Springer Monographs in Mathematics

Gunter Malle
B. Heinrich Matzat

Inverse
Galois
Theory
Second Edition
Springer Monographs in Mathematics

Editors-in-Chief
Isabelle Gallagher, Paris, France
Minhyong Kim, Oxford, UK

Series Editors
Sheldon Axler, San Francisco, USA
Mark Braverman, Toronto, Canada
Maria Chudnovsky, Princeton, USA
Sinan C. Güntürk, New York, USA
Claude Le Bris, Marne la Vallée, France
Pascal Massart, Orsay, France
Alberto Pinto, Porto, Portugal
Gabriella Pinzari, Napoli, Italy
Ken Ribet, Berkeley, USA
René Schilling, Dresden, Germany
Panagiotis Souganidis, Chicago, USA
Endre Süli, Oxford, UK
Shmuel Weinberger, Chicago, USA
Boris Zilber, Oxford, UK
This series publishes advanced monographs giving well-written presentations of the
“state-of-the-art” in fields of mathematical research that have acquired the maturity
needed for such a treatment. They are sufficiently self-contained to be accessible to
more than just the intimate specialists of the subject, and sufficiently comprehensive
to remain valuable references for many years. Besides the current state of knowledge
in its field, an SMM volume should ideally describe its relevance to and interaction
with neighbouring fields of mathematics, and give pointers to future directions of
research

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

Gunter Malle  B. Heinrich Matzat

Inverse Galois Theory

Second Edition

123
Gunter Malle B. Heinrich Matzat
FB Mathematik Interdisziplinäres Zentrum für
TU Kaiserslautern Wissenschaftliches Rechnen
Kaiserslautern, Germany Universität Heidelberg
Heidelberg, Germany

ISSN 1439-7382 ISSN 2196-9922 (electronic)

Springer Monographs in Mathematics
ISBN 978-3-662-55419-7 ISBN 978-3-662-55420-3 (eBook)
https://doi.org/10.1007/978-3-662-55420-3

Library of Congress Control Number: 2017949891

Mathematics Subject Classification (2010): 12F12, 12-XX, 20-XX

© Springer-Verlag GmbH Germany, part of Springer Nature 1999, 2018

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. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by the registered company Springer-Verlag GmbH, DE part of
Springer Nature.
The registered company address is: Heidelberger Platz 3, 14197 Berlin, Germany
Preface

Inverse Galois Theory is concerned with the question which finite groups occur as
Galois groups over a given field K. In particular this includes the question on the
structure and the representations of the absolute Galois group of K and also the
question about its finite epimorphic images, the so-called inverse problem of Galois
theory. In all these areas important progress was made in the last few years, about
which we want to report here.
The first systematic approach to the solution of the inverse problem over the field
of rational numbers Q goes back to Hilbert (1892). Using the irreducibility theorem
which he proved for this purpose, he could show that over Q and more generally
over every field finitely generated over Q there exist infinitely many Galois exten-
sions with the symmetric and the alternating groups Sn and An . E. Noether (1918)
then stated that the inverse problem for a finite group can be solved with the Hilbert
irreducibility theorem if the field of fractions of the ring of invariants of a permu-
tation representation of the group is rational, and that in this case all polynomials
with this Galois group can be parametrized. She could verify this condition for per-
mutation groups of small degree, and her student Seidelmann (1918) computed
the corresponding parametric polynomials up to degree 4. This line of research was
continued for special solvable groups by Breuer, Furtwängler and Gröbner. Unfortu-
nately, not all fields of invariants of finite groups are rational. A first counter example
was given by Swan (1969) for the field Q and by Saltman (1984) for C.
The next big step was initiated by Scholz (1937) and Reichardt (1937). By solv-
ing sufficiently many number theoretic embedding problems they could show that
all finite p-groups for odd primes p occur as Galois groups over Q. This approach
culminated in the celebrated solution of the inverse problem of Galois theory for all
solvable groups over arbitrary number fields by Šafarevič (1954d, 1989).
The next set of results was furnished by the works of Shih (1974), Fried (1977),
Belyi (1979), Matzat (1979, 1984) and Thompson (1984a), in which the concept
of rigidity was introduced and investigated. This allows to translate sufficient con-
ditions for the rationality of covers of compact Riemann surfaces with given mon-
odromy group into group theoretic criteria. Thus it is possible with purely group

v
vi Preface

theoretic considerations to prove the existence of Galois extensions with prescribed

Galois group over Q.t/ and, using the Hilbert irreducibility theorem, also over Q.
This approach has proved to be particularly effective in the case of simple and almost
simple groups, where the most far reaching results were obtained by Belyi (1979,
1983), Malle (1988b, 1992, 1996) and Reiter (1999).
Another direction of research consists in finding classes of fields for which the
inverse problem of Galois theory can be solved completely. By the Riemann exis-
tence theorem and the specialization theorem of Grothendieck for the fundamen-
tal group this is true for example for algebraic function fields in one variable over
algebraically closed fields of characteristic zero; these fields also form the start-
ing point for the rigidity method. The corresponding result in positive characteristic
was proved by Harbater (1984). According to a very recent result of Pop (1996)
it remains correct if the algebraically closed field of constants is replaced by a suf-
ficiently large field, above which every smooth curve with rational points possesses
infinitely many rational points. This result generalizes in particular the case of a
PAC (pseudo algebraically closed) field of constants treated earlier by Fried and
Völklein (1991). As an application of the latter result Fried and Völklein (1992)
and Pop (1996) could show that the absolute Galois group of a countable Hilbertian
PAC field is a free profinite group. The same result for function fields in one vari-
able over arbitrary algebraically closed fields of constants has now been proved by
Harbater (1995a) and Pop (1995); in characteristic zero this already goes back to
Douady (1964).
The aim of this book is to give a consistent and reasonably complete survey of
the results obtained in this area, with main emphasis on the rigidity method and
its applications. In order to keep the size reasonable we usually have omitted those
proofs which are worked out in other books (with the exceptions of the two intro-
ductory lecture notes of Matzat (1987) and Serre (1992)). On the other hand we
have substantially rewritten a number of results or even given new proofs.
Since the individual chapters all start with their own overview, we will only char-
acterize them briefly. In Chapter I we explain the rigidity method for coverings of
the projective line in characteristic zero. We prove the fundamental rationality crite-
ria including the translation technique. As applications we treat the abelian groups,
the symmetric and alternating groups Sn and An , the 2-dimensional linear groups
L2 .p/ and PGL2 .p/ as well as the small Mathieu groups M11 and M12 . This covers
the examples of Hilbert (1892) and Shih (1974). At the end we present the explicit
calculation of polynomials for some of these groups and give results on the special-
ization of parameters. This chapter is essentially elementary with complete proofs.
In Chapter II we give a survey on the results obtained by application of the 1-
dimensional rigidity method. This includes the almost complete realization of the
finite simple groups as Galois groups over Qab .t/ and Qab in Paragraphs 1 to 5. Here
Qab denotes the maximal abelian extension field of Q. In the following Paragraphs 6
to 9 we give a survey of realizations of simple groups over Q.t/ and over Q. Here
the proofs for the linear groups and the sporadic groups are elementary, while for
the remaining groups, some knowledge of the Deligne-Lusztig theory of characters
of reductive groups is necessary.
Preface vii

In the third chapter we develop the rigidity method for coverings of projective
spaces. This covers the approach of Fried (1977). It leads to partial generalizations
of the results in Chapter I. Here, though, the concept of rigidity of a braid orbit is not
in general sufficient to deduce existence theorems for Galois extensions over Q.t/
and Q. Usually, further arithmetic conditions like the existence of rational points
have to be satisfied. This makes the application considerably harder, as can be seen
at the example of the linear and unitary groups treated by Völklein (1993). The
second part of this chapter also contains the proof of Fried and Völklein (1991) that
over function fields k.t/ with a PAC-field of constants k of characteristic zero every
finite group occurs as a Galois group. This solves in particular the inverse problem
of Galois theory over Hilbertian PAC-fields of characteristic zero.
Chapter IV considers the question of constructing Galois extensions with com-
posite groups from given Galois extensions with simple Galois groups. This leads
to embedding problems for arithmetic function fields. In the first part of the chap-
ter we prove simple reduction theorems, and study the range of applicability of the
two elementary basic constructions for embedding problems with abelian and with
center free kernel. In continuation of the results of Chapter III the developed meth-
ods can be used to prove the result of Fried and Völklein that the absolute Galois
group of a countable Hilbertian PAC-field of characteristic zero is profinite free. In
the second part we study cohomological descriptions of the embedding obstruction.
Explicit computation of these obstructions for example gives realizations of the cen-
tral extensions of the symmetric and alternating groups as Galois groups over Q.t/.
This is followed by the investigation of concordant embedding problems in Para-
graph 8 and the remaining Hasse obstruction in Paragraph 9. Finally in Paragraph 10
we prove the Theorem of Scholz (1937) and Reichardt (1937) on the realizability
of nilpotent groups as Galois groups, more generally over global fields.
In the final Chapter V the methods of ultrametric analysis are used to prove the
results of Harbater (1984, 1995a) and Pop (1995, 1996) concerning the inverse
problem and more generally the structure of the absolute Galois group for function
fields over complete ultrametric fields of constants and, as an application, over alge-
braically closed fields in positive characteristic. Further, we introduce the notion of
large fields and prove the result of Pop (1996) that the absolute Galois group of
a countable Hilbertian large field is free profinite. The chapter ends with a short
report on the proof by Raynaud (1994) with the extension by Harbater (1994a) of
the conjecture of Abhyankar, which characterizes the possible Galois groups over
function fields in one variable over algebraically closed fields of constants of posi-
tive characteristic with restricted ramification. Unfortunately parts of the proof itself
go beyond the scope of this book.
The Appendix contains tables of polynomials with (regular) Galois groups of
small permutation rank over Q and Q.t/.
Finally we mention some topics not covered by this book. First, this concerns the
geometric version of the rigidity method for several variables with the construction
of Hurwitz schemes following Fried (1977) and Fried and Völklein (1991), which
would have led too far away from the field theoretic approach considered here. This
aspect is covered in the recent monograph of Völklein (1996). Further, we do not
viii Preface

treat the question of the present state in the Noether problem and more generally
the question of generic polynomials. Last but not least this concerns the description
of the action of the absolute Galois group or the Grothendieck-Teichmüller group
respectively on algebraic fundamental groups, considered by Ihara (1991) et al. (see
for example the survey by Nakamura (1997)). This has at present not yet led to new
Galois realizations of finite groups.
We want to thank all those which by their cooperation, their criticism and
their corrections as well as proposals for improvements have contributed to the
present form of the book. In particular these are M. Geck, D. Harbater, M. Jar-
den, J. Oesterlé, M. v. d. Put, I. R. Šafarevič, L. Schneps, J.-P. Serre, J. Sonn,
T. Szamuely, H. Völklein, A. V. Yakovlev, and our colleagues and collaborators
R. Dentzer, M. Folkers, H. Geyer, F. Häfner, G. Hiß, G. Kemper, F.-V. Kuhlmann,
F. Lübeck, R. Nauheim, F. Pop, U. Porsch, B. Przywara, S. Reiter, and M. Saı̈di.

Preface to the Second Edition

Two important new developments have taken place since the appearance of the first
edition of this book. The first is the algebraization of the Katz algorithm for (lin-
early) rigid generating systems of finite groups. The second is the emergence of
a modular Galois theory. The latter has led to new construction methods for addi-
tive polynomials with given Galois group over fields of positive characteristic. Both
methods have their origin in the Galois theory of differential and difference equa-
tions.
The algebraic version of the Katz algorithm goes back to Dettweiler and Reiter
(2000). It will be presented here with complete proofs which thus replace the long
quite involved development in Katz (1996). The chosen presentation follows the
later exposition of Dettweiler (2003). This has the advantage of being better adapted
to the notation used throughout this book and of leading to somewhat easier formu-
las in some instances. As an application we will obtain numerous linear groups
defined over IFq , in particular for high prime powers q, as Galois groups of geo-
metric field extensions over Q.t/. This would not be achievable with the simple
rigidity method described in Chapter I. The new results are given at the end of Chap-
ter III in Paragraphs 9 and 10. The original Section 9.1 is retained in Paragraph 5 as
Section 5.5. The remaining parts of Paragraphs 9 and 10 from the first edition are
superimposed by the new results and have thus become superfluous.
Finite Galois extensions N=K in positive characteristic can be generated over
K by a finite dimensional IFq -vector space V , which is moreover stable under the
Frobenius operator. Thus such Galois extensions can also be interpreted as solution
fields of difference modules over K for the Frobenius operator—Frobenius mod-
ules for short—(see van der Put and Singer (1997)). In this way the Galois group
Preface ix

of N=K comes in a natural way equipped with a faithful matrix representation over
IFq , hence with a modular representation. In analogy to the theory of differential
modules the containment of the representing matrix of the Frobenius operator in
a linear algebraic subgroup of GL.V / usually leads to a non-trivial upper bound
for the Galois group. A corresponding result in characteristic zero is not available.
There also exists an algorithm for the computation of Galois groups of additive
polynomials, respectively of Frobenius modules. This can be sped up considerably
using a modular Dedekind criterion to prove lower bounds. In favorable cases—as
they occur in our applications—the upper bound and the lower bound obtained from
the modular Dedekind criterion will agree, so that the Galois group can be deter-
mined explicitly without having to use the general algorithm for the computation of
Galois groups. Moreover it is possible, starting from the representing matrix of the
Frobenius operator, to compute directly additive polynomials whose vector space
of zeroes generates the solution field of the Frobenius module. With this method
for many families of finite groups of Lie type we will construct generating additive
polynomials of a very general form with the corresponding Galois group.
Besides the existence of a non-trivial upper bound for the Galois group, modu-
lar Galois theory enjoys a further advantage, which has no parallel in characteristic
zero due to the missing matrix representation: the field restriction for linear alge-
braic groups makes it possible to rewrite a representation of a linear group over
IFq into a (larger dimensional) representation over IFp . Using this it becomes pos-
sible for many finite linear groups defined over IFq , for which there exist geometric
Galois extensions N=K with field of constants IFq , to construct geometric Galois
extensions NQ =KQ with field of constants IFp and with the same Galois group. Due to
these facts modular Galois theory has developped to a rather attractive, active and
independent research area in modular representation theory. The above results are
presented in the new Chapter V on Additive Polynomials.
We expect that these two methods will yield geometric Galois realizations of
many further groups apart from the ones constructed here. Nevertheless, for a com-
plete solution of the inverse problem of Galois theory most likely further new fun-
damental ideas will have to be conceived.
Apart from the incorporation of the new parts we have also made a number of
additions and corrections to the first edition, and have adapted our notation in some
parts. Further the tables of example polynomials in the Appendix have been updated
and extended. For helpful advice we are indebted in particular to Annette Bachmayr,
Michael Dettweiler, Jürgen Klüners, Peter Müller and Stefan Reiter.

Kaiserslautern, Germany G. Malle

Heidelberg, Germany B. H. Matzat
Contents

I The Rigidity Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1 The Inverse Galois Problem over C.t/ and IR.t/ . . . . . . . . . . . . . . . . . 2
1.1 The Fundamental Group of the Punctured Riemann
Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Algebraic Variant of the Fundamental Group . . . . . . . . . 3
1.3 Extension by Complex Conjugation . . . . . . . . . . . . . . . . . . . . . 7
1.4 Generalization to Function Fields of Riemann Surfaces . . . . 8
2 Arithmetic Fundamental Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1 Descent to Algebraically Closed Subfields . . . . . . . . . . . . . . . 10
2.2 The Fundamental Splitting Sequence . . . . . . . . . . . . . . . . . . . . 13
2.3 The Action via the Cyclotomic Character . . . . . . . . . . . . . . . . 14
2.4 The Theorem of Belyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Fields of Definition of Galois Extensions . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 Cyclic and Projective Descent . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Fields of Definition of Geometric Field Extensions . . . . . . . . 21
3.3 Fields of Definition of Geometric Galois Extensions . . . . . . 22
4 The Rigidity Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1 The Hurwitz Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2 The Fixed Field of a Class of Generating Systems . . . . . . . . . 28
4.3 The Basic Rigidity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4 Choice of Ramification Points . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 Verification of Rigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.1 Geometric Galois Extensions over Q.t/
with Abelian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Geometric Galois Extensions over Q.t/ with Sn and An . . . . 35
5.3 Structure Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4 The Rigidity Criterion of Belyi . . . . . . . . . . . . . . . . . . . . . . . . . 40
6 Geometric Automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.1 Extension of the Algebraic Fundamental Group . . . . . . . . . . . 43
6.2 The Action of Geometric Automorphisms . . . . . . . . . . . . . . . . 45

xi
xii Contents

6.3 Rigid Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

6.4 The Twisted Rigidity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.5 Geometric Galois Extensions over Q.t/ with M12
and M11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7 Rational Translates of Galois Extensions . . . . . . . . . . . . . . . . . . . . . . . 53
7.1 Galois Rational Translates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.2 Rational Translates with Few Ramification Points . . . . . . . . . 54
7.3 Twisting Rational Translates . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.4 Geometric Galois Extensions over Q.t/ with L2 .p/ . . . . . . . 60
8 Automorphisms of the Galois Group . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8.1 Fixed Fields of Coarse Classes of Generating Systems . . . . . 63
8.2 Extension of the Galois Group by Outer Automorphisms . . . 65
8.3 Geometric Extension of the Galois Group by Outer
Automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.4 Geometric Galois Extensions over Q.t/ with PGL2 .p/ . . . . . 68
9 Computation of Polynomials with Prescribed Group . . . . . . . . . . . . . 70
9.1 Decomposition of Prime Divisors in Galois Extensions . . . . . 70
9.2 Polynomials with Groups Sn and An . . . . . . . . . . . . . . . . . . . . 72
9.3 Polynomials with the Group Aut.A6 / and Related
Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.4 Polynomials with the Mathieu Groups M12 and M11 . . . . . . . 78
10 Specialization of Geometric Galois Extensions . . . . . . . . . . . . . . . . . . 82
10.1 Local Structure Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
10.2 Reality Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
10.3 Ramification in Minimal Fields of Definition . . . . . . . . . . . . . 86
10.4 Ramification in Residue Fields . . . . . . . . . . . . . . . . . . . . . . . . . 88

II Applications of Rigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
1 The General Linear Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
1.1 Groups of Lie Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
1.2 Rigidity for GLn .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
1.3 Galois Realizations for Linear Groups . . . . . . . . . . . . . . . . . . . 98
2 Pseudo-Reflection Groups and Belyi Triples . . . . . . . . . . . . . . . . . . . . 100
2.1 Groups Generated by Pseudo-Reflections . . . . . . . . . . . . . . . . 100
2.2 An Effective Version of Belyi’s Criterion . . . . . . . . . . . . . . . . 101
2.3 Imprimitive and Symmetric Groups . . . . . . . . . . . . . . . . . . . . . 103
2.4 Invariant Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3 The Classical Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.1 Rigidity for GUn .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.2 Rigidity for CSp2n .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.3 Rigidity for SO2nC1 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4 Rigidity for COC 2n .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.5 Rigidity for CO 2n .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Contents xiii

4 The Exceptional Groups of Rank at Most 2 . . . . . . . . . . . . . . . . . . . . . 117

4.1 Divisibility Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.2 Rigidity for the Ree Groups 2G2 .q 2 / . . . . . . . . . . . . . . . . . . . 118
4.3 Rigidity for the Groups G2 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.4 Rigidity for the Groups 3D4 .q/ . . . . . . . . . . . . . . . . . . . . . . . . 122
4.5 Rigidity for the Groups 2B2 .8/ and 2F4 .2/0 . . . . . . . . . . . . . . 124
5 The Exceptional Groups of Large Rank . . . . . . . . . . . . . . . . . . . . . . . . 126
5.1 Results From Deligne–Lusztig Theory . . . . . . . . . . . . . . . . . . 126
5.2 Rigidity for the Groups F4 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.3 Rigidity for the Groups E6 .q/ and 2E6 .q/ for odd q . . . . . . . 131
5.4 Rigidity for the Groups E6 .22mC1 / and 2E6 .22m / . . . . . . . . . 135
5.5 Rigidity for the Groups E7 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.6 The Groups E8 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6 Galois Realizations of Linear and Unitary Groups over Q . . . . . . . . . 140
6.1 Extension by the Graph Automorphism . . . . . . . . . . . . . . . . . . 140
6.2 GA-Realizations over Qab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.3 GA-Realizations over Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7 Galois Realizations of Symplectic and Orthogonal Groups
over Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.1 GA-Realizations of Symplectic Groups over Q . . . . . . . . . . . 146
7.2 GA-Realizations of Odd-Dimensional
Orthogonal Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.3 Even-Dimensional Split Orthogonal Groups . . . . . . . . . . . . . 150
7.4 Even-Dimensional Non-split Orthogonal Groups . . . . . . . . . 151
7.5 The 8-Dimensional Split Orthogonal Groups . . . . . . . . . . . . . 152
8 Galois Realizations of Exceptional Groups over Q . . . . . . . . . . . . . . . 155
8.1 GA-Realizations for the Groups G2 .p/ . . . . . . . . . . . . . . . . . . 155
8.2 The Groups F4 .p/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3 The Groups E6 .p/ and 2E6 .p/ . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 The Groups E8 .p/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
9 The Sporadic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.1 The Mathieu Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.2 The Leech Lattice Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.3 The Fischer Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
9.4 The Monster Centralizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.5 The Oddments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
9.6 Galois Realizations for the Sporadic Groups . . . . . . . . . . . . . . 172
10 Summary for Simple Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
10.1 Galois Realizations over Qab . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
10.2 Galois Realizations over Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

III Action of Braids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

1 Braid Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
1.1 The Artin Braid Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
1.2 The Hurwitz Braid Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
xiv Contents

1.3 The Pure Hurwitz Braid Group . . . . . . . . . . . . . . . . . . . . . . . . . 183

1.4 The Word Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
2 Profinite Braid Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.1 The Hurwitz Braid Group as Galois Group . . . . . . . . . . . . . . . 187
2.2 Inertia Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
2.3 Structure of the Profinite Hurwitz Braid Group . . . . . . . . . . . 192
2.4 The Fixed Field of the Free Normal Subgroup . . . . . . . . . . . . 193
3 Galois Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.1 An Arithmetic Fundamental Group . . . . . . . . . . . . . . . . . . . . . 196
3.2 Hurwitz Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
3.3 The Fixed Field of a Class of Generating Systems . . . . . . . . . 199
3.4 Using the Symmetry Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4 Cyclic Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4.1 Cyclic Polynomials in Several Variables . . . . . . . . . . . . . . . . . 203
4.2 Cyclic Polynomials in One Variable . . . . . . . . . . . . . . . . . . . . . 207
4.3 Cyclic Artin-Schreier Towers . . . . . . . . . . . . . . . . . . . . . . . . . . 208
5 Rigid Braid Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.1 The Regularity Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.2 Braid Orbit Genera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.3 A Rationality Criterion for the Pure Braid Group . . . . . . . . . . 215
5.4 Rational Translation of Braid Orbits . . . . . . . . . . . . . . . . . . . . 216
5.5 Groups of Automorphisms as Galois Groups . . . . . . . . . . . . . 219
6 Unramified Rational Places . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.1 Specialization of the Fundamental Group . . . . . . . . . . . . . . . . 222
6.2 The Specialization Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.3 The Theorem of Conway and Parker . . . . . . . . . . . . . . . . . . . . 225
6.4 The Inverse Galois Problem over PAC-Fields . . . . . . . . . . . . . 229
7 Braids and Geometric Automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.1 Specialization to Two Variables . . . . . . . . . . . . . . . . . . . . . . . . 231
7.2 Action of Geometric Automorphisms . . . . . . . . . . . . . . . . . . . 232
7.3 Symmetrized Braid Orbit Genera . . . . . . . . . . . . . . . . . . . . . . . 235
7.4 A Twisted Braid Orbit Theorem . . . . . . . . . . . . . . . . . . . . . . . . 238
7.5 Geometric Galois Extensions over Q.t/ with M24 . . . . . . . . . 240
8 Ramified Rational Places . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
8.1 Decomposition Groups of Ramified Places . . . . . . . . . . . . . . . 241
8.2 Description via the Hurwitz Classification . . . . . . . . . . . . . . . 243
8.3 Braid Cycle Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.4 Prime Divisors of Odd Degree . . . . . . . . . . . . . . . . . . . . . . . . . 248
9 The Katz Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.1 The Convolution Functor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.2 Multiplicativity of Convolution . . . . . . . . . . . . . . . . . . . . . . . . . 254
9.3 Linear Rigidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
9.4 The Existence Algorithm of Katz . . . . . . . . . . . . . . . . . . . . . . . 264
9.5 Braid Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
Contents xv

10 Applications of the Katz Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

10.1 Jordan–Pochhammer Tuples . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.2 Linear and Unitary Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
10.3 Symplectic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
10.4 Orthogonal Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
10.5 Results for Groups in Characteristic Two . . . . . . . . . . . . . . . . 283

IV Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

1 Geometric Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
1.1 Hilbertian Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
1.2 Solutions of Embedding Problems . . . . . . . . . . . . . . . . . . . . . . 288
1.3 Direct Decomposition of the Kernel . . . . . . . . . . . . . . . . . . . . . 290
1.4 From Improper to Proper Solutions . . . . . . . . . . . . . . . . . . . . . 292
1.5 Fields with Projective Galois Group . . . . . . . . . . . . . . . . . . . . . 294
2 Split Embedding Problems with Abelian Kernel . . . . . . . . . . . . . . . . . 296
2.1 Wreath Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
2.2 Split Extensions with Abelian Kernel . . . . . . . . . . . . . . . . . . . 297
2.3 Semiabelian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
3 Embedding Problems with Centerless Kernel . . . . . . . . . . . . . . . . . . . 302
3.1 The Notion of GAR-Realization . . . . . . . . . . . . . . . . . . . . . . . . 302
3.2 Embedding Problems with Characteristically
Simple Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
3.3 Galois Groups of Hilbertian PAC-Fields . . . . . . . . . . . . . . . . . 306
4 Verification of the GAR-Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
4.1 GAR-Realizations in One Variable . . . . . . . . . . . . . . . . . . . . . . 310
4.2 Fields of Constants with Trivial Brauer Group . . . . . . . . . . . . 311
4.3 GAR-Realizations in Several Variables . . . . . . . . . . . . . . . . . . 312
4.4 Specialization to GAR-Realizations in Two Variables . . . . . . 314
5 Frattini Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
5.1 A Decomposition Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
5.2 The Frattini Embedding Theorem . . . . . . . . . . . . . . . . . . . . . . . 318
5.3 Centerless Frattini Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . 321
5.4 Central Frattini Extensions and 2  An . . . . . . . . . . . . . . . . . . . 323
5.5 Central Extensions of An . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
6 The Quadratic Trace Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
6.1 The Cohomological Embedding Obstruction . . . . . . . . . . . . . 329
6.2 The Trace Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
6.3 A Criterion of Serre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
6.4 Central Extensions of Sn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
7 Brauer Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.1 Regular Solutions of Brauer Embedding Problems . . . . . . . . 339
7.2 The Horizontal Local-Global Principle . . . . . . . . . . . . . . . . . . 340
7.3 The Vertical Local-Global Principle . . . . . . . . . . . . . . . . . . . . . 343
7.4 Covering Groups of Simple Groups over Qab .t/ . . . . . . . . . . . 345
xvi Contents

8 Concordant Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347

8.1 The Reduction Theorem of Kochendörffer . . . . . . . . . . . . . . . 347
8.2 The Concordance Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
8.3 Concordance over Local Fields . . . . . . . . . . . . . . . . . . . . . . . . . 354
8.4 Concordance over Global Fields . . . . . . . . . . . . . . . . . . . . . . . . 355
9 The Hasse Embedding Obstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
9.1 Kummer Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
9.2 Definition of the Hasse Obstruction . . . . . . . . . . . . . . . . . . . . . 360
9.3 Translation of the Hasse Obstruction . . . . . . . . . . . . . . . . . . . . 364
9.4 The Hasse Obstruction for Global Fields . . . . . . . . . . . . . . . . . 366
10 Nilpotent Galois Groups over Global Fields . . . . . . . . . . . . . . . . . . . . 372
10.1 Scholz Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
10.2 Scholz Embedding Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 375
10.3 The Theorem of Scholz and Reichardt . . . . . . . . . . . . . . . . . . . 379
10.4 Nilpotent Galois Groups over Global Function Fields . . . . . . 380

V Additive Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383

1 Frobenius Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
1.1 Ordinary Frobenius Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 385
1.2 Cyclic Frobenius Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
1.3 Galois Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
1.4 Effective Frobenius Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 392
2 Computation of the Galois Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
2.1 An Invariant Theoretic Criterion . . . . . . . . . . . . . . . . . . . . . . . . 394
2.2 Computation of Homogeneous Invariants . . . . . . . . . . . . . . . . 397
2.3 Specialization of Relative Resolvents . . . . . . . . . . . . . . . . . . . 398
2.4 Linear Tschirnhaus Transformations . . . . . . . . . . . . . . . . . . . . 400
2.5 The Modular Dedekind Criterion . . . . . . . . . . . . . . . . . . . . . . . 402
3 Polynomials for Split Groups of Lie Type . . . . . . . . . . . . . . . . . . . . . . 406
3.1 Linear Groups SLnC1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
3.2 Symplectic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
3.3 Odd-Dimensional Orthogonal Groups . . . . . . . . . . . . . . . . . . . 413
3.4 Even-Dimensional Orthogonal Groups SOC 2n .q/ . . . . . . . . . . 416
3.5 The Dickson Groups G2 .q/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
4 Polynomials for Twisted Groups of Lie Type . . . . . . . . . . . . . . . . . . . . 423
4.1 The Special Unitary Groups SUn .q/ . . . . . . . . . . . . . . . . . . . . 423
4.2 The Orthogonal Groups SO 2n .q/ . . . . . . . . . . . . . . . . . . . . . . . 428
4.3 The Suzuki Groups 2B2 .q 2 / . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
4.4 The Ree Groups 2 G2 .q 2 / . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
4.5 The Steinberg Triality Groups 3D4 .q/ . . . . . . . . . . . . . . . . . . . 435
5 Field Restriction in Modular Galois Theory . . . . . . . . . . . . . . . . . . . . . 440
5.1 Base Field Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
5.2 Application to Groups of Lie Type . . . . . . . . . . . . . . . . . . . . . . 442
5.3 Explicit Polynomials for SLnC1 .q/ . . . . . . . . . . . . . . . . . . . . . 443
Contents xvii

VI Rigid Analytic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447

1 Results from Rigid Analytic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 449
1.1 Tate Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
1.2 Rigid Analytic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
1.3 Analytification of Algebraic Varieties . . . . . . . . . . . . . . . . . . . 453
1.4 The GAGA-Principle for IP1 .k/an . . . . . . . . . . . . . . . . . . . . . . 455
2 The Inverse Problem over Qp .t/ and IFp .t/ . . . . . . . . . . . . . . . . . . . . 460
2.1 Induced Covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
2.2 The Inverse Problem over Complete Ultrametric Fields . . . . 462
2.3 The Inverse Problem over IFp .t/ . . . . . . . . . . . . . . . . . . . . . . . . 463
2.4 The Conjecture of Šafarevič for IFp .t/ . . . . . . . . . . . . . . . . . . . 464
3 Free Quotients of the Fundamental Group . . . . . . . . . . . . . . . . . . . . . . 467
3.1 Free Composites of Galois Extensions . . . . . . . . . . . . . . . . . . . 467
3.2 Galois Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
3.3 A Free Quotient of the Algebraic Fundamental Group . . . . . 471
4 Large Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
4.1 Existentially Closed Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
4.2 Characterization of Large Fields . . . . . . . . . . . . . . . . . . . . . . . . 475
4.3 Split Embedding Problems over Large Fields . . . . . . . . . . . . . 477
4.4 Application to Hilbertian PAC-Fields . . . . . . . . . . . . . . . . . . . 479
5 On the Fundamental Group with Restricted Ramification . . . . . . . . . 480
5.1 Projectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
5.2 Embedding Problems with p-Kernel . . . . . . . . . . . . . . . . . . . . 483
5.3 The Conjecture of Abhyankar for the Affine Line . . . . . . . . . 483
5.4 The General Case of the Conjecture of Abhyankar . . . . . . . . 485

Appendix: Example Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

1 Regular Realizations for Transitive Groups of Degree Less
than 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
2 Regular Realizations for Nonsolvable Primitive Groups . . . . . . . . . . . 499
3 Realizations over Q for Transitive Groups of Degree up to 14 . . . . . . 502

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
I The Rigidity Method

The idea of deducing the realizability of a finite group as Galois group over Q.t/
from the existence of rigid systems of generators as far as we know first appeared
implicitly in the appendix to the dissertation of Shih (1974). This concept was sub-
sequently considered independently by Fried (1977), Belyi (1979), Matzat (1979,
1984, 1985a) and later Thompson (1984a) and extended in different directions. In
this first chapter we develop the one variable treatment by Belyi, Matzat and Thomp-
son, which in essence relies on the covering theory of the punctured Riemann sphere
(resp. IP1 .C/) and the classical Riemann existence theorem. (In the third chapter, the
multi-variable approach going back to work of Fried will be presented.)
The first five paragraphs lead directly to the Basic Rigidity Theorem including its
character theoretic and representation theoretic variants and thus form the main body
of this chapter. Other treatments of the main results in this part may be found in the
lecture notes of Matzat (1987), Kap. I, as well as in those of Serre (1992), Sect. 7–9.
The next five paragraphs contain further leading results. In the sixth paragraph
we study outer automorphisms of the fundamental group originating from embed-
dings, which are here called geometric automorphisms, in contrast to earlier notation
(Matzat (1986, 1987)). These can be employed to derive the Twisted Rigidity The-
orem as the currently strongest variant of the Basic Rigidity Theorem. In the next
paragraph we introduce the translation technique of Shih (1974) and Malle (1991)
and apply it to the groups L2 .p/. Paragraph 8 contains the results needed for the
realization of automorphism groups of finite groups as Galois groups; the prototype
for this embedding theorem can be found in Matzat (1992). Finally in Paragraph 9
we show how to construct generating polynomials for field extensions whose exis-
tence was proved with the rigidity method, and in Paragraph 10 we note what can be
said about specializations of such polynomials, using the results of Fried and Dèbes
(1990) and Beckmann (1989, 1991).
As examples we consider the abelian groups, the groups An and Sn , L2 .p/ and
PGL2 .p/, as well as the smallest sporadic groups M11 and M12 . The second chapter
will then contain a systematic study of the results obtainable by the rigidity method
in the area of finite almost simple groups.

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 1

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_1
2 I The Rigidity Method

1 The Inverse Galois Problem over C.t / and IR.t /

The structure of the fundamental group of the punctured Riemann sphere is well
known. From its algebraic variant, the classical solution of the inverse problem of
Galois theory over C.t/ follows immediately. After extension of the fundamental
group by complex conjugation one also derives the solution of the inverse Galois
problem over IR.t/.

1.1 The Fundamental Group of the Punctured Riemann Sphere

Our starting point is the Riemann sphere X WD C. O From this, a set of s points
S WD fP1 ; : : : ; Ps g is removed. For any choice of base point P0 2 X n S the
top
topological fundamental group 1 .X n S I P0 / relative to P0 is generated by
homotopy classes of nonintersecting loops i from P0 counterclockwise around
Pi .

· X
P2
·
P1

· P3
P0

· Ps
· ·
·
· ·
top
Fig. 1.1 Generators of 1 .X n S I P0 /

The path obtained by concatenation of representatives for the i can clearly be

top
contracted to one point on X n S , so the generators 1 ; : : : ; s of 1 satisfy at
least the relation 1    s D 1.

Theorem 1.1 (Hurwitz (1891)). Let S be a subset of the Riemann sphere X D CO

top
of cardinality s. Then the fundamental group 1 .X n S I P0 / with respect to any
base point P0 2 X n S has the structure
top
1 .X n S I P0 / D h1 ; : : : ; s j 1    s D 1i: (1.1)

A proof of this result can be found for example in Seifert and Threllfall (1934),
47 (see also Stöcker and Zieschang (1988), Aufgabe 5.7.A2).
1 The Inverse Galois Problem over C.t / and IR.t / 3

The only continuous automorphism of the field of complex numbers C is given

by complex conjugation, denoted here by . If the set S introduced previously and
the base point P0 remain stable under , i.e., if S  D S and P0 D P0 , then 
top
acts on 1 .X n S I P0 /. Indeed, assume that S consists of r pairs of complex
conjugate points P1 ; : : : ; P2r arranged first by decreasing imaginary part and then
by decreasing real part (in case of equality of the imaginary parts), and the real
points P2rC1 <    < Ps . Choosing the real base point P0 < P2rC1 we obtain the
following diagram:

g2 · X
g1
P1
·
P2
·
·
P2r+1 Ps
r · · · ·
P0 IR
· gs
· g 2r+1

P2r−1
·

g 2r−1 P2r g 2r
·

Fig. 1.2 Action of complex conjugation

With this standard arrangement, the homotopy classes of paths i are sent to
1
2rC1i for i D 1; : : : ; 2r, and 2rCj for j D 1; : : : ; s  2r is mapped to
1 1 1
2rC1    2rCj 1 2rCj 2rCj 1    2rC1 :

This proves the following:

Theorem 1.2 (Hurwitz (1891)). If the set S and the base point P0 in Theo-
rem 1.1 are stable under complex conjugation , then  induces an automorphism
top
of 1 .X n S I P0 /. With the standard arrangement of the base point P0 and the
top
points in S as in Figure 1.2,  acts on the generators of 1 .X n S I P0 / via
1 1
1
.1 ; : : : ; s / D .2r ; : : : ; 11 ; 2rC1
1
; : : : ; .s1 /s1 2rC1 /: (1.2)

1.2 The Algebraic Variant of the Fundamental Group

top alg
The topological fundamental group 1 has an algebraic analogue 1 , for which
however the convenient visualization as group of homotopy classes of paths is lost.
4 I The Rigidity Method

Namely, let K WD C.X / be the function field of X D C, O or equivalently of the

1
projective line IP .C/. Then K is isomorphic to the field of rational functions C.t/
over C. Denote by IP.K=C/ the set of prime divisors or equivalently valuation ideals
of the function field K=C. Then the set S  X corresponds to the subset S of
primes of K=C whose valuation ideal has a common zero at one of the points Pi .
Now let NS denote the set of all finite Galois extension fields of K, ramified only
at prime divisors of S, in a fixed algebraic closure KO of K. The union of all N 2 NS
forms the maximal extension field MS of K (in K) O unramified outside S. It is again
Galois over K, and the (for jSj > 1 finite) Galois group is obtained as the projective
limit of the finite Galois groups Gal.N=K/:

Gal.MS =K/ D lim .Gal.N=K//N2NS : (1.3)



This Galois group formally depending on KO is called the algebraic fundamental

group of X n S :
alg alg O WD Gal.MS =K/:
1 .X n S / D 1 .X n S ; K/ (1.4)

For the algebraic fundamental group we get the following profinite version of Rie-
mann’s existence theorem:

Theorem 1.3 (Profinite Riemann Existence Theorem). The algebraic fundamental

alg
group 1 .X n S / is isomorphic to the profinite completion of the topological fun-
top
damental group 1 .X n S I P0 /:
alg top
1 .X n S / Š O 1 .X n S I P0 /:

Moreover for any choice of the base point P0 there exists a monomorphism
top alg
 W 1 .X n S I P0 / ! 1 .X n S /;
alg
such that 1 .X n S / is generated as topological group by the images of the i
(where .i / is identified with i ):
alg
1 .X n S / D h1 ; : : : ; s j 1    s D 1iO: (1.5)

Proof. We write X  WD X n S . The topological space X  is sufficiently connected

(in the sense of Stöcker and Zieschang (1988), Def. 6.4.3), so there exists a universal
covering
u W XO  ! X  ;
whose group of covering transformations (deck transformations) equipped with the
product of mappings is isomorphic to 1 .X  I P0 /:
top

! W Deck.u/ ! 1 .X  I P0 /
top
(1.6)
1 The Inverse Galois Problem over C.t / and IR.t / 5

(see for example loc. cit., Kor. 6.5.5, or Forster (1981), Thm. 5.6). The isomorphism
is not canonical and depends on the choice of a point PO 0 2 XO  above P0 . From the
fundamental theorem for coverings of topological surfaces (Stöcker and Zieschang
(1988), Satz 6.6.3) it follows that for each finite unramified normal covering

p W Y  ! X 

there exists a normal subgroup O of finite index in Deck.u/ such that Y  is home-
omorphic over X  to the orbit space XO  =O (equipped with the quotient topology).
Via the canonical map from XO  onto XO  =O we obtain a universal covering

v W XO  ! Y  Š XO  =O

of Y  with p  ı v D u and Deck.v/ Š O. Then we have

Deck.p  / Š Deck.u/= Deck.v/:

Now the covering p  is holomorphic with respect to the lifted analytic structures,
so by the Riemann Hebbarkeitssatz it possesses a uniquely determined holomorphic
continuation
p W Y ! X

on the compactification Y of Y (see Forster (1981), Thm. 8.4). This continuation
is unramified outside S , and moreover satisfies

Deck.p/ Š Deck.p  /:

The field N WD C.Y / is a Galois extension of K D C.X /, unramified over K outside

of the set of prime divisors S of K=C belonging to S . Its degree coincides by the
Riemann existence theorem with the number of sheets of the coverings p and p  .
From this it follows that
Gal.N=K/ Š Deck.p/
(with Gal.N=K/ acting on the right), and hence finally
Gal.N=K/ Š  .X  I P /=!.O/
top
1 0 (1.7)

with ! from (1.6) (see loc. cit., Thm. 8.12). Conversely, each finite field exten-
sion N of K unramified outside S uniquely determines a compact Riemann surface
Y , which is unramified outside of S , and with N D C.Y /. Two such extension
fields are isomorphic over K precisely when the corresponding Riemann surfaces
are homeomorphic over X (see loc. cit., Thm. 8.3 with Ex. 8.1).
Now let NS be the set of all finite Galois extensions of K inside KO unramified
S by O the set of all normal subgroups of Deck.u/ of finite
outside S, and denote
index. Then MS WD N2NS N is Galois over K and there exists an isomorphism

!O W Gal.MS =K/ D lim .Gal.N=K//N2NS !


lim .1 .X  I P0 /=!.O//O2O D O 1 .X  I P0 /

top top

6 I The Rigidity Method

since the structural homomorphisms of the respective projective systems commute

with the isomorphisms from (1.7). As 1 .X  I P0 / is free of finite rank, hence
top

residually finite (see for example Lyndon and Schupp (1977), Ch. III, Prop. 7.11,
with complement on p.195), the canonical homomorphism

O W 1 .X  I P0 / ! O 1 .X  I P0 /
top top

has trivial kernel (see Serre (1964), 1.1). Hence

 WD !O 1 ı O W 1 .X  I P0 / ! Gal.MS =K/ D 1 .X n S /
top alg

is a monomorphism with the property (1.5). t

u

It remains to study the role of the embedded homotopy classes 1 ; : : : ; s in

alg
1 .X n S /. This is achieved with:

Theorem 1.4 (Abhyankar (1957)). The images of the homotopy classes 1 ; : : : ; s

top alg
in 1 .X n S I P0 / embedded via  into 1 .X n S / generate procyclic inertia
alg
groups in the Galois group Gal.MS =K/ D 1 .X n S /. More precisely the valua-
tion ideals Pi of K=C corresponding to Pi 2 S may be embedded into valuation
ideals P O i of MS =C such that the respective inertia groups satisfy

O i =Pi / D hi iO:

I.P (1.8)

Proof. We continue to use the notation introduced in the proof of Theorem 1.3
above. In particular, PO 0 2 XO  denotes the point above P0 used in the construc-
tion of ! and . By the Main Lemma of Covering Theory (see for example Stöcker
and Zieschang (1988), Satz 6.22) any path ci representing the homotopy class i
possesses a unique lifting to a path cQi on Y  with cQi .0/ D PQ 0 WD v.PO 0 /.

Pi
ci ∈ g i

P0
Fig. 1.3 Deformation of ci

Deforming this path according to Figure 1.3 homotopically in X  [ fPi g yields

a double path ci connecting P0 and Pi . Then ci also possesses a lifting cQi on Y
1 The Inverse Galois Problem over C.t / and IR.t / 7

with respect to the extension p of p  , satisfying cQi .0/ D PQ 0 . We denote by PQ i the

intersection of the fiber p 1 .Pi / with cQi , and by di the covering transformation
of p determined by di .PQ 0 / D cQi .1/. The points PQ i are now di -invariant, and we
have d1 ı    ı ds D Id.
Let P Q i be the valuation ideal of N defined by PQ i , and i the automorphism in
G WD Gal.N=K/ induced by f i .P/ Q WD f .di .P// Q for f 2 N. As P Q i D P
Q i , the
i
element i is contained in the decomposition group and hence the inertia group
GI .PQ i =Pi /. Conversely, if  2 GI .P Q i =Pi / is given, then the corresponding cover-
ing transformation d leaves invariant the point PQ i 2 Y . Obviously all such covering
transformations are obtained when lifting the powers of i , hence i is a generating
element of GI .P Q i =Pi /. The elements .i /N2NS in the projective limit

Gal.MS =K/ D lim .Gal.N=K//N2NS



coincide by constructionS with the generators i D .i / of Gal.MS =K/ in (1.5).

Moreover, with P O i WD Q
N2NS Pi we obtain a valuation ideal of MS whose iner-
tia group in MS =K is generated as profinite group by i D .i /N2NS . t
u
The previous considerations now easily yield a solution of the inverse problem
of Galois theory over the field of rational functions C.t/:
Corollary 1.5. Every finite group occurs as Galois group over C.X / Š C.t/.
Proof. Let the finite group G be generated by the elements 1 ; : : : ; s1 , s  2. Then
there exists a continuous epimorphism

W Gal.MS =K/ ! G with .i / D i for i D 1; : : : ; s  1

/
(and .s / D .1    s1 /1 ). The fixed field N WD Mker.
S now yields a Galois
extension of K D C.X / with

Gal.N=K/ Š Gal.MS =K/= ker. / Š G: t

u

1.3 Extension by Complex Conjugation

Under the assumptions of Theorem 1.2 the field MS is not only Galois over C.t/,
but also over IR.t/. More precisely we have:
Theorem 1.6. Let S be a finite subset of X D IP1 .C/ invariant under complex
conjugation , and let MS =C.X / denote the maximal field extension unramified
outside the prime divisors in S. Then we have
alg
Gal.MS =IR.X // Š 1 .X n S /Z2 : (1.9)

If the ramification points, the base point and the homotopy classes of paths are
chosen in the standard configuration of Figure 1.2, then the generator  of the cyclic
8 I The Rigidity Method
alg
group Z2 of order 2 acts on the embedded topological generators i of 1 .X n S /
via 1 1
1
.1 ; : : : ; s / D .2r ; : : : ; 11 ; 2rC1
1
; : : : ; .s1 /s1 2rC1 /: (1.10)

Proof. By Theorem 1.2 the complex conjugation map  acts on the generators i of
the topological fundamental group in the described manner. As the corresponding
formula remains valid for the canonical projections of the i in the finite factor
groups
top
1 .X n S I P0 /=O Š Gal.N=C.X //;

 may be lifted to an automorphism O of MS =C.X / with P O O D PO 0 , which moreover

0
alg
acts like  on the embedded topological generators i of 1 .X n S /. Being a
generator of the decomposition group of P O 0 =P0 , O has order 2. Renaming O to 
we thus obtain the assertion. t
u

We call a finite field extension N=k.t/ geometric, if k is algebraically closed in

N (or equivalently, if N=k is regular). The above structure theorem now yields a
solution of the inverse Galois problem for IR.t/ even in the strong form:

Corollary 1.7 (Krull and Neukirch (1971)). Every finite group occurs as a geomet-
ric Galois group over IR.t/.

Proof. Let G be a finite group, generated by 1 ; : : : ; r . For a chosen -invariant

subset S  X in standard configuration with s D 2r, due to 1    r r1    11 D 1
there exists a continuous epimorphism
(
i for i D 1; : : : ; r ;
W Gal.MS =C.X // ! G with .i / D 1
2rC1i for i D r C 1; : : : ; 2r:

Since moreover .1 ; : : : ; r ; r1 ; : : : ; 11 / D .1 ; : : : ; r ; r1 ; : : : ; 11 / by Theo-

rem 1.6, ker. / is -invariant. Hence the semidirect product ker. /hi is a nor-
mal open subgroup of Gal.MS =IR.X // whose fixed field N over IR.X / Š IR.t/ is
Galois, and we have

Gal.N=IR.X // Š Gal.MS =IR.X //=.ker. /hi/

Š Gal.MS =C.X //= ker. / Š G:

Moreover IR is algebraically closed in N, hence the extension N=IR.X / is geomet-

ric. t
u

1.4 Generalization to Function Fields of Riemann Surfaces

The previous considerations apply to the special case of coverings of the Rie-
mann sphere. If instead we start from an arbitrary compact Riemann surface X
1 The Inverse Galois Problem over C.t / and IR.t / 9

of genus g, then for S D fP1 ; : : : ; Ps g  X the fundamental group of X n S

with respect to a base point P0 2 X n S has the form
top
1 .X n S I P0 / D
h˛1 ; ˇ1 ; : : : ; ˛g ;ˇg ; 1 ; : : : ; s j ˛1 ˇ1 ˛11 ˇ11    ˛g ˇg ˛g1 ˇg1 1    s D 1i:

Following the same procedure as in Section 1.2 we then obtain the algebraic funda-
mental group, defined as the Galois group of the maximal field extension MS =C.X /
unramified outside S, up to an isomorphism by profinite completion
alg top
1 .X n S / Š O 1 .X n S I P0 /:
top top
Via the embedding of 1 into O 1 and the above isomorphism we can again iden-
tify the generators i with generators of procyclic inertia groups of MS =C.X /.
At least for s  1 these more general fundamental groups are again free, hence
as in Section 1.2 we obtain the solution of the inverse problem of Galois theory over
all function fields of compact Riemann surfaces.
For a detailed account the reader is referred to the lecture notes Matzat (1987)
and Serre (1992).
10 I The Rigidity Method

2 Arithmetic Fundamental Groups

In the previous paragraph, the structure of the fundamental group of the punctured
Riemann sphere over the complex numbers was determined. With the Weil Rational-
N 
ity Criterion the results may be transferred to the punctured projective line IP1 .Q/
N
over the field of all algebraic numbers Q. If the ramification locus is defined over Q,
the corresponding field extension will be invariant under the absolute Galois group
N
Gal.Q=Q/ of Q. This yields a splitting group extension, called the arithmetic funda-
mental group, where the elements of Gal.Q=Q/ N act as outer automorphisms on the
alg 1 N 
algebraic fundamental group 1 .IP .Q/ /.

2.1 Descent to Algebraically Closed Subfields

The results on the algebraic fundamental group of the punctured Riemann sphere
resp. of IP1 .C/ D IP1 .C/nS can be transferred to arbitrary algebraically closed
fields kN of characteristic zero. We give the complete proof only for the case of
algebraically closed subfields of C. For the general case we refer the reader to
Grothendieck (1971), Exp. XIII, Cor. 2.12 (see also Popp (1970), 11), or van
den Dries and Ribenboim (1979).
Proposition 2.1. Let kN be an algebraically closed subfield of C, X D IP1 .C/,
X .k/N D IP1 .k/,N S D fP1 ; : : : ; Ps g a finite subset of X .k/
N and S D fP1 ; : : : ; Ps g
N N
the set of valuation ideals of k.X / Š k.t/ corresponding to S . Then for every
finite field extension N=C.t/ unramified outside S there exists precisely one function
field NN =kN which is geometric over k.t/
N and with NN C WD NN ˝ N C D N.
k

Proof. Let first N=C.t/ be a finite field extension unramified outside S. Further let
N
 be the group of k-automorphisms of C and O the group of those extensions of
elements of  to an algebraic closure C.t/ of C.t/ which fix k.t/ N pointwise. Then
O O
we have C D kN and C.t/ D k.t/.N For each ıO 2 O the field extension Nı =C.t/ is
1 N
unramified outside S since S  IP .k/, so it is a subfield of MS in (1.3) of degree
O
ŒNı W C.t/ D ŒN W C.t/ . We saw that the algebraic fundamental group
alg
1 .X n S / D Gal.MS =C.t//

is a finitely generated profinite group, so the number of normal subgroups of finite

O
index is finite. Therefore also the set fNı j ıO 2 g
O is finite. So the stabilizer

O
Q WD fıO 2 O j Nı D Ng
Q N
of N in O has finite index in O and C.t/ =k.t/ is a finite extension of constants.
Since kN was assumed to be algebraically closed we thus get C.t/ D k.t/ N respec-
Q

Q N
tively C D k.
2 Arithmetic Fundamental Groups 11

Now let a 2 X .k/N n S , Pa the corresponding valuation ideal in k.t/,

N Q a an
P
extension of Pa to N=C and

Q ıQ D P
Q a WD fıQ 2 Q j P Q ag
a

Q a . Since N=C.t/ is finite, the same is true for the number of

the stabilizer of P
extensions of Pa to N=C and hence also for the index .Q W Q a /, which implies
Q
Ca D kN as above.
By the theorem of Riemann-Roch (see Forster (1981), Thm. 16.9, for example)
N contains non-constant functions with a as the only pole. If m denotes the smallest
occurring order of pole at a, the linear space
Qm
L .P a / D fx 2 N j ordP
Q a .x/  mg

has dimension 2 over C and is generated by 1 and some ´ 2 N with ordPQ a .´/ D m:
L .PQ m / D C C C´. Since P Q a splits completely in N=C.t/, the function ´ even has
a
to generate N=C.t/, i.e., we have N D C.t; ´/.
The completion N O a of N with respect to P Q a coincides with the completion of
C.t/ with respect to the numerator divisor of .t a/, hence equals the field of formal
power series NO a D C..t  a//. In particular every x 2 L .P Qm
a / can be written in the
form X
xD ai .t  a/i with ai 2 C
i m

O a . Replacing ´ by a suitable multiple and subtracting a constant we may assume

in N
without loss of generality that am D 1 and a0 D 0 in the above representation for
´. The map
Xm
W L .PQ m / ! CŒX m ; x 7! ai X i ;
a
i D0

on the C-vector space of polynomials of degree at most m is C-linear, Q a -equivariant

and also injective, since N contains no non-constant functions without poles. Thus
every ıQ 2 Q a satisfies
Q X Q
m1
.´ı / D X m C ı
ai Xi:
i D1

ıQ Q
Q m /, so there exist constants b; c 2 C with ´ı D b´ C c.
But we have ´ 2 L .P a
Q Q
Comparison of coefficients yields aiı D ai for m < i < 0 and hence ´ı D ´. Con-
sequently ´ is a Q a -invariant generator of N=C.t/. But then the coefficients of the
minimal polynomial of ´ over C.t/ must be Q a -invariant, so by the above already
N
lie inside k.t/. N ´/=k.t/
Thus k.t; N is an extension of degree

N ´/ W k.t/
Œk.t; N D ŒN W C.t/ with C.t; ´/ D N
N
and hence geometric over k.t/. N ´/.
This proves the existence of NN WD k.t;
12 I The Rigidity Method

N with CNN 1 D CNN 2 D N. Since

Now let NN 1 ¤ NN 2 be two fields geometric over k.t/
N N
N1 and C.t/ are linearly disjoint over k.t/ the composite NN 1 NN 2 inside C.t/ satisfies
N
ŒNN 1 NN 2 W k.t/ N
> ŒNN 1 W k.t/ D ŒN W C.t/ D ŒCNN 1 NN 2 W C.t/ ;
N
contrary to the fact that a minimal polynomial for a primitive element of NN 1 NN 2 =k.t/
remains irreducible over C.t/. This shows the uniqueness of N . N t
u

The above Proposition 2.1 contains the main ingredient for the following general-
ization of Theorems 1.3 and 1.4 for algebraic fundamental groups over algebraically
closed subfields of C.

Theorem 2.2 (Grothendieck (1971)). Let kN be an algebraically closed subfield of

C, X D IP1 .C/, X .k/ N D IP1 .k/,
N S D fP1 ; : : : ; Ps g a finite subset of X .k/,
N SD
N
fP1 ; : : : ; Ps g the set of valuation ideals of k.X / corresponding to S , and MN S the
maximal algebraic extension field of k.X N N unramified outside S. Then the
/ Š k.t/
N
algebraic fundamental group 1 .X .k/nS / D Gal.MN S =k.X
alg N // has the form
alg N
1 .X .k/nS / D h1 ; : : : ; s j 1    s D 1iO: (2.1)

Moreover the elements i , i D 1; : : : ; s, are generators of inertia groups of valuation

O i of MN S =kN lying over Pi :
ideals P
O i =Pi / D hi iO:
I.P (2.2)

Proof. Denote by NN S (NS respectively) the set of finite Galois extensions of KN WD

N
k.X / (resp. K WD C.X /) unramified outside S. By Proposition 2.1 the map
N S ! NS ;
WN NN 7! N WD CNN

is bijective. As in addition the intersection and the composition of fields does not
lead outside NN S , is an isomorphism of lattices.
The map commutes with the canonical epimorphisms of the projective systems
.Gal.NN =K//
N N N and .Gal.N=K//N2NS , so the corresponding projective limits are
N 2 NS
isomorphic:

Gal.MN S =K/
N D lim .Gal.NN =K//
N N N Š lim .Gal.N=K//N2NS D Gal.MS =K/:
N 2 NS
 

This completes the proof of the first part.

For P 2 S denote by PK the unique extension to K and by P O K the extension to
MS constructed according to Theorem 1.4. The inertia group of P O WD P
O K j N over
MS
P consists of the restrictions of all elements in I.PO K =PK / to MN S , and we obviously
have
O
I.P=P/ Š I.PO K =PK /:

This yields the second part of the assertion. t

u
2 Arithmetic Fundamental Groups 13

Remark. Let kN  denote the fixed field under complex conjugation of the field kN C
in Theorem 2.2. Then if S  D S , we have

Gal.MN S =kN  .X // Š Gal.MN S =k.X

N //hi with 2 D 1: (2.3)

If moreover the ramification points are arranged in the standard configuration then
N
 acts on the generators i of Gal.MN S =k.X // as described by formula (1.10).

The above Theorem and the subsequent Remark lead to the following:
N
Corollary 2.3. Every finite group occurs as a geometric Galois group over Q.t/ and
N  .t/.
over Q

In particular the inverse Galois problem is solved in the affirmative over all ratio-
N with algebraically closed field of constants kN of characteris-
nal function fields k.t/
tic zero.

2.2 The Fundamental Splitting Sequence

From now on and for the rest of this chapter we will always take kN to be the field
N since this is one of the most interesting cases. In later
of all algebraic numbers Q,
chapters, more general base fields will also be studied. Denote the corresponding
algebraic fundamental group by

s
N
WD Gal.MN S =Q.X // D h1 ; : : : ; s j 1    s D 1iO: (2.4)

Theorem 2.4 (Splitting Theorem). If the field kN in Theorem 2.2 is taken to be the
N of all algebraic numbers, and the set S is invariant under the absolute Galois
field Q
group Q D Gal.Q=Q/ N of Q, then MN S is Galois over Q.X / Š Q.t/ and we have

Gal.MN S =Q.t// Š s  Q: (2.5)

Proof. Obviously MN S is separable and algebraic over Q.t/. By assumption each

monomorphism ' from MN S into an algebraic closure MO S of MN S permutes the ele-
ments of S, and as MN S is uniquely determined inside MO S as the maximal algebraic
N
extension field of Q.t/ unramified outside S, we clearly have MN S D MN S for all such
'

'. Hence ' is an automorphism of MN S =Q.t/, and MN S =Q.t/ is Galois, with group
WD Gal.MN S =Q.t//.
Now let P 2 IP.Q.t/=Q/ denote a prime divisor of degree one whose extension
PO to MN S remains unramified, and denote by D .P=P/ O the (profinite) decompo-
O
sition group of P=P in (for this, see Nagata (1977), Ch. 7.3). As P=PO stays
N
inert in Q.t/=Q.t/ and splits completely in MN S =Q.t/,
N D . O
P=P/ intersects s triv-
O
ially, and D .P=P/ together with s generate the full group . Since s G and
O
D .P=P/ Š Q the assertion now follows. t
u
14 I The Rigidity Method

The previous proof also yields an explicit description of complements in the

semidirect product (2.5):

Corollary 2.5. The decomposition groups of unramified prime divisors of degree

N
one are closed complements to Gal.MN S =Q.t// in Gal.MN S =Q.t//.

A generalization of Theorem 2.4 to arbitrary disclosed function fields of one vari-

able (i.e., function fields of one variable containing a prime divisor of degree 1) over
fields of characteristic zero is proved in Matzat (1987), I.5, Satz 1.

2.3 The Action via the Cyclotomic Character

Unfortunately only very little is known about the action of Q on s in the sequence
(2.5) for Gal.MN S =Q.t//. An explicit knowledge of this action would provide an
answer to the inverse problem of Galois theory. (An implicit description follows
from the work of Ihara, see for example Ihara (1991), 3.) But at least the conjugacy
classes of the images of the generators i of s can be determined.
Clearly any ı 2 Q sends the n-th root of unity n WD e 2 i=n to a primitive power
cn .ı/
n , with cn .ı/ 2 .ZZ=nZZ/ . This defines a continuous homomorphism

cW Q
O ;
! ZZ ı 7! c.ı/ WD .cn .ı//n2IN (2.6)

O (the profinite completion of

from Q onto the group of units in the Prüfer ring ZZ
ZZ). This homomorphism is called the cyclotomic character of Q . By the Theo-
rem of Kronecker and Weber (see for example Washington (1982), Thm. 14.1), the
maximal abelian extension field Qab of Q is generated by the roots of unity. Cor-
respondingly the cyclotomic character induces a canonical isomorphism from the
O :
commutator factor group Qab onto ZZ

ab O :
D Gal.Qab =Q/ Š ZZ
Q

N
If we assume that, as in Theorem 2.4, the set S of prime ideals ramified in MN S =Q.t/
N
is invariant under Gal.Q.t/=Q.t// Š Q , then the elements P1 ; : : : ; Ps of S are per-
muted by Q in the form

ı W S ! S; Pi 7! P.i /ı WD Pıi : (2.7)

This induces a permutation representation of Q on S, and hence into the symmetric

group Ss .

Theorem 2.6. Let N Š Q be a closed complement to

s D h1 ; : : : ; s j 1    s D 1iO
2 Arithmetic Fundamental Groups 15

in Gal.MN S =Q.t// as in Theorem 2.4, and ıN 2 N a lifting of an element ı 2

N N
Gal.Q.t/=Q.t// to MN S . Then iı is conjugate in s to .ic.ı/
/ı
, so the conjugacy class
in s is given by
 ıN   c.ı/ 
i D .i /ı : (2.8)

Proof. By Theorem 2.2 the i generate inertia subgroups of prime ideals P O i of

N O
MS lying over Pi 2 S, so we have I.Pi =Pi / D hi iO. The extension ı 2 N of N
N N O ıN =Pı ,
ı 2 Gal.Q.t/=Q.t// maps hi iO to hiı iO. Now this is the inertia group of P i i
N
and we have Pıi D P.i /ı , so iı is conjugate in s to a power of .i /ı , i.e., there
exists di .ı/ N 2 ZZ O  with Œ ıN D Œ di .ı/N
. The commutator factor group sab is a
i .i /ı
free abelian profinite group of rank s  1 with the two relations N1    Ns D 1 and
N N
N coincide for all
N1d1 .ı/    Nsds .ı/ D 1 for the classes Ni of i in sab . Hence the di .ı/
i D 1; : : : ; s, and do not depend on the particular choice of extension ıN of ı. So

N
d W Gal.Q.t/=Q.t// O ;
! ZZ N
ı 7! d.ı/ WD di .ı/

is a well defined homomorphism.

For the determination of d we assume that S contains at least two
N
Gal.Q.t/=Q.t//-invariant prime divisors, say P1 , Ps . This can always be guaran-
teed by enlarging S. The fixed field KN Sab of sab is the maximal abelian extension
of KN unramified outside S, hence generated by the union of all roots of functions
ti 2 KN with divisor .ti / D Pi P1
s for i D 1; : : : ; s  1, where by the choice of P1
and Ps we may assume t1 D t. Then the fixed field of hN2 ; : : : ; Ns1 iOis obtained as
[
LN WD N n/
K.´ for ´n 2 KO with .´n /n D t;
n2IN

where in addition we can require that .´nm /m D ´n , for all n; m 2 IN. In particular
N
with this choice ı possesses an extension ıN onto LN satisfying ´ın D ´n for all n 2 IN.
The field LN is N WD N1 -invariant, so the same is true for K.´
N n /, and there exists a
Q N N Q
primitive n-th root of unity n 2 Q with ´n D n ´n . Moreover we have

N ıN
ı N N d.ı/
Qnd.ı/ ´n
Q c.ı/ D Q ı D ´n D .´n / D D Qnd.ı/ ;
n n
´ınN ´ınN ´n

from which it follows that d coincides with the cyclotomic character. t

u

This has the following consequence:

Corollary 2.7. The action of Q on the commutator factor group sab of s is

entirely given in terms of the action through the cyclotomic character c, by the
formula
c.ı/ c.ı/
.N1 ; : : : ; Ns /ı D .N.1/ı ; : : : ; N.s/ı /: (2.9)
16 I The Rigidity Method

2.4 The Theorem of Belyi

The aim of this section is to show that each ı 2 Q acts as an outer automorphism
on the algebraic fundamental group s . This leads in particular to a representation
of the absolute Galois group Q into the group of outer automorphisms of the free
profinite group s .
Proposition 2.8. Let L= N QN be an algebraic function field in one variable. Then there
N
exists a function t 2 L such that at most the support of t and t  1 is ramified in
N Q.t/.
L= N
Proof. First let x 2 LN be an arbitrary function. Then only finitely many prime divi-
N
sors P1 ; : : : ; Ps 2 IP.Q.x/= N are ramified in L=
Q/ N
N Q.x/. As Q N is algebraically closed,
1 N N N which sends
there exists a canonical bijective map from IP .Q/ onto IP.Q.x/= Q/,
N
a 2 Q to the numerator divisor of x  a, and 1 to the denominator divisor of x.
Denote the preimage of fP1 ; : : : ; Ps g under this bijection by S .x/ D fa1 ; : : : ; as g.
We first show that there exists a function y 2 Q.x/ N such that the prime divisors of
N N N
Q.y/ ramified in L=Q.y/ are defined over Q.y/, i.e., such that S .y/ is a subset of
IP1 .Q/. For this, let

d.x/ WD maxfŒQ.a/ W Q j a 2 S .x/g

and
r.x/ WD jfa 2 S .x/ j ŒQ.a/ W Q D d.x/gj:
If d.x/ D 1 there is nothing to prove. For d.x/ > 1 we choose a 2 S .x/ with
ŒQ.a / W Q D d.x/. Denote the minimal polynomial of a over Q by f .X /, and let
xQ WD f .x/. Then we obviously have

Q  ff .a/ j a 2 S .x/g [ ff .a/ j f 0 .a/ D 0g [ f1g:

S .x/
N we have the estimate
For all a 2 Q

ŒQ.f .a// W Q ŒQ.a/ W Q d.x/

for the degree ŒQ.f .a// W Q , and for the zeroes of f 0 .X / the even better estimate

ŒQ.f .a// W Q deg.f 0 / < d.x/ N with f 0 .a/ D 0

for a 2 Q

holds. So d.x/ Q d.x/, and if d.x/ Q D d.x/ then r.x/ Q < r.x/. By induction in
descending lexicographical order for the pair .d.x/; r.x// the existence of the
required function y 2 Q.x/ N with S .y/  IP1 .Q/ follows.
In the second step we reduce the cardinality of S .y/. Since Aut.Q.y/= N N Š
Q/
N N N
PGL2 .Q/ acts threefold transitively on IP.Q.y/=Q/, the assertion follows imme-
diately from the first step in the case of jS .y/j 3. Otherwise we have S .y/
N with .b1 / D
fb1 ; b2 ; b3 ; b4 g. By the threefold transitivity there exists 2 PGL2 .Q/
1, .b2 / D 0, .b3 / D 1. Then already belongs to PGL2 .Q/. Permutation of b1 ,
b2 and b3 allows to assume that 0 < .b4 / < 1. Indeed, the transposition .12/ leads
2 Arithmetic Fundamental Groups 17

to the transformation .b4 / 7! .b4 /1 , and .23/ leads to .b4 / 7! 1  .b4 /.
Hence there exist positive integers n; m with .b4 / D n=.n C m/. Defining

.n C m/nCm n
g.Y / WD Y .1  Y /m
nn mm
we have g. .bi // 2 f0; 1; 1g for 1 i 4. But as

.n C m/nCmC1 n1 n
g 0 .Y / D Y .1  Y /m1 . Y/
nn mm nCm
it follows that also g.b/ 2 f0; 1; 1g for all zeroes b of g0 .Y /. With yQ WD g. .y//
we thus find S .y/ Q  IP1 .Q/, and by an argument as above jS .y/j Q < jS .y/j.
Consequently, descending induction on jS .y/j yields a function t 2 Q.y/ N with
S .t/  f0; 1; 1g. t
u

Remark. When replacing the function t by 4t.1t/ we obtain an extension ramified

in three points, with one of the ramification orders equal to 2. Such functions are
called clean Belyi functions (see Shabat and Voevodsky (1990), 0.2.2).

Proposition 2.8 allows the following characterization of algebraic function fields

N
definable over Q:

Theorem 2.9 (Belyi (1979)). An algebraic function field in one variable L=k of
characteristic zero possesses a nonsingular model over QN if and only if L contains
a rational subfield k.t/ such that no more than three prime divisors in IP.k.t/=k/
ramify in L=k.t/.
N then the existence of
Proof. If L=k has a model (i.e., a defining equation) over Q,
k.t/ L with jS .t/j 3 follows directly from Proposition 2.8. For the proof of
the reverse direction we may assume that k is algebraically closed and that S .t/
is contained in f0; 1; 1g. A model of L=k involves only finitely many coefficients,
so we may further assume that the transcendence degree of k=Q N is finite. Hence k
may be identified with an algebraically closed subfield of the complex numbers C.
Then the field CL obtained by extension of constants is a subfield of the maximal
algebraic extension MS =C.t/ unramified outside the support of t, t  1. By the The-
orem 2.2 of Grothendieck (with kN D Q) N LN WD CL \ MN S is a geometric extension
N N N
field of Q.t/ with CL D CL and k L D L. Now the canonical model of L= N
N Q.t/
N
constitutes a nonsingular model for L=k.t/ with coefficients in Q. t
u

As a second consequence of Proposition 2.8 we obtain:

Theorem 2.10 (Belyi (1979)). The action of Q on s in the arithmetic fundamental

group Gal.MN S =Q.t// gives for s  3 a faithful representation of Q into the group
of (continuous) outer automorphisms of s :

Rs W Q ! Out. s /: (2.10)
18 I The Rigidity Method

Proof. By Corollary 2.5 the representation Rs is given via the action on s of the
decomposition group of a prime divisor of degree one unramified in MN S =Q.t/. Now
let kQ Q Q
N be the fixed field of ker.Rs / and denote Q WD Gal.MN S =k.t//. As the center
of s is trivial for s  3, we have that
QD s  CQ ;
Q
where CQ denotes the centralizer in Q of s . Hence the extension of MQ WD MN SC
Q is regular over kQ with Galois group isomorphic to s . The map LQ ! Q
over k.t/ N LQ
Q to intermediate fields of MN S =Q.t/
sending intermediate fields of MQ =k.t/ N yields an
isomorphism between the lattices of intermediate fields, hence every finite subex-
N
tension of MN S =Q.t/ Q
is defined by an equation with coefficients in k.
Given a 2 Q,N there exists an elliptic curve E with invariant j.E / D a and Weier-
straß model defined over Q.a/ (see for example Silverman (1986), Ch. II, Prop. 1.4).
As s  3 the corresponding function field EN WD Q.EN / may be interpreted by Propo-
sition 2.8 as an intermediate field of MN S =Q.t/.
N By the above arguments, E= N Q.t/
N
possesses a model with coefficients in k, Q which shows a 2 k. Q This forces kQ D Q, N
and the kernel of Rs is trivial. t
u
3 Fields of Definition of Galois Extensions 19

3 Fields of Definition of Galois Extensions

In this paragraph we introduce and characterize fields of definition of field exten-

sions with additional structure of automorphisms. The resulting theorems are fun-
damental for the rationality criteria in this chapter and in Chapter III.

3.1 Cyclic and Projective Descent

For the moment, let NQ =KQ be an arbitrary field extension and GQ a subgroup of
Aut.NQ =K/.
Q A subfield K of KQ is called a field of definition of NQ =KQ if there exists
Q
a field extension N=K, linearly disjoint from K=K, with N KQ D NQ . If moreover N
Q
is G-invariant, Q or of NQ = Q KQ
then K is called a field of definition of NQ =KQ with G, G
Q
for short. In particular in the latter case G WD GjN is a subgroup of Aut.N=K/
isomorphic to G.Q
With these notations we have the general result:
Proposition 3.1. Let NQ =KQ be a Galois extension with group G and K a field of
definition of NQ =KQ over which KQ is Galois. Then also NQ =K is Galois, and the
Galois group satisfies

Gal.NQ =K/ Š Gal.NQ =K/ Q

Q Gal.K=K/: (3.1)

If K is even a field of definition of NQ =G K, Q we have the stronger direct product

decomposition
Gal.NQ =K/ Š Gal.NQ =K/ Q  Gal.K=K/:
Q (3.2)
Proof. Let N=K denote the field extension linearly disjoint from K=K Q with N KQ D
NQ , which exists by the definition. The automorphisms of K=K Q may be lifted
uniquely to automorphisms of NQ =N , so NQ =N is Galois with Gal.NQ =N / Š
Q
Gal.K=K/. By assumption the fixed field of the group hGal.NQ =K/; Q Gal.NQ =N /i
coincides with K, so NQ =K is a Galois extension. As Gal.NQ =K/ Q \ Gal.NQ =N / D
1, the group Gal.NQ =N / yields a complement of Gal.NQ =K/. Q This proves (3.1).
If moreover N=K is Galois, this complement becomes a normal subgroup in
Gal.NQ =K/, and (3.2) follows. t
u
Q
If the field extension K=K is cyclic, we have the following weak converse of
Proposition 3.1:
Theorem 3.2 (Dew (1992)). Let NQ =KQ be a Galois extension with group G and
K a subfield of K. Q Moreover assume that there exists a field extension K  =KQ of
degree exp.G/ linearly disjoint from NQ =KQ and cyclic over K. Then for the compos-
ite N  WD K  NQ (inside an algebraic closure of K) Q and G  WD Gal.N  =K  / Š G
we have
(a) If NQ =K is Galois, then K is a field of definition of N  =K  .
(b) If moreover all automorphisms of NQ =K act as inner automorphisms on G,
then K is a field of definition of N  =G  K  .
20 I The Rigidity Method

Proof. By assumption, each monomorphism from N  into an algebraic closure of

KQ fixing K maps NQ , and hence also N  , onto itself. Consequently N  =K is a
Galois extension, with group WD Gal.N  =K/. Now
Q Š Gal.NQ =K/
Gal.N  =K/ Q  Gal.K  =K/;
Q (3.3)

so Gal.N  =K/Q has exponent n WD exp.G/, and the exponent exp. / is bounded
by m WD n  ŒKQ W K . Now let ı denote a generator of the cyclic group  WD
Gal.K  =K/ Š Zm , and ıQ an extension of ı to N  . Then ıQ has order m, and hence
generates a complement Q of G  in . The fixed field N of Q thus yields a field
extension linearly disjoint from K  =K, satisfying K  N D N  . This proves (a).
If all elements of Gal.NQ =K/ act as inner automorphisms on G, the same holds
for all elements of in its action on G  . Hence is generated by G  and C .G  /.
The fixed field LQ of C .G  / is an extension of K linearly disjoint from K  =K. As
LQ NQ by (3.3) it follows that L WD KQ LQ NQ with Gal.NQ =L/ D Z .G/. Applica-
tion of (a) to the Galois extension NQ =L with subfield LQ and L0 WD K  L yields a
complement Q to Z .G  / in C .G  /. Hence we have D Q  G  , the fixed field
N of Q is Galois over KQ with group Gal.N=K/ Š G  , and we have K  N D N  ,
which proves (b). t
u

Remark. If in the previous theorem jGj and ŒKQ W K are prime to each other, then K
is even a field of definition of NQ =K, Q Q D 1).
Q NQ =G KQ respectively (since ıj
N

In the case of congruence function fields we obtain from Theorem 3.2 the fol-
lowing optimal result:

Corollary 3.3. Let K= N IF

N p be an algebraic function field in one variable over the
algebraic closure of IFp , and NN =KN a finite Galois extension with group G. Then we
have:
(a) A subfield K of KN with IFN p K D KN is a field of definition of NN =KN if and only
if NN =K is Galois.
(b) This subfield is a field of definition of NN =G KN precisely if in addition all ele-
ments of Gal.NN =K/ act as inner automorphisms on G.

Proof. By Proposition 3.1 it remains to prove that K is a field of definition of NN =KN

(resp. NN =G K).
N For this, we may assume without loss of generality that K has a
finite field of constants. First we note that there exists at least a field of definition
KQ of NN =G KN with ŒKQ W K < 1. Since KQ has cyclic extensions of any finite order,
we may now apply Theorem 3.2(a) (resp. (b)) to the corresponding Galois extension
NQ =KQ and the subfield K, and thus obtain the assertion of the corollary. t
u

Corollary 3.3 can also be regarded as a special case of the following theorem
concerning descent with a projective profinite group (see Section IV.1.5 for further
information):

Theorem 3.4. Let K=N kN be an algebraic function field over a separably closed field
N N N
k and N =K a finite Galois extension with group G. Further let K be a subfield of
3 Fields of Definition of Galois Extensions 21

N D KN such that Gal.K=K/

KN with kK N is a projective profinite group. The we have:
(a) K is a field of definition of N =K if and only if NN =K is Galois.
N N
(b) K is a field of definition of NN =G KN if and only if in addition all elements of
Gal.NN =K/ act as inner automorphisms on G.
Proof. By Proposition 3.1 we only have to show the if parts of (a) and (b). Since
NN =K is Galois with group WD Gal.NN =K/ and with  WD Gal.K=K/ N we get the
exact sequence
1 ! G ! !  ! 1: (3.4)
By assumption  is a projective profinite group. Therefore (3.4) splits (see for
example Fried and Jarden (1986), Remark 20.11) and G has a closed complement
Q inside . The fixed field NQ of Q is an extension of K which is regular over
k WD kN \ K and with kN NQ D NN . This proves (a).
Now assume that in addition each element of acts as inner automorphism on
G. Then the fixed field L of the centralizer C .G/ is an extension of K regular
N
over k. Further LN WD kL NN is a subfield of NN with Gal.NN =L/
N D Z .G/. Appli-
N N
cation of (a) to the Galois extension N =L and with the subfield L yields a closed
complement Q of Z .G/ inside C .G/. Hence we have D Q  G, and the fixed
field N of Q is Galois over K with group Gal.N=K/ Š G and satisfies kN N D NN ,
proving (b). t
u
Remark. The proof of Theorem 3.4 shows that more generally in the case that
acts by inner automorphisms on G the obstruction for K to be a field of definition
of NN =G KN lies inside the cohomology group H 2 .; Z .G// with trivial action of 
on Z .G/ (compare also Belyi (1979)).

3.2 Fields of Definition of Geometric Field Extensions

Optimal results as in Theorem 3.4 are not available in general. We therefore first
study fields of definition of geometric Galois extensions while neglecting the Galois
group.
Theorem 3.5. Let K= N kN be an algebraic function field in one variable with alge-
braically closed field of constants of characteristic zero, and NN =KN a finite Galois
N D KN is a field of definition for
extension. Then a disclosed subfield K of KN with kK
N N N
N =K if and only if N =K is Galois.
Proof. By Proposition 3.1 it remains to prove that a disclosed subfield K of KN with
field of constants k, over which NN is Galois, is a field of definition for NN =K.
N By
assumption there exists a prime divisor P 2 IP.K=k/ of degree one. For an arbitrary
extension P N let LN denote the inertia field of P=P
N 2 IP.NN =k/ N containing K,N and
N N N
Q WD PjLN . As Q=P is unramified, each prime element ´ 2 K for P is also a prime
element for QN in L. N
N Consequently the field of formal power series k..´// gives the
N N
completion of L with respect to Q. Denote by e the ramification index of P= N QN
22 I The Rigidity Method

N
(hence of P=P). N
N is obtained as k..y//
Then the completion of NN with respect to P
with y D ´. The automorphisms ı in
e

N
 WD Gal.K=K/ N
Š Gal.k=k/
N
act on the formal power series f .´/ 2 kŒŒ´ via the coefficients. Hence among the
N
extensions of ı to k..y// there exists precisely one, which we call ı, O satisfying
O
y ı D y. The set of restrictions to NN of these automorphisms

Q WD fıQ j ıQ WD ıj
O N ; ı 2 g
N

is a closed subgroup of Gal.NN =K/ isomorphic to , which moreover forms a com-

plement to Gal.NN =K/.N Hence the fixed field N of Q is a geometric extension field
of K with KN D N . We conclude that K is a field of definition of NN =K.
N N N t
u
Remark. The complement Q constructed in the proof is obviously contained in the
N
decomposition group of P=P N
and it coincides with it in the case that P=P is unram-
ified. The proof shows that in the latter case the assumption of characteristic zero is
superfluous.
The theorem has the following immediate consequence:
Corollary 3.6. If  is a characteristic open subgroup of the Galois group s D
Gal.MN S =Q.t//
N in (2.5), and NN denotes its fixed field, then NN =Q.t/
N is already defined
over Q.t/.
N
Proof. By the Splitting Theorem 2.4 each ı 2 Gal.Q.t/=Q.t// may be lifted unique-
N
ly to an automorphism ıN 2 Gal.MN S =Q.t// in a given complement of s . As  ı D  ,
ıN induces an automorphism ıQ of NN =Q.t/ extending ı, and different extensions of ı
to NN differ only by an automorphism of NN =Q.t/.
N Hence NN =Q.t/ is Galois, and the
assertion follows immediately from Theorem 3.5. t
u
Example 3.1. If in the previous corollary we have s D 2, then 2 Š ZZ O is procyclic,
and each normal subgroup of 2 is characteristic. In particular for the subfields NN
of MN S =Q.t/
N with Gal.NN =Q.t//
N Š Zn (the cyclic group of order n), the extensions
N N
N =Q.t/ are already defined over Q.t/. t
u
Example 3.2. If in Corollary 3.6 the number of generators satisfies s  3, and if we
take as  the intersection of all kernels of homomorphisms from s onto a given
finite simple group G, then we obtain: For each finite simple group G there exists
N
an integer n 2 IN and a Galois extension NN =Q.t/ N
with Gal.NN =Q.t// Š G n which is
defined over Q.t/ (not necessarily as Galois extension). t
u

3.3 Fields of Definition of Geometric Galois Extensions

The general result of Theorem 3.5 has the disadvantage that the fields of defini-
tion characterized there usually do not form fields of definition for the field exten-
3 Fields of Definition of Galois Extensions 23

sion together with the Galois group. To obtain this stronger conclusion, additional
assumptions have to be made, as the following proposition shows.

Proposition 3.7. Let NN =KN be the Galois extension in Theorem 3.5 with Galois
group G and (not necessarily disclosed) field of definition K. Then K is also a field
of definition of NN =G KN precisely if each element of WD Gal.NN =K/ acts as inner
automorphism on G and the center of G has a closed complement in the centralizer
C .G/.
N then by Proposition 3.1 the group G
Proof. If K is a field of definition for NN =G K,
possesses a closed complement Q in which acts trivially on G. This is at the same
time a closed complement for the center Z .G/ of G in C .G/ D Q  Z .G/.
For the proof of the other direction we first note that G together with C .G/
generates all of . Then a closed complement Q of Z .G/ in C .G/ also is a
direct complement of G in , hence we have D Q  G. The fixed field N of Q
now yields a geometric Galois extension N=K with KN N D NN , and the assertion is
proved. t
u

Remark. The additional assumption on the complement in Proposition 3.7 will

clearly always be satisfied if for example G is abelian, or if G has trivial center.

They are also guaranteed for example in the following case:

Corollary 3.8. The field of definition K of NN =KN in Theorem 3.5 is also a field of
definition of NN =G KN if each element of acts as inner automorphism on G and the
fixed field of C .G/ is disclosed.

