that supfhA .a C x/ j x 2 Bg is equal to hA=B .a C B/, provided the
latter is ¤ 1. An easy example for the 1 case is a separable p-group A where this
supremum for any non-zero coset modulo a basic subgroup B is !, but the coset has
height 1 in A=B.
Example 7.3.
(i) Let C be a subgroup of a separable p-group A such that A=C has an element of height !1 .
Then C is not separative in A.
(ii) A separable p-group A admits a separative chain through any of its basic subgroups. All
factor groups can be chosen to be cocyclic.
To start with, the following important properties of separative subgroups should
be pointed out.
(A) It is obvious from the definition that nice subgroups, countable subgroups,
and more generally, subgroups with heights bounded by a countable ordinal
are separative.
(B) Subgroups modulo which the factors groups have lengths < !1 are separative.
(C) Countable extensions of separative subgroups are separative. Suppose
C < B < A, where C is separative in A, and B=C is countable.
Let fbn 2 B; n < !g be a complete set of representatives of B mod C.
By hypothesis, for each n < !, there is a countable cofinal subset
fhA .a C bn C cni / j cni 2 C; i < !g in fhA .a C bn C c/ j c 2 Cg: Then
the countable set fhA .a C bn C cni / j bn 2 B; cni 2 CI n; i < !g is cofinal in
fhA .a C b/ j b 2 Bg, thus B is separative in A.
7 Subgroups of Totally Projective p-Groups 379

(D) If B0 < B1 <    <SBn < : : : is a countable chain of separative subgroups of

A, then their union n<! Bn is likewise separative in A.
(E) Separativity is transitive. In fact, suppose C < B < A; and let C be separative
in B, and B in A: Given a 2 A n B, we consider the set fhA .a C c/ j c 2 Cg:
By hypothesis, the set fhA .a C b/ j b 2 Bg contains a countable cofinal subset
fhA .a C bn / j bn 2 B; n < !g: Similarly, for each n, the set fhA .bn C c/ j
c 2 Cg contains a countable cofinal subset fhA .bn C cni / j cni 2 C; i < !g:
To show that fhA .a  cni / j cni 2 CI n; i < !g is cofinal in fhA .a C c/ j c 2 Cg;
note that there is an n < ! such that hA .a C c/
hA .a C bn /: Then
hA .a C c/
hA .bn  c/; so there is an i < ! with hA .bn  c/
hA .bn C cni /:
Hence hA .a C c/
hA .a C bn  bn  cni / D hA .a  cni /:
(F) Let C < B < A where C is nice in A. Then B=C is separative in A=C if and
only if B is separative in A. To prove this, note that by the niceness of C, for
any a 2 A; hA=C .a Cb CC/ with b 2 B is equal to hA .a Cb Cc/ for some c 2 C
(depending on b). Therefore, fhA=C .a C b C C/ j b 2 Bg contains a countable
cofinal subset if and only if so does the set fhA .a C b/ j b 2 Bg.
Absolute Separativity The next theorem shows how well separativity fits when
total projectivity is discussed.
A p-group is said to be absolutely separative if it is a separative subgroup in
every p-group in which it is contained as an isotype subgroup.
Theorem 7.4 (Hill [17]). Totally projective p-groups are absolutely separative.
Proof. Let A be a totally projective p-group, and N an H.@0 /-family of nice
subgroups of A. Suppose A is an isotype subgroup in a p-group G, but for some
g 2 G, the set fhG .g C a/ j a 2 Ag has cofinality > !. Manifestly, there is a chain
of subgroups An 2 N .n < !/ such that

supfhG .g C a/ j a 2 A1 g <    < supfhG .g C a/ j a 2 An g <   

The union C of the An also belongs to N . By our hypothesis on cofinality, there is a

b 2 A such that hG .g C b/ > supfhG .g C x/ j x 2 Cg: Hence

supfhG .b  x/ j x 2 Cg D supfhG .g C x/ j x 2 Cg:

C is nice in A, so hG .b  c/ D supfhG .b  x/ j x 2 Cg for some c 2 C: It follows

that hG .b  c/ D supfhG .g C x/ j x 2 Cg: But hG .b  c/ > hG .g C c/ implies
hG .g C b/ D hG .g C c/, a contradiction. t
u
Compatibility For subgroups A and B of a p-group G, we say that A is
compatible with B (in notation: AkB, see Hill [17]) if for every pair .a; b/ 2 A  B,
we have

hG .a C b/
hG .a C x/ for some x 2 A \ B: (11.10)
380 11 p-Groups with Elements of Infinite Height

(a) Note that the inequality (11.10) implies hG .a C b/

hG .b  x/I hence,
compatibility is a symmetric relation.
(b) Trivially, AkB whenever B
A.
Example 7.5.
(i) Let G be an unbounded separable p-group, and A a pure subgroup of G. Then AkB holds,
e.g. for a subgroup B < G for which A \ B is a summand of G. Observe that every a 2 A is
contained in a summand A0
A \ B of G, so the A0 -coordinate a C x of a C b .b 2 B/ will
satisfy h.a C b/  h.a C x/.
(ii) Let A be a finite subgroup, and B a basic subgroup in a separable p-group G such that A 6 B.
Then B is not compatible with A. Indeed, for every a 2 A n B there is a b 2 B such that
h.a C b/ > h.x/ for every x 2 A, so also h.a C x/ < h.a C b/.

(c) Compatibility is an inductive property: if B1 < B2 <    < B < S   is any

chain of subgroups of G, each compatible with A, then the union ˛ B is
likewise compatible with A.
(d) If A; B; C are subgroups of G, then Ak.B C C/ and BkC imply .A C B/kC. To
verify this, let a 2 A; b 2 B; c 2 C: Then there is an x 2 A \ .B C C/ such that
h.a C b C c/
h.b C c C x/. Setting x D b0 C c0 with b0 2 B; c0 2 C, we have
h.bCcCx/ D h.bCb0 CcCc0 /
h.yCcCc0 / for some y 2 B\C, and therefore
h.aCbCc/
h.aCb.yCc0 //: Here yCc0 2 C and yCc0 D yCxb0 2 ACB;
so .A C B/kC; indeed.
Lemma 7.6. Let A and B be subgroups of a p-group G such that B is nice in G and
compatible with A. Then the following are true:
(i) A \ B is nice in A.
(ii) If A is isotype in G; then .A C B/=B is isotype in G=B.
(iii) A subgroup B0 > B satisfies .B0 =B/k.A C B/=B exactly if B0 kA.
Proof.
(i) If a 2 A n .A \ B/ D A n B; then there is b0 2 B such that h.a C b/
h.a C b0 /
for all b 2 A \ B (by niceness). Compatibility implies h.a C b0 /
h.a C x/
for some x 2 A \ B.
(ii) Given a 2 A, for an x 2 A\B, we have inequalities hG=B .aCB/ D hG .aCb/

hG .a C x/ D hA .a C x/
h.ACB/=B.a C B/:
(iii) Suppose .B0 =B/k.A C B/=B. Given a 2 A and b0 2 B0 , we have hG .a C b0 /

hG=B .a C b0 C B/
hG=B .a C x C B/ for some x 2 B0 \ .A C B/ D .B0 \ A/ C B,
which can clearly be chosen such that x 2 B0 \ A. Now hG=B .a C x C B/ D
hG .a C x C b/ for some b 2 B, and AkB implies that the last height is

hG .a C x C y/ for some y 2 B \ A: Since x C y 2 B0 \ A; the desired B0 kA

follows.
Conversely, assume B0 kA. Let again a 2 A and b0 2 B0 . Manifestly, hG=B .aC
b C B/ D hG .a C b0 C b/ for some b 2 B, which is
hG .x C b0 C b/ for some
0

x 2 B0 \ A; which in turn is
hG=B .x C b0 C B/: Here x C B 2 .B0 \ A/ C B D
B0 \ .A C B/. t
u
The next lemma is a convincing evidence that separative subgroups are compati-
ble with many subgroups.
7 Subgroups of Totally Projective p-Groups 381

Lemma 7.7 (Hill [17]). Given a separative subgroup A, and a subgroup X of G,

there is a subgroup B of G such that
(i) X
B,
(ii) jBj
jXj@0 ; and
(iii) AkB:
In particular, every countable subset of G can be embedded in a countable subgroup
B of G such that AkB.
Proof. Set B0 D X: For each x 2 B0 choose elements an .x/ 2 A .n < !/ such
that fhG .x C an .x// j n < !g is cofinal in fhG .x C a/ j a 2 Ag. Define B1 to be
generated by B0 and all the an .x/ 2 A .x 2 B0 ; n < !/: Repeat this argument for
B1 in the role of B0 , to obtain a subgroup B2 , and keep repeating this procedure !
times. Let B be the union of the chain B0 < B1 <    < Bi <    .i < !/: Then B
obviously satisfies (i) and (ii). Given a 2 A and b 2 B, there is an i such that b 2 Bi .
By construction, there is an an .b/ 2 A such that hG .b C a/
hG .b C an .b//. Since
an .b/ 2 BiC1
B; AkB follows. t
u
When Isotype Implies Totally Projective Compatibility has been introduced
to answer our question posed above. Here is the answer.
Theorem 7.8 (Hill [17]). An isotype subgroup A of a totally projective p-group G
is totally projective if and only if G admits a G.@0 /-family of .nice/ subgroups that
are compatible with A.
Proof. Suppose A is totally projective. Let A and G be G.@0 /-families of nice
subgroups of A and G, respectively, and define

H D fX 2 G j XkA and X \ A 2 Ag:

Since compatibility is preserved under unions of chains, in order to show that H is

likewise a G.@0 /-family, it suffices to ascertain that if X 2 H and if C is a countable
subgroup of G, then there is a Y 2 H containing both X and C such that Y=X is
countable. By Lemma 7.6(ii), .ACX/=X is isotype in G=X, and since it is isomorphic
to the totally projective group A=.A \ X/, by Theorem 7.4, it is separative in G=X.
From Lemma 7.7 we infer that there is a countable subgroup Y=X of G=X that
contains .C C X/=X and is compatible with .A C X/=X. Using a familiar back-and-
forth argument, it follows that Y may be chosen so as to satisfy both Y 2 G and
Y \ A 2 A. Hence Lemma 7.6(iii) shows that YkA, i.e. Y 2 H.
Conversely, let H D fX j XkAg be a G.@0 /-family in G. Since the intersection
of two G.@0 /-families is again one, we may even assume that H is a subfamily
of G (defined in the preceding paragraph), i.e. the subgroups in H are nice in G.
Evidently, A D fX \ A j X 2 Hg is a G.@0 /-family in A. By Lemma 7.6(i), X \ A
is nice in A for each X 2 H. Since A contains a G.@0 /-family of nice subgroups, it
is totally projective. t
u
From this theorem we derive several corollaries. But first, a definition.
382 11 p-Groups with Elements of Infinite Height

Let A be a subgroup of the p-group G. We will say that S D fSi j i 2 Ig (where

A
Si
G) is a G.@0 /-family of separative subgroups over A if each Si 2 S is
separative in G and fSi =A j i 2 Ig is a G.@0 /-family in G=A.
Corollary 7.9 (Hill [17]). For an isotype subgroup A of a totally projective p-group
G the following are equivalent:
(i) A is totally projective;
(ii) G admits a G.@0 /-family S of separative subgroups over A;
(iii) there is a G.@0 /-family T of totally projective groups over A.
Proof. (i) ) (ii) If A is totally projective, then by Theorem 7.8, G admits a G.@0 /-
family H of nice subgroups compatible with A. Without loss of generality, we may
moreover assume that each X 2 H satisfies X \ A 2 A for a G.@0 /-family A of
nice subgroups of A. Manifestly, .A C X/=X Š A=.A \ X/ is totally projective, and
hence separative in G=X. Then (F) implies A C X is separative in G; consequently,
S D fA C X j X 2 Hg is a G.@0 /-family of separative subgroups over A.
(ii) ) (iii) Let S be a G.@0 /-family of separative subgroups over A, and G a
G.@0 /-family of nice subgroups of G. We now claim that the set H D fX 2 G j XjjA
and A C X 2 Sg is a G.@0 /-family in G. (F) implies that the factor groups .A C X/=X
are separative in G=X, so Lemma 7.7 guarantees, for any given countable subgroup
C of G, the existence of a countable subgroup Y=X of G=X that contains .C C X/=X
and is compatible with .A C X/=X. Using back-and-forth arguments, we may even
assume that both Y 2 G and A C Y 2 S. Hence Lemma 7.6(iii) implies YkA, thus H
is in fact a G.@0 /-family. Clearly, T D fT 2 S j T D A C X with X 2 Hg is still a
G.@0 /-family of separative subgroups over A. In view of Lemma 7.6(i), X \ A with
X 2 H is nice in A, thus A D fX \ A j X 2 Hg is a G.@0 /-family of nice subgroups.
Evidently, T 2 T is the extension of X by .A C X/=X Š A=.A \ X/ where both
groups are totally projective, the latter because of X \ A 2 A. Since X is nice in
G; T is likewise totally projective.
(iii) ) (i) is trivial, since A 2 T . t
u
An immediate corollary follows by using (A):
Corollary 7.10 (Hill [17]). Isotype subgroups of a totally projective p-group of
countable lengths are again totally projective .thus direct sums of countable
groups/. t
u
We deduce one more consequence of our discussion above; this is concerned
with countable chains of totally projective subgroups.
Corollary 7.11 (Hill [17]). Let G0 < G1 <    < Gn <    be a countable
ascending chain of p-groups each of which is isotype in their union G. If every
Gn is totally projective, then so is G.
Proof. Let G0 be a G.@0 /-family of nice subgroups of G0 . Since G0 is totally
projective and isotype in G1 , by Lemma 7.6, G1 admits a G.@0 /-family G1 of nice
subgroups compatible with G0 . Arguing in the same way, we can find a G.@0 /-
family Gn of nice subgroups in Gn , compatible with the groups G0 ; : : : ; Gn1 .
7 Subgroups of Totally Projective p-Groups 383

Observe that since Gi contains Gn if i  n; Gi is trivially compatible with every

subgroup in Gn . Define

H D fH
G j H \ Gn 2 Gn for each n < !g:

A routine back-and-forth argument convinces us that H is a G.@0 /-family in G. We

claim that each H 2 H is nice in G. Given g 2 G; g 2 Gj holds for some index j.
As Hj D H \ Gj is nice in Gj , there is an a 2 Hj such that hG .g C x/
hG .g C a/
for all x 2 Hj . If b 2 Hn .n > j/, then by compatibility hG .g C b/
hG .g C c/ for
some c 2 Hj \ Hn D Hj . Thus hG .g C b/
hG .g C a/ for all b 2 H; and therefore
g C a is proper with respect to H. t
u
More Total Projectivity For later reference, we will find useful to have the
following lemma.
Lemma 7.12. Let C be a subgroup of the reduced p-group A.
(i) If C is finite, then A is totally projective if and only if so is A=C.
(ii) If C is of finite index in A, then A is totally projective if and only if so is C.
(iii) (Wallace [1]) If jA=Cj
@0 and C is totally projective, then A is totally
projective.
Proof.
(i) Finite subgroups being nice, the claim follows easily from Lemma 2.4.
(ii) Only the case jA W Cj D p needs to be considered. Then C is an extension of
pA by C=pA. Here pA is nice both in A and C, so C is totally projective if and
only if so is pA.
(iii) There is a countable subgroup H < A such that A D C C H, and since C is
totally projective, we may assume that C \ H belongs to a nice system in C.
Then  W c C .C \ H/ 7! c C H .c 2 C/ yields an isomorphism C=.C \ H/ Š
A=H, showing that A=H is totally projective. It remains to show that H is nice
in A. Let hp .c C H/ D  for some c 2 C. Then  implies hp .c C .C \ H// D ,
so by niceness there is x 2 C \ H such that hp .c C x/ D . Thus the element
c C x 2 A has the same height as its coset mod H. t
u
Relative Balanced-Projective Resolution The following relative resolution of
a p-group with respect to an isotype subgroup reveals another close relation between
total projectivity and separative chains. (For an analogous idea, see Sect. 3 in
Chapter 12.)
Lemma 7.13. Let A be a subgroup of the p-group G. There exists a balanced-exact
sequence

˛ 
0 ! B!A ˚ C!G ! 0 (11.11)
384 11 p-Groups with Elements of Infinite Height

where   A is the inclusion map, and C is a totally projective p-group. Then

Ker  D ˛B is isomorphic to the projection B0 of ˛B into C, and C=B0 Š G=A.
Furthermore,
(i) A is isotype in G if and only if B0 is isotype in C;
(ii) A is separative in G exactly if B0 is separative in C;
(iii) G admits a G.@0 /-family a separative subgroups over A .a separative chain
from A to G/ if and only if C has such a family over B0 .
Proof. The existence of a balanced-exact sequence with C and  as stated is rather
obvious; e.g. C in a balanced-projective resolution of G will do it (B is defined as
the kernel of A ˚ C ! G). Form the commutative diagram

B === B
⏐ ⏐
⏐α ⏐β
 
0 −−−−→ A −−−−→ A ⊕ C −−−−→ C −−−−→ 0
 ⏐ ⏐
 ⏐φ ⏐
  
0 −−−−→ A −−−−→ G −−−−→ G/A −−−−→ 0,

with exact rows and middle column; the 3  3-lemma ensures that the third column
is also exact. It is clear that C=B0 Š G=A where B0 D ˇB (ˇ is the projection map
onto C).
As B (identified with ˛B) is balanced in A ˚ C; A is isotype (separative) in
G D .A ˚ C/=B if and only if A ˚ B is isotype (separative) in A ˚ C. Because of
B \ A D 0, we have B Š ˇB D B0
C: Hence A ˚ B D A ˚ B0 . Thus A ˚ B is
isotype (separative) in A ˚ C exactly if B0 is isotype (separative) in C. (iii) follows
in the same way. t
u
We can now deduce at once:
Theorem 7.14. A p-group G admits a G.@0 /-family a separative subgroups over its
isotype subgroup A if and only if in a relative balanced-projective resolution (11.11)
the kernel B is totally projective. t
u
F Notes. Theorem 7.2 can be complemented by adding that factor groups of totally
projective groups modulo fully invariant subgroups are also totally projective. This too was proved
independently by Fuchs–E. Walker (unpublished) (see [IAG]) and Linton [1].
Both the definition and most of the results on separativity are due to Hill [17] (he used different
terminology: separability, which is used here in a different context). He also proved that the class
of all balanced subgroups of reduced totally projective p-groups is closed under passing to direct
sums, direct summands, and balanced subgroups. Moreover, for an ordinal  , a group A belongs to
this class exactly if both p A and A=p A do.
For uncountable cardinals , -separativity was studied by Fuchs–Hill [1] (in the definition of
separativity replace ‘countable’ by ) in order to determine the balanced-projective dimensions
of p-groups. Several results in this section carry over almost verbatim to -separativity. If
0 ! B ! A ! C ! 0 is a balanced-exact sequence with A a totally projective p-group,
8 p -Purity 385

then B has an H./-family of -separative subgroups if and only if C has an H. C /-family of
 C -separative subgroups. It was also proved that the balanced-projective dimension of p-groups
can be any integer
0 or 1; in particular, for each n
0, any p!n -high subgroup of a p-group A
of cardinality @n satisfying p!n A ¤ 0 has balanced-projective dimension exactly n. (A corollary is
that balanced subgroups in totally projective groups need not be totally projective.)
Hill [19] proved a strong isomorphism theorem on balanced subgroups of totally projective p-
groups: If B; B0 are balanced subgroups in the totally projective p-groups A; A0 such that they have
the same UK-invariants and satisfy A=B Š A0 =B0 , then there is an isomorphism A ! A0 carrying
B onto B0 . Warfield [4] proved a criterion on the isomorphism of two isotype, -dense subgroups,
C and C0 , of a totally projective p-group G of limit length . If dim GŒp
=CŒp
D dim GŒp
=C0 Œp
,
then there is an automorphism of G inducing an isomorphism C Š C0 .
Relative balanced-projective resolutions were introduced by Bican–Fuchs [Comm. Algebra 22,
1031–1036 (1994)] for modules. They will play a more substantial role in Chapter 14.
Some classes of isotype subgroups of totally projective p-groups have well-developed theories.
These include the classes of S-groups and A-groups. Let  denote a limit ordinal not cofinal with
!. An isotype subgroup H of a totally projective p-group G is called a -elementary S-group
if G=H Š Z.p1 / and G D H C p G for all  < . H is a -elementary A-group if G=H
is (possibly non-reduced) totally projective and p .G=H/ D .H C p G/=H for all  < . An
S-group (A-group) is the direct sum of -elementary S-groups (A-groups) for various ordinals .
S-groups can also be defined as torsion subgroups of Warfield groups. To classify S- and A-groups
(within the mentioned class, of course), one needs one invariant in addition to the UK-invariants.
The proofs are long and difficult. See Warfield [6], Hill [20], Hill–Megibben [6], as well as the
literature quoted there.

Exercises

(1) Show that compatibility is reflexive, but not transitive.

(2) If AkB and A \ B D 0, then h.a C b/ D minfh.a/; h.b/g for all a 2 A and b 2 B.
(3) There exists a separative chain from A to G if there is a smooth chain A D A0 <
   < A˛ <    < A D G of separative subgroups such that the cardinalities of
the factor groups A˛C1 =A˛ .˛ < / are
@1 .
(4) Let C be a subgroup of the p-group A such that nA
C
A for some n 2 N.
Then C is a direct sum of countable groups if and only if so is A.
(5) (Hill) Let A be a direct sum of countable groups, and suppose that the UK-
invariants of A are
@1 . An isotype subgroup of A is a direct sum of countable
groups if and only if it is separative in A.
(6) If A is totally projective, then every endomorphism of p A extends to an
endomorphism of A.

8 p -Purity

There exist several versions of generalized purity, each of which is motivated by a

particular feature of pure subgroups. In this section, we deal with one that depends
on an arbitrarily chosen ordinal, and seems to be of special interest from the
homological point of view.
386 11 p-Groups with Elements of Infinite Height

Definition of p -Purity Let  denote an ordinal or 1. The exact sequence e W

0 ! B ! A ! C ! 0 is called p -pure-exact, and the subgroup B of A p -pure if

e 2 p Ext.C; B/:

Our study begins with listing a few easy consequences of the definition (Irwin–
Walker–Walker [1], Nunke [3]).
(A) For n 2 N, B is pn -pure in A exactly if pk B D B \ pk A for each k
n. This
follows at once from Theorem 5.2(i). In particular, a p1 -pure subgroup is the
same as a p-neat subgroup.
(B) The p! -pure subgroups are exactly the p-pure subgroups in the old sense.
(C) If e W 0 ! B ! A ! C ! 0 is a p -pure-exact sequence, then for every group
G, the image of the connecting map Hom.G; C/ ! Ext.G; B/ is contained in
p Ext.G; B/. Indeed, the connecting homomorphism carries  2 Hom.G; C/
into e 2 p Ext.G; B/, as e is the image of e also under the map Ext.C; B/ !
Ext.G; B/ induced by .
Similarly, for every group H, Hom.B; H/ is mapped into p Ext.C; H/:
(D) If e is like in (C), and if ˇ W B ! B0 ; W C0 ! C are arbitrary homomorphisms,
then both sequences ˇe and e are p -pure-exact. This is routine.
Anticipating Theorem 8.2, we verify a generalization of Sect. 6(B) in
Chapter 5:
(E) (Irwin–Walker–Walker [1]) Let B be a p -pure .and hence p -balanced/
subgroup in A such that A=B is divisible. Then A=p A Š B=p B for all  < .
Now B satisfies both B C p A D \< .B C p A/ and p B D B \ p A for all
 < . Hence from B C pA D A by induction we obtain B C p A D A, thus
B=p B D B=.B \ p A/ Š .B C p A/=p A D A=p A for all  < .
The main features of purity are retained by this generalization as is shown
by (F)–(K) in addition to (C)–(E) [we skip the long proofs of (H) and (J)].
(F) If B is p -pure in A, and H
B, then B=H is p -pure in A=H. The canonical
map B ! B=H induces a homomorphism Ext.A=B; B/ ! Ext.A=B; B=H/,
which yields p Ext.A=B; B/ ! p Ext.A=B; B=H/, proving the claim.
(G) If B is p -pure in A, then it is p -pure in every subgroup H between B and
A. The injection H=B ! A=B induces the homomorphism p Ext.A=B; B/ !
p Ext.H=B; B/.
(H) Assuming H
B
A; if H is p -pure in B and B is p -pure in A, then H is
p -pure in A.
(J) If H is p -pure in A, and B=H in A=H; then H is p -pure in A:
(K) The exact sequence e W 0 ! B ! A ! C ! 0 is p1 -pure-exact if and
only if it is torsion-splitting. By definition, e is p1 -pure-exact if it belongs to
p1 Ext.C; B/. This Ext is a p-adic module, the last group is the divisible part
of Ext.C; B/. Hence the claim follows from Proposition 9.1 in Chapter 9.
p -Balanced Subgroups A subgroup B of A is called p -isotype if
8 p -Purity 387

p B D B \ p A for all  < ;

and p -nice if

p .A=B/ D .p A C B/=B for all  < :

A subgroup that is both p -isotype and p -nice will be called p -balanced. It is

straightforward to verify:
(a) B is p -isotype in A if and only if, for each  < , the map B=p B ! A=p A
induced by the inclusion is monic.
(b) B is p -nice in A if and only if the map A=p A ! C=p C induced by the
surjection A ! C D A=B is epic for all  < .
(c) B is p -balanced in A if and only if the exact sequence 0 ! B ! A ! C ! 0
implies the exactness of

0 ! p B ! p A ! p C ! 0 for all  < :

A more substantial criterion for p -balancedness is recorded in the next lemma.

Lemma 8.1. A subgroup B of A is p -balanced if and only if the exact sequence

0 ! B ! A!C ! 0 satisfies

.p AŒp
/ D p CŒp
for all  < : (11.12)

Proof. Suppose B is p -balanced in A. For any c 2 p CŒp

. < /, there is an
a 2 p A satisfying a D c. Now pa 2 pC1 A \ B D pC1 B, thus pb D pa
for a suitable b 2 p B. Evidently, a  b 2 p AŒp
is mapped by  upon c, thus
p CŒp

.p AŒp
/: The inclusion in the opposite direction being trivial, (11.12)
follows.
Conversely, assume (11.12) holds. The first stage is the verification of the
inclusion p CŒpk

.p A/ for all k 2 N. Hypothesis (11.12) takes care of the
case k D 1. Assume the inclusion holds for k  1. If c 2 p C is of order pkC1 , then
there is an x 2 A such that .x/ D c, and by induction hypothesis, some y 2 pC1 A
satisfies .y/ D pc 2 pC1 CŒpk
. Choose an a0 2 p A with pa0 D y, and notice
that .x  a0 / D c  .a0 / 2 p CŒp
. Consequently, c  .a0 / D .a1 / for some
a1 2 p AŒp
, and therefore, c D .a/ with a D a0 C a1 2 p A. This shows that
 W p A ! p C is onto.
For the rest, we refer to the last part of the proof of Proposition 5.5. t
u
As every subgroup is pn -nice (n 2 N/, pn -pure subgroups are also pn -balanced.
The same implication holds for all , but it is not as trivial.
Theorem 8.2.
(i) Every p -pure-exact sequence is p -balanced-exact.
388 11 p-Groups with Elements of Infinite Height


(ii) A p -balanced-exact sequence e W 0 ! B ! A!C ! 0 with C divisible is
p -pure-exact.
Proof.
(i) Suppose e is p -pure-exact. Pick c 2 p CŒp
; and invoking Corollary 4.6 extend
the isomorphism p HC1 D hgi ! hci (where HC1 is a generalized Prüfer
group and g 7! c) to a homomorphism W HC1 ! C (recall p HC1 is nice
and HC1 =p HC1 is simply presented). By the p -pure-exactness of e and the
pC1 -projectivity of HC1 [see Sect. 9.d/], there is W HC1 ! A such that
 D . Thus .g/ 2 p AŒp
maps upon c.
(ii) To prove the second claim, assume C is divisible, and e is p -balanced-exact.
We induct on  .
/ to show that e 2 p Ext.C; B/. For  D 1, the claim

T easily˛ established. If  is a limit ordinal, then because of p Ext.C; B/ D
is
˛< p Ext.C; B/, there is nothing to prove. So let  D  C 1 and suppose
.p˛ AŒp
/ D p˛ CŒp
D CŒp
for all ˛
. From the hypothesis it follows that
there is a subgroup H
p AŒp
such that AŒp
D BŒp
˚ H and H D CŒp
:
This leads to a commutative diagram

φ
: 0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐ṗ
 δ 
 β γ
: 0 −−−−→ B −−−−→ A/H −−−−→ C −−−−→ 0

where the middle (last) vertical map is the canonical one (multiplication by p),
ˇ is the obvious one, while .a C H/ D .pa/ for a 2 A.
Since e D pe0 , the proof will be complete if we can show that the bottom
row satisfies condition (11.12), i.e. .p˛ .A=H/Œp
/ D CŒp
for all ˛
. Pick a
c 2 CŒp
; by divisibility, there is a d 2 C such that pd D c. Now (11.12) implies
that there exists an x 2 H
p˛C1 AŒp
satisfying .x/ D c; let a 2 p˛ A be such
that pa D x. Then d  .a/ 2 CŒp
, so d  .a/ D .y/ for some y 2 H. The
element a0 D a C y satisfies a0 C H 2 .p˛ A=H/Œp
and .a0 C H/ D p.a/ D
pd D c. t
u
We interrupt our discussion of p -balancedness in order to verify an interesting
corollary to Theorem 8.2 that improves on Lemma 5.1.
Theorem 8.3 (Nunke [4]). p -high subgroups of p-groups are p C1 -pure .and
therefore isotype/.
Proof. As pn -high subgroups .n 2 N/ are summands, for the proof we may assume
  !. Let H be p -high in A. We claim .A=H/Œp
D .p AŒp
C H/=H for each

. To justify this, first note that the inclusion  is obvious. So suppose a 2 AnH
satisfies pa 2 H. By maximality, hH; ai \ p AŒp
¤ 0, whence a 2 p AŒp
C

H
p AŒp
C H, in fact. Hence the exact sequence 0 ! H ! A!A=H ! 0
8 p -Purity 389

satisfies (11.12) for  C 1 (rather than for ). Since   !, A=H is divisible, so the
claim follows in view of Theorem 8.2. That H is then isotype is evident. t
u
The following example exhibits an interesting p -pure subgroup.
Example 8.4. Consider an epimorphism  W H ! Z.p1 / where H is the generalized Prüfer
group of length 
!, and  maps a generator (in a simple presentation) of H of order pn upon
the generator cn of order pn of Z.p1 /. We claim that M D Ker  is p -pure in H . To demonstrate
that the exact sequence

0 ! M ! H !Z.p1 / ! 0

has property (11.12), note that since Z.p1 / is divisible, the right-hand side of (11.12) is just the
socle of Z.p1 / for each  <  . From the definitions of H and the map , it is clear that (11.12)
is trivially satisfied whenever  is a successor ordinal: the non-zero elements of height  in H are
mapped upon non-zero elements of the socle of Z.p1 /. If  is a limit ordinal, then H is the direct
sum of groups H for  <  , and the property is obvious. Hence the sequence is p -balanced-, and
so by Theorem 8.2(ii) also p -pure-exact.

We continue with pointing out remarkable connections with Tor.


Lemma 8.5. If the exact sequence 0 ! B ! A!C ! 0 is p -balanced for some
  !, then so is the induced exact sequence


0 ! Tor.X; B/ ! Tor.X; A/! Tor.X; C/ ! 0

for every group X.

Proof. The exactness of this sequence is guaranteed by the p-purity of the first
sequence (Theorem 3.1 in Chapter 8). Keeping Lemma 8.1 in mind, it suffices to
show that this sequence satisfies condition (11.12). In view of Lemmas 4.2 and
4.3 in Chapter 8, p Tor.X; A/Œp
D Tor.p XŒp
; p AŒp
/; this group is isomorphic
to the direct sum of  copies of p AŒp
, where  D dim.p XŒp
/. Similar assertion
holds for p Tor.X; C/Œp
, and by definition, the induced map carries copies of p AŒp

upon corresponding copies of p CŒp

. Thus (11.12) for the given sequence implies
that (11.12) holds for the sequence of the Tors. t
u
Proposition 8.6 (Keef [7]). An exact sequence 0 ! B ! A ! C ! 0 of p-groups
is p -pure for an infinite ordinal  if and only if the induced sequence

0 ! Tor.H ; B/ ! Tor.H ; A/ ! Tor.H ; C/ ! 0

is splitting .H is the generalized Prüfer group of length /.

Proof. First suppose the given sequence is p -pure. Then the sequence of the
Tors is exact, since   ! (cf. Theorem 3.1 in Chapter 8). It is even p -
pure-exact, because by Theorem 3.8 in Chapter 9 Ext.Tor.H ; C/; Tor.H ; B// Š
Ext.H ; Ext.C; Tor.H ; B//, and H is p -projective (Sect. 9.d/). Taking into con-
390 11 p-Groups with Elements of Infinite Height

sideration that by Sect. 9.f / the Tors are p -projective, the claim in one direction
follows.
Conversely, assume that the sequence of Tors splits, so that there is a splitting
map Tor.H ; C/ ! Tor.H ; A/. The exact sequence in Example 8.4 yields a
surjective map Tor.H ; A/ ! Tor.Z.p1 /; A/ Š A. Then with the help of the
composite of these two maps we form a commutative square on the right, and
complete it to a commutative diagram

0 −−−−→ Tor(Mσ , C) −−−−→ Tor(Hσ , C) −−−−→ C −−−−→ 0

⏐ ⏐ 
⏐ ⏐ 
  
0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0

where the top row is by Example 8.4 and Lemma 8.5 p -pure-exact (M is from
Example 8.4). Owing to (D), the same holds for the bottom row. t
u
C -Groups p -purity plays an important role in the theory of C -groups; these
groups were introduced by Megibben [5]. Let  denote a countable limit ordinal.
A p-group A is called a C -group if, for each ordinal  < , A=p A is a direct sum
of countable groups.
Example 8.7.
(i) Every p-group A is a C! -group, since A=pn A is †-cyclic for all n < !.
(ii) A direct sum of countable p-groups of lengths <  .< !1 / is a C -group.

The following observations are obvious for a fixed .

(a) Direct sums and summands of C -groups are again C -groups.
(b) Ulm factors of C -groups are †-cyclic groups.
(c) If A is a C -group, then p A is a C0 -group for some 0 , and A=p A is a C -
group provided  is a limit ordinal.
(d) p -pure subgroups of a C -group are likewise C -groups. If G is a p -pure
subgroup of a C -group C, then it is p -balanced [see Theorem 8.2(i)], so it
satisfies G \ p C D p G for all  < . Then .G C p C/=p C is isotype in
C=p C, since

Œ.G C p C/ \ p C
=p C D Œ.G \ p C/ C p C
=p C D .p G C p C/=p C

for all  < . As C=p C is now a direct sum of countable groups, by

Corollary 7.10, G=p G Š .G C p C/=p C is also a direct sum of countable
groups.
-Basic Subgroups Let again  be a countable limit ordinal. A subgroup B of
a p-group A is said to be a -basic subgroup (Megibben [5]) if
(i) it is a direct sum of countable groups of lengths < ,
(ii) it is p -pure in A; and
8 p -Purity 391

(iii) A=B is a divisible group.

Since properties (ii) and (iii) imply that A=p A Š B=p B, it is clear that the UK-
invariants of B are determined by the UK-invariants of A=p A. Hence we derive the
uniqueness of -basic subgroups up to isomorphism, provided they exist.
Theorem 8.8 (Megibben [5]). A p-group admits a -basic subgroup if and only if
it is a C -group.
Proof. Let A be a p-group with -basic subgroup B. From the isomorphism
A=p A Š B=p B the necessity is evident. Conversely, assume A is a C -group. For
each  < , let K denote a p -high subgroup of A; there is no loss of generality
in assuming that the K form a chain. Our claim is that B D [ < K is a direct
sum of countable groups. As each K is isotype in A, the same holds for their union.
Furthermore, K is isomorphic to an isotype subgroup of a direct sum A=p A of
countable groups, so Corollary 7.10 implies that K is likewise a direct sum of
countable groups. Thus we have a countable chain of isotype subgroups of B, each
is a direct sum of countable groups of length < . It only remains to appeal to
Corollary 7.11 to conclude that B satisfies condition (i).
The rest will follow from the obvious containment AŒp

BŒp
C p A for all
 < . This, along with B \ pA D pB, implies that A=B is divisible. It also implies
that condition (11.12) holds for B for all  < , so B is p -balanced in A. Then (ii)
follows from Theorem 8.2(ii). t
u
F Notes. p -purity is a natural generalization of purity, and its importance lies also in its role
which it had in leading to the concept of total projectivity. The properties we have not proved in
details will not be needed in the rest of this volume.
Nunke [3] defines S-purity for a subfunctor S of the identity, and proves several theorems
generalizing results on p -purity. (In the definition, p Ext is replaced by S-Ext.) For more relevant
results, see Irwin–Walker–Walker [1] and Nunke [3].
Cutler [3] proves the existence of a p-group for which it is undecidable in ZFC whether or not
all of its p! -high subgroups are isomorphic. Isomorphism of high subgroups is known, however,
in special cases.
The theorems on C -groups are straightforward generalizations of the classical theory. Meg-
ibben [5] unveiled several additional features of interest, like ‘separability’: every finite subset is
contained in a countable direct summand, or that a p -pure subgroup of length <  is a summand.
Though the theory sheds more light on Kulikov’s theorems on p-groups, so far it has not produced
exciting results. The book Salce [S] contains a lot of information on C -groups. Generalizations
have been discussed by Crawley [4] and Albrecht [5] for uncountable limit . Linton [2] introduces
-large subgroups as fully invariant subgroups L of A such that L C B D A for all -basic B. If A
is a C -group, then A=L is simply presented.
A surprising result on C!1 -groups of length !1 is due to Keef [1]. He proved that the torsion
product of two such groups is a direct sum of countable groups if and only if the so-called Kurepa
hypothesis fails. (This principle, which holds in L, says that there is a tree of height !1 with
countable levels and @2 branches. It is independent of ZFC+GCH.)
392 11 p-Groups with Elements of Infinite Height

Exercises

(1) (Nunke) If 0 ! B ! A ! C ! 0 is a p -pure-exact sequence of p-groups,

then 0 ! Tor.B; X/ ! Tor.A; X/ ! Tor.C; X/ ! 0 is likewise p -pure-exact.
(2) (Keef) An exact sequence 0 ! B ! A ! C ! 0 is p -pure-exact if and only
if the natural map Tor.H ; C/ ! C factors through A ! C.
(3) If C is isotype in A, then for every ordinal , p C is isotype in p A.
(4) (Irwin–Walker–Walker) If B; C are subgroups in a p-group A such that B \ C is
p -pure in A, then C is p -pure in B C C.
(5) An infinite subgroup of a C -group is contained in a p -balanced subgroup of
the same cardinality.
(6) The UK-invariants f .H/ of a p -high subgroup H of A are equal to the
corresponding UK-invariants f .A/ of A provided  < .
(7) (Hill) The exact sequence 0 ! B ! A ! C ! 0 is p -pure-exact if
AŒp
D BŒp
C p AŒp
holds for all  < . [Hint: if   !, C is divisible,
apply Theorem 8.2(ii).]

9 The Functor p

In this section we are interested in p-groups, and therefore we will restrict

our discussion to p-groups, though sometimes we might get a more meaningful
statement without this restriction.
We call a group A p -projective ( an ordinal or 1) if

p Ext.A; G/ D 0 for all p-groups G:

In Sect. 6, we gave a slightly different definition by allowing G to be any group,

but for p-groups A, this change does not make any difference. Indeed, this is obvious
for  2 N, and for   ! it follows from the natural isomorphism Pext.A; G/ Š
Pext.A; tG/ (Sect. 5 in Chapter 9, Exercise 7).
Example 9.1. Evidently, A=p A is p -projective for every ordinal  whenever A is a totally
projective p-group.

p -Projectivity We have the following easily verified properties.

.a/ A p -projective group has the projective property with respect to p -pure-exact
sequences. From the definition it is obvious that C is p -projective if and only if
all p -pure-exact sequences of the form e W 0 ! B ! A ! C ! 0 are splitting.
If C is arbitrary and  W P ! C is any map from a p -projective p-group P, then
e is p -pure-exact as well [see Sect. 8(D)], so splitting. Hence there exists a
map P ! A required to establish p -projectivity.
.b/ Direct sums and summands of p -projective groups are again p -projective.
9 The Functor p 393

.c/ A p-group is pn -projective for an integer n if and only if it is pn -bounded. It is

p! -projective if and only if it is †-cyclic. This follows at once from Theorem 4.3
in Chapter 5.
.d/ The generalized Prüfer group H of length  is p -projective. Once we know
that H is totally projective, this becomes obvious.
.e/ If in the exact sequence 0 ! G ! F ! A ! 0; F is a †-cyclic p-group and
G is pn -bounded, then A is p!Cn -projective. To prove this, observe that in the
induced exact sequence Hom.G; ?/ ! Ext.A; ?/ ! Ext.F; ?/ ! Ext.G; ?/
the groups at both ends are pn -bounded and p! Ext.F; ?/ D 0: Therefore
p!Cn Ext.A; ?/ D 0 [cf. Sect. 1(A)–(B) in Chapter 10].
.f / If A is p -projective, then so is Tor.A; C/ for every C. In the natural isomorphism

Ext.Tor.A; C/; B/ Š Ext.A; Ext.C; B//

(see Theorem 3.8 in Chapter 9), we have p Ext.A; Ext.C; B// D 0 for every B.
.g/ The length of a p -projective p-group is at most . (This has been proved in
Sect. 6(B), here we give a different proof.) The claim is obvious whenever  is
an integer, so assume   !. From the p -pure-exact sequence in Example 8.4
we derive the exact sequence 0 ! Tor.M ; A/ ! Tor.H ; A/ ! A ! 0 which
has to be p -balanced as stated in Proposition 8.6. The length of Tor.H ; A/
cannot exceed  (see Lemma 4.2 in Chapter 8), hence the criterion (11.12) in
Lemma 8.1 ensures that p A D 0.
.h/ All torsion groups are p1 -projective. This follows from Theorem 6.5(ii) in
Chapter 9.
Enough p -Projectives By a p -projective resolution of a p-group A is meant
a p -pure-exact sequence 0 ! G ! P ! A ! 0 where P is a p -projective p-
group. .h/ shows that p1 -projective resolutions of p-groups are trivial, so we deal
only with the case when  is an ordinal.
Theorem 9.2 (Nunke [5]). For each ordinal   !, every p-group A admits a
p -projective resolution.
Proof. For   !, consider the exact sequence 0 ! Z ! N ! H ! 0 where
N is the Nunke group for the prime p (cf. Sect. 1 in Chapter 15). For any p-group
A, there is an induced exact sequence

ı
e W 0 ! Tor.N ; A/ ! Tor.H ; A/!A ! 0: (11.13)

In fact, Tor.Z; A/ D 0 holds in the induced long tensor-Tor exact sequence, and the
surjective map N ˝ A ! H ˝ A is actually an isomorphism, since the elements of
the subgroup Z have infinite heights in N , so in the tensor product they annihilate
the p-group A.
We know from .f / that Tor.H ; A/ is p -projective, so in order to confirm
that (11.13) is a p -projective resolution of A, it only remains to verify that e 2 p T
where T D Ext.A; Tor.N ; A//. The map ı in (11.13) induces a homomorphism
394 11 p-Groups with Elements of Infinite Height

ı  W T ! Ext.Tor.H ; A/; Tor.N ; A//

Š Ext.H ; Ext.A; Tor.N ; A/// D Ext.H ; T/:

ı was the natural map for A induced by the exact sequence 0 ! Z ! N ! H !

0, so ı  must also be the natural map for T. By Lemma 1.5 in Chapter 15, the kernel
of ı  is p T. We now form the extension eı via pull-back:

δ : 0 −−−−−→ Tor(Nσ , A) −−−−−→ G −−−−−→ Tor(Hσ , A) −−−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  δ
δ
: 0 −−−−−→ Tor(Nσ , A) −−−−−→ Tor(Hσ , A) −−−−−→ A −−−−−→ 0

Observe that the top row splits by Lemma 2.1 in Chapter 9 as the identity map of
Tor.H ; A/ furnishes a map between its two copies for the commutative triangle.
This tells us that e 2 Ker ı  D p T, as desired. t
u
We are at once led to the following corollary.
Corollary 9.3 (Nunke [5]). Assuming   !, a p-group A is p -projective if and
only if it is a summand of Tor.H ; A/. t
u
Example 9.4. There exist p-groups that are not p -projective for any  . A good example is
any reduced p-group A that is not starred, i.e. has larger cardinality than its basic subgroup.
It is trivially true that such an A is not pn -projective for any n  !. Let 
!. In view of
Lemma 4.8 in Chapter 8, all subgroups of Tor.H ; A/ have the same cardinality as their basic
subgroups. Therefore, A cannot be a summand of Tor.H ; A/, though it were if it was p -projective
Corollary 9.3.

On p -Projective p-Groups We record more useful results on p -projectives.

Proposition 9.5 (Nunke [5]). A p C1 -pure subgroup of a p -projective p-group is
a summand .and hence it is also p -projective/.
Proof. Consider a p C1 -pure-exact sequence 0 ! C ! A ! A=C ! 0 where A is
p -projective. In the induced exact sequence [cf. Sect. 8(C)]
˛
Hom.C; C/!p Ext.A=C; C/ ! p Ext.A; C/

the last term is 0 in view of the hypothesis on A, thus ˛ is an epimorphism.

Since C is p C1 -pure, Im ˛
p C1 Ext.A=C; C/ [see Sect. 8(C)], and therefore
p Ext.A=C; C/ D p C1 Ext.A=C; C/ is a p-divisible group. But if A=C is a p-group,
then Theorem 6.5 in Chapter 9 implies Ext.A=C; C/ is reduced. Consequently,
p Ext.A=C; C/ D 0, and the given exact sequence splits. t
u
Corollary 9.6 (Nunke [5]). If the p-group A is not p -projective, then
Tor.H C1 ; A/ is p C1 -, but not p -projective.
9 The Functor p 395

Proof. If  2 N, then the claim is pretty clear. If   !, then consider the p C1 -

pure-projective resolution 0 ! Tor.N C1 ; A/ ! Tor.H C1 ; A/ ! A ! 0. If
Tor.H C1 ; A/ were p -projective, then the sequence would split (Proposition 9.5).
Hence A would be p -projective, contradicting the hypothesis. t
u
Example 9.7. There are separable p -projective p-groups for every ordinal  . In fact, if A is any
separable p-group, then Tor.H ; A/ is separable and p -projective. Actually, if A is a group like
the one in Example 9.4, then for distinct ordinals  , Tor.H ; A/ will supply groups with different
levels of pure-projectivity, as is evident from Proposition 9.5.

Proposition 9.8 (Irwin–Walker–Walker [1]). If  is a limit ordinal, then a p -

projective p-group is a summand of a direct sum of p -projective groups for  < .
Proof. Let P be p -projective. For each  < , select a p -projective resolution

0 ! H ! P !P ! 0 with p -projective P . Form the exact sequence 0 !

K ! ˚< P !P ! 0 with  D r.˚  /; K D Ker . It is easily seen that this
sequence is p -pure-exact for all  < , so it is also p -pure-exact. Hence it splits,
and the claim follows. t
u
p -Injectivity We now move on to consider the dual notion. A p-group A is
called p -injective ( an ordinal or 1) if

p Ext.G; A/ D 0 for all p-groups G: (11.14)

It is totally injective if

p Ext.G; A=p A/ D 0 for all ordinals  and p-groups G: (11.15)

The following properties are easily verified from the definitions.

(A) A p -injective group has the injective property with respect to p -pure-exact
sequences. The proof is dual to the proof in .a/.
(B) If A is a totally injective p-group, then so are all the factor groups A=p A.
(C) Summands and torsion parts of products of p -injective .totally injective/ p-
groups are again p -injective .totally injective/.
(D) A reduced p-group is pn -injective if and only if it is pn -bounded. The reduced
p! -injective p-groups are precisely the torsion-complete groups. This is a
consequence of Theorem 3.6 in Chapter 10.
(E) The length of a reduced p -injective p-group cannot exceed . If A has length
> , then A Š t Ext.Z.p1 /; A/ shows that A cannot be p -injective.
(F) t Ext.C; A/ is p -injective for every C provided so is A. This follows from
the isomorphism Ext.B; Ext.C; A// Š Ext.Tor.B; C/; A/, since the functor
p annihilates the right side, so the left side is 0 for every p-group B. (See
Theorem 3.8 in Chapter 9.)
(G) If C is a p -projective p-group . an ordinal/, then t Ext.C; A/ is p -
injective for every group A. This is clear in view of the isomorphism
Ext.B; Ext.C; A// Š Ext.C; Ext.B; A//.
396 11 p-Groups with Elements of Infinite Height

(H) All p-groups are p1 -injective. This is obvious from Theorem 6.5(ii) in
Chapter 9.
Example 9.9. A separable p-group is totally injective .p -injective for an ordinal 
!/ if and
only if it is torsion-complete. The total injectivity (the p -injectivity) of a separable p-group A
amounts to p! Ext.X; A/ D Pext.X; A/ D 0 for all p-groups X. Because of Theorem 3.6 in
Chapter 10, this holds for A exactly if A is torsion-complete.

One more result of interest on totally injective p-groups.

Proposition 9.10 (Fuchs–Salce [1]). The Ulm factors of a reduced totally injective
p-group are torsion-complete.
Proof. For a totally injective p-group A, the exact sequence 0 ! A ! A !
A=A ! 0 yields the exact sequence

0 D Hom.Z.p1 /; A=A / ! Ext.Z.p1 /; A / !

! Ext.Z.p1 /; A/ ! Ext.Z.p1 /; A=A / D 0I

see Lemma 5.4 in Chapter 9. Consequently, Ext.Z.p1 /; A / Š Ext.Z.p1 /; A/

holds for every ordinal . Applying this isomorphism to the totally injective
group A=AC1 , and passing to the Ulm subgroup, we obtain the isomorphism
Ext.Z.p1 /; A =AC1 /1 Š Ext.Z.p1 /; A=AC1 /C1 D 0. Hence we conclude
Pext.Z.p1 /; A =AC1 / D 0, so the th Ulm factor A =AC1 (for any ordinal )
is torsion-complete (Corollary 3.7 in Chapter 10). t
u
Finally, we mention without proof that p -injective resolutions exist for all
  !.
F Notes. Since we have ignored torsion-free groups in the definition of p -projectivity and
-injectivity, some results sound different from those in the literature.
We have proved above that a p -projective p-group behaves like a totally projective group in as
much as for a limit ordinal  , it is a summand of a direct sum of p -projectives for  <  . Similar
holds for p -injectives, with direct sum replaced by torsion subgroup of the direct product.
p-groups A satisfying p Ext.Z.p1 /; A=p A/ D 0 for all  were investigated by Salce [1]. His
results are similar to the totally injective case.
Several authors discussed the p -topology for limit ordinals , especially completions in this
topology. See, e.g., Mines [1] and Dubois [1]. If cf  > !, the situation is different from the
familiar metrizable topologies (e.g., Exercises 7 and 8).

Exercises

(1) (a) A p-group A is pn -projective if and only if it is pn -bounded.

(b) It is pn -injective exactly if A D B ˚ D where pn B D 0 and D is divisible.
(2) (Irwin–Cellars–Snabb) A p -projective p-group is embeddable in a totally
projective p-group of length . [Hint: A ! D implies Tor.A; H / ! ˚ H
is monic, A a summand of Tor.]
10 p!Cn -Projective p-Groups 397

(3) Let A be a p-group such that A=p A is p -injective, and p A is pn -bounded.

Then A is p Cn -injective.
(4) A p-group A is p -injective if and only if it is isomorphic to a summand of
t Ext.H ; A/.
(5) If  is a limit ordinal, then a p -injective p-group is a summand of the torsion
subgroup of a direct product of p -injective groups for  < . [Hint: dualize the
proof of Example 9.4.]
(6) (Fuchs–Salce) A p-group A satisfying p Ext.Z.p1 /; A=p A/ D 0 for all
ordinals  is called almost totally injective. Show that A is almost totally
injective if and only if for an ordinal , both p A and A=p A are almost totally
injective. [Hint: use the long exact sequence in Corollary 9.6.]
(7) (Waller) Let A D ˚i2I Ai be of length !1 where each Ai is a countable p-group.
A is complete in the p!1 -topology.
(8) (Salce) There exists a p-group of length !1 , whose completion in the p!1 -
topology is not complete in its own p!1 -topology. [Hint: in a p!1 -balanced
projective resolution of H!1 C1 , take the completion of the kernel.]

10 p!Cn -Projective p-Groups

The initial interest in these groups was sparkled by Nunke’s result that showed
that they are composed of a †-cyclic and a bounded group. Moreover, the p!Cn -
projective p-groups (for integers n  1) share some properties which fail for
p -projectives with large .
Characterization We prove a main result of p!Cn -projective p-groups.
Theorem 10.1 (Nunke [5]). For a p-group A, the following are equivalent:
(i) A is p!Cn -projective;
(ii) A contains a pn -bounded subgroup P such that A=P is †-cyclic;
(iii) A Š F=G where F is a †-cyclic p-group and pn G D 0.
Proof.
(i) ) (ii) We start with the exact sequence

0 ! p! H!Cn ! H!Cn ! H! ! 0

which implies the exactness of

0 ! Tor.p! H!Cn ; A/ ! Tor.H!Cn ; A/ ! Tor.H! ; A/:

In this sequence, the first Tor is pn -bounded, and the third is †-cyclic. Hence
Tor.H!Cn ; A/ is an extension of a pn -bounded group P by a †-cyclic group. If
A is p!Cn -projective, then it is a summand of this Tor (see Corollary 9.3), and
as such it is likewise an extension of a pn -bounded group by a †-cyclic group.
398 11 p-Groups with Elements of Infinite Height

(ii) ) (iii) Let fai j i 2 Ig be a set of representatives of generators of A mod

P, and define C D ˚i2I hci i as a †-cyclic group where we let ci have the
same order as ai . Set F D C ˚ P0 with P0 Š P, so F is †-cyclic. Define an
epimorphism  W F ! A by sending ci to ai and mapping P0 isomorphically
upon P. It follows at once that G D Ker  is pn -bounded.
(iii) ) (i) This implication was already proved in Sect. 9.e/. t
u
Since property (ii) is inherited by subgroups, we can state:
Corollary 10.2. Subgroups of p!Cn -projective p-groups are themselves p!Cn -
projective. t
u
It should be emphasized that the pn -bounded subgroup P in condition (ii) is not
uniquely determined by A, not even up to isomorphism. But we can claim that if P1
and P2 are pn -bounded subgroups such that both A=P1 and A=P2 are †-cyclic, then
A=.P1 \ P2 / is again a †-cyclic group. In fact, it is a subgroup in the †-cyclic group
A=P1 ˚ A=P2 .
Example 10.3. For any subsocle S of a †-cyclic p-group A, the group A=S is p!C1 -projective. It
can be separable or not, depending on the closure property of S.

The next result is a noteworthy generalization of Theorem 3.5 in Chapter 10.

Lemma 10.4 (Danchev [1]). Let A be a separable p-group, and C a p!Cn -
projective pure subgroup. If A=C is countable, then A too is p!Cn -projective.
Proof. C contains a pn -bounded subgroup P such that C=P is †-cyclic. The p-adic
closure P of P in A is likewise pn -bounded (Lemma 7.6 in Chapter 1), and since
C=P is pure in A=P, and has no elements of infinite height, we have C \ P D P:
Again, C=P is pure in A=P, and it stays pure if we pass modulo .A=P/1 D P =P.
Thus .C C P /=P Š C=P is a pure †-cyclic subgroup of countable index in the
separable p-group A=P , whence by Theorem 3.5 in Chapter 10 we conclude that
A=P is †-cyclic, and thus A is p!Cn -projective. t
u
Isomorphism of p!Cn -Projective p-Groups Another noteworthy result on
!Cn
p -projectivity is an isomorphism theorem.
Theorem 10.5 (Fuchs [20]). Two p!Cn -projective p-groups, A and C, are isomor-
phic if and only if there is a height-preserving isomorphism between AŒpn
and CŒpn
.
Proof. Only sufficiency requires verification. Let W AŒpn
! CŒpn
denote a
height-preserving isomorphism. In view of Theorem 10.1, there are subgroups
P
AŒpn
and Q
CŒpn
such that A=P and C=Q are †-cyclic groups.
First, we show that P can be replaced by P0 D P\ 1 Q and Q by Q0 D P\Q,
so that .P0 / D Q0 can be assumed. We have to ascertain that A=P0 and C=Q0 are
†-cyclic. Take into account that .A=P0 /=.P=P0 / Š A=P is †-cyclic, and so P=P0 is
nice in A=P0 , and the correspondence x C P0 7! x C Q .x 2 P/ is a monomorphism
 W P=P0 ! C=Q not decreasing heights. Therefore,  extends to a homomorphism
A=P0 ! C=Q. Hence there is a map A=P0 ! A=P ˚ C=Q which is easily seen to be
monic. Thus A=P0 is †-cyclic as a subgroup of a †-cyclic group.
10 p!Cn -Projective p-Groups 399

We next show that the Hill invariants of A relative to P can be computed by

using only elements of AŒpn
. For each ordinal , we have .p C1 A C P/ \ p AŒp
D
.p C1 A C P/Œp
\ p AŒp
/, therefore it is enough to prove that .p C1 A C P/Œp

.p C1 AŒpn
C P/Œp
: But this is clear since if p.a C x/ D 0 where a 2 p C1 A and
x 2 P, then pn a D pn x D 0 implies a 2 p C1 AŒpn
:
To finish the proof, we may argue under the assumption .P/ D Q. Manifestly,
P is nice in A, and Q is nice in C. Since the Hill invariants can be computed in terms
of the pn -socles, those of P in A and of Q in C have to be equal. It only remains
to appeal to Theorem 3.6 to conclude that can be extended to an isomorphism
A ! C. t
u
It is interesting that whether or not a p-group A is p!Cn -projective is completely
determined by AŒpnC1
as a group valuated by heights in A (for valuated groups, cf.
Sect. 3 in Chapter 15).
Proposition 10.6. A p!Cn -projective p-group A can be reconstructed from the
valuated group AŒpnC1
.
Proof. Given A, consider AŒpnC1
as valuated by the heights in A. For A to be p!Cn -
projective, it is necessary that there exists a pn -bounded subgroup P such that the
socle of AŒpnC1
=P is a free valuated vector space with integer values. To see that
if such a P exists, then A is p!Cn -projective, observe that it is straightforward to
construct an extension G of P by a †-cyclic group C such that the restriction to CŒp

is isometric to the socle of AŒpnC1

=P. This G is p!Cn -projective by construction.
As P is nice in A, and the Hill invariants of P in A up to n C 1 are computable in
AŒpnC1
, and are trivially 0 for > n C 1, from Theorem 3.6 we can conclude that
A Š G. t
u
Ubiquity of p!C1 -Projective Subgroups The following result is a remarkable
evidence of the ubiquity of proper p!C1 -projective p-groups. (By a proper p!Cn -
projective group we mean a p!Cn -projective group that is not p!Cn1 -projective.)
Theorem 10.7 (Benabdallah–Irwin–Rafiq [1]). Every separable p-group which
is not †-cyclic contains a proper p!C1 -projective subgroup.
Proof. Suppose A is a separable p-group that is not †-cyclic. Splitting off a bounded
summand if necessary, we may assume that fin rk.A/ D jAj. Let B be a lower basic
subgroup of A, i.e. jA=Bj D jAj. We can represent A=B as the countable union of
divisible summands An =B where A0 D B and jA=An j D jAj .n < !/. Then the
An form a chain of pure subgroups with union A, so Theorem 5.5 in Chapter 3
guarantees the existence of an index m  1 such that Am is not †-cyclic.
Let 0 ! H ! F ! Am ! 0 be a pure-projective resolution of Am where F
is a †-cyclic p-group which is chosen such that fin rk.F/ D jAm j
jAj. Clearly,
F=.HŒp
/ is a p!C1 -projective group which—because of HŒp
D H \ FŒp
—embeds
in the direct sum F=H ˚ F=.FŒp
/. The group F=.HŒp
/ is not †-cyclic, because
otherwise the embedding HŒp
! H viewed as a height-preserving map of a nice
subgroup of F would extend to a homomorphism  W F ! H. It is readily checked
that  would then induce an automorphism of H, so that H were a summand of
F—which is obviously not the case. Since A=Am is divisible and has cardinality jAj,
400 11 p-Groups with Elements of Infinite Height

it contains a subgroup G=Am isomorphic to the †-cyclic group F=.FŒp

/. As Am is
pure in G, we get G D Am ˚ C with C Š F=.FŒp
/. Thus the proper p!C1 -projective
group F=.HŒp
/ is isomorphic to a subgroup of G, and hence of A. t
u
F Notes. The 1962 New Mexico conference (the first US conference on abelian groups)
recorded a substantial progress in the research of p -purity and p -projectivity, due to Irwin–
Walker–Walker [1] and Nunke [3]. Some of their results have been discussed in this volume.
That p!C1 -projective p-groups can be studied successfully via valuation was observed by Fuchs–
Irwin [1]. It was shown that a separable proper p!C1 -projective p-group A admits a decomposition
A D C ˚ A0 where C is †-cyclic of fin rk C D fin rk A and A0 Š A. Cutler [2] proves that a
p!Cn -projective p-group is determined in the class of all p-groups by its pnC1 -socle, but not by its
pn -socle. Cutler–Irwin–Snabb [1] show that if the reduced p-group A is not starred, then for every
n < !, A contains a proper p!Cn -projective p-group. For more on p!Cn -projective p-groups, see
Cutler–Missel [1]. p!2 -projectives have been studied, but with not much success.
It is an old unsettled problem to describe extensions of †-cyclic by †-cyclic, and
more generally, extensions of direct sums of countable groups by groups of the same
kind. In the unbounded case such extensions need not share any projective property.
Even if C is an essential subgroup of A, and if both C and A=C are †-cyclic p-
groups, A need not be a direct sum of countable groups. In fact, let A be a sepa-
rable p!Cn -projective p-group which is not †-cyclic, B a pn -bounded subgroup with
†-cyclic A=B. If the subgroup C of A is defined so as to have C=B D .A=B/Œp
, then pnC1 C D 0.
Hence C is †-cyclic, is essential in A, and A=C Š p.A=B/ is †-cyclic. But A is not a direct sum
of countable groups (and thus not totally projective), for otherwise—as a group with no elements
of infinite height—it would be †-cyclic.
Fuchs–Salce [2] consider p-groups A; C of length  C n, where  is a limit ordinal and n is a
positive integer. It is shown that A Š C provided there is an isomorphism  W A=p A ! C=p C
such that .AŒpn
=p A/ D CŒpn
=p C. Keef–Danchev [1] study the following generalization: call
a p-group A n-simply presented if it contains a pn -bounded subgroup P such that A=P is simply
presented. They call an exact sequence 0 ! A ! B ! C ! 0 n-balanced if it represents
an element of pn Bext.C; A/, and C n-balanced-projective if pn Bext.C; A/ D 0 for all A, and
prove that the n-balanced-projective p-groups are precisely the summands of n-simply presented
p-groups.
Salce [1] proves the dual of Theorem 10.1 for p!Cn -injectives, i.e. the equivalence of (i) the
p-group A is p!Cn -injective; (ii) A contains a pn -bounded subgroup P such that A=P is torsion-
complete; (iii) A Š B=H where B is torsion-complete and pn H D 0. He also verifies the isomorphy
of p!Cn -injectives whose pn -socles are isometric as valuated groups.

Exercises

(1) If a p-group A is p!Cn -projective, then so is A=A1 .

(2) (Benabdallah–Irwin–Lazaruk) If 0 ! H ! F ! A ! 0 is a pure-exact
sequence with F a †-cyclic p-group, then A is p!Cn -projective exactly if it is
isomorphic to a summand of F=HŒpn
.
(3) (Benabdallah–Irwin–Lazaruk) A p-group A is p!CnC1 -projective exactly if it
contains a subsocle S such that A=S is p!Cn -projective.
(4) (Benabdallah–Irwin–Lazaruk) If S is a subsocle of a p-group A that supports a
pure subgroup of A, and if A=S is †-cyclic, then A is †-cyclic.
(5) (Keef) If A is a p!Cn -projective p-group, and  is a height-non-decreasing
homomorphism of AŒpn
into a p-group C, then there is a homomorphism
A ! C which agrees with  on the socle of A.
11 Summable p-Groups 401

(6) (Danchev) A reduced p-group is p!Cn -projective if and only if it has a large
subgroup that is p!Cn -projective.

11 Summable p-Groups

We continue the discussion of torsion groups with an interesting class which is

somewhat more general than direct sums of countable p-groups. The study was
initiated by Honda [3], and continued by Hill–Megibben [5] under the appropriate
name ‘summable groups.’
A p-group A is said to be summable if its socle S D AŒp
is a free valuated vector
space. This means that S D ˚ S where S is a subgroup satisfying p AŒp
D
p C1 AŒp
˚ S . Hence every element ¤ 0 of S carries the same value .
Example 11.1. A †-cyclic p-group is summable, and so is the Prüfer group H!C1 whose socle is
isometric to ˚! S ; here, dim S D 1 for each   !. Also, divisible p-groups are summable.
On the other hand, no unbounded torsion-complete p-group is summable.

We mention the following properties of summability.

(A) Countable p-groups are summable. Their socles as countably valuated vector
spaces are free Proposition 8.8 in Chapter 10.
(B) Direct sums of summable p-groups are summable. Hence direct sums of
countable p-groups are summable.
(C) Summands of summable p-groups are summable. The socle of a summable
p-group may be viewed as a direct sum of countable dimensional subspaces,
so by Theorem 8.10 in Chapter 10, every summand has a socle of the same
structure.
(D) Countable isotype subgroups of summable p-groups are summable.
The generalized Prüfer group H!1 D ˚ <!1 H is summable. However, no
summable reduced p-group exists whose length exceeds !1 .
Theorem 11.2 (Honda [3]). A reduced summable p-group A satisfies p!1 A D 0.
Proof. Write AŒp
D S ˚ p!1 AŒp
where S D ˚ <!1 S , and all non-zero elements
of S have height . Working toward contradiction, assume there is an a 2 A of
height !1 . Choose a sequence bn 2 A .n < !/ such that pbn D a, and the heights
h.bn / D n are increasing with n. Thus bnC1  bn 2 AŒp
has non-zero coordinate
in Sn , and we may even assume that it has 0 coordinate in every S if   nC1 .
Since the n are countable ordinals, there is a b 2 A of countable height  > sup n
satisfying pb D a. Then b  b0 D .b  bn / C .bn  bn1 / C    C .b1  b0 / shows
that b  b0 must have infinitely many non-zero coordinate in the decomposition of
S, since bn  bn1 (of height n1 ) is the only term that has non-zero coordinate in
Sn1 , except possibly for of b  bn . However, this is impossible in a direct sum. u
t
The preceding result is strengthened in (ii) of the following proposition.
402 11 p-Groups with Elements of Infinite Height

Proposition 11.3.
(i) An extension of a summable p-group by a countable p-group is again summable.
(ii) (Hill–Megibben [5]) The equation p!1 Ext.Z.p1 /; A/ D 0 holds for every
summable p-group A.
Proof.
(i) Let G be an extension of the summable group A by the countable p-group X.
There is a countable subgroup C
A that contains all the relations of the
extension; C may be assumed to be isotype. Evidently, G contains an extension
H of C by X. Then GŒp
is obtained from AŒp
by replacing the free vector space
CŒp
by the free HŒp
of countable dimension—thus freeness is preserved.

(ii) By way of contradiction, suppose 0 ! A ! G!Z.p1 / ! 0 is a non-
splitting extension representing an element in p!1 Ext.Z.p1 /; A/: Thus this
exact sequence is p!1 -pure, and hence p!1 -balanced by Theorem 8.2. This
means that .p GŒp
/ D p Z.p1 /Œp
Š Z.p/ for all  < !1 . This is
impossible, because owing to (i), G is summable, and from the proof of (i)
it is clear that p GŒp
D p AŒp
holds for large enough . t
u
We should point out that the class of summable p-groups is strictly larger than
the class of direct sums of countable p-groups. Indeed, for every ordinal  with
! < 
!1 , there exist summable p-groups of length  that are not direct sums of
countable groups. Next we show that even !1 -pure subgroups of summable groups
need not be summable.
Example 11.4. Assume that A D ˚ HC1 where HC1 runs over the generalized Prüfer groups
of non-limit lengths  C 1 < !1 ; thus, A has length !1 . For every  , map z 2 HC1 of height
 upon a generator of the socle of Z.p1 /, and extend this map to obtain a homomorphism ˛ W
A ! Z.p1 /. As in Example 8.4, it follows that the arising exact sequence 0 ! M D Ker ˛ !
A ! Z.p1 / ! 0 is !1 -pure. Thus it represents a non-splitting extension in p!1 Ext.Z.p1 /; M/.
In view of Proposition 11.3, M is not summable.

From this example we can derive a negative answer to a question which has
stayed unanswered for several years: are subgroups of direct sums of countable p-
groups again such direct sums? (See Example 7.1 where an explicit example was
given.)
Lemma 11.5 (Nunke [5], Hill [5], Hill–Megibben [5]). A subgroup of a direct
sum of countable p-groups need not be a direct sum of countable groups. t
u
F Notes. Honda [3] started the study of summable p-groups under the name of principal
p-groups. Summability plays an important role in generalizing the classical theory of p-groups to
countable length; see Megibben [4]. Danchev [2] shows that if the p-group A has length  !1 , and
p A and A=p A are both summable for some  < !1 , then so is A.

Exercises

(1) A separable p-group is summable if and only if it is †-cyclic.

12 Elongations of p-Groups 403

(2) (Hill–Megibben) Let A be a reduced p-group, and AŒp

D S ˚ p!1 AŒp
. If the
second summand is not 0, then S is not a summable socle.
(3) (Hill–Megibben) If A is a summable p-group, and H is a p -high subgroup in
A, then H is also summable.
(4) (Hill–Megibben) A summable p-group is a summand of every p-group in which
it is p!1 -high. [Hint: Proposition 11.3(ii).]
(5) (Griffith) If G is a p!1 -high subgroup of the generalized Prüfer group H!1 C1 ,
then G is isomorphic to an isotype subgroup of H!1 C1 , but it is not summable.
[Hint: Exercise 2.]
(6) (Salce) Let B be the torsion-complete p-group with the standard basic sub-
group B. Then G D B=BŒp
is a summable group of length ! C 1 which fails to
be totally projective.

12 Elongations of p-Groups

Let  denote a limit ordinal, and E; G p-groups. If G D p A holds in the exact

sequence 0 ! G ! A ! E ! 0 (and so E Š A=p A), then we call A a -
elongation of G by E (elongation has already been defined in Sect. 6). Needless
to say, for given E; G, such an elongation exists only if the groups satisfy certain
conditions.
Existence of Elongations We will state a necessary and sufficient condition for
the existence of -elongations of p-groups. We require an ad hoc definition. If A
is a p-group of length , then by its transfinite final rank we mean the cardinal
tr-fin-rk.A/ D min < rk.p AŒp
/. If  D  C 1 is a successor ordinal, then tr-fin-
rk.A/ D rk.p AŒp
/.
Lemma 12.1 (Nunke [7]). Let E be a p-group of length , and G a reduced p-
group with basic subgroup B. There exists a -elongation of G by E if and only if
rk B
tr-fin-rk.E/.
Proof. In a -elongation A of G ¤ 0 by E, for every basis element b 2 B, and for
every  < , there is x C G 2 p EŒp
such that px D b. If b0 2 B is a different basis
element, and px0 D b0 for some x0 C G 2 p EŒp
, then x  x0 D g 2 G is impossible.
In fact, the equation pg D b  b0 2 B would contradict the purity of B in G. This
shows that p EŒp
must contain at least as many elements for every  <  as rk B.
Conversely, assume  D rk B
tr-fin-rk.E/ holds. Let H be the divisible hull
of G, thus D D H=G satisfies DŒp
Š B=pB, and therefore rk DŒp
D . As D is
divisible, it is a straightforward arithmetical exercise to construct a map W E ! D
such that .p EŒp
/ D DŒp
for all  < . With such a , we form the pull-back
diagram
404 11 p-Groups with Elements of Infinite Height

γ : 0 −−−−→ G −−−−→ A −−−−→ E −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐γ
  
: 0 −−−−→ G −−−−→ H −−−−→ D −−−−→ 0,

and claim that A is a -elongation. This is an easy consequence of the fact that
multiplication by p in H induces an isomorphism DŒp
! B=pB; indeed, the choice
of guarantees that multiplication by p in A will furnish the basis elements of B
with height  in A. t
u
The second part of the preceding proof is essentially Kulikov’s method of
constructing elongations [3]. His choice for was a map whose kernel was isotype
and dense in E.
Uniquely Elongating A p-group E has the -Zippin property if its length is ,
and if A; C are groups such that A=p A Š E Š C=p C, then every isomorphism that
exists between p A and p C extends to a suitable isomorphism A Š C. By Zippin’s
theorem 1.7, reduced countable p-groups E enjoy this property for countable .
We now focus our attention on !-elongations; this is the most exciting case.
Crawley [3] was the first to consider elongations, and raised the question as to which
separable p-groups E have the property that all !-elongations of G Š Z.p/ by E are
isomorphic. This question turns out undecidable in ZFC. A weaker form asks for the
isomorphism of all extensions of an arbitrary group G by E—this easier question is
answered in the next theorem for separable groups.
Theorem 12.2 (Crawley [3], Hill–Megibben [4], Nunke [7], Warfield [5]).
A separable p-group E has the property that all of the !-elongations of the same,
arbitrarily selected p-group by E are isomorphic if and only if E is †-cyclic.
Proof. Consider two !-elongations, A1 and A2 , of a group G by the †-cyclic p-
group E. The identity map of G may be viewed as a height-preserving isomorphism
between subgroups of the Ai . As E has no elements of infinite height, G is nice in Ai .
Furthermore, a simple calculation shows that the nth Hill invariant of Ai relative to
G is equal to the nth UK-invariant of E, so they are equal for A1 and A2 . It remains to
appeal to Theorem 3.6 to infer that the identity map of G extends to an isomorphism
between the Ai .
For necessity, we prove that if E is a separable p-group, but not †-cyclic, then
there is an elementary p-group G admitting non-isomorphic !-elongations by E.
There is no loss of generality in assuming that fin rk E D jEj D , where   @1 .
Let B be a lower basic subgroup of E, so that jE=Bj D . Owing to the arguments
in the proof of Theorem 10.7, E contains a pure subgroup C which is not †-cyclic
such that jE=Cj D . We now use the method described in Lemma 12.1 above to
elongate an elementary p-group G of the size  by E by using the socles of E=B
and E=C, respectively. The resulting extensions are not isomorphic, since the socles
12 Elongations of p-Groups 405

BŒp
˚ G and CŒp
˚ G are not isometric: the former is a free valuated vector space
(G is free with value !), but the latter is not. t
u
Crawley Groups A separable torsion group E is said to be a Crawley group if
all !-elongations of G Š Z.p/ by E are isomorphic. Megibben [7] proved that in
ZFC it is undecidable whether or not Crawley groups of cardinality @1 are †-cyclic.
We follow the Mekler–Shelah proof in solving the problem in the Constructible
Universe without cardinality restriction.
In the following arguments, H denotes an arbitrary, but fixed, p-group of
cardinality
@1 such that p! H Š Z.p/. (For convenience, temporarily we may
choose H D H!C1 , the Prüfer group of length ! C 1, but later a different choice
will be needed.)
Lemma 12.3 (Mekler–Shelah [1]). Let E0 be a separable p-group containing a
subgroup E of countable index, and 0 ! G ! A ! E ! 0 an !-elongation with
G D hgi of order p. If E0 =E is not †-cyclic, then in the commutative diagram

0 −−−−→ G −−−−→ A −−−−→ E −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  incl
0 −−−−→ G −−−−→ Ai −−−−→ E  −−−−→ 0

groups Ai .i D 1; 2/ can be chosen such that any homomorphism  W A ! H with

.g/ ¤ 0 extends to at most one of Ai .
Proof. The non-unique existence of the bottom exact sequence is a consequence of
the fact that the induced map Ext.E0 ; G/ ! Ext.E; G/ is epic, but not necessarily
monic. Suppose an elongation A1 has been selected. As E0 =E is countable, but not
†-cyclic (so not separable), it has non-zero elements of infinite heights, i.e. there are
elements xn 2 E0 n E and en 2 E .n < !/ such that px0 D e0 ; pnC1 xnC1 D x0 C enC1 .
To construct A2 , define an extension of G by hE; x0n .n < !/i .
E0 / subject to the
relations px00 D e0 C g; pnC1 x0nC1 D x00 C enC1 , and then extend it to all of E0 (this
is possible as Ext.E0 ; G/ ! Ext.E; G/ is surjective).
Suppose that  W A ! H extends to i W Ai ! H for i D 1; 2. Then
pnC1 1 .xnC1 / D 1 .x0 / C .enC1 / and pnC1 2 .x0nC1 / D 1 .x00 / C .enC1 /, as well
as p1 .x0 / D .e0 / and p2 .x00 / D .e0 / C .g/. Subtracting, we get

pnC1 .2 .x0nC1 /  1 .xnC1 // D 2 .x00 /  1 .x0 /:

Hence 2 .x00 /  1 .x0 / 2 p! H!C1 : But p.2 .x00 /  1 .x0 // D .g/ ¤ 0; a

contradiction. u
t
Next, we extend the preceding lemma to the general case when the cardinality of
E0 =E is arbitrary. This process requires the assumption of the diamond principle.
406 11 p-Groups with Elements of Infinite Height

Lemma 12.4 (Mekler–Shelah [1] (V = L)). Let again E0 be a separable p-group

containing a subgroup E, and 0 ! G ! A ! E ! 0 an !-elongation with G D
hgi of order p. Suppose E0 =E is not †-cyclic, and  W A ! H satisfies .g/ ¤ 0.
Then there is an elongation 0 ! G ! A0 ! E0 ! 0, extending the given one,
such that  cannot be extended to A0 .
Proof. Observe that the condition that E0 =E is not †-cyclic is necessary for the
existence of an elongation as required. For otherwise, every map  W A ! H extends
to A0 , as A is nice in A0 and A0 =A is †-cyclic.
We induct on the cardinality  D jE0 =Ej. The limit case being clear, we may
assume  is regular, and also uncountable, and the claim holds for smaller cardinals
Lemma 12.3. We will be done if we can find a subgroup C with E < C < E0 such
that the corresponding elongation has the stated property. Therefore, we may assume
for the rest of the proof that C=E is †-cyclic whenever jC=Ej < , and hence C=E
has a -filtration C =E . < / with †-cyclic subgroups. As E0 =E is not †-cyclic,
the set

S D f <  j C C1 =C is not †-cyclicg

must be stationary in . The diamond principle guarantees the existence of a set

f  j  2 Sg of functions  W C ! H such that, for every function W E0 ! H,
there is a  2 S satisfying  C D  .
We define a direct system of elongations 0 ! G ! A ! C ! 0 for  < 
starting with the given one (A0 D A). For the proof, an extension A is viewed
as being built on a subset on G  E0 . Only the step from  to  C 1 has to be
specified. For each  < , any extension of G by C C1 that is a continuation of the
extension by C can be selected, except when  2 S is such that  W A ! H is
a homomorphism inducing  . 2 S/. In this exceptional case, we can apply the
induction hypothesis (because A C1 =A is not †-cyclic) to get an extension A C1
such that  does not extend to any map A C1 ! H.
We claim that the direct limit sequence 0 ! G ! A0 ! E0 ! 0 is as required.
Suppose, on the contrary, there is an extension  0 W A0 ! H of . Evidently,  0
defines a function W E0 ! H whose restriction to some C . 2 S/ equals  . But
this is a contradiction to the choice of A C1 . t
u
Lemma 12.5 (Mekler–Shelah [1]). Assuming V = L, a separable p-group E is
†-cyclic exactly if every !-elongation A of Z.p/ by E admits a homomorphism
 W A ! H such that .p! A/ ¤ 0.
Proof. If A is an !-extension of G Š Z.p/ by a †-cyclic E, then a map G ! p! H
extends to a homomorphism A ! H.
Conversely, suppose E has the stated property. Theorem 12.2 takes care of the
case jEj D @0 , so suppose that jEj D  > @0 , and that groups of smaller sizes
with this property are †-cyclic. Hence all subgroups of E of cardinality <  are
†-cyclic. If  is a regular cardinal, then Lemma 12.4 implies that in a -filtration
of E with †-cyclic subgroups the set S as defined in the preceding proof cannot
12 Elongations of p-Groups 407

be stationary. This means that E is †-cyclic. The rest follows from the singular
compactness theorem. t
u
After all this preparation we can now prove:
Theorem 12.6 (Mekler–Shelah [1]). In the Constructible Universe, every Crawley
group is †-cyclic.
Proof. Let E be a separable p-group that is not †-cyclic. Owing to Lemma 12.5, we
can choose an !-elongation A of Z.p/ by E, and a group G such that p! G Š Z.p/
and jGj
2@0 .D @1 / hold true, and there is a homomorphism  W A ! G with
.p! A/ ¤ 0. But by Lemma 12.4 there exists an !-elongation A0 of Z.p/ by E such
that no homomorphism W A0 ! G exists with .p! A0 / ¤ 0. Hence A Š A0 is
impossible, and consequently, E is not a Crawley group. t
u
Megibben [7] proved: MA+ : CH implies that all @1 -separable p-groups of
cardinality @1 are Crawley groups; by Theorem 7.1 in Chapter 10 such groups
do exist in ZFC without being †-cyclic. Consequently, Crawley’s problem is
undecidable in ZFC. We forgo the details of the proof.
F Notes. By Mekler–Shelah [1], it is consistent that there exists a Crawley group of
cardinality @1 which is not @1 -separable. Also, assuming V = L, there exist a separable, non-
†-cyclic p-group E of cardinality @1 and elongations A . < 2@1 / of Z.p/ by E such that if
 W A ! A . ¤  / is a homomorphism, then .p! A / D 0. Mekler [1] shows that it is
consistent with GCH that there exist non-†-cyclic Crawley groups, so the Crawley problem is
undecidable in ZFC+GCH.
Eklof–Huber–Mekler [1] define a p-group A totally Crawley if A=p A is Crawley for all limit
ordinals . It is undecidable in ZFC that for countable lengths, every totally Crawley group is a
direct sum of countable groups.
Warfield [5] points out that !-elongations by non-†-cyclic groups may be abundant. For
@
instance, there are 22 0 non-isomorphic !-elongations by a separable p-group A of cardinality
@0
2 with countable basic subgroup; the elongated group can be any reduced p-group that admits
!-elongations by A.

Exercises

(1) (Nunke) A totally projective p-group E of limit length  is uniquely -

elongating.
(2) Summands in a uniquely elongating group are likewise uniquely elongating
provided they are of the same length.
(3) Let E be a p-group of length , a limit cardinal, and N a fully invariant subgroup
of the same length. If E is uniquely -elongating, then so is N.
(4) (Richman) A separable p-group A is Crawley if and only if, for any two
dense subgroups of codimension 1 in AŒp
, one is carried into the other by an
automorphism of A.
(5) (Nunke) Let 0 ! G ! A ! E ! 0 be an exact sequence. The connecting map
Tor.E; Z.p// ! G ˝ Z.p/ defines a homomorphism  W EŒp
! G=pG. If  is
408 11 p-Groups with Elements of Infinite Height

the length of E, then the given exact sequence is a -elongation if and only if
.p EŒp
/ D G=pG holds for all  < .
(6) Suppose that A is a p-group such that its first Ulm subgroup A1 is a direct sum
of countable groups, and A=A1 is †-cyclic. Then A is likewise a direct sum of
countable groups.

Problems to Chapter 11

PROBLEM 11.1. Investigate subgroups G in totally projective p-groups (or more

generally, in transitive groups) A such that each height-preserving endomorphism of
G extends to an endomorphism of A.
PROBLEM 11.2. Are there test groups for p -projectives ? and for p -injectives?
PROBLEM 11.3. Which p-groups contain p!C1 -projective pure subgroups?
PROBLEM 11.4. Does there exist a p!C1 -projective p-group of cardinality  that
contains a copy of every p!C1 -projective p-group of cardinality
?
PROBLEM 11.5. (Cutler) Are the (separable) p-groups A; C isomorphic if, for
every n 2 N, the subgroups AŒpn
and CŒpn
are isometric as valuated groups?
PROBLEM 11.6. Find a complete set of invariants for p-groups of finite Ulm
length whose Ulm factors are torsion-complete.
Richman [3] considers the case when the first Ulm subgroup is an elementary group.

PROBLEM 11.7. Let

u D .0 ; 1 ; : : : ; n ; : : :/

be a strictly increasing sequence of ordinals and symbols 1. The p-group A is an

u-elongation of G by E if G D A.u/ and A=G Š E.
(a) When do u-elongations exist?
(b) For which groups E are such elongations of a fixed G by E all isomorphic?
PROBLEM 11.8. Imitating the definition of pure-completeness, call a p-group
A isotype-complete if every subsocle that is isometric to the subsocle of a (totally
projective) p-group supports an isotype subgroup of A. Study isotype-complete p-
groups.
PROBLEM 11.9. Suppose A is totally projective p-group contained in a p-
group G. What can be said about the subgroups C of G if CŒp
and AŒp
are isometric
as valuated vector spaces?
Chapter 12
Torsion-Free Groups

Abstract In this chapter we start the discussion of torsion-free groups. First, we deal with
general properties along with the finite rank case, and delegate the in-depth theory of torsion-free
groups of infinite rank to the next chapter.
After presenting the basic definitions and facts, we enter the study of balancedness, a stronger
version of purity, which we have already met in the theory of torsion groups. Turning to the
problem of direct decompositions, we start with the discussion of indecomposable groups; we
do not restrict ourselves to the finite rank case as it seems more natural to deal with this important
problem without rank restrictions. Concentrating on the finite rank case, the study of pathological
direct decompositions is followed by positive results, the highlight being Lady’s theorem about the
finiteness of non-isomorphic direct decompositions.
Other aspects of direct decompositions are also discussed, including quasi- and near-
homomorphisms. Finite rank dualities will also be dealt with.

1 Characteristic and Type: Finite Rank Groups

In this section, all groups are assumed to be torsion-free, unless stated otherwise.
p1 ; p2 ; : : : ; pn ; : : : will denote the sequence of prime numbers.
Every torsion-free group A is a subgroup of a Q-vector space V such that a
maximal independent set in A is a basis of V. If fxi .i 2 I/g is a maximal independent
set in A, then every element of A depends on it, and therefore every element a 2 A
can be written uniquely as

a D r1 x1 C    C rk xk .ri 2 Q/:

Note that only certain combinations of rational coefficients are allowed, depending
on the group and the choice of the maximal independent set.
Characteristics The most typical features of elements in torsion-free groups are
crystallized in the concept corresponding to height. Let A be a torsion-free group and
a 2 A. Recall that, for a prime p, the largest integer k with pk ja, i.e., for which the
equation pk x D a is solvable in A, is called the p-height hp .a/ of a. If no such
maximal integer k exists, then we set hp .a/ D 1. The sequence of p-heights,

.a/ D .hp1 .a/; hp2 .a/; : : : ; hpn .a/; : : :/;

© Springer International Publishing Switzerland 2015 409

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_12
410 12 Torsion-Free Groups

is called the characteristic or the height-sequence of a. We will write A .a/ if

we wish to indicate the group A in which the characteristic was computed. (For
torsion-free groups, transfinite heights are unnecessary: elements of height ! are in
the p-divisible subgroup, so have height 1.)
The following observations are immediate consequences.
(a) .a/ D .a/ for all a 2 A.
(b) If .a/ D .k1 ; k2 ; : : : ; kn : : : :) (where kn  0 or D 1), then a 2 A is divisible
by the integer m D p`11    p`r r if and only if `i
ki for i D 1; : : : ; r.
(c) If we agree that 1 C 1 D 1, then

.pn a/ D .hp1 .a/; hp2 .a/; : : : ; hpn .a/ C 1; : : :/:

(d) Every sequence .k1 ; k2 ; : : : ; kn : : : :/ of non-negative integers and symbols 1 is

a realizable characteristic. In the subgroup R of Q that is generated by all p` n
n

with `n
kn , for all n, the element 1 will have this characteristic. With this R,
we can write Ra D hai if A .a/ D .k1 ; k2 ; : : : ; kn : : : :/ in a group A.
The set of all characteristics is partially ordered under the pointwise ordering; it
becomes a complete, distributive lattice of the cardinality of the continuum 2@0 . The
lattice operation ^ is the point-wise minimum, and _ is the point-wise maximum.
The minimum member of this lattice is the characteristic .0; 0; : : : ; 0; : : :/ of 1 2 Z,
while the maximum is the characteristic .1; 1; : : : ; 1; : : :/ of every element in Q.
(e) If b; c 2 A, then .b C c/  .b/ ^ .c/. Thus

A./ D fa 2 A j .a/  g

is for each characteristic  a (fully invariant) subgroup of A. Furthermore, if

A D B ˚ C and b 2 B; c 2 C, then .b C c/ D .b/ ^ .c/.
(f) A subgroup B of a torsion-free group A is pure exactly if B .b/ D A .b/ for all
b 2 B.
(g) If B is a pure subgroup of the torsion-free A, then the characteristic of a coset
mod B is computed according to the rule
_
A=B .a C B/ D A .a C b/:
b2B

(h) If ˛ W A ! C is a homomorphism between torsion-free groups, then A .a/

C .˛a/ for all a 2 A.

For a torsion-free A, the inclusion map Z ! Q induces an embedding A Š
Z ˝ A ! Q ˝ A. The latter group is torsion-free, moreover, a Q-vector space; its
dimension is equal to the rank rk.A/ of A. If B is a pure subgroup of A, then A=B is
torsion-free, its rank is called the corank of B in A.
1 Characteristic and Type: Finite Rank Groups 411

Types Two characteristics, .k1 ; k2 ; : : : ; kn ; : : :/ and .`1 ; `2 ; : : : ; `n ; : : :/, are said

to be equivalent if kn D `n for almost all n such that both kn and `n are finite
whenever kn ¤ `n . The equivalence classes of characteristics are called types.
If .a/ belongs to type t, then we say a is of type t, and write t.a/ = t, or more
explicitly, tA .a/ = t if we wish to emphasize that the type of a is computed in the
group A. Keep in mind that t.a/ = t.ma/ for all m 2 N.
A type can be represented by any member in its equivalence class; thus, we can
write t = .k1 ; k2 ; : : : ; kn ; : : :/ if the right-hand side is a characteristic in the class
of t. The lattice order on the set of characteristics induces a lattice order on the set
T of types. Accordingly, s
t for the types s, t means that there are characteristics
.k1 ; k2 ; : : : ; kn ; : : :/ in s and .`1 ; `2 ; : : : ; `n ; : : :/ in t, such that .k1 ; k2 ; : : : ; kn ; : : :/

.`1 ; `2 ; : : : ; `n ; : : :/. T is a (non-complete) distributive lattice.

(A) Linearly dependent elements a; b in a group A have the same type. In fact, then
ra D sb for some non-zero integers r; s, and thus t.a/ = t.ra/ = t.sb/ = t.b/.
(B) If b; c 2 A, then t.b C c/  t.b/ ^ t.c/, and if A D B ˚ C and b 2 B; c 2 C,
then t.b C c/ D t.b/ ^ t.c/.
(C) For ˛ W A ! B and a 2 A, we always have t.a/
t.˛a/.
(D) A torsion-free group A is called homogeneous if all of its elements ¤ 0 are of
the same type t. In this case we write t.A/ D t, and call t the type of A. With
a type t we associate four fully invariant subgroups of a torsion-free group A.
The two major ones are:

A.t/ D fa 2 A j t.a/  tg and A? .t/ D ha 2 A j t.a/ > ti:

A.t/ is always a pure subgroup in A, but A? .t/ need not be pure in A. Its
purification satisfies A? .t/
A.t/. The Baer invariants of a torsion-free
group A are the torsion-free ranks of the factor groups A.t/=A? . t/. Thus the
Baer invariants are given by a function from the set of types to the cardinal
numbers. Less useful are the fully invariant subgroups that are defined in terms
of a characteristic: A./ D fa 2 A j .a/  g and A? ./ D ha 2 A j
.a/ > i.
The other two noteworthy fully invariant subgroups are:

AŒt
D \t.A=G/t G and A? Œt
D \t.A=G/<t G

where rk A=G D 1. The ranks of these fully invariant subgroups are the Arnold–
Vinsonhaler invariants of A.
Rank One Groups The non-zero subgroups of Q (i.e., the rational groups) are
the rank 1 groups. In a rank 1 group R, all non-zero elements have the same type,
the type t.R/ of R. Our next task is to prove the fundamental theorem on rank 1
torsion-free groups.
Theorem 1.1 (Baer [6]). Two torsion-free groups of rank 1 are isomorphic exactly
if they are of the same type. Every type is realized by a subgroup of Q.
412 12 Torsion-Free Groups

Proof. Let A; B be rank one torsion-free groups of the same type. If a 2 A; b 2 B

are non-zero elements, then t.a/ D t.b/ implies that there are non-zero integers
m; n such that A .ma/ D B .nb/. The correspondence ma 7! nb extends to an
isomorphism A ! B.
The second claim is the immediate consequence of (d) above. t
u
It is useful to know the cardinality of the lattice of types.
Proposition 1.2. The cardinality of the set of non-isomorphic rank one torsion-free
groups is the power of the continuum.
Proof. The set of all characteristics is evidently of the power 2@0 , so the set of
types cannot have a larger cardinality. On the other hand, different characteristics
consisting only of 0’s and 1’s represent different types, so the cardinality in
question is at least 2@0 . From Theorem 1.1 we know that every type is realized
by a rational group. t
u
The product 1 2 of the characteristics 1 D .k1 ; k2 ; : : : ; kn ; : : :/ and 2 D
.`1 ; `2 ; : : : ; `n ; : : :/ is defined as

1 2 D .k1 C `1 ; k2 C `2 ; : : : ; kn C `n ; : : :/;

where naturally the sum of 1 and anything is 1. A characteristic  is idempotent

(i.e., 2 D ) exactly if, for every n, kn is either 0 or 1. If 1  2 , then there is a
unique largest characteristic 0 such that 2 0 D 1 I this will be denoted by 0 D
1 W 2 . The multiplication of characteristics is compatible with the equivalence
relation introduced above, so we may speak of the product t1 t2 and quotient t1 W t2
of the types t1 ; t2 , and of an idempotent type t2 D t. Type t0 D t W t is the largest
idempotent type
t: to obtain t0 from t D .k1 ; k2 ; : : : ; kn ; : : :/, replace the finite kn
by 0.
Lemma 1.3. If A and C are torsion-free groups of rank 1, then their tensor product
A ˝ C is again torsion-free of rank 1 and

t.A ˝ C/ D t.A/  t.C/:

Proof. Theorem 3.5 in Chapter 8 implies that A˝C is a subgroup of Q˝Q Š Q, so

it is of rank 1. Write .a0 / D .k1 ; k2 ; : : : ; kn ; : : :/ and .c0 / D .`1 ; `2 ; : : : ; `n ; : : :/
for a0 2 A; c0 2 C. Then a0 ˝ c0 2 A ˝ C will certainly be divisible by pm n for all
m
kn C `n , so .a0 ˝ c0 /  .a0 /.c0 /: The correspondence .ra0 ˝ sc0 / 7! rs
(where ra0 2 A; sc0 2 C and r; s 2 Q) is a bilinear map A  C ! R, where R is
the rational group with R .1/ D A .a0 /C .c0 /. From the definition of the tensor
product it is clear that .a0 ˝ c0 / > R .1/ is impossible. t
u
In the future, we shall need the following simple facts which are easy to verify
directly.
1 Characteristic and Type: Finite Rank Groups 413

Lemma 1.4. Let R and S be rational groups of types t and s, respectively. Then
(i) Hom.R; S/ D 0 unless t
s;
(ii) if t
s, then Hom.R; S/ is a rational group of type s W t;
(iii) Hom.R; R/ is a rational group of idempotent type t0 D t W t; and
(iv) Hom.Hom.R; R/; R/ is a rational group of type t. t
u
Finite Rank Groups A torsion-free group A of finite rank n is a subgroup
of an n-dimensional Q-vector space V, so we can describe A by starting with an
independent set a1 ; : : : ; an , and then adjoining more elements of V as generators of
A, responsible for divisibility properties.
Example 1.5. Let P be a set of primes, and p; q primes … P. Then

A D hp2 a; q1 b; P1 .a C b/i

is the group of rank 2 in which a; b are independent elements, and the additional generators are:
p2 a; qn b for every n 2 N, and r1 .a C b/ for every r 2 P.

Type.A/ will denote the Typeset of A, i.e. the set of types of the non-zero
elements of A. Type.A/ can be infinite even if rk.A/ is finite. We also talk of
Cotype.A/ as the set of cotypes, i.e. types of rank one torsion-free factor groups of
A; but this gives useful information only if A is of finite rank. IT.A/ and OT.A/ are
the notations for the inner and outer types of A: IT.A/ is defined as the intersection
of all types t.a/ for all a 2 A, and OT.A/ is the union of all types of rank 1 torsion-
free factor groups of A. We have:
Lemma 1.6. Let A be a torsion-free group of finite rank, a1 ; : : : ; an a
maximal independent set in A, and Ai the pure subgroup of A generated by
a1 ; : : : ; ai1 ; aiC1 ; : : : ; an . Then

IT.A/ D t.a1 / ^ : : : ^ t.an / and OT.A/ D t.A=A1 /_ : : : _t.A=An /:

Proof. What we have to show for IT.A/ is that, for every a 2 A, the type t.a/ is
larger than or equal to the stated intersection. If we write a D r1 a1 C    C rn an
with ri 2 Q, then it is clear that we must have t.a/  t.r1 a1 / ^ : : : ^ t.rn an / D
t.a1 / ^ : : : ^ t.an /:
For OT.A/ we argue that, since A1 \ : : : \ An D 0, there is an embedding A !
A=A1 ˚    ˚ A=An in a direct sum of rank 1 groups, call it AN for a moment. By
definition, _i t.A=Ai /
OT.A/. On the other hand, it is obvious that OT.A/

OT.A/.N Since every homomorphism of a direct sum of rank 1 groups into Q carries
N D
each rank 1 summand either to 0 or isomorphically into Q, it is evident that OT.A/
_i t.A=Ai /. t
u
Example 1.7. For a finite rank A, we have IT.A/ = OT.A/ if and only if A is a direct sum of
isomorphic rank 1 groups. The ‘if’ part is pretty obvious, while the converse is a consequence of
Lemma 3.6.
414 12 Torsion-Free Groups

If A is torsion-free of rank m, and if F is a free subgroup of rank m, then T D A=F

is a torsion group whose p-components Tp have ranks
m. Thus for each prime p,
we can write:

Tp D Z.pkp1 / ˚    ˚ Z.pkpm / (12.1)

where the exponents kpi are non-negative integers or symbols 1 (note that there
are exactly m summands, and Z.p0 / D 0). Another choice of F can change only
the finite exponents, and only in a finite number of p-components, into other finite
exponents. The obvious equivalence class is called the Richman type of A; this is
the same for quasi-isomorphic groups (see Sect. 9). It is easy to check that if we
assume kpn 1
  
kpn m .8 n 2 N/, then IT.A/ is represented by the characteristic
.kp1 1 ; : : : ; kpn 1 ; : : :/, and OT.A/ by the characteristic .kp1 m ; : : : ; kpn m ; : : :/.
To close this topic, we show that the cardinality of the typeset is not bounded by
the rank of the group. As a matter of fact, it can be infinite even for rank 2 groups
(cf. Exercise 10).
Lemma 1.8 (Koehler [2], Mutzbauer [2]). A finite set of types is the typeset of a
torsion-free group if and only if it is closed under intersections.
Proof. Let T D ft1 ; : : : ; tk g be a set of types, closed under intersections. The direct
sum of rational groups of these types has typeset T.
Conversely, suppose A has a finite typeset T D ft1 ; : : : ; tk g. First we show that
t1 ^ : : : ^ tk 2 T. Clearly, A D [i A.ti / (set-union) implies QA D [i QA.ti /, thus
the Q-vector space QA is the set-union of a finite number of subspaces. This can
happen only if QA D QA.tj / for some j, which means that this tj is the intersection
of all types in T. Next, let t be the intersection of some types in T. The typeset of the
subgroup A.t/ consists of those ti 2 T for which t
ti . By the preceding argument,
t is a type of a subgroup in A.t/, so it belongs to T. t
u
A type t is said to be an extractable type of the torsion-free group A if A has a
rank 1 summand of type t.
Homogeneous Groups Recall that ‘homogeneous’ means that every non-zero
element is of the same type.
Proposition 1.9 (Nongxa [2]). A homogeneous torsion-free group of type t is the
tensor product of a rank one group of type t with a homogeneous torsion-free group
of type .0; 0; : : : ; 0; : : :/.
Proof. First we deal with an idempotent type t. Let R denote a rational group of type
t, and let …1 be the set of primes at which t is 0, and …2 the set of primes at which t
is 1. Choose an essential free subgroup F in the homogeneous torsion-free group A
of type t, and decompose the torsion group A=F D A1 =F ˚ A2 =F where Ai =F is the
…i -component of A=F. The exact sequence 0 ! A1 ! A ! A=A1 Š A2 =F ! 0
induces the exact sequence 0 ! R ˝ A1 ! R ˝ A Š A ! R ˝ A2 =F D 0, whence
A Š R ˝ A1 follows. The type of every element in A1 must be
t, and in view of
the given decomposition of A=F, it cannot be larger than the type of Z.
1 Characteristic and Type: Finite Rank Groups 415

If t is an arbitrary type, and R is a rational group of type t, then the group

A0 D Hom.R; A/ is homogeneous of idempotent type t0 D t W t, and we have an
isomorphism R ˝ A0 Š A. This, combined with the preceding paragraph, completes
the proof. t
u
Subgroups Isomorphic to the Group It is evident that the subgroups
nA .n 2 N/ are isomorphic to A. Though several other subgroups may be isomorphic
to A, in the finite rank case they are not far from being an integral multiple of the
group. Indeed, we can claim:
Proposition 1.10 (Jónsson [2]). If C is a subgroup of the finite rank torsion-free
group A such that C Š A, then nA
C
A for some n 2 N.
Proof. (E. Walker) Let  W A ! C be an isomorphism, viewed as an element of
End A. It extends uniquely to the divisible hull D of A, so the extension (also denoted
by ) is a linear transformation of the finite dimensional Q-vector space D. By the
Cayley–Hamilton theorem, it satisfies an equation of the form nk k C nk1 k1 C
   C n1  C n0 D 0 .ni 2 Z/. As  is monic, so left-cancellable in End A, n0 ¤ 0
may be assumed. But then n0 A D .nk k1 C nk1 k2 C    C n1 /A
A D C,
as claimed. t
u
F Notes. It was Levi [1] who started the study of torsion-free groups; his work was practically
ignored for a long time. He wrote a thesis to obtain venia legendi, i.e. the right to lecture at the
university, but has never published his results in a journal. He was also the first to point out the
connection of abelian groups with p-adic modules. Almost 20 years later, the duality theory of
Pontryagin prompted a lot of interest in abelian groups, in particular in torsion-free groups, since
some of their properties reflect faithfully on topological properties of compact abelian groups.
Starting with the theory of p-adic modules discussed by Derry [1], Mal’cev [1] and Kurosh [2]
developed a theory, based on the equivalence of matrices, in order to find structural results on
torsion-free groups of finite rank. (As observed in [AG], the theory can easily be extended to
the countable rank case.) The theory was, unfortunately, too difficult to apply in practice, as it
translates a difficult problem to another difficult problem, without giving a better insight into the
structure. (Many years later, Richman [5] followed a more transparent approach—as described
above.) Campbell [1] tried to classify countable torsion-free groups, using matrices to describe
divisibility; this does not seem a useful approach either.
The inner and outer types were introduced in Fuchs [14], and developed by Warfield [1].
Mutzbauer [2] shows that the sum of types in the chain 0 D A0 < A1 <    < An D A of
pure subgroups (the factor groups are of rank 1) of a finite rank torsion-free group A is independent
of the chain.
Metelli [1] defines a torsion-free group A t-bihomogeneous if it satisfies both A.t/ D A
and AŒt
D 0. Inter alia she proves that countable bihomogeneous groups are completely
decomposable an analogue of Theorem 4.3 in Chapter 13.
In [IAG] it was asked what can be said about finite rank torsion-free groups A; B if Ext.A; C/ Š
Ext.B; C/ holds for all finite rank torsion-free groups C. Several papers dealt with the problem; we
point out Krylov [4] where it was shown that then A and B have to be quasi-isomorphic provided
they have no summand Š Z or Q. Goeters [1], Chekhlov–Krylov [1] investigate the dual problem:
reduced finite rank torsion-free groups A; B satisfying Ext.C; A/ Š Ext.C; B/ for all reduced
torsion-free C of finite rank. It is shown that then A and B have to be quasi-isomorphic.
The foundation of the modern theory of torsion-free groups is due to Baer; see his seminal
paper [6]. In the present chapter, several fundamental results carry his name. We should also point
out that Yakovlev [2] develops an innovative new method of dealing with finite rank torsion-free
groups, based on the theory of integral representations. The study splits, for each prime p, into a
Krull–Schmidt category.
416 12 Torsion-Free Groups

Exercises

1. (a) Let A; B be rank 1 torsion-free groups. A is isomorphic to a subgroup of B

if and only if t.A/
t.B/ if and only if Hom.A; B/ ¤ 0.
(b) Two rank 1 torsion-free groups, A and B, are isomorphic exactly if both
Hom.A; B/ ¤ 0 and Hom.B; A/ ¤ 0.
2. (a) A group of rank 1 contains either only finitely many or continuously many
pairwise non-isomorphic subgroups.
(b) For every type t¤ .1; : : : ; 1; : : :/, there exist countably many character-
istics of the same type t.
3. The completion
Q RQ of a rational group R ¤ Q is isomorphic to a summand of
ZQ Š p Jp .
4. Let G be a torsion-free group of rank 1. Find conditions on G to be written as
G Š R ˝    ˝ R, the nth ‘tensor power’ of a rank 1 torsion-free group R.
5. Give examples such that
(a) the elements of type >t do not form (only generate) a subgroup of A;
(b) A .t/ is not pure in A;
(c) A contains elements of type t and A .t/ D A.t/.
6. (Baer) If A is of finite rank, then Type.A/ satisfies both the maximum and the
minimum conditions. [Hint: ranks of A.t/.]
7. (a) For a torsion-free group G and a type t, G.t/ is the trace of a rank 1 torsion-
free group of type t in G.
(b) If C is pure in A, then C.t/ is pure in A.t/.
(c) Let 0 ! A ! B ! C ! 0 be a pure-exact sequence of finite rank groups.
If B D B.t/ for some type t, then also A D A.t/ and C D C.t/.
8. (Warfield) rk Hom.A; C/ D rk A  rk C holds for finite rank torsion-free groups
A; C if and only if IT.C/  OT.A/.
9. (Müller–Mutzbauer) The intersection of cotypes is in finite rank groups is equal
to the inner type.
10. (Baer) There exist torsion-free groups of rank 2 whose typeset is infinite. [Hint:
in Qa ˚ Qb choose A such that hp .a C pn b/ D 1 or 0 according as p D pn or
not.]
11. (a) Let B be a pure subgroup of rank 1 in the torsion-free group A of rank 2.
Then t.B/  t.A=B/ is an invariant of A: this product is independent of the
choice of B in A.
(b) Suppose that A is a homogeneous torsion-free group of rank 2. Show that
all the cotypes are the same.
12. A minimax group is defined as an extension of a group with maximum
condition on subgroups by a group with minimum condition on subgroups.
Prove or disprove: a torsion-free group is minimax if and only if it is of finite
rank.
13. (a) (Soifer [1]) A generating system S of A is irreducible if no proper subset
of S generates A. Show that a rank 1 torsion-free group A has an irreducible
2 Balanced Subgroups 417

generating system if and only if it is cyclic or its type is not idempotent.

[Hint: for sufficiency, find generators pkp qn such that in the type we have
kp > 0 an integer and kq D 1.]
(b) Use a similar trick to show that an uncountable group A admits an
irreducible system of generators whenever jA=pAj D jAj for some prime p.
14. (Höfling) (a) Let ftn .n < !/g be a countable set of types none of which contains
1. Then there is a type t such that t > tn for all n.
(b) There exists a chain containing @1 different types.

2 Balanced Subgroups

We now introduce a concept that plays a vital role in the theory of torsion-free
groups.
Balancedness Let B be a pure subgroup of the torsion-free group A. An element
a 2 A n B is called proper with respect to B if

.a C b/
.a/ for all b 2 B:

Evidently, a 2 A is proper with respect to B if and only if A .a/ D A=B .a C B/. It

is a useful fact that then .a C b/ D .a/ ^ .b/ holds for all b 2 B.
We say that B is balanced in A if every coset of B contains an element proper
with respect to B. Summands are always balanced subgroups, but balancedness is a
much stronger property than purity. If B is balanced in A, we call the exact sequence
0 ! B ! A ! A=B ! 0 balanced-exact.
Example 2.1. (a) If B is a pure subgroup of A such that A=B is homogeneous of type Z, then B
is balanced in A.
(b) A simple non-trivial example for balancedness is given in a torsion-free group A whose
elements have types
t. If A? .t/ is pure in A, then it is a balanced subgroup.
(c) An indecomposable torsion-free group of rank 2 has no balanced subgroups except for the
trivial ones.

In order to get better acquainted with balancedness, we show that it can be

characterized in several ways.
Lemma 2.2. A pure subgroup B of a torsion-free group A is balanced in A if and
only if, for every a 2 A and for every countable subset fbn gn<! of B there exists a
b 2 B such that

.a C bn /
.b C bn / for every n < !: (12.2)

Proof. Let B be balanced in A, and a 2 A. For some b 2 B, a  b is proper

with respect to B, i.e. .a  b/  .a C x/ for all x 2 B. Then .b C x/ 
minf.b  a/; .a C x/g D .a C x/ for all x 2 B.
418 12 Torsion-Free Groups

Conversely, suppose that, for every a 2 A and for fbn gn<!  B, there is a b 2 B
satisfying (12.2). Then .a b/W  .a Cbn / for each bn . If the set fbn gn<! is chosen
so as to satisfy A=B .a C B/ D n<! .a C bn /, then B .a  b/  A=B .a C B/, i.e.
a  b is proper with respect to B. t
u
The next lemma is probably the best tool for checking balancedness in most of
the proofs.
Lemma 2.3. An exact sequence of torsion-free groups is balanced-exact if and only
if all rank one torsion-free groups have the projective property relative to it.

Proof. Let e W 0 ! B ! A!C ! 0 be balanced-exact, and ˛ W R ! C a
homomorphism where R is a subgroup of Q, say, with 1 2 R. There is an a 2 A
such that a D ˛1 and C . a/ D A .˛1/  R .1/. Hence the correspondence
1 7! a extends to a homomorphism  W R ! A such that  D ˛.
Conversely, assume e has the stated property, and let x 2 A. Then the embedding
˛ W R ! C (where x D ˛1) lifts to a homomorphism  W R ! A such that
 D ˛. It is easy to check that a D 1 2 A is in the coset of x C B, and satisfies
A .a/ D C .˛1/. t
u
Some relevant properties of balancedness are reminiscent of those of purity.
Lemma 2.4 (Haimo [1]). Let A and B be pure subgroups of the torsion-free group
G such that B
A. Then we have:
(a) If B is balanced in G, then it is balanced in A as well.
(b) If B is balanced in A, and A is balanced in G, then B is balanced in G.
(c) If A is balanced in G, then A=B is balanced in G=B.
(d) If B is balanced in G, and A=B is balanced in G=B, then A is balanced in G.
Proof. (a) is trivial. To prove (b), form the commutative diagram

0 0
⏐ ⏐
⏐ ⏐
 

0 −−−−−→ B −−−−−→ A −−−−−→ A/B −−−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  

0 −−−−−→ B −−−−−→ G −−−−−→ G/B −−−−−→ 0

⏐ ⏐
⏐ ⏐
 

H H
⏐ ⏐
⏐ ⏐
 

0 0

with exact rows and columns, where the top row and the middle column are
balanced-exact. Let R be a rational group containing 1 and ˛ W R ! G=B a non-
2 Balanced Subgroups 419

trivial map. Manifestly, ˛ lifts to (A and hence to) G whenever Im ˛ is contained

in A=B. If this is not the case, then ˛ followed by G=B ! H lifts to G in the
middle column.
(c) By hypothesis, in the preceding diagram, the middle column is balanced-exact.
Thus, any map ˛ W R ! H lifts to G, and hence to G=B.
(d) Now in our diagram the middle row and the last column are assumed to be
balanced-exact. Consequently, any map ˛ W R ! H lifts to G=B, and hence
to G. Thus the middle column is balanced-exact. t
u

Lemma 2.5. For an exact sequence e W 0 ! B ! A!C ! 0 of torsion-free
groups, the following are equivalent:
(i) e is balanced-exact;

(ii) the induced sequence e./ W 0 ! B./ ! A./!C./ ! 0 is exact for each
characteristic ;

(iii) the induced sequence e.t/ W 0 ! B.t/ ! A.t/!C.t/ ! 0 is exact for every
type t.
Proof. The exactness of the sequences (ii) and (iii) at the middle terms follows at
once from the purity of B in A. Since (i) says that every c 2 C has a preimage a 2 A
with A .a/ D C .c/, it is clear that (i) is equivalent to (ii). That (ii) ) (iii) is also
evident.
To prove (iii) ) (ii), let ˛ W R ! C be a homomorphism, where R is a subgroup
of Q of type t. Then ˛R
C.t/: We refer to Lemma 3.6 infra, to argue that B.t/ is
a summand of B.t/ C ˛R, so ˛ lifts. t
u
Lemmas on Balanced Subgroups Balanced subgroups will play a most impor-
tant role in the discussion of Butler groups in Chapter 14. Therefore, let us proceed
by assembling here a few needed facts on balanced subgroups that will be needed
in the sequel.
It is far from being a trivial task to prove the existence of balanced subgroups in
large torsion-free groups.
Lemma 2.6 (Dugas–Hill–Rangaswamy [1]). If C is any subgroup of the torsion-
free A, then there exists a balanced subgroup B of A containing C such that
jBj
jCj@0 .
A torsion-free group that contains no balanced subgroups other than the trivial
ones must have cardinality
2@0 .
Proof. Ignoring a trivial case, assume  D jCj@0  2@0 . Let X be a countable
subset of C. For a 2 A, let aX denote the function from X into the lattice of all
characteristics defined by setting

aX .x/ D .a C x/ .x 2 X/:

Consider the set S of all these functions (taken for all a 2 A, and for all countable
subsets X  C); since there are only 2@0 characteristics, S has cardinality at most
2@0 D . Thus there exists a subset W in A of cardinality
 such that S D
420 12 Torsion-Free Groups

faX j a 2 W; X a countable subset of Cg. Let C1 be the pure subgroup generated

by C and this set W. Repeat this process for C1 in the place of C to obtain C2 ,
and continue transfinitely, taking unions at limit ordinals. In this way, we obtain a
smooth chain of pure subgroups C of A. After !1 steps, we get a pure subgroup
B D C!1 of A. Since each C . < !1 / has cardinality
, also jBj
.
We claim thatWthis B is balanced in A. Let a C B .a 2 A/ be a coset; its
characteristic is n .a C bn / for some fbn gn<!  B. There is a C . < !1 /
such that all bn 2 C . Construction guarantees the existence of c 2 C˛C1 satisfying
.c C bn /  .a C bn / for all n < !. Then a  c 2 a C B is proper with respect to
B, so B is balanced in A.
The second statement is a simple corollary. t
u
It is easy to see that the union of a countable chain of balanced subgroups need
not be balanced. However, for longer chains, we have:
Lemma 2.7. Let 0 D A0 < A1 <    < A <    . < / be a .not necessarily
smooth/ well-ordered ascending chain of balanced
S subgroups of the torsion-free
group G. If cf   !1 , then the subgroup A D  A is again balanced in G.

W Given g 2 G, there is a countable subset fan gn<!  A such that G=A .g C A/

Proof.
D n<! .g C an /: In view of the hypothesis cf   !1 , there exists an index  < 
with fan gn<!  A . The balancedness of A implies that .g C b / D G=A .g C A /
for some b 2 A . Then
_
.g C b / D G=A .g C A /  .g C an / D G=A .g C A/;
n

establishing the claim. t

u
Lemma 2.8 (Bican–Fuchs [1]). Suppose

0 ! B ! A !C ! 0 . < /
is a well-ordered direct system of balanced-exact sequences where the C are
reduced torsion-free groups. If the connecting maps  W C ! C .for all
 <  < / are monomorphisms with Im  pure in C , then the direct limit

0 ! B ! A!C ! 0 of the system is likewise a balanced-exact sequence.
Proof. Consider the following commutative diagram:

γσ
0 −−−−→ Bσ −−−−→ Aσ −−−−→ Cσ −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐α ⏐φ
  σ  σ
γ
0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0

Here ˛ and  stand for the canonical maps into the direct limits. To establish
balancedness, let R be a rank 1 torsion-free group. Given W R ! C and 0 ¤ x 2 R,
2 Balanced Subgroups 421

there is a  <  such that  x D x for some x 2 C . Then the correspondence

x 7! x extends to a homomorphism  W R ! C . Evidently,   D . By the
balanced-exactness of the top row, there is a map  W R ! A such that   D  .
Now ˛  W R ! A satisfies .˛  / D . ˛ /  D .  /  D  .   / D
  D , as desired. t
u
Groups of Balanced Extensions Let A; C be torsion-free groups. The equiva-
lence classes of extensions of A by C in which A is balanced form a subgroup in
Ext.C; A/, called the group of balanced extensions, and denoted Bext1 .C; A/. That
this is in fact a subgroup of Ext.C; A/ follows in the same way as for Ext, by making
use of the next lemma. (We have the affix 1 to Bext, since there are higher Bexts, as
we shall see later on.)
˛
Lemma 2.9. If e W 0 ! B!A!C ! 0 is a balanced-exact sequence, and
 W B ! B0 and W C0 ! C are homomorphisms, then the top and bottom rows in
the commutative diagram

ψ : 0 −−−−→ B −−−−→ A −−−−→ C  −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐ψ
  
α γ
: 0 −−−−→ B −−−−→ A −−−−→ C −−−−→ 0
⏐ ⏐ 
⏐φ ⏐ 
  
φ : 0 −−−−→ B  −−−−→ A −−−−→ C −−−−→ 0

are balanced-exact.
Proof. That rank 1 torsion-free groups have the projective property for the top row
follows from the pull-back property, while the same for the bottom row is almost
obvious. t
u
We conclude that  induces a map  W Bext1 .C; B/ ! Bext1 .C; B0 / and
induces  W Bext1 .C; B/ ! Bext1 .C0 ; B/: With the aid of these induced maps, we
have:
˛
Theorem 2.10. Let 0 ! B!A!C ! 0 be a balanced-exact sequence. Then
for every torsion-free group G, the induced sequences
˛
0 ! Hom.G; B/! Hom.G; A/! Hom.G; C/ !
˛
! Bext1 .G; B/! Bext1 .G; A/! Bext1 .G; C/ ! : : :
422 12 Torsion-Free Groups

and
˛
0 ! Hom.C; G/! Hom.A; G/! Hom.B; G/ !
˛
! Bext1 .C; G/! Bext1 .A; G/! Bext1 .B; G/ ! : : :

are exact.
Proof. This follows from a general theorem on relative homological algebra; see,
e.g., Mac Lane [M]. t
u
There are more functors Bexti which continue the long exact sequences in
the same fashion. As far as their definition is concerned, let us observe that if
0 ! B ! C ! G ! 0 is a balanced-projective resolution of G (with completely
decomposable C, see next section), then

BextiC1 .G; A/ D Bexti .B; A/

for every torsion-free group A, and for every i  1. This follows from the
continuation of the exact sequences in the preceding theorem by making use
of Bexti .C; A/ D 0 for completely decomposable groups C. Thus with the aid
of repeated balanced-projective resolutions the higher Bexts can be computed
recursively.
F Notes. The systematic study of balancedness started with Hunter [1], though several
sporadic results on balancedness had been proved earlier under different or no names, see, e.g.,
Lyapin [1]. The first balanced-projective resolutions appear in Albrecht–Hill [1].
Specific subgroups of various kind were subjects of publications, but none of them turned out to
be able to compete in importance with balancedness. For example, Bican [1] studied subgroups G
of a torsion-free A, called regular to mean tG .g/ D tA .g/ for all g 2 G. Regular subgroups display
a few interesting properties, and are useful in certain proofs. We also mention the dual concept: an
exact sequence 0 ! B ! A ! C ! 0 is said to be cobalanced if all rank 1 torsion-free groups
R have the injective property relative to it, i.e. if the induced map Hom.A; R/ ! Hom.B; R/ is
always surjective. Cobalancedness has been investigated by several authors, see, e.g., Goeters [2].
So far, cobalancedness has not proved to be very useful.

Exercises

(1) Localization preserves balancedness.

(2) The group Jp of the p-adic integers (p any prime) is a torsion-free group of
cardinality 2@0 that has only two balanced subgroups, viz. 0 and Jp .
(3) Let F be a free group, and G a pure subgroup such that F=G is homogeneous of
type .0; : : : ; 0; : : :/, but not free (e.g., an infinite direct product of copies of Z).
Show that G is balanced, but not a summand in F.
(4) (Kravchenko) If B is a balanced subgroup of A, then for all types s and t,
.B C A.s// \ .B C A.t// D B C A.s ^ t/.
(5) Let G D A C B be a torsion-free group, where A; B are pure subgroups of G.
If A \ B is balanced in A, then B is balanced in G. [Hint: G=B Š A=.A \ B/
3 Completely Decomposable Groups 423

implies 8 g 2 G 9 a 2 A with g  a 2 B and .a C .A \ B// D .g C B/; if

A \ B balanced, then .a C .A \ B// D .a C b/ for a b 2 A \ B.]
(6) Let G . < ! S1 / be a smooth chain of pure subgroups of a torsion-free group
G, and G D  <!1 G . If B is a subgroup of G such that B \ G D B˛ is
balanced in G for each  < !1 , then B is balanced in G.
(7) With the aid of the 3  3-lemma, prove that equivalent conditions for balanced-
ness in Lemma 2.5 are that 0 ! B=B./ ! G=G./ ! C=C./ ! 0 is exact
for each characteristic , and 0 ! B=B.t/ ! G=G.t/ ! C=C.t/ ! 0 is exact
for each type t, respectively.

3 Completely Decomposable Groups

At this point it is appropriate to inquire about balanced-projectives and balanced-

projective resolutions. A torsion-free group is balanced-projective if it has the
projective property relative to all balanced-exact sequences 0 ! B ! A ! C ! 0
of torsion-free groups.
A torsion-free group is called completely decomposable if it is a direct sum
of rank 1 groups. Free groups and divisible torsion-free groups are the most
common examples of completely decomposable groups. We intend to show that the
completely decomposable groups have the projective property relative to balanced-
exact sequences. Moreover, this projective property characterizes the completely
decomposable groups.
Balanced-Projective Resolutions Our starting point is the following theorem;
one part of its proof assumes subsequent results that do not depend on it.
Theorem 3.1 (C. Walker [3]). The balanced-projective torsion-free groups are
precisely the completely decomposable groups.
Proof. In view of Lemma 2.3 it is obvious that rank one groups are
balanced-projective. The projective property is inherited by direct sums, so the
balanced-projectivity of completely decomposable groups is obvious. That only
the completely decomposable groups are balanced-projective will follow by
standard arguments, making use of Theorems 3.2 and 3.9 below. t
u
Balanced-projectivity is destined to play for torsion-free groups a similar role
as for p-groups, as witnessed by the following analogue of Proposition 5.8 in
Chapter 11.
Theorem 3.2. (i) Every torsion-free group G has a balanced-projective resolu-
tion, i.e., it can be embedded in a balanced-exact sequence

0 ! B ! C ! G ! 0; (12.3)

where C is completely decomposable.

424 12 Torsion-Free Groups

(ii) If 0 ! B0 ! C0 ! G ! 0 is another balanced-projective resolution of the

same group, then

B ˚ C0 Š B0 ˚ C:

Proof. (i) Let R˛ run over all rank one pure subgroups of G. For each ˛, pick a
rank one group C˛ isomorphic to R˛ along with an isomorphism ˛ W C˛ ! R˛ .
Let C be the direct sum of all these C˛ , one for every R˛ , and define W C ! G
to act like ˛ on C˛ . Then, evidently, the map is surjective. Its kernel must be
balanced in C, because the criterion of Lemma 2.3 applies by construction.
(ii) (à la Schanuel’s lemma) The pull-back diagram

0 −−−−→ B −−−−→ H −−−−→ C  −−−−→ 0

 ⏐ ⏐
 ⏐ φ ⏐
  θ
γ
0 −−−−→ B −−−−→ C −−−−→ G −−−−→ 0

can be completed by a map  W C0 ! C such that  D  by Theorem 3.1. This

means that the top exact sequence is splitting, i.e. H Š B ˚ C0 . For reason of
symmetry, H Š B0 ˚ C as well. t
u
Imitating the definition of projective dimension of modules, we can introduce
a dimension concept for balancedness. Accordingly, a group G is defined to have
balanced-projective dimension n if n is the smallest integer for which

BextnC1 .G; A/ D 0 for all groups A:

We will indicate this by writing bpd G D n. A more explicit way to describe this
dimension is to form a long balanced-exact sequence

0 ! Cn ! Cn1 !    ! C1 ! C0 ! G ! 0

(with completely decomposable groups Ci ) that is shortest of this kind. Thus

bpd G D 0 if and only if G is completely decomposable. We will see later on in
Lemma 1.9 in Chapter 14 that for every n  0, there are finite rank torsion-free
groups G with bpd G D n.
Relative Balanced-Projective Resolutions A useful device in the discussion
of infinite rank Butler groups (Chapter 14) is a straightforward generalization of
balanced-projective resolutions: the relative balanced-projective resolution intro-
duced by Bican–Fuchs [2].
Let A be a pure subgroup of the torsion-free group G. We say that

0 ! K ! A ˚ C!G ! 0 (12.4)
3 Completely Decomposable Groups 425

is a balanced-projective resolution of G relative to A if this is a balanced-exact

sequence with completely decomposable C such that   A is the inclusion map
˛ W A ! G.
The existence of such a sequence is easily established: just take a balanced-
projective resolution of G, say, 0 ! H ! C!G ! 0, and choose  D r. ˚˛/ W
C ˚ A ! G and K D Ker . The corresponding exact sequence must be balanced,
because every rank 1 torsion-free group will have the projective property relative to
it (Lemma 2.3). In order to get a better understanding of relative resolutions, let us
complement this observation by adding
Lemma 3.3. Using the same notations, let the map  W K ! C be the injection
K ! A ˚ C followed by the projection into C. Then  is monic and C=K Š G=A.
Proof. Since jA is monic, so is . Noting that A ˚ K D A ˚ K, we evidently
have

C=K Š .A ˚ C/=.A ˚ K/ Š ..A C C/=K/=..A C K/=K/ Š G=A;

establishing the claim. t

u
As far as the uniqueness of such relative resolutions is concerned, we can assert:

Proposition 3.4. If (12.4) and 0 ! K 0 ! A ˚ C0 !G ! 0 are two

balanced-projective resolutions of G relative to the subgroup A .C0 is completely
decomposable/, then K ˚ C0 Š K 0 ˚ C:
Proof. Making use of the pull-back diagram

0 −−−−→ K  −−−−→ M −−−−→ A ⊕ C −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐φ
  
ψ
0 −−−−→ K  −−−−→ A ⊕ C  −−−−→ G −−−−→ 0,

the proof is as for Theorem 3.2(ii). t

u
It is an immediate corollary that the group K is completely decomposable if and
only if K 0 is completely decomposable.
Completely Decomposable Groups The following theorem provides a satis-
factory structure theorem for completely decomposable groups.
Theorem 3.5 (Baer [6]). Any two decompositions of a completely decomposable
group into direct sums of rank one groups are isomorphic.
Proof. Let A D ˚i2I Ai where the Ai are rational groups. It is readily seen that for
any type t, the subgroups A.t/ and A .t/ are the direct sums of those Ai that are of
types  t and > t, respectively. Thus At D A.t/=A .t/ is isomorphic to the direct
sum of those Ai which are exactly of type t, and the rank of At is precisely the
426 12 Torsion-Free Groups

number of components Ai of type t. Since At is defined independently of the given

direct decomposition, the claim follows from the uniqueness of the rank. t
u
As a consequence, for a completely decomposable group A, the ranks of the
factor groups A.t/=A? .t/ for all types t (the Baer invariants of A) form a complete
(and independent) set of invariants for A.
A most useful technical lemma (known as ‘Baer’s Lemma’) is as follows.
Lemma 3.6 (Baer [6]). Let B be a pure subgroup of the torsion-free group A such
that every element of the coset a C B .a 2 A n B/ has the same type in A as the coset
has in A=B. Then B is a summand of hB; ai.
Proof. As A .a/
A=B .a C B/ and t.a/ D t.a C B/, there exist a finite number
of elements a1 ; : : : ; ak 2 a C B such that A .a1 / _ : : : _ A .ak / D A=B .a C B/.
It is thus enough to show that if a  a0 mod B and A .a/; A .a0 / are equivalent
characteristics, then the coset a C B contains an element g satisfying A .g/ 
A .a/ _ A .a0 /. Hypothesis implies the existence of relatively prime integers m; n
such that A .ma/ D A .na0 /. If u; v are integers such that mu C nv D 1, then
g D mua C nva0 is  a mod B, and evidently satisfies A .g/  A .ma/ 
A .a/ _ A .a0 /. t
u
We record two important consequences of Lemma 3.6.
Corollary 3.7. In a finite rank homogeneous completely decomposable group,
every pure subgroup is a summand.
Proof. Let A D ˚niD1 Ai with rational groups Ai of the same type t, and B a pure
subgroup in A. The factor group A=B must be homogeneous of type t, since every
element of A=B is contained in the sum of the images of (finitely many) Ai . Owing
to Lemma 3.6, for each a 2 A n B, B is a summand of B1 D hB; ai . Keep going,
arguing the same way with B1 , until A is reached. t
u
Proposition 3.8 (Baer [6]). Let C be a pure subgroup of the torsion-free group A
such that
(i) A=C is completely decomposable and homogeneous of type t;
(ii) all the elements in A n C are of type t.
Then C is a summand of A.
Proof. Referring to Lemma 3.6, from (ii) we derive that C is a balanced subgroup
of A. The rest follows from Lemma 2.3. t
u
Our next result generalizes the classical theorem that subgroups of free groups
are again free (Theorem 1.6 in Chapter 3). Its proof runs in much the same way.
Theorem 3.9 (Baer [6], Kolettis [2]). In a homogeneous completely decompos-
able group, every subgroup that is homogeneous of the same type .in particular,
every pure subgroup/ is completely decomposable.
3 Completely Decomposable Groups 427

Proof. Suppose A D ˚ A with groups A of rank 1 and of type t, where the index
 ranges over the ordinals less than some ordinal . Define G D ˚< A , and
C D C \ G , where C is a t-homogeneous subgroup. Clearly, C D C C1 \ G ,
thus

C C1 =C Š .C C1 C G /=G

G C1 =G Š A :

We claim that C is a summand of C C1 : In fact, if C C1 ¤ C , then all the elements

of C C1 are of type t, and the type of C C1 =C must be t, since it is both an epic
image of C C1 and isomorphic to a subgroup of A . Hence Lemma 3.6 implies that
C C1 D C ˚ B for some rank one B . The B ’s . < / generate their direct sum,
which must be all of C. t
u
Summands of Completely Decomposable Groups The next theorem is per-
haps the most important result on complete decomposability.
Theorem 3.10 (Baer [6], Kulikov [4], Kaplansky [2]). Direct summands of com-
pletely decomposable torsion-free groups are completely decomposable.
Proof. We start with the case of finite typesets. Let A D H1 ˚    ˚ Hn D B ˚ C,
where each Hi is completely decomposable and homogeneous of type ti ; assume that
these types are different. We use induction on n. If n D 1, then A is homogeneous,
and the complete decomposability of B and C follows from Theorem 3.9. Assume
n > 1, and let tn be a maximal type in the set ft1 ; : : : ; tn g. Then Hn is fully invariant
in A, so Hn D A.tn / D B.tn / ˚ C.tn /, whence B D B.tn / ˚ B0 and C D C.tn / ˚ C0
for some B0
B and C0
C; here B.tn / and C.tn / are completely decomposable
by case n D 1. We now have A D Hn ˚ B0 ˚ C0 , so H1 ˚    ˚ Hn1 Š B0 ˚ C0 .
Induction hypothesis guarantees that B0 ; C0 are completely decomposable, and so
are therefore B and C as well.
We interrupt our argument to verify the following lemma that contains a key
technical element in our proof covering the infinite rank case.
Lemma 3.11. Let A D B ˚ C be a completely decomposable group. Then every
finite set of elements of B is contained in a finite rank completely decomposable
summand of B.
Proof. First, suppose A D B ˚ C D H ˚ G where H D H1 ˚    ˚ Hn and each
Hi is homogeneous completely decomposable of different types ti . In addition, we
assume that G has no summand of type ti . We want to show by induction on n that
H D .H \ B/ ˚ .H \ C/.
If n D 1 and t.H/ = t, then omitting summands that are of types > t, we obtain
A D B ˚ C Š H ˚ G; where bars indicate the remaining direct sum. Here H is a
fully invariant subgroup, so H D .H \ B/ ˚ .H \ C/ D .H \ B/ ˚ .H \ C/.
If n > 1, then arguing in the same way with the component Hn of a maximal type
tn , we get a similar decomposition where H1 ˚  ˚Hn1
B0 ˚C0 . This permits us
to apply the induction hypothesis to establish the claim. Once we have that H splits
428 12 Torsion-Free Groups

along B and C, we can conclude that B D .H \ B/ ˚ B0 and C D .H \ C/ ˚ C0 for

some B0
B; C0
C, and G Š B0 ˚ C0 .
A finite set of elements of B is contained in a summand of A of the form H D
H1 ˚    ˚ Hn with Hi homogeneous completely decomposable. Write A D H ˚ G.
The preceding argument shows that H \ B is a summand of B containing the given
set. It is a summand of H, so it is completely decomposable by the proof above on
the case of finite typesets. This completes the proof of the lemma. t
u
Resuming the proof of Theorem 3.10, our next step is to reduce the proof to the
case of countable rank. This is done at once with an appeal to the rank version of
Kaplansky’s lemma 2.5 in Chapter 2. So we may assume that A is of countable rank.
Let b1 ; : : : ; bk ; : : : be a complete list of elements of B. In view of Lemma 3.11, B
has a finite rank completely decomposable summand Bk containing b1 ; : : : ; bk as
well as a maximal independent set of Bk1 . Clearly, B is the union of the chain
B1
  
Bk
: : : where BkC1 D Bk ˚ Ek for some summand Ek of B. It follows
that B D ˚k<! Ek , where Ek (as a summand of BkC1 ) is completely decomposable
by the first part of the proof. Hence B is completely decomposable, indeed. t
u
Completely Decomposable Subgroups There is no reasonable criterion
for a subgroup of a completely decomposable group to be again com-
pletely decomposable. But it seems interesting that when a pure subgroup is
completely decomposable, then there is an entire chain of completely decomposable
groups between the subgroup and the containing group.
Theorem 3.12 (Fuchs–Viljoen [2]). Let A be a completely decomposable pure
subgroup of the completely decomposable group C. Then there is a smooth chain

A D B0 <    < B <    < B D C

of completely decomposable pure subgroups of C such that all the factor groups
B C1 =B are countable.

Proof. Consider the pure-exact sequence 0 ! A ! C ! G ! 0. A straightfor-
ward back-and-forth argument leads us to a smooth chain 0 D C0 <    < C <
   < C D C such that, for each  < ,
(i) C is a summand of C;
(ii) A D A \ C is a summand of A;
(iii) C is pure in G; and
(iv) C C1 =C is countable.
As a consequence of (i) and (ii), B D A C C is pure in C, and B C1 =B
is countable. Thus B is completely decomposable, because the summands in
B D A C C Š A=A ˚ C are completely decomposable. t
u
It is obvious that any torsion-free group contains pure completely decomposable
subgroups. The following result shows that such subgroups need not be too small.
3 Completely Decomposable Groups 429

Theorem 3.13 (Griffith [2]). A reduced torsion-free group A contains a pure

completely decomposable subgroup C such that jAj
jCj@0 .
Proof. Partially order the set of all pure completely decomposable subgroups Ci of
A, by declaring Ci less than Cj if Ci is a summand of Cj . Zorn’s lemma guarantees
that this set has maximal members under this partial order, and let C be one of them.
Evidently, the closure C of C in the Z-adic completion AQ of A is pure in AQ and is
complete, and hence algebraically compact. Thus AQ D C ˚ C0 for some C0 . Now A
Q so A\C ˚A\C0 is pure in A. If A\C0 were ¤ 0, then it would contain a
is pure in A,
rank 1 pure subgroup X, and C˚X would be a larger pure completely decomposable
subgroup in A. Hence the projection of A in C is monic, and jAj
jC j
jCj@0
follows (see (6.6) in Chapter 6). t
u
It should be emphasized that the groups C in the preceding lemma are not unique,
not even up to isomorphism (Exercise 5). The only uniqueness we know of is that if
they are of infinite rank, then their ranks are equal.
Chains of Completely Decomposable Groups Hill’s theorem 7.7 in Chapter 3
on chains of free groups generalizes to homogenous completely decomposable
groups.
Theorem 3.14 (Nongxa [1]). The union A of a countable chain 0 D A0 <    <
An <    .n < !/ of pure, homogeneous completely decomposable subgroups of
fixed type t is completely decomposable.
Proof. The proof runs in the same line as the proof of the quoted theorem. The
argument on projectivity is replaced by an appeal to Baer’s lemma 3.6. t
u
Balanced-Injective Groups Once balanced-projectivity has been discussed,
it is natural to have a look at the dual: the question of balanced-injectives. In
the torsion case, balanced-injectivity led to an interesting new class of groups.
Surprisingly, in the torsion-free case, we do not get anything new.
Theorem 3.15 (Hunter [1]). In the category of torsion-free groups, the balanced-
injectives are precisely the algebraically compact groups.
Proof. Evidently, pure-injective groups are balanced-injective. For the converse,
assume G is torsion-free and balanced-injective. Let S D Z./ be a free group
where the cardinal  is chosen so as to satisfy jGj
 <  @0 D 2 ; this can
be done, see the note after Proposition 7.10 in Chapter 10. Then jSj D  and
jS j D  @0 , where S denotes the Z-adic closure of S in P D Z . Every extension
by S is a balanced-extension, therefore Ext.S ; G/ D Bext.S ; G/ vanishes by the
hypothesis on G. Furthermore, the exact sequence 0 ! S ! S ! ˚ @0 Q ! 0
induces the exact sequence Hom.S; G/ ! Ext.˚ @0 Q; G/ ! Ext.S ; G/ D 0.
Here j Hom.S; G/j D jGj
  D 2 , while
@0 @0 
j Ext.˚ @0 Q; G/j  j Ext.Q; G/j  2 D 22 > 2
430 12 Torsion-Free Groups

whenever Ext.Q; G/ ¤ 0. This is a contradiction unless G is cotorsion. Thus G is

cotorsion and torsion-free, so algebraically compact. t
u
However, with the sole exception of algebraically compact groups, no torsion-
free group A can be embedded in a balanced-injective group as a balanced subgroup.
Q then it must be a summand, since A=A
In fact, if A is balanced in A, Q Š ˚ Q.
F Notes. The theory of completely decomposable groups was initiated by Baer [6]. He proved
the main characterization Theorem 3.5. He also proved Theorem 3.10, but only for the case when
the typeset satisfied the maximum condition. Kulikov [4] gave a long proof for the countable case.
The final touch was given by Kaplansky who proved that summands of direct sums of countable
groups are again of the same kind.
A homogeneous pure subgroup H of a completely decomposable group C need not be
completely decomposable (though it is if C is homogeneous). However, H will be completely
decomposable if it is *-pure in the sense that .H  .t// D H \ .C .t// for all types t; see Nongxa
[2]. On the other hand, Kravchenko [1] found necessary and sufficient conditions on a completely
decomposable group to have all pure subgroups completely decomposable. Hill–Megibben [7]
pointed out that balanced subgroups of completely decomposable groups need not be completely
decomposable again. As a matter of fact, the balanced-projective dimension of a torsion-free group
can be any integer and infinity. See Nongxa–Rangaswamy–Vinsonhaler [1] for an exposition of
balanced subgroups of completely decomposable groups.
Eklof [6] proves that a homogeneous torsion-free group of singular cardinality is completely
decomposable, provided each of its subgroups of smaller cardinality is contained in a completely
decomposable subgroup.
Baer’s ubiquitous lemma 3.6 is a most useful device needed in numerous proofs. It is the
source of several generalizations, discussed by Albrecht, Arnold, Goeters, and Lady. See, e.g.,
Proposition 12.4.

Exercises

(1) Any two direct decompositions of a completely decomposable group have

isomorphic refinements.
(2) (a) The typeset of a completely decomposable group is a semilattice under
the operation ^. It is a subsemilattice of the lattice of all types.
(b) Find the least upper bound for the cardinalities of typesets of completely
decomposable groups of rank n.
(3) A countable homogeneous torsion-free group, in which all finite rank pure
subgroups are completely decomposable, is completely decomposable.
(4) Let C be a pure subgroup of the torsion-free group A such that A=C is a
homogeneous completely decomposable group of type t. If A D A.t/ C C,
then C is a summand of A. [Hint:QBaer’s lemma.]
(5) (Griffith) In the group ZQ Š p Jp , there are maximal pure completely
decomposable subgroups of any finite rank  1 as well as of rank @0 .
(6) (Procházka) A reduced torsion-free group A has the property that an isomor-
phism A Š C ˚ A=C holds for each pure subgroup C if and only if A is a
completely decomposable group of finite rank such that the types of the rank
1 summands are totally ordered.
4 Indecomposable Groups 431

(7) (Procházka) Let C be a completely decomposable group such that the different
types of its rank 1 summands are inversely well-ordered. Show that pure
subgroups of C are completely decomposable.
(8) (Fuchs–Kertész–Szele) A torsion-free group has the property that all of its
pure subgroups are summands if and only if its reduced part is completely
decomposable homogeneous and of finite rank. [Hint: reduce to the finite rank
case; use Corollary 3.7.]
(9) Assume G D Q.p/ ˚ Q.q/ for different primes p; q. Show that each of the
completely decomposable groups A Š G.@0 / and C Š Z˚G.@0 / is isomorphic
to a pure subgroup of the other, but A and C are not isomorphic
(10) (Rangaswamy) Let C be a completely decomposable subgroup of countable
index in the torsion-free group A. There is a decomposition A D C0 ˚ B with
C0
C and jBj D @0 .
(11) (Yakovlev) Let A be a finite rank completely decomposable group. If A has a
homogeneous completely decomposable subgroup C and an endomorphism ˛
such that A D CC˛CC  C˛ n C for some n 2 N, then A is also homogeneous.
[Hint: CnC1 ! A via .c0 ; : : : ; cn / 7! c0 C ˛c1 C    C ˛ n cn .]
(12) Prove a stronger version of Theorem 3.12 by using a G.@0 /-family in place of
a smooth chain.

4 Indecomposable Groups

One of the most important questions in abelian group theory is concerned with
indecomposable groups. Recapitulating: a group is called indecomposable if it
has no direct summands other than 0 and itself. Equivalently, it has no idempotent
endomorphism different from 0 and 1. Among the torsion groups, only the cocyclic
groups are indecomposable, while no proper mixed group is indecomposable (see
Corollary 2.4 in Chapter 5). The situation is drastically different for torsion-free
groups: there exist arbitrarily large indecomposable torsion-free groups. This is
what we want to prove here.
Finite Rank Indecomposable Groups We start with the finite rank case. The
simplest way of getting indecomposable groups even up to rank 2@0 is via the p-adic
integers.
Theorem 4.1 (Baer [6]). p-pure subgroups of the group Jp of the p-adic integers
are indecomposable.
Proof. Let A be a p-pure subgroup of Jp , and let 0 ¤  2 A. If  D sk pk C
skC1 pkC1 C    .sk ¤ 0/ is the canonical form of , then by p-purity, also sk C
skC1 p C    2 A. Thus A contains a p-adic unit, and therefore A C pJp D Jp . By
purity, pA D A \ pJp , whence

A=pA D A=.A \ pJp / Š .A C pJp /=pJp D Jp =pJp Š Z.p/:

432 12 Torsion-Free Groups

If we had A D B ˚ C with B; C ¤ 0, also p-pure in Jp , we would obtain the

contradiction A=pA Š B=pB ˚ C=pC Š Z.p/ ˚ Z.p/. u
t
Our next goal is to obtain more explicit examples for indecomposable groups. In
the constructions, we will often use the practice—as a matter of convenience—of
writing p1 a as an abbreviation for what would properly be written as an infinite
set: fp1 a; : : : ; pn a; : : :g. Also, we will construct groups A by starting from a basis
fa1 ; : : : ; an g of a Q-vector space V, and then specifying additional generators of A
in V, or by starting from a direct sum, and then adjoining elements from its divisible
hull, even without mentioning V. Unexplained letters, like an ; bn ; cn , will denote
independent elements in some unspecified Q-vector space; in any case, our notation
should be self-explanatory.
Example 4.2. Every torsion-free group of rank 1 is indecomposable.
Example 4.3 (Bognár [1]). Consider a set fp1 : : : : ; pn g of primes, and an integer m relatively prime
to each pi . For every i, let Ei D hp1
i ei i and Gi D hp1 i ei ; m1 ei i be groups of rank 1; clearly,
Ei D mGi . Define A as a subgroup of G1 ˚    ˚ Gn as

A D hE1 ˚    ˚ En ; m1 .e1 C e2 /; : : : ; m1 .e1 C en /i:

To show that A is indecomposable, suppose A D B˚C. Notice that Hom.Ei ; Gj / D 0 if i ¤ j, since

every element of Ei is divisible by every power of pi , while Gj does not contain any such elements
¤ 0. Therefore, the groups Ei are fully invariant in A, so we have Ei D .B \ Ei / ˚ .C \ Ei /.
The group Ei (as a rank 1 group) is indecomposable, so either Ei  B or Ei  C. Assume e.g.
E1  B and Ei  C for some i > 1. We can write m1 .e1 C ei / D b C c with b 2 B; c 2 C. Then
e1 D mb; ei D mc, which is impossible, since none of ei is divisible in A by any prime divisor of
m. This means that all of Ei are contained either in B or in C, and hence either B D A or C D A.
Example 4.4. Modify the preceding example, choosing a prime q different from the pi , and
forming the group

G D hE1 ˚    ˚ En ; q1 .e1 C e2 /; : : : ; q1 .e1 C en /i:

The same proof applies to establish the indecomposability of this G.

Rigid Systems In the construction of indecomposable groups, the following

concept is useful. A set fGi gi2I of torsion-free groups ¤ 0 is said to be a rigid
system if
(
a subgroup of Q if i D j;
Hom.Gi ; Gj / Š
0 if i ¤ j:

That is, groups in a rigid system have no endomorphisms other than multiplications
by rational numbers, and no non-trivial homomorphisms into each other. In
particular, groups in a rigid system have no idempotent endomorphisms ¤ 0; 1,
thus they are necessarily indecomposable. A group G is called rigid if the singleton
fGg is a rigid system.
The simplest example for a rigid system is a set of rank 1 groups.
Proposition 4.5. There exists a rigid system of torsion-free groups of rank 1
consisting of 2@0 groups.
4 Indecomposable Groups 433

Proof. Suppose p1 ; q1 ; : : : ; pn ; qn ; : : : are different primes. We consider all sets Si D

fr1 ; r2 ; : : : ; rn ; : : :g where, for each n, either rn D pn or rn D qn . Clearly, no Si
contains Sj if i ¤ j; and i runs over an index set I of the power of the continuum.
For each Si , define Ai to be the rational group of type ti D .k1 ; : : : ; kn ; : : :/, where
kn D 1 exactly if the nth prime belongs to Si , and kn D 0 otherwise. Then the ti are
pairwise incomparable, hence fAi gi2I is a rigid system. t
u
Evidently, Theorem 4.1 yields homogeneous indecomposable groups of any rank

2@0 . More explicit examples are provided by the following theorem, which is a
generalized version of the example by Pontryagin [1].
Lemma 4.6. For every r  2, there exists a rigid system of 2@0 torsion-free
groups fAi gi2I of rank r such that End Ai Š Z for each i 2 I. The groups Ai are
homogeneous of type t0 D .0; : : : ; 0; : : :/.
Proof. For any prime p, a single group A of the desired kind will be defined as
a subgroup of G D Q.p/ a1 ˚    ˚ Q.p/ ar , a completely decomposable rank r
group. First we dispose of the case r D 2. Choose a p-adic unit  D s0 C s1 p C
   C sn pn C    .0
sn < p/ which is transcendental over Z; such units exist in
abundance, since the transcendence degree of Jp over the rationals is of the power
of the continuum. Set

xn D pn .a1 C n a2 / 2 G; (12.5)

where n D s0 C s1 p C    C sn1 pn1 is the nth partial sum of the standard form
of . The subgroup A
G is defined as

A D ha1 ; a2 I x1 ; x2 ; : : : ; xn ; : : : i:

A is of rank 2, and we will now show that it is rigid with End A Š Z, and
homogeneous of the type t0 of Z.
Since pxnC1 D xn C sn a2 for every n, an element of A not in B D ha1 ; a2 i
has the form kxn C k2 a2 for some integers k; k2 with gcdfp; kg D 1, and for some
n  1. Furthermore, if pm .kxn C k2 a2 / D `a2 .` 2 Z/; then substituting (12.5), the
coefficient of a1 must be 0. We conclude that ha2 i is a (p-pure, and hence a) pure
subgroup of A. To show that A is homogeneous of type t0 , only divisibility by pn
needs to be checked. Assume pn .k1 a1 C k2 a2 / 2 A for all n > 0; here k1 ¤ 0,
since pn ja2 is impossible. Then also pn .k1 a1 C k2 a2  k1 pn xn / 2 A, and substitution
from (12.5) yields pn jk1  k2 n for all n, owing to the purity of ha2 i in A. Hence
the sequence fk2  k1 n gn! is a 0-sequence, therefore, its limit k2  k1  D 0.
But this means that  is a rational number, a contradiction. Consequently, the group
A is homogeneous of type t0 .
To show that every endomorphism  ¤ 0 of A is a multiplication by a rational
number, we do this (for convenience) for a multiple m of  (which we will denote
also by ) to be able to use that B
B.  is determined by the images

.a1 / D t11 a1 C t12 a2 ; .a2 / D t21 a1 C t22 a2 .tij 2 Z/:

434 12 Torsion-Free Groups

Then we have

.xn / D pn Œ.a1 / C n .a2 /

D pn Œt11 a1 C t12 a2 C n .t21 a1 C t22 a2 /

D kn xn C `n a2 D pn kn .a1 C n a2 / C `n a2

for certain kn ; `n 2 Z. Comparing coefficients on both sides, we obtain

t11 C n t21 D kn and pn Œt12 C n t22  kn n

D `n :

It follows that the coefficient in the square brackets must be divisible by pn . Letting
n ! 1, and setting  D lim kn , we are led to the equations

t11 C t21 D  and t12 C t22 D :

Thus t12 Ct22 D .t11 Ct21 /, or t21  2 C.t11 t22 / t12 D 0: The last equation is
of degree
2 in , so by the transcendency of  we get t11 D t22 and t12 D t21 D 0.
This shows that  acts as a multiplication by the integer t11 . If we do not assume that
B
B, then we can only conclude that  is a multiplication by a rational number.
But this must carry the pure subgroup ha2 i into itself, so it has to be an integer. It
follows that End A Š Z, as we wished to show.
If we want to construct a group of rank r > 2, then we select p-adic units
2 ; : : : ; r which are algebraically independent over Q. We define A as above
as a subgroup in G by adjoining to the independent set fa1 ; a2 ; : : : ; ar g, for each
j D 2; : : : ; r, additional generators

xjn D pn .a1 C jn aj / 2 G .n < !; 2

j
r/;

where jn denotes the nth partial sum in the standard form of j . Then the subgroups
Aj D ha1 ; aj i are fully invariant in A, any endomorphism of A acts as multiplication
by an integer on each Aj , and since ha1 i is common to all of them, these integers must
be the same for all the Aj .
Using more algebraically independent units, 2 ; : : : ; r ; 2 ; : : : ; r , we can
construct, besides A, also another group A0 of rank r with the aid of 2 ; : : : ; r .
A word-by-word repetition of the proof in the preceding paragraphs will convince
us that the only  W A ! A0 must be the zero map. Since there are continuously
many algebraically independent units in Jp , one can construct a rigid system
of continuously many homogeneous groups of any finite rank r  2 whose
endomorphism rings are Š Z. t
u
Indecomposable Groups of Infinite Ranks The various methods used above
open the door to a vast supply of large indecomposable groups. Our construction of
such groups relies on the next result.
4 Indecomposable Groups 435

Lemma 4.7 (Fuchs [12]). Let fE0 ; Ei .i 2 I/g be a rigid system of torsion-free
groups, and pi .i 2 I/ primes such that there are elements ui 2 E0 and vi 2 Ei not
divisible by pi . Then the group

A D hE0 ˚ .˚i2I Ei /I p1

i .ui C vi / 8 i 2 Ii

is indecomposable. The same holds if p1 1

i is replaced by pi or by n1
i for integers
ni whose prime divisors divide neither ui nor vi .
Proof. The proof of indecomposability runs as before. Rigity ensures the full
invariance of the subgroups E0 ; Ei in A. Therefore, if A D B ˚ C, then Ei D
.B \ Ei / ˚ .C \ Ei / for i 2 I [ f0g. Hence the indecomposability of the Ei implies
that each of Ei is contained entirely either in B or in C. Assume E0
B and Ej
C
for some j 2 I, and write p1
j .uj C vj / D b C c .b 2 B; c 2 C/. Hence uj D pj b and
vj D pj c, which is impossible in Ej , and likewise in A. It follows that all of E0 ; Ei
belong to the same summand B or C, and A is indecomposable. t
u
The Blowing Up Lemma We keep in mind that our ultimate goal in this section
is to establish the existence of arbitrarily large indecomposable groups. In order to
get ready for the proof, we will need a powerful result on blowing up a small system
of indecomposable groups to extremely large systems. First, a definition.
Let S denote a Z-algebra (in our application, S D Z). A fully rigid system
fGX j X  Ig of groups for S is a set of groups GX , indexed by the subsets X of a set
I, such that
(i) GX
GY whenever X  Y, and
(ii) Hom.GX ; GY / Š S or 0 according as X  Y or not.
(If S is a subring of Q, then a fully rigid system for S is a rigid system in the sense
above.) We quote the relevant result without proof.
Lemma 4.8 (Corner [8]). Suppose there exists a fully rigid system fNX j X  Jg
of groups for the Z-algebra S over a set J of at least six elements. If I is any infinite
set of cardinality  jNJ j, then there exists a fully rigid system fMX j X  Ig for S
such that

jMX j D jIj for all X  I:

Evidently, all the groups in a fully rigid system are indecomposable if 0 and 1
are the only idempotents in S.
Arbitrarily Large Indecomposable Groups By virtue of the preceding lemma,
in order to prove the existence of arbitrarily large indecomposable groups, it only
remains to find a fully rigid system for Z over a set with at least six elements. This
is what we are going to do in the next proof.
Theorem 4.9. (i) (Shelah [1]) For any cardinal , there exist indecomposable
torsion-free groups of cardinality .
436 12 Torsion-Free Groups

(ii) For every infinite cardinal , there exist fully rigid systems of 2 indecompos-
able groups of cardinality  with Z as endomorphism ring.
Proof. If  is a finite cardinal, then we refer to Theorem 4.1 or to Example 4.3. For
infinite cardinals , we exhibit fully rigid systems for Z over finite sets (of any size)
so that Lemma 4.8 can be applied for .
The group A of rank r constructed in Lemma 4.6 depends on the choice of the
p-adic units, so let us indicate this by A D A.2 ; : : : ; r /. Corner’s proof shows
that different choices of r algebraically independent p-adic units yield to a rigid set.
Now, choose I D f2 ; : : : ; r g as our index set for r  7, and for a subset X  I,
the group A.X/ as our GX to obtain a collection of groups. It should be clear from
the proof above that this is a fully rigid system for Z. t
u
Since a set of cardinality  has a set of 2 subsets none of which contains another,
it is clear that, for every infinite cardinal , there is a rigid system consisting of 2
groups—as many groups as the cardinality of the set of all non-isomorphic groups of
cardinality . We find the abundance of arbitrarily large indecomposable groups
quite remarkable, especially because their sheer existence was in doubt for quite a
while.
The following generalization of rigid systems turned out to be very useful.
Charles [4] calls a set of groups Ai .i 2 I/ a semi-rigid system if the index set
I is partially ordered such that

Hom.Ai ; Aj / ¤ 0 if and only if i

j in I:

The typical example is the set of isomorphy classes of all rank 1 torsion-free groups.
Several results proved for rigid systems can be extended to semi-rigid systems.
Purely Indecomposable Groups We now consider a property stronger than
mere indecomposability. A group A is purely indecomposable if every pure
subgroup of A is indecomposable.
Theorem 4.10 (Armstrong [1], Griffith [1]). A reduced torsion-free purely inde-
Q thus jAj
2@0 .
composable group A is isomorphic to a pure subgroup of Z;
Proof. If A is as stated, then its completion AQ is a reduced algebraically compact
group, so every 0 ¤ a 2 A embeds in a summand G of AQ that is Š Jp for some p.
Write AQ D G ˚ H, and note that both G \ A and H \ A are pure in A, and so is their
direct sum. Hence necessarily H \ A D 0. Thus A Š G \ A, a pure subgroup of Jp .ut
F Notes. The first example of an indecomposable torsion-free group of rank
2 was
constructed by Levi [1], but it seems that only a few people were aware of it (it was not published
in any journal). Much later, Pontryagin [1] gave an example of rank 2, and subsequently, Kurosh
[2] and Mal’cev [1] developed a theory of finite rank torsion-free groups based on Derry’s paper
[1] on p-adic modules, using equivalent matrices. They succeeded in proving the existence of
indecomposable groups of any finite rank. Baer’s elegant idea (see [6]) was a big step forward,
it furnished us with a supply of indecomposable groups up to rank 2@0 . In the early 1950s,
Szele tried really hard to construct examples beyond the continuum without success, and he was
willing to conjecture that the continuum was a barrier (which turned out to be the case indeed
4 Indecomposable Groups 437

for purely indecomposable groups). In 1958/1959 (after Szele’s death), Hulanicki [2], Sa̧siada [1],
@
and Fuchs [9] independently constructed indecomposable groups of cardinality 22 0 . Corner [6]
proved the existence up to the first inaccessible cardinal, correcting an unfortunate set-theoretical
mistake in Fuchs [12]. Indecomposable groups even for some measurable (but not for all) cardinals
were constructed in Fuchs [18]. The final step was furnished by Shelah [1] who got rid of the
cardinality restrictions. His proof was an amazing technical tour de force. Corner’s Blowing-up
Lemma 4.8 relies on Shelah’s ideas and technique. Of course, once the most powerful theorem
on the existence of large groups with prescribed cotorsion-free endomorphism rings is established
(Sect. 7 in Chapter 16), the existence of arbitrarily large indecomposable groups becomes a trivial
corollary by choosing Z as endomorphism ring.
Assuming V = L, Eklof–Mekler [1] prove that for every regular, not weakly compact cardinal
, there exist 2 non-isomorphic indecomposable strongly -free groups of cardinality . Dugas
[1] shows that these exist with Z as endomorphism ring.
There are exciting results on absolutely indecomposable torsion-free groups. These are groups
that are not only indecomposable, but they remain so under any generic extension of the universe,
i.e. they do not decompose into a proper direct sum no matter what kind of forcing is applied. The
existence of absolutely indecomposable groups has been established by Göbel–Shelah [5] up to
the first Erdős cardinal (which seems to be the natural boundary for such groups).
The first rigid system of size > 2@0 was constructed by de Groot [1]. The existence of arbitrarily
large rigid sets of torsion-free groups raises the question whether or not there exists a rigid proper
class. The existence of such a proper class is not provable in ZFC, see A. Kanamori–M. Magidor
[Springer Lecture Notes 669, 99–275 (1978)]. However, such a class can be constructed in L, as is
shown by Dugas–Herden [1]. Göbel–Shelah [1] establish the existence of a semi-rigid proper class
in ZFC.

Exercises

(1) (de Groot) Let p; p1 ; : : : ; pn denote distinct odd primes. Prove that the group
G D hp1 1 1
1 a1 ; : : : ; pn an ; p .a1 C    C an /i is indecomposable.
(2) The group A D hB ˚ C; p1 .b C c/i (p any prime) is not indecomposable if
B; C are rank 1 torsion-free groups with t.B/
t.C/.
(3) (a) The pure subgroups A; C of Jp are isomorphic if and only if there is a p-adic
unit  such that multiplication by  yields an isomorphism A ! C. [Hint:
Jp is algebraically compact, and End Jp Š Jp .]
(b) Let r 2 N or r D @0 . Jp contains a rigid system (of size 2@0 ) of
indecomposable pure subgroups of rank r.
(4) Examine the proof of Lemma 4.6, and show that it would be enough to assume
that the p-adic integers j are quadratically independent.
(5) (Corner) For every integer m  2, there is a torsion-free group A of rank m with
the following properties:
(i) all subgroups of rank
m  1 are free;
(ii) all torsion-free factor groups of rank 1 are divisible;
(iii) End A Š Z. [Hint: xn D pn .a1 C 2n a2 C    C mn am /:]
(6) If a reduced group A D B ˚ C contains a subgroup G Š Jp , then either G
B
or G \ B D 0. [Hint: G \ B is pure in G, so G=.G \ B/ is divisible.]
438 12 Torsion-Free Groups

(7) (de Groot) (a) A torsion-free group is purely indecomposable if any two
independent elements have incomparable types.
(b) Let V be a Q-vector space with basis fa1 ; : : : ; an ; : : :g. Arrange the linear
combinations b D k1 an1 C    C kr anr with n1 <    < nr and integers
0 ¤ ki 2 Z; k1 2 N; gcdfk1 ; : : : ; kr g D 1 in a sequence. To the jth b
in the sequence assign the characteristic .0; : : : ; 0; 1; 0; : : :/ with 1 only
at the jth prime to obtain a subgroup A of V. Prove that A is purely
indecomposable.

5 Pathological Direct Decompositions of Finite Rank Groups

From the point of view of direct decompositions, the existence of large inde-
composable groups is a major difference between torsion and torsion-free groups.
Another striking difference between them is the occurrence of various phenomena
in the direct decompositions of torsion-free groups already in the finite rank case.
The situation is more cumbersome than one might expect, and we shall spend some
time to make it more transparent.
It is evident that a torsion-free group of finite rank decomposes into the direct
sum of a finite number of indecomposable groups. The question we pose is this: is
there any kind of uniqueness in these decompositions? The best way to describe the
answer is to quote Kaplansky who claimed that in this strange part of the subject,
anything that can conceivably happen actually does happen.
We will list several examples of pathological decompositions, involving lots of
computations (in particular, divisibility considerations). Complicated technicalities
are, however, unavoidable in view of the nature of the subject, and unfortunately,
they only provide a very superficial insight into the structure of finite rank torsion-
free groups.
Decompositions with the Same Number of Summands Our first theorem is
a variant of Jónsson’s original counterexample that set the stage for further strange
examples. Recall: for a set P primes, we use P1 a as an abbreviation of the set
fp1 a j p 2 Pg.
Theorem 5.1. There are rank 3 torsion-free groups that have many non-isomorphic
direct decompositions into indecomposable summands.
Proof. Let P; Q denote two infinite disjoint sets of primes, and p > 3 a prime p …
P [ Q. Define

A D hP1 ai ˚ hP1 b; Q1 c; p1 .b C c/i

with independent a; b; c. Pick s 2 Z with gcdfs; pg D 1, s 6 0; ˙1 mod p, and find

r; t 2 Z satisfying sr  pt D 1. Then a0 D sa C tb; b0 D pa C rb; c are independent,
and we claim that
5 Pathological Direct Decompositions of Finite Rank Groups 439

A0 D hP1 a0 i ˚ hP1 b0 ; Q1 c; p1 .sb0 C c/i

is equal to A. As hP1 ai ˚ hP1 bi D hP1 a0 i ˚ hP1 b0 i is obvious, we prove

A0 D A by showing that the given relations imply that b C c is divisible by p if and
only if sb0 C c is divisible by p. But this is rather obvious in view of sb0 C c D
spa C srb C c D spa  ptb C .b C c/: It is also clear that the rank 2 summands above
are indecomposable. It remains to prove that these summands are not isomorphic.
Any isomorphism  between them must act as b 7! ˙b0 (i.e. C or ) and c 7! ˙c,
since  must respect divisibility properties. But then .b C c/ D ˙b0 ˙ c, which is
impossible, since these elements are not divisible by p in A. t
u
Even if we restrict ourselves to direct decompositions with a fixed number
of indecomposable summands, the rank distribution of the summands can be
arbitrary—this is shown by the following striking result.
Theorem 5.2 (Corner [1]). Suppose n; k are the integers such that 1
k
n.
There exists a torsion-free group A of rank n with the following property: to every
partition n D r1 C    C rk of n into k integers ri  1, there is a decomposition
A D A1 ˚    ˚ Ak where Ai is indecomposable of rank ri .
Proof. Let p; p1 ; : : : ; pnk ; q1 ; : : : ; qnk denote different primes. Select independent
elements a1 ; : : : ; ak ; b1 ; : : : ; bnk , and define

A D hp1 a1 ; : : : ; p1 ak ; p1 1

1 b1 ; : : : ; pnk bnk ;

q1 1
1 .a1 C b1 /; : : : ; qnk .a1 C bnk /i:

To show that this group has the indicated property, let n D r1 C  Crk be a partition
of n with ri  1. If s1 ; : : : ; sk are integers such that s1 C   Csk D 1, then the system

s1 c1 C    Csk ck D a1
c1 C c2 D a2
: : : : : : : : : : : :: : : : : : : : : : : :
c1 Cck D ak

with determinant 1 is solvable for c1 ; : : : ; ck 2 A, and hc1 ; : : : ; ck i D ha1 ; : : : ; ak i

holds. Set t1 D 0, and ti D .r1  1/ C    C .ri1  1/ for i D 2; : : : ; k. Using these
ci , we define

Ai D hp1 ci ; p1
j bj ; q1
j .ci C bj / .j D ti C 1; : : : ; tiC1 /i

for i D 1; : : : ; k. In view of

a1 C bj D .ci C bj / C s1 c1 C    C si1 ci1 C .si  1/ci C    C sk ck ;

440 12 Torsion-Free Groups

it is routine to check that if the numbers s1 ; : : : ; sk are chosen such that

(
1 mod qj if j D ti C 1; : : : ; tiC1 ;
si 
0 mod qj otherwise;

then the Ai will be subgroups of A, and every generator of A will be contained in

A1 C  CAk . This yields at once A D A1 ˚  ˚Ak where by construction, r.Ai / D ri ,
as desired. That the Ai are indecomposable follows as in Example 4.3. t
u
For instance, the numbers si can be chosen as si D mti C1 qO ti C1 C    C mtiC1 qO tiC1 ,
P
where qO j D q1 : : : qj1 qjC1 : : : qnk and kjD1 mj qO j D 1.
In our endeavor to find some sort of uniqueness under reasonable restrictions
on the decompositions, we try comparing only decompositions with the same
distribution of ranks. This is still not enough to ensure isomorphy.
Theorem 5.3 (Fuchs–Loonstra [1]). Given any integer m  2, there exist inde-
composable groups, A and C, of rank 2 such that

A ˚˚A Š C ˚˚C .n summands/

if and only if n  0 mod m.

Proof. We start with two disjoint, infinite sets P1 ; P2 of primes, and a prime p (to
be suitably chosen later on) such that p … P1 [ P2 . We define the following rank 1
groups:

Xi D hP1
1 xi i; Yi D hP1
2 yi i .i D 1; : : : ; n/:

With independent xi ; yi , we form the isomorphic indecomposable groups

Ai D hXi ˚ Yi ; p1 .xi C yi /i .i D 1; : : : ; n/:

Next, we choose groups Ui Š Xi and Vi Š Yi .i D 1; : : : ; n/ and let ui 7! xi ,

vi 7! yi .ui 2 Ui ; vi 2 Vi / under some fixed isomorphisms. For any choice of
k D 1; : : : ; p  1, we form the isomorphic groups

Ci D hUi ˚ Vi ; p1 .ui C kvi /i .i D 1; : : : ; n/

which are also indecomposable.

Suppose  W A D A1 ˚    ˚ Am ! C D C1 ˚    ˚ Cm is an isomorphism.
The choice of the groups Xi ; Yi assures that  maps X D X1 ˚    ˚ Xm and Y D
Y1 ˚    ˚ Ym upon U D U1 ˚    ˚ Um and V D V1 ˚    ˚ Vm , respectively. In
particular,
5 Pathological Direct Decompositions of Finite Rank Groups 441

X
m X
m
.xi / D rij uj and .yi / D sij vj ; (12.6)
jD1 jD1

where rij ; sij are integers such that the matrices krij k; ksij k are invertible:

det krij k D ˙1 and det ksij k D ˙1: (12.7)

 has to preserve divisibility by p, therefore

X
m X
m
.xi C yi / D rij .uj C kvj / C .sij  krij /vj
jD1 jD1

taken in combination with the independence of the vj implies

sij  krij mod p for all i; j: (12.8)

On the other hand, (12.7) and (12.8) together imply that  as given in (12.6) is in
fact an isomorphism A ! C, since pjxi C yi in A for all i implies pjuj C kvj in C for
all j, and vice versa.
It remains to choose the coefficients rij ; sij , the prime p, and the integer k in a
suitable way to avoid isomorphism for n summands if m6 jn. From (12.7) and (12.8)
we obtain km  ˙1 mod p. On the other hand, in order to exclude isomorphy for
n < m, k will be chosen so as to satisfy

km  1 mod p; but kn 6 ˙1 mod p .n D 1; : : : ; m  1/: (12.9)

Observe that if k satisfies (12.9), then it satisfies kn  ˙1 mod p if and only if njm.
By Dirichlet’s theorem on primes in arithmetic progressions, we can find a prime p
such that p  1 mod 2m, and by a well-known theorem in number theory, there is a
primitive root t mod p. If we set k D t.p1/=2m , then (12.9) is satisfied.
Let ` be an integer such that k`  1 mod p, and define two m  m matrices:
0 1 0 1
1 k k0
B` 1 C B1 C
B C B k C
B0 `1 C B0 1k C
B C B C
B C B C
B` 1` 1 C; B1 k1 k C:
B C B C
B: : : : : : : : C B: : : : : : : : C
B C B C
@: : : : 1`1 A @: : : : k1k A
: : : : ` 1 ` 1 : : : : 1 k 1 k

In the first matrix, we have 1s in the main diagonal except for the last entry, below
the diagonal we have alternately ` and 0 in the first column, ` and 1 in the other
columns. The second matrix is obtained from the first by substituting k for 1 and
442 12 Torsion-Free Groups

1 for ` throughout, and finally, putting k0 D .1/mC1 .km C 1/ in the upper right
corner. A simple calculation is needed to check that if we choose the first matrix for
krij k, and the second matrix for ksij k, then all of (12.7), (12.8) and (12.9) will be
satisfied. t
u
Different Numbers of Summands The next result tells us that even the
number of indecomposable summands can be different in the decompositions. (Here
‘summand’ will mean ‘non-zero summand.’)
Theorem 5.4 (Fuchs–Gräbe [1]). For every integer n  1, there exists a torsion-
free group of rank 2n which has decompositions into the direct sum of m indecom-
posable summands for any m with 2
m
n C 1.
Proof. Let p1 ; : : : ; pn ; q2 ; : : : ; qn ; r denote different primes. For 1
k
n, define
the group

Ank Dhp1 1
1 a1 i ˚    ˚ hpk ak i
˚ hp1 1 1 1 1 1
1 b1 ; : : : ; pn bn ; r q2 .b1  b2 /; : : : ; r qn .b1  bn /i;

where the summands are indecomposable. If s; t; u; v are integers such that vsut D
1, then set ci D sai C tbi ; di D uai C vbi in order to get hp1
i ai i ˚ hp1
i bi i D
hp1
i c i i ˚ hp 1
i d i i for i D 1; : : : ; k. Define

C D hp1 1
1 c1 ; : : : ; p k ck ; p1 1 1 1
kC1 bkC1 ; : : : ; pn bn ; r qi u.c1  ci /;

r1 q1 1 1
j .uc1 C bj /; r qj .b1  bj / .1 < i
k < j
n/i;

D D hp1 1
1 d1 ; : : : ; pk dk ; r1 q1
i s.d1  di / .1 < i
k/i:

These groups are indecomposable, and we want to choose s; t; u; v so as to have

Ank D C ˚ D. To ensure that C
Ank , the divisibility relations

rqi j u.c1  ci / D usa1  usai C ut.b1  bi /;

rqj j uc1 C bj D usa1 C .ut C 1/b1 C .bj  b1 /

must hold for all indices 1 < i

k < j
n. Thus rqi j us for i D 2; : : : ; n, and
rqj j .ut C 1/ D vs for j D k C 1; : : : ; n. Hence gcdfu; vg D 1 implies rqj j s. Thus,
the conditions

rqi j us .i D 2; : : : ; k/ and rqj j s .j D k C 1; : : : ; n/ (12.10)

must be satisfied. For D

Ank to hold, we must have

rqi j s.d1  di / D sua1  suai C sv.b1  bi / .i D 2; : : : ; k/;

5 Pathological Direct Decompositions of Finite Rank Groups 443

which are satisfied whenever (12.10) holds. Since ai D vci  tdi and bi D uci C
sdi .i D 1; : : : ; k/, we have

r1 q1 1 1 1 1
i .b1  bi / D r qi u.c1  ci / C r qi s.d1  di / .i D 2; : : : ; k/;

r1 q1 1 1 1 1
j .b1  bj / D r qj .uc1 C bj / C r qj sd1 .j D k C 1; : : : ; n/;

showing that Ank

C ˚ D, i.e. Ank D C ˚ D provided (12.10) holds. It does if we
set u D q2 : : : qk and s D r  qkC1 : : : qn .
Now we choose A D Ann of rank 2n, and observe that A D Ank ˚ hp1
kC1 akC1 i ˚
   ˚ hp1
n an i for each k, where by the above, the first component has a direct
decomposition into two indecomposable summands. As 1
k
n, the number of
indecomposable summands in direct decompositions of A could be any m provided
2
m
n C 1. t
u
No Cancellation As our final attempt in search of uniqueness we consider
the cancellation property. We say that a finite rank torsion-free group A has the
cancellation property if, for any finite rank torsion-free groups, G and H, A ˚ G Š
A ˚ H implies G Š H. An equivalent formulation is: A ˚ G D A0 ˚ H with
A Š A0 implies G Š H. The next result shows that not all rank 1 groups share the
cancellation property (though some do, see Sect. 2, Exercise 12 in Chapter 3).
Theorem 5.5 (Fuchs–Loonstra [1]). For every integer m  1, there exists a
torsion-free group A of rank 3 which has m decompositions

A D Bi ˚ Ci .i D 1; : : : ; m/

such that B1 Š    Š Bm are of rank 1, while C1 ; : : : ; Cm are pairwise non-

isomorphic, indecomposable groups of rank 2.
Proof. We start off with two disjoint, infinite sets P1 ; P2 of primes, and a prime
p … P1 [ P2 . We define the groups

B1 D hP1
1 b1 i; X1 D hP1
1 x1 i; Y D hP1
2 yi:

If b1 ; x1 ; y are independent, then with the indecomposable group C1 D hX1 ˚Y,

p1 .x1 C y/i we set

A D B1 ˚ C1 :

Now choose integers qi ; ri ; si ; ti .i D 2; : : : ; m/ such that qi ti  ri si D 1, and let

bi D qi b1 C si x1 ; xi D ri b1 C ti x1 .i D 2; : : : ; m/:

Then we have Bi ˚ Xi D B1 ˚ X1 for the pure subgroups Bi D hbi i ; Xi D hxi i

isomorphic to B1 . We intend to choose Bi ; Xi .i  2/ such that, for some integers ki
we will have
444 12 Torsion-Free Groups

A D Bi ˚ Ci with Ci D hXi ˚ Y; p1 .ki xi C y/i (12.11)

where the Ci ’s are non-isomorphic (and indecomposable). Then to ensure that ki xi C

y D ki ri b1 C .ki ti  1/x1 C .x1 C y/ will be divisible by p, we must select ki such
that

ki ri  0 and ki ti  1 mod p: (12.12)

For any ki with 1 < ki < p, we choose ri D p; qi D ki ; and si ; ti so as to satisfy

qi ti  ri si D 1; then (12.12) holds, and hence so does (12.11).
We still have to ascertain that the Ci are pairwise non-isomorphic. Observe
that the only automorphisms of the Xi and Y are multiplications by ˙1, thus any
isomorphism  W Ci ! Cj must map Xi upon Xj , and Y upon itself such that
.xi / D ˙xj ; .y/ D ˙y. This means that we have .ki xi C y/ D ˙.kj xj ˙ y/.
The preservation of divisibility forces kj  ˙ki mod p. Consequently, if we choose
.k1 D 1/; k2 D 2; : : : ; km D m and a prime p > 2m  1, then we guarantee that
kj 6 ˙ki mod p for i ¤ j, and therefore, no two of the groups C1 ; : : : ; Cm are
isomorphic. t
u
This theorem also shows that, for any rank r  3, and for every integer m, there
exist torsion-free groups of rank r which have at least m pairwise non-isomorphic
decompositions into indecomposable summands. However, Lady’s theorem 6.9 will
show that no torsion-free group of finite rank admits infinitely many non-isomorphic
decompositions.
Possible Rank Distributions The question of rank distributions in direct
sums of finite rank torsion-free groups is too important to leave out completely.
The problem was completely solved by Yakovlev and Blagoveshchenskaya in a
satisfactory way. The proof is too long, it requires further structural information
to be determined, so here we cannot do more than just stating the theorem without
proof. As expected, the rank 1 summands play a special role.
Theorem 5.6 (Yakovlev [1], Blagoveshchenskaya–Yakovlev [1]). Let

n D r1 C    C rk D s1 C    C s`

be two partitions of the positive integer n  3 .ri ; sj 2 N/. Let u and v denote the
number of ri and sj , respectively, that are equal to 1.
A necessary and sufficient condition for the existence of a torsion-free group of
rank n to admit a direct decomposition into indecomposable summands of ranks ri ,
and also one with indecomposable summands of ranks sj , is that
(i) ri
n  v for all i, and sj
n  u for all j; and
(ii) if ri D n  v for some i, then exactly one sj ¤ 1 .and is equal to n  v/, and if
sj D n  u for some j, then exactly one ri ¤ 1 .and is equal to n  u/. t
u
The proof of sufficiency is based on a clever construction of special indecompos-
able groups, called flower groups.
5 Pathological Direct Decompositions of Finite Rank Groups 445

Another result deals with the precise spectrum of the numbers of indecomposable
summands in possible direct decompositions. Permitted as well as forbidden
numbers can be selected as desired. We quote it without its lengthy proof.
Theorem 5.7 (Fuchs–Gräbe [1]). Let N be any finite collection of integers  2,
and n the largest integer in N. There exists a torsion-free group of rank
2.n  1/
which has a decomposition into the direct sum of k indecomposable summands ¤ 0
if and only if k 2 N. t
u
F Notes. After reading all the above strange decompositions that torsion-free groups of
finite rank display, one might get the false impression that these are the norm rather than the
exception. As a matter of fact, the first pathological example by Jónsson [1] was quite a surprise,
and even some group theorists had doubts about the correctness of his numerical example. This
was an apparent setback in the structure theory, though Jónsson himself tried to restore an order
by introducing an equivalence relation weaker than isomorphism (see Theorem 9.9), shaking our
belief in isomorphism as an exclusive principle of classification. Thanks to E. Walker [2], we know
that the best way of viewing quasi-isomorphism is to interpret it as an isomorphism in an adequate
quotient category. Quasi-isomorphism clarified the direct decompositions, but unfortunately it has
not contributed too much to the structure problem: the results by Beaumont–Pierce [2], as well as
numerous later theorems (e.g., Mutzbauer [1], Richman [2], etc.) on rank 2 groups are a convincing
evidence that there is still a long way to go (if possible at all) to classify finite rank torsion-
free groups even up to quasi-isomorphism. There is an extensive literature on pathological direct
decompositions; one of the earliest is Jesmanowicz [1].
For more important material on direct decompositions of finite rank torsion-free groups, we
refer to Faticoni [Fa].

Exercises

(1) A finite rank torsion-free group A D B ˚ C may have a summand which is

completely decomposable of rank r > 0, even if neither B nor C has rank one
summands.
(2) There is no torsion-free group of rank 4, which can be written as a direct sum
of two rank 1 groups, and an indecomposable rank 2 group, and also as a direct
sum of a rank 1 and an indecomposable rank 3 group. [Hint: look at the types
of the rank 1 summands.]
(3) Let r1 ; : : : ; rk1 2 N and n D r1 C    C rk1 . There exists a group A of rank
2n such that A D A1 ˚    ˚ Ak1 ˚ Ak D B ˚ C, where Ai is indecomposable
of rank ri .i D 1; : : : ; k  1/, while the groups Ak ; B; C are indecomposable of
rank n. [Hint: in the proof of Theorem 5.1, replace rank 1 groups by groups
from a suitable rigid system.]
(4) (Fuchs–Loonstra) For every m  2, there exist pairwise non-isomorphic
indecomposable groups B; C1 ; : : : ; Cm of rank 2 such that B ˚    ˚ B Š
C1 ˚    ˚ Cm (m summands B), but the direct sum of n .< m/ copies of B
is not isomorphic to the direct sum of any n groups in the set fC1 ; : : : ; Cm g.
[Hint: argue as in Theorem 5.3.]
446 12 Torsion-Free Groups

(5) For every integer n  1, there is an indecomposable group A of rank n such that
A ˚ B Š A ˚ C for suitable non-isomorphic indecomposable groups B; C of
finite rank. [Hint: Theorem 5.5.]
(6) (Sa̧siada) If X D h51 x; 31 y; 21 .x C y/i, then every indecomposable
summand of X .n/ for any n > 1 is isomorphic to X. [Hint: any such summand
must have a subgroup Š h51 x; 31 yi of index 2.]
(7) (a) (Corner) Let A be torsion-free of rank n, and B; C isomorphic pure
subgroups of A of rank n  1. Then A=B Š A=C. [Hint: reduce to
n D 2; B ¤ C, and calculate the torsion subgroup of A=.B C C/ in two
ways.]
(b) Cancellation by a finite rank torsion-free group is permitted if the comple-
ments are of rank 1.

6 Direct Decompositions of Finite Rank Groups:

Positive Results

We have seen that a torsion-free group of finite rank may have numerous non-
isomorphic direct decompositions. Is there anything positive which can be stated
about such decompositions? In particular, must such a group have but finitely many
non-isomorphic such decompositions? Lady [3] has answered this question in the
positive.
This is an important result, therefore we give a detailed proof, though from here
on nearly everything is ring theory. The proof of the main Theorem 6.9 relies on
several not so well-known facts from ring theory, in particular, on a deep result by
Jordan–Zassenhaus that will be quoted without proof.
Lemmas on Rings and Modules In the next lemma, E stands for the endomor-
phism ring of A.
Lemma 6.1. Let B and B0 be direct summands of a group A, and let ;  0 denote
the projections onto them. Then B Š B0 if and only if the right ideals E and  0 E
are isomorphic as right E-modules.
Proof. See the proof of the more complete Lemma 1.6 in Chapter 16. t
u
We break down the discussion of the required ring-theoretical background into
a series of easy lemmas. The first lemma can be found in most text books on rings
and modules. R will denote a ring with 1, and J its Jacobson radical.
Lemma 6.2 (Nakayama’s Lemma). If M is a finitely generated right R-module
such that M D MJ, then M D 0.
Proof. Let fx1 ; : : : ; xk g with k  1 be a minimal generating system for M. By
hypothesis, we have x1 D x1 r1 C    C xk rk with ri 2 J. Since 1  r1 is invertible in
R, x1 can be expressed as a linear combination of the other generators, contrary to
the minimality of the generating system. Thus k D 0, and hence M D 0. t
u
6 Direct Decompositions of Finite Rank Groups: Positive Results 447

A right ideal L of R is nilpotent if Lk D 0 for some integer k  1. The union

of all nilpotent right ideals of R is a two-sided ideal N called the nilpotent radical
of R. Always, N
J. We say that a ring R is torsion-free or of finite rank if its
additive group has this property.
Lemma 6.3. For a torsion-free ring R of finite rank, the nilpotent radical N is
nilpotent and pure in R.
Proof. It is immediate that the pure closure of a nilpotent ideal is likewise nilpotent;
hence, N is pure in R. If L is a pure nilpotent right ideal of R of maximal rank, then
for every nilpotent right ideal K of R; L C K is nilpotent, and must be of the same
rank as L. As L is pure, we obtain L C K D L; i.e. K
L; and hence N
L. t
u
Lemma 6.4. If R is a finite dimensional algebra over a field F, then J D N, and
R=J is a semi-simple artinian ring.
Proof. If n is such that Jn has (as an F-vector space) the smallest dimension among
the powers of the Jacobson radical J, then JnC1 D Jn , and Nakayama’s lemma
implies Jn D 0. Hence J D N, and the rest is classical. t
u
Lemma 6.5. Let N be the nilpotent radical of the torsion-free ring R of finite rank,
and A; B two finitely generated projective right R-modules. Then A Š B if and
only if

A=AN Š B=BN:

Proof. An isomorphism  W A ! B evidently induces an isomorphism A=AN !

B=BN. Conversely, assume W A=AN ! B=BN is an isomorphism. Consider the
diagram

φ
A B
⏐ ⏐
⏐
α
⏐β

ψ
A/AN −−−−→ B/BN

where the vertical maps are the canonical ones; the existence of a map  W A ! B
making the square commute follows from the projectivity of A. As ˇ.D ˛/ is
surjective, A C BN D B follows. Applying Nakayama’s lemma to B=A (with J
= N), we obtain A D B. Thus A Š B ˚ Ker  by the projectivity of B. Hence
rk A  rk B, and by symmetry, we obtain rk A D rk B. Therefore, Ker  D 0, and 
is an isomorphism. t
u
Lemma 6.6. A torsion-free ring R of finite rank n contains a subring S whose
additive group is a free group of rank n. If R has an identity, S can be chosen to
contain it.
448 12 Torsion-Free Groups

Proof. Let fx1 ; : : : ; xn g be a maximal independent set in the additive group of R.

There are tijk 2 Q such that xi xj D tij1 x1 C    C tijn xn . If m 2 Z is a common
denominator of the tijk , then the subgroup generated by the elements yi D mxi is a
subring S of R with free additive group. If R has identity, it can be adjoined to S. u
t
The Jordan–Zassenhaus Lemma So far we have managed quite well with the
proofs, but now we have come to a deep result about the finiteness of certain right
modules, without which it seems we cannot prove Lady’s theorem. This result is
fundamental in representation theory (we state it only for the ring Z rather than for
Dedekind domains).
Theorem 6.7 (Jordan–Zassenhaus Lemma (for Z)). Let R be a finite-
dimensional semisimple Q-algebra, and S a subring of R such that the additive
groups of S and R have the same rank. For any right R-module H, there are but
finitely many non-isomorphic right S-submodules M of H such that MR D H. u t
This result fits perfectly to verify the following lemma.
Lemma 6.8. A torsion-free ring R of finite rank has but a finite number of non-
isomorphic left ideals that are summands of R.
Proof. In view of Lemma 6.5, the proof can be reduced to the case when R has
0 radical. In this case QR is a semi-simple artinian Q-algebra. Let S be a subring
of QR as stated in Theorem 6.7. Evidently, S can be replaced by nS
R with 1
adjoined, i.e. S
R can be assumed. As a semi-simple artinian ring, QR has up to
isomorphism but a finite number of right ideals. By Theorem 6.7, there exist only a
finite number of non-isomorphic S-modules in QR, say, M1 ; : : : ; Mt .
If the right ideal L is a summand of R, then L D R for an idempotent . For
some integer n > 0, we have n 2 S; and there is an S-isomorphism nS ! Mi
for some i. This extends to an R-isomorphism nR ! Mi R. As nR Š R as right
R-modules, the proof is complete. t
u
Lady’s Theorem The lemma just proved, if combined with Lemma 6.1, leads
us at once to the following major result.
Theorem 6.9 (Lady [3]). A torsion-free group of finite rank can have but a finite
number of non-isomorphic direct summands. t
u
It seem sensible at this point to illustrate how the situation is drastically simplified
if we move to Jp -modules. There exist only two non-isomorphic rank 1 torsion-free
Jp -modules, viz. Jp and its field of quotients, Qp . Both are algebraically compact,
and as a consequence, every finite rank torsion-free Jp -module is a direct sum of a
finite number of copies of these two modules, so completely decomposable. (By the
way, this also holds if the rank is countable.) Moreover, such direct decompositions
are unique up to isomorphism—this is rather obvious in this situation, and also
follows from the fact that these modules have the exchange property. We can claim
that the torsion-free Jp -modules of finite rank have the Krull–Schmidt property.
F Notes. The Lady’s theorem is one of the most important results in the theory of torsion-free
groups. A group-theoretical proof would be most welcome.
7 Substitution Properties 449

The Jordan–Zassenhaus lemma plays a prominent role in Representation Theory, see. e.g., C.W.
Curtis–I. Reiner, Representation Theory of Finite Groups and Associative Algebras, Sect. 10. As a
matter of fact, this lemma is intrinsically tied to the finiteness of non-isomorphic summands, see,
e.g., Fuchs–Vámos [J. Algebra 230, 730–748 (2000)].
Lots of interesting positive results on direct decompositions of finite rank torsion-free groups
may be found in Faticoni [Fa], using categorical approach.

Exercises

(1) Let A be a torsion-free group of infinite rank whose endomorphism ring is of

finite rank. Then
(a) A is a direct sum of a finite number of indecomposable groups;
(b) A has only a finite number of non-isomorphic direct decompositions.
(2) Let A D B ˚ C, where B; C are torsion-free of countable rank. A might have
infinitely many non-isomorphic direct decompositions even if both B and C
have but finitely many. [Hint: Theorem 1.1 of Chapter 13.]

7 Substitution Properties

The Substitution Property Another positive result for finite rank groups we
wish to draw attention to is the substitution property. A group A has the substitution
property if any pair of decompositions

G D A1 ˚ B D A2 ˚ C with A1 Š A Š A2

implies that there exists an A0

G such that

G D A0 ˚ B D A0 ˚ C:

In other words, summands isomorphic to A can be replaced by the same subgroup,

by the same copy of A, in both decompositions. (Notice the fine distinction between
equality and isomorphism.)
We take note that in general, none of the exchange, the substitution, and the
cancellation properties implies the other, with one exception, see (c). This is
evidenced by the following counterexamples.
Example 7.1. The group A D ˚n<! Z.p1 / is injective, so it has the exchange property, but
neither the substitution, nor the cancellation property. In fact, A cannot be cancelled in the direct
sums A ˚ Z.p1 / ˚ Z.p1 / Š A ˚ Z.p1 /.
Example 7.2. The additive group of the ring R of rational numbers with denominators coprime to
pq for primes p ¤ q does not have the exchange property, since its endomorphism ring R is not
450 12 Torsion-Free Groups

local (Theorem 6.2 in Chapter 6). But it does have the substitution property; this will follow from
(d) below, since such an R has 1 in the stable range.
Example 7.3. The group Z has the cancellation, but neither the exchange nor the substitution
property. No exchange property, since its endomorphism ring is not local (Theorem 6.2 in
Chapter 6). For cancellation, see Sect. 2, Exercise 12 in Chapter 3. Let p be a prime, and k 2 N
such that k 6 1; 0; 1 mod p. Let m; n 2 N be chosen such that mk  np D 1. Consider the
free group F D hxi ˚ hyi D hx0 i ˚ hy0 i where x0 D mx C ny; y0 D px C ky. If z 2 F satisfies
F D hzi ˚ hyi, then it must be of the form z D ˙x C `y for some ` 2 N. But such a z can
never satisfy F D hzi ˚ hy0 i, because y is contained in this direct sum only if k  `p D ˙1.
Consequently, the substitution property fails for Z.

(a) If the groups A; B have the substitution property, then so does their direct sum
A ˚ B. This is straightforward.
(b) Summands inherit the substitution property. Let A D B˚C have the substitution
property, and let G D B1 ˚ H D B2 ˚ K with Bi Š B. Apply the hypothesis to
G ˚ C D .B1 ˚ C/ ˚ H D .B2 ˚ C/ ˚ K to find A0 Š A such that G ˚ C D
A0 ˚ H D A0 ˚ K. Then B0 D G \ A0 can replace both Bi in the direct sums.
(c) For all groups, substitution implies cancellation. What we have to show is that
if A0 ˚ B D A0 ˚ C, then B Š C—which is pretty evident.
Stable RangeP Let R be a ring with 1. A subset fa1 ; : : : ; an g of R is called right
unimodular if niD1 ai R D R. The integer n > 0 is in the stable range of R if, for
every right unimodular subset fa1 ; : : : ; anC1 g of R, there exist b1 ; : : : ; bn 2 R such
that the set

fa1 C anC1 b1 ; : : : ; an C anC1 bn g

is right unimodular. In particular, R has 1 in the stable range if ax C by D 1

(a; b; x; y 2 R) implies the existence of a z 2 R such that a C bz is a unit in R;
equivalently, there is a z0 2 R with au C bz0 D 1 for a unit u 2 R. (It can be shown
that this definition is left-right symmetric, but this fact is irrelevant for us.)
(d) It is a well-known theorem by H. Bass that 1 is in the stable range of R
whenever R/ J is artinian; here J denotes the Jacobson radical of R.
The principal result on the substitution property is as follows.
Theorem 7.4 (Warfield [9]). A group has the substitution property if and only if
its endomorphism ring has 1 in the stable range.
Proof. This is a special case of the more general Theorem 7.8 that will be proved
below. t
u
A noteworthy consequence of the preceding theorem is as follows.
Corollary 7.5. Groups with semi-local endomorphism rings have the substitution
property.
Proof. Combine (d) with Theorem 7.4 to argue that semi-local rings have 1 in the
stable range. t
u
7 Substitution Properties 451

Example 7.6. (a) A ring R of rational numbers which is p-divisible for almost all primes p is
semi-local, so it has 1 in the stable range. A torsion-free group whose endomorphism ring is
isomorphic to this R has the substitution property.
(b) 1 is not in the stable range of Z: 2Z C 5Z D Z, but 2 C 5n D ˙1 for no n 2 Z.

In the following proof we use the fact that in the endomorphism ring E of a finite
rank torsion-free group A, one-sided units are two-sided units: if ; ı 2 E satisfy
ı D 1A , then is monic (counting the ranks), so it can be cancelled on the left in
ı D , to get ı D 1A .
Proposition 7.7 (Arnold [A]). Let A be a torsion-free group of finite rank.
(i) If 1 is in the stable range of E D End A, then for each n 2 N, every unit of
E=nE lifts to a unit of E.
(ii) If A is strongly indecomposable .see Sect. 9/, and if every unit of E=nE lifts to
a unit of E for every n 2 N, then 1 is in the stable range of E.
Proof. (i) If for ˛ 2 E, ˛ C nE 2 E=nE is a unit, then ˛ˇ C n D 1 for some
ˇ;  2 E. By hypothesis, ˛ C n is a unit in E for some  2 E.
(ii) We refer to Proposition 9.6 to claim that the quasi-endomorphism ring QE of
a strongly indecomposable A is a local ring. Suppose ˛; ˇ; ; ı 2 E are such
that ˛ˇ C ı D 1. If ı 2 J (Jacobson radical of QE), then it is nilpotent
(Lemma 6.4), and ˛ˇ D 1  ı is a unit in E, so ˛ D ˛ C  0 is a unit in E. If
ı … J, then is a unit of QE, i.e. 0 D n ¤ 0 for some 0 2 E; n 2 Z. The
artinian ring E=nE has 1 in the stable range, so (d) implies that the same holds
for E. t
u
n-Substitution A notable generalization of the substitution property is due to
Warfield [9] who defined the n-substitution property .n 2 N) for a group A as
follows: if

G D A1 ˚    ˚ An ˚ B D A0 ˚ C with Ai Š A .i D 0; : : : ; n/;

then there exists A0

G such that

G D A0 ˚ L ˚ B D A0 ˚ C with A0 Š A; L
A1 ˚    ˚ An : (12.13)

Evidently, 1-substitution is identical with the substitution property discussed above.

In the next proof, the ad hoc notation ..a1 ; : : : ; an // will be used for a column
vector.
Theorem 7.8 (Warfield [9]). A group enjoys the n-substitution property .n  1/ if
and only if n is in the stable range of its endomorphism ring.
Proof. Given A, suppose E D End A has n in the stable range, and let G D A1 ˚
  ˚An ˚B D A0 ˚C with Ai Š A for all i. We can write the projection  W G ! A0
as  D .1 ; : : : ; n ; ˇ/ with i W Ai ! A0 ; ˇ W B ! A0 , and the injection  W A0 ! G
as  D ..1 ; : : : ; n ; // with i W A0 ! Ai ; W A0 ! B (i.e. .a/ D 1 .a/ C    C
n .a/ C .a/ .a 2 A0 //. Thus 1 D  D 1 1 C    C n n C ˇ . Applying the
452 12 Torsion-Free Groups

stable range condition, we get

.1 C ˇ ı1 /1 C    C .n C ˇ ın /n D 1

for certain ıi ; i 2 E. Define the maps

 D .1 C ˇ ı1 ; : : : ; n C ˇ ın ; 0/ W G ! Im  D A0 ;
D ..1 ; : : : ; n ; .ı1 1 C    C ın n /// W A0 ! G;

so that  D 1. Therefore, G D A0 ˚ Ker . Since B

Ker  manifestly, we
can write Ker  D L ˚ B where L D .A1 ˚    ˚ An / \ Ker . Consequently,
G D A0 ˚ L ˚ B, and also G D A0 ˚ C, because  D 1 as well. P
Conversely, assume A has the n-substitution property, and nC1 iD1 i i D 1 2
E where i ; i 2 E. Let G D A1 ˚    ˚ AnC1 with Ai Š A, and consider the
monomorphism  D ..1 ; : : : ; nC1 // W A ! G. The map  D .1 ; : : : ; nC1 / W
G ! A is a splitting map for , since  D 1; thus, G D A ˚ C with C D Ker .
By hypothesis, there are A0 Š A and L < A1 ˚    ˚ An such that (12.13) holds
with B D AnC1 . The
PnC1 injection  W A0 ! G is of the form  D ..1 ; : : : ; nC1 // with
i 2 E satisfying iD1 i i D 1 2 E. Let  W G ! A0 with Ker  D L ˚ AnC1 ; then
 D 1;  D .1 ; : : : ; n ; 0/ yields  D 1 1 C    C n n C 0 D 1: Hence

.1 C nC1 nC1 1 /1 C    C .n C nC1 nC1 n /n D 1;

establishing the claim. t

u
2 in Stable Range We end this section with a brief look at a remarkable feature
of torsion-free groups of finite rank: a weaker version of the substitution property.
The result relies on the following theorem that has its own independent interest.
Theorem 7.9 (Warfield [9]). The endomorphism ring of any finite rank torsion-
free group has 2 in the stable range.
Proof. Let A be a torsion-free group of finite rank, and let 1 1 C 2 2 C 3 3 D 1
where i ; i 2 E D End A: First, assume that 1 is a monic map, thus 1 E is of finite
index in E, and so E N D E=1 E is a finite ring (see Proposition 1.10). As such, it is
artinian, so 1 is in the stable range, whence 2 2 C .1 1 C 3 3 /  1 mod 1 E
implies 2 C .1 1 C 3 3 /   mod 1 E for some ;  2 E where the coset of
N Hence
 is a unit in E.

.2 C 1 E/ C .3 3 C 1 E/ D  C 1 E:

If 2 2 E satisfies 2  1 2 1 E, then .2 C 3 3 /2 C 1 E D 1 C 1 E, which

means that 1 D 1 1 C .2 C 3 3 /2 D .1 C 3  0/1 C .2 C 3 3 /2 for
some 1 2 E:
If 1 is not monic, then we use the fact (that we will learn in Lemma 9.5) that QA
is an artinian Q-algebra, and therefore 1 is in the stable range of QE. Consequently,
7 Substitution Properties 453

10 D 1 C .2 2 C 3 3 /ˇ is a unit in QE for some ˇ 2 E, or, equivalently, 10

is monic in E. Therefore, we can argue with 10 in the same way as we did with 1
before, and the rewritten equation 10 1 C 2 .2  2 ˇ1 / C 3 3 .1  ˇ1 / D 1 will
lead us to

Œ1 C .2 2 C 3 3 /ˇ
1 C Œ2 C 3 3 .1  ˇ1 /
2 D 1

for some 1 ; 2 2 E. This is the same as

Œ1 C 3 3 .ˇ  .1  ˇ1 / 2 /
1 C Œ2 C 3 3 .1  ˇ1 /
.2 C 2 1 / D 1;

completing the proof. t

u
From Theorems 7.8 and 7.9 we can now conclude:
Corollary 7.10 (Warfield [9]). Torsion-free groups of finite rank enjoy the
2-substitution property. t
u
We will need the following simple observation.
Lemma 7.11 (Warfield [9]). Suppose A is a torsion-free group of finite rank. If
A ˚ A ˚ X Š A ˚ Y for certain groups X; Y, then there exists a subgroup L of A ˚ A
such that A ˚ L Š A ˚ A and Y Š L ˚ X:
Proof. Set G D A ˚ A ˚ X D A0 ˚ Y 0 with A0 Š A; Y 0 Š Y. Applying the 2-
substitution, we have A0 ; L such that G D A0 ˚ L ˚ X D A0 ˚ Y 0 with L
A ˚ A.
Hence Y Š Y 0 Š L ˚ X is clear. Also, A ˚ A Š A0 ˚ L Š A ˚ L follows as
A0 Š A0 Š A. t
u
F Notes. Prompted by a proof of Crawley [2], the substitution property was introduced in
the author’s paper [Monatshefte Math. 75, 198–204 (1971)], where it was shown that a ring enjoys
this property if and only if 1 is in the stable range (without using this terminology). Substitution
was developed by Warfield [9], who defined also the n-substitution property, and showed that it is
related to the stable range of the endomorphism ring.
Another noteworthy development is due to K.R. Goodearl [Pac. J. Math. 64, 387–411 (1976)].
He says that A has power-substitution if G D A1 ˚ B D A2 ˚ C with Ai Š A implies that
Gn D H ˚ Bn D H ˚ Cn for some n
1, and some group H. He goes on to prove that if
n is in the stable range of End A, then A has power-substitution. (The converse fails as is shown
by a counterexample C.X/ of continuous functions for a suitable compact Hausdorff space X.)
Torsion-free groups of finite rank have the power-substitution property.

Exercises

(1) Give a detailed proof for the claim that A D B ˚ C has the n-substitution
property if and only if both B and C have it.
(2) (Warfield) Suppose A has the substitution property, and G D A1 ˚ B D A2 ˚ C
where Ai Š A. Then there is a group D
G such that G D A1 ˚ D D A2 ˚ D.
454 12 Torsion-Free Groups

(3) (Warfield) Suppose A has the 2-substitution property. If A ˚ B Š A ˚ C, and if

B has a summand Š A, then B Š C.
(4) (Warfield) If A has the n-substitution property, then so does its endomorphism
ring E as a left E-module.

8 Finite Rank p-Local Groups

By a p-local group is meant a group that is also a Z.p/ -module; equivalently, a group
that is uniquely q-divisible for every prime q ¤ p. Thus there is only one prime to
deal with in p-local groups. The restriction to p-local groups has quite a simplifying
effect on the group structure; for instance, there are only two isomorphy classes of
rank 1 torsion-free p-local groups: Z.p/ and Q. Furthermore, like p-groups, p-local
groups may contain non-zero q-basic subgroups only for q D p, so we may call it
simply a basic subgroup.
Preliminaries on Local Groups We begin with important informations on
p-local groups.
Lemma 8.1. Let A be a p-local torsion-free group of finite rank. Then p End A
is contained in the Jacobson radical J of the endomorphism ring End A. Thus
.End A/=J is a finite semi-simple ring of characteristic p.
Proof. We show that for every  2 End A, 1  p is an automorphism of A. It
is clear that if p does not divide a 2 A, then it does not divide .1  p/a either.
Hence multiplication by 1  p preserves p-heights. Therefore, the pure subgroup
Ker.1  p/ must be 0, thus 1  p is monic, and Im.1  p/ is an essential subgroup
of A. From the preservation of heights it follows that Im.1p/ is pure in A, whence
Im.1  p/ D A. Consequently, 1  p is an automorphism, indeed. t
u
Theorem 8.2. A reduced finite rank torsion-free p-local group has semi-local
endomorphism ring. Therefore, it enjoys both the substitution and the cancellation
properties.
Proof. By Lemma 8.1, E=J is finite, so E is semi-local (cf. also Theorem 5.11 in
Chapter 16). Therefore, from Corollary 7.5 we deduce that A has the substitution,
and hence also the cancellation property. t
u
Power Cancellation A group A is said to have power cancellation if an
isomorphism An Š Cn for some n  2 implies A Š C:
In the next proof, we shall use notation and results from Sect. 12.
Proposition 8.3. Torsion-free p-local groups of finite rank have the power cancel-
lation property.
Proof. Suppose that the finite rank torsion-free groups A; C are p-local, and satisfy
An D Cn for some n  2. Evidently, A; C 2 add.G/. Denote by E the endomorphism
ring of G D A ˚ C. In the Arnold–Lady category equivalence (Theorem 12.2),
8 Finite Rank p-Local Groups 455

A 7! Hom.G; A/, where Hom is a projective right E-module. Thus Hom.G; A/n D
Hom.G; C/n . Passing modulo the Jacobson radical,

ŒHom.G; A/=J Hom.G; A/

n D ŒHom.G; C/=J Hom.G; C/
n :

Since E/J is finite, so artinian, the category of E/J-modules enjoys the Krull-
Schmidt property. Hence we conclude

Hom.G; A/=J Hom.G; A/ Š Hom.G; C/=J Hom.G; C/:

We can argue as in Lemma 6.5 to prove the isomorphy of the projective right
E-modules Hom.G; A/ and Hom.G; C/. Therefore, Hom.G; A/ ˝E G Š A is
isomorphic to Hom.G; C/ ˝E G Š C. t
u
Modules Over the p-Adic Integers The situation becomes extremely simple,
even in the countable rank case, if we move to modules over the ring Jp of p-adic
integers.
Theorem 8.4 (Derry [1], Kaplansky [2]). Let M be a torsion-free Jp -module of
countable rank. Then M is a direct sum of rank 1 modules, each isomorphic to Jp or
to its quotient field Qp .
Proof. Every rank 1 subgroup of M is contained in a pure rank 1 subgroup
isomorphic to Jp or Qp . These are algebraically compact, so summands. The
finite rank case is now obvious, while for countable rank the argument used in
Pontryagin’s theorem 7.1 in Chapter 3 establishes the claim. t
u
Let G be an arbitrary torsion-free group, and G.p/ D Z.p/ ˝ G its localization at
the prime p. It is harmless to identify G.p/ with the subgroup of the divisible hull
D of G whose elements x satisfy nx 2 G for some n 2 N coprime to p. We then
have G D \p G.p/ where the intersection is taken in the injective hull of G (see
Lemma 5.1 of Chapter 8).
If we embed the divisible hull D of G (which is a Q-vector space) in the Qp -
vector space Qp ˝ D D Qp ˝ G via d 7! 1 ˝ d, then we get an embedding of G
in Jp ˝ G which we will denote by G.p/ .
Lemma 8.5. For a torsion-free group G and a prime p,

G.p/ D D \ G.p/ ;

where the intersection is computed in Qp ˝ G.

Proof. To prove the non-trivial part of the claim that D \ G.p/
G.p/ , let x belong
to the intersection. If S D fgi gi2I is a maximal independent set in G.p/ , then we can
456 12 Torsion-Free Groups

write x D 1 g1 C    C m gm .i 2 Qp ; gi 2 S/. Since x 2 D and the coefficients

of the gi are uniqueley determined, we must have i 2 Q, whence x 2 G.p/ is
immediate. t
u

Splitting Fields We are now restricting our considerations to p-local torsion-

free groups A of finite rank. Besides rk A, there are two significant numerical
invariants for such an A: the p-corank n D rkp A D rk.A=pA/ D rk B and
m D rk A=B, where B is a basic subgroup of A. (In this section, the letters m; n will
keep this meaning.) Tensoring the exact sequence 0 ! B ! A ! A=B Š Qm ! 0
with Jp , we obtain splitting:

A.p/ Š .Jp ˝ B/ ˚ .Qp /m ;

thus A.p/ becomes completely decomposable over Jp . The Kurosh theory shows that
there is no need to go as far as Jp (which has the transcendence degree of the
continuum over Z.p/ ) in order to obtain such an attractive decomposition; indeed,
we will see in a moment that a suitable finite extension will suffice.
We start with a particular maximal independent system in our group A. Extend a
basis fb1 ; : : : ; bn g of B by elements c1 ; : : : ; cm 2 A to have a maximal independent
set in A (thus the cosets of c1 ; : : : ; cm form a basis of the Q-vector space A=B). We
choose representatives xij 2 A .i D 1; : : : ; mI j < !/ of the generators of A=B such
that

xi0 D ci ; pxi;jC1  xij mod B .i D 1; : : : ; mI j < !/:

Thus pxi;jC1 D xij Cri1j b1 C  Crinj bn with uniquely determined integer coefficients
rikj . If we express pjC1 xi;jC1 in terms of xi0 , we then get

pjC1 xi;jC1 D i1j b1 C    C inj bn C ci .j < !/ (12.14)

where ikj denotes the jth partial sum of a p-adic integer ik .i D 1; : : : ; mI k D

1; : : : ; n/. Let P D kik k be the m  n matrix with entries defined by Eq. (12.14),
and K\ the subfield of Qp obtained by adjoining all the ik to Q (a priori K\ also
depends on the basis, but—as it will be clear from Theorem 8.6—it does not). Set
R\ D Jp \ K\ ; this is a valuation domain. We now have

R\ ˝ A D R\ b1 ˚    ˚ R\ bn ˚ K\ d1 ˚    ˚ K\ dm ; (12.15)

where di D i1 b1 C    C in bn C ci .i D 1; : : : ; m/.
We will say that a subfield K of Qp is a splitting field for A if R ˝ A decomposes
like (12.15), i.e. R ˝ A is a direct sum of cyclic and rank 1 divisible R-modules,
where R = K \ Jp .
8 Finite Rank p-Local Groups 457

We are looking for a minimal splitting field. We claim:

Theorem 8.6 (Szekeres [1]). Given a p-local torsion-free group A of finite rank,
there exists a uniquely determined subfield of the field of the p-adic numbers, which
is the minimal splitting field for A, contained in any other splitting field for A.
Proof. The argument above establishes the existence of splitting fields for A (which
was not doubtful anyway). We now prove that the field K\ defined above is
independent of the choice of the basis, and is in fact the minimal splitting field
for A.
Let us choose another basic subgroup with a basis b01 ; : : : ; b0n , and extend it
to a maximal independent set by adjoining elements c01 ; : : : ; c0m . The new set
depends linearly over Q on the bk ; ci , and therefore the coefficients of the b0k in
the expressions (12.13) for the c0i , being uniquely determined, must belong to K\ .
This means that the splitting field K determined by the new choice is contained in
K\ . For reasons of symmetry, K D K\ follows. Consequently, K\ is well defined by
t
A, it must be its unique minimal splitting field contained in any splitting field of A.u
It is now clear that the splitting field K\ is determined by the m  n matrix
P D kik k, where the entries are from Jp . Actually, any choice of the matrix P
defines a p-local torsion-free group of rank n C m and p-corank n. To prove this,
and also to show how to recapture A from the data provided by the ik , consider the
following pull-back diagram

⊕nk=1 Z(p) bk
⏐
⏐

0 −−−−→ ⊕m
i=1 Z(p) ci −−−−→ A −−−−→ ⊕nk=1 Qbk −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐ 
  P ·can
can
0 −−−−→ ⊕m m m ∞
i=1 Z(p) ci −−−−→ ⊕i=1 Qci −−−−→ ⊕i=1 Z(p ) −−−−→ 0

where we used the canonical map onto the factor group ˚nkD1 Z.p1 /, followed by
the map between the two divisible factor groups given by the transpose P0 of P, to
define the vertical map on the right. Indeed, it is easily checked that the pull-back
construction yields A as defined by the equations in (12.14).
Example 8.7. Consider the indecomposable group of rank r in Lemma 4.6. Its minimal splitting
field is Q.2 ; : : : ; r /.
F Notes. The category of p-local torsion-free groups of finite rank is not a Krull–Schmidt
category. Arnold [6] points out that in order to have an example with two non-isomorphic direct
decompositions, groups of ranks at least 10 are required.
Most of the theorems on p-local torsion-free groups carry over to torsion-free modules over
discrete valuation domains, where the role of Jp is taken over by the completion of the domain, see
Lady [6]. In several papers, Lady has developed a theory of splitting fields over Dedekind domains.
458 12 Torsion-Free Groups

For those who want more results and more details, we refer to his papers [5]. Glaz–Vinsonhaler–
Wickless [1] introduce splitting rings for p-local torsion-free groups, depending on the choice of a
basic subgroup. For more on the local case, see Hill–Lane–Megibben [1].
It is of course a major problem to move from the local to the global case. The Derry–Kurosh–
Malcev theory provides a reasonable method. For a discussion of the problem from the sheaf-
theoretic point of view, see Rotman [4] and Turgi [1].

Exercises

(1) Show that quasi-isomorphic torsion-free groups of finite rank (next section)
share splitting fields.
(2) Relate the splitting field of a direct sum to those of the summands.
(3) (Lady) If a field contains the splitting fields of A and C, then it also contains the
splitting fields of both A ˝ C and Hom.A; C/.
(4) How is the p-adic completion AQ related to R\ ˝ A?
(5) (Szekeres) Show that Theorem 8.6 also holds for groups of countable rank.

9 Quasi-Isomorphism

In view of the pathological examples of Sect. 5, one might abandon in despair the
hope of finding any kind of uniqueness in the direct decompositions of finite rank
torsion-free groups. Fortunately, there is good news. In order to remedy the situation,
B. Jónsson came up with an ingenious idea: a weaker version of isomorphism that
led to a kind of uniqueness of the direct decompositions. This is a drastic departure
from the traditional approach, but the consequences are significant and far reaching,
even though they follow rather quickly.
Quasi-Morphisms By a quasi-homomorphism  W A!C P between two finite
rank torsion-free groups we mean a homomorphism of A into the divisible hull of C
such that n Im 
C for some n 2 N. Thus the group of quasi-homomorphisms is
Q Hom.A; C/ Š Q ˝ Hom.A; C/:

The groups A; C are quasi-isomorphic, written

A C;
if there are quasi-homomorphisms  W A!C P and W C!A P such that  D q1A
and  D q0 1C for some non-zero q; q0 2 Q. Then an integral multiple of  is
an isomorphism of A with a subgroup of C, and an integral multiple of is an
isomorphism of C with a subgroup of A. It is immediate that quasi-isomorphism is
an equivalence relation in the class of finite rank torsion-free groups, weaker than
isomorphism.
9 Quasi-Isomorphism 459

Actually,  D q1A implies . / D q whence cancellation by the monic 

yields  D q1C . Therefore, it suffices to assume only one half of (b) and (c) in
Lemma 9.1. For the torsion-free groups A; C of finite rank the following are
equivalent:
(a) A and C are quasi-isomorphic;
(b) there are integers m; n > 0 such that
mA
C0
A and nC
A0
C; (12.16)

where C0 Š C and A0 Š A.
(c) A is isomorphic to a finite index subgroup of C, and C is isomorphic to a finite
index subgroup of A.
Proof. (a) ) (b) Assume A C, i.e. there are maps ; as stated in the definition.
Then there are r; s 2 N such that r.A/
C and s .C/
A. Thus rsqA D
rs .A/
s .C/
A proves the first relation in (12.16). The second follows
similarly.
(b) ) (c) This implication is trivial, because if mA is of finite index in A, and nC
in C.
(c) ) (a) If (c) holds, then an isomorphism  W A ! A0
C may be regarded
as a quasi-homomorphism A!C, P and an isomorphism W C ! C0
A as
P
C!A. The composite map  W A ! A carries A isomorphically into itself, so by
Proposition 1.10 it is multiplication by some q 2 N times an automorphism ˛ of A.
Changing  to ˛ 1 , we get .˛ 1 / D q1A , and hence also .˛ 1 / D q1C . u t
Every subgroup of finite index in a rank 1 group is isomorphic to the group
itself, therefore, for torsion-free groups of rank 1, quasi-isomorphism implies
isomorphism. More generally, we have
Lemma 9.2 (Procházka [2]). Assume A is a finite rank torsion-free group such that
jA=pAj
p for every prime p. If a torsion-free group C is quasi-isomorphic to A,
then A Š C.
Proof. For the proof, there is no loss of generality in assuming that C is a subgroup
of A of prime index p. Then pA
C < A whence jA=pAj
p implies that C D
pA Š A. t
u
It is routine to check for all finite rank torsion-free groups A; B; C:
(A) If A B, then Type.A/=Type.B/.
(B) If D; D0 are divisible groups, then D D0 implies D Š D0 .
(C) If B C, then A ˚ B A ˚ C for all A.
(D) If A B, then for all C, we have both Hom.A; C/ Hom.B; C/ and
Hom.C; A/ Hom.C; B/. Similarly for Ext and the tensor product.
Quasi-Direct Decompositions If A is quasi-isomorphic to the direct sum C1 ˚
   ˚ Cn , then we say that
460 12 Torsion-Free Groups

A C1 ˚    ˚ Cn (12.17)

is a quasi-direct decomposition, and the Ci are quasi-summands of A. A group

that has only the trivial quasi-direct decomposition is called strongly indecompos-
able. Equation (12.17) and A C10 ˚    ˚ Cm0 are viewed as quasi-isomorphic
direct decompositions if n D m, and there is a bijection between the sets of their
components that preserves quasi-isomorphism.
(E) Being a quasi-summand is a transitive property.
(F) If B is a quasi-summand of A, and X is quasi-isomorphic to a subgroup of A
containing B, then X has a quasi-summand Š B.
Example 9.3. The group A in Example 4.3 is quasi-equal to the direct sum ˚n En . Thus A 
˚n En is a quasi-direct decomposition of A into strongly indecomposable summands, as stated in
Theorem 9.9 below. On the other hand, the group in Example 4.4 is strongly indecomposable.
Example 9.4. Rigid groups are strongly indecomposable. For example, the groups in Lemma 4.6
are strongly indecomposable.

Let A be torsion-free group of finite rank, and D its divisible hull. A quasi-
endomorphism of A is a homomorphism  W A ! D such that n.A/
A for
some integer n > 0. The quasi-endomorphisms of A form a subring Q End A in the
ring of all endomorphisms of D.
This definition immediately yields a number of useful results.
Lemma 9.5. For a torsion-free group of finite rank A, the following hold:
(i) Q End A is a left artinian Q-algebra;
(ii) every idempotent  2 Q End A yields a quasi-direct decomposition

A A ˚ .1  /A:

If A B ˚ C, then B A; C .1  /A for an idempotent  2 Q EndA.

Proof. (i) Q End A is clearly a Q-vector space, since  is a quasi-endomorphism
exactly if n is an endomorphism for some n > 0. The rest follows from the
finite dimensionality of Q End A.
(ii) The claim is an immediate consequence of the definitions. t
u
It is a good idea to keep in mind that A is strongly indecomposable if nA

B ˚ C
A (for some integer n > 0, and subgroups B; C) implies that either B D 0
or else C D 0.
The finite rank torsion-free groups A share the property that their quasi-direct
decompositions are in a bijective correspondence with the direct decompositions of
their quasi-endomorphism rings Q End A into left ideals (see Lemma 1.5). In our
case, this correspondence is given by

A A ˚ .1  /A $ Q End A D .Q End A/ ˚ .Q End A/.1  /;

for an idempotent  2 Q End A.

9 Quasi-Isomorphism 461

It is well known that an artinian ring with identity decomposes into a direct sum
of a finite number of ideals ¤ 0 which are local rings. Hence we obtain the following
important information:
Proposition 9.6 (Reid [2]). The quasi-endomorphism ring of a finite rank torsion-
free group is local if and only if the group is strongly indecomposable. t
u
An analogue of the well-known Fitting’s Lemma can be established in the finite
rank situation.
Lemma 9.7. Let  be an endomorphism of the finite rank torsion-free group A.
Then there is an integer n  1 and a quasi-direct decomposition

A Im n ˚ Ker n : (12.18)

Proof. Choose n 2 N such that Im n has the smallest rank. Then n is monic on
Im n , whence Proposition 1.10 implies that there exists m 2 N such that m Im n

Im 2n . Therefore, for any given 0 ¤ a 2 A, there exists b 2 A such that mn .a/ D
2n .b/. Hence ma D n .b/ C .ma  n .b// 2 Im n C Ker n . As n is monic on
Im n , we also have Im n \ Kern D 0, and the claim follows. t
u
When Quasi-Isomorphism Implies Isomorphic Endomorphism Rings In
general, quasi-isomorphic groups need not have isomorphic endomorphism rings,
though they share simultaneously some relevant properties, like commutativity. An
interesting special case when their endomorphism rings have to be isomorphic is
recorded in the following result.
Lemma 9.8 (Arnold–Hunter–Richman [1]). Assume that A; B are quasi-
isomorphic finite rank torsion-free groups. If End B is a principal ideal domain,
then the endomorphism rings End A and End B are isomorphic.
Proof. It is clear that End A is an integral domain. Let ˛ W A!B P denote a quasi-
isomorphism. The set H˛ D ˛ Hom.B; A/ is evidently a non-zero ideal in End B.
A non-zero  2 End A induces an End B-endomorphism  W H˛ ! H˛ such that
˛ˇ 7! ˛ˇ .ˇ 2 Hom.B; A//. Indeed, if ˛ˇ D 0, then .ˇ˛/2 D 0, so also ˇ˛ D 0,
thus .˛ˇ/2 D 0 and ˛ˇ D 0 as End B is a domain. In addition, there exists a
u˛ in the quotient field of End B such that  .x/ D u˛ x for all x 2 H˛ . Repeated
application of  yields .u˛ /k x 2 H˛ for all k 2 N, which can happen only if
u˛ 2 End B (a PID).
We claim that the correspondence  7! u˛ is a ring isomorphism ˛ W End A !
End B. It is evidently additive, and maps 1A upon 1B . From

˛ ./˛ ./.˛ˇ/ D ˛ ./.˛ˇ/ D ˛ˇ D ˛ ./.˛ˇ/ .;  2 End A/

P
we conclude that it preserves multiplication. Finally, if W B!C is a quasi-
isomorphism, then
462 12 Torsion-Free Groups

 ˛ ./. ˛ı/ D . ˛/ı D ˛ ./.˛ı/ D  .˛ .//. ˛ı/ .ı 2 Hom.C; A//;

˛
thus  ˛ D  ˛ . In the special case A D C,  is multiplication by , so u D 
and  ˛ is the identity. Consequently,  ˛ is the identity on End A, and so ˛ has to
be an isomorphism (interestingly, it is independent of the choice of ˛). t
u
Krull–Schmidt for Quasi-Direct Decompositions The main result on quasi-
direct decompositions is our next theorem.
Theorem 9.9 (Jónsson [2]). Let A be a torsion-free group of finite rank, and
A A1 ˚    ˚ Am ; A C1 ˚    ˚ Cn
quasi-direct decompositions of A, where each of Ai and Cj is strongly indecompos-
able ¤ 0. Then m D n, and after suitably rearranging the components, Ai Ci for
i D 1; : : : ; m.
Proof. Groups with local endomorphism rings share the exchange property. It is not
difficult to check that this also holds for local quasi-endomorphism rings in quasi-
direct decompositions. Hence it is immediate that A1 can replace one of the Ci , say,
C1 , i.e. we have A A1 ˚ C2 ˚    ˚ Cn . Passing mod A1 , we get A2 ˚    ˚ Am
C2 ˚    ˚ Cn . Induction completes the proof. t
u
Once quasi-isomorphism proved to be a useful concept, the category of finite
rank torsion-free groups under quasi-homomorphisms becomes an object of interest.
The motivation, even the need for studying this category, was pointed out by E.
Walker who interpreted the preceding theorem by considering it in this category.
Accordingly, we now define the category A0 where the objects are the torsion-free
groups of finite rank, and the morphisms are the quasi-homomorphisms, i.e.
HomA0 .A; B/ D Q Hom.A; B/
for torsion-free groups A; B of finite rank. This category can also be viewed as the
quotient category of the category of torsion-free groups of finite rank modulo the
category of bounded groups. The theorems above tell us that A0 is a Krull–Schmidt
category where the indecomposable objects have local endomorphism rings.
Procházka–Murley Groups We consider briefly a class of groups where quasi-
isomorphism equals isomorphism. A torsion-free group M of finite rank is called a
Procházka–Murley group if its p-corank is
1 for each prime p, i.e. rk.M=pM/

1 for every p.
Example 9.10. (a) Rank 1 torsion-free groups are Procházka–Murley
Q groups.
Q D p Jp .
(b) The same holds for finite rank pure subgroups of Z

We can state immediately:

.˛/ The class of Procházka–Murley groups is closed under taking pure subgroups,
torsion-free homomorphic images, and tensor products.
.ˇ/ A torsion-free group of finite rank is a Procházka–Murley group exactly if all
of its finite homomorphic images are cyclic.
9 Quasi-Isomorphism 463

. / If N is a pure subgroup of a Procházka–Murley group M, then for any prime

p, either N or M=N is p-divisible. Indeed, this follows from the exact sequence
0 ! N=pN ! M=pM ! .M=N/=p.M=N/ ! 0, where only the first two or
the last two factor groups can be non-zero.
.ı/ The Z-adic completion M Q of a Procházka–Murley group M is isomorphic to a
Q
summand of Z. More precisely, isomorphic to the direct product of a copy of Jp
for each prime p satisfying pM < M.
Lemma 9.11 (Murley [1]). A torsion-free group M of finite rank is a Procházka–
Murley group if and only if it shares the following properties:
(i) if N is quasi-isomorphic to M, then N Š M; and
(ii) if N is a subgroup of M isomorphic to M, then N D nM for some n 2 N.
Proof. Suppose M is a Procházka–Murley group. If N M, then there are an
integer k > 0 and a subgroup N 0 Š N satisfying kM
N 0
M. N 0 =kM is a
subgroup of the cyclic group M=kM, so cyclic: N 0 =kM D nM=kM for a divisor n of
k. Then N 0 D nM Š M, proving both (i) and (ii). Conversely, if M has properties
(i) and (ii), then let pM
N
M for a prime p. By (i), N Š M, while (ii) implies
that N D nM Š M for an n 2 N. Clearly, only n D p is a possibility whenever
pM ¤ M. t
u
We can now derive the following theorem as a corollary to the preceding lemma.
Corollary 9.12 (Murley [1]). Finite rank Procházka–Murley groups have the
Krull-Schmidt property: if

A D M1 ˚    ˚ Mm D N1 ˚    ˚ Nn

where the Mi and Nj are indecomposable Procházka–Murley groups, then m D n,

and there is a rearrangement such that Mi Š Ni for i D 1; : : : ; m.
Proof. From Lemma 9.11(i) we infer that indecomposable Procházka–Murley
groups are strongly indecomposable, thus the components in the given direct decom-
positions are strongly indecomposable. A simple appeal to Jónsson’s theorem 9.9
completes the proof. t
u
We conclude with pointing out a remarkable feature of indecomposable
Procházka–Murley groups.
Theorem 9.13 (Murley [1]). Any endomorphism of an indecomposable Procházka–
Murley group is an integer times an automorphism. The endomorphism ring is a
principal ideal domain.
Proof. Let M be an indecomposable Procházka–Murley group. Every endomor-
phism  ¤ 0 extends uniquely to an endomorphism of its completion M Q which
is, by .ı/, a product of p-adic integers Jp . By indecomposability, each prime occurs
at most once, thus End M Q is isomorphic to the direct product of the rings of p-adic
integers for the same primes. This is a commutative ring with no nilpotent elements,
464 12 Torsion-Free Groups

so the same holds for End M. Lemmas 9.7 and 9.11(i) imply M Š Im n ˚ Ker n
for some n 2 N. In view of indecomposability, one of the summands vanishes, so 
is either monic, or nilpotent. But nilpotency is not an option, thus  is monic, and
therefore, Im  D mM for some m 2 N, because of Lemma 9.11(ii). Hence m1  is
an automorphism of M. t
u
F Notes. The concept of quasi-isomorphism is due to Jónsson [1]. It was Reid [3] who
formulated the basic properties of quasi-direct decompositions for finite rank torsion-free groups.
Quasi-isomorphism has been studied extensively, it is frequently the natural concept when
isomorphism cannot be expected. Actually, the study of the p-corank  1 case was initiated
by Procházka [2], a few years before Murley’s paper [1] was published with more substantial
results. We have chosen to break tradition in the terminology (Murley group) to give due credit to
Procházka.
Faticoni [2] proves inter alia the following for a torsion-free group G that is quasi-isomorphic
to the direct sum of a finite number of pairwise non-quasi-isomorphic, strongly indecomposable
finite rank torsion-free groups: Every decomposition of G into the direct sum of indecomposable
groups is unique up to isomorphism.
Kozhukhov [1] investigates finite rank torsion-free groups G which have no nilpotent endo-
morphisms ¤ 0. Such a G has a unique finite index subgroup A with direct decomposition
A D A1 ˚    ˚ An , where the Ai are pure in G, are strongly indecomposable, and satisfy
Hom.Ai ; Aj / D 0 .i ¤ j/. Krylov [4] points out that if for the finite rank torsion-free groups
A; B there is a finite set fC1 ; : : : ; Cn g of finite rank torsion-free groups such that rkp Hom.A; Ci / D
rkp Hom.B; Ci / for all primes p and for all i, then A; B are quasi-isomorphic.
There are various versions of notation for quasi-isomorphism. I propose to use  for
quasi-isomorphism and  for near-isomorphism; these are close to the standard symbol Š for
isomorphism, and it is easy to remember: Š )  ) .

Exercises

(1) For every n 2 N, there exist strongly indecomposable torsion-free groups of

rank n.
(2) Let A be a torsion-free group of finite rank. There exist at most countably
many non-isomorphic groups that are quasi-isomorphic to A. [Hint: count the
subgroups between nA and A.]
(3) If A C, then IT.A/ D IT.C/ and OT.A/ D OT.C/.
(4) (Procházka) Suppose A B are finite rank torsion-free groups. Then
Ext.A; C/ Š Ext.B; C/ for every group C.
(5) If B; C are pure subgroups in a finite rank torsion-free group A such that
B \ C D 0 and A=B C, then A B ˚ C.
(6) A finite rank torsion-free group A is Procházka–Murley if and only if every
subgroup of finite index is isomorphic to A.
(7) (Procházka) Every rank 2 quotient-divisible group (see Sect. 11) is a
Procházka–Murley group.
(8) (Murley) The Kaplansky test problems for Procházka–Murley groups yield
positive results.
10 Near-Isomorphism 465

(9) (Murley) (a) Let M; N be reduced Procházka–Murley groups such that, for
every prime p, only one of pM < M and pN < N holds. Then Hom.M; N/ D
0 D Hom.N; M/.
(b) In any direct decomposition M D M1 ˚    ˚ Mk of a reduced Procházka–
Murley group M, the summands are fully invariant.
(10) Let A be as in Corollary 9.12. If a pure subgroup of A is a finite direct sum of
groups quasi-isomorphic to the Ai , then it is a summand of A. [Hint: induction.]

10 Near-Isomorphism

From Jónsson’s theorem 9.9 we can derive that if A˚B Š A˚C holds for finite rank
torsion-free groups A; B; C, then B and C are quasi-isomorphic. In fact, passing to
quasi-isomorphism and decomposing into direct sums of strongly indecomposable
groups, Theorem 9.9 applies. It turns out that a stronger statement can be made: B
and C are near-isomorphic in the sense defined below.
The Category Ap Following Arnold [A], we introduce the category Ap for each
prime p. The objects in Ap are the finite rank torsion-free groups, while

HomAp .A; C/ D Z.p/ ˝ Hom.A; C/ D Hom.A; C/.p/ ;

i.e. the localization of Hom.A; C/ at p.

Theorem 10.1 (Lady [4], Arnold [A]). The following conditions are equivalent
for the reduced torsion-free groups A and C of finite rank:
(i) for every prime p, C contains a subgroup isomorphic to A, whose index is
coprime to p, and vice versa;
(ii) for every prime p, there exist an integer k coprime to p as well as maps
˛p W A ! C and p W C ! A such that

p ˛p D k1A and ˛p p D k1C I

(iii) A and C are quasi-isomorphic as well as Ap -isomorphic for each prime p;

(iv) for every square-free integer m, there are maps m W A ! C and m W C ! A
such that

m m  1A mod m End A and m m  1C mod m End C: (12.19)

Proof. (i) ) (ii) Condition (i) means that for each p there is a map ˛p carrying
A isomorphically onto a subgroup of C, say, of index k, coprime to p. Thus kC

˛p A
C. If we denote by p W C ! A the composite map of multiplication by k
466 12 Torsion-Free Groups

followed by ˛p1 , then we have ˛p p D k1C . Hence ˛p p ˛p D k˛p , and since ˛p is

necessarily monic, we can cancel it on the left to obtain p ˛p D k1A .
(ii) ) (iii) The maps ˛p ; p in (ii) are evidently monic, so A C. They define
uniquely Ap -maps between A and C, which satisfy the same equations. As k is a
unit mod p, (ii) implies that A; C are Ap -isomorphic.
(iii) ) (iv) Condition (iii) implies that there exist p 2 HomAp .A; C/ and p 2
HomAp .C; A/ that are inverse to each other mod p. Hence (iv) is obvious if m is
a prime. To complete the proof, we show that if there are maps m ; m and n ; n
satisfying (12.19) with gcdfm; ng D 1, then there exist also mn ; mn as required.
Let mx C ny D 1 for x; y 2 Z, and set  D mxn C nym ;  D mxn C nym .
Then using mx D m2 x2 C mnxy and ny D mnxy C n2 y2 , as well as m m
 1C mod m End C and n n  1C mod n End C, we obtain

  m2 x2 n n C n2 y2 m m  mxn n C nym m
 mx1C C ny1C  1C mod mn End C:

Similarly,   1A mod mn End A, thus mn D ; mn D  are as desired.

(iv) ) (i) First we show that there is a square-free m 2 N such that mX ¤ X for
every non-zero subgroup X of B D A ˚ C. If there is a prime p such that B contains
no p-divisible subgroup, then m D p is a good choice. Otherwise, pick a prime p for
which B is not p-divisible. The p-divisible subgroup D of B has a smaller rank than
B, so by induction we can assume the existence of an n 2 N such that nX ¤ X for
all subgroups X
D. Then m D pn is as desired.
Using the integer m selected in the preceding paragraph, we next show that for the
corresponding maps m ; m , the composite m m is monic. (iv) claims that m m D
1A C m for some  2 End A. If a 2 Ker m m , then a D m.a/, thus Ker m m
is contained in the m-divisible subgroup of A, so it is 0. Hence Im m m Š A,
so it is of finite index in A (Proposition 1.10). Furthermore, m m D 1A C m
implies that Im m m is m-pure in A, thus there is a k 2 N, coprime to m, such that
kA
m m A
m C
A: Analogously, m m is monic and k0 C
m A < C for
some k0 coprime to m. Thus A; C satisfy (i). t
u
Making use of the notation A0 introduced in the preceding section, (iii) can be
rephrased by saying that A and C are both A0 - and Ap -isomorphic for all p.
From the proof it is clear that in (i), one-sided condition would suffice. It also
follows:
Lemma 10.2. The preceding theorem holds even if in conditions (i), (ii), the prime
p is replaced by an arbitrary square-free integer m. t
u
Near-Isomorphism Call the finite rank torsion-free groups A and C near-
isomorphic or of the same genus, in notation:

A  C;
10 Near-Isomorphism 467

if they satisfy the equivalent conditions of Theorem 10.1. Thus

isomorphism ) near-isomorphism ) quasi-isomorphism.

We will see that none of the arrows is reversible.

(A) Near-isomorphism is an equivalence relation. Symmetry is clear from
Theorem 10.1, while transitivity is obvious.
(B) A  B if and only if their divisible parts are isomorphic and their reduced
parts are near-isomorphic.
(C) A  B implies that A.p/ Š C.p/ for each prime p.
(D) For torsion-free groups of finite rank, we have A1 ˚    ˚ An  B1 ˚    ˚ Bn
whenever Ai  Bi for i D 1; : : : ; n.
(E) If a finite rank group A is p-divisible for almost all primes p, then A  C implies
A Š C. Let n denote the product of primes for which A is not divisible. In view
of Lemma 10.2, there are k 2 N coprime to n and an injection  W C ! A such
that kA
C
A. As kA D A,  is an isomorphism.
(F) The objects in the category Ap have the cancellation property. This is a
consequence of Theorem 8.2.
Example 10.3. (a) The rank 2 summands Ai ; Ci in the proof of Theorem 5.3 are near-isomorphic.
If the coefficient k is chosen there to be coprime to p, then the arising groups are
non-isomorphic. (The reader might enjoy in examining all the summands in Sect. 5 for near-
isomorphism.)
(b) Let p1 ; p2 ; q be different primes. With independent elements e1 ; e2 , consider the groups

A1 D hp1
1 e1 ; p1
2 e2 ; q1 .e1 C e2 /i;

A2 D hp1
1 e1 ; p1
2 e2 ; q1 .e1 C ke2 /i;

B D hp1
1 e1 ; p1
2 e2 ; q2 .e1 C e2 /i;

where k 6 1 mod q and gcdfk; qg D 1: All three groups are quasi-isomorphic, but pairwise
non-isomorphic. In addition, A1  A2 (easy to check). However, A1 is not near-isomorphic to
B, as the index of any embedding of A1 in B must be divisible by the prime q.

Near-Isomorphism and Cancellation The following theorem is an important

result on near-isomorphism, and a good reason for its study.
Theorem 10.4 (Lady [4]). Let G D A ˚ B D A0 ˚ C be two direct decompositions
of a torsion-free group of finite rank. If A  A0 , then B  C, i.e. also B and C are
near-isomorphic.
Proof. Consider the given decompositions in the categories A0 and Ap . Cancella-
tion holds in all these categories where A and A0 are isomorphic (Lemma 9.1). Hence
B and C are also isomorphic in these categories, i.e. they are near-isomorphic. ut
Our next aim is to relate near-isomorphism to power-cancellation. We require a
lemma that points out another relevant property of near-isomorphism.
Lemma 10.5 (Lady [4]). Suppose that Ai ; Bi ; C .i D 1; : : : ; k/ are near-
isomorphic finite rank torsion-free groups. Then
468 12 Torsion-Free Groups

(i) there is a group D near-isomorphic to C such that

G D A1 ˚    ˚ Ak ˚ C Š B1 ˚    ˚ Bk ˚ DI

(ii) there are but a finite number of non-isomorphic groups near-isomorphic to Ai .

Proof. (i) To settle the k D 1 case, we prove that, for near-isomorphic A; B; C, the
group G D A ˚ C has a summand Š B. We can find subgroups A0 ; C0 of B
such that A0 Š A; C0 Š C, and the indices ŒB W A0
D m and ŒB W C0
D n are
coprime. There are integers r; s satisfying rm C sn D 1. Then the monic map
b 7! .rmb; snb/ 2 A0 ˚ C0 Š G followed by W A0 ˚ C0 ! A0 C C0 D B is
the identity on B. Hence G Š B ˚ D for some D. From Theorem 10.4 it follows
that D is near-isomorphic to C.
Suppose A1 ˚    ˚ Ak ˚ C Š B1 ˚    ˚ Bk ˚ Dk for some Dk  C. By the
case k D 1, there is a DkC1  C such that AkC1 ˚ Dk Š BkC1 ˚ DkC1 . Then
A1 ˚    ˚ AkC1 ˚ C Š B1 ˚    ˚ BkC1 ˚ DkC1 ; and (i) follows.
(ii) An obvious consequence of (i) is that every group B near-isomorphic to A is
isomorphic to a summand of A ˚ A. This direct sum is a finite rank group, so
Lady’s theorem 6.9 applies: it has but finitely many non-isomorpic summands.
t
u
It is worthwhile observing that the number of non-isomorphic, near-isomorphic
finite rank torsion-free groups can be an arbitrarily large integer. This is evident,
e.g., from Theorem 5.5.
Theorem 10.6 (Arnold [5]). Finite rank torsion-free groups A; C are near-
isomorphic if and only if Ak Š Ck holds for some integer k  1.
Proof. First, suppose that Ak Š Ck for some k 2 N. We argue as in Theorem 10.4
to conclude that A  C.
Conversely, assume A  C. By Lemma 10.5(i), for every n > 1, there is Bn  C
such that An Š Cn1 ˚ Bn . From Lemma 10.5(ii) we infer that there exist integers
n < m with Bn Š Bm . Hence

Am Š Cm1 ˚ Bm Š Cmn ˚ Cn1 ˚ Bn Š Cmn ˚ An :

We appeal to the 2-substitution property, in particular, to Lemma 7.11 to find a group

L .
A ˚ A/ satisfying L ˚ A Š A ˚ A such that

L ˚ Am2 Š Cmn ˚ An1 :

We can rewrite the last isomorphism as Am1 Š Cmn ˚An1 . If we repeat this argu-
ment n  1 more times, then we arrive at the desired conclusion: Amn Š Cmn . ut
The preceding theorem, taken in combination with Theorem 10.4, gives the
following interesting result on the cancellations of finite rank torsion-free groups:
A ˚ B D A0 ˚ C with A Š A0 implies that Bk Š Ck holds for some k 2 N.
10 Near-Isomorphism 469

F Notes. The idea of near-isomorphism originates in Representation Theory. In the theory

of lattices over orders over Dedekind domains, it is equivalent to being of the same genus. It was
Lady who introduced this notion into the theory of finite rank torsion-free groups.
In contrast to quasi-isomorphism, indecomposability is preserved under near-isomorphism; see
Arnold [3]. For near-isomorphism of Butler-groups, see Arnold–Dugas [2]. O’Meara–Vinsonhaler
[1] prove a very surprising result: there exist non-isomorphic torsion-free groups A; C of finite rank
such that An Š Cn for all n
2.
The cancellation property was introduced by Jónsson–Tarski [Notre Dame Math. Lectures 5
(1947)]. It became an ubiquitous topic in the theory of modules. For abelian groups, it originated
from a question by Kaplansky, answered by E. Walker [1], Honda [2], and Cohn [1], that Z
was cancellative (see Sect. 2, Exercise 11 in Chapter 3), and continued with Fuchs–Loonstra [1]
where the rational groups with the cancellation property were characterized. See Stelzer [1] for
a systematic treatment in the finite rank case. Blazhenov [1] succeeded in getting a complete
solution: a finite rank torsion-free group A has the cancellation property if and only if A D F ˚ B,
where F is free and B satisfies the following conditions: (i) for each n 2 N, units of End B=n End B
lift to units of End B; and (ii) the quasi-summands of B satisfy the so-called weak Eichler condition.
Arnold [A] has definite results concerning self-cancellation: A ˚ A Š A ˚ B implies A Š B, while
O’Meara–Vinsonhaler [1] discuss separative cancellation: A ˚ A Š A ˚ B Š B ˚ B implies
A Š B.

Exercises

(1) Let B; C be subgroups of the finite rank torsion-free group A such that
B \ C D 0. If B is pure in A and A=B  C, then A  B ˚ C.
(2) Decide whether or not B  C implies Hom.A; B/  Hom.A; C/.
(3) If A  C for finite rank torsion-free groups, then the trace of A in C is all of C.
(4) Let A; C be near-isomorphic finite rank torsion-free groups, and A1 ; A2 sub-
groups of A isomorphic to C with coprime indices. Prove that A1 \ A2  C.
(5) Let A; B; C be finite rank torsion-free groups such that pA D A for almost all
primes. Then A ˚ B Š A ˚ C implies B Š C.
(6) (Fuchs–Loonstra) A rational group R has the cancellation property if and only
if, for every n 2 N, each automorphism of R=nR lifts to an automorphism of R.
(7) Assume B; C are subgroups of A such that there are ˇ; 2 End A for which
ˇ  B and  C are inverse isomorphisms between B and C. If A has the
cancellation property, then A=B Š A=C. [Hint: consider the push-out of ˇ W
B ! A and the injection B ! A.]
(8) A group with the exchange property has the cancellation property if and only if
isomorphic summands have isomorphic complements.
(9) (Arnold) A group A is said to have self-cancellation if A ˚ A0 Š A ˚ A implies
A0 Š A.
(a) Summands inherit self-cancellation.
(b) Let A be torsion-free and of finite rank. If A Š C ˚ C, then A has self-
cancellation. [Hint: from C2 ˚ A0 Š C4 argue as in Theorem 10.4.]
470 12 Torsion-Free Groups

11 Dualities for Finite Rank Groups

The main objective of this section is to discuss dualities for finite rank torsion-
free groups. There exist several duality theories which are of independent interest
and deserve particular attention, though they cannot compete in applicability or in
depth with dualities implemented by maps into injective objects. Our exposition is
centered around the Warfield and the Arnold dualities. Both are limited in applica-
tions, but have a number of interesting features that are worthwhile exploring.
Maps from and Into Rank 1 Groups First, we collect information about homo-
morphisms from and into groups of rank 1. In this section, our standing hypothesis
is that all groups are torsion-free and of finite rank (unless stated otherwise), and A
is of rank 1 with End A D R < Q. We observe that Hom.B; C/; B ˝ C are (both
left and right) R-modules if either B or C is an R-module. Also, it is useful to keep
in mind that, if C is a torsion-free R-module, and if B is a pure subgroup of C,
then both B and C=B are R-modules. Also, if both X and Y are R-modules, then
Hom.X; Y/ D HomR .X; Y/, i.e. all group homomorphisms between R-modules are
automatically R-homomorphisms.
The following two lemmas provide relevant information needed in the dualities.
Lemma 11.1 (Warfield [1]). Let A be as stated above. For a torsion-free group G
of finite rank, the homomorphism

 W G ! Hom.A; A ˝ G/

given by .g/.a/ D a ˝ g .a 2 A; g 2 G/ is injective. It is an isomorphism exactly

if G is an R-module.
Proof. Since rk A D 1, rk.A ˝ G/ D rk G, and manifestly rk Hom.A; A ˝ G/ D
rk.A ˝ G/, so  is injective. If  is an isomorphism, then G ought to be an R-
module, as the Hom is such a module.
Conversely, assume G is an R-module. As .G/ is essential in Hom (by counting
the ranks), for an  W A ! A ˝ G we must have an integer m ¤ 0 such that
m.a/ D a ˝ g for some g 2 G and for all a 2 A. We can get rid of a prime power
divisor pk of m for which pR = R, by replacing g by g=.pk / 2 G. If pk j m and
pR ¤ R, then choosing a 2 A n pA, pk j a ˝ g implies pk j g. Thus g can be replaced
by some g0 2 G such that m is not divisible by pk . In this way, at the end of this
process, we get .a/ D a ˝ g0 , establishing the surjectivity of . t
u
Lemma 11.2 (Warfield [1]). With the notation of Lemma 11.1, the homomor-
phism

W Hom.A; G/ ˝ A ! G

given by . ˝ a/ D .a/ . 2 Hom.A; G/; a 2 A/ is monic. Furthermore, Im D

TrA .G/, the trace of A in G.
11 Dualities for Finite Rank Groups 471

Proof. Suppose . ˝ a/ D 0, i.e. .a/ D 0 for some  2 Hom.A; G/ and 0 ¤

a 2 A. Then by torsion-freeness, .A/ D 0, and  D 0. The claim about the trace is
straightforward. t
u
Torsionless and Reflexive Groups Interesting questions arise if we repeat
homomorphisms into the same group A. (Recall A has rank 1, but 6Š Q.) Once the
group A has been fixed, we may use unambiguously the notation G D Hom.G; A/.
Observe:
(a) A Š RC and .RC / Š A.
(b) If G is of finite rank, then always rk G
rk G.
(c) .G ˚ H/ D G ˚ H  .
(d) If G ! H is an epimorphism, then the induced map H  ! G is monic.
Consider the canonical homomorphism

GW G ! G D Hom.Hom.G; A/; A/ D Hom.G ; A/

defined as G .g/./ D .g/ .g 2 G;  2 G / for G of arbitrary finite rank. Adopting

the customary terminology, G will be called A-reflexive if G is an isomorphism, and
A-torsionless in case G is a monomorphism. Note that a homomorphism  W G ! H
induces a map   W H  ! G , and hence a map   W G ! H  .
The following lemma is a simple, but fundamental prelude to the theorem below.
Lemma 11.3. The following conditions are equivalent for a torsion-free group G
of finite rank n:
(i) G is A-torsionless;
(ii) G D Hom.G; A/ has rank n;
(iii) OT.G/
t.A/;
(iv) G is isomorphic to a subgroup of An D A ˚    ˚ A.
Proof. (i) ) (ii) That rk G
rk G is obvious. On the other hand, rk G can equal
rk G only if rk G D rk G.
(ii) ) (iii) OT.G/ > t.A/ would mean that there exists a corank 1 pure
subgroup B of G such that Hom.G=B; A/ D 0. Then in the exact sequence
0 D Hom.G=B; A/ ! Hom.G; A/ ! Hom.B; A/ the last Hom has rank
n  1,
and hence so does Hom.G; A/, in contradiction to (ii).
T (iii) ) (iv) Select pure subgroups B1 ; : : : ; Bn in G, each of rankn n  1, such that
i Bi D 0. Then G embeds in ˚i G=Bi , which in turn embeds in A .
(iv) ) (i) Given a monomorphism  W G ! An , for each 0 ¤ g 2 G, there exists
a canonical projection i W An ! A such that i ..g// ¤ 0. This shows that the
canonical homomorphism G is injective. t
u
The embedding (iv) provides a framework for an analysis of A-torsionless
groups, besides it allows us to confine the study to R-submodules of An where
R = End A.
The next proposition is concerned with the closure properties of the class of
A-torsionless R-modules.
472 12 Torsion-Free Groups

Lemma 11.4. The class of A-torsionless R-modules of finite rank is closed under
finite direct sums, pure subgroups and torsion-free quotients.
Proof. Let B be a pure subgroup of an A-torsionless group G. A moment’s
reflection on the ranks of the groups in the exact sequence 0 ! Hom.G=B; A/ !
Hom.G; A/ ! Hom.B; A/ convinces us that, in view of Lemma 11.3(ii), both B and
G=B are A-torsionless. t
u
Example 11.5. A completely decomposable group G D G1 ˚    ˚ Gn of finite rank is A-
torsionless if and only if all the rank 1 summands have types  t.A/. G is A-reflexive if and
only if t.Gi / D t.A/ for all i.

We also have the pleasant fact about dual groups: G is always A-reflexive.
Lemma 11.6. For a torsion-free group G of finite rank, the map G W G ! G
has an inverse:

 W G ! G where ./.g/ D .G .g// .g 2 G;  2 G /:

Proof. Only a simple calculation is needed right from the definitions:

.G .//.g/ D G ./.G .g// D G ./ D .g/ . 2 G ; g 2 G/

whence G D 1G is immediate. Since G cannot have a larger rank than G ,
also G  D 1G , thus the maps G and  are inverse isomorphisms. t
u
Dualities Let C and D denote two categories. The contravariant functors F W
C ! D and G W D ! C define a duality between C and D if

GF Š 1C and FG Š 1D :

Here we mean isomorphism in the sense that GF.C/ Š C naturally for all C 2 C.
Note that the functors F; G are exact. If C D D and F D G, then we say F is a
duality functor on C.
Motivated by the well-known duality in Linear Algebra for finite dimensional
vector spaces over a field, we consider dualities for full subcategories of finite rank
torsion-free groups defined in terms of a fixed rank 1 torsion-free group A. Always
C D D is assumed, and the functors are F D G D Hom.; A/. The group C D
Hom.C; A/ will be called the A-dual of C. The question we are interested in is
concerned with subcategories C, for which this is a genuine duality, i.e., when the
map C of C into its A-bidual C is an isomorphism for all C 2 C.
Warfield Duality A torsion-free R-module C is said to be locally free (over
R) if the localization C.p/ D C ˝ Z.p/ is a free Z.p/ -module for all primes p with
pR ¤ R. Thus the localization of C at a prime p is either a divisible group or a free
Z.p/ -module according as pR D R or not.
Lemma 11.7. Let A and B be torsion-free groups, A of rank one, and B of finite
rank. Then C D Hom.B; A/ is locally free over R D End A.
11 Dualities for Finite Rank Groups 473

Proof. Clearly, pC D C ¤ 0 if and only if pA D A. If pA ¤ A, then there are

natural injections C.p/ ! Hom.B; A/.p/ ! Hom.B.p/ ; A.p/ / where A.p/ Š Z.p/ .
Let X denote a basic submodule of B.p/ . As B.p/ =X is divisible, but A.p/ is not, the
restriction map Hom.B.p/ ; A.p/ / ! Hom.X; A.p// is injective. Since X and A.p/ are
free over Z.p/ , so is Hom.X; A.p/ /. Therefore, C.p/ is Z.p/ -free as a submodule of a
free Z.p/ -module. t
u
This brings us to Warfield duality:
Theorem 11.8 (Warfield [1]). For A and R as above, define CA as the full
subcategory of the category of torsion-free groups of finite rank whose objects are
the locally free R-modules C satisfying OT.C/
t.A/. The category CA admits a
duality defined by the functor Hom.; A/.
Proof. We show that a torsion-free R-module C of finite rank is A-reflexive if and
only if (i) C is locally free over R, and (ii) OT.C/
t.A/.
If C is A-reflexive, then it is a locally free R-module by Lemma 11.7, and satisfies
(ii) because of Lemma 11.3.
Conversely, assume C has properties (i) and (ii). Then Lemma 11.3 guarantees
that C is A-torsionless. It also follows that Im C is an essential subgroup in C , so
it only remains to verify its purity in C . Suppose the map  W Hom.C; A/ ! A is
divisible by a prime p in C . Then Im 
pA holds. In the non-trivial case when
pA ¤ A, we localize at p to get p W Hom.C.p/ ; A.p/ / ! pA.p/: In view of (i), C.p/ is
a free Z.p/ -module, and therefore the same holds for Hom.C.p/ ; A.p/ /. Consequently
p must be divisible by p. Then  is divisible by p in Im C . t
u
Arnold Duality A more sophisticated duality for a different subclass of finite
rank torsion-free groups was established by Arnold [1]. This duality involves the
quotient-divisible groups, introduced by Beaumont–Pierce [2]. The Arnold duality
works only up to quasi-isomorphism.
A torsion-free group A of finite rank is said to be quotient-divisible if it contains
a free essential subgroup F such that A=F is a divisible (torsion) group. We observe
right away:
(A) If A is quotient-divisible, and F 0 is any free subgroup of the same rank as A,
then A=F 0 is a direct sum of a bounded and a divisible torsion group.
(B) Direct sums and pure subgroups of quotient-divisible groups are again
quotient-divisible.
(C) A group that is quasi-isomorphic to a quotient-divisible group is likewise
quotient-divisible.
Example 11.9. (a) Pure subgroups of Jp as well as finite rank local torsion-free groups are
quotient-divisible.
(b) The groups constructed in the proof of Lemma 4.6 are quotient-divisible.