Proof. Under these assumptions the fixed field L of C .G/ is a disclosed geomet-
ric extension field of K. Hence by Theorem 3.5 the group Gal.NN =kL/ N D Z .G/
possesses a closed complement in Gal.NN =L/ D C .G/. t
u

The characterizing condition for fields of definition of Galois extensions NN =G KN

in Proposition 3.7 is usually quite hard to check for arbitrary groups, so we give here
a sufficient condition for its occurrence.

Theorem 3.9. Let NN =KN be the Galois extension of Theorem 3.5 with Galois group
G and field of definition K. Moreover assume that each element of Gal.NN =K/ acts
as an inner automorphism on G and that there exists a prime divisor P 2 IP.K=k/
of degree one satisfying the following normalizer condition
(N): The center of G possesses a complement in the normalizer NG .GI / in G of
N above P.
N 2 IP.NN =k/
the inertia group GI of a prime divisor P
Then K is also a field of definition for NN =G K.
N

Proof. By the Remark following Theorem 3.5 the group G possesses a closed com-
plement Q in N
WD Gal.NN =K/ contained in the decomposition group of P=P.
Q N Q N
Its fixed field N is a geometric extension field of K with K N D N . Further let
24 I The Rigidity Method

Q  WD Q \ C .G/ with fixed field NQ  , and K  be the fixed field of hQ  ; Gi. Then
NQ  is the Galois closure of NQ =K and a geometric Galois extension of K  with
group G  Š G. Moreover H WD Gal.NQ  =K/ is isomorphic to a subgroup of the
holomorph G Aut.G/ containing G.

D̃ N̄
D̃ ∗
Ñ ∗
Ñ Z (G)
N
Z (G∗) G
G∗
C∗
K̄
K∗
K
Fig. 3.1 Galois descent using (N)

By construction each element ıQ 2 Q acts on G as an element from NG .GI /.

Hence each element of Gal.NQ  =NQ / acts on G  as an element from NG  .GI / Š
NG .GI /. By a theorem of Jordan (see for example Zassenhaus (1958), Ch. II, 4,
Thm. 5), C  WD CH .G  / is isomorphic to a subgroup U of NG .GI / containing
Z .G/. By assumption NG .GI / and hence U contains a complement to Z .G/.
Hence Z .G  / also possesses a complement in C  . The fixed field N of this com-
plement now constitutes a geometric and Galois extension field of K with

Gal.N=K/ Š G N D NN :
and KN t
u

If we allow a cyclotomic extension of constants, then Theorem 3.9 has the fol-
lowing useful variant:
Corollary 3.10. Assume that instead of (N) the prime divisor P in Theorem 3.9
satisfies the centralizer condition
(C): The center of G possesses a complement in the centralizer CG .GI /.
Then there exists a finite cyclotomic extension K 0 =K such that K 0 is a field of defi-
N
nition for NN =G K.

Proof. As in the proof of Theorem 3.9 let Q denote a closed complement to G in

WD Gal.NN =K/ contained in the decomposition group of P=P, N and NQ the corre-
Q 0 Q Q  Q
sponding fixed field. Moreover let  WD  \ C .GI /,  WD  \ C .G/, with
NQ 0 , NQ  the corresponding fixed fields, and K 0 , K  be the fixed fields of hQ 0 ; Gi
resp. hQ  ; Gi. Then NQ  is the Galois closure of NQ 0 =K 0 . Again H WD Gal.NQ  =K 0 /
is isomorphic to a subgroup of the holomorph G Aut.G/ containing G, and
G  WD Gal.NQ  =K  / Š G.
3 Fields of Definition of Galois Extensions 25

D˜
D˜  N̄
D˜ ∗
Ñ Ñ  Ñ ∗ Z (G)
N Z (G∗) G
C∗ G∗

K̄
K∗
K
K
Fig. 3.2 Galois descent using (C)

Replacing now NQ =K and NG .GI / in the proof of Theorem 3.9 by NQ 0 =K 0 and

CG .GI / resp., we get a geometric Galois extension field N 0 of K 0 with

G 0 WD Gal.N 0 =K 0 / Š G N 0 D NN :
and KN

By the choice of Q we see that = Q Q 0 becomes isomorphic to a subgroup of

NG .GI /=CG .GI /, hence K =K is a cyclic extension. As the group Q acts on the
0

cyclic group GI via the cyclotomic character, we indeed have that K 0 =K is gener-
ated by roots of unity, which yields the conclusion. t
u

Remark. Theorems 3.5 and 3.9, as well as Corollary 3.10, also hold in positive char-
acteristic with the same proof, if only K=k possesses a prime divisor of degree one
not wildly ramified in NN =K (satisfying the condition (N) resp. (C) if appropriate).
26 I The Rigidity Method

4 The Rigidity Property

In order to detect small fields of definition for Galois extensions NN =Q.t/

N we study
here the action of the absolute Galois group of Q on the subfields of MN S =Q.t/
N with
prescribed Galois group. Here the Hurwitz classification of such fields proves advan-
tageous. As the main result of the discussion we obtain the Basic Rigidity Theorem
and its variant for irrational ramification points.

4.1 The Hurwitz Classification

Our starting point is the arithmetic fundamental group in (2.5)

D Gal.MN S =Q.t// Š s  Q;

where s D h1 ; : : : ; s j 1    s D 1iOdenotes the Galois group of the maximal alge-

braic Galois extension MN S =Q.t/ N unramified outside S, where jSj D s. Each element
ı 2 Gal.Q.t/=Q.t// Š Q may be lifted uniquely to an automorphism ıQ 2 inside
N
a given closed complement of s , which then permutes the closed subgroups of s ,
hence by Galois correspondence also the intermediate fields of MN S =Q.t/. N Now let
 be an open normal subgroup of s with fixed field N and N

N
G WD Gal.NN =Q.t// D s =:

Q Q
Then both the normal subgroup  ı and the field NN ı are independent of the partic-
ular lifting ıQ of ı, i.e., of the given complement, and may hence be denoted by  ı ,
NN ı respectively. We thus obtain

NN ı D .MN S /ı D MN S :
ı
(4.1)

Now let  D .1 ; : : : ; s / 2 G s be the image of  D .1 ; : : : ; s / under the canonical

(continuous) homomorphism W s ! G. Then  is a generating system of G
satisfying the product relation 1    s D 1. We call such a system a generating
s-system of G. The set of all generating s-systems of G is denoted by

˙s .G/ WD f 2 G s j h i D G; 1    s D 1g: (4.2)

For each  2 ˙s .G/ there exists precisely one (continuous)  2 Hom. s ; G/ with
 ./ D  , the kernel of which constitutes a closed subgroup of s denoted by
ker. /. Then we have ker. / D ker./ for two generating s-systems  ;  2 ˙s .G/
if and only if there exists ˛ 2 Aut.G/ such that  ˛ D .
N
The set of all intermediate fields of MN S =Q.t/ with Galois group isomorphic to G
is now denoted by
N S .G/ WD fNN j Q.t/
N N NN MN S ; Gal.NN =Q.t//
N Š Gg: (4.3)
4 The Rigidity Property 27

The above considerations may now be collected as follows:

Theorem 4.1 (Hurwitz Classification). The fields NN 2 N N S .G/ are parameterized

by the generating s-system classes  Aut.G/ 2 ˙s .G/= Aut.G/; more precisely there
exists a bijection

N S .G/;
NS W ˙s .G/= Aut.G/ ! N  Aut.G/ 7! NN  WD MN S
ker. /
: (4.4)

The components i of the parameter  yield generators of inertia groups of prime

ideals of NN  =Q.t/
N over Pi 2 S via
N
' W G ! Gal.NN  =Q.t//; i 7! 1
 .i / ker. /: (4.5)

Remark. Clearly, NS may also be interpreted as a surjective map

N S .G/;
NS W ˙s .G/= Inn.G/ ! N Œ WD  Inn.G/ 7! NN  : (4.6)

The group Q respectively more precisely each complement QQ of s in iso-

morphic to Q now acts on Hom. s ; G/ from the right via

Hom. s ; G/ 
QQ ! Hom. s ; G/;
Q 7! :ıQ with . :ı/.
. ; ı/ Q / D .ı Q ıQ1 /:
(4.7)
With  :ıQ WD  :ıQ this induces an action of QQ on ˙s .G/ from the right. This
satisfies

˙s .G/  QQ ! ˙s .G/; Q 7!  :ıQ D  ıQ1 with  ıQ WD

. ; ı/  .
ıQ
/: (4.8)
Q Q Q 1
(The definition of  ı WD  . ı / instead of  . ı / allows an immediate transfer
of transformation formulae from  to  , see for example Paragraphs 6 and 7 and
Ch. III.3.) Taking into account that the image of Œ 2 ˙s .G/= Inn.G/ under ıQ 2 QQ
is independent of the particular lifting ıQ of ı 2 Q , we thus obtain a well-defined
action of Q on ˙s .G/= Inn.G/:

ı 1 Q1
˙s .G/= Inn.G/  Q ! ˙s .G/= Inn.G/; .Œ ; ı/ 7! Œ WD Œ ı : (4.9)

With these notations we may conclude:

N
Proposition 4.2. With the action of ı 2 Gal.Q.t/=Q.t// on ˙s .G/= Inn.G/ defined
in (4.9) we have
1
NN  ı D .NN  /ı : (4.10)
28 I The Rigidity Method

4.2 The Fixed Field of a Class of Generating Systems

N
In the action of on ˙s .G/= Inn.G/ introduced in (4.9),  WD Gal.Q.t/=Q.t// acts
on the classes of generating systems via the cyclotomic character. More precisely
we have:
N
Proposition 4.3. Let ı 2 Gal.Q.t/=Q.t//, Œ 2 ˙s .G/= Inn.G/, and denote by Ci ,
resp. Ciı , the conjugacy class of the i -th component of a representative in Œ ,
resp. Œ ı . Then we have
Ciı D C.ic.ı/
/ı
: (4.11)

Proof. For an extension ıQ of ı onto MN S , from (4.8) and Theorem 2.6 we have
Q ıQ c.ı/ c.ı/ c.ı/
Ciı D Œiı D Œ  .i / DŒ  ..i /ı / D Œ.i /ı D C.i /ı : t
u

If we assume that the prime divisors P 2 S are -invariant, hence defined over
Q.t/, then the conjugacy classes Ci WD Œi 2 Cl.G/ belonging to a generating s-
system  are mapped simultaneously onto the conjugacy classes of their c.ı/-th
c.ı/
power: Ciı D Ci . This furnishes a transitive permutation representation of  on
the set of primitive powers

C WD fCn j n 2 .ZZ=jGjZZ/ g (4.12)

of the vector C D .C1 ; : : : ; Cs / 2 Cl.G/s . Hence the kernel

C WD fı 2  j Cc.ı/ D Cg (4.13)

of this permutation representation is a closed subgroup of  with index

d.C/ WD jC jI (4.14)

this index will be called the irrationality degree of C.

Proposition 4.4. Let C D .C1 ; : : : ; Cs / 2 Cl.G/s be a class vector of G. Then the

N C is an abelian number field of degree
fixed field QC WD Q

ŒQC W Q D d.C/: (4.15)

It is generated over Q by the values of the complex irreducible characters of G on

the classes C1 ; : : : ; Cs :

QC D Q.f.Ci / j  2 Irr.G/; i D 1; : : : ; sg/: (4.16)

O  the factor group

Proof. Since c./ D ZZ

=C Š Gal.QC =Q/

4 The Rigidity Property 29

is abelian. Further, as  acts transitively on C via c, we have

ŒQC W Q D . W C / D d.C/:

Equality (4.16) results from the fact that for all conjugacy classes C of G

QC D Q.f.C / j  2 Irr.G/g/;

which follows from ı .C / D .C c.ı/ / (see Huppert (1967), Satz V.13.1). t
u

According to Proposition 4.4, a class vector C 2 Cl.G/s will be called rational

if d.C/ D 1 and hence QC D Q. For C D .C1 ; : : : ; Cs / 2 Cl.G/s let

˙.C/ WD f 2 ˙s .G/ j i 2 Ci g (4.17)

with ˙s .G/ as in (4.2), and denote by

l.C/ WD j˙.C/= Inn.G/j (4.18)

the number of generating s-system classes Πof G with components i 2 Ci . Fol-

lowing Thompson (1984a) a class vector C is called rigid if l.C/ D 1. It is called
rationally rigid if moreover C is rational.
As the stabilizer
 WD fı 2  j Œ ı D Œ g (4.19)
of Πunder the action (4.9) of   is a closed subgroup of with index at most
l.C/ in C , we conclude:

Theorem 4.5. Let G be a finite group and  a generating s-system belonging to

the class vector C. If the ramification locus S of NN  =Q.t/
N remains pointwise fixed
N
under  D Gal.Q.t/=Q.t//, then the fixed field K of  contains the cyclotomic
field QC , and we have
ŒK W QC .t/ l.C/: (4.20)
Thus we have K D QC .t/ if C is rigid, and if C is even rationally rigid it follows
that K D Q.t/.

4.3 The Basic Rigidity Theorem

The fixed field K of a class of generating s-systems Œ 2 ˙s .G/= Inn.G/ often

N
coincides with a field of definition for NN  =G Q.t/.

Proposition 4.6. The field NN  is Galois over K . Moreover the automorphisms of

NN  =K act as inner automorphisms on Gal.NN  =Q.t//.
N
30 I The Rigidity Method

N
Proof. Let ı 2 Gal.Q.t/=K Q N
 / with an extension ı onto MS . As Œ D Œ
ı
we have
by (4.10)
Q Q1
ker. /ı D ker. ı / D ker. /:
Hence ker. / is normal in Gal.MN S =K / and the fixed field NN  of ker. / is Galois
Q
over K . By assumption, two generating systems  and  ı differ by an inner auto-
Q
morphism of G. Hence there exists a  2 G with  ı D  . So for the images
N N
'.i / WD ' .i / of i in Gal.N =Q.t// we obtain by (4.5) and (4.8)
Q 1 ıQ 1 ıQ
'. /ı D  . / ker. / D  . / ker. / D '. / D '. /'. / :

Thus ıQ maps the generating system '. / of Gal.NN  =Q.t//

N onto its conjugate
'. / '. /
and therefore acts as inner automorphism on Gal.NN  =Q.t//.
N t
u
N
Remark. According to the definition, K is the smallest subfield of Q.t/=Q.t/ for
which the assertions of Proposition 4.6 hold. It is sometimes called field of moduli.

Application of Theorem 3.9 and Proposition 4.6 now yields:

Theorem 4.7. Let G be a finite group in which the center has a complement, and let
 be a generating s-system of G. Then the fixed field K of Πis a field of definition
N
of NN  =G Q.t/.

Proof. By Proposition 4.6, the extension NN  =K is Galois and each automorphism
of NN  =K acts as an inner automorphism on G D Gal.NN  =Q.t//. N Now let k be
the algebraic closure of Q in K and P 2 IP.k .t/=k / a prime divisor of degree
one unramified in NN  =k .t/. As the inertia subgroups GI of prime divisors P N 2
N N
IP.N =Q/ above P are trivial, Z .G/ possesses a complement in NG .GI / D G.
Hence P satisfies the normalizer condition (N) of Theorem 3.9, and k .t/ is a field
N
of definition for NN  =G Q.t/. In particular there exists a geometric Galois extension
N =k .t/ with
Gal.N =k .t// Š G and QN N  D NN  : t
u

In the case that l.C/ D 1 this specializes to the Basic Rigidity Theorem:

Theorem 4.8 (Basic Rigidity Theorem). Let G be a finite group in which the center
has a complement, and C 2 Cl.G/s a rigid class vector of G. Then for any arbitrar-
ily chosen set S of s prime divisors Pi 2 IP.QC .t/=QC / of degree one there exists a
Galois extension N=QC .t/ unramified outside S with

Gal.N=QC .t// Š G

such that the inertia groups over the Pi are generated by elements i 2 Ci .
If the class vector is rationally rigid, we have QC D Q.

Proof. By assumption C is rigid, so there exists a single class of generating s-

systems Œ of G in ˙.C/= Inn.G/. By the Hurwitz classification (4.4) there exists
N
a Galois extension NN  =Q.t/ with group G, unramified outside S. As l.C/ D 1,
4 The Rigidity Property 31

Πremains fixed under C , and we have K D QC .t/ by Theorem 4.5. By Theo-

N
rem 4.7, QC .t/ is a field of definition for NN  =G Q.t/, hence there exists a geometric
Galois extension N=QC .t/ with

Gal.N=QC .t// Š G N D NN  :
and QN

The additional assertion on the ramification of NN  =Q.t/

N and hence also N=QC .t/ is
contained in Theorem 4.1. Finally, by Proposition 4.4 we have QC D Q if and only
if C is rational. t
u

Remark. In the proof of Theorem 4.7, the existence of a complement to the cen-
ter was only used to deduce the existence of a prime divisor P 2 IP.k .t/=k / of
degree one satisfying the normalizer condition (N). Theorems 4.7 and 4.8 hence
remain valid under this weaker assumption, with the same proof.

4.4 Choice of Ramification Points

Under a suitable choice of the ramification points it may be possible to obtain geo-
metric Galois extensions over Q.t/ even if the corresponding class vector C is not
rational. Explicitly checkable criteria for this can be found. For their formulation,
we need some more notation.
For C 2 Cl.G/s let

Sym.C/ WD f! 2 Ss j C! 2 C g (4.21)

with .C1 ; : : : ; Cs /! WD .C1! ; : : : ; Cs ! / be the full symmetry group of C and V

Sym.C/ a symmetry group of C. For such a V let

CV WD fC! j ! 2 V g  C : (4.22)

Furthermore,
d V .C/ WD jC j=jCV j (4.23)
is called the V -symmetrized irrationality degree of C. According to our definitions
we have d V .C/ D 1 precisely when CV D C . The class vector C is then called
V -symmetric.
In analogy to (4.13) we now write

VC WD fı 2  j Cc.ı/ 2 CV g (4.24)

for the setwise stabilizer of CV in  under its action by exponentiation via the
cyclotomic character. The above definitions and Proposition 4.4 now yield
32 I The Rigidity Method
V
Proposition 4.9. The fixed field QVC WD QN C of V is an abelian number field con-
C
tained in QC , of degree
ŒQVC W Q D d V .C/: (4.25)
In particular we have QVC D Q if and only if the class vector C is V -symmetric.

Under the assumption S D S the group  possesses the permutation represen-

tation  
1 ::: s
S W  ! Ss ; ı 7! (4.26)
.1/ı : : : .s/ı
in the symmetric group Ss . By definition, Cı then belongs to C if and only if S .ı/
lies in Sym.C/. So no information on the position of the ramification points is lost
if we restrict ourselves to S ./ Sym.C/.

Theorem 4.10. Let G be a finite group and  a generating s-system belonging to

N
the class vector C. Further assume that the action (4.26) of  D Gal.Q.t/=Q.t//
N N
on the ramification locus S of N =Q.t/ satisfies

V WD S ./ Sym.C/:

Then the fixed field K of Πcontains the cyclotomic field QVC . Moreover if VC acts
1
via  WD S inversely to the cyclotomic character c on CV , i.e., if C.ı/ D Cc.ı/ ,
then
ŒK W QVC .t/ l.C/: (4.27)

Proof. By Proposition 4.3 we have Ciı D C.ic.ı/

/ı
D Ci and hence .C.ı/ /c.ı/ D C for
all ı 2  . As C.ı/ 2 CV we also have Cc.ı/ 2 CV , which entails ı 2 VC . So we
certainly have K  QVC .
1
From C.ı/ D Cc.ı/ we conclude that Cı D C for all ı 2 VC . Hence the index
.VC W  / is bounded by the cardinality l.C/ of the orbit space ˙.C/= Inn.G/. This
finally implies (4.27). t
u

As a special case of the above result we obtain the following stronger version of
the Basic Rigidity Theorem:

Theorem 4.11 (Strong Rigidity Theorem). Let G be a finite group whose center
possesses a complement and with a rigid class vector C 2 Cl.G/s . Furthermore let
V be a symmetry group of C with the property that for each ı 2 VC there exists pre-
cisely one ! 2 V with Cc.ı/ D C! . Then there exists a geometric Galois extension
N=QVC .t/ with
Gal.N=QVC .t// Š G: (4.28)
If moreover C is V -symmetric, then we have QVC D Q.
4 The Rigidity Property 33

Proof. The set of classes fC1 ; : : : ; Cs g decomposes into orbits under the action of
the symmetry group V . For each such orbit B let
c.ı/
B WD fı 2 VC j Ci D Ci for all Ci 2 Bg:

From (4.13) we know that B  C , and B =C acts regularly on B. Hence the
fixed field QB WD Q N B is contained in QC and therefore by Proposition 4.4 an abelian
number field with ŒQB W QVC D jBj. Now for an arbitrary class Ci 2 B we choose
a primitive element ai of QB =QVC . For ı 2 VC and the element ! 2 V uniquely
1
determined by ı we define P.i /! to be the numerator divisor of .t  aiı /. In this
way we obtain prime divisors Pj 2 IP.Q.t/= N N for all indices j contained in the V -
Q/
orbit of i . Collect these (without loss of generality pairwise distinct) prime divisors
in S D fP1 ; : : : ; Ps g. Then by construction the permutation representation S from
(4.26) satisfies S .ı/ D ! 1 , and hence we have
1
Cı D .C! /c.ı/ D C for all ı 2 VC : (4.29)

Now let Œ denote the unique generating s-system class in ˙.C/= Inn.G/, further
NN  the field in N
N S .G/ determined by the Hurwitz classification (4.4), and K the
fixed field of Π. Then K coincides with QVC .t/ by (4.29) and Theorem 4.10.
N
Hence by Theorem 3.9 the field QVC .t/ constitutes a field of definition of NN  =G Q.t/,
V
and there exists a geometric Galois extension N=QC .t/ with

Gal.N=QVC .t// Š G N D NN  :
and QN

Further by Proposition 4.9 we have QVC D Q precisely when the class vector C is
V -symmetric. u
t

Remark. As in the proof of the Basic Rigidity Theorem — which here corresponds
to the case V D 1 — the inertia groups over the ramification points Pi are generated
by the elements i 2 Ci . The assumptions on ! of Theorem 4.11 then translate to
the statement: Each ı 2 VC permutes the components of C D .C1 ; : : : ; Cs / via c.ı/
and of .P1 ; : : : ; Ps / via S in the same way, which of course is equivalent to (4.29).
34 I The Rigidity Method

5 Verification of Rigidity

In the first two sections the Basic Rigidity Theorem and its stronger variant are
applied to abelian and to symmetric groups. For these, the existence of rigid gener-
ating systems may be checked by direct computation. In the next two sections we
show how to find rigid class vectors using the character table or a suitable matrix
representation of the given group.

5.1 Geometric Galois Extensions over Q.t / with Abelian Groups

All class vectors C of abelian groups with ˙.C/ ¤ ; are automatically rigid, thus
these groups present the easiest case. In particular the cyclic groups Zn D hi obvi-
ously possess C D .C; C 1 /, where C D fg, as a natural rigid class vector (com-
pare Example 3.1). But C is rationally rigid only for n 2, since QC D Q. n /. If all
primitive powers of , resp. C , are collected into one class vector C D .C m j m 2
.ZZ=nZZ/ /, then we obtain a rigid class vector for any n  3, which under a suitable
choice of ramification points and symmetry group V yields QVC D Q and therefore
leads to geometric Galois extensions over Q.t/ with group Zn . This statement can
be generalized to arbitrary finite abelian groups as will be shown in the following
theorem.
For brevity, let us call a realization of a group H as Galois group Gal.N=K/ a
G-realization of H over k (in r variables), if it satisfies the property
(G): N=K is a geometric Galois extension with Galois group H over a rational
function field K=k (of transcendence degree r).
(In Serre (1992), 4.1, this property is called GalT in the case k D Q.)

Theorem 5.1. Every finite abelian group possesses a G-realization over Q.

Proof. Any finite abelian group G decomposes into a direct product of cyclic groups

G D Zn1      Znr ;

say, where we assume that the ni are in increasing order. For the first q factors
Zni D hi i with ni D 2 the pair .i ; i / forms a generating 2-system. For the
remaining factors with ni  3 the tuples .im j m 2 .ZZ=ni ZZ/ / clearly yield gener-
ating '.ni /-systems, where ' denotes the Euler '-function. Composing Pthese gen-
erating systems we obtain a generating s-system  of G with s D q C riD1 '.ni /.
As G is abelian, we trivially have l.C/ D 1 for the class vector C containing  .
Now by construction the components C1 ; : : : ; Cs of C consist of full orbits under
N
the exponentiation with c.ı/ for ı 2  D Gal.Q.t/=Q.t//, so exponentiation induces
a permutation representation  of  on the classes Ci (trivial on the first 2q classes
of involutions), and hence on f1; : : : ; sg with ker./ D C . For V WD im./ we then
have VC D  due to CV D C , and =C Š V . Hence for each ı 2  there exists
5 Verification of Rigidity 35

a unique ! 2 V satisfying Cc.ı/ D C! . Since moreover the center of an abelian

group possesses the trivial group as complement, we may apply the Strong Rigidity
Theorem 4.11 to obtain the desired conclusion. t
u
Remark. Since it is easy to obtain a G-realization of a direct product from G-
realizations of its factors (see also Chapter IV, Corollary 1.7), the assumption that
Z .G/ possesses a complement in G may from now on be replaced without loss of
generality by the assumption that Z .G/ D 1.

5.2 Geometric Galois Extensions over Q.t / with Sn and An

A further example in which the rigidity of class vectors may be shown purely com-
binatorially is given by the groups Sn . Therefore, let 2A, .n  1/A and nA denote
the conjugacy classes of transpositions, .n  1/-cycles and n-cycles respectively, in
the symmetric group Sn on n letters. The results of this section rely on the simple
fact:
Proposition 5.2. The class vector C D .2A; .n  1/A; nA/ of Sn is rationally rigid
for n  2.
Proof. All elements of Sn with the same cycle shape are conjugate, so the class
vector C is rational. In case n D 2, C D .2A; 2A/ is a rigid class vector of S2 D Z2 .
For n  3 we have to show that C contains just one class of generating systems. Let
 D .1 ; 2 ; 3 / 2 ˙.C/ be such a system. By conjugation with elements from Sn
we may assume that 3 D .1 : : : n/. Moreover conjugating 1 D .i j / with a power
of 3 we get
j 1
3i 1 1 31i D .1 j +1–i /;
1j
3 1 3 D .1 i +1–j /;

so 1 can be transformed into the shape .1 k/ with 2 k n=2 C 1. But as

3 1 D .1 : : : n/.1 k/ D .1 : : : k  1/.k : : : n/

should lie in .n  1/A, it follows that necessarily k D 2. So each  2 ˙.C/ is con-

jugate to ..12/; .2 : : : n/1 ; .1 : : : n//. Moreover it is well known that a primitive sub-
group of Sn containing a transposition is already equal to Sn , so this finally shows
that ˙.C/ ¤ ;. t
u
Using the Basic Rigidity Theorem, Proposition 5.2 now yields the following
result, originally proved by Hilbert:
Theorem 5.3 (Hilbert (1892)). The groups Sn and An possess G-realizations
over Q.
Proof. For the symmetric groups Sn this follows with the Basic Rigidity Theo-
rem 4.8 immediately from Proposition 5.2.
Now let N=Q.t/ denote this Galois extension with Gal.N=Q.t// D Sn for C D
.2A; .n  1/A; nA/. Further, let K 0 be the fixed field of An . Either the .n  1/-cycles
36 I The Rigidity Method

or the n-cycles, and hence the inertia groups generated by them, lie already inside
the alternating group An . Thus, only two prime divisors are ramified in K 0 =Q.t/,
both of order two. Consequently the different D.K 0 =Q.t// has degree 2. By the
Hurwitz genus formula (see for example Lang (1982), Ch. I, Thm. 6.1) this implies
1
g.K 0 / D 1 C 2.g.Q.t/=Q/  1/ C deg.D.K 0 =Q.t// D 0:
2
Moreover the ramified prime divisors have degree one in K 0 =Q.t/, so K 0 =Q is a
rational function field, say K 0 D Q.t 0 / (see Artin (1967), Ch. 16, Thm. 7), and
N=Q.t 0 / is a geometric Galois extension with group An . u
t

A G-realization Gal.N=K/ over k for a group H with trivial center is called a

GA-realization of H over k if in addition it has property
(A): Gal.N=K/ may be embedded into a geometric Galois extension with group
Aut.H / (under identification of H with Inn.H /).
Such realizations will prove particularly useful for the solution of embedding prob-
lems with kernel H (compare Chapter IV.3). Since Aut.An / D Sn for n ¤ 6 we may
thus deduce from Theorem 5.3 the following result:

Corollary 5.4. The groups An possess GA-realizations over Q for n ¤ 6.

In general it is extremely difficult to prove rigidity of class vectors as above just

from the definition without using further information. In the next section we show
how to profit from the knowledge of structure constants, which may be computed
from the character table. This method has proved particularly useful for the sporadic
groups and the exceptional groups of Lie type (see Chapter II).

5.3 Structure Constants

We first enlarge the set ˙.C/ of generating s-systems of G in C to the set

˙N .C/ WD f 2 G s j i 2 Ci ; 1    s D 1g (5.1)

of not necessarily generating s-systems. The group G also acts on this set by conju-
gation in the components. The quotient
N
n.C/ WD j˙.C/j=j Inn.G/j (5.2)

constitutes an estimate for the number of orbits under this action; it will be called
the normalized structure constant of C.
5 Verification of Rigidity 37

Proposition 5.5. The normalized structure constant of a class vector C 2 Cl.G/s of

a finite group G is given by
X jZ .G/j
n.C/ D : (5.3)
jCG .h1 ; : : : ; s i/j
Œ 2˙N .C/= Inn.G/

Proof. The class equation for the action of G on ˙N .C/ gives

X
N
j˙.C/j D .G W CG .h1 ; : : : ; s i// :
Œ 2˙N .C/= Inn.G/

The assertion now follows with (5.2). t

u
From Proposition 5.5 we immediately get
Corollary 5.6. For a class vector C 2 Cl.G/s of a finite group G we have l.C/
N
n.C/. Moreover equality holds if and only if ˙.C/ D ˙.C/.
For later use we now split up the normalized structure constant into the contribu-
tion from the generated subgroups. This leads to the following sum formula:
Proposition 5.7. For the class vector C of a finite group G we have
X
n.C/ D n.CI H / (5.4)
ŒH WH G

with
.H W Z .H // X
n.CI H / D lH .D/: (5.5)
.NG .H / W Z .G// D C

Here the sum in (5.4) runs over the conjugacy classes of subgroups of G and in (5.5)
over the class vectors of H fusing into C in G.
Proof. By combining those summands in (5.3) for which h i is conjugate in G to
H G we obtain
X jZ .G/j
n.CI H / D :
jCG .H /j
N .C/= Inn.G/
Œ 2˙
Œh i DŒH

For a class vector D of H fusing into C precisely .NG .H / W CG .H //=j Inn.H /j

classes of generating systems  2 ˙.D/= Inn.H / of H fuse into one class  2
N
˙.C/= Inn.G/. This further implies
X X j Inn.H /jjZ .G/j
n.CI H / D
.NG .H / W CG .H //jCG .H /j
D C Œ 2˙.D/= Inn.H /

X .H W Z .H //
D lH .D/: t
u
.NG .H / W Z .G//
D C
38 I The Rigidity Method

The normalized structure constant of C may be computed directly from the values
of the complex irreducible characters of G. This offers the possibility of determining
the important invariant l.C/ from the character tables of G and its subgroups.

Theorem 5.8. Let C D .C1 ; : : : ; Cs / 2 Cl.G/s be a class vector of a finite group G,

where s  2. Then we have
X jGjs2 Y .i /
s
n.C/ D jZ .G/j ; i 2 Ci : (5.6)
.1/s2 jCG .i /j
2Irr.G/ i D1

Proof. For  2 Irr.G/ let R W G ! GLn .C/ denote a corresponding matrix repre-
sentation. By the Schur’s Lemma for each  2 G there exists an !./ 2 C satisfying

1 X ./
R.  / D !./In ; where !./ D ;
jGj .1/
2G

as follows from the evaluation of traces. Hence for all pairs .; / 2 G 2 we have

1 X ./
R.  / D R./:
jGj .1/
2G

Induction on s now yields

1 X .1 /    .s /
R.11    ss / D R./;
jGj s .1/s
2G
s

and evaluation of traces for  D 1 then leads to

1 X .1 /    .s /
.11    ss / D :
jGj s .1/s1
s 2G

Now let
1 X
 WD .1/
jGj
2Irr.G/

be the characteristic function of the identity in G. Accordingly, multiplying the pre-

vious equation by .1/jGjs1 and summing over  2 Irr.G/ we hence obtain
X X .1 /    .s /
m.C/ WD .11    ss / D jGjs1 :
.1/s2
2G s 2Irr.G/

Here m.C/ counts the number of solutions  2 G s of 11    ss D 1. The normal-
ized structure constant
1
n.C/ D jf 2 C j 1    s D 1gj
j Inn.G/j
5 Verification of Rigidity 39

may then be expressed as

m.C/ Y
s
n.C/ D jCG .i /j1 :
j Inn.G/j
i D1

So indeed we obtain (5.6). t

u
In particular Theorem 5.8 leads to the following frequently used criterion for
rigid class vectors:
Corollary 5.9. A class vector C 2 Cl.G/s of a finite group G is rigid if the following
two conditions are satisfied:
(1) G D h1 ; : : : ; s i for some i 2 Ci with 1    s D 1,
X .1 /    .s / jCG .1 /j    jCG .s /j
(2) D :
.1/ s2 jGjs2 jZ .G/j
2Irr.G/

Proof. From (2) we get n.C/ D 1, and then Condition (1) clearly implies l.C/ D 1
by Corollary 5.6. t
u
The following example is an application of Theorem 5.8. It shows that rigidity
for groups of reasonable size may easily be checked even by hand.
Example 5.1. Let G D L2 .8/ D SL2 .8/ and C D .9A; 9B; 9C / the class vector con-
sisting of the three classes containing elements of order 9, where 9B D .9A/2 and
9C D .9A/4 , say. Then from the character table of G in the group Atlas (Conway
et al. (1985)) one calculates

jGj X .1 /.2 /.3 /

n.C/ D
jCG .1 /j3 .1/
2Irr.G/
504 1 1 1 1 1
D .1 C C C C  C 0 C 0 C 0/ D 1:
93 7 7 7 7 8
It remains to show that condition (1) in Corollary 5.9 is satisfied. Assuming the
contrary, any triple  2 C with 1 2 3 D 1 would generate a proper subgroup of
G. The only maximal subgroups of G with order divisible by 9 are dihedral groups
D18 of order 18, hence we would have h i D Z9 . But this implies 2 2 f12 ; 17 g and
3 2 f14 ; 15 g, contradicting the product relation 1 2 3 D 1. Hence .9A; 9B; 9C /
is a rigid class vector of L2 .8/.
Now QC is the maximal real subfield of the field of ninth roots of unity. With the
choice V D h.123/i and a corresponding V -configuration S we get VC D  and
=C Š V , and in particular, V acts regularly on C . Thus for each ı 2  there
exists a unique ! 2 V with Cc.ı/ D C! . From the Strong Rigidity Theorem 4.11
now follows the existence of a G-realization of L2 .8/ over Q. t
u
In the next section, we introduce a criterion for rigidity making use of an irre-
ducible matrix representation of G. This has proved extremely helpful in the case of
classical groups (see Chapter II).
40 I The Rigidity Method

5.4 The Rigidity Criterion of Belyi

In this section
R W G ! GL.V / Š GLn .k/;  7! R./; (5.7)
denotes a faithful irreducible representation of the group G into the group of auto-
morphisms of an n-dimensional vector space V over an arbitrary field k. We identify
G with its image in GL.V / under the embedding R; so in particular we need not
distinguish between 1 2 G and IdV . We then have:

Theorem 5.10 (Belyi (1979)). Let G be a finite group, embedded into GL.V / Š
GLn .k/ via the irreducible representation R. Assume that there exist 1 ; 2 2 G
with h1 ; 2 i D G, and a 2 k  , such that 1  a1 has rank one. Then all gener-
ating 3-systems in ˙.Œ1 ; Œ2 ; Œ21 11 / are conjugate under Aut.G/. If more-
over NGL.V / .G/ D G  CGL.V / .G/, then the triple .Œ1 ; Œ2 ; Œ21 11 / of conju-
gacy classes in G is rigid.

Proof. Let Q 1 ; Q 2 2 G be a generating pair of G with ŒQ i D Œi , i D 1; 2, and

ŒQ 1 Q 2 D Œ1 2 . We have to show that there exists an element ˛ 2 Aut.G/ (resp.
˛ 2 G in the second case) such that Q i D i˛ for i D 1; 2. By conjugation with a
suitable element of G we may assume Q 2 D 2 , and then ˛ has to centralize 2 . Let
 WD 1 a1 and Q WD Q 1 a1 be the elements of rank one. As 1 2 and Q 1 Q 2 D Q 1 2
are conjugate in G, in the field of rational functions k.t/ we have

det.2 C a2 C t1/ D det.

Q 2 C a2 C t1/: (5.8)

Obviously in End.V ˝k k..t/// we have the identity

1
X
.a2 C t1/1 D .a2 /i 1 .t/i ;
i D0

where k..t// denotes the field of formal power series in t over k. Multiplying (5.8)
by det.a2 C t1/1 we further obtain

det.2 .a2 C t1/1 C 1/ D det.

Q 2 .a2 C t1/1 C 1/: (5.9)

Let 2 End.W / be an element of rank at most one, where W is an arbitrary vector

space over a field. Then consideration of the Jordan normal form immediately shows
that det. C 1/ D tr. / C 1. Since the rank of 2 .a2 C t1/ is bounded by the rank
of , which equals 1, this may be applied to (5.9), yielding
1
X 1
X
tr.a 2i .t/i / D tr.aQ 2i .t/i /;
i D0 i D0
5 Verification of Rigidity 41

and therefore by comparing coefficients

tr.2i / D tr.
Q 2i / (5.10)

for i < 0. But since 2 has finite order, (5.10) holds for all i .
We now decompose the endomorphism  of V of rank one into the surjection
1 W V ! k and the injection 2 W k ! V , so  D 2 ı 1 . Then kŒ2 2 .1/ is 2 - and
-invariant, hence also stabilized by 1 . Thus it forms a non-trivial G-invariant sub-
space of V , which must coincide with V by the irreducibility of G. If we decompose
Q D Q2 ı Q1 in a similar manner, we get kŒ2 Q2 .1/ D V by the same arguments. In
particular, there exists an automorphism ˛ of the kŒ2 -module V , hence an element
of Aut.V / centralizing 2 , with Q2 D 2˛ . Thus we get

tr.2i / D tr.2i / D tr.2i 2 1 / D tr..2i 2 1 /˛ / D tr.2i Q2 1˛ /;

and from (5.10) we finally obtain

tr.2i Q2 .1˛  Q1 // D 0

for all integers i . Since we already know kŒ2 Q2 .1/ D V , it follows that 1˛ D Q1 .
Hence  ˛ D ,Q 1˛ D Q 1 , so ˛ belongs to NGL.V / .G/ and is therefore the required
element. In the second case we may clearly assume that ˛ already lies in G. t
u

Remark. The proof shows that instead of NGL.V / .G/ D G  CGL.V / .G/ it suffices
to assume in Belyi’s rigidity criterion that all ˛ 2 NGL.V / .G/ fixing the conjugacy
classes Œ1 and Œ2 already belong to G  CGL.V / .G/.

As an application of Theorem 5.10 we obtain Hecke’s characterization of the

field of modular functions of level p (Hecke (1935)).

Example 5.2. It is well known that the special linear group G WD SL2 .p/, p ¤ 2, in
its natural matrix representation over IFp is generated by
   
11 1 1
1 D and 2 D :
01 1 0

(The corresponding matrices over ZZ even generate the modular group SL2 .ZZ/.)
Then by Theorem 5.10 all generating 3-systems in ˙.Œ1 ; Œ2 ; Œ21 11 / are con-
jugate under Aut.G/. Here H WD G  CGL.V / .G/ D G  Z .GL.V // has index 2
in NGL.V / .G/ D GL2 .p/. Thus it is still true that l.C/ D 1 for the class vector
C D .Œ1 ; Œ2 ; Œ21 11 /, since conjugation by ˛ 2 GL2 .p/nH interchanges the
two classes of elements of order p in SL2 .p/, so does not fix Œ1 .
Passing from SL2 .p/ to the factor group GN WD L2 .p/ we also obtain l.C/ N D1
N N
for the image ŒN 2 ˙.C/. Hence the class of generating systems N Aut.G/
is charac-
terized by the element orders p, 3 and 2 of N 1 , N 2 and N 21 N 11 respectively. Con-
sequently NN N =Q.t/
N is the field of modular functions of level p. The two classes
2A and 3A are easily seen to be rational, while the class pA is semirational, i.e., we
42 I The Rigidity Method
p
have pB D .pA/w for any primitive root w modulo p. Thus we have QC D Q. p  /
with p  WD .1/.p1/=2 p. The Basic Rigidity Theorem now proves that the field of
modular functions
p of level p with its group of automorphism L2 .p/ is defined over
the field Q. p  ; t/. t
u

The field extensions in the above example will reappear in Paragraph 7. Using
rational translates they will be shown to give rise to geometric L2 .p/-extensions
over Q.t/ for certain primes p.
6 Geometric Automorphisms 43

6 Geometric Automorphisms

In the rigidity criteria discussed until now, possible fields of definition for Galois
N
extensions are only sought among the intermediate fields of Q.t/=Q.t/. Smaller
fields of definition and hence better results can sometimes be reached by including
the group of geometric automorphisms Aut.Q.t/= N N This allows for all subfields
Q/.
N
of Q.t/ of transcendence degree 1 over Q to be taken into consideration. The action
of geometric automorphisms on the fundamental group can be described explicitly.
Thus a numerically verifiable criterion for the existence of fields of definition can be
obtained, the so called Twisted Rigidity Theorem. An application of it proves that
the small Mathieu groups occur as Galois groups over Q.t/.

6.1 Extension of the Algebraic Fundamental Group

We start again from the algebraic fundamental group

s
N
D Gal.MN S =Q.t// D h1 ; : : : ; s j 1    s D 1iO:

To extend this group by a group H of geometric automorphisms, we have to require

at least that H leaves the set S stable. Thus we may as well restrict ourselves to
considering the group
N
HS WD f 2 Aut.Q.t/= N j S D Sg:
Q/ (6.1)

Then HS defines a permutation representation

 
1 ::: s
S W HS ! Ss ;  7! ; (6.2)
.1/ : : : .s/

into the symmetric group Ss . For each subgroup V of S .HS / we obtain an inverse
image
HSV D f 2 HS j S ./ 2 V g: (6.3)
If moreover we assume s  3, which by Example 3.1 and Theorem 5.1 is possible
without loss of information, then we have:

Proposition 6.1. For s  3 the permutation representation S of HS into Ss is faith-

ful. In particular for V S .HS / we always have

HSV Š V: (6.4)

Proof. The assertions in Proposition 6.1 follow from the well known fact that
N
Aut.Q.t/= N Š PGL2 .Q/
Q/ N acts sharply threefold transitively on IP1 .Q/
N and hence
N N
on IP.Q.t/=Q/. t
u
44 I The Rigidity Method

Possible groups of geometric automorphisms HSV thus have to be among the

N The latter are well known from the classical literature:
finite subgroups of PGL2 .Q/.
N
Theorem 6.2 (Klein (1884)). Let H be a finite subgroup of Aut.Q.t/= N Š PGL2
Q/
N Then we have:
.Q/.
(a) H is a finite rotation group, i.e., one of

Zn ; Dn ; A4 ; S4 ; A5 .n 2 IN/: (6.5)
N H =Q
(b) At most three prime divisors of Q.t/ N are ramified in Q.t/=
N N H . The
Q.t/
corresponding triples of ramification indices are

.1; n; n/; .2; 2; n/; .2; 3; 3/; .2; 3; 4/; .2; 3; 5/: (6.6)
N
(c) The Galois extension Q.t/= N H is uniquely determined by the three rami-
Q.t/
fication points and the corresponding triple of ramification indices.
N
Proof. The Hurwitz genus formula for the field extension Q.t/= N H with the
Q.t/
ramification indices e1 ; : : : ; es forces

X
s
1
N
2.jH j  1/ D deg.D.Q.t/= N H // D jH j
Q.t/ .1  /:
ei
i D1

Interpreting this as a diophantine equation in e1 ; : : : ; es and jH j one first finds that

s 3 and then that .e1 ; e2 ; e3 I jH j/ can only be one of .1; n; nI n/, .2; 2; nI 2n/,
.2; 3; 3I 12/, .2; 3; 4I 24/, .2; 3; 5I 60/. Thus H is generated by a 3-system

.1 ; 2 ; 3 / with 1e1 D 2e2 D 3e3 D 1 and 1 2 3 D 1: (6.7)

From these generators and relations an easy calculation, for example with the Todd-
Coxeter algorithm, shows that H is one of the groups listed in (a), and moreover
that the relations in (6.7) already give a presentation for H .
For (c) it suffices by the Hurwitz classification to prove that in each of the possi-
ble groups H there exists precisely one class of generating 3-systems  Aut.H / of H
modulo Aut.H / with the corresponding element orders. The details for this straight-
forward calculation are given for example in Matzat (1987), III, 1.3. t
u

By the theorem of Lüroth the function field Q.t/N H =Q N is rational, say Q.t/
N HD
N Q
Q.t /. Consequently the above Galois extensions may be identified with Galois
extensions inside MN S =Q.
N tQ/. The following result contains generating s-systems, free
up to the product relation, for the corresponding open normal subgroups  of 3 .
N
Theorem 6.3. Let MN S =Q.t/ be the maximal algebraic Galois extension, unramified
outside S D fP1 ; P2 ; P3 g,

3 D Gal.MN S =Q.t//
N D h1 ; 2 ; 3 j 1 2 3 D 1iO;
6 Geometric Automorphisms 45

and NN V =Q.t/
N the rational intermediate field of MN S =Q.t/
N corresponding to the finite
rotation group V according to Theorem 6.2. Then for V WD Gal.MN S =NN V / we have:
 1  1n  1
.a/ Zn D h1 ; 1 2 ; : : : ; 1 2 ; 2n ; 3n j 1 1 2    3n D 1iO:
(6.8)
1 1 n1 1 1
.b/ D n D h12 ; .12 /2 1 ; : : : ; .12 /.2 1 / ; 22 ; .22 /2 3 2 ; : : : ;
(6.9)
1 1 1
3 2 /n1
.22 /.2 ; .3n /2 ; 3n j 12 .12 /2 1
   3n D 1iO
(c) For V D A4 ; S4 ; A5 the group V possesses a free generating s-system of
length s D 14; 26; 62 respectively.

Proof. First let V D Zn . The two prime divisors P2 , P3 both have ramification
order n, so V \ hi iO is generated by in for i D 2; 3. The prime divisor P1 com-
pletely decomposes in NN V =Q.t/.
N The inertia groups of prime divisors of P1 in NN V
are hence obtained by h1 iO, where  runs through a full system of representatives of
i
3 =V , for example  D 2 with i D 0; : : : ; n1. With this choice the product rela-
tion is obviously satisfied. Now V , being a subgroup of index n in a free group of
rank 2, is also free of rank r D n C 1, so no further relations exist between these gen-
 1
erators, and .1 ; 1 2 ; : : : ; 3n / constitutes a free generating .n C 2/-system of V .
In a completely similar way it is shown that (b) gives a free generating .2n C 2/-
system of V for V D Dn . The corresponding free generating s-systems for A4 , S4
and A5 are not reproduced here on account of their size. t
u

6.2 The Action of Geometric Automorphisms

If V S .HS / then by definition the geometric automorphisms  2 HSV extend to

N These act as outer automorphisms on s D Gal.MN S =Q.t//.
automorphisms of MN S =Q. N
Explicit formulae for the images of ΠD .1 ; : : : ; s / Inn.s /
under this action can
easily be derived from Theorem 6.3. This is achieved for s D 3 and s D 4 in this
section.
N
Theorem 6.4. Let MN S =Q.t/ denote the maximal algebraic Galois extension unram-
ified outside S D fP1 ; P2 ; P3 g with

3 D Gal.MN S =Q.t//
N D h1 ; 2 ; 3 j 1 2 3 D 1iO:

Then HS Š S3 is independent of the choice of S. For the preimages i 2 HS of

generating elements !2 D .12/ and !3 D .123/ of S3 we have:

Œ1 ; 2 ; 3 2
D Œ2 ; 1 ; 2 3 21 ; (6.10)

Œ1 ; 2 ; 3 3
D Œ2 ; 3 ; 1 : (6.11)
46 I The Rigidity Method

Proof. The isomorphism HS Š S3 follows from the threefold transitive action of

N
Aut.Q.t/= N on IP.Q.t/=
Q/ N N Now first let V D h!2 i, Q.
Q/. N tQ/ be the fixed field of H V
S
and
N tQ// D hQ1 ; Q2 ; Q3 j Q1 Q2 Q3 D 1iO
Q3 WD Gal.MN Q =Q.
S

be the Galois group of the maximal algebraic Galois extension unramified outside
PQ 1 WD P1 j N Q , P
Q 2 WD P3 j N Q and the second prime divisor P
Q 3 2 IP.Q.
N tQ/=Q/
N ram-
Q.t / Q.t /
N N N
ified in Q.t/=Q.tQ/. According to Theorem 6.3(a) Q.t/ is the fixed field of

Q V D hQ1 ; Q2 Q1 Q21 ; Q22 ; Q32 j .Q1 Q2 /2 Q32 D 1iO:

The prime divisor P Q 3 is unramified in MN S =Q.t/,

N hence we obtain 3 from Q V by
addition of the relation Q3 D 1:
2

3 Š hQ1 ; Q2 Q1 Q21 ; Q22 j .Q1 Q2 /2 D 1iO:

This allows us to identify 1 D Q1 , 2 D Q2 Q1 Q21 and 3 D Q22 , and then 2 is the
outer automorphism of 3 induced by Q3 D Q1 Q2 D Q21 Q11 . Hence we have

Q Q 2 Q11 1  1 11  1
; .Q22 /Q1

Π2
D ŒQ1 2 ; Q1 2 D Œ2 3 ; 1 3 ; 3 1 ;

which upon conjugation by 31 22 yields formula (6.10).

Now let V D h!3 i, Q.N tQ/ WD Q.t/
N HSV and Q3 D Gal.MN Q =Q. N tQ// the Galois group
S
of the maximal algebraic Galois extension unramified outside P Q 1 WD P1 j N Q and
Q.t /
the two prime divisors PQ 2 and PQ 3 ramified in Q.t/=
N N tQ/. Again by Theorem 6.3(a)
Q.
N
we have that Q.t/ is the fixed field of

Q V D hQ1 ; Q2 Q1 Q21 ; Q22 Q1 Q22 ; Q23 ; Q33 j .Q1 Q2 /3 Q33 D 1iO:

From this we obtain 3 by addition of the relations Q23 D 1 and Q33 D 1, hence

3 Š hQ1 ; Q2 Q1 Q21 ; Q22 Q1 Q22 j .Q1 Q2 /3 Q23 D 1iO:

With 1 D Q1 , 2 D Q2 Q1 Q21 , 3 D Q22 Q1 Q22 and 3 D Q21 we thus obtain

Π3
D ŒQ2 Q1 Q21 ; Q22 Q1 Q22 ; Q23 Q1 Q23 D Œ2 ; 3 ; 1 ;

proving (6.11). t
u

In the case s D 4 not all subgroups of the symmetric group S4 occur as images
of HS , since S has to consist of full HS -orbits, while for example the group S4 can
N
be seen to have only orbits of lengths 6; 8; 12 and 24 on IP.Q.t/= N The remaining
Q/.
possibilities for s D 4 are collected in the following theorem. Here the 4-cycle !4 is
chosen such that together with !2 it generates the dihedral group D4 of order 8.
6 Geometric Automorphisms 47

Theorem 6.5. Let MN S =Q.t/

N be the maximal Galois extension unramified outside
S D fP1 ; P2 ; P3 ; P4 g with

4 D Gal.MN S =Q.t//
N D h1 ; 2 ; 3 ; 4 j 1 2 3 4 D 1iO:

Then any subgroup V of S .HS / is conjugate in S4 to a subgroup of A4 D h!3 ; !42 i

or D4 D h!2 ; !4 i with !2 D .12/, !3 D .123/ and !4 D .1324/. For the preimages
i 2 HS of these generating elements !i we have, independently of the choice of S:
Œ1 ; 2 ; 3 ; 4 2
D Œ2 ; 3 1 31 ; 3 ; 11 4 1 ; (6.12)
Œ1 ; 2 ; 3 ; 4 3
D Œ2 ; 3 ; 1 ; 11 4 1 ; (6.13)
Œ1 ; 2 ; 3 ; 4 4
D Œ3 ; 11 4 1 ; 2 ; 3 1 31 : (6.14)

Proof. The proof is entirely analogous to the one for Theorem 6.4, so we restrict
ourselves to computing the example with 4 , which has not been treated in the
literature. Let V D h!4 i, Q. N tQ/ D Q.t/
N HSV and Q4 D Gal.MN Q =Q.
N tQ// be the Galois
S
group of the maximal algebraic Galois extension unramified outside P Q 1 WD P1 j N Q
Q.t /
Q Q N N
and the two prime divisors P2 , P3 ramified in Q.t/=Q.tQ/. Then by Theorem 6.3 the
N
field Q.t/ is the fixed field of

Q V D hQ1 ; Q2 Q1 Q21 ; Q22 Q1 Q22 ; Q23 Q1 Q23 ; Q24 ; Q34 j .Q1 Q2 /4 Q34 D 1iO:

The group 4 is obtained from this by addition of the two relations Q24 D 1 and
Q34 D 1, so we get

4 D hQ1 ; Q2 Q1 Q21 ; Q22 Q1 Q22 ; Q23 Q1 Q23 j .Q1 Q2 /4 Q24 D 1iO:

As !4 D .1324/, a good choice of generators is 1 D Q1 , 3 D Q2 Q1 Q21 , 31 2 3 D

Q22 Q1 Q22 and 4 D Q23 Q1 Q23 . With 4 D Q21 we obtain
 
Œ1 ; 3 ; 2 3 ; 4 4
D Œ3 ; 2 3 ; 4 ; 1 ;

and hence

Π4
D Œ3 ; 41 3 ; 23 ; 1 D Œ3 ; 11 4 1 ; 2 ; 3 1 31 : t
u

Variants of the proofs of Theorems 6.4 and 6.5, in which the images of embedded
homotopy classes of paths are determined topologically, are contained in Matzat
(1987), Ch. III, 1, for example.

6.3 Rigid Orbits

N we obtain analogously to (4.8) via

Extending  2 HS to Q 2 Aut.MN S =Q/
1
˙s .G/  HS ! ˙s .G/; Q 7!  :Q WD  Q
. ; / with  Q D  .
Q
/ (6.15)
48 I The Rigidity Method

an action of Q on ˙s .G/, respectively of  on ˙s .G/= Inn.G/. For this we obviously

have:

Proposition 6.6. Let C 2 Cl.G/s be a class vector of G and V a symmetry group of

C. Then via (6.15) HSV acts on ˙.CV /= Inn.G/. For the inverse action Œ 7! Œ ,
the formulae for the generating elements i 2 HSV carry over unchanged from Œ
to ΠD Π ./ .

In the case where HSV Š V this action now splits ˙.CV /= Inn.G/ into HSV -
orbits. Their number shall be denoted by l V .C/. This notation generalizes the pre-
viously introduced l.C/ D l 1 .C/ for the trivial symmetry group 1. The orbits may
differ by the type of their stabilizers

HV WD f 2 HSV j ΠD Πg: (6.16)

For U HSV we now define

HSV
lUV .C/ WD jfŒ j  2 ˙.C/; HV D U ˛ for an ˛ 2 Aut.HSV /gj (6.17)

to be the number of such orbits with stabilizer equal to U up to an automorphism of

HSV Š V . The class equation then yields:

Proposition 6.7. Let U be a system of representatives of subgroups of V modulo

Aut.V /. Then for each class vector C of G we have
X X
l V .C/ D lUV .C/; l.CV / D .V W U /lUV .C/: (6.18)
U 2U U 2U

V .C/ D 1 are characterized inside ˙.C /= Inn.G/ by their

V V
Those orbits with lH

stabilizer HV . These are of particular interest and will be called rigid HSV -orbits.
In the case where ˙.CV /= Inn.G/ consists of just one single HSV -orbit, C will also
be called a V -rigid class vector. The next theorem will show that this is a suitable
generalization of the rigidity property for a class vector defined in Section 4.2. For
N
this we extend  D Gal.Q.t/=Q.t// to

VS WD h; HSV i N

Aut.Q.t/=Q/; (6.19)

and denote the stabilizer of Œ 2 ˙s .G/= Inn.G/ in this group by

V WD fı 2 VS j Œ ı
D Πg (6.20)

analogously to (4.19).
N
For brevity we call a subset S  IP.Q.t/= N of cardinality s a V -configuration for
Q/
V Ss , if S ./ is contained in V and if moreover we have HSV Š V . According
to Theorems 6.4 and 6.5 such configurations exist for example in the case s D 3 for
all V S3 and in the case s D 4 for all V A4 and V D4 .
6 Geometric Automorphisms 49

Theorem 6.8. Let C 2 Cl.G/s be a class vector of the finite group G, V a symmetry
group of C and S a V -configuration. Furthermore, for  2 ˙.C/ let KV denote the
fixed field of V and kV the algebraic closure of Q in KV . Then we have

ŒkV W QVC V
lH V .C/: (6.21)


HSV
In particular KV =QVC is regular if Πis rigid.

Proof. According to S ./ V , for each ı 2 VS we have ! WD S .ı/ 2 V . Hence

1
for each ı 2 V we have Cc.ı/ D C! 2 CV since Cı D C, which implies ı 2
Q VC WD hVC ; HSV i, with VC D fı 2  j Cc.ı/ 2 CV g from (4.24).
Now H V is normal in Q V , so ı 2 Q V maps the class of generating systems Œ 2
S C C
1
˙.CV /= Inn.G/ onto Œ 2 ˙.CV /= Inn.G/, where moreover HVı D .HV /ı is
ı

the image of HV N Q VC we

under an automorphism of HSV . Hence with KQ CV WD Q.t/
have
ŒKV W KQ CV D .Q VC W V / .HSV W HV / lH
V
V .C/:


The extension KQ CV =QVC is regular, so as N V

QK N
D Q.t/ HV
this proves the estimate

ŒKV W KQ CV .Q VC W V /

ŒkV W QVC D D V
lH V .C/:
N V W Q
ŒQK N KQ V .HSV W HV / 
C

In the case of a rigid HSV -orbit lH

V V V
V .C/ equals 1, and so K =QC is regular. t
u


Hence if Œ belongs to a rigid HSV -orbit, the general estimate Œk W QVC l.C/
from Theorem 4.10 may be improved by the much better estimate ŒkV W QVC
V V
lH V .C/. But unfortunately the corresponding fixed field K is not always a ratio-

nal function field and therefore in general does not allow us to construct Galois
extensions over kV .t/ and kV .

6.4 The Twisted Rigidity Theorem

The rationality of KV =kV can be guaranteed under a relatively simple additional
hypothesis.

Proposition 6.9. The fixed field KV of V is a rational function field over kV if V
possesses an orbit of odd length in f1; : : : ; sg.
N
Proof. The elements ı 2 V can only permute prime divisors of Q.t/= QN lying in
the same HSV -orbit. Thus by the assumption, KV =kV possesses a prime divisor of
N
odd degree, and it has genus 0, being a subfield of Q.t/. Hence (see Artin (1967),
Ch. 16, Thm. 7) KV =kV is a rational function field. t
u
50 I The Rigidity Method

Remark. In the case HV ¤ 1 it suffices in Proposition 6.9 to assume that HSV
possesses an orbit of odd length on the set of prime divisors in S restricted to
N HV .
KN V WD Q.t/

From Theorem 6.8 we thus obtain the Twisted Rigidity Theorem in the following
formulation:

Theorem 6.10 (Twisted Rigidity Theorem). Let G be a finite group with trivial cen-
ter, C 2 Cl.G/s a class vector with s  3 and V a symmetry group of C having an
orbit of odd length and a V -configuration S. If ˙.CV /= Inn.G/ contains a rigid
HSV -orbit, then there exists a geometric Galois extension N=QVC .tQ/ with

Gal.N=QVC .tQ// Š G: (6.22)

If moreover the class vector C is V -symmetric, then we have QVC D Q.

Proof. Let Œ be an element in the rigid HSV -orbit in ˙.CV /= Inn.G/ which exists
by assumption. Then by Theorem 6.8 and Proposition 6.9 the fixed field KV is a
rational function field over QVC , say KV D QVC .tQ/. As in Proposition 4.6 we obtain
that NN  =KV is Galois, and each automorphism of NN  =KV acts as an inner automor-
N
phism on G D Gal.NN  =Q.t//. Now as Z .G/ D 1 it follows from Proposition 3.7
N
that KV D QVC .tQ/ constitutes a field of definition of NN  =G Q.t/. t
u

Remark. In the case HV D 1 extension of constants of N=QVC .tQ/ by Q N in the

N
Twisted Rigidity Theorem leads to the Galois extension NN  =Q.t/, while in the case
HV ¤ 1 we arrive at a Galois extension QN=N N tQ/ with QN
Q. N ¤ NN  . The translation
occurring in this case will be studied in detail in the next paragraph.

In the final section we apply the Twisted Rigidity Theorem to realize the small
Mathieu groups as Galois groups.

6.5 Geometric Galois Extensions over Q.t / with M12 and M11

The Mathieu group M12 possesses two conjugacy classes 4A and 4B of elements
of order 4, and one, denoted 10A, of elements of order 10. Here 4A denotes the
class of double 4-cycles in a given faithful permutation representation  of M12
of degree 12. With respect to this same permutation representation the elements of
order ten consist of the disjoint product of a 10-cycle and a transposition (Conway
et al. (1985)).

Proposition 6.11. The class vector C D .4A; 4A; 10A/ of M12 is rational, and with
V D h.12/i we have
l.C/ D 2 and l V .C/ D 1: (6.23)
So the class vector C is rationally V -rigid.
6 Geometric Automorphisms 51

Proof. The classes 4A and 10A of M12 are rational, so by definition C is rational
class vector (see the Atlas of Conway et al. (1985)). Furthermore, from the char-
acter table of M12 one calculates the normalized structure constant according to
Theorem 5.8
jGj X .1 /2 .3 /
n.C/ D
jCG .1 /j2 jCG .3 /j .1/
2Irr.G/
95040 9 1 4 4 1
D 2 .1   C C  / D 2:
32 10 11 11 54 66 99
Now let  2 ˙N .C/ and U WD h i. Then U contains elements of order 10, and accord-
ing to the list of maximal subgroups of M12 in the Group Atlas either we have
U D M12 or U is contained in one of the two maximal subgroups of M12 of type
M10 Z2 or Z2  S5 .
The group M12 contains two conjugacy classes of maximal subgroups M10 Z2 .
Those in the first class act intransitively in the permutation representation , with
orbits of lengths 10 and 2, while those in the second class act transitively. But the
relation 1 2 3 D 1 together with the permutation types .4/2 .1/4 and .10/.2/ of the
elements 1 ; 2 and 3 respectively is not compatible with an intransitive action of
type .10/.2/, so the first possibility can be excluded. Next the outer automorphism
group of M12 has order 2, and any non-trivial outer automorphism ˛ exchanges the
two classes of maximal subgroups M10 Z2 , as well as the conjugacy classes 4A and
4B. In the representation , the latter class consists of elements of type .4/2 .2/2 .
Hence by first applying ˛ we may again argue as above with the product relation and
the permutation types to exclude the second class of subgroups M10 Z2 as well.
If U were contained in Z2  S5 , then the projection p1 of U onto the first factor
Z2 Š hi would have to be surjective. Since p1 .3 / D  and p1 .1 /p1 .2 /p1 .3 / D
1 this would imply fp1 .1 /; p1 .2 /g D f1; g. In any case, 1 and 2 would lie in
different conjugacy classes of 4-elements in Z2  S5 . The permutation character for
such a subgroup shows that it intersects both classes 4A and 4B of 4-elements in
M12 . But Z2  S5 has just two classes of 4-elements, so one of these has to fuse into
4B. Thus by our above observation also one of 1 , 2 would have to lie in class 4B,
contradicting the choice of the class vector .4A; 4A; 10A/.
We have now proved that ˙N .C/ D ˙.C/, which together with Corollary 5.6
yields l.C/ D n.C/ D 2. Now assume that Œ 2 ˙.C/= Inn.G/ is a fixed point under
the generating element 2 2 HSV . Then by (6.10) there exists  2 M12 with  D
2 2
.2 ; 1 ; 2 3 21 /. Since this implies 1 D 1 and 2 D 2 , so that  centralizes
all of M12 , we conclude that  2 D 1. Setting 3 WD .2 /1 this would mean 32 D
.21  1 /2 D 21 11 D 3 , contradicting the fact that M12 contains no elements
of order 20. Hence the two classes of generating systems in ˙.C/= Inn.G/ lie in a
single HSV -orbit, and we have l V .C/ D 1. t
u

The argument for the non-trivial action given at the end of the proof is a special
case of the Fixed Point Theorem in Section 7.1. From the above proposition and the
52 I The Rigidity Method

Twisted Rigidity Theorem we already obtain the first half of

Theorem 6.12. The Mathieu groups M12 and M11 possess G-realizations over Q.

Proof. By Proposition 6.11 and Theorem 6.10 there exists a geometric Galois exten-
sion N=Q.tQ/ with Gal.N=Q.tQ// Š M12 belonging to the class vector .4A; 4A; 10A/.
The permutation representation  introduced at the beginning of this section is
induced by the permutation action on the cosets of an (intransitive) maximal sub-
group of type M11 . Denote the fixed field of such a subgroup by L. Then three prime
N
divisors Pi 2 IP.Q.t/= N are ramified in QL=
Q/ N Q.t/
N N
(where we have Q.t/ N tQ/).
D Q.
According to the cycle decompositions for generators of inertia groups 1 ; 2 2 4A
and 3 2 10A the ramification behavior of these is given by

Pi D Q4i;1 Q4i;2 Qi;3 Qi;4Qi;5 Qi;6 for i D 1; 2; P3 D Q10 2

3;1 Q3;2 : (6.24)

N Q.t//
(The underlying result will be shown in Section 9.1.) So the different D.QL= N
is of degree 22, and by the Hurwitz genus formula one calculates the genus

N Q/N D 1  ŒQL
N W Q.t/
N 1 N Q.t///
N
g.QL= C deg.D.QL= D 0:
2
Hence L=Q also has genus 0. Since moreover Q3;1 cannot split in QL=L, N its restric-
tion Q3;1 jL is a prime divisor of degree 1. This proves that L=Q is a rational function
field, and Gal.N=L/ provides a G-realization of the Mathieu group M11 . t
u

Remark. By (6.24) the class vector belonging to the geometric M11 -extension con-
structed above, according to the Hurwitz classification, consists of eight components
4A and one component 5A in M11 .

We shall return to the automorphism group Aut.M12 / and to the other sporadic
groups in Chapter II.9.
7 Rational Translates of Galois Extensions 53

7 Rational Translates of Galois Extensions

If a Galois extension NN =Q.t/

N is translated via the translation theorem of Galois the-
ory, then in general the corresponding class vector will change. For example, such
a translation can sometimes be employed to construct class vectors (resp. orbits)
which are rationally rigid with respect to a symmetry group from rigid class vectors
(resp. orbits). We give here some theorems to illustrate this phenomenon. These will
finally be applied to the linear groups L2 .p/.

7.1 Galois Rational Translates

Before treating the general case, we first consider rational translates with a Galois
extension. Here rational translate means that the field used for translation is a ratio-
nal function field. Thanks to this condition, the possible translation fields are those
originating from a finite rotation group as classified in Theorem 6.2. The following
result can be read off almost immediately from Theorem 6.3.

Proposition 7.1. Let G be a finite group,  2 ˙3 .G/, and NN  2 N N S .G/ the corre-
N N
sponding field extension with group Gal.N =Q.t// Š G. Furthermore let Q.u/ N D
N N
NV 2 NS .V / be the rational function field belonging to the finite rotation group V ,
which we assume to be linearly disjoint from NN  . If now T denotes the set of prime
N
divisors over S in Q.u/, then we have

NN  .u/ D NN  2 N
N T .G/ with  D 'V . /; (7.1)

where 'V is defined by the free generating s-system 'V ./ of V given in Theo-
rem 6.3.

Proof. The proof results from Theorem 6.3 by identifying the initial group G for
N
the Hurwitz classification over Q.t/ N
and over Q.u/. We then have

ker./ D ker. / \ V : t
u

Now 'V maps ˙3 .G/= Inn.G/ bijectively onto the subset of V -invariant gener-
ating systems in ˙r .G/ with r D jTj. This leads to the following generalization of
the fixed point theorems in Matzat (1987), III, 2.4:

Theorem 7.2 (Fixed Point Theorem). Let G be a finite group with trivial center and
V a finite rotation group having no common non-trivial factor group with G. Then
the map

'V W ˙3 .G/= Inn.G/ ! ˙r .G/= Inn.G/; Œ 7! Œ'V . / ; (7.2)

defined in (7.1) is injective. The image of 'V consists precisely of the HTV -invariant
classes of generating systems Œ 2 ˙r .G/= Inn.G/.
54 I The Rigidity Method

Proof. Without loss of generality we may assume that S consists of the support of
.t/ and .t  1/. Now G and V have no non-trivial factor group in common so the
fields NN  and NN V are necessarily linearly disjoint over Q.t/
N for all  2 ˙3 .G/. This
N N
implies that Gal.N =Q.u// Š G for  D 'V . / and

GV WD Gal.NN  =Q.t//
N Š G  V: (7.3)

Since Z .G/ D 1, the field NN  is uniquely determined by NN  as the fixed field of

CGV .G/ Š V . So if NN  ¤ NN Q are different, then we also have NN  ¤ NN Q . This
proves the injectivity of 'V first on ˙3 .G/= Aut.G/, and then since

Œ'V . ˛ / D Œ'V . / ˛ for ˛ 2 Aut.G/ (7.4)

also on ˙3 .G/= Inn.G/.

By construction, the set T is a V -configuration. As the extensions of  2 HTV D
N
Gal.Q.u/= N
Q.t// act as inner automorphisms on Gal.NN  =Q.u//
N Š G by (7.3), the
class Œ D Œ'V . / is invariant under HT according to (4.5), and we have HV D
V

HTV Š V . Conversely, given  2 ˙r .G/ with HV D HTV , then Q.t/ N is a field of
N N
definition of N =G Q.u/. Hence for the corresponding Galois extension NN =Q.t/ N
N N
with N .u/ D N there exists by the Hurwitz classification a system  2 ˙3 .G/
with NN D NN  . t
u
Remark. Without the assumption on generation in Theorem 7.2, the translation 'V
only maps into ˙N r .G/= Inn.G/. This generalization will be of use in the next para-
graph (see (8.16)).
The previous result also allows the explicit determination of generators for the
inertia groups for the Galois extensions constructed in the Twisted Rigidity Theo-
rem.

7.2 Rational Translates with Few Ramification Points

The possibilities for Galois rational translates are rather restricted by Theorem 6.2.
But if the assumption that our translation field be Galois is dropped, then a wide
variety of field extensions are possible. Guralnick and Thompson (1990) conjec-
tured that apart from cyclic and alternating groups only a finite list of further simple
groups can occur as composition factors of the Galois group of the Galois closure
N
of such a field extension Q.u/= N
Q.t/. This has been proved by the work of many
authors, the last step was given by Frohardt and Magaard (2001). Frohardt et al.
(2016) have announced a complete classification of all possible non-alternating and
non-cyclic composition factors which comprises 45 further simple groups, including
for example all five Mathieu groups.
Here we will here restrict ourselves to translates where the translated Galois
extension is ramified in at most 4 points. Such Galois extensions have the advan-
tage that they can be tested on possible symmetries with the help of the formulae
7 Rational Translates of Galois Extensions 55

for geometric automorphisms proved in the previous paragraph. Without loss of

generality we may moreover restrict ourselves to primitive translates, i.e., those
N
extensions Q.u/= N
Q.t/ having no proper intermediate field, since all other translates
may be composed of these. In the language of Galois groups, this means that the
Galois group of the Galois closure LN of Q.u/=
N N
Q.t/ acts primitively on the cosets
N N
of Gal.L=Q.u//. All possibilities for such non-Galois rational translates of degree
n 4 for Galois extensions of genus g > 1 are collected in the following theorem.
Theorem 7.3. For Galois extensions of genus g > 1 with at most four ramifica-
tion points there exist the following primitive and non-Galois rational translates of
degree n 4:
(a) a translation of degree 3 with group S3 , class vector C D .2A; 2A; 3A/ and
 1 1
'S0 3 . / D .12 ; 1 2 ; .22 /1 ; 2 ; 33 /; (7.5)

(b) a translation of degree 4 with group A4 , class vector C D .2B; 3A; 3B/ and

0 1  2 22 2
'A 4
. / D .12 ; .12 /3 1
; 23 ; 2 3 ; .33 /2 ; 3 /; (7.6)

as well as one with C D .3A; 3A; 3A/ and

00  2  2  2
'A 4
. / D .13 ; 1 2 ; 23 ; 2 3 ; 33 ; 3 1 /; (7.7)

(c) a translation of degree 4 with group S4 , class vector C D .2A; 3A; 4A/ and
 1 1 1
'S0 4 . / D .12 ; 1 2 ; 12 1 ; .23 /3 1
; 2 ; 34 /: (7.8)

Here qA denotes the class of q-cycles and 2B the class of double transpositions in
the symmetric group Sn .
Proof. Clearly only the groups S3 , A4 and S4 possess primitive, faithful, non regular
permutation representations of degree n 4. The condition g.Q.u// N D 0 together
with the Hurwitz genus formula then shows that the s prime divisors P1 ; : : : ; Ps
ramified in the Galois extension NN =Q.t/
N have to split into .s  2/n C 2 prime divi-
N
sors Qj in Q.u/= N
Q.t/. Since we require that r 4, all but four of the Qj have
N
to ramify over Q.t/. Consequently the generating s-system for the Galois clo-
sure LN of Q.u/=
N N
Q.t/ in the degree n permutation representation contains elements
with altogether at most four fixed points. This restricts the possible class vectors
to Ca D .2A; 2A; 3A/ and C0a D .2A; 2A; 2A; 2A/ for S3 , Cb D .2B; 3A; 3B/ and
C0b D .3A; 3A; 3A/ for A4 and Cc D .2A; 3A; 4A/ for S4 . In the case of the class
vector C0a , due to r 4 precisely four prime divisors are ramified in NN .u/=Q.u/ N
N N N N
and in N .u/=Q.t/, all of order 2, which implies g.N =Q/ D 1. For the remaining
class vectors Ca , Cb and Cc , the corresponding Galois extension L= N Q.t/
N is uniquely
0
determined by Theorem 6.2(c), and the same holds for Cb as well.
In the case of V D S3 with the class vector Ca , the field LN D NN V (in the nota-
tion of Theorem 6.3) possesses three conjugate subfields of degree 3. Among these
56 I The Rigidity Method

we may without loss of generality choose for Q.u/ N the one for which S0 3 WD
Gal.MN S =Q.u//
N contains the inertia group h2 iO. This determines S0 3 , which then
has the following generating 5-system:
 1 1
S0 3 D h12 ; 1 2 ; .22 /1 ; 2 ; 33 j .1 2 /3 33 D 1iO:

This shows (a). The translation formulae in (b) and (c) are proved in a completely
analogous fashion. t
u

With these translation maps, classes of generating systems may be distinguished

as in the Fixed Point Theorem 7.2 via their stabilizers.

Theorem 7.4. Let G be a finite group with trivial center and ' a rational transla-
tion, whose group does not possess a non-trivial factor group in common with G.
Then the map

' W ˙s .G/= Inn.G/ ! ˙r .G/= Inn.G/; Œ 7! Œ'. / (7.9)

is injective.
N
Proof. Let Q.u/= N
Q.t/ be the field extension giving rise to the translation, and LN
its Galois closure. By assumption, the translation group H WD Gal.L= N Q.t//
N and G
do not have a common non-trivial factor group, so L=N Q.t/
N is linearly disjoint from
N
each Galois extension NN  =Q.t/ with Œ 2 ˙s .G/= Inn.G/. The composition L N NN 
N
is hence Galois over Q.t/ with group

GH WD Gal.LN NN  =Q.t//
N Š G  H:

As in the proof of Theorem 7.2 the Galois extension NN  is uniquely determined

by LN NN  and hence also by NN  .u/ as the fixed field of CGH .G/ (where here G D
Gal.LN NN  =L/).
N In particular, if NN  ¤ NN Q then also NN '. / ¤ NN '.Q / . Via the Hurwitz
classification, ' is thus seen to be injective on ˙s .G/= Aut.G/, and then as in (7.4)
since Œ'. ˛ / D Œ'. / ˛ also on ˙s .G/= Inn.G/. t
u

This immediately yields the following decomposition for the classes of generat-
ing systems in ˙.C /= Inn.G/:

Corollary 7.5. Let C 2 Cl.G/r be a class vector of G and C1 ; : : : ; Cq 2 Cl.G/s

those class vectors for which ˙.Cj / contains preimages under the translation '
in (7.9). Then we have a disjoint decomposition
q
[

˙.C/= Inn.G/ D '.˙.Cj /= Inn.G// [ R; (7.10)
j D1

where R denotes the set of classes of generating r-systems in ˙.C/= Inn.G/ which
cannot be obtained by translation with '.
7 Rational Translates of Galois Extensions 57

Such a decomposition can be utilized to obtain improved degree estimates for

fields of definition of the corresponding Galois extensions, in a similar way to the
decomposition into orbits with different stabilizer in Theorem 6.8. This is illustrated
in the following example of a rational Galois translate.
Example 7.1. Let G D S8 and C D .10A; 10A; 3B/ where the classes 10A, 3B
are characterized by the cycle types .5; 2; 1/, .32 ; 12 / respectively. Then one finds
l.C/ D 15. Under the action of the geometric automorphisms in H V with the sym-
metry group V D h.12/i, the set ˙.C/= Inn.G/ splits into three orbits of length 2
and nine fixed points. In the notation of Section 6.3 we hence have l1V .C/ D 3 and
lVV .C/ D 9. By the Fixed Point Theorem 7.2 the nine fixed points originate from a
translation with 'Z2 , which by Theorems 7.2 and 6.3 has the form

 1
'Z2 . / D .1 ; 1 2 ; 22 /; (7.11)

if we let 32 D 1, say. The possible preimages are now C1 D .10A; 6A; 2D/ with
l.C1 / D 1, C2 D .10A; 6B; 2C / with l.C2 / D 7 and C3 D .10A; 6C; 2D/ with
l.C3 / D 1, where 6A, 6B, 6C denote the classes of permutations with types
.6; 12 /, .6; 2/ resp. .32 ; 2/, and 2C , 2D those of types .23 ; 12 / resp. .24 /. The
class vectors C1 and C3 are rationally rigid, hence the images of the correspond-
ing classes Œ 2 'Z2 .˙.Ci /= Inn.G// for i D 1 and i D 3 also remain stable under
N
 D Gal.Q.t/=Q.t//. Since Z .S8 / D 1, the fixed field K D Q.t/ for these Œ
N
constitutes a field of definition for NN  =G Q.t/. t
u
In the above example, Galois extensions N =Q.t/ N with Œ 2 ˙.C/= Inn.G/
defined over Q only arise by translation of Galois extensions which themselves are
also already defined over Q.t/. To obtain genuinely new Galois extensions in this
way, jRj would have to be equal to 1 in Corollary 7.5. However, an example of this
phenomenon has yet to be found.
More successfully the rational translation could be used in connection with the
Twisted Rigidity Theorem, when the class vector of the translated Galois extension
possesses additional symmetries.

7.3 Twisting Rational Translates

A rational translation

' W ˙s .G/= Inn.G/ ! ˙r .G/= Inn.G/; Œ 7! Œ'. / ;

in particular also translates the class vectors of a group G, hence it may be coarsened
to
' W Cl.G/s ! Cl.G/r ; C 7! '.C/: (7.12)
N
Denote the corresponding field extension by Q.u/= N
Q.t/. Then moreover ' asso-
N
ciates to the set S of prime divisors ramified over Q.t/ the subset T of those prime
58 I The Rigidity Method

N
divisors Q 2 IP.Q.u/= N over S which still ramify over Q.u/.
Q/ N This will be abbrevi-
ated by T D '.S/.

Proposition 7.6. If the rational translate ' is defined over k, i.e., if it originates from
a geometric field extension k.u/=k.t/ with k  Q, N
N then for all ı 2 Gal.Q.t/=k.t//
and their extensions ıQ 2 Gal.Q.u/=k.u//
N we have

ıQ ı ' D ' ı ı: (7.13)

The prototype of the translation theorems may then be formulated as follows:

Theorem 7.7 (Translation Theorem). Let G be a finite group with trivial center,
C 2 Cl.G/s a class vector, ' a rational translation defined over Q and V a symmetry
group of '.C/ with an orbit of odd length, with respect to which '.S/ forms a V -
configuration. If there exists a Œ 2 ˙.C/= Inn.G/ with
V
'.Œ ı / 2 T WD Œ'. / H'.S/ for all ı 2 V'.C/ ; (7.14)

Q with Galois group

then there exists a geometric Galois extension N=QV'.C/ .u/

Gal.N=QV'.C/ .u//
Q Š G: (7.15)

Here QV'.C/ D Q if '.C/ is a V -symmetric class vector.

Proof. Let Q.u/=Q.t/ be the field extension affording the translation. Further, let
 WD '. /, D WD '.C/ and T WD '.S/. As T forms a V -configuration, we have H WD
HTV Š V , and T remains invariant under Q WD Gal.Q.u/=Q N V
D .u//. Thus T is also
Q Q
stable under D WD h; H i.
V

By assumption we have '.Œ ı / 2 T for all ı 2 VD D Gal.Q.t/=Q N V

D .t//, hence
Q Q Q
by Proposition 7.6 the image Œ'. / also lies in T for all ı 2  . Consequently T
ı

is a Q VD -invariant H -orbit in ˙.D/= Inn.G/. The index of the stabilizer Q V of Œ

in Q VD hence coincides with the index of the corresponding stabilizer in H . This
proves that the fixed field KV of Q V is regular over QVD .
Since by assumption V possesses an orbit of odd length, Proposition 6.9 proves
the existence of uQ 2 KV with KV D QVD .u/. Q Furthermore, as Z .G/ D 1 we may
conclude from Proposition 3.7 that KV is a field of definition of NN  =G Q.u/,
N which
completes the proof of the assertion. t
u

Remark. If the stabilizer H'. / in Theorem 7.7 is trivial, then we have moreover
N D NN '. / .
QN

If the set '.S/ in the Translation Theorem consists of just three prime divisors,
the hypotheses on the V -configuration need not be checked thanks to Theorem 6.4,
and without loss of generality S may be assumed to be defined over Q. In this case
7 Rational Translates of Galois Extensions 59

the translation formulae in Theorem 7.3 yield the explicit result:

Corollary 7.8. Let G be a finite group with trivial center and without factor groups
of type S3 or A4 . Further let C D .C1 ; C2 ; C3 / be a rigid class vector of G, consist-
ing of two rational classes C1 and C2 with C12 D 1 and C23 D 1, and the semira-
tional class C3 (with primitive power CN 3 ¤ C3 ). If the corresponding unique classes
of generating systems Œ 2 ˙.C/= Inn.G/, resp. Œ 2 ˙.C/= N Inn.G/ satisfy one of
1
CN 3 D C32 and Œ1 ; 3 1 3 ; 32 D Œ1 ; 32 ; 31
  3
DW Œ'./ ; or (7.16)
1 2
CN 3 D C33 and Œ2 ; 3 2 3 ; 33 D Œ2 ; 33 ; 32
  3
DW Œ'./ ; (7.17)
Q with
then there exists a geometric Galois extension N=Q.u/

Q ŠG
Gal.N=Q.u// N D NN '. / :
and QN (7.18)

Proof. Since the class vector C is semirational and rigid, the orbit of Πunder  WD
N
Gal.Q.t/=Q.t// consists of just Πitself and Π, and we have
N
 D Gal.Q.t/=Q C .t//.
In the first case, when CN 3 D C32 , we choose the rational translation ' WD 'S0 3 ı 02
with 'S0 3 from Theorem 7.3(a) and the geometric automorphism 02 belonging to
! D .23/. Then since 12 D 1 and 23 D 1 we get

Œ'. / D Œ'S0 3 .1 ; 2 3 21 ; 2 / D Œ1 ; 32 ; 31 3 : (7.19)

Hence we have '.C/ D .C1 ; CN 3 ; C3 / and V WD h.23/i is a subgroup of Sym.'.C//.

The assumption (7.14) in the Translation Theorem is thus satisfied provided that
0
'.Œ ı / D Œ'. / 2 for ı 2 n : (7.20)

By (7.19) this is equivalent to

 31
'.Œ / D Œ1 ; 3 1 ; 32 :

In the case CN 3 D C33 we choose the rational translation ' D 'A

0
4
from Theo-
rem 7.3(b). Then we have

0  32
Œ'. / D Œ'A 4
.1 ; 2 ; 3 / D Œ2 ; 33 ; 3 2 : (7.21)

So here we have '.C/ D .C2 ; CN 3 ; C3 / with V D h.23/i Sym.'.C// as above.

Accordingly the condition (7.14) is equivalent to
 31
'.Œ / D Œ2 ; 3 2 ; 33 :

Since in both cases the image class vector '.C/ is V -symmetric, we have
QV'.C/ D Q, and then the assertion follows from Theorem 7.7 together with the sub-
sequent remark. t
u
60 I The Rigidity Method

In the next section we study some applications of these new results.

7.4 Geometric Galois Extensions over Q.t / with L2 .p/

Our starting point here is the semirational rigid class vector .2A; 3A; pA/ of the pro-
jective special linear group L2 .p/ introduced in Example 5.2, which characterizes
the field of modular functions of level p. From this by rational translation and twist-
ing one may construct geometric L2 .p/-extensions over Q.t/, as was first proved
by Shih (1974). The first two parts of the following theorem are applications of
Corollary 7.8, while the third part requires a suitable translation of degree 8 for its
proof.

Theorem 7.9 (Shih (1974)). For all primes p satisfying either . p2 / D 1, . p3 / D 1
or . p7 / D 1, the group L2 .p/ possesses a G-realization over Q.

Proof. According to Example 5.2 the class vector C D .2A; 3A; pA/ of L2 .p/ is
N
semirational and rigid. The classes of generating 3-systems in ˙.C/ and ˙.C/,
where CN D .2A; 3A; pB/, are obtained by Example 5.2 as the images of
       
aC1 1 0 1 1a a
; ; with D ˙1 (7.22)
1 0 1 1 01 p

under the canonical epimorphism from SL2 .p/ onto L2 .p/. (Observe the reverse
ordering!) Thus we have pB D .pA/a if and only if . pa / D 1.
For a D 2 and a D 3 Corollary 7.8 may now be applied. Indeed it is easily verified
that the twisting condition (7.16) resp. (7.17) is satisfied by the two classes of gener-
ating systems in (7.22). Therefore there exist geometric Galois extensions N=Q.u/ Q
with group L2 .p/ for the class vector .2A; pB; pA/ when . p2 / D 1, respectively
for the class vector .3A; pB; pA/ when . p3 / D 1.
In the case a D 7 we employ the rational translation defined by a root field of
degree 8 for the field of modular functions of level 7. (A root field for a Galois
extension is an intermediate field whose Galois closure coincides with the whole
extension.) This is generated by a zero of the equation

.u2 C 5u C 1/3.u2 C 13u C 49/  1728tu D 0 (7.23)

(see for example Klein and Fricke (1890), Kap. 7, 4 (3)). The permutation types
for the conjugacy classes of the corresponding class vector .2A; 3A; 7A/ are .24 /,
.32 ; 12 / and .7; 1/. The translation formula for a class vector C D .C1 ; C2 ; C3 / with
C21 D 1 and C32 D 1 is then given by

 2  33
'. / D .2 ; 2 3 ; 37 ; 3 2 / (7.24)
7 Rational Translates of Galois Extensions 61

(where without loss of generality we have chosen the 2 -invariant root field). This
shows that '.C/ D .3A; 3A; pB; pA/ and Sym.'.C// D h.12/; .34/i. Any quadru-
N
ple of prime divisors consisting of two orbits of length 2 under Gal.Q.u/=Q.u//
forms at least a h.12/.34/i-configuration, hence in particular the above set '.S/.
This leads to the symmetry condition
2
Œ'. / 4 D Œ'./ ; (7.25)

where 24 denotes the geometric automorphism in H'.S/

V
belonging to !42 D .12/.34/
as in Theorem 6.5. Since '.C/ is V -symmetric, verification of (7.25) with the two
V
classes of generating systems in (7.22) shows that the fixed field K'. / is regular
over Q.
Now V does not possess an orbit of odd length on '.S/, so the rationality
V
of K'. / =Q can only be deduced from the explicit knowledge of '.S/, which is
implicit in the generating equation (7.23). Ordered according to (7.24) the prime
divisors Qj 2 '.S/ are given by the divisor equalities

Q4 Q1 Q2
.u/ D ; .u2 C 13u C 49/ D :
Q3 Q23
p  
C D Q. p / with p D .1/
.p1/=2
In Example 5.2 we saw that p Q p. With respect

to the new variable v WD p uC7 2 QC .u/ the product Q3 Q4 becomes the numer-
u7

ator divisor of .v 2 C p  /, and Q1 Q2 that of .v 2 C 27p  /. Thus the geomet-

ric automorphism 24 maps the function v onto v. Since each automorphism
N
ı 2 Gal.Q.u/=Q.u// with non-trivial restriction to QC .u/=Q.u/ maps Œ to Œ and
hence by Proposition 7.6 also Œ'. / to Œ'./ , the class Œ'. / is invariant under
24 ı ı. Consequently K'. Q
/ is the fixed field of the restriction ı of 4 ı ı to QC .u/.
V 2

Since p p p
Q
.v  p  /ı D .v  p  /ı D .v C p  /;
p Q
the numerator divisor Q3 of .v  p  / is ı-invariant. Thus K'.V
/
possesses a prime
divisor of degree 1 and is a rational function field, say K'. / D Q.u/.
V
Q Moreover
V
K'. is a field of definition of N  =Q.u/
N N with its Galois group L 2 .p/ by Proposi-
/
tion 3.7, so the proof of the assertion is also complete in the case . p / D 1.
7
t
u

In the original proof of this result, Shih used Shimura’s theory of canonical sys-
tems of models instead of the symmetrization of the translation formulae. In contrast
to this, the following result can only be proved with the help of the translation theory
presented in this chapter. The proof is rather similar to the previous ones, so we just
give a rough sketch.

Theorem 7.10. For primes p with . p5 / D 1 the group L2 .p/ possesses a G-

realization over Q.

Sketch of proof. First one verifies that under the hypotheses . p2 / D 1, which can be
assumed according to Theorem 7.9, the semirational class vector C D .2A; 4A; pA/
62 I The Rigidity Method

of L2 .p/ is rigid. With CN D .2A; 4A; pB/ the classes Œ 2 ˙.C/ and Œ 2 ˙.C/
N
of generating 3-systems are the epimorphic images of
 p   p p     
0p 2=a 2
p  2=a 10 a
; ; with D ˙1 (7.26)
a= 2 0 a= 2 0 a1 p
p
in L2 .p/, where 2 denotes a root in IFp of X 2  2. As translation one employs the
field extension Q.u/=Q.t/ generated by

.u  1/4 .u2  6u C 25/ C 256tu D 0: (7.27)

In the Hurwitz classification it belongs to the class vector .2B; 4A; 5A/ of the group
PGL2 .5/ Š S5 , with components of permutation types .23 /, .4; 12 /, resp. .5; 1/. The
translation map for a class vector C with C21 D 1 and C42 D 1 is then seen to be given
by
 2  32
'. / D .2 ; 2 3 ; 35 ; 3 2 /: (7.28)
Under the assumption D 1 the image of C D .2A; 4A; pA/ 2 Cl.L2 .p//3
. p5 /
hence equals '.C/ D .4A; 4A; pB; pA/. As in the previous proof the set '.S/ forms
a V -configuration for V WD h.12/.34/i D h!42 i, and '.C/ is V -symmetric. The ver-
ification of the symmetry condition
2
Œ'. / 4 D Œ'./ (7.29)
V
is straightforward calculation. This shows that the field of definition K'. /
of
N N
N'. / =Q.u/ is regular over Q. Finally one checks
p the rationality as in the previ-
ous proof, this time using the new variable v WD p  uC5
u5
. t
u

A detailed proof of this theorem, as well as further examples, can be found in

Malle (1991). Also in Shih (1978) and Malle (1993a) generating polynomials for
the Galois extensions constructed in Theorems 7.9 and 7.10 with group L2 .p/ were
determined at least for p 37. (For p 29 see the table in the Appendix A.2.)
For that it was useful that the results of Theorem 7.10 possess a translation into
the language of modular functions. Indeed, the constructed field extensions may be
interpreted as rational models of modular function fields belonging to congruence
subgroups inside certain Hecke groups (see Malle (1993a), Section 5).

Remark. According to Theorems 7.9 and 7.10 the groups L2 .p/ possess G-realiza-
tions over Q at least for all primes p < 311. By Hilbert’s irreducibility theorem this
remains true for ordinary Galois extensions with group L2 .p/ over Q. By a new
result of Zywina (2015) the latter statement has been shown to hold for all primes
p > 3.
8 Automorphisms of the Galois Group 63

8 Automorphisms of the Galois Group

The fixed fields of a generating system of a group G modulo Aut.G/, or more gener-
ally of an intermediate group A between Inn.G/ and Aut.G/ are natural candidates
for fields of definition for the field extension obtained from the Hurwitz classifica-
tion, but not in general with the Galois group. So instead of G, one can sometimes
realize the group A, under certain assumptions even as geometric Galois group.
With such an extension theorem we may for example embed the L2 .p/-extensions
of the last paragraph into geometric PGL2 .p/-extensions.

8.1 Fixed Fields of Coarse Classes of Generating Systems

In this paragraph, A denotes a group of automorphisms of a finite group G contain-

ing Inn.G/, and AN WD A= Inn.G/. The orbit of a class vector C 2 Cl.G/s under A is
denoted by
CA WD fC˛ j ˛ 2 Ag: (8.1)
The full symmetry group of CA is

Sym.CA / WD f! 2 Ss j C!˛ 2 C for some ˛ 2 Ag; (8.2)

and any subgroup V Sym.CA / is called a symmetry group of CA . With respect

to such a V we define HSV and VS as in (6.3), (6.19) respectively. For the coarse
classes of generating systems  A 2 ˙.CA /=A we thus obtain the fixed groups

V A WD fı 2 VS j  ıA D  A g and HVA WD HSV \ V A ; (8.3)

with corresponding fixed fields

N VA V
KVA WD Q.t/  resp. KN VA WD Q.t/
N H A : (8.4)

Theorem 8.1. If the fixed field KVA of  A is a disclosed function field, then it con-
N
stitutes a field of definition of NN  =Q.t/ (without the group action).

Proof. As in Proposition 4.6 we note that the field extension NN  =KVA is Galois.
By Theorem 3.5 the field KVA is hence a field of definition of NN  =Q.t/,
N being a
disclosed function field. t
u

Remark. According to Proposition 6.9 the field KVA is disclosed for example when
V possesses an orbit of odd length on S.

We next give an estimate for the degree of the extension of constants in KVA =Q.
Again we first determine the cyclotomic field extension contained in KVA =Q. There-
64 I The Rigidity Method

fore analogously to (4.24) we define the stabilizer of

CAV WD fC˛! j ˛ 2 A; ! 2 V g (8.5)

N
in  D Gal.Q.t/=Q.t// under the action via the cyclotomic character as

C WD fı 2  j C
AV 2 CAV g:
c.ı/
(8.6)

This has index

jC j
d AV .C/ WD . W AV
C /D ; (8.7)
jCAV \ C j
which in generalization of (4.23) is called the AV -symmetrized irrationality degree
of C. In particular C is called AV -symmetric if d AV .C/ D 1. Immediately from the
definitions we have:
AV

C WD Q
Proposition 8.2. The fixed field QAV N C is an abelian number field contained
V
in QC QC with h i
QAV
C W Q D d AV .C/: (8.8)

C D Q if the class vector C is AV -symmetric.

In particular we have QAV

The number of HSV -orbits in ˙.CAV /=A with given stabilizer U HSV up to
automorphisms is denoted by
V
lUAV .C/ WD jf AHS j  2 ˙.C/; HVA D U ˛ for some ˛ 2 Aut.HSV /gj (8.9)
V
in generalization of (6.17). Correspondingly  AHS is called a rigid HSV -orbit if
lUAV .C/ D 1 for U D HVA . If ˙.CAV /=A consists of a single HSV -orbit, the class
vector C itself is called AV -rigid. This defines all notations in the following result:

Theorem 8.3. Let G be a finite group, Inn.G/ A Aut.G/, C 2 Cl.G/s with

s  3, and V Sym.CA / a symmetry group with respect to which S forms a V -
configuration. Then the fixed field KVA of  A 2 ˙.CA /=A contains QAV
C . The
V V
degree of the algebraic closure k A of Q in K A can be estimated by

ŒkV A W QAV
C lUAV .C/ with U D HVA : (8.10)
V
C is regular if 
In particular, KVA =QAV AHS
is rigid.

Proof. The proof is entirely analogous to that of Theorem 6.8: As S ./ V we

have Cc.ı/ 2 CAV for each ı 2 V A , which entails V A hAV Q AV
C ; HS i DW C . If
V

now KQ CAV denotes the fixed field of Q AV V ı 1

C , then since H ıA D .H A /
V
we have

ŒKVA W KQ CAV D .Q AV

C W  A /
V
.HSV W HVA /  lH
AV
V .C/:
A
8 Automorphisms of the Galois Group 65

Together with (8.4) we then obtain

h i ŒKVA W KQ CAV .Q AV
C W  A /
V
kV A W QAV D D AV
lH V .C/:
C N VA W Q
ŒQK N KQ AV .HSV W HVA / A
 C

V
C is regular when 
By definition the extension KVA =QAV t
u
AHS
is rigid.
In the case of a trivial symmetry group we obtain the following generalization
of Theorem 4.5 from the preceding theorem (where we now omit the superfluous
groups from the notation):
Corollary 8.4. If in Theorem 8.3 we have V D 1 so that S remains pointwise fixed
N
under  D Gal.Q.t/=Q.t//, then it follows that
1
ŒKCA W QA
C .t/ l A .C/ D l.CA /: (8.11)
N
jAj

Proof. By the previous results, it only remains to show the equality in (8.11). This
follows from
j A j D jAj D jAjN  j Inn.G/j D jAj
N  jΠj (8.12)
and the definition (8.9):
N  l A.CA / D jAj
l.CA / D jAj N  l A.C/: t
u

Example 8.1. According to Example 5.2 the class vector C D .2A; 3A; pA/ of L2 .p/
satisfies l A .C/ D 1. By Theorem 8.1 and Corollary 8.4 the field of modular func-
C .t/ D Q.t/. Still, since QC ¤ Q, this is not
tions of level p is hence defined over QA
true if we include the Galois group. t
u
In the next section Galois extensions over KVA with group A will be constructed
for which the fixed field of Inn.G/ coincides with KV . In the case Z .G/ D 1 this
leads to an extension of Gal.N=KV / Š G by AN D A= Inn.G/ (compare with the
Twisted Rigidity Theorem 6.10).

8.2 Extension of the Galois Group by Outer Automorphisms

In this section, we make the general assumption that the group A acts on the VS -
V
orbit of ΠD  Inn.G/ , i.e., that  A forms a subset of ΠS . Then we first have the
following general result:
Proposition 8.5. Let G, A and C be as in Theorem 8.3, V Sym.CA / and  2
V
˙.C / with   Œ S . Then there exists a Galois extension N A =KVA with
A A

Gal.N A =KVA / Š A and Gal.N A =KV / Š Inn.G/: (8.13)

Here the field extension N A =KV is geometric.

66 I The Rigidity Method

Proof. It was already seen in the proof of Theorem 8.1 that the extension NN  =KVA
is Galois. Now let N A denote the fixed field of the centralizer of G D Gal.NN  =Q.t//
N
N
in this Galois extension. Then due to Q.t/N D N
A N Z .G/ N
DW N we have
A

N
Gal.N A =KV / Š Gal.NN A =Q.t// Š Inn.G/;

and in particular N A =KV is geometric and Galois. Further Gal.N A =KVA / is cer-
tainly isomorphic to a subgroup of A. By assumption AN D A= Inn.G/ acts on the
VS -orbit of Π, which thus according to (8.12) splits under AN into suborbits of
N This implies
length jAj.
h i 
KV W KVA D V A W V D jAj N

and hence Gal.N A =KVA / Š A. t

u

Adding our standard hypotheses this implies the following:

Theorem 8.6. Let G, A, C and V be as in Theorem 8.3, where moreover V pos-

sesses an orbit of odd length. Assume furthermore that the class vector C is V -rigid,
N
and that for each ˛ 2 A there exists ı 2  D Gal.Q.t/=Q.t// with C˛V D Cc.ı/V .
AV Q
Then there exists a Galois extension N =QC .t / such that N =QVC is regular and
A A

Gal.N A =QAV Q A V Q
C .t // Š A and Gal.N =QC .t // Š Inn.G/: (8.14)

C D Q if C is AV -symmetric.
Here we have QAV

Proof. The V -rigidity of C and the assumption that CAV  CV guarantee that
V
˙.CV /= Inn.G/ contains a single VS -orbit. Consequently we have  A  Œ S .
V
By Proposition 8.5 it remains to show that the function field K A is rational over
QAV
C . But under the above assumptions this follows immediately from Theorems 8.1
and 8.3. t
u

Example 8.2. We return to the case of the field N=QC .t/ of modularpfunctions
of level p already considered in Example 8.1. Here we have QC D Q. p  / and
Gal.N=QC .t// Š L2 .p/. Since CA D C we may conclude from Theorem 8.6 with
V D 1 that
Gal.N=Q.t// Š PGL2 .p/: t
u

8.3 Geometric Extension of the Galois Group by Outer

Automorphisms

The possibility of extending the Galois group becomes still more interesting when
KV =KVA is geometric. For this we get the following criterion:
8 Automorphisms of the Galois Group 67

Theorem 8.7 (Extension Theorem). Let G, A, C and V be as in Theorem 8.3, where

moreover V possesses an orbit of odd length. Suppose that ˙.CV /= Inn.G/ con-
tains a rigid HSV -orbit on which A acts. Then there exists a geometric Galois exten-
sion N A =QVC . Q́ / with

Gal.N A =QVC . Q́// Š A and Gal.N A =QVC .tQ// Š Inn.G/: (8.15)

If the class vector C is V -symmetric, then we have QVC D Q.

V
Proof. Let Œ HS be the rigid HSV -orbit in ˙.CV /= Inn.G/. For this we have
V V
 A  ΠHS D ΠS by assumption. Hence by Proposition 8.5 there exists a Galois
extension N A =KVA with group Gal.N A =KVA / Š A, in which N A =KV is geomet-
V
ric. Now since  A  ΠHS we have
h i   h i
KV W KVA D V A W V D HVA W HV D KN V W KN VA ;

which shows that the extension KV =KVA is geometric. Because of the rigidity of
the HSV -orbit it follows from Theorem 6.8 that KV =QVC is regular. Hence the same
holds for KVA =QVC . Since V possesses an orbit of odd length, Proposition 6.9 shows
that first KV =QVC and then by the Theorem of Lüroth also KVA =QVC is rational, say
KV D QVC .tQ/ and KVA =QVC . Q́ /. This completes the proof. t
u

Remark. In the case Z .G/ D 1 the Galois extension N A =QVC .tQ/ in (8.15) coincides
with the extension N=QVC .tQ/ constructed in the Twisted Rigidity Theorem 6.10.

With the next example we continue the Example 5.1.

Example 8.3. Let G D L2 .8/ and C the class vector .9A; 9B; 9C / of G. Since
l.C/ D 1, ˙.C/= Inn.G/ consists of a single class of generating systems Œ D
Œ1 ; 2 ; 3 . According to (6.11) its components i are permuted cyclically by V D
h.123/i. On the other hand, Out.G/ D Z3 also permutes the three classes 9A, 9B
and 9C cyclically. Thus for every V -configuration S and for A WD Aut.G/ D L2 .8/
V
we have  A D ΠHS . In particular,  A itself is a rigid HSV -orbit. Hence Theo-
rem 8.7 implies the existence of a geometric Galois extension N=Q. Q́ / with

Gal.N=Q. Q́// Š L2 .8/ :

The corresponding class vector is then given by C Q D .9A; 3B; 3C /. Since the fixed
field of L2 .8/ in N=Q. Q́ / is a rational function field, say Q.tQ/, Gal.N=Q.tQ// yields
a GA-realization of L2 .8/. t
u

If now NN A =Q.´/
N denotes the Galois extension with group A obtained from
N then the class of generating systems
N =QVC . Q́ / by extension of constants with Q,
A

of A parametrizing NN A can be deduced from the classifying class of generating sys-

tems Œ 2 ˙s .G/= Inn.G/ of NN =Q.t/
N with the help of the Fixed Point Theorem 7.2.
68 I The Rigidity Method

If NN =Q.´/
N N
is ramified in the r prime divisors of T  IP.Q.´/= N then NN A 2 NT .A/.
Q/,
According to the Hurwitz classification there exists Œ 2 ˙r .A/=A with NN  D NN A .
In the case Z .G/ D 1 one obtains this class using the general translation map

'V W ˙r .A/=A ! ˙N s .A/=A; Œ 7! Œ'V ./ ; (8.16)

(see the Remark following Theorem 7.2) as preimage of  A . This is a consequence

of:
Corollary 8.8. Let G, A, C and V be as in Theorem 8.3 and assume Z .G/ D 1.
Then the restriction

'VG W ˙rG .A/=A ! ˙s .G/=A;  A 7! 'V ./A ; (8.17)

of the translation map 'V from (8.16) to the inverse image ˙rG .A/=A of ˙s .G/=A is
injective. The image of 'VG consists precisely of those classes of generating systems
V
 A 2 ˙s .G/=A satisfying  A  Œ HS .
Proof. By the Fixed Point Theorem 7.2 the field NN A is uniquely determined by
N
 Aut.G/ , since it can also be obtained by a translation Q.t/= N HV under which
Q.t/
G
the group remains invariant. Hence the map 'V is injective. The characterization of
the image now follows from the proof of the Extension Theorem 8.7. t
u
This corollary may also be used to test whether a given G-realization actually is
a GA-realization or not.
Example 8.4. Let N=Q.t/ be the Galois extension with the Mathieu group G D M12
from Theorem 6.12 for the class vector C D .4A; 4A; 10A/. For Œ 2 ˙.C/= Inn.G/
V
and V D h.12/i we then have Œ HS  ˙.C/= Inn.G/. Since the outer automor-
phisms of M12 permute the classes 4A and 4B, the class  A is not contained in
V
ΠHS . Consequently, Gal.N=Q.t// is not a GA-realization for M12 . (But see The-
orem II.9.9 for a proper GA-realization of this group.) t
u

8.4 Geometric Galois Extensions over Q.t / with PGL2 .p/

Application of the Extension Theorem 8.7 leads to an embedding of the geometric

Galois extension over Q.t/ with groups L2 .p/ of the preceding paragraph into one
with groups PGL2 .p/. More precisely we have:
Theorem 8.9. The G-realizations of L2 .p/ constructed in Theorems 7.9 and 7.10
are GA-realizations.
Proof. The proof will only be given for the cases . p2 / D 1 and . p7 / D 1, the other
two being entirely similar.
Q denote the G-realization of G D L2 .p/ over Q from The-
First let Gal.N=Q.u//
orem 7.9 for . p2 / D 1. Then we have QN N D NN '. / with ' D ' 0 ı 0 and the
S3 2
8 Automorphisms of the Galois Group 69

unique class of generating systems Œ 2 ˙.C/= Inn.G/ for C D .2A; 3A; pA/.
Since '.C/ D .2A; pB; pA/ we have Sym.'.C/A / D h.23/i for A D PGL2 .p/. To
V
verify the extension condition '. /A  Œ'. / HS it therefore suffices to show the
validity of
Œ'. / ˛ D Œ'. / with  D 02
for the outer automorphisms ˛ 2 A. Due to Œ ˛ D Œ this was already shown in the
form of the twisting condition (7.20).
Now let Gal.N=Q.u// Q denote the G-realization of L2 .p/ over Q for . p7 / D 1.
Then we have QN N D NN '. / with ' from (7.24) and Πas above. We now find
'.C/ D .3A; 3A; pB; pA/, and '.S/ forms a h.12/.34/i-configuration. Hence as
above we obtain the extension condition for an outer automorphism ˛ 2 A from the
twisting condition (7.25) proved in Theorem 7.9. t
u
For later use we note the following supplementary results:
Corollary 8.10. The geometric PGL2 .p/-extensions constructed in Theorem 8.9
belong to the class vectors .2B; 4A; pA/ for . p2 / D 1, .2B; 6A; pA/ for . p3 / D 1,
.2B; 2B; 4A; pA/ for . p5 / D 1, and .2B; 2B; 3A; pA/ for . p7 / D 1. Here all ram-
ified prime divisors of the fixed field have degree 1.
Proof. This follows immediately from the translation formulae for Q.u/= N N
Q.´/ with
N
Q.´/ N
D Q.u/ H V
S .

If in the case . p2 / D 1 the prime divisors ramified in NN =Q.t/

N are denoted by
Q2 WD P1 jQ.´/
N , Q 3 WD P P j N
2 3 Q.´/ , and if Q 1 denotes the second prime divisor ram-
N N
ified in Q.u/=Q.´/, then the translation formula according to (6.8) for the ramifica-
tion orders .2; 2; 1/ reads

'Z2 ./ D .12 ; 22 ; 32 ; 3 / D .1; 1 ; 2 ; 3 /: (8.18)

Hence we find 3 D 3 2 pA, 2 2 4A since 22 D 1 2 2A, and 1 2 2B. In the case
. p3 / D 1 the class vectors may be determined with the same translation formula.
In the two remaining cases one utilizes the translation formula belonging to the
ramification orders .2; 2; 1; 1/
1 1
0
'Z 2
./ D .12 ; 22 ; 32 ; 32 4 2
; 42 ; 4 / D .1; 1; 1 ; 2 ; 3 ; 4 /: (8.19)

Because of the different ramification orders, the residue degrees of the ramified
prime divisors are always equal to 1, except possibly the divisors ramified of order 2
in the cases . p5 / D 1 and . p7 / D 1. But by the proof of Theorem 7.9 in the case
p Q
. p7 / D 1 the fixed field of L2 .p/ is K'.
V
Q with vQ D p  v (since vQ ı D v).
/ D Q.v/ Q
Now  WD 4 satisfies vQ D v,
2
Q so we may take K'. /A D Q. Q́ / with Q́ WD vQ 2 .
V

Consequently the divisors in the support of Q́ are the two prime divisors ramified
in N=Q. Q́ / of order 2. With the same argument one obtains, that also in the case
. p5 / D 1 both prime divisors ramified in N=Q. Q́ / with order 2 have degree 1. u t
70 I The Rigidity Method

9 Computation of Polynomials with Prescribed Group

For the geometric field extensions found by the rigidity method, apart from the
Galois group also the ramification points and the generators of inertia groups are
known via the Hurwitz classification. At least in principle, this enables one to com-
pute generating polynomials for these extensions. The range of calculation is only
restricted by the fact that it requires the solution of systems of non-linear alge-
braic equations in quite a number of unknowns. In this paragraph the necessary
calculations are performed for the groups Sn and An , including the exceptional case
Aut.A6 / Š PL2 .9/, and for the two small Mathieu groups M12 and M11 .

9.1 Decomposition of Prime Divisors in Galois Extensions

To compute polynomials we require knowledge of the decomposition of the prime

divisors ramified in a root field for the extension. We identify the Galois group
Gal.f / of a separable polynomial f 2 KŒX with the permutation group on the
roots x1 ; : : : ; xn of f in a splitting field N , which hence coincides with the image
of Gal.N=K/ under the permutation representation
 
x1 : : : xn
f W Gal.N=K/ ! Sn ;  7!  : (9.1)
x1 : : : xn

This is clearly equivalent to the representation of Gal.N=K/ on the cosets of the

stabilizer U of the root field L WD K.x1 /. The decomposition of prime divisors
(interpreted as valuation ideals of discrete rank 1 valuations, compare Engler and
Prestel (2005), Ch. 2.1) in L=K is then obtained from the following general result:

Theorem 9.1. Let K be a field with prime divisor P, L=K a finite separable field
extension, f 2 KŒX the minimal polynomial of a primitive element x of L=K and
N the splitting field of f over K with Galois group G D Gal.N=K/. Further let
PQ be an extension of P to N with separable residue field extension. If the set of
zeroes X D fx1 ; : : : ; xn g of f in N splits into r orbits X1 ; : : : ; Xr under the action
of the decomposition group f .GD .P=P//,Q and Xi splits into fi orbits of lengths
Q
ei under the action of the inertia group f .GI .P=P//, then P splits into a product
of prime divisors of L as follows:

Y
r
e
PD Qi i with residue degrees di D d.Qi =P/: (9.2)
i D1

O NO , denote the completions of K with respect to P, respectively of

Proof. Let K,
N with respect to P,Q and embed f as fO 2 KŒXO O
into KŒX . Then Gal.NO =K/
O Š
Q
GD .P=P/ (see for example Serre (1979), Ch. II, 3, Cor. 4), and with appropriate
9 Computation of Polynomials with Prescribed Group 71

numbering of the zeroes, also Gal.fO/ Gal.f /. If fO now has the prime decompo-
sition
Y
r
fO D fOi
i D1

O
in KŒX , then Gal.fO/ acts transitively on each set of zeroes Xi of fOi in NO , resp. N .
So X splits into r orbits Xi of lengths deg.fOi / under the action of f .GD .P=P//.
Q
The different prime ideals Qi of L lying above P correspond bijectively to the
prime polynomials fOi , and we have

deg.fOi / D e.Qi =P/d.Qi =P/ D ei di for i D 1; : : : ; r

(see loc. cit., Ch. II, Thm. 1 with Cor. 2). Now let LO i NO denote the fields generated
over KO by a zero x of fOi , let UO i WD Gal.NO =L
O i / and G=
O UO i a system of representatives
O O
of G modulo Ui . Then we have
Y Y
fOi D .X  x/ D .X  x O / with x 2 Xi :
x2Xi O UO i
O G=
2

If now PO denotes the valuation ideal of KO and QO i the one of L O i , then the inertia
indices satisfy
e.Qi =P/ D e.Q O i =P/
O D .GO I W .UO i \ GO I //

(see loc. cit., Ch. II, Thm. 1 and Ch. I, Prop. 22), where GO I denotes the inertia group
of NO =KO isomorphic to GI .P=P/.Q Consequently GO i and hence also GI .P=P/ Q
O
decompose the zero set Xi of fi into orbits of length e.Qi =P/, and this achieves
the proof. t
u
O
If instead of the embedding of f into KŒX we consider the canonical image fQ
of f in the polynomial ring over the residue field KQ WD KP, then we obtain the
following theorem of Dedekind (see for example Tschebotaröw and Schwerdtfeger
(1950), Kap. V, Satz 16):

Theorem 9.2 (Dedekind). Let K be a field with prime divisor P, ring of P-integers
O and residue field KQ WD O=P. Furthermore let f 2 OŒX be a monic polynomial
with discriminant D.f / … P, N the splitting field of f with G WD Gal.N=K/, P Q a
Q Q Q
valuation ideal of N lying above P and N WD N P a separable field extension of K.
Q Q Q
Then the permutation representation fQ of G WD Gal.N =K/ as Galois group of the
reduced polynomial fQ 2 KŒXQ Q
coincides with the restriction to GD .P=P/ of the
permutation representation f of G:

fQ .G/ Q
Q D f .GD .P=P//: (9.3)

Proof. By assumption, NQ =KQ is Galois, and the reduction induces a canonical epi-
morphism
Q
W GD .P=P/ ! G;Q  7! Q (9.4)
72 I The Rigidity Method

A
(see for example Serre (1979), Ch. I, Prop. 21 with Cor.). Since f .x/ is monic
and D.f / does not lie in P, we have D.fQ/ D D.f / ¤ 0. Hence the polynomial
fQ 2 KŒX
Q is separable of degree deg.fQ/ D deg.f /. Thus each  2 GD .P=P/
Q acts
on the zero set X of f in N in the same manner as Q on the zero set XQ of fQ in
Q
NQ . This implies that  Q ı is a faithful permutation representation of GD .P=P/,
f
equivalent to f jGD .P=P/
Q . t
u

In the case that the residue field KQ is finite, Theorem 9.2 allows us to deduce the
permutation types of elements in Gal.f /:

Corollary 9.3 (Dedekind Criterion). If under the assumptions of Theorem 9.2 the
field KQ is finite, and fQ splits in KŒX
Q into r prime polynomials fQi then Gal.f /
Q
contains permutations of type .deg.f1 /; : : : ; deg.fQr //.
Q the Galois group Gal.fQ/ is necessarily cyclic.
Proof. In the case of a finite field K,
Q
A generating element Q of Gal.f / permutes transitively the zeroes of each polyno-
mial fQi 2 KŒX
Q Q
. Thus by Theorem 9.2 the group f .GD .P=P// is generated by an
element of permutation type .deg.fQ1 /; : : : ; deg.fQr //. t
u

After these preparations we can now turn to the determination of polynomials,

first in the case of Sn and An .

9.2 Polynomials with Groups Sn and An

The starting point for our construction for Sn is the rationally rigid class vector
C D .2A; .n1/A; nA/ found in Proposition 5.2. We may choose the set S to consist
of the prime divisors in the support of .t/ and .t  1/,

P3 P1
.t/ D ; .t  1/ D : (9.5)
P2 P2

Then by Theorem 5.3 there exists an Sn -extension N=Q.t/ where precisely P1 ; P2

and P3 are ramified, of orders 2, n  1 resp. n. Moreover according to the Hurwitz
classification we have QNN D NN  , where Πdenotes the unique class of generating
systems in ˙.C/= Inn.G/. By (4.5) the N i WD ' .i / then generate inertia groups of
prime divisors PQ i of NN  =QN lying above Pi .
Now let L D N Sn1
be a root field of degree n of N=Q.t/, let y be a primitive
element of L=Q.t/ with minimal polynomial f . Then the permutation represen-
tation given by Gal.f / is equivalent to the representation of Sn on the cosets of
Sn1 , and hence to the natural permutation representation of Sn . Consequently the
generators of inertia groups have permutation types .2; 1n2 /, .n  1; 1/ and .n/ in
Gal.f /. Since the inertia groups are normal in the decomposition groups (see Serre
(1979), Ch. I, Prop. 20), and the orbits of the normalizers of hi i in Sn have lengths
.2; n  2/, .n  1; 1/ and .n/, by Theorem 9.1 there exist (not necessarily prime)
9 Computation of Polynomials with Prescribed Group 73

divisors Qi;j of L=Q with

P1 D Q21;1 Q1;2 ; P2 D Qn1

2;1 Q2;2 ; P3 D Q3 ;
n
(9.6)

where Q1;1 has degree deg.Q1;1 / D 1. Thus the different D.L=Q.t// has degree
2.n  1/, and the Hurwitz genus formula yields
1
g.L=Q/ D 1 C n.g.Q.t//  1/ C deg.D.L=Q.t/// D 0:
2
Since moreover L=Q possesses at least the prime divisor Q3 of degree 1, it is a
rational function field. An element of L may be determined up to scalar multiples
by fixing its divisor. We may hence uniquely define a generating element x of L by
the two equations
Q3 Q1;1
.x/ D and .x  1/ D : (9.7)
Q2;1 Q2;1
The linear space of Qn2;1 consists of the polynomial functions in x of degree at
most n. So we obtain
Q2;2 Q1;2
D .x  a/ and D .b.x// (9.8)
Q2;1 Qn2
2;1

with a 2 Q and a monic polynomial b.X / 2 QŒX of degree n  2.

From (9.5) up to (9.8) we deduce the divisor equalities
 n 
P3 Qn3 x
.t/ D D n1 D ;
P2 Q2;1 Q2;2 x a
 
P1 Q21;1 Q1;2 .x  1/2 b.x/
.t  1/ D D n1 D :
P2 Q2;1 Q2;2 x a

Hence there exist elements c0 ; c1 2 Q with

t.x  a/ D c0 x n ; .t  1/.x  a/ D c1 .x  1/2 b.x/; (9.9)

which after elimination of t yields

x  a D c0 x n  c1 .x  1/2 b.x/:

Since the element x is transcendental over Q, we find c0 D c1 DW c, and there

remains the polynomial identity
1
h.X / WD X n  .X  a/ D .X  1/2 b.X /:
c
The polynomial h.X / has a double zero at 1, which implies
1 a 1
h.1/ D 1  C D 0; h0 .1/ D n  D 0:
c c c
74 I The Rigidity Method

Substituting the only solution

1 n1
cD ; aD ;
n n
of this system of linear equations into (9.9), we obtain a relation between x and
t which obviously originates from the minimal polynomial of x over Q.t/. This
completes the proof of:

Theorem 9.4. The Galois group of the polynomial

f .t; X / D X n  t.nX  n C 1/ 2 Q.t/ŒX (9.10)

is the symmetric group Sn . It generates the geometric Galois extension N=Q.t/ in

Theorem 5.3 with the choice of S according to (9.5).

Remark. Some specializations of t to rational numbers preserving the Galois group

of f are given in Example 10.5.

In the proof of Theorem 5.3 it was noticed that the fixed field K 0 of the alternating
group An in N=Q.t/ is a rational function field, in which for odd n the prime divisors
P1 and P2 ramify and for even n the prime divisors P1 and P3 ramify. Thus the
field extensions K 0 =Q.t/ differ according to the parity of n. Since the calculations
in both cases behave identically, we will just give the details in the case of odd n.
Then P1 and P2 ramify in K 0 =Q.t/, say

P1 D Y21 ; ; P2 D Y22 ; (9.11)

and there exists a function y 2 K 0 with divisor Y1 Y1

2 . The divisor equality

P1 Y2
.t  1/ D D 12 D .y 2 /
P2 Y2

then proves the existence of c 2 Q satisfying

t  1 D cy 2 : (9.12)

Since y was only determined up to scalar multiples, it remains to find the class
of c in Q =.Q /2 . This can be achieved using the discriminant of the polynomial
f .t; X / in Theorem 9.4

D.f / D .n  1/n1 nn t n1 .1  t/ with  D .1/n.n1/=2 ; (9.13)

which may be computed from the well known formula for the discriminant of a
trinomial. Since the zeroes of f .1 C cy 2 ; X / generate the field N over K 0 D Q.y/
with group An , the discriminant of this polynomial must be a square, which implies

1 D D.f .1 C cy 2 ; X // D n.1  t/ D  nc;

2 2 2
9 Computation of Polynomials with Prescribed Group 75

where D2 means equality up to squares. Consequently we may choose c D n,

which then completely determines y. From (9.12) we obtain the minimal polyno-
mial
1
g.t; Y / D Y 2 C .t  1/ (9.14)
n
for y over Q.t/. Together with the corresponding computations in the case of even n
we thus obtain:

Theorem 9.5. The Galois group of the polynomial

f .1  ny 2 ; X / 2 Q.y/ŒX for n 1 .mod 2/; resp. (9.15)

1
f. ; X / 2 Q.y/ŒX for n 0 .mod 2/ (9.16)
1 C .n  1/y 2
with f .t; X / as in (9.10) and  WD .1/n.n1/=2 is the alternating group An . It
generates the geometric Galois extension N=Q.y/ with Q.y/ D K 0 in Theorem 5.3.

Since the calculations in the following examples run along the same lines, we
will not give all the details.

9.3 Polynomials with the Group Aut.A6 / and Related Groups

Apart from the case n D 6, the automorphism group of the simple group An is just
Sn . To complete the list of GA-realizations of alternating groups, we here construct
a polynomial with Galois group Aut.A6 / Š PL2 .9/. A suitable class vector can
easily be found by hand calculation, using the Group Atlas.

Proposition 9.6. The class vector .2B; 4C; 10A/ of Aut.A6 / is rationally rigid.

Proof. From the character table of Aut.A6 / in the Group Atlas it is immedi-
ately verified that the normalized structure constant of the rational class vector
C D .2B; 4C; 10A/ equals 1. Now let  2 ˙N .C/ and U WD h i. Since 1 2 S6
while 2 ; 3 … S6 , the translation formula for Z2 with ramification orders .1; 2; 2/
using (6.8) yields a generating 4-system
 1
'Z2 . / D .1 ; 1 2 ; 22 ; 32 /

of U 0 WD U \ S6 . One checks that 1 2 2B, 2 1 21 2 2C , 22 2 2A and 32 2 5A,

so U 0 contains elements from all three classes of involutions of S6 , and elements
of order 5. This excludes all proper subgroups of S6 as candidates for U 0 , and we
conclude U D Aut.A6 /. This implies that ˙N .C/ D ˙.C/ and hence by Corollary 5.6
that l.C/ D n.C/ D 1. t
u
76 I The Rigidity Method

Now as in (9.5) choose S to be the support of .t/ and .t  1/, this time with

P2 P1
.t/ D ; .t  1/ D ; (9.17)
P3 P3

and let NN  be the field extension in NS .G/ for G WD Aut.A6 /, parametrized by the
unique class of generating systems Œ 2 ˙.C/= Inn.G/. According to the Basic
N
Rigidity Theorem 4.8, NN  =G Q.t/ is defined over Q.t/, whence there exists a geo-
metric Galois extension N=Q.t/ with QN N D NN  . Furthermore let L be a root field
of N=Q.t/ of degree 10 and y a primitive element of L=Q.t/ with minimal polyno-
mial f . Then Gal.f / is the image of a permutation representation of G Š PL2 .9/
in S10 (via its action on the projective line over IF9 ), in which elements from the
classes 2B, 4C and 10A are seen to have permutation types .23 ; 14 /, .42 ; 12 /, .10/
respectively. By Theorem 9.1 the Pi split in L=Q.t/ as

P1 D Q21;1 Q1;2 ; P2 D Q42;1 Q2;2 ; P3 D Q10

3 ; (9.18)

with deg.Q1;1 / D 3 and deg.Q2;1 / D 2. So the different of L=Q.t/ has degree 18,
which shows that g.L=Q/ D 0. Thus L=Q is a rational function field, and there
exists a unique function x 2 L in the linear space of Q3 satisfying

Q2;1 Q2;2 Q1;1 Q1;2

D .x 2 Ca/; 2 D .p.x// D .x 2 C50x Cb/; 3 D.q.x// ; 4 D .r.x// ;
Q23 Q3 Q3 Q3

with a; b 2 Q and monic polynomials q.X /; r.X / 2 QŒX of degree 3, resp. 4. Here
for the final determination of x we have used the fact that the trace of p.X / cannot
vanish, since otherwise only imprimitive solutions will occur. As above this leads to
the divisor equations

P2 Q42;1 Q2;2
.t/ D D D .x 2 C a/4 .x 2 C 50x C b/ ;
P3 Q103
P1 Q21;1 ; Q1;2
.t  1/ D D D q.x/2 r.x/ :
P3 Q10 3

So there exist co ; c1 2 Q with

c0 t D .x 2 C a/4 .x 2 C 50x C b/; c1 .t  1/ D q.x/2 r.x/: (9.19)

Eliminating t from this shows c0 D c1 DW c, and there remains the polynomial iden-
tity in QŒX
.X 2 C a/4 .X 2 C 50X C b/ D q.X /2 r.X / C c: (9.20)
Formal differentiation of this with respect to X leads to a second identity, which
upon substitution into (9.20) and using that X 2 C a and q.X / are coprime admits
9 Computation of Polynomials with Prescribed Group 77

the following splitting:

10q.X / D 8X.X 2 C 50X C b/ C .X 2 C a/.2X C 50/;

10.X 2 C a/3 D 2q 0 .X /r.X / C q.X /r 0 .X /: (9.21)

Comparison of coefficients now gives a system of non-linear algebraic equations.

Only one of its six solutions is rational, with

a D 405; b D 945; c D 214 312 55 ; (9.22)

which hence leads to the polynomial with group G D Aut.A6 /.

Theorem 9.7. The Galois group of the polynomial

f .t; X / D .X 2  405/4.X 2 C 50X C 945/  214312 55 t (9.23)

is the group Aut.A6 / Š PL2 .9/. It generates a geometric Galois extension N=Q.t/
ramified over 1; 0; 1 for the class vector .2B; 4C; 10A/.

Obviously the fixed fields K1 ; K2 ; K3 of the index two subgroups S6 , M10 and
PGL2 .p/ in N=Q.t/ are also rational function fields Ki D Q.yi /, with
t 1
d1 y12 D t; d2 y22 D t  1; d3 y32 D ; (9.24)
t
for certain di 2 Q , since 1 2 S6 , 2 2 M10 , 3 2 PGL2 .9/. As 2 1 21 2 2C ,
the normalizer of 1 2 2B and hence also the decomposition groups of prime divi-
sors of N lying above P1 are already contained in S6 . So P1 splits in K1 =Q.t/,
which according to Theorem 9.1 entails d1 D2 1. Due to M10 A10 , the class of d2
modulo squares can again be determined using the discriminant of f .t; X /. This is
calculated from (9.19) up to (9.21) as

D.f / D N .f 0 .t; x// D N .10.x 2  405/3 q.x//

D 1010 .ct/6 .c.t  1//3 D 2136 3108 555 t 6 .t  1/3 ; (9.25)

where here N denotes the norm of L=Q.t/. This yields d2 D2 5. Finally, the com-
posite of the three fields Ki is Galois over Q.t/ with group Z2  Z2 , which forces
d1 d2 d3 D2 1. So we may choose

d1 D 1 and d2 D d3 D 5: (9.26)

Since P1 splits in K1 =Q.t/ and precisely the two prime divisors of P1 in K1

— these are the numerators of .y1 ˙ 1/ — ramify in N A6 =K1 , the function field
K 0 WD N A6 is also rational. A generating equation of K 0 =K1 then has the shape
y1 C 1
d´2 D with d D 5; (9.27)
y1  1
78 I The Rigidity Method

since from (9.25) we find as above that d D2 5. In conclusion this leads to:
Corollary 9.8. Specializing t in the polynomial f .t; X / in Theorem 9.7 to y12 ,
1 C 5y22 resp. 1=.1  5y32/ yields polynomials fi .yi ; X / 2 Q.yi /ŒX with groups

Gal.f1 .y1 ; X // Š S6 ; Gal.f2 .y2 ; X // Š M10 ; Gal.f3 .y3 ; X // Š PGL2 .9/:

(9.28)
 2 2
5´ C1
Specialization of t to 5´2 1 yields a polynomial g.´; X / 2 Q.´/ŒX with

Gal.g.´; X // Š A6 : (9.29)

The latter gives a GA-realization of A6 over Q.

Remark. In a completely similar fashion, GA-realizations over Q for L2 .p 2 / for all
p ˙2 .mod 5/, p ¤ 2, may be obtained, by starting from the rationally rigid class
vector C D .2B; 4C; 10A/ of the automorphism group PL2 .p 2 / Š Aut.L2 .p 2 //
(see Theorem II.7.7 with Remark).
For future use (see for example Chapter II.6 and II.7) we state the following
observation made in the proof of Corollary 9.8 as a special lemma:
Lemma 9.9. Let K D k.t/ be a rational function field of one variable and N=K a
geometric Galois extension with group G. Further let H be a normal subgroup of
G such that the fixed field N H =k has genus 0. Now assume that for a generator 
of an inertia group in G over a prime divisor P 2 IP.K=k/ of degree 1 ramified in
N=K we have
NG ./ hH; i: (9.30)
Then N H =k is a rational function field.
Remark. By Theorem 6.2 the assumption g.N H =k/ D 0 is always satisfied if for
example G=H is a dihedral group and only three prime divisors P 2 IP.K=k/ are
ramified in N H =K.

9.4 Polynomials with the Mathieu Groups M12 and M11

The final example to be given concerns the two smallest sporadic simple groups,
namely the Mathieu groups M12 and M11 . We compute polynomials for the Galois
extensions found in Theorem 6.12. Let N=Q.tQ/ denote the geometric Galois exten-
sion with Gal.N=Q.tQ// Š M12 for the class vector C D .4A; 4A; 10A/, and L
the fixed field of M11 . Since this forms a root field of N=Q.tQ/, a polynomial
with group M12 may be obtained as minimal polynomial of a primitive element
of L=Q.tQ/. From Proposition 6.11 we known that l.C/ D 2 and l V .C/ D 1 for
V D h.12/i, hence the first two ramified prime divisors P1 and P2 are permuted by
Gal.Q.N tQ/=Q.tQ//, while P3 remains fixed. Now let k.tQ/ be the splitting field of P1
in this field extension, of degree 2 over Q.tQ/.
9 Computation of Polynomials with Prescribed Group 79

The classes 4A and 10A have permutation types .42 ; 14 /, .10; 2/ in the permu-
tation representation on the cosets of an intransitive M11 , so Theorem 9.1 gives the
ramification behavior of Pi in kL=k.tQ/ as

Pi D Q4i;1Qi;2 for i D 1; 2; P3 D Q10 2

3;1 Q3;2 ; (9.31)

with deg.Qi;1 / D 2 for i D 1; 2. Since P1 ; P2 are conjugate in k.tQ/=Q.tQ/, there

exists a function t 2 Q.tQ/ with

P1 P2
D .t C a/; D .t  a/; and a2 2 Q: (9.32)
P3 P3
Thereby, the divisor .t/ of t is uniquely determined. By the proof of Theorem 6.12
the function field L=Q is rational, hence there exist x 2 L and monic polynomials
q; r 2 kŒX such that

Q3;2 Q1;1 Q1;2

D .x/; D .q.x// ; D .r.x// : (9.33)
Q3;1 Q23;1 Q43;1

Denoting images under the non-trivial automorphism of k.tQ/=Q.tQ/ by a bar, we thus

moreover have
Q2;1 Q2;2
D .q.x//
N ; D .Nr .x// : (9.34)
Q23;1 Q43;1
Together with (9.31) up to (9.34) this implies the divisor equations
 
P1 Q41;1 Q1;2 q.x/4 r.x/
.t C a/ D D 10 2 D ;
P3 Q3;1 Q3;2 x2
 
P2 Q42;1 Q2;2 N 4 rN .x/
q.x/
.t  a/ D D 10 2 D :
P3 Q3;1 Q3;2 x2

So there exist c; cN 2 k  with

c.t C a/x 2 D q.x/4 r.x/; N  a/x 2 D q.x/

c.t N 4 rN .x/; (9.35)

which after elimination of t first forces c D cN 2 Q , and then results in the polyno-
mial identity
2acX 2 D q.X /4 r.X /  q.X N /4 r.X
N / (9.36)
in kŒX . As in the proof of Theorem 9.7, differentiation of this equation, substitution
N / are necessarily coprime now yields
of ac and observing that q.X / and q.X

10q.X /3 D 4qN 0 .X /Nr .X /X C q.X

N /Nr 0 .X /X  2q.X
N /Nr .X / (9.37)

and its conjugate in kŒX =QŒX . By comparison of coefficients we obtain a sys-

N r; rN ,
tem of non-linear algebraic equations in the 12 unknown coefficients of q; q;
to which we may add the requirement tr.q q/N D 6. The latter finally determines
80 I The Rigidity Method

the function x, which was until now fixed only up to scalar multiples (note that the
N D 0 leads only to imprimitive field extensions). This system of equations
case tr.q q/
possesses just oneppair of conjugatepsolutions in a quadratic extension field k of Q,
namely in k D Q. 5/. With d WD 5 this is given by
3 9
q.X / D X 2 C .3 C d /X  d;
5 25
12 336 216 108 81
r.X / D X 4 C .8 C d /X 3 C .30 C d /X 2 C . C d /X C : (9.38)
5 25 5 5 25
Now t was only fixed up to scalar multiples, so in the equation

2ctx 2 D q.x/4 r.x/ C q.x/

N 4 rN .x/

following from (9.35) we may still replace t by ct since c 2 Q . Then x is a zero of

the polynomial
1
f .t; X / D q.X /4 r.X / C q.X
N /4 rN .X /  tX 2 :
2
Substituting q and r from (9.38) then yields:

Theorem 9.10. The following polynomial has Galois group M12 over Q.t/:
3348 9 35559 8 5832 7
f .t; X / D X 12 C 20X 11 C 162X 10 C X C X C X
5 52 5
84564 857304 5 807003 4 1810836 3
 3 X6  X C X C X
5 54 55 55
511758 2 2125764 531441
 X C XC  tX 2 : (9.39)
56 57 58
By suitable specialization of t in f .t; X / one easily obtains a polynomial with
group M11 :

Corollary 9.11. Writing the polynomial in Theorem 9.10 as f .t; X / D h.X /  tX 2 ,

the polynomial
x 2 h.X /  h.x/X 2
g.x; X / WD 2 Q.x/ŒX (9.40)
X x
has Galois group M11 over Q.x/.

Proof. Let x be a zero of f .t; X / in the splitting field N . Then since t D h.x/=x 2
the field N is generated over Q.t; x/ D Q.x/ by a zero of
 
1 h.x/
h.X /  2 X 2 2 Q.x/ŒX
X x x

and hence by g.x; X /. t

u
9 Computation of Polynomials with Prescribed Group 81

Remark. With the help of Corollary 9.3 one easily finds infinite families of spe-
cializations a; b 2 Q such that Gal.f .a; X // Š M12 resp. Gal.g.b; X // Š M11 . For
example this is true for all a 1 .mod 66/ and all b 1 .mod 133/ (see for exam-
ple Matzat (1987), IV, 6.4).

A collection of further polynomials for almost simple groups constructed by this

method is contained in the Tables in the Appendix A.1 and A.2.
82 I The Rigidity Method

10 Specialization of Geometric Galois Extensions

By the Hilbert irreducibility theorem every polynomial f .t; X / 2 k.t/ŒX over a

number field k possesses infinitely many specializations t 7! a 2 k with canonically
isomorphic Galois group (compare Chapter IV.1). We will now study local proper-
ties like reality questions and ramification behavior for Galois extensions obtained
in this way.

10.1 Local Structure Stability

By construction, irreducible generating polynomials f D f .t; X / of Galois exten-

sions NN =Q.t/
N are defined over a number field k and absolutely irreducible. Hence
they remain irreducible if k is replaced by its completion with respect to any valua-
tion. In this section let k denote a complete metric field. If for a 2 k the numerator
divisor P of .t  a/ is not contained in the ramification locus S of N=k.t/, then by
the Dedekind’s Theorem 9.2 for fa WD f .a; X / 2 kŒX we have
Q
Gal.fa / D f .GD .P=P// Q
Š GD .P=P/; (10.1)

where PQ denotes an extension of P to N . Then obviously the absolute Galois group

N N
k D Gal.k=k/ of k, where k denotes a maximal separable extension field of k,
acts on the zeroes of fa and so yields a continuous homomorphism

a W k
Q
! Gal.fa / D f .GD .P=P// Gal.f / Š G (10.2)
from k in Gal.f / Š G, unique up to inner automorphisms of G. Its kernel consists
of the fixed group in k of the field Na WD N P Q generated by fa over k. Now let S
denote the set of those a 2 IP1 .k/ for which the numerator divisor of .t  a/ (resp.
. 1t / when a D 1) lies in S, and

 WD f a j a 2 IP1 .k/nS g: (10.3)

Equipping the finite orbit space = Inn.G/ with the discrete topology we have:
Theorem 10.1 (Saltman (1982)). Let k be a complete metric field and f .t; X / 2
k.t/ŒX an irreducible polynomial. Then the map
W IP1 .k/nS ! = Inn.G/; a 7! a; (10.4)
is well defined and continuous; in particular the field extensions Na =k generated
by f .a; X / do not change if the corresponding a are sufficiently close.
Proof. In the case of a complex archimedean valuation the assertion is trivially sat-
isfied since k D 1. Since in the case of a real archimedean valuation the zeroes
of fa WD f .a; X / 2 kŒX depend continuously on a, the decomposition type of fa
can only change when a passes through an element of S . Finally, in the case of
ultrametric valuations, the decomposition type and the coefficients of the factors
10 Specialization of Geometric Galois Extensions 83

of fa again depend continuously on a by Hensel’s Lemma (see for example Artin

(1967), Ch. 2, Thm. 5a). For sufficiently close a; b 2 k the factors of fa and fb gen-
erate the same field by Krasner’s Lemma (see loc. cit., Ch. 2, Thm. 9), which proves
Na D Nb . t
u
Example 10.1. In the case f .t; X / D X 2  t 2 Qp .t/ŒX with p ¤ 2 the projective
p1g divides up into four
line IP1 .Qp /nf0; open and closed subsets belonging to the
four fields Qp . a/ with a 2 Q 2
p =.Qp / . t
u
In the next section we will first study some applications in the case of real
archimedean valuations.

10.2 Reality Questions

Let NN =Q.t/
N be a Galois extension which together with its Galois group is defined
over a real number field k. With respect to a real archimedean valuation of k,
i.e., an embedding of k into IR, by Theorem 10.1 to each connected component
O of IP1 .k/ n S there belongs a well defined conjugacy class of involutions of
G D Gal.NN =Q.t//,
N whose elements upon specialization of t to a 2 Q take the role
of complex conjugation a WD jNa in the residue fields Na =k. When G has trivial
center, this conjugacy class can be described explicitly without problems. For this
let Na WD N P Q be the residue class field of N=k.t/ modulo a fixed extension PQ of
the numerator divisor P 2 IP.k.t/=k/ of .t  a/ and

a
Q
W GD .P=P/ ! Ga (10.5)

the canonical epimorphism introduced in (9.4). Then we first have:

Proposition 10.2. The complex conjugation commutes with the epimorphism a.
The action of a on Ga D Gal.Na =k/ is given by

a W Ga ! Ga ; a ./ 7! a .

/; (10.6)
N
where  denotes the complex conjugation in QN=k.t/ from the remark following
Theorem 2.2.
N we have by Proposition 3.1
Proof. For NN WD QN

WD Gal.NN =k.t// D Gal.NN =N /  Gal.NN =Q.t//:

N

Let P N denote an extension of P Q on NN and NQ D resp. ND the decomposition fields

of P=P resp. P=P. The corresponding fields of constants are kQ and k since ND =k
N Q N
is regular. Using the translation theorem of Galois theory we obtain
Q
Gal.k=k/ Q
Š Gal.NQ D =ND / Š Gal.N=.N \ ND // D GD .P=P/;

which shows that kQ D Na and Gal.k=k/

Q D Ga .
84 I The Rigidity Method

N̄
ÑD

N̄D
N
ND
Q̄(t)
N ∩ ND ρ
Q̄ (t)

k̃(t)
k(t)
Fig. 10.1 Complex conjugation in residue class fields

N
Now let  be the complex conjugation in Q.t/=Q.t/ and Q the unique extension
N N Q N
of  onto N with P D P (compare with Theorem 1.6 and the Remark after The-
orem 2.2). Then Q has trivial restriction to N \ ND and by Theorem 1.6 acts like
Q
O D  on GD .P=P/ N
Š Gal.NN =NN D / with NN D WD Q.N \ND /. Since j
Q kQ D jkQ D a
these actions commute with the canonical epimorphism a . t
u

Using Theorem 1.6 we may deduce from this proposition the following result,
which in the case of three ramification points originates from Serre (1988):

Theorem 10.3 (Fried and Dèbes (1990)). Let N=k.t/ be a geometric Galois exten-
N
sion with group G over a real number field k, whose ramification locus S  IP1 .Q/
N N
of QN D N has the standard form as in Figure 1.2. Assume furthermore that the
prime divisor P 2 IP.k.t/=k/ belonging to the base point a WD P0 stays inert in
N=k.t/. Then the corresponding residue field extension Na =k is Galois with

Ga WD Gal.Na =k/ Š G: (10.7)

Moreover, the complex conjugation a 2 Ga acts on Ga via

a . /
a
D a .

/ (10.8)

with the canonical epimorphism a in (10.5) from G onto Ga , and

1 1
1
  D .2r ; : : : ; 11 ; 2rC1
1
; : : : ; .s1 /s1 2rC1 /: (10.9)

In the case Z .G/ D 1 this uniquely determines the conjugacy class of a 2 Ga

depending on Π.
10 Specialization of Geometric Galois Extensions 85

Proof. Let PQ denote the unique extension of P onto N . Then we have GD .P=P/
Q Š
N N
G and hence (10.7) follows. Now identifying Gal.N =Q.t// with Gal.N=k.t// we
obtain the generators i of inertia subgroups of N=k.t/ by i D .' ı  /.i /, with
the canonical epimorphisms  ; ' from the Hurwitz classification (4.5). With this
the action of a on Ga follows directly from Theorem 1.6, using (10.6) and the
Remark after Theorem 2.2. Finally in the case Z .G/ D 1 by (10.8) and (10.9) the
action of a is uniquely determined by a and  . Hence the conjugacy class of a
in Ga Š G is already determined by Œ . t
u
N ,
Remark. Obviously, the field Na in Theorem 10.3 is real, i.e., a subfield of Q
precisely when a D 1.

By the Structure Stability Theorem 10.1 the conjugacy class of a 2 Ga remains

constant as long as the reference point a 2 IP1 .k/ belonging to P varies in the
connected component O of IP1 .IR/ lying between as WD Ps and a2rC1 WD P2rC1
(compare with Figure 1.2). Otherwise the transformation formulae for the complex
conjugation  change, and with this in general also the class of a . But at least the
following result holds, which was first proved in the case s D 3 by Serre (1992),
8.4.3:

Proposition 10.4. The conjugacy class of the complex conjugation a1 .a / 2 G
for a 2 IP1 .k/ remains unchanged when a passes through a point ai 2 S for which
the generator i of the inertia group has odd centralizer order in Gal.NN  =Q.t//.
N

Proof. We give the proof for the passage from a 2 .as ; a2rC1 / to b 2 .a2rC1 ; a2rC2 /.
The complex conjugation with respect to the base point b satisfies

b . /
b
D a .
2rC1
/

and thus
1 1 .
.b / 1 a/
b
2rC1 D 2rC1 D 2rC1
a
:
So Q 2rC1 WD a1 .a / b1 .b / lies in the centralizer of 2rC1 in G, and thus has
odd order n WD o.Q 2rC1 /. The dihedral group
1 1
h a .a /; b .b /i Š Dn
1 1
hence contains a single class of involutions, which proves that a .a / and b
.b /
are conjugate in Dn and so also in G. t
u

Remark. If we have s D 3, r D 0 and Z .G/ D 1 in Proposition 10.4, then we even

have Q j D j for j D 1; 2; 3, and it suffices to assume that o.i / is odd.

Example 10.2 (Serre (1988)). If the Galois extension N =G Q.t/ with Z .G/ D 1 and
three real ramification points possesses real specializations Na =G Q, then it follows
that G Š S3 . Indeed, if a D 1 for the standard configuration, then by the above
remark we have 1 D b1 .b / and 3 D c1 .c / with b 2 .a1 ; a2 / and c 2 .a2 ; a3 /.
Thus G, being generated by two involutions, is isomorphic to a dihedral group Dn .
86 I The Rigidity Method

From this the assertion follows since the class vectors of type .2; 2; n/ of Dn are not
rational for odd n > 3. t
u

From Theorem 10.3 we may further deduce:

Corollary 10.5. Q N  .t/ is a field of definition of NN  =G Q.t/

N precisely when there
exists an involution  2 G satisfying   D  .

Proof. If Q N
N  .t/ is a field of definition of NN  =G Q.t/, then the complex conjugation
 acts as an inner automorphism on G, and there exists an involution  2 G with
  D  . Conversely from the existence of such an involution  it follows that the
fixed field N of hi is a geometric Galois extension of Q N D NN  . u
N  .t/ with QN t

Example 10.3 (Fried and Dèbes (1990)). If in Corollary 10.5 all ramification points
are real, then
1 1 1
  D .11 ; .21 /1 ; : : : ; .s1 /s1 1 / D  :

In particular G is then generated by the involutions

i WD 1    i for i D 1; : : : ; s

(with s D ). Conversely each generation by involutions G D h1 ; : : : ; s i with

 WD s , 1 WD s 1 and i WD i 1 i for i D 2; : : : ; s yields the sufficient condi-
tion   D  for real fields of definition (with respect to the standard configuration
in Figure 1.2). t
u

For a study of the ramification in the residue class field extensions the reduc-
tion of constants has to be utilized. This also yields restrictions on the primes of Q
ramified in minimal fields of definition.

10.3 Ramification in Minimal Fields of Definition

N extend-
Let o denote the valuation ring and p the valuation ideal of a valuation of Q
ing the p-valuation for some rational prime p. The corresponding residue homo-
morphism  W o ! IF N p can be extended to a place

N ! IP1 .IF
} W IP1 .Q/ N p /; a 7! a;
N (10.10)

by setting }.a/ D 1 for a … o. A finite subset S  IP1 .Q/ N DQ N [ 1 is called }-

N
stable (or also p-stable) if } is injective on S , i.e., if jS j D jS j. With these nota-
tions the following theorem holds, which was first proved by Grothendieck (1971),
Exp. XIII, Cor. 2.12 (see also Popp (1970), Satz 12.1).
N and S 
Theorem 10.6 (Grothendieck (1971)). Let p be an extension of .p/ onto Q
1 N
IP .Q/ a finite p-stable subset. Then the maximal factor groups of p-prime order
10 Specialization of Geometric Galois Extensions 87
alg N
of the algebraic fundamental groups 1 .IP1 .Q/nS
alg N p /nSN/ are
/ and 1 .IP1 .IF
isomorphic via the specialization homomorphism:
.p/ N
1 .IP1 .Q/nS
.p/ N p /nSN/:
/ Š 1 .IP1 .IF (10.11)

The isomorphism is uniquely determined by p up to an inner automorphism.

Thus also the Hurwitz classification of Galois extensions with p-prime order
of the Galois group carries over in a canonical way from Q.t/ N N p .t/. If S, SN
to IF
N
denote the sets of prime divisors in Q.t/ N p .t/ corresponding to S resp. SN,
resp. IF
and MN S , MN SN the maximal algebraic Galois extensions unramified outside of S,
.p/ .p/

N and with degree not divisible by p, then the corresponding Galois groups
resp. S,
are isomorphic by Theorem 10.6:

N
Gal.MN S.p/ =Q.t// Š Gal.MN SN.p/ =IF
N p .t//: (10.12)

N
In particular the valuation ideal pt of Q.t/ corresponding to the t-functional valua-
tion given by p with trivial value at t possesses an inert extension pO t on MN S .
.p/

Corollary 10.7. Let G be a finite group with p-prime order. Then the map defined
by

N SN .G/;
NSN W ˙s .G/= Inn.G/ ! N Œ 7! NN .p/ WD .MN SN.p/ /ker. / ; (10.13)

is surjective and compatible with reduction of constants, i.e., the restriction pN t of

the t-functional extension pO t of p on MN S.p/ satisfies

NN .p/ D NN  pN t : (10.14)

From Theorem 10.6 and Corollary 10.7 one may deduce restrictions on the ram-
ification in minimal fields of definition. This is contained in the following important
result:
Theorem 10.8 (Beckmann (1989)). Let G be a finite group,  2 ˙s .G/ and NN  2
N
NS .G/ with S D S for  D Gal.Q.t/=Q.t//. Then any prime p 2 IP ramified in the
N  either divides the group order jGj or is such
field of constants of k .t/ D Q.t/
that S is not p-stable.
Proof. Let p be a prime which does not divide jGj and for which S is p-stable.
In a first step we show that all prime divisors p of p are unramified in k =k A
with A D Aut.G/. For this, let pt be the functional extension of p on k A .t/ and
pN t an extension of pt on NN  . Since pN t remains inert in NN  =Q.t/
N by Corollary 10.7,
N
the inertia group I of pN t =pt in the Galois extension N =k A .t/ together with
N
G D Gal.NN  =Q.t// generates a subgroup Q of the splitting group D of pN t =pt .
Now I is normal in D (see for example Nagata (1977), Thm. 7.3.2), hence I is
also normal in Q with I \ G D 1, which shows that Q D I  G. Hence the fixed
Q
field k.t/ N
of Q constitutes a field of definition of NN  =Q.t/ with the property that
88 I The Rigidity Method

Q
pN t jkQ is unramified in k=k N
 A . Since any automorphism of Q.t/=k .t/ acts as inner
automorphism on G, k lies inside k. Q So all prime divisors of p and also of p are
unramified in k =k A .
In the second step let NN  denote the composite of all NN ı for ı 2 , G  WD
Gal.NN  =Q.t//
N and Π the classifying class of generating s-systems of G  : NN  D
NN   . Since NN  =Q.t/ is Galois, Q.t/ is a field of definition for NN  =Q.t/
N by The-
orem 3.5, and we have k.  /A D Q. Further, k  is obtained as the composite of
all kı with ı 2  and thus coincides with the Galois closure of k =Q. Since all
prime divisors of jG  j already divide jGj, and S does not change since S D S,
we conclude from the first step that p is unramified in k  =Q and hence a fortiori
in k =Q. t
u
Remark. If in the case s D 3 we choose the ramification locus f0; 1; 1g, the assump-
tion of p-stability in Theorem 10.8 is automatically satisfied.

10.4 Ramification in Residue Fields

Let N=k.t/ be a geometric Galois extension over a number field k and p 2 IP.k/.
We say that N=k.t/ has good reduction modulo p if the t-functional extension pt
of p on k.t/ possesses an inert extension pQ t on N=k.t/, such that the residue field
extension N pQ t =kp.t/ is geometric and moreover we have

Gal.N pQ t =kp.t// Š Gal.N=k.t//: (10.15)

If then the norm of p is coprime to the group order, and the ramification locus is
p-stable, it follows from Corollary 10.7 that

IF N /pN t
N p .N pQ t / D .QN (10.16)
N . Conversely we have only the following weaker
for each extension pN t of pQ t onto QN
result, which was first proved by Beckmann (1991), Prop. 2.3, using different meth-
ods:
Proposition 10.9. Let N=k.t/ be a geometric Galois extension, whose Galois group
has trivial center and p-prime order, and p 2 IP.k/ an extension of .p/, with respect
to which the ramification locus S  IP1 .k/ of N=k.t/ is p-stable. Then N=k.t/ has
good reduction modulo p.
Proof. From the p-stability of the ramification locus we conclude with Corol-
lary 10.7 that the t-functional valuation ideal pt of k.t/ has an extension pN t onto
N with
QN
N /pN t =IF
Gal..QN N Q.t//
N p .t// Š Gal.QN= N Š Gal.N=k.t// D G:

Now let NN WD QN
N , WD Gal.NN =k.t// and D and I the splitting group and the
N
inertia group of pN t in . Since pN t is inert in NN =Q.t/, N N N
I and G WD Gal.N =Q.t//
10 Specialization of Geometric Galois Extensions 89

generate a subgroup Q of D . This forces I E Q and I \ GN D 1, from which it

follows that Q D I  G. N Thus I is a subgroup of C .G/ N D Gal.NN =N /, and pN t
does not ramify in N=k.t/.
We have D  G, N
N so the fixed field of D is an intermediate field of Q.t/=k.t/,
say N N D
D k .t/, and we set N WD k N . Since I then also lies in Gal.NN =N 0 /,
0 0 0

pN t has relative degree n WD jGj in N 0 =k 0 .t/. Since D \ Gal.NN =N / D Gal.NN =N 0 /,

the relative degree of pN t in N 0 =N is equal to 1. So the restriction pQ t of pN t has residue
degree n in N=k.t/. This shows that pt is inert in N=k.t/, and we have

Gal.N pQ t =kp.t// Š Gal.N=k.t//:

N /pN t . If this
N p .N pQ t / D .QN
It remains to show that N pQ t =kp is regular, i.e., that IF
were not the case, N pQ t =kp.t/ and also N pQ t =kp.t/, with the restriction pQ 0t of pN t
0 0

to N 0 , would contain a non-trivial extension of constants, which since Z .G/ D 1

contradicts the fact that

Gal.NN pN t =N 0 pQ 0t / CN .Gal.NN pN t =IF

N p .t///

with N WD Gal.NN pN t =kp.t//. t

u

The assumption Z .G/ D 1 in Proposition 10.9 is necessary, as can be seen in

the following simple example:

Example 10.4. The Galois extension N=Q.t/ with group Z2 generated by x 2 D 3t

obviously has bad reduction modulo 3, although the prime 3 does not divide the
group order, and moreover S D f0; 1g is 3-stable. u
t

Under the assumption of good reduction modulo p the ramification in the spe-
cialized field extension Na =k can be described.

Theorem 10.10 (Beckmann (1991)). Let N=k.t/ be a finite geometric Galois exten-
sion with group G over k Q, N where QN N D NN  2 NS .G/, and with the set
1 N
S D fa1 ; : : : ; as g  IP .Q/ of zeroes of the Pi 2 S. If N=k.t/ has good reduc-
tion modulo p 2 IP.k/ and S is p-stable, then for a 2 IP1 .k/nS there exists at most
one i with ei WD ordp .a  ai / ¤ 0, and the inertia group of an extension pQ of p on
the residue class field Na is either trivial or it is generated by the canonical image
e
'a ./ in Ga D Gal.Na =k/ of an element  2 G conjugate to i i in G.

The proof would lead us too far astray, so we refer to the original paper Beck-
mann (1991), Sect. 3. This theorem implies the following weak but explicit version
of the Hilbert Irreducibility Theorem:

Corollary 10.11. In Theorem 10.10 we have Ga Š G, if all  2 Ce with e WD

.e1 ; : : : ; es / generate the group G, where C 2 Cl.G/s denotes the class vector of
G with  2 ˙.C/.
90 I The Rigidity Method

Example 10.5 (Beckmann (1991))). For a prime n we consider the polynomial

f .t; X / D X n  t.nX  n C 1/ 2 Q.t/ŒX

with group Sn from Theorem 9.4. If we specialize the variable t to a 2 Qnf0; 1g

and if there exists a prime p > n such that ordp .a/ is prime to n and ordp .a  1/ is
odd, then f .a; X / 2 QŒX has Galois group Sn and in particular remains irreducible.
This follows with the calculations in (9.2) immediately from Corollary 10.11, since
for prime n any group containing an n-cycle is primitive, and a primitive group
containing a transposition equals already Sn by a theorem of Jordan. t
u
II Applications of Rigidity

It is perhaps astonishing that the rigidity criteria developed in the previous chapter
do indeed apply to a wide variety of finite groups, in particular to non-abelian finite
simple groups. In fact, it turns out that most of these groups satisfy some form of
the rigidity criterion, at least over the field Qab .t/.
Assuming the classification of finite simple groups we know that the non-abelian
finite simple groups are the alternating groups, the groups of Lie type, and the 26
sporadic simple groups. It was already shown in Theorem I.5.3 that all alternating
groups occur as Galois groups over Q.t/. The classical groups of Lie type com-
prise the linear, unitary, symplectic and various orthogonal groups defined over finite
fields IFq . By construction they all possess a natural matrix representation over IFq .
Belyi (1979, 1983) found an ingenious way to make use of such matrix represen-
tations (Theorem I.5.10). His result, which proves that all simple classical groups
of Lie type occur as geometric Galois groups over suitable abelian number fields,
can be considered as the first major step in realizing nonsolvable groups as Galois
groups. A different proof of his results for the classical groups, using the classifi-
cation of irreducible pseudo-reflection groups, was later given by Walter (1984).
We present Belyi’s proof for the general linear groups in the first Paragraph. For the
other classical groups we avoid the rather heavy computations with Steinberg gen-
erators necessary in Belyi’s approach and rather follow Walter’s approach, making
use of a nice effective version of Belyi’s criterion due to Völklein (1998).
To treat the exceptional groups of Lie type we have to employ much deeper meth-
ods, in particular the Deligne-Lusztig character theory for groups of Lie type and
the classification of the finite simple groups. In Paragraphs 4 and 5 the character
theoretic form of the rigidity criterion is shown to apply to most of the exceptional
groups of Lie type, a result which is due to Malle (1988b, 1992) in good character-
istic and to Lübeck and Malle (1998) in bad characteristic different from 2.
The results for Galois realizations of simple groups over Q.t/ at present are far
from complete. We present results of Malle (1996) and Reiter (1999) concerning
the classical groups in Paragraph 6 and 7, and those of Malle (1988b, 1992) on the
exceptional groups of Lie type in Paragraph 8. See also some further results obtained

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 91

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_2
92 II Applications of Rigidity

by different methods in Chapter III.10. While the classical groups can again be
treated with the Belyi Criterion, the Deligne-Lusztig theory is needed for the other
cases. In this context we also prove some stronger results on realization of groups
of automorphisms of classical groups over Qab .t/.
Finally, the 26 sporadic groups are covered by ad hoc arguments in the final para-
graph. It turns out that all but possibly the Mathieu group M23 have GA-realizations
even over the field of rational numbers. The proper references for the sporadic
groups are given at the beginning of Paragraph 9.
For better reference we collect in Paragraph 10 the G- and GA-realizations
proved in this chapter.
1 The General Linear Groups 93

1 The General Linear Groups

The first paragraph is devoted to the case of the linear groups, where the calculations
are given in some detail. In this case the verification of rigidity comes down to
matrix computations, see also Matzat (1987), II.5.2.
Before presenting this elementary proof we first collect some notions and results
from the theory of linear algebraic groups. While this theory is not needed for the
treatment of the general linear groups in the subsequent sections, it still provides a
more natural language for the formulation of certain facts. In the later paragraphs,
the remaining series of groups of Lie type will then be considered under this point
of view.

1.1 Groups of Lie Type

It is most natural to consider the classical groups from the point of view of algebraic
groups. For a short introduction see for example Carter (1985). The details may be
found in Steinberg (1967) or Carter (1989).
Let G be a connected reductive algebraic group over the algebraic closure kN of a
finite field k D IFq . Let T be a maximal torus of G, contained in a Borel subgroup
B. Then B is the semidirect product of its unipotent radical U D Ru .B/ with T.
There exists a unique opposite Borel subgroup B determined by the property that
B \ B D T. With its unipotent radical U D Ru .B / we then have B D U T.
Let Hom.T; kN  / denote the group of algebraic homomorphisms from T to the group
of algebraic automorphisms of the additive group of k, N which is isomorphic to kN  .

The minimal non-trivial subgroups of U and U normalized by T are connected
unipotent algebraic groups of dimension one, isomorphic to the additive group of
N The action of T on such a subgroup X defines an element ˛ 2 Hom.T; kN  /, a
k.
so-called root. Distinct subgroups give rise to distinct roots, thus the minimal T-
invariant subgroups X of U and U may be indexed by roots: X D X˛ Š k. N We
write
˚ ˙ .G/ WD f˛ j there exists X  U˙ with X D X˛ g;
for the sets of positive resp. negative roots. For any positive root ˛, ˛ is also
a root. It can be shown that there exists a surjective homomorphism SL2 .k/ N !
hX˛ ; X˛ i onto the group generated by the corresponding root subgroups. The
preimage of T \ hX˛ ; X˛ i is a maximal torus of SL2 .k/, N hence isomorphic to
N  _ N 
k . This defines an element ˛ 2 Hom.k ; T/, the coroot of ˛. The sets ˚.G/ D
˚ C .G/ [ ˚  .G/ of roots and ˚ _ .G/ D f˛ _ j ˛ 2 ˚.G/g of coroots define a root
system in Hom.T; kN  / ˝ IR.
Now assume that G is defined over a finite field IFq and let F be the correspond-
ing Steinberg endomorphism F W G ! G. The group of fixed points G WD GF is
a finite group of Lie type. A result of fundamental importance in this area is the
following (see for example Steinberg (1968)):
94 II Applications of Rigidity

Theorem 1.1 (Lang–Steinberg). Let G be a connected algebraic group over an

algebraically closed field of positive characteristic and F W G ! G a surjective
homomorphism such that GF is finite. Then the map

L W G ! G;  7!  1 F ./;

is surjective.

As one of the many applications we sketch a proof of the following well-known

result which will be of importance in proving generation:

Proposition 1.2. Let G be a connected reductive algebraic group, F W G ! G a

surjective homomorphism such that GF is finite and T an F -stable maximal torus
Q ! ŒG; G be the simply-connected covering of the derived group of
of G. Let ' W G
G (see Borel et al. (1970), E-27). Then G WD GF is generated by '.GQ F / together
F
with T .

Indeed, since G is connected the map  7! Q 1 F ./ Q (where Q 2 GQ is such

Q D ) defines an isomorphism of G='.G
that './ Q F / with ker.'/=.1  F / ker.'/.
Q with '.T/
Now let TQ be the F -stable torus in G Q D T. This also contains the cen-
tral subgroup ker.'/ and is connected. Thus by the same arguments we obtain that
T =.T \ '.GQ F // Š ker.'/=.1  F / ker.'/, where T WD TF . But then G='.G QF/Š
T =.T \ '.GQ F // as claimed.

The action of F on G induces an action on the Dynkin diagram of the root sys-
tem ˚ associated to G. If this action is trivial, we call F an untwisted, otherwise a
twisted Steinberg map. If we start from simple simply-connected groups G of types
An ; Bn ; Cn or Dn , then among the resulting finite groups we find all the univer-
sal classical groups listed in Table 1.1. The first column gives the Lie notation for
G D GF , the second column identifies these groups with classical matrix groups.
The cases of B2 in characteristic 2 with F inducing the exceptional graph automor-
phism, and of D4 with F inducing the triality automorphism will be considered in
the section on exceptional groups.

Table 1.1 Classical groups

G R.G/ NGL.V / .R.G//

An .q/sc SLnC1 .q/ n1 SLnC1 .q/ GLnC1 .q/
Bn .q/sc Spin2nC1 .q/ 2 6 j q; n  3 O2nC1 .q/ Z .GL2nC1 .q// GO2nC1 .q/
Cn .q/sc Sp2n .q/ n2 Sp2n .q/ CSp2n .q/
Dn .q/sc SpinC
2n .q/ n4 C
2n .q/ COC
2n .q/
2
An .q/sc SUnC1 .q/ n2 SUnC1 .q/ Z .GLnC1 .q 2 // GUnC1 .q/
2
Dn .q/sc Spin
2n .q/ n4 
2n .q/ CO
2n .q/
1 The General Linear Groups 95

Each of the above groups possesses a natural matrix representation R on a vec-

tor space V over IFq (respectively over IFq 2 for the unitary groups) whose dimen-
sion is given by the lower index in the classical notation for G. This representation
R W G ! Aut.V / is not necessarily faithful. The image of G under R is given in the
fourth column. The last column describes the full normalizer of R.G/ in the general
linear group on V . This will be needed for the application of Belyi’s criterion. The
correctness of the entries in the last column will be shown in the subsections treat-
ing the individual cases. In the third column we have indicated restrictions on q and
n which can be made so as to avoid duplications stemming from generic isomor-
phisms. Under these restrictions, still all simple classical groups of Lie type occur
as a non-abelian composition factor of one of the groups listed (see for example
Carter (1989), 11.3 and 14.5).
In the next sections we present the elementary rigidity proof for GLn .q/ in some
detail.

1.2 Rigidity for GLn .q/

Let k D IFq be the finite field with q D p m elements, where p is prime. We first
prove rigidity for the general linear group GLn .IFq / which, following the usual con-
vention in finite group theory, we will denote by GLn .q/.
Denote by G0 D SLn .k/ N the simple, simply connected algebraic group of type
An1 over the algebraic closure kN of k, so G 0 WD G0 F is the special linear group
over IFq . The subgroup of diagonal matrices in G 0 forms a maximal torus T . With
respect to a suitable orthonormal basis f1 ; : : : ; n g of Hom.T; kN  / ˝ IR the set of
roots is given by
˚ WD fi  j j 1 i; j n; i ¤ j g:
We identify the root i  j with the ordered pair .i; j / of indices. In the case of
SLn .q/, the root subgroup Xr corresponding to r D .i; j / just consists of the matri-
ces
Xi;j D fxi;j .u/ WD Id CuIi;j j u 2 IFq g ;
where Ii;j denotes the matrix whose only non-zero entry lies at position .i; j / and
equals 1.
The following commutator formulae follow immediately from the explicit
description of the root subgroups:
8
ˆ
<xi;l .uv/ if i ¤ l, j D k;
Œxi;j .u/; xk;l .v/ D xk;j .uv/ if i D l, j ¤ k; (1.1)
:̂
1 if i ¤ l, j ¤ k.