Let Q denote the category of finite rank quotient-divisible torsion-free groups

with quasi-homomorphisms as morphisms. Then we can state:
Theorem 11.10 (Arnold [1]). There is a duality functor D W Q ! Q that preserves
ranks, and for every prime p satisfies:
474 12 Torsion-Free Groups

rkp D.A/ D rkA  rkp A:

Proof. First we deal with the p-local case, and describe how the duality works
there. Any p-local group A is necessarily quotient-divisible, so we can use the
representation of A as a pull-back, given after Theorem 8.6. The dual D.A/ is
obtained in a similar fashion, switching the roles of the bottom and the right-hand
sequences, and using the matrix P (Sect. 8):

⊕nk=1 Z(p) bk = ⊕nk=1 Z(p) bk

⏐ ⏐
⏐ ⏐
 
0 −−−−→ ⊕m
i=1 Z(p) ci −−−−→ D(A) −−−−→ ⊕nk=1 Qbk −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐can
  
P·can
0 −−−−→ ⊕m
i=1 Z(p) ci −−−−→ ⊕m
i=1 Qci −−−−→ ⊕nk=1 Z(p∞ ) −−−−→ 0

In order to verify that this correspondence preserves quasi-isomorphisms if different

elements bk ; ci are selected, it suffices to refer to our claim in Sect. 8 that such a
change keeps the quasi-isomorphy classes fixed.
In the global case, the duality is defined as the intersection of the locally defined
duals, so it is defined in terms of the matrices Pp , one for each prime p. A switch to
a different basis can change the isomorphy class for a finite number of primes only,
so the intersection of the modified local duals will yield a quasi-isomorphic dual. u t
Fomin Duality More general dualities are also available in the literature, above
all the Fomin and the Vinsonhaler–Wickless dualities. It is evident that if larger
classes of groups are included in a duality, then some pleasant features are lost in
the generalization, and the duality becomes less usable, though it can be instructive
by revealing unnoticed features.
Fomin [2] defines a category in terms of the Richman type of the group ((12.19)
of Sect. 1). Suppose r D .: : : ; kp ; : : :/ and s D .: : : ; `p ; : : :/ are fixed types
(represented by characteristics) such that r
s. The objects of the category Drs
are the torsion-free groups A of finite rank such that in their Richman types, for
each prime p, all the exponents are equal either to kp or to `p . Thus a p-component
looks like

Tp Š Z.pkp /rp ˚ Z.p`p /sp .0

kp
`p /

where rp ; sp are non-negative integers with sum rk A. It is understood that the

morphisms in this category Drs are the quasi-homomorphisms.
The definition of the dual group is similar to the one employed for the Arnold
duality, e.g. for the dual, the ranks rp and sp switch roles. The precise definition is
not simple, and since we are not going to use this duality, we do not elaborate, and
refer to Fomin [2] for details.
12 More on Finite Rank Groups 475

Example 11.11. (a) If r D .0; : : : ; 0; : : :/ (the type of Z) and s D .1; : : : ; 1; : : :/ (the type of
Q), then the objects of the category Drs are the quotient-divisible groups.
(b) If s is the type of Z.p/ , and r is as in (a), then the p-local groups are in Drs .
(c) For any type t, the objects in Dtt are homogeneous of type t.

Vinsonhaler–Wickless [4] formulate a more general duality that includes the

dualities mentioned above as special cases. They write (not uniquely) a torsion-free
group A of finite rank as A D B C C, where B is locally free, and C is quotient-
divisible. Now the duality assigns A D A.B/ C W.C/ to A, where A.B/ is the
Arnold dual of B, and W.C/ denotes the Warfield dual of C. It turns out that A is
unique up to quasi-isomorphism.
F Notes. The Warfield duality has wider appeal in module theory, see Bazzoni–Salce [J.
Algebra 185, 836–868 (1996)] for a far reaching generalization. Goeters [3] extends the Warfield
duality by allowing A to be chosen from a larger class of finite rank groups. Mader–Mutzbauer–
Vinsonhaler [1] establish a duality for the category of almost completely decomposable finite rank
groups, where the categorical isomorphism is near-isomorphism.
Faticoni–Goeters–Vinsonhaler–Wickless [1] prove the following interesting result: let C ; D be
full subcategories of torsion-free groups of finite rank, such that C is closed under isomorphisms,
and taking rank 1 pure subgroups. If there exist functors F W C ! D and G W D ! C implementing
a duality, then both F and G ought to be naturally equivalent to the functor Hom.; A/ for some
rank 1 torsion-free group A.
Fomin–Wickless [1] extend the notion of quotient-divisibility to mixed groups, and show that
the category of all quotient-divisible groups is dual to the category of torsion-free groups.
Krylov [3] studies homomorphism groups into rank 1 groups. For infinite rank groups, an
extensive study is presented, including duality as well as Z-reflexivity. For groups of infinite ranks,
duality has been the subject of a large number of publications. The theory requires advanced set
theory. There are fascinating examples, e.g. @1 -free groups with trivial duals (Pabst [1]). The reader
is referred to Mekler–Schlitt [1]. See Eklof–Mekler [EM] for an excellent presentation.

Exercises

(1) (Warfield) Let A be a rank 1 group with endomorphism ring R. A finite rank
torsion-free group H is of the form H Š Hom.G; A/ for some G exactly if H is
locally free over R.
(2) (a) If G D H ˚ K, then G D H  ˚ K  :
(b) Direct sums and summands of A-reflexive groups are A-reflexive.
(3) If G is a surjective map, then G is reflexive.
(4) Let G be an A-reflexive torsion-free group of finite rank. The endomorphism
rings of G and G are anti-isomorphic.
(5) Find the Arnold dual of a rank 2 pure subgroup of Jp .

12 More on Finite Rank Groups

The Arnold–Lady Category Equivalence Let A be a torsion-free group of finite

rank. We define two categories. One is add.A/: the full subcategory of Ab whose
objects are the finite direct sums of copies of A, and their summands. The other
476 12 Torsion-Free Groups

is proj(E) where E D End A; this is the full subcategory of the category of

right E-modules whose objects are the finitely generated projective modules. The
morphisms are Z- and E-isomorphisms, respectively.
Example 12.1. Let A be a strongly indecomposable torsion-free group of finite rank. Then the
objects in add.A/ are the finite direct sums of groups near-isomorphic to A.

Theorem 12.2 (Arnold–Lady [1]). For a torsion-free group A of finite rank, the
categories add.A/ and proj.E/ are equivalent, as witnessed by the inverse functors

Hom.A; ?/ W add.A/ ! proj.E/ and ? ˝E A W proj.E/ ! add.A/:

Proof. It is clear that F D Hom.A; ?/ and G D ? ˝E A are functors between the

indicated categories. We show that there exists a natural transformation  from GF
to the identity functor of add.A/ given by

C W Hom.A; C/ ˝E A ! C .C . ˝ a/ D .a//

where C 2 add.A//; a 2 A;  2 Hom.A; C/: Also, a natural transformation from

the identity functor of proj(E) to FG such that

PW P ! Hom.A; P ˝E A/ . P ./.a/ D  ˝ a/

where P 2 proj.E/;  2 P; a 2 A: It is evident that if C is a finite direct sum

of copies of A, then C is an isomorphism, and if P is a finitely generated free E-
module, then P is an isomorphism. The same holds for summands, and since the
isomorphisms are natural, the proof is complete. t
u
In a special case we do not have to restrict ourselves in add.A/ to finite direct
sums of copies of A and in proj(E) to finitely generated projective E-modules.
Corollary 12.3 (Arnold–Murley [1]). If A is self-small, then the category of
add.A/ is equivalent to the category proj(E) of projective right E-modules, even
if the extended categories are considered.
Proof. This is rather obvious, since if A is self-small, then Hom.A; / commutes
with infinite direct sums as well. t
u
This theory provides a fresh point of view of known facts on rank 1 groups,
besides extending several useful results from rank 1 to the finite rank case. As a
sample, we prove one such result.
The idea behind condition (i) in Proposition 12.4 was inspired by Baer’s lemma:
in case A D C is of rank 1 and of type t, Lemma 3.6 claims that the exact sequence
in (i) is splitting. (The symbol SA .G/ in the next result stands for the trace TrA .G/ D
h.A/ j  W A ! Gi of A in G.)
Proposition 12.4 (Arnold–Lady [1]). The following are equivalent for a torsion-
free group A of finite rank:
12 More on Finite Rank Groups 477

(i) an exact sequence 0 ! B ! G!C ! 0 of finite rank torsion-free groups is
splitting if G D B C SA .G/ and C 2 add.A/;
(ii) IA ¤ A for all proper right ideals I of E D End A.
Proof. (i) ) (ii) Assume the right ideal I of E satisfies IA D A, so the map  W
I ˝E A ! I A D A is epic. Choose a free right E-module P with an epimorphism  W
P ! I. Then  D . ˝1A / W P˝A ! A is likewise epic. Since SA .P˝A/ D P˝A,
(i) implies that  splits. Consequently, the induced map P Š Hom.A; P ˝ A/ !
Hom.A; A/ Š E is also splitting, thus epic. We now calculate: if  2 P; a 2 A, then
P ./.a/ D  ˝ a 7! ./ ˝ a 7! ./.a/. Hence f./ j  2 Pg D E which along
with ./ 2 I implies I = E. Thus IA D A only if I = E.
(ii) ) (i) It suffices to prove (i) for C D A, because then a splitting map for a
direct sum of copies of A can be obtained individually for the summands. Define the
right ideal I D f ı j ı W A ! Gg; by the hypothesis on G, it satisfies IA D A. Hence
(ii) implies I = E. This means there is ı W A ! G such that ı D 1A , i.e. the exact
sequence in (i) splits. t
u
Irreducible Torsion-Free Groups The concept of irreducibility was introduced
by Reid [3]. He called a torsion-free group G irreducible if every non-zero fully
invariant pure subgroup of G is equal to G. Equivalently, if every non-trivial End.G/-
submodule M of G is an essential subgroup of G (i.e. G=M is torsion).
(A) Irreducibility is invariant under quasi-isomorphism.
(B) Pure subgroups of irreducible groups are irreducible.
(C) An irreducible torsion-free group G is homogeneous. In fact, if t is a type of an
element in G, then G.t/ D G, and this is true for every type in G.
Example 12.5. A strongly indecomposable torsion-free group G of finite rank is irreducible: its
quasi-endomorphism ring Q End G is a division ring, and therefore all non-zero fully invariant
subgroups are essential in G.

It is not difficult to characterize irreducible torsion-free groups of finite rank up

to quasi-isomorphism.
Theorem 12.6 (Reid [3]). For a torsion-free group G of finite rank the following
are equivalent.
(i) G is irreducible;
(ii) G is quasi-isomorphic to a direct sum of a finite number of copies of a strongly
indecomposable, irreducible group H of finite rank;
(iii) Q End G is a full matrix ring over a division ring D.
Proof. (i) , (ii) Let G be irreducible. We have G ˚niD1 Gi with strongly
indecomposable groups Gi (Theorem 9.9). The fully invariant subgroup generated
by any of the Gi must include an essential subgroup of each of the other summands.
This being mutual, all the Gi must be quasi-isomorphic. Conversely, (ii) implies that
a Gi contains the End H-submodule generated by any of its non-zero submodules as
an essential submodule. Furthermore, there are monic maps Gi ! Gj for all i; j.
(ii) , (iii) If G ˚niD1 Gi with quasi-isomorphic strongly indecomposable sum-
mands, then the quasi-endomorphism rings of the components must be isomorphic
478 12 Torsion-Free Groups

division rings. Hence (iii) follows. Conversely, (iii) implies that there are primitive
idempotents i 2 Q End G .i D 1; : : : ; n/ such that G ˚niD1 i G with strongly
indecomposable summands Gi D i G. That Q End G is a full matrix ring over a
division ring D shows that Q Hom.Gi ; Gj / Š D for all i ¤ j, so the Gi ’s are quasi-
isomorphic. t
u
F Notes. Arnold [2] considers strongly homogeneous finite rank groups (any pure rank 1
subgroup can be carried by a suitable automorphism onto any other such subgroup), and proves that
they are exactly the groups of the form R˝H, where H is homogeneous completely decomposable,
and R is a ring in an algebraic numbers field such that every element of R is an integral multiple
of a unit. For a finite rank torsion-free group G, Reid [3] defines the pseudo-socle S as the pure
subgroup generated by all non-zero minimal pure fully invariant subgroups of G. Inter alia, he
shows that G D S if and only if Q End G is a semi-simple artinian ring.
Albrecht [3] extends several results to groups of infinite rank, but he imposes extra conditions
on their endomorphism rings.
In these final notes to Chapter 12, let us make additional comments.
There are numerous publications on the theory of torsion-free groups of finite rank, dealing
with various questions concerning a smaller or a larger subclass. Needless to say, we have made
no attempt to be exhaustive in our presentation. We had to draw a line somewhere, and tried to be
selective so that the ideas that we believe to be central to the theory would stand out.
At this time, the ultimate goal of finding a classification of finite rank torsion-free groups
in terms of reasonable invariants is still beyond our reach, and will very likely remain so for
quite a while. Eklof points out that in L all countable groups are “classifiable,” since a definable
well-ordering of the universe can be used, and a canonical member of each isomorphy class can
be chosen. Of course, this is not a kind of classification that one has in mind. There has been
considerable effort to find reasonable invariants, without much success. The amount of open
questions is staggering. New evidence has begun to emerge that questions the feasibility of a
classification theory. The aim of these remarkable investigations is to verify the conjecture that no
reasonable classification is possible for this class of groups. The closest one could get so far is the
introduction of levels of difficulty in the classification problem, and the characterization of the set-
theoretical complexity of any candidate for complete invariants, providing strong evidence against
classification. (Besides finite rank torsion-free groups, also separable p-groups of cardinality @1 as
well as @1 -separable torsion-free groups of size @1 have been investigated.) The approach is by
forming a standard Borel space, and studying a Borel equivalence relation on the space—in this
way, the complexity of the classification problem can be measured. It has been proved (that we
believe to know anyway from experience) that no classification is possible in the same fashion as
the rank 1 groups were classified via their types. More precisely, it is not provable in ZFC that
the class of torsion-free groups of rank 2 is classifiable in terms of recursive functions (Melles).
For more details, see papers especially by G. Hjorth, A.S. Kechris, G. Melles, and S. Thomas. The
interesting survey by Thomas [1] is highly recommended.

Exercises

(1) (Arnold–Lady) Every C 2 add.A/ is isomorphic to a direct sum of copies of the

finite rank torsion-free group A if and only if each finitely generated projective
right E-module is free.
(2) (Reid) An irreducible torsion-free group of finite rank is strongly indecompos-
able if and only if its quasi-endomorphism ring is a division ring.
12 More on Finite Rank Groups 479

(3) Define the level of a pure fully invariant subgroup of a finite rank torsion-free
group A as follows: f0g is of level 0, and a pure fully invariant subgroup of A is
of level n if it contains properly a level n  1 fully invariant pure subgroup of A,
but none of higher level. Show that for every n 2 N, there is a group An that has
a level n pure fully invariant subgroup.

Problems to Chapter 12

PROBLEM 12.1. Let G; H be torsion-free groups of finite rank. Call H an

immediate extension of G if G
H, and G.t/=G .t/ Š H.t/=H  .t/ for all
types t, where the isomorphism is induced by the injection map. Study immediate
extensions.
PROBLEM 12.2. When are two torsion-free groups of rank n  3 quasi-
isomorphic if they have the same collection of rank n  1 pure subgroups?
PROBLEM 12.3. (Krylov–Tuganbaev). Characterize torsion-free groups in
which every balanced subgroup is a summand.
PROBLEM 12.4. Characterize finite rank torsion-free groups for which quasi-
isomorphism and near-isomorphism are equivalent.
PROBLEM 12.5. Investigate the category of torsion-free groups modulo finite
rank groups. Hom is taken modulo homomorphisms with finite rank images.
PROBLEM 12.6. (Warfield). Find cancellation rings, i.e. rings R such that every
group A with End A Š R has the cancellation property.
PROBLEM 12.7. When does A ˚ B Š A ˚ C imply End B Š End C? (A; B; C are
torsion-free of finite rank.)
Chapter 13
Torsion-Free Groups of Infinite Rank

Abstract This chapter continues the theme of torsion-free groups, this time for the infinite rank
case. There is no shortage of relevant results.
After a short discussion of direct decompositions of countable torsion-free groups, we enter
the study of slender groups which display remarkable phenomena. We provide the main results on
this class of groups. Much can be said about separable and vector groups. These seem theoretically
close to completely decomposable groups, but are less tractable, and so more challenging. The
measurable case is quite interesting.
The theory of torsion-free groups would not be satisfactorily dealt with without the discussion
of the Whitehead problem. For a quarter of century this was the main open problem in abelian
groups. We will give a detailed proof of its undecidability, mimicking Shelah’s epoch-making
solution. We show that the answers are different in the constructible universe, and in a model of set
theory with Martin’s Axiom and the denial of CH.

1 Direct Decompositions of Infinite Rank Groups

In our discussions of torsion-free groups of infinite rank, the first program is to

study their direct decompositions. One can say quite a bit about their strange
decompositions already in the countable rank case, and we will concentrate on this
cardinality.
In contrast to the finite rank case, countable rank groups need not be direct sums
of indecomposable groups. Some rather paradoxical behavior will be shown in the
theorems that follow. In view of our experience with finite rank groups, it should
not be a surprise that nearly everything conceivable can occur in the countable case
as well.
Decompositions into Indecomposable Summands First we consider the case
when the groups are better behaved, and admit decompositions into indecomposable
summands. The first question we take up is concerned with the number of
summands.
Theorem 1.1 (Corner [1]). There exists a torsion-free group A that has two
decompositions
M
ADB˚C D En ;
n2Z

© Springer International Publishing Switzerland 2015 481

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_13
482 13 Torsion-Free Groups of Infinite Rank

where B and C are indecomposable of rank @0 , while the groups En are indecom-
posable of rank 2.
Proof. Let fpn gn ; fqn gn ; frn gn be three, pairwise disjoint, infinite sets of primes,
indexed by n 2 Z. With independent elements bn ; cn define

B Dhp1 1
n bn ; qn .bn C bnC1 / 8 n 2 Zi;

C Dhp1 1
n cn ; rn .cn C cnC1 / 8 n 2 Zi:

Both B and C are of rank @0 and indecomposable. Choose integers kn (to be specified
later), and with un D .1 C kn /bn  kn cn ; vn D kn bn C .1  kn /cn define

En D hp1 1 1 1
n un ; pnC1 vnC1 ; qn rn .un C vnC1 /i .8 n 2 Z/:

The groups En are indecomposable of rank 2. It is straightforward to check that

hbn i ˚ hcn i D hun i ˚ hvn i . From

un C vnC1 D .bn C bnC1 / C kn .bn  cn / C .knC1  1/.bnC1  cnC1 /

it is apparent that in order to have qn j un C vnC1 it is necessary to choose the kn

such that kn  0 and knC1  1 mod qn . Similarly, from

un C vnC1 D .cn C cnC1 / C .1 C kn /.bn  cn / C knC1 .bnC1  cnC1 /

it follows that for rn j un C vnC1 the kn must satisfy knC1  0 and kn  1 mod rn .
Thus if we choose the numbers kn such that
8
ˆ
<0
ˆ mod qn rn1 ;
kn  1 mod qn1 ;
ˆ
:̂1 mod r ;
n

then the En will be subgroups of A D B ˚ C. Moreover, if the kn are (and elementary

number theory tells us that they can be) chosen in this way, then qn j un C vnC1 will
imply qn j bn C bnC1 , and rn j un C vnC1 will imply rn j cn C cnC1 in the direct sum
˚n2Z En , proving that A D ˚n2Z En . t
u
We observe the interesting fact that indecomposable summands of infinite direct
sums of finite rank groups need not be of finite rank.
The method of Theorem 5.1 in Chapter 12 can be extended, with minor
modifications, to yield an analogous conclusion in the countable rank case.
Theorem 1.2. There exist torsion-free groups G that have decompositions G D
A ˚ B D C ˚ D, where B; C; D are indecomposable of rank @0 , and A is completely
decomposable of rank @0 .
1 Direct Decompositions of Infinite Rank Groups 483

Proof. We start with selecting different primes p; q; and pn .n 2 N/. For indepen-
dent an ; bn define

A D ˚1 1 1 1 1
nD1 hpn an i and B D hpn bn ; p q .bn  bnC1 / 8 n 2 Ni:

Then A is completely decomposable, and reasoning as usual we infer that B is

indecomposable. We set G D A ˚ B. Next we choose integers s; t such that
ps  qt D 1. With cn D pan C tbn ; dn D qan C sbn , we set

C Dhp1 1
n cn ; p .cn  cnC1 / 8 n 2 Ni;

D Dhp1 1
n dn ; q .dn  dnC1 / 8 n 2 Ni:

These groups are indecomposable. The equality A ˚ B D C ˚ D follows as in the

proof of Theorem 5.1 in Chapter 12. t
u
Also, Theorem 5.2 in Chapter 12 has an analogue in the countable case.
Theorem 1.3 (Corner [1]). There exists a torsion-free group A of countable rank
such that, for every choice of a sequence r1 ; : : : ; rn ; : : : of positive integers with
infinitely many rn > 1, there exist indecomposable groups An of rank rn such that
A D ˚n2N An .
Proof. Choose different primes p; pn ; qn .n 2 N/, and independent elements
un ; xn .n 2 N/, and define

A D ˚n2N Bn where Bn D hp1 un ; p1 1

n xn ; qn .un C xn /i:

Given a sequence frn gn , let m denote the number of 1’s in this sequence .0
m

@0 ), and let s1 ; : : : ; sn ; : : : be the list of those rn that are > 1. Finally, set t0 D 0; tn D
s1 C    C sn  n. Clearly,

A D ˚n2N Cn where Cn D Bt2n2 C1 ˚    ˚ Bt2n I

Cn is of rank 2s2n1 C 2s2n  2. By the proof of Theorem 5.2 in Chapter 12,

Cn decomposes into the direct sum of an indecomposable group Xn of rank s2n1
C s2n  1 (of the same form as Ai in the proof of the quoted theorem) and
s2n1 Cs2n 1 groups of rank 1 (all of the form hp1 vn i). We regroup the summands:
we let Yn be the direct sum of Xn and a rank 1 group, and keep m rank 1 groups
separately from the Xn . Applying the quoted theorem once again, we rewrite Yn as
a direct sum of two indecomposable groups of ranks s2n1 and s2n . In this way, we
get a decomposition of A into indecomposable summands exactly m of which are of
rank 1, while the rest are of ranks s1 ; : : : ; sn ; : : :. t
u
This remarkable group admits continuously many non-isomorphic direct decom-
positions into indecomposable summands; in fact, there exist continuously many
distinct infinite subsets of the positive integers  2, and owing to this theorem
484 13 Torsion-Free Groups of Infinite Rank

each subset gives rise to a direct decomposition into indecomposable summands

whose ranks are exactly the elements in the selected subset. (We could have admitted
repeated use of the same rank, but there was no need for it.)
The next theorem shows that a countable group may have continuously many
non-isomorphic indecomposable summands even if it is a direct sum of only two
indecomposable groups.
Theorem 1.4 (Fuchs [17]). There exists a torsion-free group A of rank @0 such that

A D Bj ˚ Cj with Bj Š Cj

holds for continuously many, pairwise non-isomorphic indecomposable groups Bj

of countable rank.
Proof. Let Pi ; Qi .i 2 N/ be pairwise disjoint, infinite sets of primes and p; q; r
distinct odd primes not in their union. With independent elements ai ; bi ; ci ; di .i 2 N/
we define

B DhP1 1 1 1 1
i ai ; Qi bi ; p .ai C aiC1 /; q .bi C biC1 /; r .ai C bi / 8 i 2 Ni;

C DhP1 1 1 1 1
i ci ; Qi di ; p .ci C ciC1 /; q .di C diC1 /; r .ci C di / 8 i 2 Ni:

Then B and C are isomorphic indecomposable groups, and set A D B ˚ C, For each
i choose an integer ki (to be specified later), and let

ui D ai ; vi D ki bi C .ki2  1/di ; xi D ki ai C ci ; yi D bi C ki di

for all i 2 N. The aim is to choose the integers ki such that A D U ˚ X where

U DhP1 1 1 1 1
i ui ; Qi vi ; p .ui C uiC1 /; q .vi C viC1 /; r .ui C ki vi / 8 i 2 Ni;

X DhP1 1 1 1 1
i xi ; Qi yi ; p .xi C xiC1 /; q .yi C yiC1 /; r .xi C ki yi / 8 i 2 Ni:

Investigating divisibility by p; q; and r, our usual technique shows that for A D

U ˚ X it is necessary and sufficient that the ki be subject to the conditions:

ki  kiC1 mod pq and ki2  1 mod r for all i 2 N: (13.1)

Pick an integer ` such that `  1 mod pq and `  1 mod r. If, for each i, we
choose ki D 1 or ki D `, then the sequence of such ki will satisfy (13.1), so we get
a decomposition A D U ˚ X with indecomposable U Š X.
We fix k1 D 1, and show that if the sequence k2 ; : : : ; ki ; : : : differs from the
sequence k20 ; : : : ; ki0 ; : : : (i.e., for at least one i, ki ¤ ki0 ), then the corresponding
1 Direct Decompositions of Infinite Rank Groups 485

groups U and U 0 are not isomorphic. Manifestly, divisibility consideration shows

that an isomorphism  W U ! U 0 must act on the generators as ui 7! ˙u0i and
vi 7! ˙vi0 . From

p j ui C uiC1 7! ˙.u0i ˙ u0iC1 /; q j vi C viC1 7! ˙.vi0 ˙ viC1

0
/;
r j u1 C v1 7! ˙.u01 ˙ v10 /

we infer that the signs of u0i and vi0 should be throughout the same, say C. Therefore,
ui 7! u0i ; vi 7! vi0 implies r j ui C ki vi 7! u0i C ki vi0 for every i. This is impossible if
one of ki ; ki0 is 1 and the other is `, because ` 6 1 mod r. To complete the proof,
it only remains to observe that there are continuously many ways of choosing the
sequence k2 ; : : : ; kn ; : : :. t
u
Superdecomposability So far we have dealt with groups which had direct
decompositions with indecomposable summands. The next result establishes
the existence of groups for any infinite cardinal without indecomposable
summands ¤ 0.
A group that has no indecomposable summands other than 0 will be called
superdecomposable.
Theorem 1.5 (Corner [1]). There exist superdecomposable torsion-free groups of
any infinite rank.
Proof. The proof relies on the existence theorem of groups with prescribed
endomorphism rings; see Sect. 7 in Chapter 16. We therefore begin the proof with
introducing our ring R.
Let  denote an infinite cardinal, which will be viewed as the set of ordinals < ,
so it makes sense to form subsets. We define a monoid T to consist of the finite
subsets of  with multiplication defined as

  D [

for all ;  2 T. Evidently, the empty set is the identity of T. L

Define A D ZŒT
as the semigroup algebra of T over Z; thus, A D  2T Z
with free additive group. The completion AQ of A in the Z-adic topology Pis a Z-
algebra whose elements a ¤ 0 may be written as countable sums a D i<! ri i
with ri 2 Z; Q i 2 T, where for every n 2 N almost all coefficients ri are divisible
by nŠ. By the support Œa
of a we mean the set fi j ri 6D 0g  T; this is always a
countable subset of T.
Our claim is that every Z-algebra R between the algebras A and AQ for any infinite
cardinal  is superdecomposable as an algebra. To prove this, consider an algebra
summand S ¤ 0 of R with projection e W R ! S. It is enough to show that the
idempotent e is not primitive. We distinguish two cases.
486 13 Torsion-Free Groups of Infinite Rank

Case 1. If there is an ordinal ˛ 2  not contained in any  2 Œe

, then f˛g 2 R
is an idempotent satisfying ef˛g 6D 0. It also satisfies ef˛g 6D e, since  [ ˛ …
Œe
for any  2 Œe
. The elements ef˛g and e  ef˛g are non-zero orthogonal
idempotents in R with sum e, so e cannot be primitive. S
Case 2. P P ˛ exists, then we must have  D @0 and  D Œe
. Write
If no such ordinal
e D  2Œe
r  DP  2T r  2 R .r 2PZ/ Q with r D 0 for all  2 T n Œe
.
If ef˛g D e, then  2T r . [ f˛g/ D  2T r : Pick  2 Œe
with r ¤ 0.
If ˛ … , then the comparison of the coefficients of  [ f˛g 2 T on both
sides yields r C r [f˛g D r [f˛g : Hence r D 0, contradicting the choice of .
Consequently, ef˛g ¤ e for all ˛ 2 .
PBy way of contradiction, suppose that ef˛g D 0 for all ˛ 2  n Œ
. Then
 2T r . [ f˛g/ D 0, where the coefficient of  [ f˛g is r C r [f˛g D 0.
Thus r [f˛g D r for all ˛ 2  n Œ
, which is impossible. It follows that there
is always an ˛ 2  satisfying ef˛g ¤ 0 (in addition to ef˛g ¤ e). We have
completed the proof that R has no primitive idempotent.
Since the additive group of R is a free group, by Theorem 7.1 in Chapter 16 there
exist groups A of cardinalities   with End A Š R. Such an A is superdecompos-
able, since its endomorphism ring contains no primitive idempotents. t
u
Groups Isomorphic to Certain Powers of Their Own It is easy to find torsion-
free groups A that are isomorphic to A ˚ A, and hence to the direct sum of any finite
or countably infinite copies of itself; e.g. infinite rank free or divisible groups. But
it is a not an easy task to construct an A that is isomorphic to A ˚ A ˚ A, but not to
A ˚ A.
Theorem 1.6 (Corner [3]). For every positive integer m, there exists a countable
torsion-free group A such that An1 Š An2 .ni 2 N/ if and only if n1  n2 mod m.
Proof. We remark at the outset that A has the stated property if and only if A Š
A ˚    ˚ A with m C 1, but not for a fewer number . 2/ of summands. Indeed,
the non-trivial “if” part follows by observing that one has merely to add a certain
number of copies of A to both sides of an isomorphism, and cancel m summands A
whenever there are more than m summands.
The construction of the group A is based on Theorem 3.3 in Chapter 16.
Accordingly, first we define a ring R (on a countable free group) that will be the
endomorphism ring of our group.
Let T be a monoid with 0, generated by symbols i ; i .i D 0; 1; : : : ; m/ subject
to the relations
(
1 if i D j;
j i D
0 if i ¤ j:
1 Direct Decompositions of Infinite Rank Groups 487

Let S denote the semigroup ring of T over Z such that 0 2 T is identified with
0 2 S. The additive group of S is free where the different non-zero products of the
j and i form a basis. Every such product is of the form

i1 i2 : : : ik j` : : : j2 j1 ; (13.2)

where k; `  0, and the indices i; j are taken from the set f0; 1; : : : ; mg. The principal
ideal L generated by  D 1  0 0      m m in S is additively generated by the
elements of the form

i1 i2 : : : ik j` : : : j2 j1 (13.3)

since i  D 0 D i for all i. It is a trivial observation that if the basis

elements (13.2) with k; `  1 and ik D j` D 0 are replaced by the elements of
the form (13.3), then the new system remains an additive basis for S. Consequently,
the additive group of the ring R D S/L is free (and countable).
Our next step is to define a group homomorphism W S ! Z.m/ which vanishes
on L and satisfies

.1/ D 1m and ./ D ./ for all ;  2 S:

Let map all the generators (13.2) to 0 except

.i1 i2 : : : ik ik : : : i2 i1 / D 1m for all k 2 N:

Thus trivially vanishes on the elements (13.3) except when k D ` and it D jt

for t D 0; 1; : : : ; k in which case its value is 1m  .m C 1/1m , which is likewise 0 in
Z.m/. It remains to verify ./ D ./ when  is of the form (13.2), and  is i
or i ; this is left to the reader as an exercise.
In view of Theorem 3.3 in Chapter 16, there is a countable torsion-free group A
whose endomorphism ring is isomorphic to the ring R; for convenience, we assume
End A D R. Then i i D i .i D 0; 1; : : : ; m/ are pairwise orthogonal idempotents
with sum 1, thus A D 0 A ˚    ˚ m A: But i i D i and i i D 1 imply i A Š A;
hence A is isomorphic to the direct sum of m C 1 copies of itself.
Now suppose A is isomorphic to the direct sum of n C 1 copies of itself .0 < n <
m/. In view of Lemma 1.6 in Chapter 16, this means the existence of ˛; ˇ 2 End A
such that ˛ˇ D 1 and ˇ˛ D 0 C    C n . But 1m D .˛ˇ/ D .ˇ˛/ D
.0 C    C n / D .n C 1/ 1m holds for no n if 0 < n < m. t
u
F Notes. A different kind of pathological decomposition was pointed out by Fuchs–Gräbe
[1]. Given any (finite or infinite) set N of integers > 1, there is a countable torsion-free group AN
that has a decomposition into the direct sum of n non-zero indecomposable summands if and only
if n 2 N.
Our Theorem 1.4 can be extended to groups of higher cardinalities : there is a torsion-free
group A of cardinality  admitting 2 non-isomorphic direct decompositions A D Bi ˚ Ci with
indecomposable Bi Š Ci . Also, there are groups A with the same kind of decompositions where all
488 13 Torsion-Free Groups of Infinite Rank

the Bi are isomorphic, and the Ci are non-isomorphic (Fuchs [17]). The proofs rely on the existence
of rigid systems of cardinality . A consequence is that there exist connected compact abelian
groups of cardinality 2 admitting 2 non-isomorphic closed summands which are algebraically
all isomorphic.
Honestly, for uncountable torsion-free groups, anomaly in direct decompositions is less
exciting, since their size makes it easier to use fancy constructions. However, there are certain
limitations. For instance, Kaplansky’s theorem that summands of direct sums of countable groups
are again direct sums of countable groups makes it impossible to have A D B ˚ C D ˚i Ai
with indecomposable summands, where jAj D @1 and the Ai are of finite or countable ranks. A
surprising example is a torsion-free group which is both the countable direct sum and the countable
direct product of copies of the very same group; see Eda’s Proposition 4.9.
Göbel–Ziegler [1] prove that, for every infinite cardinal , there exist groups of cardinality 
that decompose into the direct sum of  non-zero summands for each  < , but not to  non-
zero summands. For each infinite cardinal , Dugas–Göbel [5] establish the existence of torsion-
free groups such that every non-zero summand decomposes into the direct sum of  non-zero
subgroups. Answering a question by Sabbagh [1] and Eklof–Shelah [1] proved the existence, for
every m 2 N, of a torsion-free group A of cardinality 2@0 such that A ˚ Zn Š A holds if and
only if m divides n. Countable superdecomposable groups were constructed by Krylov [2] and
Benabdallah–Birtz [1], using different methods. Corner [8] proves the existence of A D B ˚ C D
D ˚ E of any size 
@0 , where A; B are indecomposable, and each summand of E decomposes
into  non-trivial summands.
Blass [1] investigates when a group A has the property that, for every subgroup B, the size of the
divisible subgroup of A=B does not exceed jBj. He proves that it suffices to check this for countable
B; then it holds for all jBj < @! . Also, it is consistent with ZFC that it fails for @! , but holds for
many larger sizes.

Exercises

(1) Every summand of the group A in the proof of Theorem 1.5 is fully invariant.
(2) (Corner)
(a) Let A be a direct sum of infinitely many indecomposable groups of finite
rank. If almost all ranks are 1, then in every decomposition of A into
indecomposable groups, almost all components must be of rank 1. [Hint:
A is completely decomposable modulo a subgroup of finite rank.]
(b) Conclude that in Theorem 1.3 the condition that infinitely many ri be > 1
is relevant.
(c) Part (a) fails to hold if the components are not all of finite rank.
(3) (Corner) There are countable torsion-free groups A; B; C such that A Š B ˚ C;
B Š A ˚ C; but A and B are not isomorphic. [Hint: choose m D 2 in Theorem
1.6.]
(4) (Corner) There exists a countable torsion-free group A such that A is isomorphic
to the direct sum of any finite number of, but not to the direct sum of infinitely
many copies of itself.
(5) If A is superdecomposable, then so are the groups A./ for every cardinal .
[Hint: direct sums of groups in Theorem 1.5; prove this for a finite sum, a
countable sum, and apply Theorem 2.5 in Chapter 2.]
2 Slender Groups 489

(6) A torsion-free group of infinite rank is a subdirect product of copies of Q.

(7) (Reid) Every torsion-free group of infinite rank is a sum of two free groups.

2 Slender Groups

A remarkable class of torsion-free groups was discovered by J. Łoś in the 1950s:

the class of slender groups. Without further ado, we embark on their very attractive
theory.
Let P denote the direct product of a countable set of infinite cyclic groups, and S
its subgroup that is the direct sum of the same groups:
1
Y 1
M
PD hen i; SD hen i; where hen i Š Z:
nD1 nD1

P1
The elements of P can be written as formal infinite sums: x D nD1 kn en , or as
infinite vectors, like x D .k1 e1 ; : : : ; kn en ; : : :/, or else simply as x D .k1 ; : : : ; kn ; : : :/,
where kn 2 Z.
Slenderness A group G is called slender if, for every homomorphism
 W P ! G, we have en D 0 for almost all indices n. From this definition it
is evident that the following hold true:
(a) Subgroups of slender groups are slender.
(b) The group P is not slender.
(c) No injective group ¤ 0 is slender, so slender groups are reduced.
(d) The group Jp of p-adic integers, and more generally, an algebraically compact
group ¤ 0, is not slender. In fact, the free group S can be mapped into Jp by a
homomorphism  such that en ¤ 0 for all n. Since S is pure in P, and Jp is
algebraically compact,  extends to a map  W P ! Jp with en ¤ 0 for all n.
(e) Slender groups are torsion-free. The argument in (d) works for Z.p/ (for any
prime p), since it is algebraically compact. A slender group cannot contain a
non-slender Z.p/, so it must be torsion-free.
The following lemma is a consequence of our remarks.
Lemma 2.1. If  W P ! G with G slender, then S D 0 implies  D 0.
Homomorphic image of P in a slender group is a finitely generated free group.
Proof. If  is as stated, then Im  is an epic image of the algebraically compact
group P=S, so it is a cotorsion group. A torsion-free cotorsion group is algebraically
compact. No such group can be a subgroup of G, hence Im  D 0. The second claim
is a simple corollary to the first, using the definition of slenderness. t
u
490 13 Torsion-Free Groups of Infinite Rank

A brief analysis of the preceding lemma makes it apparent that every homomor-
phism  of P into a slender group G is completely determined by its restriction
  S. As a consequence, we can write
X1  X1
 kn en D kn .en / 2 G;
nD1 nD1

where in the last sum almost all terms are 0.

An important observation is contained in the following simple lemma.
Lemma 2.2. The class of slender groups is closed under taking extensions.
Proof. Let G be an extension of the slender group A by the slender group B, and
let  W P ! G. Only finitely many en are contained in A, and the same holds for
G=A Š B and the images of  followed by the canonical map. Therefore, almost all
en D 0, and G is slender as well. t
u
One can exhibit a large number of examples of slender groups by making use of
the following two lemmas.
Lemma 2.3 (Sa̧siada [2]). A torsion-free group of cardinality less than the contin-
uum is slender if and only if it is reduced.

Proof. Because of (c), it suffices to verify the “if” part of the statement. Let G
be reduced torsion-free of cardinality < 2@0 , and suppose  W P ! G is a
homomorphism such that en ¤ 0 for infinitely many indices n. Omitting the
groups hen i with en D 0, we may assume en ¤ 0 for all n. Being reduced
means \k2N kG D 0, thus we can find an increasing sequence 1 D k1 <    <
kn <    of integers such that .kn Šen / … knC1 G for each n. Consider the set of
all .r1 ; : : : ; rn ; : : :/ 2 P where rn D 0 or rn D kn Š; this set is of the power of the
continuum, therefore there exist at least two elements in this set which have under
 the same image in G. Their difference is of the form a D .s1 ; : : : ; sn ; : : :/ where
sn D 0 or ˙kn Š, and clearly a D 0. We claim that this is impossible. Indeed, if j is
the first index with sj ¤ 0, then .sj ej / … kjC1 G, but

.sj ej / D .0; : : : ; 0; sjC1 ; : : :/ 2 kjC1 G;

since on the right side every coordinate is divisible by kjC1 . t

u
Lemma 2.4 (Fuchs [AG]). Direct sums of slender groups are slender.
Proof. The claim is obvious for finite direct sums. Let Gi .i 2 I/ be slender
groups, and G D ˚i2I Gi with coordinate projections i W G ! Gi . Given
 W P ! G, Lemma 2.1 implies that each i P is a finitely generated free subgroup
of Gi , and Im 
˚i2I i P. Here but a finite number of i P can be different from
0, since otherwise P had a forbidden homomorphism into a countable free group,
contradicting Lemma 2.3. Therefore, almost all i  D 0, and  may be viewed as
a map from P into the direct sum of finitely many Gi ; say, into the slender group
G1 ˚    ˚ Gm . Hence almost all of en .n 2 N/ vanish, which means that the direct
sum G is slender. t
u
2 Slender Groups 491

Example 2.5. The following groups are slender: Z, and more generally, every proper subgroup of
Q, and direct sums of these groups: the reduced completely decomposable groups. Also, direct
sums of any number of reduced countable torsion-free groups Ai . On the other hand, Q; P are not
slender.

Monotone Subgroups Though the group P is not slender, it contains a lot of

uncountable slender subgroups, Q as is shown by the next theorem.
A subgroup M of P D n2N hen i is called a monotone subgroup (Specker [1])
if it satisfies:
P P
(i) n2N kn en 2 M implies that also n2N mn en 2 M where mn D
max.1;
P jk 1 j; : : : :; jk n j/I P
(ii) n2N kn en 2 M and j`n j
jkn j .n 2 N/ imply n2N `n en 2 M.
Example 2.6.
(a) P
Every monotone subgroup contains the smallest monotone subgroup B consisting of all
n2N kn en with jknP
j bounded.
(b) The infinite sums n2N kn en , for which there is a c 2 N such that the coefficients satisfy
jkn j  cnŠ, form a monotone subgroup.

Interestingly, monotone subgroups are slender with a single exception: the

obvious one.
Q
Theorem 2.7 (G. Reid [1]). All monotone subgroups of the product P D n2N hen i
are slender, with the exception of P itself.
Proof. Let M be a monotone subgroup of P, properly contained in P, and assume
there is a homomorphism W P ! M such that en ¤ 0 for infinitely many indices
n. We may assume that en ¤ 0 for all n. Infinite cyclic groups are slender, so
for each k, there are but finitely many indices n for which the coordinate of en
in hek i is non-zero. In view of this, we may even drop more summands from P
to have en D .0; : : : ; 0; tn;in ; tn;in C1 ; : : :/ with tn;in ¤ 0 for a strictly increasing
sequence i1 <    < in <    of indices.P Pick an x D .`1 ; : : : ; `n ; : : :/ 2 P n M with
1 < `1
  
`n
   , and set y D n2N sn en 2 P by letting sn D ˙`in where
the sign is to be chosen so as not to have s1 t1;in C  Csn1 tn1;in and sn tn;in opposite
signs. As is determined by the en ; we conclude that y D .r1 ; r2 ; : : : ; rn ; : : :/
where rn D s1 t1n C    C sn1 tn1;n C sn tn;n . But jrin j  jsn tn;in j  `in and x … M
imply y … M, a contradiction. Hence M is slender, indeed. t
u
Main Theorem by Łoś The results so far are still incomplete inasmuch as we
lack information on homomorphisms of uncountable products into slender groups.
Strangely enough, measurable cardinals enter into the picture.
A main result on slender groups is the Łoś–Eda theorem. First we prove Łoś’
theorem for non-measurable cardinals, and then we move to the general case. Q
For an index set I of cardinality , we will use the ad hoc notations: P D i2I Ai
and S D ˚i2I Ai for the direct product and the direct sum, respectively, of arbitrary
non-zero torsion-free groups Ai .i 2 I/.
492 13 Torsion-Free Groups of Infinite Rank

Theorem 2.8 (Łoś). Let G be a slender group, and  W P ! G. Then

(i) Ai D 0 for almost all i,
(ii) if  is not measurable and S D 0, then  D 0.
Proof.
(i) Suppose, by way of contradiction, that Ai ¤ 0 holds for infinitely many
indices, say, for i D i1 ;Q : : : ; in ; : : :. Then choosing en 2 Ain with en ¤ 0, the
restriction of  to P0 D 1 nD1 hen i.
P / would be a forbidden homomorphism
into a slender group.
(ii) This is also an indirect proof, but more sophisticated. Suppose that S D 0,
but there exists an a 2 P with a ¤ 0. We introduce a G-valued “measure”
 on the subsets of I, by using the selected element a. For a subset J of I, we
set .J/ D aJ , where aJ has the same coordinates as a for indices j 2 J, and
0 coordinates elsewhere. It is clear that for pairwise disjoint subsets J; K of I,
we have .J [ K/ D .aJ[K / D .aJ C aK / D .J/ C .K/, so that  is
additive. In order to convince ourselves that it is moreover countably additive,
let J1 ; : : : ; Jn ; : : : be a countable set of pairwiseQ disjoint subsets of I, and J0 the
complement of their union J in I. Then P D 1 kD0 haJk i is a subgroup of P .
Applying the definition of slenderness to   P W P ! G, we conclude that
almost all of aJk D .Jk / D 0. This shows that  is countably additive. An
appeal to Lemma 6.4 in Chapter 2 shows that I is measurable, completing the
proof. t
u
If P is equipped with the product topology (keeping the components discrete),
then condition (i) in Theorem 2.8 amounts to that every homomorphism of P into
a (discrete) slender group is continuous. We should point out that (ii) is the best
possible result in the sense that for measurable cardinals , S D 0 no longer
implies  D 0. This is illustrated by the following example, generalizing Łoś’
original example.
Example 2.9. For a measurable index set I, there are non-trivial homomorphisms  of P D ZI
into any torsion-free group G ¤ 0 such that S D 0 where S D Z.I/ . To prove this, we choose
any 0 ¤ g 2 G. Let  be a f0; 1g-valued measure on I, and for a D .: : : ; ki ; : : :/ 2 P define
pairwise disjoint subsets Xn .a/ I .n 2 Z/ via Xn .a/ D fi 2 I j ki D ng. Then the union of these
subsets is I, hence exactly one of them, say, Xm .a/ has measure 1, and the rest have measure 0. We
then set a D mg, thus P ¤ 0: From the properties of  it is readily checked that  preserves
addition, and S D 0.

At this point let us pause and list a couple of corollaries to the last theorem.
Corollary 2.10. Let G be a slender group, and Ai .i 2 I/ torsion-free groups. If the
index set I is non-measurable, then there is a natural isomorphism
Y 
Hom Ai ; G Š ˚i2I Hom.Ai ; G/:
i2I
2 Slender Groups 493

Proof. From Theorem 2.8 we know that every element  of the left-hand side Hom
is actually a map from a finite sum Ai1 ˚  ˚Aik into G (where the index set depends
on ). Hence the claim is evident. t
u
The following corollary exhibits an interesting duality between direct sums and
products of infinite cyclic groups. It shows that direct sums and products of Z over
a non-measurable index set are Z-reflexive.
Corollary 2.11 (Zeeman [1]). For a non-measurable set I,
Y Y 
Hom.˚i2I Z; Z/ Š Z and Hom Z; Z Š ˚i2I Z
i2I i2I

are natural isomorphisms.

Proof. These isomorphisms are special cases of Theorem 1.7 in Chapter 7
and Corollary 2.10, respectively. u
t
Corollary 2.12. If I is a non-measurable set, then every summand of ZI is likewise
a product.
Proof. If ZI D A ˚ B, then Z.I/ D Hom.A; Z/ ˚ Hom.B; Z/, where the summands
are free groups (Corollary 2.11). Hence A Š Hom.Hom.A; Z/; Z/ is also a
product. t
u
Main Theorem by Eda Next we turn our attention to the generalization
of Theorem 2.8 to arbitrary cardinalities. We will not give a detailed proof for the
main result (Theorem 2.14), but prove a crucial lemma which will help to appreciate
the originality and the cleverness of the idea involved.
Lemma 2.13 (Eda [1,2]). Let G be a slender group, and Ai .i 2 I/ a family of
torsion-free groups, jIj D . For every homomorphism
Y
 W P D Ai ! G
i2I

there exists a finite number of @1 -complete ultrafilters U1 ; : : : ; Uk on  such that for

each a 2 P ; supp a … Uj for j D 1; : : : ; k implies .a/ D 0.
Proof. For a 2 P , and for a subset J of I define aJ 2 P to have its ith coordinate
equal to the ith coordinate ai of a or to 0 according as i 2 J or not. Let PJ denote
the set of all a 2 P with supp a  J.
Consider the set S D fJ  I j .PJ / ¤ 0g. We claim that there are no infinitely
many pairwise disjoint sets in S. In fact, if the sets J1 ; : : : ; Jn ; : : : were pairwise
disjoint membersQin S and .aJn / ¤ 0 for each n, then  would induce a forbidden
homomorphism n<! haJn i ! G.
We now restrict the set S to

S D fJ  I j 8 K  J exactly one of .PK / ¤ 0; .PJnK / ¤ 0 holdsg:

494 13 Torsion-Free Groups of Infinite Rank

For a fixed J 2 S , define

U D fX  I j .PX\J / ¤ 0g:

Claim I. U is an !1 -complete ultrafilter.

(a) The empty set does not belong to U. Subsets of I containing a member of U also
belong to U.
(b) U is a filter. Suppose X; Y 2 U, but X \ Y … UI without loss of generality,
we may assume that X; Y are subsets of J. Then for some a 2 P ; 0 ¤
.aX / D .aX\Y /C.aXnY /, where the first term vanishes by hypothesis. Hence
.PXnY / ¤ 0, and analogously, .PYnX / ¤ 0. But this contradicts J 2 S ,
because for the choice of K D Y n X both .PK / ¤ 0 and .PJnK / ¤ 0 would
hold.
(c) U is an ultrafilter. For, if X … U and a 2 PJ is such that .a/ ¤ 0, then
0 ¤ .a/ D .aX\J / C .aJnX / D .aJnX /; so I n X belongs to U.
(d) U is @1 -complete. Let the subsets X1 ; : : : ; Xn ; : : : of I belong to U such that their
intersection is empty; here again, we may assume that they are all subsets of
J, and moreover, X1      Xn     . For each n, pick an 2 PP Xn such that
.an / ¤ 0, and define a map  W P ! G via .: : : ; kn ; : : :/ D . n<! kn an /.
The existence of such a  contradicts the slenderness of G.
Next choose fJ1 ; : : : ; Jk g, a maximal set of pairwise disjoint sets in S , and denote
by Uj the ultrafilter defined by Jj for j D 1; : : : ; k, as described above.
Claim II. If K denotes the complement of the set union J1 [    [ Jk in I, then
.PK / D 0.
Assume the contrary, i.e. K 2 S. By the maximal choice of the Jj 2 S , K
cannot belong to S , so there is a subset X1 of K such that both .PX1 / ¤ 0 and
.PKnX1 / ¤ 0. Suppose that we have found pairwise disjoint subsets X1 ; : : : ; Xn of
K such that for each i D 1; : : : ; n, both .PXi / ¤ 0 and .PKn.X1 [[Xi / / ¤ 0. To find
an XnC1 with similar features disjoint from all of X1 ; : : : ; Xn , just iterate the same
argument with K replaced by K n .X1 [    [ Xn /. If we choose Q for each n an aXn
such that .aXn / ¤ 0, then  induces a forbidden homomorphism n<! haXn i ! G.
Hence it is clear that the Uj are as claimed. t
u
Now we state the theorem, and refer to the proof to Eklof–Mekler [ME],
Section III.3.
Theorem 2.14 (Eda [1]). Suppose that fAi j i 2 Ig is a family of torsion-free
groups, and G is a slender group. There is a natural isomorphism
Y  Y  
Hom Ai ; G Š ˚U Hom Ai =U; G ;

where U runs over all @1 -complete ultrafilters on the index set I. t

u
We refer to the combined Theorems 2.8 and 2.14 as the Łoś-Eda theorem.
2 Slender Groups 495

F Notes. The roots of the theory of slender groups are in Specker [1] where it is proved
under CH that Hom.Z@0 ; Z/ Š Z.@0 / , establishing the slenderness of Z. (He was motivated by the
search for the structure of the first cohomology group of an infinite complex.) Specker’s result was
extended by Ehrenfeucht–Łoś [1] and Zeeman [1]. The theory of slender groups is the creation of
J. Łoś, his theory was published in Fuchs [AG]. Two main theorems, Theorems 2.8 and 3.5, are
due to Łoś and Nunke [2], respectively.
Eda [1] proved that Hom.Z ; Z/ is a free group for all cardinals . It is an undecidable problem
whether or not the group P could be replaced by one of its subgroups of smaller cardinality to use
as a test group for slenderness. He also discussed extension of Theorem 2.8 to the Boolean powers
of Z, and to sheaves.
The theory of monotone subgroups was initiated by Specker [1]. Irwin–Snabb [1] discussed
monotone subgroups defined by analytic functions with integral coefficients. Göbel–Wald [1]
introduced M-slenderness (replacing P in the definition by a monotone subgroup M), and proved
@
that a consequence of Martin’s Axiom is that there are as many as 22 0 inequivalent notions of
slenderness. They also proved that M=S is algebraically compact for every monotone M.
For Z-reflexive groups, see Huber [2], and above all Eklof–Mekler [EM]. Göbel–Shelah [4] use
Martin’s Axiom to construct reflexive subgroups of the Baer–Specker group.

Exercises

(1) A torsion-free group G is slender if and only if, for all  W P ! G, Im  is

finitely generated. [Hint: finitely generated groups are slender.]
P1 that if xn .n D 1; 2; : : :/ are elements of P, then an infinite sum x D
(2) Show
nD1 sn xn with sn 2 Z makes sense in P if and only if, for each positive
integer i, almost all of sn xn have vanishing ith coordinates.
(3) (Nunke)
P1 Give Pa different proof that Jp is not slender by showing that
1
nD1 k n e n !
7 nD1 p
n1
kn defines a non-zero homomorphism P ! Jp .
(4) (a) Prove Lemma 2.1 by selecting any x1 D .k1 ; : : : ; kn ; : : :/ 2 P n S with
x1 ¤ 0, and showing that  induces a forbidden homomorphism of
Q
hxn i ! G, where xn D x1  k1 e1      kn1 en1 .
(b) P can be represented in many ways as a direct product of cyclic groups.
(5) (T. Yen) Let P0 denote the subgroup of P for which P0 =S is the maximal
divisible subgroup of P=S. Slenderness can be tested by using P0 in place of
P. [Hint: .a1 ; : : : ; an ; : : :/ 7! .a1 ; : : : ; nŠan ; : : :/ is a monic map P ! P0 .]
(6) (a) Let P0 D Z with a non-measurable cardinal . Show that every summand
of P0 is again a product of infinite cyclic groups.
(b) In any direct decomposition of P0 , there are but a finite number of non-zero
summands. If   @0 , then one of the summands is Š P0 .
(7) (Łoś) Let Gi .i Q 2 I/ be slender groups, where I is a non-measurable infinite
index set. Then i2I Gi has no proper summand containing ˚i2I Gi .
(8) An infinite product of reduced countable torsion-free groups can never be a
direct sum of countable groups.
(9) (Fuchs) Let G be a countable slender group, and M a monotone subgroup of
P, containing some unbounded vectors. Each homomorphism  W M ! G
satisfies en D 0 for almost all n, and  is determined by   S.
496 13 Torsion-Free Groups of Infinite Rank

(10) (Specker) Let M1 and M2 be monotone subgroups of P, and W M1 ! M2 a

homomorphism such that S
Im . Then M1
M2 .
(11) (Specker) The set of non-isomorphic monotone subgroups of P has cardinality
@
22 0 . [Hint: find continuously many elements in P such that none of them is
contained in the monotone subgroup generated by the rest.]
(12) (Eklof–Mekler) Let G denote the subgroup of Z consisting of elements with
countable support. Show that Hom.G; Z/ Š Z./ . [Hint: every countable
subset of G is contained in a subgroup Š Z@0 .]

3 Characterizations of Slender Groups

This section is devoted to the discussion of the characterizations of slenderness, both

in topological terms and by the absence of certain subgroups. The center stage is
now taken by topology as the proofs draw heavily on (rather elementary) topological
arguments. Q
We always think of P D 1 nD1 hen i as being equipped with the product topology
where each hen i carries the discrete topology. This is a metrizable linear topology, in
which P is complete, and S D ˚1 nD1 hen i is a dense subgroup of P. We observe right
away that if C is any subgroup of P, then its completion CM in the induced topology is
the same as its closure in P. Now CM consists of all x 2 P with the following property:
for every n 2 N, there is cn 2 C such that the first n coordinates of x  cn are 0.
Topological Results No further preparation is needed for the following topo-
logical characterization of slenderness.
Theorem 3.1 (Heinlein [1], De Marco–Orsatti [1]). A reduced torsion-free group
is slender if and only if it contains no subgroup that admits a complete non-discrete
metrizable linear topology.
Proof. To prove necessity, let G be a subgroup of the torsion-free group A, and
Un .n < !/ a strictly descending chain of subgroups that forms a base of
neighborhoods of 0 in a complete non-discrete metrizable linear topology of G.
Define a map  W S ! G by setting .en / D gn ; where gn is any element of
Un nUnC1 . The universal property of completions guarantees that  extends uniquely
to a continuous homomorphism W P ! G. Such a map contradicts slenderness,
showing that G, and hence A, cannot be slender.
Conversely, suppose A is reduced, but not slender. Then there is a map  W P ! A
such that en ¤ 0 for all n. Thus A contains the subgroup G D Im  Š P=K where
K D Ker . If K D 0, then G Š P, so G admits a stated kind of topology. Now, if
K ¤ 0, then its closure in P is its completion KM in the topology inherited from P. If
KM D K, then P=K is the epic image of a complete metrizable group, and as such it
is complete in the induced topology. If KM > K, then by Proposition 6.8 in Chapter
M
9 K=K is cotorsion; it is moreover algebraically compact, since it is torsion-free.
3 Characterizations of Slender Groups 497

Evidently, it is reduced; therefore, by Corollary 3.6 in Chapter 6 it must contain a

copy of a Jp , which is obviously compact in its (metrizable) p-adic topology. t
u
It is remarkable that slender groups can be characterized in an unusual way
as groups not containing certain specific types of subgroups. This is Nunke’s
theorem that we are going to prove next. First, we examine more closely what the
completions of Z are like in linear topologies.
Lemma 3.2. The completion ZM of Z in any non-discrete, linear topology is a
compact group. If ZM is torsion-free, then it contains a copy of Jp for some prime p.
Proof. The open sets about 0 in any non-discrete linear topology of Z are necessar-
ily finite index subgroups. Thus the completion ZM (as a closed subgroup in the direct
product of finite groups) is necessarily compact. It is obviously reduced, and again
we invoke (Corollary 3.6 in Chapter 6) to argue that it must contain a copy of some
Jp whenever it is torsion-free. t
u
Example 3.3.
Q Š Jp .
(a) If we choose the subgroups pn Z .n < !/ as a base of neighborhoods about 0, then Z
(b) If the subgroups nZ with square-free numbers n are chosen as a base of neighborhoods, then
Q
M Š p Z.p/.
Z
Lemma 3.4. Let A be a reduced torsion-free group that is complete in some non-
discrete metrizable linear topology. Then it contains a copy of P or a copy of Jp for
some prime p.
Proof. Suppose that A contains a cyclic group C Š Z that is not discrete in the
induced topology. Then it also contains its completion C,M and Lemma 3.2 implies
that it must contain a copy of some Jp .
In the remaining case, all cyclic subgroups of A are discrete in the induced
topology. Let V1 >    > Vn >    be a base of neighborhoods at 0. For any
0 ¤ c1 2 A, hc1 i carries the discrete topology, hence there exists n1 2 N such that
hc1 i\Vn1 D 0. Pick any 0 ¤ c2 2 Vn1 , and then an n2 > n1 such that hc2 i\Vn2 D 0.
Continuing in the same fashion, we construct a sequence ci .i 2 N/ of elements in
A, and a strictly descending chain Vni .i 2 N/ of open subgroups. The ci generate
the direct sum C D ˚i2N hci i Š S, and a quick comparison of the topologies shows
that this is even a topological isomorphism. It follows that the closure of C is its
completion, which is evidently Š P. t
u
Nunke’s Theorem on Slenderness The interest in slender groups was strongly
influenced by the light shed by the following theorem which gives a most powerful
characterization of slender groups.
Theorem 3.5 (Nunke [2]). A group is slender exactly if it does not contain any
subgroup isomorphic to Q; P; Jp , or Z.p/ for any prime p.
Proof. The necessity being obvious (the indicated groups are not slender), assume
that no subgroup of the torsion-free group G is isomorphic to Q; P or Jp . Thus G is
reduced, so by Lemma 3.4 G does not contain any subgroup that is complete in any
metrizable linear topology. An appeal to Theorem 3.1 completes the proof. t
u
498 13 Torsion-Free Groups of Infinite Rank

The argument on the preceding theorem rested on the homomorphic images of P

in torsion-free groups. Let us find out in general what epic images of P might have.
(Actually, the insight obtained from the next proposition was the key that eventually
led to Theorem 3.5.)
Proposition 3.6 (Nunke [2]). Every epic image of P is the direct sum of a cotorsion
group, and a direct product of at most countably many infinite cyclic groups.
Proof. For a group A, we use the notation A D Hom.A; Z/. This Hom is always a
subgroup of a product of infinite cyclic groups. In particular, for a free group S of
countable rank, we have S Š P and P Š S.
If A is a subgroup of P, then the exact sequence 0 ! A ! P ! P=A ! 0
induces the exact sequence 0 ! .P=A/ ! P ! A where P is a countable free
group. Let B denote the image of P in A . As A (and hence B) is contained in a
product of infinite cyclic groups, and B is countable, by Theorem 8.2 in Chapter 3, B
is a free group, consequently, P Š B ˚ .P=A/ . Hence P Š P Š B ˚ .P=A/ ,
where both summands are products of cyclic groups, and A
A
B . Thus
P=A Š B =A ˚ .P=A/ , and we want to show that B =A is cotorsion.
First, .B =A/ D 0; because the canonical map .P=A/ ! .P=A/ is an
isomorphism (Corollary 2.11), and the projection onto the second summand in
.P=A/ Š .B =A/ ˚ .P=A/ is the natural map.
As B Š P, in order to complete the proof, it suffices to show that P=A is
cotorsion whenever .P=A/ D 0. Let AM denote the closure of A in P, where P is
equipped with the product topology. Then AM is complete in the induced topology,
M
so Proposition 6.8 in Chapter 9 implies that A=A M  D 0 ensures
is cotorsion. .P=A/
M
that, for every neighborhood U of 0 in P, U \ A is essential in U. Thus the groups
.U C A/=M AM form a base of neighborhoods of 0 in P=A.M The factor group P=AM is also
complete in the induced topology, and it follows that P=AM is the inverse limit of the
finite groups P=.U C A/,M and as such it is a compact group. To conclude, it remains
M by a cotorsion group P=AM
to observe that an extension P=A of a cotorsion group A=A
is itself cotorsion. t
u
By making use of this theorem, we can retrieve our earlier result (Theorem 3.5)
as an immediate corollary. It is also easy to derive Lemmas 2.3 and 2.4. As another
application of Nunke’s theorem we prove the theorem below.
Chains with Slender Factors By Lemma 2.2, the class of slender groups is
closed under finite chains, i.e. if all factors AiC1 =Ai of a chain 0 D A0 < A1 <
   < Ak (k 2 N) are slender, then so is Ak . We now want to replace finite chains by
smooth chains with no restriction on length.
Theorem 3.7 (Fuchs–Göbel [1]). Let

0 D A0 < A1 <    < A <    . < / (13.4)

be a smooth chain of groups where all the factor groups A C1 =A . < / are
slender. Then the union A D [ < A of the chain is likewise slender.
3 Characterizations of Slender Groups 499

Proof. We first observe that all the groups A are pure in A because of the torsion-
freeness of the factor groups. The following proof is by induction on the length  of
the chain.
Case 1:  D n is an integer. We refer to the observation just before the theorem.
Case 2:  D !, i.e. we have a chain 0 D A0 < A1 <    < Ak <    .k < !/, and
A D [k<! Ak . Case 1 implies that all the Ak are slender groups. It suffices to
show that A cannot contain any copy of the groups Q; Jp ; or P.
(a) Since none of the subgroups Ak contains a copy of Q, neither can A. By the
same token, A=Ak contains no copy of Q.
(b) Suppose, by way of contradiction, that A contains a copy of Jp for some
prime p. Jp is the completion in the p-adic topology of the localization Z.p/
of Z. Let a 2 A denote the element corresponding to 1 2 Z.p/ . There is an
index k such that a 2 AkC1 nAk , thus Z.p/ embeds in AkC1 =Ak as a subgroup.
As Jp =Z.p/ is divisible, and by (a) A=AkC1 fails to contain any copy of Q,
it follows that Jp is contained already in AkC1 =Ak . But this is impossible,
since by hypothesis this factor group is slender. Similarly, A=Ak contains no
copy of Jp for any prime p.
Combining parts (a) and (b), we conclude that no factor A=Ak can contain
any non-zero cotorsion group.
(c) Again,Qarguing by contradiction, assume that there is a monomorphism  W
P D n2N hen i ! A. It suffices to consider the case where no Ak contains
infinitely many of the en . Now  followed by the canonical maps A !
A=Ak yields a homomorphism
Y
N W P ! A=Ak :
k<!

For each x 2 P, there exists an index k < ! such that x 2 Ak , which means
that necessarily Im N
˚k<! A=Ak . In view of Theorem 6.5 in Chapter 2
there must exist an index m < ! such that Im N
A=A1 ˚    ˚ A=Am .
Consequently, Im 
AmC1 , in contradiction to the slenderness of AmC1 .
The proofs in (a)–(c) establish the claim for chains of length !.
Case 3:  > !. The arguments above in (a) and (b) apply to arbitrary , thus it
remains to settle the case about P. So assume that for chains of smaller lengths
the claim holds, and by way of contradiction, suppose that A contains a copy
of P.
If the length  is a successor ordinal,  D  C 1, then we invoke Lemma
2.2. If  is a limit ordinal, and if all the generators en 2 P are contained in A
for some ordinal  < , then the proof in Case 2 should convince us that then
also P
A . If  is a limit ordinal and cf  ¤ !, then such a  (a successor
ordinal or a limit ordinal of cofinality !) must exist. In this case the induction
hypothesis implies that it is impossible that a copy of P is contained in A.
500 13 Torsion-Free Groups of Infinite Rank

Finally, if  is cofinal with !, and if there does not exist any index  <  with
en 2 A for all n, then we select a countable subchainS 0 D A0
A˛1
  

A˛k
   with union A such that we have en 2 k<! A˛k for all n < !. The
induction hypothesis guarantees that all the factor groups A˛k =A˛k1 are slender,
so an appeal to Case 2 leads us to the desired conclusion: the union A must be
slender. t
u
Cotorsion-Free Groups We close this section with drawing attention to another
class of groups which is more general than slender groups, and which earned
importance in the construction of groups with prescribed endomorphism rings.
A group A is said to be cotorsion-free if it does not contain any cotorsion
subgroup ¤ 0. These groups can be characterized as follows.
Theorem 3.8. The following are equivalent for a group A:
(i) A is cotorsion-free;
(ii) A contains no copy of Q; Jp or Z.p/ for any prime p;
(iii) Hom.Z; Q A/ D 0.

Proof. A cotorsion-free group contains none of the listed groups as a subgroup,

hence (i) and (ii) are equivalent by Theorem 3.5. The completion ZQ is a compact
group, all of its homomorphic images are cotorsion, and each of the groups in (ii) is
an epic image. Thus (iii) is equivalent to (ii). t
u
F Notes. Actually, topology is more profoundly tied to slenderness than it appears from the
results above. Namely, if we consider, for an arbitrary index set I, the group ZI equipped with the
product topology, then we can ask when every homomorphism ZI ! Z is continuous. The answer
is not difficult: this is the case exactly if I is non-measurable. (For measurable I, a non-continuous
map is exhibited in Example 2.9).
This idea is used by Kruchkov [1] who selects a topological group T, and considers the class of
groups G with the property that every group homomorphism  W T ! G is continuous. The choice
T D Z@0 (the Baer-Specker group) with the product topology on discrete components yields the
class
Q of slender groups. The thin p-groups are obtained if T is chosen as the torsion subgroup
of n2N Z.pn / furnished with the product topology. The groups G are characterized for special
choices of T.
Eda has several interesting papers generalizing almost all aspects of slender groups. For
instance, he proved the following generalization of Nunke’s theorem: for an arbitrary cardinal
Q@
, every closed subgroup of  0 Z (consisting of vectors with countable support in the direct
Q@0
product) is isomorphic to  Z for some cardinal .
Fuchs–Göbel [1] also show that the class of slender groups cannot be obtained from any set
of slender groups by taking repeatedly unions of chains of the mentioned kind (including direct
sums) and passing to subgroups.
As already pointed out above, rings with cotorsion-free additive groups are important in
constructing groups with prescribed endomorphism rings. See Sect. 7 in Chapter 16.
4 Separable Groups 501

Exercises

(1) An @1 -free group is slender if and only if it has no subgroup Š P.

(2) (Nunke) Equip P with the product topology where the components hen i carry
the discrete topology. Then: (a) all the endomorphisms of P are continuous; (b)
the topology of P is independent of the way P is represented as a direct product
of infinite cyclic groups.
(3) If G is a dense subgroup of P, then P=G is cotorsion.
(4) A subgroup of P that is closed in the product topology of P is a product of
infinite cyclic groups. [Hint: Proof of Proposition 3.6.]
(5) Suppose X is a product in P. Then P=X is a product of finite and infinite cyclic
groups.
(6) Show that Jp does not contain any subgroup Š P.
(7) Let A be a torsion-free group that contains a slender subgroup G such that A=G
is a reduced torsion group. Show that A is slender. [Hint: Im  is reduced
torsion cotorsion (Proposition 3.6), so bounded;  W P ! A,  W A ! A=G.]
(8) A countable group is cotorsion-free if and only if it is slender.

4 Separable Groups

Our next topic is a class of groups that is more general than completely decompos-
able groups, but behave “locally” as if they were completely decomposable.
A torsion-free group A is called separable (Baer [6]) if every finite subset of
elements of A is contained in a completely decomposable summand of A. Clearly,
this summand may then be assumed to be of finite rank. Recall: if A has a rank 1
summand of type t, then we say t is an extractable type in A.
Example 4.1. The proof of Theorem 8.2 in Chapter 3 shows that the Baer-Specker group Z@0 is
separable, but not completely decomposable. The same holds for the generalized Baer-Specker
group Z for any infinite cardinal .

Properties of Separability Some key properties of separability are listed below.

(A) A group is separable if and only if its reduced part is separable.
(B) Direct sums of separable groups are separable (but their direct products are
not necessarily).
(C) If A is separable, then for every type t, A.t/ and A .t/ are pure, separable
subgroups of A.
(D) If A is separable, and t is any type, then A.t/=A .t/ is separable. This will be
a consequence of the following lemma.
Lemma 4.2.
(i) Fully invariant subgroups of a separable group are separable.
(ii) Factor groups of a separable group modulo fully invariant pure subgroups are
separable.
502 13 Torsion-Free Groups of Infinite Rank

Proof.
(i) Let C be a fully invariant subgroup in the separable group A, and let c1 ; : : : ; ck 2
C. By definition, there is a decomposition A D A1 ˚    ˚ An ˚ A0 where Ai are
of rank 1 and c1 ; : : : ; ck 2 A1 ˚    ˚ An . By full invariance, C D .C \ A1 / ˚
   ˚ .C \ An / ˚ .C \ A0 / where each C \ Ai is 0 or of rank 1, and the cj are
contained in their direct sum.
(ii) Now C is pure, so A=C is torsion-free. Given a1 CC; : : : ; ak CC, there is a direct
decomposition A D A1 ˚    ˚ An ˚ A0 where Ai are of rank 1 and a1 ; : : : ; ak 2
A1 ˚  ˚An . Then A=C D .A1 CC/=C ˚  ˚.An CC/=C ˚.A0 CC/=C where
the first n summands are either 0 or of rank 1, and their direct sum contains the
given cosets mod C. t
u
Main Results on Separable Groups The next result tells us that only the
uncountable separable groups yield something new.
Theorem 4.3 (Baer [6]). A countable separable torsion-free group is completely
decomposable.
Proof. Let A be a countable separable group, and fa1 ; : : : ; an ; : : :g a generating
system of A. There is a finite rank completely decomposable summand G1 of A
that contains a1 . Assume that we have an ascending chain G1
  
Gn of
finite rank completely decomposable summands of A such that a1 ; : : : ; ai 2 Gi for
i D 1; : : : ; n. There is then a completely decomposable summand GnC1 of A which
contains both (a maximal independent set of) Gn and anC1 . The union of the chain
of the Gn .n 2 N/ is evidently A. Setting B1 D G1 and GnC1 D Gn ˚BnC1 for n  1,
we obtain A D ˚1 nD1 Bn . Here Bn is completely decomposable as a summand of the
completely decomposable group GnC1 , see Theorem 3.10 in Chapter 12. t
u
Once we know that summands of completely decomposable torsion-free groups
are completely decomposable (Theorem 3.10 in Chapter 12), it is not difficult to
prove the analogous result for separable groups.
Theorem 4.4 (Fuchs [16]). Summands of separable torsion-free groups are again
separable.
Proof. Let G be a separable torsion-free group, and G D A ˚ B a direct
decomposition. We have to show that every finite subset S of A is contained in a
finite rank completely decomposable summand H of A.
Let G0 be a finite rank completely decomposable summand of G containing S;
there are finite rank pure subgroups A0 ; B0 of A and B, respectively, such that G0

A0 ˚ B0 . There is a finite rank completely decomposable summand G1 of G that

contains a maximal independent set in A0 ˚ B0 , and hence it contains both A0 and
B0 . Furthermore, there are finite rank pure subgroups A1 ; B1 of A and B, respectively,
satisfying G1
A1 ˚ B1 . Continuing this way, we obtain an ascending chain

G0
A 0 ˚ B 0
G1
A 1 ˚ B 1
  
Gn
A n ˚ B n
   .n < !/;
4 Separable Groups 503

where the Gn are finite rank completely decomposable summands of G; and

An ; Bn are finite rank pure subgroups of A; B. The union G0 of this chain is a
countable rank subgroup of G, which is completely decomposable as the union
of the chain of completely decomposable subgroups Gn where every group in the
chain is a summand in each of the following ones with completely decomposable
complements. Moreover, by construction,
[ [
G0 D A 0 ˚ B 0 where A0 D An ; B0 D Bn :
n n

A0 ; B0 are completely decomposable as summands of the completely decomposable

group G0 . Therefore, S is contained in a finite rank completely decomposable
summand H of A0 . Then H is a summand of G0 , and since H
Gk < G0 for some
k < !, H is a summand of Gk , so also of G, and hence of A. As a summand of Gk ,
H is completely decomposable. t
u
The case when the rank 1 summands are of the same type deserves special
attention.
Lemma 4.5 (Baer [6]).
(i) A homogeneous torsion-free group is separable if and only if every finite rank
pure subgroup is a summand.
(ii) Pure subgroups of homogeneous separable groups are separable.
Proof.
(i) Let C be a finite rank pure subgroup in the homogeneous separable group A.
It can be embedded in a finite rank completely decomposable summand B of
A. By homogeneity, Corollary 3.7 in Chapter 12 implies C is a summand of B,
and hence of A. Conversely, if pure subgroups of finite rank are summands of
A, then every finite rank pure subgroup of C is a summand of C, and it follows
at once that C is completely decomposable.
(ii) Let again C be a finite rank pure subgroup in the homogeneous separable group
A. If c1 ; : : : ; cn 2 C, then hc1 ; : : : ; cn i is by (i) a summand of A, and hence
of C. (i) also implies C separable. t
u
Example 4.6 (Nunke). Subgroups Q of homogeneous separable groups that are not pure need not be
separable. As usual, set P D n2N hen i and S D ˚n2N hen i. To see that the group A D pP C S
for any prime p is not separable, we show that the pure subgroup hxi where x D .pe1 ; : : : ; pen ; : : :/
cannot be a summand. If  W A ! hxi were the projection, then ei D 0 for all i > n for some
n 2 N. Then x D x D .pe1 C    C pen / D p.e1 C    C en / would be divisible by p in A,
a contradiction.

Separability of Direct Products The following two results are concerned with
the separability of vector groups (see next section).
Q
Lemma 4.7. For an infinite index set I, let A D i2I Ai where the Ai are of rank 1
and of fixed type t .¤ t.Q//. A is separable if and only if t is idempotent, in which
case A is homogeneous of type t.
504 13 Torsion-Free Groups of Infinite Rank

Proof. Let R be the subring of Q whose additive group is of idempotent type t.

Then A of type t may be viewed as a module over the principal ideal domain R. The
argument in Lemma 4.5 carries over verbatim to provide a proof of the “if” part.
To verify the converse, let t be a non-idempotent type represented by R < Q,
and pj .j < !/ a list of primes with pj R ¤ R. For each j select an ij 2 I and an
aj 2 Aij such that aj is not divisible by pj . The indices ij may be chosen different.
Then consider the infinite vector x D .0; : : : ; aj ; : : : ; 0; : : :/ 2 A, where the ij th
coordinate is aj for every j < !, while all other coordinates are 0. Such an x has
type < t. t
u
The foregoing lemma provides a clue how to construct homogeneous separable
groups of idempotent types t: take any pure subgroup of a direct product of rank
1 groups of type t. If the type is not idempotent, then a somewhat more difficult
procedure is needed, but it yields all homogeneous separable groups.
Proposition 4.8. A homogeneous torsion-free group of type t.¤ t.Q// is separable
if and only if it is isomorphic to a pure subgroup of some
Y 
GD Ai .t/;
i2I

where all groups Ai are rational groups of type t.

Proof. Let A be separable and homogeneous of type t. For every a 2 A, the pure
subgroup hai is a summand of A as is shown in Lemma Q 4.5, so the projections
a W A ! hai give rise to an embedding  W A ! 0¤a2A hai of A in the direct
product of rank one groups of type t. By construction, it must be a pure embedding
with image in the subgroup of elements of type  t.
In order to show that the group G in the theorem is separable, represent t as a
subgroup R of Q, and pick elements ai 2 Ai corresponding to 1 2 R under arbitrary
isomorphisms Ai Š R. All 0 ¤ g 2 G have type t, since > t is impossible. We
write g D .: : : ; ri ai ; : : :/ with ri 2 R. Set ri D si ni .si ; ni 2 Q/ where the numerator
and denominator of si contain only primes p with pR D R, while ni has no such
primes. If g corresponds to 1 in an isomorphism hgi Š R, then the ni have to be
integers. If there is an index j 2 I with nj D ˙1, then clearly G D hgi ˚ Hj
where Hj is G intersected with the product of the Ai with i ¤ j. If there is no such
j, then we proceed as in the proof of Theorem Q 8.2 in Chapter 3, and argue that there
are elements x1 ; : : : ; xk 2 G such that i2I Ai D hx1 i ˚    ˚ hxk i ˚ H 0 , where
g 2 hx1 i ˚    ˚ hxk i and H 0 is the product of almost all of the Ai . We continue
withQan induction on n, to find a finite rank completely decomposable summand X
of i2I Ai containing preassigned elements g1 ; : : : ; gn 2 G. By Lemma 4.5, X \ G
is a completely decomposable summand of G, proving that G is separable. t
u
Separable torsion-free groups are abundant. A rich source of interesting examples
is the theory of groups of continuous functions of a topological space into the
discrete group Z. To make good this claim, we will prove Proposition 4.9 below.
4 Separable Groups 505

In the following result, C.T; Z/ denotes the group of continuous functions from
the topological space T into the discrete group Z. We focus on the case when T is the
set of rational numbers in an open interval of the real line (in the interval topology).
Clearly, for every open interval .a; b/ with irrational endpoints a < b, there is an
isomorphism C..a; b/ \ Q; Z/ Š C.Q; Z/. The same holds if a D 1 or b D 1.
Proposition 4.9 (Eda [2]). The group C.Q; Z/ is separable, and is isomorphic
both to C.Q; Z/@0 and to C.Q; Z/.@0 / .
Proof. Let ı denote an irrational number, and let In D .n C ı; n C 1 C ı/ \ Q, i.e. the
set of rational numbers in the given interval. Then Q D [n2Z In is a disjoint union,
and therefore
Y
C.Q; Z/ D C.In ; Z/ Š C.Q; Z/Z :
n2Z

Let ı0 >    > ıi >    be a decreasing sequence of positive irrational numbers

converging to 0. Define

J0 D..1; ı0 / [ .ı0 ; 1// \ Q;

Ji D..ıi ; ıiC1 / [ .ıiC1 ; ıi // \ Q 8i 2 N:

We claim that the subgroup C0 D ff 2 C.Q; Z/ j f .0/ D 0g is equal to

˚i<! C.J
Qi ; Z/. Indeed, the inclusion  is obvious. For the converse, note that
C0
i<! C.Ji ; Z/, but by continuity, for an f 2 C0 there is a k 2 N such
that f  Ji D 0 for all i > k. Hence C0 D ˚i<! C.Ji ; Z/ Š C.Q; Z/.!/ , and
C.Q; Z/ D Z ˚ C0 Š Z ˚ C.Q; Z/.!/ . We obtain

C.Q; Z/ Š Z ˚ .Z ˚ C.Q; Z//.!/ Š .Z ˚ C.Q; Z//.!/ Š C.Q; Z/.!/ :

The separability of C.Q; Z/ follows immediately from the observation that it is

a pure subgroup of the homogeneous separable group ZQ . t
u
Dual Groups We intend to draw attention to a large variety of separable torsion-
free groups arising from the so-called dual groups. A will denote the group
Hom.A; Z/ which will be called the dual group of A. If we say that G is a dual
group, then we mean that G Š Hom.A; Z/ for some group A.
Lemma 4.10. A dual group is torsion-free, homogeneous, and separable. Its type
is t.Z/.
Proof. It is pretty obvious that every dual group is a subgroup in a direct product of
infinite cyclic groups. t
u
Example 4.11. For non-measurable cardinality, free groups are dual groups, and so are the direct
products of infinite cyclic groups.
Example 4.12. There are in general many groups with the same dual. For instance, Z.@0 / is the
dual of Z@0 , and also the dual of the Z-adic closure of Z.@0 / in Z@0 .
506 13 Torsion-Free Groups of Infinite Rank

If G is a dual group, then so are G and G./ for any non-measurable cardinal
. The Reid class of dual groups is generated by Z, using Q alternately
Q directQsum
andQnon-measurable direct product constructions: ˚ Z; Z; .˚ Z/; ˚. Z/;
˚. .˚Z//, etc. It was an interesting challenging question raised by G. Reid
whether the groups in this class are all different. Now it seems clear that they are
indeed different. Dual groups that do not belong to this class have been subjects of
extensive research; see Eklof–Mekler [EM].
@1 -Separability There is an immediate generalization of separability to higher
cardinals . A group A is called -separable if every subset of A whose cardinality
is <  is contained in a completely decomposable summand of A. Thus @0 -
separability is just ordinary separability. Our focus is on @1 -separability.
Example 4.13. An @1 -separable torsion-free group A of cardinality @1 has a filtration
fA j  < !1 g such that AC1 is a countable completely decomposable summand of A for each
 < !1 . Using this description, one can construct @1 -separable groups that are not completely
decomposable.

For a pure subgroup X of a t-homogeneous torsion-free group A we use the ad

hoc notation X  D \WA!R; XD0 Ker  where R is a rational group of type t. Observe
that X
X  , and X  is pure in A.
Lemma 4.14. A homogeneous torsion-free group A is @1 -separable if and only
if for every countable subgroup X, X  is a countable completely decomposable
summand of A.
Proof. Suppose A is t-homogeneous, @1 -separable, and X is a countable subgroup.
By definition, A has a countable completely decomposable summand E such that
X
E. It is obvious that X
X 
E. Evidently, E=X  is a countable pure subgroup
of the t-homogeneous separable group A=X  , thus it is completely decomposable of
type t. By Baer’s lemma, X  is a summand of E, and therefore X  is a completely
decomposable summand of A. The converse implication is trivial. t
u
Proposition 4.15. Summands of homogeneous @1 -separable torsion-free groups
are again @1 -separable.
Proof. If A D B ˚ C is such a group, and if X is a countable subgroup of B, then
obviously X 
B. Hence X  is a summand of B. t
u
Coseparable Groups A group A will be called -coseparable if every pure
subgroup of corank <  contains a summand of A whose complement is completely
decomposable of rank < . Our main interest lies in the  D @1 case.
Lemma 4.16. A torsion-free group all of whose pure subgroups are @1 -separable
is @1 -coseparable.
Proof. Assume that the pure subgroups of A are @1 -separable. Let B be pure in A
such that A=B is of countable rank, so A D B C C for a countable C. B is @1 -
separable, thus B D G ˚ H where B \ C
G, and G is completely decomposable of
countable rank. The subgroup C C G is countable, and clearly A D .C C G/ C H. If
4 Separable Groups 507

c C g D h 2 H with c 2 C; g 2 G, then c D h  g 2 C \ B
G, so h 2 G \ H D 0:
This proves A D .C C G/ ˚ H: Here H
B, and the countable summand C C G
must be completely decomposable, since A is @1 -separable. t
u
Theorem 4.17 (Griffith [3]). A group A satisfies Ext.A; S/ D 0 for a countable
free group S if and only if it is @1 -free and @1 -coseparable.
Proof. Suppose A satisfies Ext.A; S/ D 0. Every countable subgroup X of A also
satisfies Ext.X; S/ D 0, and therefore the free presentation 0 ! S ! F ! X ! 0
of X with jFj D @0 splits. This means X is free, and the @1 -freeness of A follows. To
verify @1 -coseparability, let A be uncountable, and B a pure subgroup of countable
index in A. Then there is a countable (and thus free) subgroup H
A such that
A D B C H. Form the exact sequence 0 ! C ! B ˚ H ! A ! 0, where B ˚ H
denotes the outer direct sum and C D f.c; c/ j c 2 B \ Hg. C is countable and
free, so C Š S, and hypothesis implies splitting: B ˚ H D A ˚ C0 with C0 Š C.
Let  W A ! H be the obvious projection; then, Im  is free, thus A D Ker  ˚ A0
with a countable free A0 . Since obviously Ker 
B, we can conclude that A is
@1 -coseparable.
˛
Conversely, let A be @1 -free and @1 -coseparable. Let e W 0 ! S ! H !A ! 0
represent an element of Ext.A; S/; we may view S
H. Let B be a subgroup of
H maximal with respect to B \ S D 0. Then B is of countable index in H, so
jA=˛Bj
@0 , and hence there is a summand G
˛B such that A D G ˚ F with a
countable (free) F. Let C D ˛ 1 G \ B; D D ˛ 1 F. The exact sequence 0 ! S !
˛
D!F ! 0 splits, so D D S ˚ F 0 where F 0 Š F. We have .C C F 0 / \ S D 0, since
cCx D s .c 2 C; x 2 F 0 ; s 2 S/ implies c D sx 2 C\D D C\.C\D/ D C\S

B \ S D 0; then also s D x D 0. From ˛.C C F 0 / D .G \ ˛B/ C F D G C F D A

we obtain H D .C C F 0 / C S. Therefore, S is a summand of H, i.e. the sequence e
splits, and Ext.A; S/ D 0. t
u
F Notes. A word of caution: in some publications (especially where only separable groups of
the type of Z are discussed) “separability” means that every finite set of elements is embeddable in
a free summand of finite rank.
Cornelius [1] has an interesting observation: for the separability of A, it is enough to know that
every single element of A is contained in a finite rank completely decomposable summand of A.
This is in fact an unexpected result.
That summands of separable groups are again separable has several proofs, and a number of
generalizations. We mention here that Rangaswamy [4], Hill–Megibben [7] introduce far-reaching
generalizations of separability, proving various theorems similar to those discussed above. Their
results show how more complex group structures can be approached by using the right concepts.
A number of remarkable theorems on @1 -separable groups (@1 -free case) can be found in Eklof–
Mekler [EM]; they also discuss filtration equivalence which seems to be quite a significant concept.
Dugas–Irwin [1] investigate @1 -separable subgroups in the Baer–Specker group Z@0 , and show
inter alia that it is undecidable in ZFC if every @1 -separable @1 -free group of cardinality @1 is
embeddable in Z@0 . In the proof, the same set-theoretical models are used as we will in Sects. 6
and 7. Blass–Irwin [1] have interesting results on subgroups of the Baer-Specker group Z@0 .
It was conjectured that every reflexive group is of the form Hom.C; Z/ for some C, but was
disproved by Eda–Ohta [1]. Albrecht [4] discusses A-reflexivity when A is a non-measurable
slender group.
508 13 Torsion-Free Groups of Infinite Rank

Rychkov [4] considers families fAi j i < 2 g of -separable torsion-free groups for an
uncountable regular cardinal  that is not weakly compact. Assuming V = L, he establishes the
existence of such families satisfying j Im j <  for each homomorphism  W Ai ! Aj with
i ¤ j in I. Grinshpon–Krylov [1] study the lattice of fully invariant subgroups of separable
torsion-free groups. This lattice—as expected—is determined by the extractable types of the
group. Grinshpon–Krylov [1] describe the fully transitive separable groups. For the transitivity
of completely decomposable and separable groups, see also Metelli [2].
Hill–Megibben [7] generalize the notion of separability by defining K-groups. The class
of these groups is closed under direct sums and direct summands, they display several useful
properties similar to separable groups. By making use of this concept, knice subgroups are
introduced, and torsion-free groups are studied that admit H.@0 /-families of knice subgroups.
There is a considerable body of literature on dual groups. Their in-depth discussion requires
sophisticated set-theoretical machinery; we refer to [EM] for many interesting results. See also
Mekler–Schlitt [1] where dual groups are discussed from the logical point of view. Göbel–Pokutta
[1] construct dual groups in Z@0 using MA+ :CH. It was for a while an open problem if the
groups in the Reid classes were all different. The problem was solved by Zimmermann-Huisgen
[1], Ivanov [3], and Eda [3] in the affirmative.
Nunke–Rotman [1] show that every integral singular cohomology group is of the form
Ext.G; Z/ ˚ Hom.H; Z/, where G; H can be arbitrary groups. Thus they are direct sums of a
cotorsion group and a dual group.
Coseparability is more difficult to deal with, and the reader might wonder why we have not
given any explicit examples for non-free coseparable groups. The reason is simple: it is undecidable
in ZFC whether or not all coseparable groups are free. We do not wish to enter into the discussion of
these groups, just mention that Chase [Pac. J. Math. 12, 847–854 (1962)] proved that CH implies
that there exist non-free coseparable groups of cardinality @1 , while Mekler–Shelah [3] showed
that coseparable groups of cardinality @1 are free in the model of ZFC with @2 Cohen reals added.
It is an interesting fact that all @1 -coseparable groups are reflexive (Huber [2]).
Coseparability can also be defined to depend on a cardinality. Usually the definition is restricted
to -free groups. Accordingly, A is -coseparable if it is  C @1 -free and every pure subgroup of
corank <  contains a summand of A of corank < .

Exercises

(1) (Baer) A torsion-free group A is separable if and only if

(i) every finite subset of A belongs to a direct summand that is the direct sum
of homogeneous groups; and
(ii) A.t/=A .t/ is separable (in the usual sense) for every type t. [Hint:
summands in (i) are separable.]
(2) For every group A, the group Hom.A; Z/ is separable.
(3) The tensor product of two separable groups is separable.
(4) A homogeneous separable torsion-free group A may have fully invariant
subgroups other than nA .n D 0; 1; 2; : : :/.
(5) Fully invariant subgroups of @1 -separable torsion-free groups are likewise @1 -
separable.
(6) (Sa̧siada) There are non-isomorphic dual groups that are isomorphic to pure
subgroups of each other. [Hint: Z@0 and Z.@0 / ˚ Z@0 .]
5 Vector Groups 509

(7) Show that there are dual groups of large non-measurable cardinalities that are
isomorphic to their own duals. [Hint: A ˚ A .]
(8) (Metelli)
(a) Summands as well as direct sums of @0 -coseparable groups are again @0 -
coseparable.
(b) If A is @0 -coseparable, and C is a pure fully invariant subgroup of A, then
A=C is likewise @0 -coseparable.
(9) (Metelli) Call a torsion-free group A t-bihomogeneous if A.t/ D A and
AŒt
D 0. Show that
(a) A is t-bihomogeneous if and only if it is a t-homogeneous subgroup of a
t-homogeneous separable group;
(b) countable bihomogeneous groups are completely decomposable.
(10) The group Hom.G; Z/ need not be coseparable even if G is coseparable.
(11) (Griffith) Generalize Theorem 4.17 as follows. Let F be a free group of infinite
rank . A group G satisfies Ext.G; F/ D 0 if and only if (i) it is  C -free; and
(ii) every subgroup of G of index
 contains a summand of G of index
.
[Hint: same proof.]

5 Vector Groups

An important class of torsion-free groups is the class of vector groups. So far

insufficient attention has been paid to it, though there are a number of noteworthy
features which are worthwhile exploring. As a note of caution, we should point
out at the outset that one has to be very careful with the statements, since some
results depend on non-measurability, while some others hold without any cardinal
restrictions.
By a vectorQgroup we mean a direct product of torsion-free groups of rank 1, i.e.
a group V D i2I Ri where the Ri are rational groups and I is an arbitrary index set.
(Usually, we assume Ri 6Š Q.)

Q Basic Results on Vector Groups First, we will consider vector groups V D

R
i2I i such that all the components R i have the same type t; that is, V Š R I
where
R is a rational group of type t. Such a group V will be called an elementary vector
group of type t.
Elementary vector groups need not be homogeneous; in fact,
Q
Lemma 5.1. An elementary vector group V D i2I Ri of type t is homogeneous if
and only if t is an idempotent type.
Proof. This follows from Lemma 4.7. t
u

Example 5.2. The generalized Baer-Specker group Z for any infinite cardinal  is a homogeneous
vector group. Its type is t.Z/.
510 13 Torsion-Free Groups of Infinite Rank

Example 5.3. Let R denote the group of rational numbers with square-free denominators.
The group R@0 is an elementary vector group, but it is not homogeneous: the vector
.21 ; : : : ; p1 : : : :/ 2 R@0 has characteristic .0; : : : ; 0; : : :/.
Q
Let V D i2I Ri be an arbitrary vector group where the groups Ri are torsion-
free of rank 1. For a fixed type t, let Vt denote the direct product of those Ri which
are exactly of type t. Thus
Y
VD Vt
t

with elementary vector group components, t running over all types.

In the study of vector groups where the rank 1 components are of distinct types,
it is essential to get information about their rank 1 epic images, in particular,
about the extractable types. The next lemma (valid for all cardinals) gives valuable
information about this situation.
Lemma 5.4.
Q
(a) If the vector group V D i2I Ri has a non-trivial homomorphism into a rational
group R ¤ 0, then

t.Ri /
t.R/ for some index i:
Q
(b) (Mishina [1], Eda [1]) A rank one summand of the vector group V D i2I Ri is
isomorphic to some Ri .
Proof.
(a) If R D Q; there is nothing to prove. So assume R is slender, and let  W V ! R
be a non-zero homomorphism. Let v 2 V be such that v ¤ 0, and write
v D .: : : ; vi ; : : :/ with vi 2 Ri . Using this v, we collect
Q all the components Ri for
which the characteristic .vi /QD  is fixed: V D .vi /D Ri . In this way, we
obtain a decomposition V D  V . The last product has at most continuously
many non-zero components, so V is a product over a non-measurable index
set. By the slenderness of R, there are but finitely many V , say of indices
1 ; : : : ; n , whose images under  are non-zero, while the product of the rest
of the V is mapped by  to 0. One of the coordinates w1 ; : : : ; wn of v in
the last decomposition of V has a non-zero image under , since evidently
v D w1 C    C wn . From .wi /  .wi / we conclude that the type of
R is  the type of one of w1 ; : : : ; wn which is the type of some Ri .
(b) Without loss of generality, we may assume that V is reduced. From the
arguments in (a) it is clear that a rank 1 summand R of V must be a summand of
some V1 ˚    ˚ Vn . Assume n is chosen minimal, and let t1 ; : : : ; tn denote the
corresponding (not necessarily different) types. Then by (a) we have t.R/  tj
for every j 2 f1; : : : ; kg. On the other hand, R projects non-trivially into a
component in Vj , so t.R/
tj for all j. Hence t.R/ D tj for some j. t
u
Assertion (b) can
Q be rephrased by saying that the only extractable types of a
vector group V D i2I Ri are the types t.Ri / .i 2 I/.
5 Vector Groups 511

Isomorphism of Vector Groups At this point the natural question is: when are
two vector groups isomorphic? To answer this question, we prove the following
theorem. We denote by t0 the largest idempotent type with t0
t, i.e. t0 D t W t.
Theorem 5.5. Suppose
Y Y
VD Vt and W D Wt
t t

are vector groups where Vt and Wt are elementary vector groups of type t, and t
runs over the different types.
(i) (Sa̧siadaQ[3]) V Š W if and
Q only if Vt Š Wt for every type t.
(ii) If Vt D i2I Ri Š Wt D j2J Rj with a type t 6Š t.Q/, then jIj D jJj whenever
I or J is non-measurable.
Proof.
(i) The “if” part being obvious, assume there is an isomorphism  W V ! W. Let
t W V ! Vt and t W W ! Wt denote the coordinate projections in the two
direct products. By Lemma 5.4, every homomorphism Q of Vt into Ws is trivial
unless the types satisfy s  t. Therefore, Q V t
s
t Ws . If v 2 Vt , then
v D w C w0 with w 2 Wt and w0 2 s>t Ws : Again by Lemma 5.4, the
last group has only the trivial homomorphisms into Vt and into Wt , whence
t  1 w0 D 0 D t w0 follows. This shows that v D t v D t  1 w and w D
t w D t v; consequently, the map t  1 t  is the identity on Vt . Changing
the roles of Vt and Wt , the stated isomorphism is immediate.
(ii) Let R and R0 be rational groups of types t and t0 , respectively. In view
of Corollary 2.10, we have Hom.Vt ; R/ Š ˚i2I Hom.Ri ; R/ D ˚i2I R0 and
Hom.Wt ; R/ Š ˚j2J R0 . Both are completely decomposable groups whence
Vt Š Wt implies jIj D jJj. ((ii) holds also for measurable index sets,
see Theorem 6.1 below.) t
u
Summands of Non-measurable Vector Groups One of the most interesting
facts on vector groups is concerned with direct summands in the non-measurable
case.
Theorem 5.6 (O’Neill [2]).
(i) Summands of reduced vector groups are direct products of summands of
elementary vector groups of different types.
(ii) (Balcerzyk–Bialynicki-Birula–Łoś [1]) Summands of reduced non-measurable
vector groups are again vector groups.
Proof.
Q
(i) Let V D t Vt with elementary vector groups Vt of type t. The projections in
this decomposition will be denoted by t . For each type t, the subgroups
Y 
Y
Vt D Vs and V t D Vs
s
t s>t


are fully invariant, and satisfy V t D Vt ˚ V t .
512 13 Torsion-Free Groups of Infinite Rank

Suppose V D A ˚ B with projections ˛; ˇ. In view of full invariance, for

  
each type t, we have V t D At ˚ Bt and V t D At ˚ Bt where At D A \ V t ,
   
At D A \ V t (and the same for B). Hence At D At ˚ At and Bt D Bt ˚ Bt ;
 
where At D At \ .Vt ˚ Bt / and Bt D Bt \Q .Vt ˚ At Q /: Manifestly,
Q we have
Vt D t At ˚ t Bt ; so we can write V D t Vt D t t At ˚ t t Bt : If

a D v C b 2 At with v 2 Vt and b 2 Bt ; then t a D v and a D ˛v imply that
˛t a D a; i.e. ˛t is the identity on At . Q
Let 2 End V denote the projection of V onto A0 D t t At . Consider the
map  ˛ W V ! V. From what has been proved it follows that this map
vanishes on each t At , thus also on their direct sum. The direct product of the
t At is over at most 2@0 summands, thus it is over a non-measurable index set.
Since the Ri are slender, the projection of the direct product A0 into each Ri is
also 0,Qi.e. .  ˛ /A0 D 0. Therefore, A0 D A0 D ˛ A0
A. Similarly,
B0 D t t Bt
B. Evidently, A0 ; B0 cannot be proper subgroups Qin A; B, since
A ˚ B D V D A0 ˚ B0 . Hence we derive the isomorphism A Š t t At .
(ii) For the second claim, Q we first show that every summand of an elementary
vector group V D i2I Ri of type t is again a vector group, provided the
index set I is non-measurable, and t is different from the type of Q. Let V D
A ˚ B, and R; R0 rational groups of types t, t0 , respectively. By the slenderness
of R, we have Hom.V; R/ Š ˚i2I Hom.Ri ; R/ Š ˚i2I R0 . Hence V Š
Hom.Hom.V; R/; R/. Observe that Hom.A; R/ and Hom.B; R/—as summands
of a completely decomposable group—are likewise completely decomposable.
They are complementary summands in Hom.V; R/, so Hom.Hom.A; R/; R/;
and Hom.Hom.B; R/; R/ are products of copies of R and complementary
summands in Hom.Hom.V; R/; R/. By the naturality of the isomorphisms, these
groups are isomorphic to A; B and V, respectively. Thus for non-measurable
vector groups, in the decomposition
Q used in the proof of (i), each t At is a
vector group. Hence A Š t t At is likewise a vector group. t
u
Separability of Vector Groups We do not intend to pursue the exploration
of vector groups much further, but we feel we should mention the main result on
their separability. Separability of vector groups in general has been successfully
investigated
Q by Mishina [2] and Król [1]. According to their result, a vector group
V D i2I Ri is separable if and only if the types t.Ri / D ti satisfy the following
criteria:
(i) the descending chain condition holds for the types ti ;
(ii) every set of incomparable types ti is finite; and
(iii) if in some type tj , non-zero finite heights occur at an infinite set N of primes,
then only finitely many types ti exists which are  tj and contain infinitely
many non-zero finite heights at infinitely many primes in N.
F Notes. Though in this section we have concentrated on the non-measurable case, we tried
to prove the results, if possible, to include measurable vector groups as well.
6 Powers of Z of Measurable Cardinalities 513

Mishina [1] proves that every slender summand of a vector group is a finite direct sum of its
summands. As pointed out above, separable vector groups have been characterized by Mishina [2]
and Król [1] independently. (I wish to apologize for quoting their result incorrectly in [IAG].) See
also Król–Sa̧siada [1]. Albrecht–Hill [2] provide a better proof.
Huber [1] investigates homomorphisms  W RI ! RJ for rational groups R and non-measurable
index sets I; J. The kernel and cokernels are characterized: they are of the form Ker  D Hom.A; R/
and Coker  D RK ˚ Ext.A; R/ for some group A and index set K. Ivanov [7] shows that
Kaplansky’s test problems have positive solutions for non-measurable vector groups.

Exercises

(1) For non-measurable I, derive Lemma 5.4(b) directly from Corollary 2.10,
without referring to (a).
(2) (Mishina, Łoś) A vector group over an infinite index set is completely decom-
posable if and only if almost all components are Š Q.
(3) A reduced vector group contains no cotorsion subgroups ¤ 0.
(4) Let R be a rational group of idempotent type, and  a non-measurable cardinal.
There exists a group A ofQcardinality   such that Hom.A; R/ Š A.
(5) (Beaumont) Let V D i2I Ri be an elementary vector group of type t with
infinite I. Show that V contains elements of any type s satisfying t0
s
t.
(Here, as above, t0 D t W t.)
(6) Prove that .Z@0 /n Š Z@0 for each n 2 N, but .Z@0 /.@0 / 6Š Z@0 .
(7) (Ivanov) Two non-measurable vector groups are isomorphic if each is isomor-
phic to a summand of the other.

6 Powers of Z of Measurable Cardinalities

The discussion of vector groups above suggests that there is a fundamental

difference between the behavior of vector groups of measurable and non-measurable
cardinalities. Our study would be incomplete without mentioning some results that
are critical in understanding why non-measurability was so relevant an assumption
in the above theorems. Already the measurable powers ZI display phenomena whose
understanding requires heavy set-theoretical machinery. To study ZI fully in detail
would take us too far afield, so we restrict ourselves to pointing out some typical
results.
In this section, we are working in a model of set theory in which measurable
cardinals exist. We will use the symbol  for the smallest measurable cardinal,
and the notation P for the th power Z of Z which will also be written more
explicitly as
Y
P D hei i where hei i Š Z
i<
514 13 Torsion-Free Groups of Infinite Rank

with projection maps i W P ! hei i.P As usual, we set S D ˚i< hei i. An element
of P will be written in the form x D i< ni ei with ni 2 Z.
Let U be a (non-principal) @1 -complete ultrafilter on the power set P.I/ of an
index set I of cardinality . The homomorphism U W P ! Z is defined as follows.
For x 2 P , let In D fi 2 I j i .x/ D nei g for n 2 Z, so that fIn j n 2 Zg is a
partition of I. If m is the unique integer such that Im 2 U, then we set U .x/ D m: It
is straightforward to check that U is a genuine homomorphism.
It is known that if the cardinal  is measurable, then there are at least 2 -
complete ultrafilters on .
Uniqueness of  in P If we consider powers Z , then probably one of the
first questions that comes to mind is to what extent  is determined. We verify the
uniqueness of this cardinality in a more general setting.
Theorem 6.1 (Eda [1]). Let G ¤ 0 denote a slender group. GI Š GJ holds for
infinite sets I; J if and only if jIj D jJj.
Proof. To verify the “only if” part, let  W GI ! GJ be an isomorphism, and j W
GJ ! G the jth projection map. Pick a 0 ¤ g 2 G, and for each i 2 I, let gi 2 GI
be the vector with 0 coordinates everywhere except with g as its ith coordinate. As
G is slender, for any fixed j 2 J, the set of i 2 I satisfying j ..gi // ¤ 0 must be
finite. Consequently, jIj
jJj, and by symmetry, we obtain the desired equality. u t
Odd Decompositions of Z In order to illustrate how drastically the measurable
case differs from the non-measurable situation, we exhibit examples showing that
for a measurable cardinal , P admits non-trivial direct decompositions where S
is contained in one of the summands.
Example 6.2 (O’Neill [3]).
(a) Let  be a measurable cardinal, and let U denote a non-principal @1 -complete ultrafilter on
the power set P ./. Then P D A ˚ hei, where A D fx 2 P j supp x … U g, and e 2 P is
the vector whose ith coordinate is ei for each i < . All the basis vectors ei belong to A, so
the projection  of P onto hei satisfies S D 0, though P ¤ 0.
(b) More generally, let S D fUi j i < g be a set of non-principal @1 -complete ultrafilters on P ./
such that, for each i < , there is an Xi  satisfying: XiP
2 Uj exactly if i D j. (The existence
of such a set S is established in set theory.) Setting xj D i2Xj ei , we have a decomposition
Y
P D B ˚ hxj i where B D fx 2 P j supp x … [i< Ui g:
j<

Summable Subsets A subset X of P is said to be summable (relative to the

given decomposition of P ) if, for each i < , there are but a finite number of
elements x 2 XPsuch that i .x/ ¤ 0. The subset X will be called completely
independent if x2X nx x D 0 .nx 2 Z/ implies nx D 0 for all x 2 X (O’Neill).
Example 6.3.
(a) For each i < , let xi D ei C    C eiCn 2 P with a fixed integer n. Then fxi j i < g is a
completely independent summable
P subset in P .
(b) The set fei .i < /I e D i< ei g is summable, independent, but not completely independent.
6 Powers of Z of Measurable Cardinalities 515

A few properties of summable sets are recorded in (A)–(E).

(A) Subsets of summable sets are summable. An infinite subset of P is summable
if and only if its countable subsets are summable. P
(B) If X is a summable subset in P , then any infinite linear combination x2X nx x
with nx 2 Z represents an element of P . Q
(C) If X is a completely independent summable set, then x2X hxi is a subgroup
of P .
(D) Let X D fxj j j 2 Jg be a completely independent summable subset in P D
A ˚ B. The non-zero projections of elements of X in A form a summable subset.
Indeed, if ˛ W P ! A is the projection map, then by (C), for each i 2 I, only
finitely many indices j 2 J satisfy i ˛.xj / ¤ 0.
(E) If P D A ˚ B, and X D fxj j j 2 Jg is an infinite summable subset in A, then
thereQis a completely independent summable subset Y  A such that jYj D jXj
and y2Y hyi is a summand of P . For the proof we may assume that each hxj i
is pure (and hence a summand) in A. For every x 2 X, there is a finite subset
Sx   such that x D ux C vx where hux i is a summand of ˚i2Sx hei i and
i .vx / D 0 for all i 2 Sx . Thus
Y Y
P D hux i ˚ Cx ˚ hei i D hxi ˚ Cx ˚ hei i
i2nSx i2nSx

for some finitely generated Cx

˚i2Sx hei i. Consider a maximal set M of
pairwise disjoint finite sets Sx with x 2 X. Since X was infinite, we have
jMj D jXj. It is clear that the set Y of indices x for all Sx 2 M will be a
completely independent summable subset in A. This yields a decomposition
Y Y Y
P D hyi ˚ Cy ˚ hei i;
y2Y y2Y i2nS

Q
where S D [y2Y Sy . (Note that it is not claimed that y2Y hyi
A:)
Measurable Vector Groups So far, it has not been proved that summands
of measurable vector groups are again of the same sort. The difficulty lies in
extending Theorem 5.6 to elementary vector groups of size > . However, we will
show that one of the summands is a vector group.
We require some preliminary lemmas. Let  be an endomorphism of P such that

.ei / D ei for each i < : (13.5)

By the Łoś-Eda Theorem 2.14, for each j < , there exists a finite set Fj of @1 -
Q ultrafilters on  such that if S   and S … U for any U 2 Fj , then
complete
j . i2S hei i/ D 0.
In the following arguments we have to deal separately with the ultrafilters that
contain subsets of smaller cardinalities. For each j < , let Nj denote the set of those
U 2 Fj that contain also subsets of cardinality < , and let Mj D Fj n Nj . In each
U 2 Nj , we choose an element (i.e., a subset of ) of minimal cardinality, and denote
516 13 Torsion-Free Groups of Infinite Rank

by Xj the union of these minimal subsets. Visibly, j 2 Xj , jXj j < , and if S   is

disjoint from Xj , then S … U for each U 2 Nj . Finally, we let M D [j2 Mj . As every
Mj is finite, jMj
 is manifest.
We claim that there exists a T   such that jTj D , and T … U for any U 2 M.
By Lemma 5.7 in Chapter 1, there exists a set C of subsets of  such that jCj > ,
jXj D  for each X 2 C, and jX \ Yj <  whenever X ¤ Y in C. If X ¤ Y are
in C, then X and Y cannot both be in the same U 2 M, since then X \ Y 2 U, in
contradiction to the definition of M. From jMj < jCj it is clear that there is a T 2 C
not contained in any member of M.
The following lemma holds for all infinite cardinals, but we prove it only for
regular cardinals.
Lemma 6.4 (O’Neill [4]). Let  be a regular cardinal. If  2 End P satis-
fies (13.5), then there is a subset K of  of cardinality  such that
Y
k . hei i/ D 0 for each k 2 K:
k<i2K

Proof. We keep the notation used above. Suppose j < , and for each ordinal i < j
we have already chosen an element .i/ 2 T n [`<i X` . Since jX` j <  and  is
regular, we have jT n [`<j X` j D . Pick any element .j/ in this set, and define
K D f.i/ 2 T j i < g; thus, jKj D . For simplicity, rename .i/ <  as i < .
Suppose now that j 2 K and Sj D fi 2 K j i > jg. Evidently, Sj  T; so Sj is not
contained in any U 2 Mj . In view of the
Q choice of the elements .i/, Sj is not in any
U 2 Fj D Mj [ Nj either. Hence k . k<i2K hei i/ D 0 for k 2 K. t
u
Lemma 6.5 (O’Neill [4]). Let  be an endomorphism of P such that (13.5)
holds. If
Y
j . hei i/ D 0 for every j < ;
i>j

then  is an isomorphism.
P
Proof. First,  is injective. For, if .x/ D 0, but x D i<Qni ei ¤ 0, then let j be
the first index with nj ¤ 0. Now j .x/ D j .nj ej / C j . i>j hei i/ D nj ej ¤ 0; a
contradiction. P
To see that  is surjective, pick x D i< ni ei 2 P . Let j < Q
, and assume that
for all k <Qj we have found mk 2 Z such that for all i < j, i . ki mk ek / D ni ei .
NowQj . k<j mk ek / D tj ej for some tj 2 Z. Setting mj D nj  tj , we have
j . P kj mk ek / D nj ej . In this way, we can find integers mi for allPi < . Now
y DP i< mi ei 2 P satisfies .y/ D x, since j .y/ D j . kj mk ek / C
j . i>j mi ei / D nj ej . t
u
6 Powers of Z of Measurable Cardinalities 517

Lemma 6.6 (O’Neill [4]). Let P D A ˚ B for an infinite cardinal . Then P has
a decomposition
Y
P D hcj i ˚ E;
j2J

where cj ¤ 0, jJj D , and either A or B contains all the cj .

Proof. Let ˛ W P ! A be the natural projection, and ˛i D i ˛. Let J denote
either the set fi <  j ˛.ei / ¤ 0g or the set fi <  j .1  ˛/.ei / ¤ 0g whichever
has cardinality ; say, J is the first set. As hei i is a slender group, the subset
f˛.ei / j i 2 Jg is summable (see (D)). An appeal to (E) completes the proof. t
u
We are now prepared to prove the two main results of this section.
Theorem 6.7 (O’Neill [3]). Suppose  is an infinite cardinal, and Z D A ˚ B.
Then either

A Š Z or B Š Z :
Q
Proof. In view of Lemma 6.6, we may assume P D j2J haj i ˚ E whereQjJj D 
and aj 2 A for all j 2 J. Let j W P ! haj i; ˛ W P ! A and W P ! Q j2J haj i
denote
Q the natural projections. Consider the composite map  D ˛ W j2J haj i !
j2J haj i.
First, let  be a regular cardinal. Q In view of Lemma 6.4, there is a subset
K
Q  J of cardinality
Q  such
Q that  j . j<k2K hak i/ D 0 for all j 2 K. Next write
j2J haj i D k2K hak i ˚ j2JnK ha Qj i, and denote by  the projection to the first
summand.QLemma 6.5 (applied Q to Q k2K ha k i Š P ) tells us that the composite
map  W k2K hak i ! j2J haj i ! k2K Q k i is an isomorphism. In view of this
ha
isomorphism, we can conclude that C D k2K hak i is a summand: A D C˚C0 . Thus
A has a summand C Š P . The rest is easy: A Š P ˚ C0 Š .C ˚ C0 ˚ B/@0 ˚ C0 Š
.C ˚ C0 ˚ B/@0 Š .A ˚ B/@0 Š P .
If  is singular, then the arguments in the preceding paragraph apply to every
regular cardinal  < . Thus, for every such , selecting a subset of , A contains
a summand C Š Z . If we carefully check the maps involved, we find that they
are independent of the choice of the subsets, so they can be combined to obtain a
summand C Š Z , as before. The conclusion is the same: A Š Z . t
u
Theorem 6.8 (O’Neill [4]). For the first measurable cardinal , every summand of
Z is again a power of Z.
Proof. From Theorem 6.7 we know that if P D A ˚ B (with projections ˛; ˇ), then
one of the components, say A, is isomorphic to P . First, suppose jBj < . Define
an equivalence relation on the index set by declaring i and j equivalent if and only
if i .b/ D j .b/ for each b 2 B. This gives a partition of the index P < 
Q set into
equivalence classes Xj .j 2 J/, and B is evidently a summand of j2J h i2Xj ei i
which is a non-measurable vector group. Therefore, by Theorem 5.6, B is a vector
group.
518 13 Torsion-Free Groups of Infinite Rank

If jBj D , then fˇei j i < g  B is a summable set of cardinality . Application

of (E) yields a completely independent summable set of the same cardinality in B,
and the arguments in the proof of Theorem 6.7 for B (rather than for A) complete
the proof. t
u
Subgroup in Large Product The final topic in this section is concerned with
relatively small subgroups in ZI for arbitrary index set I.
Proposition 6.9 (O’Neill [5]). Let A denote a subgroup of cardinality  in the
product ZI . Then A is contained in a subgroup C of ZI such that jCj
2 .
Proof. Assume  < jIj, and let a D .: : : ; ai ; : : :/ 2 A. Define i; j 2 I equivalent if
ai D aj for all a 2 A. This gives rise to a partition … of I where i; j belong to the same
equivalence class u 2 … if they are equivalent. Each u defines a homomorphism
u W A ! Z by u .a/ D ai if i 2 u. Clearly, j…j
jZ j D 2 . Let eu 2 ZI denote
the vector whose ith coordinate is 1 or 0 according as i 2 u or not. Let J be a subset
Q one representative from each u. Then Z D C ˚ Z where the
I InJ
of I that contains
subgroup C D u2… heu i contains A. t
u
Q F Notes. Eda [1] extends Lemma 5.4 by showing that every slender summand of a product
i2I Ai of non-measurable torsion-free groups Ai is isomorphic to a summand of a direct sum of
finitely many Ai .
Fink [1] calls a set F D fUi j i 2 Ig of @1 -complete ultrafilters on the power set P./ completely
independent if there is a partition  D [i2I Xi such that Xi 2 Ui n [j¤i Uj . He shows that if
Z D A ˚ B, and if the set F of ultrafilters associated with the natural projection ˛ W Z ! A is
completely independent, then A Š ZjFj . The results on Z can be modified and extended to powers
of any rational group < Q. Another result of interest is due to Huber [1]: assuming V = L, for a
rational group R < Q, a subgroup A of a power RI is a summand if and only if RI =A is likewise a
power of R.

Exercises

(1) A summable subset generates a free subgroup.

(2) A subset of P that is summable relative to a decomposition of P is also
summable relative to any other decomposition. P
(3) (a) Every element x 2 P can be written in the formPx D n2Z neXn , where the
Xn .n 2 Z/ are disjoint subsets of I, and eXn D i2Xn ei ;
(b) for an @1 -complete ultrafilter U on the power set P./, there is exactly one
index m 2 Z such that Xm 2 U.
(4) Let A be a summand of P ,  any infinite cardinal. Then A ˚ P Š P :
(5) If G ¤ 0 is a slender group, then GI is isomorphic to a subgroup of GJ if and
only if jIj
jJj.
(6) (O’Neill)
P Let X D fxj j j 2 Jg  P be a summable subset. Write xj D
neYjn where fYjn j n 2 Zg is a partition of I for each j 2 J, and
n2Z P
eK D k2K ek for K  I. Then feYjn j j 2 J; n 2 Zg is a summable subset.
7 Whitehead Groups If V = L 519

(7) (a) (O’Neill) A subgroup of P of cardinality  is contained in a summand of

P that is Š P 0 with  0
2 .
(b) Derive from (a) that a non-measurable summand of any P is a product.

7 Whitehead Groups If V = L

By a Whitehead group or, in short, by a W-group (Rotman) is meant a group A

satisfying

Ext.A; Z/ D 0;

i.e. every extension of Z by A is splitting. Evidently, free groups are W-groups, and
Whitehead’s question was whether or not the converse is true, i.e. are all W-groups
free? After several unsuccessful attempts by various algebraists, it was S. Shelah
who took the bold step of approaching the problem in a revolutionary way, and
found a flabbergasting answer to this question: it is undecidable in ZFC, i.e. it can
be neither proved nor refuted within the axiom system ZFC. We direct our efforts
towards a proof of this result, without exploring consequences or related problems.
But how can one prove that a problem is undecidable? An “easy” way would be
to show that it is equivalent to a known undecidable problem—this approach is hard
to come by in a case when no such problem exists that is related. It remains to follow
the standard road to proving undecidability: to create two models of set theory, one
in which the answer is positive, and another one which leads to a negative answer.
Such models are constructed by adjoining appropriate new axioms to the existing
axioms of ZFC, new axioms that are consistent with the old axioms. To carry out
this task for the Whitehead problem, we follow Shelah [1], and first assume Jensen’s
Diamond Principle (see Sect. 4 in Chapter 1) to obtain an affirmative answer; then
we move to a model of ZFC with Martin’s Axiom and :CH adjoined, in which the
answer will be in the negative.
Elementary Results on W-Groups The proofs are not easy and involve lots of
technicalities. One has to understand well what kind of influence the added axiom
has on the problem. We devote this section and the next one to the presentation of a
complete proof of undecidability.
In order to assemble basic facts, we list the following properties which can easily
be derived staying in the realm of ZFC.
(a) Subgroups of W-groups are again W-groups. If B is a subgroup of a W-group
A, then the inclusion map B ! A induces an epimorphism 0 D Ext.A; Z/ !
Ext.B; Z/.
(b) Direct sums of W-groups are again
Q W-groups. This follows at once from the
isomorphism Ext.˚i2I Ai ; Z/ Š i2I Ext.Ai ; Z/.
520 13 Torsion-Free Groups of Infinite Rank

(c) The class of W-groups is closed under extension: if A; C are W-groups in the
exact sequence 0 ! A ! B ! C ! 0; then so is B. This is immediate
consequence of the exactness of the induced sequence 0 D Ext.C; Z/ !
Ext.B; Z/ ! Ext.A; Z/ D 0:
(d) W-groups are torsion-free. In view of (a), it suffices to show that a cyclic group
Z.p/ is not a W-group. But this follows at once from the observation that Z is a
non-splitting extension of pZ Š Z by Z=pZ Š Z.p/.
(e) A W-group of finite rank is free. Let A be a torsion-free group of finite rank
n, and F a free subgroup of rank n; thus, A=F is a torsion group. If A is not
finitely generated, then A=F is infinite, and in this case Ext.A=F; Z/ has to be
(at least) of the power of the continuum. This follows at once from Corollaries
3.6 in Chapter 9 and 1.8 in Chapter 7, but it is easy to prove it directly. (In
fact, if A=F contains a copy of Z.p1 / for some prime p, then Ext.A=F; Z/
contains Ext.Z.p1 /; Z/ Š Hom.Z.p1 /; Z.p1 // Š Jp . If A=F contains no
copy of Z.p1 / for any prime p,Q Q socle, so Ext.A=F; Z/
then A=F has an infinite
maps upon an infinite product p Ext.Z.p/; Z/ Š p Z.p/, and therefore its
cardinality is  2@0 .) Consider the exact sequence 0 ! F ! A ! A=F ! 0;
it induces the exact sequence Hom.F; Z/ ! Ext.A=F; Z/ ! Ext.A; Z/: Here
Hom.F; Z/ D ˚ Hom.Z; Z/ D ˚ Z is finitely generated free, so countable.
Hence if A is a W-group, i.e. if Ext.A; Z/ D 0; then Ext.A=F; Z/ cannot be
uncountable. Thus then A=F must be finite, so A is finitely generated free.
(f) (Rotman [2]) W-groups are separable. Let F be a finite rank pure subgroup of a
W-group A. The exact sequence in (e) implies that there is an epimorphism
Hom.F; Z/ ! Ext.A=F; Z/. In our case, Hom is finitely generated, while
the Ext is divisible (see Sect. 3(F) in Chapter 9), so it must be 0. Hence also
Ext.A=F; F/ D 0, i.e. F is a summand of A.
We have come to the first major result of this section.
Theorem 7.1 (Ehrenfeucht [1]). Countable Whitehead groups are free. Hence all
Whitehead groups are @1 -free.
Proof. In view of (f), countable W-groups are completely decomposable (cf. Theo-
rem 4.3). By (e), their finite rank subgroups are free, so they are free as well. t
u
Uncountable W-Groups We now move to the case of W-groups of uncountable
cardinalities, at this point, still no additional axiom is required. Obstacles to
extend Theorem 7.1 quickly present themselves already at cardinality @1 . Since it
suffices to prove undecidability at @1 , we will focus our attention on W-groups of
this cardinality, but will prove results on all uncountable regular cardinals whenever
no extra effort is required.
We shall need the following lemmas.
Lemma 7.2. Let 0 ! A ! B ! C ! 0 be an exact sequence of torsion-free
groups, where B is a W-group, but C is not. There is a homomorphism  W A ! Z
that fails to extend to a homomorphism B ! Z:
7 Whitehead Groups If V = L 521

Proof. Form the exact sequence Hom.B; Z/ ! Hom.A; Z/ ! Ext.C; Z/ !

Ext.B; Z/ D 0: By hypothesis, the first Ext does not vanish, so the map between
the two Homs is not epic. Hence the claim is obvious. t
u
˛ ˇ
Lemma 7.3 (Shelah [1]). As before, let 0 ! A!B!C ! 0 be an exact
sequence where B is a W-group, but C is not. Consider the splitting exact sequences
in the rows of the following diagram where ;  0 denote the projections onto the
second summands. Let  be the obvious injection map A ! Z ˚ A: Then there is a
homomorphism  W Z ˚ A ! Z ˚ B making the diagram

π
0 −−−−→ Z −−−−→ Z ⊕ A −−−−→ A −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
 χ α

π
0 −−−−→ Z −−−−→ Z ⊕ B −−−−→ B −−−−→ 0

commute such that there is no splitting map  W B ! Z ˚ B for  0 with  D ˛:

Proof. Let  be the map as stated in the preceding lemma. Define  to map .n; a/ 2
Z ˚ A onto .n C a; ˛a/ 2 Z ˚ B: It is readily checked that—due to the choice of
—there is no splitting map  W B ! Z ˚ B with the required property. t
u
Consequences of V D L The stage is now set to move to a model of set theory
in which the Axiom of Constructibility holds. Our first order of business is to discuss
its immediate consequences concerning W-groups. The reader is reminded that the
Diamond Principle } is a consequence of V D L.
In the next proof, it will be convenient to consider an extension G of Z by A
to be a group on the set Z  A: Accordingly, if fA˛ g˛< is a filtration of A; then
fZ  A˛ g˛< is a filtration of G:
Lemma 7.4 (Shelah [1]). Assuming }, let  be an uncountable regular cardinal,
and A a torsion-free group of cardinality : Suppose that A has a filtration fA˛ g˛<
where
(i) each A˛ has cardinality < I
(ii) each A˛ is a pure W-subgroup of AI
(iii) E D f˛ <  j A˛C1 =A˛ is not a W-groupg is stationary in :
Then A is not a Whitehead group.
Proof. Hypothesis } implies that there is a family fg˛ g˛2E of functions g˛ W A˛ !
Z  A˛ .˛ 2 E/ such that, for every function g W A ! Z  A; the set E0 D f˛ 2
E j g  A˛ D g˛ g is stationary in .