Next, for ˛ 2 ˚ and u 2 IF

q let

w˛ .u/ WD x˛ .u/x˛ .u1 /x˛ .u/; h˛ .u/ WD w˛ .u/w˛ .1/1 : (1.2)

96 II Applications of Rigidity

Then the h˛ .u/ lie in T , the w˛ WD w˛ .1/ are contained in NG 0 .T / and for v 2 IFq
we have

w˛ .u/x˛ .v/w˛ .u/ D x˛ .u2 v/; (1.3)

w˛ xˇ .u/w˛1 D xw˛ .ˇ / .˙u/;
1
h˛ .v/ xˇ .u/h˛ .v/ D xˇ .v <ˇ;˛> u/; (1.4)

where u 2 IFq ; v 2 IF

q . Here < ˇ; ˛ >WD 2.˛; ˇ/=.ˇ; ˇ/, where .˛; ˇ/ denotes the
inner product on the root space.
We denote by h.v/ the element diag.v; 1; : : : ; 1/ of GLn .q/. It is easily verified
that

h.v/1 x1;i .u/ h.v/ D x1;i .v 1 u/; h.v/1 xi;1 .u/ h.v/ D xi;1 .vu/
h.v/1 w1;i .u/ h.v/ D w1;i .uv 1 / (1.5)

for 2 i n, while h.v/ commutes with all xi;i C1.u/ for i > 1. Due to the different
forms of defining relations (1.1), (1.3), we have to distinguish between the rank one
case and the general case:

Proposition 1.3. Let v be a generator of IF

q.
(a) The group GL2 .q/, q  3, is generated by 1 WD h.v/ and 2 WD x1;2 .1/w1;2 .
(b) The group GLn .q/, n  3, is generated by the elements 1 WD x1;2 .1/ and
Q
2 WD wn1;n    w1;2 h.v/ D . n1
i D1 wni;ni C1 /h.v/.

Proof. Let H WD h1 ; 2 i.

We first consider (a). There we have

2 1 21 D x1;2 .1/ w1;2 h.v/ w1;2 .1/ x1;2 .1/

D x1;2 .1/ h.v/ w1;2 .v 1 / w1;2 .1/ x1;2 .1/ by (1.5)
1 1
D h.v/ x1;2 .v / h1;2 .v /x1;2 .1/ by (1.2)
1
D h.v/ h1;2 .v / x1;2 .v  1/ by (1.4);

so the commutator
Œ1 ; 2 1 21 D h.v 1 / x1;2 .1  v/ h.v/ x1;2 .v  1/
1v .v  1/2
D x1;2 . / x1;2 .v  1/ D x1;2 .u/; where u WD ¤ 0;
v v
is contained in H . Conjugating x1;2 .u/ by powers of 1 we obtain

1i x1;2 .u/1i D h.v i / x1;2 .u/ h.v i / D x1;2 .v i u/;

so by the definition of v all of X1;2 is contained in H , whence

x1;2 .1/2 D w1;2 2 H:

1 The General Linear Groups 97

1
But w1;2 X1;2 w1;2 D X2;1 by (1.3), hence also X2;1  H . It is well known that
SL2 .q/ is generated by X1;2 ; X2;1 so the group H contains SL2 .q/, and since
det.1 / D v we conclude that H D GL2 .q/.
In case (b) we first verify that (1.4) here takes the form

w2;3 .1/x1;2 .1/w2;3 D x2;3 .1/ x3;2 .1/ x2;3 .1/ x1;2 .1/ x2;3 .1/x3;2 .1/x2;3 .1/
D x2;3 .1/ x3;2 .1/ x1;2 .1/ x1;3 .1/ x3;2 .1/ x2;3 .1/
D x2;3 .1/ x1;2 .1/ x1;2 .1/ x1;3 .1/ x2;3 .1/ D x1;3 .1/

using (1.2) and (1.1), and similarly

w1;2 .1/ x1;3 .1/ w1;2 D x1;2 .1/ x2;3 .1/ x1;3 .1/ x1;2 .1/ D x2;3 .1/;

while all wi;i C1 with i  3 commute with x1;2 .1/. With this we obtain

21 1 2 D h.v 1 / w1;2 .1/    wn1;n .1/ x1;2 .1/ wn1;n    w1;2 h.v/
D h.v 1 / w1;2 .1/ w2;3 .1/ x1;2 .1/ w2;3 w1;2 h.v/
D h.v 1 / w1;2 .1/ x1;3 .1/ w1;2 h.v/ D h.v 1 / x2;3 .1/ h.v/ D x2;3 .1/;

and then inductively

2i 1 2i D xi C1;i C2 .1/ 2 H; for i D 0; : : : ; n  2: (1.6)

Finally,

2nC1 1 2n1 D 21 xn1;n .1/ 2

Dh.v 1 / w1;2 .1/    wn1;n .1/ xn1;n .1/ wn1;n    w1;2 h.v/
Dh.v 1 / w1;2 .1/    wn2;n1 .1/ xn;n1 .1/ wn2;n1    w1;2 h.v/
Dh.v 1 / w1;2 .1/    wn3;n2 .1/ xn;n2 .1/ wn3;n2    w1;2 h.v/
D : : : D h.v 1 / xn;1 ..1/n / h.v/ D xn;1 ..1/n v/ 2 H:

Since xn;1 .v/ D xn;1 .v/1 we certainly have xn;1 .v/ 2 H . From the commutator
relations (1.1)
Œx1;i .1/; xi;i C1 .1/ D x1;i C1 .1/
we inductively see that x1;i C1 .1/ 2 H for i D 2; : : : ; n  1, using (1.6). Also, since
n > 2, for all u 2 IFq we have

Œx1;n .1/; Œxn;1 .v/; x1;2 .u/ D Œx1;n .1/; xn;2 .vu/ D x1;2 .vu/;

so again inductively, starting with u D 1, it follows that X1;2  H . As in (1.6) this

forces Xi C1;i C2  H for i D 1; : : : ; n  2, and by repeated application of (1.1) all
Xi;j with i < j are seen to lie in H . Finally

ŒXi;n ; xn;1 .v/ D Xi;1 ; for i D 2; : : : ; n  1;

98 II Applications of Rigidity

and (1.1) allows to conclude that indeed all root subgroups are contained in H .
Again, since SLn .q/ is generated by the groups Xi;j this shows that SLn .q/ H ,
and as det.2 / D v, it then follows that H D GLn .q/. t
u

1.3 Galois Realizations for Linear Groups

Using the Rigidity Criterion of Belyi we obtain the following result from Proposi-
tion 1.3:

Theorem 1.4. The groups GLn .q/, q D p m , .n; q/ ¤ .2; 2/, possess G-realizations
over abelian number fields k.n; q/ Qab for the class vector .Œ1 ; Œ2 ; Œ21 11 /
with i , i D 1; 2, as defined in Proposition 1.3.

Proof. Obviously the group G WD GLn .q/ acts irreducibly in its natural matrix
representation. Moreover the elements 1 in Proposition 1.3 all possess an n  1-
dimensional eigenspace for the eigenvalue 1. Hence by the criterion of Belyi (The-
orem I.5.10) we have l.Œ1 ; Œ2 ; Œ21 11 / D 1 since the condition on the normal-
izer is trivially satisfied.
For the application of the Basic Rigidity Theorem I.4.8 it remains to check the
normalizer condition (N) (see the remark after Theorem I.4.8). Let N WD NG .h1 i/
be the normalizer of the inertia group over the first ramification point, and further-
more let E be the n  1-dimensional eigenspace for the eigenvalue 1 of 1 . Since
clearly E is invariant under N , this defines a canonical homomorphism

 W N ! GL.V =E/ Š GL1 .q/ Š IF

q;

which maps the center Z .G/ surjectively onto IF q . Hence ker./ is a complement
to Z .G/ in N , and N satisfies the normalizer condition (N). t
u

Remark. The trivial case of the solvable group GL2 .2/ Š S3 of order six follows
for example from Theorem I.5.3.

Passing to suitable subgroups and quotients we obtain:

Corollary 1.5. The groups SLn .q/, PGLn .q/ and Ln .q/ possess G-realizations over
the same fields of definition k D k.n; q/ Qab as for GLn .q/ in Theorem 1.4.

Proof. Denote by N=k.t/ the geometric Galois extension for GLn .q/ given by The-
orem 1.4, and by K 0 the fixed field of SLn .q/. So Gal.N=K 0 / Š SLn .q/, and clearly
we are done if we can show that K 0 is a rational function field. The extension
K 0 =k.t/ is geometric with

Gal.K 0 =k.t// Š GLn .q/= SLn .q/ Š IF

q;

and ramification occurs at most at the three prime divisors of degree 1 ramified
in N=k.t/. But for n D 2 the element 2 already lies in SLn .q/, while for n > 2
1 The General Linear Groups 99

we have 1 2 SLn .q/. Hence precisely two divisors ramify in K 0 =k.t/. From the
Hurwitz genus formula it follows that the ramification is of order q  1 for both of
them. Thus the genus g.K 0 / is equal to 0 and the prime divisors lying above the two
ramification points have degree one, which proves that K 0 is rational (see the proof
of Theorem I.5.3 for a similar argument).
As PGLn .q/ D GLn .q/=Z .GLn .q// and Ln .q/ D SLn .q/=Z .SLn .q// are both
quotients of groups already realized as geometric Galois groups over k, the remain-
ing assertions follow trivially. t
u

Since Aut.L2 .p// D PGL2 .p/ we may also deduce the following:

Corollary 1.6. For all primes p > 3 the groups L2 .p/ possess GA-realizations over
Qab .

Remark. The corresponding result for all linear and unitary groups Ln .p/ and
Un .p/ for n  3 and primes p > 2 will be shown in Corollary 6.6.

This completes the study of linear groups. It should be clear from the above proof
that indeed all groups between GLn .q/ and Ln .q/ may be realized as Galois groups
over the original field of definition, i.e., all factor groups of intermediate groups
SLn .q/ H GLn .q/. In fact, the fixed field of H is rational, being contained in
a rational function field of transcendence degree one. In particular this applies to all
groups isogenous to SLn .q/.

Remark. In the papers of Belyi (1979, 1983) all series of classical groups of Lie
type are shown to be rigid by using similar methods, the matrix computations being
replaced by calculations with Steinberg generators and relators. Since these compu-
tations become quite involved in some cases, we prefer to present a more elegant,
though less elementary proof which relies on the classification of finite irreducible
pseudo-reflection groups.
100 II Applications of Rigidity

2 Pseudo-Reflection Groups and Belyi Triples

In order to prove rigidity for the remaining classical groups we first present an ele-
gant effective version of Belyi’s criterion, which was proved by Völklein (1998),
completing partial results by Malle (1996) and Reiter (1999), and the classification
of irreducible pseudo-reflection groups.

2.1 Groups Generated by Pseudo-Reflections

For a finite dimensional vector space V a non-trivial element  2 GL.V / is called

a pseudo-reflection if it leaves a hyperplane pointwise fixed. If  is diagonal-
izable, so has an eigenvalue a ¤ 1, then it is called a homology. If moreover
a D 1, then  is called a reflection. The non-diagonalizable pseudo-reflections
are also called transvections. By their definition pseudo-reflections naturally occur
in applications of the Theorem of Belyi. The irreducible linear groups generated by
pseudo-reflections are known by work of Wagner (1978, 1980), Kantor (1979) and
Zalesskiı̌ and Serežkin (1980). We will state their classification, starting with the
case of imprimitive groups.
We recall that a linear group G GL.V / is called a primitive linear group if it
does not stabilize any non-trivial direct sum decomposition of the underlying vector
space V . Let now m; n > 1 and G.m; 1; n/ be the group of all monomial n  n-
matrices whose entries are m-th roots of unity, of order mn nŠ. For the following
easy result see for example Wagner (1978), Lemma 2.2.

Proposition 2.1. Let G GLn .q/ be an irreducible imprimitive linear group gen-
erated by pseudo-reflections. Then G is conjugate to a subgroup of G.m; 1; n/ for
some mj.q  1/, the pseudo-reflections in G have order dividing m or 2 and the nor-
malizer in GLn .q/ of G is again imprimitive unless n 2 or .m; n/ 2 f.3; 3/; .2; 4/g.
In the latter cases jNGLn .q/ .G/j divides 648.q  1/ respectively 1152.q  1/.

In the primitive case we subdivide the classification according to the type of

generating pseudo-reflections.

Theorem 2.2 (Kantor (1979)). Let n  4, q D p m , and H an irreducible subgroup

of SLn .q/ generated by transvections. Then one of the following holds:
(a) H D SLn .q/Q or H D Spn .q/ Q for qjq,
Q or
(b) H D SUn .q/ Q for qQ 2 jq,
or p D 2 and one of
(c) n is even, H D GOn.˙/ .q/ Q for qjq,
Q or
(d) n is even, H D SnC1 or H D SnC2 , or
(e) n D 6, H D 31 :U4 .3/:22 , or
(f) q  4 and H is imprimitive.
2 Pseudo-Reflection Groups and Belyi Triples 101

Theorem 2.3 (Wagner (1978)). Let n  3, q D p m and H a primitive subgroup

of GLn .q/ generated by non-involutionary homologies. Then one of the following
holds:
Q
(a) SLn .q/ H GLn .q/Q for qjq,
Q or
Q
(b) SUn .q/ H GUn .q/ Q for qQ 2 jq,
or n 4 and H Š GUn .2/. In the latter cases the homologies have order 3.

In the following result we write W .F4 /; W .H4 /; : : : for the Coxeter groups of
types F4 ; H4 ; : : : in their natural reflection representation.

Theorem 2.4 (Wagner (1980), Zalesskiı̌ and Serežkin (1980)). Let n  3, q D p m

for p ¤ 2, and H a primitive subgroup of GLn .q/ generated by reflections. Then
one of the following holds:
(a) SLn .q/Q H GLn .q/ Q for qjq,
Q or
(b) SUn .q/ Q H GUn .q/ Q for qQ 2 jq, or
.˙/
(c) n .q/ Q < H GOn.˙/ .q/ Q for qjq,
Q or
(d) H D SnC1 and p6 j.n C 1/, or H D SnC2 and pj.n C 2/, or
(e) n D 3, H 2 fA5  2; L2 .7/  2; 3:A6  2g or .H; p/ D .3:A7  2; 5/,
(f) n D 4, H 2 fW .F4 /; Œ26 :S5 ; W .H4 /; Œ26 :S6 g or .H; p/ D .4: L3 .4/:22 ; 3/;
(g) n D 5, H D O5 .3/  2,
(h) n D 6, H 2 f61 :U4 .3/:22 ; W .E6 /g,
(i) n D 7, H D W .E7 / D 2: O7 .2/,
( j) n D 8, H D W .E8 / D 2: SOC 8 .2/.

For the purely group theoretical proofs of these theorems the reader is referred to
the original papers.

2.2 An Effective Version of Belyi’s Criterion

In this section we present an effective version of the Belyi criterion, which asserts
that for certain class vectors of length three the structure constant is in fact non-zero
(and thus equal to one by the Theorem I.5.10 of Belyi).

Lemma 2.5. Let k be a field and a1 ; : : : ; an ; b1 ; : : : ; bn 2 k  . Then there exist unique-

ly determined lower (resp. upper) triangular matrices ;  2 GLn .k/ with diagonal
entries a1 ; : : : ; an (resp. b1 ; : : : ; bn ) in this order such that  1   Id has identical
rows.

Proof. Let  be the lower triangular matrix with diagonal entries a1 ; : : : ; an and
off-diagonal entries xij , where xij D 0 if j > i . Let  denote the matrix such that
all rows of   Id are equal to .y1 ; : : : ; yn /. The assertion of the Lemma is now
equivalent to the following: there exist unique solutions for the xij ; yj such that 
is upper triangular with diagonal entries b1 ; : : : ; bn . This in turn is equivalent to the
102 II Applications of Rigidity

unique solvability of the system of equations

´i yj C xij D 0; ´i yi C ai D bi ; for 1 i n; 1 j i  1; (2.1)

where we have set

i 1
X
´i WD ai C xi l : (2.2)
lD1

This can be solved recursively on i to give the unique solutions

b1    bi 1 a1    ai
yi D .bi  ai /; ´i D ; xij D ´i yj (2.3)
a1    ai b1    bi 1
for 1 i n, 1 j i  1, where moreover we have

X
i
b1    bi
1C yl D : (2.4)
a1    ai
lD1

Indeed, for i D 1 this immediately follows from (2.1). So now assume that i > 1.
Adding up the equations ´i yj C xij D 0 for j D 1; : : : ; i  1 we get

´i .y1 C : : : C yi 1 / C ´i  ai D 0

which, using (2.4) inductively, gives the asserted value of ´i . Then the values of yi
and xij are obtained from (2.1), and (2.4) follows by induction. t
u
A triple .1 ; 2 ; 3 / of elements i 2 GLn .k/ with 1 2 3 D 1 is called a Belyi
triple if 1 is a pseudo-reflection (i.e., rk.1  Id/ D 1) and the group h1 ; 2 i gen-
erated by the triple is an irreducible subgroup of GLn .K/. We can now state the
existence theorem for Belyi triples.
Theorem 2.6 (Völklein (1998)). Let f; g 2 IFq ŒX be coprime monic polynomials
of degree n with non-vanishing constant coefficient. Then there exists a Belyi triple
.1 ; 2 ; 3 / in GLn .q/ such that the characteristic (and the minimal) polynomial of
21 (resp. 3 ) equals f .X / (resp. g.X /).
Proof. Let IF N q be the algebraic closure of IFq , and denote the zeroes of f .X / (resp.
g.X /) in IF N q by a1 ; : : : ; an (resp. b1 ; : : : ; bn ). By Lemma 2.5 there exist matrices
;  2 GLn .IF N q / with characteristic polynomials f .X /, g.X / respectively such that
1
   Id has rank 1. Assume that G WD h; i is reducible on V WD IF N nq , with non-
trivial G-invariant subspace 0 < W < V . Since rk. 1   Id/ D 1 the element  1 
acts as identity on W or on V =W . In either case  D  on that space, showing
that  and  have a common eigenvalue, which contradicts our assumption on f; g.
Thus  D .1 ; 2 ; 3 / with 1 WD . 1 /1 , 2 WD  1 , 3 WD  is a Belyi triple
in GLn .IF N q /. The intersection of any eigenspace of 2 or 3 with the hyperplane of
fixed points of 1 is a G-invariant subspace. Hence, since G is irreducible, all such
eigenspaces are at most 1-dimensional which means that f; g are also the minimal
polynomials of 21 ; 3 .
2 Pseudo-Reflection Groups and Belyi Triples 103

It remains to descend to the finite field IFq . Let F W GLn .IF N q / ! GLn .IF N q / be
the Frobenius-morphism raising each matrix entry to its q-th power. Then F . / D
N q /. Since f; g 2 IFq ŒX are
.F .1 /; F .2 /; F .3 // is again a Belyi triple in GLn .IF
F -invariant the elements i and F .i / have the same characteristic polynomial for
i D 2; 3, which equals the minimal polynomial by the above observation. Further-
more, since rk.1  Id/ D 1 it follows that also 1 and F .1 / have the same minimal
polynomial. By the theory of Jordan normal forms this shows that i is conjugate
to F .i / for i D 1; 2; 3. By the Theorem I.5.10 of Belyi there exists a  2 GLn .IF N q/

with i D F .i /. But by the Theorem 1.1 of Lang-Steinberg any  2 GLn .IFq / can N
be decomposed as  D F . /1 with a suitable  2 GLn .IF N q /. Then F .  / D   is
i i
F -invariant for i D 1; 2; 3, hence  is a Belyi triple in GLn .q/ with the prescribed


characteristic polynomials. t
u
Remark. The condition in Theorem 2.6 of f; g being coprime is actually necessary
and sufficient for the existence of associated Belyi triples, see Völklein (1998).

2.3 Imprimitive and Symmetric Groups

The following result shows that for a wide class of Belyi triples the generated group
is primitive.
Proposition 2.7. Let  D .1 ; 2 ; 3 / be a Belyi triple in GLn .q/ and assume that
G WD h i is imprimitive with respect to a decomposition IFnq D V1 ˚ : : : ˚ Vr with
r > 1.
(a) If r < n then tr.2 / D tr.3 / D 0, r divides gcd.o.2 /; o.3 //, the orders
of 2 ; 3 are bounded above by r.q n=r  1/, and the characteristic polynomials of
2 ; 3 are of the form

Y
r
f . i X /; where r
D 1: (2.5)
i D1

(b) If r D n then G G.q  1; 1; n/, o.1 / D 2, either 2 or 3 has order divid-

ing n.q  1/ and trace 0 while the other has order at most n2 .q  1/=4, and the
characteristic polynomials of 2 ; 3 are of the form

X n  c1 ; .X k  c2 /.X nk  c3 /; where c1 ; c2 ; c3 2 IF

q; (2.6)

for some 1 k n  1.
Proof. The G-invariant decomposition IFnq D V1 ˚ : : : ˚ Vr induces a permutation
representation r W G ! Sr of G into the symmetric group on r letters whose image
is transitive (since G acts irreducibly). Since rk.1  Id/ D 1 we necessarily have
r .1 / D 1 in case (a). So both r .2 /; r .3 / are r-cycles, which proves the first
part of (a). Moreover the r-th power of i stabilizes each of V1 ; : : : ; Vr , for i D 2; 3.
This proves (a).
104 II Applications of Rigidity

In case (b) by definition we have G G.q  1; 1; n/ Š IF q o Sn . As r .G/

is transitive and r .1 / is either the identity or a transposition, at least one of
r .2 /; r .3 / must be an n-cycle, so its order divides n.q  1/ and its trace van-
ishes, while the other has at most two orbits. It is clear that elements of G.q 1; 1; n/
which map to a permutation of type .k/.n  k/ have characteristic polynomial as
stated. t
u

By similar considerations we can handle symmetric groups as follows:

Proposition 2.8. Let  D .1 ; 2 ; 3 / be a Belyi triple in GLn .q/ with 12 D 1 and
assume that G WD h i is a symmetric group Sm with m 2 fn C 1; n C 2g. Then one
of o.2 /; o.3 / is bounded above by m, the other by m2 =4.

Proof. Firstly, as 1 is a pseudo-reflection of order 2, it represents a transposition

of Sm , m 2 fn C 1; n C 2g. Thus 2 ; 3 both map to permutations with at most two
orbits. In fact, since transpositions have odd signature and the product 1 2 3 D 1,
one of 2 ; 3 must be an m-cycle. This proves the assertion. t
u

2.4 Invariant Forms

In this section we construct quadratic and bilinear forms invariant under the Belyi
triples corresponding to suitable polynomials f; g in Theorem 2.6. This allows an
easy recognition of the group generated by  . For properties of spaces with forms
see for example Aschbacher (1986), Ch. 7. We will make use of the following easy
fact (see for example Wagner (1978), Lemma 2.1).

Proposition 2.9. Let G GLn .q/ be an irreducible subgroup containing a pseudo-

reflection. Then G is absolutely irreducible.

The only case where the existence of invariant forms can not be proved abstractly
from Belyi’s criterion is that of quadratic forms in characteristic 2. Here Völklein
(1998), Lemma 5 and 6, showed that his effective form of Belyi’s criterion can
be used to write down such forms explicitly (Parts (b) and (c) of the following
result). This allows us to distinguish between Belyi triples generating the orthogonal
resp. symplectic groups in characteristic 2.

Theorem 2.10. Let f; g 2 IFq ŒX be coprime monic polynomials of degree n with

non-vanishing constant coefficient and G the group generated by a Belyi triple con-
structed from f; g as in Theorem 2.6.
N q of f and g are of the form
(a) Let n D 2m C 1 be odd. If the roots in IF
1
a1 ; : : : ; am ; 1; am ; : : : ; a11 ; 1
b1 ; : : : ; bm ; 1; bm ; : : : ; b11 (2.7)

then G leaves invariant a non-degenerate quadratic form.

2 Pseudo-Reflection Groups and Belyi Triples 105

N q of f and g are of the form

(b) Let n D 2m. If the roots in IF
1
a1 ; : : : ; am ; cam ; : : : ; ca11 ; 1
b1 ; : : : ; bm ; cbm ; : : : ; cb11 (2.8)

then G leaves a non-degenerate alternating bilinear form invariant up to scalars.

(c) Let n D 2m and q even. If the roots in IF N q of f and g are of the form (2.8)
with c D 1 then G leaves a non-degenerate quadratic form invariant if and only if
one of the ai or bi is equal to 1.
N q of f and g are of the form
(d) Let n D 2m and q odd. If the roots in IF
1
a1 ; : : : ; am1 ; c; c; c 2 am1 ; : : : ; c 2 a11 ; 1
b1 ; : : : ; bm ; c 2 bm ; : : : ; c 2 b11 (2.9)

then G leaves a non-degenerate quadratic form invariant up to scalars.

Proof. Let ;  be the matrices constructed in the proof of Lemma 2.5 with respect
to the ordering of the roots as in the statement, and  WD  1 . Let  be the matrix
whose only non-zero entries are yi in position .i; n  i C 1/. It is invertible since
all yi are non-zero. An easy calculation then shows that  1 .t  Id/ has iden-
tical rows. But if Q  Id has identical rows for some Q 2 GLn .q/, this also holds
for Q1  Id, thus  1 .t  Id/ has identical rows. Furthermore, we obtain that
 1  t  is lower triangular with diagonal entries an1 ; : : : ; a11 , and  1  t  is
upper triangular with diagonal entries bn1 ; : : : ; b11 .
In case (a) it follows from (2.3) that under the condition (2.7) the entries yi of
  Id satisfy

b1    bni b1    bi 1
yni C1 D .bni C1  ani C1 / D .ai  bi / D yi
a1    ani C1 a1    ai

for 1 i m. Thus the matrix  is symmetric and invertible. By the preceding

considerations  0 WD  1  t ,  0 WD  1  t  are lower (upper) triangular with the
same diagonal entries as  (resp. ) and the product . 0 /1  0 has identical rows. By
the uniqueness assertion in Lemma 2.5 this implies  0 D ,  0 D , so

 t  D ;  t  D ;

i.e., G WD h; i leaves invariant the symmetric bilinear form defined by  since G
is absolutely irreducible by Proposition 2.9.
For (b) it follows from (2.3) that the entries yi of   Id satisfy yi D yni C1
for 1 i m. Thus now the matrix  is skew-symmetric and invertible. Let Q WD
p 1 p 1
c , Q WD c . Then as before the uniqueness part of Lemma 2.5 shows that
h;
Q Q i GLn .IF N q / leaves invariant the skew-symmetric form . ; / defined by . But
then ;  2 GLn .q/ leave . ; / invariant up to scalar multiples.
In (c), let V D IF2n q and Q W V ! IFq a quadratic form left invariant by G. Then

. ; /0 W V  V ! IFq ; .v; w/0 WD Q.v C w/ C Q.v/ C Q.w/;

106 II Applications of Rigidity

is an alternating bilinear G-invariant form on V , hence by the absolute irreducibility

of G (Proposition 2.9) must be equal up to scalars to the form . ; / defined by the
matrix  from (b). Without loss of generality we may assume that they are actually
equal. With respect to the standard basis e1 ; : : : ; en of V we may write
P P
Q..u1 ; : : : ; un /t / D niD1 u2i Q.ei / C i <j ui uj .ei ; ej /
Pn P
D i D1 u2i Q.ei / C m i D1 ui uni C1 yi ; (2.10)

so Q is determined by the values Q.ei /, 1 i n. We let  WD  1 . Then

Q..ei // D Q.ei C yi enC1 / D Q.ei / C yi2 .Q.enC1 / C 1/

where we have set enC1 WD e1 C: : : Cen . Thus Q is -invariant only if Q.enC1 / D 1.

Furthermore, .ei ; ej / D 0 for i; j  m C 1, so for i  m C 1 we have Q..ei // D
Q.ei / if and only if

X
n
.1 C ai2 /Q.ei / C 2
xj;i Q.ei / D 0 :
j Di C1

We may assume that ai ¤ 1 for m C 1 i n, so the above has the only solu-
tion Q.emC1 / D : : : D Q.en / D 0. If moreover all roots bi are different from 1
then similarly we obtain Q..ei // D Q.ei / for 1 i m if and only if Q.e1 / D
: : : D Q.em / D 0. But then Q.enC1 / D y1 C: : : Cym D 1 Cb1    bm =.a1    am / ¤ 1
by (2.4), so Q is not -invariant. This proves the first part of the assertion.
Now assume that bm D 1 and set

b1    bm X
m
Q..u1 ; : : : ; un /t / WD u2m C ui uni C1 yi : (2.11)
a1    am
i D1

Since Q.ei / D 0 for i  m C 1 we have that Q..ei // D Q.ei / for i  m C 1.

Also, Q.enC1 / D 1, so Q is -invariant. Finally the definition of Q shows that
Q. 1 .ei // D Q.ei / for 1 i m, so

.Q/.ei / D 1 .Q/.ei / D .Q/.ei / D Q. 1 .ei // D Q.ei /

for 1 i m, hence Q is also -invariant, thus G-invariant.

In (d) we show that G leaves a non-degenerate quadratic form invariant if c D 1,
from which the general assertion follows by considering Q WD c 1 , Q WD c 1 .
Let  be the n  n-matrix with only non-zero entries yi in position .i; n  i C 1/
for 1 i m  1, ym in position .m; m C 1/; .m C 1; m/ and 2=´mC1 in position
.m C 1; m C 1/. Then  is symmetric and invertible. Moreover it is easily checked
using (2.3) that  1 .t  Id/ has identical rows while  1  t  is lower triangu-
1
lar with diagonal entries a1 ; : : : am1 ; 1; 1; am1 ; : : : ; a11 , and  1  t  is upper
1 1
triangular with diagonal entries bn ; : : : ; b1 . By the same argument as in (b) this
proves that G leaves invariant the non-degenerate quadratic form defined by . u t
3 The Classical Groups 107

3 The Classical Groups

In this paragraph we verify rigidity for the remaining classical groups over finite
fields, i.e., the unitary, symplectic and orthogonal groups, using the classification of
irreducible pseudo-reflection groups.

3.1 Rigidity for GUn .q/

The V be a finite dimensional vector space over IFq 2 and  the generator of
Gal.IFq 2 =IFq /. Then on V there exists a unique Hermitian form with respect to
 up to equivalence (see Aschbacher (1986), (21.6.2)). The subgroup of GLn .q 2 /
leaving such a form invariant is called the general unitary group GUn .q/ on V .
Proposition 3.1. Let q D p m , n  3, a 2 IF
q 2n
of multiplicative order q n  .1/n
and
Y
n1
i
f .X / WD .X  a.q/ /; g.X / WD .X  1/n : (3.1)
i D0

If  D .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then

G WD h i D GUn .q/.
Proof. By definition we have f; g 2 IFq 2 ŒX and f; g are coprime. Thus G WD h i
GLn .q 2 / is irreducible. Also, G is not contained in GLn .q/ Q for any proper subfield
IFqQ of IFq 2 since f … IFqQ ŒX . Let  be the matrix constructed in the first part of the
proof of Theorem 2.10. Then Q WD . 1  t /q is lower triangular with the same
diagonal entries as  (not necessarily in the same order), and similarly for Q WD
. 1  t /q . Furthermore Q 1 Q  Id has rank 1. Thus Q WD ..Q 1 /
Q 1 ; Q 1 ; Q / is a
Belyi triple in GLn .IF N q / whose elements have the same characteristic polynomials
as for the Belyi triple  . By the Theorem of Belyi Q is conjugate to  by some
 2 GLn .IF N q /. Thus

 D . /1 . t /q ;  D . /1 . t /q 

and G fixes some non-degenerate Hermitian form. This shows that G GUn .q/.
Since 3 2 SLn .q 2 / and the determinant

Y
n1
i n .1/n /=.qC1/
det.2 / D a.q/ D a.q
i D0

of 2 has multiplicative order q C 1 the pseudo-reflection 1 is a homology of

order q C 1. Since moreover by construction 2 is semisimple while 3 is unipo-
tent, the group G acts primitively by Proposition 2.7. Hence the normal subgroup
H of G generated by the conjugacy class in G of the homology 1 acts irreducibly.
According to Proposition 2.1 either H is also primitive or n D q C 1 D 3. In the
108 II Applications of Rigidity

latter case it is easily checked that G D GU3 .2/. In the former case by Theorem 2.3
the only remaining possibility is G D H D GUn .q/. t
u
Thus we obtain:
Theorem 3.2. The groups GUn .q/, q D p m , n  3, possess G-realizations over
abelian number fields k.n; q/ Qab for the class vector .Œ1 ; Œ2 ; Œ3 / defined
in Proposition 3.1. The groups SUn .q/, PGUn .q/ and Un .q/ possess G-realizations
over the same fields of definition.
Proof. This is a straightforward application of the Belyi criterion to the class vector
C D .Œ1 ; Œ2 ; Œ3 /. The elements in the first class have an .n  1/-dimensional
eigenspace. The normalizer of G D GUn .q/ in the general linear group GLn .q 2 / is
generated by G and Z .GLn .q//, so by Theorem I.5.10 we have l.C/ D 1. Let N WD
NG .h1 i/ be the normalizer of the inertia group above the first ramification point,
and let E be the 1-eigenspace of 1 . As in Theorem 1.4 this defines a canonical
homomorphism

 W N ! GU.V =E/ Š GU1 .q/ Š IF

q2
=IF
q;

which maps the center Z .G/ surjectively onto IF q2

=IF
q . Hence the kernel ker./ is
a complement to Z .G/ in N , and N satisfies the normalizer condition (N). Theo-
rem I.4.8 then proves the first part of the assertion.
The descent to the other unitary groups is achieved by the same arguments as in
the proof of Corollary 1.5. t
u
Remark. In a completely similar fashion it can be shown that GLn .q/ possesses a
G-realizations over abelian number fields by choosing

Y
n1
i
f .X / D .X  aq /; g.X / D .X  1/n ;
i D0

where a is a generator of IF

q n , thus giving an alternative proof of Theorems 1.4
and 1.5.

3.2 Rigidity for CSp2n .q/

Let V be an even-dimensional vector space over a finite field IFq . All non-singular
alternating bilinear forms on V are equivalent (see Aschbacher (1986), (21.6.1)).
The subgroup of GL.V / leaving invariant such a form is the symplectic group on V ,
denoted by Sp2n .q/, where dim.V / D 2n. Since Sp2 .q/ Š SL2 .q/ we will assume
n  2 throughout this section. The symplectic group coincides with the image of the
universal Chevalley group of type Cn .q/sc in its natural matrix representation (see
Table 1.1). The conformal symplectic group CSp2n .q/ is by definition the subgroup
of elements of GL.V / leaving the symplectic form invariant up to scalars.
3 The Classical Groups 109

Proposition 3.3. Let q D p m , n  2, .n; q/ ¤ .2; 2/, a 2 IF

q 2n
of multiplicative order
n C1
.q n C 1/.q  1/, c WD aq , and

Y
2n1
i
f WD .X  aq /; g.X / WD .X  1/n .X  c/n : (3.2)
i D0

If  D .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then

G WD h i D CSp2n .q/.
n n 2n1
Proof. We have aq D ca1 for c D aq C1 2 IFq , so the roots a; aq ; : : : ; aq of
f 2 IFq ŒX are of the form (2.8). By Theorem 2.10 this implies that G WD h i
CSp2n .q/. Also, G is not contained in GL2n .q/ Q for any proper subfield IFqQ of IFq
p p 1
by the choice of a. As thep orders of c2 ; c 3 are coprime, G is primitive by
Proposition 2.7 since o. c2 / D .q n C 1/ > n2 .q  1/. Thus the normal subgroup
H of G generated by the transvections in G acts irreducibly.
If p is odd then by Theorem 2.2 we have H  Sp2n .q/. If p D 2 then by
Lemma 2.1 either n D 4, H G.2; 1; 4/, or H is one of the primitive groups listed
in Theorem 2.2. Moreover if H G.2; 1; 4/ then jGj divides 648.q  1/ which is
only possible in the excluded case q D 2. Furthermore, H is neither an orthogonal
group by Theorem 2.10(c), nor a symmetric group by Proposition 2.8. By compar-
ison of orders, 31 :U4 .3/:22 cannot be contained in Sp6 .2/. But for q  4 the order
of 2 is too large.
Finally, since 2 multiplies the symplectic form by the generator c of IF q the
group G is the full conformal group CSp2n .q/. t
u

Theorem 3.4. The groups CSp2n .q/, q D p m , .n; q/ ¤ .2; 2/, possess G-realizations
over abelian number fields k.n; q/ for the class vector of  defined in Proposi-
tion 3.3. The groups Sp2n .q/, PCSp2n .q/ and S2n .q/ possess G-realizations over
the same fields of definition.

Proof. By construction  is a Belyi triple. By its definition the group CSp2n .q/ is
self-normalizing in GL2n .q/, so the stronger second part of Theorem I.5.10 applies,
showing rigidity. The normalizer condition (N) for h1 i is verified by exactly the
same argument as in the proof of Theorem 1.4, using the fact that 1 has a .2n  1/-
dimensional eigenspace and that Z .CSp2n .q// Š IF q . Application of Theorem I.4.8
now yields the Galois realization.
The subgroup Sp2n .q/ has a rational fixed field in the Galois extension for
CSp2n .q/ by the same arguments as in the proof of Corollary 1.5, so occurs as
Galois group over the same field of definition. Finally, the groups PCSp2n .q/ D
CSp2n .q/=Z .CSp2n .q// and S2n .q/ D Sp2n .q/=Z .Sp2n .q// are factor groups, so
the remaining assertion is clear. t
u

Remark. The group Sp4 .2/ excluded in the Theorem is isomorphic to the symmetric
group S6 on six letters, and hence trivially known to occur as geometric Galois
group over Q.t/.
110 II Applications of Rigidity

3.3 Rigidity for SO2nC1 .q/

Let V be an odd-dimensional vector space over a finite field IFq of odd order q.
All non-degenerate symmetric forms on V are similar (see for example Aschbacher
(1986), (21.6.4)). The group leaving such a form invariant is the orthogonal group,
denoted by GO2nC1 .q/, where dim.V / D 2n C 1. The special orthogonal group
SO2nC1 .q/ consists of the elements of GO2nC1 .q/ of determinant 1. In Lie notation,
these are the groups Bn .q/ad . The image of the universal Chevalley group Bn .q/sc
in this matrix representation is the simple group O2nC1 .q/, the kernel of the spinor
norm
spin W SO2nC1 .q/ ! IF  2
q =.IFq / (3.3)
which is defined as follows: Any  2 SO2nC1 .q/ can be written as a product  D
1    r of reflections. Let vi denote an eigenvector of i for the eigenvalue 1 for
i D 1; : : : ; r. Then spin./ D Q.v1 /    Q.vr /.IF 2
q / where Q is the quadratic form
left invariant by SO2nC1 .q/ (see Aschbacher (1986), (22.11)).

Lemma 3.5. Let q be odd and  2 G WD SO2nC1 .q/ be of order q n C 1. Then 

generates a (cyclic) maximal torus of G. In particular, spin./ ¤ 1.

Proof. Let G WD SO2nC1 .IF N q / be defined with respect to the standard quadratic
P
form Q..x1 ; : : : ; x2nC1 / / D nC1
t
i D1 xi x2nC2i . Up to conjugation all semisimple
elements  2 G lie in the diagonal maximal torus of G. Thus the eigenvalues of 
are of the form a1 ; : : : ; an ; ˙1; an1 ; : : : ; a11 . If  2 G D SO2nC1 .q/ then the set of
eigenvalues must be defined over IFq . Thus q n C 1 is the largest possible order of
a semisimple element in SO2nC1 .q/. Furthermore, if  is such an element, then the
centralizer in G of  consists only of its powers. Hence hi is a maximal abelian
subgroup of G consisting of semisimple elements, so it is a maximal torus. The
remaining assertion now follows from Proposition 1.2. t
u

We shall assume from now on that n  3, since otherwise the orthogonal groups
are isomorphic to symplectic groups, and these were already considered in the pre-
vious section.

Proposition 3.6. Let q be odd, n  3, a 2 IF

q 2n
of multiplicative order q n C 1 and

Y
2n1
i
f .X / WD .X  1/ .X  aq /; g.X / WD .X C 1/2nC1 : (3.4)
i D0

If Q WD .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then

G WD h1 ; 2 i D SO2nC1 .q/.
n 2n1
Proof. We have aq D a1 , so the roots a; aq ; : : : ; aq ; 1 of f 2 IFq ŒX are of the
form (2.7). By Theorem 2.10 this shows that GQ WD hQ i GO2nC1 .q/. By construc-
tion the orders of 2 ; 3 are coprime, hence by Proposition 2.7 the group G acts
primitively since o.2 / D q n C 1 > .q  1/.2n C 1/2 =4. Thus the normal subgroup
3 The Classical Groups 111

H of GQ generated by the class of the reflection 1 in G acts irreducibly. By Proposi-

tion 2.1 it even acts primitively whence H is a primitive reflection group contained
in GO2nC1 .q/. By Theorem 2.4 either H  O2nC1 .q/ or H D S2nC2 ; S2nC3 or
n D 7, H D 2: O7 .2/. The symmetric groups can be excluded by Proposition 2.8,
while H D 2: O7 .2/ is ruled out by the order of 2 . Now 1 has determinant 1
and 2 lies in SO2nC1 .q/ n O2nC1 .q/ by Lemma 3.5. Hence GQ D GO2nC1 .q/ and
G D h1 ; 2 i D SO2nC1 .q/. t
u

Theorem 3.7. The groups SO2nC1 .q/, q D p m odd, n  3, possess G-realizations

over abelian number fields k.n; q/ Qab for the class vector .Œ1 ; Œ2 ; Œ3 /
defined in Proposition 3.6. The simple groups O2nC1 .q/ possess G-realizations over
the same fields of definition.

Proof. We have seen in Proposition 3.6 that  is a Belyi triple, so the criterion
of Belyi is applicable. By Kleidman and Liebeck (1990), Prop. 2.10.6, the group
SO2nC1 .q/ acts absolutely irreducibly in its natural representation, so by loc. cit.,
Cor. 2.10.4, its normalizer in GL2nC1 .q/ consists of those matrices leaving the
quadratic form invariant up to scalars, and this is the direct product of SO2nC1 .q/
with Z .GLn .q//. Hence the second condition of Theorem I.5.10 is also satisfied
and the result follows from the Basic Rigidity Theorem I.4.8 since Z .G/ D 1.
The fixed field of the subgroup O2nC1 .q/ of index two in SO2nC1 .q/ is rational,
which proves the second assertion. t
u

3.4 Rigidity for COC

2n .q/

Let V be an even-dimensional vector space over a finite field IFq . There exist two
types of non-degenerate quadratic forms on V (see Aschbacher (1986), (21.6.3)).
The subgroup of GL2n .q/ leaving invariant the quadratic form

X
n
Q.x/ D xi x2nC1i (3.5)
i D1

of maximal Witt index is called the orthogonal group of plus type GOC 2n .q/. The
conformal orthogonal group COC 2n .q/ consists of those matrices leaving the form
invariant up to scalar multiples.
Now first assume that q is odd. The special orthogonal group SOC 2n .q/ is defined
as the intersection of GOC
2n .q/ with SL 2n .q/. In Lie notation these groups have type
Dn . The embeddings of SOC 2n .q/ into CO C
2n .q/ and GO C
2n .q/ give rise to the exact
diagram
112 II Applications of Rigidity

1 1 1
? ? ?
y y y
1 ! SOC Cı
! IF
2n .q/ ,! CO2n .q/ ! q ! 1
? ? ?
y y y
mult
1 ! GOC C
! IF
2n .q/ ,! CO2n .q/ ! q ! 1
(3.6)
? ? ?
ydet y y
1 ! f˙1g Š Z2 ! 1
? ?
y y
1 1
Here COCı 2n .q/ denotes the finite group associated to the connected component of
the identity in the conformal orthogonal group over an algebraic closure of IFq .
A generator of the orthogonal group GOC 2n .q/ defined with respect to (3.5) over
SOC2n .q/ is the element 0 1
0 1
 WD @ Id2n2 A (3.7)
1 0
which induces the graph automorphism of order two on the diagram Dn . With this
the non-split orthogonal group GO 2n .q/ can be defined as the group of fixed points
of the product of the field automorphism of IFq 2 =IFq times .
The image of the universal Chevalley group Dn .q/sc in the above matrix repre-
sentation is the group C C
2n .q/, the kernel in SO2n .q/ of the spinor norm

spin W SOC   2
2n .q/ ! IFq =.IFq /

which is defined as in odd dimension. Clearly, the full normalizer of C2n .q/ in
GL2n .q/ is the conformal group COC2n .q/. The somewhat complicated connection
between various orthogonal groups for odd q is most conveniently depicted in the
following exact diagram:

1 1 1
? ? ?
y y y
spin
1 ! Z .C C
! ˙.IF
2n .q// ,! Z .SO2n .q// !
 2
q / =.IFq / ! 1
2
? ? ?
y y y
spin (3.8)
1! C
2n .q/ ,! SOC
2n .q/ !
! IF  2
q =.IFq / !1
? ? ?
y y y
1! OC
2n .q/ ,! PSOC ! IF
2n .q/ !
 2
q = ˙ .IFq / !1
? ? ?
y y y
1 1 1
3 The Classical Groups 113

Here the group OC

2n .q/ is simple for n  3. In analogy to Lemma 3.5 we have the
following result:

Lemma 3.8. Let  2 G WD SOC 2n .q/ be of order q  1. Then  generates a (cyclic)

n
C
maximal torus of G. In particular,  … 2n .q/.

The next result of Reiter (1999) allows to distinguish between the two types of
orthogonal groups in even dimension.

Lemma 3.9. Let  2 COCı 2n .q/ be semisimple with characteristic polynomial

Q
f .X / D 2ni D1 .X  ai / such that ai2 ¤ mult./ for i D 1; : : : ; 2n. Then COı
2n .q/
does not contain an element with characteristic polynomial f .X /.
N q/
Proof. Every semisimple element of the connected algebraic group G WD COı2n .IF
lies in a maximal torus, and all maximal tori are conjugate. A particular maximal
torus of G is
N
T WD fdiag.t1 ; : : : ; tn ; utn1 ; : : : ; ut11 / j u; ti 2 IFqg

(see Digne and Michel (1991), p. 147). From the action of the Weyl group of T it
follows that all semisimple elements in T with the same characteristic polynomial
are conjugate in G.
Now let 1 ; 2 2 G with characteristic polynomial f .X /. By the above we have
2 D 1 for some  2 G. Let F W G ! G be the standard Frobenius morphism.
Assume that F .1 / D 1 while F .2 / D 2 with  from (3.7). Then

F ./1 1 F ./ 1  1 D 1 ;

so F ./ 1  1 2 CCO2n .IFN q / .1 /. Taking determinants (respectively quasi-determi-

nants (3.11) in the case of characteristic 2) shows that det.F ./ 1  1 / D 1, so
F ./ 1  1 2 CO2n .IF N q / n G. But this gives a contradiction since the centralizer
in CO2n .IFN q / of a semisimple element satisfying the assumptions of the Lemma is
a product of full linear groups, thus it is connected and already contained in G. This
proves the assertion. t
u

Proposition 3.10. Let q be odd, n  4, a a generator of IF 

q n , c a generator of IFq
and
Y
n1
i i
f .X / WD .X  aq /.X  caq /; g.X / WD .X  1/n1 .X  c/n1 .X 2  c/ :
i D0
(3.9)
If  WD .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then
G WD h i D COC 2n .q/.

Proof. We have f; g 2 IFq ŒX and the roots of f; g are as in (2.9). Thus G WD h i

CO˙2n .q/ by Theorem 2.10. Furthermore, by Lemma 3.9 we have G COC2n .q/
C n C
since SO2 .q / SO2n .q/ clearly contains elements with characteristic polynomial
114 II Applications of Rigidity
Qn1 qi i
i D0 .X  a /.X  aq /. If G acts imprimitively with respect to a decomposition
V1 ˚ : : : ˚ Vr then r D n by Proposition 2.7(a). But we have o.2 / D q n  1 >
n2 .q  1/ so by Proposition 2.7(b) the group G acts primitively. Thus the normal
subgroup H of G generated by the class of the reflection 1 in G acts irreducibly.
By Proposition 2.1 it even acts primitively whence H is a primitive reflection group
contained in GOC C
2n .q/. By Theorem 2.4 either H > 2n .q/ or H D S2nC1 ; S2nC2 or
C
n D 8, H D 2: SO8 .2/. The symmetric groups can be excluded by Proposition 2.8,
while H D 2: SOC 8 .2/ is ruled out by the order of 2 . Now 1 has determinant 1
and 2 lies in SOC C
2n .q/ n 2n .q/ by Lemma 3.8. Finally, 3 multiplies the form by
the generator c of IFq , so G D h i D COC

2n .q/. t
u

Theorem 3.11. The groups COC 2n .q/, q D p

m
odd, n  4, possess G-realizations
ab
over abelian number fields k.n; q/ Q .t/ for the class vector .Œ1 ; Œ2 ; Œ3 /
constructed in Proposition 3.10. The groups GOC C C C
2n .q/, SO2n .q/, PSO2n .q/, 2n .q/
C ab
and O2n .q/ possess G-realizations over Q .
Proof. Let C D .Œ1 ; Œ2 ; Œ3 / be the class vector consisting of the conjugacy
classes of the elements i defined in Proposition 3.10. Since 1 has a .2n  1/-
dimensional eigenspace for the eigenvalue 1 and since, as remarked above, COC 2n .q/
is self-normalizing in GL2n .q/, we conclude from the Belyi criterion I.5.10 that
l.C/ D 1 and therefore C is rigid. The center Z WD Z .COC C
2n .q// of CO2n .q/ con-
sists of the scalar multiples of the identity matrix, hence the argument in the proof
of Theorem 1.4 may be used to verify the normalizer condition (N) for the inertia
group over 1 . The first part of the assertion then follows with Theorem I.4.8.
It remains to achieve the descent to the simple group OC Q
2n .q/. For this, let K
C
be the fixed field of 2n .q/  Z (see the diagrams (3.6) and (3.8)) in the Galois
extension N=k.t/ for COC 2n .q/, and

Q
H WD Gal.K=k.t// Š COC C
2n .q/=.2n .q/  Z/:

The first element 1 in the class vector is an involution, and the square of the sec-
ond also lies in C 2n .q/  Z. Hence H has a generating system of elements with
orders .2; 2; k/, and so is a dihedral group. In particular, KQ has genus zero by Theo-
rem I.6.2, and since the absolute Galois group of Qab is projective (see Serre (1964),
Ch. III, 2.4, Ex. 3, or also Theorem IV.1.11(d)), Qab is a field of definition of the
Galois extension QN=N N by Theorem I.3.4. Thus the group C .q/  Z has a G-
QK 2n
ab
realization over Q . Factoring by its center we obtain the desired realization for the
simple group OC 2n .q/.
The fixed fields of GOC C Q
2n .q/  Z and SO2n .q/  Z are subfields of K, so the pre-
ceding arguments also yield Galois extensions for GO2n .q/, SO2n .q/ and PSOC
C C
2n .q/
over the same field of definition as OC 2n .q/. t
u
In even characteristic the relationship between the various types of orthogonal
groups becomes much easier. So the four orthogonal groups in diagram (3.8) all
coincide, while the diagram (3.6) simplifies to

COC m C m 
2n .2 / Š GO2n .2 /  IF2m : (3.10)
3 The Classical Groups 115

The simple group OC m

2n .2 / is now obtained as the kernel of the quasi-determinant

qdet W GOC
2n .2 / ! f˙1g;
m
 7! .1/rk. 1/
(3.11)

(see Conway et al. (1985), p.xii).

Proposition 3.12. Let q D 2m , n  4, a a generator of IF

q n and

Y
n1
i i
f .X / WD .X  aq /.X  aq /; g.X / WD .X  1/2n : (3.12)
i D0

If  WD .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then

G WD h i D GOC 2n .q/.

The proof is analogous to the one of Proposition 3.10 above, using that an ele-
ment with minimal polynomial g.X / has quasi-determinant 1.

Theorem 3.13. The groups GOC 2n .q/, q D 2 , n  4, possess G-realizations over

m
ab
abelian number fields k.n; q/ Q for the class vector .Œ1 ; Œ2 ; Œ3 / defined in
Proposition 3.12. The groups OC
2n .q/ possess G-realizations over the same field of
definition.

Proof. Since the first element 1 of the class vector satisfies the assumptions of
Belyi’s criterion, we obtain rigidity for the group GOC m C m
2n .2 /. Since GO2n .2 / has
trivial center the first assertion follows with Theorem I.4.8. The descent to the simple
group is possible since GOC C
2n .q/= O2n .q/ is of order 2. t
u

3.5 Rigidity for CO

2n .q/

We consider the group of non-maximal Witt-index as the subgroup of GOC 2

2n .q /
fixed by the product of the field automorphism of IFq 2 =IFq times the non-trivial
graph automorphism of order 2 induced by the element  defined in (3.7). Thus the
C
orthogonal group of minus type GO 2n .q/ consists of those elements  in GO2n .q /
2
C
which satisfy  D ,

N where N denotes the automorphism of GO2n .q / raising each
2

matrix entry to its q-th power. The groups CO 

2n .q/ and SO2n .q/ are the stabi-
lizers of this automorphism in the corresponding untwisted groups COC 2
2n .q / and
C 2
SO2n .q /. Taking fixed points we thus obtain the analogues of diagrams (3.6)
and (3.8) for the various orthogonal groups of minus type.

Proposition 3.14. Let q D p m , n  4, a 2 IF

q 2n
of multiplicative order .q n C 1/.q 
n C1
1/, c WD aq , and

Y
2n1
i
f .X / WD .X  aq /; g.X / WD .X  1/n1 .X  c/n1 .X 2  c/ : (3.13)
i D0
116 II Applications of Rigidity

If  WD .1 ; 2 ; 3 / is a Belyi triple constructed from f; g as in Theorem 2.6 then

G WD h i D CO 2n .q/.

Proof. The roots of the polynomials f; g 2 IFq ŒX satisfy (2.9) hence G CO˙ 2n .q/.
By Lemma 3.9 we even have G CO .q/ since SO  n
.q / SO 
.q/ contains an
Q2n1
2n 2
q i C1 q i
2n
element with characteristic polynomial i D0 .X  a /. For the remainder of
the argument we may proceed as in the proof of Proposition 3.10 to show that indeed
G D CO  
2n .q/, since by definition det.1 / D 1, 2 lies in SO2n .q/ n 2n .q/, and
the element 3 multiplies the form by the element c of multiplicative order q  1.
t
u

Theorem 3.15. The groups CO 2n .q/, q D p , n  4, possess G-realizations over

m
ab
abelian number fields k.n; q/ Q for the class vector .Œ1 ; Œ2 ; Œ3 / defined in
Proposition 3.14. The groups GO    
2n .q/, SO2n .q/, PSO2n .q/, 2n .q/ and O2n .q/
ab
possess G-realizations over Q .

Proof. The proof is entirely analogous to those of the preceding results. The Belyi
criterion yields l.Œ1 ; Œ2 ; Œ3 / D 1, the normalizer condition (N) is verified as in
Theorem 3.2 using the fact that jZ .CO 2n .q//j D q  1, and then the Galois real-
ization for CO 2n .q/ follows from Theorem I.4.8. The other groups are obtained by
descent arguments as in the proof of Theorem 3.11 using the corresponding substi-
tutes for the diagrams (3.6) and (3.8) in the twisted case. Again, if q is odd the simple
group can only be shown to occur over Qab .t/ by using the fact (Theorem IV.1.11(d))
that Qab has projective absolute Galois group together with Theorem I.3.4. t
u
4 The Exceptional Groups of Rank at Most 2 117

4 The Exceptional Groups of Rank at Most 2

The exceptional groups of Lie type comprise ten families of finite simple groups
derived from the simple exceptional Lie algebras or from classical groups as fixed
points under an exceptional automorphism. Unlike the classical groups of Lie type,
they do not possess a ‘nice’ matrix representation, and in particular, Belyi’s crite-
rion is not applicable to them. Hence rigidity has to be proved in some other way.
Unfortunately, at present no uniform way of treating all exceptional groups of Lie
type is known; moreover, it is not even known whether all of them can be realized
as Galois groups over Qab .t/: the only finite simple groups not known to occur as
Galois groups over abelian number fields are among the exceptional groups of Lie
type in characteristic 2.
Nevertheless, quite a number of these groups can be covered by the criteria intro-
duced in the previous chapter. The key lies in the character theoretic form of the
Rigidity Criterion. The five series of exceptional groups of Lie rank at most two,
namely 2B2 , 2G2 , G2 , 3D4 and 2F4 , are particularly suited for an application of
this criterion. The conjugacy classes of elements, the character tables and the lists
of maximal subgroups are explicitly known in all these cases. Moreover, all these
results are generic, which is to say that they are almost independent of q. Indeed,
the entries in the character tables are polynomials in q, and there exists a fixed
number of maximal subgroups, apart from the so-called subfield groups obtained
by restricting the field of definition of the relevant group. So basically each series
may be treated more or less like one single group; only sometimes the effect of bad
characteristic has to be taken into account, forcing a subdivision into several cases.
Thanks to the almost complete knowledge of the small rank groups, only two
divisibility criteria are needed in addition. They are presented in the first section.

4.1 Divisibility Criteria

In the application of the character theoretic form of the Basic Rigidity Criterion
to a class vector C of a finite group G, first the structure constant n.C/ has to be
calculated. Then it remains to prove that a system  2 ˙N .C/ generates all of G.
This is usually done by showing that H WD h i cannot lie in any of the maximal
subgroups of G. One easy way to exclude a maximal subgroup M is Lagrange’s
theorem, that is, to exhibit a prime divisor of jH j which does not divide the order of
M.
The orders of the finite groups of Lie type G.q/ can be written as polynomials in
q D p m . Let r .X / denote the r-th cyclotomic polynomial in X , i.e., the polynomial
whose roots are the primitive r-th roots of unity, which can be defined recursively
by Y
Xr  1 D s .X / : (4.1)
sjr
118 II Applications of Rigidity

Then, more precisely, the order of G.q/ is a product of certain cyclotomic factors
r .q/ times a power of q. Factors of this type tend to have large prime divisors.
The main result in this direction is due to Zsigmondy (see Huppert and Blackburn
(1982), Ch. IX, Thm. 8.3):

Proposition 4.1. For all positive integers r  3 and all primes p there exists a prime
divisor l of p r  1, dividing none of the p s  1 with s < r, except for the case r D 6,
p D 2.

Such a prime l is called a primitive prime divisor of p r  1. As an easy conse-

quence we get:
Q
Corollary 4.2. If n .p m / with nm  3 divides the product kiD1 .p si  1/, then
there exists an index i with mnjsi , or else p D 2 and mn D 6.

Proof. If .p; mn/ ¤ .2; 6/ then there exists a primitive prime divisor l of p mn  1
by Proposition 4.1. Because
ˇ Y
mn1
ˇ
.p mn  1/ ˇˇ n .p m / .p j  1/;
j D1

l must already divide n .p m /. Again by the Proposition there now exists an i with
mn si . But gcd.p mn  1; p s  1/ D p gcd.mn;s/  1, so l can only divide p s  1 if
s is a multiple of mn. t
u

A quite different criterion also makes use of the orders of the elements to identify
a group generated by a triple of elements. We call a system .1 ; 2 ; 3 / of elements
with orders o.i / an .o.1 /; o.2 /; o.3 //-triple for short.

Proposition 4.3. Let H WD h1 ; 2 ; 3 i, where 1 2 3 D 1, such that the orders

o.1 /, o.2 / and o.3 / are pairwise coprime. Then H is perfect.

Proof. Assume H is not perfect. Then the commutator subgroup H 0 is a proper

normal subgroup of H with abelian factor group. In particular there exists a normal
subgroup N with factor H=N cyclic of prime order p. This factor H=N must have a
generating .m1 ; m2 ; m3 /-system, where mi jo.i / for i D 1; 2; 3. As the three orders
o.i / are pairwise coprime, at most one of them is divisible by p, so at most one of
the mi is different from 1. The product relation then forces all three to be equal to 1,
which contradicts the fact that the images should generate H=N . t
u

4.2 Rigidity for the Ree Groups 2G2 .q 2 /

The easiest series to treat among the exceptional groups are the Ree groups 2G2 .q 2 /
in characteristic 3. They are obtained from a simple algebraic group of type G2
N 3 as fixed points under a twisted Frobenius map, and exist only when
defined over IF
4 The Exceptional Groups of Rank at Most 2 119

q 2 D 32mC1 is an odd power of 3. The group 2G2 .3/ is isomorphic to the automor-
phism group L2 .8/ of L2 .8/, while the other groups are simple. The conjugacy
classes of elements of G D 2G2 .q 2 / were determined by Ward (1966), as well as a
large portion of the character tables. A complete list of maximal subgroups has been
obtained by Kleidman (1988b).
Let C be a class vector for G consisting of the unique class of involutions C2 ,
the class C3 of 3-elements lying central in a Sylow 3-subgroup (denoted by ŒX in
Wardp(1966)), and Cs , one of the classes containing semisimple elements of order
q 2  3q C 1 (which, despite its appearance, is an integer).

Table 4.1 Character values for 2G2 .32mC1 /

1 C2 C3 Cs
1 1 1 1 1
p
3
p p
5 6
q.q 2  1/.q 2 C 3q C 1/  12 .q 2  1/  16 .3q 2 C 3q/ 1
p
3
p p
7 6
q.q 2  1/.q 2 C 3q C 1/  12 .q 2  1/  16 .3q 2 C 3q/ 1

Proposition 4.4. The class vector C D .C2 ; C3 ; Cs / of 2G2 .32mC1 /, m  1, is rigid.

Proof. We claim that C satisfies the character theoretic rigidity criterion in Corol-
lary I.5.9. The relevant part of the character table of 2G2 .q 2 / is reproduced in
Table 4.1 with Ward’s notation for the characters. It contains the three irreducible
characters 1 ; 5 ; 7 not vanishing on any of the three classes of C and thus contribut-
ing to the normalized structure constant n.C/. Now using j2G2 .q 2 /j D pq .q  1/
6 2

.q 6 C 1/, jCG .1 /j D q 2 .q 4  1/, jCG .2 /j D q 6 and jCG .3 /j D q 2  3q C 1, this
shows that n.C/ D 1.
It remains to check that each  2 ˙N .C/ generates all of 2G2 .q 2 /. Let  be such a
system, and set H WD h i. According to Kleidman (1988b), the isomorphism types
of maximal subgroups of 2G2 .q 2 / for q 2 > 3 are as follows:

L2 .q 2 /  2; .22  D q2 C1 /:3; Œq 6 :.q 2  1/;

p 2
p
.q 2 C 3q C 1/:6; .q 2  3q C 1/:6;
2
G2 .r 2 / for r 2 D 32kC1 with .2m C 1/=.2k C 1/ prime:

Since the elements i , i D 1; 2; 3, have pairwise coprime orders, the group H is

perfect by Proposition 4.3. The only nonsolvable maximal subgroups in the list are
L2 .q 2 /2 and the subfield groups 2G2 .r 2 /. But the order of thep first is clearly prime
to o.3 /. The maximal element order p in 2
G 2 .r 2
/ equals r 2
C 3r C 1 2r 2 C 1,
which is smaller than o.3 / D q  3q C 1  1=2q C 1 if r ¤ q. Hence H is not
2 2

contained in any of the maximal subgroups of 2G2 .q 2 / and thus coincides with the
full group, which proves the assertion. t
u
120 II Applications of Rigidity

Theorem 4.5. The groups 2G2 .q 2 /, q 2 D 32mC1 > 3, possess G-realizations over
Qab for the class vector C D .C2 ; C3p; Cs /. More precisely, a field of index six in the
cyclotomic field Q. s / of s WD .q 2  3q C 1/-th roots of unity is a field of definition
for this Galois extension. The group 2G2 .3/0 Š L2 .8/ possesses a GA-realization
over Q.

Proof. This follows from the Basic Rigidity Theorem I.4.8. The precise field of
definition can be read off from the character table as described in Proposition I.4.4.
Since the first two classes of C are rational, only the third class contributes to the
irrationalities. The GA-realization for L2 .8/ was already obtained in Example I.8.3.
t
u

The same result holds if the class Cs is replaced by one of its powers such that the
corresponding element order is a primitive prime divisor in the sense of Zsigmondy.
In some cases, this allows to diminish the degree of a field of definition.

4.3 Rigidity for the Groups G2 .q/

For the groups G2 .q/, qDp m , the character tables were determined by Chang and
Ree (1974) in characteristic p  5, by Enomoto (1976) in characteristic 3, and by
Enomoto and Yamada (1986) in even characteristic. The maximal subgroups were
obtained by Cooperstein (1981) for p D 2 and by Kleidman (1988b) for p  3. It
turns out that one may choose a class vector yielding rigidity for almost all groups
G2 .q/.
For this let C1C be a class of elements of order q 2  q C 1 if q 1 .mod 3/,
and C1 (resp. C10 ), a class of elements of order q 2 C q C 1 if q 1 .mod 3/
(resp. q 0 .mod 3/). Also let C2C , C20 denote a class of elements of order q C 1
with centralizer order q.q C 1/.q 2  1/ if q 1 .mod 3/, resp. q 0 .mod 3/,
and C2 a class of elements of order q  1 with centralizer order q.q  1/ .q 2  1/
if q 1 .mod 3/. Finally let C3 denote the class of semisimple 3-elements with
centralizer order q 3 .q 2  1/.q 3  "/ if q " D ˙1 .mod 3/, and the unipotent class
of 3-elements central in a Sylow 3-subgroup if 3jq. Unfortunately the notations for
conjugacy classes employed by the different authors vary considerably.

Table 4.2 Conjugacy classes for G2 .q/

CR E EY
"D1 " D 1 "D0 "D1 " D 1
C1" h6 h3 E5 E4 E3
C2" h2b h1a D11 D21 C21
C3 k3 A2 B0
4 The Exceptional Groups of Rank at Most 2 121

Table 4.2 gives a dictionary between our notation and that in the papers of Chang
and Ree (1974) (abbreviated CR), of Enomoto (1976) (E) and of Enomoto and
Yamada (1986) (EY). From the information given in the literature cited above one
verifies that the three classes are nonempty for all q ¤ 2. The character table of
the composite group G2 .2/ Š U3 .3/:2 is reproduced in the Atlas of finite groups
(Conway et al. (1985)), and we shall use the names of conjugacy classes given
there for this group.

Proposition 4.6. (a) The class vector C D .C1" ; C2" ; C3 / of G2 .q/, 2 ¤ q D p m "
.mod 3/, is rigid.
(b) The class vector C D .2B; 4D; 12A/ of G2 .2/ is rationally rigid.

Proof. In the general case (a), the relevant part of the character table is given in
Table 4.3 for q " .mod 3/, where " 2 f˙1g, (here the first column gives the char-
acter names in CR, the second one those in EY), and in Table 4.4 for 3 dividing q.
(Here the notation for the irreducible characters is taken from Enomoto (1976)).
Together with the group order jG2 .q/j D q 6 .q 2  1/.q 6  1/ and the centralizer
orders given above this yields n.C/ D 1.

Table 4.3 Character values for G2 .q/, q " .mod 3/

CR EY 1 C3 C2" C1"
11 0 1 1 1 1
12 5 q6 "q 3 "q 1

Table 4.4 Character values for G2 .q/, q D 3m

1 C3 C20 C10
0 1 1 1 1
10 1
6
q12 6 1
6
q.2q  1/1 1 1

N
Now let H WD h i for  2 ˙.C/. The three classes were chosen such that the
corresponding element orders are pairwise prime. Hence H is perfect by Proposi-
tion 4.3. Moreover we may apply Corollary 4.2 to a primitive prime divisor of the
order o.3 / D q 2 ˙ q C 1. Note that the exceptions do not occur due to the con-
gruence conditions in the definition of C1" . By Kleidman (1988b) and Cooperstein
122 II Applications of Rigidity

(1981) the nonsolvable maximal subgroups of G2 .q/ with order divisible by a prim-
itive prime divisor of o.3 / are

SL3 .q/W 2 if q 1 .mod 3/;

SL3 .q/W 2 if q 0 .mod 3/ (two classes);
SU3 .q/W 2 if q 1 .mod 3/;
L2 .13/ if q D 3 or 4:

Let us first assume that q 1 .mod 3/. Then we have to exclude SU3 .q/W 2, and
for q D 4 the L2 .13/. But the latter is easy, since for q D 4 elements in C2C have
order 5, which does not divide j L2 .13/j. If H SU3 .q/W 2, then since H is perfect
we even have H SU3 .q/. Note that in our case SU3 .q/ D U3 .q/ since q 1
.mod 3/. It now suffices to show that C3 does not intersect this subgroup. In U3 .q/
there exists just one class 3A of elements of order 3, and from the character table
of U3 .q/ in Simpson and Frame (1973) one calculates the normalized structure
constant n.3A; 3A; C1C / D q 2 C q C 1. But in G2 .q/ the corresponding structure
constant is found to vanish. So indeed 3A does not fuse into C3 , and the result
follows for q 1 .mod 3/. For q 1 .mod 3/ the proof is entirely similar, with
U3 .q/ replaced by L3 .q/.
So let’s finally assume 3jq. In this case, the two classes of maximal subgroups
SL3 .q/ are those generated by all long root subgroups, respectively all short root
subgroups. Since C3 is the class of long root elements, its intersection with the
second class is empty. The exceptional graph automorphism in characteristic 3
exchanges long and short roots, and it interchanges the two classes of maximal tori
in G2 .q/ of order q 2  1. Thus it also interchanges our class C20 D D11 with the
class denoted by D21 in Enomoto (1976). Since SL3 .q/ has just one class of tori
of order q 2  1, it can intersect at most (in fact, precisely) one of D11 , D21 . From
the description of the semisimple classes by Enomoto it now follows that the SL3 .q/
generated by long root elements intersects D21 and so the possibility SL3 .q/ is ruled
out. If q D 3, then by the Atlas the unique 3-class in L2 .13/ fuses into 3D of G2 .3/,
which is not central in a Sylow 3-subgroup. This completes the proof for (a).
For G D G2 .2/ in (b) this is a special case of Theorem 6.8(b) due to the excep-
tional isomorphism G2 .2/ Š Aut.U3 .3//. t
u

Theorem 4.7. The groups G2 .q/, q D p m , possess G-realizations over Qab for the
class vector C of Proposition 4.6.

4.4 Rigidity for the Groups 3D4 .q/

The conjugacy classes, the character table and the list of maximal subgroups for
3
D4 .q/ were determined by Spaltenstein (1982), Deriziotis and Michler (1987) and
Kleidman (1988a). We chose the three conjugacy classes C1 WD Œu1 , C2 WD Œu5
4 The Exceptional Groups of Rank at Most 2 123

and C3 WD Œs14 , where the names of the representative elements are taken from
Deriziotis and Michler (1987).
Proposition 4.8. The class vector C D .C1 ; C2 ; C3 / of 3D4 .q/, q D p m , is rigid.
Proof. From the character table in Deriziotis and Michler (1987) one finds that the
only irreducible characters not vanishing on any of the three classes in C are as in
Table 4.5. Here the last line stands for a whole family of characters constant on all
classes apart from C3 , and such that the sum of their values on that last class equals
1 (this is indicated by the notation f1g in the table). In particular, since

jCG .1 /j D q 12 .q 6  1/; jCG .2 /j D q 6 ; jCG .3 /j D q 4  q 2 C 1

for  2 ˙.C/, and

j3D4 .q/j D q 12 .q 2  1/.q 6  1/.q 8 C q 4 C 1/

we obtain the value n.C/ D 1 for the normalized structure constant.

Table 4.5 Character values for 3D4 .q/

1 C1 C2 C3
1 1 1 1 1
14 12 22 32 62 1 2 3 6 1 f1g

Now let H WD h i. Using the list of maximal subgroups of 3D4 .q/ in Kleidman
(1988a) and Corollary 4.2 applied to a primitive prime divisor of o.3 / D q 4  q 2 C
1 D 12 one finds that either H D 3D4 .q/ or H NG .h3 i/ D .q 4  q 2 C 1/:4.
The elements 1 and 2 are unipotent, hence have p-power order, so they cannot
be contained in the latter subgroup if p is odd. For p D 2 one checks that 2 has
order 4, while 1 is an involution. This follows either from the Chevalley commuta-
tor formulae, or by considering the element orders in the character table of the group
3
D4 .2/ over the prime field in the Atlas. If H were in NG .h3 i/, then factoring by
its cyclic normal subgroup of order q 4  q 2 C 1 reveals that the corresponding factor
HN would be generated by a .2; 4/-system, which is a contradiction to the product
relation. So we have proved H D 3D4 .q/ as required. t
u
This immediately implies the following result, using that the first two classes in
C are rational, while elements in the third are conjugate to four of their primitive
powers:
Theorem 4.9. The groups 3D4 .q/, q D p m , possess G-realizations over Qab for the
class vector C D .C1 ; C2 ; C3 / defined in Proposition 4.8. More precisely, a field of
index four in the cyclotomic field Q. s / of s WD .q 4  q 2 C 1/-th roots of unity is a
field of definition for this Galois extension.
124 II Applications of Rigidity

4.5 Rigidity for the Groups 2B2 .8/ and 2F4 .2/0

In the preceding three sections, we have treated all series of exceptional groups of
Lie rank at most two, apart from the Suzuki and Ree groups in characteristic 2,
2
B2 .q 2 / and 2F4 .q 2 /. No Galois realizations for them are known at present, except
for the smallest group in each series. In particular for the Suzuki groups 2B2 .22mC1 /
one can check that the Basic Rigidity Criterion does not apply to any class triple C
if m is large.
The Suzuki groups G D 2B2 .22mC1 / have a very transparent structure. Their
character tables and maximal subgroups were determined by Suzuki. Thus all ingre-
dients for the application of the character theoretic rigidity criterion are known. Only
six different types of conjugacy classes exist in G. This makes it possible to compute
n.C/ for all class vectors C of length three. It turns out that the structure constants
are nonconstant polynomials in q 2 D 22mC1 , except if C contains at least two invo-
lution classes (and hence  2 ˙.C/ N cannot generate G). Furthermore it is not hard
to prove that the contributions to n.C/ coming from proper subgroups are too small
to diminish this growth rate for l.C/. From this it follows that the Suzuki groups for
large m cannot be realized as Galois groups via the Basic Rigidity Theorem applied
to a class triple C.
However in the smallest simple case 2B2 .8/, a G-realization may be found (note
that 2B2 .2/ Š 5W 4 is solvable). Here the class names are taken from the Atlas.

Theorem
p 4.10. The group Aut.2B2 .8// D 2B2 .8/W 3 possesses a G-realization over
Q. 3/ for the class vector C D .2A; 3A; 15A/. This yields a GA-realization of
2
B2 .8/ for the class vector .2A; 2A; 2A; 5A/.

Proof. From the Atlas one finds n.C/ D 1 in Aut.2B2 .8//. The only maximal sub-
groups of order divisible by 15 are the centralizers of the outer automorphisms, of
type 2B2 .2/  3 D 5W 4  3. Factoring by the normal Sylow 5-subgroup we would
arrive at a .2; 3; 3/-system inside 4  3 Š Z12 , which is a contradiction. So the gen-
eration is clear as well, proving that C is rigid by Corollary I.5.9. The second asser-
tion follows by descent to the simple subgroup of index 3, similar to the argument
proving the Galois realization for An from the one of Sn . t
u

The situation for the Ree groups 2F4 .22mC1 / is not much better. Since now there
are 52 families of conjugacy classes, a complete study is no longer possible. But
all ‘promising’ class triples show a similar behavior to those of the Suzuki groups.
Still, the smallest case can again be treated. Here 2F4 .2/ Š Ti:2, where Ti D 2F4 .2/0
(the Tits group) is a simple group not appearing elsewhere in the classification.
p
Theorem 4.11. The group Ti D 2F4 .2/0 possesses G-realizations over Q. 13/ for
the class vectors C1 D .2A; 3A; 13A/ and C2 D .2A; 5A; 13A/.

Proof. From the Atlas character table one verifies that n.C1 / D n.C2 / D 1. The
maximal subgroups of Ti with order divisible by 13 are L3 .3/  2 and L2 .25/  2.
N
Since the group H generated by a triple  2 ˙.C/ is perfect by Proposition 4.3,
4 The Exceptional Groups of Rank at Most 2 125

only L3 .3/ and L2 .25/ themselves remain as candidates. But the involution classes
in both groups possess centralizer order divisible by 3, hence they must both fuse
into 2B of G. So the class 2A intersects neither of them, and we have H D G. This
p The character table reveals that 2A, 3A and 5A
proves rigidity by Corollary I.5.9.
are rational while Q13A D Q. 13/. t
u

The second class vector will again turn up in the Galois realization of the sporadic
Rudvalis group Ru.
126 II Applications of Rigidity

5 The Exceptional Groups of Large Rank

Apart from those cases treated in the previous paragraph, the character tables of
the exceptional groups are not yet completely known. Still, the theory of Deligne
and Lusztig on the characters of reductive groups over finite fields furnishes enough
information in most cases for the computation of structure constants, if the class
vector is chosen carefully enough. Also, no complete lists of maximal subgroups
for these groups have been obtained at present. But it is possible to enumerate the
maximal subgroups lying above certain cyclic maximal tori (see for example Weigel
(1992)), and this will suffice for our purpose.

5.1 Results From Deligne–Lusztig Theory

As in the previous paragraph, and in contrast to the case of classical groups, we shall
not work with explicitly given triples of elements, but instead apply the character
theoretic form of the Basic Rigidity Theorem. The determination of the structure
constant for a chosen class vector requires the knowledge of at least part of the table
of complex irreducible characters of G. This is furnished by the Deligne–Lusztig
theory of characters of reductive groups over finite fields.
Let G be a connected reductive algebraic group over the algebraic closure of
IFq , and F W G ! G a Frobenius map. An important tool in the character theory of
the groups GF are the characters RT ./ which were constructed in the fundamen-
tal paper of Deligne and Lusztig (1976). To each F -stable maximal torus T and
to each irreducible (hence linear) character  of T WD TF a generalized character
RT ./ can be defined. Two such characters have disjoint sets of constituents if the
corresponding pairs .T; / and .T 0 ;  0 / are not geometrically conjugate. Since every
irreducible character of GF appears in some RT ./, this defines a partition of the
set Irr.GF / into a disjoint union of Lusztig series. However, the RT ./ do not span
the whole space of class functions on GF , but a very large part of it. Elements in
the Q-span of the RT ./ are called uniform functions. Similarly, a conjugacy class
C of GF is called uniform, if its characteristic function can be expressed as a linear
combination of RT ./s. For semisimple classes we have the following nice result:
Theorem 5.1 (Deligne and Lusztig). Let G be a connected reductive group with
Frobenius endomorphism F W G ! G and s 2 GF semisimple.
(a) If .s/ ¤ 0 for  2 Irr.GF /, then there exist geometrically conjugate pairs
.T; /, .T 0 ;  0 / with RT ./.s/ ¤ 0 and .; RT 0 . 0 // ¤ 0.
(b) If RT ./.s/ ¤ 0 then there exists a g 2 GF with Tg CG .s/.
The first part follows directly from the fact that semisimple classes are uniform,
the second is a consequence of the character formula (see Carter (1985), Prop. 7.5.5,
Thm. 7.2.8). This shows that if s has a small centralizer, then only few irreducible
characters of GF will take non-zero values on s.
5 The Exceptional Groups of Large Rank 127

In general, the characters RT ./ will themselves be irreducible. Only when  is

not in general position with respect to the Weyl group of T , then RT ./ has norm
bigger than one. In particular, this is the case for  D 1. The irreducible constituents
of the RT .1/ are called unipotent characters. In Lusztig’s classification of the irre-
ducible characters of the groups GF they play a crucial role, similar to the role of
the unipotent classes in the classification of all conjugacy classes of GF . In our
applications, apart from the case E7 , we will only need to know values of unipotent
characters.
We give an explicit formula for these at least on semisimple classes. For its for-
mulation, some more notation has to be introduced. For an arbitrary algebraic group
H, let Hı be the connected component of the identity. Furthermore, for a connected
reductive group G let "G be the sign associated to G via the relative rank (see Carter
(1985), p.199), and for a semisimple element s 2 GF , write "s WD "CGı .s/ . Deligne
and Lusztig (1976) proved the following result:

Proposition 5.2. The value of the unipotent character  of the group GF on the
semisimple element s 2 GF is given by
X
.s/ D "s jCGı .s/F j1
p  "T .; RT .1//: (5.1)
T Ws2T

The decomposition of the RT .1/ into unipotent characters, and hence the scalar
products .; RT .1//, were determined by Lusztig (1980) for exceptional groups.
It is possible to evaluate (5.1) effectively. For this, one can use the fact that all
tori TF containing s lie in CGı .s/F (see Carter (1985), Prop. 3.5.2). By loc. cit.,
Thm. 3.5.4, CGı .s/ is a reductive algebraic group itself. So the sum in (5.1) runs
precisely over the maximal TF of CGı .s/F . Now the GF -conjugacy classes of F -
stable maximal tori T of G are parametrized by F -conjugacy classes Œw F  W .G/
in the Weyl group. This is indicated by writing Tw for a torus parametrized by
w 2 W . Hence with C WD CGı .s/F , the C -class ŒTw contains exactly jC j=jNC .Tw /j
different maximal tori conjugate to Tw . Writing W0 for the Weyl group of C we
therefore have

jNC .Tw /j D jTF

w j  .NC .Tw / W Tw / D jTw j  jCW0 .w/j;
F F

where CW0 .w/ denotes the F -centralizer of w in W0 . Moreover, for GF of untwisted

type, define the almost characters
X
R WD jW j1 .w/RTw .1/ for  2 Irr.W /: (5.2)
w2W

Then (5.1) becomes

128 II Applications of Rigidity

Corollary 5.3. If GF is of untwisted type, the value of the unipotent character  of

the group GF on the semisimple element s 2 GF is given by
X X .w/
.s/ D "s jCGı .s/F jp0  .; R /  "Tw : (5.3)
jTF
w jCW0 .w/j
j 
2Irr.W / Œw W0

This looks more complicated than (5.1) but in fact all ingredients in this formula
are explicitly known, and the sum may thus be effectively computed for all unipotent
characters  of GF . The multiplicities .; R / are determined by the .; RT .1//
via (5.2). Tables of these numbers are given in Carter (1985), Sect. 13.6.
The theory for the values on arbitrary elements is not yet complete. But for uni-
form classes, by definition the character values are linear combinations of RT ./,
and an analogue of (5.1) holds. It turns out that in these cases it suffices to know
the Green functions of GF , which are defined as the restrictions of the RT ./ to the
set of unipotent elements. They do not depend on , but only on the GF -class of T.
Tables of them have been calculated for exceptional groups at least in good char-
acteristic. For details, the reader is referred to the books of Lusztig (1984), Carter
(1985) or Digne and Michel (1991).
Important parts of this theory have been generalized to disconnected groups G by
Digne and Michel (1994). This will be useful in the treatment of groups of type E6 .

5.2 Rigidity for the Groups F4 .q/

The character table of the groups F4 .q/ is not yet known in general. Only the small-
est group F4 .2/ has been included in the Atlas. Also, no complete lists of maximal
subgroups have been obtained for these groups. There are partial results, in good
characteristic, but in general the problem is still open. Through a good choice of
class vector, both of the above problems may be circumvented.
The conjugacy classes of elements in G D F4 .q/ were determined by Shoji
(1974) in characteristic p  3, and by Shinoda (1974) for even q. We define Cp
to be the class of (unipotent) p-elements in long root subgroups. They are central
elements of a Sylow p-subgroup of G, and are characterized by this property if p is
odd. This class is called Œx1 in Shoji (1974), and Œx2 in Shinoda (1974).
Furthermore, G contains two families of semisimple classes such that its ele-
ments have centralizer type A2 C A1 , one of them having h10 as representative in
Shoji (1974), the other h16 (resp. h16 , h17 in Shinoda (1974)). We let CqC1 WD
Œh16 . For odd q this class may also be characterized by the fact that its .q C 1/=2-th
power is the involution with centralizer structure B4 . In even characteristic, the two
families of classes are interchanged by the exceptional graph automorphism. Finally,
choose CT to contain elements of order q 4  q 2 C 1 D 12 , generating a Coxeter
torus of F4 .q/. Representatives for this class are denoted by h99 , h76 respectively,
in loc. cit. For F4 .2/, we use the class names from the Atlas.
5 The Exceptional Groups of Large Rank 129

Proposition 5.4. (a) The class vector C D .Cp ; CqC1 ; CT / of F4 .q/, q D p m ¤ 2,

is rigid.
(b) The class vector C D .2A; 8A; 17A/ of F4 .2/ is rigid.