We construct a non-split exact sequence e W 0 ! Z ! G!A ! 0 as a direct
˛
limit of splitting exact sequences e˛ W 0 ! Z ! G˛ !A˛ ! 0 for ˛ <  such that
the underlying set for G˛ is Z  A˛ ; and whenever ı < ˛; there is a commutative
diagram
522 13 Torsion-Free Groups of Infinite Rank

π
δ: 0 −−−−→ Z −−−−→ Gδ −−−δ−→ Aδ −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
   (13.6)
πα
α: 0 −−−−→ Z −−−−→ Gα −−−−→ Aα −−−−→ 0

where the right vertical map is the inclusion map. Let ˛ < , and assume that
the exact sequences eˇ have been defined for all ˇ < ˛ such that all the required
diagrams commute.
Case 1. If ˛ is a limit ordinal, then e˛ is defined as the direct limit of the exact
sequences eˇ with ˇ < ˛: This is a splitting sequence, since A˛ is a W-group.
Case 2. Let ˛ D ı C 1: If ı … E or if gı is not a splitting map for ı ; then let
˛
e˛ W 0 ! Z ! G˛ !A˛ ! 0 be an extension of eı with any choice of
Gı ! G˛ such that (13.6) commutes, and the underlying set for G˛ is ZA˛ : e˛
is a splitting exact sequence because of condition (ii).
Case 3. Let ˛ D ı C 1; ı 2 E; and assume that gı W Aı ! Z  Aı (= the underlying
set for Gı ) is a splitting map for ı . From Lemma 7.3 we conclude that there is
˛
an extension e˛ W 0 ! Z ! G˛ !A˛ ! 0 with a commutative diagram (13.6)
such that there is no splitting map  W A˛ ! G˛ for ˛ matching ı . Again, we
view G˛ to be a group on the set Z  A˛ :

We now define e W 0 ! Z ! G!A ! 0 as the direct limit of the splitting
˛
exact sequences e˛ W 0 ! Z ! G˛ !A˛ ! 0 for ˛ <  where G D Z  A as
sets. By way of contradiction, assume there is a splitting homomorphism g W A ! G
for : Note that we must have g.A˛ /
G˛ for every ˛ < : By the choice of the g˛ ;
there is a ı 2 E (actually, stationarily many of them) such that g  Aı D gı : This
means that eıC1 has been constructed according to Case 3 above. Since g  AıC1
is both a splitting map for eıC1 and an extension of g  Aı (which is impossible by
construction), we reached a contradiction to the existence of g: Thus A cannot be a
W-group. t
u
We shall now state an auxiliary lemma that contains an important step in the
proof of Theorem 7.6.
Lemma 7.5. Assume }. A W-group A of uncountable regular cardinality  satis-
fies:
(? ) Every subgroup B of cardinality <  is contained in a pure subgroup C of
cardinality <  such that D=C is a W-group for every subgroup D of cardinality
<  satisfying C < D < A.
Proof. Assume toward a contradiction that .? / fails for A, i.e. there is a (pure)
subgroup B of cardinality <  such that, for every pure subgroup C > B of
cardinality < , we can find a subgroup D > C of cardinality <  such that D=C is
not a W-group. Define a smooth chain fB˛ g˛< of pure subgroups in A as follows.
Start with B0 D B. If ˛ <  and Bˇ has been defined for all ˇ < ˛, then
7 Whitehead Groups If V = L 523

S
1. if ˛ is a limit ordinal, then set B˛ D ˇ<˛ Bˇ I
2. if ˛ D ı C 1; then using (? ), take for B˛ a pure subgroup of A of cardinality < 
that contains Bı such that B˛ =Bı is not a W-group.
S
As a subgroup of A, B0 D ˛< B˛ is a W-group. On the other hand, the preceding
lemma with E D  applies, showing that B0 cannot be a W-group. We have reached
a contradiction, so (? ) holds. t
u
We are now able to prove:
Theorem 7.6 (Shelah [1]). Assume again }. Let  be an uncountable regular
cardinal, and A a Whitehead group of cardinality . If all Whitehead groups of
cardinalities <  are free, then A is free.
Proof. We use (? ) in Lemma 7.5 as a guidance to create a -filtration fA˛ g˛< of
A (now D=C is a W-group of cardinality < , thus free). To start with, choose a
maximal independent set fa˛ g˛< in A. Set A0 D 0, and suppose that ˛ <  and the
pure groups Aˇ have been defined for all ˇ < ˛ such that
(i) they are of cardinalities < ,
(ii) they form a smooth chain, and
(iii) Aˇ contains a for all < ˇ.
S
Then for a limit ordinal ˛, set A˛ D ˇ<˛ Aˇ , as required. In case ˛ is a successor
ordinal, say, ˛ D ı C 1, then let B be the pure subgroup in A that is generated by Aı
and aı , and choose for A˛ the subgroup C obtained by applying (? ) to this B.
Since A is a W-group, the set E D f˛ <  j A˛C1 =A˛ is not freeg cannot be
stationary in , as is obvious in view of Lemma 7.4. Thus we can be sure that
there is a cub F in  which does not intersect E; write F D ff .˛/g˛< . To finish
the proof it suffices to show that fAf .˛/ g˛< is a filtration of A with free factor
groups Af .˛C1/ =Af .˛/ . But this follows at once from that Af .˛/C1 =Af .˛/ is free by
construction, and that Af .˛C1/ =Af .˛/C1 is free as Af .˛/C1 is a C in the application
of (? ). t
u
We have obtained enough information on our way toward the proof of the main
theorem of this section.
Theorem 7.7 (Shelah [1]). In the Constructible Universe, Whitehead groups of
any cardinality are free.
Proof. We induct on the cardinality  of the W-group A. Theorem 7.1 takes care of
the countable case, so assume that A is a W-group of uncountable cardinality , and
all W-groups of cardinalities <  are free. If  is a regular cardinal, then Theorem
7.6 guarantees that A is free. If  is a singular cardinal, then the freeness of A is a
consequence of the singular compactness theorem (Theorem 9.2 in Chapter 3). u t
F Notes. According to A. Ehrenfeucht, J.H.C. Whitehead raised the problem in 1952 during
his visit to Warsaw. Unaware of the solution by K. Stein [Math. Annalen 123, 201–222 (1951)] of
an equivalent problem in the countable case (applied to the second Cousin problem), Ehrenfeucht
[1] published his solution for countable groups in the equivalent form: if H is a subgroup of a
countable free group F such that every  W H ! Z extends to a W F ! Z, then H is a summand
of F.
524 13 Torsion-Free Groups of Infinite Rank

It should also be mentioned that J. Dixmier [Bull. Sci. Math. France 81, 3–48 (1957)] unveils
an intimate relation between the Pontryagin duals of Whitehead groups, and the arcwise (locally
arcwise) connected locally compact abelian groups.

Exercises

(1) Let A be torsion-free of finite rank. If A is not free, then j Ext.A; Z/j D 2@0 .
(2) A group A is free if it satisfies Ext.A; F/ D 0 for all free groups F.
(3) For a reduced group A, the following are equivalent:
(a) Ext.A; Z/ is torsion-free;
(b) A has the projective property relative to the exact sequences 0 ! nZ !
Z ! Z=nZ ! 0 for all integers n. [Hint: apply the functor Hom.A; / to
this exact sequence.]
(4) In every model of ZFC, the class of W-groups is closed under isomorphisms,
direct sums, summands, and extensions.

8 Whitehead Groups Under Martin’s Axiom

Having proved that W-groups are free in some model of ZFC, in order to verify the
claimed undecidability of the Whitehead problem, the issue is to find a suitable
hypothesis which, if added to the axioms of ZFC, will provide us examples of
non-free W-groups. As we have already mentioned above, such a hypothesis
is Martin’s Axiom (MA), which we have formulated in Sect. 4 of Chapter 1.
(Application of MA makes sense only for uncountable cardinals < 2@0 , so we will
have to deny CH.)
Consequences of Martin’s Axiom Consequently, we place ourselves in the
following setting: ZFC + MA + :CH, and we are in search for an example of a
non-free W-group in this model of set theory. Of course, we are free to use the
results which we have proved in the preceding section for every model of ZFC.
We turn to the proof of the crucial lemma which already requires the full strength
of Martin’s Axiom. The hard work goes into showing that the selected poset has the
requisite properties.
Lemma 8.1. Assume ZFCCMAC:CH. Any separable @1 -free group of cardinal-
ity @1 that satisfies condition .? / in Lemma 7.5 for  D @1 is a W-group.
Proof. Suppose A is a separable @1 -free group of cardinality @1 (this cardinality is
now less than the continuum) satisfying .? /, and let

e W 0 ! Z ! G!A ! 0
8 Whitehead Groups Under Martin’s Axiom 525

be an exact sequence. We want to show that e is splitting by establishing the

existence of a map ı W A ! G satisfying ı D 1A .
To get prepared for the application of MA, we have to select a suitable partially
ordered set. This will be the set P of all pairs .S; ıS / where S is a finitely generated
pure subgroup (thus a summand) of A, and ıS W S ! G satisfies ıS D 1S . Define a
partial order on P in the obvious way:

.S; ıS /
.S0 ; ıS0 /

should mean that S

S0 and ıS D ıS0  S.
Let us point out right away that every finitely generated pure subgroup S of A
occurs in a pair .S; ıS / 2 P, since every S is a free summand of A. However, there
are but countably many pairs .S; ıS / 2 P with the same S; in fact, the action of ıS
is completely determined on the generators of S, and the image of an element of S
under ıS always belongs to the same coset mod Z. Hence it should be clear that, for
a fixed S, ıS can act in at most countably many different ways.
In order to be able to apply MA, we have to study the upper bound situation in
P. If both S and S0 are pure subgroups of finite rank in A, then S is a summand of
the purification S? of S C S0 which is a summand of A. Now if .S; ıS /; .S0 ; ıS0 / are in
P, then ıS W S ! G and ıS0 W S0 ! G have a common extension ı 0 to S C S0 exactly if
they coincide on S \ S0 . As ı 0 D 1SCS0 , it is easy to see that ı 0 extends uniquely to
a map ıS? W S? ! G such that ıS? D 1S? : Hence we conclude that for every choice
of a 2 A; the set

Ca D f.S; ıS / 2 P j a 2 Sg

is cofinal in P. If we can show that P satisfies the countable antichain condition,

then we can appeal to MA to obtain a directed subset D of P that intersects every
Ca . This will define a map ı W A ! G satisfying ı D 1A if we set ıa D ıS a for
a 2 A whenever .S; ıS / 2 D \ Ca : From the fact that D is directed it follows that ı
is in fact a homomorphism, and thus it is a splitting map—and the proof will then
be finished.
Therefore, it remains to verify the countable antichain condition on P. Suppose
Q D f.S ; ı / j  < !1 g is an uncountable subset of P. We want to show that Q (or
a subset of Q) contains two elements which have a common upper bound in P. This
will be accomplished in three steps.
Step 1. There is a pure free subgroup A0 of A that contains S for uncountably many
.S ; ı / 2 Q: To verify this claim, note that each S has finite rank, so there is an
integer m such that uncountably many S have rank m; moreover, by dropping
to a subset of Q, we may assume that all of S have the same rank m. Pick a pure
subgroup T of A of possibly highest rank that is contained in uncountably many
S (maybe T D 0). By a remark above, ı  T must be one of at most countably
many maps, so we may as well suppose that the ı  T are equal for all  < !1 .
T is a summand of A, so S D T ˚ U for a subgroup U of A. Set A0 D T, and
526 13 Torsion-Free Groups of Infinite Rank

suppose that, for some  < !1 , we have defined a smooth chain A of countable
pure free subgroups of A, along with ordinals C1 , such that UC1
AC1
for all  < . If  is a limit ordinal, we let A be the union of all A with  < .
If  D C 1, then appeal to .? /, and pick a countable @1 -pure subgroup B of
A that contains A . There is an index  such that  > C1 for all  < 
and U \ B D 0; otherwise, we get a contradiction to the maximal choice of
T. Define A as the purification of A C U in A. Now U \ B D 0 implies
A \ B D A , so A =A is free as it is isomorphic to a countable subgroup S of
A=B . Completing this induction for all  < !1 , we arrive at A0 D <!1 A
and an uncountable subset Q0 D f.SC1 ; ıC1 / j  < !1 g of Q. Evidently, A0
contains all SC1 with  < !1 , and is free, since the factors in the chain are
free.
Step 2. Let Q D f.S ; ı / j  < !1 g be an uncountable subset of P such that every
S . < !1 / is contained in a pure free subgroup A0 of A. If X is a basis of A0 ,
then there is a subset Q0 D f.S0 ; ı0 / j  < !1 g of P such that .S ; ı /
.S0 ; ı0 /
for each  < !1 , and each S0 is generated by a subset of X. This follows from
the observation that every .S; ı/ can be extended to an .S0 ; ı0 / such that S0 is
generated by basis elements in X.
Step 3. If every S . < !1 / in Q (as in Step 2) is generated by a subset of a basis
X of a pure free subgroup A0 of A, then Q contains two pairs with a common
upper bound in P.
As in Step 1, we may assume that all S are of the same rank, and all contain a
summand T such that all the ı  T are equal for  < !1 . Pick any .S0 ; ı0 / 2 Q;
evidently, T ¤ S0 . By the choice of T, for every pure subgroup T 0 generated by
a subset of X with T < T 0
S0 , there are at most countably many  < !1 with
T 0
S . Also, there are but countably many such T 0 , so after omitting all these
.S ; ı / from Q, there are still uncountably many pairs left. If .S ; ı / 2 Q is
one of these, then necessarily S0 \ S D T. This guarantees that ı0 and ı have
a common extension to S0 C S . This group is pure in A0 , and hence in A, as it
is generated by a subset of X.
This completes the proofs of the countable antichain condition and the lemma. u
t
The Basic Example For the proof of the main theorem we have to ascertain the
existence of a non-free example for Lemma 8.1; this is furnished by the following
lemma (valid in ZFC).
Lemma 8.2. There exists a separable torsion-free group of cardinality @1 which
satisfies .? /, but is not free.
Proof. Follow the proof of Theorem 8.9 in Chapter 3 in constructing a group F of
cardinality @1 that is the union of a smooth chain F . < !1 / groups satisfying
(i) F is a countable free group for all  < !1 ;
(ii) F C1 =F Š Q if  < !1 is a limit ordinal;
(iii) F =F is free for all  <  < !1 if  is a successor ordinal.
Such an F is @1 -free, but not free, and in addition it satisfies .? /. t
u
8 Whitehead Groups Under Martin’s Axiom 527

We now formulate our findings in

Theorem 8.3 (Shelah [1] (ZFC+MA+ :CH)). There exist Whitehead groups of
cardinality @1 which are not free. u
t
Undecidability of the Whitehead Problem From Theorems 7.7 and 8.3 we
now conclude:
Corollary 8.4 (Shelah [1]). The Whitehead problem is undecidable in ZFC. t
u
F Notes. The history of W-groups is replete with unsuccessful attempts, furnishing meager
information about the groups. The solution of the Whitehead problem in 1973 by Shelah was
the first undecidable result in abelian group theory, and became the starting point of a new era in
the theory, dominated by set-theoretical methods. The discovery of several unsolvable problems
followed suit. The methods were extended to module theory, and today there is a vast amount of
literature dealing with problems that can be traced back to the Whitehead problem, as witnessed
by the excellent monograph by Eklof–Mekler [EM].
For a relatively easy proof for the undecidability, Eklof [3] is recommended.
Numerous publications deal with variants of the Whitehead problem, imposing various
conditions on the model of set theory. For example Shelah [3], where CH is assumed.
Rychkov [2] considers the problem of vanishing Pext that is reminiscent to the Whitehead
problem. Assume A; C are groups satisfying jAj < jCj D . If  is a regular cardinal, and
C . < / is a filtration of C with pure subgroups, then Pext.C; A/ D 0 holds exactly if
Pext.C ; A/ D 0 and Pext.CC1 =C ; A/ D 0 for all  < . He also establishes a kind of
singular compactness result by proving that if  is singular, then Pext.C; A/ D 0 if and only if
Pext.X; A/ D 0 for all pure subgroups X of C of cardinality < . On the other hand, assuming
ZFC+MA+ :CH, if jAj D @0 , and if C is of cardinality @1 such that every countable subgroup is
contained in a countable †-cyclic direct summand, then Pext.C; A/ D 0.

Exercises

(1) (Chase) If j Ext.G; Z/j < 2@0 , then G is a direct sum of a finite group and an
@1 -free group. [Hint: argue as in Sect. 7(e), t.G=F/ must be finite.]
(2) (Rotman, Nunke) W-groups are slender. [Hint: Ext.Z@0 ; Z/ ¤ 0; refer to
algebraic compactness of P=S.]
(3) (Griffith) An @1 -free @1 -coseparable torsion-free group is a W-group.
[Hint: Theorem 4.17.]
(4) Prove that MA+:CH implies that for every uncountable cardinal  there exist
non-free Whitehead groups.
(5) What can be said about a group A which has the property that every homomor-
phism A ! Q=Z lifts to a map A ! Q?

Problems to Chapter 13

PROBLEM 13.1. Are summands of @n -free @n -separable torsion-free groups

again @n -separable?
In the case n D 1, the answer is given in a few special cases in Mekler [2].
528 13 Torsion-Free Groups of Infinite Rank

PROBLEM 13.2. Let  <  denote infinite cardinals, and A the subgroup in Z
that consists of vectors whose supports are of cardinality < . Is every summand of
A of the same form? In particular, if  is measurable, and  is not.
PROBLEM 13.3. When is the tensor product of two reflexive groups again
reflexive?
PROBLEM 13.4. (Eklof–Mekler) Does there exist a reflexive group of measur-
able cardinality?
PROBLEM 13.5. Characterize vector groups A such that A D B ˚ C implies that
either B Š A or C Š A.
PROBLEM 13.6. (J. Martinez) Which torsion-free groups admit archimedean
lattice-order?
Chapter 14
Butler Groups

Abstract The theory of Butler groups is one of the most elaborate branches of abelian groups.
We devote a whole chapter to its study. This may seem excessive, since Butler groups are very
special, but the fact that the results and the methods provide more than a superficial glimpse into
a fascinating theory (the most extensive one today on torsion-free groups of arbitrary rank) is a
compelling reason for including a broader discussion.
In the finite rank case, Butler groups are torsion-free generated by a finite number of rank 1
groups. It is very tempting to think that a kind of finite generation makes them susceptible to a
satisfactory classification; however, so far no comprehensive theory has emerged, though many
encouraging results are available. We prove the basic results on them, but we are unable to dig
deeply into the theory without getting involved in overcomplicated details. An important part of the
theory aims at finding more tractable classes of finite rank Butler groups. The fast developing, very
successful theory of almost completely decomposable groups is well documented in Mader[Ma].
We investigate more thoroughly Butler groups of large cardinalities; their theory has gained
considerable popularity in the last quarter of the twentieth century, and even today they remain
under intense scrutiny. It is an illuminating experience to see the effect of set theory on their
algebraic structure.

1 Finite Rank Butler Groups

The only class of finite rank torsion-free groups that is larger than the class of
completely decomposable groups and has a well developed theory is the class of
Butler groups. This class is more manageable than finite rank torsion-free groups
in general, but still general enough to generate a variety of challenging problems.
This is still a fertile research area. The theory originates from Butler [1] and Bican
[3], using different approaches. The fundamental facts on this class of groups are
due to them.
Basic Properties of Butler Groups After these prefatory remarks, we now state
the precise definition. A torsion-free group of finite rank is called a Butler group if
it is a pure subgroup of a completely decomposable group (of finite rank).
Evidently, all completely decomposable groups of finite rank, in particular, all
rank 1 groups, are Butler groups. The main interest lies, of course, in Butler groups
that fail to be completely decomposable.
Example 1.1. Let A be the direct sum: A D A1 ˚ A2 ˚ A3 where each Ai is torsion-free of rank 1
such that they contain elements ai 2 Ai of characteristics .1; 1; 0; 0; : : : /; .1; 0; 1; 0; 0; : : : /,

© Springer International Publishing Switzerland 2015 529

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_14
530 14 Butler Groups

and .0; 1; 1; 0; 0; : : : /; respectively. The pure subgroup generated by the elements a1  a2 ;

a2  a3 ; a3  a1 is a rank two indecomposable Butler group.
Example 1.2. Let fq; p1 ; : : : ; pn g be a set of primes, and V a Q-vector space with basis
fe1 ; : : : ; en g. Then the group

G D hp1
1 e1 ; : : : ; p1
n en ; q1 .e1 C e2 /; : : : ; q1 .e1 C en /i  V

is an indecomposable Butler group of rank n. (If we replace q1 by q1 , the arising group is still
Butler.)
Example 1.3. No pure subgroup of rank n
2 in the group Jp of the p-adic integers is a Butler
group.

In a completely decomposable group A D A1 ˚    ˚ An ; where each Ai is of

rank 1, the types of elements are evidently finite intersections of types of the Ai ,
so Type.A/ is finite. Since the typeset of a pure subgroup is always a subset of the
typeset of the group, it follows at once that finite rank Butler groups have finite
typesets.
Butler’s Theorem The next theorem will show that there is an equivalent way
of defining a Butler group. The two definitions are used interchangeably.
Let A D B1 C  CBm be a torsion-free group where the Bj ’s are subgroups of rank
one. This is equivalent to saying that A is an epimorphic image of the outer direct
sum A0 D B1 ˚    ˚ Bm , which is a completely decomposable torsion-free group of
finite rank. Here we may assume without loss of generality that the subgroups Bj are
pure in A (they can always be replaced by their purifications), thus A is generated
by a finite set of rank 1 pure subgroups.
The classical result on Butler groups reads as follows.
Theorem 1.4 (Butler [1]). For a torsion-free group A the following are equivalent:
(i) A is a pure subgroup of a completely decomposable group of finite rank;
(ii) A is an epimorphic image of a completely decomposable group of finite rank.
Proof. (i) ) (ii) Let A be a pure subgroup of a completely decomposable group
C D C1 ˚    ˚ Cn , where each Ci is of rank 1. For an a 2 A write a D c1 C    C cn
with ci 2 Ci , and define the support of a 2 A as usual by supp a D fi j ci ¤ 0g.
Note that if S D supp a D supp b (with a; b ¤ 0 in A), then there are s; t 2 N such
that supp .sa  tb/ is properly contained in S. Hence sa D tb if S is minimal. This
shows that, for every minimal support S, fa 2 A j supp a D Sg [ 0 is a rank 1 pure
subgroup BS of A.
Let fBj j j 2 Jg be the set of all rank one pure subgroups of A whose elements
have minimal supports. Evidently, this is a finite set, say, J D f1; : : : ; mg. In order
to verify (ii) for A, it suffices to show that every x 2 A belongs to B D B1 C   C Bm .
Clearly, for each x 2 A there is a y 2 A such that supp y is minimal and is a subset
of supp x. In view of the purity of A in C, for any given prime p, we can find an i 2
supp y such that the ith coordinates xi and yi of x and y have the same heights at p.
Then there are integers s; t with .p; s/ D 1 such that supp .sx  ty/ is a proper subset
of supp x. Using induction, we may assume sx  ty 2 B. Thus y 2 B implies sx 2 B,
1 Finite Rank Butler Groups 531

and so A=B is a torsion group. If we had an x 2 A n B satisfying px 2 B for some

prime p, then the existence of an integer s (prime to p) with sx 2 B would imply
x 2 B, a contradiction. Hence A D B, and (ii) follows.
(ii) ) (i) Suppose A D B1 C    C Bm where the Bj are different pure subgroups
of rank 1. If m D 1, there is nothing to prove. So assume m > 1, and the claim holds
for groups with smaller m. Form the factor groups Xj D A=Bj , and consider the map

 W A ! X D X1 ˚    ˚ Xm

defined by .a/ D .a C B1 ; : : : ; a C Bm /. Since rk A > 1, the intersection of two

different Bj is 0, so  is monic. Assume, by way of contradiction, that .A/ is
not pure in X, i.e., there are a prime p and an element a 2 A n pA with .a/ D
p.x1 C B1 ; : : : ; xm C Bm / for some xj 2 A. Then bj D a  pxj 2 Bj n pBj for each
j. Since the Bj are of rank 1, pBj and bj generate Bj , and so pA and a (or any of
the bj ) generate A. Hence A=pA is cyclic of order p. This implies that none of the
Bj is p-divisible, but every Xj D A=Bj is p-divisible. This is impossible if the rank
of A is > 1, since Xi is generated by the images of the Bj with j ¤ i, and since a
finite number of not p-divisible rank one subgroups cannot generate a p-divisible
subgroup. Hence we conclude that  embeds A as a pure subgroup in X. Again,
Xi being generated by the images of the Bj with j ¤ i, the induction hypothesis
applies to the Xi : they are pure subgroups in completely decomposable groups of
finite rank. This property carries over to their direct sum X, and hence it follows that
A is likewise pure in a completely decomposable group. t
u
An easy consequence worthwhile recording is as follows.
Corollary 1.5. A homogeneous finite rank Butler group is completely decompos-
able.
Proof. A homogeneous finite rank Butler group B is an epic image of a finite rank
completely decomposable group C that may be chosen to be homogeneous of the
same type. The kernel of such an epimorphism C ! B is a pure subgroup, so
by Corollary 3.7 in Chapter 12 it is a summand of C. Therefore, Theorem 3.10 in
Chapter 12 implies that B is likewise completely decomposable. t
u
The following claim is immediate from the definition and Theorem 1.4.
Lemma 1.6. The class of finite rank Butler groups is closed under the formation of
finite direct sums, pure subgroups and torsion-free epimorphic images. t
u
In a Butler group B, of importance are the critical types. The factor group
B.t/=B? . t/ is 0 or homogeneous of type t (we will see that it is completely decom-
posable). t is a critical type of B if this factor group is ¤ 0. Evidently, the critical
types of a completely decomposable group are the types of its rank 1 summands.
The critical types play an important role in the following characterization of Butler
groups.
532 14 Butler Groups

Theorem 1.7 (Butler [1]). A finite rank torsion-free group A is a Butler group if
and only if it satisfies the following conditions:
(a) Type.A/ is finite;
(b) for each critical type t, A? .t/ has finite index in its purification A? .t/ I
(c) for each critical type t, there is a direct decomposition

A.t/ D At ˚ A? .t/ (14.1)

where At is a homogeneous completely decomposable group of type t.

Proof. First suppose A is a Butler group. (a) has been observed above. As a pure
subgroup of A; A.t/ is a Butler group, and hence At is likewise Butler. Since a
homogeneous Butler group must be completely decomposable (see Corollary 1.5)(c)
follows in view of Lemma 3.6 in Chapter 12. Considering that A? .t/ is a Butler
group, we have A? .t/ D A1 C  CAk CA? .t/ for certain rank one pure subgroups Ai
of A. If none of Ai can be dropped, then the Ai are all of type t, and their intersections
with A? .t/ are likewise of type t. Hence each Ai \ A? .t/ is of finite index in Ai , and
therefore A? .t/ is of finite index in A? .t/ .
Conversely, let A satisfy (a)–(c). (b) implies that A? .t/ D ?
P Ft C AP. t/ for a
0
finitely generated free subgroup Ft . Hence the subgroup A D t At C t Ft is by
(a) a finite sum of rank one groups, so it is a Butler group. If A n A0 is not empty, then
choose an element a 2 A n A0 whose type s is maximal among the types of elements
in A n A0 . Then

a 2 A.s/ D As ˚ A? .s/ D As ˚ .Fs C A? .s//:

By the maximal choice of s, we have A? .s/

A0 , so also A.s/
A0 , in contradiction
to the choice of a. t
u
Bican’s Theorem The following remarkable characterization of Butler groups
relies on the partition of primes. If  is a set of primes, then the localization Z of
Z at  is the group of all rational numbers whose denominators are relatively prime
to every prime in . It should be observed that an immediate consequence of the
definition is that localizations of Butler groups are likewise Butler groups.
Theorem 1.8 (Bican [3]). A finite rank torsion-free group B is a Butler group if
and only if there is a partition

… D …1 [    [ …k

of the set … of prime numbers such that for each ` .` D 1; : : : ; k/, the tensor product
B ˝ Z…` .localization of B at …` / is a completely decomposable group with totally
ordered typeset.
1 Finite Rank Butler Groups 533

Proof. Let B be a Butler group, viewed as a pure subgroup of the completely

decomposable group C D C1 ˚    ˚ Cn where the Ci are regarded as subgroups of
Q containing 1. Let …ij .1
i < j
n/ denote the set of primes p for which hp .1/
is not larger inTCi than in Cj , and define f…1 ; : : : ; …k g as the set of all non-empty
intersections i<j Pij where, for each pair i < j; Pij is equal either to …ij or to the
complement of …ij in …. For each `, the subgroups Ci` D Ci ˝ Z…` .i D 1; : : : ; n/
of Q are totally ordered by inclusion, whence it follows that for each `; B ˝ Z…` (as
a pure subgroup of C ˝ Z…` ) has a totally ordered typeset. Such a Butler group has
to be completely decomposable, since in its decomposition (14.1) with its minimal
type t, the first summand is completely decomposable, and so is the second by an
induction hypothesis. T
Conversely, it suffices to show that B D k`D1 .B ˝ Z…` / is isomorphic to a pure
subgroup of the completely decomposable group ˚k`D1 .B˝Z…` /, or more generally,
X1 \    \ Xk is isomorphic to a pure subgroup of X1 ˚    ˚ Xk whenever X1 ; : : : ; Xk
are subgroups of a given torsion-free group. For k D 2 this is obvious from the
exactness of the sequence 0 ! X1 \ X2 ! X1 ˚ X2 ! X1 C X2 ! 0, and the
general case follows by a straightforward induction on k. t
u
Balanced-Projective Dimensions In the next theorem, which provides an
upper bound for the balanced-projective dimension (bpd) of a Butler group (for
the definition of bpd see Sect. 3 in Chapter 12), we consider strictly increasing
sequences t1 <    < tn of types.
Lemma 1.9 (Arnold–Vinsonhaler [1]). Let B be a finite rank Butler group, and
n the maximal length of strictly increasing sequences of types in B. Then bpd B

n  1.
Proof. We argue by induction on n. n D 1 means that B is homogeneous, and
therefore completely decomposable by Corollary 1.5; then bpd B D 0. Suppose
n > 1, and let 0 ! A ! C ! B ! 0 be a balanced-projective resolution of
B, where C is completely decomposable. Hence we have Type.A/
Type.C/

Type.B/; the second inequality holds if C is chosen with minimal typeset. Setting
A D C0 ˚ A0 where C0 is completely decomposable, and A0 has no rank 1 summand,
we claim that Type.A0 / cannot contain any t0 that is maximal in Type.B/. For if X is
a pure rank 1 subgroup of A0 of maximal type t0 , then X is pure in C, so it must be
a summand of C.t0 /. It is then a summand of C, and hence of A0 , contradicting the
choice of A0 . Consequently, the maximal length of increasing sequences of types in
A0 is strictly less than n  1, so bpd A0
n  2 by induction hypothesis. As bpd B =
bpd A C 1 = bpd A0 C 1 (by standard dimension arguments), the assertion follows.
t
u
Arnold–Vinsonhaler [1] exhibit an example for a Butler group B with type chains
of maximal length n and bpd B D n  1.
534 14 Butler Groups

Example 1.10. Let p1 < p2 < p3 be primes, and Z.pi ;::: / < Q the localization of Z at the primes
listed in the subscript. Define B as the kernel of the map ˇ W Z.p1 / ˚ Z.p2 / ˚ Z.p3 / ! Q, acting as
an embedding on each summand. Clearly, rk B D 2, and B is not completely decomposable. One
can check that the sequence
˛
0 ! Z.p1 ;p2 ;p3 / !Z.p2 ;p3 / ˚ Z.p1 ;p3 / ˚ Z.p1 ;p2 / !B ! 0

is balanced-exact where ˛.x/ D .x; x; x/. It follows that bpd B D 1.

Almost Completely Decomposable Groups Lady [1] defines an almost com-

pletely decomposable group as a finite extension of a finite rank completely
decomposable group. These groups form a class of well treatable Butler groups.
They have a fast growing, well-developed theory; for details, we refer to the very
informative treatise Mader [Ma]. Here we only prove a typical result.
Let A be a finite rank Butler group. For each critical type t, choose a completely
decomposable subgroup Rt .D At / in the decomposition (1), and set R D ˚R t ,
summation over the critical types. Such an R is called a regulating subgroup.
Theorem 1.11 (Lady [1]). In an almost completely decomposable group, all reg-
ulating subgroups are isomorphic completely decomposable groups, and have the
same finite index.
Proof. Suppose R and S are regulating subgroups in the almost completely decom-
posable group A. By definition, they are completely decomposable, and since they
have the same rank as A, they are isomorphic. The equality of the indices jA W Rj and
jA W Sj is proved by induction on the number of critical types. Let s be a minimal
type, and let R0 D ˚t¤s Rt , so R D Rs ˚ R0 , and similarly, S D Ss ˚ S0 , and set
A0 D A? .s/ . In view of the equality jA W Rj D jA W .Rs ˚ A0 /j  j.Rs ˚ A0 / W Rj D
jA W .As C A0 /j  jA0 W R0 j and the same for S, the claim follows, as R0 ; S0 are regulating
subgroups in A0 , so jA0 W R0 j D jA0 W S0 j by induction hypothesis. t
u
Representation by Posets It would be a mistake to leave the theory of finite
rank Butler groups without at least mentioning its close connection with the
representation of partially ordered sets. This is an important branch in the theory
of finite rank Butler groups.
Let S be a poset (partially ordered set), and K a field. By a K-representation of
S we mean a K-vector space V along with subspaces Vi .i 2 S/ such that i
j in S
implies Vi
Vj . The K-representations V D ŒVI Vi .i 2 S/
of S form an (additive)
category Rep(K, S/ if we define a morphism

V D ŒVI Vi .i 2 S/
! W D ŒWI Wi .i 2 S/

as a K-linear map  W V ! W satisfying .Vi /

Wi for all i 2 S.
1 Finite Rank Butler Groups 535

Let B be a finite rank Butler group, and T D Type.B/. Consider the set T0 of
join-irreducible types t 2 T under the reverse(!) ordering. Recall that t 2 T is join-
irreducible if t D t1 _ t2 .ti 2 T/ implies t D t1 or t D t2 . In the present case,
K D Q, and the canonical Q-representation of B is given by the correspondence

B 7! ŒQ ˝ BI Q ˝ B.t/ j t 2 T0
:

This defines a functor FT0 from the quasi-isomorphism category of Butler groups
with typeset in T to the category Rep(Q; T0 /. It was already shown by Butler [1]
that FT0 is an isomorphism. We observe that Rep(Q; S/ is a pre-abelian category
with finite direct sums.
A plethora of interesting results are available for these representations; we refer
to Arnold [A1], Arnold–Dugas [3] for details.
Special Butler Groups It is not our task to develop a full-fledged theory of
finite rank Butler groups. Our program is less ambitious, but the theory is so
extensive and fascinating that a couple of additional remarks has to be made. First,
there are several treatable subclasses that deserve special attention. Needless to
say, none of these subclasses offers a theory as impressive or as complete as that
of completely decomposable groups. One of these classes consists of the almost
completely decomposable groups mentioned above.
When it comes to investigating the relationship between being an image or a pure
subgroup of a completely decomposable group, then the classes of B .m/-groups are
in the center of attention. In the Metelli classification of finite rank Butler groups,
a group B is said to be a B .m/ -group if it is the quotient of a finite rank completely
decomposable group by a pure subgroup of rank m  0. In particular, B .0/ is the
class of completely decomposable groups of finite rank, and obviously, we have a
strictly increasing chain B .0/  B .1/      B .m/  : : : of classes. So far only
the B .1/-groups have been studied extensively. A more correct picture of finite rank
Butler groups is likely to emerge when we will be able to find the right tools to deal
with B .m/-groups for m > 1. The dual classification puts a Butler group B in class
B.m/ if it is the kernel of a homomorphism of a finite rank completely decomposable
group into the direct sum of m copies of Q.
F Notes. The groups discussed in this section were named by L. Lady Butler groups to credit
Butler who initiated their study in 1965, thereby sawing the seeds of a rich and fascinating theory
(Butler called them diagrammatic groups). These groups were rediscovered and discussed from
the point of view of finite rank 1 generation by Bican later in 1980. Cf. also Koehler [1] in the rank
2 case.
Regulating subgroups play an important role in the theory of almost completely decomposable
groups. The intersection of all regulating subgroups is the regulator, a fully invariant completely
decomposable subgroup of finite index (Burkhardt [1]). For almost completely decomposable
groups, Mader [Ma] is highly recommended.
Richman [6] defines a duality for Butler groups: the category of finite rank Butler groups whose
typeset is contained in a fixed finite distributive lattice T is dual to the category where the typeset
T 0 is anti-isomorphic to T. The duality is implemented by valuated homomorphisms into vector
spaces over Q. Another kind of duality was discussed by Arnold–Vinsonhaler [4], using groups Ai
536 14 Butler Groups

that are rational groups of types in T. The kernels of maps A1 ˚    ˚ An ! Q are the duals to the
cokernels of diagonal embeddings \niD1 Ai ! A1 ˚  ˚An , called bracket groups G ŒA1 ; : : : ; An
.
The quest for complete sets of invariants has not been successful for Butler groups in general,
but more or less satisfactory results have been obtained in very special cases. An in-depth
study of B.1/ -groups started with Richman [5] in a special case, and continued in the papers
Arnold–Vinsonhaler [5], Goeters–Ullery–Vinsonhaler [1], and Hill–Megibben [10]. The study
of B.1/ -groups was initiated in Arnold–Vinsonhaler [4], continued by W.Y. Lee [1], Fuchs–
Metelli [1], Metelli [3, 4], and developed in several papers by De Vivo–Metelli [1, 2], where
the remarkable connection to matrices with 0; 1 entries is explored in depth. Cf. also Yom [1]
for a discussion of both cases. Arnold–Vinsonhaler [5] succeed in establishing complete sets of
invariants for strongly indecomposable B.1/ - and B.1/ -groups. Vinsonhaler–Wallutis–Wickless [1]
have encouraging results on a special subclass of B.2/ -groups.
In a different vein, classes K.n/ .n < !/ of finite rank Butler groups were defined by
Kravchenko [2]: K.0/ is the class of all Butler groups. If 0 ! A ! C ! B ! 0 is a balanced-
exact sequence with completely decomposable C, then let A 2 K.n C 1/ provided B 2 K.n/. For
more on this classification, see Nongxa–Vinsonhaler [1]. Certainly, the study of finite rank Butler
groups has a great deal of potential.
Arnold–Vinsonhaler [6] is an excellent survey. Butler groups have been studied from the
constructive point of view by Richman [7] and Mines–Vinsonhaler [1]. For the classification
problem, see Thomas [2].

Exercises

(1) Fully invariant subgroups of Butler groups are Butler groups.

(2) (Arnold) Balanced extensions of Butler groups by Butler groups are again
Butler groups.
(3) A torsion-free group of finite rank with finite typeset need not be a Butler
group.
(4) If A; B are Butler subgroups of a torsion-free group G, then both A C B and
A \ B are Butler groups. [Hint: 0 ! A \ B ! A ˚ B ! A C B ! 0.]
(5) Let B be a finite rank Butler group, and t a type. Then B.t/ D B.t’/ for some
type t’ 2 Type.B/.
(6) Let A be a corank 1 pure subgroup of the finite rank Butler group B. Then
t.B=A/ D sup t.b/ for b 2 B n A.
(7) Give a counterexample to the claim that a finite rank torsion-free group B is
Butler if B ˝ Z.p/ is completely decomposable for every prime p.
(8) A Butler group with linearly ordered typeset is completely decomposable.
(9) The Metelli classes B .m/ are strictly increasing with increasing m.
(10) The tensor product of two finite rank Butler groups is again a Butler group.
(11) Homomorphism groups between finite rank Butler groups are Butler groups.
In particular, the additive group of the endomorphism ring of a finite rank
Butler group is again a Butler group.
(12) (Giovannitti) Let p; q denote different primes. Show that Bext1 .Q; Z.p;q/ / D 0,
and there is an exact sequence 0 ! Z.p;q/ ! Z.p/ ˚ Z.q/ ! Q ! 0 (not
balanced).
2 Prebalanced and Decent Subgroups 537

2 Prebalanced and Decent Subgroups

The further discussion of Butler groups is greatly enhanced by the introduction of

new concepts. We examine briefly three salient properties which prove relevant to
Butler groups both in the finite and in the infinite rank cases: prebalanced and decent
subgroups, as well as subgroups with the torsion extension property (TEP).
Balanced subgroups play a decisive role in the theory of torsion-free groups,
but the homological machinery they provide is unsatisfactory for the theory of
Butler groups. The reason is that finite rank Butler groups need not contain
any non-obvious balanced subgroup. However, there is a natural generalization
of balancedness (due to Richman [6]) that can furnish us with a sought-after
homological machinery. We also define the companion notion of decent subgroup
(Albrecht–Hill [1]).
Prebalanced Subgroups Let A be a subgroup of a (not necessarily torsion-free)
group G with torsion-free quotient G=A. A is said to be prebalanced in G if the
following condition is satisfied: for each rank 1 pure subgroup H=A of G=A, there
is a finite rank Butler subgroup B of G such that H D A C B.
Furthermore, A is called decent in G if the same conclusion holds whenever
H=A is any finite rank pure subgroup of G=A. Evidently, decent subgroups are
prebalanced, but balanced subgroups need not be decent as is shown by (c) in the
following example.
Example 2.1.
(a) Every pure subgroup in a finite rank Butler group B is decent, and hence prebalanced: if
C < C0 are pure subgroups of B with finite rk C0 =C, then C0 D C C C0 with C0 a finite rank
Butler group.
(b) The indecomposable group A in Example 4.3 in Chapter 12 has no proper balanced subgroups
¤ 0, but all of its proper pure subgroups are prebalanced and decent.
(c) Let C be any finite rank torsion-free group that is not Butler; e.g. of Pontryagin type
(Lemma 4.6 in Chapter 12). If 0 ! A ! G ! C ! 0 is a balanced-projective resolution of
C, then A is (pre)balanced, but not decent in G.
It might be helpful if we insert the following remarks before proceeding with the
discussion.
(A) A pure-exact sequence 0 ! A ! G ! J ! 0 of torsion-free groups
with rk J D 1 is prebalanced exactly if there are a finite rank completely
decomposable group X D X1 ˚    ˚ Xn .rk Xi D 1/ and an epimorphism
 W X ! J that lifts to a homomorphism  W X ! G. This follows at once from
the definition.
(B) A pure subgroup A of the torsion-free group G is prebalanced if and only if, for
each g 2 G there is a finite subset fa1 ; : : : ; an g  A such that

hA; gi D A C hg C a1 i C    C hg C an i :

That is, G=A .g C A/ is of the same type as G .g C a1 / _    _ G .g C an /. As

G=A .g C A/ is the supremum of all G .g C a/ with a 2 A, we can increase
538 14 Butler Groups

the subset of the ai by adjoining finitely many elements anC1 ; : : : ; anCk 2 A to

conclude that

G=A .g C A/ D G .g C a1 / _    _ G .g C anCk /:

Thus if A is prebalanced in G, then G=A .g C A/ is the join of a finite number

of characteristics G=A .g C ai / .ai 2 A/.
(C) A subgroup A of corank 1 in a torsion-free group G is prebalanced if and only
if, in the relative balanced-projective resolution (Sect. 3 in Chapter 12)

0!K !A˚C !G!0

the completely decomposable group C can be chosen to be of finite rank. For

a proof, it suffices to take into consideration that G is generated by A and the
image of C; the latter is a Butler group provided rk C < 1.
We leave the proof of Lemma 2.2 as an Exercise 2.
Lemma 2.2. Let B
A be pure subgroups of the torsion-free group G. Then the
following holds:
(i) If B is prebalanced in G, then it is prebalanced in A.
(ii) If B is prebalanced in A, and A is prebalanced in G, then B is a prebalanced
subgroup in G.
(iii) If A is prebalanced in G, then so is A=B in G=B.
(iv) If B is prebalanced in G, and A=B is prebalanced in G=B, then A is
prebalanced in G. t
u
Prebalanced-Exact Sequences An immediate consequence of this lemma is
that prebalanced-exact sequences form a ‘proper class’: in the group Ext.C; A/
of extensions of A by a torsion-free group C, the extensions represented by
prebalanced-exact sequences form a subgroup; it will be denoted by PBext1 .C; A/.
Evidently, it contains Bext1 .C; A/, the group of balanced extensions, discussed in
Sect. 2 in Chapter 12.
This brings us to the exact sequence of central importance. The maps between
the PBexts are precisely the restrictions of induced maps between the Exts. Like for
Bext, here we have higher functors PBextn to restore exactness to the right, but we
will ignore them for n > 1.
Theorem 2.3. Let 0 ! A ! B ! C ! 0 be a prebalanced-exact sequence of
torsion-free groups. For a torsion-free G and arbitrary H, we have the following
exact sequences:

0 ! Hom.G; A/ ! Hom.G; B/ ! Hom.G; C/ !

! PBext1 .G; A/ ! PBext1 .G; B/ ! PBext1 .G; C/ ! : : :
2 Prebalanced and Decent Subgroups 539

and

0 ! Hom.C; H/ ! Hom.B; H/ ! Hom.A; H/ !

! PBext1 .C; H/ ! PBext1 .B; H/ ! PBext1 .A; H/ ! : : : :

Proof. We refer to Mac Lane[M] for relative homological algebra. t

u
Unfortunately, there are not enough prebalanced-projectives (only the free groups
are such), and therefore we are unable to form prebalanced-projective resolutions.
This deficiency, however, does not influence much the usefulness of this notion in
the theory of Butler groups: we can always rely on balanced-projective resolutions.
More information about PBext is furnished by the following lemma.
Lemma 2.4 (Giovannitti [1]). Let G be a torsion-free group. Then PBext1 .G; B/
is a divisible subgroup of Ext.G; B/ that contains the torsion part of Ext.G; B/.
Proof. The group property of PBext1 .G; B/ has been established above. By
Sect. 3(F) in Chapter 9, Ext.G; B/ is divisible. To complete the proof, it suffices
to show that if me 2 PBext1 .G; B/ for some e 2 Ext.G; B/ and integer m > 0;
then e 2 PBext1 .G; B/ likewise (this will include the me D 0 case). Consider the
commutative diagram with exact rows:

ξ
X −−−−→ J
⏐ ⏐
⏐
η
⏐inj

α
m : 0 −−−−→ B −−−−→ A −−−−→ G −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
 φ ṁ
α
: 0 −−−−→ B −−−−→ A −−−−→ G −−−−→ 0,

and suppose that the upper row is prebalanced-exact. Let J be a rank 1 pure
subgroup of G. By (A), we find a finite rank completely decomposable group X,
and an epimorphism  W X ! J that factors through A0 , i.e. ˛ 0  D  for some
 W X ! A0 : Now ˛ maps X onto mJ which is of finite index in J. Thus, using a
finitely generated free group F, we can extend ˛ W X ! mJ to a surjective map
X ˚F ! J. This map clearly lifts to X ˚F ! A, proving the prebalanced-exactness
of e. t
u
Though the group PBext1 .C; A/ is in general much larger than its subgroup
Bext1 .C; A/, surprisingly, for torsion groups A these two groups coincide. We give
here a trivial proof for this frequently needed fact.
Lemma 2.5 (Fuchs–Metelli [3]). A prebalanced-exact sequence

0!T !G!C!0
540 14 Butler Groups

with C torsion-free is necessarily balanced-exact whenever T is torsion.

Proof. We apply the definition of prebalancedness to A D T and to a pure subgroup
H of G such that T < H and rk H=T D 1. Then the sum H D T C B (for a Butler
group B) must be direct, since T \ B D 0. Here B has to be of rank 1, so T is
balanced in H (and hence in G). t
u
This lemma is applied to record a remarkable property of Butler groups. The
proof for countable rank is as simple as for the finite rank case.
Proposition 2.6. Every pure subgroup B of a countable completely decomposable
group satisfies Bext1 .B; T/ D 0 for all torsion groups T.
Proof. Let B be as stated, and 0 D B0 < B1 <    < Bn < : : : a chain of pure
subgroups with union B where rk BnC1 =Bn D 1 .n < !/. Then the Bn are Butler
groups, and 0 ! Bn ! BnC1 ! BnC1 =Bn ! 0 is a prebalanced-exact sequence, so
Theorem 2.3 yields the exactness of

Bext1 .BnC1 =Bn ; T/ ! Bext1 .BnC1 ; T/ ! Bext1 .Bn ; T/

for every torsion group T, where we have applied Lemma 2.5 by switching PBext
to Bext. The first Bext vanishes since BnC1 =Bn is of rank 1. A simple induction on n
leads to Bext1 .Bn ; T/ D 0 for n < !. Like in the proof of Lemma 4.1 in Chapter 9,
hence we can conclude that Bext1 .B; T/ D 0 for all T. t
u
Decent Subgroups On several occasions, the stronger version of prebalanced-
ness will be required: decent subgroups. Again, we leave the proof to the reader
that decent subgroups enjoy the same properties (Lemma 2.2) as balancedness and
prebalancedness (Exercise 3).
We have already observed that decent subgroups are prebalanced. However, for
the converse an additional condition is needed, as is clear from the following result,
where by a finitely Butler group we mean a group all of whose finite rank pure
subgroups are Butler groups.
Lemma 2.7 (Fuchs–Viljoen [1]). A pure subgroup A of a torsion-free group G is
decent if and only if
(i) A is prebalanced in G, and
(ii) G=A is finitely Butler.
Proof. Suppose A is decent in G. Then (i) is obvious. To check (ii), let C=A be
a finite rank pure subgroup of G=A, and H a finite rank Butler group such that
C D A C H. Then C=A Š H=.A \ H/ is Butler, as guaranteed by Theorem 1.4.
Conversely, assume (i) and (ii) are satisfied, and let C=A be a finite rank pure
subgroup of G=A. By (ii), it is a Butler group, so C=A D C1 =A C    C Ck =A with
rank 1 subgroups Ci =A. By (i), for each i, Ci D A C Bi with a finite rank Butler
group Bi . But then C D A C B where B D B1 C    C Bn is a finite rank Butler
group. t
u
3 The Torsion Extension Property 541

F Notes. Prebalancedness was introduced by Richman [6] under the name ‘semi-
balancedness,’ and rediscovered by Fuchs–Viljoen [1]. Its homological properties were explored
by Bican–Fuchs [2]. Giovannitti [2] investigates, for Butler groups B; C, certain subgroups of
PBext.C; B/ that are defined in terms of lattices of types.
Easy examples show that the union of an ascending chain of prebalanced subgroups need not
be prebalanced. All these counterexamples are essentially countable unions, because the following
holds. Suppose that 0 D B0 < B1 <    < B < : : : . < S
/ is a smooth properly ascending chain
of prebalanced subgroups of the group G. The union B D < B is prebalanced in G whenever
cf 
!1 (cp. Lemma 2.7 in Chapter 12).
The notion of regular subgroup is due to Bican [2]: regularity means that the elements have
the same types in the subgroup as in the entire group.

Exercises
˛
(1) Let 0 ! B ! G ! C ! 0 . < / be a direct system of prebalanced-
exact sequences such that the connecting maps  W C ! C . <  < / are
all monic with Im  pure in C . Then the direct limit of the system is likewise
prebalanced-exact.
(2) Prove Lemma 2.2.
(3) Let B < A be pure subgroups in the torsion-free group G.
(a) If B is decent in G, then it is decent in A.
(b) If B is decent in A, and A is decent in G, then B is decent in G.
(c) If A is decent in G, then A=B is decent in G=B.
(d) If B is decent in G, and A=B is decent in G=B, then A is decent in G.
(4) Prove the analogue of (C) for decent subgroups.
(5) Give an explicit example of
(a) a pure subgroup that is not prebalanced;
(b) a prebalanced subgroup that is not decent; and
(c) a prebalanced subgroup that is not balanced.
(6) Extend (Lemma 2.5) to the case when T is a decent subgroup.
(7) (Kravchenko) Assume B is a finite rank Butler group. An exact sequence 0 !
A ! B ! C ! 0 is balanced-exact if and only if, for all types t, A C B.t/ is
pure in B.

3 The Torsion Extension Property

We turn our attention to a third vital ingredient in our study, the torsion extension
property. Unlike prebalancedness, it is not a generalization of balancedness.
TEP-Subgroups A subgroup A of a torsion-free group G is said to be a
TEP-subgroup, or to have the TEP in G, if every homomorphism A ! T .T
any torsion group) extends to a homomorphism G ! T; that is, if the induced map
542 14 Butler Groups

Hom.G; T/ ! Hom.A; T/ is surjective for all torsion T. This is easily seen to be

equivalent to the following: for every subgroup C of A for which A=C is torsion,
A=C is a summand of G=C.
Observe that only a pure subgroup can be a TEP-subgroup. In fact, if A is not
pure in G, then there is a prime p such that pA < A \ pG. It is then easy to find a
homomorphism of A into the cyclic group Z.p/ of order p which cannot be extended
to G.
Needless to say, direct summands are TEP-subgroups. A few elementary remarks
on the new concept are in order.
Lemma 3.1. Let B < A be pure subgroups of the torsion-free group G. Then the
following holds:
(i) If B is TEP in G, then it is TEP in A.
(ii) If B is TEP in A, and A is TEP in G, then B is a TEP-subgroup in G.
(iii) If A is TEP in G, then so is A=B in G=B.
(iv) If B is TEP in G, and A=B is TEP in G=B, then A is TEP in G.
Proof. All claims are straightforward with the exception of (iv). So let T be a torsion
group, and ˛ W A ! T a homomorphism. By hypothesis, the restriction ˛  B W
B ! T extends to a homomorphism W G ! T. Now ˛ induces a map A=B ! T,
which extends to a homomorphism ı W G=B ! T. If  W G ! G=B denotes the
natural map, then  ı W G ! T satisfies .  ı/a D a  .  ˛/a D ˛a for
every a 2 A. t
u
Example 3.2.
(a) Let G be a finite rank torsion-free group that is not Butler, and 0 ! A ! C ! G ! 0
a balanced-projective resolution of G, with completely decomposable C. The subgroup A is
balanced in C, but fails to have the TEP, as it will be evident from Lemma 3.3 below.
(b) Pure subgroups of finite rank Butler groups are TEP-subgroups, since all mixed factor groups
are splitting. This is made precise in Corollary 3.6.

The Basic Lemma Most significant is the property of finite rank Butler groups
which is the content of the next theorem. Parts of its proof are separated as
preliminary lemmas; they cover countable groups as well; at one point in the proof
of Theorem 3.5 we will need this more general result.
Lemma 3.3 (Dugas–Rangaswamy [1]). Let 0 ! A ! G ! B ! 0 be a
prebalanced-exact sequence, where G satisfies Bext1 .G; T/ D 0 for all torsion
groups T. Then Bext1 .B; T/ D 0 holds for all torsion T if and only if A has TEP
in G.
Proof. The given sequence induces the sequence Hom.G; T/ ! Hom.A; T/ !
Bext1 .B; T/ ! Bext1 .G; T/ D 0 which is exact for any torsion T (Theorem 2.3);
we have already replaced PBext by Bext as permitted by Lemma 2.5. The map
between the two Bexts is monic for each torsion T exactly if A is TEP in G. t
u
The key technical ingredient in the proof of Theorem 3.5 is the following lemma.
3 The Torsion Extension Property 543

Lemma 3.4 (Dugas–Rangaswamy [2]). Suppose G is a countable torsion-free

group such that Bext1 .G; T/ D 0 for all torsion T. Every TEP-subgroup in G that is
of finite corank is decent in G.
Proof. Assume A is a pure subgroup of finite corank in G. Let G D fJi j i < !g be
the set of all rank 1 pure subgroups in G, and E a finitely generated free subgroup
of G such that G=K is a torsion group, where K D A C E. Let ˇ W G ! G=K denote
the canonical projection. Given a family fni j i < !g of positive integers, for each i
define Ci D Ji =ni .Ji \ K/, and i as the composite map

i W Ci D Ji =ni .Ji \ K/ ! Ji =.Ji \ K/ ! .K C Ji /=K ! G=K

(all maps are canonical). Finally, set

C D ˚i<! Ci ; D ˚i<! i W C ! G=K; and T D Ker :

Let H be the pull-back of and ˇ in the diagram

β : 0 −−−−→ T −−−−→ H −−−−→ G −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐β
  
γ
: 0 −−−−→ T −−−−→ C −−−−→ G/K −−−−→ 0.

To see that eˇ is balanced-exact, take any map  W J ! G where J is torsion-

free of rank 1. Clearly, ˇ maps J into some i Ci
G=K, and hence  lifts to a
homomorphism  W J ! C such that  D ˇ. Then by the pull-back property,
there exists a map J ! H establishing the balanced-exactness of eˇ. In view of the
hypothesis on G, the top row splits, giving rise to a morphism ˛ W G ! C such that
˛ D ˇ.
Next suppose A is TEP-subgroup in G. Since ˛A
T; the map ˛  A extends
to a homomorphism ˛ 0 W G ! T. Clearly, ˛  K need not coincide with ˛ 0  K.
But if  W T ! T=F is the canonical projection with kernel F D .˛  ˛ 0 /E, then
.˛  K/ D .˛ 0 /  K. We conclude that in the commutative diagram

0 −−−−→ K −−−−→ G −−−−→ G/K −−−−→ 0

⏐ ⏐ 
⏐α K ⏐α 
  
γ
: 0 −−−−→ T −−−−→ C −−−−→ G/K −−−−→ 0
⏐ ⏐ 
⏐ζ ⏐ 
  
δ
ζ : 0 −−−−→ T /F −−−−→ C/F −−−−→ G/K −−−−→ 0
544 14 Butler Groups

with exact rows and columns the bottom sequence splits, thus there is a map  W
G=K ! C=F with ı D 1G=K .
If G=K D .K C J1 C    C Jj /=K holds for some j < !, then G D A C B with
a Butler group B D E C J1 C    C Jj , and we are done. Otherwise, for each i < !
we can choose a gi 2 G with gi … Hi D .K C J1 C    C Ji /=K; say, of order qi . We
show that this leads to a contradiction.
We regress to the construction of C, and choose the integers ni in a particular
way. With the aid the qi just defined, we set ni D q1    qi for all i < !. As F
is a finite group, we have F
C1 ˚    ˚ Cj for some j < !. For such a j, let
.gj / D cjC1 C    C ck C F .ci 2 Ci /. From qj gj D 0 we obtain qj .gj / D
qj .cjC1 C    C ck / 2 F, whence qj .cjC1 C    C ck / D 0 follows. By construction,
qj jni for each i  j, whence qj ci D 0 implies

ci 2 Ci Œni
D .Ji \ K/=ni .Ji \ K/ D Ker i
Ker :

Then gj D ı.cjC1 C  Cck CF/ D .cjC1 C  Cck / 2 Hj ; an obvious contradiction.

t
u
Butler Groups and Bext After all this preparation, it is easy to verify the
following theorem whose extension to the infinite case reshaped the theory of Butler
groups.
Theorem 3.5 (Bican [2], Bican–Salce [1]). A finite rank torsion-free group B is a
Butler group if and only if it satisfies

Bext1 .B; T/ D 0 for all torsion groups T:

Proof. The necessity is a special case of Proposition 2.6. Conversely, assume that B
is of finite rank and satisfies Bext1 .B; T/ D 0 for all torsion T. Consider a balanced-
projective resolution 0 ! A ! C ! B ! 0 of B. Here C is a countable completely
decomposable group, and by Lemma 3.3 B satisfies Bext1 .B; T/ D 0 for all torsion
T only if A is a TEP-subgroup of C. Lemma 3.4 applies to conclude that A is then
decent in C. An appeal to Lemma 2.7 leads us to the conclusion that B Š C=A is a
(finitely) Butler group, completing the proof. t
u
We give an alternative proof for the necessity part of the preceding theorem. Let
B be a finite rank Butler group, and … D …1 [    [ …k a partition of the set
of primes such that the groups B ˝ Z…` are completely decomposable groups (cf.
Theorem 1.8). If 0 ! T ! A ! B ! 0 is a balanced-exact sequence with T
torsion, then, for every ` D 1; : : : ; k, the sequence

0 ! T ˝ Z…` ! A ˝ Z…` ! B ˝ Z…` ! 0

3 The Torsion Extension Property 545

is also balanced-exact, so it splits. If ` W A ˝ Z…` ! T ˝ Z…` is a splitting map for

the inclusion T ˝ Z…` ! A ˝ Z…` , then the composition


A ! ˚k`D1 A ˝ Z…` ! ˚k`D1 T ˝ Z…` D T

is a splitting map for the inclusion T ! A, where  D 1 ˚    ˚ k .

Finally, we prove a corollary. If B is a finite rank Butler group, then for each pure
subgroup C, the factor group B=C is a Butler group, so Bext1 .B=C; T/ D 0. From
Lemma 3.3 we conclude:
Corollary 3.6 (Bican [3]). In finite rank Butler groups, all pure subgroups are
TEP-subgroups. t
u
F Notes. That finite rank Butler groups have trivial balanced extensions was observed by
Bican [2], using complicated computations, and the idea of homological characterization in terms
of Bext is due to Bican–Salce [1]. This started an extensive study of infinite rank Butler groups
that was in the center of investigations for more than two decades in the twentieth century. See the
Notes to Sect. 1.
The Torsion Extension Property was observed by Procházka [1], and used by Bican [3] in
an equivalent form. It was more fully developed by Dugas–Rangaswamy [1] and Dugas–Hill–
Rangaswamy [1].

Exercises

(1) Supply the details for the proofs of (i)–(iii) in Lemma 3.1.
(2) The condition “all pure subgroups in A are TEP-subgroups” does not imply that
a finite rank torsion-free group A has to be a Butler group. [Hint: Pontryagin’s
example for a rank two indecomposable group.]
(3) (a) A completely decomposable group of countable rank contains pure sub-
groups that do not have the TEP.
(b) Every torsion-free group of infinite rank contains a pure subgroup that is
not a TEP-subgroup.
(4) For torsion-free groups A; C, the extensions of A by C in which A is a TEP-
subgroup form a subgroup of Ext.C; A/.
(5) (a) In every extension of Z by a torsion-free group, Z is a TEP-subgroup.
(b) Let R denote the group of rationals with square-free denominators. Except
for the splitting extension, none of the extensions of R by Q contains R as
a TEP-subgroup.
546 14 Butler Groups

4 Countable Butler Groups

Bican–Salce [1] introduced two generalizations of finite rank Butler groups to

groups of arbitrary ranks. There is no general agreement which of these should be
called Butler groups, so we follow the convention, and call them B1 - and B2 -groups,
respectively.
B1 - and B2 -Groups We will say that a torsion-free group B (of any rank) is a
B1 -group if

Bext1 .B; T/ D 0 for all torsion groups TI (14.2)

and a B2 -group if there is a smooth chain of pure subgroups,

0 D B0 < B1 <    < B <    < B D [ < B ; (14.3)

with finite rank factors such that, for each  < , B is decent in B C1 , i.e. B C1 D
B C G for some finite rank (pure) Butler subgroup G . Note that the factor groups
B C1 =B in (14.3) are finite rank Butler groups.
There is a third concept which we have already introduced before and is relevant
to the theory: a torsion-free group all of whose finite rank pure subgroups are Butler
groups is called finitely Butler.
An important remark is called for regarding terminology. My opinion is that it is not justified to
call infinite rank groups satisfying condition (14.2) or (14.3) Butler groups: Butler’s contribution
is immense, but limited to the finite rank case. It is tempting to rename them Bican–Salce groups,
since Bican and Salce brought the idea of balancedness into the picture. However, I hesitated to
change the adopted terminology, because this would cause difficulty in relating our presentation to
the respective literature. Notwithstanding, I feel strongly that the terminology in the infinite rank
case is incorrect. (May I quote Ovid: Video meliora proboque, deteriora sequor.)
Example 4.1. The union B of an ascending chain 0 D B0 < B1 <    < Bn < : : : .n < !/ of
finite rank Butler groups is a B2 -group provided each Bn is pure in BnC1 . B is evidently finitely
Butler, and from Theorem 4.2 below it will follow that it is also a B1 -group.

Countable Butler Groups Focusing now our attention on the countable case,
we prove the following main result which summarizes the precise relation between
the three concepts under consideration.
Theorem 4.2 (Bican–Salce [1]). For a countable rank torsion-free group B the
following are equivalent:
(i) B is finitely Butler;
(ii) B is a B2 -group;
(iii) B is a B1 -group.
Proof. (i) ) (ii) B is the union of a countable chain 0 D B0 < B1 <    < Bn < : : :
of pure subgroups where Bn is of rank n. By hypothesis, they are Butler groups. As
Bn is decent in BnC1 , (ii) holds for B.
4 Countable Butler Groups 547

(ii) ) (iii) B is the union of a countable chain (as in (i)), where the factor groups
are finite rank Butler groups, so by Theorem 3.5 they satisfy (iii). We argue just
as in the proof of Lemma 4.1 in Chapter 9, using the fact that T stays balanced in
its extension by BnC1 , if Bn factored out. Thus B satisfies (iii).
(iii) ) (i) Let 0 ! A ! C ! B ! 0 be a balanced-projective resolution of the
B1 -group B, where C is a countable completely decomposable group. Pick a finite
rank pure subgroup G of B, and form the commutative diagram

0 −−−−→ A −−−−→ H −−−−→ G −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  
0 −−−−→ A −−−−→ C −−−−→ B −−−−→ 0

where the right-hand square is a pull-back. H is finitely Butler as a pure subgroup

of the completely decomposable group C, hence it is a B1 -group by the first two
steps of this proof. Owing to Lemma 3.3, A has TEP in C, so it is a TEP-subgroup
of finite corank in H. Lemma 3.4 implies that A is decent in H, and hence G Š H=A
is a Butler group. t
u
This theorem gives a clear picture of the groups under consideration in the
countable case. As a consequence, there is no ambiguity in calling a countable group
a Butler group to mean either a B1 - or a B2 -group, or else a finitely Butler group.
Since the property of being finitely Butler is evidently inherited by pure
subgroups, we obtain at once:
Corollary 4.3 (Bican–Salce [1]). Pure subgroups of countable Butler groups are
themselves Butler groups. u
t
In Corollary 3.6 we have seen that pure subgroups of finite rank Butler groups
are TEP-subgroups. Now we raise the question as to when a pure subgroup of
a countable Butler group is a TEP-subgroup. Posing this question gives us the
opportunity of gaining a better insight into the interplay of three important concepts.
Corollary 4.4 (Dugas–Rangaswamy [1]). Let B be a countable Butler group, and
A a pure subgroup of B. The following are equivalent:
(a) A is a TEP-subgroup of B;
(b) A is decent in B;
(c) A is prebalanced in B and B=A is a Butler group.
Proof. Assuming (a), let B0 be pure in B such that A
B0 and B0 =A is of finite rank.
Then B0 is Butler by Corollary 4.3, and (b) follows from Lemma 3.4. Because of
Theorem 4.2, the equivalence of (b) and (c) is a consequence of Lemma 2.7. If A
satisfies (c), then Lemma 3.3 implies that also (a) holds. t
u
F Notes. Countable Butler groups are important transitions from finite rank Butler groups to
the general case: they share some of the relevant properties of the finite rank groups, most of which
get lost for large cardinalities as we shall see in the following sections. We would like to make
548 14 Butler Groups

it clear: countable Butler groups are not necessarily pure subgroups in completely decomposable
groups. Indeed, Arnold–Rangaswamy [1] give an example of a countable rank Butler group that
fails to be a pure subgroup in any completely decomposable group.

Exercises

(1) The union of a chain of pure finitely Butler subgroups is also finitely Butler.
(2) A TEP-subgroup in a B2 -group is a B2 -group.
(3) (Bican–El Bashir) Let A be a torsion-free group, and A D B C C where B; C are
B1 -groups. If B \ C is TEP in B, then A is also a B1 -group.
(4) Let 0 D G0 < G1 <    < Gn < : : : be a countable chain of countableS Butler
groups such that each link is pure in the next one. Then the union G D n<! Gn
is likewise a Butler group. [Hint: Theorem 4.2.]
(5) (Dugas–Rangaswamy) A pure subgroup of a Butler group of countable rank is
a decent subgroup if and only if it is a TEP-subgroup.
(6) The tensor product of two finitely Butler groups is again finitely Butler.

5 B1 - and B2 -Groups

In the preceding section, we have defined B1 - and B2 -groups, and now we wish to
learn more about them without cardinality restriction. So forget about countability,
and refer to a transfinite chain of pure subgroups in the B2 -group B,

0 D B0 < B1 <    < B <    < B D [ < B D B (14.4)

where the factors are of finite rank, and B is decent in B C1 . < /.
The following claims are evident from the definitions.
(A) Completely decomposable groups are both B1 -, B2 -groups, and also finitely
Butler.
(B) Direct sums of B1 - .B2 /-groups .finitely Butler groups/ are again of the same
kind.
(C) Summands of B1 -groups .finitely Butler groups/ are again B1 -groups .finitely
Butler/. (For the same for B2 -groups, see Corollary 5.5.)
(D) For each  < , both the subgroups B and the factor groups B=B in (14.4)
are B2 -groups. For factor groups, this is a consequence of the preservation of
decency when passing to the stated factor group.
(E) A torsion-free group B is a B1 -group exactly if it has the projective property
relative to all balanced-exact sequences 0 ! T ! G ! C ! 0 with T
torsion and C torsion-free. This is simply a reformulation of the definition of
B1 -groups.
5 B1 - and B2 -Groups 549

(F) Homogeneous B2 -groups are completely decomposable. Indeed, if the finite

rank Butler pure subgroups G in the definition of B2 -groups (recall: B C1 D
B C G in (14.4)) are homogeneous of the same type t, then by Corollary 1.5
they are completely decomposable. Furthermore, then the intersection B \G
is a completely decomposable summand in G , G D .B \G /˚H for some
H which is likewise t-homogeneous completely decomposable. Therefore,
B C1 D B ˚ H ; and B is the direct sum of the H . < /.
(G) Prebalanced extensions of B2 -groups by B2 -groups are B2 -groups. By
Lemma 2.7, a prebalanced A is decent if G=A is B2 , so a chain of decent
subgroups of A can be continued by such a chain of G=A to reach G.
(H) Let 0 ! A ! B ! C ! 0 be a prebalanced-exact sequence, and B a
B1 -group. C is a B1 -group exactly if A is TEP in B. Cp. Lemma 3.3.
If we wish, we may admit in (14.4) factors of countable rank:
Lemma 5.1 (Dugas–Rangaswamy [1]). A torsion-free group B of cardinality  is
a B2 -group if and only if it admits a smooth chain (14.4) such that
(i) B C1 =B is countable for every  < ;
(ii) B is decent in B C1 .or in B/ for every  < .
Proof. Note that by transfinite induction it follows that all B in (14.4) are decent
in the B2 -group B.
Necessity being obvious, in order to prove sufficiency suppose that (14.4) is a
smooth chain satisfying (i)–(ii). There is a finite or countably infinite chain of pure
subgroups between B and B C1 , say, B D C0 < C1 <    < Cn <    <
C! D B C1 with rank 1 factors CnC1 =Cn . If Xn is a finite rank Butler group with
CnC1 D B C Xn , then also CnC1 D Cn C Xn , thus Cn is decent in CnC1 . Hence any
chain (14.4) with (i)–(ii) can be refined to a chain as required in the definition of
B2 -groups. t
u
Prebalancedness is particularly suited to deal with B1 -groups as well.
Lemma 5.2. Prebalanced extensions of B1 -groups by B1 -groups are B1 -groups.
Moreover, if (14.4) is a chain in B such that
(i) each subgroup B is prebalanced in its immediate successor, and
(ii) all the factors B C1 =B are B1 -groups,
then B is likewise a B1 -group.
Proof. If 0 ! A ! B ! C ! 0 is a prebalanced-exact sequence of torsion-
free groups, then for a torsion group T, the induced sequence Bext1 .C; T/ !
Bext1 .B; T/ ! Bext1 .A; T/ is exact by Theorem 2.3; in view of Lemma 2.5, we
could replace PBext by Bext. If A and C are B1 -groups, then the two extremal Bexts
vanish, so Bext1 .B; T/ D 0 for all torsion T.
The second claim follows from the first if B is the union of a finite chain. For
infinite chains, we argue as in Eklof’s Lemma 4.1 in Chapter 9. t
u
550 14 Butler Groups

The next result deals with the relation between B1 - and B2 -groups in general. One
direction the implication is rather easy.
Theorem 5.3. B2 -groups of any rank are B1 -groups.
Proof. For a countable chain, the claim follows from Proposition 2.6. For transfinite
chains, we again refer to the proof of Lemma 4.1 in Chapter 9. t
u
(J) In the chain (14.4), the decent subgroups B are TEP-subgroups of the
B2 -group B. This is an immediate consequence of Lemma 3.3 in view of the
preceding theorem.
Characterizations of B2 -Groups The next result establishes a larger supply of
decent subgroups in B2 -groups.
Theorem 5.4 (Albrecht–Hill [1]). A torsion-free group G is a B2 -group if and only
if it admits an H.@0 /-family B of decent subgroups. Equivalently, it has an H.@0 /-
family of decent TEP-subgroups.
Furthermore, if A; B 2 B and B < A, then A=B is a B2 -group.
Proof. One way the claim is obvious, since from an H.@0 /-family of decent
subgroups we can extract a smooth chain with countable factors which are
B2 -groups. Such a chain can be refined to a desired chain with finite rank factors.
For the converse, assume that G is a B2 -group. We appeal to Theorem 5.5 in
Chapter 1 to obtain an H.@0 /-family B, and to its proof to argue that, for blocked
subsets S, the subgroups G.S/ are decent and TEP-subgroups in G. So let C=G.S/
be a finite rank pure subgroup of G=G.S/. There is a finite blocked subset S0 that
contains representatives of a maximal independent set in C=G.S/. Then C D G.S/C
.C\G.S0 //, where the intersection is a pure subgroup of G.S0 /, so a finite rank Butler
group. Thus G.S/ is decent in G. It is also a TEP-subgroup, because by creating an
increasing sequence of blocked subgroups, we can find a chain like (14.4) passing
through G.S/ and climbing up to G; finally, we refer to (D).
The last argument implies that all the factors in B are B2 -groups. t
u
Equipped with this theorem, we can derive a not so obvious corollary (that we
could not claim in (C)).
Corollary 5.5. Summands of B2 -groups are B2 -groups.
Proof. Let A D G ˚ H be a direct decomposition of a B2 -group, and B an H.@0 /-
family of decent subgroups of A. Define the subfamily

B ? D fB 2 B j B D B \ G ˚ B \ Hg:

In view of Lemma 5.4 in Chapter 1, this is an H.@0 /-family of decent subgroups of

A, and the set fB \ G j B 2 B ? g is an H.@0 /-family of decent subgroups of G. u
t
It might help the reader to navigate a passage through the theory of general Butler groups,
through this rich and fascinating subject, if we explain briefly our approach and our limited
goal, before getting involved more sophisticated arguments. A principal aim of the theory is to
6 Solid Subgroups 551

understand the fine distinction between B1 - and B2 -groups of arbitrary cardinalities. It turns out
that the question whether or not the classes of B1 - and B2 -groups are identical has no definite
answer: it is undecidable in ZFC. We have shown above that B2 -groups are always B1 -groups,
and will provide a necessary and sufficient condition under which B1 -groups are B2 -groups (see
Theorem 8.2 below). Though it is of no practical use, still it will be discussed, because it is
instrumental in understanding why CH is a sufficient condition (Theorem 9.9). However, the proof
that it is undecidable in ZFC whether or not all B1 -groups are B2 -groups would require a significant
amount of preparations, which is beyond the scope of this book.
The equivalence problem of B1 - and B2 -groups is intrinsically tied to another, almost equally
important question as to when Bext2 .G; T/ vanishes for all torsion-free G and all torsion T. To
answer this question, we will give a couple of equivalent conditions in Proposition 8.8. A third
relevant question that is concerned with the subgroup problem will be addressed briefly: for which
subgroups is the B2 -property hereditary?
F Notes. It is still not known if Bext1 .B; T/ D 0 for all direct sums T of finite cyclic groups
implies that B is a B1 -group (problem raised by Arnold). To answer the question as to whether
or not Bext1 .B; T/ D 0 for all countable T entails that B is a B1 -group, we point out that this is
undecidable in ZFC (Dugas–Rangaswamy [1]). Indeed, choose B homogeneous of type Z, then
Bext1 .B; T/ D Ext1 .B; T/, so B is a Baer group, and for such groups the problem is undecidable
in ZFC, see Sect. 2 in Chapter 15.

Exercises

(1) In a B2 -group, every countable subgroup is contained in a countable TEP-

subgroup.
(2) Let 0 ! A ! B ! C ! 0 be a prebalanced-exact sequence of torsion-free
groups. A is a B1 -group if both B and C are B1 -groups. [Hint: (H).]
(3) (a) Prebalanced extension of a finitely Butler group by another such group is
finitely Butler.
(b) Let 0 D C0 < C1 <    < Cn < : : : .n < !/ be a sequence of finitely
Butler groups. If Cn is prebalanced in CnC1 for each n < !, then the union
C D [Cn is also finitely Butler.
(4) Corank 1 prebalanced subgroups are decent.
(5) (Dugas–Hill–Rangaswamy) A torsion-free group G of cardinality @1 is a B2 -
group if it admits a smooth chain of TEP-subgroups with countable quotients.
(6) Let 0 ! A ! C ! B ! 0 be a balanced-projective resolution of a B1 -group
B. If A is a B2 -group, then B is finitely Butler. [Hint: if G=A is finite rank pure in
C=A Š B, then A has TEP in G, apply Lemma 3.4, A is decent in G; thus G=A
is Butler.]

6 Solid Subgroups

We now step into the more complicated realm of uncountable Butler groups. We
need additional tools to understand what happens for larger cardinalities. As it turns
out, the crucial device in approaching the problem is a kind of chain that is more
552 14 Butler Groups

general than the one in the definition of B2 -groups. Our choice is based on a concept
that is an extension of prebalancedness to arbitrary cardinals. We shall need only the
countable version, so our focus will be on such extensions.
We start with a simple observation. Let G be a torsion-free group, and A a pure
subgroup of corank 1 in G. Consider the types t.J/ of those rank 1 pure subgroups J
of G which are not contained in A (thus disjoint from A). In the lattice T of all types,
we form the lattice ideal IGjA generated by all these types t.J/. Thus IGjA consists
of all types t satisfying t
t.J1 / _    _ t.Jm / for some finite set fJ1 ; : : : ; Jm g of
rank 1 pure subgroups disjoint from A. As the cardinality of T is the power of the
continuum, every ideal of T can be generated by at most 2@0 types.
Example 6.1.
(a) If A is prebalanced in G, then IGjA is finitely generated.
(b) For each  (@0    2@0 ), T contains an ideal I that can only be generated by exactly 
types. Indeed, consider a set † of almost disjoint countable subsets fS j  < g of the set
… of primes. (The existence of such a † was proved in Lemma 5.6 in Chapter 1.) For each
S 2 †, define the type t D .k2 ; : : : ; kp ; : : : / by setting kp D 1 or 0 according as p 2 S
or p … S . It is easily seen that

t  t1 _ : : : _ tk if and only if  2 f1 ; : : : ; k g:

Consequently, in the lattice T of all types, the ideal I generated by all t with  <  is as
desired.

Solid Subgroups Let A be a corank 1 pure subgroup in the torsion-free group G.

A is said to be solid in G, if the lattice ideal IGjA can be countably generated. More
explicitly, A is solid in G means that there is a countable set S D fJn j n < !g of
rank 1 pure subgroups in G, but not in A, such that for every rank 1 pure subgroup
J of G, not in A, we can find a finite subset fJn1 ; : : : ; Jnk g  S such that t.J/

t.Jn1 / _    _ t.Jnk /.
Example 6.2.
(i) Decent subgroups are solid.
(ii) If the typeset of a torsion-free group G is countable, then every pure subgroup of G is solid.

Pus point out right away that, if G; A are as before, then we have always G D
Let
A C n<! Jn where the groups Jn are rank 1 pure subgroups of G not in A. Indeed,
the characteristic of a coset g C A 2 G=A is the supremum of the characteristics of
suitably chosen countably many elements in the coset (for each prime p, the correct
p-height need to be obtained). But this does not mean that the ideal IGjA is at most
countably generated. A counterexample is given to substantiate this claim.
Example 6.3. Let A denote a free abelian group of rank @1 , say, with basis fa j  < !1 g, and I
an ideal of T that is @1 -generated. For each t 2 I select a characteristic  2 t . If B D Zb is an
infinite cyclic group, define G as the subgroup of the divisible hull of A ˚ B generated by A; B and
by elements making a C b of characteristic  . Then IGjA D I is not countably generated.

In some proofs it is useful to have an element-wise criterion available for solid

subgroups. Clearly, A is solid in G if for each g 2 G n A there exists a subset
6 Solid Subgroups 553

fan j n < !g of A (to which we shall refer as a solid set for g over A) along with
non-zero integers rn such that for each a 2 A

t.g C a/
t.r1 g C an1 / _    _ t.rk g C ank /

for some k < !.

We should note that the use of the rn is unavoidable (Rangaswamy). Indeed, let
G D Q.p/ x ˚ Q.q/ y with primes p ¤ q, and A D Z.x C y/ a pure subgroup. If
g D p1 x, then hq .g C A/ D 1, but for no a 2 A is g C a divisible by all powers
of q.
The following characterization of solid subgroups will be required.
Lemma 6.4. A pure subgroup A of corank one in the torsion-free group G is solid
in G if and only if G D A C H where H is a countable pure subgroup of G such that
every g 2 G n A satisfies

t.g/
t.x1 / _    _ t.xn /

for some finite set fx1 ; : : : ; xn g  H n A.

Proof. Let fgn gn<! be a complete set of representatives of G mod A. If A is solid
in G, then choose H as the purification of the subgroup generated by all gn and
all solid sets for the gn . Conversely, if H is as stated, then the set fa 2 A j 9 r 2
N with rg C a 2 Hg is a solid set for g 2 G n A. t
u
If G=A is torsion-free of arbitrary rank, then we define A to be solid in G
whenever A is solid in every pure subgroup C of G that contains A as a corank 1
subgroup. Solidity will be most relevant in our future discussions, we list some of
the main properties in the following two lemmas. For proofs we refer to Exercise 3.
Lemma 6.5 (Bican–Fuchs [2]). Let B < A be pure subgroups of the torsion-free
group G. Then the following holds:
(a) If B is solid in G, then it is solid in A.
(b) If B is solid in A, and A is solid in G, then B is solid in G.
(c) Let B be prebalanced in G. A is solid in G if and only if A=B is solid in G=B. u
t
Other properties of importance are as follows.
Lemma 6.6 (Bican–Fuchs [2]).
(i) Assume B < A are pure subgroups of the torsion-free group G with A=B
countable. If B is solid in G, then so is A as well.
(ii) The union of a countable ascending chain of solid subgroups is likewise solid.
Proof.
(i) Let fai gi<! be a complete set of representatives of A mod B. Given g 2 G n A,
for each of the countably many g C ai , select a solid set fbi1 ; : : : ; bik ; : : : g  B.
A simple calculation convinces us that fai C bi1 ; : : : ; ai C bik ; : : : .i < !/g  A
is a solid set for g over A.
554 14 Butler Groups

(ii) Let A1 <    < An < : : : be a countable ascending chain of solid subgroups
of G, and let A denote their union. If g 2 G n A, then a moment’s reflection
shows that the union of the solid sets for g over An (for each n) is a solid set for
g over A. t
u
The following generalization of Lemma 3.4 will be needed.
Lemma 6.7. Assume G is finitely Butler, and A is a solid TEP-subgroup in G. Then
A is decent in G.
Proof. The proof is identical with Lemma 3.4, only a single adjustment is needed: if
rk.G=A/ D 1, then choose for G the set of rank one pure subgroups in G generating
the ideal IGjA : t
u
Solid Subgroups in Completely Decomposable Group The following result
reveals a close connection between solidity and the property of being a B2 -subgroup.
Theorem 6.8 (Fuchs–Metelli–Rangaswamy[1]). Suppose G is a corank one pure
subgroup of the completely decomposable group C D ˚ < C .each C is of
rank 1/.
G is a B2 -group if and only if the lattice ideal I generated by the types t.C / with
C 6
G can be countably generated, i.e. G is solid in C.
Proof. Let G be as stated, and  W C ! C=G D R the canonical map where R is a
rational group containing 1. For the proof we may assume that G does not contain
any C . For  < , we set

C./ D ˚ < C and G./ D G \ C./:

For each  < ; pick c 2 C such that .c / D 1 2 R, and let C D hc  c i
for  <  < . Evidently, C is of type t.C / ^ t.C /, and G is generated by all
the C . Next we change the well-ordering of the C by first listing those countably
many C that are needed to generate C along with G, i.e. G C C.!/ D C.
For the proof of sufficiency, suppose the ideal I is countably generated. We may
in addition assume that the well-ordering is chosen such that the types t.C / with
 < ! generate I. Then from ! on, every G./ is prebalanced in G. C 1/. As
G.!/ is a countable B2 -group, the chain G.!/ <    < G./ < : : : . < / ensures
that G is likewise a B2 -group.
To prove necessity, let  > ! and G a corank 1 pure B2 -subgroup of C with
G \ C D 0 for all . By Theorem 5.4, G has an H.@0 /-family F of decent sub-
groups. Let F1 be a countable member of F that contains G.!/. There is a countable
index set I1 such that F1 < ˚ 2I1 C : We can find a countable F2 2 F containing
G\˚ 2I1 C˛ , and then a countable index set I2 such that I1  I2 and F2 < ˚ 2I2 C .
Continuing this back-and-forth argument, after ! steps we take unions to obtain an
F 2 F which is clearly equal to G \ ˚ 2I C for a countable index set I.
6 Solid Subgroups 555

Again changing the well-ordering if necessary, we may in addition assume that

there is a countable ordinal ı for which

F D G \ ˚ 2I C D G.ı/:

This subgroup being prebalanced in G, we infer that for each  2 C with   ı

there are a finite number of indices 1 ; : : : ; k <  such that G. C 1/ D G./ C
C1 C    C Ck . Then

t.C / D t.G. C 1/=G.// D .t.C / ^ t.C1 // _    _ .t.C / ^ t.Ck //

hence t.C /
t.C1 / _    _ t.Ck / follows. Since the generation of G./ involves
C only for  < , this inequality remains true no matter how C is selected from
the set of the remaining C (i.e.,   ). Hence the ideal I is in fact generated by
the types t.C / with  < ı. t
u
Example 6.9. If in the preceding theorem, the subgroup G is solid in C, then it is an example of a
solid subgroup that is not prebalanced.

We will use now relative balanced-projective resolutions (Sect. 3 in Chapter 12).

Proposition 6.10. For a corank 1 pure subgroup A of the torsion-free group G, the
following are equivalent:
(i) A is solid in G;
(ii) there is a relative balanced-projective resolution


0 ! K ! A ˚ C!G ! 0 (14.5)

with a countable completely decomposable group C;

(iii) the kernel K is a B2 -group in a prebalanced-exact sequence of the form (14.5)
where C is completely decomposable, and   A D 1A .

Proof. (i) , (ii) If H is as stated in Lemma 6.4, then let 0 ! B ! C!H ! 0
be a balanced-projective resolution of H with countable C. In this case, r.1A ˚ / W
A ˚ C ! G yields a sought-after resolution. Conversely, if C is as in (ii), then the
purification H of its image in G satisfies Lemma 6.4.
(iii) ) (i) If (14.4) is as stated in (iii), then starting from the middle row, we form
the commutative diagram
556 14 Butler Groups

A A
⏐ ⏐
⏐ ⏐
 
φ
0 −−−−→ K −−−−→ A ⊕ C −−−−→ G −−−−→ 0
 ⏐ ⏐
 ⏐ ⏐
  
0 −−−−→ K −−−−→ C −−−−→ G/A −−−−→ 0,

where the bottom row is obtained by factoring out A from A ˚ C and from G; this is
allowed because  is the identity on A. (We have also used the fact that A \ K D 0.)
As rk G=A D 1, K is a corank 1 pure subgroup in the completely decomposable
group C. By Theorem 6.8, K is a B2 -group if and only if the T-ideal I generated by
the rank 1 summands of C can be countably generated. As the summands of C are
coming from those rank 1 subgroups of G that are disjoint from A, this ideal I is
countably generated if and only if A is solid in G.
(i) ) (iii) If A is solid in G, then choose (14.4) to be a relative balanced-
projective resolution of G over A. The final argument in the preceding paragraph
convinces us that K must be a B2 -group. t
u
Example 6.11 (Bican [1]). The following group G is pure in a completely decomposable group,
but is not B2 . Consider the set of types t . < 2@0 / defined in Example 6.1 with the aid of an
almost disjoint set of primes. Select a rank 1 group C of type t , and let C D ˚<!1 C . If G is
a pure subgroup in C of corank 1 such that it does not contain any C , then it is homogeneous of
type .0; : : : ; 0; : : : /, but is not a B2 -group (Theorem 6.8). Evidently, G is not solid in C.
F Notes. Solid subgroups were introduced under the name of @0 -prebalanced subgroups by
Bican–Fuchs [2], where -balancedness was also discussed for cardinals  with @0    2@0 .
Here we have adopted a more colloquial terminology for the only case  D @0 treated here.
Earlier literature on infinite Butler groups used subgroups called ‘separable in the sense of Hill’
or ‘separative’ which probably are a bit easier to define than solid subgroups: a pure subgroup A
is separative in G if, for every g 2 G, there is a countable subset X A (depending on g) such
that for any a 2 A there is an x 2 X satisfying .g C a/  .g C x/ (cf. Sect. 7 in Chapter 11).
Thus separative subgroups are solid. We do not spend time here to sort out the precise relation
between separative and solid subgroups, we just wish to point out that assuming :CH, there exist
torsion-free groups of cardinality @2 with solid, but no separative chains; see Dugas–Thomé [1].

Exercises

(1) Suppose that  is the supremum of the characteristics fi j i 2 Ig. Then there is
a countable subset J in I such that  D supfi j i 2 Jg.
(2) Let C D ˚p Q.p/ for all primes p, and  W C ! Q such that   Q.p/
is the embedding. Use the idea of Theorem 6.8 to show that G D Ker  is
prebalanced, but not decent in C.
(3) Give detailed proof for Lemma 6.5.
7 Solid Chains 557

(4) A corank 1 solid subgroup of a B1 -group is a B1 -group.

(5) Assuming CH, a corank 1 pure subgroup of a B2 -group is a B2 -group if and only
if it is a solid subgroup.

7 Solid Chains

The solid exact sequences fail to form a proper class, consequently, they do not
support a relative homological machinery. However, solid chains will provide us
with powerful tools in the study of Butler groups of large cardinalities.
Solid Chains Let A denote a pure subgroup of the torsion-free group G. By a
solid chain from A to G we understand a smooth chain

A D G0 < G1 < : : : < G < : : : < G D G (14.6)

of solid subgroups of G, where all the factors G C1 =G are torsion-free of rank one.
We will say that G admits a solid chain if there is a solid chain from 0 to G.
(a) All B2 -groups admit solid chains. This is obvious, as decent subgroups are solid.
(b) If in the chain (14.6) , the subgroups G of G are solid, and the factors G C1 =G
have cardinalities
@1 , then the chain can be refined to a solid chain. In fact,
countable extensions of solid subgroups are again solid (see Lemma 6.6(i)), so
we can refine the factor groups to factors of rank 1.
(c) Torsion-free groups of cardinalities
@1 admit solid chains.
(d) If Type.G/ is countable, then G has a solid chain from every pure subgroup up
to G.
Lemma 7.1. Let A be a pure subgroup of a completely decomposable group C. If
the cardinality of Type.A/ is
@1 , then A admits a solid chain.
Proof. Without loss of generality, we may assume that C has typeset of cardinality

@1 . Such a C is the union of a smooth chain of completely decomposable

summands with countable typesets. The intersections of these summands with A
provide a solid chain for A. t
u
If we deny CH, then we can give an example of a torsion-free group of cardinality
@2 that does not admit a solid chain; see Lemma 8.6 below.
Existence of Solid Chains The following result is crucial for the rest of our
discussion of Butler groups.
Theorem 7.2 (Fuchs [21]). Let A be a pure subgroup of the torsion-free group G.
The following conditions are equivalent:
558 14 Butler Groups

(i) there exists a solid chain from A to G;

(ii) there is a balanced-exact sequence


0 ! B ! A ˚ C!G ! 0 (14.7)

where C is completely decomposable,   A is the inclusion, and B is a B2 -

group;
(iii) the same as (ii) with prebalanced-exact (14.7).
Proof. (i) ) (ii) We place ourselves in the following setting: we have a solid chain
A D G0 <    < G < G C1 <    < G with factors of rank 1. We build a relative
balanced-projective resolution of G with the aid of this chain as follows: from a
relative balanced-projective resolution 0 ! B ! A ˚ C ! G ! 0 of G
we form a relative balanced-projective resolution 0 ! B C1 ! A ˚ C C1 !
G C1 ! 0 of G C1 by choosing C C1 D C ˚ C for a suitable completely
decomposable group C and mapping C into G C1 such that in an arbitrarily
chosen direct decomposition of C , no rank 1 summand maps into G .
In this way, we obtain commutative diagrams with exact rows and columns:

0 0 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
0 −−−−→ Bσ −−−−→ A ⊕ Cσ −−−−→ Gσ −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
0 −−−−→ Bσ+1 −−−−→ A ⊕ Cσ ⊕ C ∗ −−−−→ Gσ+1 −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
0 −−−−→ Bσ+1 /Bσ −−−−→ C∗ −−−−→ Gσ+1 /Gσ −−−−→ 0
⏐ ⏐ ⏐
⏐ ⏐ ⏐
  
0 0 0

Evidently, B is balanced in B C1 , since it is balanced in A˚C which is a summand

of A ˚ C ˚ C . It is routine to check that at limit ordinals the direct limits provide
relative balanced-projective resolutions.
Now, if G is solid in G C1 , then the ideal I generated by the types of summands
in C has a countable set of generators, thus jC j D @0 may be assumed. In
the diagram, B C1 =B is a corank one pure subgroup of the countable completely
decomposable group C , so Theorem 6.8 guarantees that B C1 =B is a B2 -group.
Consequently, B is the union of a smooth chain 0 D B0 <    < B < B C1 < : : :
of subgroups such that each is balanced in its successor, and each of the factors
7 Solid Chains 559

B C1 =B admits a chain of decent subgroups. These chains lift to B creating a

desired chain for B, so (ii) holds even with balanced (14.7).
(ii) ) (iii) is trivial.
(iii) ) (i) Suppose (14.7) is as stated in (iii). B admits an H.@0 /-family B of
decent TEP-subgroups (Theorem 5.4). C is a completely decomposable group, so it
has an H.@0 /-family C of summands; furthermore, by (5.2), the torsion-free group
G=A has an H.@0 /-family G of pure subgroups. Dropping to suitable subfamilies
(by the usual straightforward, but tedious back-and-forth argument), filtrations of
B; C can be found, and one for G from A, such that the sequences 0 ! B !
A ˚ C ! G ! 0 (where B 2 B; C 2 C and G 2 G) are all prebalanced-exact,
and the factor groups C C1 =C are all countable. Setting C C1 D C ˚ C (with
countable completely decomposable C ), consider the commutative diagram with
prebalanced-exact rows

Bσ Bσ
⏐ ⏐
⏐ ⏐
 
0 −−−−→ Bσ+1 −−−−→ A ⊕ Cσ ⊕ C ∗ −−−−→ Gσ+1 −−−−→ 0
⏐ ⏐ 
⏐ ⏐ 
  
0 −−−−→ K −−−−→ Gσ ⊕ C ∗ −−−−→ Gσ+1 −−−−→ 0

where the vertical arrows are monic in the first set, and epic in the second set; the
first column is exact by the 3  3-lemma. Since B ; B C1 2 B, it follows from
Theorem 5.4 that K Š B C1 =B is a B2 -group. The bottom row is like the one in
Proposition 6.10, so we can conclude that G is solid in G C1 , and the proof is
complete. t
u
The special case A D 0 in the preceding theorem leads at once to a criterion we
have been looking for:
Theorem 7.3. A torsion-free group G admits a solid chain if and only if there is a
prebalanced-exact sequence 0 ! B ! C ! G ! 0 .where C is a B2 -group/ in
which B is a B2 -group.
Proof. If G admits a solid chain, then by Theorem 7.2 there is a balanced-exact
sequence with torsion-free C and B2 -group B. Conversely, let B; C be B2 -groups
in the bottom prebalanced-exact sequence. Using a surjective map  W C0 ! C
from a completely decomposable C0 with balanced kernel, we form the commutative
diagram with exact rows:
560 14 Butler Groups

0 −−−−→ B  −−−−→ C  −−−−→ G −−−−→ 0

⏐ ⏐ 
⏐ ⏐φ 
  
0 −−−−→ B −−−−→ C −−−−→ G −−−−→ 0.

Ker  is balanced in C0 , and hence also in B0 . By Theorem 7.2 it is a B2 -group, and

thus B0 is B2 as a balanced extension of the B2 -group Ker  by B0 . By Lemma 2.2(iv),
the top row is prebalanced-exact. Hence Theorem 7.2 implies that G admits a solid
chain. t
u
We have a noteworthy corollary to this theorem:
Corollary 7.4. If a torsion-free group admits a solid chain, then its summands also
admit solid chains.
Proof. Suppose G D G1 ˚ G2 has a solid chain. Choose balanced-projective
resolutions 0 ! Ai ! Ci ! Gi ! 0 .i D 1; 2/, where the Ci are completely
decomposable. Then 0 ! A1 ˚ A2 ! C1 ˚ C2 ! G ! 0 is a balanced-projective
resolution of G. By Theorem 7.3, A1 ˚ A2 is a B2 -group, so by Corollary 5.5 the
groups Ai are also B2 -groups. Apply Theorem 7.3 again to conclude that the groups
Gi admit solid chains. t
u
Next we verify an analogue of Lemma 6.5.
Lemma 7.5. Suppose B < A are pure subgroups of the torsion-free G. Then the
following holds:
(a) If there are solid chains from B to A and from A to G, then there is also one from
B to G.
(b) Assume B is balanced in G. There is a solid chain from A to G if and only if
there is one from A=B to G=B.
Proof. The proofs are similar for both claims, making use of Theorem 7.2. We give
details for (b) only.
(b) Starting from the relative balanced-exact resolution in the top row, we factor
out B to obtain the bottom row in the commutative diagram

0 −−−−→ K −−−−→ A⊕C −−−−→ G −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  
0 −−−−→ K −−−−→ A/B ⊕ C −−−−→ G/B −−−−→ 0.

As B is balanced in G, the bottom row is also balanced-exact, so it is a relative

balanced-exact resolution of A=B in G=B. Both solid chains in question exist exactly
if K is a B2 -group (Theorem 7.2). t
u
The following lemma is a crucial ingredient in the proof of Corollary 7.7.
7 Solid Chains 561

Lemma 7.6 (Bican–Fuchs [2]). Suppose 0 ! T ! G ! A ! 0 is a balanced-

exact sequence where T is a torsion group. If A is a subgroup in a torsion-free
group B such that there is a solid chain leading from A up to B, then there exists a
commutative diagram with balanced-exact rows

0 −−−−→ T −−−−→ G −−−−→ A −−−−→ 0

 ⏐ ⏐
 ⏐ ⏐
  
0 −−−−→ T −−−−→ H −−−−→ B −−−−→ 0.

Proof. The claim is equivalent to the statement that the map Bext1 .B; T/ !
Bext1 .A; T/ induced by the inclusion A ! B is surjective. Consider a relative
balanced-projective resolution 0 ! K ! A ˚ C ! B ! 0; where C is completely
decomposable. The existence of a solid chain from A to B implies that K is a B2 -
group. In the induced exact sequence

Bext1 .B; T/ ! Bext1 .A ˚ C; T/ D Bext1 .A; T/ ! Bext1 .K; T/

(for any torsion group T) the last term vanishes in view of Theorem 5.3. The
balancedness of the bottom sequence follows straightforwardly (it is easier if we
add ˚C to both G and A). Hence the claim follows. t
u
A sufficient criterion for a pure subgroup of a B1 -group to be again such a group
can be given immediately.
Corollary 7.7. Let B be a B1 -group, and A a pure subgroup of B. If there is a solid
chain from A to B, then A is likewise a B1 -group.
Proof. Suppose there is a solid chain from A to B. Lemma 7.6 implies that any
balanced-exact sequence 0 ! T ! G ! A ! 0 can be embedded in a
commutative diagram like the one in Lemma 7.6 with balanced-exact bottom row.
Now, if B is a B1 -group, then the bottom row splits, and hence so does the top row.
Consequently, A is a B1 -group. t
u
Absolutely Solid Groups As a brief excursion, we consider the following
concept. Call a torsion-free group absolutely solid if it is solid in every torsion-
free group in which it is contained as a pure subgroup. The proofs of a few basic
properties of this notion are delegated to the exercises. Clearly, all countable torsion-
free groups are absolutely solid. We are interested in more exciting examples: in
B2 -groups.
Theorem 7.8 (Rangaswamy [7], Bican–Fuchs [2]). B2 -groups are absolutely
solid.
Proof. Let B be a B2 -group, and G a torsion-free group containing B as a pure
subgroup. For the proof that B is solid in G, we may assume without loss of
generality that G=B is of rank 1, so there is a countable pure subgroup S of G such
562 14 Butler Groups

that G D B C S. By enlarging S if necessary, we may assume that S \ B is decent in

B; see Theorem 5.4. The isomorphism G=S Š B=.S \ B/ convinces us that G=S is a
B2 -group, and as such it admits a solid chain. This chain lifts to a solid chain from S
to G, as S is countable, so solid in G. Thus, G admits a solid chain. By Theorem 7.3,
the kernel H in a balanced-projective resolution 0 ! H ! A ! B ! 0
of B (with completely decomposable A) is a B2 -group; owing to Sect. 5(H), it is
also a TEP-subgroup in A. After choosing a relative balanced-projective resolution
0 ! K ! B ˚ C ! G ! 0 of B in G, we form the commutative diagram

H H
⏐ ⏐
⏐ ⏐
 
0 −−−−→ L −−−−→ A ⊕ C −−−−→ G −−−−→ 0
⏐ ⏐ 
⏐ ⏐ 
  
0 −−−−→ K −−−−→ B ⊕ C −−−−→ G −−−−→ 0,

where the vertical arrows are monic in the first set, and epic in the second set. Since
the bottom row and the middle column are balanced-exact, the middle row is also
balanced-exact, and hence it is a balanced-projective resolution of G. As G admits
solid chains, L must be a B2 -group. Therefore, K is a B1 -group as the quotient of a
B2 -group by a balanced TEP-subgroup (cp. Sect. 5(H)). Moreover, it is a B2 -group,
as H is a B2 -group (see Exercise 6). It remains to refer to Theorem 7.2 to conclude
that B is solid in G. t
u
F Notes. Theorem 7.8 was proved by Bican–Fuchs [2] for B1 -groups in L, and the general
version by Rangaswamy [7].
Fuchs–Rangaswamy [3] show that the union of a countable chain of pure B2 -subgroups is again
a B2 -group. The same holds for smooth chains of length !1 provided the subgroups in the chain
are decent and of cardinalities  @1 . Bican–Rangaswamy [1] extend the results to longer chains
under appropriate conditions on the factor groups. See also Bican–Rangaswamy–Vinsonhaler [1].

Exercises

(1) A torsion-free group of cardinality

@1 admits a solid chain.
(2) The direct sum of groups with solid chains also has a solid chain.
(3) Furnish the details of proof in Lemma 7.5.
(4) If a torsion-free group has a solid chain, then it also has an H.@0 /-family of
solid subgroups.
(5) If the torsion-free group G contains a pure B2 -subgroup B of infinite index ,
then it contains a pure subgroup S of cardinality  such that G=S is a B2 -group.
[Hint: Theorem 7.2.]
(6) Show that direct sums and summands of absolutely solid groups are likewise
absolutely solid.
8 Butler Groups of Uncountable Ranks 563

(7) A solid subgroup A of an absolutely solid group G is likewise absolutely solid.

[Hint: push-out for A ! G and 0 ! A ! H ! H=A ! 0.]
(8) Any extension of an absolutely solid group by a countable group is again
absolutely solid.

8 Butler Groups of Uncountable Ranks

We are at last on the final route toward a necessary and sufficient condition that
makes a B1 -group into a B2 -group. Armed with the arsenal of solid subgroups, the
only missing ingredient is the following key lemma which is a recast of Lemma 4.4
in Chapter 9.
Lemma 8.1 (Dugas–Hill–Rangaswamy [1]). Let  be an uncountable regular
cardinal, and 0 D C0 < C1 <    < C < : : : . < / a smooth chain of
pure subgroups of the torsion-free group C such that
S
(a)  < C D C;
(b) jC j <  for all  < ;
(c) for each  < ; C is a B1 -group.
If C is a B1 -group, then the set

E D f <  j 9  >  such that C is not a TEP–subgroup in C g

is not stationary in .
Proof. Apply Lemma 4.4 in Chapter 9 (see the notation there) when the B are
torsion groups, noting that the condition on Ext in the definition of the set S amounts
to that the map Hom.C C1 ; B / ! Hom.C ; B / is not surjective; see the proof
there. Surjectivity fails in the present case for suitable B whenever C is not TEP
in C C1 . Consequently, Lemma 4.4 in Chapter 9 implies what we wish to prove. u t
When B1 Implies B2 We are going to use this lemma to prove the following
theorem that is one of the cornerstones in our investigation of the relation between
B1 - and B2 -groups.
Theorem 8.2 (Fuchs [21]). A B1 -group is a B2 -group if and only if it admits a solid
chain.
Proof. Necessity is evident, but the verification of sufficiency requires a delicate
argument. Let B be a B1 -group of cardinality , and 0 D B0 <    < B < B C1 <
: : : . < / a solid chain leading up to B. It is clear that this chain contains a -
filtration of the group B; we drop to such a filtration, but keep the notation. Observe
that all the subgroups B in the chain are B1 -groups—as it is manifest in view of
Corollary 7.7.
564 14 Butler Groups

The claim is verified by transfinite induction on . If  D @0 : we know that

countable B1 -groups are B2 -groups (Theorem 4.2). So let  > @0 .
First, dispose of the case of regular cardinals . The preceding lemma assures the
existence of a cub S in  such that B is a TEP-subgroup for all  2 S. Omitting
all the B with  … S, we relabel the indices of the remaining B by ordinals < .
The new smooth chain contains only B1 -subgroups which are TEP in B, and so by
Sect. 5(H) even all the factors are B1 -groups. We appeal to Lemma 6.7 to argue that
B is decent (and hence prebalanced) in B C1 for every  < , thus segments of the
original solid chain provides solid chains for the factors [Lemma 6.5(c)]. Therefore,
the factors (being of cardinalities < ) are B2 -groups by induction hypothesis.
Decency being transitive, we can insert a chain of decent subgroups (available in the
factor B2 -groups) between the links. This yields a smooth chain of decent subgroups
from 0 all the way up to B, and we can conclude that B itself is a B2 -group.
We turn to the consideration of the singular cardinal case. Let B be a B1 -group of
singular cardinality  with a solid chain, say, with rank 1 factors. By Corollary 7.7,
all the subgroups in the chain are B1 -groups, and hence B2 -groups by induction
hypothesis. We refer to Theorem 5.5 in Chapter 1 to conclude that B has an H.@0 /-
family F of subgroups each of which has a smooth chain of decent subgroups. At
this point, we have an ample supply of B2 -subgroups in B of cardinalities <  which
makes the application of Theorem 9.3 in Chapter 3 possible. Using the notation
introduced there, choose  D @0 . Now F consists of B2 -groups, and a ‘basis’ of a
G 2 F is a smooth chain of decent subgroups. A subgroup H < G is a ‘free factor’
of G if both H and G=H belong to F . It is straightforward, but a bit lengthy, to check
that our situation is covered by Theorem 9.3 in Chapter 3; we leave the details to
interested readers. t
u
Unfortunately, the preceding result fails to give any clue on how to build a solid
chain in a B1 -group. More can be done if CH is assumed, but then the problem would
be oversimplified, and would become less exciting. We have no explanation for the
reason why the B1 -property forces certain solid chains to become decent chains.
A quick consequence of the preceding theorem is recorded for homogeneous
groups.
Corollary 8.3. Homogeneous B1 -groups are completely decomposable.
Proof. Homogeneous groups obviously have solid chains, so by Theorem 8.2 they
are B2 -groups provided they are B1 -groups. The rest follows from Sect. 5(F). t
u
A more relevant result is obtained if we combine Theorem 8.2 with Sect. 5(D):
Corollary 8.4. A B1 -group whose typeset has cardinality
@0 is a B2 -group. t
u
Another useful corollary to Theorem 8.2 is as follows.
Corollary 8.5. Let 0 ! A ! B ! C ! 0 be a prebalanced-exact sequence where
B is a B2 -, and A; C are B1 -groups. A is B2 if and only if C is.
Proof. From Theorem 7.3 it follows that A is a B2 -group if and only if C has a solid
chain. As C is B1 , this is the case exactly if C is a B2 -group. t
u
8 Butler Groups of Uncountable Ranks 565

The Vanishing of Bext2 We turn our attention to the other principal problem.
The following two results are preludes to our main result (Theorem 8.8) on Bext2 ;
they are concerned with the case when CH fails. Under the hypothesis :CH, they
provide an explicit mechanism for constructing groups without solid chains.
The next lemma is a natural continuation of Theorem 6.8: it has the same setup,
only the cardinality is assumed to be > @1 , but
2@0 . For its proof, we will need the
existence of two sets, † and †0 , of the cardinality of the continuum, whose members
are almost disjoint subsets of a countable set S with properties listed in Lemma 5.8
in Chapter 1. We may assume that the elements of S are prime numbers. A set X in
† or in †0 defines a rational group of type t D .k2 ; k3 ; : : : ; kp ; : : : / where kp D 1 or
0 according as p 2 X or not.
Lemma 8.6 (Fuchs [21]). Suppose that 2@0 > @1 . Let
M
CD C
 <!2

be a completely decomposable group where the set of types of the rank one
summands C is the union of the sets of types defined by the almost disjoint sets
† and †0 of cardinality @2 . Then no corank 1 subgroup of C that contains none of
the C admits a solid chain.
Proof. The notations C./; G./; C defined in the proof of Theorem 6.8 will have
the same meaning here, the only difference is that the range of indices is now !2 .
We may suppose that the well-ordering of the C is done alternately from the two
systems † and †0 .
Consider the chain fG./g for  < !2 . If  > !1 , then G./ is not solid in G;
as a matter of fact, not even in G. C 1/. In fact, the lattice ideal generated by the
types t.C /^t.C / . < /.@1 of them) is @1 -generated. For, if the type t does not
belong to the same system († or †0 ) as t , then t ^ t
.t ^ t1 / _    _ .t ^ tk /
can hold only if  2 f1 ; : : : ; k g, while adjoining intersections of t with types
from the system containing t does not change the type of the union at all.
Assume the claim is false: G has a solid chain. The cardinal @2 being regular,
there would then exist a cub E in !2 such that G./ is solid in G for all  2 E. But
this is absurd, so G cannot have a solid chain. t
u
As an immediate consequence of the preceding results, we verify:
Proposition 8.7 (Dugas–Thomé [1]). If 2@0 > @1 , then there exist a torsion-free
group H and a torsion group T such that
Bext2 .H; T/ ¤ 0:

In other words, if in a model V of ZFC, we have Bext2 .G; T/ D 0 for all torsion-free
groups G and torsion groups T, then CH must hold in V.
Proof. Hypothesis :CH implies that some corank 1 subgroup H (of cardinality
@2 ) of C in Lemma 8.6 fails to admit a solid chain. Such a group H is generated
by the rank 1 subgroups C , so choosing groups X Š C (where t.X / D
566 14 Butler Groups

t ^ t ) with fixed isomorphisms  W X ! C , we set X D ˚ X . There is


a balanced-projective resolution 0 ! K ! X !H ! 0 with  D ˚  ; hence
Bext2 .H; T/ Š Bext1 .K; T/. By virtue of Theorem 7.2, the kernel K is not a B2 -
group. Taking into account that by construction, any intersection of different types
t.X / is equal to the type of Z, it follows that K is homogeneous of type of Z. A
homogeneous K is a B1 -group if and only if it is B2 . Consequently, Bext1 .K; T/ ¤ 0
must hold for a suitable torsion group T. t
u
At this point we pause to look at the picture that emerges from the analysis. The
following theorem will shed more light on the connection between the vanishing of
Bext2 and the existence of solid chains.
Theorem 8.8. In any model of ZFC, the following are equivalent:
(i) Bext2 .G; T/ D 0 for all torsion-free G and torsion T.
(ii) Every torsion-free group admits a solid chain.
(iii) Balanced subgroups of completely decomposable groups are B2 -groups.
Either of (i)–(iii) implies the Continuum Hypothesis.
Proof. Before entering into the proof, we point out that (i) is equivalent to saying
that in a balanced-projective resolution 0 ! B ! C ! G ! 0 (with completely
decomposable C) of any torsion-free group G, the kernel B is a B1 -group. This is
an obvious consequence of the exact sequence 0 D Bext1 .C; T/ ! Bext1 .B; T/ !
Bext2 .G; T/ ! Bext2 .C; T/ D 0.
(i) ) (ii) Assuming (i), from Lemma 8.6 we deduce that CH holds, and therefore
by Lemma 7.1 the kernels of balanced-projective resolutions admit solid chains. An
appeal to Theorem 7.2 implies (ii).
(ii) , (iii) is an immediate consequence of Theorem 7.3.
(iii) ) (i) Evidently, (iii) implies that in balanced-projective resolutions the
kernels B are B2 -groups. Then Bext1 .B; T/ D 0 always for kernels B, and thus
Bext2 .G; T/ D 0 for all G and for all torsion T.
As far as CH is concerned, we have already observed above that it follows
from (i). t
u
To answer the speculation about whether the stated conditions (i)–(iii) follow
from CH, we point out that not even GCH implies the vanishing of Bext2 . In fact,
Magidor–Shelah [2] succeeded in establishing the existence of a model of ZFC
with GCH in which (i) fails. However, the hypothesis V D L turns out to suffice
to have Bext2 .G; T/ D 0 for all torsion-free G and torsion T (see Fuchs–Magidor
[1]). Consequently, in the constructible universe L, the properties B1 and B2 are
equivalent.
Rangaswamy’s Criterion for B2 -Groups Nevertheless, CH has a remarkable
consequence.
8 Butler Groups of Uncountable Ranks 567

Corollary 8.9 (Rangaswamy [6]). Assuming CH, a torsion-free group B is a B2 -

group if and only if

Bext1 .B; T/ D 0 and Bext2 .B; T/ D 0

for all torsion groups T.

Proof. If B is a B2 -group, then it admits a solid chain, so Bext2 .B; T/ D 0 by
Theorem 8.8, and the stated condition is necessary. Conversely, the vanishing of
the two Bexts implies that B is a B1 -group with a solid chain, thus—in view of
Theorem 8.2—it is a B2 -group. t
u
It is now time to think of the higher Bexts as well. The only relevant information
we have is that if B is a B2 -group, then Bexti .B; T/ D 0 for all i  1, and for
all torsion groups T. This is a simple consequence of the long Hom-Bext exact
sequence, using induction on i.
Subgroups of B2 -Groups As far as the subgroup problem is concerned, it boils
down to that of the existence of solid chains, this time from the subgroup to the
group. The following theorem improves on Corollary 7.7.
Theorem 8.10 (Fuchs [21]). For a pure subgroup A of a B2 -group G, these are
equivalent:
(i) A is a B2 -group;
(ii) there is a solid chain from A to G.
Proof. (i) ) (ii) Let 0 ! B ! A˚C ! G ! 0 be a balanced-projective resolution
of G relative to A, where C is completely decomposable. This gives rise to the exact
sequence

0 D Bext1 .G; T/ ! Bext1 .A; T/ ! Bext1 .B; T/ ! Bext2 .G; T/ D 0

for each torsion T; the terms involving G vanish, G being a B2 -group. Hence A
is a B1 -group exactly if so is B. Now, if A is a B2 -group, then so is A ˚ C,
and Corollary 8.5 implies that B is likewise a B2 -group. Hence (ii) follows from
Theorem 7.3.
(ii) ) (i) Hypothesis implies that B is a B2 -group, and so the same holds for
A ˚ C as a balanced extension of B2 -groups. Summands preserve B2 -property, so (i)
follows. t
u
F Notes. The theory of Butler groups culminates in the undecidability statement that in ZFC
alone it is impossible to prove or disprove that B1 -groups are B2 -groups. The very sophisticated
arguments by Shelah–Strüngmann [1] exceed the scope of this book; however, a remark is called
for regarding the method of proof. They start with a model of ZFC in which GCH holds, and
use forcing with Cohen reals for <  (where 
@4 ) in order to extend the model. A group is
constructed, and it is shown that in the extended model, the group is a B1 -, but not a B2 -group.
Most important developments in the problem of B1 -B2 relations are recent papers by Mader–
Strüngmann [1] and Strüngmann [3]. In the joint paper, a new class of groups, called Hawaiian
groups, are introduced to provide examples for B1 - and B2 -groups of arbitrary cardinalities. Let
  2@0 be a cardinal. A -Hawaiian group is built on a completely decomposable subgroup with
additional data in the form of  new generators satisfying appropriate conditions. Strüngmann
568 14 Butler Groups

[3] shows that for certain data the Hawaiian groups are B1 -, but not B2 -groups, when adding 
Cohen reals to the universe. In some models of ZFC with 2@0 D @4 , he obtains indecomposable
B1 -groups that are pure subgroups in completely decomposable groups, and have endomorphism
rings Š Z, without having the B2 -property.
Let us indicate briefly the milestones in the history of Butler groups of arbitrary cardinalities.
As it has already been mentioned, it started with Bican–Salce [1] who defined these groups, and
proved that B2 -groups are B1 -groups without cardinality restrictions. They succeeded in showing
that the converse holds in the countable case. The converse was established for groups of cardinality
@1 by Albrecht–Hill [1]. For higher cardinalities, the proof of the coincidence of the two classes
requires additional set-theoretical hypothesis, and it was proved up to @! assuming CH by Dugas–
Hill–Rangaswamy [1]. Rangaswamy [6] obtained the homological characterization (Theorem 8.9).
Restricting the size of the groups, rather than expanding the ZFC models, Fuchs [21] showed that
a torsion-free B of cardinality @n .n
1/ is a B2 -group if and only if it satisfies Bexti .B; T/ D 0
for all i  n C 1. Dugas–Thomé [1] proved that Bext2 .G; T/ need not vanish for all torsion-
free G if CH is denied. Magidor–Shelah [2] have the stronger claim that Bext2 .G; T/ ¤ 0 may
hold even if GCH is assumed. A result by Fuchs–Magidor [1] states: in the constructible universe,
the class of B1 -groups coincides with the class of B2 -groups, and Bext2 .G; T/ D 0 holds for all
torsion-free groups G and torsion groups T. The proof makes use of the so-called Box Principle, a
set-theoretical device, known to be a consequence of the hypothesis V = L. As mentioned above,
the heralded independence result in ZFC is due to Shelah–Strüngmann [1].

Exercises

(1) There are finitely Butler groups of cardinality @1 which fail to be B1 -groups.
[Hint: union of free groups which is not free.]
(2) Suppose A is a TEP-subgroup of the finitely Butler group G. If A is a B2 -group,
then it is decent in G.
(3) Assume
0 −−−−−→ B  −−−−−→ G −−−−−→ C −−−−−→ 0
⏐ ⏐ 
⏐ ⏐γ 
  

0 −−−−−→ B −−−−−→ G −−−−−→ C −−−−−→ 0

is a commutative diagram with exact rows, where G is a B1 -group and is epic

with Ker prebalanced in G0 . Then the top row has TEP exactly if the bottom
row has it.
(4) (a) Let B be a B1 -group of cardinality
@1 . Show that it is a B2 -group by
proving that it has a solid chain.
(b) A B1 -group that has a smooth chain of prebalanced subgroups with factors
of cardinalities
@1 is a B2 -group.
(5) (Fuchs–Rangaswamy) A pure subgroup of a completely decomposable group is
a B2 -group provided it is a B1 -group.
(6) (Fuchs–Rangaswamy) Derive Griffith’s Theorem 2.1 in Chapter 15 from Theo-
rem 8.10 by writing a Baer group B as B D F=H with free groups F; H. Then
argue that the Hom-Ext sequence implies that H is TEP in F, so B is a B2 -group.
Finite rank subgroups of B are free, so B is homogeneous, and therefore free.
9 More on Infinite Butler Groups 569

9 More on Infinite Butler Groups

A few additional results of interest on Butler groups are worthwhile recording.

Rigid Systems of B2 -Groups One is inclined to think that B2 -groups —being
so special—possess more transparent structures than torsion-free groups in general.
This is probably true, but in spite of this, they can display several unusual features
known for general torsion-free groups. For instance, on one hand, there exist
indecomposable B2 -groups of arbitrarily large cardinalities, and on the other hand,
there exist countable superdecomposable B2 -groups. As one of our final projects on
Butler groups, we discuss next the question of large indecomposable B2 -groups. We
will not only establish the existence of such groups, but will moreover show that
there is an ample supply of them.
To begin with, we construct a fully rigid system of finite rank Butler groups for
Z on the index set I D f1; : : : ; ng for any integer n  2.
Let fA1 ; A0 ; : : : ; An g be a set of rank one torsion-free groups; the type of Ai will
be denoted by ti . Assume that the types in the set

ftij D ti ^ tj j i ¤ jI i; j D 1; 0; : : : ; ng

are pairwise incomparable. Fix any map  W A1 ˚ A0 ˚    ˚ An ! Q such that

.Ai / ¤ 0 for all i. We claim: H D Ker  is a strongly indecomposable Butler
group of rank n C 1 whose endomorphism ring is isomorphic to a subring of Q.
The rank 1 subgroups H.tij / are fully invariant in H, so any  2 End H acts as a
multiplication by a rational number rij on H.tij /. For distinct indices i; j; k, H.tik /
intersects H.tij / ˚ H. tjk / non-trivially, thus we must have rij D rjk D rik , proving
the claim.
End H will be Š Z, if e.g. the types are chosen such that Z is the endomorphism
ring of a rank 1 group of type t1 ^ t0 .
It is easy to find types of the desired kind. For instance, choose

t1 D .1; 0; 1; 1; 1; : : : / and t0 D .0; 1; 1; 1; 1; : : : /;

while for i D 1; 2; : : : ; n, let ti be the type which starts with 1; 1 and continues
with a periodically repeating string of the form .1; : : : ; 1; 0; 1; : : : ; 1/ of length n,
with a single 0 in the ith place.
For a subset X  I D f1; : : : ; ng, define the group HX as the kernel of the
restriction of  to the direct sum ˚Ai with i 2 f1; 0g [ X. Then HI D H, and
the subgroups HX are fully invariant in H. It is readily seen that the set fHX j X  Ig
is in fact a fully rigid system for Z. We have all the ingredients for the application
of Corner’s Lemma 4.8 in Chapter 12 to derive (Exercise 3):
Theorem 9.1 (Fuchs–Metelli [2]). For every infinite cardinal , there exists a
fully rigid system consisting of 2 .pairwise non-quasi-isomorphic/ B2 -groups of
cardinality  each of which has endomorphism ring isomorphic to Z. t
u
570 14 Butler Groups

Superdecomposable Butler Groups We now turn our attention to the dual

problem: the existence of superdecomposable Butler groups. Recall that a superde-
composable group was defined to be a group which had no indecomposable
summands except for 0. (The superdecomposable groups constructed in Theo-
rem 1.5 in Chapter 13 are not Butler.)
The method we adopt here is based on the construction of a special tree. Define
a tree T of length ! whose nth level Tn contains 2n vertices:

Tn D f.n; 0/; .n; 1/; : : : ; .n; 2n  1/g:

The edges are directed, connecting vertex .n; m/ to vertices .n C 1; 2m/ and
.n C 1; 2m C 1/ for all 0
m < 2n . To each vertex .n; m/ of T D [n<! Tn assign a
torsion-free group Anm , and to each directed edge a monomorphism
0 1
nmW Anm ! AnC1;2m ; nm W Anm ! AnC1;2mC1 :
L
Next, for each n < ! define the group Gn D .n;m/2Tn Anm , and the map nW Gn !
GnC1 such that its restriction to Anm is the map
0 1
nm ˚ nm W Anm ! AnC1;2m ˚ AnC1;2mC1 :

Let G denote the limit of the arising direct system fGn j n .n < !/g. All the
canonical maps n W Gn ! G are monic, so the Anm and the Gn may be identified
with their images in the direct limit G.
Lemma 9.2 (Benabdallah–Birtz [1], Meinel [1]). The direct limit G is a superde-
composable group if fAnm j .n; m/ 2 Tn g is a family of non-trivial torsion-free groups
satisfying the following two conditions:
(i) Hom.Anm ; Aij / D 0 if there is no directed path from vertex .n; m/ to vertex .i; j/;
(ii) for every homomorphism W Anm ! G there is an integer k such that  factors
through a map kW Anm ! Gk , i.e.,  D k k .
Proof. Suppose we have a collection of torsion-free groups Anm satisfying (i)–(ii).
Define Bnm .n < !I m D 0; : : : ; 2n  1/ as the union of the Aij for all vertices .i; j/
above .n; m/ (including .n; m/ itself), which are connected to .n; m/ by a directed
path. Just as the Aij , these Bnm may be viewed as subgroups of G. It is readily
checked:
(a) B00 D G. Furthermore, Bnm \ Gn D Anm .
(b) Bnm D BnC1;2m ˚ BnC1;2mC1 for all .n; m/ 2 T, as is clear from the way Anm was
embedded in the direct sum AnC1;2m ˚ AnC1;2mC1 .
(c) Each Bnm is fully invariant in G. Indeed, this is an immediate consequence of
conditions (i) and (ii).
(d) The intersection of the Bnm for vertices .n; m/ in a branch Y D f.n; in / j n < !g
of T is 0 (where inC1 D 2in or 2in C 1). In fact, let y 2 G belong to each
9 More on Infinite Butler Groups 571

Bn;in 2 Y. There is an n < ! with y 2 Gn , thus y 2 Bn;in \ Gn D An;in . Also,

y 2 BnC1;inC1 \ GnC1 D AnC1;inC1 , whence y 2 An;in \ AnC1;inC1 D 0:
To verify the lemma, suppose C is an indecomposable summand of G, say, G D
C ˚ D for some D
G. From (a)–(c) we derive G D B00 D B10 ˚ B11 , whence
B1i D .B1i \ C/ ˚ .B1i \ D/ for i D 1; 2. Thus C D .B10 \ C/ ˚ .B11 \ C/. By
indecomposability, either C
B10 or C
B11 , and so C is a summand of one of
them. In the first alternative, C D .B20 \ C/ ˚ .B21 \ C/, so again, C is a summand
of either B20 or B21 . Continuing in the obvious fashion, we see that C is contained
in each Bn;in for vertices .n; in / in a branch of T. Hence (d) implies C D 0. t
u
We are now prepared to prove:
Theorem 9.3 (Dugas–Thomé [2], Fuchs–Metelli [2]). There exist superdecom-
posable Butler groups of countable rank.
Proof. We want to find finite rank Butler groups Anm satisfying conditions (i) and
(ii). We do not have to pay attention to (ii) as it is automatically satisfied by groups
of finite rank.
Let ft1 ; t0 ; t1 ; : : : ; tn ; : : : g be a countable set of types subject to the conditions
that the types ti ^ tj .i ¤ j/ are pairwise incomparable and Z is the endomorphism
ring of a rank 1 group of type t1 ^ t0 . These conditions are fulfilled, e.g., if we
define t1 and t0 to have 1’s at even numbered places, and alternately 1 and 0 at
odd numbered places:

t1 D .1; 1; 0; 1; 1; 1; 0; 1; : : : / and t0 D .0; 1; 1; 1; 0; 1; 1; 1; 0; : : : /;

while tn .n  1/ has 0 at even numbered, and 1 at odd numbered places, except for
the 2n  1st and 2nth odd numbered places, where 0’s are chosen:

t1 D .0; 0; 0; 0; 1; 0; 1; 0; 1; : : : /; t2 D .1; 0; 1; 0; 0; 0; 0; 0; 1; 0; : : : /; : : :

We proceed to select torsion-free groups An .n D 1; 0; 1; 2; : : : / of rank one such

that t.An / D tn , and monomorphisms n W An ! Q. Define a bijection f W ! ! T
with f .0/ D .0; 0/. For .n; m/ 2 T, denote by Xnm the set of vertices of T which
lie on the path from .0; 0/ to .n; m/. Using the maps n , define Anm as the group HI
constructed above with the aid of the Ai with i 2 I D f1g [ ff 1 .Xnm /g.
If there is no directed path connecting .n; m/ and .i; j/, then Anm and Aij are
incomparable members of the fully rigid system consisting of the subgroups of HJ
on the index set J D f1g [ ff 1 .Xnm [ Xij /g; hence condition (i) will be satisfied.
0 1
For nm ; nm choose embeddings of Anm in AnC1;2m ; AnC1;2mC1 as pure subgroups.
Then the groups Gn will be finite rank Butler groups, and the map n W Gn ! GnC1
yields a pure embedding of Gn in GnC1 . Since the union of a countable chain of pure
finite rank Butler subgroups is a Butler group, the proof is completed. t
u
F Notes. The construction for Theorem 9.1 shows that the groups can be chosen so as to have
their typesets contained in a lattice generated by eight elements. Arnold–Dugas [1] succeeded
572 14 Butler Groups

in sharpening this result to five generators; they use free groups with distinguished subgroups to
derive their result. The same paper contains several results on Butler groups with finite typesets.
Blagoveshchenskaya–Göbel–Strüngmann [1] investigate a new class of groups: epic images of
local almost completely decomposable groups. A generalization of near-isomorphism to infinite
ranks is used as the basis of classification.

Exercises

(1) Any direct decomposition of the group G in the proof of Theorem 9.3 has only
a finite number of summands ¤ 0.
(2) Every endomorphism of G in Theorem 9.3 is determined by its action on A00 .
(3) Prove that the groups HX referred to in Theorem 9.1 are B2 -groups. [Hint: they
are finite sums of B2 -groups tensored by free groups.]