Proof. In (a), the elements in class CT generate a cyclic maximal torus T of order
12 , which is easily seen to be a Hall subgroup of G D F4 .q/ by the order formula.
In particular all of its non-identity elements are regular. By Theorem 5.1 this means
that the only Deligne–Lusztig characters not vanishing on CT are among the RT ./,
 a linear character of T in general position, and the RT 0 .1/ for arbitrary maximal
tori T 0 . But again by Theorem 5.1 the first take value zero on elements in CqC1 ,
since their centralizer does not contain a conjugate of T for order reasons. It remains
to consider the constituents of the RT 0 .1/, hence the 37 unipotent characters of G.
For a list of these, see Carter (1985), Sect. 13.9. With Corollary 5.3, their values on
CT may easily be computed, and it turns out that only twelve among them do not
vanish on CT . Next, the same can be done for the semisimple class CqC1 . Precisely
five unipotent characters are found not to vanish on CqC1 and CT . Their values are
given in Table 5.1.

Table 5.1 Character values in F4 .q/

1 Cp CqC1 CT
1;0 1 1 1 1
24 4
1;24 q 0 q 1
B2 ; 1 1
2
q12 32 8  12 q1 3 .q 4  q 3 C 1/ 12 1
B2 ;  1 13 2 2
2
q 1 3 8  12 q 13 1 3 q 2
12 1
6;600 1 4 2 2 2
12
q 3 4 6 8 1 4
12
q 3 4 6 .3q 4 C 2q 2 C 1/ 2q6 1

It remains to prove the correctness of the values on the unipotent class Cp . It

follows from Shoji (1982) in characteristic p  5, from Porsch (1993) in charac-
teristic 3 and from Malle (1993c) in characteristic 2 that the class Cp is uniform,
and then the values may be computed from the Green functions given in the cited
references. The desired structure constant n.C/ D 1 now follows from Table 5.1 and
the centralizer orders

jCG .1 /j D q 24 .q 2  1/.q 4  1/.q 6  1/;

jCG .2 /j D q 4 .q C 1/.q 2  1/2 .q 3 C 1/; jCG .3 /j D q 4  q 2 C 1:

Next we have to prove generation. For this let H WD h i for  2 ˙.C/. N Then
H contains the maximal torus T D h3 i. Now for q  4 the possible overgroups
of T in G have been completely classified by Weigel (1992) (see his Table I and
Figs. 4 and 5). From the proof given there it follows that this remains true for q D 3
130 II Applications of Rigidity

except for a possible additional overgroup U3 .9/. But such a group cannot contain
H , since in 2A2 .32 / any element of order 4 D q C 1 has centralizer order divisible
by 5, which is not the case for jCG .2 /j. Hence for q  3 the group H has one of the
structures given in loc. cit. Now NG .T / Š .q 4  q 2 C 1/:12, which can be ruled out
since H is perfect by Proposition 4.3. Thus we are left with the candidates 3D4 .q/
and F4 .q/ for H .
It remains to exclude that H D 3D4 .q/. First assume that p ¤ 2. Then by Der-
iziotis and Michler (1987) H has a single class of involutions C . This would have
to fuse into the class C2 of 2.qC1/=2 , which by definition is the one with central-
izer structure B4 in G. Since the normalizer of a torus of order 12 has structure
12 :4 in H , H contains a dihedral group 12 :2 with involutions in C2 . But in G one
calculates n.C2 ; C2 ; CT / D 0 (similarly to the determination of n.C/ above), so we
obtain a contradiction to the assumption that C fuses into C2 , and H D G follows
for odd q.
If p D 2, the groups F4 .2m / possess exactly two classes of (maximal) subgroups
3
D4 .2m /W 3, which are interchanged by the graph automorphism  (see loc. cit.). We
now study the fusion of some classes from these maximal subgroups into the classes
Cp and CqC1 of G.
By Spaltenstein (1982), 3D4 .2m / contains two classes of involutions, denoted
there by A1 and 3A1 . Of these, A1 can only fuse into one of C2 or C2 , as can be
seen from the centralizer orders. On the other hand, elements from 3A1 cannot fuse

into either C2 , C2 , since the first are contained in the normalizer of the torus T , but
not the second. This can be seen by calculating the corresponding structure constant
n.3A1 ; 3A1 ; CT / in the group 3D4 .2/, resp. .C2 ; C2 ; CT / in F4 .2/, in the Atlas, since
representatives for 3A1 , resp. C2 , are already defined over IF2 . Further 3D4 .2m /
contains just one class of maximal tori of order .q C 1/.q 3 C 1/, while G possesses
two, denoted by C3 C A1 and D4 in Table 1 of Shinoda (1974), and interchanged
by the outer automorphism  . If both types of tori contained elements conjugate to
h16 , then both types would also occur in CG .2 / Š 2A2 .q/:A1 .q/:.q C 1/. But in
the latter group, all tori of that order are conjugate, which yields a contradiction.
Hence the element h16 lies in one of the two types of tori, the element h17 in the
other. The above discussion shows that one class of maximal subgroups 3D4 .2m /W 3
of G only contains elements from C2 and Œh17 , say, while the other only contains
elements from C2 and Œh16 D Œh17 . Thus H cannot lie in either of the two, and the
proof is complete.
For G D F4 .2/, the character table in the Atlas yields n.C/ D 1. The classes of
maximal subgroups of G were enumerated by Norton and Wilson (1989). The only
ones with order divisible by 17 are two groups S8 .2/. The permutation characters
for these subgroups are easily determined to be

1 C 4 C 6 C 10 C 12 ; 1 C 3 C 6 C 9 C 11 ;

which both vanish on the class 8A. (Alternatively, one knows that these S8 .2/ are
generated by long, resp. short root elements, while the class 8A contains products
5 The Exceptional Groups of Large Rank 131

of root elements for long and short roots.) This completes the proof also in the case
q D 2. t
u

Theorem 5.5. The groups F4 .q/, q D p m , possess G-realizations over abelian num-
ber fields k.q/ Qab for the classpvector C of Proposition 5.4. For q D 2 this yields
a G-realization of F4 .2/ over Q. 17/.

The rationality assertion for F4 .2/ follows from the Atlas table.
Galois realizations over Q of infinitely many groups F4 .p/ for primes p are con-
tained in Paragraph 8.

5.3 Rigidity for the Groups E6 .q/ and 2E6 .q/ for odd q

The circumstances for groups of type E6 are less favorable than for type F4 , since
here, as in the case of classical groups, they exist in different isogeny types. This
entails that the simple group E6 .q/ in general is a subgroup of index three in the
group of adjoint type E6 .q/ad . Or, if we start from the simply connected group
E6 .q/sc , then E6 .q/ is obtained by factoring out a center of order three. Both sit-
uations turn out to be hostile to the existence of rigid class triples. But the Dynkin
diagram of E6 has a graph automorphism of order two. We will prove G-realizations
for the group E6 .q/sc W 2, obtained by extending the simply connected group by this
graph automorphism, at least in characteristic p  3. For p D 2 this approach fails,
and we will treat this case in the next section.
So first let G WD E6 .q/sc W 2 be the extension by the graph automorphism of order
two of the simply connected group of type E6 , and assume that q D p m is odd.
Then G may be regarded as the group of fixed points under a Frobenius map F
of a disconnected reductive algebraic group. It was shown by Digne and Michel
(1994) that the Deligne–Lusztig theory has a nice and straightforward generaliza-
tion to disconnected groups. So there exist generalized Deligne–Lusztig characters
for such groups, indexed by the classes of maximal tori in the centralizer of the
graph automorphism. Also, the character formulae given in the first section remain
valid, with a suitable interpretation. The decomposition of the generalized Deligne–
Lusztig characters in the case of disconnected groups of type E6 was determined
in Malle (1993b), so all ingredients are available to explicitly compute character
values also for G.
Let C2 be the class of the graph automorphism  in G. Now  clearly central-
izes an F4 .q/ and we have CG . / D F4 .q/  2. If q 1 .mod 3/ then E6 .q/sc
has a center of order three. Since Z .F4 .q// D 1,  acts non-trivially on that cen-
ter, and G is a group with trivial center. Let u be a unipotent element in CG . /
with jCF4 .u/j D q 24 13 23 3 4 6 , which hence is central in a Sylow p-subgroup of
F4 .q/; since p ¤ 2, the corresponding class is uniquely determined by this prop-
erty. The element u then has order 2p and centralizer order 2q 24 13 23 3 4 6 in
G. From the list of conjugacy classes in Mizuno (1977) it follows that u must
132 II Applications of Rigidity

have centralizer order q 36 15 23 32 4 5 6 in G, because only that centralizer of a
unipotent element contains a subgroup C3 .q/ (compare also with the lists in Carter
(1985), p.402). Denote the class of u in G by C2p .
Finally, let T be a maximal torus of G of order q 6 C q 3 C 1 D 9 , and let CT
be the conjugacy class of a generating element 3 in this torus. Then we have
jCG .3 /j D 9 , since the outer automorphism acts fixed point freely on T .

Proposition 5.6. The class vector C D .C2p ; C2p ; CT / of E6 .q/sc W 2, q D p m , p ¤

2, is rigid.

Proof. We first calculate the structure constant for C. If an irreducible character 

of G does not vanish on the class CT , then clearly the same holds for its restriction
to G 0 WD E6 .q/sc . Thus we first classify the irreducible characters of the connected
group G 0 not vanishing on CT . Since CG 0 ./ D T for all elements  2 T not lying in
Z .G 0 /, these are the constituents of RT ./ and of RT 0 . 0 /, where T 0 is an arbitrary
torus with character  0 of order dividing 3 corresponding to the central elements of
G 0 . Since  acts non-trivially on Z .G 0 /, those with  0 ¤ 1 are not  -stable. It then
follows that none of their constituents extends to G, so they will not contribute to
n.C/. The irreducible RT ./ are either not  -stable, or vanish on the class of  ,
since the centralizer of  does not contain T , as follows from the character formula
for disconnected groups (see Digne and Michel (1994), Prop. 2.4). Hence it remains
to consider the constituents of the RT 0 .1/, the unipotent characters of G 0 . Nine of
them do not vanish on CT by (5.1). Of these, another two take value zero on the outer
quasi-central class C2 , again by the character formula in loc. cit. The remaining
seven characters and their values on classes in C and on the class C2 , which will be
needed in the proof, are given in Table 5.2.

Table 5.2 Character values in E6 .q/sc W 2

1 C2 C2p CT
1;0 1 1 1 1
20;2 2
q 4 5 8 12 q4 8 12 q.q C q C 1/
6 4
1
20;20 q 20 4 5 8 12 q 13 4 8 12 q 13 1
36 24
1;36 q q 0 1
90;8 1 7 3
3
q 3 5 62 8 12 1 4 2 2
3
q 3 6 8 12 3
q 3 6 .2q 4 C 1/
1 4
1
E6 Π1 7 6 4 2
3
q 1 2 4 5 8 1 4 4 4 2
3
q 1 2 4 8  13 q 4 13 23 4 1
E6 Π2 1 7 6 4 2
3
q 1 2 4 5 8 1 4 4 4 2
3
q 1 2 4 8  13 q 4 13 23 4 1

Note that each line represents the two extensions of a unipotent character from
G 0 to G. Our notation for the unipotent characters is the same as in Carter (1985),
Sect 13.9. Here the entries on the mixed class C2p may again be determined from
5 The Exceptional Groups of Large Rank 133

the character formula in Digne and Michel (1994). This requires knowledge of the
2
Green functions on C2p , which can be found in Beynon and Spaltenstein (1984)
for p  5, and in Porsch (1993) for p D 3. Note that by Digne and Michel (1994),
2
Props. 2.4 and 4.10, the class C2p is uniform if C2p is uniform, and the latter holds
by the criterion of Lusztig (1977) (remark after 2.16). Using the centralizer orders
given before, the result n.C/ D 1 follows from Table 5.2.
Now let H WD h i for  2 ˙N .C/ and HN WD H=.H \ Z .G 0 //. The element 3
generates a maximal torus of G of order q 6 C q 3 C 1. By Weigel (1992), Fig. 6, if
we can exclude HN A2 .q 3 /W 6 then HN D GN follows. We cannot have HN D A2 .q 3 /
since otherwise HN \ GN 0 would be a subgroup of index two in that simple group.
On the other hand, TN is already contained in A2 .q 3 /, hence HN has to be an exten-
sion of degree 2 of this group. More precisely, this has to be the extension with the
graph automorphism. Namely, if q is not a square, A2 .q 3 / has only this non-trivial
degree 2 extension. If on the other hand q D r 2 is a square, then the field automor-
phism centralizes an A2 .r 3 /, the graph field automorphism a 2A2 .r 3 /. But neither of
these is contained in CG .1p / D F4 .q/  2, as is seen using a primitive prime divisor
of 9 .r/, 18 .r/ respectively.
The particular extension of A2 .q 3 / with the graph automorphism is again a dis-
connected group, and as above, but much easier, the following table of values is
obtained:

Table 5.3 Character values in A2 .q 3 /W 2

1 c2 cT
3 1 1 1
21 q2 0 1
111 q3 q 1

It contains all characters of A2 .q 3 / not vanishing on regular elements in its torus

of order q 6 C q 3 C 1 and on its unique class c2 of outer involutions. We find that
n.c2 ; c2 ; cT / D 1 in A2 .q 3 /W 2, but from Table 5.2 we see n.C2 ; C2 ; CT / D 0, which
shows that C2 does not intersect an A2 .q 3 /W 2. So finally we have proved that HN D
N and since the extension by the center is non-split, also that H D G.
G, t
u
Application of the Rigidity Criterion yields:
Theorem 5.7. The groups E6 .q/sc W 2, q D p m , p ¤ 2, possess G-realizations over
abelian number fields k.q/ Qab for the class vector C of Proposition 5.6.
By descent from E6 .q/sc W 2 to E6 .q/sc as in the proof of Corollary 1.5, and then
by taking factor groups modulo the center, this immediately gives Galois realiza-
tions for the simple groups:
Corollary 5.8. The groups E6 .q/sc , E6 .q/W 2 and E6 .q/, q odd, possess G-realiza-
tions over the same fields k.q/ as for E6 .q/sc W 2 in Theorem 5.7.
134 II Applications of Rigidity

As a corollary to the proof of Proposition 5.6 we also obtain:

Theorem 5.9. The groups E6 .q/ad W 2 and E6 .q/ad , q odd, possess G-realizations
over Qab .
Proof. The only cases to consider are those where E6 .q/sc 6Š E6 .q/ad , i.e., where
q 1 .mod 3/. Denote by C N the canonical image in GN D E6 .q/W 2 of the rigid class
vector C of G D E6 .q/sc W 2. Here the image CN 2p of the class C2p is again an outer
class, i.e., not contained in GN 0 D E6 .q/. The image TN of the maximal torus T of
G defined above extends to a maximal torus TQ of GQ D E6 .q/ad Š E6 .q/W 3. We
denote by CQ T the class of a generating element of this cyclic subgroup TQ of G. Q
Since the unipotent characters of a group of Lie type are classified independently
of the isogeny type of the group, precisely the same arguments as in the proof of
Proposition 5.6 show that the class vector C Q D .CN 2p ; CN 2p ; CQ T / of the extension GW
Q 2
of GQ by the graph automorphism has structure constant n.C/ Q D 1. Also, the proof
of generation remains valid, which yields rigidity of C. Q Since GW Q 2=GN 0 Š S3 the
descent to all groups above and including GN 0 D E6 .q/ is possible by the standard
descent argument. t
u
Since for q D p the group E6 .p/ad W 2 is the full automorphism group of the
simple group E6 .p/, we may conclude from this:
Corollary 5.10. For p ¤ 2 the groups E6 .p/ possess GA-realizations over Qab .
Remark. Theorem 5.7 realizes E6 .q/sc W 2 as a Galois group for the class vector C D
.C2p ; C2p ; CT /. By the Fixed Point Theorem I.7.2, we may obtain a further Galois
realization for a class vector C0 WD .C; C2p ; CT / of G, with C some class of outer
involutions different from C2 . The generating systems contained in ˙.C0 / arise
because the generating triple in C is a fixed point under the geometric automorphism
interchanging the first two (identical) classes. Indeed, the structure constant for C0
may be calculated as in Proposition 5.6, and turns out to be equal to 1. This gives a
second proof for the existence of Galois realizations.
The twisted groups 2E6 .q/ may be treated in an entirely analogous way. The
extension by the graph automorphism G D 2E6 .q/W 2 also exists for the twisted
groups. The classes for C may be chosen as before, except that we now take CT
to contain generators of a torus of order 18 D q 6  q 3 C 1. Then the class vector
.C2p ; C2p ; CT / for G is rigid when q is odd. The proof can be copied word for word
from the one for Proposition 5.6, using the relevant part of the character table given
in Table 5.4.
It may be noticed that Table 5.4 is obtained from Table 5.2 by simply replacing q
by q everywhere. This observation is explained by the Ennola duality for groups
of Lie type.
For generation, we again cite Weigel (1992), Fig. 7, who treats the case q >
3. The table for 2A2 .q 3 /W 2 corresponding to Table 5.3 is again obtained by just
replacing q by q, and the arguments go through as above. For q D 3, the methods
of loc. cit. again lead to the only possibility 2A2 .33 /W 2, for which we proceed as
before. We can thus prove:
5 The Exceptional Groups of Large Rank 135

Theorem 5.11. The groups 2E6 .q/sc W 2, q D p m , p ¤ 2, possess G-realizations over

abelian number fields k.q/ Qab for the class vector C D .C2p ; C2p ; CT /. The
groups 2E6 .q/sc , 2E6 .q/W 2 and 2E6 .q/ possess G-realizations over the same fields
k.q/.

Table 5.4 Character values in 2E6 .q/W 2

1 C2 C2p CT
1;0 1 1 1 1
4;1 2
q 4 8 10 12 q4 8 12 q.q C q C 1/
6 4
1
4;13 q 20 4 8 10 12 q 13 4 8 12 q 13 1
36 24
1;24 q q 0 1
6;600 1 7 2 3
3
q 3 6 8 10 12 1 4 2 2
3
q 3 6 8 12 3
q 3 6 .2q 4 C 1/
1 4
1
E6 Π1 7 4 6 2
3
q 1 2 4 8 10 1 4 4 4 2
3
q 1 2 4 8  13 q 4 13 23 4 1
E6 Π2 1 7 4 6 2
3
q 1 2 4 8 10 1 4 4 4 2
3
q 1 2 4 8  13 q 4 13 23 4 1

As for E6 .q/ we also obtain the result for the adjoint group from the above proof:

Theorem 5.12. The groups 2E6 .q/ad W 2 and 2E6 .q/ad , q odd, possess G-realizations
over Qab . For q D p this yields a GA-realization of 2E6 .p/, p odd, over Qab .

5.4 Rigidity for the Groups E6 .22mC1 / and 2E6 .22m /

The G-realizations given in the previous section needed the assumption p ¤ 2

already in the definition of the class C2p . It turns out that a rather similar class
also exists in characteristic two, but it contains involutions, so that the class vector
C would consist of two involution classes and hence could not contain generating
systems for our non-solvable groups.
In characteristic 2, at present rigid class vectors are only known when the adjoint
group coincides with the simply connected group, i.e., for E6 .22mC1 / and 2E6 .22m /.
This result will now be presented. First let G D E6 .22mC1 / where m  0. We refer
to the list of conjugacy classes of G given by Mizuno (1977). Let C2 D Œx1 be the
class of involutions central in a maximal unipotent subgroup of G, with centralizer
order q 36 15 23 32 4 5 6 , C8 D Œx11 the class of unipotent elements with central-
izer order q 15 1 , and finally let CT be our well known class of generators of a torus
with order 9 D q 6 C q 3 C 1. From the representative element x11 given by Mizuno
it is easily verified that elements in C8 indeed have order 8. All three classes exist
for all q D 22mC1 .
136 II Applications of Rigidity

Proposition 5.13. The class vector C D .C2 ; C8 ; CT / of E6 .22mC1 /, m  0, is rigid.

Proof. First one determines the irreducible characters not vanishing on the class
CT . As in the proof of Proposition 5.6 these are at most the unipotent characters
and the irreducible RT ./. Their values on the semisimple class CT follow readily
with (5.1), while for the determination of those on the unipotent classes C2 and C8
we need the Green functions in characteristic 2 calculated in Malle (1993c). There
the classes are denoted as Œu1 and Œu18 , and it is proved that both are uniform.
Hence the values can be computed from Table 6 in loc. cit. It turns out that only
three unipotent characters take non-zero values on both C8 and CT . They are listed
in Table 5.5, together with the family of irreducible RT ./. The structure constant
n.C/ D 1 then follows.

Table 5.5 Character values in E6 .22mC1 /

1 C2 C8 CT
1;0 1 1 1 1
20;2 2
q 4 5 8 12 2
q 5 q 22
1
64;4 q 4 23 42 62 8 12 q 4 22 4 5 6 .q 3  q C 1/ q4 1
RT . / 16 24 33 42 5 62 8 12 15 23 32 4 5 6 8 1 .q  q  1/
3
f1g

If we let H WD h i for  2 ˙N .C/, then as in the proof of Proposition 5.6 we see that
Weigel (1992), Fig. 6, applies. But none of the proper subgroups listed there con-
tains elements of order 8. This is clear for SL3 .22mC1 /, since its unipotent classes
all have representatives in SL3 .2/ Š L2 .7/. But then it also holds for the odd order
extension SL3 .22mC1 /:3. Since all other candidates different from G are contained
in the latter, this concludes the proof. t
u

Application of the Basic Rigidity Criterion now yields:

Theorem 5.14. The groups E6 .22mC1 /, m  0, possess G-realizations over abelian

number fields for the class vector C D .C2 ; C8 ; CT /.

Again this may immediately be transferred to the twisted groups. Namely, let
G WD 2E6 .22m /. Then the simple group again coincides with the one of adjoint
type. A classification of unipotent classes for these groups is contained in Malle
(1993c), Table 9. Let C2 D Œu1 be the unipotent class with centralizer order
q 36 13 25 3 4 62 10 , C8 D Œu18 the class of unipotent elements with centralizer
order q 15 2 , and CT a class of generators for the anisotropic torus of order 18 . All
three classes exist for all q.
With the same arguments as above, one proves that C D .C2 ; C8 ; CT / for G is
rigid. This leads to:
5 The Exceptional Groups of Large Rank 137

Theorem 5.15. The groups 2E6 .22m /, m  1, possess G-realizations over abelian
number fields for the class vector C D .C2 ; C8 ; CT /.

5.5 Rigidity for the Groups E7 .q/

The hardest case of all exceptional groups occurs for the groups of type E7 . Since
the Euler '-function of any integer n  3 is even, one may deduce that there exist
no good maximal tori T in groups of odd rank. Still, instead of a maximal torus we
can find a subtorus on which few characters do not vanish. The proofs in this section
will not be given in full detail.
Let G WD E7 .q/ad with q D p n the power of an odd prime p. Then the simple
group G 0 D E7 .q/ has index 2 in G. The semisimple conjugacy classes of G can
be found in Deriziotis (1983) or in Fleischmann and Janiszczak (1993), while
the unipotent classes were determined by Mizuno (1980). A class bearing some
similarity to the classes CT used so far is CTı , containing generators of the maximal
torus of order .q  ı/.q 6 C ıq 3 C 1/ with q ı .mod 3/, ı 2 f1; 1g. Define Cp
to be the unipotent class 4A1 in Mizuno (1980) or in Carter (1985), p.403. To
assure generation of G, choose the third class so as to contain involutions from
GnG 0 . According to Borel et al. (1970), Part F, 11 and 12, those have centralizer
structure A7 or 2A7 . More precisely, in the case q 1 .mod 4/ an involution with
centralizer A7 .q/ is not contained in G 0 , and in the case q 1 .mod 4/ the one
with centralizer A7 .q/ lies outside G 0 . For q " .mod 4/, " 2 f1; 1g, let the class
C2" consist of such elements.

Proposition 5.16. . The class vector C D .C2" ; Cp ; CTı / of E7 .q/ad , q D p m , p ¤ 2,

is rigid.

Proof. We first classify the Deligne–Lusztig characters RT 0 ./ not vanishing on the
third class CTı . This is equivalent to finding the semisimple conjugacy classes in the
dual T  of T . These are easily determined, and we list them in Table 5.6.

Table 5.6 Semisimple classes in T  .

o.s/ # of classes CG  .s/ # of chars.

1 1 G 76
2 1 G 76
j.q  ı/; ¤ 1; 2 .q  2  ı/=2 .q  ı/  Eı6 .q/ 30
6 j.q  ı/ .q  ı/.q 6 C ıq 3 /=18 T 1

Here the notation Eı6 .q/ is shorthand for E6 .q/ if ı D 1 and 2E6 .q/ if ı D 1 (note
that for the congruence q ı .mod 3/ the adjoint and the simply connected types
138 II Applications of Rigidity

of these groups coincide). The last column indicates how many irreducible char-
acters of G occur as constituents of the RT 0 ./ parametrized by the semisimple
element s.
The first two classes yield the extensions to E7 .q/ad D E7 .q/W 2 of the 76 unipo-
tent characters of G 0 . The last family consists of the irreducible RT ./, and they
vanish on C2" by Theorem 5.1, since T is not contained in the centralizer of such
involutions. The values of the remaining characters on our three classes can be cal-
culated from the Green functions and informations on the Weyl groups of types E6 ,
E7 and A7 . We omit the cumbersome details, and also the resulting table of val-
ues (they are given in Malle (1986), respectively in Lübeck and Malle (1998) for
p 5). It turns out that indeed n.C/ D 1.
For generation we may again turn to Weigel (1992), Fig. 9 and Fig. 10. First
assume ı D 1. Then we have to exclude the possibility .q  1/  E6 .q/  2 and the
parabolic subgroups of type Q  .q  1/  E6 .q/, with the unipotent radical Q. Denote
N
by H the group generated by a triple  2 ˙.C/. In the first case H .q 1/ E6 .q/
2, factoring by the normal subgroup H \ E6 .q/ obviously yields a p 0 -group, so it
has a .2; 1; 2/-system or reduces to the identity. In any case this would imply that
E6 .q/ contains elements of order .q  1/9 , which is not the case since its torus of
order 9 is self-centralizing. Similarly, the second possibility may be excluded, after
first factoring out the normal p-subgroup H \ Q.
The same arguments also apply when ı D 1, only here the second possibility
doesn’t even arise. This completes the proof of generation since C2" was chosen to
lie in E7 .q/ad n E7 .q/. t
u

Theorem 5.17. The groups E7 .q/ad , q D p m , p ¤ 2, possess G-realizations over

abelian number fields k.q/ Qab for the class vector C of Proposition 5.16.

As immediate corollary we find Galois realizations for the simple groups (here
Cp , Cp0 denote classes of E7 .q/0ad fusing into Cp of E7 .q/ad ):

Corollary 5.18. The groups E7 .q/ D E7 .q/0ad , q D p m , p ¤ 2, possess G-reali-

zations over the same fields k.q/ Qab for the class vector .Cp ; Cp0 ; .CTı /2 /
obtained by descent from the E7 .q/ad -realizations in Theorem 5.17.

5.6 The Groups E8 .q/

Let G WD E8 .q/, q D p m with p  3 odd. There exists a rough classification of the

conjugacy classes of G, namely the unipotent classes were found by Mizuno (1980)
and the types of semisimple classes by Deriziotis (1983). The Green functions
in good characteristic were calculated by Beynon and Spaltenstein (1984) under
the assumption that q is large enough; this additional hypotheses was subsequently
shown by Lusztig to be unnecessary.
Denote by C2 the class of involutions of G with centralizer of Lie type D8 and
by Cp the class of unipotent elements with representative ´189 in Mizuno (1980),
5 The Exceptional Groups of Large Rank 139

which corresponds to the class denoted 4A1 in Beynon and Spaltenstein (1984).
Finally, the class CT will contain generating elements of the cyclic torus of order
30 D q 8 C q 7  q 5  q 4  q 3 C q C 1.
Proposition 5.19. The class vector C D .C2 ; Cp ; CT / of E8 .q/, q D p m , p  3, is
rigid.
Proof. An element 3 2 CT generates a cyclic Hall-subgroup T of G D E8 .q/, as
follows from the order formula for G. In particular, all elements  2 T n1 are regular.
By Theorem 5.1(b) this means that among the Deligne–Lusztig characters RT 0 ./,
where T 0 denotes a maximal torus of G, only those where T 0 is G-conjugate to T
or where  D 1, do not vanish on the elements in CT . The first ones, with  ¤ 1, are
the irreducible Deligne–Lusztig characters for T , the second have as constituents
the unipotent characters of G. By Theorem 5.1(a) this implies that we have only to
consider the RT ./ and the unipotent characters for a possible contribution to the
structure constant n.C/.
Now T is self-centralizing and of odd order, so cannot be contained in the cen-
tralizer of an involution from class C2 . Thus the irreducible RT ./ vanish on C2 .
There remain the 166 unipotent characters . Their values on the class CT of regular
elements can easily be computed by formula (5.3) and the known character table of
the Weyl group W .E8 /. It turns out that precisely thirty unipotent characters do not
vanish on CT . So at most these may contribute to n.C/. (These thirty characters are
the non-exceptional characters in the principal l-block of G for any prime l > 31
dividing 30 .)
Next comes the more cumbersome step of calculating .C2 / for the remaining
thirty unipotent , which by Corollary 5.3 requires also knowledge of W .D8 / and its
fusion into W .E8 /. We omit the details of the straightforward computations. In the
end, only 14 of the remaining unipotent characters do not vanish on C2 . The values
of these on the class Cp now follow from the Green functions given in Beynon and
Spaltenstein (1984), using the fact that the unipotent class Cp is uniform. These
values consist of rather unpleasant polynomials of degree up to 25 in q, which are
hence omitted here (see Malle (1986) and Lübeck and Malle (1998) for the tables).
With the centralizer orders

jCG .1 /j D q 56 .q 2  1/.q 4  1/.q 6  1/.q 8  1/2 .q 10  1/.q 12  1/.q 14  1/;

jCG .2 /j D q 100 .q 2  1/.q 4  1/.q 6  1/.q 8  1/

N
and jCG .3 /j D 30 for  2 ˙.C/ one then obtains n.C/ D 1.
For the generation, we may employ Weigel (1992), Fig. 11, since H contains
the maximal torus T of order 30 generated by 3 . As the i have pairwise prime
orders, H is perfect by Proposition 4.3. It thus cannot lie in NG .T / D 30 :30, and
hence coincides with G. t
u
Theorem 5.20. The groups E8 .q/, q D p m , p ¤ 2, possess G-realizations over
abelian number fields k.q/ Qab for the class vector C of Proposition 5.19.
Some G-realizations of E8 .p/ over Q are contained in Paragraph 8.
140 II Applications of Rigidity

6 Galois Realizations of Linear and Unitary Groups over Q

When seeking G-realizations over Q, in addition to rigidity one has to require ratio-
nality of the class vectors. This imposes strong restrictions on the element orders
in the class triple, which then in turn complicate the proof of generation. In general
the conditions obtained in that way do not allow to realize a whole series of groups
of Lie type, not even all groups in a series defined over the prime field, but only
those where the defining characteristic satisfies some additional number theoretic
condition.
Before considering rational realizations, we extend the Galois realizations of
Belyi for the linear and unitary groups over Qab (Corollary 1.5 and Theorem 3.2)
to also include the graph automorphism of order 2, at least if the characteristic of
the ground field is odd. Under favorable circumstances this even allows to obtain
GA-realizations over Q.

6.1 Extension by the Graph Automorphism

Before we return to the linear and unitary groups over finite fields we first formulate
a kind of converse of the translation theorems.

Proposition 6.1. Let G be a finite group, C D .C1 ; C2 ; C3 / a rigid class vector of

G and  D .1 ; 2 ; 3 / 2 ˙.C/. Let Z Z .G/ be a subgroup of the center of G
N WD .CN 1 ; CN 2 ; CN 3 / is a rigid class
and denote by CN i the images of Ci in G=Z. Then C
vector of G=Z if for all 2 Z \ G there exist 1 ; 2 ; 3 2 Z such that 1 2 3 D 1
0

and i i is conjugate to i for i D 1; 2; 3.

N Thus there exist 0 2 Z such that 0  0 is con-
Proof. Let  0 D .10 ; 20 ; 30 / 2 ˙.C/.
Q3 i i i
jugate to i and moreover WD i D1 i0 i0 2 Z. But we also have 2 G 0 since i0 i0
is conjugate to i . By assumption there exist i 2 Z such that
0 0
i i i D i i i

and 1 2 3 D 1 . Thus . 1 10 10 ; 2 20 20 ; 3 30 30 / 2 ˙.C/, hence it is conjugate to

 . This implies that  0 is conjugate to  in G=Z. t
u

Let G be a finite group, C D .C1 ; C2 ; C3 / a rigid class vector of G and  an


automorphism of G of order 2 such that C1 D C1 , while C2 and C3 are inter-
changed by  . Let .1 ; 2 ; 3 / 2 ˙.C/. Since 2 and 3 are conjugate in G there
exists an element 1 2 G with 2 1 D 3 . Clearly a permutation of classes does not
change rigidity, so .C1 ; C3 ; C2 / is also rigid. But this implies that the two triples
.1 1 ; 3 ; 3 1 / and .21 1 2 ; 3 ; 2 / are conjugate by an element 2 2 G:

.1 ; 2 ; 3 / 1 2
D .1 1 ; 3 ; 3 1 / 2
D .21 1 2 ; 3 ; 2 /: (6.1)
6 Galois Realizations of Linear and Unitary Groups over Q 141

2
The element  WD  1 2 hence satisfies i D i for i D 2; 3. Since 2 2 G cen-
tralizes the generators 1 ; 2 ;  of the semidirect product GQ WD G:h i, it lies in
Z .G/Q \ G. The class vector C Q WD .Œ ; Œ 1 ; Œ2 / of GQ thus has the following
2
properties:
Q
h; 2 i D G; Q \ G;
2 D 2 Z .G/ .21 /2 D 31 21 D 1 : (6.2)
Q is rigid:
In one particular case it is easy to see that C

Corollary 6.2. Let G be a finite group, C D .C1 ; C2 ; C3 / a rigid class vector of

G,  an automorphism of G of order 2 such that C1 D C1 , C2 D C3 , and GQ D
G:h i the semidirect product. If Z .G/ D 1 then there exists a rigid class vector
CQ D .CQ 1 ; CQ 2 ; CQ 3 / of GQ such that C is the translation image of C
Q under the passage
Q
from G to G, i.e.,

CQ 12 D 1; CQ 22 D C1 ; CQ 3 D C2 [ C3 :

Proof. Since Z .G/ D 1 it follows from (6.2) that the class vector C Q defined above
Q
has C as its translation image under the passage from G to G. In the notation intro-
Q Since C is rigid we may assume after con-
duced before let .0 ; 0 21 ; 2 / 2 ˙.C/.
jugation that . 2 / D 1 . But then 0 1 centralizes the generators of G, so
0 1 2

0 D . Hence C Q is rigid. t
u

We now specialize to the situation we are most interested in. Let 2 ¤ p 2 IP be

an odd prime and q a power of p. Let

 W GLn .q/ ! GLn .q/; A 7! .At /1 ; (6.3)

be the graph automorphism of GLn .q/, which sends a matrix to the inverse of its
transpose. For n > 2 the corresponding semidirect product GLn .q/:h i is a non-
trivial extension of GLn .q/. If q D qQ 2 is a square then  normalizes (a suitable con-
jugate of) the subgroup GUn .q/ Q < GLn .q/, hence gives rise to a semidirect product
Q
GUn .q/:h i.

Lemma 6.3. Let q be odd and n  4 be even. Then all involutions in the coset
N q / are conjugate to   antidiag.1; : : : ; 1/. For the finite groups,  GLn .q/
 GLn .IF
and  GUn .q/ both contain two classes of involutions, with centralizers GOC n .q/
and GO n .q/.

Proof. If ./2 D 1 for  2 GLn .k/ then  D  t . Furthermore, ./ D   t , so

the classes of involutions in  GLn .k/ are in bijection with the equivalence classes
of non-degenerate symmetric forms on k n . If k D IF N q there is just one such class,
with representative   antidiag.1; : : : ; 1/ (see (3.5)). By definition the centralizer in
GLn .k/ is the orthogonal group GOn .k/. Since the connected component of this
centralizer has index 2, the class splits into two conjugacy classes in the finite groups
GLn .q/:h i or GUn .q/:h i. By (3.5) the centralizers in GLn .q/ of such involutions
are orthogonal group GO˙ 2n .q/. t
u
142 II Applications of Rigidity

Proposition 6.4. Let n  3, q odd, G GLn .q/ be absolutely irreducible and C D

.C1 ; C2 ; C3 / a rigid class vector of G. Assume that G is normalized by some  0 2
0
GLn .q/: and C1 D C1 , while C2 and C3 are interchanged by  0 .
(a) If either n is odd or C1 is a class of transvections, then there exists a rigid
class vector C Q D .CQ 1 ; CQ 2 ; CQ 3 / of the group extension GQ D G:h 0 i such that C is the
translation image of C Q under the passage from GQ to G, i.e.,

CQ 12 D 1; CQ 22 D C1 ; CQ 3 D C2 [ C3 :

Q with qQ 2 D q and let

(b) Assume moreover that G contains SLn .q/ or SUn .q/
GN WD ˙G=f˙1g. If C1 and C2 [ C3 are rational in G, then the image C NQ of C
Q is a
N
Q N 0
rationally rigid class vector of G D G:h i.

Proof. Let  2 ˙.C/ and  the element constructed above. We first claim that
2 D 1. Since G is absolutely irreducible and 2 centralizes the generators 1 ; 2 ; 
Q it lies in the center f˙1g of GLn .q/:h i. The squares of elements from the
of G,
coset  GLn .q/ have determinant 1, so for odd n we have 2 D 1. If n is even and
2 D 1 then .21 /2 D 31 21 D 1 , so the semisimple part of 21 is an
outer involution centralizing the transvection 1 . By Lemma 6.3, the centralizer of
an involution in  GLn .q/ is an orthogonal group, which gives a contradiction since
orthogonal groups in odd characteristic do not contain transvections (see Theo-
rem 2.2). So 2 D 1 in all cases, and by (6.2) the class vector C Q WD .Œ ; Œ 1 ; Œ2 /
2
Q
of G has the original class vector C as a translation image in G.
Q with .0  1 /2 D 1
As in the proof of Corollary 6.2, if .0 ; 0 21 ; 2 / 2 ˙.C/ 2
then   centralizes G, so  D a for some a1 2 Z .G/. By assumption 0 21 D
0 1 0

a21 is conjugate to 21 , so there exists a  2 G with .a21 / D 21 . To

show rigidity of C Q it suffices to show that a D b 2 for some b1 2 Z .G/ since then
. ;  2 ; 2 / D .; 21 ; 2 /b1 . If n is odd then evaluation of the induced from
0 0 1

G to GQ of the determinant on the previous equality yields that a is a square. For

even n note that .21 /2 D 1 by (6.2), hence  2 N WD NG .1 /. By the proof of
Theorem 1.4 the center of G has a complement M in N , N D Z .G/M , and it can
be checked that M remains normal in NGQ .1 /. In the factor group N=M Š Z .G/
two elements are conjugate via the outer automorphism only if they differ by a
square. This shows that a D b 2 for some b1 2 Z .G/.
For (b) we note that the classes of the involution  and of 2 are rational in GNQ
by assumption. Furthermore, the image of 21 in GNQ has order 2p, with rational
p-part, so it is rational as well. Finally,  is conjugate to  by Lemma 6.3 since
both are outer involutions with the same centralizer in GLn .q/. By Proposition 6.1
this shows that C NQ remains rigid in G. NQ t
u

In the next section we apply the preceding proposition to the case of general
linear and unitary groups extended by the graph automorphism.
6 Galois Realizations of Linear and Unitary Groups over Q 143

6.2 GA-Realizations over Qab

In odd characteristic the following result extends Belyi’s realization of linear and
unitary groups as regular Galois groups over Qab (Theorems 1.4 and 3.2) by includ-
ing the outer graph automorphism.

Theorem 6.5. Let n  3, p be an odd prime and q D p m .

(a) The group PGLn .q/:h i possesses a G-realization over Qab . The fixed field of
the simple group Ln .q/ is a rational function field.
(b) The group PGUn .q/:h i possesses a G-realization over Qab . The fixed field
of the simple group Un .q/ is a rational function field.

Proof. For (a) assume first that q ¤ 3. Let q 0 WD q in case (a) respectively q 0 WD q 2
in case (b), and

f .X / WD .X  a/.X  1/n1 with a 2 IF

q 0 n f˙1; 1  ng; a ¤ 1:
3
(6.4)

(Such an a exists since q 0  5.) By the choice of a the sets of roots of f .X / and
f .X / are disjoint, so by Theorem 2.6 there exists a Belyi triple  D .1 ; 2 ; 3 /
in GLn .q 0 / such that 2 and 31 both have characteristic polynomial f .X /. More-
over by Proposition 2.7 the group G WD h i is primitive since f has trace different
from 0. Since a suitable p-power of 2 is a homology of order larger than 3, G
contains either SUn .q/Q or SLn .q/
Q as a normal subgroup by Theorem 2.3.
In case (a) we now choose a to be a generator of the cyclic group IF q . Since
GUn .q/ Q does not contain an element with minimal polynomial f .X / for any qjq Q 2
we may conclude that G D GLn .q/ in this case. In case (b) we choose a to be of
multiplicative order q C 1 in IFq 2 , so aq D a1 . Then the action of the Frobenius
morphism (taking q-th powers) interchanges the classes of 2 and 3 , while it fixes
the class of 1 . Since the same is true for  , the group G is now centralized by a
suitable conjugate of the product of the Frobenius map with  . It thus cannot be a
linear group which implies G D GUn .q/.
If n is odd then 1 is a reflection and we have CG .1 / D GL1 .q/  GLn1 .q/
(respectively GU1 .q/  GUn1 .q/ in case (b)) so the center Z .G/ has a comple-
ment in the normalizer of h1 i. For n even 1 is a transvection and the center of G
has a complement in the normalizer of h1 i by the proofs of Theorems 1.4 and 3.2.
Thus the Basic Rigidity Theorem I.4.8 yields the existence of a G-realization of G
over Qab . Furthermore, by Proposition 6.4(a) there exists a rigid class vector C Q of
G:h i possessing C as a translation image. Thus by Proposition 6.4(b) and Theo-
rem I.4.8 there exists a G-realization of the group G:h i=f˙1g with trivial center.
Trivially this furnishes a G-realization of the factor group PGLn .q/:h i (respec-
tively PGUn .q/:h i in case (b)) with respect to the canonical image of the class
vector C.Q Furthermore, the fixed field K of the simple group Ln .q/ (resp. Un .q/) is
rational, since the subextension corresponding to K is a dihedral extension ram-
ified at three points with ramification indices .2; 2; k/, where k D gcd.q  1; n/
(resp. k D gcd.q C 1; n/).
144 II Applications of Rigidity

If q D 3 in case (a) we start with the polynomial

f .X / D .X  1/n2 .X 2 C X C 2/

instead, the roots of the quadratic factor being an 8-th root of unity and its third
power. As before there exists a Belyi triple  . The group G WD h i now at least
contains a transvection or a reflection, so we may apply Theorems 2.2 or 2.4. In even
dimension, the symplectic group may be excluded by the following observation. If 
is a semisimple element in the conformal symplectic group CSpn over an algebraic
closure of IFq then it has eigenvalues ft1 ; : : : ; tm ; ut11 ; : : : ; utm
1
g for some ti ; u ¤ 0
(see Digne and Michel (1991), 15.2, for example). Thus the semisimple part of 2
cannot lie inside the normalizer of a symplectic group. The orthogonal groups in
odd dimension may again be ruled out by the eigenvalue distribution of the element
2 . If G were a symmetric group, then 1 would have sign 1, and 2 ; 3 would
have to have the same sign. But this is impossible because of the product relation.
The four 3 and 5-dimensional exceptional reflection groups do not possess elements
of order 8, and in the 7-dimensional group 2: O7 .2/ all classes are rational. So G
is none of the exceptional cases, and hence contains SLn .3/ as a normal subgroup.
Then the rest of the proof goes through without change. t
u

Since Aut.Ln .p// D PGLn .p/:h i and Aut.Un .p// D PGUn .p/:h i for n  3,
we have the immediate consequence (extending Corollary 1.6):

Corollary 6.6. Let p be an odd prime. Then the simple groups Ln .p/ and Un .p/
possess GA-realizations over Qab .

6.3 GA-Realizations over Q

We now consider cases were the above approach leads to realizations over the field
of rational numbers Q. The element 2 used for the Galois realizations in the pre-
vious section is semirational only if a has multiplicative order 4 or 6. This leads
to:

Theorem 6.7. Let n > 2 be even and p > 2.

(a) For gcd.n; p  1/ D 2 and either n 2 .mod 4/ and p 5 .mod 8/, or
n 0 .mod 4/ and p 7 .mod 12/, the group Ln .p/ possesses a GA-realization
over Q.
(b) For gcd.n; p  1/ D 2 and either n 2 .mod 4/ and p 3 .mod 8/, or
n 0 .mod 4/ and p 5 .mod 12/, the group Un .p/ possesses a GA-realization
over Q.

Proof. We first consider case (a). If p 1 .mod 4/ there exists an element a 2 IFp
with a2 D 1. If p 1 .mod 6/ there exists an element a 2 IFp with a3 D 1, a ¤
1. These are different from ˙1 and 1  n. Thus the polynomial f .X / D .X  a/
.X  1/n1 is as in (6.4) and there exists a Belyi triple .1 ; 2 ; 3 /, where 1 is
6 Galois Realizations of Linear and Unitary Groups over Q 145

a transvection and 2 is semirational. The arguments in the proof of Theorem 6.5

now give a semirationally rigid class vector C in GLn .p/ for a group G containing
SLn .p/. Since under the given congruences a is a non-square in IFp , the quotient
GN of G by its center is an extension of Ln .p/ of degree 2. Under the assumption
gcd.n; p  1/ D 2 this coincides with all of PGLn .p/. By Proposition 6.4 the class
vector C is the translation image of a rigid class vector of GLn .p/:h i, such that
after factoring out f˙1g we obtain a rationally rigid class vector. Taking the quotient
modulo the center we obtain a G-realization N=Q.t/ of PGLn .p/:h i D Aut.Ln .p//
over Q. Since the quotient by the simple group Ln .p/ is a four group, we may now
apply Lemma I.9.9 to conclude that the fixed field L of the simple group is a rational
function field. Indeed, the order of the centralizer of 3 in PGLn .p/ equals the one
in PGLn .p/:h i, so the full normalizer of the inertia group of P3 lies already in
PGLn .p/.
In case (b) for p 3 .mod 4/ there exists a 2 IFp2 n IFp with a2 D 1, and for
p 1 .mod 6/ there exists an a 2 IFp2 n IFp with a3 D 1, a ¤ 1. We can now
argue as before, using that a is a non-square precisely under the conditions stated in
the Theorem. t
u

Remark. The assumption n 0 .mod 4/ when p 7 .mod 12/ is necessary in

the previous construction, since otherwise the fixed field of the simple group can be
shown not to be rational over Q. In particular the stronger assertion in Malle (1996),
Satz 4.8, is not correct.

In odd dimension we obtain a similar result:

Theorem 6.8. Let n be odd.

(a) For gcd.n; p  1/ D 1, p > 3 and p 6 1 .mod 12/ the group Ln .p/ pos-
sesses a GA-realization over Q.
(b) For gcd.n; p C1/ D 1, p > 2 and p 6 1 .mod 12/ the group Un .p/ possesses
a GA-realization over Q.

Proof. In case (a), if p 1 .mod 4/ or p 1 .mod 6/ there exists an element

a 2 IFp with a2 D 1 or with a3 D 1, a ¤ 1. Again, f .X / D .X  1/n1 .X  a/
is as in (6.4) and there exists a semirational Belyi triple .1 ; 2 ; 3 /. The proof
now proceeds as the one for Theorem 6.7, except that the condition on a being a
non-square is not needed here because gcd.n; p  1/ D 1 implies Aut.Ln .p// D
Ln .p/:h i. Similarly for case (b) there exist suitable elements a 2 IFp2 n IFp under
the given congruences. t
u

Remark. For more general results on realizations of linear and unitary groups cov-
ering further congruence classes see Reiter (1999). Linear groups have also been
studied by Folkers (1995) starting from rather different class vectors. See also Chap-
ter III.10 for further G-realizations of linear and unitary groups not necessarily over
the prime field.
146 II Applications of Rigidity

7 Galois Realizations of Symplectic and Orthogonal Groups

over Q

For symplectic and orthogonal groups over the prime field we can apply the tech-
niques of Paragraph 2 to construct Galois realizations over Q. The results are essen-
tially due to Reiter (1999), in characteristic 2 also to Völklein (1998). In the last
section, G-realizations over Qab for the groups OC 8 .q/ with the exceptional triality
graph automorphism are proved, using Lusztig’s character theory and the classifi-
cation of the finite simple groups. For infinitely many primes p this leads to GA-
realizations over Q for the groups OC8 .p/ over the prime field.

7.1 GA-Realizations of Symplectic Groups over Q

In this section we find Galois realizations over Q for infinite series of symplectic
groups. But first we record a result on rationality of conjugacy classes.

Proposition 7.1. (a) A semisimple element in SO2nC1 .q/, CO˙ 2n .q/ or CSp2n .q/
is rational if and only if it is rational considered as element of the corresponding
general linear group.
(b) Let  be an element of G 2 fSO2nC1 .q/; CO˙ 2n .q/; CSp2n .q/g whose minimal
polynomial coincides with its characteristic polynomial and which is rational in the
corresponding general linear group. Then  is rational in G.

A proof can be found in Reiter (1999), Bem. 3.10 and Folg. 3.4. The first part is
a consequence of the Theorem of Lang-Steinberg. For the second part it suffices by
the first to show that the unipotent part u of  is rational inside the centralizer of
the semisimple part s . This is true since the centralizer of s is a group of classical
type, in which u is a regular unipotent element. But the regular unipotent elements
in groups of type SO2nC1 .q/, CO˙ 2n .q/, CSp2n .q/ are rational.

Theorem 7.2. Let n  2. For odd primes p 6 ˙1 .mod 24/, p6 jn, the symplectic
groups S2n .p/ possess GA-realizations over Q.

Proof. We define
8
ˆ
<.X  a8 / .X  a8 /
n 3 n
if p 3 .mod 8/;
f .X / D .X  1/n .X  a4 /n if p 5 .mod 8/;
:̂
.X  1/n .X  a6 /n if p 7 .mod 12/;

where ai denotes a primitive i -th root of unity in IFp2 . Note that by the chosen
congruences we have f .X / 2 IFp ŒX . Then by Theorem 2.6 there exists a Belyi
triple  D .1 ; 2 ; 3 / such that 1 is a transvection and 2 , 31 have f .X / as
their characteristic polynomial. The group G WD h i acts primitively by Propo-
sition 2.7, hence by Proposition 2.1 and Theorem 2.2 we have Sp2n .p/ E G or
7 Galois Realizations of Symplectic and Orthogonal Groups over Q 147

SL2n .p/ E G. By the criterion of Belyi we obtain n.C/ D 1 for the class vector
C D .Œ1 ; Œ2 ; Œ3 /. Now the roots of f are of the form (2.8) for
8
ˆ
<1 if p 3 .mod 8/;
c WD a4 if p 5 .mod 8/;
:̂
a6 if p 7 .mod 12/;

so G CSp2n .p/. Furthermore, 2 is conjugate to c21 in GL2n .p/. Thus Œ2

is not rational, hence semirational in CSp2n .p/ by Proposition 7.1. Again by the
choice of a it follows that GZ .GL2n .p// D CSp2n .p/. By the criterion of Belyi
this implies l.C/ D 1. The normalizer condition is verified as in the proof of Theo-
rem 3.4 and an application of Theorem I.4.8 yields the Galois realization. The factor
group of G by Sp2n .p/ is cyclic, so the fixed field of Sp2n .p/ is a rational function
field. Taking the quotient by the center gives the assertion.
For primes p 5 .mod 12/ we choose
p
f .X / WD .X  a6.p1/ /n .X  a6.p1/ /n ; g.X / WD .X 2  ap1 /n ;

where again ai denotes a primitive i -th root of unity in IFp2 . As before the group
G generated by a Belyi triple  D .1 ; 2 ; 3 / corresponding to f; g contains the
symplectic group and is contained in CSp2n .p/, since the roots of f are of the
5p7
form (2.8) for c WD a6.p1/ . Since 3 is conjugate to 3 , the image N of  in
CSp2n .p/=Z .CSp2n .p// D PCSp2n .p/ is still rigid by Proposition 6.1. By the
choice of 2 we have GN WD hN i D PCSp2n .p/. The images of 2 ; 3 in GN are semi-
rational by Proposition 7.1, so we may argue as above to obtain a GA-realization of
S2n .p/ over Q. t
u

To conclude this topic we prove a related result in characteristic 2.

Theorem 7.3. The symplectic groups S2n .2/ possess GA-realizations over Q.

Proof. For n D 2 we have S4 .2/ Š S6 which was treated in Corollary I.9.8. For
n  3 we choose

f .X / WD .X 2 C X C 1/n ; g.X / WD 5 .X /.n3/=2 7 .X /;

if n is odd, respectively

f .X / WD .X 2 C X C 1/n3 9 .X /; g.X / WD 5 .X /n=2 ;

if n is even, with the cyclotomic polynomials 5 , 7 and 9 . Then f; g are coprime

of degree 2n with non-vanishing constant coefficient. By Theorem 2.6 there exists
a corresponding Belyi triple  D .1 ; 2 ; 3 / where 1 is a transvection. The group
G WD h i GL2n .2/ is contained in Sp2n .2/ by Theorem 2.10(b). Furthermore, G
is primitive by Proposition 2.7 since 2 has trace different from 0. The normal sub-
group H of G generated by the conjugates of 1 is thus one of the irreducible groups
148 II Applications of Rigidity

in Theorem 2.2. In particular H is primitive. The orthogonal groups can be ruled

out by Theorem 2.10(c). The minimal polynomials show that H is not a symmetric
group. Finally, the order of S6 .2/ is not divisible by the order of 31 :U4 .3/:22 .
The elements i were chosen such that any primitive power of i has the same
eigenvalues as i . Thus  is rationally rigid by Proposition 7.1. Since for n  3
the group S2n .2/ coincides with its automorphism group we obtain GA-realizations
over Q. t
u

Remark. For further GA-realizations of symplectic groups over Q see Reiter (1999)
and Chapter III.10.

7.2 GA-Realizations of Odd-Dimensional Orthogonal Groups

In odd dimension we obtain Galois realizations of orthogonal groups. These general-

ize the GA-realization for the class vector .2B; 4A; pA/ for . p2 / D 1, .2B; 6A; pA/
for . p3 / D 1 of L2 .p/ Š O3 .p/ in Corollary I.8.10 to arbitrary dimensions.

Theorem 7.4. Let n  1. For odd primes p 6 ˙1 .mod 24/ the odd-dimensional
orthogonal groups O2nC1 .p/ possess GA-realizations over Q.

Proof. With a primitive 4-th or 6-th root of unity a in IFp2 we define

f .X / WD .X  1/2n1 .X  a/.X  a1 /; g.X / WD .X  1/2nC1 :

By Theorem 2.6 there exists a Belyi triple  D .1 ; 2 ; 3 / such that 1 is a reflection
and 2 ; 3 have minimal polynomials f .X /; g.X /. By Theorem 2.10(a) we have
G WD h i GO2nC1 .p/. If 2n C 1 6 0 .mod p/ then tr.3 / ¤ 0 so G is primitive
since pjo.2 /6 j.2n C 1/.p  1/. On the other hand if pj.2n C 1/ then tr.2 / ¤ 0 so
by Proposition 2.7 either G is primitive or it is contained in the monomial subgroup
G.p  1; 1; 2n C 1/. But the latter case is ruled out by the form of f .X /. Thus G is
primitive and the normal subgroup H of G generated by the class of the reflection 1
acts irreducibly. By Proposition 2.1 it even acts primitively whence H is a primitive
reflection group contained in GO2nC1 .q/. The symmetric groups are ruled out by
using that g.X / is not the characteristic polynomial of a permutation matrix. In
dimension 2n C 1 D 3 the orthogonal group SO2nC1 .p/ is isomorphic to PGL2 .p/
for which the assertion was shown in Corollary I.8.10. For 2n C 1  5 the element
2 has order 4p, resp. 6p, and this rules out the exceptional cases in Theorem 2.4.
Hence G must contain the orthogonal group O2nC1 .p/ as a normal subgroup. By
the chosen congruences the semisimple part of 2 does not lie in the commutator
group of SO2nC1 .p/, thus we have G D SO2nC1 .p/.
The element 3 is regular unipotent and the unipotent part of 2 is a regular
unipotent element of SO2n1 .p/, thus both classes are rational by Carter (1985),
Prop. 5.1.7. So C WD .Œ1 ; Œ2 ; Œ3 / is a rational class vector, which is rigid
by the criterion of Belyi. Since the center of G is trivial, Theorem I.4.8 yields a
7 Galois Realizations of Symplectic and Orthogonal Groups over Q 149

Galois realization of SO2nC1 .p/ over Q, which is the full automorphism group of
the simple subgroup O2nC1 .p/ of index 2. t
u

We even obtain GA-realizations over Q for groups O2nC1 .p 2 /, generalizing a

result of Feit (1984) for L2 .p 2 / Š O3 .p 2 /:

Theorem 7.5. Let n  1.

(a) For odd primes p the odd-dimensional orthogonal groups O2nC1 .p 2 / possess
GA-realizations over Qab .
(b) For odd primes p ˙2 .mod 5/ the odd-dimensional orthogonal groups
O2nC1 .p 2 / possess GA-realizations over Q.

Proof. For (a) let a be a generating element of IF

p4
, for (b) let a 2 IF
p4
be of
multiplicative order 10. Let

f .X / WD .X  1/2n1 .X  a/.X  a1 /; g.X / WD .X  1/2n1 .X  ap /.X  ap /;

so that g is the image of f under the generating automorphism of IFp2 =IFp . Let
 D .1 ; 2 ; 3 / be a Belyi triple corresponding to this choice of characteristic poly-
nomials according to Theorem 2.6. By Theorem 2.10(a) we have G WD h1 ; 2 i
GO2nC1 .p 2 /. Also, G is not contained in GL2nC1 .p/ since f; g are not rational
over IFp by the choice of a. As in the proof of Theorem 7.4 it can then be checked
that G D SO2nC1 .p 2 /.
Now let F W GL2nC1 .p 2 / ! GL2nC1 .p 2 / be the Frobenius morphism sending
each matrix entry to its p-th power. Replacing G by a suitable conjugate we may
assume that it is normalized by F . Then F fixes the class of 1 and exchanges the
classes of 2 and 3 . Thus we may apply Corollary 6.2 with the automorphism  of
order 2 induced by F to obtain a rigid class vector C Q of GQ WD SO2nC1 .p 2 /:h i Š
Aut.O2nC1 .p 2 //. Since i , i D 2; 3, in (b) are semirational, CQ is rationally rigid. We
next claim that the coset G contains a single class of involutions, the class of  .
Indeed, assume . /2 D 1 D F ./. Then by the Theorem 1.1 of Lang-Steinberg
there exists  2 SO2nC1 .IF N q / with  1 F ./ D , which even lies in G because of

F 2 ./ D F ./ D F ./ D :

Q is the
But then  D  1 F ./ D  1   is conjugate to  . Thus the first class of C
class of  . Since NG . / D CG . / D SO2nC1 .p/ is contained in G 0 D O2nC1 .p 2 /
by definition of the spinor norm (3.3) it follows by Lemma I.9.9 that the G-
realization of GQ is a GA-realization of O2nC1 .p 2 /. t
u

Remark. For other results on GA-realizations of odd-dimensional orthogonal groups

see Malle (1996), Satz 6.9, Reiter (1999) and Chapter III.10.
150 II Applications of Rigidity

7.3 Even-Dimensional Split Orthogonal Groups

The outer automorphism groups of the split even-dimensional orthogonal groups

always have order divisible by 4, so the descent from realizations of the automor-
phism group to the simple group is more difficult even over the prime field. We
first look at the case where j Out.OC 2n .p//j D 4, i.e., where n is odd and p 3
.mod 4/. Let G WD PCOC 2n .p/ denote the projective conformal split orthogonal
group, G ı WD PCOCı 2n .p/ the finite group associated to the connected component
of the identity in the projective conformal orthogonal group, which is normal of
index 2 in G, and G 0 WD OC ı
2n .p/ the commutator subgroup of G . Since n is odd,
0 0
we have G=G Š 2 and G is simple for all n  3.
2

Theorem 7.6. Let n  3 be odd and p a prime with p 3 .mod 8/ or p 7

.mod 12/, p6 jn. Then there exist regular Galois extensions of Q.t/ with groups
PCOC C C C
2n .p/, PSO2n .p/ and O2n .p/. In particular, O2n .p/ then possesses a GA-
realization over Q.

Proof. If p 3 .mod 8/ then let

f .X / WD .X 2  1/.X  a/n1 .X  ap /n1 ; g.X / WD .X 2 C 1/n ; (7.1)

with a 2 IF
p2
of multiplicative order 8, while for p 7 .mod 12/ let

f .X / WD .X  1/n .X  b/n ; g.X / WD .X 2  b/n ; (7.2)

with b 2 IF p of multiplicative order 6. By Theorem 2.10(d) the group G WD h i

GL2n .p/ generated by a Belyi triple belonging to f; g is a subgroup of CO˙ 2n .p/.
By Lemma 3.9 applied to the semisimple part of 2 we see that G COC 2n .p/. The
group G is primitive by Proposition 2.8. Now by Lemma 2.1 the subgroup H gen-
erated by the conjugates of the reflection 1 is among the groups in Theorem 2.4.
The symmetric groups can be excluded since g.X / has no eigenvalue 1. The excep-
tional groups in dimensions 5 and 7 can be excluded by the order of 2 . Hence we
have that G contains C 2n .p/. By the chosen congruences the element 2 gener-
ates COCı2n .p/ over  C
2n .p/. Since G also contains the reflection 1 we obtain that
G D COC 2n .p/. Now  3 is conjugate to 3 , so by Proposition 6.1 the image N of
 in GN WD G=Z .G/ Š PCOC 2n .p/ is still rigid. By their distribution of eigenvalues
all three elements of N are rational in GN by Proposition 7.1. We may thus apply the
rigidity criterion to obtain a G-realization N=Q.t/ with group PCOC 2n .p/. We may
descend to the normal subgroup OC 2n .p/ of index 4 by virtue of Lemma I.9.9 since
the centralizer of the reflection 1 is contained in the subgroup of order 2 generated
over OC 2n .p/ by 1 . Thus the Galois realization is a GA-realization of the simple
group OC 2n .p/. t
u

In the case where n is even or p 1 .mod 4/ the outer automorphism group of

OC
2n .p/ is a dihedral group of order 8.
7 Galois Realizations of Symplectic and Orthogonal Groups over Q 151

7.4 Even-Dimensional Non-split Orthogonal Groups

As for the split orthogonal groups we subdivide the cases according to the outer
automorphism group of O 2n .p/. If n is even or p 1 .mod 4/ then Out.O 2n .p//
is of order 4.

Theorem 7.7. Let n  3 be odd and p 5 .mod 12/, or n  6 be even and p 3

.mod 8/, or n 2 .mod 4/ and p ˙2 .mod 5/. Then there exist regular Galois
extensions of Q.t/ with groups PCO  
2n .p/, PSO2n .p/ and O2n .p/. In particular,

O2n .p/ then possesses a GA-realization over Q.

Proof. For n odd, p 5 .mod 12/ choose

p
f .X / WD .X  a6.p1/ /n .X  a6.p1/ /n ; g.X / D .X 2  ap1 /n ;

for n even, p 3 .mod 8/ let

f .X / WD .X 2 1/.X a8 /n1 .X a83 /n1; g.X / WD .X 2 C1/n1 .X a85 /.X a87 /;

and for n 2 .mod 4/, p ˙2 .mod 5/, let

2 3
f .X / D ..X  b/.X  b p /.X  b p /.X  b p //n=2 ; g.X / D .X 2  ap1 /n ;

with ai ; b 2 IF
p4
of multiplicative order i , 10.p  1/ respectively. We can now argue
as in the proof of Theorem 7.6 to verify the assertions of the Theorem. t
u

Remark. For n D 2 we have PCO 4 .p/ Š PL2 .p /, and the GA-realization of

2

L2 .p 2 /, p ¤ 2, p ˙2 .mod 5/, obtained above is the one constructed by Feit

(1984).
For n D 3 Theorem 7.7 yields Galois realizations for the unitary groups U4 .p/ Š
O6 .p/, p 5 .mod 12/.

Over the prime field with two elements we obtain complete results:

Theorem 7.8. For n  5 the orthogonal groups O˙

2n .2/ possess GA-realizations
over Q.

Proof. For the orthogonal groups of plus-type let

(
.X 2 C X C 1/n3 7 .X / if n is odd;
f .X / WD .X  1/2n ; g.X / WD
.X 2 C X C 1/n if n is even;

and  D .1 ; 2 ; 3 / a Belyi triple corresponding to f; g. By Theorem 2.10(c) the

group G WD h i is contained in GO˙ 2n .2/. An element with characteristic polyno-
mial g.X / can be found in OC 2 .4/ n=2
OC C
2n .2/ if n is even and in O2 .4/
.n3/=2

C C C
O6 .2/ O2n .2/ if n is odd, thus G GO2n .2/ by Lemma 3.9. If n is odd then
G is primitive since 3 has trace different from 0. If n is even then we are not in
152 II Applications of Rigidity

the case (a) of Proposition 2.8. But case (b) cannot occur since q D 2. Thus G is
primitive and the normal subgroup H of G generated by the transvections conjugate
to 1 is one of the irreducible groups in Theorem 2.2. The symmetric groups cannot
occur by the distribution of eigenvalues of 2 ; 3 . Since n  5 this only leaves the
orthogonal group.
The elements 2 ; 3 with minimal polynomials f , g are rational in GO 2n .2/ by
Häfner (1992), Bem. 2.3. Since 1 is a transvection of order 2, this shows that  is
a rationally rigid system of generators of G D GOC 0
2n .2/. The simple group G has
0
index 2 in G, so the Galois realization for G is obtained by the standard descent
argument. Finally, G is the automorphism group of G 0 , and the constructed Galois
extension gives a GA-realization.
For the non-split group we choose
(
.X 2 C X C 1/n if n is odd;
f .X / WD .X  1/ ;2n
g.X / WD
.X 2 C X C 1/n2 5 .X / if n is even:

Elements with characteristic polynomial g.X / can be found in O 2 .2/

n
O
2n .2/
  
if n is odd, and in O2 .2/ n2
 O4 .2/ O2n .2/ if n is even, so the group G gen-
erated by a Belyi triple  corresponding to f; g is contained in GO2n .2/ by Theo-
rem 2.10(c) and Lemma 3.9. The argument now proceeds as in the split type. t
u

Remark. For earlier results on GA-realizations of orthogonal groups over IF2 see
Thompson (1984b) and Häfner (1992), Satz 4.5 and 4.6. Further G-realizations of
orthogonal groups will be presented in Chapter III.10.

7.5 The 8-Dimensional Split Orthogonal Groups

Let G WD PCOC 8 .q/:hıi, q D p , be the extension of the projective conformal 8-

m

dimensional split orthogonal group over the finite field IFq with the triality graph
automorphism ı of order 3. Then we have
(
C S4 ; if p > 2;
G= O8 .q/ Š
S3 ; if p D 2:

Let C2 be the class of the graph automorphism  in G of order 2 with centralizer

of type GO7 .q/ D SO7 .q/  2. The graph automorphism ı of order 3 centralizes a
subgroup of type G2 .q/; let C3C be the class of the product of ı with a (regular)
element of order q 2 C q C 1 in its centralizer, and analogously C3 the class of
the product with a (regular) element of order q 2  q C 1. Finally denote by T a
 -invariant maximal torus of the connected component G ı WD PCOC ı
8 .q/ < G of
order .q C 1/ . This has the structure of a direct product of two cyclic groups of
2 2

order q 2 C 1. The graph automorphism  interchanges the two factors, hence there
exist elements of order 2.q 2 C 1/ in the coset T: . Let C4 be the class of such an
7 Galois Realizations of Symplectic and Orthogonal Groups over Q 153

element. The centralizer of  2 C4 in G has order 2.q 2 C1/q.q 2 1/. Let C˙ denote
the class vector C˙ WD .C2 ; C3˙ ; C4 / of G.

Proposition 7.9. The class vectors C˙ of G satisfy l.C˙ / D 1.

Proof. We evaluate the formula for the normalized structure constant of C˙ . This
again requires results from the character theory of disconnected groups. No non-
trivial semisimple element of G has centralizer order divisible by both 3 and 4 ,
respectively by 6 and 4 . By the character formula in Digne and Michel (1994)
this implies that apart from extensions of unipotent characters of G ı , no other char-
acters take non-zero values on all three classes of C˙ . These extensions are the
constituents of the generalized Deligne–Lusztig characters RT;1 . For disconnected
groups of type D4 W 2 and D4 W 3, the decomposition of these RT;1 into unipotent
characters was determined in Malle (1993b), Thms. 6 and 7. From this the decom-
position in the case D4 W S3 can immediately be deduced. The values of the exten-
sions of unipotent characters on the three classes of C˙ may now be calculated
using the character formula in Digne and Michel (1994). One finds that only four
unipotent characters give non-zero contributions to n.C˙ /. Their values are given in
Table 7.1. From the definition of the classes C2 ; C3˙ ; C4 we can read off the central-
izer orders jCG .1 /j D 2q 9 .q 2  1/.q 4  1/.q 6  1/, jCG .2 /j D 3.q 2 ˙ q C 1/ and
.˙/
jCG .3 /j D 2q.q 4  1/ for elements i 2 Ci . With this information the assertion
˙
n.C / D 1 follows.

Table 7.1 Character values for OC

8 .q/:S4

1 C2 C3C C3 C4

4; 1 1 1 1 1
1 3 4 1 2 3
12;1 2
q 2 6 2
q 2 6 1 : 2
cusp 1 3 4
2
q 1 3  12 q 2 13 3 : 1 1
12 9
14 ; q q 1 1 q

For the proof of generation we make use of the explicit determination of the
conjugacy classes of maximal subgroups of groups of type D4 by Kleidman (1987).
Let p1 and p2 be primitive prime divisors of q 2 ˙ q C 1 and q 2 C 1 in the sense of
Zsigmondy (Proposition 4.1). It can be checked that among the maximal subgroups
of G ı none of order divisible by both p1 and p2 extends to the extension by the
group of graph automorphisms. Thus any triple  2 ˙.C/ N generates G and the
assertion follows from Corollary I.5.9. t
u

Theorem 7.10. (a) The groups PCOC 8 .q/:hıi, q D p , possess G-realizations over
m
ab C
Q . The fixed field of the simple normal subgroup O8 .q/ is a rational function field.
(b) The groups OC ab
8 .p/ possess GA-realizations over Q .
154 II Applications of Rigidity

Proof. By Proposition 7.9 the class vectors C˙ of G are rigid. By the Basic Rigidity
Theorem I.4.8 this implies the existence of a G-realization N=Qab .t/ with group G.
The fixed field K of the simple group OC ab
8 .q/ has Galois group S4 over Q .t/ and
class vector .2; 3; 4/, if p is odd, respectively group S3 and class vector .2; 3; 2/ if
p is even. Thus K has genus 0 and is rational over Qab .
In the case q D p we have G D Aut.OC 8 .p// and assertion (b) follows from the
above considerations. t
u

Replacing the conjugacy classes C3˙ and C4 by suitable rational classes, we

obtain GA-realizations over Q.

Theorem 7.11. Let p 2 IP with p ˙2 .mod 5/ and p ˙2; ˙3 .mod 7/. Then
OC
8 .p/ possesses a GA-realization over Q.

Proof. This is an easy consequence of the proof of Proposition 7.9. If p 6 ˙1

.mod 5/ then 5j.p 2 C 1/, and 5 is a primitive prime divisor of 4 .p/. Furthermore,
if p 6 ˙1 .mod 7/, then 7 divides either q 2 C q C 1 or q 2  q C 1 and is a primitive
prime divisor of the corresponding cyclotomic polynomial 3 .p/ or 6 .p/. Thus if
we replace the class C4 by an analogously defined class C10 of elements of order 10,
and C3˙ by the corresponding class C21 of elements of order 21, then by the charac-
ter formula for unipotent characters extended to the disconnected case (Digne and
Michel (1994)) the values in Table 7.1 remain correct for the new classes, hence
the normalized structure constant n.C2 ; C21 ; C10 / also equals 1. In the proof of gen-
eration we had only used properties of primitive prime divisors, and since 5 and 7
are such primitive divisors, the generation property also remains true. So under the
assumptions of the theorem, the class vector C Q WD .C2 ; C21 ; C10 / is also rigid.
Elements of order 5 in T are rational, so the same holds for elements in C10 .
Further ı is a rational element, and elements of order 7 in the centralizer G2 .p/
of ı are rational. Finally this G2 .p/ is even contained in the centralizer of an
S3 containing ı, which proves the rationality of C21 and the assertion of the
Theorem. t
u
8 Galois Realizations of Exceptional Groups over Q 155

8 Galois Realizations of Exceptional Groups over Q

In this paragraph we derive Galois realizations for certain exceptional groups over
Q.t/. The methods are rather similar to those employed in the previous paragraph.
We first deal with the easier case of G2 .p/. In the second section we give a result on
tori which are trivial intersection subgroups, which will allow to prove generation
also in the cases of the larger groups treated later on.

8.1 GA-Realizations for the Groups G2 .p/

The first series of groups we shall consider are the groups of type G2 defined over
the prime field IFp . It was shown independently by Thompson (unpublished) and
by Feit and Fong (1984) that they occur as geometric Galois groups over Q for
all primes p  5. While Thompson’s proof utilizes generators and relations as in
Belyi’s treatment of the classical groups (see also Malle (1986), 3.3), Feit and
Fong quote the character table and the classification of finite simple groups. We
will follow this second approach, which is more akin to the methods for the other
exceptional groups.
Let G D G2 .p/, p  5, and let C1 be the unique conjugacy class of involutions in
G, C2 the class of long root elements, and C3 the class of regular unipotent elements.
Corresponding elements are denoted by k2 , u2 , resp. u6 in Chang and Ree (1974).

Theorem 8.1. The groups G2 .p/, p  5 prime, possess GA-realizations over Q for
the class vector C D .C1 ; C2 ; C3 /.

Proof. Since G D G2 .p/ has trivial outer automorphism group for primes p  5,
we are reduced to showing that C is rationally rigid for G.
First from the character table of G in Chang and Ree (1974) one computes
that n.C/ D 1 and verifies that all three classes are rational. Note that, contrary
to the case of the class vector for G2 .q/ used in Proposition 4.6, most families of
irreducible characters of G contribute to n.C/. This comes from the fact that the
only semisimple class C1 in C has large centralizer. We omit the list of the relevant
eleven character families. They can be found in Feit and Fong (1984).
Now let H D h i for a system  2 ˙N .C/. Since the two rational p-elements 2
and 3 are not conjugate in G, the order of H is divisible at least by p 2 . By the list
of maximal subgroups of G in Kleidman (1988b) we are left with the cases

Œp 5 W GL2 .p/ (2 classes); .SL2 .p/ ı SL2 .p//  2; SL3 .p/W 2; SU3 .p/W 2:

If H were contained in one of the parabolic subgroups Œp 5 W GL2 .p/, then, since
it is generated by p-elements, it would already lie in Œp 5 W SL2 .p/. But now upon
factorization by the normal p-subgroup H \ Œp 5 the image of 1 can only be the
unique and hence central involution of SL2 .p/. Then by the product relation one of
the other two elements orders would also have to be even, which is a contradiction.
156 II Applications of Rigidity

The centralizer of an involution .SL2 .p/ ı SL2 .p//  2 does not contain regular
unipotent elements, since those have centralizer order p 2 . This rules out the second
case, and also the centralizer of a 3-element, which is of type SL3 .p/ when p 1
.mod 3/, and of type SU3 .p/ otherwise.
Since H is generated by its p-elements, we are left with H L3 .p/ or H
U3 .p/, depending on the congruence of p mod 3. Both the latter groups possess
exactly two classes of p-elements and one involution class. Hence the fusion from
either containing H into G is clear. But the corresponding structure constants vanish
in both groups. This excludes the last possible maximal subgroups, and the proof is
complete. t
u

We already constructed a GA-realization for G2 .2/0 D U3 .3/ over Q in Theo-

rem 6.8(b), so only the case p D 3 remains to be treated.

Theorem 8.2. The group G2 .3/ possesses a GA-realization over Q for the class
vector .3A; 4C; 6E/ of its automorphism group Aut.G2 .3// Š G2 .3/W 2.

Proof. From the character table in the Atlas one finds n.3A; 4C; 6E/ D 1. Now
let H be the group generated by a corresponding triple of elements. Since 6E 2 D
3C , the group H intersects two distinct classes of 3-elements, hence its Sylow 3-
subgroup has order at least nine. By the list of maximal subgroups of Aut.G2 .3//
in the Atlas, it thus lies in one of Œ36 W D8 , L2 .8/W 3  2 or Œ25 :.S3  S3 /. But H is
no subgroup of the first candidate, as can be seen by factoring out the normal Sylow
3-subgroup. Also, the second type of maximal subgroup consists of centralizers of
2B-involutions, but no element from 3A centralizes such an involution. Finally, the
maximal subgroup Œ25 :.S3  S3 / is the centralizer of a 2A-involution, but this now
does not centralize 3C -elements. Hence all three possibilities are ruled out, and we
conclude that H D G. Descent of degree two now yields the result for the simple
group. t
u

8.2 The Groups F4 .p/

Our choice of class vector for F4 .p/ is similar to the one in Section 5.2. The first
class Cp again consists of long root elements. For the second class, when q 1
.qC1/=6
.mod 3/, we take C6 WD Œh16 containing rational elements of order 6, a power
of the class CqC1 in Section 5.2. If q 1 .mod 3/, similar elements of order six are
.q1/=6
given by C6 WD Œh15 .
In Proposition 5.4 we chose CT to contain a generator of the cyclic torus T . It
turns out that it suffices to include an element with order a primitive prime divisor of
jT j into the class structure to obtain rigidity. Now T has normalizer NG .T / D T:12,
so any non-identity rational elements in it has order 13. Such elements exist when
q is a primitive root modulo 13. Now when q D p m is a primitive root, the same
certainly holds for p, so that our element of order 13 will already be contained in
8 Galois Realizations of Exceptional Groups over Q 157

the group over the prime field F4 .p/. We can hence only expect generation when
q D p. Let C13 be the class of an element of order 13 in T .

Theorem 8.3. For primes p 2; 6; 7; 11 .mod 13/, p  19, the groups F4 .p/ have
GA-realizations over Q for the class vector C D .Cp ; C6 ; C13 /.

Proof. First note that the class vector is rational by Shoji (1974). So it suffices
to show l.C/ D 1. In the case p 1 .mod 3/, the arguments in Proposition 5.4
to find the nonvanishing characters also work with C13 and C6 . Moreover, elements
from CT and C13 , resp. from CqC1 and C6 have conjugate centralizers, so the values
of unipotent characters on them are the same by the character formula (5.3). So in
this case we immediately get n.C/ D 1 from Table 5.1. If p 1 .mod 3/, similar
arguments to those in the proof of Proposition 5.4 also allow the verification of
n.C/ D 1. The relevant character values are omitted.
But for generation we cannot use Weigel (1992) as above, since there only the
overgroups of the whole torus T are classified. Instead we apply Malle and Tester-
man (2011), Theorem 29.5. Note that H D h i for  2 ˙.C/ N is perfect by Propo-
sition 4.3. Corollary 4.2 applied to the primitive prime divisor 13 of p 4  p 2 C 1
excludes the possibilities (ii)–(iv) and this forces H to be simple of Lie type in
characteristic p or one of the groups in Liebeck and Seitz (1999), Theorem 2. The
following of these have order divisible by 13:

L2 .13/; L2 .25/; L2 .27/; L3 .3/; L4 .3/; 2B2 .8/; 3D4 .2/:

All of these groups have order only divisible by primes l 13, while the order of
H is also divisible by p  19.
Thus we are left with the groups of Lie type in the same characteristic p. We
only need to consider those of Lie rank at most 4, since otherwise they cannot be
contained in G. Again since 13 is a primitive prime divisor of 12 .p/, we may
apply the divisibility criterion in Corollary 4.2 to the order of candidates H.p m / in
the form
12 .p/ jH.p m /j jF4 .p/j:
Apart from G this leaves the six cases

A1 .p 6 /; A2 .p 4 /; 2A2 .p 2 /; B2 .p 3 /; G2 .p 2 /; 3D4 .p/:

Most of those cannot be contained in G: The centralizer of an element of order 13

has order .p 6 C 1/=2 in A1 .p 6 /, p 8 C p 4 C 1 in A2 .p 4 / and p 6 C 1 in B2 .p 3 /, but
only p 4 p 2 C1 in G. Also, G2 .p 2 / contains a maximal torus of order p 4 Cp 2 C1,
while there are no elements of that order in G. The remaining two groups 2A2 .p 2 /
and 3D4 .p/ are excluded precisely as in the proof of Proposition 5.4, and the proof
is complete. t
u

Using a different class vector Guralnick, Lübeck and Yu (2016) were able to
show a more comprehensive result for F4 .p/. To state it, let C2 denote the class
of involutions with centralizer of type C3 C A1 , C2p the class of elements of order
158 II Applications of Rigidity

2p whose square is a long root element and whose pth power is an involution with
centralizer type B4 , and Cu be the class of regular unipotent elements. All three
classes are rational.

Theorem 8.4. For all primes p > 3 the groups F4 .p/ have GA-realizations over Q
for the class vector C D .C2 ; C2p ; Cu / defined above.

The proof requires a more detailed analysis involving estimates on character val-
ues as well as precise information on subgroups of F4 .p/ containing regular unipo-
tent elements.

8.3 The Groups E6 .p/ and 2E6 .p/

Let again G WD E6 .q/sc W 2 be the extension by the graph automorphism of order two
of the simply connected group of type E6 . Define the class C2p as in Section 5.3.
This class is rational. Next, we have to find rational elements in the torus of order
q 6 C q 3 C 1 D 9 . Since jNG .T /=T j D 18, these will necessarily have order 19,
and exist when q is the square of a primitive root modulo 19. Let C19 denote the
class of such elements.

Theorem 8.5. For primes p 4; 5; 6; 9; 16; 17 .mod 19/, the group E6 .p/sc W 2 has
a G-realization over Q for the class vector C D .C2p ; C2p ; C19 /.

Proof. As in the case of F4 .p/ it first follows from the Deligne–Lusztig theory of
disconnected groups (Digne and Michel (1994)) that the character values on CT
given in Table 5.2 coincide with those on C19 . This yields n.C/ D 1.
So now let H D h i with  2 ˙N .C/, and HN WD H=.H \ Z .G 0 //. For genera-
tion, we may proceed as in the case of F4 .p/. In particular, by Malle and Testerman
(2011), Theorem 29.5, it remains to consider simple groups of Lie type in charac-
teristic p and the groups in Liebeck and Seitz (1999), Theorem 2.
Of the latter, only L2 .19/; J1 and J3 have order divisible by 19. But J1 pos-
sesses a Frobenius subgroup of type 11W 10. By Cohen et al. (1992) such can
only exist in G when p D 31 resp. p D 11. But neither p satisfies the congru-
ences. The only possible case for the occurrence of L2 .19/ is for p D 5. As H
contains outer elements of G, H \ G 0 is a subgroup of HN of index two. If we
had L2 .19/ HN Aut.L2 .19// D L2 .19/W 2, then consequently HN D L2 .19/W 2.
But this is impossible since L2 .19/W 2 does not possess outer elements of order
10 D 2p. The remaining case R D J3 might only happen for p 2 f5; 17g, as we
have jJ3 j D 27  35  5  17  19. The case p D 5 is excluded as above. Finally from the
Atlas one finds that the .34; 34; 19/-structure constants of J3 W 2 are too large to come
from .C34 ; C34 ; C19 / in G.
The groups of Lie type are treated as in the proof of Proposition 8.3. Using Corol-
lary 4.2, only A2 .p 3 / remains apart from E6 .p/. Now the end of the proof of Propo-
sition 5.6 may be used to exclude that last case, and the assertion follows. t
u
8 Galois Realizations of Exceptional Groups over Q 159

Descent to the normal subgroup of index 2, taking factor groups, and using that
Aut.E6 .p// D E6 .p/W 2 for p 6 1 .mod 3/ leads to:

Corollary 8.6. The groups E6 .p/W 2, E6 .p/sc and E6 .p/ possess G-realizations
over Q for primes p 4; 5; 6; 9; 16; 17 .mod 19/. If moreover p 1 .mod 3/,
this yields a GA-realization of E6 .p/ over Q.

A completely similar treatment is possible for the twisted groups. The choice of
classes for the class vector is precisely the same as above, except that now we look
for rational elements of order 19 in the torus of order 18 , which exist if p is a
primitive root mod 19.

Theorem 8.7. The groups 2E6 .p/sc W 2, 2E6 .p/W 2, 2E6 .p/sc and 2E6 .p/ possess G-
realizations over Q for primes p 2; 3; 10; 13; 14; 15 .mod 19/, p > 3, where the
class vector for 2E6 .p/sc W 2 is C D .C2p ; C2p ; C19 /. If moreover p 6 1 .mod 3/
this yields a GA-realization of 2E6 .p/ over Q.

Sketch of proof. The structure constant n.C/ D 1 follows from Table 5.4. Arguing as
in the preceding proof, we only need to consider simple subgroups. We first consider
those not of Lie type not in characteristic p, viz. L2 .19/, and J3 . Now the group
generated by a triple from C clearly has order divisible by p, hence p must be
one of p D 2; 3; 5; 19 for L2 .19/ or p D 2; 3; 5; 17; 19 for J3 . None of these primes is
allowed in the theorem (note that PGL2 .19/ possesses a .6; 6; 19/ generating system,
so p D 3 has to be excluded at this point). Again, the groups of Lie type in the same
characteristic are excluded as usual.
The final assertion follows since Aut.2E6 .p// D 2E6 .p/W 2 for p 6 1 .mod 3/.
t
u

Remark. As with E6 .q/, we obtain a second Galois realization for the groups in
Theorems 8.5, 8.6 and 8.7 from the Fixed Point Theorem I.7.2, belonging to a class
vector .C2 ; C2p ; C19 /.

8.4 The Groups E8 .p/

Recall the maximal torus T of E8 .q/ from Section 5.6. If it contains rational ele-
ments, we can hope to obtain Galois realizations over Q, at least for E8 .p/. Since
NG .T / D T:30, rational elements in T n1 necessarily have order 31. Now 31 divides
30 precisely when p is a primitive root modulo 31. This happens for roughly every
fourth prime. There exists another t.i.-torus in G containing rational elements of
order 31, namely the one with order 15 . Let C31 denote the class of an element of
order 31 in E8 .p/ when 31 15 or 31 30 . To ease the notation, write

P1 WD fp 2 IP j p 3; 11; 12; 13; 17; 21; 22 or 24 .mod 31/g;

P2 WD fp 2 IP j p 7; 9; 10; 14; 18; 19; 20 or 28 .mod 31/g;

160 II Applications of Rigidity

for the set Pi of primes p of multiplicative order 30= i modulo 31. Then with the
classes C2 and Cp introduced in Section 5.6 we have:

Theorem 8.8. The groups E8 .p/ have GA-realizations over Q for primes p 2 P1 [
P2 , p  7 for the class vector .C2 ; Cp ; C31 /.

Note that elements from the third class either lie in the torus of order 30 used
above, or in one of order 15 with rather similar properties. Hence the proof can be
completed as the one for F4 .p/ in Theorem 8.3, using the list in Liebeck and Seitz
(1999) of possible simple subgroups of E8 .q/. The details can be found in Malle
(1988b).
As in the case of F4 .p/ a more comprehensive result is available whose proof
relies on more precise discussion of character values and on a knowledge of over-
groups of regular unipotent elements. Let C2 be the class of involutions in E8 .p/
with centralizer of type D8 , Cp the class of unipotent elements of type 4A1 , and Cu
the class of regular unipotent elements. The following is shown in Guralnick and
Malle (2014):

Theorem 8.9. For all primes p  7 the groups E8 .p/ have GA-realizations over Q
for the class vector C D .C2 ; Cp ; Cu / defined above.
9 The Sporadic Groups 161

9 The Sporadic Groups

The character theoretic form of the rationality criterion is most suited for application
to the sporadic groups. The character tables of all 26 sporadic groups are known (see
Conway et al. (1985) and also Breuer, Malle and O’Brien (2016)), and moreover
the lists of maximal subgroups are almost complete by the work of a number of
authors.
Due to the irregular nature of the sporadic groups, no unified proof of rigidity
for them is known. Still, they can be treated in a case by case analysis, and it turns
out that with at most the exception of the Mathieu group M23 , they all occur as
geometric Galois groups over Q.t/. This result is due to Thompson (1984a) for the
monster M, to Matzat (1985a) for M12 and M22 , to Hoyden-Siedersleben (1985)
for M23 , M24 and J1 , to Matzat and Zeh-Marschke (1986) for M11 , to Hunt (1986)
for J2 , Suz, HS, Co3 , Co2 , Co1 , Fi23 , Fi024 and Th, to Hoyden-Siedersleben and
Matzat (1986) for HN, ON, and to Pahlings (1988, 1989) for McL, He, Fi22 , J3 ,
Ly, Ru and J4 . No correct Galois realization for the baby monster B is contained in
the literature (note that the structure constant given for B in Hunt (1986) is equal
to 3, not 1 as stated).
The proofs appearing in the literature sometimes make extensive use of computer
calculations, of character tables for maximal subgroups and their fusion maps. We
have tried to present proofs which may be checked just by using the Atlas. There-
fore, in addition to the Galois realization for B, new realizations are given for the
sporadic groups Suz, Co3 , Co1 , He, Fi22 , Fi23 and Ru in order to get simplified
proofs. Nevertheless for the groups B, J3 and J4 no easy arguments could be found,
and our proofs are sketchy and still rely on a certain amount of computer calcula-
tions, most notably in the cases of J4 and B. Despite considerable efforts the group
M23 has at present only been realized as Galois group over several quadratic exten-
sion fields of Q.
The sporadic groups may be subdivided into several families of more or less
similar groups, as proposed in the Atlas. We follow this subdivision in our proof.

9.1 The Mathieu Groups

The names for the conjugacy classes in the class vectors are taken from the Atlas,
as well as all results on character tables, on classes, powermaps and maximal sub-
groups, if no other source is indicated.
In the proofs we will make frequent use of the following observations. Let
G D h1 ; 2 i be generated by a 3-system  D .1 ; 2 ; 21 11 / with element orders
.n1 ; n2 ; n3 /. If H G G is a normal subgroup, then clearly G=H is generated by the
3-system N of images, with orders .m1 ; m2 ; m3 /, where mi jni , i D 1; 2; 3. If G=H
has a transparent structure, this may be used to deduce restrictions on possible tuples
.n1 ; n2 ; n3 / for G. One strong application was already given by Proposition 4.3.
162 II Applications of Rigidity

Proposition 9.1. (a) The class vector .2C; 3A; 12A/ of Aut.M12 / is rationally rigid.
(b) The class vector .2B; 4C; 11A/ of Aut.M22 / is rationally rigid.
(c) The class vector .2A; 3B; 23A/ of M24 is rigid.

Proof. We use the character theoretic rigidity criterion in Corollary I.5.9 and the
tables of complex irreducible characters of the sporadic simple groups G and their
automorphism groups Aut.G/ in the Atlas. From these tables the structure constants
n.C/ for the class vectors above are easily calculated to be equal to 1 in all cases.
To be able to apply the Basic Rigidity Theorem I.4.8, we have to prove l.C/ D 1. In
the case n.C/ D 1 this is equivalent to showing that a triple from C cannot generate
a proper subgroup of Aut.G/. Let H denote the subgroup of Aut.G/ generated by
a triple of elements from the class vector C.
In case (a), the squares of elements from class 12A lie in class 6A, so clearly H
contains elements from classes 3A and 6A. In the Atlas, the permutation characters
for ten of the eleven classes of maximal subgroups of M12 are given. From these one
checks that none contains elements from both 3A and 6A. So the only remaining
possibility is H M D S4  S3 . If M has a .2; 3; 12/-system, then the last class has
to be the product of the 4-cycles in S4 times the 3-cycles in S3 . It then follows that

C \ M D ..2/  .1/; .3/  .3/; .4/  .3//

in obvious notation. Now M has three classes of elements of order three, .1/  .3/,
.3/  .1/ and .3/  .3/, with centralizers in M of orders 72, 18 and 9 respectively.
If nM .C/ ¤ 0, then clearly the first of the three classes fuses into 3B D .6A/2 ,
while the third fuses into 3A. But now, depending on the fusion of .3/  .1/, the
permutation character  D 1G M takes either the values .3A/ D 12, .3B/ D 5, or
.3A/ D 18, .3B/ D 1. This contradicts the fact that by congruence properties of
character values we should have .3A/ .3B/ .mod 3/. So H 6 M , completing
the proof for part (a).
In case (b), from the Atlas we see that PGL2 .11/ D L2 .11/W 2 is the only type of
maximal subgroup of Aut.G/ different from G with order divisible by 11. Assume
that H PGL2 .11/. The intersection of this maximal subgroup with G is equal to
L2 .11/, so elements from the single outer class of involutions of PGL2 .11/ would
have to fuse into 2B. The centralizer order for this class in PGL2 .11/ is divisible
by 5, in contrast to the centralizer order for 2B in Aut.M22 /. Thus we have H 6
PGL2 .11/ (this can also be seen from the permutation character given in the Atlas),
and the result follows.
In the third case, the maximal subgroups of G of order divisible by 23 are M23
and L2 .23/. Both of these have just one class of elements of orders 2 and 3. More-
over both contain elements of order 6. Since .6A/2 D 3A, .6A/3 D 2A, while
.6B/2 D 3B, .6B/3 D 2B, the classes 2a, 3a, from the two maximal subgroups
either fuse into 2A, 3A, or into 2B, 3B; in particular none of the two types inter-
sects both 2A and 3B. This proves that H D M24 . t
u
9 The Sporadic Groups 163

9.2 The Leech Lattice Groups

The sporadic groups J2 , Suz, HS, McL, Co3 , Co2 and Co1 are most conveniently
described as stabilizers of vectors in the Leech lattice and hence subsumed under
the name of Leech lattice groups.
Proposition 9.2. (a) The class vector .3A; 8C; 14A/ of Aut.J2 / is rationally rigid.
(b) The class vector .2C; 8D; 13A/ of Aut.Suz/ is rationally rigid.
(c) The class vector .2C; 5C; 30A/ of Aut.HS/ is rationally rigid.
(d) The class vector .3A; 4B; 10B/ of Aut.McL/ is rationally rigid.
Proof. This can again be proved with Corollary I.5.9. From the complex character
tables one verifies n.C/ D 1 in cases (a)–(c) and n.C/ D 3 for Aut.McL/. It remains
to check generation in the first three cases, while in the last we have to account for
a summand of 2 in the structure constant coming from proper subgroups.
The list in the Atlas reveals that only the maximal subgroups U3 .3/W 2 Š G2 .2/
and PGL2 .7/  2 of Aut.J2 / D J2 W 2 different from J2 have order divisible by 7.
But G2 .2/ does not contain elements of order 14. Assume H PGL2 .7/  2.
The factor group H=.H \ L2 .7// is a subgroup of Z2  Z2 and has a .1; 2; 2/-
generation (obtained from C), hence is isomorphic to Z2 . So we would already have
H PGL2 .7/ or H L2 .7/  2, but the first contains no elements of order 14, the
second none of order 8. This excludes the last possible maximal subgroup, proving
generation in case (a).
The maximal subgroups of Aut.Suz/ different from Suz of order divisible by 13
are G2 .4/W 2 and L2 .25/W 2. The latter is actually equal to L2 .25/W 22 , since L2 .25/:23
is non-split, and the outer involutions in L2 .25/W 21 centralize an element of order 13.
But L2 .25/W 22 contains no element of order 8. The permutation character  of Suz
on G2 .4/ is well known; in fact, Suz may be constructed as a rank three permuta-
tion group with point stabilizer G2 .4/. The permutation character of Aut.Suz/ on
G2 .4/W 2 may easily be deduced from that. It restricts to Suz as  D 1 C 780 C
1001 , and since it has to take non-negative values, it is exactly given by the sum
of the first, the fourth and the fifth irreducible character of Aut.G/ in the Atlas. But
this permutation character vanishes on class 8D, so we have H D Aut.Suz/ and the
result follows for part (b).
In case (c), as H contains representatives from the two rational conjugacy classes
5B D .30A/6 and 5C , its order is divisible by at least 25. The only maximal sub-
groups of Aut.HS/ apart from G WD HS with order divisible by 25 are U3 .5/W 2,
51C2
C W Œ2
5
and M D 5W 4  S5 . But the first subgroup contains no outer elements,
while the order of the second is prime to three. The third one, a direct product of the
Frobenius group of order 20 with the symmetric group S5 , obviously has three con-
jugacy classes of elements of order five, C1 D 1  5, C2 D 5  1, C3 D 5  5, where
we have indicated the projections of these classes onto the two factors of the direct
product M . The corresponding centralizer orders in M are 100, 600 and 25, respec-
tively. The first class fuses into 5A of G by the description M D N .2B; 3A; 5A/ in
the Atlas. The class C2 has to fuse into 5B since that is the only class with central-
Aut.G/
izer order divisible by three. Now the values of the permutation character M on
164 II Applications of Rigidity

1A and on 5B have to be congruent modulo 5, hence divisible by 5, as immediately

follows from consideration of the cycle type of the element of order 5 in this permu-
Aut.G/
tation representation. The inclusion C2  5B contributes 1 to M .5B/, so the
class C3 also has to fuse into 5B and M \ 5C is empty. Thus H 6 M , proving the
result for assertion (c).
Excluding the maximal subgroup of Aut.McL/ of order not divisible by 5, we
are left with U4 .3/W 23 , U3 .5/W 2, 31C4 W 4S5 , 34 W .M10  2/, L3 .4/W 22 , 2 S8 , M11  2
and .51C2
C W 3W 8/:2. Factoring the last group by its normal subgroup of order 3  5
3

it becomes evident that it cannot contain .3; 4; 10/-triples. The subgroup 3 1C4
W 4S5
is identified as N .3A/, so the powermap shows that its 5-elements lie in 5A of
G WD McL. But as .10B/2 D 5B it does not intersect class 10B. Similarly we have
2 S8 D N .2A/, hence its 5-elements again lie in 5A. The group M11  2 centralizes
the outer involution 2B, so its unique class of 3-elements fuses into 3B of G. Also,
U3 .5/W 2 has outer elements of order six, which hence fuse into 6C of G. But the
squares of such elements lie in 3B, and it follows that 3A \ U3 .5/W 2 D ;. Precisely
the same argument applies to L3 .4/W 22 . In any complement of 34 in the semidirect
product 34 W .M10  2/ the central involution is outer, hence lies in class 2B. So again
all 3-elements of the complement fuse into 3B. As this is true for any complement,
3A-elements could only lie in the normal subgroup of 34 W .M10  2/. Factoring by
this 34 we see that C cannot contain triples from that maximal subgroup.
We are left with M D U4 .3/W 23 . Since this group has uniquely determined outer
classes 4j of 4-elements and 10c of elements of order 10, these fuse into 4B, 10B
of G, respectively. Further M contains a full Sylow 3-subgroup of G. The class
3A of G consists of third powers of elements of order nine and all third powers of
elements of order 9 in M fall into 3a, so the intersection of M with 3A is just 3a.
The structure constant for the intersections of C with M now equals

nM .3a; 4j; 10c/ D 2:

From the list of maximal subgroups of M it is easily seen that any such triple has
to generate all of M , so this maximal subgroup contributes 2 to n.C/ by Proposi-
tion I.5.7. This yields l.C/ D n.C/  2 D 1. t
u

Proposition 9.3. (a) The class vector .2A; 5B; 30A/ of Co3 is rationally rigid.
(b) The class vector .2A; 5A; 28A/ of Co2 is rationally rigid.
(c) The class vector .3A; 5C; 13A/ of Co1 is rationally rigid.

Proof. The structure constant for Co3 is found to be n.C/ D 7, while in the other
two cases one computes n.C/ D 1.
In G D Co3 , the sixth powers of elements from 30A lie in class 5A of G, so
H contains elements from the two rational classes 5A, 5B, and its order must be
divisible by 25. This rules out all maximal subgroups apart from McLW 2, HS and
U3 .5/W S3 . The group HS has no elements of order 30. Clearly H has no factor
group S3 , and H U3 .5/W 2 can be ruled out since U3 .5/ does not contain elements
of order 15. Finally consider M D McLW 2. The class 5a of M consists of sixth
powers of elements from 30a (the unique class of 30-elements in M ), so fuses into
9 The Sporadic Groups 165

5A of G. As the full 5-part of jGj divides jM j, the other class 5b of 5-elements in M

must fuse into 5B. Finally 2a D .30a/15 in M , so 2a fuses into 2A D .30A/15 , and
since 2a; 5b already lie in the simple group M 0 D McL, we only have to consider
subgroups of McL. The structure constant in M equals

nM .2a; 5b; 30a/ D 6:

The only maximal subgroup of McL containing elements of order 30 and with order
divisible by 25 is a 51C2
C W 3W 8, but clearly the above triple cannot lie in that group. So
any of the six triples in M generates M 0 , which yields a contribution of 6 to n.C/
by Proposition I.5.7. Since this was the only maximal subgroup left to consider, we
obtain l.C/ D n.C/  6 D 1:
In part (b), for divisibility reasons only the maximal subgroups 210 W M22 W 2, McL,
1C8
2C W 2 S6 .2/, HSW 2, Œ211 W A8 , U4 .3/:D8 , M23 or U6 .2/W 2 might contain H . Of
these, McL, HSW 2, M23 and U6 .2/W 2 do not contain elements of order 28. The
maximal subgroup 21C8 C W 2 S6 .2/ is the centralizer in G of 2A-involutions, and as
such does not intersect the class 5A as can be seen from the powermap of ele-
ments of order 10 in G. Similarly, Œ211 W A8 centralizes a 2B-involution, but the
self-centralizing subgroup generated by an element from 28A only contains 2A
involutions, hence does not lie in CG .2B/. If H were contained in U4 .3/:D8 , then
it would already lie in an extension of U4 .3/ of degree two, as can be seen from C N
in H=.H \ U4 .3//. But as U4 .3/ has no elements of order 14, no element in such a
degree 2 extension has order 28. Finally assume H 210 W M22 W 2. We claim that M22
contains no 5A-elements. Indeed, the Sylow 11-subgroup of G is self-centralizing,
with normalizer 11W 10. The maximal subgroup U6 .2/W 2 contains a subgroup 11W 5.
Its elements of order 30 are rational, so they fuse into 30A of G. The sixth power of
30A is 5B, showing that the 11-normalizer 11W 10 contains 5B-type elements. Now
M22 also contains 11W 5, hence its single class of elements of order 5 fuses into 5B,
which completes the proof in case (b).
In the third case the maximal subgroups with order divisible by 13 are 3 SuzW 2
and .A4  G2 .4//W 2. The centralizer order of 5C -elements in G is not divisible by 9,
but in A4  G2 .4/, all 5-elements have centralizer order 12  300. The group 3 SuzW 2
arises as the normalizer N .3A/, but from the powermap of elements of order 15 it
follows that 5C does not intersect the centralizer C .3A/. Hence we have H DG. u t

9.3 The Fischer Groups

Proposition 9.4. (a) The class vector .2D; 5A; 42A/ of Aut.Fi22 / is rationally rigid.
(b) The class vector .2A; 7A; 27A/ of Fi23 is rationally rigid.
(c) The class vector .2C; 8D; 29A/ of Aut.Fi024 / is rationally rigid.

Proof. For all three groups, the normalized structure constant n.C/ equals 1.
In case G D Fi22 , among the maximal subgroups of Aut.G/ different from G,
only 2 U6 .2/:2, G2 .3/W 2, OC
8 .2/W S3  2, 2 W M22 , 2 W S6 .2/ and S3  U4 .3/:.2 /122
10 7 2
166 II Applications of Rigidity

have order divisible by 7 (Wilson (2017)). As 2 U6 .2/, G2 .3/, M22 and S6 .2/ do
not contain elements of order 21, we are left with the cases M1 D OC 8 .2/W S3  2 and
M2 D S3  U4 .3/:.22 /122 . These may be excluded at the same time. For this denote
by Ni the simple normal subgroup OC 8 .2/, resp. U4 .3/ of Mi . If H is contained in
Mi , then H=.H \ Ni / has order at most two, since the second class of C clearly lies
inside Ni . So H Ni :2, which contradicts the fact that in neither case the group Ni
contains elements of order 21. This completes the proof of (a).
By Wilson (2017), only the maximal subgroups of Fi23 isomorphic to OC 8 .3/W S3 ,
Œ2 3 :2S4 or Œ310 :.2  L3 .3// might contain elements of order 27. But the order
7 12

of the latter two is not divisible by 7. If H were contained in OC 8 .3/W S3 , then, as

the orders of the three elements in C are coprime, it would already lie in OC 8 .3/
by Proposition 4.3. But that group has no elements of order 27. So it follows that
H D G for case (b).
By Wilson (2017), the only maximal subgroup of Fi024 of order divisible by 29
is the local subgroup 29W 14, which extends to 29W 28 in Fi24 . Since 8 does not divide
its order, we obviously have H D Aut.G/. t
u

9.4 The Monster Centralizers

The four sporadic groups He, HN, Th and B occur as centralizers of elements in the
monster group M. We treat B and M separately, since at least for M the information
on maximal subgroups is still incomplete, although strong restrictions on possible
subgroups follow from the classification, and so different methods of proof must be
employed.

Proposition 9.5. (a) The class vector .2C; 3A; 30A/ of Aut.He/ is rationally rigid.
(b) The class vector .2C; 5A; 42A/ of Aut.HN/ is rationally rigid.
(c) The class vector .2A; 3A; 19A/ of Th is rationally rigid.

Proof. For Aut.He/ we have n.C/ D 7=6, while in the two other cases the structure
constant is calculated as n.C/ D 1.
For G D He, among the maximal subgroups of Aut.G/, only those of type
S4 .4/W 4, S5 o 2, 22 L3 .4/:D12 , 3 S7  2 and 52 W 4S4 have orders divisible by 5. Of
these, S4 .4/W 4, 22 L3 .4/:D12 and 52 W 4S4 do not contain elements of order 30. For
the second of these this follows from its characterization as N .2A2 /: the square of
the element of order 30 has odd centralizer order in G, so acts non-trivially on 2A2 .
Thus the cube of the element of order 30 has to centralize the whole four group. But
.30A/3 D 10B, and the centralizer of such an element is of order 60 and contains at
least one element from .10B/2 D 2C , so cannot contain a 2A2 , as claimed. Assume
that 52 W 4S4 has elements of order 30. Then an element  of order three in the com-
plement 4S4 has to centralize some element of order 5 in the elementary abelian
normal subgroup 52 , hence also the cyclic subgroup generated by this element. This
leaves 20 elements in 52 , of which obviously  has to centralize at least another one.
But then we would have 25 jCG ./j, which is not the case.
9 The Sporadic Groups 167

Next assume H 3 S7  2. Then by factoring out H \ 3 A7 we see that H


already lies inside 3 A7  2. The projection of C onto the second factor now has
the type .2; 1; 2/, hence it projects as .2; 3; 15/ onto 3 A7 . The relevant structure
constants in that group are easily computed and by Proposition I.5.7 contribute at
most 1=3 C 1=6 to n.C/, depending on the exact fusion.
We are left with M D S5 o 2. Again, if H M , then H already lies in a degree 2
extension of A5  A5 . Elements of order 30 in the wreath product S5 o 2 are eas-
ily seen to generate their proper centralizers, and then it follows that a degree two
extension containing these must be S5  A5 . The projection of C onto the factors
must then equal .2; 3; 6/  .2; 3; 5/, and again we find a structure constant of 1=6. In
conclusion, we have 7=6  l.C/  n.C/  4=6 D 3=6, and this proves l.C/ D 1 in
case (a).
The maximal subgroups of Aut.HN/ with order divisible by 5 and 7 are of types
S12 , 4  HS:2 and 5W 4  U3 .5/W 2. But neither HS nor U3 .5/ does contain elements
of order 21, hence the same holds for the second and third maximal subgroup listed
above. So we are left with the symmetric group S12 . From the character table of S12
it may be calculated that the only non-vanishing .2; 5; 42/-structure constants are

n.2e; 5b; 42a/ D 3 and n.2f; 5b; 42a/ D 51=2:

But S12 is self normalizing in Aut.HN/, so if indeed one of the two above class
vectors would fuse into C in Aut.HN/, then the structure constant could be at most
as large as in Aut.HN/ by Proposition I.5.7. This contradiction rules out H S12 ,
proving the result in case (b).
For G D Th, only the maximal subgroups of type U3 .8/W 6 and L2 .19/W 2 contain
elements of order 19 (Wilson (2017)). But in L2 .19/, the elements of order three
are third powers, which is not the case for elements from 3A in G. If H U3 .8/W 6,
then we already have H U3 .8/, as H is perfect by Proposition 4.3. The class 3c of
U3 .8/ has third roots, so it cannot fuse into 3A of G. Thus C would have to be equal
to one of .2a; 3ab; 19abcdef /. But the corresponding structure constants vanish in
U3 .8/, so we have H D G, proving part (c). t
u

Proposition 9.6. (a) The class vector .2C; 3A; 55A/ of B is rationally rigid.
(b) The class vector .2A; 3B; 29A/ of M is rationally rigid.

Proof. For the baby monster B, we proceed as follows: The .2C; 3A; 55A/-structure
constant of B equals 1. Using the matrix representation of dimension 4370 of B over
IF2 constructed by himself, R. Wilson verified the existence of a generating triple of
elements in the class vector C. To be more precise, he found three matrices, lying
in the respective classes of C and with product one, such that the orders of some
randomly produced elements contained all prime divisors of the group order. Then
an easy application of the classification shows that the three elements already have
to generate B (see the list of maximal subgroups in Wilson (2017)).
The maximal subgroups of M are not yet known completely, but as in the previ-
ous case, we can make use of the classification. The three element orders in C are
pairwise coprime, so H is perfect by Proposition 4.3. Let L denote a top non-abelian
168 II Applications of Rigidity

simple composition factor of H . Then the condition 29 jLj together with the list of
possible simple sections of M in the Atlas shows that L 2 fL2 .29/; L2 .59/; Fi024 g
or L D H D M. Now the class 2A of M consists of f3; 4; 5; 6g-transpositions, i.e.,
of transpositions such that the product of any two has order at most 6. The groups
L2 .29/ and L2 .59/ contain dihedral subgroups of orders 14, 58 respectively, so their
unique class of involutions cannot fuse into 2A. Hence if H ¤ M, then its top com-
position factor is isomorphic to the Fischer group Fi024 .
Assume that this is the case. Let S be a minimal normal subgroup of H . Then S
is a direct product of isomorphic simple groups. The element of order 29 in H acts
on S . If S D S1      Sr were non-abelian, then either 29 j Aut.S1 /j, hence S D L
by the classification above, or S consists of at least 29 factors, since the element
of order 29 has centralizer order 3  29. This is impossible since only the prime 2
occurs with multiplicity more than 20 in jMj. If S D p r is elementary abelian, then
again the element of order 29 acts fixed point freely unless p D 3 or p D 29. Now
2 and 3 are primitive roots mod 29, while 5 has order 14 and 7 has order 7, so since
the top composition factor L of H already has a 2-part 221 , the possibility of fixed
point free action can be ruled out. Also, p D 29 is impossible since 29 divides jMj
just once and is already contained in the top composition factor L D Fi024 . We are
left with the possibility of a central 3-extension of Fi024 , non-split since H is perfect.
This can only be the universal covering group 3 Fi024 .

Table 9.1 Character values in M

1A 11A 17A 23AB 29A

196883 16 6 3 2

Table 9.2 Character values in 3 Fi024

1A 11A 17A 23AB 29A

1 1 1 1 1 1
8671 8671 3 1 0 0
57477 57477 2 0 0 1
783C N 783 1566 4 2 2 0
N
64584 C 64584 129168 6 2 0 2

To deal with the remaining cases Fi024 and 3 Fi024 , we determine the possible
restrictions of the smallest non-trivial character  of degree 196883 of M to the uni-
versal covering 3 Fi024 . The values of  on some classes of M are given in Table 9.1.
The rational characters of 3 Fi024 of degree at most 196883 with their values on the
9 The Sporadic Groups 169

corresponding classes are as given in Table 9.2 where the last two are faithful while
the first three factor through the simple group. Note that we only have to consider
rational characters as constituents of the restriction, since  is rational. The linear
system of equations for the coefficients in the restriction has the unique positive
solution

j3 Fi024 D 1C 8671 C 57477 C . 783 C

N 783 / C . 64584 C
N 64584 /:

Now .3B/ D 53, thus comparison with the character table of 3 Fi024 shows that only
3-classes above 3b, 3c and 3d of Fi024 fuse into 3B of M. But the structure constant
n.2a; 3bcd; 29a/ for these 3-classes turns out to vanish in 3 Fi024 . This eliminates
the last possible proper subgroup for case (b) and proves l.C/ D 1. t
u

9.5 The Oddments

These last six sporadic groups do not fit into any convenient family and were there-
fore baptized the oddments in the Atlas.

Proposition 9.7. (a) The class vector .2A; 5A; 5B/ of J1 is rigid.
(b) The class vector .2B; 4A; 22A/ of Aut.ON/ is rationally rigid.
(c) The class vector .2A; 5A; 14A/ of Ly is rationally rigid.
(d) The class vector .2A; 5A; 13A/ of Ru is rationally rigid.

Proof. For J1 we have n.C/ D 5=2. The maximal subgroups of J1 with order divis-
ible by 5 are 11W 10, D6  D10 , 2  A5 and L2 .11/. Factoring the first one by its
normal Sylow 11-subgroup one sees that it cannot contain .2; 5; 5/-systems. The
same is true for the second, since all elements of order five are contained in the
normal Sylow 5-subgroup. In the third case, as elements of order five only occur in
the second factor, we would already have H A5 . It is easily seen that no proper
subgroup of A5 has a .2; 5; 5/-system, so in this case H D A5 . The corresponding
structure constant is equal to 1 in A5 . The last type of maximal subgroups contains
two classes of maximal subgroups A5 . Each of them contributes 1 to the struc-
ture constant nL2 .11/ .2a; 5a; 5b/ D 2 in L2 .11/, so the contribution from proper
subgroups is precisely the one coming from subgroups isomorphic to A5 . But the
number of classes of such subgroups may be computed from the .2; 3; 5/-structure
constant of G, since any .2; 3; 5/-triple generates an alternating group A5 . We have
nJ1 .2A; 3A; 5A/ D 3=2, and if this is subtracted from the normalized structure con-
stant n.C/ D 5=2, we get l.C/ D 1 in part (a).
In case (b) the structure constant n.C/ equals 1. The only maximal subgroups of
Aut.ON/ different from ON with order divisible by 11 are of type J1  2. But J1 ,
hence J1  2, does not contain elements of order four. So H D Aut.ON/, proving
part (b).
For G D Ly we compute n.C/ D 3=2. Only the maximal subgroups G2 .5/,
3 McLW 2, and 2 A11 of G have orders divisible by 7. But G2 .5/ contains no elements
170 II Applications of Rigidity

of order 14. The maximal subgroup 2 A11 is the centralizer of a 2A-element, so it

intersects both classes 5A, 5B of G non-trivially, as can be seen from the powermap.
From the centralizer orders in A11 it is clear that 5a from 2 A11 fuses into 5A of
G, while 5b fuses into 5B. The group 2 A11 contains just one class of elements
of order 14, namely 14a, and two of order 2, one of them consisting of the cen-
tral element, and the other containing the preimages of 2b of A11 in 2 A11 . Clearly
the .2; 5; 14/-structure constant involving the central involution of 2 A11 vanishes,
while the second is found to be equal to

n2 A11 .2b; 5a; 14a/ D 1=6:

We finally consider the possibility H 3 McLW 2 DW M . This group is the central-

izer of a 3A-element, so also contains elements from both 5A and 5B. Again the
two classes 5a, 5b of M are seen to fuse into 5A, 5B respectively. There exists
just one class 14a of elements of order 14 in M . Since 5a and 14a lie in 3 McL,
only the inner involution class 2a of M can contribute to nM .2; 5a; 14a/. One
finds nM .2a; 5a; 14a/ D 1=2. By Proposition I.5.7 it is clear that not all of n.C/
is accounted for in proper subgroups, so l.C/ D 1 follows as claimed.
The .2b; 5a; 14a/-system of 2 A11 actually generates a 2 A8 , with centralizer of
order six, which explains the 1=6. This group also occurs as a maximal subgroup in
McL, so the contribution from 2 A11 is also contained in the structure constant 1=2
for the 3 McLW 2, but this is not relevant for the proof.
For G D Ru the structure constant for C equals n.C/ D 28, so a contribution
of 27 has to be shown to come from proper subgroups. Only the maximal subgroups
2
F4 .2/, .22  2B2 .8//W 3 and L2 .25/:22 have order divisible by 5 and 13. Clearly
the group .22  2B2 .8//W 3 contains elements of order 15, so its 5-elements fuse into
class 5B of G, which rules out this maximal subgroup. The two involution classes
2a, 2b of 2F4 .2/ have centralizer order divisible by 211 , 3 respectively, so both fuse
into 2A of G. The elements of order 5 in 2F4 .2/ are fourth powers of elements
of order 20, so fuse into 5A of G. (The fusion could also be deduced from the
permutation character given in the Atlas.) Hence the contribution from 2F4 .2/ to
n.C/ equals
n.2a; 5a; 13a/ C n.2b; 5a; 13a/ D 1 C 26 D 27:
Finally assume H L2 .25/:22 . Then, since the second and third class in C con-
tain elements of odd order, clearly H L2 .25/. The relevant structure constant in
L2 .25/:22 is calculated as

n.2a; 5ab; 13abcdef / D 3;

where we have written 5ab for 5a [ 5b and similarly for 13abcdef . As 27 C 3 >
n.C/ this means by Proposition I.5.7 that either L2 .25/:22 does not intersect one of
the three classes, so contributes 0, or that its normal subgroup L2 .25/ already lies in
2
F4 .2/, and the contribution was already accounted for in the latter group. In either
case we get l.C/ D 28  27 D 1, and the result in case (d) follows. (Actually, the
9 The Sporadic Groups 171

second possibility holds, since 2F4 .2/ has a maximal subgroup L2 .25/ 2, but this is
not important for the proof.) t
u

For the two Janko groups J3 and J4 we have to appeal to computer results to
verify rigidity:

Proposition 9.8. (a) The class vector .2B; 3B; 8B/ of Aut.J3 / is rationally rigid.
(b) The class vector .2A; 4C; 11A/ of J4 is rationally rigid.

Proof. For the automorphism group J3 W 2 of the group J3 one verifies that n.C/ D 1
from the Atlas. The group H generated by a triple of elements from C cannot lie
inside the maximal subgroup L2 .16/W 4, since all involutions of the latter lie in the
normal subgroup of index two. Also, the possibility 21C4  W S5 can be excluded, since
by the description in the Atlas, elements of order 3 in that group are of 3A-type.
Next, the order of 19W 18 is not divisible by 8. Assume that H lies in .3  M10 /W 2.
Then the factor group by the almost central 3 also has a .2; 3; 8/ generating system.
Now two of the A6 -cosets of M10 W 2 lie in the outer half of Aut.J3 /, and just one
of them contains involutions, while only the other one contains elements of order 8.
This contradicts the possibility of a non-zero .2; 3; 8/ structure constant for M10 W 2.
To deal with the remaining cases, we have to use computer results. The permutation
characters for the maximal subgroups 24 W .3  A5 /:2, L2 .17/  2 and 35 W 8:2 were
calculated by Pahlings (1989). It turns out that none of these groups intersect class
8B. Finally, the character table of 22C4 W .S3  S3 / was also determined in loc. cit.,
and it proves that the .2B; 3B; 8B/-structure constant of that maximal subgroup
vanishes. This rules out the last proper subgroup, thus proving rigidity.
The structure constant for C D .2A; 4C; 11A/ in J4 equals 3=2. According to Wil-
son (2017), among the maximal subgroups of J4 , only 211 W M24 , 21C12 C  3  M22 W 2,
1C2
11C W .52S4 /, U3 .11/, M22 W 2, L2 .32/W 5 and L2 .23/W 2 have order divisible by 11.
Of these, 111C2
C W .5  2S4 / obviously can not possess a .2; 4; 11/-system. The group
L2 .32/W 5 contains no elements of order 4, in L2 .23/W 2 all elements of order 4 are
squares, while this is not true for elements of 4C in J4 . In U3 .11/W 2, the outer ele-
ments of order 4 have centralizer order divisible by 11, hence cannot fuse into 4C ,
while the inner ones are again squares. The elements of order 11 in M22 W 2 are of
type 11B by loc. cit., Cor. 6.2.2. To exclude 211 W M24 , we note that the possible
irreducible constituents (1 ; : : : ; 21 in Atlas-notation) of the permutation character
on this maximal subgroup all take nonnegative values on 11B, hence this subgroup
has trivial intersection with 11A. (This also follows from the aforementioned result
in loc. cit.) Finally, to exclude the possibility 21C12
C  3  M22 W 2 we have to employ
the unpublished character table of that group, which was calculated by B. Fischer
(see Pahlings (1988)), together with its fusion into J4 . It turns out that this subgroup
contributes exactly 1/2 to n.C/, leaving l.C/ D 1. t
u
172 II Applications of Rigidity

9.6 Galois Realizations for the Sporadic Groups

Collecting the rigidity results of the previous sections we obtain GA-realizations for
most sporadic simple groups:
Theorem 9.9. All sporadic simple groups G with the possible exception of the
Mathieu groups M23 and M24 possess G-realizations over Q. A class vector C of
Aut.G/ yielding such a realization is given for each case apart from M11 in Propo-
sitions 9.1–9.8. p
Both M23 and M24 occur as geometric Galois groups over Q. 23/.t/.
Proof. The Mathieu group M11 was treated in Theorem I.6.12 by descent from M12 .
Now let G be a sporadic group different from M11 , M23 , M24 and J1 . Then we have
proved rational rigidity for the class vector C of Aut.G/ in Propositions 9.1–9.8.
Since these groups have trivial center, they occur as regular Galois groups over Q.t/
by the Basic Rigidity Theorem I.4.8.
If G D Aut.G/ the assertion of the theorem follows immediately. In all other
cases we have .Aut.G/ W G/ D 2. The class vectors all contain two outer classes, so
the fixed field L of G in the Aut.G/-extension N=Q.t/ is ramified at precisely two
places. Thus L is a rational function field L D Q.u/, and N=Q.u/ yields a geometric
Galois extension with group G. p
If G D J1 , then C D .2A; 5A; 5B/ is not rational, since QC D Q. 5/. But C is
V -symmetric for V D h.23/i, and this leads to QVC D Q in this case as well. By the
Strong Rigidity Theorem I.4.11 this guarantees the existence of a geometric Galois
extension of Q.t/ with group J1 . p
For G D M24 , the class vector C satisfies QC D Q. 23/, and Theorem I.4.8
implies the assertion for that group. The Mathieu group M23 may now be obtained
by descent from the Galois realization for M24 , much the same way as for M11 in
the proof of Theorem I.6.12.
p Namely, let K denote the fixed fieldpof M23 in the
Galois extension
p N=Q. 23; t/ with group M24 of degree ŒK W Q. 23; t/ D 24
over Q. 23; t/. The permutation types of elements in the three classes in the per-
mutation representation of degree 24 are .2/8 .1/8 , .3/8 , .23/.1/ respectively. This
describespthe ramification behavior of the three ramified prime divisors P1 ; P2 ; P3
in K=Q. 23; t/, and from the Hurwitz genus formula it now follows that g.K/ D
above P3 there lie prime
0. Since p p divisors of degree 1, K is a rational function field
K D Q. 23; u/, and N=Q. 23; u/ yields the required Galois realization. The
class vector belonging to this extension can also be deduced from the permutation
types in the same way as for M11 in Theorem I.6.12. t
u
The stronger assertion for Aut.G/ instead of G is of importance in the context
of embedding problems treated in Chapter IV. Indeed, from Theorem 9.9 and the
definitions we immediately see:
Corollary 9.10. The sporadic groups different from M23 and M24 possess GA-
realizations
p over Q. The Mathieu groups M23 and M24 possess GA-realizations
over Q. 23/.
A GA-realization over Q for M24 will be proved in Theorem III.7.12.
10 Summary for Simple Groups 173

10 Summary for Simple Groups

For the convenience of the reader and for future use we collect the G- and GA-
realizations for finite simple groups proved in this Chapter.

10.1 Galois Realizations over Qab

The picture for G-realizations of simple groups over Qab is almost complete:

Theorem 10.1. The finite simple groups possesses G-realizations over Qab except
possibly the following exceptional groups of Lie type in characteristic 2:
2
B2 .22mC1 /; m  2; 2
F2 .22mC1 /; m  1; E6 .22m /; m  0;
E6 .22mC1 /; m  0;
2
E7 .2m /; m  1; E8 .2m /; m  1:

Proof. Theorem I.5.3 proves a G-realization for An , n  5. A G-realization for

Ln .q/ is given in Corollary 1.5, for Un .q/ in Theorem 3.2, for S2n .q/ in Theo-
rem 3.4, for O2nC1 .q/ in Theorem 3.7, for OC 2n .q/ in Theorems 3.11 and 3.13, and
for O2n .q/ in Theorem 3.15.
For the exceptional groups of Lie type 2G2 .32mC1 / the G-realization was con-
structed in Theorem 4.5, for G2 .q/ in Theorem 4.7, for 3D4 .q/ in Theorem 4.9,
for 2B2 .8/ in Theorem 4.10 and for 2F4 .2/0 in Theorem 4.11. For F4 .q/ the result
can be found in Theorem 5.5, for E6 .q/ with q ¤ 22m in Corollary 5.8 and Theo-
rem 5.14, for 2E6 .q/ with q ¤ 22mC1 in Theorems 5.11 and 5.15, for E7 .q/ with
odd q in Corollary 5.18, for E8 .q/ in Theorem 5.20. Finally the assertion for the
sporadic groups is contained in Theorem 9.9. t
u

Theorem 10.2. The following finite simple groups possess GA-realizations over the
field Qab .
(a) The non-abelian simple alternating groups An .
(b) The groups of Lie type G.p/ for 2 < p 2 IP with the possible exception
3
D4 .p/.
(c) The groups S2n .2/, OC 
2n .2/, O2n .2/.
(d) The sporadic simple groups.

Proof. The GA-realizations used in the theorem are contained in the following
results: Corollaries I.5.4 and I.9.8 prove GA-realizations for An for n ¤ 6 respec-
tively n D 6 over Q and hence over Qab .
GA-realizations for the groups Ln .p/ and Un .p/ were constructed in Corol-
lary 6.6. Using Aut.O2nC1 .p// D SO2nC1 .p/ we obtain GA-realizations of
O2nC1 .p/ over Qab from Theorem 3.7. The corresponding result for S2n .p/ fol-
lows from Theorem 3.4 since Aut.S2n .p// D PCSp2n .p/ for .n; p/ ¤ .2; 2/. GA-
realizations for OC2n .p/ with n  5 are obtained from Theorem 3.11 for p ¤ 2
174 II Applications of Rigidity

and Theorem 3.13 for p D 2, since all automorphisms of OC 2n .p/ are induced by
COC 2n .p/. The case O C
8 .p/ with the exceptional graph automorphism of order 3
was treated in Theorem 7.10(b). Correspondingly, we have a GA-realization for
O2n .p/ and n  4 over Q from Theorem 3.15 since here again all automorphisms
ab

of O2n .p/ are induced by CO


2n .p/. (The graph-field automorphism is already real-
ized inside the conformal group.)
A GA-realization for 2G2 .3/ Š Aut.L2 .8// was obtained in Example I.8.3. Theo-
rems 4.7, 8.1 and 8.2 give GA-realizations for G2 .p/, for F4 .p/ it follows from The-
orem 5.5 since Out.F4 .p// D 1 for p ¤ 2. GA-realizations over Qab for E6 .p/ with
odd p and 2E6 .p/ with odd p were obtained in Corollary 5.10 and Theorem 5.12.
The result for E7 .p/ follows from Theorem 5.18 because Aut.E7 .p// D E7 .p/ad .
Finally for E8 .p/ the result follows from Theorem 5.20 since Out.E8 .p// D 1.
By Theorem 9.9 and Corollary 9.10 all sporadic groups apart from M23 and p M24
have GA-realizations over Q and the latter two exceptions at least over Q. 23/
since their outer automorphism groups are trivial. This completes the proof of the
theorem. t
u

10.2 Galois Realizations over Q

Most of the known G-realizations of finite simple groups over Q are in fact GA-
realizations, so we just collect the latter (including a GA-realization for M24 which
will be proved in Theorem III.7.12):
Theorem 10.3. The following finite simple groups possess GA-realizations over the
field Q.
(a) The non-abelian simple alternating groups An .
(b) The linear groups Ln .p/ for gcd.n; p  1/ D 1, p > 3 and p 6 1 .mod 12/,
and for gcd.n; p  1/ D 2, n 2 .mod 4/ and p 5 .mod 8/, or n 0 .mod 4/
and p 7 .mod 12/.
(c) The unitary groups Un .p/ for gcd.n; p C 1/ D 1, p > 2 and p 6 1 .mod 12/,
and for gcd.n; p C 1/ D 2, n 2 .mod 4/ and p 3 .mod 8/, or n 0 .mod 4/
and p 5 .mod 12/.
(d) The symplectic groups S2n .p/ for odd primes p 6 ˙1 .mod 24/, p6 jn, or for
p D 2.
(e) The orthogonal groups O2nC1 .p/ for n  1 and odd primes p 6 ˙1
.mod 24/.
(f) The orthogonal groups O2nC1 .p 2 / for odd primes p ˙2 .mod 5/.
(g) The orthogonal groups OC 2n .p/ for p D 2, or for n  3 odd and p 3
.mod 8/ or p 7 .mod 12/, p6 jn.
(h) The orthogonal groups O 2n .p/ for p D 2, or for n  3 odd and p 5
.mod 12/, or n  6 even and p 3 .mod 8/, or n 2 .mod 4/ and p ˙2
.mod 5/.
10 Summary for Simple Groups 175

(i) The orthogonal groups OC

8 .p/ for p ˙2 .mod 5/, p ˙2; ˙3 .mod 7/.
( j) The groups G2 .p/.
(k) The groups F4 .p/ for p  5.
(l) The groups E6 .p/ for p 4; 5; 6; 9; 16; 17 .mod 19/, p 1 .mod 3/.
(m) The groups E6 .p/ for p 2; 3; 10; 13; 14; 15 .mod 19/, p 6
2
1 .mod 3/,
p > 3.
(n) The groups E8 .p/ for p  7.
(o) The sporadic simple groups with the possible exception of M23 .

Proof. This was shown in Corollaries I.5.4 and I.9.8 for the alternating groups.
Part (b) is contained in Theorems 6.7(a) and 6.8(a), the assertion (c) is proved in
Theorems 6.7(b) and 6.8(b). Part (d) was shown in Theorems 7.2 and 7.3, part (e)
in Theorem 7.4, part (f) in Theorem 7.5, part (g) in Theorems 7.6, 7.8 and 7.11,
part (h) in Theorems 7.7 and 7.8, part (i) in Theorem 7.11. For the exceptional
groups, the result is contained in Theorems 8.1, 8.2 and 6.8 for part (j), in Theo-
rem 8.4 for part (k), in Corollary 8.6 for part (l), in Theorem 8.7 for part (m), in
Theorem 8.9 for part (n) and in Corollary 9.10 for the sporadic groups except M24 .
A GA-realization over Q for M24 will be proved in Theorem III.7.12. t
u

GA-realizations for groups L2 .p 2 / are implicitely contained in Theorem 10.3(f)

because of L2 .p 2 / Š O3 .p 2 /. The group L2 .8/ is handled in Chapter I, Exam-
ple 8.3.
Further G- and GA-realizations of classical simple groups of Lie type are
obtained in the next Chapter III.10, and Chapter IV.4.3, respectively.
III Action of Braids

In this chapter we pursue an approach to the solution of the inverse problem of

Galois theory using several variables, which building on ideas of Hurwitz (1891),
goes back to results of Fried (1977, 1984) and Fried and Biggers (1982). A variant
more suitable for the realization of groups as Galois groups was proposed by Matzat
(1989) and subsequently developed in papers of Matzat (1991a) and in geometric
context, of Fried and Völklein (1991) (see also the survey article Matzat (1991c)).
This version will be presented here.
The advantage of this approach consists in the fact that the rather restrictive group
theoretic rigidity criterion in the case of one variable is replaced by a much weaker
rigidity condition on braid orbits. But this then only implies the regularity of the
field of definition over Q. The rationality needed for an application of Hilbert’s irre-
ducibility theorem must therefore be guaranteed by additional arithmetic conditions.
The first five paragraphs contain the basic material of the chapter developed in the
papers mentioned above. In Paragraphs 1 and 2 we collect the necessary group the-
oretic results on braid groups and braid actions. In Paragraph 3 the rigidity theorems
of Chapter I are transferred to Galois realizations of several variables and used to
construct cyclic polynomials in Paragraph 4. Paragraph 5 finally contains the basic
investigations on the rationality of fields of definition, a result concerning rational
translation, and ends with a criterion for GA-realizations in several variables.
In the following three paragraphs we consider questions of specialization and fur-
ther results. The specialization theorem in Paragraph 6 yields the connection with
Chapter I by characterizing Galois realizations in one variable as specializations
by unramified rational places of Galois realizations in several variables. This leads
in particular to the solution of the inverse problem over Hilbertian PAC-fields of
characteristic zero (Fried and Völklein (1991)). In Paragraph 7 we treat the addi-
tional use of geometric automorphisms of Matzat (1991a) and in Paragraph 8 the
relevance of the decomposition groups, which was first studied in loc. cit. and con-
tinued in Matzat (1993). Here the two biggest Mathieu groups serve as illustrating
examples.

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 177

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_3
178 III Action of Braids

The last two paragraphs present the Katz algorithm with some applications. Para-
graph 9 contains an algebraic version due to Dettweiler and Reiter (2000) of the
Katz algorithm for rigid local systems introduced in Katz (1996) with complete
proofs. With this it is possible to connect linearly rigid generating systems of linear
groups algorithmically with linear rigid systems of GLn with small n for example
over finite fields IFq . By compatibility with the braid action the same holds for lin-
early rigid braid orbits. In the applications in Paragraph 10, starting with rationally
(braid) rigid generating systems of subgroups of GL1 .q/ of GL2 .q/ we obtain ratio-
nally linear (braid) rigid generating systems for many classical linear groups over
Q. This complements the results of Theorem II.10.3 in particular for higher prime
powers q.
1 Braid Groups 179

1 Braid Groups

In this first paragraph we introduce the Artin and the Hurwitz braid groups as funda-
mental groups and collect some of their properties which will be needed later on. We
omit the proofs which can already be found in the monograph of Birman (1975).

1.1 The Artin Braid Group

For an arbitrary topological space X we call

Xr WD f.x1 ; : : : ; xr / 2 X r j xi ¤ xj for i ¤ j g

the r-fold uncomplete product of X . On this the symmetric group Sr acts in a

natural way by transformation of coordinates. The orbit space of Xr under this
action
XQr WD Xr =Sr
is called the r-fold uncomplete symmetric product of X .
If for X we choose the affine line A1 .C/ D C (with the topology of the Euclidean
plane), then the fundamental group of Xr with respect to the base point P0 D
.1; : : : ; r/ 2 Xr is called the pure (or unpermuted) Artin braid group Br and the
fundamental group of XQr with respect to the Sr -orbit PQ 0 of P0 the (full) Artin
braid group BQ r :

Br WD 1 .A1 .C/r I P0 /; .C/r I PQ 0 /:

top Q 1
BQ r WD 1 .A
top
(1.1)

Obviously Br is a normal subgroup of BQr with BQr =Br Š Sr .

Every element ˇ 2 BQ r possesses a unique lifting to a path b in Xr with b.0/ D
P0 D .1; : : : ; r/ and b.1/ D .1! ; : : : ; r ! / for an ! 2 Sr . The coordinate functions of
b.t/ D .b1 .t/; : : : ; br .t// then form a geometric braid in the sense of Artin (1925).
With the elements ˇi of BQ r depicted in Figure 1.1 we have:

Theorem 1.1 (Artin (1925)). (a) The full Artin braid group BQ r is generated by
ˇ1 ; : : : ; ˇr1 subject to the relations

ˇi ˇj D ˇj ˇi for 1 i <j r  1 and j ¤ i C 1;

ˇi ˇi C1 ˇi D ˇi C1 ˇi ˇi C1 for 1 i r  2:
(b) The pure Artin braid group Br is generated in BQ r by the elements
1 1
ˇij WD .ˇi2 /ˇi C1 ˇj 1 D .ˇj21 /ˇj 2 ˇi for 1 i <j r:
180 III Action of Braids

1 i i+1 r 1 i j r
0 A1(C) 0 A1(C)
b1 br
... ...

1 A1(C) 1 A1(C)
1 i i+1 r 1 i j r
Fig. 1.1 Artin braids ˇi : : : : : : and ˇij

The proof can be found in Birman (1975), Thm. 1.8 with Lemma 1.8.2, where
moreover defining relations are given for the pure Artin braid group (see also Hansen
(1989), Appendix 1, for a correction of the braid relations).
From the presentation of BQ r in Theorem 1.1 we immediately obtain a canonical
epimorphism
qr W BQ r ! Sr ; ˇi 7! .i; i C 1/ (1.2)
with ker.qr / D Br . Now let

BQ r WD qr1 .Sr1 / D fˇ 2 BQr j .r/qr .ˇ/ D rg (1.3)

denote the preimage of the stabilizer Sr1 of r under qr and

Fr1 WD h1 ; : : : ; r1 i with i WD ˇi r : (1.4)

Thus we obtain the fundamental structure theorem for the Artin braid group:

Theorem 1.2 (Chow (1948)). For r  3 we have:

(a) Fr1 is a free normal subgroup of BQ r of rank r  1.
(b) BQ r decomposes into a semidirect product of Fr1 with BQ r1 , where the action
of the generators ˇ1 ; : : : ˇr2 of BQ r1 on Fr1 is given componentwise by

.1 ; : : : ; r1 /ˇi D .1 ; : : : ; i 1 ; i i C1 i1 ; i ; i C2 ; : : : ; r1 /: (1.5)

The proof of this statement can either be found in the original paper or in the
proof of Lemma 1.8.2 and Cor. 1.8.3 in Birman (1975). As a consequence we obtain
the following result, which was also proved by Chow (1948) (see also Birman
(1975), Cor. 1.8.4):

Corollary 1.3. For r  3 the center of the Artin braid group BQ r is the infinite cyclic
group generated by .ˇ1    ˇr1 /r .

Remark. In the case r D 2 we have BQ 2 D hˇ1 i D Z .BQ 2 /.

1 Braid Groups 181

We defined the Artin braid groups as fundamental groups of the uncomplete sym-
metric products of C, respectively of the space of regular orbits under Sr on Cr .
The corresponding construction may be applied to all finite Coxeter groups (see for
example Suzuki (1982), Ch. 3, 4, or Aschbacher (1986), 29), and leads to groups
having similar properties to the Artin braid group. These have entered the literature
under the name of Artin groups (see Brieskorn and Saito (1972)).

1.2 The Hurwitz Braid Group

For the definition of the Artin braid groups the topological space X was chosen
to be the affine line A1 .C/ D C, and the base point P0 D .1; : : : ; r/. If instead we
replace it by the projective line IP1 .C/ D CO (with the topology of the sphere in
3-dimensional Euclidean space) and again use P0 D .1; : : : ; r/ by identifying the
O then we obtain the pure (or unpermuted)
points in IP1 .C/ with the elements of C,
Hurwitz braid group Hr and the ( full) Hurwitz braid group HQ r :

Hr WD 1 .IP1 .C/r I P0 /; HQ r WD 1 .IP .C/r I PQ 0 /:

top top Q 1
(1.6)

As above these satisfy Hr G HQ r with HQ r =Hr Š Sr if PQ 0 is chosen to be the Sr -orbit

of P0 .
In analogy to the Artin braid group the elements of HQ r can be represented as
geometric braids, but now between two spheres (see Figure 1.2). The right hand part
of Figure 1.2 also shows that in addition to the relations of the Artin braid group,
which are called braid relations, at least one further relation ˇ1    ˇr1 ˇr1    ˇ1 D
1 holds in HQ r .

i j 2
1 r
1 r
0 0
IP1(C) IP1(C)

i j 2 r
1 1 r 1 1
IP1(C) IP1(C)
Fig. 1.2 Hurwitz braids ˇij : : : : : : and ˇ1    ˇr1 ˇr1    ˇ1

Theorem 1.4 (Fadell and Van Buskirk (1962)). Let Nr be the normal subgroup of
BQ r generated by ˇ1    ˇr1 ˇr1    ˇ1 2 BQ r . Then we have

HQ r Š BQ r =Nr and Hr Š Br =Nr :

182 III Action of Braids

In particular a presentation of the full Hurwitz braid group resp. the pure Hurwitz
braid group is obtained from the one of BQ r resp. Br in Theorem 1.1 by addition of
the relation
ˇ1    ˇr1 ˇr1    ˇ1 D 1: (1.7)

Calling (1.7) (which coincides with ˇ12    ˇ1r D 1) or more generally the rela-
tions ˇ1j    ˇj 1;j ˇj;j C1    ˇjr D 1 sphere relations, we get the more general

Corollary 1.5. The normal subgroup Nr of BQr is generated by the sphere relations:

Nr D hˇ1j    ˇj 1;j ˇj;j C1    ˇjr j 1 < j < ri: (1.8)

The proof for these results closely follows the one of Theorem 1.1 (see for exam-
ple Birman (1975), Thm. 1.11). In contrast, the structure theorem for the Hurwitz
braid group corresponding to Theorem 1.2 was not included in the monograph of
Birman. For that, in analogy to (1.2)–(1.4), let

qr W HQ r ! Sr ; ˇi 7! .i; i C 1/ (1.9)

denote the canonical epimorphism with kernel Hr ,

HQ r WD qr1 .Sr1 / D fˇ 2 HQ r j .r/qr .ˇ/ D rg (1.10)

the preimage of Sr1 under qr and

Gr1 WD h1 ; : : : ; r1 i with i D ˇi r : (1.11)

Theorem 1.6 (Gillette and Van Buskirk (1968)). For r  4 we have:

(a) Gr1 is a free normal subgroup of HQ r of rank r  2 with the defining relation
1    r1 D 1.
(b) HQ r is an extension of Gr1 by HQ r1 , where the action of the generators
ˇ1 ; : : : ; ˇr2 of HQ r1 on Gr1 is given by the following formula (up to simultaneous
conjugation in Gr1 )

Œ1 ; : : : ; r1 ˇi
D Œ1 ; : : : ; i 1 ; i i C1 i1 ; i ; i C2 ; : : : ; r1 : (1.12)

Proof. The sphere relation for j D r gives

ˇ1r    ˇr1;r D 1    r1 D 1: (1.13)

Hence a presentation of Gr1 is obtained from one of Fr1 by addition of the rela-
tion (1.13). According to Theorem 1.2(a) the group Gr1 is thus free of rank r  2.
Being the image of Fr1 under the canonical surjection from BQ r onto HQ r , Gr1
is a normal subgroup of HQ r with factor group isomorphic to .BQ r1 / D HQ r1 . Since
the action of BQ r1 on Fr1 commutes with , (1.12) follows from (1.5). t
u
1 Braid Groups 183

Remark. For r 3 the Hurwitz braid groups HQ r are finite and therefore require a
separate treatment. We have

HQ 2 D hˇ1 j ˇ12 D 1i Š Z2 ; (1.14)

HQ 3 D hˇ1 ; ˇ1 ˇ2 j ˇ12 D .ˇ1 ˇ2 ˇ1 /2 D .ˇ1 ˇ2 /3 i Š Z3 Z4 (1.15)

and hence H2 D 1 and H3 Š Z2 .
With the help of the Schreier transversal

i WD ˇr1    ˇi for i D 1; : : : ; r

of HQ r in HQ r (resp. BQ r in BQr ) we obtain, using the Reidemeister rewriting process

(see Lyndon and Schupp (1977), Ch. II, Prop. 4.1), the following simple and useful
presentation of HQ r .
Proposition 1.7. For r  3 the group HQ r is generated by ˇ1 ; : : : ; ˇr2 . The braid
relations of BQ r1 together with

.ˇ1    ˇr2 /2.r1/ D 1

constitute a complete set of defining relations.

A proof for this assertion and for the subsequent Corollary 1.8, which also
originates from Gillette and Van Buskirk (1968), is obtained from the proof of
Lemma 4.2.3 in Birman (1975).
Corollary 1.8. For r  3 the center of the Hurwitz braid group HQ r has order 2 and
is generated by
r WD .ˇ1    ˇr1 /r D .ˇ1    ˇr2 /r1 : (1.16)
The factor group HQ r =Z .HQ r / is known under the name of mapping class group
of the r-pointed sphere (see Birman (1975), Thm. 4.5). For later use we mention
the following structure result for the first infinite Hurwitz braid group HQ 4 and for
HQ 4 =Z .HQ 4 / (see Birman (1975), Lemma 5.4.1 and 5.4.3):
Proposition 1.9. HQ 4 has a quaternion group Q8 of order 8 as normal subgroup
with
HQ 4 =Q8 D PSL2 .ZZ/ for Q8 D hˇ1 ˇ31 ; .ˇ1 ˇ2 ˇ3 /2 i: (1.17)
The induced group extension of HQ 4 =Z .HQ 4 / splits:

HQ 4 =Z .HQ 4 / Š Z22  PSL2 .ZZ/: (1.18)

1.3 The Pure Hurwitz Braid Group

The fundamental difference between the structure theorem for the Artin braid group
and the Hurwitz braid group lies in the fact that the underlying exact sequence splits
in the case of the Artin braid group. This is impossible for the Hurwitz braid group,
184 III Action of Braids

since HQ r does not even contain subgroups isomorphic to HQ r1 (see Gillette and Van
Buskirk (1968), Thm. 4.12). Nevertheless the corresponding result for the pure Hur-
witz braid group remains true. The first proof of this result used mainly topological
methods (Fadell (1962)); the purely algebraic proof given here was communicated
to us by L. Schneps.
Theorem 1.10 (Fadell and Van Buskirk (1962)). For r  4 the pure Hurwitz braid
group Hr is a semidirect product of the free normal subgroup Gr1 with Hr1 .
Proof. Since the sphere relation makes it impossible to embed Hr1 canonically
into Hr , we will denote the elements of Br1 and Hr by ˇij and the corresponding
elements of Br2 and Hr1 by ˇQij . Furthermore, we let

Hr WD hˇij 2 Hr j 1 i <j r  1; .i; j / ¤ .1; 2/i;


Br1 WD hˇij 2 Br1 j 1 i <j r  1; .i; j / ¤ .1; 2/i:
According to Corollary 1.5 upon replacing ˇij 2 Hr by ˇij 2 Br1

the groups Hr

and Br1 then possess identical presentations, which shows

Hr Š Br1

: (1.19)

In particular Hr contains no non-trivial elements of finite order by Theorem 1.2.

The generator r of the center of Hr then satisfies

r D .ˇ1    ˇr2 /r1 D ˇ12 .ˇ13 ˇ23 /    .ˇ1;r1    ˇr2;r1 /;

thus ˇ12 lies in the group generated by Z .Hr / and Hr , yielding Hr D hHr ;
Z .Hr /i. As r is of order 2, we moreover have Hr \ Z .Hr / D 1, so that

Hr D Hr  Z .Hr /: (1.20)

By (1.19) and (1.20) the inclusion '  W Br2

 
! Br1 can be extended to a monomor-
phism
 
' W Hr1 Š Br2  Z .Hr1 / ! Br1  Z .Hr / Š Hr
by sending the non-trivial central element r1 D Qr onto r . Composing ' with the
canonical projection

W Hr ! Hr1 ; ˇ 7! ˇQ D ˇGr1

we obtain for .ˇQ  ; Q / 2 Hr  Z .Hr / with ˇQ  2 Hr1


and Q 2 fQr ; 1g

. ı '/.ˇQ  ; /
Q D .ˇ  ; / D ˇ  Gr1 D ˇQ  Q

and thus ı ' D IdHr1 . Consequently we have Gr1 \ '.Hr1 / D 1 and also
Hr D hGr1 ; '.Hr1 /i, which finally implies

Hr D Gr1 '.Hr1 / Š Gr1 Hr1 : t

u
1 Braid Groups 185

1.4 The Word Problem

For an arbitrary group G let OG denote the system of normal subgroups of finite
index:
OG WD fO G G j .G W O/ < 1g: (1.21)
An arbitrary normal subgroup U of G is called a thick normal subgroup if

OUG WD fO \ U j O 2 OG g (1.22)

constitutes a cofinal system in OU , i.e., if for any O 2 OU there exists a OQ 2 OUG

with OQ O.
Thick normal subgroups are useful for example for inductive proofs of residual
finiteness. Namely, we have the

Proposition 1.11. If the group G possesses a residually finite thick normal subgroup
U with residually finite factor group G=U , then G itself is residually finite.

Proof. Let \
K WD O:
O2OG

Then since U is residually finite and OUG is cofinal in OU we have

\ \
K \U D .O \ U / D OU D 1:
O2OG G
OU 2OU

Now let GN WD G=U with canonical epimorphism  W G ! G, N and KN WD .K/.

N 1 N
Then for all O 2 OGN the inverse image  .O/ is normal in G of finite index and
by definition contains K. By the residual finiteness of GN it now follows that
\
KN ON D 1
N
O2O N
G

and hence K U , which together with K \ U D 1 finally yields K D 1. t

u

To apply this result to the Hurwitz braid group, one first shows the following
result, which also proves useful in a different context.

Proposition 1.12. The group Gr1 is a thick normal subgroup of HQ r and therefore
also of the pure Hurwitz braid group Hr .

Proof. Let first G D Hr and U D Gr1 . Then from Theorem 1.10 we know G Š
U H with H Š Hr1 . Since U is a finitely generated normal subgroup of G, the
group \
OU WD O
2G
186 III Action of Braids

has finite index in U for any O 2 OU (Lyndon and Schupp (1977), Ch. IV,
Thm. 4.7), and hence is an element of OU satisfying OU G G. For any OH 2 OH
the group hOU ; OH i D OU OH has finite index in G, thus it follows that
\
OG WD .OU OH / 2 OG
2G

with OG \ U D OU . We deduce that O OU 2 OUG , which shows that OUG is a

cofinal subsystem of OU .
Now G has finite index .r  1/Š in GQ WD HQ r , hence for all OU 2 OUG it follows
that \ Q
OQ WD OU 2 OUG :
Q
2G
Q
Thus OUG also constitutes a cofinal subsystem of OU . t
u

From Propositions 1.11 and 1.12 we may now deduce the main result of this
section.

Theorem 1.13. The full Hurwitz braid group HQ r is a finitely generated residually
finite group.

Proof. The pure Hurwitz braid group Hr is finite for r 3 and by Theorem 1.6 for
r  4 possesses a free normal subgroup Gr1 of rank r  2 with Hr =Gr1 Š Hr1 .
Being a free group, Gr1 is residually finite (see for example Lyndon and Schupp
(1977), Ch. III, Prop. 7.11 with supplement on p. 195). By Proposition 1.12, Gr1
is hence a residually finite thick normal subgroup of Hr . Thus at least for r D 4 all
assumptions in Proposition 1.11 are satisfied, proving that H4 is residually finite.
Induction on r then proves the result for all r  4.
Since .HQ r W Hr / < 1, the group Hr is a thick normal subgroup of HQ r , which
again with Proposition 1.11, now applied to G D HQ r and U D Hr entails the residual
finiteness of HQ r . t
u

As a corollary to Theorem 1.13, using a result of McKinsey (Lyndon and Schupp

(1977), Ch. IV, Thm. 4.6) we obtain a nice consequence which was first proved by
Fadell and Van Buskirk (1962) by different methods:

Corollary 1.14. The Hurwitz braid group HQ r has solvable word problem.

In a completely analogous way one may prove the solvability of the word prob-
lem for the Artin braid groups (compare with Artin (1925, 1947)).
2 Profinite Braid Groups 187

2 Profinite Braid Groups

In analogy to Section I.1.2, the universal covering of the uncomplete (symmetric)

product of IP1 .C/ defines a Galois extension, whose structure reflects that of the
Hurwitz braid groups. As in the case of one variable the canonical generators of the
braid group play the role of generators of inertia groups.

2.1 The Hurwitz Braid Group as Galois Group

For a field k and an algebraic variety X let k.X / denote the field of k-valued
functions on X .
Proposition 2.1. For X D IP1 .C/ we have:
(a) The field C.Xr / is purely transcendental over C of degree r and is generated
by the functions

ti W Xr n Ui ! C; .x1 ; : : : ; xr / 7! xi for i D 1; : : : ; r;

where Ui denotes the hyperplane defined by xi D 1.

(b) The field C.XQr / is purely transcendental over C of degree r and is generated
by the elementary symmetric functions
X
tQj WD ti1    tij for j D 1; : : : ; r:
1i1 <<ij r

Proof. Since Xr is an open and dense subset (with the Zariski topology) of the
product X r , the function fields of Xr and X r coincide, i.e., we have

C.Xr / D C.X r / D C.t1 ; : : : ; tr /:

From the definition of the Sr -covering Xr ! XQr in Section 1.1 we thus obtain

C.XQr / D C.tQ1 ; : : : ; tQr /: t

u

Now let k be an algebraically closed field and X a quasiprojective normal vari-

ety. Then a field extension L=k.X / is called unramified over X , if the normal-
ization Y of X in L is unramified over X . The set of finite extension fields of
K WD k.X / unramified over X is closed under composition and taking of the
Galois hull (in a given algebraic closure of K). Thus, the union of all such fields
forms a maximal extension field M of K unramified over X , which is moreover
Galois over K (see Grothendieck (1971), Exp. I.10, Exp. V.4+8 and Exp. XII.5, or
Popp (1970), 1. Vorl.). The corresponding Galois group is then called the algebraic
fundamental group of X :
alg
1 .X / WD Gal.M=K/:
188 III Action of Braids

In our case we thus obtain:

Theorem 2.2. Let Mr be a maximal field extension of C.Qt/ D C.tQ1 ; : : : ; tQr / unrami-
fied over the r-fold uncomplete symmetric product of X D IP1 .C/. Then the Galois
group of Mr =C.Qt/ is isomorphic to the profinite completion of the full Hurwitz braid
group HQ r :
Gal.Mr =C.Qt// D 1 .XQr / Š .HQ r /O:
alg
(2.1)
Here C.t/ D C.t1 ; : : : ; tr / is the fixed field of the profinite completion of the pure
Hurwitz braid group:

Gal.Mr =C.t// D 1 .Xr / Š HO r :

alg
(2.2)

Proof. The proof runs entirely along the lines of the proof of Theorem I.1.3, with
the only difference that here we have to cite the higher dimensional version of the
Riemann Hebbarkeitssatz and the Riemann Existence Theorem.
Obviously Xr is sufficiently connected (in the sense of Stöcker and Zieschang
(1988), Def. 6.4.3). Thus there exists a universal covering

u W XOr ! Xr

whose group of covering transformations is isomorphic to the fundamental group of

Xr with respect to the base point P0 D .1; : : : ; r/ and hence to the pure Hurwitz
braid group via
! W Deck.u/ ! 1 .Xr I P0 / D Hr ;
top
(2.3)
depending on a point PO 0 2 XOr with u.PO 0 / D P0 . From the fundamental theorem
for coverings of topological spaces it now follows that for each finite unramified
normal covering
p  W Yr ! Xr
there exists a unique normal subgroup O of finite index in Deck.u/ such that Yr is
homeomorphic to the orbit space XOr =O (equipped with the quotient topology). Via
the canonical map of XO  onto XO  =O we obtain a universal covering
r r

v W XOr ! Yr Š XOr =O

satisfying p  ı v D u and Deck.v/ Š O, where moreover

Deck.p  / Š Deck.u/= Deck.v/:

By a theorem of Grauert and Remmert (see for example Grothendieck (1971),

Exp. XII, Thm. 5.4, or also Popp (1970), 1. Vorl.), the covering p  possesses a
unique continuation
p W Yr ! X r
to a normal projective manifold Y , in which Y  lies dense and for which moreover
r r

Deck.p/ Š Deck.p  /:
2 Profinite Braid Groups 189

The field N WD C.Yr / is a Galois extension of K WD C.t/ unramified over Xr ,

whose degree equals the number of sheets of the covering p respectively p  by
the Grothendieck version of the Riemann existence theorem (Grothendieck (1971),
Exp. XII, Thm. 5.1). This implies

Gal.N=K/ Š Deck.p/

(with Gal.N=K/ acting from the right) und hence finally

Gal.N=K/ Š Hr =!.O/:

On the other hand, every finite Galois extension field N of K unramified over
Xr determines via the normalization of Xr in N a normal unramified covering Yr
of Xr with C.Yr / D N. Thus the above mapping between the set Or of normal
subgroups of finite index of Deck.u/ and the set Nr of finite Galois extensions
of K unramified over Xr becomes a Galois correspondence. Since the maximal
extension field Mr of K unramified over Xr can be regarded as the union of all N 2
Nr , the Galois group of Mr =K is the projective limit of the Galois groups Gal.N=K/,
hence
Gal.Mr =K/ Š lim .Hr =!.O//O2Or D HO r :


The canonical map qQ W Xr ! XQr is unramified, so uQ WD qQ ı u becomes a universal

covering of XQr , which finally proves

Q Š .HQ r /O;
Gal.Mr =K/
Q WD C.Qt/.
where we have set K t
u

In what follows we call the profinite completions of Hr resp. HQ r profinite Hur-

witz braid groups. Lacking a suitable different letter in the Greek alphabet we will
again denote them by Hr resp. HQ r . This should not give rise to any confusion, since
the discrete Hurwitz braid groups will from now on always occur as fundamental
groups 1 .Xr I P0 / or 1 .XQr I PQ 0 / respectively, for P0 D .1; : : : ; r/. (If neces-
top top

sary they will be distinguished from the profinite groups by using the symbols HL r
or HQ r_ respectively.)
We see that by Theorem 2.2 the profinite Hurwitz braid groups also satisfy

HQ r =Hr Š Sr : (2.4)

Furthermore Theorem 1.13 immediately implies (compare Serre (1964), Ch. I,

1.1):

Corollary 2.3. The canonical map from the (discrete) Hurwitz braid group into the
profinite Hurwitz braid group is injective.

Consequently in what follows, we do not distinguish between the elements of

1 .XQr I PQ 0 / and their embedded images in HQ r (depending on PO 0 by (2.3)).
top
190 III Action of Braids

2.2 Inertia Groups

As in the 1-dimensional case the embedded generators ˇi of HQ r resp. ˇij of Hr can

be interpreted as generators of inertia groups in Mr =C.Qt/ resp. Mr =C.t/.

Theorem 2.4. Let K D C.t/ and Dij be the valuation ideal of K defined by .ti  tj /
in the corresponding local ring. Further let Mr =K be a maximal field extension
unramified over IP1 .C/r with the Galois group 1 .IP1 .C/r / D Hr . Then there
alg

exists a valuation ideal D O ij of Mr lying above Dij such that ˇij generates the
O ij =Dij :
inertia group of D
I.DO ij =Dij / D hˇij iO: (2.5)

Proof. The notations introduced in the proof of Theorem 2.2 are tacitly used. Fur-
ther let Dij be the hyperplane xi D xj in X r and Dij the complement of the inter-
sections Dij \ Dkl for fk; lg ¤ fi; j g in Dij .
Let u W XOr ! Xr be an universal covering of Xr and PO 0 2 XOr the preimage
of the base point P0 D .1; : : : ; r/ 2 Xr under u used for (2.3). For a closed path
bij in the homotopy class ˇij by the main lemma of covering theory there exists a
unique lifting bQij to Yr with bQij .0/ D PQ 0 WD v.PO 0 /.

i j
1 r
0
IP1(C)

i j
1 1 r
IP1(C)
Fig. 2.1 Deformation of bij

We now deform the path bij according to Figure 2.1 homotopically in Xr [ Dij
to a path bij in X r , which apart from the intersection point Pij D .1; : : : ; j  1;
i; j C 1; : : : ; r/ with Dij lies completely inside Xr . Then bij also possesses a lift-
ing bQij in Yr with respect to the continuation p of p  . (The preimage PQ ij of Pij
is uniquely determined, since the continuation p of p  is continuous.) Denote the
irreducible component of the fiber p 1 .Dij / containing PQ ij by DQ ij , and by dij the
covering transformation of p determined by dij .PQ 0 / D bQij .1/. Then DQ ij remains
pointwise fixed under dij due to dij .PQ ij / D PQ ij .
As a closed submanifold of Yr of codimension 1, DQ ij defines a discrete ultramet-
ric valuation on N continuing the valuation of K belonging to Dij . The correspond-
ing valuation ring and valuation ideal are denoted by O Q ij resp. D
Q ij . For any f 2 O
Q ij
2 Profinite Braid Groups 191

and the automorphism ij 2 Gal.N=K/ defined by f ij .P/ Q D f .dij .P//
Q we have
f ij
f 2 DQ ij , hence the element ij is contained in the inertia group of D Q ij =Dij .
If conversely  lies in the inertia group of D Q ij =Dij , then DQ ij belongs to the fixed
point manifold of the corresponding covering transformation d of Yr =X r (see for
example Popp (1970), Lemma 1.22 and 1.23). In particular the point PQ ij is fixed
by d . On the side of the braid group HL r D 1 .IP1 .C/r I P0 / this means that d cor-
top

responds to a braid ˇ which becomes trivial by identifying the coordinates i and j

of P0 along bij (see Figure 2.2). (From the point of view of homotopy groups this
procedure describes a deformation of the loops b 2 ˇ to loops inside Dij by moving
the base point P0 to the base point Pij 2 Dij along bij , see also Paragraph 8.1.)
This implies that ˇ is a power of ˇij . Thus ij even generates the inertia group of
DQ ij =Dij .

bi j bi∗j

· · · · · · · ·
1 r
i−1 i i+1 j−1 j j+1

Fig. 2.2 Moving P0 to Pij

By construction we thus have

[
.ij /O2Or D ˇij 2 Hr ; O ij WD
and D Q ij
D
N2Nr

is a valuation ideal of Mr lying above Dij whose inertia group is generated by ˇij
as a topological group. t
u

Theorem 2.4 has the following consequence for the Galois extension Mr =C.Qt/:

Corollary 2.5. Let KQ D C.Qt/ be the field of rational functions of the uncomplete
symmetric product XQr of X D IP1 .C/ and D Q i the valuation ideal of KQ con-
tained in the valuation ideal Di;i C1 of K D C.t/. Then Mr =KQ is Galois with group

1 .XQr / D HQ r , and we have
alg

O i;i C1=D
I.D Q i / D hˇi iO: (2.6)

Proof. The Galois group of Mr =KQ had already been determined in Corollary 2.1.
By Theorem 2.4 we have
O i;i C1=Di;i C1/ D hˇ 2 iO;
I.D i
192 III Action of Braids

Q i . It consists of those
hence it only remains to identify the inertia group of Di;i C1 =D
elements of
Gal.K=K/ Q Š Sr D qr .HQ r /

whose corresponding covering transformation of X r over X r =Sr leaves the hyper-

plane Di;i C1 pointwise fixed, hence of the powers of the transposition .i; i C 1/. The
assertion now follows since qr .ˇi / D .i; i C 1/. t
u
Before we study the field extension Mr =C.tQ/ in more detail, we prove some fur-
ther results on the structure of the Hurwitz braid group.

2.3 Structure of the Profinite Hurwitz Braid Group

The key to the following structure theorems lies in the result proved in Proposi-
tion 1.12 asserting that Gr1 is a thick normal subgroup of the Hurwitz braid group.
Namely more generally we have:
Proposition 2.6. Let G be a finitely generated group and U a thick normal subgroup

1
of G. Then we have:
(a) UO is a closed normal subgroup of GO and G=
O UO Š G=U .
(b) If U possesses a complement H in G, then GO Š UO  HO .
Proof. By assumption UO is not only the projective limit of the groups U=O for
O 2 OU but we also have

UO D lim .U=O/
Q Q G
O2O
 U

with the OUG defined in (1.22). For any O 2 OG the exact sequence
ı
1 ! UO=O ! G=O ! G=O UO=O ! 1

together with the isomorphism theorems immediately implies the exactness of

1 ! U=.U \ O/ ! G=O ! G=UO ! 1:

Since passage to the projective limit is an exact functor (Ribes (1970), Ch. I,
Prop. 3.6), we obtain the exact sequence of profinite groups asserted in (a).
Now let H be a complement to U in G. Obviously for each O 2 OG we have
OU WD O \ U 2 OU and OH WD O \ H 2 OH . Since OU G G the subgroup of G
generated by OU and OH is a semidirect product of finite index in G, and it follows
that \
OQ WD .OU OH / 2 OG :
2G

If as above we set OQ U WD OQ \ U and OQ H WD OQ \ H , then clearly OQ U D OU and

.OH W OQ H / ..OU OH / W O/. Q Consequently OQ coincides with the semidirect
2 Profinite Braid Groups 193

product generated by OU and OQ H . Since OQ has finite index in G, OQG WD fOQ j

O 2 OG g forms a cofinal subsystem of OG . For OQ 2 OQG the split exact sequence

1 ! U ! G ! H ! 1

first implies the sequence


Q OQ ! G=OQ !
1 ! U O= Q OQ ! 1
H O=
Q OQ Š U=OU and H O=
with U O= Q OQ Š H=OQ H . Since

Q D .U W OU /.H W OQ H / D .U OQ W O/.H
.G W O/ Q OQ W O/
Q

this is exact and clearly also splits. Passage to the projective limit yields the splitting
sequence of profinite groups claimed in (b). t
u

We denote the profinite completion of Gr1 by r1 :

r1 WD GO r1 D h1 ; : : : ; r1 j 1    r1 D 1iO: (2.7)

Then Theorems 1.6 and 1.10 yield:

Theorem 2.7. For r  4 we have:

(a) r1 is a closed normal subgroup of HQ r WD fˇ 2 HQ r j qr .ˇ/ 2 Sr1 g, free of
rank r  2, with
HQ r = r1 Š HQ r1 : (2.8)
The action of the generators ˇ of the factor group isomorphic to H Q  on
r1 i r1
r1 is given by the formulae in Theorem 1.6.
(b) r1 possesses in Hr a complement isomorphic to Hr1 :

Hr Š r1 Hr1 : (2.9)

In the next section we investigate the fixed field of r1 .