Problems to Chapter 14

PROBLEM 14.1 (Metelli). Find invariants for the quasi-isomorphy classes of

B .n/-groups, especially in the n D 2 case.
PROBLEM 14.2. Determine the projective dimensions of finite rank Butler
groups over their endomorphism rings.
PROBLEM 14.3. When is the tensor product of two B1 -groups again a B1 -group?
PROBLEM 14.4. When is a finitely Butler group a B2 -group? Is it if it has a solid
chain?
PROBLEM 14.5. Characterize inverse limits of (finite rank) Butler groups.
Chapter 15
Mixed Groups

Abstract Mixed groups form the most general class of abelian groups, and traditionally, the
main question is to find out how the torsion and torsion-free parts are put together to create a
mixed group.
For a long time the theory of mixed groups was a neglected area; papers dealing with mixed
groups were embarrassingly sparse. The reason was perhaps that no powerful methods were
available to deal with more intricate situations. The enormous complexity of their structure made
an in-depth study of mixed groups virtually impossible. As a result, the main concern was to find
out when a group was not an honest mixed group, but a splitting group: a direct sum of a torsion
and a torsion-free group. The characterizations of those torsion and the torsion-free groups that
force splitting were the main results on mixed groups up to the 1970s.
The theory was suddenly revitalized: the impetus was the revision of the traditional view as an
extension of a torsion group by a torsion-free group. Mixed groups could also be regarded in the
diagonally opposite way as an extension of a torsion-free group by a torsion group. This point of
view, inspired by J. Rotman, turned out a more profound approach to mixed groups, though there
was no canonical way of treating them from this angle. New life was brought to the study of mixed
groups after it was observed, in addition, that newly developed methods in the theory of torsion
groups may be used, mutatis mutandis, to deal with some special interesting mixed groups. The
theory of valuated groups—as the new vehicle for research—was developed under the leadership
of the New Mexico group theorists.
The quintessence of the new approach, first systematically applied by Warfield, is the idea
of ignoring temporarily the torsion subgroup, but at the same time enriching a free subgroup
of maximal rank with the height functions (as valuations) of its elements. In this way, relevant
information about the absent torsion subgroup is incorporated into the free subgroup considered,
which sufficed in interesting special cases.
In the course of developing a theory of mixed groups, we start with the conventional splitting
problem. We then introduce the machinery needed for the discussion of tractable classes of mixed
groups. Some tools developed for torsion and torsion-free groups need to be readjusted, and new
tools have to be acquired to deal with the new entities. Existence and extension of homomorphisms
between mixed groups are in the center of discussion. The highlights are the structure theorems on
simply presented and Warfield groups.

1 Splitting Mixed Groups

It is easy to find examples for mixed groups: just take the direct sum of a torsion and
a torsion-free group. But there exist non-splitting mixed groups, i.e. mixed groups
that do not split into such a direct sum, and needless to say, we are mainly concerned
with this kind of mixed group.

© Springer International Publishing Switzerland 2015 573

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_15
574 15 Mixed Groups

Non-Splitting Mixed Groups We begin with examples of non-splitting mixed

groups. Our first example provides a stunningly simple method of manufacturing
non-splitting mixed groups, while the other two are key examples exhibiting non-
splitting mixed groups of torsion-free rank 1.
Example 1.1. Let T be an unbounded reduced torsion group. Then Ext.Q=Z; T/ D T  is a non-
splitting reduced mixed group (the cotorsion hull of T) with T as torsion subgroup.
Example 1.2. Let p1 ; : : : ; pn ; : : : denote different primes, and define

T1 D ˚1
nD1 han i with o.an / D pn :
Q1
Thus T1 is the torsion subgroup of the direct product B D nD1 han i, a mixed group. Let b0 D
.a1 ; : : : ; an ; : : : / 2 B. For i ¤ n, the equation pn x D ai is uniquely solvable in hai i, thus B contains
a unique element bn such that pn bn D .a1 ; : : : ; an1 ; 0; anC1 ; : : : / D b0  an : T1 is the torsion part
of the group A1 D hT1 ; b1 ; : : : ; bn ; : : : i such that A1 =T1 is torsion-free of rank 1. We claim A1 is not
splitting. Otherwise, we had A1 D T1 ˚ G for some G < A1 , so bn D tn C gn with tn 2 T1 ; gn 2 G
for each n < !. Thus pn tn Cpn gn D pn bn D b0 an D .t0 an /Cg0 : Equating the T1 -coordinates
we get pn tn D t0  an for n D 1; 2; : : : . There is a prime pj for which the equation pj x D t0 is
solvable in T1 , and if x D t 2 T1 is a solution, then we get pj .t tj / D aj , an obvious contradiction.
Example 1.3. Let p be any prime, and set

T2 D ˚1
nD1 han i with o.an / D p2n :
Q1
Now we define bn D .0; : : : ; 0; an ; panC1 ; p2 anC2 ; : : : / 2 nD1 han i. These bn are of infinite
order, and satisfy pbnC1 D bn  an .n D 1; 2; : : : /. It is readily checked that T2 is the torsion part
of the p-local group A2 D hT2 ; b1 ; : : : ; bn ; : : : i such that A2 =T2 is p-divisible torsion-free of rank
1. If A2 D T2 ˚ QG held for some G < A2 , then G would be a p-divisible subgroup of A2 , contrary
1
to the fact that nD1 han i has no such subgroup ¤ 0. Consequently, A2 is not splitting.

One nice aspect of Example 1.1 is that the method employed there can be used
to show that every non-splitting mixed group of torsion-free rank 1 can be found in
the cotorsion hull of its torsion subgroup.
Proposition 1.4. Let A be a reduced, non-splitting mixed group of torsion-free rank
1, and T D t.A/. The canonical map T ! Ext.Q=Z; T/ extends to a monomorphism
A ! Ext.Q=Z; T/.
Proof. As A=T is torsion-free and Ext.Q=Z; T/ is cotorsion, Theorem 9.2 in Chap-
ter 9 allows us to extend the canonical map T ! Ext.Q=Z; T/ to a homomorphism
 W A ! Ext.Q=Z; T/. Evidently, Ker  is torsion-free and reduced. Hence, if
Ker  ¤ 0, then Im  must have a torsion subgroup > T. But this is impossible. u
t
We stop for a moment to interpret this result as saying in effect that the cotorsion
hull of a reduced torsion group T is a largest mixed group with torsion part T that has
no non-trivial direct decomposition where T is contained in one of the summands.
Nunke Groups By making use of the generalized Prüfer p-groups H (which
will now be denoted as H .p/ to indicate dependence on p), we introduce non-
splitting mixed groups of torsion-free rank 1, following Nunke [3]. The Nunke
group N .p/ of length  is obtained as follows.
1 Splitting Mixed Groups 575

For every prime p and every ordinal , a mixed group N .p/ is constructed such
that the sequence

0 ! Z ! N .p/ ! H .p/ ! 0 (15.1)

is exact, where Z is an infinite cyclic group, and the groups N .p/ satisfy the
following conditions:
(i) p N .p/ D Z ,
(ii) N C1 .p/ D pN .p/ and Z C1 D pZ , and
(iii) for a limit ordinal ; N .p/=Z Š ˚< N .p/=Z .
The construction is an imitation of the definition of the H in Sect. 1 in Chapter 10,
the only difference being that the starting point is not a cyclic group of order p, but
an infinite cyclic group. The details are left to the reader.
We will refer several times to the following important observation.
Lemma 1.5 (Nunke [5]). For every group A, the exact sequence (15.1) induces the
long exact sequence
ı
0 ! Hom.H .p/; A/ ! Hom.N .p/; A/ ! Hom.Z ; A/ Š A!
ı
! Ext.H .p/; A/ ! Ext.N .p/; A/ ! 0

where Ker ı D p A.
Proof. We get the sequence from Theorem 2.3 in Chapter 9. The image of the map
between the last two Homs comes from the restrictions of maps N .p/ ! A to
Z D p N .p/, so it is contained in p A. On the other hand, as H .p/ is totally
projective, every Z ! p A extends to a map N .p/ ! A. t
u
Splitting Criteria We continue with the discussion of conditions under which
a mixed group A splits, i.e. A D T ˚ G, where T D t.A/ is unique, while the
torsion-free G Š A=T is unique up to isomorphism.
First, we ask whether a mixed group ought to split if all of its localizations at
primes are splitting; the answer is “no:” Example 1.2 is a counterexample (see also
Examples 2.2 and 2.3). However, a fixed homomorphism all of whose localizations
are splitting maps is itself a splitting map.
The next theorem is a frequently quoted characterization of those torsion groups
that force the splitting of mixed groups whenever they appear as torsion parts.
Theorem 1.6 (Baer [4], S. Fomin [1]). A torsion group T satisfies Ext.G; T/ D 0
for all torsion-free groups G if and only if T D B ˚ D for a bounded group B and a
divisible group D.
Proof. Considering that divisible subgroups are always summands, it suffices to
deal with reduced groups T. Our Example 1.1 above shows that T ought to be
bounded to have the stated property. On the other hand, if T is bounded, then because
of its purity it is a summand in every mixed group in which it is the torsion subgroup.
t
u
576 15 Mixed Groups

The dual problem of characterizing the torsion-free groups G such that extensions
of torsion groups by G are always splitting is more difficult; it will be discussed in
the next section.
There are several necessary and sufficient criteria for splitting, but none is totally
satisfactory. One of the best is given in the next theorem.
Theorem 1.7 (May [2]). A mixed group A is splitting if and only if it is the union
of a countable chain

0 D A0
A1
  
An
: : :

of subgroups such that, for all n < !, the torsion subgroup tAn is bounded and
t.A=An / D .tA C An /=An :
Proof. If A is splitting, A D tA ˚ G, then the subgroups An D AŒnŠ
˚ G satisfy
the stated conditions. Conversely, suppose A has subgroups An as stated. Then An D
tAn ˚ Gn with torsion-free Gn .n < !/. We are done if we can choose the Gn to form
a chain, because then we can put G D [n<! Gn . Given a partial chain 0 D G0

  
Gn , we argue

AnC1 =.tAnC1 ˚ Gn / D AnC1 =Œ.tA ˚ Gn / \ AnC1

Š .tA C AnC1 /=.tA C An / Š
Š Œ.tA C AnC1 /=An
=Œ.tA C An /=An
Š Œ.tA C AnC1 /=An
=t.A=An /:

The last factor group is torsion-free, thus tAnC1 ˚ Gn is pure in AnC1 . It follows that
.tAnC1 ˚ Gn /=Gn is a bounded pure subgroup in AnC1 =Gn , thus

AnC1 =Gn D .tAnC1 ˚ Gn /=Gn ˚ GnC1 =Gn

for some GnC1  Gn : Therefore, AnC1 D tAnC1 ˚ GnC1 . t

u
Quasi-Splitting We turn our attention to mixed groups that are close to being
splitting in the following sense. A group A is quasi-splitting if it has a splitting
subgroup B such that nA
B
A for some integer n > 0. Note that if B D S ˚ G,
where S is torsion and G is torsion-free, then BCT D T ˚G also splits for the torsion
part T of A. Consequently, in the definition of quasi-splitting we may assume that
B has the same torsion subgroup as A. The next example exhibits a quasi-splitting
mixed group that fails to split.
Example 1.8. Let T D ˚1 Q
nD1 Z.p / and G D F , the p-adic completion of a free group F
n

of countable rank. Then G is algebraically compact such that G=pG Š T=pT, so there is an
epimorphism  W G ! T=pT with kernel pG. Let A be the pull-back in the commutative diagram
with exact rows:
0 −−−−−→ pT −−−−−→ A −−−−−→ G −−−−−→ 0
 ⏐ ⏐
 ⏐ ⏐η
  
can
0 −−−−−→ pT −−−−−→ T −−−−−→ T /pT −−−−−→ 0,
1 Splitting Mixed Groups 577

so A may be thought of as the set of all .t; g/ 2 T ˚G with g D t CpT. Then pA  pT ˚pG  A,
so A is quasi-splitting. If A were splitting, then there would exist a map  W G ! T such that
g D g C pT for all g 2 G. But G is cotorsion and reduced torsion, so it must be of bounded
order. Therefore, the image of G in T=pT would be finite, while G is infinite. Thus A cannot
split.

Next we prove a lemma in a more general setting than needed for the proof of
Theorem 1.10. Let C be a group and n > 0 an integer. Multiplication by n in C
 
factors as C!nC!C where  W c 7! nc and  is the inclusion map. Given the
bottom exact sequence, we build a commutative diagram with exact rows:

α β
νμ : 0 −−−−−→ A −−−−−→ B −−−−−→ C −−−−−→ 0
 ⏐ ⏐
 ⏐ ⏐μ
  
α β
ν : 0 −−−−−→ A −−−−−→ nB + A −−−−−→ nC −−−−−→ 0
 ⏐ ⏐
 ⏐ ⏐ν
  
α β
: 0 −−−−−→ A −−−−−→ B −−−−−→ C −−−−−→ 0.

Lemma 1.9 (C. Walker [1]). If e splits, then e represents an element of

Ext.C; A/Œn
. If CŒn
D 0, then the converse is also true.
Proof. Observe that e D ne. Now, if e splits, then so does ne, and therefore e
belongs to Ext.C; A/Œn
. Conversely, e 2 Ext.C; A/Œn
means that e is contained in
the kernel of the endomorphism of Ext.C; A/ induced by multiplication by n in C.
Therefore, if CŒn
D 0, then in view of Lemma 5.1(iii) in Chapter 9 this is equivalent
to the splitting of e. t
u
The following theorem identifies the torsion subgroup of Ext.G; T/ as the set of
quasi-splitting extensions.
Theorem 1.10 (C. Walker [1]). A mixed group A is quasi-splitting if and only if
the exact sequence

0!T !A!G!0 (15.2)

.where T D t.A/ and G is torsion-free/ represents an element of finite order in

Ext.G; T/.
Proof. If this exact sequence represents an element of finite order in Ext.G; T/, then
the second part of the preceding lemma implies that the exact sequence

0 ! T ! nA C T ! .nA C T/=T ! 0 (15.3)

is splitting. Since nA
nA C T
A, this means that A is quasi-splitting.
578 15 Mixed Groups

Conversely, assume A is quasi-splitting, say, n > 0 is an integer such that some

subgroup B
A with nA C T
B is splitting. Then T is a summand of nA C T,
and (15.3) is a splitting sequence. From Lemma 1.9 we conclude that then (15.2)
represents an element of Ext.G; T/Œn
. t
u
Non-Splitting Groups Close to Splitting From a different point of view, we
might regard a non-splitting mixed group to be close to being splitting if all of its
subgroups of smaller cardinalities are splitting. Concerning such groups we have the
following result.
Proposition 1.11. If, for the cardinal , there is a non-free -free group of
cardinality , then there exists a non-splitting mixed group of cardinality  all of
whose subgroups of smaller cardinalities are splitting.
Proof. Let A be -free, non-free group of cardinality . Theorem 2.1 in the next
section implies that A is not a Baer group, thus there exists a torsion group T with
Ext.A; T/ ¤ 0. Lemma 2.4 shows that jTj
 may be assumed. Choose a mixed
group M which is a representative of a non-zero element in this Ext. Then every
subgroup N of M of cardinality <  must be splitting, since N=.N \T/ Š .N CT/=T
is a free group. t
u
In particular, in view of Corollary 8.10 in Chapter 3, we can conclude that for
every integer n > 1 there exist mixed groups of cardinality @n which do not split,
but all of their subgroups of cardinalities
@n1 are splitting.
F Notes. It was Levi [1] who constructed the first non-splitting mixed group. It took nearly
two decades until S. Fomin [1] and Baer [4] found a sufficient and a precise condition, respectively,
for a torsion group to force splitting. The theory of quasi-splitting is due primarily to C. Walker
[1].
Several authors, including A. Fomin [3], investigate quotient-divisible mixed groups. For more
on these groups, see Albrecht–Breaz–Vinsonhaler–Wickless [1].
The obvious question as to when non-equivalent extensions are isomorphic has not been
thoroughly investigated. Exception is Mader [3] who studied isomorphic extensions of a torsion
group by a torsion-free group.

Exercises

(1) Summands of splitting groups are splitting.

(2) (Procházka) If A is a mixed group such that, for some n 2 N, the subgroup nA
splits, then A also splits.
(3) (a) (S. Fomin) A D ha1 ; : : : ; an ; : : : j pa1 D    D pn an D : : : i is a non-
splitting mixed group.
(b) (Kulikov) The mixed group A D ha1 ; : : : ; an ; : : : j p2 .a1  pa2 / D    D
p2n .an  panC1 / D    D 0i does not split.
2 Baer Groups are Free 579

(4) (Oppelt) Let T D T1 ˚    ˚ Tn be a torsion group with pi -components Ti , and

let A be a mixed group with torsion part T. A splits if and only if, for every i,
the group G=.T1 ˚    ˚ Ti1 ˚ TiC1 ˚    ˚ Tn / splits.
(5) (Corner) For a torsion group T, these are equivalent:
(a) T is an endomorphic image of every mixed group with torsion part T;
(b) T has a basic subgroup that is an endomorphic image of every mixed group
in which T is the torsion part;
(c) T D B ˚ D with bounded B and divisible D.
(6) There exist epimorphisms A1 ! T1 and A2 ! T2 (see Examples 1.2 and 1.3).
[Hint: ai 7! 0; b0 7! 0; bn 7! an ; and a4n 7! an ; a4nCi 7! 0 for i D 1; 2; 3.]
(7) (Mishina) Let A be a mixed group with no elements of order 2. A is splitting if
it has an automorphism which acts as multiplication by 1 on its torsion part
T, and induces the identity map on A=T.
(8) (Stratton) Let A be a mixed group, and F a pure free subgroup of finite rank.
A is splitting exactly if A=F is splitting. [Hint: the canonical A ! A=F yields
an isomorphism between torsion groups.]
(9) Give an example showing that the torsion part need not be an endomorphic
image of a mixed group. [Hint: Example 1.1, or B in Example 1.2.]
(10) (Strüngmann) Let T be a torsion group with only a finite number of non-
trivial bounded p-components (no restriction on the unbounded ones). There
is a countable completely decomposable group C such that every countable
torsion-free group G with Ext.G; T/ D 0 embeds in C. [Hint: try C D R.@0 /
with R < Q, hp .1/ D 1 for fp j Tp not unboundedg.]
(11) (Joubert–Ohlhoff–Schoeman) A mixed group A with torsion subgroup T is
quasi-splitting if and only if there is n 2 N such that the map nP W T ! T
extends to A ! T.

2 Baer Groups are Free

We now attack the problem dual to the one settled in Theorem 1.6 above: which are
the torsion-free groups G such that every mixed group A whose torsion subgroup
T satisfies A=T Š G is splitting? This problem was posed by Baer [4], and stayed
open for 30 years, until it was solved by Griffith [4].
Baer Groups Actually, we will deal with this problem in a slightly more general
form by not assuming in advance that G is torsion-free. Accordingly, we say that B
is a Baer group if

Ext.B; T/ D 0 for all torsion groups T:

Evidently, free groups are Baer groups, and our goal is to show that there are no
other Baer groups.
580 15 Mixed Groups

We start with a list of a few elementary properties.

(a) Subgroups of Baer groups are Baer groups. This follows from the fact that if
C ! B is monic, then the induced map Ext.B; T/ ! Ext.C; T/ is epic.
(b) A Baer group B must be torsion-free. In fact, if B contains a cyclic subgroup C of
order p for some prime p, then from the exactness of the sequence Ext.B; C/ !
Ext.C; C/ ! 0 and from Ext.C; C/ ¤ 0 we conclude Ext.B; C/ ¤ 0, so such a
B cannot be Baer.
(c) A finite rank Baer group is free. Let T0 denote the direct sum of countably many
copies of the group ˚p ˚k<! Z.pk / (summation over all primes). Suppose B is
torsion-free of finite rank n, and F is a free subgroup of rank n in B. If B is not
finitely generated, then B=F is a countably infinite torsion group, in which case
Ext.B=F; T0 / has to be of the power of the continuum. In fact, if B=F contains
a copy of Z.p1 / for some prime p, then Ext.B=F; T0 / contains a copy of the
p-adic integers. If B=F is infinite, but contains no copy of Z.p1 /, then B=F
has an infinite socle, so Ext.B=F; T0 / maps upon an infinite product of cyclic
groups of (not necessarily different) prime orders.
The exact sequence 0 ! F ! B ! B=F ! 0 induces the exact
sequence Hom.F; T0 / ! Ext.B=F; T0 / ! Ext.B; T0 /: Here Hom.F; T0 / D
˚n Hom.Z; T0 / D ˚n T0 is a countable group. Hence if Ext.B; T0 / D 0; then
Ext.B=F; T0 / cannot be uncountable. This implies that B=F must be finite, thus
B is finitely generated, and hence free.
(d) Countable Baer groups are free. By Pontryagin’s criterion, it suffices to show
that their finite rank subgroups are free. These subgroups are Baer by (a), so the
claim follows from (c).
Only Free Groups are Baer We can now verify the main result on Baer groups.
Theorem 2.1 (Griffith [4]). A group is a Baer group if and only if it is free.
Proof. Only the necessity requires a proof. So assume B is a Baer group, say, of
cardinality . To start a transfinite induction on the cardinality, refer to (d) for the
case  D @0 . Assume  is an uncountable cardinal, and Baer groups of cardinalities
<  are free. We distinguish two cases according as  is a regular or a singular
cardinal.
Case I:  is a regular cardinal. Pick a filtration fB g < of B with pure subgroups
of cardinalities < . By (a), the B are Baer groups. We now apply Lemma 4.4
in Chapter 9 with large †-cyclic groups A to conclude that there is a cub C in
 such that for  2 C, B =B is a Baer group of cardinality <  for all  > 
(cf. Lemma 2.4), and hence free by the induction hypothesis. Keeping only the
B with  2 C, we obtain a subfiltration fB g 2C where all the factor groups
are free: this cannot be anything else than a filtration of a free group. Thus B is
free.
Case II:  is a singular cardinal. In view of (a) and the induction hypothesis, B is
a -free group. The singular compactness theorem 9.2 in Chapter 3 guarantees
that B is free. t
u
2 Baer Groups are Free 581

We note in passing that a group G, satisfying Ext.G; T/ D 0 for all p-groups T

and all primes p, need not be a Baer group. This is demonstrated by the following
examples.
Example 2.2. A prototype example is the group R of rational numbers with square-free denomi-
nators. The exact sequence 0 ! hp1 ZiQ! R ! ˚q¤p Z.q/ ! 0 (summation over all primes
q ¤ p) induces the exact sequence 0 D q Ext.Z.q/; T/ ! Ext.R; T/ ! Ext.hp1 Zi; T/ D 0
for any p-group T, showing that Ext.R; T/ D 0 for every p-group T and for every prime p.
However, R is not a Baer group.
Example 2.3. There is a non-free @1 -free group A, for which Ext.A; T/ D 0 holds for all p-groups
T, for every prime p. We use the construction described in Theorem 8.9 in Chapter 3 to get a group
A of cardinality @1 as the union of a smooth chain fF . < !1 /g of countable free groups. If
 is a successor ordinal, then we put FC1 D F ˚ Z, and if  is a limit ordinal, then we let
F D [< F (which will be free by Theorem 7.5 in Chapter 3). Finally, if  D  C 1 for a
limit ordinal , then define F to be countable free and to fit into the exact sequence 0 ! F !
FC1 ! R ! 0 (free resolution of R) where R is the group of rational numbers with square-free
denominators. Then A D [<!1 F is @1 -free, and Theorem 7.5 in Chapter 3 proves that it is not
free. Using the fact that Ext.R; T/ D 0 for p-groups T (any p), Lemma 4.1 in Chapter 9 shows
Ext.A; T/ D 0 holds for all p-groups T, for any prime p.

Test Groups for Baer Groups We now ask the question: is there a ‘test group’
for Baer groups? In other words, does there exist a single torsion group TP such that
P D 0 implies that B is a Baer group? If we limit the size of B, then a direct
Ext.B; T/
sum of cyclic groups of adequate size is such a test group—as is shown by the next
lemma, but there is no such group that works in ZFC for all groups B. However,
the situation changes drastically in the constructible universe: in L, there exist test
groups even of countable cardinality; see Theorem 2.5.
Lemma 2.4. A group B of cardinality
 is a Baer group provided that
Ext.B; T / D 0 holds for the torsion group

T D ˚ Œ˚p ˚k<! Z.pk /

:

Proof. Assume B has cardinality

 and satisfies Ext.B; T / D 0. To prove B is a
Baer group, we first show that Ext.B; S/ D 0 for all †-cyclic groups S. The factor set
defining an extension of S by B has at most  different elements, so it is contained in
a summand of S of cardinality
. This summand is isomorphic to a summand of
T , hence hypothesis implies that such a factor set ought to be a transformation set.
Once we know that Ext.B; S/ D 0 for all †-cyclic torsion groups S, then we
consider a pure-projective resolution 0 ! U ! S ! T ! 0 of an arbitrary torsion
T; here S is torsion and †-cyclic. The exactness of Ext.B; S/ ! Ext.B; T/ ! 0
implies that B is a Baer group. t
u
Next we intend to show that in the constructible universe, there do exist test
groups for the Baer property.
Theorem 2.5 ((V = L) (Eklof [4])). Let TP denote the direct sum of countably many
copies of the group ˚p ˚k<! Z.pk /. A group B is a Baer group if and only if
P D 0.
Ext.B; T/
582 15 Mixed Groups

P D 0. We induct
Proof. Necessity being obvious, assume that B satisfies Ext.B; T/
on the cardinality  of the group B.
Case I:  is a regular cardinal. If  D @0 , the claim follows from Lemma 2.4. So
suppose  is uncountable, and B satisfies Ext.B; T/ P D 0. Then Ext.B0 ; T/
P D0
0 0
for all subgroups B
B, so by induction, all subgroups B of cardinalities < 
are Baer, thus there is a smooth chain 0 D B0 < B1 <    < B < : : : of Baer
groups with union B such that jB j <  for all  < . Then Ext.B ; T/ P D0
for all , so—in view of the hypothesis V = L—we can appeal to Lemma 4.3
in Chapter 9 to argue that there is a cub C   such that Ext.B =B ; T/P D0
for all  <  in C. By induction hypothesis, then Ext.B =B ; T/ D 0 holds for
all torsion T. It remains to refer to Lemma 4.3 in Chapter 9 to derive that also
Ext.B; T/ D 0. Thus B is a Baer group.
Case II:  is a singular cardinal. We argue as in the proof of Theorem 2.1. t
u
Eklof [4] has shown that Theorem 2.5 is false if the hypothesis V = L is dropped:
P D 0 is a Baer group.
it is undecidable in ZFC that a group B satisfying Ext.B; T/
F Notes. As mentioned above, the problem on Baer groups was open for 30 years; during
this time, several partial results were published. Finally, Griffith [4] solved the problem using an
excellent idea. First he constructed, for every infinite cardinal , a mixed group G such that all
torsion-free subgroups of G were free, but G modulo its torsion part T (which was like the
one in Lemma 2.4 above) was divisible of cardinality . Then he considered the exact sequence
0 ! T ! G ! G =T ! 0 which induced the exact sequence

Hom.B; G / ! Hom.B; G =T / ! Ext.B; T / D 0

for a Baer group B. He argued that since the map between the Homs is epic, if B embeds in G =T ,
then it also embeds in G . Consequently, it must be free.
A more difficult problem is to describe the pairs .F; T/ of torsion-free F and torsion T that
satisfy Ext.F; T/ D 0. So far the best approach is via Baer cotorsion-pairs .F ; T /. They are
defined to consist of a class F of torsion-free and a class T of torsion groups such that Ext.F; T/ D
0 for all F 2 F , T 2 T , and these classes are saturated. Assuming V = L, Strüngmann [2] classified
these pairs in terms of sets P of primes with ˚p2P Z.p/ 2 T along with sets Q of primes p with
˚n<! Z.pn / 2 T . In another paper [1], he considered the class CG of all torsion groups T which
satisfy Ext.G; T/ D 0 for a torsion-free group G. He showed that if G is countable, then CG D CC
for a suitable countable completely decomposable group C. Again assuming V = L, he also proved
the existence of -universal groups G for every torsion group T that has but finitely many bounded
p-components ¤ 0. Here -universal for a T means that jGj D , Ext.G; T/ D 0, and every group
H of cardinality   that satisfies Ext.H; T/ D 0 embeds in G.
Under the hypothesis V = L, Eklof [4] shows that, for a torsion group T and for a torsion-
free group G of singular cardinality , Ext.G; T/ D 0 holds if and only if Ext.C; T/ D 0 for all
subgroups C < G of cardinality < .
Kaplansky raised the question of Baer modules B over integral domains R: which R-modules
B satisfy Ext1R .B; T/ D 0 for all torsion modules T? It took much more time than for abelian
groups to get the complete solution. After sporadic publications by various authors dealing with
special cases, Eklof–Fuchs–Shelah [Trans. Amer. Math. Soc. 322, 547–560 (1990)] reduced the
problem to countably generated modules B, and finally, Angeleri-Hügel–Bazzoni–Herbera [ibid.
360, 2409–2421 (2008)] succeeded in proving that countable Baer modules are projective. Thus
all Baer modules over any integral domain are projective.
2 Baer Groups are Free 583

One does not need a sharp eye to notice the difference when we ask for groups that force Ext
to vanish for a class of groups (like in the Baer problem) or just for a single group (Whitehead
problem). The Baer problem is an excellent example to demonstrate the difference when we
admit all torsion groups or just one of fixed (countable) cardinality. The common thread in both
cases is that advanced set-theoretical methods are required, but the latter case is more prone to
undecidability.

Exercises

(1) (Sa̧siada) Let B be a basic subgroup of a torsion group T, and G a torsion-free

group.
(a) Show that Ext.G; T/ D 0 if and only if Ext.G; B/ D 0.
(b) Conclude that in studying pairs T; G for which Ext.G; T/ D 0, it suffices
to consider for T only †-cyclic torsion groups.
(2) Let A be a mixed group with torsion part T such that A=T is a separable
torsion-free group, not completely decomposable. A need not split even if all
of its subgroups split that contain T and have finite torsion-free rank.
(3) (Griffith) If G satisfies Ext.G; S/ D 0 for all p-groups S, then Ext.G; T/ is
torsion-free for all torsion groups T. [Hint: if Tp D 0, then multiplication by p
is an automorphism on T, and hence on Ext.]
(4) (Griffith) Let G be a torsion-free group such that every extension of every
torsion group by G is quasi-splitting. Then G is free.
(5) (Strüngmann) Let G be torsion-free of finite rank. There is a rational group R
such that, for all torsion groups T, Ext.G; T/ D 0 is equivalent to Ext.R; T/ D
0. [Hint: outer type.]
(6) (Baer) Let T and G be torsion and torsion-free groups, respectively, such that
Ext.G; T/ D 0. Then:
(a) Suppose p1 ; : : : ; pn ; : : : is an infinite sequence of different primes satis-
fying pn T < T for all n. G cannot contain any pure finite rank subgroup S
such that G=S has elements ¤ 0 divisible by all the pn .
(b) If for some prime p, the p-basic subgroup of T is unbounded, then G
contains no pure subgroup S of finite rank such that G=S has elements
¤ 0 of infinite p-height.
(7) (Baer) If G in the preceding exercise is countable, then conditions (a) and (b)
are necessary and sufficient for Ext.G; T/ D 0 for any torsion T. [Hint: write
G as the union of a chain of finite rank pure subgroups, and use induction.]
(8) (Strüngmann) Let G be countable and torsion-free. Ext.G; T/ D 0 for some
torsion T if and only if Ext.H; T/ D 0 for all finite rank subgroups H of G.
[Hint: Lemma 4.1 in Chapter 9.]
(9) (Pext version of Baer’s problem) A group G satisfying Pext.G; T/ D 0 for all
torsion groups T is a †-cyclic group. [Hint: pure-projective resolution.]
584 15 Mixed Groups

(10) (Megibben) Call a mixed group separable if every finite set of elements is
contained in a direct sum of a group with minimum condition and a completely
decomposable torsion-free group.
(a) A is separable if and only if so are tA, A=tA, and tA is balanced in A.
(b) A separable mixed group need not be splitting. [Hint: extend a torsion by
an @1 -free.]
(11) (Megibben) A torsion-free group A satisfies Bext.A; T/ D Ext.A; T/ for all
torsion groups T if and only if it is homogeneous of type .0; : : : ; 0; : : : /.

3 Valuated Groups. Height-Matrices

The fundamental properties of heights are formalized in the concept which we

are going to study here: valuation. Valuated groups, as we shall see soon, are
indispensable tools in the theory of mixed groups. Their theory was developed in
the fourth quarter of the last century.
Ordinal Valuation Consider the class  of ordinals with the symbol 1
adjoined as the maximum element, and let p be an arbitrary, but fixed prime. Let
vp be a function on the group A to  subject to the following conditions (we agree
that  < 1 for all  2 ):
(i) vp .a/ 2  for all a 2 A; vp .0/ D 1;
(ii) vp .a C b/  min fvp .a/; vp .b/g for all a; b 2 A;
(iii) vp .pa/ > vp .a/ for all a 2 A n f0g;
(iv) vp .na/ D vp .a/ for all a 2 A and all integers n prime to p.
In this case, vp will be called a p-valuation of A. A p-valuated group is a group
together with a p-valuation vp .
A subgroup C of a p-valuated group A carries the inherited valuation. The factor
group A=C is equipped with the supremum valuation, i.e. to the coset aCC we assign
the value supc2C vp .a C c/. In this way, the canonical homomorphism  W A ! A=C
becomes a valuated morphism in the sense explained below. (Actually, the assigned
values in A=C are the smallest for which  is a genuine morphism.)
Example 3.1. An immediate example for a p-valuation is the p-height function hp .a/ on a group
A. From the definition of hp .a/ it is immediate that vp .a/
hp .a/ holds for any p-valuation vp of
a 2 A.
Example 3.2. A similar example is obtained if A is embedded as a subgroup in a larger group G,
and for a 2 A, vp .a/ is defined as the p-height hp .a/, this time computed in G. (We shall see below
in Theorem 3.6 that actually every p-valuation arises in this way by a suitable embedding.)
Example 3.3. Let 0 < 1 <    < n < : : : be any strictly increasing sequence of ordinals. We
define a p-valuation on an infinite cyclic group hgi by setting vp .mpn g/ D n for each n 2 N and
for every integer m prime to p. hgi is freely p-valuated (see below) if 0 is arbitrarily chosen and
n D 0 C n for each n < ! (thus the values are totally determined once vp .g/ is fixed for the
generator g).
3 Valuated Groups. Height-Matrices 585

Example 3.4. If Ai .i Q2 I/ are p-valuated groups, then their direct sum (i.e., coproduct) ˚ Ai and
their direct product Ai become p-valuated groups if we set vp .a/ D inf fvp .ai / j i 2 supp ag
where a D .: : : ; ai ; : : : / .ai 2 Ai /. (There is no danger of confusion by denoting the p-valuation in
all occurring groups by the same symbol vp .)

Category Vp For p-valuated groups A and B, a morphism  W A ! B is a group

homomorphism satisfying

vp ..a//  vp .a/ for all a 2 A:

An isometry W A ! B between valuated groups is a group isomorphism that

preserves p-values (i.e. vp . .a// D vp .a/ holds for all a 2 A), and an embedding
is an injective group homomorphism that preserves p-values.
For the elements a of a p-valuated group whose orders are finite and prime to p,
we have necessarily vp .a/ D 1 (this follows from comparing conditions (iii) and
(iv) in the definition). Consequently, no p-valuation can give any useful information
about such elements. Therefore, it is reasonable to disregard them, and focus our
attention on p-local groups, i.e., on Z.p/ -modules. Thus we will be dealing with the
category Vp of p-local p-valuated groups with the indicated morphisms.
By a free p-valuated group F is meant a free group F D ˚hai i where each
infinite cyclic group hai i carries a free p-valuation; this means—we repeat—that
vp .ai / D i can be arbitrary for each i 2 I, and then vp .pn ai / D i C n for all n < !.
The valuation on F is defined by taking infima of the values of the coordinates (like
in Example 3.4 above). In other words, F is the direct sum of freely p-valuated
infinite cyclic groups.
Lemma 3.5. For every p-valuated p-local group A there is a free p-valuated group
F such that A is isometric to the p-valuated quotient F=H for some subgroup H of F.
Proof. For every 0 ¤ a 2 A, choose a freely p-valuated infinite cyclic group hxa i
such that vp .xa / D vp .a/, and let F be the p-valuated direct sum of all the hxa i. Then
the surjective mapping  W F ! A (induced by xa 7! a) is easily seen not to increase
values. Therefore, the induced map F= Ker  ! A must be an isometry. t
u
Note that Ker  in the last proof is a p-nice subgroup of F in the sense to be
defined in the next section; i.e. each coset of F mod Ker  contains an element that
has the same value as the coset.
Theorem 3.6 (Richman–Walker [5]). Let Vp be the category defined above. There
exist a functor TW Vp ! Vp and a natural transformation 1 T such that, for each
group A 2 Vp with p-valuation vp ,
(a) T.A/ has the p-height valuation hp ;
(b) A is a p-nice subgroup of T.A/, and hp .a/ D vp .a/ for a 2 A;
(c) T.A/=A is a totally projective p-group.
Proof. For every 0 ¤ a 2 A, consider the Nunke group Na .p/ where  D vp .a/,
and let xa 2 Na .p/ denote its generator of infinite order (of height ). As hxa i is
p-nice in Na .p/, F D ˚0¤a2A hxa i will be a p-nice subgroup of G D ˚0¤a2A Na .p/.
586 15 Mixed Groups

We denote by W F ! A the map induced by xa 7! a, and set H D Ker . Define

T.A/ D G=H, so that A D F=H is a subgroup in T.A/. If T.A/ is furnished with
the height valuation, then—as is straightforward to check—the restriction to A will
be identical with vp . The p-niceness of A in T.A/ follows from that of F in G. The
functorial behavior of the process is evident from the definition. (c) is an obvious
consequence of the definition of Nunke groups. t
u
Global Valuation Turning to the global case, we introduce a valuation where
the values are no longer ordinals and symbols 1, but countable sequences  D
.1 ; 2 ; : : : ; n ; : : : / of ordinals and 1 (like characteristics for torsion-free groups).
Suppose A has p-valuation vp for each prime p. For a 2 A, we set

.a/ D .v2 .a/; v3 .a/; : : : ; vpn .a/; : : : /

where pn stands for the nth prime. The point-wise ordering makes the collection of
these sequences a lattice-ordered class with .0; 0; : : : ; 0; : : : / as minimum and with
.1; 1; : : : ; 1; : : : / as maximum element. This ‘global valuation’  (which we
might call ‘characteristic’) satisfies conditions (i)–(iv) listed above for p-valuations.
The characteristics  (as above) and 0 D .10 ; 20 ; : : : ; n0 ; : : : / will be regarded
equivalent, in notation:  0 , if n D n0 for almost all n. For instance,
.nx/ .x/ holds for all n 2 N. We will write Œ
for the equivalence class of
the characteristic . The collection of these Œ
carries a lattice-order.
Let C D hxi be an infinite cyclic group that is furnished with a free global
valuation, given by an arbitrarily chosen characteristic .x/ D .2 ; 3 ; : : : ; p ; : : : /.
The cyclic groups C D hxi and C0 D hx0 i are of the same type if the characteristics
.x/ and .x0 / are equivalent.
Height-Matrices In mixed groups A, we have to consider the overall divisibil-
ity; this can be accomplished by listing the indicators of an element for all primes.
Let p1 ; : : : ; pn ; : : : denote the sequence of the prime numbers in order of magnitude.
With an element a 2 A we associate the !  ! height-matrix
0 1 0 1
hp1 .a/ hp1 .p1 a/ : : : hp1 .pk1 a/ ::: up1 .a/
B ::: B C
::: ::: ::: : : :C
CDB ::: C
H.a/ D B
@hp .a/ hp .pn a/ : : : hp .pk a/ B C D knk kI
n n n n : : :A @upn .a/A
::: ::: ::: ::: ::: :::

(see Rotman [1], Megibben [3], Myshkin [1]) where the first column is .a/, and
the nth row represents the pn -indicator of a. Keep in mind that the entry nk 2 H.a/
in the .n; k/-position records the pn -height of pkn a, for all n 2 N and k < !.
The class of height-matrices is lattice-ordered in the obvious way:
0
H D knk k
knk k D H0
3 Valuated Groups. Height-Matrices 587

0
means that nk
nk for all n; k < !. Thus the lattice operations ^; _ between
height-matrices are the point-wise minima and maxima.
The following observations are evident from the definition.
(a) H.a/ D H.a/ for all a 2 A.
(b) H.pn a/ is obtained from H.a/ by deleting the first entry in the nth row, and
shifting the rest of the row to the left.
(c) If a is of finite order, then almost all entries in H.a/ are 1.
(d) Every entry in H.a/ is 1 if and only if a is contained in the divisible subgroup
of A.
(e) pk11    pknn a is divisible by m D p`11    p`nn exactly if `i
iki for i D 1; : : : ; n.
(f) If A D B ˚ C and a D b C c .b 2 B; c 2 C/, then H.a/ D H.b/ ^ H.c/:
Example 3.7. Let B D ˚1 nD1 hbn i with o.bn / D p
2n1 Q , the p-adic completion of B.
, and A D B
Let j0 <    < jk < : : : denote a strictly increasing sequence of non-negative integers. Our claim
is that A contains an element a with hp .pk a/ D jk for every k < ! if and only if a gap jk C 1 < jkC1
implies that jk is even. To verify, we argue as in Lemma 1.1 in Chapter 10.

Equivalence of Height-Matrices If the elements a; b of infinite order in the

mixed group A satisfy ra D sb for some integers r; s ¤ 0, then from (b) it is easily
seen that the nth rows of H.a/ D knk k and H.b/ D knk k can differ only if pn jrs.
If this is the case for a prime pn , then there must exist integers i; j  0 such that

n;iCk D n;jCk for all k < !: (15.4)

In view of this, we define two ! ! matrices M D knk k and N D knk k equivalent

(in notation: M N) if, for almost all n, the nth rows are identical, and for each
distinct row there exist integers i; j  0 (depending on the row) such that (15.4)
holds. Equivalently, if pH for a prime p and a height-matrix H is defined as a matrix
obtained from H by shifting the p-row one position to the left, and nH for n 2 N by
induction, then M N means that there are m; n 2 N satisfying mM D nN.
We note that the height-matrices of elements in the same coset modulo the torsion
subgroup are always equivalent.
Characterization of Height-Matrices One of the fundamental questions relat-
ing height-matrices is concerned with conditions under which a matrix of ordinals
and symbols 1 can be realized as a height-matrix in some mixed group. The answer
is not difficult, but it is not easy either. (Recall that we have agreed that 1 < 1.)
Theorem 3.8 (Megibben [3], Rotman [1]). An !  ! matrix M D knk k
consisting of ordinals and symbols 1 is the height-matrix of an element in some
mixed group A if and only if, for every n, it satisfies

n0 < n1 <    < nk < : : : :

Proof. Necessity being evident, we turn to the proof of sufficiency. Given the matrix
M D knk k satisfying the stated condition, the strategy is to build a mixed group
of torsion-free rank 1 by equipping the elements of an infinite cyclic group with
the correct p-heights, for every prime p. This is first done separately for each prime.
588 15 Mixed Groups

Thus for each prime pn and integer k  0, we select an infinite cyclic group, and call
ank its generator. We build a group Ank around hank i. If nk D 1, then let Ank be the
rational group with Ank =hank i Š Z.p1
n /. If nk is an ordinal, then let Ank Š Nnk .pn /
be the Nunke group (Sect. 1), with ank as the generator of height nk in its infinite
cyclic subgroup. Choose an extra infinite cyclic group hai, and define A by the push-
out diagram

0 −−−−−→ ⊕n,k an,k  −−−−−→ ⊕n,k An,k −−−−−→ ⊕n,k An,k /an,k  −−−−−→ 0
⏐ ⏐ 
⏐ ⏐ 
δ  
α
0 −−−−−→ a −−−−−→ A −−−−−→ ⊕n,k An,k /an,k  −−−−−→ 0,

where ı is defined via ı W an;k 7! pkn a for all n 2 N; k < !. It is rather obvious that
the height-matrix of ˛.a/ 2 A is  M, and it is easy to see that equality must hold.
For instance, if we factor out hpn ˛.a/i from A, then we get a huge direct sum of
generalized Prüfer groups H .pn /’s or Z.p1 n /, and ˛.a/ lies in a single summand,
not effected by the other summands, and therefore it will carry the correct pn -height.
t
u
With every (mixed) group A and every height-matrix H we associate two fully
invariant subgroups; they are defined as

A.H/ D fa 2 A j H.a/  Hg and A .H/ D ha 2 A j H.a/ > Hi:

Obviously, A .H/
A.H/, and A.H/
A.H0 / whenever H  H0 . In general, there
exist also other fully invariant subgroups in A.
Example 3.9. Let A be a non-splitting mixed group such that all of the p-components of tA are
unbounded. Then the minimal height-matrix H for which tA  A.H/ holds is H D knk k where
nk D k for all n. For this height-matrix, we have A.H/ D A.
F Notes. By now it should be clear that besides the order, the height is the most valuable
information about a group element, and valuation is just a natural abstract version of height. The
theory of valuation of groups developed slowly. Rotman [2] observed that a finite rank torsion-free
group is determined by an essential free subgroup furnished with the height-valuation. Richman–
Walker [5] is the pioneering work on valuations. Undoubtedly, much of the credit for a systematic
theory of valuated groups goes to the New Mexico group, above all to Hunter, Richman, and E.
Walker.
A few more words about Vp . It is a pre-abelian category, i.e. it is additive, and morphisms have
˛ ˇ
kernels and cokernels. An exact sequence 0 ! A!B!C ! 0 makes perfectly good sense if
it means that ˛ is an embedding, and ˇ induces an isometry B=A Š C (thus A is like a balanced
subgroup). There are enough projectives (the free valuated groups) and injectives (algebraically
compact groups with height-valuation). Interested reader should consult Richman–Walker [5] for
further information.
Theorem 3.6 generalizes the ‘crude existence theorem’ by Rotman–Yen [1] in which A was an
infinite cyclic group.
Valuated groups might be the source of easily tractable invariants that would help in the future
to obtain reasonable classification of groups in some classes. We are spoiled by the tremendous
success of numerical and cardinal invariants in the traditional structure theorems, and perhaps
3 Valuated Groups. Height-Matrices 589

do not try hard enough to obtain other kinds of more complicated, but still simply recognizable
invariants. We have to learn more about valuation, and look for invariants in the category Vp of
p-valuated p-local groups.
It might, perhaps, be appropriate to mention another kind of valuation used in the literature: the
‘coset valuation.’ Let C be a subgroup of A. The p-height of the coset a C C in A=C is calculated
ordinarily as supc2C hp .a C c/ (or 1). In some situations, it is desirable to distinguish according
as this supremum is equal to one of hp .a C c/ (i.e., the coset contains an element proper with
respect to C) or not. This can be done, e.g. by defining the coset valuation as follows: kp .a C C/ D
supc2C .hp .a C c/ C 1/.
Grinshpon–Krylov [1] published an extensive study of transitivity and full transitivity in mixed
groups. A is fully transitive if for all x; y 2 A with H.x/  H.y/, there is an  2 End A such that
.x/ D y. Inter alia, the full transitivity of direct sums is investigated. See also Misyakov [1] and
Grinshpon–Misyakov [1].
Let us mention here some literature on universal embeddings. A group U is (purely) universal
for a set S of groups, if U 2 S and every group in S embeds as a (pure) subgroup in U. Kojman–
Shelah [1] consider various classes of interest; their main set-theoretic tools are club guessing
sequences.

Exercises

(1) Let A be p-local, and vp a p-valuation in A. For  2 , define A./ D fa 2 A j

vp .a/  g. Show that the collection of the A./ determine vp .
(2) Let A be a p-local mixed group. The indicators of elements in A satisfy the gap
condition (see Sect. 1 in Chapter 10).
(3) Prove the inequality for height-matrices: H.a C b/  H.a/ ^ H.b/ for all
a; b 2 A.
(4) Define A D ha0 ; a1 ; : : : ; an ; : : : i subject to the relations pn a0 D pkn an for n  1,
where 0 < 1 < k1 <    < kn < : : : are integers. Describe the height-matrices
H.a0 / and H.an /.
(5) a 2 A belongs to the th Ulm subgroup A of A if and only if H.a/ D knk k
satisfies n0  ! for all n.
(6) Examine what can happen to a height-matrix when passing (a) to a subgroup;
(b) to a factor group.
(7) (Megibben) Let A be a countable mixed group of torsion-free rank 1, a 2 A of
order 1. The following conditions on H.a/ are necessary and sufficient for A
to split:
(a) almost every row is free of gaps;
(b) no row has infinitely many gaps;
(c) if a row contains entries  !, then it also contains 1.
(8) Find a necessary and sufficient condition on the height-matrix to belong to a
quasi-splitting mixed group of torsion-free rank 1.
590 15 Mixed Groups

4 Nice, Isotype, and Balanced Subgroups

Before embarking on the discussion of a variety of mixed groups, now away from
splitting questions, we should agree to keep without modification those concepts
and notations used for torsion and torsion-free groups that make sense for the mixed
case. Accordingly, e.g. Ulm subgroups, UK-invariants (separately for each prime)
as well as Hill invariants will be used without explanation.
However, we have to adapt to mixed groups certain concepts developed for
torsion groups. We take a similar approach for the extended concepts, and since
some proofs are almost, if not totally, identical to the torsion case, we will just wave
hands, and refer to the appropriate places for verification.
Nice Subgroups As before, for a prime p, by a p-nice subgroup of A is meant
a subgroup N such that

.p A C N/=N D p .A=N/ for all ordinals : (15.5)

The p-niceness of N is equivalent to the representability of every coset of N in A by

an element of the p-height of the coset. Such a representative a is called p-proper
with respect to N. It satisfies

hp .a/  hp .a C x/ for all x 2 N:

A subgroup N of A will be called nice in A if, for every prime p, the localization
N.p/ D N ˝ Z.p/ is p-nice in A.p/ D A ˝ Z.p/ .
Next we list some relevant properties of niceness (p-niceness).
(a) Finite subgroups are always nice.
(b) Let Ni be a subgroup of the group Ai , for i 2 I. The subgroup ˚i2I Ni is nice
.p-nice/ in ˚i2I Ai if and only if Ni is nice .p-nice/ in Ai for all i 2 I.
(c) If N is contained in a subgroup B of A, then N need not be nice in B even if N is
nice in A.
(d) Niceness is not transitive in general. See Sect. 2(E) in Chapter 11.
Lemma 4.1. Let N
M
A where N is nice in A. Then M is nice in A if and only
if M=N is nice in A=N.
Proof. We refer to the proof of Lemma 2.4 in Chapter 11. t
u
The next lemma was known for a while for Jp -modules, valid for finitely
generated G. For groups, only a weaker version can be established.
Lemma 4.2 (Rotman [1], Megibben [3], Wallace [1]). Let A be a mixed group of
torsion-free rank 1, and G D hgi an infinite cyclic subgroup. If A=G is a p-group,
then G is p-nice in A.
Proof. We prove that cosets contain elements proper with respect to G. Let a 2 AnG
satisfy pt a D ps rg for some t  1; s  0, and gcdfr; pg D 1. Suppose there is
4 Nice, Isotype, and Balanced Subgroups 591

a strictly increasing sequence hp .a C ni g/ .i < !/ of ordinals, where ni 2 Z and

n0 D 0. By the triangle inequality, hp .ni g/ D hp .a/, so there is k 2 N such that
ni D pk mi with gcd.mi ; p/ D 1 for all i. If s ¤ k C t, then

hp .a C ni g/ < hp .pt a C pkCt mi g/ D hp .ps rg C pkCt mi g/ D hp .pj g/

where j D minfs; k C tg. Since hp ..niC1  ni /g/ D hp .a C ni g/ is increasing with

i, it is clear that the coefficients niC1  ni are divisible by increasing powers of p,
so it is impossible to have hp .a C ni g/ < hp .pj g/ for all i. Consequently, there is
a maximal height in the set fhp .a C ni g/ j i < !g, and the coset a C hgi contains
a proper element. If s D k C t, then we replace a by a  pk rg in our argument, so
pt a D 0 may be assumed. In this case, hp .a C ni g/ < hp .pkCt mi g/ D hp .pkCt g/
leads to a similar contradiction. t
u
Isotype Subgroups A straightforward generalization of isotypeness in p-groups
is the following concept for mixed groups. A subgroup G of the (mixed) group A is
called p-isotype if

p G D G \ p A for all ordinals : (15.6)

G is isotype if it is p-isotype for each prime p.

Isotype subgroups have properties (a)–(d) listed in Sect. 5 in Chapter 11.
Example 4.3 (Megibben [3]). A subgroup G of A with torsion-free factor group A=G is isotype in
A. In view of the definition of p A for limit ordinals  , for a transfinite induction it will suffice to
prove that if (15.6) holds for an ordinal  , then it also holds for  C 1. So pick g 2 G \ pC1 A,
and let a 2 p A satisfy pa D g. By the torsion-freeness of A=G, we have a 2 G, thus g 2 pC1 G.

The following lemma is readily verified from the definitions.

Lemma 4.4. If C is p-isotype in A, then

C.u/ D C \ A.u/ for all p-indicators u;

and if C is isotype in A, then

C.H/ D C \ A.H/ for all height-matrices H: t

u
˛ ˇ
Balanced Subgroups A short exact sequence 0 ! A!B!C ! 0 is called
balanced-exact if the induced sequence

˛ ˇ
0 ! A./!B./!C./ ! 0 (15.7)

is exact for every characteristic ; we recall: A./ D fa 2 A j .a/  g. (15.7) is

called H-exact if, for every height-matrix H, the induced sequence

ˇ
0 ! A.H/ ! B.H/!C.H/ ! 0 (15.8)
592 15 Mixed Groups

is exact. H-exact sequences are evidently balanced-exact. However, the converse is

false (see infra ), though they are equivalent for torsion and torsion-free groups. In
fact, for p-groups this follows from Proposition 5.5 in Chapter 11, while for torsion-
free groups, we observe that height-matrices are determined by their first columns,
i.e. by the characteristics.
A useful criterion for balanced-exactness is given in the next lemma.
ˇ
Lemma 4.5. The sequence 0 ! A ! B!C ! 0 is balanced-exact if and only if
(i) ˇ.B.// D C./ for every characteristic , and
(ii) ˇ.p BŒp
/ D p CŒp
for every prime p and ordinal .
Proof. Condition (i) asserts that (15.7) is exact at C./. The rest of the proof is the
same as in Proposition 5.5 in Chapter 11. t
u
An entirely analogous proof applies to verify:
ˇ
Lemma 4.6. The sequence 0 ! A ! B!C ! 0 is H-exact if and only if
(i) ˇ.B.H// D C.H/ for every height-matrix H, and
(ii) ˇ.p BŒp
/ D p CŒp
for every prime p and every ordinal . t
u
F Notes. This section contains three fundamental concepts needed in the study of mixed
groups. Our discussion above shows the overwhelming influence of p-groups on these concepts.
A brief comment on the definition of niceness. Naturally, one is tempted to define global
niceness as p-niceness for every prime p. The bad news is that under this definition, e.g. pZ would
not be nice in Z (the elements of Z=pZ have infinite q-heights for every prime q ¤ p). The good
news is that this can easily be corrected: localizations provide a useful concept (so that pZ is nice
in Z).
For mixed groups over a complete discrete valuation domain (like Jp ) the situation is more
favorable: they are more tractable as we work with one prime only, and the important Lemma 4.2
holds more generally for finitely generated submodules G (Rotman [1], Rotman–Yen [1], and C.M.
Bang [Proc. Amer. Math. Soc. 28, 381–388 (1971)]).

Exercises

(1) p A is p-nice in a mixed group A for every ordinal .

(2) Prove that isotypeness is transitive also for mixed groups.
(3) If C is an isotype subgroup of the mixed group A, then the height-matrix of any
c 2 C is the same whether computed in C or in A.
(4) If 0 ! A ! B ! C ! 0 is a balanced-exact sequence, then the induced
sequence 0 ! DA ! DB ! DC ! 0 of the divisible subgroups is splitting
exact. [Hint: choose  in (15.7) properly.]
(5) (Irwin–Walker–Walker) An exact sequence 0 ! A ! B ! C ! 0 represents
an element of p1 Ext.C; A/ if and only if the induced sequence 0 ! Ap !
Bp ! Cp ! 0 of p-components is splitting exact.
5 Mixed Groups of Torsion-Free Rank One 593

(6) For every increasing sequence 0 < 1 <    < n < : : : of ordinals, there is a
Z.p/ -module A which contains an element x of infinite order such that
(a) hp .pn x/ D n for each n < !;
(b) X D Z.p/ x is a nice submodule of A;
(c) A=X is a totally projective p-group of length supn n . [Hint: Theorem 3.8.]

5 Mixed Groups of Torsion-Free Rank One

At this point we reduce generality, and restrict our considerations to mixed groups
of torsion-free rank 1. The reason for our focusing on this special case lies in our
wanting to learn the basics of how an infinite cyclic group and a torsion group blend
to form a tractable mixed group. With the help of height-matrices and UK-invariants,
a satisfactory structure theorem will be obtained for a class of such groups without
much effort (see Theorem 5.2). Already this result will give the flavor of both the
beauty and the difficulties in dealing with mixed groups.
The first thing we observe is that in a mixed group of torsion-free rank 1, any
two elements of infinite order are dependent, so their height-matrices are equivalent.
Consequently, there is a well-defined equivalence class of matrices associated with
such a group A; we shall denote it by H.A/, and call it the height-matrix of A.
In what follows, we will often deal with height-preserving isomorphisms  W
G ! H between subgroups G of A and H of C, i.e. hp .g/ D hp .g/ for all g 2 G
and for all primes p, where the heights are computed in the containing groups C
and A, respectively. This situation will be crystalized in introducing a category in
Sect. 8.
The following lemma provides an important step in the proof of Theorem 5.2
inasmuch as it tells how to compute the Hill invariants relative to an infinite cyclic
subgroup from the UK-invariants of the group. Recall that f .A/ and f .A; G/ denote
the th UK-invariant of A for a given prime, and its th Hill invariant with respect
to the subgroup G, respectively.
Lemma 5.1. Let A be a group of torsion-free rank 1, and G an infinite cyclic
subgroup such that A=G is a p-group. Then
(
f .A; G/ C 1 if the p–row of H.A/ has a gap at ;
f .A/ D
f .A; G/ if there is no such gap:

Proof. Recall that we have f .A; G/ D p AŒp

=G./, and suppose that G./ D
.p C1 ACG/\p AŒp
¤ 0, i.e. there is g 2 G such that hp .g/ D  and aCg 2 p AŒp

for some a 2 p C1 A: Then pg D pa 2 p C2 A, thus g represents a gap in the p-row

at . Evidently, dim G./
1, showing that f .A/ D f .A; G/ C 1. Conversely, if
the p-row of H.g/ has a gap at , then the reverse argument yields G./ ¤ 0. Hence
the claim follows. t
u
594 15 Mixed Groups

We have come to the classification theorem on mixed groups of torsion-free

rank 1. It was proved by Rotman [2], Megibben [3], Myshkin [1] for countable
groups, and generalized by Warfield [7].
Theorem 5.2 (Warfield [7]). Let A and C be mixed groups of torsion-free rank 1,
G in A and H in C infinite cyclic subgroups such that A=G and C=H are totally
projective torsion groups. A Š C if and only if
(i) the UK-invariants of A and C are equal for all primes p; and
(ii) the height-matrices H.A/ and H.C/ are equivalent.
Proof. It is obvious that the stated conditions are necessary. To verify sufficiency,
assume (i)–(ii). By (ii), there are a 2 G and c 2 H, both of infinite order with
the same height-matrix, thus a 7! c induces a height-preserving isomorphism  0 W
hai ! hci.
Let A.p/ denote the subgroup of A such that A.p/=hai D .A=hai/p ; thus A.p/=hai
is a totally projective p-group (observe that A=G and A=pG are simultaneously
totally projective). Let C.p/ have similar meaning for C. In view of Lemma 4.2, hai
is p-nice in A.p/, and hci is p-nice in C.p/. Furthermore, because of the equality of
height-matrices, Lemma 5.1 guarantees that the relative UK-invariants f .A.p/; hai/
and f .C.p/; hci/ are the same for all p and . Thus we have all the ingredients
to apply Lemma 4.5 in Chapter 11, and to conclude that there is an isomorphism
p W A.p/ ! C.p/ extending  0 . The difficult work having been done, it only remains
to paste together the isomorphisms p for all primes. This can be accomplished by
the commutative diagram

α
⊕p A(p) −−−−→ A
⏐
⏐
⊕φp  φ

γ
⊕p C(p) −−−−→ C

where the maps ˛; are induced by the component-wise embeddings, and  is

uniquely determined by the commutativity of the diagram. It is straightforward to
verify that  is an isomorphism. t
u
It is, perhaps, worthwhile pointing out that the last theorem may fail whenever
the factor groups A=G and C=H are not totally projective; this is demonstrated by
the following example.
Example 5.3 (Megibben [3]). Let B be an unbounded countable direct sum of cyclic p-groups, and
B its torsion-completion. Let T be a pure subgroup of B containing B such that B=T Š Z.p1 /.
The exact sequence 0 ! Hom.Z; T/ Š T ! Ext.Q=Z; T/ ! Ext.Q; T/ Š ˚Q ! 0 and
the reducedness of the first Ext imply that Ext.Z.p1 /; T/ contains a mixed group G such that
T < G; G=T Š Q and G1 ¤ 0. In a similar fashion, we obtain a subgroup H of Ext.Z.p1 /; B/
such that B < H; H=B Š Q and H 1 ¤ 0. There are elements g 2 G; h 2 H of infinite order
5 Mixed Groups of Torsion-Free Rank One 595

such that hp .pk g/ D ! C k D hp .pk h/ for k < !, so that A D G ˚ B and C D H ˚ T have

equivalent height-matrices H.A/  H.C/; besides being of torsion-free rank 1, and having the
same UK-invariants. In order to justify the claim that A 6Š C, we prove:

.a/ G=G1 Š B; .b/ H=H 1 Š B; .c/ B ˚ T 6Š B ˚ B:

(a) follows from the fact that if T is pure in a separable p-group E such that E=T Š Z.p1 /,
then E Š B (Sect. 3 in Chapter 10, Exercise 15). Since H=H 1 is a countable separable p-group,
and B is isomorphic to a pure subgroup with Z.p1 / as quotient, (b) follows from Theorem 3.5 in
Chapter 10. Finally, the proof of Theorem 9.5 in Chapter 10 shows that B ˚ T Š B ˚ B could hold
only if B ˚ T had a summand isomorphic to B, which is evidently not the case.

We observe that in Theorem 5.2 the assumption that G and H were infinite cyclic
groups was used in the proof only in the form that they were nice and isomorphic.
Consequently, the proof works also for the following more general result, it needs
no adaptation.
Theorem 5.4. Let A and C be mixed groups, G < A and H < C nice subgroups
such that
(i) there is a height-preserving isomorphism  W G ! H; and
(ii) A=G and C=H are isomorphic totally projective torsion groups.
Then there is an isomorphism   W A ! C such that    G D . t
u
F Notes. The first relevant result on the structure of non-splitting mixed groups is due to
Kaplansky–Mackey [1] who proved that two countably generated modules of torsion-free rank
1 over a complete discrete valuation domain are isomorphic if and only if they have equivalent
characteristics and the same UK-invariants. The completeness hypothesis was later removed by
Megibben [3], making the Kaplansky–Mackey result an honest abelian group theorem. Rotman–
Yen [1] proved the isomorphy of countably generated modules of finite torsion-free ranks over
a complete discrete valuation domain, under the hypothesis that they have the same height-
matrices and the same UK-invariants. Stratton [1] showed that the Rotman–Yen result fails if the
completeness condition is abandoned. Wallace [1] proves that two p-local groups of torsion-free
rank 1 with totally projective torsion subgroups are isomorphic exactly if they have equivalent
height-matrices and the same UK-invariants.
For countable mixed groups of torsion-free rank 1, the isomorphism theorem was proved
independently by Rotman [2], Megibben [3], Myshkin [1].
Files [2] considers transitivity and full transitivity for mixed groups of torsion-free rank 1. Such
a group enjoys both properties whenever its torsion subgroup is separable. Transitive properties
more generally were dealt with by Grinshpon–Krylov [1]; in this systematic study, a wealth of
interesting results on mixed groups was proved. Self-small mixed groups of finite torsion-free
ranks were discussed by Albrecht [6] (in particular, the A-projective groups), also by Albrecht–
Breaz–Wickless [1].

Exercises

(1) (Pierce) Let A be a p-group of length . There exists a mixed group G such that
p G Š Z and G=p G Š A.
596 15 Mixed Groups

(2) (Megibben) Prove Theorem 5.2 by replacing (i) by the condition that t.A/ Š
t.C/ are torsion-complete p-groups.
(3) Two countable mixed groups of torsion-free rank 1 are isomorphic if each is
isomorphic to an isotype subgroup of the other.
(4) (Rotman) Assume A; C are countable mixed groups of torsion-free rank 1, and
T is a torsion group with finite UK-invariants. Then A ˚ T Š C ˚ T implies
A Š C.
(5) (Rotman) Again, let A and C be countable mixed groups of torsion-free rank 1.
If A ˚ A Š C ˚ C, then A Š C.
(6) Find relations between the height-matrix and the UK-invariants of a countable
mixed group of torsion-free rank 1.
(7) (Warfield) Verify the one-sided version of Theorem 5.4 for a homomorphism
  W A ! C. Now  W G ! H is a homomorphism not decreasing heights, and
only the existence of a homomorphism A=G ! C=H is assumed.

6 Simply Presented Mixed Groups

Having gotten acquainted with mixed groups of torsion-free rank 1, we move

to groups of higher torsion-free ranks. Our next concern is the theory of simple
presented mixed groups which combines two prominent theories at an elevated
level: those on simply presented p-groups, and on completely decomposable torsion-
free groups. It requires new machinery to deal with some unexpected, if not
recondite, aspects.
A group A was called simply presented if it could be defined in terms of
generators and relations in such a way that each relation involves at most two
generators. Now the relations are of the form nx D 0 or mx D ny with generators
x; y and m; n 2 Z. Just as in the case of p-groups, no generality is lost by assuming
that all the relations are of the form px D 0 or px D y where x; y are generators, and
p is a prime (not fixed). Henceforth it will be assumed that the simple presentations
are faithful in the sense defined in Sect. 3 in Chapter 11, and ‘generator’ means one
in a faithful simple presentation.
Example 6.1. (a) The Nunke groups N are simply presented mixed groups of torsion-free
rank 1.
(b) A faithfully simply presented group is torsion if, for every generator x, there are a sequence
x D x0 ; : : : ; xk of generators, and a sequence of (not necessarily distinct) primes p0 ; : : : ; pk
such that pi1 xi1 D xi for i D 1; : : : ; k and pk xk D 0.
(c) We repeat: a torsion-free group is simply presented if and only if it is completely decompos-
able.

Our discussion starts with a most informative lemma.

Lemma 6.2 (Warfield [4]). A simply presented group is the direct sum of simply
presented groups whose torsion-free rank is at most 1.
6 Simply Presented Mixed Groups 597

Proof. Let X be the set of generators in a faithful simple presentation of A, and


0 ! H ! F !A ! 0 an exact sequence, where F is the free group on X, and
 acts as the identity map of X. An equivalence relation is defined for the elements
of X such that x y means that there are integers n; m 2 Z for which m.x/ D
n.y/ ¤ 0 holds in A. If Vi .i 2 I/ denote the different equivalence classes, then
every hVi i is a summand of F, and evidently F D ˚i2I hVi i. The subgroup H is
generated by the elements of the form mx  ny 2 F, each of which is contained in
some hVi i. Therefore, A Š F=H Š ˚i2I hVi i=.hVi i \ H/, where the summands are
either torsion or of torsion-free rank 1. t
u
A few useful facts are collected in the following statements. (The claims bear
more than a passing resemblance to the torsion case.)
(a) If A is simply presented on the set X, then for a subset Y  X, both hYi and
hX n Yi are simply presented.
(b) If A is simply presented on the set X, then hYi is a nice subgroup for every
Y  X. Because of Lemma 6.2, it suffices to prove this for rk0 A
1. If o.y/ D
1 .y 2 Y/, then A=hyi is simply presented torsion, so hYi=hyi is nice in it, as it
is p-nice for each prime (see Lemma 3.4 in Chapter 11). Since by Lemma 4.2
hyi is nice in A, from Lemma 4.1 we conclude that hYi is nice in A.
(c) For every height-matrix H, there exists a simply presented group whose height-
matrix is equivalent to H. A close examination of the proof of Theorem 3.8
shows that the mixed group constructed there is simply presented.
(d) The torsion subgroup of a simply presented group need not be simply presented.
This is illustrated by the following example.
Example 6.3 (Warfield). Consider the Nunke group G D N!1 .p/; thus p!1 G Š Z and G=p!1 G
is a simply presented p-group. Since G=tG Š Q.p/ , tG embeds in the totally projective p-group
G0 D G=p!1 G as a dense isotype subgroup H with cokernel Š Z.p1 /: G0 is of length !1 , so
G0 D ˚i2I Gi holds with countable summands Gi . To see that H cannot be simply presented,
observe that HŒp
< G0 Œp
, and by density G0 Œp
D HŒp
C p G0 Œp
for all  < !1 . If H were
simply presented, then H D ˚j2J Hj would hold with countable Hj . Then G0 D H C .˚i2I 0 Gi /
for a countable index set I 0 . If  < !1 is the length of ˚i2I 0 Gi , then p G0 Œp
D p HŒp
whence
G0 Œp
D HŒp
would follow, a contradiction.

Theorem 6.4 (Warfield [4]).

(i) A torsion summand of a simply presented mixed group is simply presented.
(ii) A mixed group A of torsion-free rank 1 is a summand of a simply presented
group if and only if there is an element a 2 A of infinite order such that A=hai
is a simply presented torsion group.
Proof. (i) Let G D T ˚ K be a simply presented mixed group where T is a torsion
group. From Lemma 6.2 we know that G D ˚i2I Gi where rk0 .Gi / D 1 for
every i 2 I. Let gi 2 Gi be a generator of infinite order; it can be replaced
by an integral multiple, so we can choose gi 2 K. Then G=F Š T ˚ .K=F/
where F denotes the (free) subgroup of G generated by the gi .i 2 I/. Hence
T is a summand of the simply presented torsion group G=F, so a reference to
Lemma 4.2 in Chapter 11 confirms that T is simply presented.
598 15 Mixed Groups

(ii) For the proof of sufficiency, assume a 2 A is as stated. In view of Theorem 3.8,
there exists a simply presented group G with A.H/ D G.H/. Choose a simply
presented torsion group T with large UK-invariants so that the UK-invariants of
A0 D A ˚ T and G0 D G ˚ T are equal to those of T. Since the height-matrices
have not changed by adding a torsion summand, we can appeal to Theorem 5.2
to conclude A0 Š G0 . It follows that A is isomorphic to a summand of the simply
presented G0 .
For necessity, assume G D A˚C is a simply presented group where rk0 .A/ D
1. Let Gi ; gi and F have the same meaning as in the proof of (i). Pick 0 ¤
a 2 A \ F, and select a finite subset J  I such that a 2 ˚i2J hgi i D F 0 .
Putting G0 D ˚i2J Gi , obviously, G0 =F 0 is simply presented and torsion. As
F 00 D hai ˚ .C \ F 0 / has finite index in F 0 , G0 =F 00 is also simply presented. It
follows that A=hai is isomorphic to a summand of the simply presented group
G0 =F 00 ˚ ˚i2InJ Gi . An appeal to (i) completes the proof. t
u
It should be pointed out that Theorem 6.4(ii) becomes false if ‘summands of’
is deleted. In fact, not all summands of simply presented mixed groups are simply
presented; see Example 7.1.
One might expect simply presented mixed groups to have the same kind of
classification as their torsion and torsion-free counterparts. However, as our next
example demonstrates, no such result can exist.
Example 6.5 (Warfield [4]). There exist simply presented groups A; B; C; D of torsion-free rank 1
with inequivalent height-matrices such that

A ˚ B Š C ˚ D:

In this example, the groups are countable and have no elements ¤ 0 of infinite height.
To define A; B, start with a 2 A; b 2 B of infinite orders. Adjoin generators in order to have:
h2 .2n a/ D h3 .3n a/ D 2n or 2n  1 according as n is even or odd; h2 .2n b/ D h3 .3n b/ D 2n for
all n; and hp .pn a/ D hp .pn b/ D n for all primes p
5. By Theorem 3.8, such A; B of torsion-
free rank 1 do exist. In the group G D A ˚ B, choose c D 9a C b; d D 8a C b. We then
have h2 .2n c/ D h2 .2n a/; h2 .2n d/ D h2 .2n b/ and h3 .3n c/ D h3 .3n b/; h3 .3n d/ D h3 .3n a/, while
hp .pn c/ D hp .pn d/ D n for p
5.
The height-matrices H.a/; H.b/; H.c/; H.d/ are inequivalent. It is a long, but rather straight-
forward calculation to verify that there exist subgroups C; D with c 2 C; d 2 D and G D C ˚ D.
(Actually, it would suffice to prove that fc; dg is a decomposition basis for G (see below), because
then it is easy to show that G has the stated decomposition.)
F Notes. Following the triumph with the classification of simply presented torsion groups,
the focus shifted to torsion-free and mixed groups. It was clear that one should attempt to adjust
the powerful and most successful methods acquired in the torsion case to mixed groups. Mixed
groups whose torsion subgroups are classifiable were the obvious targets, though this seemed a
remote possibility. The change in viewing a mixed group as an extension of a torsion-free group by
a torsion one, albeit not canonically, opened the door to a successful application of the arsenal. It is
fortunate that Warfield took the lead in laying down the foundations of a theory for tractable mixed
groups and in developing it into a remarkable theory. I think it is amazing that the UK-invariants
of the ignored torsion part can be coded as a valuation in a torsion-free group.
7 Warfield Groups 599

Exercises

(1) Every simply presented mixed group admits an H.@0 /-family of nice sub-
groups.
(2) Determine the height-matrix of the global Nunke group N .
(3) (a) Find the height-matrices of the summands in Example 6.5.
(b) Check that H.a/ C H.b/ D H.c/ C H.d/.

7 Warfield Groups

The study of the so-called Warfield groups is one of the most challenging tasks in the
theory of mixed groups. They are slightly more general than the simply presented
groups, but their theory is less transparent.
Let A be a mixed group. The natural homomorphism ˛p W A ! A.p/ maps a 2 A
to a ˝ 1 2 A.p/ D Q A ˝ Z.p/ . The maps ˛p combined for all primes p define an
embedding ˛ W A ! p A.p/ . If F is a freeQ subgroup of A such that A=F is torsion,
then ˛.A/=˛.F/ is the torsion subgroup of p A.p/ =˛.F/. Hence we deduce that the
cokernel of ˛ is torsion-free and divisible.
(a) If the torsion-free rank of A is 1, and a 2 A is of infinite order such that A=hai
is a totally projective p-group, then for every b 2 A of infinite order, A=hbi is
also totally projective. This follows from the linear dependence of a and b, and
from the total projectivity of an extension of a finite group hai=hmai by a totally
projective p-group A=hai (Lemma 7.12 in Chapter 11).
(b) If A is of torsion-free rank 1, and a 2 A is of infinite order, then A=hai is totally
projective whenever the torsion subgroup tA is a totally projective p-group. In
fact, tA is isomorphic to a subgroup of countable index in A=hai, and a p-group
is itself totally projective if it has a totally projective subgroup of countable
index (see Lemma 7.12 in Chapter 11).
Warfield Groups By a Warfield group is meant a summand of a simply
presented group. Thus simply presented torsion groups as well as completely
decomposable torsion-free groups are Warfield (but we usually think of mixed
groups when Warfield groups are mentioned). That not all mixed Warfield groups
are simply presented was first observed by Rotman–Yen [1] (in a different context);
we furnish the following counterexample due to Warfield [7].
Example 7.1. Consider the Z.p/ -module A generated by the symbols an .n < !/ subject to the
defining relations p2n an D pn a0 .n < !/. Its torsion subgroup tA D ˚n<! Z.p/ .an  panC1 /
satisfies A=tA Š Q. The p-indicator of a0 is u D .0; 2; 4; : : : /, while for other primes the indicators
are constant 1. A is simply presented as a group.
Suppose A D B ˚ C where rk0 B D 1 and C is a p-group. The indicator of any b 2 B of infinite
order is equivalent to u above, whence the gap condition implies that B ought to contain, for almost
all n, elements of order p and of height 2n. As tA D .tA \ B/ ˚ C, it is manifest that C must be
finite.
600 15 Mixed Groups

Define G to be generated by the symbols g; an .n < !/ with the defining relations

pg D 0; p2nC1 .an  panC1 / D g .n < !/

to satisfy

p! G D hgi Š Z.p/ and G=p! G Š A:

Evidently G is Warfield (since G=ha0 i is a countable p-group, so simply presented), and the claim
is that G is not simply presented. For, suppose by way of contradiction that it is. Then G D H ˚ K,
where H is simply presented of torsion-free rank 1 and K is torsion. From the properties of A, it
follows that p! G  K would imply that K=p! G is finite, which is impossible because p! K D p! G
in this case. Therefore, by full invariance, p! G  H. If x; y are generators of H, and o.x/ D 1,
then the p-height of y C hxi in H=hxi is equal to hp .y/. It follows that H=hxi admits a simple
presentation such that all the generators are of finite heights; thus, p! .H=hxi/ D 0, i.e. p! H  hxi.
But this is impossible, as x is of infinite order, while p! G has order p. Consequently, H cannot be
simply presented.

It is fairly easy to describe Warfield groups of torsion-free rank 1 in various ways.

Theorem 7.2 (Warfield [8]). For a group A of torsion-free rank 1 the following
conditions are equivalent:
(i) A is a Warfield group;
(ii) A is an extension of Z by a simply presented torsion group;
(iii) A has an H.@0 /-family of nice subgroups.
Proof. (i) , (ii) This is proved in Theorem 6.4.
(ii) ) (iii) Let A be an extension of an infinite cyclic group hai by a simply
presented torsion group B. From Lemma 4.2 we derive that hai is nice in A. We also
know from Theorem 5.9 in Chapter 11 that B admits an H.@0 /-family N of nice
subgroups. By Lemma 4.2, hai ˝ Z.p/ is p-nice for each p, hence hai is nice in A. It
remains to refer to Lemma 4.1 to argue that the extensions of hai by the members
of N will be an H.@0 /-family of nice subgroups in A.
(iii) ) (i) Suppose A has an H.@0 /-family N of nice subgroups. Factoring out
a (nice) infinite cyclic subgroup hai from the members of N that contain hai, we
get an H.@0 /-family of nice subgroups in A=hai, so this factor group is totally
projective. We can find a simply presented group C of torsion-free rank 1 with
H.C/ D H.A/. Furthermore, there exists a torsion group T with large UK-invariants
such that the groups T; A ˚ T; C ˚ T all have the same UK-invariants. Applying
Theorem 5.2 to A ˚ T and C ˚ T, the theorem is proved. t
u
Decomposition Subgroups For Warfield groups of higher torsion-free ranks, a
more sophisticated approach is required, as the situation is by far more complicated.
The role of a single infinite cyclic group is replaced by a rank-wise maximal free
subgroup. These free subgroups are not arbitrary, they are very special subgroups,
absolutely fundamental in the study of Warfield groups.
The definition will be stated for arbitrary mixed groups A. A free subgroup F
of A is said to be a decomposition subgroup and a basis X D fxi gi2I of F a
decomposition basis if
7 Warfield Groups 601

(i) A=F is torsion; i.e. X is a maximal independent set of elements of infinite order;
(ii) for any prime p and for any finite subset J of I, we have
X
hp . ni xi / D min hp .ni xi / .ni 2 Z/:
i2J
i2J

If, in addition, F is a nice subgroup, it will be called a nice decomposition

subgroup. We observe:
(A) All groups of torsion-free rank 1 have decomposition subgroups, and so do their
direct sums.
(B) If X D fxi gi2I is a decomposition basis for A, and if mi ¤ 0 are integers, then
Y D fmi xi gi2I is likewise a decomposition basis, called a subordinate to X.
(C) A subordinate to a nice decomposition basis is likewise nice. The conclusion
follows from the fact that if Y is a subordinate to a decomposition basis X, then
hYi is nice in hXi.
Example 7.3. A simply presented group admits a nice decomposition subgroup: decompose the
group as in Lemma 6.2, and form the direct sum of infinite cyclic groups, one from each non-
torsion summand.

Two decomposition bases are considered equivalent if there is a bijection

between the sets such that corresponding elements have equivalent height-matrices.
That in general decomposition bases are inequivalent is shown by Example 6.5.
Though decomposition bases are always nice if they are finite (Exercise 3), they
need not be nice in general.
The next result has a long sophisticated proof, we record it without giving details.
Lemma 7.4 (Hunter–Richman [1]). Every decomposition basis of a Warfield
group A has a nice subordinate Y such that A=hYi is totally projective torsion. u
t
The next result is also fundamental, it is proved by using a general decomposition
theorem on certain direct sums in additive categories. We quote it without proof.
Theorem 7.5 (Arnold–Hunter–Richman [1]). Summands of a group with decom-
position basis have decomposition bases. t
u
We do not want to go through the detailed proofs of Lemma 7.4 and Theorem 7.5,
we would rather concentrate on the principles that make the theory work.
Next, we characterize Warfield groups in terms of decomposition bases.
Theorem 7.6. A group A is Warfield if and only if it contains a nice decomposition
basis X such that A=hXi is totally projective torsion.
Proof. First suppose A has a nice decomposition basis X as stated. There exists
a simply presented group C whose decomposition basis Y is equivalent to X. By
making use of a totally projective torsion group T with large UK-invariants, we
argue as in the last part of the proof of Theorem 7.2 to conclude that A is a summand
of the simply presented C ˚ T.
602 15 Mixed Groups

Conversely, let A ˚ B D C where C D ˚i2I Ci is simply presented with

rk0 Ci D 1, and assume Z D fci j ci 2 Ci g is a nice decomposition basis for C. By
Theorem 7.5, both A and B have decomposition bases, say, X and Y, respectively.
Then X [ Y is a decomposition basis for C, and Lemma 7.4 implies that it has a
subordinate X 0 [Y 0 .X 0  X; Y 0  Y/ that generates a nice subgroup of C with totally
projective C=hX 0 [ Y 0 i. The last group is clearly isomorphic to A=hX 0 i ˚ B=hY 0 i, so
X 0 is a nice decomposition basis in A with totally projective A=hX 0 i. t
u
Warfield Invariants In order to classify local Warfield groups of arbitrary
cardinalities, an additional cardinal invariant is required. We follow Stanton [1] in
introducing the Warfield invariants in general. The advantage of this method is that
the definition is not restricted to Warfield groups.
Let A be p-local, for a fixed p, and let u D .0 ; 1 ; : : : ; n ; : : : / denote a strictly
increasing sequence of ordinals (now no 1 is permitted). As before, we set A.u/ D
fa 2 A j hp .pn a/  n g, and define

A .u/ D ha 2 A.u/ j hp .pn a/ > n for infinitely many ni:

The elements in the definition of A.u/ form a subgroup, but this is not the case
for A .u/, in general. Both are fully invariant subgroups of A, and obviously,
tA
A .u/ for all u. It is equally clear that pA.u/
A .u/, whence we see that
A.u/=A .u/ is a Z=pZ-vector space.
For n < !, the map

n W A.u/=A .u/ ! A.pn u/=A .pn u/

acting as a C A .u/ 7! pn a C A .pn u/ .a 2 A.u// is a well-defined linear map

between vector spaces. We claim n is monic for each n; for the proof, the case
n D 1 will suffice. Let a CP A .u/ 2 Ker 1 where a 2 A.u/. To verify that then
 
a 2 A .u/, consider pa D m jD1 rj xj .rj 2 Z/ where xj are generators of A .pu/,
thus hp .pk xj /  kC1 for all k < ! with strict inequalities for infinitely many k’s.
We can find yj 2 A of height  0 such that Ppyj D xj . Now pu.yj / D u.xj / implies
yj 2 A .u/. Taking the containment a  m jD1 rj yj 2 tA
A .u/ into account,
P Pm 
a D .a  m jD1 rj y j / C jD1 rj y j 2 A .u/ follows.
Recall that u v means pn u D pm v for some n; m < !. Hence we conclude that
the direct limit

lim A.pn u/=A .pn u/ D wA .u/

!
n<!

(with the n as connecting maps) depends only on the equivalence class Œu

of
indicators, and not on the individual indicators. This vector space (or its dimension)
is called the uth Warfield invariant of A. For the indicator w D .1; : : : ; 1; : : : /,
A.w/ is the divisible part D of A, and A .w/ D 0, so that the wth Warfield invariant
is the (dimension of) D.
7 Warfield Groups 603

Lemma 7.7 (Stanton [1]). Let A be a reduced p-local group.

(i) If the torsion-free rank of A is 1, then wA (u)D rk0 A(u), i.e. it is 0 or 1.
(ii) If A D ˚i2I Ai , then wA (u)D ˚i2I wAi (u).
Proof. (i) Suppose A has torsion-free rank 1, and u.a/ D .0 ; : : : ; n ; : : : / for an
a 2 A of infinite order. If b 2 A but … A .u/, then because of u.a/ u.b/
there is m 2 N such that pm a D kpm b for some integer k prime to p. Now
a  kb 2 tA
A .u/ implies a C A .u/ D kb C A .u/. This proves that
A.u/=A .u/ Š Z.p/. For A.pn u/=A .pn u/ we have the same proof and the same
conclusion, so wA .u/
1. On the other hand, if the elements of infinite order in
A have characteristics v 6 u, then A.u/ D 0 and wA .u/ D 0.
(ii) This is a simple consequence of the definition and (i). t
u
Thus for Warfield groups, we have two sets of invariants: the UK- and the
Warfield invariants, both are cardinal numbers.
Isomorphism of Local Warfield Groups All the groundwork has been laid for
the classification of Warfield groups in the local case.
Theorem 7.8. The p-local Warfield groups A; C are isomorphic if and only if
(i) they have the same UK-invariants;
(ii) their Warfield invariants are equal.
Proof. Only sufficiency requires a proof. So assume A; C satisfy (i)–(ii). Owing to
Theorem 7.6, decomposition bases X in A and Y in C can be selected such that hXi is
nice in A, hYi is nice in C, and A=hXi; C=hYi are totally projective. From Lemma 7.7
it is clear that condition (ii) implies the equality of the cardinalities of the sets of
elements in X and Y whose indicators belong to a given equivalence class Œu
. Hence
we can replace X; Y by subordinates, if necessary, (we keep the notations X; Y for
the subordinates) to obtain a bijection f W X ! Y such that u.x/ D u.f .x// for all
x 2 X. By (C), these subordinates generate nice subgroups, and as an extension
of a nice totally projective subgroup by a totally projective group is again totally
projective, it is evident that the total projectivity of the quotients A=hXi; C=hYi is
preserved. Refer to an obvious extension of Lemma 5.1 to higher ranks, and argue
that the equality of the relative invariants f .A; hXi/ and f .C; hYi/ is a consequence
of (i). Finally, we invoke Theorem 5.4 to complete the proof. t
u
The situation is far less transparent in the global case. The classification of global
Warfield groups is not possible in terms of equivalence of matrices only, as is evident
from Example 6.5. Stanton [2] introduced a coarser equivalence relation for height-
matrices which he called ‘compatibility,’ but I prefer to use ‘quasi-equivalence’
instead to stress its strong parallelism to quasi-isomorphism of torsion-free groups.
Accordingly, we will call the height-matrices M and N quasi-equivalent if there
exist positive integers m; n such that

mM  N and nN  M:
604 15 Mixed Groups

By making use of the equivalence classes under this relation (in addition to the
equivalence classes of matrices), Stanton introduces new invariants as a vehicle for
a classification of global Warfield groups.
Example 7.9. Consider the groups A; B; C; D defined in Example 6.5. All their height-matrices are
quasi-equivalent. For instance, 6 H.C/
H.A/ and 6 H.A/
H.C/.
In this section, the reader has been burdened with unproved propositions that have been utilized
in subsequent proofs. This is not a usual practice in this volume, and I apologize for having done so.
Mixed groups have enormously rich structure, some theorems require not only delicate reasoning,
but also lots of technical calculations, consuming lots of pages. In spite of the missing details (for
which we refer to the literature), I hope that the material presented above lays down the groundwork
and conveys properly a feeling for the theory of mixed groups. A streamlined version of the existing
results would be most welcome, it certainly calls for further investigations in order to simplify and
to extend the results.
F Notes. Each structure theorem should be accompanied with an existence theorem. Such
a theorem on Warfield groups was provided by Hunter–Richman–Walker [1]; its proof requires
delicate arguments on cardinals. Conditions on the UK- and Warfield invariants are stated to
ensure the existence of a Warfield group for a prescribed nice decomposition basis. It is also shown
that a direct sum of infinitely many copies of any local Warfield group is simply presented, and
every local Warfield group is the direct sum of a simply presented group and a Warfield group of
countable torsion-free rank.
Arnold–Hunter–Richman [1] proved that a finite torsion-free rank summand of a group with
decomposition basis also has such a basis. For arbitrary summands, the proof is due to Stanton
[3] and Arnold–Hunter–Richman [1]. Hunter–Richman [1] construct a valuated free group G with
decomposition bases X; Y such that no subgroup of G is generated both by a subordinate to X and
by a subordinate to Y. An important characterization of Warfield groups is due to Hill–Megibben
[9]. If the torsion-free rank is > 1, then the existence of an H.@0 /-family of nice subgroups is not
enough to get a Warfield group, unless the existence of a decomposition basis is taken for granted.
Alternately, the notion of niceness can be strengthened to so-called knice subgroups; then Warfield
groups are characterized as groups admitting H.@0 /-families of knice subgroups. In terms of a nice
composition series, Moore [1] and Loth [2] establish different kinds of characterization of Warfield
groups. In the paper Hill–Megibben [8], there are several interesting results on the existence of
automorphisms in Warfield groups, carrying elements to elements with the same height-matrix. In
his dissertation, Stanton proves that if A is a Warfield group, then for some torsion T, A ˚ T is a
direct sum of groups of torsion-free rank 1.
In a different vein, Warfield [8] considered cancellation for countable mixed groups of finite
torsion-free ranks. Results analogous to those in Sect. 10 in Chapter 12 are established. Hill–Ullery
[1] gave necessary and sufficient conditions on an isotype subgroup of a local Warfield group to be
likewise Warfield; the results were generalized by Megibben–Ullery [1] to the global case. Loth
[1] characterized the Pontryagin duals of Warfield groups.
A final comment on Warfield groups is in order. The best approach to their theory is
undoubtedly via valuated groups as demonstrated by Richman–Walker [5]: a free subgroup
with valuation is standing for the entire group. As a consequence, the study boils down to the
investigation of valuated free groups. This approach is easy to understand, but not so easy to work
with, because there is no canonical way to it, and besides, a less tractable non-abelian category
is involved. We did not discuss Warfield groups from this point of view to avoid entering into an
unexplored area.
8 The Categories WALK and WARF 605

The interested reader’s attention is called to a series of papers by R. Göbel, C. Jacoby, K.

Leistner, P. Loth, and L. Strüngmann, discussing model theoretic versions of some topics discussed
here. E.g. in paper [1] by the last four authors, a classification theorem is proved using cardinal
invariants that were deduced from the UK- and Warfield-invariants.

Exercises

(1) Every countably generated group of torsion-free rank 1 is a summand of a

simply presented group.
(2) (Hunter–Richman–Walker) Let C be a subgroup of A such that A=C is torsion.
Then A and C have the same Warfield invariants.
(3) (Hunter–Richman) Every finite subset Y in a decomposition basis generates a
nice subgroup. [Hint: localize, induct on the size of Y using Lemma 4.2 and
4.1.]
(4) Given u, construct a Warfield group with w(u)D 1.
(5) (Jarisch–Mutzbauer–Toubassi) Let A be a p-local group such that tA is
†-cyclic. Then A is Warfield if and only if A=p! A is simply presented and
p! A is completely decomposable.

8 The Categories WALK and WARF

We proceed to consider two categories that can help a better understanding of the
behavior of mixed groups. Their definition is based on the excellent idea that we
have already seen to work effectively: ignoring torsion subgroups, but retaining the
information on divisibility in the surviving free subgroup.
Category WALK The category WALK was introduced by E. Walker. Its objects
are groups, its morphisms are ordinary group homomorphisms, with the proviso that
two maps: ˛; ˇ W A ! C are viewed equal whenever Im.˛  ˇ/ is a torsion group.
In other words, a homomorphism ˛ W A ! C with Im ˛
tC is 0 in this category:

HomWALK .A; C/ D Hom.A; C/= Hom.A; tC/:

Apply the Hom functor to the exact sequence 0 ! tC ! C ! C=tC ! 0

to get 0 ! Hom.A; tC/ ! Hom.A; C/ ! Hom.A; C=tC/. This shows
that HomWALK .A; C/ is isomorphic to a subgroup of Hom.A; C=tC/ D
Hom.A=tA; C=tC/, i.e. it is a torsion-free group.
It is easily checked that WALK is an additive category with kernels and infinite
direct sums. WALK-isomorphism can be described in traditional terms.
606 15 Mixed Groups

Lemma 8.1. The groups A; C are isomorphic in WALK, A ŠWALK C, if and only if
there exist torsion groups X; Y such that

A ˚ X Š C ˚ Y:

Proof. Observe that in WALK, the injection A ! A ˚ X and the projection

A ˚ X ! A are inverse morphisms provided that X is a torsion group. Hence the
stated group isomorphism implies that we have A ŠWALK A ˚ X Š C ˚ Y ŠWALK C:
Conversely, assume A ŠWALK C. This means that there are homomorphisms:
˛ W A ! C and W C ! A such that ˛ D 1C C  and ˛ D 1A C  where
 2 Hom.C; tC/;  2 Hom.A; tA/. Define G by the pull-back diagram

0 −−−−→ tC −−−−→ G −−−−→ A −−−−→ 0

 ⏐ ⏐
 ⏐ α ⏐can
  
can
0 −−−−→ tC −−−−→ C −−−−→ C/tC ∼
= A/tA −−−−→ 0

The map ˛ renders the lower triangle commutative, so the top row splits, thus G Š
A ˚ tC. By symmetry, G Š C ˚ tA, completing the proof. t
u
Actually, there is no loss of generality in assuming X D Y in Lemma 8.1: if they
are not isomorphic, then both can be replaced by .X ˚ Y/.@0 / .
Category WARF In the category WARF introduced by Warfield, the objects
are groups, and a morphism ˛ W A ! C between two objects is an honest
homomorphism from a subgroup B of A into C such that
(i) A=B is a torsion group, and
(ii) HA .b/
HC .˛b/ for all b 2 B of infinite order.
The morphisms ˛; ˇ W A ! C are considered equal if they have the same domain,
and Im.˛  ˇ/ is a torsion group.
To be more specific, we give a precise description of HomWARF . For a subgroup
B of A, let Hom.AjB; C/ denote the subgroup of Hom.B; C/ consisting of all
homomorphisms ˛ W B ! C (in Ab) that do not increase heights, computed in
A and C, respectively. It is clear that an injective map ˇ W B0 ! B induces a group
homomorphism ˇ 0 W Hom.AjB; C/ ! Hom.AjB0 ; C/ via ˇ 0 .˛/ D ˛ˇ. It follows that

HomWARF .A; C/ Š lim Hom.AjB; C/

!
where the direct limit is taken over all B
A for which A=B is torsion.
It is easy to tell what the WARF-isomorphy of two arbitrary mixed groups A
and C means: this is just another way of saying that there exist subgroups B
A
and D
C such that A=B; C=D are torsion and there is a height-preserving group
isomorphism (i.e., in Ab) between B and D.
Note that WARF is an additive category with kernels and infinite sums. It fails to
admit cokernels, so it is not an abelian category.
8 The Categories WALK and WARF 607

The definition of this category is motivated by the following lemmas.

Lemma 8.2. Two mixed groups, A and C, of torsion-free rank 1 are isomorphic in
WARF if and only if H.A/ H.C/.
Proof. For sufficiency, note that the equivalence of height-matrices means that there
are a 2 A and c 2 C such that H.a/ D H.c/. The homomorphisms ˛ W hai ! C and
W hci ! A (where ˛.a/ D c; .c/ D a) are clearly inverse morphisms in WARF.
Conversely, if there exist morphisms ˛ W A ! C; W C ! A that are inverse to each
other in WARF, then there exist a 2 A; c 2 C with H.a/
H.c/ and a0 2 A with
H.a0 /  H.c/. Hence H.a/ H.a0 /, and the claim follows. t
u
Lemma 8.3. A group has a decomposition basis if and only if it is WARF-
isomorphic to a simply presented group.
Proof. Suppose X D fxi gi2I is a decomposition basis of A. There exist simply
presented groups Ci of torsion-free rank 1 such that some ci 2 Ci satisfy
H.ci / D H.xi /. The group C D ˚i2I Ci is then WARF-isomorphic to A via the
correspondence ci 7! xi .
Conversely, let A be WARF-isomorphic to C D ˚i2I Ci where each Ci is simply
presented of torsion-free rank 1. Let Y D fci gi2I with ci 2 Ci be a decomposition
basis of C. By definition, there are subgroups B
A; D
C, and a height-
preserving isomorphism  W B ! D, where A=B; C=D are torsion. Replace Y by a
subordinate to get Y  D. Then xi D  1 .ci / is a decomposition basis of A. t
u
Some direct decompositions are unique in WARF.
Theorem 8.4. Let A be a p-local simply presented group, and

A D ˚i2I Ai D ˚i2I Ci

direct decompositions where Ai ; Ci are of torsion-free rank 1. There is a permutation

 of the index set I such that

Ai ŠWARF C.i/ for all i 2 I:

Proof. In the local case, the WARF-isomorphism of groups of torsion-free rank 1

means that the associated indicators are equivalent; this follows from Lemma 8.2.
For the equivalence class of u, the number of components Ai (also the number
of Ci ) with indicators u is given by the Warfield-invariant wA .u/ as stated in
Lemma 7.7. (The maximum indicator u D .1; 1; : : : / requires a different, but
trivial argument.) t
u
F Notes. The idea of regarding groups equivalent by making them isomorphic via adding
suitable summands goes back to an old paper by R. Baer. For torsion summands, this was first
applied by Rotman–Yen [1] before Walker formalized the idea by defining an honest category.
The WALK- and WARF-isomorphisms are in general different; however, for countably
generated Jp -modules of finite torsion-free ranks, they coincide. A category of mixed groups
similar to WALK (defined in an earlier paper) was investigated by Yakovlev [3] with the purpose of
transferring results on decompositions of torsion-free groups to mixed groups. Franzen–Goldsmith
608 15 Mixed Groups

[1] construct a full embedding of the category of reduced torsion-free groups in WALK to derive
new results on mixed groups.
Evidently, when dealing with Warfield groups, there is an apparent advantage in interpreting
the results in the category WARF. This simplifies and clarifies the situation, which is good enough
motivation for introducing this category. It is safe to say that an additional benefit is that it facilitates
to unveil further features of the groups. However, one should not forget that it cannot tell the whole
story about the groups themselves. That the category WARF is so useful in the local case is due to
the fact that in this category p-local groups of torsion-free rank 1 have local endomorphism rings.
In contrast, in the global situation, the endomorphism rings are principal ideal domains, that makes
the objects less tractable.

Exercises

(1) Give a detailed proof that WALK is an additive category with infinite direct
sums, and that morphisms have kernels, but not necessarily cokernels.
(2) Prove that WARF is an additive category with infinite direct sums.
(3) (Warfield) Show that, in the category WARF, the small objects are the groups
of finite torsion-free rank.
(4) (Warfield) Two countable groups of finite torsion-free rank are isomorphic if
and only if they are WALK-isomorphic and have the same UK-invariants.

9 Projective Properties of Warfield Groups

We now embark on a brief study of the projective properties of Warfield groups:

they can be described as projectives relative to a class of short exact sequences.
This is entirely analogous to the balanced-projectivity of totally projective p-groups,
discussed in Sect. 5 in Chapter 11.
˛ ˇ
H-Projectivity An exact sequence 0 ! A!B!C ! 0 was called H-exact
if the induced sequence

˛ ˇ
0 ! A.H/!B.H/!C.H/ ! 0

was exact for all height-matrices H. Accordingly, a group G will be called

H-projective if it has the projective property with respect to all H-exact sequences;
i.e. the map Hom.G; B/ ! Hom.G; C/ induced by ˇ is an epimorphism.
(a) We observe right away that for a p-group G, H-projectivity is identical with
balanced-projectivity. Indeed, all the maps from G land in the p-components,
the sequence of the p-components is balanced-exact, and the rest follows from
Theorem 5.9 in Chapter 11.
9 Projective Properties of Warfield Groups 609

(b) It is obvious that H-projective groups share the general properties typical for
any class of projective objects, like preservation under arbitrary direct sums
and summands.
A result that is very special for this class of groups is the following.
Theorem 9.1 (Warfield [7]).
(i) All simply presented groups are H-projective.
(ii) An exact sequence relative to which every simply presented group .of torsion-
free rank 1/ has the projective property is H-exact.
Proof. (i) Owing to Lemma 6.2 it suffices to show that a simply presented group
˛ ˇ
G of torsion-free rank
1 is H-projective. Let 0 ! A!B!C ! 0 be an
H-exact sequence, and  W G ! C. The torsion case has been settled by
(a) above, so assume g 2 G is a generator of infinite order, hgi is nice in
G, and G=hgi is simply presented torsion. By H-exactness, there is a b 2 B
such that ˇb D g and H.b/ D H.g/. The correspondence g 7! b extends
to a homomorphism W G ! B. Now   ˇ induces a homomorphism
 0 W G=hgi ! C, so by the torsion case there is some ı 0 W G=hgi ! B such
that ˇı 0 D  0 . Hence if ı W G ! B denotes the canonical map followed by ı 0 ,
then ı C W G ! B satisfies ˇ. C ı/ D , as desired.
(ii) Suppose that all simply presented groups of torsion-free rank 1 are projective
˛ ˇ
relative to the exact sequence 0 ! A!B!C ! 0. If c 2 C is of infinite
order, then let G be simply presented of torsion-free rank 1 whose generator g
of infinite order satisfies H.g/ D H.c/; such a G exists by virtue of Sect. 6(c).
The correspondence g 7! c extends to a homomorphism W G ! C, and
hypothesis implies the existence of a map  W G ! B satisfying ˇ D .
It is clear that H.g/ D H.c/, which shows that ˇ maps B.H/ onto C.H/.
If c 2 CŒp
, then choose a simply presented G of torsion-free rank 1 whose
generator g satisfies hp .g/ D hp .c/. The same argument leads to the conclusion
that ˇ.p BŒp
/ D p CŒp
. An appeal to Lemma 4.6 completes the proof. t
u
H-Projectivity and Warfield Groups To show that there are enough
H-projective groups, we prove:
Theorem 9.2 (Warfield [7]). Every group A can be embedded in an H-exact
˛
sequence 0 ! K ! G!A ! 0 where G is H-projective .simply presented/.
Proof. Let f W A ! X be a bijection between A and a set X. Define the group
G to be generated by X, subject exactly to those relations that hold between the
corresponding elements in A and involve at most two elements. Then G is simply
presented, and f 1 induces an epimorphism ˛ W G ! A. We let K D Ker ˛. To
show that the sequence at hand is H-exact, observe that hp .f .a// D hp .a/ for all
a 2 A and for all p, as it follows by a straightforward induction using the fact that
a relation pb D a between a; b 2 A implies pf .b/ D f .a/ in G. Hence we conclude
that H.f .a// D H.a/ for all a 2 A, and thus ˛.G.H// D A.H/. Furthermore, the
generator f .a/ in G has the same height as a in A, and if o.a/ D p, then the order of
610 15 Mixed Groups

f .a/ in G is also p. Consequently, ˛.p GŒp

/ D p AŒp
holds for all p and , so the
sequence is H-exact (Lemma 4.6). t
u
We are now prepared for an important characterization of H-projective groups.
Theorem 9.3 (Warfield [7]). A group is H-projective if and only if it is a Warfield
group.
Proof. In view of Theorem 9.1 and the definition of Warfield groups, the ‘if’ part is
evident. It remains to show that an H-projective group G is Warfield. The proof is
standard. By Theorem 9.2, G embeds in an H-exact sequence 0 ! K ! B ! G !
0 where B is simply presented. H-projectivity forces splitting, thus G is isomorphic
to a summand of B, i.e. G is Warfield. t
u
Balanced-Projectivity Example 6.5 makes it evident that no good classification
is expected for global Warfield groups. However, there is a more tractable subclass
that is more conveniently defined via its projective property, and that includes both
the torsion and the torsion-free simply presented groups. And more importantly,
luckily, this subclass does admit a satisfactory classification.
˛ ˇ
The exact sequence 0 ! A!B!C ! 0 was defined to be balanced-exact if
the induced sequence

˛ ˇ
0 ! A./!B./!C./ ! 0

was exact for all characteristics . A group G that has the projective property for all
balanced-exact sequences will be called balanced-projective.
Lemma 9.4. Balanced-projective groups are H-projective.
Proof. This follows from the fact that H-exact sequences are balanced-exact.
(Theorem 9.10 will show that the reverse implication fails.) t
u
Example 9.5. For torsion as well as for torsion-free groups, balanced-projectivity as defined above
coincides with balanced-projectivity as defined in previous chapters. This is evident for torsion-free
groups, and follows from Theorem 5.9 in Chapter 11 for torsion groups.
Example 9.6. Divisible groups are balanced-projective.

Given a characteristic  D .2 ; 3 ; : : : ; p ; : : : / (where for each prime p the entry

p is an ordinal or the symbol 1), let Z denote the localization of Z at the set  of
primes p for which p ¤ 1 (this notation for  will be used in the balance of this
section). We define a group M as a group of torsion-free rank 1 with an element x
such that M ./ D Z x, M =Z x is a simply presented reduced torsion group, and
H.x/ is the “gapless” height-matrix
0 1
2 2 C 1 : : : 2 C k : : :
B: : : : : : : : : : : : : : : C
H D B C
@ p p C 1 : : : p C k : : :A D kp C kk:
::: ::: ::: ::: :::
9 Projective Properties of Warfield Groups 611

Lemma 9.7. The totally projective torsion groups as well as the groups M are
balanced-projective.
Proof. It suffices to verify the claim for M . The proof of Theorem 9.1 applies
almost verbatim. t
u
The groups M are called -elementary balanced-projective. Thus a Warfield
group of torsion-free rank 1 is elementary balanced-projective if and only if its
height-matrix is equivalent to a gapless matrix.
Example 9.8. (a) If  D .2 ; 3 ; : : : ; p ; : : : / where each p is either a finite ordinal or the
symbol 1, then M is a rational group.
(b) If all p D 0 with the exception of a single q (which is an arbitrary ordinal), then we get the
q-local Nunke group of length q .

Characterization of Balanced-Projective Groups The existence of balanced-

projective resolutions follows easily by standard arguments.
Lemma 9.9 (Warfield [7]). For every group A, there is a balanced-exact sequence

0 ! K ! G ! A ! 0 where G is a direct sum of a totally projective torsion
group and elementary balanced-projective mixed groups.
Proof. By Proposition 5.8 in Chapter 11, the torsion subgroup tA admits a balanced-
0
projective resolution 0 ! K 0 ! G0 ! tA ! 0 where G0 is totally projective
torsion. For each a 2 A of infinite order, select an elementary balanced projective
group Ga D M.a/ along with a homomorphism a W Ga ! A, mapping a generator
of the infinite cyclic Ga ..a// upon a. We continue by setting G D G0 ˚ ˚a2AntA Ga
and D 0 ˚ ˚a a , to obtain an exact sequence as stated above, where is
surjective and does not increase heights.
It remains to verify balancedness. We argue that Lemma 4.5(i) is fairly obvious
from the construction, while in Lemma 4.5(ii) only the torsion part is involved for
which G0 was selected so as to satisfy balancedness. t
u
Hence we can derive at once:
Theorem 9.10. A .mixed/ group is balanced-projective exactly if it is a summand
of a direct sum of a totally projective torsion group and -elementary balanced-
projective groups for various characteristics . t
u
F Notes. H-projectivity was studied by Warfield [4] under the name sequentially-pure-
projectivity. For any mixed group, H-projective dimension makes sense. They were studied by
Megibben–Ullery [2] who gave upper bounds for the H-projective dimensions of certain groups.
The theory of balanced-projective groups was developed by Warfield Q [7]; the global case
was investigated by Wick [2]. Warfield introduced the global heights h D p pp , and used the
functorial notation hA for our A./. Lane [1] characterized p-local balanced-projective groups in
terms of H.@0 /-families of so-called K-nice subgroups, and in his paper [2], he obtained definitive
results on the balanced-projective dimension of p-local groups. Lane–Megibben [1] extended these
results to the global case.
612 15 Mixed Groups

Exercises

(1) (Warfield) A group A is WARF-isomorphic to an elementary balanced-

projective group exactly if it is of torsion-free rank 1, and contains an element
a such that .na/ D n.a/ for all n 2 N.
(2) (Warfield) A is WARF-isomorphic to a direct sum of an elementary balanced-
projective groups if and only if it contains a free subgroup F D ˚hxi i such that
A=F is torsion, and for all finite subsets fx1 ; : : : ; xk g and for all p

hp .n1 x1 C    C nk xk / D minfhp .n1 x1 /; : : : ; hp .nk xk /g:

(3) For every characteristic  there exists a -elementary balanced-projective

group. [Hint: start with Z x and furnish x with the correct heights.]
(4) If A is a -elementary balanced-projective group, then EndWARF A Š Z .

Problems to Chapter 15

PROBLEM 15.1. Using the theory of totally projective p-groups and height-
matrices, develop a theory of mixed groups A such that t.A/ is totally projective,
and A=t.A/ is completely decomposable (or divisible).
PROBLEM 15.2. Are two mixed groups of torsion-free rank 1 necessarily iso-
morphic if their endomorphism rings (automorphism groups) are isomorphic and
the torsion-free parts are Š Q?
PROBLEM 15.3. Are there interesting groups that can be presented such that
every generator occurs in at most n relations for a fixed n 2 N?
Examples for n D 2 include divisible groups and completely decomposable torsion-free
groups.
Chapter 16
Endomorphism Rings

Abstract With an abelian group A one associates the ring End A of its endomorphisms. This is an
associative ring with 1 which frequently reflects several relevant features of the group. Information
about direct decompositions is certainly stored in this ring. It is quite challenging to unveil hidden
relations between a group and its endomorphism ring.
Quite a lot of information is available for the endomorphism rings of p-groups. A celebrated
theorem by Baer and Kaplansky shows that torsion groups with isomorphic endomorphism
rings ought to be isomorphic. Moreover, the endomorphism rings of separable p-groups can be
characterized ring-theoretically. However, in the torsion-free case, we can offer nothing anywhere
near as informative or complete as for torsion groups. As a matter of fact, on one hand, there
exists a large variety of non-isomorphic torsion-free groups (even of finite rank) with isomorphic
endomorphism rings, and on the other hand, endomorphism rings of torsion-free groups seem
to be quite general: every countable rank reduced torsion-free ring with identity appears as an
endomorphism ring of a torsion-free group. The situation is not much better even if we involve the
finite topology. The mixed case is of course more difficult to survey.
Though the problem concerning the relations between group and ring properties has attracted
much attention, our current knowledge is still far from being satisfactory. The main obstacle to
developing a feasible in-depth theory is probably the lack of correspondence between relevant
group and relevant ring properties. However, there is a great variety of examples of groups with
interesting endomorphism rings, and we will list a few which we think are more interesting. In
some cases we have to be satisfied with just stating the results in order to avoid tiresome ring-
theoretical arguments. In some proofs, however, we had no choice but to refer to results on rings
which can be found in most textbooks on graduate algebra. There are very good surveys on
endomorphism rings by Russian algebraists, e.g. Krylov–Tuganbaev [1], and especially, the book
by Krylov–Mikhalev–Tuganbaev [KMT].
In this chapter, we require some, but no more than a reasonable acquaintance with standard
facts on associative rings.

1 Endomorphism Rings

Rings of Endomorphisms It is a familiar fact that the endomorphisms ˛; ˇ; : : :

of an abelian group A form a ring under the addition and multiplication of
homomorphisms:

.˛ C ˇ/a D ˛a C ˇa and .˛ˇ/a D ˛.ˇa/ .8a 2 A/:

© Springer International Publishing Switzerland 2015 613

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_16
614 16 Endomorphism Rings

The endomorphism ring End A of A is an associative ring with identity. By abuse

of notation, we will use the same symbol for the endomorphism ring and for its
additive group: the endomorphism group.
Example 1.1.
(a) If A D hei Š Z, then every ˛ 2 End A is completely determined by ˛e. Since e is a free
generator, every correspondence e 7! ne for any n 2 Z extends to an endomorphism. The
operations with endomorphisms are like with integers, so End A Š Z (ring isomorphism).
(b) A similar argument shows that if A is a cyclic group of order m, then we have End A Š Z=mZ
(as rings).
Example 1.2.
(a) From Example 1.4 in Chapter 7 we obtain the isomorphism End.Z.p1 // Š Jp .
Q
(b) End.Q=Z/ Š p Jp D Z: Q This follows from (a).

Example 1.3. Example 1.6 in Chapter 7 shows that End Jp Š Jp for every prime p.
Example 1.4. Let R denote a rational group, 1 2 R. Here again, an endomorphism ˛ is fully
determined by ˛1 D r 2 R, so ˛ is simply a multiplication by the rational number r 2 R. As
endomorphisms respect divisibility, rP can be an endomorphism of R only if every prime factor p of
the denominator of r (in its reduced form) satisfies pR D R. That the converse is also true is seen
immediately. Thus End R is a subring of Q whose type is the largest idempotent type  t.R/, i.e.
t.R/ W t.R/.

We continue with a few elementary observations.

(A) A group isomorphism  W A ! C induces a ring isomorphism   W End A !
End C via   W ˛ 7! ˛ 1 .
(B) Suppose A D B ˚ C with  W A ! B the projection map. Then the identification
End B D .End A/ can be made. Indeed, for ˛ 2 End A, ˛ 2 End B, while
if ˇ 2 End B, then ˇ D ˇ may be regarded as an element in End A.
(C) If  is a central idempotent in End A, then A is a fully invariant summand of A.
(D) If A D A1 ˚    ˚ An is a direct decomposition with fully invariant summands,
then

End A Š End A1 ˚    ˚ End An :

If the summands are not fully invariant, then we get only a matrix representa-
tion, see Proposition 1.14.
An idempotent  ¤ 0 is said to be primitive if it cannot be written as a sum
of two non-zero orthogonal idempotents. The following claim is rather obvious in
view of (D).
(E) For an idempotent  ¤ 0 of End A, the summand A is indecomposable if and
only if  is a primitive idempotent.
We have already made frequent use of the fact that direct decompositions corre-
spond to idempotent endomorphisms. This interplay between direct decompositions
and endomorphisms is constantly used. At this point, we insert the following lemma
that will be indispensable throughout.
1 Endomorphism Rings 615

Lemma 1.5. There is a bijection between the finite direct decompositions A D A1 ˚

   ˚ An of a group A, and decompositions of End A into finite direct sums of left
ideals,

End A D L1 ˚    ˚ Ln : (16.1)

If Ai D i A with pairwise orthogonal idempotents i , then Li D .End A/i for

i D 1; : : : ; n.
Proof. Suppose A D 1 A ˚    ˚ n A with mutually orthogonal idempotents i .
The well-known Peirce decomposition of End A yields End A D .End A/1 ˚    ˚
.End A/n : Conversely, if (16.1) holds with left ideals Li of End A, then—as is
readily checked—we have Li D .End A/i where i is the ith coordinate of the
identity of End A. These i are orthogonal idempotents with sum 1, hence A D
1 A ˚    ˚ n A follows at once. It is pretty clear that the indicated correspondence
is a bijection. t
u
As far as isomorphic summands are concerned, the basic information is recorded
in the next lemma.
Lemma 1.6. Suppose that B1 ; B2 are summands of A, corresponding to idempotents
1 ; 2 2 E D End A. The following are equivalent:
(i) B1 Š B2 ;
(ii) there exist ˛; ˇ 2 E such that ˛ D 1 ˛2 ; ˇ D 2 ˇ1 , and ˛ˇ D 1 , ˇ˛ D 2 ;
(iii) E1 Š E2 as left E-modules;
(iv) 1 E Š 2 E as right E-modules.
Proof.
(i) ) (iii) Let W B1 ! B2 and ı W B2 ! B1 be inverse isomorphisms. Thus
ı D 2 and ı2 D ı imply that Eı D E2 , because each of ı and 2
is contained in the left E-ideal generated by the other. Multiplication
by ı and yield E1 ! E1 ı ! Eı D E2 ! Eı D E1 .
The composite is multiplication by 1 which is the identity on E1 , so
multiplication by is an isomorphism.
(iii) , (ii) Choose an isomorphism  W E1 ! E2 , and let .1 / D ˛,
 1 .2 / D ˇ with ˛; ˇ 2 E. Then ˛ D 1 ˛2 ; ˇ D 2 ˇ1 , and also
 1  W 1 7! ˛ D ˛2 7! ˛ˇ, thus ˛ˇ D 1 ; similarly, ˇ˛ D 2 .
Conversely, if ˛; ˇ 2 E are as stated, then E˛ D E2 . Thus 1 7! ˛
induces an epimorphism  W E1 ! E2 .  is an isomorphism, because
if .1 / D ˛ D 0 for some  2 E, then also 1 D ˛ˇ D 0.
(ii) , (iv) Condition (ii) is left-right symmetric, so by the last proof (ii) is also
equivalent to 1 E Š 2 E.
(iv) ) (i) Let  W 1 E ! 2 E be an isomorphism. Define a map B1 ! B2 as
1 .a/ ! .1 /.a/ .a 2 A/. It has the obvious inverse, so it has to be an
isomorphism. t
u
616 16 Endomorphism Rings

In the next result we are referring to the completion AQ and the cotorsion hull A
of a group A.
Proposition 1.7.
(i) Let A be a group with A1 D 0 .i.e., Hausdorff in the Z-adic topology/. For every
 2 End A, there is a unique Q 2 End AQ such that Q  A D .
(ii) For a reduced torsion group T, there is a natural isomorphism End T Š End T 
between the endomorphism rings.
Proof.
(i) The hypothesis A1 D 0 allows us to regard A as a pure subgroup of A. Q By the
Q  can be extended to an Q W AQ ! AQ which must be unique
pure-injectivity of A,
Q
in view of the density of A in A.
(ii) The exact sequence 0 ! T ! T  ! D ! 0 with torsion-free divisible
D induces the exact sequence 0 D Hom.D; T  / ! Hom.T  ; T  / !
Hom.T; T  / ! Ext.D; T  / D 0. Hence and from the obvious Hom.T; T  / Š
Hom.T; T/ the claim is evident. t
u
Inessential Endomorphisms When studying the endomorphisms of reduced
p-groups, we always have to deal with endomorphisms to and from cyclic sum-
mands. These as well as the small endomorphisms are ‘inessential’ endomorphisms:
they do not reveal much about the group structure. The relevant information
about the group encoded in the endomorphism ring is actually in the other
endomorphisms. Also, certain torsion-free groups (like separable groups) admit
endomorphisms that provide hardly any information about group. The idea of
formalizing this phenomenon is due to Corner–Göbel [1]. We define this concept
in the local case.
Let A be a reduced p-local group with A1 D 0, and B a basic subgroup of A.
Evidently, B
A
B,Q and every  2 End A extends uniquely to an Q 2 End B.
Q Now
 2 End A is called inessential if . Q
Q B/
A: The inessential endomorphisms form
an ideal of End A, denoted Ines A.
Q < .Jp /N . In this case, Ines A consists of
Example 1.8. Let A Š .Jp /.N/ . Thus B Š .Z.p/ /.N/ and B
those  2 End A for which Im  is contained in a finite direct sum of the Jp .
Example 1.9. Let A be a separable p-group, and B its basic subgroup. Now the ideal Ines A
coincides with set of the small endomorphisms (Sect. 3 in Chapter 7).
If  2 Ends A, then for every k 2 N, there is an n 2 N such that .pn AŒpk
/ D 0. Then also
nQ k Q , so Q.B
Q/ D
Q.p BŒp
/ D 0, thus Q is also small. By Sect. 3(E) in Chapter 7, B C Ker Q D B
Q.B/ D .B/  A. Thus Ends A  Ines A.
Suppose  is not small, i.e. for some k 2 N and for all n 2 N, .pn AŒpk
/ ¤ 0. We can select
independent elements ai 2 A of orders  pk and of increasing heights ni such that ai ¤ P0. We
may, in addition, assume that the elements .ai / are also independent. Then the sum c D i<! ai
converges in BQ , but Q.c/ ¤ 0. Consequently, Ines A  Ends A, and so Ines A D Ends A.

Annihilator Ideals in End A There is a fundamental correspondence between

certain subgroups of a group, and certain left ideals of its endomorphism ring.
1 Endomorphism Rings 617

For a subgroup G of the group A, we set

G? D f 2 End A j .g/ D 0 8g 2 Gg;

and for a left ideal L of End A, we define

L? D fa 2 A j .a/ D 0 8 2 Lg:

Evidently, G? is a left ideal in End A, called annihilator left ideal, and L? is a

subgroup of A, called kernel subgroup.
(a) For every subgroup G
A, we have G??? D G? , and L??? D L? for every
left ideal L in End A.
(b) A subgroup G
A is a kernel subgroup exactly if G?? D G, and a left ideal L
is an annihilator left ideal if and only if L?? D L.
(c) The correspondences G 7! G? and L 7! L? are inclusion reversing inverse
maps between the set of kernel subgroups of A and the set of annihilator left
ideals of End A.
Example 1.10.
(a) Consider the group Z.p1 / and its endomorphism ring Jp . The annihilator ideals are pn Jp , and
Z.pn /. Z.p1 // are the kernel subgroups.
(b) Let A Š Q ˚    ˚ Q, an n-dimensional Q-vector space. Then E D End A is isomorphic
to the n  n-matrix ring over Q. Every subspace is a kernel subgroup and every left ideal an
annihilator.
Finite Topology Matrix representation of linear transformations is an important
issue in linear algebra. For groups it is possible to establish a similar, though much
less informative, representation of endomorphisms of direct sums. Since we do not
wish to restrict our study to finite direct decompositions, we need to introduce a
natural topology in endomorphism rings.
Actually, endomorphism rings admit various topologies defined in terms of the
underlying groups. They play an increasingly important role in the study of groups
(also modules, rings, etc.). We will discuss the most significant topology, the so-
called finite topology that was introduced into the theory of abelian groups by
Szele [8].
The terminology comes from the fact that the open neighborhoods are defined in
terms of finite subsets of the group: every finite subset X in A defines an open set
about ˛ 2 End A, viz.

UX .˛/ D f 2 End A j x D ˛x 8x 2 Xg:

It is clear that UX .˛/ D \x2X Ux .˛/; also, UX .˛/ D ˛ C UX .0/ for each ˛ 2 End A.
Thus, the finite topology can more conveniently be defined with the aid of a subbase
of neighborhoods of 0, by the open sets UX D f 2 End A j X D 0g, taken for all
finite subsets X
A. The finite topology is evidently Hausdorff, and moreover, it is
linear: the open sets UX are left ideals in End A. As a consequence, the continuity of
the addition in End A is immediate. Moreover, we can state:
618 16 Endomorphism Rings

Theorem 1.11. The endomorphism ring End A of a group A is a complete topolog-

ical ring in the finite topology.
Proof. To prove that ring multiplication is continuous, let ˛; ˇ 2 End A, and let
˛ˇ C UX be an open neighborhood of ˛ˇ. Since UX is a left ideal and since UˇX ˇ

UX , the desired continuity follows from

.˛ C UˇX /.ˇ C UX /  ˛ˇ C UˇX ˇ C UX  ˛ˇ C UX :

Therefore, End A is a topological ring.

To verify completeness, let f˛X gX2I denote a Cauchy net where the index set
I (the set of finite subsets of A) is partially ordered by inclusion. Reminder: the
Cauchy net satisfies: given X 2 I, ˛Y  ˛Z 2 UX holds for all Y; Z 2 I containing
some X0 2 I; here X  X0 may be assumed, so that ˛Y x D ˛Z x .8x 2 X/ for large
Y; Z. Therefore, the common value of the ˛Y x for large Y defines an element x0 2 A,
and it is readily seen that the assignment ˛ W x 7! x0 is an endomorphism of A. This
is the limit of our Cauchy net, as is evident from ˛  ˛Y 2 UX for all Y  X0 . t
u
Observe that the factor group .End A/=UX is isomorphic to a subgroup of a direct
sum of copies of A, because UX is the kernel of the homomorphism End A ! AX
defined as  7! .x1 ; : : : ; xn /, where X D fx1 ; : : : ; xn g.
The finite topology raises several questions. An immediate one is whether or not
the finite topology can be discrete for an infinite group.
Theorem 1.12 (Arnold–Murley [1]). The endomorphism ring of a group A is
discrete in the finite topology only if A is self-small. The converse holds for countable
End A.
Proof. Suppose A is not self-small, so there exists a map  W A ! ˚i<! Ai
where Ai Š A and i  ¤ 0 for all i; here i denotes the ith projection. Define
Gn D fa 2 A j i .a/ D 0 8i  ng, an increasing chain of proper subgroups with
union A. Then f 2 End A j .Gn / D 0g ¤ 0 for all n
!. Any finite subset X of
A is contained in some Gn , so UX ¤ 0. Thus no open set in the finite topology is 0,
and End A is not discrete in the finite topology.
Conversely, suppose that End A is not discrete in the finite topology. Let U0 >
   > Un > : : : be a strictly descending chain of neighborhoods in End A; if End A
is countable, then there is no harm in assuming that \n<! Un D 0 as well. Choose
n 2 Un such that n P … UnC1 . Define a homomorphism  W A ! ˚i<! Ai with
Ai Š A by setting  D n<! n . This means that A is not self-small. t
u
Example 1.13.
(a) The finite topology on the endomorphism ring of A is discrete if A is finite, or torsion-free of
finite rank, or a rigid group, but not discrete if A is an infinite p-group.
(b) Jp is self-small, but its endomorphism ring is not discrete in the finite topology. The same
holds for torsion-complete p-groups with finite UK-invariants.
When the Finite Topology is Compact Compactness being always of particu-
lar interest, let us turn our attention to the question as to when End A is compact in
the finite topology.
1 Endomorphism Rings 619

Before stating the relevant result, consider the fully invariant subgroup Ox D
fx j  2 End Ag of A which we will call the orbit of x 2 A. The evaluation map
 7! x is a group homomorphism of End G onto Ox whose kernel is Ux . This leads
to the group isomorphism .End A/=Ux Š Ox (which is moreover an End A-module
map).
Proposition 1.14. The endomorphism ring End A of a group A is compact in the
finite topology if and only if A is a torsion group whose p-components are finitely
cogenerated.
Proof. Knowing that End A is complete in the finite topology, for the compactness
of End A it is necessary and sufficient that all neighborhoods Ux be of finite indices.
This is the case exactly if the orbits are finite.
First, suppose End A is compact. As hxi is a subgroup of Ox , A ought to be a
torsion group. By Corollary 2.3 in Chapter 5, every non-zero p-component Ap of A
contains a cocyclic summand; let Cp be one of minimal order in Ap . The orbit of its
cogenerator is the socle of Ap , so it must be finite. Thus Ap is finitely cogenerated
(hence a finite direct sum of cocyclic groups).
Conversely, if A D ˚p Ap with finitely cogenerated Ap , then for every n 2 N, AŒn

is finite. But x 2 AŒn

implies Ox
AŒn
, thus all Ux are of finite indices in End A
(and the same holds for all neighborhoods UX ). Consequently, End A is compact in
the finite topology. t
u
Though the endomorphism ring of a p-group is rarely algebraically compact as
a ring, we can still claim that its endomorphism group is algebraically compact.
In fact, from Theorem 2.1 in Chapter 7 we know that Hom.A; / is algebraically
compact whenever A is a p-group. The invariants can be computed by making use
of theorems by Pierce [1].
Matrix Representations of Endomorphisms Once we have a ring topology P in
End A, it makes sense to form convergent infinite sums. An infinite sum i2I ˛i
with ˛i 2 End A is convergent if, forPeach x 2 A, almost all ˛i x D 0, in which case
˛ 2 End A is its limit where ˛x D i ˛i x for x 2 A. A matrix k˛ji k withPentries
in End A is said to be column-convergent if for each column i, the sum j ˛ji is
convergent.
Suppose A D ˚i2I Ai is a (finite or infinite) direct sum, and i are the associated
projections, viewed as P mutually orthogonal idempotents in End A. Every a 2 A
can be written as P a D i i aP where almost all terms vanish. For ˛ 2 End A, we
then have ˛a D i ˛i a D i;j .j ˛i /a. In this way, with every ˛ 2 End A we
associate an I  I-matrix k˛ji k where ˛ji D j ˛i . If kˇji k with ˇji D j ˇi is the
matrix associated with ˇ 2 End A, then the matrices associated with ˛  Pˇ and ˛ˇ
are precisely the difference matrix k˛ji  ˇji k and the product matrix k k ˛jk ˇki k,
respectively. We conclude that  W ˛ ! k˛ji k is a ring homomorphism whose kernel
is evidently 0. P
For a fixed index i, ˛i a D j ˛ji a converges for every a 2 A, indicating that
the matrix k˛ji k is column-convergent. Conversely, if a matrix with entries ˛ji 2
j .End A/i is column-convergent, then it must come from an ˛ 2 End A, namely,
620 16 Endomorphism Rings

P
P ˛ D
from i;j ˛ji . This is a convergent sum, because for each a 2 A, the sum
i;j ˛ji a is finite. If j .End A/i is identified with Hom.Ai ; Aj /, then we can state our
findings as follows.
Proposition 1.15. Let A D ˚i2I Ai be a direct sum decomposition of A. Then End A
is isomorphic to the ring of all column-convergent I  I-matrices k˛ji k where ˛ji 2
Hom.Ai ; Aj /. t
u
Needless to say, if the Ai are small objects, then every column contains but a
finite number of non-zero entries, and if the group is a finite direct sum, then its
endomorphism ring will be the full matrix ring with the indicated entries.
Example 1.16. Let A D ˚i2I hai i be a free group. In the matrix representation of End A, the
entries are integers, and in each column, almost all entries are 0.
Example 1.17. If D D ˚i2I Di is a torsion-free divisible group with Di Š Q, then we are in the
same situation as in the preceding example, the only difference is that the entries of the matrix can
be arbitrary rational numbers.
Example 1.18. Let A D A0 ˚ .˚p Ap /, where A0 is torsion-free, while the Ap are p-groups
belonging to different primes. Then the matrices representing the endomorphisms are !  !-
matrices of the form
0 1
˛00 0 0 ::: 0 :::
B˛20 ˛22 0 : : : 0 : : :C
B C
B˛30 0 ˛33 : : : 0 : : :C
B C
B : : : : : : : : : : : : : : : : : :C
B C
@˛ 0 0 ::: ˛ : : :A
p0 pp
::: ::: ::: ::: ::: :::

where ˛00 2 End A0 ; ˛p0 2 Hom.A0 ; Ap /; and ˛pp 2 End Ap .