2.4 The Fixed Field of the Free Normal Subgroup

The following theorem explains inductively the stepwise structure of the Galois
extension Mr =K with K D C.t/:

Theorem 2.8. Let Mr =K for r  4 be the field extension described in Theorem 2.2
with Gal.Mr =K/ D Hr . Then the fixed field of the free normal subgroup r1 of Hr
is the field Mr1 .tr /:
M r
r1
D Mr1 .tr /: (2.10)
Moreover Mr1 is algebraically closed in Mr .
194 III Action of Braids

Proof. Let X WD IP1 .C/ and N WD M

r
r1
. Since
O i r =Di r / D hˇi r iO
I.D r1

the valuation ideals Di r of K are unramified in N=K for i D 1; : : : ; r  1. Conse-

quently N=K is unramified even over Xr1   X , which since
top  top 
1 .Xr1  X I .1; : : : ; r// Š 1 .Xr1 I .1; : : : ; r  1// implies that N is a subfield
of Mr1 .tr / for which moreover

Gal.N=K/ Š Hr = r1 Š Hr1 Š Gal.Mr1 .tr /=K/:

Obviously the kernel of the canonical epimorphism from Gal.Mr1 =K/ to Gal.N=K/
lies in the intersection of all normal subgroups of finite index and hence equals the
trivial group, which forces N D Mr1 .tr /.
If L is an intermediate field of Mr =Mr1 algebraic over Mr1 , then L is algebraic
over C.t1 ; : : : ; tr1 / and unramified over Xr \ X r1 D Xr1
 . Thus it is a subfield
of Mr1 which finally proves that L D Mr1 . t
u
Descending induction over s then immediately leads to:
Corollary 2.9. For 3 s r the group
.r/
s WD hˇij j 1 i < j; s < j riO Hr (2.11)

is normal in Hr and has fixed field

.r/
M
r
s D Ms .tsC1 ; : : : ; tr /: (2.12)

The following description of the field extension Mr =Mr1 builds the connection
to the 1-dimensional case (compare Theorem I.1.3).
Theorem 2.10. Let S be the set of numerator divisors of .ti  tr / for i D 1; : : : ; r  1
and r  4 in the divisor group of Mr1 .tr /=Mr1 . Then Mr is a maximal Galois
extension field of Mr1 .tr /, regular over Mr1 and unramified outside S. Further
Mr is characterized by these properties among the extension fields of Mr1 .tr /.
Proof. By Theorem 2.8 the field extension Mr =Mr1 is regular. Further at most
the valuation ideals Mr1 Di r of Mr1 .tr / ramify in Mr =Mr1 .tr /. These are the
numerator divisors of .ti  tr / in the divisor group of Mr1 .tr /=Mr1 . After exten-
sion of constants by an algebraic closure M O r1 of Mr1 (in an algebraic closure of
K) we obtain a Galois extension M O r1 Mr =M O r1 .tr / with Galois group isomorphic
O
to r1 , unramified outside the set S of extensions of S to M O r1 .tr /. Thus M
O r1 Mr
O O
is a subfield of the maximal extension field Mr of Mr1 .tr / unramified outside S, O
whose Galois group over M O r1 .tr / is free profinite of rank r  2 by Theorem I.1.3.
Consequently the Galois extension M O r =M
O r1 Mr induces a canonical epimorphism
from r1 onto itself, whose kernel lies in the intersection of all open normal sub-
groups of r1 and hence is trivial. So we have M O r1 Mr D M O r , which proves the
maximality of Mr .
2 Profinite Braid Groups 195

O r =Mr1 .tr / is Galois with

By what precedes M

O r =Mr1 .tr // D Gal.M

Gal.M O r =Mr /  Gal.M
O r =M
O r1 .tr //:

The center of r1 is trivial, so Mr is characterized inside the whole Galois exten-
sion as the fixed field of the centralizer of Gal.M O r1 .tr //.
O r =M t
u

Remark. By a similar consideration the first layer in the above tower of fields M3 =K,
where K D C.t1 ; t2 ; t3 /, with group H3 Š Z2 , can be seen to be
p
M3 D K. .t1  t2 /.t1  t3 /.t2  t3 //: (2.13)
196 III Action of Braids

3 Galois Descent

Using a similar approach to that of the first chapter, we now study the fields of
definition of intermediate fields of the Galois extension Mr =Mr1 .t/ with the free
Galois group r1 by means of the Hurwitz classification. Among other results this
then leads to variants in several variables of the Basic Rigidity Theorem and the
Strong Rigidity Theorem.

3.1 An Arithmetic Fundamental Group

For the descent from C to Q N we need a generalization of Proposition I.2.1 to function

fields of several variables. This can be obtained from Weil’s rationality criterion for
regular extensions of fields of constants (see Weil (1956), Thm. 4). The following
variant is proved in the book of Serre (1992), Thm. 6.3.3.
Proposition 3.1. Let kN be an algebraically closed subfield of C, X an algebraic
variety over kN and KN WD k.XN / respectively K WD C.X / the function field of X
N
over k, C respectively. Then for every finite field extension N=K unramified over X
there exists precisely one field extension NN =KN unramified over X and regular over
kN with NN C WD NN ˝kN C D N.
Using the steps of the proof of Theorem I.2.2 we thus obtain here:
Theorem 3.2. Let MN r be a maximal extension field of Q. N Qt/ unramified over the r-
N 1 N
fold uncomplete symmetric product of X .Q/ D IP .Q/. Then the Galois group of
N Qt/ is isomorphic to the profinite Hurwitz braid group:
MN r =Q.

N Qt// D  alg .XQr .Q//

Gal.MN r =Q. N Š HQ r : (3.1)
1

N
Herein Q.t/ is the fixed field of the profinite pure Hurwitz braid group:

Gal.MN r =Q.t//
N D 1 .Xr .Q//
N Š Hr :
alg
(3.2)

Via this isomorphism the interpretation of the generators ˇi of HQ r resp. ˇij of Hr as

generators of inertia groups according to Theorem 2.4 and Corollary 2.5 is trans-
N
ported to the field extensions over Q.
In the following we will hence identify HQ r with Gal.MN r =Q. N tQ// and Hr with
N
Gal.MN r =Q.t//.
The Splitting Theorem I.2.4 with Theorem I.2.6 carries over to the present situa-
tion in the following form:
Theorem 3.3 (Splitting Theorem). The maximal extension field MN r of Q. N tQ/ unram-
1
N r is Galois over Q.Qt/ with
Q .Q/
ified over IP

Gal.MN r =Q.Qt// Š HQ r  Q: (3.3)

3 Galois Descent 197

Herein the absolute Galois group Q of Q acts on the generators ˇi of HQ r and ˇij
of Hr via the cyclotomic character c, i.e., for ı 2 Q we have

Œˇi ı
D Œˇic.ı/ and Œˇij ı
D Œˇijc.ı/ : (3.4)

Proof. Obviously the ramification locus XQ r .Q/ N n XQr .Q/

N of MN r =Q.
N Qt/ remains
invariant under all automorphisms ı of Q. N Qt/=Q.Qt/. Now MN r is uniquely determined
in a fixed algebraic closure of Q. N Qt/ as the maximal extension field of Q.N Qt/ unramified
over XQr .Q/,
N so any lifting ıN of ı maps MN r onto itself. Hence ıN is an automorphism
of MN r =Q.Qt/, and MN r =Q.Qt/ is Galois with group Q , say.
Clearly HQ r and Hr lie normal in Q . Let P denote the valuation ideal of Q.t/
of functions vanishing at a point P of Xr .Q/. Then any valuation ideal P O of MN r
above P is unramified in MN r =Q.t/. Since P=P O moreover does not decompose in
N
Q.t/=Q.t/, and completely decomposes in MN r =Q.t/, N the decomposition group of
O
P=P yields a complement to Hr isomorphic to Q in WD Gal.MN r =Q.t//. This
then also forms a closed complement to HQ r in Q D .MN r =Q.Qt//.
For the proof of the second assertion we may utilize the proof of Theorem I.2.6.
Denote by Mr1 .tr / the fixed field of a complement to Hr1 in Gal.MN r1 .tr /=
Q.t//. Then we have

Gal.MN r =Mr1 .tr // Š r1  Q :

By Theorem 2.10 the extension MN r =MN r1 .tr / is maximal Galois unramified outside
the set S of numerator divisors Pi of .tr  ti / for i D 1; : : : ; r  1 with Galois group
N
r1 . Since Gal.Mr1 .tr /=Mr1 .tr // leaves this set pointwise fixed, the proof of
Theorem I.2.6 yields
ı ı c.ı/ c.ı/
Œˇi r D Œi D Œi D Œˇi r :

Exchanging the variables ti we thus obtain the corresponding result for all genera-
tors ˇij of Hr .
By Corollary 2.5 the braid ˇi generates an inertia group over a valuation ideal
of Q.Qt/ in Q , so the conjugacy class Œˇi is also mapped to a power of itself. Since
ˇi2 D ˇi;i C1 this finally implies Œˇi ı D Œˇic.ı/ . t
u

As in Chapter I we refer to the Galois groups Q D Gal.MN r =Q.Qt// and D

Gal.MN r =Q.t// as arithmetic fundamental groups (belonging to HQ r resp. Hr ).

3.2 Hurwitz Classification

Theorem 2.10 and Proposition 3.2 show that with the maximal Galois extension
MN r =MN r1 .tr /, unramified outside the set of numerator divisors Pi of .tr  ti / for
i D 1; : : : ; r  1 in the divisor group of MN r1 .tr /=MN r1 we are in the same situation
198 III Action of Braids

as with the Galois extension MN S =Q.t/

N in Chapter I. From now on we always let
s WD r  1 with s  3 and, if necessary, t WD tr .
For a finite group G we correspondingly set
N s .G/ WD fNN j MN s .t/
N NN MN sC1 ; Gal.NN =MN s .t// Š Gg: (3.5)

For  2 ˙s .G/ let  be the continuous homomorphism from s D Gal.MN sC1 =

MN s .t// onto G with kernel ker. / defined by  ./ D  (compare I, (4.2)). Then
in complete analogy to Theorem I.4.1 we have:

Theorem 3.4 (Hurwitz Classification). For s  3 the fields NN 2 NN s .G/ are parame-
terized by the classes of generating s-systems  Aut.G/ 2 ˙s .G/= Aut.G/. More pre-
cisely there exists a bijection

N s .G/;  Aut.G/ 7! NN  WD MN sC1 :

ker. /
Ns W ˙s .G/= Aut.G/ ! N (3.6)

The components i of the parameter  generate inertia groups of prime ideals of

NN  =MN s .t/ over Pi via

' W G ! Gal.NN  =MN s .t//; i 7! 1

 .i / ker. /: (3.7)

Remark. Ns may also be interpreted as a surjective map

N s .G/;
Ns W ˙s .G/= Inn.G/ ! N Œ 7! NN  : (3.8)

Due to  the group

G HQ sC1 is not only normal in D Gal.MN sC1 =Q.t//, but
s s
also in
Q  WD Gal.MN sC1 =Q.tQ1 ; : : : ; tQs ; t//; (3.9)
where the tQi denote the elementary symmetric functions in t1 ; : : : ; ts . As in I, (4.8),
we now define an action of Q  on ˙s .G/ from the right via

˙s .G/  Q  ! ˙s .G/; N 7!  :ıN D  ı N 1 N ıN

. ; ı/ with  ı D  . /: (3.10)

Then formula (3.4) carries over to the components of the image  D  ./. As a
consequence the actions of Q  respectively of

Q s WD Gal.MN s .t/=Q.tQ1 ; : : : ; tQs ; t// Š Q  = s (3.11)

N s .G/ are inverse to each other.
on ˙s .G/= Inn.G/ and N

Proposition 3.5. With the action of ıQ 2 Q s on ˙s .G/= Inn.G/ defined in (3.10) we

have
Q1
NN  ıQ D .NN  /ı : (3.12)

The proof is exactly the same as the one for Proposition I.4.2.
3 Galois Descent 199

3.3 The Fixed Field of a Class of Generating Systems

Via the action of Qs on G defined in (3.10) the group Q s acts on the conjugacy
classes of G via the cyclotomic character. More precisely we have the following
analogue of Proposition I.4.3:

Proposition 3.6. According to the decomposition of Q s into a semidirect product

HQ s  Q write ıQ 2 Q s as ˇı with ˇ 2 HQ s . Then ıQ 2 Q s acts on the set of conjugacy
classes Ci of the i -th component of Œ 2 ˙s .G/= Inn.G/ via
Q
Ciı D C.ic.ı/
/ˇ ; (3.13)

with the cyclotomic character c and .i /ˇ WD .i /qs .ˇ/.

Proof. By Theorem 1.6 and Corollary 2.3 the generators ˇj of HQ s act on the gen-
erators i of s and hence according to (3.10) on the generators i of G via

Œ1 ; : : : ; s ˇj
D Œ1 ; : : : ; j 1 ; j j C1 j1 ; j ; j C2 ; : : : ; s :

Consequently Ci D Œi is mapped onto C.i /ˇ under ˇ 2 HQ s . Using Theorem 3.3 we

thus obtain
Q
Ciı D Ciˇ ı D C.iı /ˇ D C.ic.ı/
/ˇ : t
u

If in Proposition 3.6 the element ıQ D ˇı belongs to the subgroup

s WD Gal.MN s .t/=Q.t1 ; : : : ; ts ; t// D = s (3.14)

Q
of Q s , then we even have Ciı D Ci
c.ı/
since Hs D ker.qs /. By (3.13), the stabilizer

C WD fı 2 s j Cı D Cg (3.15)

of the class vector C D .C1 ; : : : ; Cs / under this action of s (compare I, (4.13)) is a

closed subgroup with fixed field

MN s .t/C D QC .t/ (3.16)

where QC is the field introduced in Proposition I.4.4.

The fixed field K under the closed subgroup
ı
 WD fı 2 s j Œ D Œ g (3.17)

for Œ 2 ˙.C/= Inn.G/ then satisfies the analogue of Theorems I.4.5 and I.4.7:

Theorem 3.7. Let G be a finite group and Πa class of generating s-systems of G

with s  3 belonging to the class vector C.
200 III Action of Braids

(a) The fixed field K of Πcontains the field QC .t/ D QC .t1 ; : : : ; ts ; t/ and we
have
ŒK W QC .t/ l.C/: (3.18)
(b) If Z .G/ D 1 then there exists a geometric Galois extension N =K with

Gal.N =K / Š G and MN s N D NN  : (3.19)

Proof. Assertion (a) follows from

.C W  / j˙.C/= Inn.G/j D l.C/:

According to Proposition I.4.6 the field NN  is Galois over K and the corresponding
Galois group acts by inner automorphisms on Gal.NN  =MN s .t// Š G. Thus the field
N in part (b) is obtained as the fixed field of the centralizer of Gal.NN  =MN s .t// in
Gal.NN  =K /. t
u

In the special case l.C/ D 1 we recover from Theorem 3.7 a variant of the Basic
Rigidity Theorem I.4.8 in several variables:

Corollary 3.8. Let G be a finite group with trivial center and with a rigid class vec-
tor C 2 Cl.G/s . Then there exists a geometric Galois extension N=QC .t1 ; : : : ; ts ; t/
with group
Gal.N=QC .t// Š G:
If moreover C is rationally rigid, then we even have QC D Q.

In the next section we deduce the higher dimensional analogue of the Strong
Rigidity Theorem using symmetry groups of C.

3.4 Using the Symmetry Group

As in Section I.4.4, let V denote a symmetry group of the class vector C of G and
CV D fC! j ! 2 V g. Furthermore, for ıQ D ˇı according to Proposition 3.6 let
Q
Q VC WD fıQ 2 Q s j Cı 2 CV g D fıQ 2 Q s j Cc.ı/ 2 CV g: (3.20)

The fixed field of this closed subgroup is the subfield of all V -invariant func-
tions in QVC .t/ D QVC .t1 ; : : : ; ts ; t/. We write tV for a system of basic invariants of
t1 ; : : : ; ts ; t D tsC1 under the action of V Ss (by permutation of the indices), and
then obtain
QV
MN s .t/C D .QVC .t//V D QVC .tV /: (3.21)
Here in general QVC .tV / is not rational but only a unirational function field, i.e., a
subfield of a purely transcendental field with the same field of constants. The fixed
3 Galois Descent 201

field KV of the closed subgroup

ıQ
V WD fıQ 2 Q s j Œ D Œ g (3.22)

of Q s then satisfies the following generalization of Theorem 3.7 with exactly the
same proof:

Theorem 3.9. Let G be a finite group, Πa class of generating s-systems with s  3

of G with class vector C and V a symmetry group of C.
(a) The fixed field KV of Πcontains the field QVC .tV /, and we have

ŒKV W QVC .tV / l.CV /: (3.23)

(b) If Z .G/ D 1 then there exists a geometric Galois extension N =KV with
group
Gal.N =KV / Š G and MN s N D NN  : (3.24)

For the proof of the analogue of the Strong Rigidity Theorem it is easiest to use a
semilinear rationality criterion introduced by Speiser (1919) (see also Borel (1991),
Ch. AG, 14.2). This will be employed several times in what follows.

Proposition 3.10 (Speiser’s Lemma). Let K=k be an algebraic function field of

transcendence degree r and K= N kN an extension of constants with a separable alge-
braic closure kN of k. If K=
N kN possesses a linear subspace UN invariant by  WD
N
Gal.K=K/ N UN / D KN and dim.UN / D r, then the fixed point set U WD UN  is a
with k.
vector space over k with K D k.U /. In particular, K=k is a rational function field
and any basis of the vector space U=k yields a transcendence basis of K=k.

Proof. For u 2 UN the index of the stabilizer u of u in  is finite. The intersection

Q of the -conjugates of u is hence a normal subgroup of finite index in , whose
fixed field in KN resp. kN will be denoted by KQ resp. k.
Q Let 1 D 1; 2 ; : : : ; n be the
Q Q
elements of = and a1 ; : : : ; an a k-basis of k. Then the

X
n
ai j uj 2 UN

vi WD for i D 1; : : : ; n (3.25)
j D1

are -invariant and thus lie in U . From the linear independence of the automor-
phisms j over kQ it follows that the matrix A D .ai j /ij is invertible in kQ nn , with


inverse B D .bhi /hi say. We calculate that

X
n X
n X
n

X
n
b1i vi D b1i ai j uj D ı1j uj D u (3.26)
i D1 i D1 j D1 j D1

Q
with the Kronecker symbol ıij is a k-linear combination of elements of U . This
N N
implies k ˝k U D U and thus K D k.U /. t
u
202 III Action of Braids

The analogue of the Strong Rigidity Theorem is now given by:

Theorem 3.11. Let G be a finite group with trivial center and rigid class vector
C 2 Cl.G/s where s  3. Further let V be a symmetry group of C with the property
that for each ı 2 VC WD s \ Q VC there exists a unique ! 2 V satisfying Cc.ı/ D C! .
Then there exists a function field QVC .v1 ; : : : ; vs ; t/, purely transcendental over QVC ,
and a geometric Galois extension N=QVC .v; t/ with group

Gal.N=QVC .v; t// Š G:

Moreover if C is V -symmetric, then QVC D Q.

Proof. From rigidity of C we obtain that ˙.C/= Inn.G/ consists of a single class of
generating systems Π. This then satisfies

V D fı 2 Q VC j Cı D Cg D Q C  C ;

which first yields QVC .tV / KV QC .t/. Since by assumption for each ı 2 VC
there exists ! 2 V with Cc.ı/ D C! , the extension KV =QVC is regular. More-
over, from the uniqueness of ! it follows that QK N V D Q.t/.
N As V Q VC , any
V N
ı 2  D Gal.Q.t/=K V
 / permutes the transcendence
N
basis t1 ; : : : ; ts of Q.t/= N
Q.t/
L
N
and hence acts on the Q-vector space UN D siD1 QtN i . By Proposition 3.10 the space
UN possesses a basis v1 ; : : : ; vs with vi 2 KV which leads to

KV D QVC .v; t/: (3.27)

From this the assertion follows with Theorem 3.9(b). t

u

In Paragraph 5 the structure of KV is studied in the nonrigid situation. Before,

in the next paragraph we use Theorem 3.11 to construct polynomials with cyclic
Galois groups.
4 Cyclic Polynomials 203

4 Cyclic Polynomials

In this paragraph we treat the higher dimensional analogue of the construction of

cyclic field extensions from Section I.5.1, as an example. With its help it is pos-
sible to compute generating polynomials of geometric cyclic field extensions of
degree n over fields of characteristic not dividing n. These results are complemented
by the construction of polynomials which over any field of characteristic p gener-
ate a cyclic Galois extension of p-power degree. Thus every finite abelian group
possesses G-realizations over every field.

4.1 Cyclic Polynomials in Several Variables

Since for n D 2 the polynomial

f2 .t; X / WD X 2  t 2 k.t/ŒX (4.1)

generates a geometric Z2 -extension over k.t/ for any field k of characteristic dif-
ferent from 2, we will assume until further notice that n > 2.

Proposition 4.1. Let 2 < n 2 IN, Zn D hi a cyclic group of order n, C D .Πi j i 2
.ZZ=nZZ/ / and V WD .ZZ=nZZ/ . Then for the generating system  2 C there exists
a geometric Galois extension N =KV with

Gal.N =KV / Š Zn and KV D Q.v; t/; (4.2)

where v D .v1 ; : : : ; v'.n/ / and t are independent transcendentals over Q.

Proof. To simplify the formulae we set

Ci WD Πi D f i g; and C D .Ci j i 2 .ZZ=nZZ/ / (4.3)

as above. (For groups Zp with p 2 IP this notation coincides with the one used so
far.) Since Zn is abelian, the class vector C is rigid. By the definition I, (4.12), we
have CV D C , hence CV is even V -symmetric with

where c.ı/ 2 .ZZ=nZZ/ D V:

c.ı/
Ciı D Ci

Thus the additional assumption in Theorem 3.11 is satisfied and we have QVC D Q
and KCV D Q.v; t/.
Since cyclic groups have non-trivial center, Theorem 3.11 does not immedi-
ately apply, and it remains to show the existence of a geometric Galois extension
N =KV . But this can easily be verified (compare the 1-dimensional analogue The-
orem I.4.11). As in the proof of the Splitting Theorem 3.3 we obtain from the
decomposition group of an unramified rational point of Q.v; t/ a complement to
204 III Action of Braids

Gal.NN  =Q.t//
N D Zn in Gal.NN  =Q.v; t//. Since by construction this group acts triv-
ially on Zn , it is even a direct complement, whose fixed field N is geometric over
Q.v; t/ and Galois with group Gal.N =Q.v; t// Š Zn . t
u

We proceed to compute generating polynomials for N =Q.v; t/. For this we

start with the extension of constants NQ  WD QC N of N , with QC D Q. n /, over
QC .v; t/ D QC .t/, where following the notation in (4.3) we write t D .: : : ; ti ; : : : ; t j
i 2 .ZZ=nZZ/ /. By Kummer theory NQ  is a subfield of QC .x; t/ with xin D t  ti for
i 2 .ZZ=nZZ/ . The correct subfield can now be determined from the ramification
structure (4.3), taking (3.12) in Proposition 3.5 into account: We have
Y
NQ  D QC .t; y/ with y WD xirn .1= i / ; (4.4)
i 2.ZZ=nZZ/

where rn .1= i / denotes the smallest positive representative of the residue class of
1= i in .ZZ=nZZ/ . By Proposition 4.1 it follows that

Gal.NQ  =KV / Š Gal.NQ  =QC .t//  Gal.QC =Q/ Š Zn  V:

It only remains to find a generating element of NQ  =QC .t/ invariant under V . For
this the Lagrangian resolvents
X
ui WD ij
n yj for i D 1; : : : ; n with (4.5)
j 2.ZZ=nZZ/

Y
yj WD xirn .j= i / for j 2 .ZZ=nZZ/ ; (4.6)
i 2.ZZ=nZZ/

are suitable.

Lemma 4.2. The polynomial

X
n Y
n
hn .X / D .1/i si X ni WD .X  ui / 2 Q.y; t/ŒX ; (4.7)
i D1 i D1

where y D .yj j j 2 .ZZ=nZZ/ / from (4.6), satisfies:

(a) hn .X / has coefficients si 2 ZZŒt .
(b) hn .X / is an Eisenstein polynomial with respect to the numerator divisor of
.t  ti / for every i 2 .ZZ=nZZ/ .

Proof. We denote by

X
n
qm WD um
i for m 2 f1; : : : ; ng (4.8)
i D1
4 Cyclic Polynomials 205

the m-th power sum of the ui from (4.5). According to Newton’s identities

X
m1
qm C .1/i si qmi C .1/m msm D 0 for m D 1; : : : ; n (4.9)
i D1

it suffices for (a) to prove that qm lies in ZZŒt . This follows by an elementary calcu-
lation: We have
X
n X m
X Y
m X
n Y
m
qm D ij
n yj D yjl ijl
n
i D1 j 2.ZZ=nZZ/ j2..ZZ=nZZ/ /m lD1 i D1 lD1

P
for 1
P m n. As niD1 nij equals n for njj and 0 else, only the terms with
m
lD1 jl 0 .mod n/ do not vanish in the above sum. These satisfy

Y
m Y
m Y Y Pm
r .jl = i / lD1 rn .jl = i /
yjl D xi n D xi 2 ZZŒt
lD1 lD1 i 2.ZZ=nZZ/ i 2.ZZ=nZZ/

due to
X
m X
m
rn .jl = i / jl = i 0 .mod n/ for i 2 .ZZ=nZZ/ :
lD1 lD1

This proves (a). Further this implies that xin D t  ti divides qm and according
to (4.9) also sm for m D 1; : : : ; n.
For (b) it remains to prove that .t  ti / divides sn precisely once. Again by (4.9) it
suffices to show the corresponding
Q assertion for qn . The latter follows from the fact
that only for the summand n nlD1 yjl of qn with jl D i for l D 1; : : : ; n we have

Y
n
1X
n
ord.t ti / n yjl D rn .jl = i / D 1: t
u
n
lD1 lD1

By Proposition 4.1 the polynomial hn .X / 2 Q.t/ŒX is not Galois since Q.t/ ¤

KV . Instead we have to consider the minimal polynomial of u1 over KV . To sim-
plify the formulae we will not employ the transcendence basis of KV =Q coming
from Speiser’s Lemma (Proposition 3.10), but

vj WD Vj .; t/=V ./ for j D 1; : : : ; '.n/ (4.10)

with the Vandermonde determinant V ./ for  WD . ni j i 2 .ZZ=nZZ/ / and the deter-
minant Vj .; t/ which is obtained from V ./ by replacing the j -th column by the
transposed of .ti j i 2 .ZZ=nZZ/ / (compare Matzat (1987), Kap. IV, 3.1). Then by
the Cramer’s rule we have the inversion formula

X
'.n/
i.j 1/
ti D n vj for i 2 .ZZ=nZZ/ : (4.11)
j D1
206 III Action of Braids

Substituting ti according to (4.11) in hn .t; X / 2 Q.t/ŒX from Lemma 4.2 we obtain

the polynomial

X
'.n/
i.j 1/
gn .X / D gn .v; t; X / WD hn . n vj ; t; X / 2 Q. n ; v; t/ŒX (4.12)
j D1

with the following properties:

Theorem 4.3. The polynomial gn .X / in (4.12) satisfies:

(a) gn .X / has coefficients in ZZŒv; t .
(b) gn .X / generates a geometric Galois extension Nn =Q.v; t/ with the group
Zn .
(c) The extension Nn . n /=Q. n ; v; t/ is completely ramified in the numerator
divisor Pi of .t  ti / for i 2 .ZZ=nZZ/ and unramified elsewhere.
(d) The denominator divisor P1 of .t/ splits completely in Nn =Q.v; t/.

Proof. Every automorphism ! of Q. n ; t/=Q.v; t/ permutes the ti and the ni for

i 2 .ZZ=nZZ/ in the same way. Obviously these automorphisms can be extended
uniquely to a group of automorphisms !Q of QC .x; t/=Q.v; t/, which permute the
generators xi also like the ni . These extensions !Q act on the yj from (4.6) via
Y Q
!
Y Y
yj!Q D xirn .j= i / D xirln .j= i / D xirn .jl= i / D yjl
i 2.ZZ=nZZ/ i 2.ZZ=nZZ/ i 2.ZZ=nZZ/

with l 2 .ZZ=nZZ/ determined by . ni /! D ni l . Thus the restrictions of the !Q to

NQ  D Q. n ; t; y/ form a group VQ Š V , and the Lagrangian resolvents ui from (4.5)
Q
lie in the fixed field N WD NQ V of VQ . In particular gn .X /, being the minimal poly-
nomial of ui , is contained in QC .t/V ŒX D Q.v; t/ŒX by Lemma 4.2(a) with coef-
ficients in ZZŒv; t by (4.11). Further, since

Gal.NQ  =Q.v; t// Š Gal.NQ  =N /  Gal.NQ  =QC .t// Š V  Zn ;

gn is Galois with group Zn , and the splitting field Nn WD N of gn is regular over

Q.v/. This proves (a) and (b). According to the choice of the ramification points
assertion (c) follows directly from the knowledge of the class vector C in (4.3).
It remains to show (d), which cannot be deduced from the general theory. For this
we replace t by t 0 WD t 1 and use n-th roots xi0 of 1  t 0 ti instead of the generators
xi . Now forming yj0 , u0i , h0n and gn0 according to the formulae (4.5)–(4.7) and (4.12)
with xi replaced by xi0 , then the u0i also generate the Galois extension Nn =Q.v; t/,
and gn0 is the minimal polynomial of the u0i over Q.v; t 0 / D Q.v; t/. Upon special-
ization of t 0 to tN0 D 0 the xi0 become the same n-th root of unity for all i due to the
compatibility with VQ , and we have yNj0 D 1 for j 2 .ZZ=nZZ/ . Thus uN 0n D '.n/ is a
simple root of the specialized polynomial gN n0 . By Theorem I.9.1 the decomposition
groups of the corresponding extension P Q 1 of P1 and hence of all extensions of
P1 onto Nn are trivial. t
u
4 Cyclic Polynomials 207

Corollary 4.4. The assertions of Theorem 4.3 remain true for the residue class poly-
nomial gN n .X / 2 IFp .v; t/ŒX of gn .X / from (4.12) modulo a prime p not dividing n,
with IFp in place of ZZ respectively Q.

Proof. It suffices to show that gN n .X / stays irreducible. With a primitive n-th root of
unity n over IFp we have, using (4.10),

gN n .: : : ; Vj .; t/=V ./; : : : ; t; X / D hN n .X / 2 IFp .t/ŒX

with hn .X / from (4.7). By the proof of Lemma 4.2 the reduction hN n .X / is an

Eisenstein polynomial with respect to the numerator divisor Pi of .t  ti / and thus
irreducible even over IF N p .t/. Since hN n and gN n have the same splitting field over
IF N p .t/, gN n is also irreducible over IF
N p .v; t/ D IF N p .v; t/ and a fortiori over IFp .v; t/.
t
u

Remark. By specializing .t  ti / to ti one obtains from gn resp. gN n the polynomials

given by Dentzer (1995a) (see also Smith (1991) for odd prime powers n).

Since the polynomials gn .X / in '.n/ C 1 parameters from Theorem 4.3 are hard
to compute explicitly, we specialize them to one-variable polynomials in the next
section.

4.2 Cyclic Polynomials in One Variable

Specializing the independent variables v D .v1 ; : : : ; v'.n/ / in gn .v; t; X / from (4.12)

to .0; 1; 0; : : : ; 0/, we obtain a polynomial

fn .t; X / WD gn .0; 1; 0; : : : ; 0; t; X / 2 ZZŒt; X (4.13)

with the following properties (see for example Dentzer (1995a), Thm. 1):

Theorem 4.5. The polynomials fn .t; X / from (4.13) satisfy:

(a) fn .t; X / generates a geometric Galois extension Nn =Q.t/ with group Zn .
(b) The extension Nn . n /=Q. n ; t/ is completely ramified in the numerator divi-
sors Pi of .t  ni / for i 2 .ZZ=nZZ/ and unramified elsewhere.
(c) The denominator divisor P1 of .t/ splits completely in Nn =Q.t/.
(d) If the prime p does not divide n, then (a)–(c) continue to hold for the reduction
fNn .X / 2 IFp Œt; X with IFp in place of Q.

Proof. By Theorem 4.3 it suffices to show that the specialized polynomials fn .X /

and fNn .X / are irreducible and that under the specialization ti goes to ni . The latter
follows immediately from (4.11). The irreducibility of fn and fNn is an easy conse-
quence of the fact that fn and fNn are Eisenstein polynomials over Q.t/
N respectively
N
over IFp .t/ with respect to the numerator divisors Pi of .t  n /.
i
t
u
208 III Action of Braids

Remark. The polynomials fn .t; X / generate the cyclic Galois extension over Q.t/
obtained in Theorem I.5.1 with the choice of ni for i 2 .ZZ=nZZ/ as ramification
N
points over Q.t/.

Numerical evaluation of the formula yields:

Corollary 4.6. With the n-th cyclotomic polynomials n D n .t/ we have:

f3 .t; X / DX 3  33 X  .2t C 1/3 ;

f4 .t; X / DX 4  44 X 2 C 44 ;
f5 .t; X / DX 5  105 X 3  5.4t 2 C 2t  1/5 X 2
5.3t 3 C 6t 2 C 4t C 2/1 5 X  .4t 6 C 6t 5  5t 4  10t 3  10t 2  9t  1/5 :

Remark. A list of further cyclic polynomials up to degree 16 can be found in

Dentzer (1995a) and up to degree 11, as well as for 16, in the tables of the Appendix.

Theorems 4.3 and 4.5 give constructions of G-realizations of the cyclic group Zn
over all fields of characteristic prime to n. In the next section we treat the case of
cyclic groups of p-power order over fields of characteristic p.

4.3 Cyclic Artin-Schreier Towers

Since in contrast to the cyclic Kummer-extensions the Artin-Schreier-extensions

exist already without adjunction of suitable roots of unity, the results of this section
can be read off from the general theory without crossing by a wreath-product.
Let K first be an arbitrary field of characteristic p and W .K/ the ring of Witt
vectors x D .x1 ; : : : ; xn / of length n over K (see Jacobson (1980), Thm. 8.26).
Then cyclic extensions of degree p n over K are classified by the cyclic subgroups
of order p n of the additive group W C .K/ of W .K/ modulo the subgroup fxp  x j
x 2 W .K/g (loc. cit., Thm. 8.31). These are generated by the Witt vectors x with
x1 … fx p  x j x 2 Kg (loc. cit., Thm. 8.32). We can thus state the analogue of
Theorem 4.3:

Theorem 4.7. Let K.t/, t D .t1 ; : : : ; tn /, be a rational function field over a field K
of characteristic p. Then the field

Np;n WD K.t; x/; with xp  x D t; (4.14)

generated by the components xi 2 K.t/ of a solution x of xp  x D t in the Witt ring

W .K.t// satisfies:
(a) Np;n is regular over K and Galois over K.t/ with group Zpn .
(b) Np;n =K.t/ ramifies in the denominator divisors Pi of .ti / with ramification
index p nC1i and is unramified elsewhere.
(c) The numerator divisor P0 of .t1 / splits completely in K.t; x1 /=K.t/.
4 Cyclic Polynomials 209

Proof. The assertion Gal.Np;n =K.t// Š Zpn follows immediately from the above
since t1 … fx p  x j x 2 K.t/g. The regularity of Np;n =K is a trivial consequence
of (b), since a separable extension of constants is unramified.
For part (b) we consider the tower of fields

L0 WD K.t/; Li WD Li 1 .xi / with Ln D Np;n :

Here the Li =Li 1 are Artin-Schreier-extensions

p
Li D Li 1 .xi / with xi  xi D hi .x1 ; : : : ; xi 1 / C ti (4.15)

(Jacobson (1980), Ch. 8.11, Proof of Lemma 2) where hi 2 KŒX1 ; : : : ; Xi 1 has

constant term 0 (loc. cit., Ch. 8.10, (50)). Hence only the denominator divisor P1 of
.t1 / ramifies in L1 =K.t/. But now the inertia group of any extension of P1 to Np;n
is the full group, so P1 is even completely ramified in Np;n =K.t/. In the case i > 1,
(4.15) implies that in Li =Li 1 apart from the denominator divisors of x1 ; : : : ; xi 1
and thus of t1 ; : : : ; ti 1 precisely the denominator divisor Pi of .ti / ramifies, which
is inert in Li 1 =K.t/, and as above this ramifies completely in Np;n =Li 1 .
By the Dedekind Criterion (Corollary I.9.3) the numerator divisor P0 of .t1 /
splits completely in L1 =K.t/. Since its extensions to L1 remain inert in Np;n =L1
the decomposition group of P0 equals Gal.Np;n =L1 / Š Zpn1 . t
u

Remark. The generating polynomials of Np;n =K.t/ are generic polynomials over K
in the sense of Saltman (1982) or Jensen, Ledet and Yui (2002). The same holds for
the polynomials gn .X / in Theorem 4.3 for odd prime powers n (see Smith (1991),
Thm. 5).

The analogue of Theorem 4.5 is obtained by choosing K D IFp .t/ in Theorem 4.7
and specializing .t1 ; : : : ; tn / to .t; 0; : : : ; 0/:

Corollary 4.8. Let x be a solution of xp x D .t; 0; : : : ; 0/ in the Witt ring W .IFp .t//.
Then the field Np;n generated by the components xi 2 IFp .t/ of x over IF.t/ satisfies:
(a) Np;n =IFp .t/ is a geometric Galois extension with group Zpn .
(b) Np;n =IFp .t/ is completely ramified in the denominator divisor P1 of .t/ and
unramified elsewhere.
(c) The numerator divisor P0 of .t/ splits completely in Np;n =IFp .t/.

Proof. As t … fx p  x j x 2 IFp .t/g the specialized field extension Np;n =IFp .t/
possesses the Galois group Zpn . Thus (a) and (b) follow immediately from Theo-
rem 4.7(a) and (b). By Theorem 4.7(c) the divisor P0 splits completely in L1 =K.t/
and the numerator divisor of .x1 / is an extension P.1/0 of P0 . Since the constant
term of h1 in (4.15) vanishes, by the Dedekind Criterion (Corollary I.9.3) P.1/
0 splits
completely in L2 =L1 and the numerator divisor of .x2 / is divisible by an extension
P.2/ .1/ .n/
0 of P0 . Induction then yields an unramified prime divisor P0 of P0 in Np;n
of degree 1 and thus with trivial decomposition group. Hence here P0 even splits
completely in Np;n =K.t/. t
u
210 III Action of Braids

We finish the paragraph by giving the two simplest generating polynomials for
Np;n =IFp .t/.

Corollary 4.9. The fields Np;n =IFp .t/ in Corollary 4.8 for n 2 by (4.15) are gen-
erated by the Artin-Schreier-polynomials
p
fp;1 .t; X1 / DX1  X1  t 2 IFp .t/ŒX1 ;
!
1X
p1
pi p
fp;2 .t; X2 / DX2p  X2  .1/ x ipCpi 2 IFp .t; x1 /ŒX2 ;
p i 1
i D1

with a zero x1 of fp;1 .t; X1 / in IFp .t/.

A generating polynomial of Np;2 =IFp .t/ can be computed from this using the
Buchberger algorithm.

Remark. Using Corollary IV.1.7 from the next Chapter it follows from Theorem 4.5
and Corollary 4.8 that every finite abelian group possesses a G-realization over any
prime field and hence over any field.
5 Rigid Braid Orbits 211

5 Rigid Braid Orbits

In this paragraph we study the question when the fixed field of a class of generat-
ing systems is regular over Q and rational. The first question may be answered by
proving rigidity of the corresponding braid orbit, the second by calculation of braid
orbit genera. As main result we obtain the Rigid Braid Orbit Theorem, which is
then demonstrated on the example of L2 .25/. The last two sections contain a trans-
lation theorem for braid orbits and a higher dimensional analogue of the Extension
Theorem I.8.7.

5.1 The Regularity Criterion

For the action defined in (3.10) of HQ s on G and hence also on ˙s .G/ we have by
Theorem 1.6 with Corollary 2.3:

Proposition 5.1. Let G be a finite group. Then for s  3 the generators ˇi of HQ s act
on ˙s .G/= Inn.G/ via (3.10) as

Œ1 ; : : : ; s ˇi
D Œ1 ; : : : ; i 1 ; i i C1 i1 ; i ; i C2 ; : : : ; s : (5.1)

In particular the pure Hurwitz braid group Hs permutes the classes of generating
systems in ˙.C/= Inn.G/ for each class vector C of G.

Now let V be a symmetry group of C. The orbit of Œ 2 ˙.CV /= Inn.G/ under

the action (5.1) of

HsV WD HQ s \ Q VC D fˇ 2 HQ s j Cˇ 2 CV g (5.2)

will from now on be symbolized by

HsV
B V . / WD Π(resp. B. / for V D 1) (5.3)

and will be called the V -symmetrized braid orbit of Π. The number of such braid
orbits in ˙.CV /= Inn.G/ is denoted by

hV .CV / WD j˙.CV /= Inn.G/HsV j (resp. h.C/ for V D 1) (5.4)

in analogy to I, (6.18) (resp. I, (4.18) in the case V D 1).

For a V -symmetrized braid orbit B  ˙.CV /= Inn.G/ let further

VB WD fı 2 Q VC j Œ ı
2 B for Π2 Bg (5.5)

be the stabilizer of the braid orbit. The corresponding fixed field then satisfies:
212 III Action of Braids

Proposition 5.2. The fixed field of the V -symmetrized braid orbit B D B V . / of

Œ 2 ˙.C/= Inn.G/ of G has the form

MN s .t/B D kV .tV /;

V
(5.6)

where kV denotes the algebraic closure of Q in KV and tV is defined by (3.21).
Proof. Let
HV WD HsV \ V (5.7)
be the stabilizer of Œ in HsV . Then since HV VB the HsV -orbit B V . / of Œ
equals the VB -orbit. Since this implies

ŒKV W MN s .t/B D .VB W V / D .HsV W HV / D ŒQK

N V W MN s .t/Hs ;
V V

the algebraic closure of Q in MN s .t/B coincides with the one in KV . Now VB
V

Q VC , so by (3.21) QVC .tV / is a subfield of MN s .t/B . Consequently we also have

V

kV .tV / MN s .t/B , which, as

V

N MN s .t/VB D MN s .t/HsV D Q.t

Q N V/

proves (5.6). t
u
Proposition 5.2 furnishes an interpretation for the extension of constants in
KV =QVC .tV / as well as for the purely geometric part. The degree of the latter obvi-
ously equals the orbit length jB V . /j. To obtain a good estimate for the extension
of constants we first have to introduce some more notation. In analogy to I, (6.17),
for an open subgroup U HsV let

hVU .C/ WD jfB V . / j  2 ˙.CV /; HV D U ˛ for an ˛ 2 Aut.HsV /gj (5.8)

be the number of braid orbits of those Œ 2 ˙.CV /= Inn.G/ whose stabilizer HV
in HsV coincides with U up to an automorphism. Those braid orbits B V . / with
hVH V .C/ D 1 are distinguished by their stabilizers and are called rigid HsV -orbits.

Furthermore, C is called a HsV -rigid class vector if ˙.CV /= Inn.G/ consists of a
single HsV -orbit.
Theorem 5.3 (Regularity Theorem). Let G be a finite group, C 2 Cl.G/s with s  3
a class vector of G and V a symmetry group of C. Also for  2 ˙.C/ let KV denote
the fixed field of V and kV the algebraic closure of Q in KV . Then we have

ŒKV W kV .tV / D jB V . /j and ŒkV W QVC hVH V .C/: (5.9)



In particular KV =QVC is regular if B V . / is a HsV -rigid braid orbit.

Proof. By what precedes it remains to prove the inequality in (5.9). By the class
equation ŒkV W QVC equals the number of braid orbits B V . / in ˙.CV /= Inn.G/
5 Rigid Braid Orbits 213

permuted by Q VC . Now let ı 2 Q VC , B D B V . / be a braid orbit in ˙.CV /= Inn.G/

and B ı D B V . ı / the image of B under ı. Then the stabilizer of Œ ı in HsV satis-
fies HVı D .HV /ı by Proposition 3.5. As HsV G Q VC , it is the image of HV under
1

an automorphism of HsV . Thus using the definition (5.8) of hVU .C/ for U D HV we
obtain
ŒkV W QVC D jfB V . ı / j ı 2 Q VC gj hVU .C/: t
u
In contrast to the situation in Theorem I.6.8 the extension K =QVC .tV / of degree

ŒKV W QVC .tV / lUV .C/ for U D HV

is here decomposed into a purely geometric part of degree jB V . /j and an extension

of constants kV =QVC , whose degree is bounded by the (generally much smaller)
number of braid orbits hVU .C/. A disadvantage now comes from the fact that in
general KV is not a rational function field any more. In the next section we give
explicit criteria for rationality of KV in the case V D 1.

5.2 Braid Orbit Genera

As above let G be a finite group, C 2 Cl.G/s a class vector of G with s  3 compo-

nents,  2 ˙.C/ and B. / D Œ Hs the orbit of Œ under the pure braid group Hs .
Also let

Bs . / WD Πs1
and Bj . / WD Bj C1 . /j 1 for 1 j s 1 (5.10)

be the orbit of Bj C1 . / under j 1 WD j.s/ 1 = j

.s/
(with j.s/ from Corollary 2.9),
where the elements of the orbit Bj . / are orbits of type Bj C1 . Denote the number
of cycles of ˇij j.s/ on Bj . / with cij . Then

j 1
1X
gj . / WD 1  jBj . /j C .jBj . /j  cij / (5.11)
2
i D1

is called the j -th braid orbit genus of  .

In analogy to the previous section we call a braid orbit Bj . / rigid if there are no
further braid orbits Bj ./ in Bj C1 . /Hj such that the stabilizers of the elements of
Bj . / and Bj ./ differ in Hj only by an automorphism of Hj .
Proposition 5.4. If Bj . / is a rigid j 1 -orbit, then

jBi . /j D 1 and gi . / D 0 for i D 1; : : : ; j  1: (5.12)

Proof. If Bj . / is a rigid j 1 -orbit in Bj C1 . /Hj , then it stays invariant under

all factor groups i 1 D i.s/1 = i
.s/
for i < j . Consequently we have jBi . /j D 1
and gi . / D 0 for i < j . t
u
214 III Action of Braids

Remark. Since 2 D H3 Š Z2 acts trivially on all 3 -orbits inside a H4 -orbit, as

H4 Š 3 H3 , (5.12) certainly holds for all i 3.

The case s D 4 also possesses another particularity:

Proposition 5.5. In the case s D 4 we have g4 . / D g4 . ˇ / for all ˇ 2 HQ 4 .

1
Proof. It is easily verified that for s D 4 the pairs ˇ14 and ˇ23 , ˇ24 and ˇ14 ˇ13 ˇ14 ,
and ˇ34 and ˇ12 all have the same action on ˙4 .G/ since they only differ by the
central involution 4 D .ˇ1 ˇ2 /3 . Thus the permutation types of fˇ14 ; ˇ24 ; ˇ34 g coin-
cide with those of fˇ1! 4! ; ˇ2! 4! ; ˇ3! 4! g for all ! 2 S4 and hence also with those
of fˇ14 ; ˇ24 ; ˇ34 g for all ˇ 2 HQ 4 .
ˇ ˇ ˇ
t
u

The corresponding assertion for s  5 is false in general. The next theorem con-
tains the fundamental characterization of the braid orbit genera by intermediate
fields of K =k .

Theorem 5.6. Let G be a finite group, C 2 Cl.G/s a class vector of G with s  3

and  2 ˙.C/. Also let Kj be the algebraic closure of Q.t1 ; : : : ; tj / inside K . Then
for 1 j s the field extensions Kj =Kj 1 .tj / satisfy

ŒKj W Kj 1 .tj / D jBj . /j and g.Kj =Kj 1 / D gj . /: (5.13)

Proof. The proof uses descending induction on j . First, clearly we have K D

Ks .t/ with t D tsC1 and Ks1 D Ks \ MN s1 . Thus application of Theorem 2.8
yields

ŒKs W Ks1 .ts / D ŒKN s W MN s1 .ts / D . s1 W. s1 \ H // D jBs . /j

with KN s WD MN s1 Ks . Similarly for 1 < j < s with KN j WD MN j 1 Kj and

j WD Gal.MN s =KN j .tj C1 ; : : : ; ts // D .s/

j 1 \ H j
.s/

we obtain the sequence of equalities

ŒKj W Kj 1 .tj / D ŒKN j W MN j 1 .tj / D . .s/

j 1 = j
.s/
W j = .s/
j / D jBj . /j:

Here we have K0 D k by Theorem 5.3. This proves the first part of the assertion.
Let f .X / be the minimal polynomial of a primitive element of KN j =MN j 1 .tj /.
The Galois group of f .X / is equivalent as permutation group on the zeroes of f .X /
to the permutation representation of j.s/ on the cosets of j in j.s/
1 , respectively
of j.s/
1 = j
.s/
on the cosets of j = j.s/ , and hence to the permutation representa-
tion of j 1 on Bj . /. By Theorem 2.10, and since KN j MN j only the valuation
ideals Dij generated by the .ti  tj / for 1 i < j ramify in KN j =MN j 1 .tj /. Here
by Theorem 2.4 the inertia group of a suitable extension of Dij onto MN j is gener-
ated by the element ˇij j 2 Gal.MN j =MN j 1 .tj //. According to Theorem I.9.1 the
.s/

ramification type of Dij in KN j =MN j 1 .tj / and hence the degree of the different of
5 Rigid Braid Orbits 215

KN j =MN j 1 .tj / may be read off from the cycle decomposition of ˇij
.s/
j on Bj . /;
it equals
X
j 1 X
cij
X
j 1
.eijk  1/ D .jBj . /j  cij /;
i D1 kD1 i D1
.s/
where cij denotes the number of cycles of ˇij j on Bj . / and eijk for 1 k cij
their lengths. The genus formula then yields
j 1
1X
g.KN j =MN j 1 / D 1  ŒKN j W MN j 1 .tj / C .jBj . /j  cij / D gj . /:
2
i D1

Since the genus of Kj =Kj 1 does not change by the extension of constants to
MN j 1 , the second part of the assertion follows. u
t
In the further text the field Ks in Theorem 5.6 will play a special role and then
be denoted by K , so that K D K .t/.

5.3 A Rationality Criterion for the Pure Braid Group

With the help of Theorem 5.6 it is now easy to formulate sufficient criteria for
the rationality of K =k . One is given by the following oddness condition, whose
validity implies the existence of a divisor of odd degree in Kj =Kj 1 :
.s/
(Oj ) In the action of ˇij j on Bj . / there occurs for some i < j a cycle length
eijk an odd number of times.
With this we get:
Theorem 5.7 (Braid Orbit Theorem). Let G be a finite group with trivial center,
C 2 Cl.G/s with s  3,  2 ˙.C/ and B. / the Hs -orbit of Œ . For j D 4; : : : ; s
assume further that gj . / D 0 and that the oddness condition (Oj ) is satisfied.
Then K D k .u1 ; : : : ; us ; t/ is a rational function field over k and there exists a
geometric Galois extension N =k .u; t/ with

Gal.N =k .u; t// Š G and MN s N D NN  : (5.14)

Proof. By the Remark following Proposition 5.4 we have K3 D k .t1 ; t2 ; t3 /. For

j with 4 j s let now Kj 1 =k be a rational function field. Then by (5.13) the
genus g.Kj =Kj 1 / D gj . / equals zero. Since only prime divisors of equal degree
are conjugate in KN j =Kj , the oddness condition (Oj ) together with Theorem I.9.1
imply that the valuation Dij ideal generated by .ti  tj / possesses an extension onto
Kj with odd residue class degree. Thus Kj =Kj 1 and hence by induction Kj =k
is a rational function field. The first part of the assertion then follows from the fact
that K D K .t/ with K D Ks according to the notation introduced at the end of
the last section. The second part is obtained using Theorem 3.9(b) for V D 1. t
u
216 III Action of Braids

Under the additional assumption that B. / is a rigid braid orbit, the Regularity
Theorem 5.3 implies:

Corollary 5.8 (Rigid Braid Orbit Theorem). If in Theorem 5.7 the braid orbit
B. / is additionally supposed rigid, then there even exists a rational function field
QC .u; t/=QC and a geometric Galois extension N =QC .u; t/ with

Gal.N =QC .u; t// Š G and MN s N D NN  : (5.15)

In particular we have QC D Q if C is a rational class vector.

This result is now employed to construct G-realizations over Q for the simple
group L2 .25/. This is the smallest among the groups L2 .q/ with q ¤ 16 for which
such a realization could not be found with the rigidity theorems in the first chapter.

Example 5.1 (Przywara (1991)). Let G D P†L2 .25/ be the group obtained from
the simple group L2 .25/ by extension with the field automorphism. Then the class
vector C D .2A; 2C; 2D; 12A/ of G in Atlas notation is rational with l.C/ D 12 and
h.C/ D 1, i.e., B D ˙.C/= Inn.G/ forms a rigid braid orbit. The permutation types
of ˇi 4 on B may be calculated as .32 ; 22 ; 12 / for ˇ14 and .5; 3; 22 / for ˇ24 and ˇ34 .
Thus we have g4 . / D 0 for Π2 B. Since clearly the oddness condition (O4 ) is
also satisfied, by the Rigid Braid Orbit Theorem K D Q.u1 ; : : : ; u4 ; t/ is a rational
function field over Q and there exists a geometric Galois extension N=Q.u; t/ with
the group P†L2 .25/.
The fixed field K0 of the subgroup G 0 Š L2 .25/ of G is of degree 2 over Q.u; t/
and is regular over Q.u/. Since 2A  G 0 and 12A  G 0 only the numerator divi-
sors Pi of .t  ti / for i D 2; 3 are ramified in K0 =Q.u; t/. By the genus formula
K0 =Q.u/ has genus zero and is even a rational function field, since for example the
prime divisor of P2 in K0 =Q.u; t/ is of degree 1. Thus we have K0 D Q.u; ´/, say,
and Gal.N=Q.u; ´// yields a G-realization of L2 .25/ over Q. t
u

Example 5.1 has been generalized by Shiina (2003b) to groups L2 .p 2 / with

p 6 ˙1 .mod 24/ and by Dettweiler (2003) for S2n .p 2 / with the same p. A more
far reaching result by Thompson and Völklein (1998) is presented here in Theo-
rem 10.9 as an application of the Katz algorithm.

5.4 Rational Translation of Braid Orbits

Next we want to investigate the behavior of braid orbits under translation. Because
of the number of generators of the acting braid group a braid orbit can only be
expected to be the full image of another braid orbit, if the translation maps a class
vector of length s into one of length at most s. For s  4 this is only possible
under rather restrictive conditions. Let NN =MN s .t/ be a Galois extension with group
G contained in MN sC1 .t/ parametrized by Πand with class vector C, i.e., NN D NN 
with  2 ˙.C/= Inn.G/, and L= N MN s .t/ the rational field extension chosen to achieve
5 Rigid Braid Orbits 217

the translation. Then the number of ramified points of NN L=

N L,
N multiplied by their
residue class degrees, satisfy

ns  2.n  1/  i s

by the Hurwitz relative genus formula, where i denotes the number of prime divi-
sors of LN ramified in L= N MN s .t/ but unramified in NN L= N L.
N Since i 2.n  1/ this
forces s D 4. But then we have i D 2.n  1/ and all ramification orders in L= N MN s .t/
are equal to 2. If we further assume that the class vector C does not only consist of
involution classes (which rules out only fields NN of genus g.NN =MN s / D 1), then we
N MN s .t/ is a Z2 -translation of type .1A; 1A; 2A; 2A/,
either have .n; i / D .2; 2/ and L=
N N
or .n; i / D .4; 6/ and L=Ms .t/ is an E4 -translation of type .1A; 2A; 2A; 2A/. If
without loss of generality we restrict ourselves to primitive field extensions, then a
single possibility for the translation field remains. This proves part (a) of the follow-
ing theorem:
Theorem 5.9. (a) For a finite Galois extension NN =MN s .t/ inside MN sC1 =MN s .t/ of
genus g > 1 with at least s  4 ramification points there exists only one possi-
ble primitive translation into a Galois extension with at most s ramification points,
namely a Z2 -translation 'Z2 with the class vector .1A; 1A; 2A; 2A/.
(b) The translation 'Z2 from (a) maps a generating 4-system  belonging to the
class vector C D .C1 ; C2 ; C3 ; C4 / with two involution classes C3 and C4 to
 21
Q WD 'Z2 . / D .1 ; 1 3 Q
; 2 ; 23 / 2 ˙N .C/ (5.16)

with CQ D .C1 ; C1 ; C2 ; C2 /. If G possesses no subgroup of index 2 then Q is a gener-

ating system of G.
(c) If moreover G has trivial center, then the translation map
Q Inn.G/
'Z2 W ˙.C/= Inn.G/ ! ˙.C/=

is injective and maps H4Z2 -orbits onto full H4Z2 -orbits for

H4Z2 WD fˇ 2 HQ 4 j q.ˇ/ 2 Z2 D h.34/ig D H4 [ H4 ˇ3 :

Proof. By the preceding considerations, only (b) and (c) remain to be proved.
As i D 2 the original class vector C contains two classes of involutions, which
by the choice of the translation class vector .1A; 1A; 2A; 2A/ are the classes C3 and
C4 . The translation formula can now be obtained as in the proof of Theorem I.6.3:
The Galois group Gal.MN sC1 =L/,N which in accordance with Theorem I.6.3 will be
denoted by Z2 E 4 , is generated by the generators of inertia groups 1 ; 2 ; 32 ; 42 ,
together with two elements conjugate to i , i D 1; 2, by non-trivial coset represen-
 1  1  1
tatives of Z2 in 4 . For example, with the choice Q1 D 1 3 2 , Q2 D 2 3 , the
generating system .1 ; Q1 ; 2 ; Q2 ; 32 ; 42 / satisfies the product relation, and we have

 1 21  1
Z2 D h1 ; 1 3 ; 2 ; 2 3 ; 32 ; 42 j 1    42 D 1iO; (5.17)
218 III Action of Braids

since Z2 is a free profinite group of rank 5. Application of the canonical epimor-
phism
W 4 ! G; i 7! i 2 Ci ;
yields the translation formula (5.16) due to 32 D 42 D 1. By construction Q gen-
erates a subgroup of G isomorphic to Gal.NN L= N L/.
N Since this can have index at
most 2 in G, the additional assumption forces G D hQ i.
If the center of G is trivial (or has odd order), then the translation map 'Z2 is
injective by Theorem I.7.4. It remains to verify the compatibility with the H4Z2 -
orbits. For this we compute the 'Z2 -image of the H4Z2 -orbit B Z2 . / of  . Since
the action of the central element 4 D .ˇ1 ˇ2 /3 2 H4 on B Z2 . / is trivial, it suffices
to determine the image of the generators of

3[ 3 ˇ3 D hˇ14 ; ˇ24 ; ˇ3 i:

This equals
1 1
'Z2 . ˇ14 / D 'Z2 . /ˇ2 ˇ3 ˇ2 ˇ3 ˇ2 ;
2 2
'Z2 . ˇ24 / D 'Z2 . /ˇ2 ˇ3 ˇ2 ;
1
'Z2 . ˇ3 / D 'Z2 . /ˇ2 ˇ34 ˇ2 : (5.18)
2
Hence the subgroup U of HQ 4 generated by ˇ2ˇ3 ˇ2 , ˇ3 2 , ˇ34
ˇ ˇ2
and the central element
1
4 acts on the image 'Z2 .B Z2 . //. Now U ˇ2 obviously contains the elements ˇ34
ˇ
and .ˇ2 3 /2 D .ˇ22 /ˇ3 D ˇ24 , and thus by the sphere relation (1.13) also ˇ14 , so we
1 1
conclude H4 U ˇ2 and thus U ˇ2 D hH4 ; ˇ2ˇ3 ; ˇ3ˇ2 i. The relation

ˇ32 ˇ2ˇ3 .ˇ3ˇ2 /1 ˇ22 D 1

1
finally implies U ˇ2 D hH4 ; ˇ3ˇ2 i, whence the result

ˇ2
U D hH4ˇ2 ; ˇ3 2 i D hH4 ; ˇ3 i D H4Z2 : t
u

Example 5.2. For the group A5 the sets ˙.Ci /= Inn.G/ for C1 D .3A; 2A; 2A; 2A/,
C2 D .3A; 3A; 2A; 2A/ and C3 D .3A; 3A; 3A; 3A/ each form a single H4 -orbit of
length 18. The above theorem now shows that the orbits B2 and B3 can be obtained
by a rational translation from B1 . But this translation does not preserve the braid
orbit genera: We have g4 . 1 / D 1, g4 . 2 / D 2, g4 . 3 / D 1 for  i 2 Bi . t
u

Example 5.3. For G D S5 the set B WD ˙.C/= Inn.G/ for C D .6A; 6A; 2A; 2A/
forms a single H4 -orbit of length 24. Although the translation map 'Z2 to ˙.C/= Q
Inn.G/ with C Q D .6A; 6A; 6A; 6A/ is injective, since A5 contains no elements of
order 6, 'Z2 .B/ splits into two H4 -orbits BQ 1 , BQ2 , of lengths 12, whose union is a
single H4Z2 -orbit. This example shows that the translation map in Theorem 5.9(c)
does not in general preserve H4 -orbits. On the other hand, this can be used for a
5 Rigid Braid Orbits 219

reduction of braid orbit genera. In this case we have g. / D 2 and g.Q / D 0 for
 2 B, Q 2 BQ1 [ BQ 2 . t
u

In view of Theorem 5.9 it is natural to ask under which conditions a braid orbit
B D B. / is mapped at least into a single braid orbit under a rational translation.
As the Example 5.3 shows, this will in general require additional assumptions which
guarantee that the field K and the field L generated over Q.t/N by the ramification
points of L= N
N MN s .t/ are linearly disjoint over Q.t/. With a specialization argument it
can be shown that this hypothesis is already sufficient.

5.5 Groups of Automorphisms as Galois Groups

At the end of this paragraph we prove a higher-dimensional analogue of the Exten-

sion Theorem I.8.7. Let V be a symmetry group of the class vector C 2 Cl.G/s with
s  3. According to (5.2) and (5.3)

B V . / WD ΠHsV
with HsV WD fˇ 2 HQ s j Cˇ 2 CV g

denotes the V -symmetric braid orbit of Π. For an intermediate group A between

Inn.G/ and Aut.G/ set, in analogy to the notation in Chapter I.8,

V A WD fı 2 Q VC j  ıA D  A g and HVA WD HQ s \ V A (5.19)

with Q VC from (3.20) and

VA V
N VA :
KVA WD MN s .t/ and KN VA WD MN s .t/  A D QK
H


(5.20)

Then we first have:

Theorem 5.10. Let G be a finite group, C 2 Cl.G/s with s  3, V a symmetry group

of C and B V . / the V -symmetrized braid orbit of  2 ˙.CV /. Furthermore let A
be an intermediate group between Inn.G/ and Aut.G/ acting on B V . /. Then there
exists a geometric Galois extension NA =KVA with

Gal.NA =KVA / Š A and Gal.NA =KV / Š Inn.G/: (5.21)

If here B V . / is rigid in ˙.CV /= Inn.G/, then KVA =QVC is regular.

Proof. Let NN  =MN s .t/ be the Galois extension with Gal.NN  =MN s .t// Š G belong-
ing to Πby the Hurwitz classification (Theorem 3.4). The fixed field of the center
Z .G/ herein is denoted by NN A . By definition NN  is Galois over KVA in (5.20),
and every automorphism of NN  =KVA acts on the normal subgroup Gal.NN  =MN s .t//
as an element of A. The fixed field NA of the centralizer of Gal.NN  =MN s .t// in
220 III Action of Braids

Gal.NN  =KVA / thus has the following properties: it satisfies MN s NA D NN A and
Gal.NA =KVA / is isomorphic to a subgroup of A.
By the assumptions, A and hence also AN WD A= Inn.G/ acts on B V . / and
decomposes B V . / according to Corollary I.8.4 (see I, (8.12)) into orbits of length
N which implies
jAj,
N
ŒKV W KVA D .V A W V / D jAj
with V from (3.22). From

Gal.NA =KV / Š Gal.NN A =MN s .t// Š Inn.G/

we thus immediately obtain (5.21). Furthermore from the action of AN on B V . /

N which first proves ŒKV W K VA D ŒKN V W KN VA and thus
follows .HVA W HV / D jAj,  

N A W KN VA :
ŒNA W KVA D ŒQN 

The latter implies that NA =KVA contains no extension of constants and hence is a
geometric Galois extension. t
u

Remark. In the case Z .G/ D 1 the field NA coincides with N in Theorem 3.9.

The question of rationality of KVA =QVC can at least in the case V D 1 be decided
with the methods of Paragraph 4, by investigating the coarser class of generating
systems  A instead of the braid orbit B. / of Π. This is worked out in Matzat
(1991a), 5.2. Another special case has been proved by Völklein.

Theorem 5.11 (Völklein (1992b)). Let G be a finite group, C 2 Cl.G/s with s  3,

V D Sym.C/, W D f! 2 V j C! D Cg and A WD f˛ 2 Aut.G/ j C˛ D Cg. If then for
a  2 ˙.C/ the braid orbit B W . / is rigid in ˙.C/= Inn.G/and A acts transitively
on it, then there exists a geometric Galois extension NA =QVC .v1 ; : : : ; vs ; t/ over the
purely transcendental field QVC .v; t/ with

Gal.NA =QVC .v; t// Š A: (5.22)

If here C is V -symmetric, then we have QVC D Q.

Proof. By assumption the braid orbit B W . / is rigid in ˙.C/= Inn.G/ and thus
B V . / is rigid in ˙.CV /= Inn.G/. From Theorem 5.10 the existence of a geometric
Galois extension NA =KVA with

Gal.NA =KVA / Š A and kV A D QVC

now follows. It remains to show the rationality of KVA =QVC . Since A acts transitively
on B W . /, at least we have

QVC .tV / KVA KWA D QW W

C .t /
5 Rigid Braid Orbits 221

(compare Theorem 3.9 for A D Inn.G/). Here, since CW D C, the vector tW D

.tQ1 ; : : : ; tQs ; t/ consists of t and the elementary symmetric functions tQj of those ti
with equal conjugacy class Ci . By the main theorem on symmetric polynomials
at least the field KWA =QW C is a purely transcendental function field. By definition
of W for each ı 2 Gal.Q.t N W /=QVC .tW // there exists exactly one !N 2 V =W with
Cc.ı/ D C!N , which then permutes the systems of equal conjugacy classes in C. As
in the proof of Theorem 3.11 this implies that the field extension KVA =QVC is reg-
ular and we have QK N VA D Q.tN W /. Since by construction V =W permutes the class

vectors belonging to CV the group Gal.Q.t N W /=QVC .tV // permutes the elementary
Q Q
symmetric polynomials t1 ; : : : ; ts and leaves t invariant. Since the same holds for the
subgroup Gal.Q.t N W /=K VA /, we obtain with Speiser’s Lemma (Proposition 3.10)

as in the proof of Theorem 3.11 transcendentals v1 ; : : : ; vs independent over QVC .t/
with KVA D QVC .v1 ; : : : ; vs ; t/ and moreover

Gal.QVC .v; t/=QVC .tV // Š V =W: (5.23)

t
u

Unfortunately the condition that A acts transitively on B W . / is rather restric-

tive. Nevertheless Völklein (1992b, 1993) was able to find interesting examples by
the investigation of centerless groups with very many outer automorphisms. Here we
will use Theorems 5.10 and 5.11 in Section IV.4.3, to construct higher dimensional
GA-realizations fulfilling some additional rationality condition.
222 III Action of Braids

6 Unramified Rational Places

This paragraph provides a close connection with Chapter I by showing that the exis-
tence of unramified rational places of the fixed field of a class of generating systems
of G in k.t/ and the existence of geometric Galois extensions over k.t/ are intercon-
nected via specialization. Using a theorem of Conway and Parker on the existence
of rigid braid orbits this implies in particular the solution of the inverse problem of
Galois theory over k.t/ for PAC-fields k of characteristic zero.

6.1 Specialization of the Fundamental Group

In the subsequent sections we will several times use the following specialization
theorem for the algebraic fundamental group (see Grothendieck (1971), Exp. X,
Cor. 3.9, and compare with Theorem I.10.6 for the mixed characteristic case).
Theorem 6.1 (Grothendieck (1971)). Let KN be an algebraically closed field of char-
N a place of KN onto a subfield k.
N ! IP1 .k/
acteristic 0 and } W IP1 .K/ N Then for any
1 N N
finite }-stable subset S of IP .K/ with S WD }.S / we have
alg
1 .IP1 .K/ N n SN/:
N n S / Š  alg .IP1 .k/ (6.1)
1
alg
The isomorphism is uniquely determined by } up to an inner automorphism of 1 .
Sketch of proof. Since kN KN we may consider SN as a subset of IP1 .K/.
N Since S
N
and S have equal cardinality (compare Theorem I.2.2 for subfields of C) we see
alg
1 .IP1 .K/ N n SN/:
N n S / Š  .IP1 .k/ alg
1

This implies the assertion with the proof of Theorem I.2.2 if there we replace C
N
by K. u
t
In what follows let }a denote the place defined by
N 1 ; : : : ; ts ; t// ! IP1 .Q.t//;
}a W IP1 .Q.t N
N s;
.t1 ; : : : ; ts / 7! a D .a1 ; : : : ; as / 2 IP1 .Q/ (6.2)
oa the corresponding valuation ring and pa its valuation ideal. The latter is unrami-
N
fied in MN s .t/=Q.t/ when a 2 IP1 .Q/ N s . Furthermore, let pO a denote an extension of pa
to MsC1 and pN a its restriction to MN s .t/. Then the specialization theorem for the
N
fundamental group leads to:
Proposition 6.2. For a 2 IP1 .Q/ N s , s  3, we have MN s .t/pN a D Q.t/,
N  WD
and MN sC1
MN sC1 pO a is the maximal field extension of Q.t/N unramified outside the set S of numer-
ator divisors of .t  ai / for i D 1; : : : ; s. In particular we have
 =Q.t//
Gal.MN sC1 N Š D h1 ; : : : ; s j 1    s D 1iO: (6.3)
s
6 Unramified Rational Places 223

Proof. Since MN s pN a =Q
N is algebraic, we have MN s pN a D Q
N and MN s .t/pN a D Q.t/.
N Now
let T be the set of numerator divisors of .t  ti / for i D 1; : : : ; s in Ms .t/=MN s . As in
N
Theorem 2.10 let further MO s denote the algebraic closure of MN s in an algebraically
closed hull MO sC1 of MN sC1 and TO the set of extensions of T to MO s .t/. By Theo-
rem 2.10 the Galois extension MO s MN sC1 =MO s .t/ is maximal (inside MO sC1 ) unrami-
fied outside TO and we have canonically

Gal.MN sC1 =MN s .t// Š Gal.MO s MN sC1 =MO s .t// D 1 .IP1 .MO s / n T /
alg

with T D ft1 ; : : : ; ts g. By Theorem 6.1 the extension of pO a to MO s MN sC1 induces an

isomorphism of fundamental groups. With S D fa1 ; : : : ; as g we thus have

1 .IP1 .MO s / n T / Š 1 .IP1 .Q/

N n S / D Gal.MN S =Q.t//:
N
alg alg

Composition of the two isomorphisms yields (6.3), using Theorem I.2.2. t

u

As a consequence we obtain in particular (compare also Corollary I.10.7):

Corollary 6.3. Let G be a finite group. Then the Hurwitz classifications Ns and NS
of ˙s .G/= Aut.G/ (and also ˙s .G/= Inn.G/) are compatible on Ns .G/ resp. NS .G/
with the specialization homomorphism of the fundamental group, i.e., with the
restrictions pQ a of pO a to NN  we have

NN  WD NN  pQ a D MN Sker. / : (6.4)

6.2 The Specialization Theorem

Now let K V be the fixed field of  introduced in Theorem 3.9 and kK V for k V
k Q N be an extension of constants of K V . Also let } be a place of IP1 .kK V /
into IP1 .k.t//, whose extensions }N to IP1 .MN s .t// are unramified over Q.t/,
N i.e.,
1 N 
for which there exists a 2 IP .Q/s with }N D }N a . Such a place is abbreviated as an
unramified rational place of kKV =k.t/. Let k D Gal.Q=k/ N and S . k / denote the
image of the permutation representation of k on the set S of numerator divisors of
N
.t  ai / in Q.t/= N (compare I, (4.26)).
Q

Theorem 6.4. Let G be a finite group with trivial center, C 2 Cl.G/s with s  3, V
a symmetry group of C and  2 ˙.C/. Then we have:
(a) If kK V =k.t/ for k V k Q N possesses an unramified rational place, then
there exists a geometric Galois extension N=k.t/ satisfying

Gal.N=k.t// Š G; N D NN  2 NS .G/
QN and S . k/ V: (6.5)
N D NN 
(b) If there exists a geometric Galois extension N=k.t/ with group G, QN
and S . k / V , then we have k  k  , and kK  =k.t/ possesses an unramified
V V

rational place.
224 III Action of Braids

Proof. By Theorem 3.9(b) there exists a geometric Galois extension N =K V with

Gal.N =K V / Š G and MN s N D NN  :

Hence we also have Gal.kN =kK V / Š G. Now let } be an unramified rational

place of kK V =k.t/, p the corresponding valuation ideal, pO D pO a an extension of
p to MN sC1 with corresponding restrictions pQ to kN and pN to kKV respectively,
and N WD kN pQ the residue class field of kN modulo pQ . Since the extension of
constants with QN commutes with the residue class map and pN has residue degree 1
N N V
in Ms .t/=Q.t /, we have
N  pQ D NN  pQ D NN  :
N D QN
QN

The sequence of inequalities

N W Q.t/
jGj D ŒkN W kK V  ŒN W k.t/  ŒQN N D jGj

now shows that N=k.t/ is geometric and Galois of degree jGj, which implies
that Gal.N=k.t// Š G. Finally from k.tV /p D k.t/ we obtain aV 2 k and hence
S . k / V .
Now let N=k.t/ be a geometric Galois extension with Gal.N=k.t// Š G and
N D NN  . To this corresponds via the Hurwitz classification the ordered set S D
QN
fP1 ; : : : ; Ps g of unramified prime divisors of NN  =Q.t/,
N where Pi is the numerator
divisor of .t  ai /. For S we have by assumption S . k / V . Here V is a symmetry
group of C D .Œ1 ; : : : ; Œs / since the inertia groups of Pi 2 S conjugate over k.t/
coincide and k acts on ˙s .G/= Inn.G/ via the cyclotomic character.
By Corollary 6.3 the field NN  is the image of NN  2 Ns .G/ under the restriction
of a place }O of IP1 .MN sC1 / in IP1 .Q.t//
N O i / D ai for i D 1; : : : ; s. (Different
with }.t
such places differ by an element of HQ s .) Now let ı 2 Gal.NN  =N / and ıQ an extension
Q
of ı to NN  =k.t/. Since then the residue class fields of NN  and NN ı by the respec-

tive restrictions of the valuation ideal pO of }O both yield NN  D .NN  /ı , we obtain
NN ı D NN  from Corollary 6.3. Since ı centralizes the group Gal.NN  =Q.t//,
Q N ıQ also
Q
centralizes the group Gal.NN  =MN s .t//. This implies Œ ı D Œ which finally proves
ıN WD ıj
Q N
Ms .t / 2  and hence k  k  .
V V

Now let N be the field introduced in (3.24) with Gal.N =K V / Š G. Then ı, Q

belonging to the centralizer of Gal.NN  =MN s .t//, lies in Gal.NN  =kN / and thus
leaves kN pointwise fixed. Hence also N  WD kN pQ and K  WD kK V pN are left
pointwise fixed by ı. This forces K  D k.t/ and N  D N , since N  =k.t/ is Galois
with QNN  D NN  and N is uniquely determined as the fixed field of the centralizer of
Gal.NN  =Q.t//
N in Gal.NN  =k.t//. In particular the restriction } of }O to IP1 .kK V / is
an unramified rational place of kK V =k.t/. t
u

Composition of Theorems 5.3 and 6.4 yields:

Theorem 6.5 (Specialization Theorem). Let G be a finite group with trivial center,
C 2 Cl.G/s a class vector with s  3 and V a symmetry group of C. Furthermore,
6 Unramified Rational Places 225

assume that the braid orbit B V . / of  2 ˙.C/ is rigid under HsV . Then there
exists a geometric Galois extension N=QVC .t/ with

Gal.N=QVC .t// Š G N D NN 
and QN (6.6)

if and only if K V =QVC .t/ possesses an unramified rational place.

Here we have QVC D Q if C is V -symmetric.
The main problem with the application of Theorem 6.5 is in general the difficult
proof of existence of unramified rational places. This handicap vanishes by defini-
tion if the field of constants k is assumed to be pseudo algebraically closed (see
Section 6.4 for the definition). Then it only remains to ensure the existence of rigid
braid orbits, which is essentially possible by a group theoretic result of Conway and
Parker.

6.3 The Theorem of Conway and Parker

For the proof of the theorem of Conway and Parker a number of preparatory remarks
have to be made. For a finite group G let W .G/ denote the semigroup generated by
the words in G, where the word consisting of  2 G is denoted by w for better
distinction. The natural mapping

e W W .G/ ! G; w1    ws 7! 1    s ; (6.7)

is called the evaluation function. Furthermore, for w D w1    ws 2 W .G/ let
Q
wN WD B.w/ Q 1    ws /
D B.w (6.8)

denote the orbit of w under the full Artin braid group BQ s via the action on the
indices (1.5) and
H.G/ WD fwN j w 2 W .G/g (6.9)
the set of such orbits.
Proposition 6.6. (a) Equipped with the product wN 1  wN 2 WD w1 w2 the set H.G/
becomes a semigroup, with
wN  wN D wN wN   : (6.10)
(b) The evaluation function e is a class function on H.G/. Every wN 2 H.G/ with
N D 1 lies in the center of H.G/.
e.w/
Proof. Obviously the product wN 1 wN 2 is well defined and using the braid action of ˇ1
we obtain
wN  wN D w w D w w  D wN wN   :
Since by (1.5) the action of ˇi on G s leaves invariant the product of the components,
N Now let wN 2 H.G/ with e.w/
e is constant on the classes w. N D 1, say wN D wN 1 s .
226 III Action of Braids

Then iterating (6.10) we find for  2 G

wN wN 1 s D wN wN 1    wN s D wN 1    wN s wN D wN 1 s wN

since 1    s D 1. Thus wN commutes with all generators wN of W .G/ and therefore

lies in the center of H.G/. t
u
By Proposition 6.6(b) the braid orbit
Q
uN WD wN o./ 2 H.G/ (6.11)
2Gnf1g

lies in the center of H.G/. We define a congruence relation on H.G/ by wN 1 wN 2

if there exist ni 2 IN with uN n1 wN 1 D uN n2 wN 2 . Then the factor group
GQ WD H.G/= (6.12)
with multiplication defined on the representatives satisfies:
Proposition 6.7. (a) GQ is a central group extension of G. The canonical epimor-
phism ' W GQ ! G is induced by the evaluation function e.
Q and we have './
(b) The congruence classes Q of wN  generate G, Q D  and
Q Q D Q Q Q : (6.13)

Proof. Let  2 G n f1g. Since e.wN o./ / D 1 the element wN o./ lies in Z .H.G// and
may hence be taken as the first factor of u. N Thus there exists wN 2 H.G/ with uN D
N As above iteration of (6.10) yields wN  wN D wN wN  since e.u/
wN  w. N D 1. Consequently
Q Now GQ is generated by the elements Q for  2 G n f1g and thus
Q is invertible in G.
forms a group. Further by Proposition 6.6(b) it is a central extension of G, for which
the canonical epimorphism ' is induced by the evaluation function. This proves (a),
and (b) follows by Proposition 6.6(a). t
u
For G a finite group let F WD F .G/ D h j  2 Gi be the free group over G and

W F ! G;  7! ;

the corresponding epimorphism with kernel R WD ker. /. The Schur multiplier of

G is defined as the quotient

M.G/ WD .ŒF; F \ R/=ŒF; R

(see Huppert (1967), Kap. V, Satz 23.5). If the Schur multiplier M.G/ is generated
by commutators, then from (6.13) we obtain an interesting lifting property of the
evaluation function. Here for  D .1 ; : : : ; s / 2 G s , resp. wN  D wN 1    wN s 2 H.G/
we use the notation

s. / WD .mC . //C 2Cl.G/ with mC . / D jfj j j 2 C gj (6.14)

for the vector of conjugacy class multiplicities (the shape function).

6 Unramified Rational Places 227

Proposition 6.8. Let G be a finite group whose Schur multiplier M.G/ is generated
by commutators, and GQ a central extension of G. Further let fQ j  2 Gg be a
system of representatives of G in GQ satisfying (6.13). Then for all  ;  2 G s with
s. / D s./ and 1    s D 1    s we have Q 1    Q s D Q1    Qs .

Proof. Let F; R; be defined for G as above, and Q the epimorphism

Q W F ! G;
Q  7! Q with ı Q D ;

where  denotes the canonical epimorphism from GQ to G. Then since Q .R/

ker./ the group Q .R/ is a subgroup of Z .G/, Q which shows that Q .ŒF; R / D 1.
By assumption the Schur multiplier M.G/ of G is generated by commutators,
hence modulo ŒF; R by elements Œ; ı 2 ŒF; F \ R. Since the commutator of  WD
. Q . // and  WD . Q .ı// in G is trivial, and hence  D , it follows from (6.13)
Q whence ŒQ ; Q D 1. Consequently we have Q .ŒF; F \ R/ D 1, and Q
that Q Q D Q ,
induces an epimorphism
N W FN WD F=.ŒF; F \ R/ ! GQ with N .N / D :
Q

For RN WD R=.ŒF; F \ R/ we thus have on the one hand RN D ker. ı N / and hence
N D ker./, and on the other ŒFN ; FN \ RN D 1, which forces ŒG;
N .R/ Q GQ \ N .R/
N D 1.
Q Q Q
Hence .ŒG; G / is isomorphic to a subgroup of G, and G may be embedded into
Q G;
the direct product of G with A WD G=ΠQ GQ , where the embedding is given by

 W GQ ! G  A; Q 7! .; ˛ / with Q GQ
˛ WD Q ŒG;

for the system of representatives fQ j  2 Gg. According to (6.13) any ;  2 G

satisfy
Q GQ D Q Q ŒG;
˛  D Q ŒG; Q GQ D Œ Q GQ D ˛ ;
Q G;
and ˛ only depends on the conjugacy class of . The assertion now follows from
the equality
Q
s Q
s Q
s Q
s Q
s Q
s Q
s
Q i D .i ; ˛i / D . i ; ˛i / D . i ; ˛ i/D Qi : t
u
i D1 i D1 i D1 i D1 i D1 i D1 i D1

In the following lemma we collect some final prerequisites for the proof of the
theorem of Conway and Parker.

Lemma 6.9. Let G be a finite group.

(a) If 1 ; 2 2 G are conjugate elements of G of order n D o.i / , and if G is
generated by 1 ; : : : ; s , then we have

wN n1 wN 1    wN s D wN n2 wN 1    wN s :

N vN 2 H.G/ satisfy s.w/

(b) If w; N  s.vN u/
N then there exists xN 2 H.G/ with wN D vN x.
N
228 III Action of Braids

(c) There exists m 2 IN such that for all i  m left multiplication

uN W H.G/i ! H.G/i C1 ; wN 7! uN w;
N

is bijective on H.G/i WD fwN 2 H.G/ j s.w/

N D s.uN i /g.
Proof. By Proposition 6.6(b) we have wN n1 2 Z .H.G//. Hence for  WD 1 we have

wN n wN 1    wN s DwN 1    wN j 1 wN n wN j    wN s
DwN 1    wN j wN nj wN j C1    wN s D wN nj wN 1    wN s

which proves (a) by induction since G D h1 ; : : : ; s i.

For the proof of (b) first let vN D wN  with  2 G, n WD o./ and C WD Π.
Then if wN D wN 1    wN s and s.w/ N  s.wN  u/, N there exist more than njC j indices j
with j 2 C , and thus at least one  2 C occurs more than n times among the
j . For this  we have that wN D wN n wN 1    wN r with i 2 G and 1 D  by (6.10).
Since Œ1 ; : : : ; Œs covers the non-trivial conjugacy classes of G, the same holds for
Œ1 ; : : : ; Œr , proving G D h1 ; : : : ; r i (see Fried and Jarden (1986), Lemma 12.4).
So by (a) we deduce

wN D wN n wN 1    wN r D wN n wN 1    wN r D wN  xN

for a suitable xN 2 H.G/. This proves (b) by induction on the length of v.N
From (b) it follows that for the choice of vN D uN left multiplication by uN from
H.G/i to H.G/i C1 is surjective for all i  1. Since all the H.G/i are finite sets,
there exists m 2 IN such that this map is even bijective for all i  m. This completes
the proof. t
u
After these preparations the proof of the theorem of Conway and Parker is easily
completed (see also Fried and Völklein (1991), Appendix).
Theorem 6.10 (Conway and Parker). Let G be a finite group whose multiplier
M.G/ is generated by commutators. Furthermore let C 2 Cl.G/s be a class vector
of G containing each nontrivial class of G sufficiently often. Then all  2 ˙.CSs /
lie in a single orbit under the full Artin braid group BQ s .
Proof. Let  ;  2 G s with s. / D s./ and 1    s D 1    s and wN 1 WD wN 1    wN s ,
wN 2 WD wN 1    wN s . By Propositions 6.7 and 6.8 this implies that Q 1    Q s D Q1    Qs .
Hence there exist ni 2 IN with uN n1 wN 1 D uN n2 wN 2 , where since s.wN 1 / D s.wN 2 / we
even have n1 D n2 DW n. By Lemma 6.9(b) there exist vN 2 H.G/ and xN i 2 H.G/m
with wN 1 D vN xN 1 and wN 2 D vN xN 2 . Since GQ forms a group, there also exist l 2 IN0 and
yN 2 H.G/ with yN vN D uN l .
Thus from uN n wN 1 D uN n wN 2 it follows that uN n vN xN 1 D uN n vN xN 2 and after multiplication
by yN from the left also uN nCl xN 1 D uN nCl xN 2 . Now xN i 2 H.G/m , so by Lemma 6.9(c)
this forces xN 1 D xN 2 . Hence we have

wN 1 D vN xN 1 D vN xN 2 D wN 2 ;

i.e.,  and  lie in the same HQ s -orbit. t

u
6 Unramified Rational Places 229

Remark. In the proof of Theorem 6.10 the assumption e. / D 1 was not used.