F Notes. There is an extensive literature on endomorphism rings; this is one of the most
investigated areas in the theory. Readers interested in the subject are referred to Krylov–Mikhalev–
Tuganbaev [KMT] where a large amount of material on the endomorphism rings of abelian groups
is presented in a systematic manner with full proofs. Important results are also available on
endomorphism rings of groups (or modules) with distinguished submodules; these are instrumental
in deriving results on endomorphism rings.
The finite topology provides additional information about the relation between the group and
its endomorphism ring. It is a rare possibility of defining the finite topology intrinsically (i.e.,
solely in the endomorphism ring, without referring to the group), but it seems it is a most relevant
feature in cases when the endomorphism ring determines the group, like for torsion groups, certain
completely decomposable groups, etc. Recently, May [7] discusses the use of finite topology.
There are a few ideals in the endomorphism ring that are of interest from the group theoretical
point of view. The most widely studied ideal is the Jacobson radical J.A/. In general, there is no
satisfactory characterization for J.A/ in terms of the group A, only in special cases. Another notable
ideal is Ines A, the set of inessential endomorphisms. For a study of right ideals, see Faticoni [2].
No detailed discussion will be given here to the ideals of End A.
Let us point out some results by Mishina on endomorphism rings. In her paper [3], she
characterizes the groups A for which every endomorphism of each subgroup extends to an
endomorphism of A: these groups are either divisible, or torsion with homogeneous p-components.
In another paper [4], she shows that all the endomorphisms of all factor groups A=C lift to A if
and only if A is either free, or torsion with homogeneous p-components, or else the direct sum
of a divisible torsion group and a finitely generated free group. (Similar results are proved for
automorphisms, see Sect. 1, Exercise 8 in Chapter 17.)
1 Endomorphism Rings 621

Recently, several publications deal with the so-called algebraic entropy which was recently
introduced in abelian groups. The paper Dikranjan–Goldsmith–Salce–Zanardo [1] contains lots of
interesting results on the entropy of endomorphisms.

Exercises
2
(1) If jAj D pn , then j End Aj
pn .
(2) (a) If G
A, then Hom.A; G/ is a right ideal in End A.
(b) If G is fully invariant, then Hom.A; G/ is a two-sided ideal.
(3) (Lawver) All endomorphic images of A are fully invariant exactly if, for every
a 2 A and for all ;  2 End A, there is a b 2 A such that ./a D b.
(4) Show that the finite topology of End A for a separable torsion-free group A can
be defined intrinsically (i.e. without reference to A).
(5) Describe the finite topology of Jp as an endomorphism ring of Z.p1 / and as
that of Jp .
(6) (a) The direct sum and the direct product of elementary p-groups Tp for
different primes p have isomorphic endomorphism rings.
(b) However, these endomorphism rings are not isomorphic as topological
rings (equipped with the finite topology).
(7) For an infinite group A, End A is always infinite. Give examples where jAj <
j End Aj, and where jAj > j End Aj.
(8) (a) Let fGi gi2I be a system of subgroups of A which is directed upwards
under inclusion such that [i2I Gi D A. Define a topology in End A by
declaring the set of left ideals Li D f 2 End A j Gi D 0g as a base
of neighborhoods about 0. Show that End A is a complete group in this
topology, and if the Gi are fully invariant in A, then End A is a topological
ring.
(b) (Pierce) Let A be a p-group and Gi D AŒpi
.i < !/. Then End A is a
complete topological ring in this topology.
(9) A self-small group is not the direct sum of infinitely many non-zero groups,
but it may be decomposed into the direct sum of any finite number of non-zero
summands. [Hint: torsion-complete with standard basic.]
(10) A group is self-small if all of its endomorphisms ¤ 0 are monic.
(11) (Dlab) Let D be a divisible p-group of countable rank. Representing endomor-
phisms by matrices over Jp , show that in each column almost all entries are
divisible by pk for every k > 0.
(12) Suppose A D ˚i2I Ai with countable summands Ai . In the matrix represen-
tations of endomorphisms, every column contains at most countably many
non-zero entries.
(13) The set EndE A of E-endomorphisms of A is the center of the ring E D End A.
(14) For any group A, the center of the ring End.A ˚ Z/ is Š Z.
622 16 Endomorphism Rings

2 Endomorphism Rings of p-Groups

It is a rather intriguing question to find general properties shared by the endomor-

phism rings of p-groups. Fortunately, substantial information is available, and we
wish to discuss some relevant results.
Role of Basic Subgroups We start with the observation that underscores the
relevance of basic subgroups also from the point of view of endomorphisms: any
endomorphism of a reduced p-group is completely determined by its action on a
basic subgroup. Actually, a stronger statement holds: any homomorphism of a p-
group A into a reduced group C is determined by its restriction to a basic subgroup
B of A. The exact sequence 0 ! B ! A ! A=B ! 0 induces the exact sequence
0 D Hom.A=B; C/ ! Hom.A; C/ ! Hom.B; C/ which justifies our claim.
The Finite Topology It should be pointed out that in case of reduced p-groups,
the finite topology of the endomorphism ring can be defined intrinsically, without
reference to the underlying group. If x 2 A, then there are a projection  W A ! hci
onto a cyclic summand and a  2 E such that x D c. Manifestly, the neighborhood
Ux (annihilating x) is nothing else than the left annihilator ideal ./? .
Proposition 2.1. Let A be a separable p-group.
(i) The finite topology of its endomorphism ring E can be defined by taking the left
annihilators of the primitive idempotents.
(ii) In the finite topology of E, the left ideal E0 of E generated by the primitive
idempotents is dense, and its completion is E.
Proof. As Ux
Upx and every element is contained in a finite summand, the U for
primitive idempotents form a subbase.
For (ii) we show that for every  2 E and for every neighborhood Ux , the coset
 C Ux intersects E0 . Now Ux D .pk /? (if hxi D pk A) is a left ideal, so 1   2 Ux
implies .1  / 2 U whence  2  C U follows. t
u
Structure of End for p-Groups We are in the fortunate situation that a lot is
known about the endomorphism rings of p-groups. As a matter of fact, Liebert [3]
gave a complete characterization in the separable case. A much less informative, but
perhaps more attractive information is recorded in the next theorem.
Before stating the theorem, we recall a definition: a ring E is a split extension of
a subring R by an ideal L of E if there exists a ring homomorphism  W E ! R that
is the identity on R, and L D Ker. We write E D R ˚ L (direct sum in the group
sense).
Theorem 2.2 (Pierce [1]). For a p-group A, End A is a split extension

End A Š R ˚ Ends A;

where R is a ring whose additive group is the completion of a free p-adic module,
and Ends A is the ideal of small endomorphisms of A.
2 Endomorphism Rings of p-Groups 623

Proof. This is an immediate consequence of Theorem 3.3 in Chapter 7. t

u
Baer–Kaplansky Theorem The endomorphism ring of an abelian group may
contain more or less information about the group itself. There are arbitrarily large
torsion-free groups whose endomorphism rings are just Z, or a subring of Q, so their
endomorphism rings reveal no more about their structure than indecomposability.
But, on the other hand, if A is cocyclic, then End A (now Š Z=pk Z or Š Jp )
completely characterizes A among the torsion groups. Indeed, in this case End A
has only two idempotents: 0 and 1, so any torsion group C with End C Š End A
must be indecomposable, and hence cocyclic. It is a trivial exercise to check that
only C Š Z.pk / .k 2 N or 1/ is a possibility.
Example 2.3. It can very well happen that a torsion group and a torsion-free group have
isomorphic endomorphism rings. Example: Jp is the endomorphism ring of both Z.p1 / and Jp .
Exercise 6 in Sect. 1 provides an example of the same situation for a torsion and a mixed group.

It is natural to wonder in which cases the group can entirely be recaptured from
its endomorphism ring, or, more accurately, when the isomorphy of endomorphism
rings implies that the groups themselves are isomorphic. We will show that this is
always the case for torsion groups. The proof relies on a distinguished feature of
torsion groups: they have lots of summands, even indecomposable summands, so an
adequate supply of idempotent endomorphisms is at our disposal.
We start with a preparatory lemma.
Lemma 2.4 (Richman–Walker [1]). Let E D End A where A is a p-group.
(i) If A is a bounded or a divisible group, then A Š E for an idempotent  2 E.
(ii) If A has an unbounded basic subgroup, then there are idempotent endomor-
phisms i .i 2 N/ such that

A Š lim Ei
!
with connecting maps i W i 7! i iC1 . 2 E; i 2 N/:
Proof. Let A be a bounded p-group, and  2 E a primitive idempotent of maximal
order pk . Then A D hci is a summand of A. We claim that ˛ W E ! A is a group
isomorphism where ˛ W  7! .c/ D .c/ . 2 E/. It is clearly a homomorphism,
and it is surjective, since c can be mapped onto every element of A by a suitable .
Moreover, it is monic, because if we have .c/ D 0 for some  2 E, then also
 D 0. Similar argument applies if A is divisible.
Next, let A have an unbounded basic subgroup. Then there exist primitive
idempotents i 2 E of orders pni with n1 <    < ni < : : : such that A D
C1 ˚    ˚ Ci ˚ Ai with Cj < Ai for all j > i, where Ci D i A is cyclic of order pni .
Choose endomorphisms i W A ! Ci satisfying i  Ci D 1Ci , i .CiC1 / D Ci , while
i .Cj / D 0 for j ¤ i; iC1 and i .AiC1 / D 0. Then for every i 2 N, i D i Ci iC1 is
an idempotent endomorphism with Im i D Ci . Evidently, we can select generators
ci of Ci inductively so as to satisfy i .ciC1 / D ci for all i.
624 16 Endomorphism Rings

Consider the E-modules Ei .i 2 N/. The E-map i W Ei ! EiC1 defined

by i .i / D i iC1 is monic; indeed, if i iC1 D 0 . 2 E/, then i .ci / D
i iC1 .ciC1 / D 0 implies i D 0 (since i is 0 everywhere else). We thus have
a direct system Ei .i 2 N/ of left E-modules with connecting maps i . As above,
we can show that the map ˛i W i 7! .ci / of Ei to AŒpni
is an E-isomorphism.
Moreover, the diagram

α
i
Eπi −−−− → A[pni ]
⏐ ⏐
⏐γi ⏐β
  i
αi+1
Eπi+1 −−−−→ A[pni+1 ]

(where ˇi is the inclusion map) commutes for each i, because ˇi ˛i .i / D .ci / and
˛iC1 i .i / D ˛iC1 .i iC1 / D i .ciC1 / D .i C i iC1 /.ciC1 / D i .ciC1 / D
.ci /: Since A D lim AŒpni
, we obtain A Š lim Ei . t
u
! !
Theorem 2.5 (Baer [9], Kaplansky [1]). If A and C are torsion groups whose
endomorphism rings are isomorphic, then A Š C.
Moreover, every ring isomorphism W End A ! End C is induced by a group
isomorphism  W A ! C; i.e. W  7!  1 .
Proof. The proof can at once be reduced to p-groups. So suppose A; C are p-groups,
and W End A ! End C is a ring isomorphism. To simplify notation, write E D
End A.
If A is bounded or divisible, then by Lemma 2.4(i) we have A Š E for
an idempotent  2 E. Similarly, C Š .E/ ./ whence the existence of an
isomorphism  W A ! C is immediate.
If A D B˚D where B is bounded and D is divisible, then End A has a torsion-free
part Š End D that is an ideal, whose two-sided annihilator is Š End B. These ideals
are carried by to the corresponding ideals of End C. The settled cases imply that
A and C have isomorphic bounded and divisible subgroups.
If A has unbounded basic subgroup, then in view of Lemma 2.4(ii) there are
idempotents i 2 E such that A Š lim Ei , and clearly C Š lim .E/ .i / with
! !
corresponding connecting maps. Thus again A Š C follows.
We proceed to the second claim. For simplicity we identify A with E or with
lim Ei , as the case may be, and similarly for C. Let  W A ! C be the isomorphism
!
induced by from E to .E/ ./, or by between the direct limits. Then the
endomorphism ./ . 2 E/ acting on C can also be obtained by using  1 ,
applying  and followed by , i.e. ./ D  1 , as claimed. t
u
Since all the primitive idempotents are contained in the ideal Ends A of small
endomorphisms, from the foregoing proof we conclude:
Corollary 2.6 (Pierce [1]). The torsion groups A and C are isomorphic exactly if
the rings Ends A and Ends C are isomorphic. t
u
2 Endomorphism Rings of p-Groups 625

An immediate corollary to Theorem 2.5 is the following remarkable fact:

Corollary 2.7 (Baer [9]). Every automorphism of the endomorphism ring of a
torsion group is inner.
Proof. Let ˛ be an automorphism of End A, where A is a torsion group. Theorem 2.5
asserts that it must act as ˛ W  7!  1 for some automorphism  of A
( 2 End A). Here  is viewed as a unit in End A. t
u
Liebert’s Theorem The following characterization of the endomorphism rings
of separable p-groups is an important document, though it seems difficult to use it.
We state it without proof.
Theorem 2.8 (Liebert [3]). For a ring E, there exists a separable p-group A such
that End A Š E if and only if the following conditions are satisfied:
(i) the sum E0 of all minimal non-nil left ideals is a ring whose additive group is a
p-group, and whose left annihilator in E is 0;
(ii) if ;  are primitive idempotents in E, then the additive group of E is cyclic;
(iii) if ;  are primitive idempotents in E such that o./
o./, then the left
annihilator of E is contained in the left annihilator of E, and EE D
EŒo./
;
(iv) a proper right ideal of E0 whose left annihilator in E0 is nilpotent is not a
summand in E0 ;
(v) E is complete in its finite topology. t
u
Center of the Endomorphism Ring We proceed to identify the center of the
endomorphism ring of a torsion group. The general case immediately reduces to
primary groups, and it is more elegant to formulate the result for p-groups.
Theorem 2.9 (Charles [1], Kaplansky [K]). The center of the endomorphism ring
of a p-group consists of multiplications by p-adic integers or by elements of the
residue class ring Z=pk Z according as the group is unbounded or bounded by
pk .with minimal k/.
Proof. Multiplication by a rational or a p-adic integer is an endomorphism of any
p-group A; it evidently commutes with every  2 End A.
Conversely, in the center of End A means that the map W A ! A is an End A-
module homomorphism. We now appeal to Lemma 6.1(i) below to infer that must
act as multiplication by a  2 Jp . t
u
The Jacobson Radical The Jacobson radical J.A/ of the endomorphism ring
E of a p-group A has been extensively studied, but so far there is no satisfactory
characterization. Pierce [1] compared it to the set

H.A/ D f 2 E j .pn AŒp

/
pnC1 AŒp
g;

which is an ideal of E located between pE and E (also called Pierce radical); it

consists of all endomorphisms that strictly increase the heights of elements of finite
heights in the socle. From the point of view of endomorphism rings of p-groups,
626 16 Endomorphism Rings

H.A/ seems to be more tractable than the Jacobson radical. We now prove results
that involve H.A/.
Lemma 2.10 (Pierce [1]).
(i) For every reduced p-group A, H.A/ contains the Jacobson radical J.A/ of
End A.
(ii) For a separable p-group A,P J.A/ D H.A/ if and only if, for all a 2 AŒp
and all
 2 H.A/, the infinite sum n<! n .a/ converges .in the p-adic topology of A/.
(iii) If A is torsion-complete, then J.A/ D H.A/.
Proof.
(i) Assume  2 J.A/, but  … H.A/. Thus for some n < !, there is an a 2 AŒp
of
finite height n such that also hp .a/ D n. Then a is in the socle of a summand
hbi, and a is in the socle of a summand hci, both of order pnC1 . Evidently,
there is a  2 End A mapping hci onto hbi such that .a/ D a. Now  2 J.A/
implies  2 J.A/, so 1 is an automorphism of A. However, .1/a D 0,
a contradiction.
(ii) If  2 H.A/ and H.A/ D J.A/, then 1 is an automorphism of A. Thus, given
a 2 AŒp
, there is a b 2 AŒp
such that .1  /b D a. It follows that the partial
sums bn D a C .a/ C    C n .a/ D .1  nC1 /b satisfy b  bn D nC1 .b/.
Since  2 H.A/ guarantees that hp .nC1 .b// > n, the sequence bn converges
to b.
Conversely, if we know that all the infinite sums of the stated kind converge,
then we can show that 1   is an automorphism of A for all  2 H.A/. For
each a 2 AŒp
, we have hp ..1  /a/ D hp .a/, and therefore Ker.1  / D 0.
To see that Im.1  / D A, the proof goes by induction on the order of a 2 A,
to verify thePexistence of a c 2 A such that .1  /c D
P a. If o.a/ D p, then for c
we choose n<! n .a/. If o.a/ D pkC1 , then b0 D n<! n .pk a/ must belong
to pk A, because all the partial sums belong to this (closed) subgroup. Hence
b0 D pk u for some u 2 A, and we have .1  /pk u D pk a. Now a  .1  /u
is of order
pk , so by the induction hypothesis there is a v 2 A satisfying
.1  /v D a  .1  /u. Then c D u C v is mapped by 1   upon a.
(iii) If A is a torsion-complete p-group, then the sums in (ii) always converge, and
therefore, J.A/ D H.A/. t
u
Example 2.11. Let A D ˚n<! han i where o.an / D pn . Then J.End A/ ¤ H.End A/. Indeed, the
correspondence  W an 7! panC1 defines an endomorphism in H.A/, and the sequence .1 C  C
   C k /a1 .k < !/ does not converge.

Proposition 2.12 (Pierce [1]).

(i) For a reduced p-group A, there is a ring embedding
Y
W .End A/=H.A/ ! Mfn .A/ (16.2)
n<!
2 Endomorphism Rings of p-Groups 627

where Mfn .A/ denotes a matrix ring over the prime field Fp whose dimension
is the nth UK-invariant fn .A/ of A. Im contains the direct sum of the matrix
rings.
(ii) For a separable p-group A, is an isomorphism if and only if A is torsion-
complete.
Proof.
(i) Every  2 End A induces a linear transformation n of the fn .A/-dimensional
vector space pn AŒp
=pnC1 AŒp
. Thus  7! .n /n<! induces a ring homomor-
phism from End A to the right-hand side of (16.2), whose kernel is exactly H.A/.
If we write the basic subgroup of A as B D ˚n<! Bn where Bn is the direct
sum of cyclic groups of order pnC1 , then pn AŒp
=pnC1 AŒp
Š Bn Œp
. It is clear
that any linear transformation n on Bn Œp
extends to an endomorphism of Bn ,
and then to an endomorphism  of A. This is always true for a simultaneous
extension of a finite number of n s, and also for infinitely many provided A is
torsion-complete.
(ii) That is an isomorphism only if the separable p-group A is torsion-complete
can
P be seen from the representation of elements in separable p-groups as
n<! bn with bn 2 Bn . If an arbitrarily
P chosen collection n 2 End Bn .n < !/
extends to  2 End A, then all n<! .bn / must be contained in A, thus B
A.
t
u
The proof shows that Im is a subdirect product of the matrix rings.
F Notes. The study of the endomorphism rings of p-groups A was initiated by Pierce [1]; in
this important paper, he proved several relevant results. In a sequel [3] to [1], he characterizes
End A within a class of rings when A is separable with a prescribed basic subgroup. A more
satisfactory realization theorem is due to Liebert [1, 3] who characterized the endomorphism
rings as rings, first for bounded, and later for separable p-groups. (Though it is an important
contribution, it still falls short of the true significance, since the conditions are not illumi-
nating.) See also Liebert [4]. Goldsmith [2] examines endomorphism rings of non-separable
p-groups.
Generalizing Corner [5], Dugas–Göbel [3] show that for every reduced ring R over Jp , whose
additive group is torsion-free and algebraically compact, there exists a separable p-group A such
that End A is a split extension of R and Ends A. They also prove, for every cardinal , the existence
of  separable p-groups, all with such a fixed R, so that all homomorphisms between them are
small.
Ideals in endomorphism rings have been discussed by Liebert [2] and Monk [1]. According
to Hausen [4] and Ivanov [1], the sum of nilpotent ideals is the collection of all  2 End A, for
which there is a finite chain 0 D A0 < A1 <    < An D A of fully invariant subgroups such
that AiC1  Ai . Hausen [5] generalizes the ideal H.A/ by defining IA as the collection of all
 2 End A for which there is a finite sequence 0 D 0 < 1 <    < n D  of ordinals such that
.pi AŒp
/  piC1 AŒp
for i D 0; 1; : : : ; n  1, where  denotes the length of A. Her theorems are
concerned with totally projective groups.
Several authors discussed a modified version of the Baer–Kaplansky theorem: when the
isomorphism of the Jacobson radicals of the endomorphism rings implies the isomorphism of
the groups. See, e.g., Flagg [1], Hausen–Johnson [1], Hausen–Praeger–Schultz [1], Schultz [2].
Puusemp [1] showed that the conclusion of Theorem 2.5 remains true if only the isomorphy
of the multiplicative endomorphism semigroups is assumed. There are theorems similar to
Theorem 2.5 on certain mixed groups. For the endomorphism semigroup, see also Sebel’din [3]. In
several papers, May–Toubassi (see, e.g., [1]) showed that two mixed local groups of torsion-free
628 16 Endomorphism Rings

rank 1 with totally projective torsion subgroups are isomorphic if their endomorphism rings are
isomorphic; for a survey, see May [4]. Files–Wickless [1] extended the Baer–Kaplansky theorem
to a class of global mixed groups. Files [1] proved the isomorphy of reduced local Warfield groups
with isomorphic endomorphism rings.
Nunke [6] has a nice generalization of Theorem 2.8: If A; C are unbounded p-groups, then
End A and End C embed in End.Tor.A; C// in such a way that they are centralizers of each other,
and their intersection is precisely the center of End.Tor.A; C//. The special case C Š Z.p1 /
yields Theorem 2.9.
A fairly large literature deals with the problem as to when End A equals the subring generated
by its units (i.e., by Aut A). Castagna [1] gives an example where this subring is a proper subring.
Hill [6] shows that if p > 2, then every endomorphism of a totally projective p-group is the sum
of two automorphisms. Hill–Megibben–Ullery [1] prove the same for local Warfield groups. See
also Goldsmith–Meehan–Wallutis [1] where the unit sum numbers (number of units needed to be
added to get the endomorphisms) are investigated.
Bunina–Mikhalëv [1] investigate when the endomorphism rings of two p-groups are elemen-
tarily equivalent.

Exercises

(1) Give more examples to show that a torsion-free group and a torsion group may
have isomorphic endomorphism rings.
(2) (Levi) Find the endomorphisms which map every subgroup into itself. In
particular, for p-groups.
(3) If A is a torsion group, then the Z-adic topology of End A is finer than its finite
topology. [Hint: n.End A/
Ux if nx D 0.]
(4) Let A be a separable p-group with basic subgroup B. Then End A is a closed
subring in End B (in the finite topology).
(5) (Szele–Szendrei) A torsion group A has commutative endomorphism ring
exactly if A
Q=Z.
(6) Find two non-isomorphic p-groups with isomorphic endomorphism groups.
(7) Verify the analogue of the Baer–Kaplansky theorem for adjusted cotorsion
groups.
(8) For a divisible p-group D, the Jacobson radical of End D is equal to p End D.
(9) (a) Assume A is a torsion group. The endomorphisms of A with finitely
cogenerated images form an ideal V.A/ in End A. It is the ideal generated
by the primitive idempotents.
(b) Follow the proof of the Baer–Kaplansky theorem to conclude that a ring
isomorphism V.A/ Š V.C/ implies A Š C provided that C, too, is torsion.
(10) Let A be a torsion-complete p-group, and  2 End A with Ker  D 0. If  maps
a basic subgroup into a basic subgroup, then  2 Aut A.
(11) (D’Este [1]) A group C is said to be an E-dual of A if End A and End C are
anti-isomorphic rings. A reduced p-group has an E-dual if and only if it is
torsion-complete with finite UK-invariants. [Hint: summands Bn in a basic
subgroup B D ˚n Bn are finite; A is separable and B
A.]
3 Endomorphism Rings of Torsion-Free Groups 629

3 Endomorphism Rings of Torsion-Free Groups

In contrast to torsion groups, non-isomorphic torsion-free groups may very well

have isomorphic endomorphism rings. Another major difference in the behavior
of endomorphism rings between torsion and torsion-free groups lies in the fact
that only minor restrictive conditions hold for the torsion-free case. We will see in
Theorem 7.1 that only slight restriction on the ring (to be cotorsion-free) is enough
to guarantee that it is an endomorphism ring of a torsion-free group.
Example 3.1 (Corner [2]). A ring whose additive group is isomorphic to Q ˚ Q cannot be the
endomorphism ring of an abelian group. For, such a group must be torsion-free divisible, and it
cannot be of rank 1, neither of rank
2, because then the rank of its endomorphism ring is 1, and

4, respectively. (Similar argument holds for Q ˚ Q ˚ Q.)

Example 3.2 (Sa̧siada [4], Corner [2]). There exist torsion-free groups of finite rank whose endo-
morphism groups are isomorphic, but their endomorphism rings are not. Indeed, by Theorem 3.3
below, there exist such groups with endomorphism rings isomorphic to the ring Z ˚ Z and to the
ring of the Gaussian integers Z C Zi.

Corner’s Theorem We enter our study of torsion-free endomorphism rings with

the following striking theorem. Though it is a corollary to Theorem 7.1 that is by far
more general, we offer a full proof of this historically important result that opened
new prospects in the theory; this proof is needed, because we will give no detailed
proof for Theorem 7.1. The idea of localization in the proof is due to Orsatti [2], so
Corner’s method will be needed merely in the more tractable local case.
Theorem 3.3 (Corner [2]). Every countable reduced torsion-free ring is isomor-
phic to the endomorphism ring of some countable reduced torsion-free group.
Proof. Let R be a countable, p-local reduced torsion-free ring. It is a Z.p/ -algebra,
Hausdorff in its p-adic topology, so it is a pure subring in its p-adic completion
Q which is a torsion-free Jp -algebra with the same identity. Choose a maximal set
R,
fzn .n < !/g in R that is linearly independent over Jp . Thus, for every a 2 R, there
is a dependence relation

pk a D 1 z1 C    C m zm .i 2 Jp /;

for some k and m, where the coefficients i are uniquely determined up to factors
pj . We denote by S the pure subring of Jp generated by Z.p/ and the i for all a 2 R.
Clearly, S is still countable. We proceed with the construction which requires several
steps.
(a) Suppose 1 ; : : : ; n 2 Jp are linearly independent over S. We claim: 1 r1 C
   C n rn D 0 .ri 2 R/ implies that all ri D 0. In fact, for sufficiently
`
P P` 2 N, there are relations p ri D i1 z1 C    C im zm P
large with ij 2 S, thus
i j i ij zj D 0. By the independence of the zj , we have i i ij D 0 .j D
1; : : : ; m/ whence, by hypothesis on the i , all ij D 0, and ri D 0.
630 16 Endomorphism Rings

(b) For each non-zero a 2 R choose a pair a ; a of p-adic integers such that the set
fa ; a j 0 ¤ a 2 Rg is algebraically independent over S. This is possible, for S
is countable, and the transcendence degree of Jp over S is the continuum. Set

Q
ea D a 1 C a a 2 R; (16.3)

and define the group A as the pure subgroup

Q
A D hR; Rea 8a 2 Ri
R: (16.4)

Obviously, A is countable, reduced, and torsion-free.

(c) It is evident from the definition that A is a left R-module. It is faithful, as
different elements of R act differently on 1 2 A. Consequently, R is isomorphic
to a subring of End A.
(d) In order to show that it is not a proper subring, select an  2 End A. Since R is
pure and dense in A (which is pure and dense in R), Q it follows that AQ D R.
Q By
Proposition 2.10 in Chapter 6,  extends uniquely to a Jp -endomorphism Q of
Q Then
R.

ea D .
Q a 1 C a a/ D a .1/
Q C a .a/
Q D a .1/ C a .a/

for any a 2 A. More explicitly, write

X
n X
n X
n
pk .ea / D b0 C bi eai ; pk .1/ D c0 C ci eai ; pk .a/ D d0 C di eai
iD1 iD1 iD1

for ai ; bi ; ci ; di 2 R, and for some k; n 2 N. Substitution yields

X
n
b0 C bi .ai 1 C ai ai /
iD1
" #
X
n X
n
D a Œc0 C ci .ai 1 C ai ai /
C a d0 C di .ai 1 C ai ai /
iD1 iD1

where we may assume that a1 D a. Comparing the corresponding coefficients

on both sides, we use algebraic independence to argue that b1 D c0 , b1 a D d0 ,
while all other bi ; ci ; di vanish. This means pk .1/ D c0 ; pk .a/ D c0 a which
thus holds for all a 2 R. Therefore, with the notation 1 D c 2 R, we have
a D ca for all a 2 R, showing that  acts on R as left multiplication by c 2 R.
The same holds for Q and for  D Q  A. This completes the proof of the local
case.
3 Endomorphism Rings of Torsion-Free Groups 631

(e) Moving to the global case, suppose R is as stated in the theorem. We get
RQ D Q R Q Q
p .p/ where R.p/ is the p-adic completion of the reduced part of the
localization R.p/ D Z.p/ ˝ R. For a 2 R we write a D .: : : ; ap ; : : : / with
ap 2 R.p/ .
Just as in (b), for each a 2 R choose a ; a now as a D .: : : ; pa ; : : : /; a D
.: : : ; pa ; : : : / with pa ; pa 2 Jp algebraically independent over S.p/ for each p;
note that if R.p/ D 0, then pa D pa D 0 can be chosen. Defining ea as in (16.3)
and A as in (16.4), A becomes a countable subgroup of R. Q As in (c), we argue
that R is isomorphic to a subring of End A. Every  2 End A extends uniquely
to Q 2 End.R Q C / which must act coordinate-wise in each R Q .p/ , because these are
Q
fully invariant subrings in R. By the local case, Q is left multiplication by the
R.p/ -component of 1 D c 2 R, thus Q must agree with the left multiplication
by c on all of R, Q in particular on A. This establishes the claim that End A Š R.
t
u
Since there is a set of cardinality 2@0 of pairwise disjoint finite subsets of
algebraically independent elements of Jp , and since these define non-isomorphic
torsion-free groups in the above construction, it is clear that there are 2@0 non-
isomorphic solutions in Theorem 3.3.
The Topological Version Another point of interest emerges if the endomor-
phism rings are equipped with the finite topology. Then all endomorphism rings
of countable reduced torsion-free groups can be characterized, even if they are
uncountable. Note that the necessity of the condition stated in the next theorem
is immediate: each left ideal L in Theorem 3.4 is defined to consist of all  2 End A
that annihilate a fixed a 2 A. However, the proof of sufficiency involves more ring
theory than we care to get into, and therefore we state the theorem without proof.
Theorem 3.4 (Corner [4]). A topological ring R is isomorphic to End A for a
countable reduced torsion-free group A if and only if it is complete in the topology
with a base of neighborhoods of 0 consisting of left ideals Ln .n < !/ such that the
factor groups R=Ln are countable, reduced and torsion-free. t
u
Quasi-Endomorphism Ring This is a most useful tool in dealing with finite
rank torsion-free groups. The set of quasi-endomorphisms of a torsion-free A (see
Sect. 9 in Chapter 12) is a Q-algebra

Q End A D Q ˝ End A:

The fully invariant pure subgroups of A form a lattice F where G \ H and .G C H/
are the lattice operations .G; H 2 F/.
Lemma 3.5 (Reid [3]). Let A be a torsion-free group. The correspondences

G 7! Q ˝ G and M 7! M \ A
632 16 Endomorphism Rings

are inverse to each other between the lattice F of fully invariant pure subgroups G
and the Q End A-submodules M of Q ˝ A.
Proof. Straightforward. t
u
Isomorphic Endomorphism Rings It seems that it does not make much sense
to pose the question as to when the isomorphy of endomorphism rings of torsion-
free groups implies the isomorphy of the groups themselves. Surprisingly it has an
answer, albeit in a very special case, by the following theorem.
Theorem 3.6 (Sebel’din [1]). Suppose that End A Š End C where A and C are
direct sums of rational groups each of which is p-divisible for almost all primes p.
Then A Š C.
Proof. If A has  summands of type Z.p/ for a prime p, then End A has  orthogonal
primitive idempotents of this type. Hence from End A Š End C it follows that A and
C must have the same numbers of summands of types Z.p/ . Factor out the ideals of
elements of types Z.p/ for all p, and repeat the same argument for types Z.p;q/ for
different primes p; q (i.e., for rational groups r-divisible for all primes r ¤ p; q).
We can then conclude in the same way the equality of the numbers of summands of
these types. If we keep doing this, including more and more primes, then the claim
follows. t
u
Sebel’din points out that this is a sharp result: the hypothesis on A in the
preceding theorem cannot be weakened: if A is of rank 1 and if its type t is
finite at infinitely many primes, then there are non-isomorphic C of rank one with
isomorphic endomorphism ring.
In an another special case, more can be stated:
Theorem 3.7 (Wolfson [2]). Let A; C be homogeneous separable torsion-free
groups of type t.Z/. If W End A ! End C is a ring isomorphism, then there exists
an isomorphism  W A ! C such that ./ D  1 for all  2 End A.
Proof. If  2 End A is a primitive idempotent, then A Š .End A/, and the same
holds for C. Hence the existence of  is immediate. The rest follows as in the proof
of Theorem 2.5. t
u
F Notes. Theorem 3.3 is one of the most significant theorems on endomorphism rings.
It has been generalized, see Theorem 7.2. Corner [2] also proves that for finite rank groups,
Theorem 3.3 has a noteworthy improvement: a reduced torsion-free ring of rank n is isomorphic to
the endomorphism ring of a torsion-free group of rank  2n. This is the sharpest result in general.
Zassenhaus [1] found conditions for a ring of rank n to be the endomorphism ring of a group of the
same rank. For a generalization of this result, see Dugas–Göbel [6].
Theorem 3.3 has been generalized by several authors, see Sect. 7. It was Göbel [1] who
observed that Corner’s theorem should be valid for uncountable groups under suitable assumptions.
The first generalization to arbitrary cardinalities was given by Dugas–Göbel [2] for cotorsion-free
rings under the hypothesis V D L.
There are numerous results on the endomorphism rings of a few selected classes of torsion-free
groups. For example, Dugas–Thomé [2] discuss the Butler version, while the endomorphism rings
of separable torsion-free groups were characterized by Metelli–Salce [1] in the homogeneous case,
3 Endomorphism Rings of Torsion-Free Groups 633

by Webb [1] for homogeneously decomposable groups, and by Bazzoni–Metelli [1] in the general
case. The conditions are similar to those given by Liebert [3] for p-groups. Blagoveshchenskaya
[1] investigates the case of countable torsion-free groups.
Faticoni [Fat] observes and demonstrates that for torsion-free finite rank groups, the endomor-
phism ring modulo the nilradical is more tractable than the ring itself. Chekhlov [3] considers
groups whose idempotent endomorphisms are central, including the separable and cotorsion cases.
For a reduced torsion-free A, Krylov [5] defines the ideal H.A/ of E D End A as the set of all
 2 E for which hp .x/ > hp .x/ for all x 2 A if the latter height is finite (this being the torsion-
free analogue of the ideal denoted by the same symbol in the torsion case). If A is of finite rank,
then the Jacobson radical J of E contains H.A/ and is nilpotent mod it. For more on End, see also
Krylov [6].

Exercises

(1) If A is torsion-free, then p End A D End A exactly if pA D A.

(2) Every reduced torsion-free ring of rank one is the endomorphism ring of a
torsion-free group of rank n 2 N.
(3) (Corner) If End A is countable, reduced and torsion-free, then A must be
reduced and torsion-free.
(4) (Corner) A ring whose additive group is Š Jp ˚ Jp cannot be the endomor-
phism ring of any group.
(5) For an arbitrarily large cardinal , there exist torsion-free groups of cardinality
 whose endomorphism rings have cardinality 2 .
(6) For a finite rank torsion-free A, End A has no divisors of zero if and only if all
endomorphisms are monic if and only if Q End A is a division ring.
(7) The center of the endomorphism ring of a homogeneous separable torsion-free
group is a subring of Q.
(8) (Hauptfleisch) Suppose that End A Š End C where A and C are homogeneous
separable torsion-free groups of types t and s, respectively.
(a) If t D s, then every isomorphism between the endomorphism rings is
induced by an isomorphism between the groups.
(b) If t ¤ s, then A ˝ S Š C ˝ T where T; S are rational groups of type t and
s, respectively.
(9) Is Theorem 3.7 true for any type t?
(10) (J. Reid, Orsatti) Call a ring R subcommutative if for all r; s 2 R there is
t 2 R such that rt D sr.
(a) The ring of the integral quaternions q D aCbiCcjCdk with a; b; c; d 2 Z
is subcommutative.
(b) If End A is subcommutative, then endomorphic images of A are fully
invariant subgroups.
(c) Conclude that the full invariance of endomorphic images does not imply
the commutativity of the endomorphism ring.
634 16 Endomorphism Rings

4 Endomorphism Rings of Special Groups

Our next concern is with the endomorphism rings of some important types of
groups, like projective, injective, etc. groups.
Endomorphism Rings of Free Groups We start with free groups. It might
be helpful to note in advance that if F is a free group, then the image of an
endomorphism  is also a free subgroup, and therefore Ker  is always a summand
of F.
At this point, we need a definition: a ring R is called Baer ring if the left (or,
equivalently, the right) annihilator of a non-empty subset of R is generated by an
idempotent.
Theorem 4.1 (Wolfson [1], Rangaswamy [2], Tsukerman [1]). The endomor-
phism ring of a free group is a Baer ring if and only if the group is countable.
Proof. We show that, for a free group F, End F is a Baer ring if and only if F
enjoys the strong summand intersection property, i.e. intersections of any number
of summands are summands. Then the claim will follow from Proposition 7.12 in
Chapter 3.
Assume F has the strong summand intersection property. Consider the right
annihilator N
End F D E of a subset X  E, and let K D \2X Ker. Then
K is a summand of F by hypothesis, thus K D F for an idempotent  2 E. Hence
 D 0 for all  2 X, moreover, ˛ D 0 for an ˛ 2 E if and only if ˛F
F—
which holds if and only if ˛ D ˛, i.e. ˛ 2 E. Thus N D E, and End F is a Baer
ring.
Conversely, suppose End F D E is a Baer ring, and K D \2X Ker  for some
set X  E, where the summands are written in the form Ker . Define N as the
right annihilator ideal of X, thus N D E with an idempotent . Then K D F is a
summand of F. t
u
Endomorphism Rings of Algebraically Compact Groups Next, we consider
algebraically compact groups. We know from Theorem 2.11 in Chapter 7 that their
endomorphism groups are also algebraically compact, but we can now prove much
more.
Theorem 4.2.
(a) The endomorphism ring of an algebraically compact group is a right alge-
braically compact ring.
(b) The same holds if the group is torsion-complete.
Proof.
(a) We show: if A is algebraically compact, then E D Hom.A; A/ is an algebraically
compact right E-module (the first A being viewed as a left E-module). Let N be
a pure submodule in a right E-module M. It suffices to prove that the group
homomorphism

HomE .M; Hom.A; A// ! HomE .N; Hom.A; A//

4 Endomorphism Rings of Special Groups 635

induced by the inclusion map N ! M is surjective. In view of a well-known

natural isomorphism, this can be rewritten as

Hom.M ˝E A; A/ ! Hom.N ˝E A; A/:

Since N ! M is a pure inclusion as right E-modules, N ˝E A ! M ˝E A is

a pure inclusion group-theoretically. Therefore, by the pure-injectivity of A, the
map in question is in fact surjective.
(b) A reduced torsion group A has the same endomorphism ring as its cotorsion
hull A . For a torsion-complete group, the cotorsion and pure-injective hulls are
identical. t
u
Endomorphism Rings of Divisible Groups We now turn to injective groups;
their endomorphism rings are very special, indeed.
Theorem 4.3. The endomorphism ring E of a divisible group D is a right alge-
braically compact ring whose Jacobson radical J consists of those  2 E for which
Ker  is an essential subgroup in D.
Idempotents mod J lift, and E=J is a von Neumann regular ring.
Proof. By Theorem 4.2, E is an algebraically compact ring. Let  2 E be such
that Ker  is essential in D. Then Ker  is also essential in D for every  2 E.
Since Ker  \ Ker.1  / D 0, we obtain that Ker.1  / D 0, so 1  
is a monomorphism. Evidently, Ker 
Im.1  /, therefore Im.1  / is
an essential divisible subgroup, whence it follows that, for all , 1   is an
automorphism of D, i.e. an invertible element in E. Thus  2 J. Conversely, if
 2 J, then let K be a subgroup in D with K \ Ker  D 0. By the injectivity
of D, there is a  W D ! D such that x D x for all x 2 K. This means that
K
Ker.1  / where 1   2 Aut D (as  2 J). Thus K D 0, and Ker is
essential in D.
If  2 E satisfies 2   2 J, then N D Ker.2  / is essential in D. Now
N \ Ker  D 0, for if a .a 2 N/ is annihilated by , then 0 D .a/ D a.
The subgroup N ˚ Ker  is essential in D, since it contains N: every a 2 N can be
written in the form a D a C .1  /a 2 N ˚ Ker . Thus D D B ˚ C where
N
B; Ker 
C. If  W D ! B denotes the projection with C D 0, then   
annihilates every a D a C .1  /a 2 N. Therefore,    2 J, showing that  lifts
to  mod J.
Assume now that  2 E, and K
D is maximal with respect to the property
K \ Ker  D 0. Then K C Ker 
Ker.  / where  2 E is as above. Hence
Ker.  / is essential in D, and so    2 J. We conclude that E/J is a von
Neumann regular ring. t
u
Although we won’t do it here, with some extra effort we can even prove that the
factor ring E/J is self-injective; see Notes.
Endomorphism Rings of Cotorsion Groups Finally, we prove something
concerning endomorphism groups of cotorsion groups that should not be surprising.
636 16 Endomorphism Rings

Theorem 4.4. The endomorphism group of a cotorsion group is cotorsion.

Proof. If C is an adjusted cotorsion group, then End C Š End tC by Proposi-
tion 1.7(ii). As tC is torsion, End tC is algebraically compact.
In general, let G D C ˚ A ˚ D where C is adjusted cotorsion, A is reduced
torsion-free algebraically compact, and D is divisible. Since homomorphism groups
into algebraically compact groups are algebraically compact, Hom.G; A ˚ D/
is algebraically compact (Theorem 2.11 in Chapter 7). To find the structure of
Hom.G; C/, observe that Hom.D; C/ D 0, thus it remains to check Hom.A; C/.
If B is a basic subgroup of the complete group A, then we have an exact sequence
0 ! B ! A ! ˚ Q ! 0. This implies the exactness of 0 D Hom.˚ Q; C/ !
Hom.A; C/ ! Hom.B; C/ ! Ext.˚ Q; C/ D 0, thus the middle terms are
isomorphic. As B is the direct sum of copies of Z.p/ for various primes p, Hom.B; C/
will be the direct product of copies of C.p/ for various primes p, so cotorsion. t
u
F Notes. In general, endomorphism rings of cotorsion groups need not be cotorsion rings,
but they are in the torsion-free case (Theorem 4.2). For the endomorphism ring of an algebraically
compact group, a result similar to the divisible case can be established: idempotents mod its
Jacobson radical J lift, and E/J is a von Neumann regular self-injective ring. This follows from
a ring-theoretical result by Zimmermann–Zimmermann-Huisgen [Math. Z. 161, 81–93 (1978)]
stating that the radical of a pure-injective ring has this property.

Exercises

(1) (Szélpál) (a) End A is a torsion ring if and only if A is a bounded group.
(b) End A is torsion-free if and only if the reduced part of A is torsion-free, and
pA ¤ A implies AŒp
D 0.
(2) The endomorphism ring of a cotorsion group need not be algebraically compact
as a group.
(3) The endomorphism ring of the additive group of any injective module is
algebraically compact.
(4) (Rangaswamy) For a torsion group A, End A is a Baer ring if and only if, for
every prime p, Ap is either divisible or elementary.

5 Special Endomorphism Rings

So far our investigations relating a group and its endomorphism ring have mainly
been concerned with the question as to how the group structure is reflected in the
endomorphism ring. In this section we approach this problem from the opposite
direction, and the dominant theme will be the kind of influence properties of
endomorphism rings have on the underlying group. The results in this section would
5 Special Endomorphism Rings 637

seem to suggest that specific endomorphism rings are indeed special. Needless to
say, we are primarily interested in conventional ring properties.
It seems sensible at this point to stress that one should not have high expectations
of the interrelations between interesting group and important ring properties. We
keep emphasizing that significant group and ring properties seldom match: groups
whose endomorphism rings are of special interest in ring theory are few and far
between.
We are going to survey several cases of interest.
Elementary Properties We start assembling a few simple observations that will
simplify subsequent arguments. We will abbreviate E D End A.
(a) Suppose ˛ 2 E. If nj˛, then ˛A
nA, and if n˛ D 0, then ˛A
AŒn
for n 2 N.
Indeed, if ˇ 2 E satisfies nˇ D ˛, then ˛A D nˇA
nA, and if n˛ D 0, then
n˛A D 0, so ˛A
AŒn
.
(b) If End A D R1 ˚ R2 is a ring direct sum, then A D B1 ˚ B2 where Ri Š End Bi
and Hom.Bi ; Bj / D 0 for i ¤ j. This is a simple consequence of Sect. 1(D).
(c) If A has a sequence of direct decompositions

A D A0 ˚    ˚ An ˚ Cn with Cn D AnC1 ˚ CnC1 .n < !/

where An ¤ 0 .n < !/, then neither the left nor the right .principal/ ideals
of E satisfy the minimum condition. To verify this, let n W A ! Cn denote
the obvious projection. Then n nC1 D nC1 , but no ˛ 2 E exists satisfying
nC1 ˛ D n , because Im nC1 ˛
Im nC1 < Im n . This establishes the
proper inclusion nC1 E < n E for every n. Also, nC1 n D nC1 , and there is
no ˇ 2 E with ˇnC1 D n , since clearly Ker n < Ker nC1
Ker ˇnC1 :
This shows that EnC1 < En .
(d) If is in the center of E, then Im and Ker are fully invariant subgroups of
A. For each  2 E, we have  .A/ D .A/
Im , and if a 2 Ker , then
.a/ D . .a// D 0 means .a/ 2 Ker .
(e) In a torsion ring .with or without identity/, the divisible subgroup is contained
in the annihilator of the ring. Thus 1 2 R implies R is reduced.
Recall that an element ˛ 2 R is regular (in the sense of von Neumann) if ˛ˇ˛ D
˛ for some ˇ 2 R, and -regular if ˛ m is regular for some m 2 N.
Lemma 5.1 (Rangaswamy [1]). An endomorphism ˛ of A is a .von Neumann/
regular element in End A if and only if both Im ˛ and Ker ˛ are summands of A.
Proof. Assume ˛ 2 E D End A satisfies ˛ˇ˛ D ˛ for some ˇ 2 E. Since ˛ˇ
and ˇ˛ are idempotents, they are projections of A, so their images and kernels are
summands of A. The inclusions Im ˛ˇ˛
Im ˛ˇ
Im ˛ and Ker ˛
Ker ˇ˛

Ker ˛ˇ˛ imply Im ˛ D Im ˛ˇ and Ker ˛ D Ker ˇ˛, so Im ˛; Ker ˛ are summands
of A.
Conversely, suppose Im ˛ D B and Ker ˛ D K are summands of A, say, A D
B˚C D K˚H for some C; H
A. Because of K\H D 0, ˛ maps H isomorphically
638 16 Endomorphism Rings

into, and evidently onto, B. This means, there is a ˇ 2 End A that annihilates C and
is inverse to ˛  H on B. Writing a D g C h 2 K ˚ H, we get .˛ˇ˛/a D .˛ˇ/˛h D
˛h D ˛a for every a 2 A, whence ˛ˇ˛ D ˛ follows. t
u
Simple Endomorphism Rings Our survey of special endomorphism rings starts
with division rings.
Proposition 5.2 (Szele [2]). The endomorphism ring of A is a division ring if and
only if A Š Q or A Š Z.p/ for some prime p.
A division ring is the endomorphism ring of a group if and only if it is the additive
group of a prime field.
Proof. If End A is a division ring, then every non-zero ˛ 2 End A is an automor-
phism, which entails, for every prime p, either pA D A or pA D 0. It also follows
that 0; 1 are the only idempotents in End A. After ruling out Z.p1 /, it is clear that
the only possibility is either A Š Z.p/ for some p, or else A Š Q. t
u
By a simple ring is meant a ring R whose only ideals are 0 and R.
Theorem 5.3. End A is a simple ring if and only if, for some integer n, either A Š
˚n Q or A Š ˚n Z.p/ for a prime p.
A simple ring is the endomorphism ring of a group if and only if it is a complete
matrix ring of finite order over a prime field.
Proof. Sufficiency is obvious from the matrix representation of direct sums.
Conversely, if E D End A is a simple ring, then for every prime p, either pE D 0
or pE D E. In the first alternative, (a) above implies that A
AŒp
, i.e. A is an
elementary p-group. If pE D E for every p, then (a) applied to ˛ D 1 shows that
A is divisible. In this case, the socle of E, if not zero, would be a non-trivial ideal,
therefore E must be torsion-free, and division by every p is an automorphism of
A. Consequently, A is likewise torsion-free divisible. Our conclusion is that either
A Š ˚ Z.p/ or A Š ˚ Q. The endomorphisms mapping A onto a finite rank
subgroup form an ideal in E, by simplicity this must be all of E. Hence A is a
finite direct sum, and we are done, since the second assertion is pretty clear from
our argument. t
u
Artinian Endomorphism Rings The preceding theorem shows that simple
rings can be endomorphism rings only if they are left and right artinian. It is not
difficult to identify all artinian endomorphism rings.
Theorem 5.4.
(i) End A is left .right/ artinian if and only if

ADB˚D

where B is finite and D is torsion-free divisible of finite rank.

(ii) (Szász [1]) The same holds for left .right/ perfect rings.
5 Special Endomorphism Rings 639

Proof.
(i) If E D End A is left (right) artinian, then in the set of ideals fnE j n 2 Ng there
is a minimal one, say, mE; clearly, it must be divisible. (a) above shows that
nmjm1A implies mA
nmA for every n 2 N, i.e. mA is divisible. Consequently,
A D B ˚ D where mB D 0 and D is divisible. From (c) we derive that both B
and D are of finite rank; in particular, B is a finite group. The presence of Z.p1 /
is ruled out, since a ring with 1 cannot have a non-zero annihilator.
For the converse, assume A has the stated decomposition. Then the two
summands are fully invariant, so End A D End B ˚ End D. Hence End B is
finite and End D is a complete matrix ring of finite order over Q, so End A is
artinian.
(ii) The same proof goes through under the milder hypothesis that the principal left
(right) ideals satisfy the minimum condition, i.e. End A is a left (right) perfect
ring. t
u
PID Endomorphism Rings We have seen that the endomorphism rings of
Procházka-Murley groups were PID. Clearly, a group with PID endomorphism
ring is indecomposable. More specific results can be stated about the behavior of
summands with PID endomorphism rings. A typical result is recorded in the next
proposition. We recall that quasi-isomorphic groups whose endomorphism rings are
PID have isomorphic endomorphism rings (cf. Lemma 9.8 in Chapter 12).
Proposition 5.5. Let A D A1 ˚    ˚ An where the summands are quasi-isomorphic
finite rank torsion-free groups and have endomorphism rings isomorphic to the same
principal ideal domain D.
(i) For every endomorphism  of A, Ker  is a summand of A. It is the direct sum of
D-modules quasi-isomorphic to the Ai .
(ii) A pure subgroup of A that is quasi-isomorphic to one .and hence to each/ of Ai ,
is a summand of A.
Proof.
(i) An endomorphism  of A can be represented as an n  n matrix M D kij k
with entries in the field Q of quotients of D. For the proof, we multiply M by a
suitable 0 ¤ 2 D so as to have all the entries contained in D. By hypothesis,
D is a PID, so there exist invertible matrices A D k˛ki k; B D kˇj` k 2 Mn .D/
(the full matrix ring of order n over D) such that AMB D kık` k is a diagonal
matrix: ık` D 0 for k ¤ ` (and ıkk jı`` if k < `). An easy calculation shows that
its diagonal entries are
X
ıkk D ˛ki . ij /ˇjk 2 D .k
n/;
i;j

where the indicated inclusion ıkk 2 D is justified by the fact (as we shall see
in the next paragraph) that the diagonal elements represent endomorphisms of
quasi-isomorphic modules, so by Lemma 9.8 in Chapter 12 they have to be
elements of End Ai D D. (One can also argue that AMB represents a map
640 16 Endomorphism Rings

A ! A which must evidently be times of a map A ! A.) Consequently, the

matrix of  admits a diagonal form with entries in Q.
This fact translated to maps asserts that there are automorphisms ;  of A
(corresponding to A; B) such that  D  is represented by a diagonal matrix.
The zero columns in this matrix correspond to the kernel of ; consequently, the
kernel is of the form Ker  D B1 ˚    ˚ Bm with m
n. Each component Bj is
obtained as a sum of isomorphic copies of some of the components Ai , so it is
evidently quasi-isomorphic to the Ai . Hence for Ker  D Ker D Ker 1 D
.Ker/ we have

Ker  D .B1 ˚    ˚ Bm / D B1 ˚    ˚ Bm :

(ii) Let i W A ! Ai denote the projections in the given direct decomposition, and
let us fix quasi-isomorphisms ˛i W Ai ! C .i D 1; : : : ; n/ where C is a pure
subgroup in A. Evidently, ˛i i  C is an endomorphism of C, say, it is equal to
some i 2 D (Lemma 9.8 in Chapter 12). Let  D gcdf1 ; : : : ; n g be calculated
in D; thus, we have  D 1 1 C    C n n for suitable i 2 D. Then  D
P
i i ˛i i W A ! C is an endomorphism of A whose restriction to C acts like .
We use induction on n to complete the proof. If n D 1, then C
A1 , and
by purity equality holds. Next suppose n > 1. Evidently,  D    is a non-
zero endomorphism of A that annihilates C. Owing to (i), Ker  is a summand
of A, and also a direct sum of groups quasi-isomorphic to the Ai . Since  ¤ 0,
this summand has less than n components, so induction hypothesis applies. The
claim that C is a summand follows at once. t
u
Noetherian Endomorphism Rings It is a trivial observation that if End A
is noetherian, then A decomposes into the direct sum of a finite number of
indecomposable groups. It seems difficult to say much more in general about
groups with noetherian endomorphism rings: just consider arbitrarily large torsion-
free groups whose endomorphism rings are Š Z. The only hope is to put aside
torsion-free groups, and concentrate on torsion groups. Luckily, we have a complete
description in this case.
Theorem 5.6. Suppose A is a torsion group. End A is left .or right/ noetherian
exactly if A is finitely cogenerated.
Proof. If End A is noetherian, then A is a finite direct sum of indecomposable (i.e.
cocylic) groups, so it is finitely cogenerated. Conversely, if A is finitely cogenerated,
then by Theorem 5.3 in Chapter 4, A D B ˚ D with finite B and finite rank divisible
D. In the additive decomposition End A D Hom.B; D/ ˚ End D, the first summand
is finite, while the second summand is a finite direct sum of complete matrix rings
over the p-adic integers, for various primes. Thus End A is a finite extension of a
two-sided noetherian ring. t
u
Notice a kind of duality: a torsion group A has the maximum (minimum)
condition of subgroups if and only if End A has the minimum (maximum) condition
on left (or right) ideals.
5 Special Endomorphism Rings 641

Regular Endomorphism Rings Our next project in this section is concerned

with (von Neumann) regular rings.
Example 5.7.
(a) The endomorphism ring of an elementary group is regular. This is obvious from the fact that
every subgroup isQ
a summand; see Lemma 5.1.
(b) A direct product p Ap of elementary p-groups Ap , for different
Q primes p, has also regular
endomorphism ring. This follows from the isomorphism End. p Ap / Š End.˚p Ap /.
Example 5.8. Torsion-free divisible groups have regular endomorphism rings.

Theorem 5.9. If End A is a von Neumann regular ring, then A D D ˚ G where

(i) D is torsion-free divisible; it is 0 if G is notQ
torsion;
(ii) G is a pure subgroup between ˚p2P Tp and p2P Tp , where Tp is an elementary
p-group, and P is a set of primes.
Proof. If End A is a regular ring, then to every ˛ 2 End A there is a ˇ 2 End A
such that ˛ˇ˛ D ˛. Specifically, if ˛ is multiplication by p, then pˇp D p, which
implies that A contains no elements of order p2 , so Ap , the p-component of A, is an
elementary group. Since also p D pˇpˇp D pˇpˇpˇp D : : : ; it is clear that, for
every a 2 A, hp .pa/ D 1. Therefore, pG is p-divisible. Write A D D ˚ G with D
divisible and G reduced; at this point we already know that D is torsion-free, and
T D tG is an elementary group. We have G D Tp ˚ G.p/ with p-divisible G.p/, and
it follows that G.p/ D pG. As G=Tp is p-divisible, G=T is divisible.
The intersection \p pG is clearly a torsion-free subgroup of G. For every a 2 G,
pa belongs to every Gq (for every prime q), and is divisible in Gq by every pk .k  2/
uniquely. This means \p pG is divisible, and hence 0. Thus the intersection of the
kernels of the projections p W G Q ! Tp is 0, and consequently, G is isomorphic to
a subgroup of the direct product p2P Tp , containing T D ˚p2P Tp with divisible
G=T.
If G is not torsion and D ¤ 0, then A has an endomorphism mapping G ! D non-
trivially whose kernel contains T, but is not a summand (G has no proper summand
containing T). This is in violation to Lemma 5.1. t
u
Semi-Local Endomorphism Rings Finally, we consider local and semi-local
endomorphism rings. Recall that a ring R is said to be semi-local if R=J is a semi-
simple artinian ring; as usual, here J denotes the Jacobson radical of R. R is local
if R=J is a division ring.
Lemma 5.10 (Orsatti [1]). A torsion group has local endomorphism ring if and
only if it is cocyclic.
Proof. A local ring has only two idempotents: 0 and 1, therefore a group with local
endomorphism ring is indecomposable. A torsion group is indecomposable if and
only if it is cocyclic. Since End Z.pn / Š Z=pn Z and End Z.p1 / Š Jp are local
rings, the claim is evident. t
u
642 16 Endomorphism Rings

For the semi-local case we prove:

Theorem 5.11 (Cǎlugǎreanu [1]). A group A has semi-local endomorphism ring if
and only if A D T ˚ G where T is a finitely cogenerated group, and G is torsion-free
with semi-local End G.
In case G has finite rank, End G is semi-local exactly if pG D G for almost all
primes p.
Proof. Assume End A is semi-local. A semi-local ring has but a finite number of
orthogonal idempotents, hence A is a finite direct sum of indecomposable groups.
This must be true for tA D T as well, whence A D T ˚ G where T and G are
as stated (because summands inherit semi-local endomorphism rings). If pG ¤ G,
then also p End G ¤ End G, and there is at least one maximal left ideal Lp between
p End G and End G. Since Lp ¤ Lq if p ¤ q are primes, and since in a semi-local
ring the Jacobson radical is the intersection of finitely many maximal left ideals,
pG < G can hold only for a finite number of primes.
Conversely, if A is as stated, then both T and G have semi-local endomorphism
rings, and it is easy to check that the same is true for A (the radical of End A is then
the direct sum of the radicals of End T and End G). If G is of finite rank, then so
is End A, and because End A=p End A is not 0 but for finitely many primes p, and
then it is finite, the same must hold for .End A/=J. Hence J is a finite intersection of
maximal left ideals, and .End A/=J is a finite direct sum of simple artinian rings. u t
F Notes. It was Szele who initiated a systematic study of groups whose endomorphism
rings belong to a class of rings of interest, and since then this has been a recurrent theme in
the literature. While the theorems above convince us that some familiar ring properties of the
endomorphism ring might impose severe restrictions on the groups, it is unlikely that classes of
groups whose endomorphism rings share some other prominent ring properties, like commutativity,
hereditariness etc. will admit a satisfactory description in the near future, since these do not impose
much restriction on the groups. There exists an extensive literature on special endomorphism rings
of restricted classes of groups, many of these results are technical or involve more advanced ring
theory.
For more results on cases when End A is regular, or when its principal right ideals are projective,
see Glaz–Wickless [1]. Karpenko–Misyakov [1] describe the groups whose endomorphism rings
have regular center. Mader [5] investigates the maximal regular ideal in endomorphism rings.
Groups with -regular endomorphism rings were studied in Fuchs–Rangaswamy [1]; the results
are similar to Example 5.7. See F. Kasch–A. Mader, Regularity and substructures of Hom (2009)
where regular homomorphisms were studied.
Misyakova [1] examines the case when the endomorphism ring is semi-prime (i.e., the
intersection of prime ideals is 0). Salce–Menegazzo [1] investigate groups with linearly compact
endomorphism rings. Several papers are devoted to self-injective endomorphism rings, see
Albrecht [2], Ivanov [6], Rangaswamy [3].

Exercises

(1) The endomorphism ring of a separable torsion-free group A is left (or right)
noetherian if and only if A is completely decomposable of finite rank.
6 Groups as Modules Over Their Endomorphism Rings 643

(2) Find groups in which every endomorphism is either an automorphism or

nilpotent.
(3) (Szele–Szendrei)
(a) If End A is commutative, then the p-components Ap are cocyclic, and A=t.A/
is p-divisible for every p with Ap ¤ 0.
(b) A splitting mixed group A has commutative endomorphism ring exactly
if both End t.A/ and End.A=t.A// are commutative, and A satisfies the
condition in (a).
Q
(4) The endomorphism ring of A D ˚p Z.p/ ˚ p Z.p/ is not regular.
(5) (Rangaswamy) Kernels and images of all the endomorphisms of A are pure
subgroups in A if and only if t.A/ is elementary and A=t.A/ is divisible.
(6) (Rangaswamy) If the kernel of an  2 End A is a summand of A, then the right
annihilator of  is a projective right ideal of End A. [Hint: if  W A ! Ker  is
projection, then Ann  D  End A.]
(7) (Cǎlugǎreanu) A divisible group has semi-local endomorphism ring exactly if
it is of finite rank.
(8) If A is a p-group and End A is Dedekind-finite (i.e. ˛ˇ D 1 .˛; ˇ 2 End A/
implies ˇ˛ D 1), then the UK-invariants of A are finite.

6 Groups as Modules Over Their Endomorphism Rings

Since every group A is a left module over its own endomorphism ring E D End A, it
is tempting to strive for a better understanding as to how A behaves as an E-module.
A large number of special cases have already been investigated (mostly restricted to
subcategories of Ab), but so far no systematic study is available. Here we collect a
few special cases of interest in order to give a flavor of this interesting topic.
Preliminaries A few trivial facts to keep in mind:
(A) A group A is a faithful E-module (i.e. no non-zero element of A is annihilated
by E).
(B) The E-submodules are exactly the fully invariant subgroups.
(C) If  W A ! N is an E-homomorphism into an E-module N, then Ker  is a fully
invariant subgroup of A.
Our starting point is a brief analysis of the E-homomorphisms into and from A.
The basic facts for the primary case are summarized in the following lemma.
Lemma 6.1. Let A be a p-group with a fully invariant subgroup H.
(i) An endomorphism is an E-map if and only if it is multiplication by a p-adic
integer.
(ii) If A is either bounded or has an unbounded basic subgroup, then a homomor-
phism ˛ W A ! A=H is an E-map if and only if it is the canonical homomorphism
followed by multiplication by a p-adic integer.
644 16 Endomorphism Rings

Proof. The ‘if’ parts of both claims are obvious, so we proceed to prove necessity.
Let A D C ˚ U be a decomposition of A where C is a cocyclic group. If C is cyclic
of order pn , then let c 2 C denote a generator, and if C Š Z.p1 /, then pick any
non-zero element c 2 C, say, of order pn . Choose an  2 E so as to satisfy:

 W c 7! c C x; u 7! tu .8 u 2 U/ (16.5)

where t is any integer; x 2 AŒpn

is chosen arbitrarily if C is cyclic, while if C is
quasi-cyclic, then x can be any element of order
pn in the divisible subgroup of A.
(i) We need a brief calculation. Suppose  2 E is an E-map, and write .c/ D
kc C v 2 A with k 2 Z; v 2 U. Then

.c/ D kc C kx C tv and .c/ D kc C v C .x/:

The equality  D  holds for all possible choices of t and x; this yields v D 0
and .x/ D kx. Hence we conclude that if C is cyclic, then  acts on AŒpn
as
multiplication by k. If C is quasi-cyclic, then the same can be said only about the
action of  in the divisible subgroup of H. The integer k depends on the order
of c, but for different elements the numbers must match, so it follows by usual
arguments that there is a  2 Jp such that  acts as multiplication by . In case
A is bounded by pn , then only multiplications mod pn need to be considered.
(ii) Assume A is as stated above in (ii), so it has a cyclic summand C D hci of
order pn with maximal n or with arbitrarily large n. If A D C ˚ U, then clearly,
A=H D C=.H \ C/ ˚ U=.H \ U/. For an E-map ˛ W A ! A=H, we set
˛c D kNc C vN with k 2 Z; v 2 U where bars indicate cosets mod H. If  is as
above in (16.5), then Nc D cN C xN and Nu D tNu. We then have

˛.c/ D .kNc C v/
N D kNc C kNx C tv;
N ˛.c/ D ˛.c C x/ D kNc C vN C ˛x:

These have to be equal for all permissible choices of x and t, whence vN D 0;

˛c D kNc, and ˛x D kNx. Hence ˛ acts as multiplication by the integer k on AŒpn

mod H, in particular, on hNci Š hci=.H \ hci/. As in (i), we argue that these

numbers k define a  2 Jp such that ˛x D Nx for all x 2 A. Consequently,
˛ D  where  W A ! A=H is the canonical map.
(From Theorem 6.11 below it will follow that (ii) fails for p-groups of the
form B ˚ D with bounded B and divisible D ¤ 0.) t
u
From the preceding lemma it is obvious that the ring of E-endomorphisms of a
p-group is isomorphic to the center of E.
To formulate the precise statements in the following theorems, we will need new
definitions. To simplify, following [KMT], we adopt an easy terminology: a group
will be called endo-P if it has property P as a left E-module.
Endo-Finitely-Generated Groups First, we exhibit a few examples for endo-
finitely generated groups. Some are even endo-cyclic.
6 Groups as Modules Over Their Endomorphism Rings 645

Example 6.2. An immediate example for an endo-finitely-generated group is any bounded group
B. If b 2 B is an element of maximal order, then B is generated by b over E D End B. Thus B D Eb
is endo-cyclic.
Example 6.3. A torsion-free divisible group is also endo-cyclic. The direct sum of a torsion-free
divisible group and a bounded group is endo-finitely generated.
Example 6.4. An E-group A (see Sect. 6 in Chapter 18 for definition) is endo-cyclic: if a 2 A
corresponds to 1 2 E, then A D Ea.

Endo-finitely generated torsion groups admit a full characterization.

Proposition 6.5. A torsion group is endo-finitely-generated if and only if it is
bounded.
Proof. As pointed out in Example 6.2, a bounded group is endo-cyclic. On the other
hand, if A is unbounded torsion, then a finite set can generate over End A only a
bounded subgroup of A. t
u
Endo-Artinian and Endo-Noetherian Groups It is not difficult to characterize
the groups that are artinian over their endomorphism rings; however, only little has
been established for the noetherian case.
Theorem 6.6 (Krylov–Mikhalev–Tuganbaev[KMT]).
(i) A group A is endo-artinian if and only if A D B˚D where B is a bounded group,
and D is a divisible group with a finite number of non-zero p-components.
(ii) A is endo-noetherian exactly if A D B ˚ C where B is a bounded group, and C
is torsion-free endo-noetherian.
Proof.
(i) Assume A is endo-artinian. Then there is a minimal subgroup among the
subgroups of the form nA .n 2 N/, say, mA is minimal. This mA is divisible,
so A D B ˚ mA with mB D 0. Clearly, mA can have but a finite number of
p-components ¤ 0, because these are E-submodules.
Conversely, if A is of the form stated in the theorem, then B has only a finite
number of fully invariant subgroups. The torsion-free part of D has no proper
fully invariant subgroup ¤ 0, while the fully invariant proper subgroups of
a p-component Dp of D are of the form Dp Œpk
, so they satisfy the minimum
condition.
(ii) Supposing A is endo-noetherian, there is a maximal subgroup in the set of
subgroups of the form AŒn
.n 2 N/. This must coincide with the torsion
subgroup tA of A, thus A D AŒm
˚ C for some m 2 N and torsion-free C.
Such a C must be endo-noetherian, because the fully invariant subgroups of A
containing AŒm
correspond to those of C.
For the converse, note that all fully invariant subgroups of A D B ˚ C are
among the direct sums of fully invariant subgroups of B and C. t
u
Endo-noetherian torsion-free groups are abundant, and it seems impossible to
characterize them reasonably.
646 16 Endomorphism Rings

Example 6.7. The strongly irreducible torsion-free groups of finite rank are endo-noetherian, since
their non-zero fully invariant subgroups have finite indices.

Endo-Projective Groups As far as endo-projective groups are concerned, only

the torsion groups have a satisfactory characterization, the general case seems to be
out of reach at this time. It is an easy exercise to see that a torsion group is endo-
projective exactly if its p-components are, thus it suffices to deal with p-groups.
Theorem 6.8 (Richman–Walker [1]). A p-group is endo-projective if and only if
it is bounded.
Proof. Suppose the p-group A is E-projective. If A is bounded, then by
Lemma 2.4(i), A Š E with an idempotent , thus A is a summand of the free
module E.
If A is unbounded, then we concentrate on the group homomorphism

 W HomE .A; E/ ˝E A ! HomE .A; A/

acting as . ˝ a/.a0 / 7! .a0 /.a/, where  2 HomE .A; E/ and a; a0 2 A. Since A is

a p-group, HomE .A; E/ ˝E A is also a p-group, while HomE .A; A/ is the center of
E that is isomorphic to Jp (cf. Theorem 2.9). Thus  D 0. It is clear that .a0 /.a/ D
0 for all a; a0 2 A only if HomE .A; E/ D 0. But if A is E-projective, then A is
a summand of a direct sum of copies of E, so HomE .A; E/ ¤ 0. Therefore, no
unbounded A can be endo-projective. t
u
Example 6.9. Examples of endo-projective torsion-free groups are Z and Q. Also, Q ˚ Q is endo-
projective as it is isomorphic to E for an idempotent .

Endo-Projective Dimension It is natural to inquire about the endo-projective

dimension of groups. We will denote by p.d.A the E-projective dimension of A.
Theorem 6.10 (Douglas–Farahat [1]).
(i) A torsion group is either endo-projective or has endo-projective dimension 1.
(ii) The same holds for a divisible group.
Proof.
(i) It suffices to deal with p-groups A. Let n denote the projection of A onto
a summand Cn Š Z.pn /. Then AŒpn
is E-projective, being isomorphic to
Hom.Cn ; A/ Š En . Thus p.d.A D 0 if A is bounded.
Next suppose A has an unbounded basic subgroup. Then A is the union of
a countable ascending chain of E-projective subgroups AŒpn
for integers n for
which A has cyclic summands of order pn . A well-known lemma of Auslander
states that if the groups in such a chain have projective dimensions
m, then
the projective dimension of the union is
m C 1; in our case, m D 0. Thus
p.d.A
1 in this case.
Finally, let A D B ˚ D where B is pn -bounded, and D ¤ 0 is divisible; we
may assume B ¤ 0, since the divisible case will be settled in (ii). Consider the
submodules Ak D AŒpnCk
for k < ! whose union is A. For a cyclic summand C
6 Groups as Modules Over Their Endomorphism Rings 647

of order pn , we have A0 Š Hom.C; A/ as E-modules, so p.d.A0 D 0. If we prove

that p.d..Ak =Ak1 / D 1, then another form of Auslander’s lemma on projective
dimensions will assure that p.d.A cannot exceed 1. It is immediately seen that
the factor group Ak =Ak1 is generated over E by any element d 2 D of order
pnCk , so it is isomorphic to E=L where L D f 2 E j d 2 Ak1 g is a left
ideal of E. Write A D D0 ˚ A0 where d 2 D0 Š Z.p1 /, and define  2 E as
multiplication by p on D0 and the identity on A0 . It follows that L D E Š E,
so L is E-projective. Hence p.d..Ak =Ak1 / D p.d..E=L/ D 1, and p.d.A
1.
(ii) If D is a divisible group that is not torsion, then D Š E where  denotes the
projection onto a summand Š Q, since D D Ed for any d 2 D of infinite order.
In this case, D is evidently endo-projective.
If D is a divisible p-group, then form a projective resolution of Z.p1 / as a
Jp -module: 0 ! H ! F ! Z.p1 / ! 0 where F; H are free Jp -modules
(submodules of free are free as Jp is a PID). Now E D End D is a (torsion-free,
hence) flat Jp -module, so the tensored sequence

0 ! E ˝Jp H ! E ˝Jp F ! E ˝Jp Z.p1 / D D ! 0

is exact. Moreover, it is exact even as an E-sequence. The first two tensor

products are free E-modules, whence p.d.D
1 is the consequence of a
Kaplansky inequality for projective dimensions in an exact sequence. t
u
In contrast, the endo-projective dimension of a torsion-free group can be any
integer or 1; see the Notes.
Endo-Quasi-Projective Groups Turning our attention to the quasi-projective
case, we can prove the following characterization for p-groups.
Theorem 6.11 (Fuchs [19]). A p-group is endo-quasi-projective if and only if it is
bounded or has an unbounded basic subgroup.
Proof. Evidently, a group A is quasi-projective over E if and only if, for each fully
invariant subgroup H of A, every E-homomorphism ˛ W A ! A=H factors as ˛ D 
where  W A ! A=H is the canonical map, and  W A ! A is a suitable E-map, i.e.
multiplication by some  2 Jp .
To prove necessity, we have to rule out the case when A D B ˚ D where pm B D 0
for some m 2 N and D ¤ 0 is divisible. Let  W A ! A=H be the canonical map
with H D AŒpm
. Choose ˛ W A ! A=H so as to satisfy ˛B D 0 and ˛  D W D !
D=.DŒpm
/ an isomorphism; clearly, ˛ is an E-map. There is no  2 Jp such that
˛ D , since ˛DŒpm
¤ 0, but DŒpm
D 0 for all  2 Jp . Consequently, A must
be as stated.
For sufficiency, note that a bounded group is by Theorem 6.8 endo-projective,
while for groups with unbounded basic subgroups, an appeal to Lemma 6.1(ii)
completes the proof. t
u
Endo-Flat Groups In order to describe the endo-flat torsion groups, we may
again restrict our consideration to p-groups.
648 16 Endomorphism Rings

Theorem 6.12 (Richman–Walker [1]). A p-group is endo-flat if and only if it is

bounded or has an unbounded basic subgroup.
Proof. If A is bounded, then it is E-projective, so a fortiori E-flat. It is shown in
Theorem 2.4 that if A is a p-group with an unbounded basic subgroup, then AŒpn
is
E-projective (for any n 2 N if there is a primitive idempotent of order pn ). Hence A
is E-flat as the direct limit of E-projective groups.
It remains to show that A D B ˚ D is not endo-flat if B is bounded and D ¤ 0
is divisible. If pn B D 0 (with smallest n), then E D End A is additively the direct
sum of a pn -bounded group and a torsion-free group, and the same must hold for E-
projective modules. Direct limits of groups that are direct sums of pn -bounded and
torsion-free groups are again of the same kind. Therefore, if D ¤ 0, then A cannot
be flat. t
u
We find it rather surprising that a torsion group is endo-flat if and only if it is
endo-quasi-projective.
Example 6.13. Let A be a torsion-free group whose endomorphism ring is a PID (e.g. any rigid
group). Then A is endo-flat, because it is torsion-free as an E-module. For more examples see
Faticoni–Goeters [1].

Endo-Injective Groups We do not wish to discuss endo-injectivity in general,

we only wish to prove that finite groups share this property. We will then state the
general result without proof (in order to avoid the search for additional structural
information).
Proposition 6.14. Finite groups are endo-injective.
Proof. We prove that a finite p-group F is endo-injective. For convenience, let us
abbreviate Hom.F; Z.p1 // D F ı : If F is viewed as a left E-module, then module
theory tells us that F ı becomes a right E-module, and F ıı a left E-module. For finite
F, we have the canonical E-isomorphism F ! F ıı . Since the action of E on F on
the left is the same as the action of E on F ı on the right, from Theorem 6.8 we infer
that F ı is a projective right E-module. We appeal to the natural isomorphism

Ext1E .C; HomZ .B; Q=Z// Š HomZ .TorE1 .B; C/; Q=Z/

that holds if C is a left and B is a right E-module (see Cartan–Eilenberg [CE]).

Choose B D F ı , then its projectivity implies that the right-hand side is 0, so the
same holds for the Ext on the left for each C. Hence HomZ .F ı ; Q=Z// D F ıı Š F
is an injective left E-module. t
u
The main result on endo-injectivity is the following theorem.
Proposition 6.15 (Richman–Walker [3]). A group A is endo-injective if and only
if it is of the form
Q
A D p Ap ˚ D
6 Groups as Modules Over Their Endomorphism Rings 649

where each Ap is a finite p-group, D is a divisible group of finite rank, and either
(i) D D 0; or
(ii) D is not a mixed group, and almost all Ap D 0. t
u
F Notes. There is an extensive literature on groups (and more generally, on modules) that
are of special kind over their own endomorphism rings. Significant advances have been made, and
the proofs are getting more and more involved in ring theory. This area is certainly a gold mine
for research problems. More on this subject may be found in the monograph Krylov–Mikhalev–
Tuganbaev [KMT].
Some sample results from the rich selection:
(i) Reid [5, 6] has an in-depth study of the endo-finitely generated case.
(ii) Niedzwecki–Reid [1] described cyclic endo-projective groups A. Their result states that
A D RC ˚ M where R is an E-ring and M is an E-module, i.e. an R-module with
HomR .R; M/ D HomZ .R; M/.
(iii) See Arnold–Pierce–Reid–Vinsonhaler–Wickless [1] for more on endo-projectivity. They
show that a torsion-free group is endo-projective whenever it is endo-quasi-projective, and
an endo-projective group is 2-generated.
(iv) For arbitrary torsion-free groups, endo-quasi-projectivity was investigated by Vinsonhaler–
Wickless [1], Vinsonhaler [1], Bowman–Rangaswamy [1].
(v) A is endo-flat exactly if for every finitely generated left ideal L of End A, the canonical
map L ˝ A ! LA is an isomorphism (Goeters–Reid [1]). Cf. also Albrecht–Faticoni [1],
Albrecht–Goeters–Wickless [1].
(vi) A completely decomposable group is endo-flat exactly if in the set of types of summands,
two types with upper bound have also a lower bound (Richman–Walker [6]). This result is
generalized by Rangaswamy [5] to separable groups.
(vii) Detachable p-groups are endo-quasi-injective (Richman [4]). Here ‘detachable’ means that
every element a in the socle is contained in a summand C such that p CŒp
D hai for
 D hp .a/. (Separable and totally projective p-groups are detachable.) For more endo-
quasi-injective groups, see Poole–Reid [1].
(viii) Endo-torsion and other module properties were discussed by Faticoni [3].
(ix) A torsion-free A is endo-noetherian if and only if it is quasi-isomorphic to a finite direct
sum of strongly irreducible groups (Paras [1]).
(x) Endo-uniserial groups (the fully invariant subgroups form a chain) have been investigated
by Hausen [7], [8], and Dugas–Hausen [1].
(xi) Endomorphism rings with chain conditions were studied by Albrecht [1].
(xii) The question when a group is endo-slender was considered by Huber [2], Mader [4],
Eda [5].
Q Eda proved that a necessary and sufficient condition for a reduced vector group
A D i2I Ri to be endo-slender is that for each infinite subset J of I, there exists an index
i 2 I such that the set fj 2 J j t.Rj /  t.Ri /g is infinite.
(xiii) Self-injective endomorphism rings are subjects of several of papers, see e.g. Rangaswamy
[3], Ivanov [6], Albrecht [2].
(xiv) Goldsmith–Vámos [1] deal with the so-called clean endomorphism rings.
After it has been observed that injective modules over any ring have algebraically compact
additive groups, Richman–Walker [3] characterized the algebraically compact groups that are
injective over the endomorphism ring. Vinsonhaler–Wickless [2] consider the endo-injective hull
of separable p-groups: they are reduced algebraically compact groups, whose p-basic subgroups
can be calculated.
Projective dimensions over the endomorphism ring were studied by Douglas–Farahat [1]. Inter
alia they prove that, for any group A and for its powers AI , the endo-projective dimensions are
equal. That the endo-projective dimension of a finite rank torsion-free group can be any integer or
1 was shown by Bobylev [1] for countable rank torsion-free groups. In the finite rank case, the
same was proved for finite dimensions by Angad-Gaur [1], employing Corner type constructions.
650 16 Endomorphism Rings

Vinsonhaler–Wickless [3] discussed the flat dimension of completely decomposable torsion-free

groups, and proved that it can be any integer.
Wickless [2] studied the torsion subgroup as an endo-submodule. Properties of
Hom.C; A/; Ext.C; A/; A ˝ C; Tor.A; C/, viewed as modules over End A or End C, are popular
subjects in the Russian literature. This is an interesting area of research which already produced
several noteworthy results.

Exercises

(1) If A is endo-finite, and H is a fully invariant subgroup of A, then A=H is also

endo-finite. [Hint: cosets of an E-generating set for A.]
(2) Subgroups and finite direct sums of endo-artinian groups are likewise endo-
artinian.
(3) Finite direct sums and summands of endo-noetherian groups are again endo-
noetherian.
(4) Find the groups that are both endo-artinian and endo-noetherian.
(5) For any group C, the group A D Z ˚ C is endo-cyclic and endo-projective.
(6) (Niedzwecki–Reid) If A is generated by n elements over End A, then An is
cyclic over End.An /.
(7) There are torsion-free groups A of arbitrarily large cardinality  such that—as
End A-modules—they cannot be generated by less than  elements.
(8) (Wu–Jans) If A is endo-quasi-projective, then A=H is also endo-quasi-pro-
jective for any fully invariant subgroup H.
(9) (Poole–Reid) Let A be a direct sum of fully invariant subgroups Ai . Then A is
endo-quasi-injective if and only if each Ai is endo-quasi-injective.
(10) (a) (Brameret, Feigelstock) A p-group is endo-uniserial if and only if it is
divisible or a direct sum of copies of Z.pn / and Z.pnC1 / for some n 2 N.
(b) (Hausen) A torsion-free endo-uniserial group of finite rank is p-local for
some p, and every non-zero fully invariant subgroup is of finite index.
(11) Find groups, other than finite groups, that are both endo-projective and endo-
injective.

7 Groups with Prescribed Endomorphism Rings

It is natural to ask the question: if we are given a ring, under what conditions is
it an endomorphism ring? And if it is an endomorphism ring, how can we find
groups to fit? These questions have been answered for separable p-groups partially
by Pierce [1], and completely by Liebert [3]; the conditions are quite complicated
(see Theorem 2.8), and so are the constructions.
We now consider the questions in general. For torsion-free groups, we will have
a mild sufficient condition on the ring to satisfy to be an endomorphism ring, but
7 Groups with Prescribed Endomorphism Rings 651

the real challenge is to construct groups once the target ring has been selected.
Torsion-free groups of totally different, even arbitrarily large cardinalities may have
isomorphic endomorphism rings. We will return to torsion groups, but prefer to
think of the endomorphism rings in the form of Theorem 2.2, and our aim will be
to find groups once the ring R is prescribed. Unfortunately, for mixed groups, only
very little can be said.
The Torsion-Free Case Corner’s Theorem 3.3 provides a satisfactory condition
for countable rings. Of course, one prefers conditions that are applicable in arbitrary
cardinalities. It turns out that the study of this problem in full generality requires
sophisticated machinery that would go beyond the scope of this volume. Therefore,
we just state the theorems without elaborating the proofs. Before doing so, let us
recall that the endomorphism ring a group of cardinality  has cardinality at most 2 .
For cotorsion-free rings the result is most satisfactory.
Theorem 7.1 (Dugas–Göbel [2], Shelah [4]). Let R denote a cotorsion-free ring,
and  a cardinal satisfying  @0 D  and  C  jRj. There exists a group A of
cardinality  C such that

End A Š R:

Moreover, there is a rigid system of 2 groups with this property. t

u
There are various proofs of this theorem, all rely on some set-theoretical
principle, like the Black Box or the Shelah elevator. The group A is constructed
as a sandwich group between the direct sum B of copies of RC and its Z-adic
completion B. Q In this way, we could get too many endomorphisms, and the essence
of the construction is to find out how to get rid of unwanted maps. This is a difficult
and tiring technical process—and additional principles are needed to accomplish
this task. For details, we refer to the excellent expositions Eklof–Mekler [EM],
Göbel–Trlifaj [GT], or the original papers.
In the publications, slightly different conditions are stated for the cardinals
involved. The main issue in the preceding theorem is that, under fairly general
conditions, there exist arbitrarily large groups having the given ring as their
endomorphism ring. Interestingly, for small cardinals a different approach seems
to be necessary.
We should mention a topological version that is actually stronger than The-
orem 7.1 inasmuch as it gives an explicit necessary and sufficient condition
(recall that all endomorphism rings are complete in the finite topology). It extends
Theorem 3.4 to arbitrary cardinalities.
Theorem 7.2 (Dugas–Göbel [5]). A topological ring R is isomorphic to the finitely
topologized endomorphism ring of a cotorsion-free group if and only if R is
complete and Hausdorff in a topology that admits a basis of neighborhoods of 0
consisting of right ideals N with cotorsion-free factor groups R=N. t
u
652 16 Endomorphism Rings

The Torsion Case The problem for torsion groups is a different ball game.
The endomorphism rings are extremely restricted: they are built on algebraically
compact groups, and admit unavoidable small endomorphisms. The big problem
is that these small endomorphisms form an ideal that already totally characterizes
the group itself. Thus the only hope is to have some control on the endomorphism
rings modulo small endomorphisms. We know from Theorem 2.2 that End A is a
split extension of the ideal Ends A by an algebraically compact ring that faithfully
reflects direct decompositions with unbounded summands. We have something to
say about this ring.
Theorem 7.3 (Corner [5], Dugas–Göbel [3]). Let R denote a ring whose additive
group is the p-adic completion of a free group. There exists a separable p-group A
such that

End A Š R ˚ Ends A: t
u

Moreover, if the ring R is given as stated, then for every infinite cardinal , there
exists a family fA j  < g of separable p-groups such that for each , End A Š
R ˚ Ends A , and Hom.A ; A / D Homs .A ; A / whenever  ¤ .
Reduced Mixed Groups Corner–Göbel [1] gave a unified treatment of con-
structing groups from prescribed endomorphism rings that included mixed groups
as well. On a different vein, Franzen–Goldsmith constructed a functor from the
category of reduced torsion-free groups to the category of reduced mixed group to
transfer results from one category to the other. They proved:
Theorem 7.4 (Franzen–Goldsmith [1]). For every reduced countable torsion-free
ring R there exists a reduced countable mixed group A such that A=tA is divisible,
and

End A Š R ˚ Hom.A; tA/: t

u

F Notes. Starting with Corner’s epoch making Theorem 3.3, a significant amount of work
has been done on constructing torsion-free groups, even for arbitrarily large cardinalities, with
prescribed endomorphism rings. Using a consequence of V = L, Dugas–Göbel [2] proved that
every cotorsion-free ring is an endomorphism ring. For the general version, see Shelah [4]. The
condition of cotorsion-freeness appears in almost all of the publications.
The consequences of Theorem 7.2 are legion, especially in constructing fascinating examples.
It seems easier to construct a ring with required properties than large groups satisfying prescribed
conditions. We have already taken advantage of this procedure.
May [5] considers cases when the rings are not necessarily cotorsion-free. Arnold–Vinsonhaler
[3] proved that every reduced ring R whose additive group is a finite rank Butler group is the
endomorphism ring of a finite rank Butler group provided that pR ¤ R holds for at least five
primes p. Another kind of realization theorem is due to Dugas–Irwin–Khabbaz [1] who considered
subrings of the Baer-Specker ring P D Z@0 as split extensions of the ideal of finite rank
endomorphisms as endomorphism rings of pure subgroups of P.
7 Groups with Prescribed Endomorphism Rings 653

Exercises

(1) Prove that there exists no group whose endomorphism ring is isomorphic to
Q ˚ Q or to Z=pZ ˚ Z=pZ or to Jp ˚ Jp for any prime p.
(2) Prove the existence of a mixed group with divisible torsion-free part such that
its endomorphism ring has a homomorphic image of a ring with countable free
additive group. [Hint: Theorem 7.4.]

Problems to Chapter 16

PROBLEM 16.1. Is the anti-isomorphic ring of an endomorphism ring also an

endomorphism ring?
PROBLEM 16.2. Find the groups such that every endomorphism is the sum of
(two) monomorphisms.
There are several publications on the unit sum number, see Notes to Sect. 2.

PROBLEM 16.3 (Megibben). Investigate the endomorphism rings of @1 -free @1 -

separable torsion-free groups.
PROBLEM 16.4. Which groups have endo-projective cover?
PROBLEM 16.5. Relate the endo-flat cover of a group to the group.
PROBLEM 16.6. Discuss pure-projective and pure-injective dimensions over
endomorphism rings.
PROBLEM 16.7. Find the stable range of endomorphism rings.
PROBLEM 16.8. Let S be a ring containing End A as a subring, sharing the same
identity. When does a group G exist that contains A and satisfies End G Š S?
PROBLEM 16.9. The endo-Goldie-dimension of a torsion-free group can be
arbitrarily large. Given the endomorphism ring, can we say something about the
minimum?
PROBLEM 16.10. Which groups A have the property that every automorphism
of Aut A extends to an automorphism of End A?
PROBLEM 16.11. For a subgroup C of A define CP D \2S Ker  where
S D f 2 End A jC D 0g. Which subgroups satisfy CP D C?
PROBLEM 16.12. Any polynomial identity (besides commutativity) in endomor-
phism rings?
Chapter 17
Automorphism Groups

Abstract Needless to say, automorphism groups contain in general much less information
about the group structure than the endomorphism rings. There is an enormous contrast between
torsion and torsion-free groups, perhaps even larger than from the point of view of endomorphism
rings. Here again, the case of p-groups is more favorable (at least for p > 2) inasmuch as
their automorphism groups determine the groups up to isomorphism. The torsion-free case seems
uncontrollable, but a close examination shows that, interestingly, only a handful of finite groups
may occur as automorphism groups of torsion-free groups of finite rank.
We remark at the outset that we will not give complete proofs of the two most important
theorems on automorphism groups: that Aut A determines A if A is a p-group (p > 2), and the
full description of finite automorphism groups of finite rank torsion-free groups. The proofs are
too long and require lot of arguments on non-commutative groups.

1 Automorphism Groups

Automorphisms The automorphisms of a group A form a group under composition

(written multiplicatively): the product ˛ˇ of two automorphisms ˛; ˇ of A is
defined as

.˛ˇ/a D ˛.ˇa/ for all a 2 A:

This is, in general a non-commutative group, the automorphism group of A;

denoted Aut A. It is exactly the group of units in the ring End A.
The groups of order 1 and 2 have no automorphisms other than the identity.
But every group of order > 2 has other automorphisms as well: a 7! a is an
automorphism different from 1A except when A is an elementary 2-group. If A is an
elementary 2-group of order  4, then permutations of the elements of a basis give
rise to automorphisms ¤ 1A .
The automorphism groups of some important groups are listed next.
Example 1.1. Aut Z Š Z.2/. Besides 1Z , multiplication by 1 is the only automorphism.
Example 1.2. An automorphism of Z.n/ carries a generator a into another generator b, and clearly
b D ka with gcd.k; n/ D 1. Conversely, any such k gives rise to a well-defined automorphism of
Z.n/ mapping a to ka. It follows that Aut Z.n/ is isomorphic to the multiplicative group of the
residue classes of integers coprime to n. Thus Aut Z.n/ is commutative, and its order is given by
Euler’s totient function '.n/.

© Springer International Publishing Switzerland 2015 655

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_17
656 17 Automorphism Groups

Example 1.3. As End Z.p1 / Š Jp , it follows that Aut Z.p1 / is isomorphic to the multiplicative
group of p-adic units. (This group will be explicitly described in Sect. 8 in Chapter 18.) The same
holds for Aut Jp .
Example 1.4. Let R be a rational group of type t D .k1 ; : : : ; kn ; : : : /. Then a 7! pn a .a 2 R/ is
an automorphism of R if and only if kn D 1. Since each endomorphism is a multiplication by a
non-zero rational number, it follows that Aut R must be isomorphic to the multiplicative group of
all rational numbers whose nominators and denominators are divisible by primes pn only for which
kn D 1. Thus Aut R is isomorphic to the additive group which is the direct sum of Z.2/ and as
many copies of Z as the number of kn D 1 in t.
Example 1.5. Let A be an elementary p-group of order pm . Then A is an m-dimensional vector
space over the prime field of characteristic p. Since automorphisms of A are linear transformations
of the vector space, Aut A is isomorphic to the general linear group GL.m; p/ (non-commutative).

Elementary Facts We list below some facts that we have found useful to keep in
mind.
(a) If C is a characteristic subgroup of A and ˛ 2 Aut A, then ˛  C is an
automorphism of C, and a C C 7! ˛a C C .a 2 A/ is an automorphism of
A=C.
(b) An isomorphism  W A ! C between groups induces an isomorphism   W
Aut A ! Aut C via   .˛/ D ˛ 1 .
(c) If A is a direct sum, and if we represent the endomorphisms by matrices, then
the automorphisms correspond to the invertible matrices.
(d) If A D B ˚ C, then Aut B may be regarded as a subgroup of Aut A, identified
with the set of all ˛ 2 Aut A satisfying ˛  C D 1C . (This identification depends
on the choice of the complement C.)
(e) If A D ˚i2I Ai , then under the identification indicated in (d), each Q of Aut Ai may
be considered as a subgroup of Aut A. The cartesian product i2I Aut Ai is a
subgroup of Aut A, it consists of those ˛ 2 Aut A that carryQeach Ai into itself.
If all the Ai are characteristic subgroups Q in A, then Aut A Š i2I Aut Ai . Hence:
(f) If A is a torsion group, then Aut A Š p Aut Ap .
(g) Every automorphism of pn A extends to an automorphism of A. It suffices to
prove this for n D 1, and when A has no summand of order p. Let be an
automorphism
Pm of pA. If fbi gi2I is a p-basis of A, then write a 2 A in the form
a D iD1 i i C pc with ki 2 Z n pZ and c 2 A. Note that the terms ki bi
k b
and pc are uniquely determined by a. For each i 2 I, choose ai 2 A such that
pai D .pbi/ 2 pA, and define  W A ! A by

X
m
.a/ D ki ai C .pc/:
iD1

It is clear that  is an endomorphism of A such that   pA D . To see that

 is monic, assume .a/ D 0 for some a 2 A. Then .pa/ D 0, so pa D 0.
Hypothesis on A implies a 2 pA, therefore, .a/ D .a/ D 0 is impossible.
Finally, to show that  is epic, for a given x 2 A pick a y 2 A such that .py/ D
px. Then x  .y/ is of order p, thus x  .y/ D pc for some c 2 A. If z 2 A
satisfies pz D .pc/, then x D .y C z/. Consequently,  2 Aut A.
1 Automorphism Groups 657

(h) (Baer [3]) If C is a characteristic subgroup of A, then

Fix C D f˛ 2 Aut A j ˛c D c 8c 2 Cg

is a normal subgroup of Aut A. Conversely, if  is a normal subgroup of Aut A,

then

Fix  D fa 2 A j ıa D a 8ı 2 g

is a characteristic subgroup of A. This is a Galois-type correspondence.

Stabilizer Let 0 D C0 < C1 <    < Cn D A be a chain of subgroups of A.
By the stabilizer of this chain we mean the subgroup consisting of all ˛ 2 Aut A
such that ˛ induces the identity on all of the factor groups Ci =Ci1 ; in other words,
˛ci  ci 2 Ci1 .8 ci 2 Ci / for all i  1. By the stabilizer of a subgroup C is meant
the stabilizer of the chain 0 < C < A.
Lemma 1.6. The stabilizer † of a subgroup C of A is isomorphic to Hom.A=C; C/,
so it is commutative. It is a normal subgroup of Aut A whenever C is a characteristic
subgroup of A.
Proof. If ˛; ˇ 2 †, then for a 2 A we have ˛a D a C c and ˇa D a C d for
some c; d 2 C. Hence .˛ˇ/a D ˛.a C d/ D a C c C d D .ˇ˛/a. The mappings
˛ W a C C 7! .˛  1/a D c and ˇ W a C C 7! .ˇ  1/a D d are well-defined
homomorphisms A=C ! C such that ˛ˇ W a C C 7! c C d. Therefore, ˛ 7! ˛ is
a homomorphism † ! Hom.A=C; C/ with trivial kernel. It is onto, since the map
˛ W a 7! .a C C/ belongs to † for each  2 Hom.A=C; C/.
The second assertion is straightforward to check. t
u
Example 1.7. Let † denote the stabilizer of the infinite chain

0 < AŒp
< AŒp2
<    < AŒpn
<    < A

for an unbounded p-group A. Define

†n D f˛ 2 † j ˛.a/ D a 8a 2 AŒpn
g .n < !/:

Each †n is a normal subgroup of Aut A, and in the descending normal chain † D †0 B †1 B

   B †n B    B f1g the factor group †n =†nC1 is a subgroup of the stabilizer of AŒpn
in
AŒpnC1
, so it is an elementary p-group as it is clear from Lemma 1.6. A group admitting a normal
chain with elementary p-groups as factor groups is often called (in non-abelian group theory)
generalized p-nilpotent.

Of special interest is the case when A is a reduced p-group (of length ), and the
chain is

AŒp
 pAŒp
     p AŒp
     p AŒp
D 0; (17.1)

subgroups in the socle of A. The dimensions of the factor groups S D

p AŒp
=p C1 AŒp
are given by the UK-invariants of A, so Aut S is the general
658 17 Automorphism Groups

linear group GL.f .A/; p/. As the groups in the chain are fully invariant, every
˛ 2 Aut A induces an automorphism

˛ W a C p C1 AŒp
7! ˛a C p C1 AŒp
.a 2 p AŒp
/

Q a homomorphism W ˛ 7! .˛0 ; : : : ; ˛ ; : : : / of Aut A

of S . In this way, we obtain
into the cartesian product  < Aut S whose kernel is the stabilizer † of the chain
(17.1). As every automorphism of Sn .n < !/ extends to an automorphism of A,
Im will contain all of Aut Sn .n < !/. Consequently, for a separable p-group,
Im is a subdirect product of the groups Aut Sn .n < !/.
Proposition 1.8 (Freedman [1], Hill [10]). The homomorphism
Y
W Aut A ! Aut S
 <

is surjective if the p-group A is †-cyclic, torsion-complete, or totally projective.

Proof. The claim is trivial for †-cyclic groups. For any p-group A, a collection
of automorphisms ˛n 2 Aut Sn .n < !/ lifts to an automorphism of a basic
subgroup B, and every such map can be extended to A whenever A is torsion-
complete Theorem 3.11 in Chapter 17.
If A is totally projective, then we apply transfinite induction on the length
 of A. We are done if 
!. Suppose A has length  C 1, and we select an
arbitrary sequence .˛0 ; : : : ; ˛ / of automorphisms of S .
/. By induction,
A=p A admits an automorphism ˛ 0 inducing the given ˛ on S . < /. We apply
Lemma 4.5 in Chapter 11, choosing both groups as A=p C1 A D A and both nice
subgroups as S D p A=p C1 A. The height-preserving isomorphism  we now use
between the two copies of S is the given automorphism ˛ of S , which is now
extended by the theorem to an automorphism of A such that all the monic maps
between the Hill invariants are induced by the given ˛ . < / (see the proof of
Lemma 4.5 in Chapter 11). This yields an ˛ 2 Aut A inducing the prescribed set of
the ˛ . Applying an element in the stabilizer of (17.1), we can even guarantee that
˛ induces ˛ 0 on A=p A. This will allow us to build matching automorphisms on the
factor groups A=p A leading to a desired automorphism of A. t
u
Involutions Recall that direct decompositions are recognizable in terms of idem-
potent endomorphisms. There is a similar, though not so effective, tool for automor-
phisms: the involutions, i.e. automorphisms  such that  2 D 1. As we shall see,
they do not provide as much information as the projections, and, in addition, to make
it work and to avoid unpleasant nuisance, one has to assume that multiplication by
2 is an automorphism of the group.
Throughout (j)-.`/, we will tacitly assume that A is a group such that a 7! 2a is
an automorphism of A. (In particular, A has no 2-component.)
1 Automorphism Groups 659

(j) An involution  defines two subgroups

A
 D fa 2 A j a D ag and AC
 D fa 2 A j a D ag

C
such that A D A
 ˚ A . Thus an involution  ¤ ˙1 gives rise to a non-trivial
direct decomposition. The projections associated with this direct sum are

 D 12 .1  / and 1   D 12 .1 C /:

Conversely, the involutions corresponding to the projections ; 1   are 1 

2; 2  1.
Example 1.9. That without the hypothesis 2A D A the decomposition (j) may fail is shown e.g.
by the torsion-free group A D hp1 a; q1 b; 12 .a C b/i, defined for different odd primes p; q.
The correspondence a 7! a; b 7! b induces an involution, but A is indecomposable.

(k) Two involutions ;  of A commute if and only if

C C C C C
A    
 D .A \ A / ˚ .A \ A / and A D .A \ A / ˚ .A \ A /;

C
and similarly for A ; A . Note that if these equalities hold, then the two
involutions commute on all four intersections, so also on A.
.`/ A direct decomposition

A D C1 ˚    ˚ Cn .Ci ¤ 0/ (17.2)

defines a set f1 ; : : : ; n g of commuting involutions such that i  Cj D ıij 1Cj

where ıij is the Kronecker delta. Conversely, a set f1 ; : : : ; n g of pairwise
commuting involutions determines a unique decomposition (17.2).
F Notes. Our knowledge about the automorphism groups of abelian groups is rather
fragmentary. Several examples of groups are known that can never be automorphism groups of
abelian groups, and it is hard to visualize what sort of condition will make a group into an
automorphism group. (In contrast, e.g. for commutative rings, every group can be an automorphism
group.) No systematic analysis of all automorphism groups of abelian groups is available as yet.
Whatever the outcome of such a study will be, it is clear that the normal structure of automorphism
groups will be a major factor in understanding these groups. This justifies the widespread interest
in search for normal subgroups in Aut A.
A troublesome aspect in the study of automorphism groups is that the methods of group theory,
albeit extremely powerful and sharp, seem to be inapt in this case. We feel that automorphism
groups of abelian groups deserve much more attention by non-commutative group theorists than
they have received in the past: it is a fertile area that might also have interesting repercussions to
general group theory.
660 17 Automorphism Groups

Exercises

(1) A summand that is a characteristic subgroup is necessarily fully invariant.

(2) (a) Every automorphism of A extends to an automorphism of its injective hull.
The extension is unique if A is torsion-free.
(b) Assume A1 D 0. Every automorphism of A extends uniquely to an
automorphism of its Z-adic completion A. O
(c) Every automorphism of a reduced group can be extended uniquely to its
cotorsion hull.
(3) The elements of the stabilizer subgroup of a chain carry each link into itself.
(4) (Leptin) ˛ 2 Aut A induces the identity on A=nA and leaves the elements AŒn

fixed if and only if it is of the form ˛ D 1  n for some  2 EndA. [Hint: for
n D pk , .1  ˛/A
pk A; .1  ˛/AŒpk
D 0, consider p-basic in A.]
(5) The group ring Z.G/ of G D Aut A with integral coefficients is mapped into
End A in the obvious way. Show that the image of this ring-homomorphism is
the subring of End A generated by G.
(6) The automorphisms of A and the automorphisms of C induce commuting
automorphisms on the groups Hom.C; A/; Ext.C; A/; C ˝ A; Tor.C; A/:
(7) Let hai i be cyclic of prime order pi , and T D ˚i<! hai i for different pi .
Q
(a) Every automorphism of T extends uniquely to G D i<! hai i.
(b) j Aut Gj D 2@0 .
(c) Two pure subgroups of G are isomorphic if and only if there exists an ˛ 2
Aut G carrying one onto the other.
(8) (Mishina) A group A has the property that every automorphism of every
subgroup extends to an automorphism of A if and only if A is one of the
following groups: (a) divisible group; (b) torsion group with homogeneous p-
components; (c) direct sum of a divisible torsion group and a rank 1 torsion-free
group. [Hint: consider †-cyclic subgroups.]
(9) To give an example of a characteristic subgroup that fails to be fully invariant,
pextension of degree 2 of Q such that
choose a ring R of algebraic integers in an
˙1 are the only units in R (e.g., R D ZŒ 5
). Pick a torsion-free group A of
rank 4 with End A Š R, and show that (a) all subgroups of A are characteristic;
(b) all endomorphisms of A are monic; and (c) A contains no fully invariant
subgroup of rank 1.

2 Automorphism Groups of p-Groups

With the exception of groups of order

2, all p-groups have non-trivial automor-
phisms; moreover, if the group is infinite, then it has at least 2@0 automorphisms (cf.
Exercise 3).
2 Automorphism Groups of p-Groups 661

The Center For torsion groups A, our first concern is to identify the center of Aut A.
We may restrict our considerations to p-groups.
Theorem 2.1 (Baer [5]). If p ¤ 2, then the center of the automorphism group of a
p-group A consists of
(i) multiplications by p-adic units, if A is unbounded, and
(ii) multiplications by integers k with 1
k < pn coprime to p, if A is bounded, and
pn is the smallest bound.
Proof. It follows from Theorem 2.9 in Chapter 16 that the multiplications listed in
(i) or (ii) belong to the center. It is more difficult to prove that there are no other
elements in the center. Fortunately, the arguments in Lemma 6.1 in Chapter 16
can be applied with  2 Aut A, because  can be chosen as an automorphism by
requiring that gcd.t; p/ D 1. (In the present situation, the hypothesis p ¤ 2 is
needed in the proof to conclude that v D 0.) t
u
There is a single special case:

A D Z.2m / ˚ B ˚ Z.21 / with 2m1 B D 0 .m  1/ (17.3)

that is responsible for making the prime p D 2 exceptional in the preceding theorem.
For this group, the center of Aut A is the direct product of the multiplicative group
of the 2-adic units and Z.2/ D h˛i, where ˛ is the identity everywhere except on the
generators c of the largest cyclic summand Z.2m /; in this situation, ˛.c/ D c C v,
where v denotes the unique element of order 2 and of infinite height.
To see this, we take a closer look at the proof of Lemma 6.1(i) in Chapter 16. Let
A be a 2-group that we write as A D hci˚B˚D, where o.c/ D 2m , D is the divisible
part, and so far no assumption on B. Now .c/ D kc C v .v 2 B ˚ D/, and suppose
v ¤ 0. Everything works in the proof (in particular,  has to be multiplication by
a dyadic unit  in D), except that v could be of order 2, and that k must be odd to
have  2 Aut A. Replacing  by an odd multiple, k D 1 may be assumed. If  W
hci ! B ˚ D is any homomorphism, then  has to commute with the automorphism
ˇ defined by ˇ.c/ D c C .c/; ˇ.x/ D x .x 2 B ˚ D/, therefore

c C .c/ C v D ˇ.c C v/ D ˇ.c/ D ˇ.c/ D .c C .c// D c C v C .c/:

Thus  fixes all homomorphic images of c in B ˚ D which is .B ˚ D/Œ2m

. This
conclusion applied to a cyclic summand of order  2m in B would imply v D 0,
so 2m1 B D 0 follows. If v is contained in a cyclic summand hbi < B, then 
commutes with the automorphism that is the identity on the complement of hbi
and satisfies .b/ D .b/ C b with a monomorphism  W hbi ! hci. This leads to
c C v D .c/ D  .c/ D .c/ D .c C v/ D c C .v/ C v, whence .v/ D 0,
a contradiction. Therefore, v 2 D. If there is another w 2 D of order 2, then there
is an automorphism of D (and hence of A) that switches v and w; this would not
commute with . Consequently, D Š Z.21 /, A has to be of form (17.3), and the
automorphisms have to be as stated.
662 17 Automorphism Groups

Accordingly, from Theorem 8.6 in Chapter 18 it follows that for the center
z.Aut A/ of the automorphism group of a p-group A we have: if p > 2; then

z.Aut A/ Š Z.p  1/ ˚ Jp or Š Z.p  1/ ˚ Z.pm1 /

according as A is unbounded or bounded by pm , while if p D 2, then

z.Aut A/ Š J2 or Š Z.2m1 / or Š Z.2/ ˚ J2

according as A is unbounded, bounded by 2m , or of exceptional type.

Since a group is abelian exactly if it coincides with its center, from the preceding
theorem it is immediate that the automorphism group of a p-group is commutative
if and only if it is cocyclic or isomorphic to Z.21 / ˚ Z.2/:
Involutions of p-Groups It appears that involutions enter the picture naturally.
Involutions contain a lot of information about p-groups, but we have to learn how to
read them. We now go on to collect some relevant observations about involutions in
p-groups.
If S  Aut A,

c.S/ D f˛ 2 Aut A j ˛ D ˛ 8 2 Sg

will denote the centralizer of S in Aut A. Of course, c.Aut A/ D z.Aut A/. An

involution  is said to be extremal if either AC 
 or A is indecomposable ¤ 0.
In (a)–(e) we keep assuming that p ¤ 2.
(a) The p-group A is bounded, and pn is the l.u.b. for the orders of its elements if
and only if z.Aut A/ Š Z.pn1 .p  1//.
(b) If A is a p-group, and if  2 End A is such that Ker  D 0 and .pn AŒp
/ D
pn AŒp
for all n < !, then  is an automorphism. Evidently,  is monic and
AŒp

Im . To prove by induction on n that AŒpn

Im , suppose a 2 A
has order pn .n > 1/. By induction hypothesis, there are a b 2 AŒpn
such that
.pn1 b/ D pn1 a, and a c 2 A with .c/ D a  .b/. Hence a D .b C c/, and
 is epic as well.
(c) Assume A D C1 ˚    ˚ Ck .Ci ¤ 0/ is a direct decomposition of the p-group
A. Let i denote the involutions that belong to this decomposition. Then the
centralizer

cf1 ; : : : ; k g D Aut C1      Aut Ck ;

and if k  3, then the center of this centralizer contains exactly 2k involutions.

That under the identification stated in Sect. 1 (d), all Aut Ci belong to the
centralizer of the set of the i is evident. Conversely, if ˛ 2 Aut A commutes
with each of i , then ˛Ci D ˛A 
i
Ai D Ci for every i, whence ˛Ci D Ci , so
˛ 2  Aut Ci : The final claim is evident.
2 Automorphism Groups of p-Groups 663

(d) If an involution  of A is extremal, then for every commuting involution ,

in the direct decomposition stated in Sect. 1 (h) there are at most 3 non-zero
summands.
Example 2.2 (Leptin [2]). It can happen that, for different primes p; q, a p-group and a q-group
have isomorphic automorphism groups, though this is a very rare phenomenon. This is the case if
and only if the groups are cyclic: Z.pk / and Z.q/ such that 3  p < q and q  1 D pk1 .p  1/:
(E.g., q is a Fermat prime, and p D 2.)

The main result on p-groups is the following theorem which we state without
proof. It is an important contribution to the theory of p-groups.
Theorem 2.3 (Leptin [2], Liebert [5]). Assume p > 2 and A; C are p-groups. If
Aut A Š Aut C, then A Š C. u
t
F Notes. Automorphism groups of finite abelian groups have been investigated by several
authors, starting with Shoda [1], and as a result, a lot is known about them. A further step was taken
by Baer [3] who studied the normal structure of Aut A for infinite p-groups A; he also investigated
the correspondence between characteristic subgroups of A and normal subgroups of Aut A. That
the results gave special status to the prime number 2 should come as no surprise. Later, the study
of the normal structure of Aut A has advanced considerably, see Faltings [1], Freedman [1], Leptin
[3], Mader [1, 2], Hill [10], and above all Hausen [1, 2, 3].
The amount of open questions as to how the normal structure can be described is staggering.
Several papers were devoted to the question of characteristic subgroups whose automorphisms
extend to automorphisms of the containing group, as well as to pairs of subgroups whose
isomorphism extends to an automorphism of the entire group. Solvable automorphism groups were
studied by Shlyafer [1] and Brandl [1]: for a p-group A, Aut A is solvable only if A has the minimum
condition on subgroups. See also papers by Hausen listed in the Bibliography.
The multiplicative analogue of the Baer–Kaplansky theorem is more subtle; it has not been
completely settled: the case p D 2 is still open. The work started with Freedman [1] who proved
that for p
5 two countable reduced p-groups A; C are isomorphic whenever p! A Š p! C and
Aut A Š Aut C. It was a big step forward when Leptin [2] succeeded in relaxing the requirements
of countability and the isomorphy of the first Ulm subgroups. In cases p D 2; 3, involutions
created serious problems. Only 25 years later was able Liebert [5] to find a proof, using completely
different methods, that included also the prime p D 3. Unfortunately, the proofs do not provide a
method to encode relevant structural information about the group from the automorphism group.
Schultz [3] claimed a proof for p D 2, but his proof relied on a lemma that held only for p
3.
The case p D 2 seems awfully difficult, and the Leptin-Liebert theorem might not extend to this
special case. For p
3, Liebert [5] describes the possible isomorphisms between Aut A and Aut A0
for isomorphic p-groups A; A0 ; there exist some not induced by any group isomorphism.
Leptin [2] has an interesting in-depth study of commuting involutions. For an illuminating
survey on results concerning automorphism groups up to the year 1999, we refer to Schultz [4]
(the claim for p D 2 should be ignored).
The elementary equivalence of automorphism groups of p-groups .p
3/ were discussed by
Bunina–Roǐzner [1].

Exercises

(1) (a) An elementary group of order pm has .pm  1/.pm  p/    .pm  pm1 /
automorphisms.
(b) j Aut Aj D 2 if A is an elementary p-group of cardinality .
664 17 Automorphism Groups

(2) If A is a p-group of order pm , then j Aut Aj is divisible by pm1 .

(3) (Boyer, E. Walker, Khabbaz, Mader, Hill)
(a) If A is an infinite reduced p-group, then j Aut Aj D 2jBj , where B is a basic
subgroup of A.
(b) If A is a non-reduced p-group, then j Aut Aj D 2jAj .
(4) (Baer) The subgroup of elements in a p-group A that are left fixed under all
automorphisms of A is not f0g if and only if p D 2 and A has a unique element
g of order 2 of maximal height (in which case this subgroup is hgi).
(5) (Baer) Let A be a separable p-group and p  3. The set of elements of A left
fixed under all automorphisms that fix a subgroup C elementwise is C , the
closure of C in the p-adic topology. (If p D 2, the exceptional subgroup hgi of
the preceding example, if exists, must be adjoined to C .)
(6) (Hill) Let A be a †-cyclic p-group. For n  1, every automorphism of AŒpn

that preserves heights (computed in A) extends to an automorphism of A.

[Hint: extend it to AŒpnC1
.]
(7) (Megibben) Every automorphism of a large subgroup of a p-group A that
preserves heights (computed in A) extends to an automorphism of A.
(8) If A is a p-group, and ˛ 2 Aut A induces the identity on p A=p C1A for some
ordinal , then it also induces the identity on p Cn A=p CnC1 A for all integers
n  1.
(9) (Mader) Let A be a divisible p-group, and †n the normal subgroup of Aut A
that consists of those ˛ 2 Aut A which leave AŒpn
element-wise fixed. Show
that .Aut A/=†n Š Aut AŒpn
.
(10) (Hausen) The automorphism group of a finitely cogenerated group is residu-
ally finite (i.e., every element ¤ 1 is contained in a normal subgroup of finite
index). [Hint: consider ˛ 2 Aut A fixing AŒpn
element-wise.]
(11) (Tarwater) Let A be a homogeneous †-cyclic p-group. It has an automorphism
˛ carrying the subgroup G into the subgroup H if and only if G Š H and
AŒp
=GŒp
Š AŒp
=HŒp
.
(12) (Baer) For a p-group A with p  3, all the torsion subgroups of Aut A are finite
exactly if A is finitely cogenerated. [Hint: an infinity of independent summands
yields an infinite torsion group in the centralizer of the system of involutions;
conversely, reduce to divisible groups and consider matrix representation.]
(13) Suppose A D B ˚ C, and B is fully invariant in A. Then Aut A is the semidirect
product of the stabilizer of the chain 0 < B < A by the subgroup Aut BAut C.
[Hint: map Aut A onto Aut B  Aut C.]

3 Automorphism Groups of Torsion-Free Groups

The most relevant results we know concerning automorphism groups of torsion-

free groups are about the finite automorphism groups. Before tackling this special
3 Automorphism Groups of Torsion-Free Groups 665

situation, let us consider a few facts without putting any restrictive condition on the
automorphism groups.
(A) If a group G is isomorphic to Aut A for a torsion-free group A, then G is
isomorphic to the group of units in a subring R of a Q-algebra S. The converse
holds if G is finite.
Indeed, in the torsion-free case, Aut A is the unit group in End A that is
a subring of the Q-algebra Q ˝ End A. If G is a finite subgroup in the unit
group of a Q-algebra R, then it is the group of units in the subring R  ZŒG

generated by G. In view of Theorem 3.3 in Chapter 16, there exists a group A

with End A Š R.
(B) Monic endomorphisms of a finite rank torsion-free group are automorphisms.
More generally, if the additive group of a ring R is a finite rank torsion-free
group, then the one-sided units of R are two-sided.
Assume r; s 2 R satisfy rs D 1, but sr ¤ 1. Then r.sr  1/ D 0, so the right
annihilator Ann r ¤ 0. Therefore, the right annihilator Ann r
R contains a
pure subgroup of rank  1. Then an obvious argument with ranks shows that
rs D 1 is impossible.

The automorphism group of a torsion-free group rarely contains any relevant

information about the structure of the group itself—this is evident from several
examples. Not even the rank 1 groups can be recaptured from their automorphism
groups, as is illustrated by the following example.
Example 3.1. Let A be a rational group of type .k1 ; : : : ; kn ; : : : /. If all the kn are finite, then
Aut A Š Z.2/. In all other cases, Aut A is a direct product of Z.2/ and a free abelian group whose
rank is equal to the number of infinite kn ’s.

Finite Automorphism Groups Our knowledge of (non-commutative) groups that

can be realized as automorphism groups of torsion-free groups does not go much
further than the finite case (and sporadic examples). In fact, up to now, the most
important result concerning the automorphism groups of torsion-free groups is the
classification of finite automorphism groups. The full discussion involves more
non-commutative group theory than we care to get into; we refer to the excellent
presentation by Corner [10]. However, we can extract essential facts from the sole
hypothesis that the automorphism groups are torsion—we will see that even this
milder condition alone means severe restriction.
In the following statements we use the notations: G D Aut A and E D End A
where A is a reduced torsion-free group. In the arguments, we take advantage of the
fact that G is the unit group in E.
(a) If G is a torsion group, then E contains no nilpotent elements ¤ 0.
For the proof we may suppose that the nilpotent  2 E satisfies 2 D 0. Then
the endomorphisms 1C and 1 are inverse to each other, so they belong to G.
But .1 C /n D 1 C n ¤ 1 for n 2 N unless  D 0 (due to the torsion-freeness
of E).
666 17 Automorphism Groups

(b) If G is torsion, then every involution  2 G is in the center of G.

For every ˛ 2 G, the endomorphisms .1 C /˛.1  /; .1  /˛.1 C / are
nilpotent, so they are 0 by (a). Therefore 2.˛  ˛/ D .1 C /˛.1  / 
.1  /˛.1 C / D 0, whence ˛ D ˛ follows.
(c) If G is torsion, and ˛ 2 G is of odd order m > 1, then m D 3.
Manifestly, the proof can be confined to the case m D pk for a prime p  3. The
endomorphism ˇ D 1  ˛ C ˛ 2  C    C ˛ m3 has an inverse:
(
1 ˛ 2  ˛ 4 C     C ˛ mC1 if m  1 mod 4;
ˇ D
3 5
˛ ˛ C C ˛ m
if m  1 mod 4I

thus ˇ 2 G. Since ˛ m  1 annihilates A, there must exist a minimal polynomial

k
f .x/ 2 ZŒx
such that f .˛/ annihilates A. Clearly, f .x/ divides xp  1, but
k1
not xp  1, whence from the irreducibility (over Z) of the pk th cyclotomic
k k1
polynomial pk .x/ D .xp  1/=.xp  1/, we deduce the divisibility relation
pk .x/jf .x/. Consequently, the mapping ˛ 7!  (where  is a primitive pk th
complex root of unity) extends to a ring homomorphism  W ZŒ˛
! QŒ
of the
subring of E generated by ˛ into the algebraic extension of Q by . The image of
ˇ under  is the complex number 1 C 2 C   C m3 D .1C p 2 /=.1C/
k

whose absolute value is ¤ 1 unless  p 2 equals  or its complex conjugate

k

N D 1=. In the first alternative we have pk D 3, while the second alternative:

 p 1 D 1 is impossible. Since .ˇ/ must be a root of unity, the only possibility
k

is that ˛ is of order 3.
(d) If G is a torsion group, then it does not contain any element of order 8.
In fact, if ˛ 2 G is of order 8, then

ˇ D 1 C .1  ˛ 4 /.1 C ˛  ˛ 3 / and D 1 C .1  ˛ 4 /.1  ˛ C ˛ 3 /

p the subring ZŒ˛

of E
are inverse to each other in E. As in (c), we can show that
generated by ˛ maps into QŒ
via ˛ 7!  D .1 Cpi/= 2 = a primitive 8th root
of unity. This homomorphism sends ˇ into 3 C 2 2 which is a unit of infinite
order in QŒ
; thus ˇ cannot be of finite order.
At this point, from (c) and (d) we deduce that G, if torsion, contains only
elements whose orders are divisors of 12.
(e) If G is torsion, then the involution 1 is not contained in any cyclic subgroup
of order 12.
We show that the assumption that ˛ 2 G of order 12 satisfies ˛ 6 D 1 is
contradictory. Because then

ˇ D 1  ˛ C ˛2  ˛3 C ˛4 and D 1  ˛ 2  ˛ 3  ˛ 4  ˛ 5

are inverse automorphisms, and—as we wish to prove—of infinite order. We

again examine the polynomial g.x/ 2 ZŒx
of minimal degree such that
g.˛/A D 0. Now x6 C 1 D .x2 C 1/.x4  x2 C 1/ with irreducible factors.
3 Automorphism Groups of Torsion-Free Groups 667

As ˛ 2 ¤ 1 is obvious, it follows that the cyclotomic polynomial 12 .x/ D

x4  x2 C 1 is a divisor of g.x/. As above, we get a ring homomorphism carrying
ˇ into 1   C  2   3 C  4 D .1 C  5 /=.1 C /, where  denotes a primitive
12th root of unity. But j1 C  5 j ¤ j1 C j, so ˇ must be of infinite order.

Primordinal Groups We stop here with the listing of properties of finite (torsion)
automorphism groups G, but will formulate the precise result. The theorem below
refers to six groups, which are called primordinal groups. The following is the
list of the primordinal groups (the non-cyclic groups are given by generators and
defining relations):

cyclic groups: Z.2/; Z.4/; Z.6/;

quaternion group: Q8 D f˛; ˇ j ˛ 2 D ˇ 2 D .˛ˇ/2 g;

dicyclic group: DC12 D f˛; ˇ j ˛ 3 D ˇ 2 D .˛ˇ/2 g;

binary tetrahedral group: BT24 D f˛; ˇ j ˛ 3 D ˇ 3 D .˛ˇ/2 g;

where the indices indicate the group orders. For the proof of the following theorem
we refer to Corner [10]. (B-blocks are groups constructed from BT24 .)
Theorem 3.2 (Hallett–Hirsch [1], Corner [10]). A finite group G is the automor-
phism group of some torsion-free group if and only if it is a subdirect product of
primordinal groups such that either
(i) G has a direct factor of order 2; or
(ii) there exists a direct decomposition

G D G0  G1      Gr

where G1 ; : : : ; Gr are B-blocks, and G0 is a subdirect product of copies of the groups

Z.4/; Q8 ; DC12 in such a way that it contains the diagonal involution 1. t
u
Examples for Primordinal Automorphism Groups The following examples
exhibit torsion-free groups with primordinal automorphism groups.
Example 3.3. Z.2/ as an automorphism group. We have numerous examples for groups with Z.2/
as automorphism groups, the simplest is Z. All the examples are indecomposable.
Example 3.4. Z.4/ as an automorphism group. As the unit group of the Gaussian integers is cyclic
of order 4, the existence of an example is guaranteed by Theorem 3.3 in Chapter 16. An explicit
example is as follows.
Let p1 ; : : : ; pi ; : : : be different primes  1 mod 4. Then 1 is a quadratic residue mod pi , so
we can find ki 2 N such that ki2  1 mod pi for each i. Define

A D ha; b; p1
i .a C ki b/ .i 2 N/i: (17.4)
668 17 Automorphism Groups

Then a 7! b; b 7! a induces an automorphism ˛ of order 4. We show that A has no

automorphisms other than the powers of ˛. Every ˇ 2 Aut A acts as ˇa D ra C sb; ˇb D ta C ub
with integers r; s; t; u satisfying ru  st D ˙1. As ˇ.a C ki b/ D .r C ki t/a C .s C ki u/b ought to
be divisible by pi , it must be an integral linear combination of pi a; pi b; a C ki b. This leads to the
relation sCki u  ki .rCki t/ mod pi whence sCt  ki .ru/ mod pi , and so .sCt/2  .ru/2
mod pi . This holds for all i, so congruence becomes equality, whence t D s; r D u follows. This,
in view of ru  st D ˙1 leaves only four possibilities: r D ˙1; s D 0 and r D 0; s D ˙1. These
correspond to the powers of ˛, in particular, the choice r D 0; s D 1 to ˛.
Example 3.5. Z.6/ as an automorphism group. To give an explicit example, consider an infinite
set p1 ; : : : ; pi ; : : : of primes  1 mod 6. Then there exist integers ki such that ki2 C ki  1
mod pi . Consider the group A as in (17.4) with the new set of primes. The correspondence
a 7! b; b 7! a C b induces an automorphism ˛ of order 6. For ˇ 2 Aut A, write as
before ˇa D ra C sb; ˇb D ta C ub with r; s; t; u 2 Z satisfying ru  st D ˙1. Then
pi jˇ.a C ki b/ D .r C ki t/a C .s C ki u/b implies ki .r C ki t/  s C ki u mod pi . Replacing s by
ski ski2 and canceling by ki , we obtain rCki t  .1Cki /sCu mod pi , or rCsu  ki .sCt/
mod pi : Eliminating ki , we get

.r C s  u/2  .r C s  u/.s C t/ C .s C t/2  0 mod pi :

Again, this holds for all i, thus this congruence becomes equality. As the equation x2  xy C y2 D 0
has only the trivial solution in real numbers x; y, it follows that u D r C s; t D s. Therefore,
r2 C rs C s2 D ru  st D ˙1, whence the only possibilities are r D ˙1; s D 0I r D 1; s D 1,
and those obtained by switching r and s. Consequently, A has at most six automorphisms.
Example 3.6. The quaternion group Q8 as an automorphism group. We select two disjoint infinite
sets of primes, p1 ; : : : ; pi ; : : : and q1 ; : : : ; qi ; : : : , all  1 mod 4, and ki ; `i 2 Z such that ki2  1
mod pi , `2i  1 mod qi . Our group will now be

A D ha; b; c; d; p1 1 1 1
i .a C ki b/; pi .d C ki c/; qi .a C `i c/; qi .b C `i d/ .i 2 N/i:

Using the same argument as in previous examples, from the divisibility relations it follows that
2 Aut A must act like this:

a D ra C sb C tc C ud; b D sa C rb C uc  td;

c D ta  ub C rc C sd; d D ua C tb  sc C rd

for some r; s; t; u 2 Z, such that the determinant of the coefficient matrix—which is equal to
.r2 Cs2 Ct2 Cu2 /2 —is a unit in Z, thus it is equal to 1. Evidently, the equation r2 Cs2 Ct2 Cu2 D 1
has exactly 8 solutions in integers, and the correspondences

˛ W a 7! b; b 7! a; c 7! d; d 7! c; ˇ W a 7! c; b 7! d; c 7! a; d 7! b

(with the choices s D 1; r D t D u D 0, resp. t D 1; r D s D u D 0) induce automorphisms such

that ˛ 2 D ˇ 2 D .˛ˇ/2 D 1. Thus Aut A Š Q8 .
Example 3.7. The dicyclic group DC12 as an automorphism group. Again, we start with two
disjoint infinite sets pi and qi of primes such that pi  1 mod 4 and qi  1 mod 6 for all i 2 N.
Choose integers ki ; `i so as to satisfy ki2  1 mod pi and `2i C `i  1 mod qi , and define

A D ha; b; c; d; p1 1 1 1
i .a C ki b/; pi .c C ki d/; qi .a C `i c/; qi .d C `i b/ .i 2 N/i:

It is easy to check that the correspondences

˛ W a 7! c; b 7! d; c 7! a C c; d 7! b C d; ˇ W a 7! d; b 7! c; c 7! b; d 7! a
3 Automorphism Groups of Torsion-Free Groups 669

give rise to automorphisms of A such that ˛ 3 D ˇ 2 D .˛ˇ/2 D 1. The same technique with
divisibility as before leads to the conclusion that, for every 2 Aut A we must have

a D ra C sb C tc C ud; b D sa C rb  uc  td;

c D ta C .u C s/b C .r C t/c  sd; d D .u C s/a  tb C sc C .r C t/d

with suitable r; s; t; u 2 Z. The determinant of the coefficient matrix must be a unit in Z, thus
.r2 C rt C t2 C s2 C su C u2 /2 D 1. This equation has 12 integral solutions: r D ˙1; s D t D u D
0I r D t D 1; s D u D 0, as well as those obtained from these two by permissible permutations.
Example 3.8. The binary tetrahedral group BT24 as an automorphism group. Let pi ; qi ; ki :`i have
the same meaning as in the preceding example. This time, A is defined as

A D ha; b; c; d; p1 1 1 1
i .a C ki b/; pi .a  c C ki d/; qi .a C `i d/; qi .c C `i b/ .i 2 N/i:

The maps

˛ W a 7! d; b 7! a C c; c 7! a  b C c C d; d 7! a C d;

ˇ W a 7! c; b 7! b  d; c 7! a C c; d 7! b
induce automorphisms of A satisfying ˛ 3 D ˇ 3 D .˛ˇ/2 D 1. It is a tiring but straightforward
calculation to show that ˛; ˇ generate a subgroup of order 24, and there are no other automor-
phisms.

Direct Products of Primordinal Groups Finally, we show how to find a torsion-

free group whose automorphism group is a finite direct product of primordinal
groups.
We start off with the observation that our Examples 3.3–3.8 can easily be
modified by replacing the group A by the tensor product A ˝ R, where R is a rational
group whose type .k1 ; : : : ; kn ; : : : / contains only integers, and kn D 0 whenever the
nth prime occurs among the primes used in the example. Needless to say, only the
case of infinitely many kn ¤ 0 bears any interest.
Now let G D G1   Gn where each Gj is isomorphic to one of the primordinal
groups. For each Gj select a torsion-free group Aj with Aut Aj Š Gj . The primes
p 6 1 mod 4 (or mod 6) are infinite in number, so we can select a rigid system
R1 ; : : : ; Rn of rational groups of the types mentioned above such that kn > 0 only if
pn 6 1 mod 4 and mod 6. It is readily seen that then also the groups Bj D Aj ˝ Rj
form a rigid system. Hence

Aut.B1 ˚    ˚ Bn / D Aut B1      Aut Bn Š G1      Gn :

Orderable Groups as Automorphism Groups It was Corner who pointed out that
groups admitting total orders are candidates for examples that can be realized as
automorphism groups of torsion-free groups modulo the ubiquitous Z.2/. The basic
observation is that for such a group G, the group ring ZŒG
with integer coefficients
has no units other than
P the group elementsP and their negatives. The reason is that
the product of x D kiD1 ni gi and y D `jD1 mj hj (where ni ; mj 2 Z; gi ; hj 2 G)
can never be 1 unless each of x and y is ˙ a group element. In fact, if ni ¤ 0 ¤ mj
670 17 Automorphism Groups

for all i; j, and if g1 <    < gk ; h1 <    < h` under an arbitrarily selected total
order, then in the product xy the terms n1 m1 g1 h1 and nk m` gk h` never cancel, and are
different unless k D 1 D `. As a result, for an orderable group G, the unit group of
the group ring ZŒG
is isomorphic to G  Z.2/.
Theorem 3.9. For every orderable group G, there exist torsion-free groups whose
automorphism groups are isomorphic to G  Z.2/.
Proof. We appeal to Theorem 7.1 in Chapter 16 to derive the existence of torsion-
free groups whose endomorphism rings are isomorphic to ZŒG
. Each of these
groups has the desired property. t
u
Example 3.10. An abelian group is orderable if and only if it is torsion-free. All non-commutative
free groups are orderable, and by a theorem of Malcev, every locally nilpotent torsion-free group
admits a total order.
F Notes. The first relevant results on automorphism groups of torsion-free groups are in the
memoir Beaumont–Pierce [2] on rank 2 groups. They show how complicated these groups can be
even in this extremely special case. See also Król [2].
It was Hallett–Hirsch [1] who started the systematic investigations of finite automorphism
groups of torsion-free groups. It is really surprising that these groups are made up of only a handful
groups by using subdirect products. Their results were corrected and substantially improved by
Corner in a monumental paper [10], published posthumously. Robust arguments were needed
to verify the precise conditions on the admissible subdirect products. It might be helpful to call
attention to the semidirect product representations DC12 Š Z.3/ Ì Z.4/ and BT24 Š Q8 Ì Z.3/.
Hallett–Hirsch [1] also point out that if a torsion-free group H, commutative or not, has finite
automorphism group, then it ought to be abelian. Because if Aut H is finite, then z.H/ is of finite
index in H. Schur’s theorem tells us that then the commutator subgroup H 0 is finite, so it must be
trivial in a torsion-free group.
May [6] has more results on commutative automorphism groups of torsion-free groups; he
succeeded in describing those of countable rank.
p-adic modules display a simpler behavior: Corner–Goldsmith [1] establish the isomorphy of
reduced torsion-free modules over Jp .p ¤ 2/ provided their automorphism groups are isomorphic.
We have had several occasions to observe that there exist torsion-free groups of arbitrarily large
cardinalities with automorphism groups Š Z.2/—which is clearly the smallest possible group
that can be the automorphism group of a torsion-free group ¤ 0. The question concerning the
existence of indecomposable torsion-free groups of cardinality  whose automorphism groups are
of largest possible cardinality, viz. of size 2 , was answered for arbitrarily large  in the affirmative
in Fuchs [22].
The study of automorphism groups is certainly an attractive ground for research, there are lots
of unanswered questions.
Bekker has several papers on the holomorph of abelian groups; Bekker–Nedov [1] show that in
some special cases the holomorph characterizes the group.

Exercises

(1) The stabilizer of any chain of subgroups of a torsion-free group is a torsion-free

group.
(2) Let G be an elementary 2-group of order 2n . There is a finite rank indecompos-
able torsion-free group A such that Aut A Š G.
3 Automorphism Groups of Torsion-Free Groups 671

(3) For any cardinal , the elementary 2-group of cardinality 2 is the automor-
phism group of some torsion-free group of rank .
(4) (Hallett–Hirsch) A finite abelian group F is isomorphic to Aut A for a torsion-
free A if and only if (i) jFj is even; (ii) ˛ 12 D 1 for all ˛ 2 F; and (iii) not every
˛ 2 F of order 2 is contained in a cyclic group of order 12.
(5) (de Vries–de Miranda) Let p1 ; p2 ; q1 ; q2 be different primes. The automorphism
group of

A D hp1 1 1 1
1 a; p1 b; p2 c; p2 d;

q1 1 1 1
1 .a C c/; q1 .b C d/; q2 .a C d/; q2 .b  c C d/i

is Š Z.6/. [Hint: a 7! b; b 7! a C b; c 7! d; d 7! c C d is a generator.]

(6) (Hallett–Hirsch) Suppose A is a torsion-free group with Aut A Š Q8 D
h˛; ˇ j ˛ 2 D ˇ 2 D .˛ˇ/2 i.
(a) For every 0 ¤ a 2 A, the elements a; ˛.a/; ˇ.a/; ˛ˇ.a/ are independent.
(b) If rk A is finite, then it is a multiple of 4.

Problems to Chapter 17

PROBLEM 17.1. Study the normal structure of automorphism groups of p-

groups.
PROBLEM 17.2. For a p-group A, what are the connections between properties
of Aut A and the UK-invariants of A?
PROBLEM 17.3. For which finite rank torsion-free groups A are all the automor-
phisms of End A inner?
PROBLEM 17.4. If Aut A (the unit group of End A) does not generate End A, are
there then characteristic subgroups in A that are not fully invariant?
PROBLEM 17.5. Characterize Aut A for separable p-groups A, or for simply
presented p-groups.
PROBLEM 17.6. Find a simple proof for the Leptin–Liebert theorem. Is Theo-
rem 2.3 true also for p D 2?
Chapter 18
Groups in Rings and in Fields

Abstract The most frequent occurrence of abelian groups, apart from vector spaces, is
undoubtedly the groups found in rings and fields. This chapter is devoted to their study.
While in general we deal exclusively with associative rings with 1, in the first two sections
of this chapter we also include rings without identity as well as not necessarily associative rings
(called ‘narings’ for short) in order to make the discussion smoother. As a matter of fact, the
collection of narings on a group A displays more pleasant features than the set of associative rings,
as demonstrated by the group Mult A. This group, suggested by R. Baer, crystallizes the idea of
building narings on a group (thus from the additive point of view, associativity in rings seems less
natural).
The paper devoted to the additive groups of rings, published by Rédei–Szele [1] on the special
case of torsion-free rings of rank 1, was the beginning of Szele’s ambitious program on the
systematic study of additive groups. Our current knowledge on the additive groups of rings, apart
from artinian and regular rings, is still more fragmentary than systematic, though a large amount
of material is available in the literature. The inherent problem is that interesting ring properties
rarely correspond to familiar group properties. Due to limitation of space, we shall not pursue
this matter here; we refer the reader to Feigelstock’s two-volume treatise [Fe]. The problem of
rings isomorphic to the endomorphism ring of their additive group (called E-rings) attracted much
attention; we present a few miscellaneous results on them.
Our final topic concerns multiplicative groups: groups of units in commutative rings and
multiplicative groups of fields. While the theory for rings has not reached maturity, there are several
essential results in the case of fields, mostly due to W. May. Unfortunately, we cannot discuss them
here, because they require more advanced results on fields.

1 Additive Groups of Rings and Modules

In this section we collect a few fundamental facts on the additive groups of rings.
But first of all, conventions on the terminology are in order.
Rings and Their Additive Groups In this chapter, by a ring we shall mean
an associative ring (with or without identity), and by a naring a not necessarily
associative ring (there are many of those of importance, like Lie rings, alternative
rings, etc.). As customary, we will attribute to the ring properties of its underlying
additive group in cases when no confusion may arise. Consequently, terms like p-
ring, torsion or torsion-free ring, divisible and reduced ring, pure ideal, etc. will
make perfectly good sense without any further comment. If desirable, distinction
will be made between the ring R and its additive group RC ; however, extreme
caution is necessary in the context of direct decompositions. A ring R whose

© Springer International Publishing Switzerland 2015 673

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6_18
674 18 Groups in Rings and in Fields

additive group is isomorphic to the group A will be called a ring on A; we also

say R is supported by A.
Lemma 1.1. For all a; b 2 R and m; n 2 N, we have:
(a) If ma D 0 and nb D 0, then dab D 0 where d D gcdfm; ng.
(b) If mja and mb D 0, then ab D 0.
(c) mja and njb imply mnjab.
Proof. Follow the proof of Lemma 1.2 in Chapter 8. t
u
Some immediate corollaries of this lemma are listed below.

(A) In every ring R, the following are (two-sided) ideals:

(i) nR and RŒn
for every n 2 N;
(ii) the torsion part t.R/, and its p-components;
(iii) the Ulm subgroups, the subgroup p R for every ordinal , including the
p-divisible part p1 R and the divisible subgroup of R;
(iv) the fully invariant subgroups of the additive group RC ;
(v) if R is torsion-free, then R(t) is an ideal for every type t.
Similar statements hold if R is replaced by a left or right ideal. (iv) follows
from the fact that in any ring R, multiplication by an element from the left (or
from the right) is an endomorphism of RC .
(B) If R is a torsion ring, then there is a ring direct decomposition R D ˚p Rp into
p-components.
(C) If pje for an idempotent e, then pk je for all k 2 N. If the ring R has an identity
1, then mj1 .m 2 N/ implies mk ja for all a 2 R and k 2 N:
(D) From (C) we obtain that the p-heights satisfy

hp .ab/  hp .a/ C hp .b/ for all a; b 2 R:

(E) Consequently, if R is a torsion-free ring, then the characters obey the rule
.ab/  .a/.b/ for all a; b 2R.
(F) Elements in the torsion subgroup of a ring are annihilated by elements in the
first Ulm subgroup, in particular, by the elements in the divisible part of the
additive group. The additive group of a p-ring is therefore separable whenever
it has an identity; actually, it must then be a bounded group.
Example 1.2. Let R be a torsion-free ring. If a 2 R is not nilpotent, then either t.a/ is idempotent
or all the types of an .n 2 N/ are different.

Extending Ring to a Larger Group A more relevant problem is concerned with

the extension of the ring structure on a group to a larger group. The first question
which comes to mind is whether or not a ring can be extended to a ring on the
injective hull of its additive group. From (F) it is obvious that in the divisible hull of
a torsion group all products ought to be 0. Thus the real interest lies in the torsion-
free case, which can be dealt with satisfactorily.
1 Additive Groups of Rings and Modules 675

Theorem 1.3. The ring structure on a torsion-free group can be extended uniquely
to a ring on the divisible hull of its additive group.
Proof. There is only one way to extend the multiplication  from a ring on a torsion-
free A to its divisible hull D: if x; y 2 D and m; n 2 N such that mx; my 2 A, then
the extension  (if exists) must be

.x; y/ D .mn/1 mx  ny 2 D:

This is independent of the choice of m; n, and makes D into a ring—this is easy to

check. t
u
Notice that the ring on D inherits commutativity (or associativity), and preserves
identity if the ring on A had one.
Example 1.4. Let R be an integral domain such that RC is torsion-free of finite rank n. Then the
unique extension ring on its divisible hull D is a field. In fact, given r ¤ 0; s in R, the equation
rx D s 2 R is solvable in D, since rR is of rank n, so for some x 2 R, s must be equal to rx.

There are two more noteworthy embeddings of a group that should be examined:
one into its Z-adic completion, and one into its cotorsion hull. Before entering into
the analysis of when and in how many ways a ring structure can be extended in these
cases, we prove an auxiliary lemma.
Lemma 1.5. Let A be a reduced group. A ring structure on a pure dense subgroup
of A can be extended at most in one way to a ring on A.
Proof. From a pure-exact sequence 0 ! C ! A ! A=C ! 0, a repeated
application of Theorem 3.1 in Chapter 8 yields the pure-exact sequences 0 !
C ˝ C ! A ˝ C ! .A=C/ ˝ C ! 0 and 0 ! A ˝ C ! A ˝ A ! A ˝ .A=C/ ! 0,
leading to the exact sequence

0 ! C ˝ C ! A ˝ A ! Œ.A=C
˝ C
˚ ŒA ˝ .A=C/
! 0: (18.1)

Clearly, the terms involving the divisible group A=C are divisible, thus the induced
map Hom.A ˝ A; A/ ! Hom.C ˝ C; A/ is monic. This means that any C ˝ C ! C
can have at most one extension A ˝ A ! A. t
u
We are now prepared to look at the embeddings in completions and in cotorsion
hulls.
Theorem 1.6. Let G denote the pure-injective or the cotorsion hull of A. A ring
structure on A can be extended to a ring on G, which is unique whenever A is
reduced.
Proof. First, let G be the completion of A. As above, from the pure-exact sequence
0 ! A ! G ! G=A ! 0 we derive the conclusion that 0 ! A ˝ A ! G ˝ G
is pure-exact, and by pure-injectivity, that the induced map Hom.G ˝ G; G/ !
Hom.A ˝ A; G/ is surjective. Hence the extensibility is manifest. If A is reduced, the
preceding lemma takes care of uniqueness.
676 18 Groups in Rings and in Fields

If G is the cotorsion hull of A, then in the exact sequence 0 ! A ˝ A !

G ˝ G ! Œ.G=A
˝ A
˚ ŒG ˝ .G=A/
! 0 (that is like (18.1)) the direct sum
is torsion-free and divisible. A reference to Theorem 9.2 in Chapter 9 shows that the
map Hom.G ˝ G; G/ ! Hom.A ˝ A; G/ is surjective, completing the proof. t
u
F Notes. The pioneering papers: Beaumont [1], Szele [1], Rédei–Szele [1], Beaumont–
Zuckerman [1] (all published around 1950) provide only a superficial analysis of the relation
between the structure of a ring and its additive group, in very special cases. They stimulated
interest in the additive groups of rings, and in the next decade several more substantial papers
were published, including Beaumont–Pierce [1] which initiated a more systematic study of rings
on finite rank torsion-free groups.

Exercises

(1) Describe all (associative) rings on the following groups: Z.pn /; Z.p1 /; Z.
(2) (a) In a torsion-free ring R with identity 1, .1/
.a/ for each a 2 R:
(b) An unbounded p-ring cannot have identity.
(3) If U; V are fully invariant subgroups (ideals) of R, then their product is
contained in their intersection U \ V.
(4) (Steinfeld) Let R be a ring without divisors of zero. Then either pR D 0 for
some prime p, or R is torsion-free whose typeset is directed upward.
(5) In a torsion-free ring R, the nil radical is a pure ideal.
(6) Let M denote a maximal (left) ideal of the torsion-free ring R. Then either
pR
M for some prime p, or R/M is torsion-free divisible.
(7) The ring constructed in Theorem 1.3 has the property that every torsion-free
divisible ring containing a subring isomorphic to the ring on A also contains a
ring isomorphic to the one on D.
(8) Let R be a ring with additive group A such that A1 D 0. Then the
unique extension ring on the completion AQ inherits polynomial identities (in
Q
particular, associativity, commutativity) from R, and is a Z-algebra. [Hint: AQ
is an inverse limit.]
(9) Let A and D be as in Theorem 1.3.
(a) Establish a bijection between the pure left ideals of A and D.
(b) This correspondence preserves primeness.
(c) D can have an identity even if A has none.
(d) If D has an identity, then every left ideal of D is pure.
(e) D has a non-zero nilpotent ideal exactly if A has got one.
(10) A ring on a splitting mixed group need not be a direct sum of a torsion and a
torsion-free ring. [Hint: on hai ˚ hbi Š Z ˚ Z.p/ define a2 D b; ab D ba D
b2 D 0.]
2 Rings on Groups 677

(11) For a ring R with identity, by a universal R-module on the group A is meant
an R-module U along with a group homomorphism  W A ! U such that for
every group homomorphism ˛ W A ! M from A into an R-module M, there
is a unique R-homomorphism W U ! M such that  D ˛. Verify the
existence and uniqueness of universal R-modules. [Hint: R ˝ A.]

2 Rings on Groups

In the preceding section we discussed additive groups of rings, in this section we

reverse our approach, and will focus on ring structures on given groups. Here it
seems reasonable to consider not only the associative rings, but to include also
narings in the discussion.
Every group A can be furnished with a ring structure in a trivial way by defining
all products to be 0; such a ring is called the zero-ring on A. In general, there exist
several other rings on A.
Rings on a Given Group By multiplication on a group A, we mean a function
WAA!A
that is additive in both arguments. The multiplications in A are in a bijective
correspondence with the naring structures on A. Observe that left multiplications
by a 2 A are endomorphisms of the group, thus the multiplication  in R defines a
homomorphism  W A ! End A such that .a; x/ D .a/.x/ .a; x 2 A/. Vice versa,
every such  defines a naring on A. In this way, the naring structures on A can be
identified with the elements of

Mult A D Hom.A; End A/ Š Hom.A ˝ A; A/:

Mult A is called the group of multiplications on A (R. Baer).

Example 2.1. (a) Let A D hai be a cyclic group. Then Mult A Š A generated by  W a 7! 1A
. 2 Hom.A; End A//. It is a rare phenomenon that all the multiplications on a group are
associative and commutative.
(b) Let A D hai ˚ hbi, both cyclic groups of the same order. The multiplication defined by the
following map is not associative: .a/ W a 7! b; b 7! bI .b/ W a 7! a; b 7! a (the rest follows
by distributivity).
Example 2.2. Let A be a torsion-free group ¤ Z with End A Š Z. Then Mult A D 0; A is a nil
group (see below).

We call a group N a nil group if there is no naring on N except for the zero-ring,
i.e. Mult N D 0, and call N a quasi-nil group if it admits but a finite number of
non-isomorphic rings. In the torsion case, both the nil, and the quasi-nil groups can
be described without difficulty; moreover, with a little extra effort, even the case of
mixed nil groups can be settled.
678 18 Groups in Rings and in Fields

Proposition 2.3 (Szele [1]). A torsion group is nil if and only if it is divisible. There
exists no genuine mixed group that is nil.
Proof. A summand of a nil group is nil, and since cyclic groups of orders pk .k 2 N/
are not nil, it is clear that the torsion part of a nil group must be divisible. On the
other hand, for a torsion divisible group D we have D ˝ D D 0, so Mult D D 0.
For a mixed nil group N, tN is divisible, and if N ¤ tN ¤ 0, then there is a
non-trivial map 0 ¤ N ˝ N ! tN, so Mult N ¤ 0. t
u
Proposition 2.4 (Fuchs [7]). A torsion group N is quasi-nil exactly if N D B ˚ D
where B is finite and D is divisible.
Proof. The basic subgroup B of a quasi-nil torsion group N must be finite.
Otherwise, it is easy to construct rings R on N such that the squares R2 have different
finite cardinalities. Thus N must have the indicated structure.
Conversely, assume N D B ˚ D, where B D hb1 i ˚   ˚ hbk i is a finite and D is a
divisible group. The narings on N are completely determined by the products bi bj of
basis elements; the products might have non-zero D-coordinates, but all of them are
contained in a finite rank summand of D, so as far as isomorphism is concerned, this
summand may be regarded the same for all narings. This means that there are but a
finite number of choices for the products bi bj , and therefore N is quasi-nil, indeed.
t
u
Example 2.5. (a) A rank 1 torsion-free group is nil if and only if its type is not idempotent. A
completely decomposable torsion-free group that is homogeneous of non-idempotent type is
a nil group.
(b) (Corner) There are homogeneous torsion-free nil groups of type .0; : : : ; 0; : : : /. [Hint: Sect. 4,
Exercise 5 in Chapter 12.]

Proposition 2.6 (Wickless [1]). All rings on a group A are nilpotent if and only if
A D D ˚ C, where D is torsion divisible, and C is reduced torsion-free supporting
only nilpotent rings.
Proof. If all rings on A are nilpotent, then A cannot have any finite cyclic
summand, or contain a copy of Q, since these support non-nilpotent rings. Hence
A D D ˚ C, where D is torsion divisible and C is reduced torsion-free. It is clear
that only nilpotent rings can exist on C. Conversely, if A has such a decomposition,
then D is an ideal in any ring R on A, and R N D R=D is nilpotent. Then .R/N mD0
2m 2
(the mth power of the ring) for an m 2 N, thus .R/
.D/ D 0: t
u
Rings on Finite Rank Torsion-Free Groups Substantial results were obtained by
Beaumont–Pierce concerning rings on finite rank torsion-free groups. The highlight
is a generalization of the famous Wedderburn principal theorem which is stated next
without detailed proof. We write A D Q ˝ A, and identify 1 ˝ a with a 2 A; A is
a Q-algebra if A is a ring.
Theorem 2.7 (Beaumont–Pierce [1]). Assume A is a torsion-free ring of finite
rank, and let A D S ˚N be the Wedderburn decomposition of the Q-algebra A ,
where S is a semi-simple subalgebra and N is the radical of A . Then S D A\S
2 Rings on Groups 679

is a subring, and N D A \ N is the largest nilpotent ideal of A; they satisfy

Q ˝ S D S ; Q ˝ N D N .
Furthermore, S ˚ N is a subring of finite index in A.
Proof. Everything follows without difficulty from the classical Wedderburn theo-
rem, except for the last statement. That S ˚ N is a finite index subring requires a
long proof. (We can add that the additive group of S is quotient-divisible.) t
u
Role of Basic Subgroups We turn our attention to an interesting phenomenon
underlining the relevance of basic subgroups.
Proposition 2.8 (Fuchs [5]). Let A be a p-group, and B D ˚i2I hbi i a basic
subgroup of A. Any multiplication  2 Mult A is completely determined by the
set of the .bi ; bj / for all i; j 2 I.
Conversely, every choice of the elements .bi ; bj / for all i; j 2 I .subject to the
necessary condition that o..bi ; bj //
minfo.bi /; o.bj /g/ gives rise to a naring
on A.
Proof. There are several proofs to choose from, the following is most elementary
and direct. Needless to say, that if  is given for the basis elements of B (of course,
the values need not be in B), then it extends linearly to all of B. To find .a; a0 / for
all a; a 2 A, let o.a0 / D pk , and write accordingly a D b C pk x with b 2 B; x 2 A, to
obtain

.a; a0 / D .b; a0 / C .pk x; a0 / D .b; a0 / C .x; pk a0 / D .b; a0 /:

In a similar fashion, .b; a0 / D .b; b0 / for a suitable b0 2 B, establishing the first

part of our claim.
For the second part, note that, every b 2 B being a unique linear combination
of the bi , the ring postulates for .b; b0 / for all b; b0 2 B are readily checked. If 
is extended to the whole of A (in the way shown in the preceding paragraph), then
just a routine verification is needed to show that the extended  is well defined and
satisfies the requisite postulates. t
u
Example 2.9. Let B be the torsion-completion of the direct sum B D ˚i2I hbi i of cyclic p-groups.
Rings can be defined on B in various ways, we need to specify the products bi bj .
(a) Putting bi bj D bi or bj according as o.bi /  o.bj / or not.
(b) Order the index set I, and in case o.bi / D o.bj /, set bi bj D bi or D bj according as i  j or
i > j.
(c) Of course, bi bj can be chosen arbitrarily, e.g. bi bj D bk at random (subject to the necessary
order condition) or a linear combination.
In (a)–(c), B is a subring of the ring on B such that all the products belong to B. In case (b)–(c),
associativity is in doubt.
Example 2.10. Let A be a p-group with basic subgroup B. Then Mult A Š Hom.B ˝ B; A/; this
follows from Proposition 2.8.

Absolute Properties When dealing with rings on a given group A, an inevitable

problem is to find those subgroups of A that enjoy certain property P, like being an
ideal, or an annihilator, in every ring supported by A. We shall call them absolute P
on A.
680 18 Groups in Rings and in Fields

We only consider absolute ideals, i.e. subgroups that are definitely (one- or two-
sided) ideals in every ring on A; evidently, problems for left-, right-, or two-sided
ideals are identical, because the anti-isomorphic ring is on the same group. We
observed above that the fully invariant subgroups of A enjoy this absolute property,
but we will see in a moment that not only these subgroups. To identify the absolute
ideals, we define an ideal of End A which plays a decisive role in the answer. This
ideal was introduced by Fried [1], so we shall refer to it as the Fried ideal F of
End A. It is defined as the trace of A in End A:

F D h Im  j  2 Hom.A; End A/i:

To show that F is in fact an ideal in End A, notice that for  2 End A, the mappings
a 7! .a/ and a 7! .a/ are homomorphisms A ! End A, as is readily checked.
Thus .a/ and .a/ are among the generators of F for all a 2 A, so F is in fact
an ideal of End A.
Example 2.11. (a) (Fried [1]) Let A be a reduced p-group, so End A is a reduced algebraically
compact group. In this case, F is the torsion subgroup of End A.
(b) Let A D Z.p1 /, so End A Š Jp . Thus F D 0.
(c) If A is a torsion-free group of rank one, then F D End A or 0 according as t.A/ is idempotent
or not.

Theorem 2.12 (Fried [1]). A subgroup C of a group A is an absolute ideal if it is

invariant under the Fried ideal F of End A, i.e. FC
C. The converse holds if we
admit narings on A as well.
Proof. Let R be a (na)ring on A. We associate with a 2R the left multiplication
a W x 7! ax .x 2 A/. The correspondence  W a 7! a is a homomorphism of A
into End A, and hence into F. A subgroup C of A is a left ideal in R exactly if every
a carries C into itself. The same argument applies to the right multiplications. The
conclusion is that if FC
C, then C is an absolute ideal of A.
Conversely, every  2 Hom.A; End A/ gives rise to a ring multiplication by
defining the product of a; c 2 A as ac D .a/c (which may be non-associative).
Therefore, if C is an ideal in every naring on A, then .a/c 2 C for c 2 C, i.e.
FC
C. t
u
Example 2.13. (a) F D 0 means that all subgroups are absolute ideals. This is the case, e.g., for
Jp ; Z.p1 /, and for rational groups of non-idempotent types.
(b) In a p-group A, all subgroups of A1 are absolute ideals.
F Notes. Borho [1] investigates the number n of non-isomorphic associative rings (identity
not required) on finite rank torsion-free groups. He shows that only n D 1; 2; 3; and 1 are
possible. n D 1 holds for nil groups, n D 2 only for Q, and n D 3 for Q ˚ R where R is a
rational group of non-idempotent type. For more information about rings on torsion-free groups of
finite rank, see Freedman [2], Niedzwecki [1], and on mixed groups, Jackett [2].
Arnold [4] proves that if RC is a finite rank Butler group, then NC is Butler and RC =NC is
almost completely decomposable (N denotes the nil radical). Feigelstock [1] considers rings R that
are T-nilpotent (i.e., for any sequence ai 2 R .i < !/, there is n < ! such that a1 a2    an D 0),
and shows that a torsion-free group cannot have the property that all rings on it are T-nilpotent.
2 Rings on Groups 681

Beaumont–Lawver [1] study groups supporting semi-simple rings, Haimo [2] investigates groups
that support radical (non-zero) rings. For rings on completely decomposable groups, see Ree–
Wisner [1] and Gardner [1]. Kompantseva [1] studies Mult on reduced algebraically compact
groups.
Absolute properties of subgroups were investigated in several papers. In particular, absolute
Jacobson radicals were described in special cases. Eklof–Mez [1] show that on a reduced p-adic
algebraically compact group A, the subgroup pA is the absolute Jacobson radical. For absolute
nil-ideals, see Kompantseva [2].
Bergman [1] and Hill [14] investigate the additive groups of rings generated by idempotents.
Interestingly, they are always ˙-cyclic.

Exercises

(1) Mult A is a bimodule over End A.

(2) (a) The group Mult D is torsion-free and divisible for divisible D.
(b) All multiplications on a torsion-free divisible D are associative if and only
if rk D
1.
(3) (a) Mult Jp Š Jp for every prime p (cf. Example 1.10 in Chapter 8).
(b) All rings (identity not required) on Jp are isomorphic to one of the
following rings: pk Jp .k < !/ and the zero ring on Jp .
(4) (a) A homomorphism  W A ! End A defines an associative ring structure
on the group A if and only if it satisfies .x/.y/ D ..x/.y// for all
x; y 2 A.
(b) In order to get an associative ring on A, Im  ought to be a subring in
End A. If  is injective, then this condition is sufficient as well.
(5) (Fried) For a subgroup B of A, define B D fa 2 A j Fa
Bg.
(a) Show that B is a subgroup of A.
(b) B is an ideal in every naring over A if and only if B
B if and only if
there is a fully invariant subgroup C of A such that C
B
C .
(6) The absolute annihilator of a torsion group A contains A1 , and there is an
associative and commutative ring on A whose annihilator is exactly A1 .
(7) (a) Every element in the Frattini ideal pR of a p-ring R generates a nilpotent
ideal of R, but the ideal pR itself need not be nilpotent.
(b) (Jackett) There is an associative and commutative ring on a p-group A
such that pA is exactly the set of its nilpotent element. (Thus pA is the nil
radical.)
(c) The absolute Jacobson radical of a p-group A contains pA.
(8) Let A be a p-group such that there are at most countably many non-isomorphic
rings on A. Prove that A is a nil group. [Hint: argue with the basic subgroup.]
(9) Let R be a p-ring, and B a basic subgroup of RC . If every element of B is
nilpotent, then the same holds for R. [Hint: for a 2 R there is a b 2 B with
ak D bk .]
(10) (Feigelstock) A p-group A has the property that every associative ring on A is
commutative if and only if the reduced part of A is cyclic.
682 18 Groups in Rings and in Fields

3 Additive Groups of Noetherian Rings

We proceed to investigate the additive structures of some important types of rings.

Our study opens with the class of noetherian rings.
Nilpotent Noetherian Rings We begin with the nilpotent case.
Lemma 3.1 (Szele [6]). A group is the additive group of a nilpotent left or right
noetherian ring if and only if it is finitely generated.
Proof. Let R be a noetherian ring with .R/k D 0 (kth power) for some k 2 N. The
subgroups between .R/iC1 and .R/i .i D 1; : : : ; k  1/ are ideals of R, thus they
satisfy the maximum condition. Equivalently, .R/i =.R/iC1 is finitely generated as
a group. Therefore, also RC is finitely generated. Conversely, the zero-ring over a
finitely generated group is nilpotent noetherian. t
u
Mixed Noetherian Rings Split We cannot tell much about the additive structure
of a general noetherian ring R, besides its torsion subgroup T. The subgroups RŒn

are ideals for all n 2 N, thus there is a maximal one, say, RŒm
, among them.
Then necessarily T D RŒm
; mT D 0, thus RC D TC ˚ C for some torsion-
free subgroup C. This C is isomorphic to the additive group of the noetherian ring
R/T (but it need not be a ring as products of its elements need not belong to C).
Hence
Lemma 3.2. A group A is the additive group of a left or right noetherian ring if
and only if A D T ˚ C, where T is a bounded group, and C is torsion-free that can
carry a noetherian ring structure.
Proof. The converse will follow from Lemma 4.3. t
u
In view of this lemma, the study of additive groups of noetherian rings may
be restricted to the torsion-free case. A further reduction to the reduced case is
immediate as soon as we notice that there are noetherian rings, even fields, on every
torsion-free divisible group.
Example 3.3. The additive group of an algebraic number field of degree n gives an example of a
field with additive group Š Q.n/ , while for an infinite cardinal , the quotient field of a polynomial
ring over Q with  indeterminates yields an example with additive group Q./ .

Our present knowledge of noetherian rings on reduced torsion-free groups does

not go beyond a few elementary remarks.
(A) For every left ideal L of a noetherian ring R, we have nL
L for some n 2 N:
L =L is bounded for the purification L of L.
This follows from the ascending chain condition on L =L.
(B) The types of elements in R satisfy the minimum condition.
In fact, to a properly descending chain t1 >    > tn > : : : of types of elements
in R, there corresponds a properly ascending chain R.t1 / <    < R.tn / < : : :
of ideals of R.
3 Additive Groups of Noetherian Rings 683

Noetherian Rings on Free Groups The next theorem shows that arbitrarily large
free groups support noetherian domains.
Theorem 3.4 (O’Neill [1]). For every cardinal , the free group on  generators
supports a commutative noetherian domain.
Proof. If  D n is an integer, then a ring of algebraic integers of degree n is a
Dedekind domain whose additive group is a free group on n generators.
So suppose  is infinite, and x . < / are indeterminates. Let P denote
the polynomial ring in these indeterminates with coefficients in Z. The primitive
polynomials (i.e., with 1 as the gcd of coefficients) form a multiplicative semigroup
S in P, and define the ring R as the localization of P at S. That is, R consists of
all fractions f =g where f ; g 2 P and g is primitive. Since every polynomial in R is
the product of an integer and primitive polynomials, from the definition it should be
evident that the only ideals of R are nR for integers n  0. This shows that R is a
(commutative) noetherian integral domain.
It remains to show that RC is a free group. That the additive group PC is
free follows from the unique representation of polynomials as sums of different
monomials. We break the proof into several steps.
Step 1. Subrings R and subgroups G . For every ordinal 0 <  < , let R be the
set of all fractions f =g 2 R where f ; g contain explicitly only indeterminates x
with  < . In addition, let P D P \ R . Furthermore, let G be the subgroup
of R C1 which is generated by the fractions f =g subject to the conditions:
(i) f ; g 2 P C1 ;
(ii) f ; g have no common factors except ˙1;
(iii) either g contains a term with x , or else f is a monomial with x as a factor.
Then every fraction f =g 2 G satisfies (iii) even if f is not monomial, because
if the common denominator of a finite sum of generators contains no x , then
all the numerators in this sum must contain x as a factor.
Step 2. We claim: RC D ˚ < G . That the G generate RC requires no comment,
only the
P direct sum property needs a verification. To this end, we prove that
G \ < G D 0. Assume f =g D u=v ¤ 0 where either f or g contains
x , but u; v have only indeterminates x with  < . It is well known that P
is a unique factorization domain, so we may assume that f ; g have no common
factor ¤ ˙1. Then f v D gu shows that the factors of f are factors of u, so f
cannot contain x , and neither can g, a contradiction. Thus the sum is direct.
This argument also shows that RC C
 C1 D R ˚ G .
Step 3. G is free. In a finite rank subgroup H of R, let g 2 P be a common
denominator of elements in a maximal independent set. Then gH is a finite rank
subgroup in the free group PC , so it is finitely generated free. Here g can be
cancelled to argue that H itself is finitely generated free. Owing to Pontryagin’s
criterion, this allows us to conclude that G is free for countable ordinals .
We also observe that for   !, G Š G C1 is obvious, since they are
generated by using equipotent sets of indeterminates, so freeness is inherited
684 18 Groups in Rings and in Fields

when moving to the next ordinal. If  is a limit ordinal, then R D [< R

along with RC C
C1 D R ˚ G guarantees that R D ˚< G . This completes
the proof. t
u
F Notes. There are several minor results on the additive groups of noetherian rings in the
literature, but so far no relevant information has been found. Theorem 3.4 is an exception, since it
shows that no serious restriction can be expected.

Exercises

(1) Let R be a torsion-free ring, and D the ring on the divisible hull of RC . D is
noetherian if so is R, but it can be noetherian even if R is not.
(2) A homogeneous completely decomposable torsion-free group supports a
noetherian domain if and only if its type is idempotent.
(3) A finitely generated free group admits countably many pairwise non-isomorphic
rings; all noetherian.
(4) For every subgroup G of a finitely generated free group F, there is a ring R on
F such that the additive group of .R/2 is G.

4 Additive Groups of Artinian Rings

Next, we consider left artinian rings (with or without identity), i.e. rings in which
the left ideals satisfy the minimum condition. We will see that the minimum
condition has a profound impact on the additive structure—this should not be
surprising, since the artinian condition is very restrictive. Our main result will
establish a necessary and sufficient condition for a group to be the additive group of
an artinian ring.
Nilpotent Artinian Rings To begin with, we consider nilpotent artinian rings R,
that is, .R/k D 0 for some k 2 N (called the exponent of R). Their additive groups
are easily characterized.
Proposition 4.1 (Szele [6]). A group A is the additive group of a nilpotent artinian
ring if and only if A satisfies the minimum condition on subgroups. Thus A is a
torsion group characterized in Theorem 5.3 in Chapter 4.
Proof. In a nilpotent ring R of exponent k, every subgroup C satisfying .R/iC1

C
.R/i .i D 1; : : : ; k  1/ is an ideal of R. Hence if R is artinian, the minimum
condition is satisfied by the subgroups of .R/i =.R/iC1 for all i, and hence by those
of R. Conversely, the zero-ring on a group with minimum condition on subgroups
is artinian (and nilpotent). t
u
Preliminary Lemmas Removing the hypothesis of nilpotency, we turn our atten-
tion to artinian rings in general. Their study starts with two preliminary lemmas.
4 Additive Groups of Artinian Rings 685

Lemma 4.2. The additive group of a field is a torsion-free divisible group or an

elementary p-group according as the characteristic of the field is 0 or p. Every such
group is the additive group of a field.
Proof. The fields are vector spaces over Q or Z=pZ, depending on their charac-
teristic. The existence of finite algebraic extensions of degree n and extensions of
transcendence degree  of the prime fields (for any cardinal ) prove the second
assertion. (See also Example 3.3.) t
u
The following lemma is less obvious.
Lemma 4.3. Let A Š ˚ Z.pk / with an infinite cardinal , for a fixed k 2 N. There
exists a commutative ring R with identity on A such that R; pR; : : : ; pk R D 0 are
the only ideals of R.
Proof. Let G be a totally ordered abelian group with jGj D . Define R as the formal
Laurent series
P ring over the ring Z=pk Z for G. That is, R consists of all formal power

series f D 2G n x with coefficients n 2 Z=pk Z, subject to the condition that
the set of exponents of the non-zero terms is well-ordered in the total order of G.
Of course, the operations in R are the same as for power series. It is well known that
these formal Laurent series form a field provided that the coefficients are taken from
a field. Using this property for Z=pZ, it is clear that there is no ideal between pR and
R, and hence it follows that the only ideals in R are the obvious R; pR; : : : ; pk R D 0.
A somewhat simpler argument is as follows. Let I be an index set of cardinality
, T the polynomial ring over Z=pk Z in indeterminates xi with i 2 I, and R the ring
T localized at the semigroup of all primitive polynomials; here ‘primitive’ simply
means that not all the coefficients are divisible by p. It is immediate that the additive
group of T, and hence of R, is a direct sum of as many copies of Z.pk / as jIj, and it
is also easy to verify that the only ideals of R are R; pR; : : : ; pk R D 0. t
u
Characterization of the Additive Structure The next result gives us full insight
into the additive structure of artinian rings.
Theorem 4.4 (Szele–Fuchs [1]). A group A is the additive group of a .left/ artinian
ring .commutative artinian ring/ if and only if it is a direct sum A D B ˚ C ˚ D
where
(i) B is a bounded group;
(ii) C is a finite direct sum of quasi-cyclic groups Z.p1 / for equal or different
primes p;
(iii) D D ˚ Q is a -dimensional Q-vector space,  any cardinal.
Proof. Suppose R is an artinian ring. In view of the minimum condition, the set of
ideals nR .n D 1; 2; : : : / contains a minimal one, say, mR D L. This must satisfy
pL D L for each prime p, showing that L is divisible as a group. As such, it is,
group-theoretically, a summand: RC D LC ˚ B holds for some subgroup B. Then
LC D mRC D mLC ˚ mB, whence mB D 0 follows. We conclude that RC is a
direct sum of a divisible group LC and a bounded group B. Write LC D C ˚ D,
where C is torsion divisible and D is torsion-free divisible. To complete the proof
686 18 Groups in Rings and in Fields

of necessity, we need only show that C is of finite rank. But this is an immediate
consequence of Sect. 1 (F), since C is annihilated by L and B, so all subgroups in C
are ideals, thus C must satisfy the minimum condition.
Conversely, assume A has the indicated structure. We then write A D B1 ˚    ˚
Bn ˚C ˚D, where C; D are as in (ii)–(iii), while each Bi is a homogeneous ˙-cyclic
group. To build a ring over A, we invoke Lemma 4.3 to get rings with finitely many
ideals on each Bi , and define the zero-ring on C, and a field on D. This yields a
commutative artinian ring on A (which has an identity exactly if C D 0). t
u
Note that the summand C is contained in the annihilator ideal of any artinian ring
on A. From the proof it is clear that the one-sided ideals of a (left) artinian ring have
the same kind of additive structure as the ring.
We can derive some not so obvious ring-theoretical consequences from the last
theorem.
Theorem 4.5 (Szele–Fuchs [1], Szász [2]). Every left artinian ring R is the ring-
theoretical direct sum of a torsion-free left artinian ring S, and a finite number of
left artinian p-rings Tp , belonging to different primes p:

R D S ˚ Tp1 ˚    ˚ Tpm :

Every torsion-free left artinian ring, in particular, S, has a left identity.

Proof. We start with the proof of the second claim. Since a torsion-free artinian
ring S cannot be nilpotent, ring theory tells us that it must contain an idempotent
e ¤ 0 such that e C N is the identity element in the semi-simple artinian ring
S=N (N denotes the nilpotent radical of S). By Theorem 4.4, the additive group
of every left ideal of S is divisible, so for every a 2 S, the left ideal generated
by a0 D a  ea is divisible. Therefore 12 a0 D na0 C ba0 for some n 2 Z; b 2 S,
whence .2n  1/a0 D 2ba0 and a0 D ca0 with c D 2.2n  1/1 b 2 S. We obtain
e.ca0 / D ea0 D 0, which implies a0 D .c  ec/a0 D    D .c  ec/k a0 for every
k > 1. Since c  ec 2 N and N is nilpotent, we get .c  ec/k D 0 for some k 2 N,
and so a0 D 0. This means that a D ea for every a 2 S, i.e. e is a left identity for S.
To prove the first claim, note that the torsion part T of a left artinian ring R
is an ideal in R, and is the direct sum of its p-components Tp , where in view
of Theorem 4.4 only a finite number of them can be different from 0. Thus RC D
D ˚ TC where D is a torsion-free divisible subgroup. By what has been proved,
the factor ring R=T has a left identity e C T where e 2 D may be assumed (so that
eT D 0 D Te). Setting S D eR, we have the Peirce-decomposition a D eaC.aea/
where ea 2 S; a  ea 2 T; thus SC Š D as both are complements of T. Now S is a
subring—this is clear from .S/2 D .eR/2
eR D S: The direct sum R D S ˚ T
holds also in the ring-theoretical sense, since the torsion and torsion-free parts of an
artinian ring are subrings annihilating each other. t
u
Thus every left artinian ring is the direct sum of a finite number of (indecom-
posable) left artinian rings, each of which is either torsion-free or a p-ring. An
immediate consequence is that the structure theory of artinian rings can be reduced
4 Additive Groups of Artinian Rings 687

to those of torsion-free and p-rings. Since quasi-cyclic groups contribute next to

nothing to the structure of artinian rings, among artinian p-rings only the bounded
ones are of real interest.
Embedding in Artinian Ring with Identity We turn our attention to two basically
ring-theoretical questions concerning artinian rings: when can an artinian ring be
embedded in an artinian ring with identity? And, when is an artinian ring
noetherian? Though these are actually questions that should be treated within ring
theory, their discussion in this volume is justified by the unusual situation that not
only their full solutions rely heavily upon group-theoretic reasoning on the additive
structure of the ring, but even the actual conditions are in group-theoretic terms.
It is a well-known classical result that every ring without identity can be
embedded in a ring with identity (extending it by Z). However, in most cases
such an embedding changes the ring structure to a great extent (it certainly will
not be artinian). A different method helps to get some extended rings also artinian,
however, it turns out that it is not possible to embed every artinian ring in an artinian
ring with identity. The precise result on embeddability is as follows.
Theorem 4.6 (Szele–Fuchs [1]). A left artinian ring can be embedded in a left
artinian ring with .one- or two-sided/ identity if and only if it does not contain any
quasi-cyclic subgroup.
Proof. To prove the ‘only if’ part, suppose R is artinian with an identity. From The-
orem 4.4 we conclude that the quasi-cyclic subgroups of R are always contained in
the annihilator of the ring. The annihilator ideal is evidently 0 whenever a left or
right identity is present. Thus the stated condition is necessary.
Conversely, suppose R is artinian without quasi-cyclic subgroups. By Theo-
rem 4.5, R is then the direct sum of a torsion-free divisible artinian ring S and a
finite number of artinian rings Tp whose additive groups are direct sums of cyclic
groups of the same prime power order. If S or Tp fails to contain an identity, then we
can adjoin one as follows. The set S D .Q; S/ under the coordinate-wise addition
and under the usual multiplication rule

.m; r/.n; s/ D .mn; ms C nr C rs/ .r; s 2 S; m; n 2 Q/

is a ring that is artinian with identity .1; 0/. Similarly, for a pk -bounded Tp , the
set .Z=pk Z; Tp / under similar operations is artinian with identity .1 C pk Z; 0/. The
direct sum of these rings with identity provides the desired embedding. t
u
When Artinian Implies Noetherian Passing to the second question we raised
above, the answer is provided by the following theorem.
Theorem 4.7 (Szele–Fuchs [1], Fuchs [8]). A left artinian ring is left noetherian
if and only if it contains no quasi-cyclic subgroup. Equivalently, if and only if its
annihilator is finite.
Proof. Each quasi-cyclic subgroup in an artinian ring R is contained in the
annihilator of the ring, and every subgroup of the annihilator is an ideal of R. Since
688 18 Groups in Rings and in Fields

quasi-cyclic groups violate the maximum condition on subgroups, a left artinian

ring that is left noetherian cannot contain any quasi-cyclic subgroup.
Next suppose R is left artinian and has no quasi-cyclic subgroup. In order to
prove that it is left noetherian, we show that R as a left R-module is of finite length
(it has a finite composition series). Let N denote the nilpotent radical of R; hence,
.N/k D 0 for a minimal k 2 N. If N D R, then R is nilpotent, so Proposition 4.1
implies the minimum condition on subgroups. From the absence of quasi-cyclic
subgroups we conclude that R is finite in this case. Our attention now narrows to the
case when

R D N0 > N > .N/2 >    > .N/k1 > .N/k D 0

is a properly descending chain of ideals. The factor modules Mi D .N/i1 =.N/i .i D

1; : : : ; k/ are annihilated by N, they can therefore be viewed as left modules over the
semi-simple artinian ring R=N which we shall denote by R0 . It is a classical result
that a module over a semi-simple artinian ring decomposes into the direct sum of
simple (unital) submodules and a submodule annihilated by the ring. This applies
to the Mi where the submodules correspond to left ideals of R. Thus every Mi is a
direct sum of a finite number of simple R0 -modules and a finite submodule. Hence
it is obvious that R is of finite length, so it is definitely noetherian.
The annihilator ideal of an artinian-noetherian ring must satisfy both the mini-
mum and the maximum conditions on subgroups, so it must be finite. Conversely,
if the annihilator is finite in an artinian ring, then it cannot contain any quasi-cyclic
subgroup (that are annihilators in artinian rings), so the ring is noetherian by the
preceding arguments. t
u
An obvious corollary is a frequently cited theorem of C. Hopkins: a left artinian
ring with a left or right-sided identity is noetherian.
F Notes. In the theory of additive structures of rings, the most complete results are related
to artinian rings. The results shed more light also on some of their ring theoretical properties. It is
unusual that important ring properties can be derived from the additive structure.
The ring direct decomposition in Theorem 4.5 was extended to perfect rings by Huynh [1]. (A
ring is left perfect if its principal left ideals satisfy the minimum condition.)

Exercises

(1) If the ring is not artinian, then its quasi-cyclic subgroups need not belong to the
annihilator of the ring.
(2) (Szász)
(a) In a perfect ring, quasi-cyclic subgroups belong to the annihilator of the
ring.
(b) A group A is the additive group of a perfect ring exactly if it is the direct
sum of a divisible torsion-free group and a torsion group.
5 Additive Groups of Regular Rings 689

(3) If the torsion-free ring S in Theorem 4.5 is of finite rank, then it has a two-sided
identity.
(4) A nilpotent artinian ring is embeddable in an artinian ring with identity if and
only if it is finite.
(5) Let M be a (not necessarily unital) left module over a ring R such that the
submodules of M satisfy the minimum condition.

(a) The group of M C is the direct sum of a divisible and a bounded group.
(b) If R is artinian, then M has the maximum condition on submodules exactly
if it contains no quasi-cyclic subgroup.
(6) (Fuchs) Let R be a ring with restricted minimum condition on left ideals (i.e.,
minimum condition holds modulo every non-zero left ideal).
(a) If R contains elements ¤ 0 of finite order, then its additive structure is the
same as in Theorem 4.4. [Hint: RŒp
.]
(b) If R is torsion-free, then it is homogeneous of idempotent type. [Hint: ideals
of the form \i pki i R.]

5 Additive Groups of Regular Rings

Our program for this section is to study the additive structures of (von Neumann)
regular and generalized regular rings (associativity will be assumed, but not
identity). Significant information can be obtained in these cases about the additive
groups. To begin with, we recall the relevant definitions.
Structure of the Additive Group An element a of a ring R is said to be regular if
axa D a for some x 2 R, and m-regular for m 2 N if am is a regular element. A ring
R is regular or m-regular if each of its elements has this property. R is -regular
if every element of R is m-regular for some m 2 N (depending on the element).
Theorem 5.1 (Fuchs [5]). The additive group of a regular ring is the direct sum of
a torsion-free divisible group and a reduced group C sandwiched between the direct
sum and the direct product of its p-components:
Y
˚p Tp
C
Tp :
p

TC C
p are elementary p-groups, and C=T is torsion-freeQ divisible.
There is a bijection between the regular subrings of p Tp =˚p Tp and the regular
rings sandwiched between the direct sum and direct product of the p-components Tp .
Proof. Let R be a regular ring. If pk ja 2 R, then p2k jaxa D a. Hence an element a in
a regular ring is either not divisible by the prime p or divisible by every power of p.
In the latter alternative, its order cannot be a power of p, since ax is then annihilated
690 18 Groups in Rings and in Fields

by a. Hence the p-component Tp of R must be an elementary p-group, and as such

it is a summand: R D Tp ˚ Rp . This is, moreover, a ring-direct sum, since Rp is
p-divisible. This is so, since for every b 2 Rp , the element pb is divisible by every
power of p, and as Rp contains no elements of order p, b itself must be divisible by
p. It also follows that Rp is a ring, and that a torsion-free regular ring is divisible.
Clearly, if q is a prime ¤ p, then the q-component Tq is contained in Rp , and a
repeated application of the last direct sum decomposition yields the ring-direct sum
decomposition

R D Tp1 ˚    ˚ Tpk ˚ R0 ; (18.2)

where p1 ; : : : ; pk are different primes, and multiplications by these primes are

automorphisms of the complement R0 D Rp1 \    \ Rpk .
Next consider D D \p Rp with p running over all primes; it is obviously a torsion-
free ideal in R. From what has been said before, it is clear that D is divisible: each
d 2 D is divisible by p, and thus p-divisible. Write RC D DC ˚ C where C denotes
a reduced group containing T D ˚p Tp . For each p, (18.2) defines a projection p W
R ! Tp such that \p Ker p D \p Rp D D: We conclude that R=D is isomorphic to
a ring-theoretical subdirect sum of the rings Im p D Tp . Now R=T—as a torsion-
free regular ring—is divisible, and so is its pure subgroup C=T. Thus we arrive at
the first part of the claim.
From the above discussion it is clear that every regular ring R between the direct
sum and the direct product defines a regular ring in R/T. Conversely, the complete
inverse image of such a ring is a regular ring between the direct sum and the direct
product, as a ring that contains a regular ring as an ideal with regular factor ring is
itself a regular ring. t
u
Q
Example 5.2. If C D ˚p Tp or C D p Tp ; then there is a regular ring with additive group C: just
choose for Tp any field of characteristic p. (See also the ring M infra.)

Generalized Regular Rings In sharp contrast to regular rings, the additive groups
of m-regular (m  2/ and -regular rings can be arbitrary, since zero-rings share all
these properties. However, the situation changes drastically if the rings are supposed
to have an identity, or are just (left) ideals in m-regular, etc. rings with identity. Here
we will restrict ourselves to the discussion of -regularity.
Theorem 5.3 (Fuchs–Rangaswamy [1]). If L is a left ideal in a -regular ring R
with identity, then
(i) for every prime p, the p-component Lp of L is bounded, and L is a ring-direct
sum L D Lp ˚ pm L For some m 2 N;
(ii) for the torsion subgroup T of L, L=T is a divisible torsion-free -regular ring.

Proof. (i) The -regularity of R implies .p1/mx.p1/m D .p1/m , i.e. p2 m x D pm 1

holds for some x 2 R and m 2 N. Hence, p2 m xa D pm a for every a 2 L,
5 Additive Groups of Regular Rings 691

showing that every pm a is divisible by p2 m in the principal left ideal Ra. We

derive that p2 m L D pm L, and pm L is p-divisible. Hence if a 2 Lp is of order
pk , then hp .pm a/  !, thus for some y 2 R, pm a D pk .ya/ D y.pk a/ D 0,
whence pm Lp D 0 is obtained. Hence Lp is bounded, so LC D LC p ˚ C for some
subgroup C of L. Notice that pm L D pm C is p-divisible, and division by p in
C is unique. Consequently, C D pm C, and we obtain that L D Lp ˚ pm L is a
ring-theoretical direct sum. This completes the proof of (i).
(ii) All that we have to do is to observe that L/T is an epic image of the p-divisible
rings pm L, and -regularity is inherited by surjective images. t
u

Note that the integer m in the preceding theorem depends only on prime p, and is
the same for all left ideals of the -regular ring.
Embedding in Regular Ring with Identity We wish to solve the problem of
embedding of regular rings in rings with identity preserving regularity. We wish to
show that there is always such an embedding. Oddly enough, this is another purely
ring-theoretical question that is answered by taking full advantage of our knowledge
of the additive groups.
To start with, we construct a commutative regular ring M with identity as follows.
For every prime p, take the prime field Fp Q D Z=pZ of characteristic p; let p
denote its identity element. Evidently, F D p Fp is a commutative regular ring
with identity  D .: : : ; p ; : : : /. The quotient F= ˚p Fp is a torsion-free divisible
regular ring in which the coset of  generates a pure subgroup M= ˚p Fp isomorphic
to Q as a ring as well. This M is a regular ring: it contains the regular ring ˚p Fp as
an ideal modulo which it is regular.
It might be of interest to point out that the ring F is the completion of M, and its
ring structure is completely determined by M.
Proposition 5.4 (Fuchs–Halperin [1]). Every regular ring is a unital algebra over
the commutative regular ring M.
Proof. Let R be a regular ring. To define the action of x 2 M on a 2 R, we write
x D .: : : ; xp ; : : : / with xp 2 Fp , and notice that, by construction, there is a rational
number mn1 .m; n 2 N/ such that nxp  m mod p for almost all primes p. Select
a finite set fp1 ; : : : ; pk g of primes which includes all prime divisors of m; n as well
as the primes for which the last congruence fails to hold. With such a set of primes,
we make a ring-decomposition

M D Fp1 ˚    ˚ Fpk ˚ M0 ; (18.3)

where M0 is an ideal of M such that multiplication by each pi .i D 1; : : : ; k/ is an

automorphism of M0 . Accordingly, we have

x D xp1 C    C xpk C x0 .xpi 2 Fpi ; x0 2 M0 /:

692 18 Groups in Rings and in Fields

Now R has a decomposition R D Tp1 ˚    ˚ Tpk ˚ R0 ; corresponding to (18.3).

If we write accordingly a D ap1 C    C apk C a0 with api 2 Tpi ; a0 2 R0 , then we
can define

xa D xp1 ap1 C    C xpk apk C mn1 a0 : (18.4)

Take into account that Tp is an Fp -vector space, so that xp ap makes sense for every
p, and so does mn1 a0 by virtue of the choice of the primes (multiplication by n
is an automorphism on R0 ). We still have to convince ourselves that xa does not
change if we select a larger set of primes or use a different form for mn1 . But this
follows right away from our selection of primes which guarantees that each xp with
p … fp1 ; : : : ; pk g acts on Fp as a multiplication by mn1 . We leave it to the reader to
check the algebra postulates to conclude that R is an M-algebra, indeed. t
u
By taking full advantage of the last result, it becomes easy to verify the following
theorem.
Theorem 5.5 (Fuchs–Halperin [1]). Every regular ring can be embedded as an
ideal in a regular ring with identity.
Proof. Given a regular ring R (without identity), we form the set of pairs .x; r/ 2
M  R, and define operations as usual to make it into a ring R : addition: .x; r/ C
.y; s/ D .x C y; r C s/, and multiplication:

.x; r/  .y; s/ D .xy; xs C yr C rs/ with x; y 2 M; r; s 2 R:

Then .; 0/ is the identity in R , and r 7! .0; r/ .r 2 R/ embeds R in R as an ideal.

As both the ideal and the factor ring are regular, so is R . t
u
F Notes. We refer to additional literature to the embedding problem by various authors, see,
e.g., Jackett [1] where more information is given about the embedding between the direct sum
and the direct product. Proposition 5.4 and Theorem 5.5 are excellent examples to demonstrate
the relevance of the additive group in rings. No method of proof is known to me that avoids the
additive structure.
Results on -regular rings, similar to those in Proposition 5.4 and Theorem 5.5, were proved by
Fuchs–Rangaswamy [1], provided the additive group satisfies a necessary condition. The methods
were similar. For additional results, consult Feigelstock [Fe], where more additive structures are
examined. Feigelstock [2] deals with the additive groups of self-injective rings.

Exercises

(1) A group is the additive group of a boolean ring (every element is idempotent) if
and only if it is an elementary 2-group.
6 E-Rings 693

(2) A group is isomorphic to the additive group of a noetherian regular ring exactly
if it is a finite direct sum of a torsion-free divisible group and elementary
p-groups.
(3) Show that the torsion subgroup of a Baer ring is an elementary group (for
definition, see Sect. 4 in Chapter 16).
(4) The ring M above is the best choice in the following sense. If M’ is any
commutative regular ring R with identity such that every regular ring is a
unital M’-algebra, then there is an identity-preserving surjective homomorphism
 W M0 ! M with x0 a D .x0 /a for all x0 2 M0 ; a 2 R. Q
(5) Given n 2 N, construct a regular ring between ˚p Fp and p Fp whose torsion-
free rank is n 2 N (notation as before Proposition 5.4). [Hint: vectors with
solutions of the congruence xn  1 mod p as coordinates.]
(6) A torsion-free divisible m-regular ring without identity is an ideal in an m-
regular ring with identity.

6 E-Rings

Every ring with identity is a subring of the endomorphism ring of its additive group.
In [AG], one of the problems asked for the characterization of rings R for which
End RC Š R. In his dissertation, Schultz [1] made a good progress towards the
solution of the problem.
Properties of E-Ring Following Schultz, we say that R is an E-ring if every
endomorphism of RC is a left multiplication rP by an element r 2 R. Accordingly,
a group A is called an E-group if it supports an E-ring; this ring is uniquely
determined by A as it is isomorphic to End A.
We observe right away that every E-ring has an identity; furthermore, for an
E-ring R, the correspondence r 7! rP is a ring isomorphism R ! End RC , and
 7! .1/ . 2 End RC / is its inverse. Actually, the identity has a distinguished
role: R is an E-ring if and only if the only endomorphism of RC that carries 1 2 RC
to 0 is the zero endomorphism, see Theorem 6.3(ii).
Example 6.1. (a) The following are easy examples of E-groups: Z; Q; Z=nZ; and Q ˚ Z=nZ
for every n 2 N. Also, the rational groups of idempotent types are E-groups.
(b) Subrings of a finite algebraic extension of Q (containing 1) are E-rings. (Proof is omitted.)
(c) Examples of E-rings include all subrings of ZQ with 1, in particular, Jp for any prime p.

Example 6.2 (Douglas–Farahat). We can define a finite rank torsion-free E-ring on the group
A D hp11 a1 ; : : : ; p1
n an ; 21 .a1 C    C an /i (with different odd primes pi ) for any n
2, by
setting a2i D 2ai and ai aj D 0 for all i ¤ j. The ring identity is the element 21 .a1 C    C an /.

As a starting point, we prove the following lemma that gives several necessary
and sufficient conditions on a ring to be an E-ring.
694 18 Groups in Rings and in Fields

Theorem 6.3 (Bowshell–Schultz [1]). These are equivalent for a ring R with 1:
(i) R is an E-ring;
(ii) If  2 End.RC / satisfies .1/ D 0, then  D 0;
(iii) HomZ .R; R/ D HomR .R; R/ where R is viewed as a right module over itself;
(iv) End.RC / is commutative.
Proof. (i) ) (ii) By (i),  2 End.RC / is multiplication by some s 2 R. Thus
0 D .1/ D s  1 in RC , also in R. Thus s D 0, which means  D 0.
(ii) ) (iii) Let  2 HomZ .R; R/. If .1/ D s 2 R, then .  s/.1/ D 0, thus (ii)
implies  D sP. Hence .rt/ D srt D .r/t for all r; t 2 R.
(iii) ) (iv) First we prove that R is commutative: for r; s 2 R we have by (iii)
rs D sP.r/ D sP.1  r/ D sP.1/r D sr (dots denote now right multiplication). To
show that ;  2 End.RC / commute, note that .r/ D ..1/r/ D .1/.1/r
which is by the commutativity of R equal to .r/.
(iv) ) (i) For r 2 R;  2 End.RC /, we have .r/ D .Pr.1// D .Pr/.1// which
is—by commutativity—equal to rP .1/ D .1/r. Hence  is left multiplication
by .1/ 2 R, and R is an E-ring. t
u

We continue with listing a few pertinent properties of E-rings.

(A) Let R be an E-ring .so commutative/, and A D RC . There is a bijection

between the elements of A and rings .with or without units/ on A. For 0 ¤
a 2 A, define a new multiplication r ı s D ras (last product computed in R).
This gives rise to a ring structure on A with 1 ı 1 D a, so all these rings are
different. To see that there is no other ring structure S on A, suppose ? is a
ring operation in S. Then for a fixed r 2 R, s 7! r ? s is an endomorphism
of A, thus it is multiplication by some rN 2 R. Also, the correspondence r 7! rN
is an endomorphism of A, so rN D ra for some a 2 R. We conclude that
r ? s D rN s D ras, as claimed.
(B) If A is an E-group, and if  2 Mult A defines an E-ring on A, then a .x; y/ D
.a; .x; y// for all a 2 A yields all the elements of Mult A. This follows
from (A).
Observe that (B) shows that if A is an E-group, then every element of Mult A
represents an associative and commutative ring structure on A.
(C) For an E-group A, the Fried ideal is all of End A. Thus a subgroup in an E-
group is an absolute ideal if and only if it is fully invariant.
(D) Endomorphic images of an E-group are fully invariant subgroups. This is a
simple consequence of the commutativity of the endomorphism ring.
(E) An endomorphic image of the additive group of an E-ring R is the additive
group of a principal ideal of R, it is itself an E-group. Every endomorphism
is a multiplication by a ring element a 2 R, so it is of the form r 7! ra. If we
define multiplication on Ra via ra ı sa D ras .D rsa/ .r; s 2 R/, then we get
a ring (with identity a) on Ra.
6 E-Rings 695

(F) Every endomorphism of the additive group of a principal ideal Ra extends to

all of the E-group RC . By (E), Ra is an E-ring, so each endomorphism of its
additive group is a multiplication by some ra 2 Ra. The multiplication by ra
in R is a desired extension.
(G) Every direct decomposition of the additive group of an E-ring R is also a ring-
direct decomposition. By (D), the summands are fully invariant. Products of
elements from different summands must be 0 by the full invariance.
(H) The direct sum R ˚ S of two rings is an E-ring exactly if both R and S are
E-rings and Hom.RC ; SC / D 0 D Hom.SC ; RC /: One way, the implication
follows from (D) and (F). For the reverse implication observe that if R and S
are as stated, then the endomorphism ring of the direct sum is just the direct
sum of the endomorphism rings.
(J) The additive group of an E-ring cannot contain any pure subgroups of the
form Q ˚ Q, Z.pm / ˚ Z.pn / with m; n 2 N or Jp ˚ Jp for any prime p. These
subgroups would then be ring-direct summands and E-rings (see (D), (G)), but
their endomorphism rings are not commutative.
(K) If S is a ring with identity such that SC RC for a finite rank torsion-free
E-ring R, then S is likewise an E-ring. Since Q ˝ End SC Š Q ˝ End RC ,
both endomorphism rings are simultaneously commutative Theorem 6.3.
Special Cases In a few cases, E-rings with particular kinds of additive groups can
be completely classified.
Lemma 6.4 (Schultz [1]). Suppose R is an E-ring.
(i) If R is torsion, then R Š Z=nZ for some n 2 N.
(ii) If R is not reduced, then R Š Q ˚ Z=nZ for some n 2 N.
(iii) If RC is finitely generated, then R Š Z or Š Z=nZ with n 2 N.
Proof. (i)–(ii) If R is an E-ring, then by (H) all of its summands are E-rings. This
rules out quasi-cyclic groups Z.p1 /, and by (J) also more than one copy of Q. Again
by (H), if Q is a summand, then its complement must be a reduced torsion group.
Such a torsion group cannot contain two cyclic p-groups as summands for the same
p (see (J)), and so it must be a subgroup of Q=Z. The existence of identity shows
that it must be cyclic. RC in (i) (in (ii)) has commutative endomorphism ring, so R
is an E-ring.
(iii) In case RC is finitely generated, we argue that no direct decomposition may
contain two isomorphic copies of Z, or more than one cyclic p-group for the same
prime. The possibility Z ˚ Z=nZ is ruled out by calculating its endomorphism ring,
so RC must be as stated. t
u
We mention the following characterization of torsion-free E-rings that are of
finite rank.
Theorem 6.5 (Bowshell–Schultz [1]). A finite rank torsion-free E-ring R is quasi-
isomorphic to a ring-direct sum

R R1 ˚    ˚ Rn
696 18 Groups in Rings and in Fields

of E-rings Ri such that QRi is an algebraic number field, and, for all i ¤ j,
Hom.Ri ; Rj / D 0.
Proof. Sufficiency is clear in view of Example 6.1 and (H).
From Theorem 2.7 it follows that the direct decomposition of the Q-algebra
R D R ˝ Q D S ˚ N yields a subring A D S ˚ N of finite index in R, where S
is a subring such that S is a semi-simple Q-algebra, and N is the nil-radical of A.
If R is an E-ring, then by (K), so is A, and then by (G), so are S and N. As N cannot
have identity, it must be 0. The commutativity of R implies that S is a ring-direct
sum of finite rank fields Fi .i D 1; : : : ; n/. Such fields ought to be algebraic number
fields, so setting Ri D Fi \ R, the proof is complete. t
u
Satisfactory result exists also for cotorsion E-rings. Note that the next theorem
implies that the cardinality of such a ring cannot exceed 2@0 .
Theorem 6.6 (Bowshell–Schultz [1]). A reduced cotorsion E-ring R is a direct
sum of two cotorsion E-rings:
Q
(a) R1 D Qp2X Z.pkp / with kp 2 N, and
(b) R2 D p2Y Jp
where X; Y are disjoint sets of primes. Conversely, a direct sum of rings in (a) and
(b) with disjoint X; Y is an E-ring.
Proof. Assume R is a cotorsion E-ring. By Theorem 7.3 in Chapter 9 and (G), R
is the direct sum of an adjusted cotorsion ring R1 , and a torsion-free ring R2 on an
algebraically compact group. If T denotes the torsion subgroup of R1 , then by (J)
the p-components of T are cyclic, and therefore its cotorsion hull is as stated in (a).
Again by (J), R2 cannot have any summand of the form Jp ˚ Jp for any prime p,
whence (b) is clear.
For the converse, it is obvious that the rings in (a) and (b) are E-rings.
Since an E-ring cannot contain a direct sum Jp ˚ Z.pk / as a summand (it has
non-commutative endomorphism ring), the index sets X; Y must be disjoint. On the
other hand, if they are disjoint, the direct sum R1 ˚ R2 is obviously an E-ring. u t
Example
Q 6.7 (Bowshell–Schultz [1]). Let Ri .i 2 I/ be a set of E-rings such that Hom
. j¤i2I Ri ; Rj / D 0 for each j 2 I. In addition, suppose that there exist a prime p, and
Q
surjective ring maps i W Ri ! Z=pZ for each i. Then the subgroup of i2I Ri consisting
of all .: : : ; ri ; : : : / .ri 2 Ri / with i .ri / D j .rj / for all i; j 2 I is an E-ring R. This ring is
indecomposable if and only if all the Ri are indecomposable. Sufficiency follows from the easily
established fact that R contains no idempotents except for 0; 1.
F Notes. Pierce–Vinsonhaler [1] has an important collection of results on finite rank torsion-
free E-rings. See Vinsonhaler [2] for an informative survey of E-rings and E-modules.
The question of existence of arbitrarily large E-rings was for a while a major problem; it has
been answered by Dugas–Mader–Vinsonhaler [1] in the affirmative by using Shelah’s Black Box.
Dugas [4] proves that there exist arbitrarily large E-modules. Pierce [4] has numerous results on
E-modules. A theorem by Faticoni [1] shows that being an E-ring is not as restrictive as one might
think: every countable reduced torsion-free commutative ring is contained as a pure subring in an
E-ring.
Hausen–Johnson [2] define the E-ring core of a ring R by a transfinite process as follows. Let
C0 D R. If C has been defined for an ordinal , then let CC1 be the ring of those r 2 C for
7 Groups of Units in Commutative Rings 697

which multiplication by r is in the center of End.C /. For a limit ordinal  , set C D \< C .
There is an ordinal  with C D CC1 , and this subring (evidently an E-ring) is called the E-ring
core of R.
Absolute E-rings (i.e., remain E-rings in any generic extension of the universe) were investi-
gated by Göbel–Herden–Shelah [1]. Herden–Shelah [1] proved that absolute E-rings of cardinality
 exist if and only if  is less than the first Erdős cardinal.
Generalized E-rings were studied by Feigelstock–Hausen–Raphael [1]. A slight generalization
is when the isomorphy A Š .End A/C is not required to be induced by multiplication. Further
generalization is when only the existence of an epimorphism A ! .End A/C is assumed. It turns
out that this is not a proper generalization in case A is torsion-free of finite rank: A is then an
E-ring. Göbel–Shelah–Strüngmann [1] construct a group A, that is not an E-group, but satisfies
A Š .End A/C with a non-commutative endomorphism ring—quite an interesting phenomenon.
Dugas–Feigelstock [1] define A-ring as a ring R with 1 having the property that every
automorphism of its additive group is left multiplication by a unit of R. They show that A-rings
retain some of the features of E-rings, e.g. their unit groups U.R/ are abelian.

Exercises

(1) The torsion subgroup of an E-ring must be separable.

(2) Cotorsion E-groups are algebraically compact.
(3) (Schultz) Let A be a completely decomposable torsion-free E-group of finite
rank. Then the rank 1 summands are of incomparable idempotent types.
(4) Show that R1 in Lemma 6.4 contains pure subrings that are not cotorsion, but
are still E-rings.
(5) (Goeters) Let R be a torsion-free ring with rkp R D 1. If it is p-reduced, then
it is an E-ring. [Hint: RC embeds as a p-pure subgroup in Jp ,  2 End RC is
multiplication by  2 Jp .]
(6) (Faticoni) LetQ Ep be a collection of p-local E-rings for different primes p. Then
the ring E D p Ep is again an E-ring.

7 Groups of Units in Commutative Rings

In this section, all rings are associative and commutative with identity.
Unit Groups In every ring R with identity 1, the elements which have a (multi-
plicative) inverse with respect to 1 form a group under multiplication, called the
unit group of R. It will be denoted by U.R/. Thus 1 is the identity of the group
of units, and the symbol  will stand for the direct product. We will consider unit
groups sometimes as a multiplicative, some other time as being converted into an
additive group. From the context it will always be clear which operation is meant.
The correspondence F W R 7! U.R/ is functorial. In fact, if  W R ! S is a
ring homomorphism preserving identities, then  induces a group homomorphism
N W U.R/ ! U.S/: it is just the restriction map. It is remarkable that F has a
698 18 Groups in Rings and in Fields

left adjoint functor G which is defined by letting the group ring over the integers
correspond to a group, i.e. G W A 7! ZŒA
. Indeed, for each group A and each ring R,
there is a natural bijection between the set of homomorphisms A ! U.R/ and the
set of identity preserving ring maps ZŒA
! R.
Example 7.1. (a) The units in Z=pk Z, for a prime power pk , are the cosets coprime to p. Number
theory tells us that this is a cyclic group of order '.pk / D pk pk1 generated by any primitive
root mod pk (' is the Euler totient function).
(b) The units of the ring Z=nZ are represented by the residue classes that are coprime to n. The
unit group is a direct product of cyclic groups, its order is '.n/. It is cyclic if and only if
n D pk ; 2pk for odd primes p, or else n D 2; 4.
Example 7.2. The unit group U.Jp / is isomorphic to Z.p  1/ ˚ Jp for every odd prime p. This
will follow from Theorem 8.6.
Example 7.3. If R is the ring of integers in a finite algebraic extension of Q, then by Dirichlet’s
theorem on units, U.R/ is the direct product of a finite cyclic group of even order and a finitely
generated free group.

Before listing some informative facts about unit groups, we repeat: all rings to be
considered are commutative with 1, even if this is not stated explicitly.
(A) The unit group of a cartesian product of rings is the cartesian product of the
unit groups.
(B) The polynomial ring S D RŒxi
i2I over a domain R with any number of
indeterminates xi satisfies U.S/ Š U.R/. In fact, only a constant polynomial
may be a unit.
(C) The situation for formal power series rings is completely different. Let S D
RŒŒx

be the power series ring over a domain R in a single indeterminate x.

Then we have a direct product U.S/ Š U.R/  E where E is the multiplicative
group of all power series with constant term 1. This follows from the fact that
every unit of S is the product of a unit of R and an element of E.
(D) We now consider the localization RS of R at a semigroup S; suppose S contains
only non-zero divisors, and 1 2 S. Then U.RS / Š U.R/G where G denotes
the group of quotients of S.
(E) Let M be an R-module. The ‘idealization’ R.M/ is a ring with additive group
R ˚ M with the following rules of operations:

.r; a/C.s; b/ D .rCs; aCb/; .r; a/.s; b/ D .rs; rbCsa/ .r; s 2 R; a; b 2 M/:

Then .1; 0/ is the identity of R.M/, and .r; a/ is a unit if and only if r 2 U.R/.
As a result, we have U.R.M// Š U.R/  M.
An immediate consequence of (E) (with R D Z) is the following theorem.
Theorem 7.4. For every group A, there exists a ring with unit group isomorphic to
the additive group Z.2/ ˚ A. t
u
(F) Of special interest is the role the Jacobson radical J of R plays in connection
with U.R/. For all r 2 J, 1 C r 2 U.R/, and if J is regarded as a group under
the ‘circle’ operation r ı s D r C s C rs, then the correspondence r 7! 1 C r
is an isomorphism of .J; ı/ with a subgroup of U.R/.
7 Groups of Units in Commutative Rings 699

(G) The element 1 is always a unit of order 2, except when the additive group of
the ring is a 2-group. It is very seldom that U.R/ has odd order (see Ditor [1]).
Indeed, in this case RC is a 2-group, and the elements 1 C x with x 2 J form
a subgroup whose order is the order of the Jacobson radical J; hence J D 0.
This means that R is a semi-simple artinian ring on an elementary 2-group.
Therefore, R is a direct sum of a finite number of Galois-fields, and U.R/ is a
finite direct product of cyclic groups of orders 2n  1 for various n  1.
Groups that Can Be Unit Groups We shall now proceed to the problem of
describing those groups that can occur as unit groups in a commutative ring, and
give a brief account of the main results available in the literature.
It is almost trivial to prove (we do not need the not-so-obvious Theorem 7.4)
that every group A (now we have to use the multiplicative notation) embeds as a
subgroup in the unit group of some ring: just form the group ring ZŒA
of A over the
integers (or over any ring with 1). The elements a 2 A are evidently units in this
ring. Notably, the canonical embedding  W A ! U.ZŒA
/ (where .a/ D a for all
a 2 A) has the ‘universal’ property: if R is any ring with 1, and if ˛ W A ! U.R/
is any homomorphism, then there exists an identity preserving ring homomorphism
˛N W ZŒA
! R such that ˛ D ˛. N For the proof, it suffices
P to point P out that there is
one and only one way of extending ˛ to ZŒA
, viz. ˛N W ni ai 7! ni ˛.ai / .ni 2 Z/:
It is a delicate, difficult unsolved problem to characterize the groups that are the
unit groups in rings. In the domain case, more can be said.
Lemma 7.5. The torsion part of the unit group of an integral domain D is locally
cyclic. Conversely, every locally cyclic torsion group with an element of order 2 is
the torsion subgroup of the unit group of some domain.
Proof. Let u be a unit of order n 2 N in D. It is a root of an equation xn D 1 of
degree n. Just as in a field, also in a domain an equation of degree n can have at most
n roots, whence it follows that the p-socle of U.D/ contains at most p elements. This
forces a torsion group to be locally cyclic, i.e. isomorphic to a subgroup of Q=Z.
Conversely, let T be a torsion locally cyclic group with an element of order 2.
If T is infinite, then it is the union of cyclic groups of orders 2 mi (with increasing
mi ) which we can represent as the union of complex roots of unity of degrees 2 mi .
Correspondingly, we consider the tower of splitting fields Fi of polynomials x2 mi 1,
and define D as the set union of algebraic integers in these fields. Then D will contain
exactly the desired roots of unity, so it is a domain as desired. t
u
We have more specific information on unit groups of rings of algebraic integers
in finite algebraic extensions K of Q. By Dirichlet’s theorem, it is a direct product
of a finite cyclic group of even order and r1 C r2  1 infinite cyclic groups. Here r1
denotes the number of real, and r2 the number of pairs of conjugate complex roots
of the defining polynomial of the field K.
Example 7.6. (a) The unit group of the ring of Gaussian integers Z C iZ is cyclic of order 4: it
consists of the complex numbers ˙1; ˙i.
(b) The unit group of the ring of Eisenstein integers ZCZ (where 2 CC1 D 0) is isomorphic
to Z.6/ ˚ Z.
700 18 Groups in Rings and in Fields

We shall require a lemma, generalizing a result by Cohn [2].

Lemma 7.7. Let U be the unit group of an integral domain, and assume A is a
multiplicative group containing U such that A=U is torsion-free. Then A, too, is the
unit group of some domain.
Proof. Let R be a domain with U.R/ D U. We view A as an extension of U by A=U,
and select a representative ax 2 A in each coset x modulo U. We obtain a factor set
ux;y 2 U as defined by

ax ay D ux;y axy for all x; y 2 A=U: (18.5)

Define S as an algebra over R with the set fax gx2A=U as basis such that the basis
elements multiply according to the rule (18.5). Since the ux;y are coming from A,
they satisfy the obligatory associativity conditions to guarantee that S will become
an R-algebra. It is manifest that the multiples uax of the basis elements ax (where
u 2 U) form a subgroup of U.S/ canonically isomorphic to A.
It still remains to show that S is a domain, and U.S/ contains no elements other
than those of the form uax . Every torsion-free abelian group admits a linear order
compatible with the group operation. Choose an arbitrary, but fixed Pmlinear order
on A=U, and write a non-zero element  2 S in the form  D iD1 ri axi with
0 ¤ ri 2PR, and axi in the above set of representatives such that x1 <    < xm .
n
If  D jD1 sj ayj with 0 ¤ sj 2 R and y1 <    < yn is another element of
S, then in the product , ax1 y1 will be the smallest basis element with coefficient
r1 s1 ux1 ;y1 ¤ 0. Thus S has no zero-divisors. Since the largest basis element axm ;yn
in the product  will also have a non-vanishing coefficient, it follows that  D 1
only if m D 1 D n, and in addition, r1 ; s1 are units in R. t
u
Group Rings For group rings, the main difficulty lies in the emergence of the
so-called non-trivial units: units in RŒA
that are not products of units in R with
elements of A. There is an extensive literature on the unit groups of groups rings of
a group over a commutative ring. In this volume, we do not discuss group rings.
Example 7.8. Let A D hai be cyclic of order 5, and let Q.3/ denote the ring of rationals whose
denominators are powers of 3. Then Q.3/ ŒA
has non-trivial units. To prove this, consider D
1 C a C a2 C a3 C a4 2 Q.3/ ŒA
: Then a D , whence 2 D 5 follows. An easy calculation
2
shows that 1  2 is a unit with inverse 1  9 .
F Notes. Rings with cyclic groups of units were described by Gilmer [1] for finite rings, and
by Pearson–Schneider [1] for infinite rings.
It was G. Higman [Proc. London Math. Soc. 46, 231–248 (1938/39)] who started a systematic
study of unit groups in group rings; he investigated the units of group rings over finite algebraic
extensions of Z. In general, the study of the unit groups in (commutative) group rings is an
interesting area of research. Several significant results are available. The main contributors include
S. Berman, P. Danchev, G. Karpilovsky, W. May, T. Mollov, N.A. Nachev, W. Ullery. From the
point of view of abelian group theory, the results by Danchev are especially interesting: how the
structure of the unit group of the group ring over a p-group depends on the structure of the p-group
(e.g., if the group is totally projective).
8 Multiplicative Groups of Fields 701

To characterize the unit groups of commutative rings seems to be quite a difficult problem.
Recently several publications dealt with very special cases, even these turned out to be far from
easy. The relation between the unit group and the additive structure of the Jacobson radical is
certainly one of the crucial points to be investigated.

Exercises

(1) If R is an integral domain and A is torsion-free, then U.RŒA

/ Š U.R/  A.
(2) If U; V are unit groups of some rings, then so is their direct product U  V.
(3) The canonical embedding A W A ! ZŒA
is functorial: every group homomor-
phism ˛ W A ! B induces a ring homomorphism ˛N W ZŒA
! ZŒB
making the
following diagram commute:

Aχ
A −−−−−→ Z[A]
⏐ ⏐
⏐
α
⏐
ᾱ
Bχ
B −−−−−→ Z[B]

(4) The list of prime numbers that can be orders of unit groups in rings is: 2 and
p D 2n  1.
(5) Let R be a domain of characteristic 0 whose unit group is torsion. Then U.R/ Š
Z.2/ or Z.4/.
(6) Find the unit group in the following regular rings: F and M (for the notation,
see Sect. 5).
(7) What is the unit group of the bounded artinian ring in Lemma 4.3?

8 Multiplicative Groups of Fields

Our study of the multiplicative groups of units continues with the most important
special case when the ring is a field. Intuitively, it is expected that the characteristic
of the field will play a decisive role in the structure of its multiplicative group.
If K denotes a field, then the symbol K is used for its multiplicative group. For
obvious reason, we have to adhere to the multiplicative notation in K .
Torsion Groups in Fields It is easy to describe the torsion subgroups of fields.
Proposition 8.1. A torsion group is isomorphic to the torsion subgroup of K for
some field K of characteristic 0 if and only if it is locally cyclic with non-trivial
2-component.
Proof. See the proof of Lemma 7.5. t
u
702 18 Groups in Rings and in Fields

Turning to the case of finite characteristics, let us point out that if char K D p,
then 1 is the only root of the polynomial xp  1 D .x  1/p . Hence the p-component
of K is trivial. This is just an indication that the positive characteristic case must
be handled differently.
Special Fields In the following, we examine the multiplicative groups of the most
important fields.
1. Prime fields: Q and Fp D Z=pZ for each prime p. The fundamental theorem of
arithmetic states that every rational number ¤ 0 can be written uniquely in the
form ˙pk11 : : : pknn where pi are different primes and ki ¤ 0 are integers. Hence
Q is the direct product of h  1i and the cyclic groups hpi for all primes p.
We can now state that Q is the direct product of a cyclic group of order 2 and
a countably generated free group.
For prime characteristics, the existence of primitive roots mod p implies that
the multiplicative group is cyclic, so F p Š Z.p  1/.

2. Finite algebraic extensions of prime fields. Recall the fundamental theorem of

ideal theory from algebraic number theory: in the ring R of algebraic integers
in a finite algebraic extension of Q, a fractional ideal L ¤ 0 is a uniquely
determined product L D Pk11    Pknn with 0 ¤ ki 2 Z, where Pi are different
prime ideals. In other words, the non-zero ideals form a multiplicative group
that is a countably generated free group.
Theorem 8.2 (Skolem [1]). The multiplicative group of a finite algebraic extension
K of a prime field has the following form:
(i) If char K D 0, then K Š Z.2 m/  F, where m 2 N and F is a countably
generated free group;
(ii) If char K D p, then K Š Z.pn  1/ with n 2 N.
Conversely, every group of the form (i) or (ii) can be realized as the multiplicative
group of a finite algebraic extension of a prime field.
Proof. (i) Evidently, a; b 2 K generate the same (proper or fractional) ideal if
and only if ab1 is a unit in the ring R of algebraic integers in K. Hence  W a 7!
Ra is a multiplicative homomorphism of K into the group of ideals such that
Ker  is the group U of units in R. Consequently, K is an extension of U by a
subgroup of a (countable) free group, so it is of the indicated form.
(ii) The situation is totally different for prime characteristics, since a finite algebraic
extension of degree n of Fp is a Galois field of pn elements. The multiplicative
group of a finite field is cyclic, so K Š Z.pn  1/, as stated.
To verify the second part, note that the existence theorem on Galois fields settles
case (ii) at once. For case (i), it suffices to point out that from the proof of Lemma 7.5
it is clear that if  is a complex, 2 mth primitive root of unity, then the torsion
subgroup of Q./ is cyclic of order 2 m. t
u
8 Multiplicative Groups of Fields 703

3. Algebraically closed fields. In an algebraically closed field A, for every a 2 A

and for every prime p, the equation xp D a has p roots in A, except when
char A D p. Therefore, A must be a divisible group, and hence a direct product
of groups isomorphic to Q and Z.p1 /. The precise result is as follows.

Theorem 8.3. A group is isomorphic to the multiplicative group of an algebraically

closed field A if and only if it is of the form
(
 Q=Z ˚ D if char A D 0I
A Š
1
˚q¤p Z.q / ˚ D if char A D pI

where D is a torsion-free divisible group. D can be of any infinite cardinality, or, in

the second alternative only, also D D 0.
Proof. If char A = 0, then A must contain n roots of unity for every n, so it contains
a subgroup Š Q=Z. On the other hand, if char A D p, then—as noticed above—
the p-component of A is trivial. But otherwise, the same argument applies, with
the exception that A may be torsion (for algebraic extensions of the prime field).
As far as jAj is concerned, observe that the algebraic closure of a field of infinite
cardinality has exactly the same cardinality. t
u

4. Real-closed fields. Recall the definition: a real-closed field C is a field that

can be linearly ordered, but no proper algebraic extension admits linear order;
here, the order is meant to be compatible with the field operations: sums and
products of positive elements are positive. The only possibility is char C D 0.
p
From the theory of real-closed fields it is known that A D C. 1/, a quadratic
extension, is an algebraically closed field. As a consequence, all polynomials of
odd degrees must have a factor of degree 1 in CŒx
; in particular, extraction of
pth roots for odd primes p is possible throughout in C. But these pth roots are
unique, since ˙1 are the only roots of unity in C as it is shown by the following
argument: if 1 ¤  2 A is a pth root of 1, then 1 C  2 C    C  2.p1/ D
. 2p  1/=. 2  1/ D 0 shows that 1 is a sum of squares if  is in the field, thus
 … C. As far as square roots in general are concerned, we know that in a linear
order of C, the positive elements are complete squares, i.e. for every c 2 C, one
of c; c has a square root in C. The following theorem summarizes the facts.
Theorem 8.4. A group is isomorphic to the multiplicative group of a real-closed
field exactly if it is of the form

Z.2/ ˚ D

where D is a torsion-free divisible group of cardinality   @0 . Every infinite

cardinal  can occur. t
u
5. p-adic number fields. Recalling that every p-adic number ¤ 0 can be written
uniquely as pk  with k 2 Z and  a p-adic unit, it is clear that the multiplicative
704 18 Groups in Rings and in Fields

group of the p-adic number field is a direct product ZUp where Up is the group
of units in Jp . The units of the form  D 1 C s1 p C s2 p2 C : : : (0
si < p)
form a subgroup E of Up , and every unit is uniquely the product of an element
in the reduced residue class mod p and an element  2 E. Thus Up Š V  E
where V Š Z.p  1/ is the group of residue classes mod p prime to p.
We concentrate on E, and prove an auxiliary lemma. It is based on an ingenious
idea that goes back to K. Hensel, representing elements of E by their ‘logarithm’
values in Jp .
Lemma 8.5. Let p > 2 and e D gp1 , where g is a primitive root mod pn for
all n 2 N. Furthermore, let k denote the kth partial sum of the p-adic integer
 D s1 p C    C sn pn C : : : .0
sn < p/. The sequence ek .k < !/ converges to
an element  2 E, denoted e , and the correspondence e 7!  is an isomorphism
pJp ! E.
Proof. By number theory, there exists a primitive root g mod p that is also primitive
root modulo all powers of p (if g is an arbitrary primitive root for p, then either g or
g C p works for all pn ). Our choice guarantees that e D gp1 satisfies em  1 mod pk
for some m 2 N exactly if pk1 jm. For a p-adic integer  D s1 p C    C sn pn C    2
pJp , consider the sequence e0 ; es1 p ; : : : ; es1 pCCsk p D ek ; : : : . It converges to a p-
k

adic integer , since ek1  ek mod pk for all k 2 N. As   ek1 mod pk , we
have  2 E. We now show that the correspondence  7!  is a bijection.
For every  2 E and for every k  1, there is a unique integer k1 such that
0
k1 < pk1 and ek1   mod pk (with 0 D 0). Then uniqueness implies
k1  k mod pk1 , thus the integers k converge to a p-adic integer  such that
  k1 mod pk for all k 2 N. It is routine to check that the correspondence
 W  7! e is a homomorphism from the multiplicative group E into the additive
group pJp . t
u
In case p D 2, the number e D 5 has the property that em  1 mod 2k is tan-
tamount to the divisibility relation 2k2 jm. We can establish the same ‘logarithmic’
correspondence, the only difference is that it will be between E and 4J2 .
The preceding lemma and the remarks provide a proof for the following theorem.
Theorem 8.6 (K. Hensel). The multiplicative group of the p-adic number field Qp
is isomorphic to the additive group
(
 Z ˚ Z.p  1/ ˚ Jp if p ¤ 2;
K Š
Z ˚ Z.2/ ˚ J2 if p D 2:

F Notes. The problem of finding the multiplicative structure of fields is as old as algebraic
number theory. Dirichlet’s theorem on units was the first remarkable result, followed by Galois
fields, and by the multiplicative groups p-adic number fields by Hensel. Skolem’s theorem is of
much later origin, presumably in a less precise form it was known before (as all the ingredients
were available long before).
Naturally, the big question is to survey all groups that do occur as multiplicative groups of
fields. So far no satisfactory characterization has been given in terms of what algebraists would
8 Multiplicative Groups of Fields 705

regard satisfactory, namely, by a set of sentences of the first-order language of group theory. As a
matter of fact, S.R. Kogalowski [Dokl. Akad. Nauk SSSR 140, 1005–1007 (1961)] has shown that
the class of multiplicative groups of fields is not arithmetically closed in the sense of A.I. Malcev,
and so not axiomatizable; cf. also Sabbagh [1].
There are numerous remarkable results on the multiplicative structure of fields. To discuss any
of these results in detail would take us too far afield: the proofs are intricate arguments based on
theorems in field theory. A serious attempt to clarify the multiplicative structures of fields was
undertaken by W. May who proved several substantial theorems, of which the most significant is
perhaps the one that describes the multiplicative group up to a free factor: Given a group A whose
torsion subgroup is a subgroup of Q=Z and has a non-trivial 2-component, there exists a field K of
characteristic 0 such that K Š A  F with a free group F. Regretfully, we cannot go more deeply
into the subject, we cannot even sketch here the proof of this result; see May [1]. Another result of
his tells us about the change of unit groups under certain field extensions [3].

Exercises

(1) The additive and the multiplicative groups of a field are never isomorphic.
(2) (a) A finite group is isomorphic to the multiplicative group of a field exactly if
it is cyclic of order pn  1 for some prime p and integer n  1.
(b) A finite group is the torsion subgroup of the multiplicative group of a field
exactly if it is cyclic of even order or of the form indicated in (a).
(3) (Schenkman) Let N be the field generated by all algebraic numbers of degree

n, for a fixed n  2. Then N is a direct product of cyclic groups. [Hint: N
does not contain mth root of 1 for m > 4nŠ; use Pontryagin’s theorem.]
(4) A field K of characteristic p is called perfect if irreducible polynomials in KŒx

have only simple roots. Show that to be perfect it is necessary and sufficient that
K be p-divisible.
(5) (Sabbagh) The group Q ˚ Z.2/ is not isomorphic to the multiplicative group of
any field.

Problems to Chapter 18

PROBLEM 18.1. If a torsion-free group can support a left noetherian ring, can it
also support a two-sided (or a commutative) noetherian ring?
PROBLEM 18.2 (Niedzwecki–Reid [1]). Study the additive structure of a ring
modulo the pure subgroup generated by 1.
PROBLEM 18.3. Given A, define groups An .n < !/ by the rule: A0 D A, and
AnC1 D .End An /C . How long can this sequence be before it becomes stationary?
It becomes stationary at 0 if A is an E-group, and at 1 if A is a rigid group.

PROBLEM 18.4. Study the Picard groups of E-rings.

706 18 Groups in Rings and in Fields

PROBLEM 18.5. Characterize the unit groups of finite commutative rings.

See Pearson–Schneider [1].

PROBLEM 18.6. Characterize the unit groups of commutative von Neumann

regular and noetherian rings.
PROBLEM 18.7. Study the change of the unit groups under ring (field) exten-
sions.
References

Books

D.M. Arnold — [A] Finite Rank Torsion-free Abelian Groups and Rings. Lecture Notes in
Mathematics, vol. 931 (Springer, New York, 1982)
D.M. Arnold — [A1] Abelian Groups and Representations of Finite Partially Ordered Sets
(Springer, New York, 2000)
H. Cartan, S. Eilenberg — [CE] Homological Algebra (Princeton University Press, Princeton,
1956)
P.C. Eklof, A.H. Mekler — [EM] Almost Free Modules. Set-theoretic Methods, revised edition
(Elsevier, Amsterdam, 2002)
T. Faticoni — [Fa] Direct Sum Decompositions of Torsion-Free Finite Rank Groups. Pure Applied
Mathematics, vol. 285 (Chapman & Hall, Boca Raton, 2007)
T. Faticoni — [Fat] Modules over Endomorphism Rings. Encyclopedia of Mathematics and Its
Applications, vol. 130 (Cambridge University Press, Cambridge, 2010)
S. Feigelstock — [Fe] Additive Groups of Rings, vol. I and II. Research Notes in Mathematics,
vols. 83 and 169 (Pitman Advanced Publishing Program, Boston, 1983, 1988)
L. Fuchs — [AG] Abelian Groups. (Akadémiai Kiadó, Budapest, 1958, Pergamon Press, London,
1960, 1967)
L. Fuchs — [IAG] Infinite Abelian Groups, vols. I and II. Pure Applied Mathematics, vol. 36
(Academic, New York, London, 1970, 1973)
R. Göbel, J. Trlifaj — [GT] Approximations and Endomorphism Algebras and Modules. Exposi-
tions in Mathematics, vol. 41 (W. de Gruyter, Berlin, New York, 2006)
P. Griffith — [G] Infinite Abelian Groups. Chicago Lectures in Mathematics (University of Chicago
Press, Chicago, 1970)
T. Jech — [J] Set Theory. Pure and Applied Mathematics, vol. 79 (Academic, New York, London,
1978)
C.U. Jensen — [Je] Les Fonctions Dérivés de lim et leur Applications en Théorie des Modules.

Lecture Notes in Mathematics, vol. 254 (Springer, New York, 1972)
I. Kaplansky — [K] Infinite Abelian Groups (University of Michigan Press, Ann Arbor, 1954,
1969)
P.A. Krylov, A.V. Mikhalev, A.A. Tuganbaev — [KMT] Endomorphism Rings of Abelian Groups
(Kluwer Academic, Dordrecht, Boston, London, 2010)
S. Mac Lane — [M] Homology (Springer, New York, 1963)

© Springer International Publishing Switzerland 2015 707

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6
708 References

A. Mader — [Ma] Almost Completely Decomposable Groups. Algebra, Logic and Applications,
vol. 13 (Gordon and Breach, Amsterdam, 2000)
L. Salce — [S] Struttura dei p-Gruppi Abeliani (Pitagora, Bologna, 1980)

Papers

U. Albrecht — [1] Chain conditions in endomorphism rings. Rocky Mt. J. Math. 15, 91–106
(1985). — [2] Abelian groups with self-injective endomorphism rings. Commun. Algebra 15,
2451–2471 (1987). — [3] Endomorphism rings and a generalization of torsion-freeness and
purity. Commun. Algebra 17, 1101–1135 (1989). — [4] A-reflexive abelian groups. Houst. J.
Math. 15, 459–480 (1989). — [5] Endomorphism rings, tensor products and Fuchs’ problem,
vol. 47, in Contemporary Mathematics, vol. 130 (American Mathematical Society, Providence,
RI, 1992), pp. 17–31. — [6] A-projective resolutions and an Azumaya theorem for a class of
mixed abelian groups. Czechoslov. Math. J. 51, 73–93 (2001). — [7] Fuchs’ problem 34 for
mixed abelian groups. Proc. Am. Math. Soc. 131, 1021–1029 (2003)
U. Albrecht, S. Breaz, C. Vinsonhaler, W. Wickless — [1] Cancellation properties for quotient
divisible groups. J. Algebra 317, 424–434 (2007)
U. Albrecht, S. Breaz, W. Wickless — [1] Self-small abelian groups. Bull. Aust. Math. Soc. 80,
205–216 (2009)
U. Albrecht, T. Faticoni — [1] Abelian groups flat as modules over their endomorphism ring.
Commun. Algebra 21, 3403–3423 (1993)
U. Albrecht, H.P. Goeters, W. Wickless — [1] The flat dimension of mixed abelian groups as
E-modules. Rocky Mt. J. Math. 25, 569–590 (1995)
U. Albrecht, J. Hausen — [1] Mixed modules with the summand intersection property, in Abelian
Groups and Modules. Lecture Notes in Pure and Applied Mathematics, vol. 182 (Marcel
Dekker, New York, 1996), pp. 123–132
U. Albrecht, P. Hill — [1] Butler groups of infinite rank and Axiom 3. Czechoslov. Math. J. 37,
293–309 (1987). — [2] Separable vector groups, in Abelian Group Theory, Contemporary
Mathematics, vol. 87 (American Mathematical Society, Providence, RI, 1989), pp. 155–160
H.W.K. Angad-Gaur — [1] The homological dimension of a torsion-free abelian group of finite
rank as a module over its ring of endomorphisms. Rend. Sem. Mat. Univ. Padova 57, 299–309
(1977)
J.W. Armstrong — [1] On the indecomposability of torsion-free abelian groups. Proc. Am. Math.
Soc. 16, 323–325 (1965)
D.M. Arnold — [1] A duality for quotient divisible abelian groups of finite rank. Pac. J. Math. 42,
11–15 (1972). — [2] Strongly homogeneous torsion-free abelian groups of finite rank. Proc.
Am. Math. Soc. 56, 67–72 (1976). — [3] Genera and direct sum decompositions of torsion-free
modules, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 616 (Springer, Berlin,
1977), pp. 197–218. — [4] Pure subgroups of finite rank completely decomposable groups,
in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874 (Springer, Berlin, 1981),
pp. 1–31. — [5] Endomorphism rings and subgroups of finite rank torsion-free abelian groups.
Rocky Mt. J. Math. 12, 241–256 (1982). — [6] Direct sums of local torsion-free abelian groups.
Proc. Am. Math. Soc. 130, 1611–1617 (2002)
D.M. Arnold, M. Dugas — [1] Butler groups with finite typesets and free groups with distinguished
subgroups. Commun. Algebra 21, 1947–1982 (1993). — [2] Locally free finite rank Butler
groups and near isomorphism, in Abelian Groups and Modules (Kluwer Academic, Dordrecht,
1995), pp. 41–48. — [3] A survey of Butler groups and the role of representations, in Abelian
Groups and Modules. Lecture Notes in Pure and Applied Mathematics, vol. 182 (Marcel
Dekker, New York, 1996), pp. 1–13
D.M. Arnold, J. Hausen — [1] A characterization of modules with the summand intersection
property. Commun. Algebra 18, 519–528 (1990)
References 709

D.M. Arnold, R. Hunter, F. Richman — [1] Global Azumaya theorems in additive categories.
J. Pure Appl. Algebra 16, 223–242 (1980)
D.M. Arnold, E.L. Lady — [1] Endomorphism rings and direct sums of torsion-free abelian groups.
Trans. Am. Math. Soc. 211, 225–237 (1975)
D.M. Arnold, C.E. Murley — [1] Abelian groups A such that Hom.A; / preserves direct sums of
copies of A. Pac. J. Math. 56, 7–20 (1975)
D.M. Arnold, R. Pierce, J. Reid, C. Vinsonhaler, W. Wickless — [1] Torsion-free abelian groups
of finite rank projective as modules over their endomorphism rings. J. Algebra 71, 1–10 (1981)
D.M. Arnold, K.M. Rangaswamy — [1] A note on countable Butler groups. Boll. Unione Mat.
Ital. Sez. B Artic. Ric. Mat. 10(8), 605–611 (2007)
D.M. Arnold, C. Vinsonhaler — [1] Pure subgroups of finite rank completely decomposable
groups. II. Lecture Notes in Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 97–143. —
[2] Representing graphs for a class of torsion-free abelian groups, in Abelian Group Theory
(Gordon and Breach, London, 1987), pp. 309–332. — [3] Endomorphism rings of Butler
groups. J. Aust. Math. Soc. 42, 322–329 (1987). — [4] Duality and invariants for Butler groups.
Pac. J. Math. 148, 1–10 (1991). — [5] Isomorphism invariants for abelian groups. Trans. Am.
Math. Soc. 330, 711–724 (1992). — [6] Finite rank Butler groups: a survey of recent results,
in Abelian Groups. Lecture Notes in Pure and Applied Mathematics, vol. 146 (Marcel Dekker,
London, 1993), pp. 17–41
R. Baer — [1] The decomposition of enumerable, primary, abelian groups into direct summands.
Q. J. Math. Oxford 6, 217–221 (1935). — [2] The decomposition of abelian groups into direct
summands. Q. J. Math. Oxford 6, 222–232 (1935). — [3] Types of elements and characteristic
subgroups of abelian groups. Proc. Lond. Math. Soc. 39, 481–514 (1935). — [4] The subgroup
of elements of finite order of an abelian group. Ann. Math. 37, 766–781 (1936). — [5] Primary
abelian groups and their automorphisms. Am. J. Math. 59, 99–117 (1937). — [6] Abelian
groups without elements of finite order. Duke Math. J. 3, 68–122 (1937). — [7] Dualism in
abelian groups. Bull. Am. Math. Soc. 43, 121–124 (1937). — [8] Abelian groups that are direct
summands of every containing abelian group. Bull. Am. Math. Soc. 46, 800–806 (1940). — [9]
Automorphism rings of primary abelian operator groups. Ann. Math. 44, 192–227 (1943). —
[10] Die Torsionsuntergruppe einer abelschen Gruppe. Math. Ann. 135, 219–234 (1958)
S. Balcerzyk — [1] On algebraically compact groups of I. Kaplansky. Fund. Math. 44, 91–93
(1957). — [2] On factor groups of some subgroups of a complete direct sum of infinite cyclic
groups. Bull. Acad. Polon. Sci. Cl. III 7, 141–142 (1959). — [3] On classes of abelian groups.
Bull. Acad. Polon. Sci. Cl. III 9, 327–329 (1961); Fund. Math. 51, 149–178 (1962); Fund.
Math. 56, 199–202 (1964)
S. Balcerzyk, A. Bialynicki-Birula, J. Łoś — [1] On direct decompositions of complete direct sums
of rank 1. Bull. Acad. Polon. Sci. Cl. III 9, 451–454 (1961)
J. Barwise, P. Eklof — [1] Infinitary properties of abelian torsion groups. Ann. Math. Logic 2,
25–68 (1970/71)
S. Bazzoni, C. Metelli — [1] On abelian torsion-free separable groups and their endomorphism
rings. Symposia Math. 23, 259–285 (1979)
R.A. Beaumont — [1] Rings with additive group which is the direct sum of cyclic groups. Duke
Math. J. 15, 367–369 (1948)
R.A. Beaumont, D.A. Lawver — [1] Strongly semisimple abelian groups. Pac. J. Math. 53, 327–
336 (1974)
R.A. Beaumont, R.S. Pierce — [1] Torsion-free rings. Ill. J. Math. 5, 61–98 (1961). — [2] Torsion
free groups of rank two. Mem. Am. Math. Soc. 38 (1961)
R.A. Beaumont, H.S. Zuckerman — [1] A characterization of the subgroups of the additive
rationals. Pac. J. Math. 1, 169–177 (1951)
I.Kh. Bekker, V.N. Nedov — [1] On groups with isomorphic holomorphs [Russian]. Mat. Zametki
62, 343–350 (1997)
K. Benabdallah, A. Birtz — [1] Sur une famille de groupes abéliens super-décomposables. Can.
Math. Bull. 24, 213–218 (1981)
710 References

K. Benabdallah, J.M. Irwin, M. Rafiq — [1] A core class of abelian p-groups. Symposia Math. 13,
195–206 (1974)
K. Benabdallah, T. Okuyama — [1] On purifiable subgroups of primary abelian groups. Commun.
Algebra 19, 85–96 (1991)
G.M. Bergman — [1] Boolean rings of projective maps. J. Lond. Math. Soc. 4, 593–598 (1972)
L. Bican — [1] Completely decomposable abelian groups any pure subgroup of which is
completely decomposable. Czechoslov. Math. J. 24, 176–191 (1974). — [2] Splitting in abelian
groups. Czechoslov. Math. J. 28, 356–364 (1978). — [3] Purely finitely generated abelian
groups. Comment. Math. Univ. Carol. 21, 209–218 (1980)
L. Bican, L. Fuchs — [1] On abelian groups by which balanced extensions of a rational group split.
J. Pure Appl. Algebra 78, 221–138 (1992); II: Czechoslov. J. Math. 44, 649–660 (1994). — [2]
Subgroups of Butler groups. Commun. Algebra 22, 1037–1047 (1994)
L. Bican, K.M. Rangaswamy — [1] Smooth unions of Butler groups. Forum Math. 10, 233–247
(1998)
L. Bican, K.M. Rangaswamy, C. Vinsonhaler — [1] Butler groups as smooth ascending unions.
Commun. Algebra 28, 5039–5045 (2000)
L. Bican, L. Salce — [1] Butler groups of infinite rank. Abelian Group Theory, Lecture Notes in
Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 171–189
E.A. Blagoveshchenskaya — [1] Definability of torsion-free abelian groups of countable rank of a
certain class by their endomorphism rings [Russian]. Fundam. Prikl. Mat. 13, 31–43 (2007); J.
Math. Sci. 152, 469–478 (2008)
E. Blagoveshchenskaya, R. Göbel, L. Strüngmann — [1] Classification of some Butler groups of
infinite rank. J. Algebra 380, 1–17 (2013)
E.A. Blagoveshchenskaya, A.V. Yakovlev — [1] Direct decompositions of torsion-free abelian
groups of finite rank [Russian]. Algebra i Analiz 1, 111–127 (1989); Leningr. Math. J. 1,
117–136 (1990)
A. Blass — [1] On the divisible parts of quotient groups, in Abelian Group Theory and Related
Topics, Contemporary Mathematics, vol. 171 (American Mathematical Society, Providence,
RI, 1994), pp. 37–50
A. Blass, J. Irwin — [1] Is there a core class for almost free groups of size @1 ? Commun. Algebra
32, 1189–1200 (2004). — [2] Special families of sets and Baer-Specker groups. Commun.
Algebra 33, 1733–1744 (2005)
A.V. Blazhenov — [1] Genera and cancellation of torsion-free modules of finite rank [Russian].
Algebra i Analiz 7(6), 33–78 (1995); St. Petersburg Math. J. 7(6), 891–924 (1996)
I.V. Bobylev — [1] Projective dimension of an abelian group over its endomorphism ring
[Russian]. Usp. Mat. Nauk 28, 229–230 (1973)
M. Bognár — [1] Ein einfaches Beispiel direkt unzerlegbarer abelscher Gruppen. Publ. Math.
Debrecen 4, 509–511 (1956)
W. Borho — [1] Über die abelschen Gruppen auf denen sich nur endlich viele wesentlich
verschiedene Ringe definieren lassen. Abh. Math. Sem. Univ. Hamburg 37, 98–107 (1972)
H. Bowman, K.M. Rangaswamy — [1] Torsion-free separable abelian groups quasi projective over
their endomorphism rings. Houst. J. Math. 11, 447–453 (1985)
R.A. Bowshell, P. Schultz — [1] Unital rings whose additive endomorphisms commute. Math.
Ann. 228, 197–214 (1977)
D.L. Boyer — [1] Enumeration theorems in infinite abelian groups. Proc. Am. Math. Soc. 7,
565–570 (1956)
D. Boyer, A. Mader — [1] Functorial topologies on abelian groups. Rocky Mt. J. Math. 10,
695–708 (1980)
R. Brandl — [1] Abelian torsion groups with soluble automorphism groups. Rend. Circ. Mat.
Palermo Suppl. 23, 43–44 (1990)
E.I. Bunina, A.V. Mikhalëv — [1] Elementary equivalence of endomorphism rings of abelian
p-groups [Russian]. Fundam. Prikl. Mat. 10, 135–224 (2004); J. Math. Sci. 137, 5275–5335
(2006)
E.I. Bunina, M.A. Roǐzner — [1] Elementary equivalence of automorphism groups of abelian
p-groups. Fundam. Prikl. Mat. 15, 81–112 (2009); J. Math. Science 169, 614–635 (2010)
References 711

R. Burkhardt — [1] On a special class of almost completely decomposable torsion-free abelian

groups, in Abelian Groups and Modules. CISM Courses Lectures, vol. 287 (Springer, Berlin,
1984), pp. 141–150
M.C.R. Butler — [1] A class of torsion-free abelian groups of finite rank. Proc. Lond. Math. Soc.
15, 680–698 (1965)
G.G. Cǎlugǎreanu — [1] Abelian groups with semi-local endomorphism ring. Commun. Algebra
30, 4105–4111 (2002)
M.O’N. Campbell — [1] Countable torsion-free abelian groups. Proc. Lond. Math. Soc. 10, 11–23
(1960)
F. Castagna — [1] Sums of automorphisms of a primary abelian group. Pac. J. Math. 27, 463–473
(1968)
W. Chachólski, E.D. Farjoun, R. Göbel, Y. Segev — [1] Cellular covers of divisible abelian groups,
in Alpine perspectives on algebraic topology, Contemporary Mathematics, vol. 504 (American
Mathematical Society, Providence, RI, 2009), pp. 77–97
B. Charles — [1] Le centre de l’anneau des endomorphismes d’un groupe abélien primaire.
C. R. Acad. Sci. Paris 236, 1122–1123 (1953). — [2] Étude des groupes abéliens primaires
de type  !. Ann. Univ. Saraviensis 4, 184–199 (1955). — [3] Méthodes topologiques en
théorie des groupes abéliens, in Proceedings of the Colloquium on Abelian Groups (Akadémiai
Kiadó, Budapest, 1964), pp. 29–42. — [4] Sous-groupes fonctoriels et topologies, in Studies
on Abelian Groups (Dunod, Paris, 1968), pp.75–92
S.U. Chase — [1] Direct product of modules. Trans. Am. Math. Soc. 97, 457–473 (1960)
A.R. Chekhlov — [1] Intersection of direct summands of abelian p-groups [Russian], in Abelian
Groups and Modules (Tomsk. Gos. Univ., Tomsk, 1981), pp. 240–244. — [2] Quasipure
injective torsion-free abelian groups [Russian]. Mat. Zametki 46, 93–99 (1989). — [3] Abelian
groups with normal endomorphism rings [Russian]. Algebra Logika 48, 520–539 (2009);
Algebra Logic 48, 298–308 (2009)
A.R. Chekhlov, P.A. Krylov — [1] On L. Fuchs’ problems 17 and 43. J. Math. Sci. 143, 3517–3602
(2007)
J. Cohen, H. Gluck — [1] Stacked bases for modules over principal ideal domains. J. Algebra 14,
493–505 (1970)
P.M. Cohn — [1] The complement of a finitely generated direct summand of an abelian group.
Proc. Am. Math. Soc. 7, 520–521 (1956). — [2] Eine Bemerkung über die multiplikative
Gruppe eines Körpers. Arch. Math. 13, 344–348 (1962)
E.F. Cornelius Jr. — [1] A sufficient condition for separability. J. Algebra 67, 476–478 (1980)
A.L.S. Corner — [1] A note on rank and direct decompositions of torsion-free abelian groups. Proc.
Cambridge Philos. Soc. 57, 230–233 (1961); Proc. Cambridge Philos. Soc. 66, 239–240 (1969).
— [2] Every countable reduced torsion-free ring is an endomorphism ring. Proc. Lond. Math.
Soc. 13, 687–710 (1963). — [3] On a conjecture of Pierce concerning direct decompositions
of abelian groups. Proceedings of the Colloquium on Abelian Groups (Akadémiai Kiadó,
Budapest, 1964), pp. 43–48. — [4] Endomorphism rings of torsion-free abelian groups, in
Proceedings of the International Conference on the Theory of Groups (Gordon and Breach,
New York, 1967), pp. 59–69. — [5] On endomorphism rings of primary abelian groups.
Q. J. Math. Oxford 20, 277–296 (1969). — [6] Endomorphism algebras of large modules
with distinguished submodules. J. Algebra 11, 155–185 (1969). — [7] The independence of
Kaplansky’s notions of transitivity and full transitivity. Q. J. Math. Oxford 27, 15–20 (1976).
— [8] On the existence of very decomposable abelian groups, in Abelian Group Theory. Lecture
Notes in Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 354–357. — [9] Fully rigid
systems of modules. Rend. Sem. Mat. Univ. Padova 82, 55–66 (1989). — [10] Groups of units
of order in Q-algebras, in Models, Modules and Abelian Groups (W. de Gruyter, Berlin, 2008),
pp. 9–61
A.L.S. Corner, R. Göbel — [1] Prescribing endomorphism algebras, a unified treatment. Proc.
Lond. Math. Soc. 50, 447–479 (1985)
712 References

A.L.S. Corner, B. Goldsmith — [1] Isomorphic automorphism groups of torsion-free p-adic

modules, in Abelian Groups, Module Theory, and Topology. Lecture Notes in Pure and Applied
Mathematics, vol. 201 (Dekker, New York, 1998), pp. 125–130
P. Crawley — [1] Solution of Kaplansky’s Test Problems for primary abelian groups. J. Algebra 2,
413–431 (1965). — [2] The cancellation of torsion abelian groups in direct sums. J. Algebra
2, 432–442 (1965). — [3] Abelian p-groups determined by their Ulm sequences. Pac. J. Math.
22, 235–239 (1967). — [4] On the structure of a class of abelian p-groups. Symposia Math. 23,
33–45 (1979)
P. Crawley, A.W. Hales — [1] The structure of torsion abelian groups given by presentations. Bull.
Am. Math. Soc. 74, 954–956 (1968). — [2] The structure of abelian p-groups given by certain
presentations. J. Algebra 12, 10–23 (1969); 18, 264–268 (1971)
P. Crawley, B. Jónsson — [1] Refinements for infinite direct decompositions of algebraic systems.
Pac. J. Math. 14, 797–855 (1964)
P. Crawley, C. Megibben — [1] A simple construction of bizarre abelian groups, unpublished
D.O. Cutler — [1] Abelian p-groups A and B such that Tor.A; G/ D Tor.B; G/, G reduced.
Proc. Am. Math. Soc. 91, 12–14 (1984). — [2] Abelian groups determined by their p-socle,
in Abelian Groups and Modules (Gordon and Breach, New York, 1987), pp. 111–116. — [3]
The existence of certain pure dense subgroups of abelian p-groups is not decidable in ZFC. J.
Algebra 127, 269–278 (1989)
D. Cutler, J. Irwin — [1] Essentially fully indecomposable abelian p-groups. Quaest. Math. 9,
135–148 (1986)
D. Cutler, J. Irwin, J. Pfaendtner, T. Snabb — [1] Direct sums of cyclic summands. Comment.
Math. Univ. St. Pauli 32, 171–176 (1983)
D. Cutler, J. Irwin, T. Snabb — [1] Abelian p-groups containing proper p!Cn -projective subgroups.
Comment. Math. Univ. St. Pauli 33, 95–97 (1984)
D. Cutler, C. Missel — [1] The structure of C-indecomposable p!Cn -projective abelian p-groups.
Commun. Algebra 12, 301–319 (1984)
D.O. Cutler, R.W. Stringall — [1] A topology for primary abelian groups, in Studies on Abelian
Groups (Dunod, Paris, 1968), pp. 93–100
P.V. Danchev — [1] A note on the countable extensions of separable p!Cn -projective abelian p-
groups. Archiv Math. (Brno) 42, 251–254 (2006). — [2] On some fully invariant subgroups of
summable groups. Ann. Math. Blaise Pascal 15, 147–152 (2008)
G. De Marco, A. Orsatti — [1] Complete linear topologies on abelian groups. Symposia Math. 13,
153–161 (1974)
D. Derry — [1] Über eine Klasse von abelschen Gruppen. Proc. Lond. Math. Soc. 43, 490–506
(1937)
G. D’Este — [1] Abelian groups with anti-isomorphic endomorphism rings. Rend. Sem. Mat. Univ.
Padova 60, 55–75 (1978)
C. De Vivo, C. Metelli — [1] Admissible matrices as base changes of B.1/ -groups: a realizing
algorithm, in Abelian Groups and Modules. Trends in Mathematics (Birkhäuser, Basel, 1999),
pp. 135–147. — [2] Finite partition lattices and Butler groups. Commun. Algebra 27, 1571–
1590 (1999)
J. Dieudonné — [1] Sur les p-groupes abéliens infinis. Portugaliae Math. 11, 1–5 (1952)
D. Dikranjan, B. Goldsmith, L. Salce, P. Zanardo — [1] Algebraic entropy for abelian groups.
Trans. Am. Math. Soc. 361, 3401–3434 (2009)
S.Z. Ditor — [1] On the group of units of a ring. Am. Math. Monthly 78, 522–523 (1971)
A.J. Douglas, H.K. Farahat — [1] The homological dimension of an abelian group as a module over
its ring of endomorphisms. Monatshefte Math. 69, 294–305 (1965); II: Monatshefte Math. 76,
109–111 (1972); III: Monatshefte Math. 80, 37–44 (1975)
P.F. Dubois — [1] Generally p˛ -torsion-complete abelian groups. Trans. Am. Math. Soc. 159,
245–255 (1971)
References 713

M. Dugas — [1] Fast freie abelsche Gruppen mit Endomorphismenring Z. J. Algebra 71,
314–321 (1981). — [2] On the Jacobson radical of some endomorphism rings. Proc. Am. Math.
Soc. 102, 823–826 (1988). — [3] Localizations of torsion-free abelian groups. J. Algebra 278,
411–429 (2004); 284, 811–823 (2005). — [4] Co-local subgroups of abelian groups. II. J. Pure
Appl. Algebra 208, 117–126 (2007)
M. Dugas, S. Feigelstock — [1] A-rings. Colloq. Math. 96, 277–292 (2003)
M. Dugas, R. Göbel — [1] Algebraisch kompakte Faktorgruppen. J. Reine Angew. Math. 307/8,
341–352 (1979). — [2] Every cotorsion-free ring is an endomorphism ring. Proc. Lond.
Math. Soc. 45, 319–336 (1982). — [3] On endomorphism rings of primary abelian groups.
Math. Ann. 261, 359–385 (1982). — [4] Every cotorsion-free algebra is an endomorphism
algebra. Math. Z. 181, 451–470 (1982). — [5] Torsion-free abelian groups with prescribed
finitely topologized endomorphism ring. Proc. Am. Math. Soc. 90, 519–527 (1984). — [6] An
extension of Zassenhaus’ theorem on endomorphism rings. Fund. Math. 194, 239–251 (2007)
M. Dugas, J. Hausen — [1] Torsion-free E-uniserial groups of infinite rank, in Abelian Group
Theory, Contemporary Mathematics, vol. 87 (American Mathematical Society, Providence, RI,
1989), pp. 181–189
M. Dugas, G. Herden — [1] Arbitrary torsion classes of abelian groups. Commun. Algebra 11,
1455–74 (1983)
M. Dugas, P. Hill, K.M. Rangaswamy — [1] Butler groups of infinite rank. II. Trans. Am. Math.
Soc. 320, 643–664 (1990)
M. Dugas, J. Irwin — [1] On pure subgroups ofQcartesian products of integers. Results Math. 15,
35–52 (1989). — [2] On basic subgroups of Z. Commun. Algebra 19, 2907–2921 (1991)
M. Dugas, J. Irwin, S. Khabbaz — [1] Countable rings as endomorphism rings. Q. J. Math. Oxford
39, 201–211 (1988)
M. Dugas, A. Mader, C. Vinsonhaler — [1] Large E-rings exist. J. Algebra 108, 88–101 (1987)
M. Dugas, K.M. Rangaswamy — [1] Infinite rank Butler groups. Trans. Am. Math. Soc. 305,
129–142 (1988)
M. Dugas, B. Thomé — [1] The functor Bext under the negation of CH. Forum Math. 3, 23–33
(1991). — [2] Countable Butler groups and vector spaces with four distinguished subspaces.
J. Algebra 138, 249–272 (1991)
M. Dugas, R. Vergohsen — [1] On socles of abelian p-groups in L. Rocky Mt. J. Math. 18, 733–752
(1988)
M. Dugas, B. Zimmermann-Huisgen — [1] Iterated direct sums and products of modules, in
Abelian Group Theory. Lecture Notes in Mathematics, vol. 874 (Springer, Berlin, 1982), pp.
179–193
K. Eda — [1] A Boolean power and a direct product of abelian groups. Tsukuba J. Math.
6, 187–193 (1982). — [2] On a Boolean power of a torsion-free abelian group. J. Algebra
82, 84–93 (1983). — [3] On Z-kernel groups. Arch. Math. 41, 289–293 (1983). — [4] A
characterization of @1 -free abelian groups and its application to the Chase radical. Isr. J. Math.
60, 22–30 (1987). — [5] Slender modules, endo-slender abelian groups and large cardinals.
Fund. Math. 135, 5–24 (1990)
K. Eda, H. Ohta — [1] On abelian groups of integer-valued continuous functions, their Z-duals and
Z-reflexivity, in Abelian Group Theory (Gordon and Breach, New York, 1987), pp. 241–257
A. Ehrenfeucht — [1] On a problem of J.H.C. Whitehead concerning abelian groups. Bull. Acad.
Polon. Sci. Cl. III 3, 127–128 (1955)
A. Ehrenfeucht, J. Łoś — [1] Sur les produits cartésiens des groupes cycliques infinis. Bull. Acad.
Polon. Sci. Cl. III 2, 261–263 (1954)
S. Eilenberg, S. MacLane — [1] On the homology theory of abelian groups. Can. J. Math. 7, 43–53
(1955)
P.C. Eklof — [1] The structure of ultraproducts of abelian groups. Pac. J. Math. 47, 67–79 (1973).
— [2] On the existence of -free abelian groups. Proc. Am. Math. Soc. 47, 65–72 (1975).
— [3] Whitehead’s problem is undecidable. Am. Math. Monthly 83, 775–788 (1976). —
[4] Applications of logic to the problem of splitting abelian groups, in Logic Colloquium,
714 References

vol. 76 (North Holland, Amsterdam, 1977), pp. 287–299. — [5] Set-theoretical Methods in
Homological Algebra and Abelian Groups (Les Presses University, Montreal, 1980). — [6] On
singular compactness. Algebra Univ. 14, 310–316 (1982)
P.C. Eklof, M. Huber — [1] On the rank of Ext. Math. Z. 174, 159–185 (1980)
P. Eklof, M. Huber, A. Mekler — [1] Totally Crawley groups. J. Algebra 112, 370–384 (1988)
P.C. Eklof, A.H. Mekler, S. Shelah — [1] Almost disjoint abelian groups. Isr. J. Math. 49, 34–54
(1984)
P. Eklof, H.C. Mez — [1] Additive groups of essentially closed rings, in Abelian Groups and
Modules. CISM Courses Lectures, vol. 287 (Springer, Berlin, 1984), pp. 243–252
P.C. Eklof, S. Shelah — [1] On groups A such that A ˚ Zn D A, in Abelian Group Theory (Gordon
and Breach, New York, 1987), pp. 149–163
E. Enochs — [1] Isomorphic refinements of decompositions of a primary group into closed groups.
Bull. Sci. Math. France 91, 63–75 (1963). — [2] Torsion free covering modules. Proc. Am.
Math. Soc. 14, 884–889 (1963). — [3] Extending isomorphisms between basic subgroups.
Arch. Math. 15, 175–178 (1964)
M. Erdélyi — [1] Direct summands of abelian torsion groups [Hungarian]. Acta Univ. Debrecen
2, 145–149 (1955)
J. Erdős — [1] Torsion-free factor groups of free abelian groups and a classification of torsion free
abelian groups. Publ. Math. Debrecen 5, 172–184 (1957)
M.V. Eremina, P.A. Krylov — [1] The tensor product of abelian groups as a noetherian module
over an endomorphism ring [Russian]. Izv. Vyssh. Uchebn. Zaved. Mat. 45(4), 16–23 (2001);
Russian Math. 45(4), 14–21 (2001)
K. Faltings — [1] Automorphismengruppen endlicher abelscher p-Gruppen, in Studies on Abelian
Groups (Dunod, Paris, 1968), pp. 101–119
T.G. Faticoni — [1] Each countable reduced torsion-free commutative ring is a pure subring of an
E-ring. Commun. Algebra 15, 2545–2564 (1987). — [2] On the lattice of right ideals of the
endomorphism ring of an abelian group. Bull. Aust. Math. Soc. 38, 273–291 (1988). — [3]
Torsion-free abelian groups torsion over their endomorphism rings. Bull. Aust. Math. Soc. 50,
177–195 (1994)
T.G. Faticoni, P. Goeters — [1] Examples of torsion-free groups flat as modules over their
endomorphism rings. Commun. Algebra 19, 1–27 (1991)
T. Faticoni, H.P. Goeters, C. Vinsonhaler, W.J. Wickless — [1] Torsion-free duality is Warfield.
Proc. Am. Math. Soc. 125, 961–969 (1997)
T.H. Fay, G.L. Walls — [1] Maximal functorial topologies on abelian groups. Arch. Math. 38,
167–174 (1992)
S. Feigelstock — [1] The type set of groups and nilpotence. Comment. Math. Univ. St. Pauli 25,
159–165 (1977). — [2] Additive groups of self-injective rings. Soochow J. Math. 33, 641–645
(2007)
S. Feigelstock, J. Hausen, R. Raphael — [1] Abelian groups mapping onto their endomorphis
rings, in Abelian Groups and Modules. Trends in Mathematics (Birkhäuser, Basel, 1999),
pp. 231–239
S.T. Files — [1] Endomorphisms of local Warfield groups, in Abelian Group Theory and Related
Topics. Contemporary Mathematics, vol. 171 (American Mathematical Society, Providence, RI,
1994), pp. 99–107. — [2] On transitive mixed abelian groups, in Abelian Groups and Modules.
Lecture Notes in Pure and Applied Mathematics, vol. 182 (Marcel Dekker, New York, 1996),
pp. 243–251
S. Files, B. Goldsmith — [1] Transitive and fully transitive groups. Proc. Am. Math. Soc. 126,
1605–1610 (1998)
S. Files, W. Wickless — [1] The Baer-Kaplansky theorem for a class of global mixed groups.
Rocky Mt. J. Math. 26, 593–613 (1996)
T. Fink — [1] A note on direct decompositions of large powers of the group of integers. Commun.
Algebra 26, 3553–3562 (1998)
M.A. Flagg — [1] Jacobson radical isomorphism theorem for torsion-free modules, in Models,
Modules and Abelian groups (W. de Gruyter, Berlin, 2008), pp. 309–314
References 715

A.A. Fomin — [1] Tensor product of torsion-free abelian groups. Sibirsk. Mat. Z̆. 16, 1071–1080
(1975). — [2] Duality in certain classes of torsion-free abelian groups of finite rank [Russian].
Sibirsk. Mat. Z̆. 27(4), 117–127 (1986). — [3] Quotient divisible mixed groups, in Abelian
Groups, Rings and Modules. Contemporary Mathematics, vol. 273 (American Mathematical
Society, Providence, RI, 2001), pp. 117–128
A.A. Fomin, W.J. Wickless — [1] Quotient divisible abelian groups. Proc. Am. Math. Soc. 126,
45–52 (1998)
S.V. Fomin — [1] Über periodische Untergruppen der unendlichen abelschen Gruppen. Mat.
Sbornik 2, 1007–1009 (1937)
B. Franzen — [1] Algebraic compactness and filtered quotients, in Abelian Group Theory. Lecture
Notes in Mathematics, vol. 874 (Springer, Berlin, 1981), pp. 228–241
B. Franzen, B. Goldsmith — [1] On endomorphism algebras of mixed modules. J. Lond. Math.
Soc. 31, 468–472 (1985)
H. Freedman — [1] The automorphisms of countable primary reduced abelian groups. Proc. Lond.
Math. Soc. 12, 77–99 (1962). — [2] On the additive group of a torsion-free ring of rank two.
Publ. Math. Debrecen 20, 85–87 (1973)
E. Fried — [1] On the subgroups of an abelian group that are ideals in every ring, in Proceedings
of the Colloquium on Abelian Groups (Akadémiai Kiadó, Budapest, 1964), pp. 51–54
G. Frobenius, L. Stickelberger — [1] Über Gruppen von vertauschbaren Elementen. J. Reine
Angew. Math. 86, 217–262 (1878)
L. Fuchs — [1] The direct sum of cyclic groups. Acta Math. Acad. Sci. Hungar. 3, 177–195
(1952). — [2] On the structure of abelian p-groups. Acta Math. Acad. Sci. Hungar. 4, 267–
288 (1953). — [3] On a property of basic subgroups. Acta Math. Acad. Sci. Hungar. 5,
143–144 (1954). — [4] On abelian torsion groups which can not be represented as the direct
sum of a given cardinal number of components. Acta Math. Acad. Sci. Hungar. 7, 115–124
(1956). — [5] Ringe und ihre additive Gruppe. Publ. Math. Debrecen 4, 488–508 (1956). —
[6] Über das Tensorprodukt von Torsionsgruppen. Acta Sci. Math. Szeged 18, 29–32 (1957).
— [7] On quasi nil groups. Acta Sci. Math. Szeged 18, 33–43 (1957). — [8] Wann folgt die
Maximalbedingung aus der Minimalbedingung? Arch. Math. 8, 317–319 (1957). — [9] On
a directly indecomposable abelian group of power greater than continuum. Acta Math. Acad.
Sci. Hungar. 8, 453–454 (1957). — [10] On character groups of discrete abelian groups. Acta
Math. Acad. Sci. Hungar. 10, 133–140 (1959). — [11] Notes on abelian groups. I. Annales
Univ. Sci. Budapest 2, 5–23 (1959). — [12] The existence of indecomposable abelian groups
of arbitrary power. Acta Math. Acad. Sci. Hungar. 10, 453–457 (1959). — [13] Note on factor
groups in complete direct sums. Bull. Acad. Pol. Sci. Cl. III 11, 39–40 (1963). — [14] Recent
results and problems on abelian groups, in Topics in Abelian Groups (Chicago, 1963), Scott,
Foresman & Co. pp. 9–40. — [15] Note on linearly compact abelian groups. J. Aust. Math. Soc.
9, 433–440 (1969). — [16] Summands of separable abelian groups. Bull. London Math. Soc.
2, 205–208 (1970). — [17] Note on decompositions of torsion-free abelian groups. Comment.
Math. Helvetici 46, 409–413 (1971). — [18] Indecomposable abelian groups of measurable
cardinalities. Symposia Math. 13, 233–244 (1974). — [19] On torsion abelian groups quasi-
projective over their endomorphism rings. Proc. Am. Math. Soc. 42, 13–15 (1974). — [20] On
p!Cn -projective p-groups. Publ. Math. Debrecen 23, 309–313 (1976). — [21] Butler groups of
infinite rank. J. Pure Appl. Algebra 98, 25–44 (1995). — [22] Large indecomposable modules
with many automorphisms. Houst. J. Math. 23, 959–966 (2007)
L. Fuchs, R. Göbel — [1] Union of slender groups. Arch. Math. 87, 6–14 (2006). — [2] Cellular
covers of abelian groups. Results Math. 53, 59–76 (2009)
L. Fuchs, P. Gräbe — [1] Numbers of indecomposable summands in direct decompositions of
torsion-free abelian groups. Rend. Sem. Mat. Univ. Padova 53, 135–148 (1975)
L. Fuchs, I. Halperin — [1] On the imbedding of a regular ring in a regular ring with identity. Fund.
Math. 54, 285–290 (1964)
L. Fuchs, P. Hill — [1] The balanced-projective dimension of abelian p-groups. Trans. Am. Math.
Soc. 293, 99–112 (1986)
716 References

L. Fuchs, K.H. Hofmann — [1] Extensions of compact abelian groups by discrete ones and their
duality theory. I. J. Algebra 196, 578–594 (1997)
L. Fuchs, J.M. Irwin — [1] On p!C1 -projective p-groups. Proc. Lond. Math. Soc. 30, 459–470
(1975)
L. Fuchs, F. Loonstra — [1] On direct decompositions of torsion-free abelian groups of finite rank.
Rend. Sem. Mat. Univ. Padova 44, 175–183 (1970)
L. Fuchs, M. Magidor — [1] Butler groups of arbitrary cardinality. Isr. J. Math. 84, 239–263 (1993)
L. Fuchs, C. Metelli — [1] On a class of Butler groups. Manuscripta Math. 71, 1–28 (1991). —
[2] Indecomposable Butler groups of large cardinalities. Arch. Math. 57, 339–344 (1991). —
[3] Countable Butler groups, in Contemporary Mathematics, vol. 130 (American Mathematical
Society, Providence, RI, 1992), pp. 133–143
L. Fuchs, C. Metelli, K.M. Rangaswamy — [1] Corank one subgroups of completely decompos-
able abelian groups. Commun. Algebra 22, 1031–1037 (1994)
L. Fuchs, K.M. Rangaswamy — [1] On generalized regular rings. Math. Z. 107, 71–81 (1968). —
[2] Quasi-projective abelian groups. Bull. Soc. Math. France 98, 5–8 (1970). — [3] Unions
of chains of Butler groups, in Abelian Group Theory and Related Topics, Contemporary
Mathematics, vol. 171 (American Mathematical Society, Providence, RI, 1994), pp. 141–146.
— [4] Chains of projective modules. J. Algebra Appl. 10, 167–180 (2011)
L. Fuchs, L. Salce — [1] Almost totally injective p-groups. Quaest. Math. 1, 225–234 (1976). —
[2] Abelian p-groups of not limit length. Comment. Math. Univ. St. Pauli 26, 25–33 (1977)
L. Fuchs, L. Salce, P. Zanardo — [1] Note on the transitivity of pure-essential extensions. Colloq.
Math. 78, 283–291 (1998)
L. Fuchs, G. Viljoen — [1] Note on extensions of Butler groups. Bull. Aust. Math. Soc. 41,
117–122 (1990). — [2] Completely decomposable pure subgroups of completely decompos-
able abelian groups. Rend. Sem. Mat. Univ. Padova 92, 63–69 (1994)
S. Gacsályi — [1] On algebraically closed abelian groups. Publ. Math. Debrecen 2, 292–296 (1952)
B.J. Gardner — [1] Rings on completely decomposable torsion-free abelian groups. Comment.
Math. Univ. Carolin. 15, 381–392 (1974)
O. Gerstner — [1] Algebraische Kompaktheit bei Faktorgruppen von Gruppen ganzzahliger
Abbildungen. Manuscr. Math. 11, 103–109 (1974)
R.W. Gilmer — [1] Finite rings having a cyclic multiplicative group of units. Am. J. Math. 85,
447–452 (1963)
A.J. Giovannitti — [1] Extensions of Butler groups, in Abelian Group Theory. Lecture Notes
in Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 164–170. — [2] A note on proper
classes of exact sequences, in Methods in Module Theory. Lecture Notes in Pure and Applied
Mathematics, vol. 140 (Marcel Dekker, New York, 1992), pp. 107–116
S. Glaz, C. Vinsonhaler, W. Wickless — [1] Splitting rings for p-local torsion-free groups, in Zero-
dimensional commutative rings. Lecture Notes in Pure and Applied Mathematics, vol. 171
(Dekker, New York, 1995), pp. 223–239
S. Glaz, W. Wickless — [1] Regular and principal projective endomorphism rings of mixed abelian
groups. Commun. Algebra 22, 1161–1176 (1994)
R. Göbel — [1] Darstellung von Ringen als Endomorphismenringe. Arch. Math. 35, 338–350
(1980)
R. Göbel, D. Herden, S. Shelah — [1] Absolute E-rings. Adv. Math. 226, 235–253 (2011)
R. Göbel, W. May — [1] Cancellation of direct sums of countable abelian p-groups. Proc. Am.
Math. Soc. 131, 2705–2710 (2003)
R. Göbel, S. Pokutta — [1] Construction of dual groups using Martin’s axiom. J. Algebra 320,
2388–2404 (2008)
R. Göbel, R. Prelle — [1] Solution of two problems on cotorsion abelian groups. Arch. Math. 31,
423–431 (1978)
R. Göbel, S.V. Rychkov, B. Wald — [1] A general theory of slender groups and Fuchs-44-groups,
in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874 (Springer, Berlin, 1981),
pp. 194–201
References 717

R. Göbel, S. Shelah — [1] Semirigid classes of cotorsion-free abelian groups. J. Algebra 93,
136–150 (1985). — [2] On the existence of rigid @1 -free abelian groups of cardinality @1 , in
Abelian Groups and Modules (Kluwer Academic Publishers, Dordrecht, 1995), pp. 227–237.
— [3] Cotorsion theories and splittters. Trans. Am. Math. Soc. 352, 5357–5379 (2000). —
[4] Reflexive subgroups of the Baer-Specker group and Martin’s axiom, in Abelian Groups,
Rings and Modules. Contemporary Mathematics, vol. 273 (American Mathematical Society,
Providence, RI, 2001), pp. 145–158. — [5] Absolutely indecomposable modules. Proc. Am.
Math. Soc. 135, 1641–1649 (2007)
R. Göbel, S. Shelah, L. Strüngmann — [1] Almost-free E-rings of cardinality @1 . Can. J. Math.
55, 750–765 (2003)
R. Göbel, S. Shelah, S. Wallutis — [1] On the lattice of cotorsion theories. J. Algebra 238, 292–313
(2001)
R. Göbel, B. Wald — [1] Wachstumstypen und schlanke Gruppen. Symposia Math. 23, 201–239
(1979)
R. Göbel, S. Wallutis — [1] An algebraic version of the strong black box. Algebra Discrete Math.
1, 7–45 (2003)
R. Göbel, M. Ziegler — [1] Very decomposable abelian groups. Math. Z. 200, 485–496 (1989)
H.P. Goeters — [1] Torsion-free abelian groups with finite rank endomorphism rings. Quaest.
Math. 14, 111–115 (1991). — [2] Cobalanced torsion-free abelian groups. Commun. Algebra
21, 2715–2726 (1993). — [3] An extension of Warfield duality for abelian groups. J. Algebra
180, 848–861 (1996)
H.P. Goeters, J.D. Reid — [1] On the p-rank of Hom.A; B/, in Abelian Group Theory. Con-
temporary Mathematics, vol. 87 (American Mathematical Society, Providence, RI, 1989),
pp. 171–179
H.P. Goeters, W. Ullery, C. Vinsonhaler — [1] Numerical invariants for a class of Butler groups,
in Abelian Group Theory and Related Topics, Contemporary Mathematics, vol. 171 (American
Mathematical Society, Providence, RI, 1994), pp. 159–172
B. Goldsmith — [1] Essentially indecomposable modules over a complete discrete valuation
ring. Rend. Semin. Mat. Univ. Padova 70, 21–29 (1983). — [2] On endomorphism rings of
nonseparable abelian p-groups. J. Algebra 127, 73–79 (1989)
B. Goldsmith, C. Meehan, S.L. Wallutis — [1] On unit sum numbers of rational groups. Rocky
Mt. J. Math. 32, 1431–1450 (2002)
B. Goldsmith, P. Vámos — [1] A note on clean abelian groups. Rend. Semin. Mat. Univ. Padova
117, 181–191 (2007)
J. Gregory — [1] Abelian groups infinitarily equivalent to free ones. Notices Am. Math. Soc. 20,
A-500 (1973)
P. Griffith — [1] Purely indecomposable torsion-free groups. Proc. Am. Math. Soc. 18, 738–742
(1967). — [2] Decompositions of pure subgroups of torsion free groups. Ill. J. Math. 12,
433–438 (1968). — [3] Separability of torsion free groups and a problem of J.C. Whitehead. Ill.
J. Math. 12, 654–659 (1968). — [4] A solution to the splitting mixed problem of Baer. Trans.
Am. Math. Soc. 139, 261–269 (1969). — [5] Extensions of free groups by torsion groups.
Proc. Am. Math. Soc. 24, 677–679 (1970). — [6] On a subfunctor of Ext. Arch. Math. 21,
17–22 (1970). — [7] @n -free abelian groups. Q. J. Math. Oxford 23, 417–425 (1972)
S.Ya. Grinshpon, P.A. Krylov — [1] Fully invariant subgroups, full transitivity, and homomorphism
groups of abelian groups. J. Math. Sci. 128, 2894–2897 (2005)
S.Ya. Grinshpon, V.M. Misyakov — [1] Complete transitivity of direct products of abelian
groups [Russian], in Abelian Groups and Modules, vol. 10 (Tomsk. Gos. Univ., Tomsk, 1991),
pp. 23–30
J. de Groot — [1] Indecomposable abelian groups. Proc. Ned. Akad. Wetensch. 60, 137–145 (1957)
F. Haimo — [1] Preservation of divisibility in quotient groups. Duke Math. J. 15, 347–356 (1948).
— [2] Radical and antiradical groups. Rocky Mt. J. Math. 3, 91–106 (1973)
G. Hajós — [1] Über einfache und mehrfache Bedeckung des n-dimensionalen Raumes mit einem
Würfelgitter. Math. Z. 47, 427–467 (1942)
718 References

J.T. Hallett, K.A. Hirsch — [1] Torsion-free groups having finite automorphism groups. J. Algebra
2, 287–298 (1965)
D.K. Harrison — [1] Infinite abelian groups and homological methods. Ann. Math. 69, 366–391
(1959). — [2] Two of the problems of L. Fuchs. Publ. Math. Debrecen 7, 316–319 (1960). —
[3] On the structure of Ext, in Topics in Abelian Groups (Chicago, 1963), Scott, Foresman &
Co. pp. 195–209
J. Hausen — [1] The hypo residuum of the automorphism group of an abelian p-group. Pac. J.
Math. 35, 127–139 (1970). — [2] Automorphisms of abelian torsion groups of finite p-ranks.
Arch. Math. 22, 128–135 (1971). — [3] On the normal structure of automorphism groups of
abelian p-groups. J. Lond. Math. Soc. 5, 409–413 (1972). — [4] The normal structure of the
automorphism groups of an abelian p-group. Symposia Math. 13, 21–24 (1974). — [5] The
Jacobson radical of endomorphism rings of totally projective groups of finite type. J. Reine
Angew. Math. 292, 19–24 (1977). — [6] A cardinal-determined projectivity condition for
abelian groups and modules, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874
(Springer, Berlin, 1981), pp. 109–113. — [7] Abelian groups which are uniserial as modules
over their endomorphism rings, in Abelian Group Theory. Lecture Notes in Mathematics, vol.
1006 (Springer, Berlin, 1983), pp. 204–208. — [8] E-uniserial torsion-free abelian groups
of finite rank, in Abelian Groups and Modules. CISM Courses Lectures, vol. 287 (Springer,
Berlin, 1984), pp. 181–187. — [9] Modules with the summand intersection property. Commun.
Algebra 17, 135–148 (1989)
J. Hausen, J.A. Johnson — [1] Abelian groups with many automorphisms. Rend. Sem. Mat. Univ.
Padova 55, 1–5 (1976). — [2] A note on constructing E-rings. Publ. Math. Debrecen 38, 33–38
(1991)
J. Hausen, C.E. Praeger, P. Schultz — [1] Most abelian p-groups are determined by the Jacobson
radical of their endomorphism rings. Math. Z. 216, 431–436 (1994)
T.J. Head — [1] Dense submodules. Proc. Am. Math. Soc. 13, 197–199 (1962)
G. Heinlein — [1] Vollreflexive Ringe und schlanke Moduln, Dissertation, Erlangen, 1971
D. Herden, S. Shelah — [1] An upper cardinal bound on absolute E-rings. Proc. Am. Math. Soc.
137, 2843–2847 (2009)
P. Hill — [1] Certain pure subgroups of primary groups, in Topics in Abelian Groups (Chicago,
1963), pp. 311–314. — [2] Pure subgroups having prescribed socles. Bull. Am. Math. Soc.
71, 608–609 (1965). — [3] A classification of direct sums of closed groups. Acta Math. Acad.
Sci. Hungar. 17, 263–266 (1966). — [4] The isomorphic refinement theorem for direct sums
of closed groups. Proc. Am. Math. Soc. 18, 913–919 (1967). — [5] Isotype subgroups of
direct sums of countable groups. Ill. J. Math. 13, 281–290 (1969). — [6] Endomorphism rings
generated by units. Trans. Am. Math. Soc. 141, 99–105 (1969); Trans. Am. Math. Soc. 157, 511
(1971). — [7] A countability condition for primary groups presented by relations of length two.
Bull. Am. Math. Soc. 75, 780–782 (1969). — [8] On the decomposition of groups. Can. J. Math.
21, 762–768 (1969). — [9] On the freeness of abelian groups: a generalization of Pontryagin’s
theorem. Bull. Am. Math. Soc. 76, 1118–1120 (1970). — [10] The automorphisms of primary
abelian groups. Proc. Lond. Math. Soc. 22, 24–38 (1971). — [11] On the classification of
abelian groups. Period. Math. Hungar. 69, 41–52 (2014). — [12] Two problems of Fuchs
concerning Tor and Hom. J. Algebra 19, 379–383 (1971). — [13] Primary groups whose
subgroups of smaller cardinality are direct sums of cyclic groups. Pac. J. Math. 42, 63–67
(1972). — [14] The additive group of commutative rings generated by idempotents. Proc.
Am. Math. Soc. 38, 499–502 (1973). — [15] Countable unions of totally projective groups.
Trans. Am. Math. Soc. 190, 385–392 (1974). — [16] The third axiom of countability for
abelian groups. Proc. Am. Math. Soc. 82, 347–350 (1981). — [17] Isotype subgroups of totally
projective groups, in Abelian Group Theory, Lecture Notes in Mathematics, vol. 874 (Springer,
Berlin, 1981), pp. 305–321. — [18] The recovery of some abelian groups from their socles.
Proc. Am. Math. Soc. 86, 553–560 (1982). — [19] Balanced subgroups of totally projective
groups. J. Algebra 88, 346–349 (1984). — [20] On the structure of abelian p-groups. Trans.
Am. Math. Soc. 288, 505–525 (1985)
References 719

P. Hill, M. Lane, C. Megibben — [1] On the structure of p-local abelian groups. J. Algebra 143,
29–45 (1991)
P. Hill, C. Megibben — [1] Minimal pure subgroups in abelian groups. Bull. Soc. Math. France
92, 251–257 (1964). — [2] Quasi-closed primary groups. Acta Math. Acad. Sci. Hungar. 16,
271–274 (1965). — [3] On primary groups with countable basic subgroups. Trans. Am. Math.
Soc. 124, 49–59 (1966). — [4] Extending automorphisms and lifting decompositions in abelian
groups. Math. Ann. 175, 159–168 (1968). — [5] On direct sums of countable groups and
generalizations, in Studies in Abelian Groups (Dunod, Paris, 1968), pp. 183–206. — [6] On the
theory and classification of abelian p-groups. Math. Z. 190, 17–38 (1985). — [7] Torsion-free
groups. Trans. Am. Math. Soc. 295, 735–751 (1986). — [8] The local equivalence theorem, in
Abelian Group Theory, Contemporary Mathematics, vol. 87 (American Mathematical Society,
Providence, RI, 1989), pp. 201–219. — [9] Mixed groups. Trans. Am. Math. Soc. 334, 121–142
(1992). — [10] The classification of certain Butler groups. J. Algebra 160, 524–551 (1993).
P. Hill, C. Megibben, W. Ullery — [1] Every endomorphism of a local Warfield module is the
sum of two automorphisms, in Abelian Groups, Rings, Modules and Homological Algebra.
Lecture Notes in Pure and Applied Mathematics, vol. 249 (Chapman & Hall, London, 2006),
pp. 175–181
P. Hill, W. Ullery — [1] Isotype subgroups of local Warfield groups. Commun. Algebra 29,
1899–1907 (2001)
H.L. Hiller, M. Huber, S. Shelah — [1] The structure of Ext.A; Z/ and V = L. Math. Z. 162, 39–50
(1978)
K.Y. Honda — [1] On primary groups. Comment. Math. Univ. St. Pauli 2, 71–83 (1954). — [2]
On a decomposition theorem of primary groups. Comment. Math. Univ. St. Pauli 4, 53–66
(1955). — [3] Realism in the theory of abelian groups. Comment. Math. Univ. St. Pauli 5,
37–75 (1956); Math. Univ. St. Pauli 9, 11–28 (1961); Math. Univ. St. Pauli 12, 75–111 (1964)
M. Huber — [1] On cartesian powers of a rational group. Math. Z. 169, 253–259 (1979). — [2]
On reflexive modules and abelian groups. J. Algebra 82, 469–487 (1983)
M. Huber, R.B. Warfield Jr. — [1] On the values of the functor lim1 . Arch. Math. 33, 430–436
(1979/80)
A. Hulanicki — [1] Algebraic structure of compact abelian groups. Bull. Acad. Polon. Sci. Cl.
III 6, 71–73 (1958). — [2] Note on a paper of de Groot. Proc. Ned. Akad. Wetensch. 61, 114
(1958). — [3] The structure of the factor group of an unrestricted sum by the restricted sum of
abelian groups. Bull. Acad. Polon. Sci. Cl. III 10, 77–80 (1962)
R.H. Hunter — [1] Balanced subgroups of abelian groups. Trans. Am. Math. Soc. 215, 81–89
(1976)
R.H. Hunter, F. Richman — [1] Global Warfield groups. Trans. Am. Math. Soc. 266, 555–572
(1981)
R.H. Hunter, F. Richman, E. Walker — [1] Simply presented valuated abelian p-groups. J. Algebra
49, 125–133 (1977). — [2] Existence theorems for Warfield groups. Trans. Am. Math. Soc.
235, 345–362 (1978). — [3] Ulm’s theorem for simply presented valuated p-groups, in Abelian
Group Theory (Gordon and Breach, New York, 1987), pp. 33–64
D. van Huynh — [1] Die Spaltbarkeit von MHP-Ringen. Bull. Acad. Polon. Sci. Cl. III, 25,
939–941 (1977)
J. Irwin, J. O’Neill — [1] On direct products of abelian groups. Can. J. Math. 22, 525–544 (1970)
J.M. Irwin, F. Richman, E.A. Walker — [1] Countable direct sums of torsion complete groups.
Proc. Am. Math. Soc. 17, 763–766 (1966) Q
J. Irwin, T. Snabb — [1] A new class of subgroups of @0 Z, in Abelian Group Theory. Lecture
Notes in Mathematics, vol. 874 (Springer, Berlin, 1981), pp. 154–160
J.M. Irwin, C. Walker, E.A. Walker — [1] On p˛ -pure sequences of abelian groups, in Topics in
Abelian Groups (Chicago, 1963), Scott, Foresman & Co. pp. 69–119
J.M. Irwin, E.A. Walker — [1] On N-high subgroups of abelian groups. Pac. J. Math. 11,
1363–1374 (1961). — [2] On isotype subgroups of abelian groups. Bull. Soc. Math. France
89, 451–460 (1961)
720 References

A.V. Ivanov — [1] A problem on abelian groups [Russian]. Mat. Sbornik 105(4), 525–542 (1978);
Mat. USSR Sbornik 34, 461–474 (1978). — [2] Countable direct sums of complete and torsion-
complete groups [Russian]. Trudy Moskov. Mat. Obshch. 40, 121–170 (1979). — [3] Direct
sums and complete direct sums of abelian groups [Russian], in Abelian Groups and Modules
(Tomsk. Gos. Univ., Tomsk, 1980), pp. 70–90. — [4] Countable direct sums of certain groups
[Russian]. Trudy Moskov. Mat. Obshch. 41, 217–240 (1980). — [5] A class of abelian groups
[Russian]. Mat. Zametki 29, 351–358 (1981); Math. Notes 29, 182–185 (1981). — [6] Abelian
groups with self-injective rings of endomorphisms and with rings of endomorphisms with the
annihilator condition [Russian], in Abelian Groups and Modules (Tomsk. Gos. Univ., Tomsk,
1981), pp. 93–109. — [7] On a test problem of Kaplansky [Russian]. Trudy Moskov. Mat.
Obshch. 42, 200–220 (1981); Trans. Moscow Math. Soc. 199–218 (1982)
D.R. Jackett — [1] Some topics in the theory of ring structures on abelian groups. Bull. Aust.
Math. Soc. 18, 155–156 (1978). — [2] Rings on certain mixed abelian groups. Pac. J. Math.
98, 365–373 (1982)
C. Jacoby, K. Leistner, P. Loth, L. Strüngmann — [1] Abelian groups with partial decomposition
bases in Lı1! , Part I, in Groups and Model Theory, Contemporary Mathematics, vol. 576
(American Mathematical Society, Providence, RI, 2012), pp. 163–175
S. Janakiraman, K.M. Rangaswamy — [1] Strongly pure subgroups of abelian groups, in Group
Theory. Lecture Notes in Mathematics, vol. 573 (Springer, Berlin, 1976), pp. 57–65
C.U. Jensen — [1] On Ext.A; R/ for torsion-free A. Bull. Am. Math. Soc. 78, 831–834 (1972)
L. Jeśmanowicz — [1] On direct decompositions of torsion-free abelian groups. Bull. Acad. Polon.
Sci. Cl. III 8, 505–510 (1960)
B. Jónsson — [1] Unique factorization problem for torsionfree abelian groups. Bull. Am. Math.
Soc. 51, 364 (1945). — [2] On direct decomposition of torsion free abelian groups. Math.
Scand. 5, 230–235 (1957); and 7, 361–371 (1959)
F.F. Kamalov — [1] Intersection of direct summands of torsion-free abelian groups. Izv. Vyssh.
Uchebn. Zaved. Mat. 5, 45–56 (1977).
I. Kaplansky — [1] Some results on abelian groups. Proc. Nat. Acad. Sci. USA 38, 538–540 (1952).
— [2] Projective modules. Ann. Math. 68, 372–377 (1958)
I. Kaplansky, G.W. Mackey — [1] A generalization of Ulm’s theorem. Summa Brasil. Math. 2,
195–202 (1951)
A.V. Karpenko, V.M. Misyakov — [1] On the regularity of the center of the endomorphism ring
of an abelian group [Russian]. Fundam. Prikl. Mat. 13(3), 39–44 (2007); J. Math. Sci. 154,
304–307 (2008)
L. Kaup, M.S. Kleane — [1] Induktive Limiten endlich erzeugter freier Moduln. Manuscripta
Math. 1, 9–21 (1969)
P.F. Keef — [1] On set theory and the balanced-projective dimension of C -groups, in Abelian
Group Theory, Contemporary Mathematics, vol. 87 (American Mathematical Society, Provi-
dence, RI, 1989), pp. 31–42. — [2] On iterated torsion products of abelian p-groups. Rocky Mt.
J. Math. 21, 1035–55 (1991). — [3] On products of primary abelian groups. J. Algebra 152,
116–134 (1992). — [4] A class of primary abelian groups characterized by its socles. Proc.
Am. Math. Soc. 115 (1992), no. 3, 647–653. — [5] Extending homomorphisms on the socles
of primary abelian groups. Commun. Algebra 21, 3439–53 (1993). — [6] Primary abelian
groups admitting only small homomorphisms. Commun. Algebra 23, 3615–26 (1995). — [7]
Representable preradicals with enough projectives, in Abelian Groups and Modules (Kluwer
Academic Publishers, Dordrecht, 1995), pp. 301–311. — [8] Abelian groups and the torsion
product, in Abelian Groups and Modules. Lecture Notes Pure Applied Mathematics, vol. 182
(Marcel Dekker, New York, 1996), pp. 45–66. — [9] Mahlo cardinals and the torsion product
of primary abelian groups. Pac. J. Math. 259, 117–139 (2012)
P.F. Keef, P.V. Danchev — [1] On n-simply presented primary abelian groups. Houst. J. Math. 38,
1027–1050 (2012)
T. Kemoklidze — [1] On the full transitivity and fully invariant subgroups of cotorsion hulls of
separable p-groups. J. Math. Sci. 153(4), 506–517 (2008); J. Math. Sci. 155(5), 748–786 (2008)
References 721

A. Kertész — [1] On the decomposability of abelian p-groups into the direct sum of cyclic groups.
Acta Math. Acad. Sci. Hungar. 3, 122–126 (1952)
A. Kertész, T. Szele — [1] On the existence of non-discrete topologies in infinite abelian groups.
Publ. Math. Debrecen 3, 187–189 (1953)
S.A. Khabbaz — [1] Abelian torsion groups having a minimal system of generators. Trans. Am.
Math. Soc. 98, 527–538 (1961)
F. Kiefer — [1] The duals of totally projective groups, in Abelian Group Theory. Lecture Notes in
Mathematics, vol. 874 (Springer, Berlin, 1981), pp. 297–304
M.A. Kil’p — [1] Quasi-injective abelian groups. Vestnik Moskov. Univ. Mat. Mekh. 3, 3–4 (1967)
J. Koehler — [1] Some torsion-free rank two groups. Acta Sci. Math. Szeged 25, 186–190 (1964).
— [2] The type set of a torsion-free group of finite rank. Ill. J. Math. 9, 66–86 (1965)
M. Kojman, S. Shelah — [1] Universal abelian groups. Isr. J. Math. 92, 113–124 (1995)
G. Kolettis, Jr. — [1] Direct sums of countable groups. Duke Math J. 27, 111–125 (1960). — [2]
Homogeneously decomposable modules, in Studies on Abelian Groups (Dunod, Paris, 1968),
pp. 223–238
E.I. Kompantseva — [1] Torsion-free rings [Russian]. J. Math. Sci. 171, 213–247 (2010). — [2]
Absolute nil-ideals of an abelian group. Fundam. Prikl. Mat. 17:8, 63–76 (2012); J. Math. Sci.
197, 625–634 (2014)
L. Kovács — [1] On subgroups of the basic subgroup. Publ. Math. Debrecen 5, 261–264 (1958)
T. Koyama — [1] On quasi-closed groups and torsion complete groups. Bull. Soc. Math. France
95, 89–94 (1967)
S.F. Kozhukhov — [1] Torsion-free abelian groups of finite rank without nilpotent endomorphisms
[Russian]. Sibirsk. Mat. Z̆. 29, 58–69 (1988); Siberian Math. J. 29, 45–53 (1988)
A.A. Kravchenko — [1] Completely decomposable groups [Russian]. Mat. Zametki 31,
171–185 (1982); Math. Notes 31, 88–95 (1982). — [2] Balanced and cobalanced Butler groups
[Russian]. Mat. Zametki 45, 32–37 (1989); Math. Notes 45, 369–373 (1989)
M. Król — [1] Separable groups. I. Bull. Acad. Polon. Sci. Cl. III 9, 337–344 (1961). — [2]
Automorphism groups and endomorphism rings of torsion-free abelian groups of rank two,
Dissertationes Mathematicae, vol. 55, Warsaw, 1967
M. Król, E. Sa̧siada — [1] The complete direct sums of torsion-free abelian groups of rank one
which are separable. Bull. Acad. Polon. Sci. Cl. III 8, 1–2 (1960)
N.I. Kruchkov — [1] Some generalizations of slender abelian groups. J. Math. Sci. 154, 344–349
(2008)
P.A. Krylov — [1] Radicals of rings of endomorphisms of torsion-free abelian groups [Russian].
Mat. Sbornik 95(2), 214–228 (1974); Math. USSR Sbornik 24, 209–222 (1976). — [2] Torsion-
free abelian groups with cyclic p-basic subgroups [Russian]. Mat. Zametki 20, 805–813 (1976).
— [3] The group of homomorphisms into a torsion-free group of rank 1 [Russian], in Abelian
Groups and Modules (Tomsk. Gos. Univ., Tomsk, 1979), pp. 104–121. — [4] Two problems
on the extension groups of abelian groups [Russian]. Mat. Sbornik 185, 73–94 (1994). — [5]
Jacobson radical of the endomorphism ring of a torsion-free abelian group, in Abelian Groups
and Modules, vol. 11–12 (Tomsk. Gos. Univ., Tomsk, 1994), pp. 77–104. — [6] The Jacobson
radical of an endomorphism ring of an abelian group [Russian]. Algebra Logika 43, 60–76
(2004); Algebra Logic 43, 34–43 (2004)
P.A. Krylov, A.A. Tuganbaev — [1] Modules over discrete valuation domains. J. Math. Sci. 145,
4997–5117 (2007); II: 151, 3255–3371 (2008)
L.Ya. Kulikov — [1] Zur Theorie der abelschen Gruppen von beliebiger Mächtigkeit [Russian].
Mat. Sbornik 9, 165–182 (1941). — [2] On the theory of abelian groups of arbitrary power
[Russian]. Mat. Sbornik 16, 129–162 (1945). — [3] Generalized primary groups. I [Russian].
Trudy Moskov. Mat. Obshch. 1, 247–326 (1952); II. Trudy Moskov. Mat. Obshch. 2, 85–167
(1953). — [4] On direct decompositions of groups [Russian]. Ukrain. Mat. Zh. 4, 230–275
(1952)
722 References

A.G. Kurosh — [1] Zur Zerlegung unendlicher Gruppen. Math. Ann. 106, 107–113 (1932). — [2]
Primitive torsionsfreie Gruppen vom endlichen Range. Ann. Math. 38, 175–203 (1937)
E.L. Lady — [1] Countable torsion products of abelian p-groups. Proc. Am. Math. Soc. 37, 10–16
(1973). — [2] Almost completely decomposable torsion-free abelian groups. Proc. Am. Math.
Soc. 45, 41–47 (1974). — [3] Summands of finite rank torsion-free abelian groups. J. Algebra
32, 51–52 (1974). — [4] Nearly isomorphic torsion-free abelian groups. J. Algebra 35, 235–
238 (1975). — [5] Splitting fields for torsion-free modules over discrete valuation rings. I. J.
Algebra 49, 261–275 (1977). II: J. Algebra 66, 281–306 (1980). III: J. Algebra 307–320. —
[6] On classifying torsion-free modules over discrete valuation rings, in Abelian Group Theory.
Lecture Notes in Mathematics, vol. 616 (Springer, Berlin, 1977), pp. 168–172
M. Lane — [1] A new characterization of p-local balanced projective groups. Proc. Am. Math.
Soc. 96, 379–386 (1986). — [2] The balanced-projective dimension of p-local abelian groups.
J. Algebra 109, 1–13 (1987)
M. Lane, C. Megibben — [1] Balanced projectives and axiom 3. J. Algebra 111, 457–474 (1987)
H. Lausch — [1] Eine geometrische Beschreibung des Tensorproduktes zweier torsionsfreier
abelscher Gruppen endlichen Ranges. Geom. Dedicata 13, 419–428 (1983)
J. Lawrence — [1] Countable abelian groups with a discrete norm are free. Proc. Am. Math. Soc.
90, 352–354 (1984)
W.Y. Lee — [1] Co-diagonal Butler groups. Chinese J. Math. 17, 259–271 (1989)
H. Leptin — [1] Zur Theorie der überabzählbaren abelschen p-Gruppen. Abh. Math. Sem. Univ.
Hamburg 24, 79–90 (1960). — [2] Abelsche p-Gruppen und ihre Automorphismengruppen.
Math. Z. 73, 235–253 (1960). — [3] Einige Bemerkungen über die Automorphismen abelscher
p-Gruppen, in Proceedings of the Colloquium on Abelian Groups (Akadémiai Kiadó, Budapest,
1964), pp. 99–104
F.W. Levi — [1] Abelsche Gruppen mit abzählbaren Elementen, Habilitationsschrift (Leipzig,
1917)
W. Liebert — [1] Charakterisierung der Endomorphismenringe beschränkter abelscher Gruppen.
Math. Ann. 174, 217–232 (1967). — [2] Die minimalen Ideale der Endomorphismenringe
abelscher p-Gruppen. Math. Z. 97, 85–104 (1967). — [3] Endomorphism rings of abelian p-
groups, in Studies on Abelian Groups (Dunod, Paris, 1968), pp. 239–258. — [4] Endomorphism
rings of abelian p-groups, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 1006
(Springer, Berlin, 1983), pp. 384–399. — [5] Isomorphic automorphism groups of primary
abelian groups. I. Abelian Group Theory (Gordon and Breach, New York, 1987), pp. 9–31; II:
Abelian Group Theory, Contemporary Mathematics, vol. 87 (American Mathematical Society,
Providence, RI, 1989), pp. 51–59
R.C. Linton — [1] On fully invariant subgroups of primary abelian groups. Mich. Math. J. 22,
281–284 (1975). — [2] -large subgroups of C -groups. Pac. J. Math. 78, 477–486 (1978)
J. Łoś — [1] Abelian groups which are direct summands of every abelian group which contains
them as pure subgroups. Bull. Acad. Polon. Sci. Cl. III 4, 73 (1956); Fund. Math. 44, 84–90
(1957). — [2] Linear equations and pure subgroups. Bull. Acad. Polon. Sci. Cl. III 7, 13–18
(1959)
P. Loth — [1] The duals of Warfield groups. Pac. J. Math. 181, 333–356 (1997). — [2]
Characterizations of Warfield groups. J. Algebra 204, 32–41 (1998). — [3] Classifications of
abelian groups and Pontryagin duality, in Algebra, Logic and Applications, vol. 10 (Gordon
and Breach, Amsterdam, 1998)
E.S. Lyapin — [1] On the decompositions of abelian groups into direct sums of rational groups
[Russian]. Mat. Sbornik 8, 205–231 (1940); Am. Math. Soc. Transl. 7, 1–56 (1950)
S. MacLane — [1] Group extensions by primary abelian groups. Trans. Am. Math. Soc. 95, 1–16
(1960)
References 723

A. Mader — [1] On the automorphism group and the endomorphism ring of abelian groups. Ann.
Univ. Sci. Budapest 8, 3–12 (1965). — [2] On the normal structure of the automorphism
group and the ideal structure of the endomorphism ring of abelian p-groups. Publ. Math.
Debrecen 13, 123–137 (1966). — [3] Extensions of abelian groups, in Studies on Abelian
Groups (Paris, 1968), pp. 259–266. — [4] Groups and modules that are slender as modules over
their endomorphism rings, in Abelian Groups and Modules, CISM Courses Lectures, vol. 287
(Springer, Berlin, 1984), pp. 315–327. — [5] Regularity in endomorphism rings. Commun.
Algebra 37, 2823–2844 (2009)
A. Mader, O. Mutzbauer, C. Vinsonhaler — [1] Local-global relations for almost completely
decomposable groups. Rocky Mt. J. Math. 29, 1429–1453 (1999)
A. Mader, L. Strüngmann — [1] A class of Butler groups and their endomorphism rings. Hokkaido
Math. J. 37, 399–425 (2008)
M. Magidor, S. Shelah — [1] When does almost free imply free? (For groups, transversals, etc.). J.
Am. Math. Soc. 7, 769–830 (1994). — [2] Bext2 .G; T/ can be nontrivial even assuming GCH,
in Abelian Group Theory and Related Topics, Contemporary Mathematics, vol. 171 (American
Mathematical Society, Providence, RI, 1994), pp. 287–294
A.I. Mal’cev — [1] Torsion-free abelian groups of finite rank [Russian]. Mat. Sbornik 4, 45–68
(1938)
J.M. Maranda — [1] On pure subgroups of abelian groups. Arch. Math. 11, 1–13 (1960)
W. May — [1] Multiplicative groups of fields. Proc. Lond. Math. Soc. 24, 295–306 (1972). —
[2] Isomorphism of group algebras. J. Algebra 40, 10–18 (1976). — [3] Multiplicative groups
under field extension. Can. J. Math. 31, 436–440 (1979). — [4] Endomorphism rings of mixed
abelian groups, in Abelian Group Theory, Contemporary Mathematics, vol. 87 (American
Mathematical Society, Providence, RI, 1989), pp. 61–74. — [5] Endomorphism algebras of not
necessarily cotorsion-free modules. Contemporary Math., vol. 130 (American Mathematical
Society, Providence, RI, 1992), pp. 257–264. — [6] Abelian automorphism groups of countable
rank, in Abelian Groups and Modules. Trends in Mathematics (Birkhäuser, Basel, 1999), pp.
23–42. — [7] The use of the finite topology on endomorphism rings. J. Pure Appl. Algebra
163, 107–117 (2001)
W. May, E. Toubassi — [1] Classifying endomorphism rings of rank one mixed groups, in Abelian
Groups and Modules. CISM Courses and Lectures, vol. 287 (Springer, Berlin, 1984), pp. 253–
263
C.K. Megibben — [1] On subgroups of primary abelian groups. Publ. Math. Debrecen 12, 293–
294 (1965). — [2] Large subgroups and small homomorphisms. Mich. Math. J. 13, 153–160
(1966). — [3] On mixed groups of torsion-free rank one. Ill. J. Math. 11, 134–144 (1967).
— [4] The generalized Kulikov criterion. Can. J. Math. 21, 1192–1208 (1969). — [5] A
generalization of the classical theory of primary groups. Tôhoku Math. J. 22, 347–356 (1970).
— [6] Generalized pure injectivity. Symposia Math. 13, 257–271 (1974). — [7] Crawley’s
problem on the unique !-elongation of p-groups is undecidable. Pac. J. Math. 107, 205–212
(1983). — [8] !1 -separable p-groups, in Abelian Group Theory (Gordon and Breach, New
York, 1987), pp. 117–136
C.K. Megibben, W. Ullery — [1] Isotype Warfield subgroups of global Warfield groups. Rocky Mt.
J. Math. 32, 1523–1542 (2002). — [2] The sequentially pure projective dimension of global
groups with decomposition bases. J. Pure Appl. Algebra 187, 183–205 (2004)
K. Meinel — [1] Superdecomposable modules over integral domains. Arch. Math. 39, 11–18
(1982)
A.H. Mekler — [1] Proper forcing and abelian groups, in Abelian Group Theory. Lecture Notes
in Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 285–303. — [2] @1 -separable groups
of mixed type, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874 (Springer,
Berlin, 1981), pp. 114–126
A. Mekler, G. Schlitt — [1] The theory of dual groups. Fund. Math. 144, 129–142 (1994)
724 References

A. Mekler, S. Shelah — [1] A solution to Crawley’s problem. Pac. J. Math. 121, 133–134 (1986).
— [2] When -free implies strongly -free, in Abelian Group Theory (Gordon and Breach,
New York, 1987), pp. 137–148. — [3] Every coseparable group may be free. Isr. J. Math. 81,
161–178 (1993)
C. Metelli — [1] Bihomogeneous groups, in Abelian Group Theory, Contemporary Mathematics,
vol. 87 (American Mathematical Society, Providence, RI, 1989), pp. 161–169. — [2] On
automorphisms of completely decomposable torsion-free abelian groups. Commun. Algebra
18, 529–549 (1990). — [3] On direct sums of B.1/ -groups. Comment. Math. Univ. Carolin.
34, 587–591 (1993). — [4] An adjointness relation for finite partition lattices, in Abelian
Groups, Module Theory, and Topology. Lecture Notes in Pure and Applied Mathematics, vol.
201 (Dekker, New York, 1998), pp. 301–310
C. Metelli, L. Salce — [1] The endomorphism ring of an abelian torsion-free homogeneous
separable group. Arch. Math. 26, 480–485 (1975)
R. Mines — [1] A family of functors defined on generalized abelian groups. Pac. J. Math. 26,
349–360 (1968)
R. Mines, C. Vinsonhaler — [1] Butler groups and Bext: a constructive view, in Contemporary
Mathematics, vol. 130 (American Mathematical Society, Providence, RI, 1992), pp. 289–299
A.P. Mishina — [1] On the direct summands of complete direct sums of torsion-free abelian
groups of rank one [Russian]. Sibirsk. Mat. Zh. 3, 244–249 (1962). — [2] Separability of
complete direct sums of torsion-free abelian groups of rank one [Russian]. Dokl. Akad. Nauk
SSSR 143, 275–276 (1962); Mat. Sbornik 57, 375–383 (1962). — [3] On automorphisms and
endomorphisms of abelian groups [Russian]. Vestnik Moskov. Univ. 1962, Nr. 4, 39–43. — [4]
On automorphisms and endomorphisms of abelian groups [Russian]. Vestnik Moskov. Univ.
27, 62–66 (1972)
V.M. Misyakov — [1] On complete transitivity of reduced abelian groups [Russian], in Abelian
Groups and Modules, vol. 11–12 (Tomsk. Gos. Univ., Tomsk, 1994), pp. 134–156; J. Math.
Sci. 154, 350–373 (2008)
A.V. Misyakova — [1] Abelian groups with semiprime endomorphism ring [Russian]. Mat.
Sbornik 202, 117–126 (2011); Mat. Sbornik 202, 739–748 (2011)
G.S. Monk — [1] One-sided ideals in the endomorphism ring of an abelian p-group. Acta Math.
Acad. Sci. Hungar. 19, 171–185 (1968). — [2] Essentially indecomposable abelian p-groups.
J. Lond. Math. Soc. 3, 341–345 (1971)
J.H. Moore — [1] A characterization of Warfield groups. Proc. Am. Math. Soc. 87, 617–620 (1983)
A.I. Moskalenko — [1] Cotorsion hull of a separable p-group [Russian]. Algebra Logika 28,
207–226 (1989); Algebra Logic 28, 139–151 (1989). — [2] Definability of a countable periodic
group by groups of extensions [Russian], in Abelian Groups and Modules, vol. 10 (Tomsk. Gos.
Univ., Tomsk, 1991), pp. 86–90
C.E. Murley — [1] The classification of certain classes of torsion-free abelian groups. Pac. J. Math.
40, 647–665 (1972)
O. Mutzbauer — [1] Klassifizierung torsionsfreier abelscher Gruppen des Ranges 2. I: Rend. Sem.
Mat. Univ. Padova 55, 195–208 (1976). II: Rend. Sem. Mat. Univ. Padova 58, 163–174 (1977).
— [2] Type invariants of torsion-free abelian groups, in Abelian Group Theory, Contemporary
Mathematics, vol. 87 (American Mathematical Society, Providence, RI, 1989), pp. 133–154
O. Mutzbauer, E. Toubassi — [1] Quasibases of p-groups. Rend. Sem. Mat. Univ. Padova 102,
77–95 (1999)
V.I. Myshkin — [1] Countable abelian groups of rank 1 [Russian]. Mat. Sbornik 76, 435–448
(1968)
G.P. Niedzwecki — [1] Rings on groups, Thesis, Wesleyan University, 1982
G.P. Niedzwecki, J. Reid — [1] Abelian groups projective over their endomorphism rings.
J. Algebra 159, 139–149 (1993)
G. Nöbeling — [1] Verallgemeinerung eines Satzes von Herrn E. Specker. Invent. Math. 6, 41–55
(1968)
References 725

L.G. Nongxa — [1] A note on homogeneous torsion-free abelian groups. Q. J. Math. Oxford 35,
183–190 (1984). — [2] *-pure subgroups of completely decomposable abelian groups. Proc.
Am. Math. Soc. 100, 613–618 (1987)
L.G. Nongxa, K.M. Rangaswamy, C. Vinsonhaler — [1] Balanced subgroups of completely
decomposable groups, in Abelian Groups and Modules. Lecture Notes in Pure and Applied
Mathematics, vol. 182 (Marcel Dekker, New York, 1996), pp. 331–351
L.G. Nongxa, C. Vinsonhaler — [1] Balanced Butler groups. J. Algebra 180, 546–570 (1996)
R.J. Nunke — [1] Modules of extensions over Dedekind rings. Ill. J. Math. 3, 222–241 (1959). —
[2] Slender groups. Bull. Am. Math. Soc. 67, 274–275 (1961); Acta Sci. Math. Szeged 23, 67–
73 (1962). — [3] Purity and subfunctors of the identity, in Topics in Abelian Groups (Chicago,
1963), Scott, Foresman & Co. pp. 121–171. — [4] On the structure of Tor, in Proceedings
of the Colloquium on Abelian Groups (Akadémiai Kiadó, Budapest, 1964), pp. 115–124; Pac.
J. Math. 22, 453–464 (1967). — [5] Homology and direct sums of countable abelian groups.
Math. Z. 101, 182–212 (1967). — [6] A note on endomorphism rings of abelian p-groups,
in Studies on Abelian Groups (Dunod, Paris, 1968), pp. 305–308. — [7] Uniquely elongating
modules. Symposia Math. 13, 315–330 (1974)
R.J. Nunke, J.J. Rotman — [1] Singular cohomology groups. J. Lond. Math. Soc. 37, 301–306
(1962)
T. Okuyama — [1] On purifiable subgroups in arbitrary abelian groups. Commun. Algebra 28,
121–139 (2000). — [2] Purifiability in pure subgroups. Hokkaido Math. J. 36, 365–381 (2007)
K.C. O’Meara, C. Vinsonhaler — [1] Separative cancellation and multiple isomorphism in torsion-
free abelian groups. J. Algebra 221, 536–550 (1999)
J.D. O’Neill — [1] Noetherian rings with free additive groups. Proc. Am. Math. Soc. 92, 323–324
(1984). — [2] Direct summands of vector groups. Acta Math. Acad. Sci. Hungar. 55, 207–209
(1990). — [3] Direct summands of Z for large . in Abelian Group Theory and Related Topics,
Contemporary Mathematics, vol. 171 (American Mathematical Society, Providence, RI, 1994),
pp. 313–323. — [4] A result on direct products of copies of the integers. Commun. Algebra
23, 4825–4830 (1995). — [5] Summable subsets of direct products of infinite cyclic groups,
in Abelian Groups and Modules. Lecture Notes in Pure and Applied Mathematics, vol. 182
(Marcel Dekker, New York, 1996), pp. 353–361
A. Orsatti — [1] Alcuni gruppi abeliani il cui anello degli endomorfismi è locale. Rend. Sem. Mat.
Univ. Padova 35, 107–115 (1965). — [2] A class of rings which are the endomorphism rings
of some torsion-free abelian groups. Ann. Scuola Norm. Sup. Pisa 23, 143–153 (1969)
B.L. Osofsky — [1] Lifting direct sum decompositions of bounded abelian p-groups, in Abelian
Groups, Rings and Modules, Contemporary Mathematics, vol. 273 (American Mathematical
Society, Providence, RI, 2001), pp. 253–260
M.A. Ould-Beddi, L. Strüngmann — [1] Stacked bases for a pair of homogeneous completely
decomposable groups with bounded quotient, in Abelian Groups and Modules. Trends in
Mathematics (Birkhäuser, Basel, 1999), pp. 199–209
S. Pabst — [1] On @1 -free modules with trivial dual. Commun. Algebra 28, 5053–5065 (2000)
E.A. Palyutin — [1] Indecomposable !1 -free abelian groups [Russian]. Sibirsk. Mat. Zh. 19,
1415–1417 (1978)
A.T. Paras — [1] Abelian groups as noetherian modules over their endomorphism rings, in
Abelian Group Theory and Related Topics, Contemporary Mathematics, vol. 171 (American
Mathematical Society, Providence, RI, 1994), pp. 325–332
L.D. Parker, E.A. Walker — [1] An extension of the Ulm-Kolettis theorems, in Studies on Abelian
Groups (Dunod, Paris, 1968), pp. 309–325
K.R. Pearson, J.E. Schneider — [1] Rings with a cyclic group of units. J. Algebra 16, 243–251
(1970)
R.S. Pierce — [1] Homomorphisms of primary abelian groups, in Topics in Abelian Groups
(Chicago, 1963), Scott, Foresman & Co. pp. 215–310. — [2] Centers of purity in abelian
groups. Pac. J. Math. 13, 215–219 (1963). — [3] Endomorphism rings of primary abelian
726 References

groups, in Proceedings of the Colloquium on Abelian Groups (Akadémiai Kiadó, Budapest,

1964), pp. 125–137. — [4] E-modules, in Abelian Group Theory, Contemporary Mathematics,
vol. 87 (American Mathematical Society, Providence, RI, 1989), pp. 221–240
R.S. Pierce, C. Vinsonhaler — [1] Classifying E-rings. Commun. Algebra 19, 615–653 (1991)
L. Pontryagin — [1] The theory of topological commutative groups. Ann. Math. 35, 361–388
(1934)
G.D. Poole, J.D. Reid — [1] Abelian groups quasi-injective over their endomorphism rings. Can.
J. Math. 24, 617–621 (1972)
L. Procházka — [1] Über die Spaltbarkeit der Faktorgruppen torsionsfreier abelscher Gruppen
endlichen Ranges [Russian]. Czechoslov. Math. J. 11, 521–557 (1961). — [2] A note on quasi-
isomorphism of torsion free abelian groups of finite rank. Czechoslov. Math. J. 15, 1–8 (1965)
H. Prüfer — [1] Untersuchungen über die Zerlegbarkeit der abzählbaren primären abelschen Grup-
pen. Math. Z. 17, 35–61 (1923). — [2] Theorie der abelschen Gruppen. I. Grundeigenschaften.
Math. Z. 20, 165–187 (1924); II. Ideale Gruppen. Math. Z. 22, 222–249 (1925)
P. Puusemp — [1] On the torsion subgroups and endomorphism semigroups of abelian groups.
Algebras Groups Geom. 14, 407–422 (1997)
R. Rado — [1] A proof of the basis theorem for finitely generated abelian groups. J. Lond. Math.
Soc. 26, 74–75 (1951); 160
K.M. Rangaswamy — [1] Abelian groups with endomorphic images of special types. J. Algebra 6,
271–280 (1967). — [2] Representing Baer rings as endomorphism rings. Math. Ann. 190, 167–
176 (1970). — [3] Abelian groups with self-injective endomorphism rings, in Proceedings of
the Second International Conference on the Theory of Groups. Lecture Notes in Mathematics,
vol. 372 (Springer, Berlin, 1974), pp. 595–604. — [4] An aspect of purity and its dualisation
in abelian groups and modules. Symposia Math. 23, 307–320 (1979). — [5] Separable abelian
groups as modules over their endomorphism rings. Proc. Am. Math. Soc. 91, 195–198 (1984).
— [6] A homological characterization of abelian B2 -groups. Proc. Am. Math. Soc. 121,
409–415 (1994). — [7] A property of B2 -groups. Comment. Math. Univ. Carolin. 35, 627–631
(1994)
L. Rédei, T. Szele — [1] Die Ringe “ersten Ranges”. Acta Sci. Math. Szeged 12A, 18–29 (1950)
R. Ree, R.J. Wisner — [1] A note on torsion-free nil groups. Proc. Am. Math. Soc. 7, 6–8 (1956)
G.A. Reid — [1] Almost free abelian groups, Lecture notes, Tulane University, New Orleans, 1967
J.D. Reid — [1] A note on torsion-free abelian groups of infinite rank. Proc. Am. Math. Soc. 13,
222–225 (1962). — [2] On quasi-decompositions of torsion-free abelian groups. Proc. Am.
Math. Soc. 13, 550–554 (1962). — [3] On the ring of quasi-endomorphisms of a torsion-free
group, in Topics in Abelian Groups (Chicago, 1963), Scott, Foresman & Co. pp. 51–68. —
[4] Quasi-pure-injectivity and quasi-pure-projectivity, in Abelian Group Theory. Lecture Notes
in Mathematics, vol. 616 (1977) pp. 219–227. — [5] Abelian groups finitely generated over
their endomorphism rings, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874
(Springer, Berlin, 1981), pp. 41–52. — [6] Abelian groups cyclic over their endomorphism
rings, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 1006 (Springer, Berllin,
1983), pp. 190–203
F. Richman — [1] Thin abelian groups. Pac. J. Math. 27, 599–606 (1968). — [2] A class of rank
2 torsion-free groups, in Studies on Abelian Groups (Dunod, Paris, 1968), pp. 327–333. — [3]
Extensions of p-bounded groups. Arch. Math. 21, 449–454 (1970). — [4] Detachable p-groups
and quasi-injectivity. Acta Math. Acad. Sci. Hungar. 27, 71–73 (1976). — [5] An extension
of the theory of completely decomposable torsion-free abelian groups. Trans. Am. Math. Soc.
279, 175–185 (1983). — [6] Butler groups, valuated vector spaces and duality. Rend. Sem. Mat.
Univ. Padova 72, 13–19 (1984). — [7] The constructive theory of torsion-free abelian groups.
Commun. Algebra 18, 3913–3922 (1990)
F. Richman, E.A. Walker — [1] Primary abelian groups as modules over their endomorphism
rings. Math. Z. 89, 77–81 (1965). — [2] Extending Ulm’s theorem without group theory. Proc.
Am. Math. Soc. 21, 194–196 (1969). — [3] Modules over PID’s that are injective over their
endomorphism rings, in Ring Theory. Proceedings of the Park City Conference (Academic,
New York, 1972), pp. 363–372. — [4] Homological dimension of abelian groups over their
References 727

endomorphism rings. Proc. Am. Math. Soc. 54, 65–68 (1976). — [5] Valuated groups. J.
Algebra 56, 145–167 (1979). — [7] Cyclic Ext. Rocky Mt. J. Math. 11, 611–615 (1981)
L.A. Rogers — [1] Ulm’s theorem for partially ordered structures related to simply presented
abelian p-groups. Trans. Am. Math. Soc. 227, 333–343 (1977)
J. Rotman — [1] Mixed modules over valuation rings. Pac. J. Math. 10 (960), 607–623. — [2]
Torsion-free and mixed abelian groups. Ill. J. Math. 5, 131–143 (1961). — [3] On a problem
of Baer and a problem of Whitehead in abelian groups. Acta Math. Acad. Sci. Hungar. 12,
245–254 (1961). — [4] Sheaves and abelian groups. Symposia Math. 13, 11–20 (1974)
J. Rotman, T. Yen — [1] Modules over a complete discrete valuation ring. Trans. Am. Math. Soc.
98, 242–254 (1961)
S.V. Rychkov — [1] On direct products of abelian groups [Russian]. Dokl. Akad. Nauk SSSR
252, 301–302 (1980). — [2] The problem of splitting of pure extensions of abelian groups and
axiomatic set theory [Russian]. Uspekhi Mat. Nauk 40, 195–196 (1985). — [3] Indecomposable
abelian p-groups that are almost direct sums of cyclic groups [Russian]. Mat. Zametki 43,
705–712 (1988); Math. Notes 43, 405–408 (1988). — [4] Abelian k-separable groups [Rus-
sian], in Abelian Groups and Modules, vol. 7 (Tomsk. Gos. Univ., Tomsk, 1988), pp. 110–120
S.V. Rychkov, A.A. Fomin — [1] Abelian groups with a countable number of subgroups [Russian],
in Abelian Groups and Modules, vol. 10 (Tomsk. Gos. Univ., Tomsk, 1991), pp. 99–105
S.V. Rychkov, B. Thomé — [1] Slender groups and related concepts. Commun. Algebra 14,
333–387 (1986)
G. Sabbagh — [1] How not to characterize the multiplicative group of fields. J. Lond. Math. Soc.
1, 369–370 (1969)
L. Salce — [1] I -groups and almost totally injective groups. Ann. Univ. Ferrara Sez. VII 25,
197–204 (1979). — [2] Cotorsion theories for abelian groups. Symposia Math. 23, 11–32
(1979)
L. Salce, F. Menegazzo — [1] Abelian groups whose endomorphism ring is linearly compact.
Rend. Sem. Mat. Univ. Padova 53, 315–325 (1975)
E. Sa̧siada — [1] Construction of a directly indecomposable abelian group of the power higher than
that of the continuum. Bull. Acad. Polon. Sci. Cl. III 5, 701–703 (1957); Bull. Acad. Polon.
Sci. Cl. III 7, 23–26 (1959). — [2] Proof that every countable and reduced torsion-free abelian
group is slender. Bull. Acad. Polon. Sci. Cl. III 7, 143–144 (1959). — [3] On the isomorphism
of decompositions of torsion-free abelian groups into complete direct sums of groups of rank
one. Bull. Acad. Polon. Sci. Cl. III 7, 145–149 (1959). — [4] On two problems concerning
endomorphism groups. Ann. Univ. Sci. Budapest 2, 65–66 (1959)
C.L. Schochet — [1] A Pext primer: pure extensions and lim1 for infinite abelian groups. NYJM
Monographs, vol. 1 (State University of New York, Albany, 2003)
P. Schultz — [1] The endomorphism ring of the additive group of a ring. J. Aust. Math. Soc.
15, 60–69 (1973). — [2] When is an abelian p-group determined by the Jacobson radical
of its endomorphism ring? in Abelian Group Theory and Related Topics, Contemporary
Mathematics, vol. 171 (1994), pp. 385–396. — [3] Automorphisms which determine an abelian
p-group, in Abelian Groups, Module Theory, and Topology. Lecture Notes in Pure and Applied
Mathematics, vol. 201 (Dekker, New York, 1998) pp. 373–379. — [4] Automorphism groups of
abelian groups, in Abelian Groups, Rings and Modules, Contemporary Mathematics, vol. 273
(American Mathematical Society, Providence, RI, 2001) 51–62. — [5] Self-splitting abelian
groups. Bull. Aust. Math. Soc. 64, 71–79 (2001)
A.M. Sebel’din — [1] Conditions of the isomorphism of completely decomposable torsion-
free abelian groups with isomorphic rings of endomorphisms [Russian]. Mat. Zametki 11,
403–408 (1972). — [2] Homomorphism groups of complete direct sums of torsion-free abelian
groups of rank 1 [Russian]. Tomsk. Gos. Univ., Tomsk 1, 121–122 (1974). — [3] Determination
of abelian groups by their semigroups of endomorphisms [Russian], in Abelian Groups and
Modules, vol. 10 (Tomsk. Gos. Univ., Tomsk, 1991), pp. 125–133
S. Shelah — [1] Infinite abelian groups, Whitehead problem and some constructions. Isr. J.
Math. 18, 243–256 (1974). — [2] Existence of rigid-like families of abelian p-groups, in
Model Theory and Algebra. Lecture Notes in Mathematics, vol. 498 (Springer, Berlin, 1975),
728 References

pp. 384–402. — [3] Whitehead groups may not be free even assuming CH. I: Isr. J. Math.
28, 193–204 (1977). II: Isr. J. Math. 35, 257–285 (1980). — [4] A combinatorial theorem
and endomorphism rings of abelian groups. II: Abelian Groups and Modules. CISM Courses
Lectures, vol. 287 (Springer, Berlin, 1984), pp. 37–86
S. Shelah, L. Strüngmann — [1] It is consistent with ZFC that B1 -groups are not B2 . Forum Math.
15, 507–524 (2003). — [2] A characterization of Ext.G; Z/ assuming .V D L/. Fund. Math.
193, 141–151 (2007)
M. Shiffman — [1] The ring of automorphisms of an abelian group. Duke Math. J. 6, 579–597
(1940)
A.Z. Shlyafer — [1] Solvability of automorphism groups of abelian groups, in Abelian Groups and
Modules, vol. 7 ((Tomsk. Gos. Univ., Tomsk, 1988), pp. 148–155
K. Shoda — [1] Über die Automorphismen einer endlichen abelschen Gruppe. Math. Ann. 100,
674–686 (1928)
T. Skolem — [1] On the existence of a multiplicative basis. Norske Vid. Selsk. Forh. 2, 4–7 (1947)
A.Yu. Soifer — [1] A theorem on direct decompositions of abelian groups [Russian]. Mat. Zametki
14, 879–884 (1973); Math. Notes 14, 1064–1067 (1973)
E. Specker — [1] Additive Gruppen von Folgen ganzer Zahlen. Portugaliae Math. 9, 131–140
(1950)
R.O. Stanton — [1] An invariant for modules over a discrete valuation ring. Proc. Am. Math. Soc.
49, 51–54 (1975). — [2] Decomposition bases and Ulm’s theorem, in Abelian Group Theory.
Lecture Notes in Mathematics, vol. 616 (Springer, Berlin, 1977), pp. 39–56. — [3] Almost
affable abelian groups. J. Pure Appl. Algebra 15, 41–52 (1979)
J. Stelzer — [1] A cancellation criterion for finite rank torsion-free abelian groups. Proc. Am.
Math. Soc. 94, 363–368 (1985)
J. Stepráns — [1] A characterization of free abelian groups. Proc. Am. Math. Soc. 93, 347–349
(1988)
A.E. Stratton — [1] Mixed modules over an incomplete discrete valuation ring. Proc. Am. Math.
Soc. 21, 201–218 (1970)
R.W. Stringall — [1] Decompositions of abelian p-groups. Proc. Am. Math. Soc. 28, 409–410
(1971)
L. Strüngmann — [1] On problems by Baer and Kulikov using V = L. Ill. J. Math. 46, 477–490
(2002). — [2] Baer cotorsion pairs. Isr. J. Math. 151, 29–51 (2006). — [3] On endomorphism
rings of B1 -groups that are not B2 -groups. Proc. Am. Math. Soc. 137, 3657–3668 (2009)
F.A. Szász — [1] Die abelschen Gruppen, deren volle Endomorphismenringe die Minimalbedin-
gung für Hauptideale erfüllen. Monatshefte Math. 65, 150–153 (1961). — [2] Über Artinsche
Ringe. Bull. Acad. Polon. Sci. Cl. III 11, 351–354 (1963)
G. Szekeres — [1] Countable abelian groups without torsion. Duke Math. J. 15, 293–306 (1948)
T. Szele — [1] Zur Theorie der Zeroringe. Math. Ann. 121, 242–246 (1949). — [2] Grup-
pentheoretische Beziehungen der Primkörper. Mat. Aineiden Aika. 13, 80–85 (1949). — [3]
Ein Analogon der Körpertheorie für abelsche Gruppen. J. Reine Angew. Math. 188, 167–192
(1950). — [4] On direct decompositions of abelian groups. J. Lond. Math. Soc. 28, 247–250
(1953). — [5] On the basic subgroups of abelian p-groups. Acta Math. Acad. Sci. Hungar. 5,
129–141 (1954). — [6] Nilpotent Artinian rings. Publ. Math. Debrecen 4, 71–78 (1955). — [7]
On quasi-decomposable abelian groups. Acta Math. Acad. Sci. Hungar. 7, 109–114 (1956).
— [8] On a topology in endomorphism rings of abelian groups. Publ. Math. Debrecen 5, 1–4
(1957)
T. Szele, L. Fuchs — [1] On Artinian rings. Acta Sci. Math. Szeged 17, 30–40 (1956)
W. Szmielew — [1] Elementary properties of abelian groups. Fund. Math. 41, 203–271 (1955)
S.G. Tellman — [1] Images of induced endomorphisms in Ext.H; G/. Acta Sci. Math. Szeged 23,
290–291 (1962)
S. Thomas — [1] Borel superrigity and the classification problem for torsion-free abelian groups
of finite rank, in International Congress of Mathematicians, vol. II (European Mathematical
Society, 2006), pp. 93–116. — [2] The classification problem for finite rank Butler groups, in
Models, Modules and Abelian groups (W. de Gruyter, Berlin, 2008), pp. 329–338
References 729

G.M. Tsukerman — [1] Rings of endomorphisms of free modules [Russian]. Sibirsk. Mat. Z̆. 7,
1161–1167 (1966)
M. Turgi — [1] A sheaf-theoretical interpretation of the Kurosh theorem, in Abelian Group Theory.
Lecture Notes in Mathematics, vol. 616 (Springer, Berlin, 1977), pp. 173–196
M.A. Turmanov — [1] On pureness in abelian groups. J. Sci. Math. 137, 5336–5345 (2006)
H. Ulm — [1] Zur Theorie der abzählbar-unendlichen abelschen Gruppen. Math. Ann. 107,
774–803 (1933)
C. Vinsonhaler — [1] Torsion-free abelian groups quasi-projective over their endomorphism
rings. II. Pac. J. Math. 74, 261–265 (1978); Pac. J. Math. 79, 564–565 (1979). — [2] E-
rings and related structures, in Non-Noetherian commutative ring theory. Mathematics and its
Applications, vol. 520 (Kluwer Academic, Dordrecht, 2000), pp. 387–402
C. Vinsonhaler, S. Wallutis, W.J. Wickless — [1] A class of B.2/ -groups. Commun. Algebra 33,
2025–2035 (2005)
C. Vinsonhaler, W.J. Wickless — [1] Torsion-free abelian groups quasi-projective over their
endomorphism rings. Pac. J. Math. 68, 527–535 (1977). — [2] The injective hull of a
separable p-group as an E-module. J. Algebra 71, 32–39 (1981). — [3] Balanced projective
and cobalanced injective torsion-free groups of finite rank. Acta Math. Acad. Sci. Hungar. 46,
217–225 (1985). — [4] Dualities for torsion-free abelian groups of finite rank. J. Algebra 128,
474–487 (1990)
B. Wald — [1] On -products modulo -products, in Abelian Group Theory. Lecture Notes in
Mathematics, vol. 1006 (Springer, Berlin, 1983), pp. 362–370
C.P. Walker — [1] Properties of Ext and quasi-splitting of abelian groups. Acta Math. Acad. Sci.
Hungar. 15, 157–160 (1964). — [2] Relative homological algebra and abelian groups. Ill.
J. Math. 10, 186–209 (1966). — [3] Projective classes of completely decomposable abelian
groups. Arch. Math. 23, 581–588 (1972)
E.A. Walker — [1] Cancellation in direct sums of groups. Proc. Am. Math. Soc. 7, 898–902 (1956).
— [2] Quotient categories and quasi-isomorphisms of abelian groups, in Proceedings of the
Colloquium on Abelian Groups (Akadémiai Kiadó, Budapest, 1964), pp. 147–162. — [3] Ulm’s
theorem for totally projective groups. Proc. Am. Math. Soc. 37, 387–392 (1973). — [4] The
groups Pˇ . Symposia Math. 13, 245–255 (1974)
K.D. Wallace — [1] On mixed groups of torsion-free rank one with totally projective primary
components. J. Algebra 17, 482–488 (1971)
J.D. Waller — [1] Generalized torsion complete groups, in Studies on Abelian Groups (Dunod,
Paris, 1968), pp. 345–356
R.B. Warfield, Jr. — [1] Homomorphisms and duality for torsion-free groups. Math. Z. 107,
189–200 (1968). — [2] A Krull-Schmidt theorem for infinite sums of modules. Proc. Am.
Math. Soc. 22, 460–465 (1969). — [3] An isomorphic refinement theorem for abelian groups.
Pac. J. Math. 34, 237–255 (1970). — [4] Simply presented groups, in Proceedings of the Special
Semester on Abelian Groups (University of Arizona, Tucson, 1972). — [5] The uniqueness of
elongations of abelian groups. Pac. J. Math. 52, 289–304 (1974). — [6] A classification theorem
for abelian p-groups. Trans. Am. Math. Soc. 210, 149–168 (1975). — [7] Classification theory
of abelian groups. I: Balanced projectives. Trans. Am. Math. Soc. 222, 33–63 (1976). II: Local
theory, in Abelian Group Theory. Lecture Notes in Mathematics, vol. 874 (Springer, Berlin,
1981), pp. 322–349. — [8] The structure theory of mixed abelian groups, in Abelian Group
Theory. Lecture Notes in Mathematics, vol. 616 (Springer, Berlin, 1977), pp. 1–38. — [9]
Cancellation of modules and groups and stable range of endomorphism rings. Pac. J. Math. 91,
457–485 (1980)
M.C. Webb — [1] The endomorphism ring of homogeneously decomposable separable groups.
Arch. Math. 31, 235–243 (1978)
B.D. Wick — [1] A projective characterization for SKT-modules. Proc. Am. Math. Soc. 80, 39–43
(1980). — [2] A classification theorem for SKT-modules. Proc. Am. Math. Soc. 80, 44–46
(1980)
W.J. Wickless — [1] Abelian groups which admit only nilpotent multiplications. Pac. J. Math. 40,
251–259 (1972). — [2] T as an E -submodule of G. Pac. J. Math. 83, 555–564 (1979)
730 References

G.V. Wilson — [1] Modules with the summand intersection property. Commun. Algebra 14, 21–38
(1986).
K.G. Wolfson — [1] Baer rings of endomorphisms. Math. Ann. 143, 19–28 (1961). — [2]
Isomorphisms of the endomorphism rings of a class of torsion-free modules. Proc. Am. Math.
Soc. 14, 589–594 (1963)
S.M. Yahya — [1] p-pure exact sequences and the group of p-pure extensions. Ann. Univ. Sci.
Budapest 5, 179–191 (1962)
A.V. Yakovlev — [1] Direct decompositions of torsion-free abelian groups of finite rank [Russian].
Zap. Nauchn. Sem. Leningr. Otdel. Mat. Inst. Steklov 160, 272–285 (1987); J. Sov. Math.
52, 3206–3216 (1990). — [2] Torsion-free abelian groups of finite rank and their direct
decompositions [Russian]. Zap. Nauchn. Sem. Leningr. Otdel. Mat. Inst. Steklov. 175 (1989);
Koltsa i Moduli 3, 135–153, 165; J. Sov. Math. 57, 3524–3533 (1991). — [3] Direct
decompositions of mixed abelian groups [Russian]. Vestnik St. Petersburg Univ. Math. 43,
3–11 (2010)
H. Yamabe — [1] A condition for an abelian group to be a free abelian group with a finite basis.
Proc. Jpn. Acad. 77, 205–207 (1951)
P.D. Yom — [1] A characterization of a class of Butler groups. I. Commun. Algebra 25, 3721–3734
(1997); II: Abelian Group Theory and Related Topics, Contemporary Mathematics, vol. 171
(American Mathematical Society, Providence, RI, 1994), pp. 419–432
H. Zassenhaus — [1] Orders as endomorphism rings of modules of the same rank. J. Lond. Math.
Soc. 42, 180–182 (1967)
E.C. Zeeman — [1] On direct sums of free cycles. J. Lond. Math. Soc. 30, 195–212 (1955)
B. Zimmermann-Huisgen — [1] On Fuchs’ Problem 76. J. Reine Angew. Math. 309, 86–91 (1979)
B. Zimmermann-Huisgen, W. Zimmermann — [1] Algebraically compact rings and modules.
Math. Z. 61, 81–93 (1978)
L. Zippin — [1] Countable torsion groups. Ann. Math. 36, 86–99 (1935)
F. Zorzitto — [1] Discretely normed abelian groups. Aequationes Math. 29, 172–174 (1985)
Author Index

A Bican, L., 385, 420, 422, 424, 529, 532, 535,

Abel, N.H., vii 541, 544–548, 553, 556, 561, 562,
Albrecht, U., x, 111, 219, 391, 422, 430, 478, 568
507, 513, 537, 550, 568, 578, 595, Birtz, A., 488, 570
642, 649 Blackburn, N., 190
Angad-Gaur, H.W.K., 649 Blagoveshchenskaya, E.A., 444, 572, 633
Angeleri Hügel, L., 582 Blass, A., 118, 129, 172, 488, 507
Armstrong, J.W., 436 Blazhenov, A.V., 469
Arnold, D.M., 111, 430, 451, 457, 461, 465, Bobylev, I.V., 649
468, 469, 473, 476, 478, 533, 535, Bognár, M., 432
536, 548, 551, 571, 601, 604, 618, Borho, W., 680
649, 652, 680 Bourbaki, N., 172
Azumaya, G., 210, 211 Bowman, H., 649
Bowshell, R.A., 694–696
Boyer, D.L., 38, 168, 664
B Braconnier, J., 153
Baer, R., 20, 96, 113, 135, 139, 168, 172, 257, Brameret, M.P., 650
260, 261, 270, 275, 301, 307, 349, Brandl, R., 663
411, 415, 416, 425–427, 430, 431, Breaz, S., 578, 595
501–503, 508, 575, 578–580, 583, Brown, R., 337
607, 624, 625, 657, 661, 663, 664, de Bruijn, N.G., 91
673, 675 Bunina, E.I., 628, 663
Balcerzyk, S., 74, 183, 187, 189, 199, 511 Burkhardt, R., 535
Bang, C.M., 592 Butler, M.C.R., 529, 530, 532, 535, 546
Barwise, J., 351
Bass, H., 79, 450
Baumslag, G., 190 C
Bazzoni, S., 475, 582, 633 Cǎlugǎreanu, G.G., 642, 643
Beaumont, R.A., 455, 473, 513, 676, 678, Campbell, M.O’N., 415
681 Cartan, H., 139, 216, 233, 239, 240, 269, 648
Bekker, I.Kh., 670 Castagna, F., 628
Benabdallah, K., 153, 325, 399, 488, 570 Cellars, R.M., 396
Bergman, G.M., 126, 140, 681 Chachólski, W., 228
Berman, S.D., 700 Charles, B., 35, 38, 140, 169, 306, 319–321,
Bialynicki-Birula, A., 511 338, 436, 625

© Springer International Publishing Switzerland 2015 731

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6
732 Author Index

Chase, S.U., 66, 68, 508, 527 Enochs, E.E., 53, 286, 296–298, 301, 316, 318,
Chekhlov, A.R., 111, 165, 415, 633 341
Claborn, L., 18 Erdélyi, M., 156
Cohen, J., 99, 100, 102 Erdős, J., 98, 103-105
Cohen, P.J., 20
Cohn, P.M., 86, 158, 181, 189, 469, 700
Cornelius, E.F., Jr., 507 F
Corner, A.L.S., 227, 237, 306, 435, 437, 439, Faltings, K., 225, 317, 663
446, 481, 483, 485, 486, 488, 579, Farahat, H.K., 646, 649, 693
616, 627, 629, 631–633, 652, 665, Farjoun, E.D., 228
667, 669, 670, 678 Faticoni, T.G., 445, 449, 464, 475, 620, 633,
Crawley, P., 207, 210, 211, 319, 321, 329, 351, 648, 649, 696, 697
355, 358, 360, 361, 363, 391, 404, Fay, T.H., 38
453 Feigelstock, S., 650, 673, 680, 681, 692, 697
Curtis, C.W., 449 Files, S.T., 306, 595, 628
Cutler, D.O., 105, 159, 178, 250, 320, 321, Fink, T., 518
330, 341, 342, 391, 400, 408 Flagg, M.A., 627
Fleischer, I., 194
Fomin, A.A., 146, 244, 474, 475, 578
D Fomin, S.V., 575, 578
Danchev, P.V., 112, 398, 400-402, 700 Franzen, B., 189, 607, 652
Dauns, J., 74 Freedman, H., 658, 663
De Marco, G., 285, 496 Fried, E., 680, 681
Derry, D., 415, 436, 454 Frobenius, G., 50, 81, 85, 86
D’Este, G., 628
De Vivo, C., 536
Dickson, S.E., 241 G
Dieudonné, J., 99, 243 Gacsályi, S., 144, 145, 153
Dikranjan, D., 621 Gardner, B.J., 681
Ditor, S.Z., 699 Gauss, C.F., 49, 85
Dixmier, J., 119, 524 Generalov, A.I., 105
Dlab, V., 99, 621 Gerdt, I.V., 219
Douglas, A.J., 646, 649, 693 Gerstner, O., 189
Dubois, P.F., 396 Gilmer, R.W., 700
Dugas, M., 30, 67, 68, 172, 189, 228, 341, 342, Giovannitti, A.J., 536, 539, 541
418, 437, 469, 488, 507, 535, 542, Glaz, S., 458, 642
543, 545, 547–549, 551, 556, 563, Gluck, H., 99, 100, 102
565, 568, 571, 627, 632, 649, 651, Göbel, R., 25, 119, 159, 187, 189, 227, 228,
652, 696, 697 265, 286, 293, 333, 351, 376, 437,
Dung, N.V., 139 488, 495, 498, 500, 508, 572, 605,
616, 627, 632, 651, 652, 697
Gödel, K., 20, 23
E Goeters, H.P., 415, 422, 430, 475, 536, 648,
Eda, K., 49, 68, 74, 119, 493–495, 500, 505, 649, 697
507, 508, 510, 514, 518, 649 Golan, J., 5
Ehrenfeucht, A., 495, 520, 523 Goldsmith, B., x, 159, 306, 607, 621, 627, 628,
Eilenberg, S., 35, 139, 216, 233, 239, 240, 255, 649, 652, 670
260, 261, 263, 268, 269, 648 Goodearl, K.R., 453
Eklof, P.C., 25, 50, 107, 109, 114, 116, 117, Gräbe, P.J., 442, 445, 487
119–121, 198, 269, 270, 272, 275, Grätzer, G., 53
342, 351, 407, 430, 437, 475, 478, Gregory, J., 116, 119
488, 494–496, 506, 507, 527, 528, Griffith, P., 110, 116, 119, 333, 370, 403, 429,
581, 582, 651, 681 430, 436, 507, 509, 527, 568, 579,
El Bashir, R., 548 580, 582, 583
Author Index 733

Grinshpon, S.Ya., 508, 589, 595 Ivanov, A.V., 66, 68, 211, 341, 508, 513, 627,
de Groot, J., 85, 125, 126, 147, 437, 438 642, 649
Gruson, L., 117–119

J
H Jackett, D.R., 680, 681, 692
Haimo, F., 418, 681 Jacoby, C., 605
Hajós, G., 86–90 Janakiraman, S., 153
Hales, A.W., 351, 355, 358, 360, 361, Jans, J.P., 129, 650
363 Jarisch, R., 605
Hallett, J.T., 667, 670, 671 Jech, T., 124, 312, 330, 332
Halperin, I., 691, 692 Jenda, O.M.G., 298
Harrison, D.K., 220, 222, 242–244, 278, 281, Jensen, C.U., 64, 280, 292
282, 287–289, 291, 293 Jensen, R., 23, 24
Hauptfleisch, G.J., 633 Johnson, J.A., 627, 696
Hausen, J., 79, 111, 112, 627, 649, 650, 663, Johnson, R.E., 137, 139
664, 696, 697 Jónsson, B., 207, 210, 211, 319, 415, 445, 458,
Head, T.J., 237, 323, 325 462, 464, 469
Heinlein, G., 496 Joubert, S.V., 579
Hensel, K., 704
Herbera, D., 582
Herden, G., 437, 697 K
Higman, G., 700 Kakutani, S., 223
Hill, P., 26, 28, 31, 96, 98, 104, 105, 109, Kamalov, F.F., 111
116, 119, 122, 126, 152, 180, 219, Kanamori, A., 437
250, 270, 293, 301, 306, 322–325, Kaplansky, I., 52, 79, 84, 85, 99, 111, 153, 172,
329, 338–342, 350, 352–354, 358, 189, 191, 192, 194–196, 198, 300,
360–364, 368, 369, 373, 376–379, 302, 307, 309, 345, 350, 351, 427,
381, 382, 384, 385, 392, 401–403, 430, 438, 455, 469, 582, 595, 624,
404, 419, 422, 430, 458, 507, 508, 625
513, 536, 537, 545, 550, 551, 563, Karpenko, A.V., 642
568, 604, 628, 658, 663, 664, 681 Karpilovsky, G., 700
Hiller, H.L., 293 Kasch, F., 642
Hirsch, K.A., 667, 670, 671 Kaup, L., 125
Hjorth, G., 478 Kechris, A.S., 478
Hodges, W., 121, 122 Keef, P.F., 69, 180, 248–251, 326–328, 341,
Höfling, B., 417 389, 391, 392, 400
Hofmann, K.H., 205, 206 Keller, O.H., 90
Honda, K.Y., 5, 86, 153, 154, 401, 402, 469 Kemoklidze, T., 298, 317
Hopkins, C., 688 Kertész, A., 37, 111, 139, 142, 143, 145, 431
Huber, M., 269, 275, 285, 286, 293, 407, 495, Khabbaz, S.A., 140, 156, 179, 180, 331, 652,
508, 513, 518, 649 664
Hulanicki, A., 187, 222, 437 Kiefer, F., 376
Hunter, R.H., 351, 370, 422, 429, 461, 588, Kil’p, M.A., 138
601, 604, 605 Kleane, M.S., 125
Huynh, D.V., 139, 688 Koehler, J., 414, 535
Kogalowski, S.R., 705
Kojman, M., 589
I Kolettis, G., Jr., 350, 375, 376, 426
Irwin, J.M., 53, 105, 118, 129, 155, 172, 178, Kompantseva, E.I, 681
180, 281, 301, 315, 321, 323, 325, Kovács, L., 170
330, 341, 351, 365, 370, 372, 373, Koyama, T., 321, 323
376, 377, 386, 391, 392, 395, 396, Kozhukhov, S.F., 464
399, 400, 495, 507, 592, 652 Kravchenko, A.A., 422, 430, 541
734 Author Index

Król, M., 512, 513, 670 587, 589–591, 594–596, 604, 611,
Kruchkov, N.I., 500 628, 653, 664
Krylov, P.A., 165, 415, 464, 475, 479, 488, Meinel, K., 570
508, 589, 595, 613, 620, 633, 645, Mekler, A.H., 114, 119–121, 198, 281, 341,
649 342, 405–407, 437, 475, 494–496,
Kulikov, L. Ya., 20, 53, 94, 97, 147, 149, 506–508, 527, 528, 651
155–157, 159, 167, 171–175, 293, Melles, G., 478
303, 305, 306, 311, 312, 316, 318, Menegazzo, F., 642
320, 330, 350, 365, 370, 427, 430, Metelli, C., x, 415, 508, 509, 536, 539, 554,
578 569, 571, 572, 632, 633
Kurosh, A.G., 96, 146, 305, 415, 436 Mez, H.C., 681
Mikhalëv, A.V., 613, 620, 628, 645, 649
Mines, R., 396, 536
L Minkowski, H., 86, 90
Lady, E.L., 324, 430, 446, 448, 457, 458, 465, de Miranda, A.B, 671
467, 469, 476, 478, 534, 535 Mishina, A.P., 139, 154, 510, 512, 513, 579,
Lane, M., 458, 611 620, 660
Lausch, H., 253 Missel, C., 400
Lawrence, J., 123, 126 Misyakov, V.M., 589, 642
Lawver, D.A., 621, 681 Misyakova, A.V., 642
Lazaruk, J., 400 Mohamed, S.H., 139
Lee, W.Y., 536 Mollov, T.Zh., 700
Lefschetz, S., 224 Monk, G.S., 211, 227, 333, 627
Leistner, K., 605 Moore, J.H., 604
Leptin, H., 197, 199, 315, 316, 321, 660, 663 Morris, S.A., 206
Levi, F.W., 415, 436, 578, 628 Moskalenko, A.I., 269, 298
Liebert, W., 622, 625, 627, 633, 650, 663 Müller, B.J., 139
Linton, R.C., 377, 384, 391 Müller, E., 416
Loonstra, F., 440, 443, 445, 469 Murley, C.E., 463–465, 476, 476, 618
Łoś, J., 48, 67, 164, 183, 189, 489, 492, 495, Mutzbauer, O., 172, 414–416, 445, 475, 605
511, 513 Myshkin, V.I., 586, 594, 595
Loth, P., 206, 604, 605
Lyapin, E.S., 422
N
Nachev, N.A., 700
M Nedov, V.N., 670
Mackey, G.W., 345, 350, 595 Neumann, B.H., 6
MacLane, S., 13, 35, 255, 260, 261, 268, 422, Nicholson, W.K., 211
539 Niedzwecki, G.P., 649, 650, 680, 705
Mader, A., 38, 159, 475, 529, 534, 535, 567, Nöbeling, G., 125, 126
578, 642, 649, 663, 664, 696 Noether, E., 8
Magidor, M., 120, 437, 566, 568 Nongxa, L.G., 414, 429, 430, 536
Mal’cev, A.I., 415, 436, 670, 705 Nunke, R.J., 35, 154, 180, 245–250, 270, 276,
Maranda, J.M., 164, 165, 183, 189, 199, 200 280, 282, 304, 369, 371–374, 376,
Martinez, J., 528 377, 386, 388, 391–394, 397, 400,
Matlis, E., 139, 286 402–404, 407, 495, 497, 498, 501,
May, W., 351, 376, 620, 627, 628, 652, 670, 503, 508, 527, 574, 575, 628
673, 700, 705
McCoy, N.H., 5
Meehan, C., 628 O
Megibben, C.K., 104, 105, 152, 188, 227, 270, Ohlhoff, H.J.K., 579
301, 306, 316, 317, 322–329, 339, Ohta, H., 507
377, 385, 390, 391, 401–405, 407, Okuyama, T., 153
430, 458, 507, 508, 536, 584, 586, O’Meara, K.C., 469
Author Index 735

O’Neill, J.D., 315, 321, 511, 514, 516–519, Rogers, L.A., 351
683 Roizner, M.A., 663
Oppelt, J.A., 579 Rotman, J., 458, 508, 519, 520, 527, 573,
Ore, O., 5 586–588, 590, 592, 594–596, 599,
Orsatti, A., 285, 496, 629, 633, 641 607
Osofsky, B.L., 105 Rychkov, S.V., 119, 146, 187, 189, 327, 328,
Ould-Beddi, M.A., 105 333, 508, 527

P S
Pabst, S. See Wallutis, S.L. Sabbagh, G., 202, 488, 705
Palyutin, E.A., 119 Salce, L., x, 162, 172, 198, 201, 286, 391, 396,
Papp, Z., 139 397, 400, 403, 475, 544–547, 568,
Paras, A.T., 649 621, 632, 642
Parker, L.D., 375, 376 Samelson, H., 206
Pearson, K.R., 700, 706 Sands, A.D., 90, 91
Pfaendtner, J., 105 Sa̧siada, E., 189, 437, 446, 490, 508, 511, 513,
Pierce, R.S., 153, 180, 219, 220, 225–228, 583, 629
308–310, 329, 445, 473, 595, 619, Schenkman, E., 705
621, 622, 624–627, 649, 650, 670, Schlitt, G., 475, 508
676, 678, 696 Schmidt, E.T., 53
Pokutta, A.T., 508 Schneider, J.E., 700, 706
Pontryagin, L., 106, 119, 433, 436 Schochet, C.L., 64, 281
Poole, G.D., 649, 650 Schoeman, M., 281, 579
Praeger, C.E., 627 Schöneborn, H., 197
Prelle, R., 265 Schreier, O., 255, 261
Procházka, L., 430, 431, 459, 462, 464, 545, Schultz, P., 293, 627, 663, 693–697
578 Scott, W.R., 142
Prüfer, H., 96, 146, 149, 151, 153, 155, 156, Sebel’din, A.M., 219, 627, 632
158, 172, 175 Segev, Y., 228
Puusemp, P., 627 Shelah, S., 20, 25, 107, 119, 120, 122, 269,
286, 293, 332, 333, 341, 342,
405–407, 435, 437, 488, 495, 508,
R 519, 521, 523, 527, 566–568, 582,
Rado, R., 81 589, 651, 652, 696, 697
Rafiq, M., 399 Shiffman, M., 307
Rangaswamy, K.M., x, 115, 128, 129, 153, Shlyafer, A.Z., 663
154, 298, 419, 430, 431, 507, 542, Shoda, K., 663
543, 545, 547–549, 551, 553, 554, Sierpinski, W., 29
561–563, 567, 568, 634, 636, 637, Skolem, T., 702
642, 643, 649, 690, 692 Skornyakov, L.A, 154
Raphael, R., 697 Smith, H.J.S., 85
Raynaud, M., 117–119 Smith, P.F., 139
Rédei, L., 90, 673, 676 Snabb, T., 105, 396, 400, 495
Ree, R., 681 Soifer, A.Yu., 333, 416
Reid, G.A., 114, 506 Solovay, R., 25
Reid, J.D., 165, 461, 464, 477, 478, 489, 491, Specker, E., 113, 125, 126, 491, 495, 496
496, 631, 633, 649, 650, 705 Stanton, R.O., 602–604
Reiner, I., 449 Stein, K., 114, 523
Remak, R., 50 Steinfeld, O., 676
Richman, F., 250, 317, 323, 325, 326, 341, Stelzer, J., 469
350, 351, 407, 408, 415, 445, 461, Stenström, B., 154
535–537, 541, 585, 588, 601, 604, Stepráns, J., 124, 126
605, 623, 646, 648, 649 Stickelberger, L., 50, 81, 85
736 Author Index

Stratton, A.E., 579, 595 Walker, E.A., 53, 86, 142, 159, 172, 180, 281,
Stringall, R.W., 306, 320, 321 301, 323, 341, 350, 351, 356, 358,
Strüngmann, L., x, 105, 269, 567, 568, 572, 365, 370–373, 375–377, 384, 386,
579, 582, 583, 605, 697 391, 392, 395, 400, 415, 445, 462,
Szabó, S., 90 469, 585, 588, 592, 604, 605, 623,
Szász, F.A., 638, 686, 688 646, 648, 649, 664
Szekeres, G., 457, 458 Wallace, K.D., 383, 590, 595
Szele, T., 37, 90, 99, 139, 140, 142, 145, 151, Waller, J.D., 316, 397
155, 169, 172, 176, 177, 181, 331, Walls, G.L., 38
431, 436, 617, 628, 638, 642, 643, Wallutis, S.L. (S. Pabst), 25, 286, 475, 536,
673, 676, 678, 682, 684–687 628
Szélpál, I., 134, 636 Warfield, R.B., Jr., 158, 189, 207, 208, 210,
Szendrei, J., 628, 643 211, 285, 286, 370, 371, 385, 404,
Szmielew, W., 351 407, 415, 416, 450–454, 470, 473,
475, 479, 573, 594, 596–600, 604,
608–612
T Webb, M.C., 633
Tarski, A., 469 Whitehead, J.H.C., 523
Tarwater, D., 664 Whitney, H., 236
Tellman, S.G., 276 Wick, B.D., 611
Tenenbaum, S., 25 Wickless, W.J., 458, 475, 536, 578, 595, 628,
Thomas, S., 478, 536 642, 649, 650, 678
Thomé, B., 30, 327, 328, 556, 565, 568, 571, Wilson, G.V., 111, 112
632 Wisbauer, R., 139
Toubassi, E.H., 172, 605, 627 Wisner, R.J., 681
Trlifaj, J., 286, 651 Wolfson, K.G., 632, 634
Tsukerman, G.M., 634 Wong, E.T., 139
Tuganbaev, A.A., 479, 613, 620, 645, 649 Wu, L.E.T., 129, 650
Turgi, M., 458
Turmanov, M.A., 181
Y
Yahya, S.M., 146, 161
U
Yakovlev, A.V., 415, 431, 444, 607
Ullery, W., 536, 604, 611, 628, 700
Yamabe, H., 126
Ulm, H., 346, 350
Yen, T., 495, 588, 592, 595, 599,
607
V Yom, P.D., 536
Vámos, P., 449, 649
Vergohsen, R., 341, 342
Viljoen, G., 428, 540, 541 Z
Vinsonhaler, C., 181, 430, 458, 469, 475, 533, Zanardo, P., 201, 621
535, 536, 562, 578, 649, 650, 652, Zassenhaus, H., 632
696 Zeeman, E.C., 493, 495
de Vries, H., 671 Zheludev, M.V., 105
Ziegler, M., 333, 488
Zimmermann, W., 198, 209, 211, 636
W Zimmermann-Huisgen, B. 66–68, 198, 209,
Wald, B., 187, 189, 287, 495 211, 332, 508, 636
Walker, C.P., 52, 54, 154, 165, 281, 372, 373, Zippin, L., 347, 350
377, 386, 391, 392, 395, 400, 423, Zorzitto, F., 126
577, 578, 592 Zuckerman, H.S., 676
Subject Index

A Almost completely decomposable groups, 534

Abelian group, 1 Almost disjoint subsets, 29, 30
Absolute Almost free groups, 112, 116
direct summand, 53 Annihilator ideals in End, 617
E-ring, 697 A-ring, 697
ideals, 680, 681 Arnold duality, 473
Absolutely Arnold-Lady category equivalence, 476
indecomposable group, 437 Arnold-Vinsonhaler invariants, 411
pure group, 159 Artinian endomorphism rings, 638
separative group, 379 Artinian ring, 684
solid group, 561 Associative law, 1
Additive functor, 33 Autoduality, 205
Additive group, 673-676 Automorphism, 7
of artinian ring, 684 Automorphism group (Aut), 7, 655–659
of injective modules, 185 of p-groups, 661–663
of noetherian ring, 682 of torsion-free groups, 665–670
of rings generated by idempotents, 126 Axiom
of von Neumann regular ring, 689 of choice, 20
Adjoint functors, 34, 236 of constructibility .V D L/, 20
Adjusted cotorsion group, 287 of first countability, 70
A-group, 385 Axiom-3 family, 31
Aleph, 21
@n -cyclic group, 117, 118
@n -free group, 116, 119 B
@1 -algebraically compact group, 188 Back and forth argument, 22
@1 -coseparable group, 506, 507 Baer cotorsion-pairs, 582
@1 -free group, 113, 119 Baer group, 579–582
@1 -separable group, 316, 329, 506 Baer invariants, 411, 426
Algebraically closed field, 703 Baer-Kaplansky theorem, 624
Algebraically compact Baer ring, 634
(endomorphism) rings, 209, 634, 635 Baer’s criterion for injectivity, 135
factor groups, 186, 187 Baer’s lemma, 426
groups, 183–187, 191, 195, 198, 210, 279 Baer-Specker group, 113, 115,
homomorphism groups, 220 501
Algebraic entropy, 621 Baer sum of extensions, 260

© Springer International Publishing Switzerland 2015 737

L. Fuchs, Abelian Groups, Springer Monographs in Mathematics,
DOI 10.1007/978-3-319-19422-6
738 Subject Index

Balanced- Cellular cover, 228

exact sequence, 366, 367, 417, 591 Center
injective groups, 370, 429 of automorphism group, 661
projective dimension, 385, 424, 430, 533, of endomorphism ring, 625
611 of purity, 153
projective groups, 369, 423, 610 Centralizer, 662
projective resolution, 369, 383, 423, 424, Character group (Char), 203, 221–223, 268
433 Characteristic ((*)), 410
Balanced subgroup, 366–369, 417–421, 419, subgroup, 7, 307, 657, 660
591 Circle group (T), 73, 141, 220
Base of open neighborhoods, 36 C -groups, 390
Basic subgroup, 169, 172 Class, 20
Basis, 75, 83 Closed subset of ordinals, 21
Bext functor, 421 Closed subsocle, 300
Bext2 , 422, 565 Coarser topology, 36
Bican’s theorem, 532 Cobalanced subgroup, 422
Bifunctor, 33 Coboundary, 256
Bilinear function, 229 Cocylic
Blocked subset, 28 group, 15, 48, 156, 164, 183
Blowing up lemma, 435 summand, 156
B1 -group, 546, 548–550, 563 Codiagonal map (r/, 48
B2 -group, 546, 548, 550, 563, 567 Codomain of map, 6, 31
B.n/ -groups, 535, 536 Cofinality, 21, 24
Boolean power, 49 Cogenerator
Bounded group, 32, 96 of category Ab, 141, 142
Bounded pure subgroups, 156 of group, 15, 145
Box topology, 70 Cokernel of map (Coker '), 6
Bracket groups, 536 Colimit, 57
Butler groups Column-convergent matrix, 619
of countable rank, 546–548 Commutative
of finite rank, 529–532, 544 diagram, 8
of uncountable rank, 563–568 law, 1
Butler’s theorem, 530 Compact
endomorphism rings, 619
groups, 36, 183, 203, 221, 497
C Compatibility of subgroups, 379
Cancellable map, 7 Complementary summand, 44, 51
Cancellation property, 351, 443, 468, 469 Complete
Canonical groups, 69–72, 190–192, 194
homomorphism, 8 set of invariants, 84
maps, 57, 60 set of representatives, 3
Cardinal, 21 topology, 69
Cartesian product .˘ /, 32, 47 torsion-free groups, 289
Category, 31 Completely decomposable groups, 423,
equivalence for cotorsion groups, 289 425–429
of abelian groups (Ab), 32 Completely independent subset, 514
of p-valuated groups .Vp /, 585 Completion, 71–73, 191, 192
of valuated vector spaces .V /, 335 Connecting homomorphism, 12, 56, 60, 240,
WALK, 605 263
WARF, 606 Consistent system of equations, 143, 144
Cauchy Constructible Universe (L), 23
neat net, 69 Continuous
net, 69 chain, 26
sequence, 69, 320 filtration, 22
Subject Index 739

homomorphism, 37, 91 of infinite rank torsion-free groups,

well-ordered ascending chain, 26 481–488
Continuum Hypothesis (CH), 21 of p-local simply presented mixed groups,
Contravariant functor, 32 607
Convergence, 69, 619 of Procházka-Murley groups, 463
Coordinate in direct sum, product, 43, 46 of reduced cotorsion groups, 286, 288
Corank, 410 of separable p-groups, 328–332
Corner’s theorem, 629 of ˙-cyclic groups, 97
Coseparable group, 506, 508 of torsion-complete p-groups, 318
Coset, 3 Direct
valuation, 589 limit .lim/, 57
Cotorsion 
!
limit of exact sequences, 59, 160
direct sum, 286 product, 47, 67
group, 282–286, 289 product of subsets, 87
hull, 295 sum, 43
pair, 296 sums of cyclic groups, 94
Cotorsion-free, 500, 651 sums of countable p-groups, 374, 375
Cotype, 413 sums of torsion-complete p-groups, 340
Countable antichain condition, 25 system, 56
Countably additive measure, 23 Direct summand, 44, 50
Covariant functor, 32 of completely decomposable group, 427
Crawley group, 405–407 of separable torsion-free group, 502
Crawley-Hales theorem, 358 of totally projective p-groups, 372
Critical type, 531 of simply presented mixed groups,
Cub, 21, 22 597
Cyclic Directed set, 24, 56
group, 15, 45 Discrete
subset, 87 norm, 123
topology, 36
Disjoint subgroups, 2, 43
D Divisibility of elements, 131
Decent subgroup, 537, 540 Divisible
Decomposition groups, 132–136, 140
basis, 600, 604 hull, 136, 140
of torsion groups, 45 torsion groups, 289
subgroup, 600 Domain of map, 6, 31
Defining relations, 79 Dual groups, 472, 474, 475, 505, 508
Dense subsocle, 300, 339 Duality functor, 472
Dependence relation, 92
Derived functor lim1 , 64, 285
Diagonal map ./, 48 E
Diagram, 8 Eda’s theorem, 494
chasing, 11 E-dual, 628
Diamond Principle (}/, 23, 24 E-group, 693
Direct decompositions, 44 Eilenberg map, 64
into summands with local endomorphism Eklof’s lemmas, 270–273
rings, 210 Eklof-Shelah criterion of freeness, 107
of complete groups, 194 Elementary
of direct products, 332 balanced-projective, 611
of divisible groups, 140 divisor, 83
of finite, finitely generated groups, 81, group, 4, 16, 46
84 vector group, 509–512
of finite rank torsion-free groups, 438–448, Elongation, 375, 403
467 E-map, 643
740 Subject Index

Embedding for ˝ and Tor, 233, 240

in algebraically compact group, 186 Exchange property, 206
in artinian ring with 1, 687 for algebraically compact groups, 210
in cotorsion group, 285 for (quasi-)injective groups, 207
in divisible (injective) group, 133 for torsion-complete p-groups, 319
in pure-injective group, 200 of indecomposable groups, 207
in pure subgroup, 151 Exchange ring, 208
in regular ring with 1, 691 Extension
Endo- lemma, 368
artinian ring, 645 of groups, 255–260
finitely-generated group, 645, 649 of maps, 7
flat group, 648, 649 External direct sum, 46
injective group, 648 Ext functor, 260
injective hull, 649 Extractable type, 414
noetherian group, 645, 649
projective dimension, 646, 649
projective group, 646, 649 F
quasi-injective, 649 Factor
quasi-projective, 647, 649 group, 3
slender, 649 set, 256
uniserial, 649 Factorization of finite groups, 86–91
Endomorphism, 7, 44 Faithful simple presentation, 355
group (End), 213, 614 Field of p-adic numbers, 224, 704
ring (End), 7, 614 Filter, 22
Endomorphism ring Filtered direct product, 49
generated by units, 628 Filtration, 22
of divisible groups, 635 Final rank (fin rk), 176
of p-groups, 622–627, 652 Finer topology, 36
of special groups, 634–636 Finite
of torsion-free groups, 629–632, 651 automorphism groups, 665
Epimorphism, 6 groups, 80, 81, 86
Equational class, 99 index topology, 37
Equivalent rank torsion-free groups, 413, 431–434
categories, 34 topology of End, 617–620, 622, 651
characteristics, 411, 586 Finitely
decomposition bases, 601 Butler group, 540, 546
extensions, 257, 258 cogenerated group, 145
height-matrices, 587 generated group, 81, 82, 84
presentations, 103, 104 First axiom of countability, 36
systems of equations, 144 Fitting’s lemma, 461
Erdős cardinal, 437 5-lemma, 13
E-ring, 693–697 Fomin duality, 474
Essential subgroup, 5 Frattini subgroup, 19
Essentially Free
finitely indecomposable group, 330 filter, 71
indecomposable group, 329, 333 (abelian) group, 75
Exact functor (left, or right), 33, 34 resolution, 77
Exact sequence, 8, 33 set of generators, 76
for Hom and Bext, 421 valuated group, 585
for Hom and Ext, 217, 263 valuated vector space, 336
for Hom and PBext, 538 ‘Free’ group, 121
for Hom and Pext, 278 Fried ideal, 680
for Homs , 227 Fuchs-5 group, 98
for H, 591, 608 Fuchs-44 group, 68
Subject Index 741

Full rational group (Q/, 17 High subgroup, 20, 135, 153, 365, 388
Full subcategory, 32 Hill invariants (f .A, G//, 344
Fully Hill’s theorem on freeness, 109
invariant subgroup, 7, 51, 137, 307 Hill-Walker theorem, 358
rigid system, 435 H()-family, 26, 27
transitive groups, 302, 306, 307, 376, 589, from a chain, 28
595 Homogeneous
Functor, 32 B1 -group, B2 -group, 549, 564
Functorial completely decomposable groups, 426
isomorphism, 34 indecomposable groups, 433
subgroup, 35 separable torsion-free groups, 503, 504
topology, 38 systems of equations, 143
torsion-free groups, 411, 414, 509, 632
valuated vector spaces, 334, 336
G Hom-Ext exact sequence, 233
Gap, 300, 307 Homomorphism (!/, 6
condition, 300, 307 groups, 213–217
Generalized Continuum Hypothesis (GCH), groups with distinguished subgroups, 219
21 of direct products, 66–68
Generalized Prüfer groups (H /, 304, 357, 363, of direct systems, 58
369, 370 of inverse systems, 62
Generating system, 2, 79 Homomorphism over a subgroup, 6
Generator, 2, 79 Hopfian group, 143
of category, 76 H-projective groups, 608
Genus, 466
G()-family, 26, 27
I
Global valuation, 586
Ideals in endomorphism rings, 620, 627
Gödel’s Axiom of Constructibility (V = L), 23
Idempotent
Griffith’s theorem, 580
charateristic, 412
Group
endomorphisms, 44, 50, 52, 614, 615,
of balanced extensions (Bext), 421
623
of extensions (Ext), 257
type, 412, 509
of multiplications (Mult), 677
Identity
of p-adic integers, 18, 224, 431, 625
functor, 32
of prebalanced extensions (PBext), 538
map, morphism, 7, 9, 32
of pure extensions (Pext), 276
Image of map (Im ), 6
of type p1 , 16
Inaccessible cardinal, 23
rings, 700
Indecomposable groups, 44, 156, 196, 207,
with discrete norm, 124
431–436, 569
Group-valued measure, 67
Independent
set, 91
system of invariants, 84
H Index of subgroup, 3
Hajós’ theorem, 88 Indicator (u), 300, 301
Harrison category equivalence, 289 Indiscrete topology, 36
Hausdorff topology, 36, 37 Induced topology, 37
Hawaiian group, 567 Inductive
Height, 4, 300 set, 20
-matrix, 586, 593 topology, 319
-preserving isomorphism, 345 Inessential
-sequence, 410 endomorphism, 616
valuation, 337 homomorphism, 227
H-exact sequence, 591, 608 Injection map, 7, 47
742 Subject Index

Injective -filtration, 26
group, 134, 135 -free group, 112
hull, 136 -indecomposable group, 331
limit, 57 -product, 49
map, 6, 47 -pure subgroup, 153
Inner type (IT), 413 -separable group, 316, 506
Internal direct sum, 43 -separativity, 384
Invariants of -Shelah game, 122
algebraically compact groups, 195 Keef class, 341
compact groups, 222 Kernel-cokernel sequence, 12
completely decomposable groups, 426 Kernel
countable p-groups, 346 of a map (Ker '), 6
direct sums of countable p-groups, 375 of subdirect sum, 49
divisible groups, 140 subgroup, 617
finitely cogenerated groups, 146 Knice subgroup, 508, 604
finitely generated groups, 84 Kolettis’ theorem, 350, 375
free groups, 76 K-product, 49
mixed groups of torsion-free rank 1, 594 K-representation, 534
p-local Warfield groups, 603 Krull-Schmidt property, 211, 462
quasi-injective groups, 138 Kulikov’s theorems, 94, 97
quasi-projective groups, 128 Kurepa hypothesis, 391
˙-cyclic groups, 97
simply presented p-groups, 358
Tor, 246 L
torsion-complete p-groups, 312 Lady’s theorem, 448
Inverse, 1 -basic subgroup, 390
limit .lim/, 60 -indecomposable group, 331
 Large subgroup, 308
system, 60
Involution, 658 topology, 319
Irreducible torsion-free group, 477 Lattice of subgroups, 3
Isometry, 335, 585 Length
Isomorphic arbitrarily large, 303
automorphism groups, 663 of chain, 26
direct decompositions, 45, 97 of p-group (`(*)), 299
endomorphism rings, 624 Liebert’s theorem, 625
objects, 32 Limit, 69
Isomorphism (Š/, 6, 32 cardinal, ordinal, 21
theorems, 8 Linear
Isotype subgroup, 365, 591 combination, 2, 75
in totally projective group, 381, 382 independence, 91
topology, 36, 69
Linearly compact groups, 197, 224
J Local
Jacobson radical, 454, 625 endomorphism ring, 207, 210, 641
Jónsson group, 142 groups, 40, 251, 454, 457, 585,
Jordan-Zassenhaus lemma, 448 607
Warfield groups, 602, 603
Localization, 17, 251, 252, 532
K map, 251
Kaplansky duality, 224 Locally
Kaplansky-Mackey lemma, 345 compact extension, 203
Kaplansky’s test problems, 85 compact groups, 203–206
-complete filter, 23 cyclic group, 17
-cyclic group, 117, 249 free group, 472
Subject Index 743

Łoś-Eda theorem, 492–494 Nice

Lower basic subgroup, 176 composition chain, 362
subgroup, 352, 590
system, 362–364
M Nil group, 677
Map (!/, 6, 31 Nilpotent radical, 447
Martin’s Axiom (MA), 24 Noetherian
Matrix representation of End .H D jjaik jj/, endomorphism ring, 640
619 rings, 682
Maximal Non-splitting mixed groups, 574
divisible subgroup, 133 Norm, 123
independent system, 92 n-substitution property, 451
(maximum) element, 20 Nunke groups (N /, 574
Maximum condition on subgroups, 82
Measurable
cardinal, 23, 67 O
vector group, 515 Object in category, 31
Measure, 23 !-elongation, 404
Metelli classification of Butler groups, Order
535 of element, 2
Metrizable topology, 36 of group, 1
Minimum condition on subgroups, 146, 197, Ordinal, 21
198 Orthogonal idempotents, 44, 615
Mittag-Leffler group, 117, 118 Outer type (OT), 413
Mixed groups, 3
of torsion-free rank 1, 593
Model of set theory (V), 23 P
Modular law, 3 p-adic
Module, 39 algebraically compact group, 195
Monic map, monomorphism, 6 completion, 196
Monotone subgroups, 491, 495 component, 196
Morphism, 31 integers, 18, 224, 431, 625
in valuated groups, 585 modules, 40, 224, 175, 455
in valuated vector spaces, 334 number field, 224, 704
Multiplications on a group (Mult), 677 topology, 37
Multiplicative group of field, 701–705 Partially ordered set, 20
p-basic subgroups, 166–171, 174, 242
p-basis, 166, 167
N PBext functor, 538
Nakayama’s lemma, 446 p-component, 46
Naring (non-associative ring), 673 p-corank (rkp (A)), 94
Natural p-divisible group, 132
equivalence, 34 Periodic subset, 87
homomorphism, 8 Pext, 276–281
isomorphism, 34 p-group, 3
morphism, 34 p-height, 4, 300, 409
transformation, 34 PID endomorphism rings, 639
Near-isomorphism ./, 466 Pierce condition, 309
Neat Pierce radical, 625
Cauchy net, 69 p-independent system, 166
subgroup, 153 -regular, 637, 690, 692
Neatly convergent sequence, 69 p -topology, 73, 396
Net, 69 p-local groups, 40, 251, 454, 589
744 Subject Index

p-nice subgroup, 590 p -pure subgroup, 386

Pontryagin p!C1 -projective group, 399
dual, 203 p!Cn -injective group, 400
theorem, 106 p!Cn -projective group, 397
Poset, 20 Pull-back diagram, 54
Power Pure subgroup, 4, 149–152, 155–158
cancellation, 454 generated by, 151
set, 22 Pure-
substitution, 453 complete p-group, 322
Powers of Z, 514–518 essential extension, 199
p-pure-exact sequence, 159 essential subgroup, 199, 201
p-pure subgroup, 149 exact sequence, 159, 161–163, 242, 278
p-rank (rkp .A//, 93 extensions, 277
Prebalanced-exact sequence, 538 filtration, 26, 152
Prebalanced subgroup, 537 independence, 172
Prescribed endomorphism rings, 629, 651 Pure-injective, 164, 165, 184
Presentation, 79 hull, 200, 201
with stacked basis, 99–102 resolution, 164
Primary Purely indecomposable group, 436
component, 46 Pure-projective, 163, 164
group, 3 resolution, 163
Primitive idempotent, 614 Pure-simple group, 158
Primordinal groups, 667 Purifiable subgroup, 152
Principal filter, 23 Push-out diagram, 55
Procházka-Murley groups, 462–464 p-valuation, 584
Product
of characteristics, 412
of maps, 7 Q
of morphisms, 31 Quasi-
of types, 412 basis in p-groups, 171
topology, 37, 70 complete p-group, 323–325
Projection, 44, 47 cyclic p-group, 16
invariant subgroup, 53 direct decomposition, 460
Projective endomorphism, 460
cover, 78 endomorphism ring (QEnd), 631
group, 78 equivalent height-matrices, 603
limit, 60 homomorphism, 458
resolution, 78 indecomposable group, 330
Proper injective group, 137, 138, 207
subgroup, 2 isomorphism (/, 458
with respect to a subgroup, 343, 417, 590 nil group, 677
Prüfer projective group, 127, 128
group, 167, 175, 356 pure-injectivity, -projectivity, 165
theorems, 96 splitting mixed group, 576, 577
topology, 37, 72, 147 summand, 460
pseudo-socle (in torsion-free groups), 478 Quotient-divisible groups, 473, 475, 578
p -balanced subgroup, 387
p -high subgroup, 365, 388
p -injective group, 395 R
p -isotype subgroup, 386 Radical functor, 241
p -nice subgroup, 387, 585 Range of map, 6, 31
p -projective Rank (rk(A//, 92
group, 371, 392 distribution in direct decompositions, 444
resolution, 393 of a free group, 76
Subject Index 745

Rank 1 torsion-free groups, 411, 412 Singular

Rational groups, 17, 411 cardinal, ordinal, 21
Real-closed field, 703 compactness theorem, 120
Reduced Skeleton (of a category), 34
group, 133, 136 Slender groups, 489–494, 496–500
product, 49 Small
Refinement of direct decomposition, 45 group, 218
Reflexive groups, 471, 495 homomorphism, 225
Regular Smooth chain, 20
cardinal, ordinal, 21 of B1 -groups, 563
endomorphism ring, 641 of free groups, 106–109, 115
ring (von Neumann), 689 of nice subgroups, 362
subgroup, 422 of separative subgroups, 378
Regulating subgroup, 534 of ˙-cyclic groups, 157
Regulator, 535 of slender groups, 498
Reid class, 506 of solid subgroups, 557
Relative of totally projective p-groups, 382
balanced-projective resolution, 383, 425 Smooth filtration, 22
UK-invariants, 344 Snake lemma, 12
Representation by posets, 534 Socle, 4
Representative of coset, 3 Solid
Restriction of a map, 7 chain, 557–561
R-homomorphism, 40 subgroup, 552–556
Richman Solvability of systems of equations, 144, 158
duality, 535 Specker group, 126
type, 414 Split extension, 622, 627, 652
Rigid system, 432, 437 Splitter, 293
of B2 -groups, 569 Splitting
Ring exact sequence, 45, 161
of p-adic integers (Jp /, 18 extension, 256
on (supported by) a group, 677 field, 456, 457
R-module, 39 map, 45
mixed group, 575, 576
Stabilizer, 657
S Stable range, 450–453
Self-cancellation, 469 Stacked basis, 83, 99, 102
Self-injective endomorphism ring, 636, 642, Standard basic subgroup, 169
649 Starred p-group, 179, 248
Self-small group, 218 Stationary set, 21
Semi-local endomorphism ring, 450, 642 Strongly
Semi-rigid system, 436 indecomposable group, 460
Separable -free, 114, 120
in the sense of Hill, 378 Subcategory, 32
mixed group, 584 Subdirectly irreducible group, 50
torsion-free groups, 501-505 Subdirect product, 48
torsion group, 299, 301, 311, 333 Subfunctor of the identity, 35
Separative cancellation, 469 Subgroup, 2
Separative chain, subgroup, 378, 556 generated by .hi/, 2
Set, 20 of cyclic group, 15
S-group, 385 of free groups, 77
Shelah’s Compactness Theorem, 120 of ˙-cyclic groups, 97
˙-cyclic groups, 94–97, 102, 157, 164, 277 of totally projective p-groups, 377,
Simple endomorphism ring, 638 381–383, 385
Simply presented groups, 355, 596, 607 Subordinate decomposition basis, 601
746 Subject Index

Subsocle, 300 Totally projective p-groups, 371, 375

Substitution property, 449 Transfinite
Successor ordinal, 21 chains, 25
Summable final rank, 403
p-group, 401 height, 300
subset, 514 Transformation set, 256
Summand, 44, 50 Transitive group, 302
intersection property, 111 Transversal, 255
property of bounded pure subgroups, 155, Trivial subgroup, 2
156 Type (t), 411
property of injective groups, 135, 136 Typeset, 413
Superdecomposable
Butler groups, 570
groups, 485 U
Superfluous subgroup, 2 UK-invariants, 344, 351
Supplement subgroup, 54 of countable p-groups, 346
Support, 47 of direct sums of countable p-groups, 350
of a subgroup in a p-group, 300 of p-local Warfield groups, 603
of a valuated vector space, 334 of simply presented p-groups, 358, 360
Surjective map, 6 of Tor, 246
System of equations, 143, 184 of valuated vector spaces, 336
Szele’s theorem, 177 UK-matrix, 349
Ulm
factors (A /, 4
T length, 4
 -admissible function, 359 sequence, 305
Tensor map, 231 subgroups (A /, 4
Tensor product (˝/, 230 Ulm factors, 305
of torsion groups, 243 of cotorsion groups, 291
Tensor-torsion exact sequence, 240 of countable p-groups, 346, 349
TEP-subgroup, 541–544 of generalized Prüfer groups, 304
Test problems, 85 of simply presented p-groups, 357
Thick groups, 327 of totally injective p-groups, 396
Thin groups, 325, 326 Ulm-Kaplansky invariants (f .A//, 344
3  3-lemma, 13 Ulm’s theorem, 346
Topological group, 36 Ultrafilter, 23
Tor functor, 237 Ultraproduct, 49
Torsion-complete p-groups, 311–316 Unbounded set, 21
Torsion-completion, 312, 321 Undecidable problems, 269, 342, 391, 407,
Torsion 495, 507, 508, 527, 567, 582
extension property, 541 Unimodular set, 450
group, 3, 4 Unit group, 697, 700
part, 4 Universal for a set of groups, 582, 589
product (Tor), 237 Universal property
subgroup (tA), 3 of free groups, 76
theory, 241 of localization, 251
Torsion-free of tensor product, 230
group, 3, 409 Upper basic subgroup, 175
cover, 296 u-topology, 36
rank .rk0 .A//, 93
Torsionless group, 471
Torsion-splitting sequence, 294, 295, 386 V
Torsion subgroup (t.A/ = tA), 4 Valuated
Totally injective p-groups, 395 group, 584
Subject Index 747

vector space, 334 invariant, 602

Valuation, 334 Weakly compact cardinal, 23, 24
Variety, 99 Whitehead group (W-group), 519–527
Vector, 46
group, 509–512
Vinsonhaler-Wickless duality, 474 Z
Z-adic
completion, 73, 192
W topology, 37, 190, 191
WALK category, 605 Zermelo-Fraenkel axioms with AC, 20
Walker groups (Pˇ /, 356 Zero-ring on a group, 677
WARF category, 606 ZFC axioms, 20
Warfield Zippin property, 404
duality, 473, 475 Zippin’s theorem, 347
groups, 599–604 Zorn’s lemma, 20