Corollary 6.11. Under the hypothesis of Theorem 6.10, ˙.CSs /= Inn.G/ is a single
orbit under the full Hurwitz braid group HQ s . In particular C is then an HsV -rigid
class vector with respect to the full symmetry group V of C.

Proof. This result follows immediately from Theorem 6.10 since the Artin braid
ˇ1    ˇs ˇs    ˇ1 acts trivially on ˙s .G/= Inn.G/. t
u

We are now in a position to give the solution by Fried and Völklein (1991) of the
inverse problem of Galois theory over k.t/ for pseudo algebraically closed fields k.

6.4 The Inverse Galois Problem over PAC-Fields

A field k is called pseudo algebraically closed or PAC for short if every absolutely
irreducible algebraic variety X defined over k possesses a k-rational point, and
hence the corresponding function field k.X /=k possesses a rational place. The set
X .k/ of k-rational points of X then lies dense in the Zariski topology of X (see
Fried and Jarden (1986), Prop. 10.1), and the field k.X / possesses rational places
outside of any proper subvariety of X . Perhaps the most p interesting non-trivial
example of a PAC-field known at present p is the field Q tr
. 1/ generated over Q by
all totally real algebraic numbers and 1 (following from Pop (1996), Thm. S).
Now let K V denote the algebraic closure of Q.t1 ; : : : ; ts /V in K V ; then we
have K V D K V .t/. By the above, the regular field extensions kK V =kk V resp.
kK V =kk V .t/ obtained by extension of constants by a PAC-field k then possess
unramified rational places (in the sense of Section 6.2). Furthermore, we have
kk V D k if the class vector C of  is V -symmetric and HsV -rigid. Thus all assump-
tions of Theorem 6.4(a) are satisfied for kK V =k.t/ if only the center of G is trivial.
It remains to show how this restriction on the center and the additional hypothesis
on the multiplier in the theorem of Conway and Parker can be overcome.

Proposition 6.12. Let G be a finite group. Then we have:

(a) The Schur multiplier M.H / of any representation group H of G is generated
by commutators.
(b) G is a factor group of a finite group H with Z .H / D 1, whose Schur multi-
plier M.H / is generated by commutators.

Proof. Let MG Š M.G/ denote the kernel of the surjection H ! G. Then clearly
MG H 0 \ Z .H /. For a representation group R of H let MR Š M.H / denote the
kernel of R ! H satisfying MR R0 \ Z .R/. Let RN WD R=ŒR; R . Then MN R WD
MR =ŒR; R contains no nontrivial commutators in RN and MN R RN 0 \ Z .R/. N The
N N
surjection R ! H induces a map ' W R ! H with kernel MR . From MG H 0 we
get UN WD ' 1 .MG / RN 0 MN R D RN 0 . Clearly R=
N UN Š G. Furthermore, ŒR;
N UN MN R .
N N N
Since MR contains no nontrivial commutators we even have ŒR; U D 1. Therefore
230 III Action of Braids

RN is a central extension of G with kernel UN RN 0 . It follows that jUN j jM.G/j,

hence MN R D 1 and MR D ŒR; R , proving (a).
By (a), replacing G by a representation group if necessary, we may assume that
the Schur multiplier of G is generated by commutators. Write m D jGj and choose
S to be a nonabelian simple group with trivial multiplier (S D L2 .8/ will do, see
Huppert (1967), Satz 25.7). Then clearly the regular wreath product H D S o G has
trivial center. Any central extension of S splits since M.S / D 1. By induction, the
same holds for every central extension of S m . Thus every representation group of
H D S m G contains a normal subgroup isomorphic to S m such that the quotient
is a representation group of G. Therefore M.H / Š M.G/ is generated by commu-
tators. t
u

With this supplement to Theorem 6.10 the Specialization Theorem 6.5 leads to:

Theorem 6.13 (Fried and Völklein (1991)). Let k be a PAC-field of characteristic 0.

Then for every finite group G there exists a geometric Galois extension N=k.t/ with
group Gal.N=k.t// Š G.

Proof. Let G be a finite group. By Proposition 6.12(b) G is the factor group of a

finite group H with trivial center and whose multiplier is generated by commutators.
If we now take a class vector C 2 Cl.H /s containing each class sufficiently often and
the same number of times, then C is V -symmetric with respect to the full symmetry
group V and by Corollary 6.11 also HsV -rigid. Hence for  2 ˙.C/ due to k V D
QVC D Q we certainly have kk V D k. Since k is pseudo algebraically closed and
thus kK V =k.t/ possesses unramified rational places, by Theorem 6.4(a) there exists
a geometric Galois extension N=k.t/ with Gal.N=k.t// Š G. If now U is a normal
subgroup of H with H=U Š G, then the result follows by passage from N to the
fixed field N U . t
u

With the Hilbert irreducibility theorem this implies the following:

Corollary 6.14. Over a Hilbertian PAC-field of characteristic 0 every finite group

occurs as Galois group.

The structure of the absolute Galois group k of such a Hilbertian PAC-field k

will be completely determined in IV.3.3. Using rigid analytic patching methods the
results of Theorem 6.13 and Corollary 6.14 can be generalized to arbitrary charac-
teristic, see Theorem VI.4.8 and Corollary VI.4.9.
7 Braids and Geometric Automorphisms 231

7 Braids and Geometric Automorphisms

To facilitate the search for rational places, we here specialize the fixed fields of
classes of generating systems to function fields in two variables. This then opens
the possibility of replacing braid orbit genera by the usually smaller symmetrized
braid orbit genera obtained by using geometric automorphisms. In the case s D 4
we moreover give explicit formulae for these. In analogy to the Twisted Rigidity
Theorem I.6.10 this finally leads to a Twisted Braid Orbit Theorem, which will sub-
sequently be employed for the realization of the Mathieu group M24 as geometric
Galois group over Q.t/.

7.1 Specialization to Two Variables

In this paragraph in analogy to (6.2) we let

(
0 1 N 1 N
N  ;
.t1 ; : : : ; ts1 / 7! .a1 ; : : : ; as1 / 2 IP1 .Q/
}a W IP .Q.t/ ! IP .Q.u; t//; s1
.ts ; tsC1 / 7! .u; t/;
(7.1)
be an unramified place of Q.t/ N according to Section 5.2 into the field of rational
functions in u and t over Q, N }N a0 an extension of }a0 to IP1 .MN s .t//, }O a0 a further
1 N
extension to IP .MsC1 / and p0a resp. pN 0a , pO 0a the corresponding valuation ideals with
residue class fields MN s0 .t/ WD MN s .t/pN 0a resp. MN sC1
0
WD MN sC1 pO 0a . Then in analogy to
Proposition 6.2 we have the result:
Proposition 7.1. For s  4 the residue class fields MN i0 satisfy:
N
(a) MN s0 =Q.u/ is a maximal Galois extension unramified outside the set T D
fQ1 ; : : : ; Qs1 g of numerator divisors Qi of .u  ai / with

Gal.MN s0 =Q.u//
N Š Gal.MN s =MN s1 .ts // Š s1 : (7.2)

(b) MN sC1 0
=MN s0 .t/ is a maximal Galois extension unramified outside the set S0 D
fP1 ; : : : ; Ps g of numerator divisors Pi of .t  ai / for i D 1; : : : ; s  1 and Ps of
.t  u/ with
Gal.MN sC1
0
=MN s0 .t// Š Gal.MN sC1 =MN s .t// Š s : (7.3)
(c) MN sC1
0 N
=Q.u; t/ is Galois with
0
HsC1 WD Gal.MN sC1
0 N
=Q.u; t// Š Gal.MN sC1 =MN s1 .u; t// Š s  s1 : (7.4)

Proof. Since MN s1

0 N 1 ; : : : ; as1 / is algebraic, we have MN 0 D Q.
=Q.a s1
N From this,
(a) and (b) follow as in the proof of Proposition 6.2 from Theorem 2.10 using The-
orem 6.1. With these results (c) is obvious. In fact, according to Corollary 2.9 we
have
0 .sC1/
HsC1 Š s1 Š s  s1 : t
u
232 III Action of Braids

By Proposition 7.1 we thus obtain a Hurwitz classification compatible with

N 0s .G/ of intermediate fields of
Ns from (3.6) and NS from I, (4.4), for the set N
N 0 N 0
MsC1 =Ms .t/ with Galois group isomorphic to G. In fact, with the place given by

N
}as W IP1 .Q.u; N
t// ! IP1 .Q.t//; u 7! as N n fa1 ; : : : ; as1 g; (7.5)
with as 2 Q

and its extension }O as onto IP1 .MN sC1

0
/ and the set S of numerator divisors of .t  ai /
for i D 1; : : : ; s we obtain:

Corollary 7.2. The specialization at the restriction to IP1 .MN sC1

0
/ of the place }O a0
from (7.1) induces a bijection
N 0s .G/;
N0s W ˙s .G/= Aut.G/ ! N  Aut.G/ 7! NN 0 WD NN  pQ 0a ; (7.6)

with pQ 0a WD pO 0a \ NN  and

}O a0 ı Ns D N0s and }O as ı N0s D NS : (7.7)

Proof. For suitable extensions of }a from (6.2), }a0 and }as we have }O a D }O as ı }O a0 .
Thus the assertions result from Corollary 6.3. t
u

Remark. The Hurwitz classification (7.6) may also be refined to a mapping from
N 0s .G/ without destroying the identity (7.7).
˙s .G/= Inn.G/ onto N

Supplementing (7.4) we will in future use the notations

Hs0 WD Gal.MN s0 .t/=Q.u;

N t// and 0s WD Gal.MN s0 .t/=Q.u; t//: (7.8)

7.2 Action of Geometric Automorphisms

Now let G be a finite group, C 2 Cl.G/s a class vector of G and V a symmetry

group of C. Furthermore, for the set S0 in Proposition 7.1(b) let

VS0 WD fı 2 Aut.MN s0 .t/=Q/ j S0 .ı/ 2 V g; HSV0 WD Aut.MN s0 .t/=Q/

N \ V0 ; (7.9)
S

where S0 .ı/ is the permutation representation of ı on S0 . (This notation in particular

implies the assumptions that ı permutes the set S0 and fixes Ps .) In analogy to
Section I.6.3, we now call S0 a V -configuration, if

S0 .0s / V and HSV0 =Hs0 Š V: (7.10)

Thus the elements ı of 0s resp. of VS0 can be extended to automorphisms ıN of

MN sC1
0
and hence naturally act according to (7.6) and (3.10) on ˙s .G/= Inn.G/.
Correspondingly in this paragraph for  2 ˙.C/ we will write

V WD fı 2 VS0 j Œ ı
D Πg and HV WD HSV0 \ V : (7.11)
7 Braids and Geometric Automorphisms 233

In accordance with the previous use of the rigidity notion we now call the HSV0 -orbit
B D BSV0 . / rigid in ˙.CV /= Inn.G/ if in ˙.CV /= Inn.G/ there exists no further
HSV0 -orbit whose stabilizer differs from that of B only by automorphisms of HSV0 .
Then let
KV WD MN s0 .t/ and KV WD .MN s0 / :
V V
(7.12)

Theorem 7.3. Let G be a finite group, C 2 Cl.G/s , s  4, with symmetry group V

and S0 a V -configuration. Then for  2 ˙.C/ we have:
(a) KV =KV is a rational function field, say KV D KV .tQ/.
(b) If the HSV0 . /-orbit BSV0 . / of Œ is rigid in ˙.CV /= Inn.G/, then the field
extension KV =QVC is regular.

Proof. By assumption ı 2 V permutes the set S0 of prime divisors of MN s0 .t/=MN s0

in Proposition 7.1(b) and hence in particular maps the field of constants MN s0 onto
itself. Thus KV D KV \ MN s0 is the fixed field of the ı 2 V restricted to MN s0 . Now
MN s0 KV , being a subfield of MN s0 .t/, by Lüroth’s theorem has genus zero over MN s0 .
The same then also applies to KV =KV , since in characteristic zero the genus does
not change under extension of constants. Now Ps is the only prime divisor in S0 not
corresponding to an algebraic place, so it has to remain invariant under V . Thus Ps
does not split in MN s0 .t/=KV and therefore possesses residue degree 1 in KV =KV .
Consequently KV =KV is even a rational function field, which proves (a).
If the HSV0 -orbit BSV0 . / is rigid in ˙.C/= Inn.G/, then it remains invariant under
the full group

VS0 .QVC / WD fı 2 Aut.MN s0 .t/=QVC / j S0 .ı/ 2 V g D hHSV0 ; Gal.MN s0 .t/=QVC .u; t//i

which then also proves (b). t

u

The following proposition now shows a connection to the geometric automor-

phisms investigated in Chapter I.6.

Proposition 7.4. Under the assumptions of Theorem 7.3 for KNV WD QK

N V we
have:
(a) The field extension KN =KNV is Galois with

WN WD Gal.KN =KNV / Š Gal.KN  =KN V / Š W V: (7.13)

Here W D V if BSV0 . / consists of a single Hs0 -orbit.

N
(b) The restrictions Q of all elements N 2 WN to Q.u/ form a group of geometric
automorphisms WQ isomorphic to W , which are determined by their action on the
numerator divisors Qi of .u  ai / for i D 1; : : : ; s  1. Here the Qi are permuted
by Q in the same way as the numerator divisors Pi of .t  ai / by the  2 HV with
jKN D .N
(c) Each Q 2 WQ possesses a unique extension onto Q.u; N t/ leaving invariant the
numerator divisor of .t  u/. The extensions N of Q onto KN  D KN .t/ with this
property thus only differ by an automorphism of KN  =Q.u;N t/.
234 III Action of Braids

Proof. As the kernel of the permutation representation S0 W HV ! V , H is normal

in HV with factor group W V . Thus we first see

Gal.KN  =KN V / Š HV =H Š W V:

Now let  2 HSV0 with jMN s0 D 1. As Ps D Ps there exists a v 2 MN s0 with

t  u D .t  u/ D v.t  u/;

which implies v D 1 and hence t D t resp.  D 1. Consequently the restriction of

HSV0 onto MN s0 is faithful and we have

Gal.KN =KNV / Š Gal.KN  =KN V /;

which proves (7.13). If BSV0 . / consists of a single Hs0 -orbit, then by (7.10) we
further have
.H 0 W H /
.HV W H / D .HSV0 W Hs0 / Vs D jV j
.HS0 W HV /
and so W D V . This completes the proof of (a).
Now let Pi denote the numerator divisors of .t  ai / resp. .t  u/ in KN  =KN .
Then for each N 2 WN there exists ! 2 W satisfying PiN D Pi ! . For 1 i < j s 1
we obtain from this the equations

t N  u N D v.t  u/; t N  ai D vi .t  ai ! /; t N  aj D vj .t  aj ! /;

with v; vi ; vj 2 KN . From this results

ai  aj N
v D vi D vj D 2 Q;
ai !  aj !

hence the restrictions of N onto Q.t/N N

and also onto Q.u/ are automorphisms. In
N N
particular the restrictions Q of N 2 W onto Q.u/ form a group of geometric auto-
morphisms in the sense of Chapter I.6. As

.u  ai / Q D u N  ai D t N  v.t  u/  ai D v.u  ai ! /;

they permute the numerator divisors Qi of .u  ai / also via QiQ D Qi ! , and since
s  4 they are even uniquely determined by this by Proposition I.6.1. This proves (b).
N
Part (c) of the assertion follows from the fact that an extension of Q onto Q.u; t/
is already determined by the image of t. t
u
7 Braids and Geometric Automorphisms 235

7.3 Symmetrized Braid Orbit Genera

By Theorem 7.3 the function field KV =QVC is rational only if the function field in
one variable KV =QVC is rational. If kV denotes the precise field of constants of
KV , then the genus of KV =kV resp. of KNV =Q
N is called the s-th V -symmetrized
braid orbit genus of  :

gsV . / WD g.KV =kV / D g.KNV =Q/:

N (7.14)

This only depends on the HSV0 -orbit BSV0 . /, since the fields KNV are conjugate over
Q
N
.MN s0 /HS0 D Q.u/
V
W
.
The symmetrized braid orbit genus gsV . / is in general considerably smaller than
the usual braid orbit genus gs . / in (5.11). Namely, we have:

Proposition 7.5. Under the assumptions of Theorem 7.3 and with the group W of
Proposition 7.4(a) we have:

.a/ gs1 . / D g.KN =Q/

N D gs . / with gs . / as in (5.11); (7.15)
1
.b/ gs . /  1 D jW j.gsV . /  1/ C deg.D.KN =KNV //: (7.16)
2
Proof. The place }N a0 is inert and genus preserving in MN s =MN s1 .ts / by Proposi-
tion 7.1(a). Using the fields Kj from Theorem 5.6 we get from MN s1 Ks and MN s1
the residue class fields KN and QN respectively and thus

gs . / D g.Ks =Ks1 / D g.MN s1 Ks =MN s1 / D g.KN =Q/

N D g.K =k / D gs1 . /:

By Proposition 7.4(a) the field extension KN =KNV is Galois with group WN isomor-
phic to W . The equation in (b) now follows from the Hurwitz genus formula. t
u

Remark. Since the groups V  W in Proposition 7.4(a) satisfy

gsV . / D gsW . /; (7.17)

we may if necessary assume V D W for the computation of gsV . /.

In the following, we illustrate on the example of s D 4 how the V -symmetrized

braid orbit genera introduced above may be calculated explicitly. Here we can
restrict ourselves to V D Z2 and V D Z3 , since by Theorem I.6.5 these are the only
groups of geometric automorphisms V S4 with fixed points. The next proposition
is only for the purpose of preparation. As already announced in Section 2.1, we here
denote the discrete full Hurwitz braid group by HQ s_ , and consequently

HL sV WD fˇ 2 HQ s_ j q.ˇ/ 2 V g:

Proposition 7.6. For V D Z2 , respectively V D Z3 , two elements of finite order in

HL 4V =Z .HL 4 / are conjugate if and only if their canonical images in V coincide.
236 III Action of Braids

Proof. By Proposition 1.9 we have

HQ 4_ =Z .HQ 4_ / Š Z22  PSL2 .ZZ/ Š E4 .Z2 Z3 /

with the free product Z2 Z3 (see Lyndon and Schupp (1977), p.25) and

E4 D hˇ1 ˇ31 ; .ˇ1 ˇ2 ˇ3 /2 i=h.ˇ1 ˇ2 ˇ3 /4 i:

Since for V D Z2 and V D Z3 the subgroups HL 4V =Z .HL 4 / possess elements of

order 2 respectively 3, namely  WD ˇ1 ˇ14 Z .HL 4 / respectively  WD ˇ1 ˇ2 Z .HL 4 /,
the isomorphism HL 4 =Z .HL 4 / Š G3 proved in Theorem 1.6 implies

HL 4V =Z .HL 4 / Š G3 V: (7.18)

Since here G3 is a free group (of rank 2), the elements of finite order ˇQ WD
ˇZ .HL 4 / 2 HL 4V =Z .HL 4 / certainly satisfy o.ˇ/ Q D o.q.ˇ//
Q with q.ˇ/
Q WD q.ˇ/.
Q Q
Now first let V D Z3 and ˇ; ˇ 2 HL 4 =Z .HL 4 / be elements of order 3 with
0 V

Q D q.ˇQ 0 /. Then their images ˇ;

q.ˇ/ N ˇN 0 in PSL2 .ZZ/ are conjugate, so that there
exists an N 2 PSL2 .ZZ/ with ˇN N D ˇN 0 . As q.E4 / D h.12/.34/; .13/.24/i, there exists
a preimage Q 2 HQ 4_ =Z .HQ 4_ / with

Q Š S4 =q.E4 /:
Q 2 NS4 .hq.ˇ/i/
q./

Consequently there exists Q 0 2 E4 with ˇQ Q D Q 0 ˇQ 0 . This then satisfies q.Q 0 / D 1 and

so Q 0 D 1. Thus ˇQ and ˇQ 0 are conjugate in HL 4 =Z .HL 4 / and hence in particular in
HL 4V =Z .HL 4 /.
The assertion can be verified in a similar manner for elements of order 2 (for this,
see also Gillette and Van Buskirk (1968), Thm. 4.17). t
u

For V S4 and a V -configuration S0 , the preceding proposition and the isomor-

phisms
HSV0 Š .HL 4V =Z .HL 4 //OŠ 3 V (7.19)
following from (7.18) and Proposition 2.6(b) immediately imply:

Corollary 7.7. For s D 4 two closed elements (generators of closed subgroups) of

finite order in HSV0 are conjugate if and only if their canonical images in V coincide.

In the following, let S0 be a V -configuration for V D Z2 respectively V D Z3

and
N u/
Q. Q WD .MN s0 /HS0
V
N
Q.u/ (7.20)
be the fixed field of HSV0 , TQ WD fQ Q 1; : : : ; Q
Q sQ g the set of prime divisors of Q.
N u/
Q ram-
ified in MN s =Q.
0 N u/, Q Q
Q and ˇ1 ; : : : ; ˇsQ generators of the corresponding inertia groups.
Being elements of HSV0 these act on B WD BSV0 . / (compare Theorem 7.3). The cor-
responding permutation representation of ˇQi will be denoted by B .ˇQi /.
7 Braids and Geometric Automorphisms 237

Theorem 7.8. Let G be a finite group, C 2 Cl.G/4 , V a symmetry group of C, S0 

IP.MN 40 .t/=MN 40 / a V -configuration and B WD BSV0 . /  ˙.CV / a HSV0 -orbit. Then
we have
1X
3
gsV . / D 1  jBj C .jBj  ci /; (7.21)
2
i D1

where the ci count the number of cycles in B .ˇQi /. Furthermore we have:

(a) B .ˇQ1 /D B .ˇ14 /, B .ˇQ2 /D B .ˇ1 /, B .ˇQ3 /D B .ˇ1 ˇ14 / for V D h.12/i,
(b) B .ˇQ1 / D B .ˇ14 /, B .ˇQ2 / D B .ˇQ3 / D B .ˇ1 ˇ2 / for V D h.123/i.
Proof. Obviously exactly three prime divisors Q Q i ramify in MN 0 =Q.
N u/.
Q The permu-
4
tation representation of HS0 on B is equivalent to the coset representation of HSV0
V

on HV D Gal.MN 40 =KNV /, thus the cycle lengths eij of ˇQi with say j D 1; : : : ; ri
coincide by Theorem I.9.1 with the ramification indices of the prime divisors of Q Qi
in KNV =Q. Q As we obtain the degree of the different of KNV =Q.
N u/. N u/
Q by addition of
the ramification indices eij diminished by 1, this gives

X
3 X
ri
X
3
deg.D.KNV =Q.
N u///
Q D .eij  1/ D .jBj  ci /:
i D1 j D1 i D1

Thus (7.21) follows from the Hurwitz genus formula for KNV =Q. N u/.
Q There remains
Q
the task of determining the permutation representations B .ˇi / for the groups Z2
and Z3 .
Therefore first let V D h.12/i, VQ the group of geometric automorphisms induced
N N u/ N VQ
by V on Q.u/ by Proposition 7.4 and Q. Q WD Q.u/ . Then obviously Q1 and Q2
N
are conjugate in Q.u/= Q. Q while Q3 is ramified. So in KNV =Q.
N u/, N u/
Q at most the
prime divisors of
Q 1 D Q1 Q2 ; Q
Q Q 2 D Q2 ; Q Q 3 D Q2
3 4

are ramified, with a certain prime divisor Q4 of Q.u/=N N Hence the generators of
Q.
N 0 N
inertia groups over Q1 in Gal.M4 =Q.u// Q also generate inertia groups over Q Q 1 . If
0 0
ˇ14 denotes the image of ˇ14 in H4 , then by Theorem 2.4 and Proposition 7.1 the
element ˇQ1 WD ˇ14 0
is a generator of the inertia group of an extension of Q Q 1 in
MN 4 =Q.u/.
0 N Since the permutation representation of ˇ14 0
on B coincides with that of
ˇ14 , we certainly have B .ˇQ1 / D B .ˇ14 /.
Next we determine the permutation action on B of a generating element ˇQ3 of the
inertia group of an extension of Q Q 3 onto MN 0 . Since Q4 is unramified in MN 0 =Q.u/,
N
4 4
Q
the element ˇ3 has order 2. From Theorem I.6.5 we conclude that the geometric
automorphism 2 corresponding to .12/ acts on ˙4 .G/= Inn.G/ as ˇ1 ˇ14 and hence
as ˇ10 ˇ14
0
2 HSV0 . By Corollary 7.7 the element ˇQ3 is conjugate to ˇ10 ˇ14 0
in HSV0 ,
which proves that B .ˇQ3 / D B .ˇ1 ˇ14 /.
For V D h.12/i it only remains to determine the permutation representation on
B of a generator ˇQ2 of the inertia group of an extension Q2 onto MN 40 . Obviously
ˇQ2 lies in the centralizer C of ˇ340
in HSV0 D H40 [ ˇ10 H40 . Since the intersection
238 III Action of Braids

of C with H40 has index at most 2 in C , ˇ10 lies in C and .ˇ10 /2 D ˇ12 0
as part of
0 0
a free generating system of H4 generates its own centralizer in H4 , we conclude
that C \ H40 D hˇ12 0
iO. By the proof of Proposition 5.5 the braids ˇ12 and ˇ34 differ
only by the central involution .ˇ1 ˇ2 /3 with .ˇ10 ˇ20 /3 D 1. From this it follows that
0
ˇ12 D ˇ34 0
and hence C D hˇ10 iO. Now Q Q 2 D Q2 , so ˇ 0 generates a subgroup of
3 34
index 2 in hˇQ2 iO, and we may choose ˇQ2 D ˇ10 , which shows B .ˇQ2 / D B .ˇ1 /.
N u/ N VQ
Now let V D h.123/i and Q. Q WD Q.u/ . Then Q1 , Q2 and Q3 are conjugate
N
in Q.u/= N u/,
Q. Q and two prime divisors of Q. N u/ N
Q fully ramify in Q.u/= N u/,
Q. Q say Q Q2
and Q Q 3 . With certain prime divisors Q4 and Q5 of Q.u/= N N we thus have
Q
Q 1 D Q1 Q2 Q3 ;
Q Q 2 D Q3 ;
Q Q 3 D Q2 :
Q
4 5

As in the case of the group h.12/i it is now immediately seen that ˇQ1 D ˇ14 0
generates
the inertia group of an extension of Q1 onto M4 , which then implies that B .ˇQ1 / D
Q N 0

B .ˇ14 /. For i D 2 and i D 3 the inertia group generators ˇQi have order 3. Since the
geometric automorphism 3 for h.123/i in Theorem I.6.5 acts on ˙4 .G/= Inn.G/
as ˇ1 ˇ2 and hence as ˇ10 ˇ20 2 HSV0 , either ˇQi or ˇQi1 is conjugate to ˇ10 ˇ20 in HSV0 .
Consequently ˇQ2 and ˇQ3 may be chosen such that they satisfy B .ˇQ1 / D B .ˇQ2 / D
B .ˇ1 ˇ2 /. t
u

In a completely similar fashion the symmetrized braid orbit genera may also be
computed in the case s > 4 (see Przywara (1991), 3).

7.4 A Twisted Braid Orbit Theorem

By definition KNV =QN is a rational function field if and only if gsV . / D 0. For
the proof that K =kV is a rational function field, besides the assumption that
V

gsV . / D 0 we need the existence of a prime divisor of odd degree. This again can
be guaranteed by an oddness condition:
(O’): Under the action of ˇQ1 ; : : : ; ˇQsQ on BSV0 one of the cycle lengths, summed over
all ˇQi of the same permutation type, occurs an odd number of times.

Proposition 7.9. Let G be a finite group, C 2 Cl.G/s with s  4 and V a sym-

metry group of C. Furthermore for  2 ˙.C/ assume that with respect to a V -
configuration S0 we have gsV . / D 0 and the oddness condition (O0 ) is satisfied.
Then KV =kV is a rational function field, say KV D kV .v/.
Q

Proof. The permutation representation of HSV0 on B WD BSV0 . / is equivalent to the

coset representation of HSV0 on HV D Gal.MN s0 =KNV /. By Theorem I.9.1 the cycle
lengths eij of ˇQi coincide with the ramification indices of the prime divisors of Q
Qi
N V N
in K =Q.u/.Q Now among these prime divisors at most those can be conjugate
in KNV =KV for which the corresponding Q Q i are conjugate in Q.
N u/=K
Q V
B , where
7 Braids and Geometric Automorphisms 239

KBV denotes the fixed field of the orbit B in VS0 , and moreover the ramification
indices eij agree. Therefore condition (O0 ) implies that KV possesses a divisor of
odd degree. Together with g.KV =kV / D gsV . / D 0 this yields the rationality of
KV =kV . t
u
Combining the previous result with Theorem 7.3 we hence obtain the following
main result of this paragraph:
Theorem 7.10 (Twisted Braid Orbit Theorem). Let G be a finite group with trivial
center, C 2 Cl.G/s with s  4, V a symmetry group of C and S0  IP.MN s0 .t/=MN s0 /
a V -configuration. Furthermore assume that the HSV0 -orbit BSV0 . / of  2 ˙.C/ is
rigid in ˙.CV /= Inn.G/, that gsV . / D 0 and the oddness condition (O0 ) is satis-
Q tQ/ is a rational function field over QVC and there exists a
fied. Then KV D QVC .v;
geometric Galois extension N =QVC .v; Q tQ/ with

Q tQ// Š G
Gal.N =QVC .v; and MN s0 N D NN 0 : (7.22)

Here QVC D Q if C is V -symmetric.

Proof. The rationality of KV =QVC already follows from Theorem 7.3 and Proposi-
tion 7.9. The field N is then obtained as in Theorem 3.7(b) as the fixed field of the
centralizer of Gal.NN 0 =MN s0 .t// Š G in Gal.NN 0 =KV /. t
u
If we only want to secure the existence of such a Galois extension over QVC .tQ/,
instead of proving the rationality of KV =QVC it suffices to find an unramified ratio-
nal place } of KV =QVC .tQ/. Here such a place } is called unramified if any extension
}N of } to IP1 .MN s0 .t// is of type }N as with }as from (7.5).
Corollary 7.11. Let G, C, V , S0 and BSV0 . / be as in Theorem 7.10. If KV =QVC .tQ/
possesses an unramified rational place, then there exists a geometric Galois exten-
sion N =QVC .tQ/ with

Gal.N =QVC .tQ// Š G N

and Q.t/N N
 D N : (7.23)

Proof. For this proof we let NQ  be the fixed field of the centralizer of Gal.NN 0 =
MN s0 .t// in Gal.NN 0 =KV /. For V D 1 we then have Q
N NQ  D NN 0 , and hence by (7.7)

N  D NN 0 pQ D NN  2 NS .G/;
QN

where pQ denotes the valuation ideal of the unique extension }Q of }N to NN 0 . In par-

ticular, N =QC is regular with group Gal.N =QC .tQ// Š G.
N
Since tQ 2 Q.t/ we correspondingly have
N
Q.t/N N 0 Q D NN 
 D N p

for arbitrary V , from which as above the assertion follows in the case V ¤ 1. t
u
In the next section by way of example the biggest Mathieu group M24 is realized
as a geometric Galois group over Q.t/ using the Twisted Braid Orbit Theorem.
240 III Action of Braids

7.5 Geometric Galois Extensions over Q.t / with M24

Let G be the Mathieu group M24 in its natural permutation representation of

degree 24, and conform to Atlas notation 2A the class of involutions of type .28 ; 18 /
and 12B the class of elements of order 12 of permutation type .122 /. Then the
class vector C D .2A; 2A; 2A; 12B/ is rational, and one finds l.C/ D 144 and
h.C/ D 1. In particular C is a H4 -rigid class vector of G with the single braid
orbit B WD ˙.C/= Inn.G/. The types of the permutations B .ˇi 4 / are .53 ; 339 ; 26 /
for i D 1; 2; 3. Consequently the fourth braid orbit genus for Π2 B equals
3
g4 . / D 1  144 C .3  4 C 39  2 C 6/ D 1:
2
Thus the untwisted braid orbit theorem (Theorem 5.7) is not applicable. Neverthe-
less we have:
Q tQ/ with
Theorem 7.12. There exists a geometric Galois extension N =Q.v;
Q tQ// Š M24
Gal.N =Q.v; N  D NN 0 ;
and QN (7.24)
where  2 ˙.2A; 2A; 2A; 12B/. In particular, M24 possesses a G-realization over
Q.
Proof. For V D h.12/i, Theorem 7.8 yields the permutation types .53 ; 339 ; 26 /,
.619 ; 53 ; 43 ; 3/ and .272 / for B .ˇQi /, from which
1
g4V . / D 1  144 C .96 C 118 C 72/ D 0
2
follows. Since ˇQ1 satisfies the oddness condition (O0 ), Theorem 7.10 shows the exis-
tence of a geometric Galois extension N =Q.v; Q tQ/ with (7.24). t
u
Another proof of Theorem 7.12 is given by Dettweiler (2004), Thm. 3.10, using
a criterion for rational curves in Hurwitz spaces generalizing the Twisted Braid Orbit
Theorem.
An equation for the M24 -extension in Theorem 7.12 has been computed by
Granboulan (1996) using dessins d’enfants. By his computations he found that the
fixed field of M23 is the function field of a conic without rational points over Q.
Therefore this field extension cannot be specialized to M23 -extensions over Q as it
was possible in the case of M11 in Theorem I.6.12. So by Theorem II.9.9 the Math-
ieu group M23 remains the only sporadic simple group not yet realized as Galois
group over Q.
Remark. Further G-realizations of simple groups have been discovered by Shiina
(2003a), Thm. 0.2. Using the (Twisted) Braid Orbit Theorem he proved geometric
Galois extensions over Q.t/ for the simple groups

L3 .9/; L5 .2/; U4 .3/; U5 .2/; U6 .5/; S4 .4/; S6 .3/; 3D4 .2/; 2F4 .2/0 ; G2 .4/:

His results supplement Theorem II.10.2.

8 Ramified Rational Places 241

8 Ramified Rational Places

In this paragraph we study the specialization into ramified places. Here the exis-
tence of a rational ramified place in the ramification locus of the field of constants
of the Galois extension implies the Galois realization of the decomposition group at
this place inside the Galois extension. The latter may easily be described using the
Hurwitz classification. Since the decomposition group at inert places is isomorphic
to the original group, these again lead to existence theorems for Galois extensions
like for example the Rigid Braid Cycle Theorem contained in Section 8.3. In con-
trast, rational places in the ramification locus of the Galois extension lead at most
to realizations of proper factor groups of subgroups of the original group. But they
can sometimes be used to deduce the existence of prime divisors of odd degree and
therefore to verify oddness conditions. This is described in Section 8.4.

8.1 Decomposition Groups of Ramified Places

Let }ij for 1 i <j s C 1 denote the place

N
}ij W IP1 .Q.t// N _
! IP1 .Q.t 1 N
j // WD IP .Q.t1 ; : : : ; tj 1 ; tj C1 ; : : : ; tsC1 //; tj 7! ti ;
(8.1)

O ij onto
with the valuation ideal Dij . This by Theorem 2.4 possesses an extension D
N
MsC1 with inertia group
I.DO ij =Dij / D hˇij iO: (8.2)
O ij =Dij :
Here we moreover need to know the decomposition group of D

O ij =Dij / in MN sC1 =Q.t/

Theorem 8.1. The decomposition group D.D N with s  3 is
generated as profinite group by

ˇij ; ˇklfor k < l < i < j or k < i < j < l

or i < k < l < j or i < j < k < l;
ˇ ˇ
ˇkli l for k < i < l < j and ˇklj l for i < k < j < l: (8.3)

In particular it coincides with the centralizer and the normalizer of hˇij iOin HsC1 D
Gal.MN sC1 =Q.t//:
N

O ij =Dij / D CH .ˇij / D NH .hˇij i/:

D.D (8.4)
sC1 sC1

Proof. As in the proof of Theorem 2.4 let Dij denote the hyperplane in X sC1
defined by xi D xj and Dij the complement of the intersections Dij \ Dkl for
O ij =Q.t
fk; lg¤fi; j g in Dij . Then the residue class field extension MN sC1 D N 1 ; : : : ; tsC1 /
Dij is a maximal algebraic field extension unramified over Dij . In particular the
242 III Action of Braids

Galois group of this field extension is isomorphic to the algebraic fundamental group
of Dij :
Gal.MN O =Q.t/D
N / Š  .D  /:
alg
D sC1 ij ij 1 ij

Hence we also have

D.D O ij =Dij / Š  .D  /:
O ij =Dij /=I.D alg
(8.5)
1 ij

Here the preimage D.D O ij =Dij / of  alg .D  / is generated as a profinite group by

1 ij
those elements of the embedded group 1 .XsC1
top  I P / of homotopy classes of
0
paths with respect to the base point P0 D .1; : : : ; s C 1/, which can simultaneously
and continuously be deformed to homotopy classes in Dij with respect to the base
point Pij D .1; : : : ; j 1; i; j C1; : : : ; s C1/ such that ˇij becomes homotopy equiv-
alent to zero. Hence profinite generators of D.D O ij =Dij / are given by the elements
listed in (8.3) (see Figure 8.1).

l j k j
i i
k s+1 1
l
0 0
IP1(C) IP1(C)

1 1
1
Fig. 8.1 ˇil ˇkl ˇil ˇj1
l ˇkl ˇj l

Since all these generators commute with ˇij and even generate the full centralizer
of ˇij in HsC1 , we first have
O ij =Dij / D CH .ˇij /:
D.D sC1

Immediately from the relations between the generators ˇij of HsC1 or also from
Theorem 3.3 together with Theorem I.2.10 we conclude that ˇij cannot be conjugate
to any of its non-trivial powers in HsC1 . Hence the centralizer of ˇij coincides with
the normalizer in HsC1 of the subgroup generated by this element, which completes
the proof of the theorem. t
u
For the study of residue class fields of intermediate fields of MN sC1 =MN s .t/ only
the intersection of D.D O ij =Dij / with s D Gal.MN sC1 =MN s .t// is relevant. For this
the above leads to the following:
Corollary 8.2. The intersection ij of D.D O ij =Dij / with s is given by
 
ij D h1 ; : : : ; i 1 ; i j ; i C1
j j
; : : : ; j 1 ; j C1 ; : : : ; s iO (8.6)
for 1 i <j s, respectively by

ij D hi iO for 1 i < j D s C 1: (8.7)

8 Ramified Rational Places 243

8.2 Description via the Hurwitz Classification

Q ij denote the restriction of D

Now let D O ij to an intermediate field NN  of MN sC1 =MN s .t/
given by the Hurwitz classification (Theorem 3.4).

Theorem 8.3. Let G be a finite group, C 2 Cl.G/s with s  3 and  2 ˙.C/. Then
Q ij of D
the decomposition group of the restriction D O ij onto NN  =MN s .t/ is isomorphic
to the subgroup

Gij WD h1 ; : : : ; i 1 ; Q i ; i C1 ; : : : ; j 1 ; j C1 ; : : : ; s i for 1 i <j s (8.8)

of G with Q i WD .i    j /.i C1    j 1 /1 , respectively to

Gij WD hi i for 1 i < j D s C 1: (8.9)

Q ij =D
Proof. The decomposition group D.D N ij / is obtained by Corollary 8.2 as

Q ij =D
D.D N ij / D ij =. ij \ ker. //:

To describe the isomorphism type in the parametrizing group G we observe that

by (3.10) and Proposition 3.5, HsC1 acts inversely on s and on ˙s .G/. So it cen-
1 1
tralizes the image ij of ˇij WD .ˇi2 /ˇi C1 ˇj 1 instead of ˇij D .ˇi2 /ˇi C1 ˇj 1 .

g1
g1
1 i j
· · · · ·s · · · i ·k · gk ·s ·
k gs s+1 1 gs s+1
g ig j j
g˜i
g kg j

Fig. 8.2 Ns .hˇij i/ Ns .hˇij i/

According to Figure 8.2 we thus get Gij D h ij i with the generating system

 ij D .1 ; : : : ; i 1 ; i    j .i C1    j 1 /1 ; i C1 ; : : : ; j 1 ; j C1 ; : : : ; s /:

The proof for j D s C 1 is immediate. t

u
O ij =Dij / D hˇij iO with
Since the intersection of I.D s is trivial for j s, Theo-
rem 8.3 immediately implies:

Corollary 8.4. For 1 i < j s the Galois group of the residue class field exten-
Q ij =MN s .t/D
sion NN  D N ij is isomorphic to Gij :

Gal.NN  D N ij / Š Gij :
Q ij =MN s .t/D (8.10)
244 III Action of Braids

Denoting the generating system of Gij in (8.8) by  ij we have more precisely

Q ij D NN  2 N
NN  D N s1 .Gij /: (8.11)
ij

Here the containment relation NN  ij 2 NN s1 .Gij / makes sense only for s  4,
N
since N2 is not defined. The residue class map described in (8.11) is moreover com-
patible with the Galois action of the decomposition group, i.e., we have:
N s .G/ commutes with the residue
Proposition 8.5. The action of  2 CHsC1 .ˇij / on N
N O N
class map. Thus the image N 2 Gal.MsC1 Dij =Q.t/Dij / of  satisfies

Q ij D NN N :
NN  D (8.12)
ij

On the level of classes of generating systems (8.11) furnishes a canonical map

ij from the orbit of Œ 2 ˙s .G/= Inn.G/ under ij WD CHsC1 .ˇij / s = s to
˙s1 .Gij /=A, where A denotes the group of automorphisms of Gij induced by
NG .Gij /:
ij
ij W Œ ! ˙s1 .Gij /=A; Œ 7!  Aij : (8.13)
In the special case NG .Gij / D Gij we have A D Inn.Gij /. Then ij becomes
a map constant on the ˇij -cycles in Œ ij , whose image is a full Hs1 -orbit in
˙s1 .Gij /= Inn.Gij /.

8.3 Braid Cycle Orbits

Reduction of the Galois extension N =K given in Theorem 3.7 by the restriction
of a valuation ideal DQ ij with j s onto N yields a Galois extension N =K ,
ij ij
which after extension of constants to MN ijs WD MN s .t/DN ij becomes the Galois exten-
sion (8.10) and thus under suitable assumptions possesses a Galois group isomor-
phic to Gij .
The rationality of K ij can be checked following the procedure of Section 5.2.
In analogy to the notations used there for Œ 2 ˙s .G/= Inn.G/ let Bij . / denote
the orbit of the braid cycle Œ ij WD Œ hˇij i under ij , i.e.,
ij
Bij . / WD Œ ij with ij WD CHsC1 .ˇij / s= s: (8.14)

.s/
This orbit of braid cycles can now be structured via the groups k defined in
Corollary 2.9 like the braid orbit B. / in Section 4.2. For this let
 ij;s1
Bijs . / WD Πij with ij;s1 WD ij \ s1 (8.15)

and inductively for s > k  1

.s/ .s/ .s/
Bijk . / WD Bij;kC1 . /ij;k1 with ij;k1 WD .ij k \ k1 /= k : (8.16)
8 Ramified Rational Places 245

Here each Bijk . / is considered as a set of ijk -orbits. With this we can define the
reduced braid orbit genera

1 X
k
gijk . / WD 1  jBijk . /j C .jBijk . /j  cijkl /; (8.17)
2
j ¤lD1

.s/
where cijkl counts the number of ˇlk k -orbits on Bijk . /. In analogy to Theo-
rem 5.6 these satisfy:

Theorem 8.6. Let G be a finite group, C 2 Cl.G/s with s  3 and  2 ˙.C/.

Further let Kk denote the algebraic closure of Q.t1 ; : : : ; tk / in K and Kijk for
1 i < j s the residue class field of Kk by the restriction of D N ij . Then these
satisfy Kij;j 1 D Kijj and

ŒKijk W Kij;k1 .tk / D jBijk . /j and g.Kijk =Kij;k1 / D gijk . / (8.18)

for the k with 1 k s different from j .

In particular we have K ij D k .t_j / if the braid cycle Πij remains ij -
invariant.

Proof. The proof runs completely along the same lines as that of Theorem 5.6.
One only has to notice that by Theorem 8.1 the Galois group of the residue class
N ij =Q.t/D
field extension MN s .t/D N ij is isomorphic to ij =hˇij iO. Finally, the addi-
tional remark follows from K ij D Kijs .t/. t
u

To obtain information on Gal.N ij =K ij / we now first determine the decompo-

Q ij in the Galois extension NN ı =KN  obtained from
sition group by the restriction of D 
N =K by extension of constants with Q. N

Q ij to NN ı =KN  is
Proposition 8.7. The decomposition group of the restriction of D 
isomorphic to a subgroup GQ ij of G with

GQ ij Š CHsC1 .ˇij / \ ker. /   = CHsC1 .ˇij / \ ker. /  C .G/ (8.19)

with  D Gal.NN  =KN  /. In particular we have

Gij GQ ij NG .Gij / for 1 i <j s; (8.20)

Gij GQ ij CG .i / for 1 i < j D s C 1: (8.21)

Proof. The relevant decomposition groups are obtained as factor groups of the inter-
sections of CHsC1 .ˇij / with Gal.MN sC1 =KN  /D ker. /  and with Gal.MN sC1 =NN ı /
D ker. /  C .G/. This immediately yields (8.19) and (8.21). For (8.20) we use in
addition that the Galois extension NN  =MN s .t/ is obtained from NN ı =KN  by exten-
sion of constants with MN s , and hence GQ ij contains a normal subgroup isomorphic
to Gij . t
u
246 III Action of Braids

Since DQ ij for j s is unramified in NN  =MN s .t/ and hence also in NN ı =KN  , Propo-

sition 8.7 immediately gives:

Corollary 8.8. Let NN ıij =KN  ij be the residue field extension obtained from NN ı =KN 
upon reduction by the restriction of D Q ij with j s. Then with GQ ij from (8.19) we
have
Gal.NN ı =KN  / Š GQ ij and MN ijs NN ıij D NN  ij : (8.22)

Here in the case NG .Gij / D Gij we have GQ ij D Gij .

Our goal now is, starting from the Galois extension in (8.22), to arrive at a geo-
metric Galois extension over a rational function field over QC with Galois group
isomorphic to Gij or at least GQ ij . In general this is only possible under additional
rigidity and rationality assumptions. The latter will again be formulated as oddness
conditions:
.s/
(Oijk ): In the action of ˇlk k on Bijk . / for some l < k there occurs a cycle
length an odd number of times.
With this we obtain:

Theorem 8.9 (Braid Cycle Theorem). Let G be a finite group with trivial center and
C 2 Cl.G/s , s  3, a class vector of G. Further let  2 ˙.C/ and a pair .i; j / with
1 i < j s be chosen such that B. / is a rigid Hs -orbit and Bij . / is a rigid ij -
orbit of ˇij -cycles in B. /. Assume gijk . / D 0 and the oddness condition (Oijk )
for all k D 4; : : : ; s. If moreover Gij is self-normalizing in G, then there exists a
geometric Galois extension N ij =QC .u; Q t/ with uQ D .uQ 1 ; : : : ; uQ s1 / and

Q t// Š Gij
Gal.N ij =QC .u; and MN ijs N ij D NN  ij : (8.23)

Proof. The Rigid Braid Orbit Theorem 5.8 yields the existence of a Galois extension
N =K with K D QC .u1 ; : : : ; us ; t/ and Gal.N =K / Š G. Let N ij WD N DQ ij
and K ij WD K D N ij with the restrictions of Dij to N and K respectively. Then
N
we have QN ij D NN ıij and QK N  D KN  , so Gal.N =K / by Corollary 8.8
ij ij ij ij

contains a normal subgroup isomorphic to GQ ij D Gij . Since Gij was assumed to be

self-normalizing in G, the extension N ij =K ij is geometric and Galois with group
Gal.N ij =K ij / Š Gij .
The rigidity of Bij . / inside B. / implies that

ŒK ij W QC .t_ N N _
j / D ŒK ij W Q.tj / ;

whence the regularity of K ij =QC . Further from the vanishing of the reduced
braid genera gijk . / we first obtain that by (8.18) the algebraic closures Kijk of
QC .t1 ; : : : ; tk / in K ij have genus g.Kijk =Kij;k1 / D 0 for j ¤ k and Kijj D
Kij;j 1 . By the oddness condition (Oijk ) the extensions Kijk =Kij;k1 .tk / for j ¤ k
possess prime divisors of odd degree and thus are rational over Kij;k1 and there-
fore over QC . Consequently in K ij WD Kijs there exist s  1 independent tran-
8 Ramified Rational Places 247

Q and therefore K ij D QC .u;

scendentals uQ 1 ; : : : ; uQ s1 with K ij D QC .u/ Q t/. This
completes the proof of the first part of (8.23).
The second part follows by construction immediately from Corollary 8.8. t
u
Remark. If instead of NG .Gij / D Gij we only have NG .GQ ij / D GQ ij with GQ ij
from (8.19), then the Braid Cycle Theorem still holds for GQ ij at the place of Gij .
We now call Πij  B. / a rigid braid cycle if no further braid cycle Πij 
B. / possesses a stabilizer in ij which differs from the one of Πij only by an
automorphism of ij . Then as a special case of Theorem 8.9 we obtain the easily
applicable
Corollary 8.10 (Rigid Braid Cycle Theorem). Let G, C,  and B. / be as in The-
orem 8.9. If Πij is a rigid braid cycle with NG .Gij / D Gij , then there exists a
geometric Galois extension N ij =QC .t_j / with

j // Š Gij
Gal.N ij =QC .t_ and MN ijs N ij D NN  ij : (8.24)

Proof. Since a rigid braid cycle in particular remains invariant under ij , we have
KN  ij D Q.t
N _ / by Theorem 8.6 and hence in the proof of Theorem 8.9 also K D
j ij
_
QC .tj /. t
u
Remark. A braid cycle Πij is certainly rigid if inside B. / it is characterized by
its length, the inserted conjugacy class CQ i D ŒQ i of G and the reduced group Gij .
(Such cycles were called stable in Matzat (1991a).)
Finally the use of the Braid Cycle Theorem is demonstrated in an example com-
puted by F. Häfner:
Example 8.1. Let G D M24 and C be the rational class vector .2A; 2A; 4A; 5A/ in
Atlas-notation. Then ˙.C/= Inn.G/ is a rigid H4 -orbit of length 2000. On this, ˇ23
possesses precisely twelve 12-cycles. Among these only one possesses Aut.M22 /
as the reduced group G23 , with the class vector C23 D .2A; 12A; 5A/. This is hence
a rigid braid cycle and since Aut.M22 / as a maximal subgroup of M24 is self-
normalizing, by the Rigid Braid Cycle Theorem it leads to a geometric Galois exten-
sion N 23 =Q.t_ 3 / with the group Aut.M22 /. Passing to the rational fixed field of M22
we thus obtain a geometric Galois extension with the simple group M22 for the class
vector .6A; 5A; 5A/ over Q (see Theorem II.9.9 for another G-realization of M22 ).
Two of the 12-cycles lead to Galois extensions with the maximal subgroup G23 Š
26 W 3: S6 of M24 with C23 D .2A; 12B; 5A/. Since here the two classes of generating
systems of G23 coincide — we even have l.C23 / D 1 — and moreover Z .G23 / D 1,
by the Rigid Braid Orbit Theorem 5.8 there now exists a geometric Galois extension
3 / with the group 2 W 3 S6 . From this one can further deduce geometric
6 :
over Q.t_
Galois extensions defined over Q with the exceptional covering groups 3: S6 and
3: A6 of the groups S6 and A6 . The remaining nine 12-cycles produce the same
class vector C23 D .2A; 12B; 5A/ but the reduced group G23 coincides with the full
group G D M24 . Therefore in the last case D Q 23 remains inert in NN  =MN 4 .t/ and also
N ı N
in N =K . t
u
248 III Action of Braids

Remark. The reduction by the places }ij can in an obvious way be iterated and if
necessary be combined with the use of geometric automorphisms.

8.4 Prime Divisors of Odd Degree

The explicit knowledge of the decomposition groups GQ i;sC1 from (8.19) seems to
be of less importance for the realization of groups as Galois groups, since the Galois
groups of the corresponding residue class field extensions are proper factor groups
of subgroups of NG .hi i/. More precisely we have by Proposition 8.7:

Proposition 8.11. Reduction of the Galois extension NN ı =KN  by the corresponding

Q i;sC1 yields a Galois extension NN ı
restrictions of D =KN  i;sC1 with
i;sC1

Gal.NN ıi;sC1 =KN  i;sC1 / Š GQ i;sC1 =hi i: (8.25)

But these decomposition groups can sometimes be used to deduce the existence
of prime divisors of odd degree or even of degree 1 in root fields LN  of NN ı =KN 
respectively L of N =K . This comes from the following result, which is an
immediate consequence of Theorem I.9.1.

Proposition 8.12. Let LN  be an intermediate field of NN ı =KN  and U the permuta-

tion representation of Gal.NN ı =KN  / Š G on the cosets of U D Gal.NN ı =L
N  /. Then
Q N
Gi;sC1=hi i acts on the cycles of U .i / as on the prime divisors of Di;sC1jKN 
in LN  . The corresponding permutation representation of GQ i;sC1 =hi i is faithful at
least in the case of the regular representation U D 1.

Remark. With the help of Proposition 8.12 it is actually rather easy to obtain a lower
estimate for GQ i;sC1 by determining the image of the permutation representation of
C .G/ on the cycles of U .i /. (Since  D G  H , this action coincides with
the one given in Section 3.3 of Dèbes and Fried (1990).)

Thus Proposition 8.12 entails the main Theorem 3.14 in Dèbes and Fried (1990)
in the following form:

Theorem 8.13 (Dèbes and Fried (1990)). Let G be a finite group with trivial center,
C 2 Cl.G/s with s  3 and  2 ˙.C/. Using the assumptions and notation of Propo-
sitions 8.7 and 8.12 and with the intermediate field L of N =K with QL N  D LN 
we have:
(a) If in the action of GQ i;sC1 =hi i (resp. CG .i /=hi i) on the cycles of U .i /
there occurs a cycle orbit of odd length, then D N i;sC1 j N possesses a prime divisor
K
(resp. divisor) of odd degree in LN  .
(b) If in the action of GQ i;sC1=hi i (resp. CG .i /=hi i) on the cycles of U .i /
an odd cycle orbit length occurs an odd number of times, then D N i;sC1jK possesses
a prime divisor of odd degree in L .
8 Ramified Rational Places 249

Proof. Part (a) follows immediately from Theorem I.9.1 with Proposition 8.12.
Part (b) is obtained from the fact that at most prime divisors of the same degree
N i;sC1j N in LN  can be conjugate in L
of D N  =L . t
u
K

Using the field K D Ks defined in Section 4.2 we thus deduce:

Corollary 8.14. If in Theorem 8.13 we have g.L =K / D 0, then under the assump-
N  =QK
tions to (a) the field QL N  and under the assumptions to (b) even the field
L =K is a rational function field.

This result cannot apply to the fixed field of M23 in the M24 -extension of Theo-
rem 7.12, as shown by the result of Granboulan (1996) cited above. This is verified
by direct computations in the following example.

Example 8.2. Let N =Q.v; Q tQ/ be the M24 -extension belonging to the class vector
.2A; 2A; 2A; 12B/ and V D h.12/i constructed in Theorem 7.12. Then in KV D
Q tQ/ the prime divisors P3 WD D
Q.v; N 35 j V and P4 WD D N 45 j V both possess the
K K
residue degree 1 (and D N 15 j V D D N 25 j V the residue degree 2). Using Proposi-
K K
tion 8.12 we obtain with G D M24

GQ 35 Š CG .3 / Š 21C6 : L3 .2/:

Further GQ 35 acts transitively both on the eight fixed points and on the eight transpo-
sitions of 3 as the group 23  L3 .2/. Thus P3 splits in the fixed field L of M23 in
two prime divisors of degree 8. For P4 we obtain

GQ 45 Š CG .4 / Š h4 i;

so that the two 12-cycles remain fixed under 4 . But here the two prime divisors
of P4 are conjugate in LN  =L under complex conjugation. This follows from the
fact that the intersection of NG .h4 i/ (containing the decomposition group of P4
in N =KV ) with M23 contains no involution  with 4 D 41 (compare Corol-
lary I.10.5). Hence P4 possesses a prime divisor of degree 2 in L with decompo-
sition group of type GQ 45 W 2. t
u
250 III Action of Braids

9 The Katz Algorithm

It is the aim of this paragraph to introduce the convolution functor and to describe
the Katz algorithm based upon it. With this for example all rigid generating systems
of irreducible linear groups can be related to generating systems of GL1 , or can
be derived from such. Typical examples for this procedure will be presented in the
subsequent paragraph.

9.1 The Convolution Functor

We present the convolution functor introduced by Katz (1996) in the algebraic

version of Dettweiler and Reiter (2000) as slightly varied in Dettweiler (2003).
The latter is better adapted to the notation used here and moreover sometimes leads
to easier formulas than in the original form.
In the following let k be a field, V Š k n a k-vector space and  D .1 ; : : : ; s /
an s-tuple in GL.V / Š GLn .k/. We will use the notation
1
1 WD 1    s and 0 WD 1 :

Further let G WD h1 ; : : : ; s i be the subgroup of GL.V / generated by  . (In the

applications of Katz’s algorithm we will have 0 D Id, so that  D .1 ; : : : ; s / is an
element of ˙s .G/.)
Furthermore let Fs WD h1 ; : : : ; s i be the free group on s generators and
Rep.kŒFs / the category of finite dimensional kŒFs -modules. Its objects

. ; V / D .1 ; : : : ; s ; V /

consist of a finite dimensional k-vector space V together with images 1 ; : : : ; s 2

GL.V / of 1 ; : : : ; s . The morphisms

 W . ; V / ! .Q ; VQ / with  ı i D Q i ı 

are linear maps from V to VQ with the compatibility conditions  ı i D Q i ı  for

i D 1; : : : ; s. An element . ; V / 2 Rep.kŒFs / is called irreducible respectively abso-
lutely irreducible if V is irreducible, respectively absolutely irreducible as kŒFs -
module.
9 The Katz Algorithm 251

For . ; V / 2 Rep.kŒFs / with V Š k n and c 2 k  the matrices

0 1
n
B :: C
B : C
B C
B n C
B C
Pc .i / WD Bc.1  n / : : : c.i 1  n / ci i C1  n : : : s  n C (9.1)
B C
B n C
B C
@ ::: A
n

in GLs .V / Š GLsn .k/ with n 2 GL.V / the identity matrix, are the Pochhammer
transforms (in honor of Pochhammer (1870), who for the first time gave such a rela-
tion between hyperelliptic differential equations, compare also Haraoka (1994)).
The Pochhammer transforms define a functor

Pc W Rep.kŒFs / ! Rep.kŒFs /; . ; V / 7! .Pc . /; V s /; (9.2)

with Pc . / WD .Pc .1 /; : : : ; Pc .s //, which here will be called the Pochhammer
transformation (called convolution in Dettweiler (2003)). The product

Pc .1 / WD Pc .1 /    Pc .s /

satisfies the easily verifiable formula

0 1
1  n : : : s  n
B :: C
Pc .1 /  csn D c diag.2    s ; : : : ; s ; n /  @ ::: : A (9.3)
1  n : : : s  n

which will turn out to be quite useful. Moreover we observe the following:

Proposition 9.1. Let . ; V / 2 Rep.kŒFs /, c 2 k  and .Pc . /; V / a Pochhammer

transform. Then both
t
X WD XV WD ker.1  n /; : : : ; ker.s  n / Vs and
\
s
Y WD YV WD ker Pc .i /  sn Vs
i D1
˝ ˛
are H WD Pc .1 /; : : : ; Pc .s / -invariant subspaces of V s .

Thus the functor Pc induces a residue class functor PNc onto a lower dimensional
vector space W WD V s =.X C Y /, which will here be called convolution (middle
convolution in Dettweiler (2003)):

PNc . ; V / WD .PNc . /; W / with W WD V s =.XV C YV /: (9.4)

252 III Action of Braids

Remark. A morphism  W . ; V / ! .Q ; VQ / in Rep.kŒFs / induces canonical mor-

phisms

Pc ./ W .Pc . /; V s / ! .Pc .Q /; VQ s /; .v1 ; : : : ; vs /t 7! ..v1 /; : : : ; .vs //t ;

PNc ./ W .PNc . /; W / ! .PNc .Q /; WQ / with W D V s =.X C Y /; WQ D VQ s =.XQ C YQ /;

on Rep.kŒFs /. In this way the Pochhammer transformation Pc and the convolution

PNc become endofunctors on Rep.kŒFs /.

The dimension of the image W WD V s =.X C Y / of PNc is given by a simple for-

mula:

Proposition 9.2. Let . ; V / 2 Rep.kŒFs / with dimk .V / D n and i ¤ n for i D

1; : : : ; s, and 1 ¤ c 2 k  . Then the dimension of the image W WD V s =.X C Y / of the
convolution PNc is given by

X
s
dimk .W / D rk.i  n /  n  rk.c1  n / :
i D1

Proof. We first show that Y D YV is given by

t
Y D f 2    s .v/; 3    s .v/; : : : ; s .v/; v j v 2 ker.c1  n /g: (9.5)

To see this let us write Z for the right hand side of this equation. From
t
.Pc .i /  n / 2    s .v/; : : : ; s .v/; v D .0; : : : ; v C c1 .v/; : : : ; 0/t

with non-trivial i th row it already follows that Y  Z. Now let y D .y1 ; : : : ; ys /t 2 Y .

Then by definition .Pc .i /  sn /y D 0 for all i D 1; : : : ; s and hence also

´i WD c.1  n /y1 C : : : Cc.i 1  n /yi 1 C .ci  n /yi

C.i C1  n /yi C1 C : : : C .s  n /ys D 0:

Thus the differences

´i C1  ´i D .1  c/yi C .c  1/i C1 .yi C1 /

also vanish, and so yi D i C1 .yi C1 / as c ¤ 1. Plugging this into ´s we obtain

0 D ´s D c.1    s  2    s /ys C : : : C .cs  n /ys D .c1  n /ys

and so ys 2 ker.c1  n /. This shows that Y Z and hence our claim Y D Z.

Next we show that X \ Y D 0. For this let

v D .2    s .v/; : : : ; v/t 2 X \ Y:

9 The Katz Algorithm 253

Then .i  n /.i C1    s /.v/ D 0 for all i and hence

i    s .v/ D i C1    s .v/ D : : : D v for i D 1; : : : ; s:

Thus we have
\
s
v2 ker.i  n / \ ker.c1  n /:
i D1

Due to v D 1    s .v/ and c ¤ 0 this leads to v D 0 and hence to X \ Y D 0.

Our assertion about the dimension follows immediately from these two formulas:

X
s
dimk .W / Ds n  dimk .ker.i  n //  dimk .Y /
i D1
X
s
D rk.i  n /  .n  rk.c1  n //: t
u
i D1

Remark. In the case c D 1 one obtains directly from the definition that

X
s
Y D f.v1 ; : : : ; vs /t j .i  n /vi D 0g:
i D1

The next result shows that the convolution behaves well with respect to invariant
subspaces and their direct sums.

Proposition 9.3. Let . ; V / 2 Rep.kŒFs / and G WD h1 ; : : : ; s i. Further let PNc be

a convolution with c 2 k  and H WD hPNc .1 /; : : : ; PNc .s /i.
(a) If U V is a G-invariant subspace, then U s is an H -invariant subspace
of V s .
(b) If V D U1 ˚ U2 is a direct sum of two G-submodules then also

PNc . ; V / D PNc . ; U1 / ˚ PNc . ; U2 /:

Proof. We first show (a). By definition of PNc we see that U s V s is H -invariant.

It remains to show the formula U s \ .XV C YV / D XU C YU . In the case c D 1 this
follows immediately from Proposition 9.1. For c ¤ 1 we certainly have XU C YU
U s \ .XV C YV /. If now v D .v1 ; : : : ; vs /t is an element of U s \ .XV C YV / then by
Proposition 9.1 and the proof of Proposition 9.2 we get

vi D wi C i C1    s .u/ with wi 2 ker.i  n /; u 2 ker.c1  n /:

From this it follows that

X
s X
s
1
.i  n /vi D .i    s  i C1    s /.u/ D .1  n /.u/ D .  1/u 2 U
c
i D1 i D1
254 III Action of Braids

and hence u 2 U due to c ¤ 1. Thus also all of the components wi lie in U , which
implies that v 2 XU C YU .
Assertion (b) follows immediately from assertion (a). t
u

9.2 Multiplicativity of Convolution

In order to simplify our assumptions we introduce a hypothesis which will always be

satisfied in our application cases of irreducible linear groups. We say that . ; V / 2
Rep.kŒFs / satisfies hypothesis (H) if for all a 2 k  the following two statements
hold:
\
.H1/ W ker.ai  n / \ ker.j  n / D 0 for i D 1; : : : ; s;
j ¤i
X
.H2/ W Ui .a/ WD im.ai  n / C im.j  n / D V for i D 1; : : : ; s:
j ¤i

Let us point out the following:

Remark. (a) The k-vector spaces Ui .a/ are h i-invariant, as is

X
s
U1 WD im.j  n /: (9.6)
j D1

(b) Under hypothesis (H) we have U1 D V .

The next statement is equally obvious.

Proposition 9.4. For . ; V / 2 Rep.kŒFs / the hypothesis (H) holds in any of the two
cases:
(1) . ; V / is irreducible and dimk .V / > 1.
(2) We have dimk .V / D 1 and .1 ; : : : ; s / has at least two components with
i ¤ 1 .

We now show that hypothesis (H) is preserved by convolution.

Proposition 9.5. If hypothesis (H) is satisfied for . ; V / 2 Rep.kŒFs /, then it also

holds for PNc . ; V / for all c 2 k  .

Proof. We start with the special case a D 1. Here for part (H1) we have to check the
validity of

\
s
ker.i  ns / C X D Y C X; for X WD XV ; Y WD YV
i D1
9 The Katz Algorithm 255

where  D Pc . /. Let us denote the left hand side by U . Then certainly U  Y CX .

On the other hand v D .v1 ; : : : ; vs /t 2 U satisfies

.j  ns /v 2 .0; : : : ; ker.j  n /; : : : ; 0/t :

This implies that .i  n /vj 2 ker.j  n / for i ¤ j as well as

.cj  n /vi 2 ker.j  n /. So
\
.j  n /vj 2 ker.cj  n / \ ker.i  n / D 0;
i ¤j

T
which implies v 2 siD1 ker.i  ns / C X D Y C X and hence U Y C X.
For part (H2) in the case a D 1 we need to verify

X
s
im.i  ns / D V s :
i D1

Again denoting the left hand side by U , we trivially have U V s . By (H2) for
. ; V / for any v 2 V s there exist elements wij 2 V with
X X
vi D c.j  n /wij C .ci  n /wi i C .j  n /wij for i D 1; : : : ; s:
j <i j >i

So the vector vi WD .0; : : : ; vi ; : : : ; 0/t has the form

vi D .i  ns /wi with wi D .wi1 ; : : : ; wi s /t :

Consequently,
X
s X
s
vD vi D .i  ns /wi 2 U:
i D1 i D1

In the case a ¤ 1 for fixed i we let U denote the intersection of

\
ker.j  n / C X with ker.ai  n / C X C Y:
j ¤i

Then for (H1) we need to show that U D X C Y . Clearly U  X C Y . Now let

v D .v1 ; : : : ; vs /t 2 U , so that we have

.j  ns /v 2 X for j ¤ i and .ai  ns /v 2 X C Y:

The latter leads to a linear system of equations

i
Av D w with w D .w1 ; : : : ; 0; : : : ; ws /t 2 X1 ;
Bv D u with u 2 X C Y;
256 III Action of Braids

with the matrices

0 1
.a  1/n 0
0 1 B :: C
c1  n : : : s  n B : C
B :: C ; B C
AD@ ::
:
::
: : A B DB
B ac.1   n / : : : aci   n : : : a.s   / C
n C:
B : C
c1  n : : : cs  n @ :: A
0 .a  1/n

The second system of equations implies

uj D .a  1/vj for j ¤ i and ui D .a  1/vi C a´i with

´i D c.1  n /v1 C : : : C .ci  n /vi C : : : C .s  n /vs :

From this one obtains u D .a  1/v C azi with zi WD .0; : : : ; ´i ; : : : ; 0/t , which then
yields .a  1/v C azi 2 X C Y . For zi we obtain analogously
t
Azi D .i  n /´i ; : : : ; 0; : : : ; c.i  n /´i 2 X;
Bzi D .0; : : : ; .aci  n /´i ; : : : ; 0/ 2 X C Y:

According to (H1) for . ; V / the image of zi in UN WD U=.X C Y / is zero, which

implies zi 2 X C Y and so finally also v 2 X C Y as a ¤ 1.
To check (H2) in the case a ¤ 1 we have to verify that
X
im.ai  ns / C im.j  ns / D V s for i D 1; : : : ; s:
j ¤i

Denoting again the left hand side by U we certainly have U V s . For v 2 V s the
assumption (H2) for . ; V / yields the existence of vectors wij 2 V with
X X
vi D ac.j  n /wij C .aci  n /wi i C a.j  n /wij :
j <i j >i

For vj WD .0; : : : ; vj ; : : : ; 0/t and wj D .wj1 ; : : : ; wjs /t it then follows that vi D

.ai  ns /wi as well as vj D a.j  ns /wj for j ¤ i . So we have

X
s X
vD vj D a.j  ns /wj C .ai  ns /wi 2 U: t
u
i Dj j ¤i

Theorem 9.6. Assume that . ; V / 2 Rep.kŒFs / satisfies hypothesis (H) and let
c1 ; c2 2 K  with c1 c2 D c. Then we have an isomorphism

PNc . ; V / Š PNc2 PNc1 . ; V /

9 The Katz Algorithm 257

as kŒFs -modules and hence in particular

PN1 . ; V / Š . ; V /:

Proof. In a first step we verify the highlighted special case PN1 . ; V / Š . ; V /. From
the representation of Y for c D 1 in the Remark after Proposition 9.2 it follows that
X Y and so X C Y D Y . As U1 D V by the Remark (b) before Proposition 9.4
the linear map

X
s
 W V s ! V; .v1 ; : : : ; vs /t 7! .i  n /vi
i D1

is surjective with kernel Y . So PN1 D  ı P1 induces an isomorphism of k-vector

spaces. The isomorphy in kŒFs then follows upon verification of  ı P1 .i / D i ı
.
In a second step for  D Pc1 . / we show the formula

PNc2 PNc1 . ; V / Š PNc2 .; V s /=PNc2 .; XV C YV .c1 //; (9.7)

where the added ci is meant to indicate the action with Pci . According to the first
N W / WD PNc1 . ; V /
step we may here assume that c1 ¤ 1 ¤ c2 . Let us abbreviate .;
and
X1 WD XV Y1 WD YV .c1 /; X2 WD XV s ; Y2 WD YV s .c2 /:
Clearly assertion (9.7) would follow from

XW D .X2 CX1s CY1s /=.X1s CY1s / and YW D .Y2 CX1s CY1s /=.X1s CY1s /:

To prove the formula for XW we denote the right hand side by U . Then certainly
U XW . By definition we now have
t
XW D ker.N1  m /; : : : ; ker.Ns  m / with
ker.Ni  m / D fwi C X1 C Y1 2 W j .i  sn /wi 2 X1 C Y1 g

with m D dimk .W /. Using (9.5) we obtain

.i  sn /wi D .xi1 C 2    s .yi /; xi 2 C 3    s .yi /; : : : ; xi s C yi /

with xij 2 ker.j  n / and yi 2 ker.c1 1  n /. According to (H1) the right hand
side has the form .0; : : : ; ; : : : ; 0/. So we have yi 2 ker.j  n / for j ¤ i and yi 2
ker.c1 i  n /. This shows that

.i  sn /.wi / 2 .0; : : : ; ker.i  n /; : : : ; 0/t ;

which finally implies that wi 2 ker.i  sn / C X1 and hence XW U.

258 III Action of Braids

Now let U denote the right hand side in the stipulated formula for YW . Then we
have U YW . An element . Ń 1 ; : : : ; Ń s / from

\
s
YW D ker.Ni  sm / with .N1 ; : : : ; Ns / D PNc2 ./
i D1

is a solution of the homogeneous system of linear equations

c2 .N1  m / Ń 1 C : : : C .c2 Ni  m / Ń i C : : : C .Ns  m / Ń s D 0 .1 i s/:

Here the components Ń j have the form Ń j D ´j C X1 C Y1 with ´j 2 V s . This

implies for  D Pc2 ./ that

.i  s 2 n /.´1 ; : : : ; ´s /t 2 X1s C Y1s for i D 1; : : : ; s;

from which we conclude .´1 ; : : : ; ´s /t 2 Y2 C X1 C Y1 and thus finally . Ń 1 ; : : : ; Ń s /

2 U.
In the third and last step of the proof we show the claimed general multiplication
formula for the convolution. For this it suffices to show that the linear map

N W PNc2 PNc1 . ; V / ! PNc . ; V /

induced by

X
s
 W Pc2 Pc1 . ; V / ! Pc . ; V /; .v1 ; : : : ; vs / 7!
t
.i  ns /vi ;
i D1

is an isomorphism in Rep.kŒFs /. By the first step we may assume that c1 ¤ 1 ¤

c2 . The surjectivity of  follows directly from (H2), case a D 1, in the proof of
Proposition 9.5. We now use the notation X1 ; Y1 ; X2 ; Y2 from the second step and
in addition
X WD XV and Y WD YV .c/:
Then from the second step we first obtain the isomorphisms

PNc2 PNc1 . ; V / ŠPNc2 .; V s /=PNc2 .; X1 C Y1 /

Š.; .V s /s /=.; .X1 C Y1 /s C X2 C Y2 /

for the underlying vector spaces. Here by definition we have X2 D ker./ and
Y1s X2 . So the canonical homomorphism is certainly well-defined and surjective.
Furthermore,

X
s
.X1s / D .i  ns /X1 D .: : : ; .c1 i  n / ker.i  n /; : : :/t Š X1 D X:
i D1
9 The Katz Algorithm 259

Using (9.5) this gives

˚
.Y2 / D .2    s .v/; : : : ; s .v/; v/t j v 2 ker.c2 1  ns /
Df1 .v/  v j v 2 ker.c2 1  ns /g D ker.c2 1  ns /:

Here v D .v1 ; : : : ; vs /t 2 ker.c2 1  ns / if and only if .cPc1 .1 /  c1 ns /v D 0,

respectively if c.Pc1 .1 /  c1 ns /v D c1 .1  c/v. By (9.3) this is equivalent to
0 1
1  n : : : s  n
B :: C v D .1  c/v:
c diag.2    s ; : : : ; s ; n / @ ::: : A
1  n : : : s  n

The last line of this linear system of equations reads

´s D c.1  n /v1 C : : : C .cs  n /vs D 0

with the ´s from the proof of Proposition 9.2. The lines before that yield vi D
i C1 .vi C1 / for i D 1; : : : ; s  1, which as in the proof of Proposition 9.2 gives that
´i D 0 for i D 1; : : : ; s  1, as well as

´s D .c1  n /vs D 0 respectively vs 2 ker.c!1  n /;

where ! WD PNc . /. With this we have

.Y2 / D YV .c/ D Y:

Taken together we obtain

.X2 C Y2 C X1s / D .Y2 C X1s / D .Y2 / C .X1s / D Y C X:

Comparing dimensions then shows that

dimk .X2 C Y2 C X1s / D dimk .X2 / C dimk .X C Y /

with dimk .X2 / D dimk .ker.// D .s  1/sn. Thus,

dimk .V s /s =.X2 C Y2 C X1s / D dim V s =.X C Y / ;

which shows that N is an isomorphism.

It remains to check the compatibility of N with the group operations, that is, the
validity of
 ı .i  ns / D .!i  ns / ı  for i D 1; : : : ; s:
260 III Action of Braids

This follows with vj WD .0; : : : ; vj ; : : : ; 0/t from

8
ˆ
<c2 .i  ns /.j  ns /vj j <i
.i  sn /vj D .i  ns /.c2 j  ns /vj for j D i D .0; : : : ; wi ; : : : ; 0/t DW wi
:̂
.i  ns /.j  ns /vj j >i

in combination with

.!i  ns /.vj / D .!i  ns /.j  ns /.vj / D wi : t

u

Corollary 9.7. Assume that . ; V / 2 Rep.kŒFs / satisfies the hypothesis (H). Then
PNc . ; V / is irreducible if and only if . ; V / is.
Proof. This follows immediately from Proposition 9.3(a) together with the invert-
ibility of PNc shown in Theorem 9.6. t
u

9.3 Linear Rigidity

The concept of linear rigidity is closely related with the rigidity of a tuple of
group elements and has proved highly useful in connection with the Katz algo-
rithm. We call an s-tuple .1 ; : : : ; s / 2 GLn .k/s linearly rigid (physically rigid in
Katz (1996)), if 1    s D n and the conjugacy classes CQ i of i in GLn .k/ sat-
isfy the following rigidity condition: If .1 ; : : : ; s / 2 GLn .k/s is a further s-tuple
with i 2 CQ i , then there exists  2 GLn .k/ with i D i for all i . So the group G
generated by 1 ; : : : ; s does not enter the picture.
If now Ci is the conjugacy class of i in G WD h1 ; : : : ; s i GLn .k/ then C D
.C1 ; : : : ; Cs / is called GL-stable (or stable for short), if it satisfies the following
stability condition:
(S) Every ˛ 2 NGLn .k/ .G/ with Ci˛ D Ci for i D 1; : : : ; s lies in CGLn .k/ .G/  G.
The connection between linear rigidity and rigidity is explained by the next result
(see Theorem I.5.10 and the subsequent Remark):
Proposition 9.8. Let  2 GLn .k/s with 1    s D n , G WD h1 ; : : : ; s i and Ci WD
Œi in G. Then we have:
(a) If  is linearly rigid and C D .C1 ; : : : ; Cs / is GL-stable then  , respectively
˙s . / is rigid.
(b) If NGLn .k/ .G/ D CGLn .k/ .G/  G then every linearly rigid  2 ˙s .G/ is rigid.
Conversely it is not true that the rigidity of  2 ˙s .G/ implies linear rigidity.
This results from the fact that for linearly rigid tuples the generating system class
number l .C/ agrees in G with the normalized structure constant n.C/, which is not
usually satisfied for rigid tuples (see for example the oddments in Section II.9.5).
A crucial role for the criterion for linear rigidity used in Katz’s algorithm is
played by the so-called Lemma of Scott (1977):
9 The Katz Algorithm 261

Proposition 9.9 (Lemma of Scott). Let k be any field, V D k n and .1 ; : : : ; s / 2

GL.V /s with 1    s D IdV . Further let c.i / be the codimension
T of the fixed point
space F .i / in V , c. / the codimension of F . / WD siD1 F .i /, and c.i / and
c.  / the corresponding codimensions for the dual action. Then we have

X
s
c.i /  c. / C c.  /:
i D1

Proof. We define

U WD fv D .v1 ; : : : ; vs /t 2 V s j vi 2 .1  i /V g

and linear maps

t
 W V ! U; v 7! .1  1 /v; : : : ; .1  s /v ;
W U ! V; v 7! v1 C 1 .v2 / C : : : C 1    s1 .vs /:

Then the identity

n  1    s D .1  1 / C 1 .1  2 / C : : : C 1    s1 .1  s /

in kŒGL.V / immediately yields .1  1 /v; : : : ; .1  s /v D 0 and thus im./

ker. /. Now let G WD h1 ; : : : ; s i. Then

X
s
.U / D .1  1 /V C 1 .1  2 /V C : : : C 1    s1 .1  s /V D .1  i /V
i D1

is the smallest G-submodule W of V such that the action of G is trivial on V =W .

Thus we have
dimk . .U // D c.  / D dimk .U= ker. //:
Now ker./ is the subspace of fixed points of G, so by definition it has dimension
dimk ..U // D c. /. So we get the two expressions

X
s
dimk .U / D c.i / with c.i / D dimk ..1  i /V /
i D1

and
dimk .U / D dimk .U= ker. // C dimk .ker. /= im.// C dimk .im.//
c.  / C c. /;

which together prove the claim. t

u
262 III Action of Braids

Corollary 9.10. If  D .1 ; : : : ; s / 2 GLn .k/s with 1    s D n is irreducible, then

X
s
rk.i  n /  2n:
i D1

Proof. In the case of irreducibility of h1 ; : : : ; s i we have c. / D n D c.  /, and

the claim follows from Proposition 9.9. t
u

Remark. In the case of permutation representations, Scott’s formula just says that
the genus (as computed by the Hurwitz relative genus formula) of a cover of the
Riemann sphere ramified at s points with inertia generators i is non-negative.

The next theorem contains the previously announced criterion for linear rigidity
in the form presented by Strammbach and Völklein (1999):

Theorem 9.11. Let  2 GLn .k/s be absolutely irreducible with 1    s D n . Then

we have P
(a) rid. / WD siD1 codimk .Ck nn .i //  2.n2  1/  0.
(b) If k D kN is algebraically closed then moreover rid. / D 0 if and only if  is
linearly rigid.

Proof. For part (a) let 1 ; : : : ; s 2 GLn .k/s with 1    s D n with i conjugate in
GLn .k/ to i , that is, there exist i 2 GLn .k/ such that i D i i i1 for i D 1; : : : ; s.
Further let W D k nn be the vector space of n  n-matrices over k and define linear
maps
i W W ! W;  7! i i1 ;
of W into itself. We first show that

c.i / D n2  dimk .F .i // D n2  dimk .CW .i //: (9.8)

For !i 2 F .i / we have !i D i !i i1 . As i i i1 D i D !i1 i !i this is equiv-

alent to !i i 2 CW .i /, respectively !i 2 CW .i /i1 . From this (9.8) follows due
to dimk .CW .i /i1 / D dim.CW .i //.
Next we show that rid. /  0. For this we use the dual vector space W  via

W  W  ! k; .; ı/ 7! tr.ı/;

and the map

i W W  ! W  ; ı 7! i ıi1 ;
dual to i . Specializing i to conjugation with i , that is, choosing i D i , we
obtain
dimk .F .// D dimk CW .h1 ; : : : ; s i/ D 1
9 The Katz Algorithm 263

because of the absolute irreducibility of . ; V /, and similarly dimk .F . // D 1. So

from the Lemma of Scott and (9.8) we get

X
s X
s
2.n2  1/ D c./ C c.  / c.i / D codimk .CW .i //
i D1 i D1

as claimed.
N Then by (a)
For part (b) we first assume that rid. / D 0, but not yet that k D k.
we have
X s Xs
codimk .CW .i // D c.i / D 2.n2  1/;
i D1 i D1

so by the Lemma of Scott c./ C c. / 2.n2  1/. Thus either F ./ ¤ 0 or
F . / ¤ 0. Assume that F ./ ¤ 0 and T 0 ¤  2 F ./. Then  is invertible due to
the irreducibility of . ; V /, As F ./ D siD1 F .i / we see that i ./ D  for all
i D 1; : : : ; s. But this is equivalent to i i1 D , that is, 1 i  D i for all i . So
 is linearly rigid.
For the converse we assume that k D k; N s be linearly rigid. Fur-
N so let  2 GLn .k/
ther let Ci WD Œi be the conjugacy class of i in the algebraic group G D GLn .k/. N
We define a morphism

 W C1      Cs ! G0 WD G=Z .G/; .1 ; : : : ; s / 7! 1    s .mod Z .G//:

As  is linearly rigid, the fibre O WD  1 .n / forms a single G-orbit under simulta-
neous conjugation. Consequently we have

dimk .O/ dimk .G0 / D dimk .G/  dimk .Z .G// D n2  1

(by Springer (1998), Ch. 2.1). On the other hand by Springer (1998), Ch. 5.1,
Ch. 1.8 and Ch. 2.1 we have
X
s
dimk .O/  dimk .C1      Cs /  dimk .G0 / D dimk .Ci /  dimk .G0 /
i D1
X
s
D codimk .CW .i //  dimk .G0 /:
i D1

This ensues
X
s
codimk .CW .i // dimk .G0 / C dimk .O/ 2.n2  1/:
i D1

This implies that rid. / 0 and hence by (a) the desired equality rid. / D 0. t
u

The invariant rid. / in Theorem 9.11 is called the rigidity defect; Katz (1996)
uses the notion rigidity index for 2  rid. /.
264 III Action of Braids

9.4 The Existence Algorithm of Katz

For the convolution operator the Jordan normal forms of the components of the
image PNc . / can easily be calculated from those of 1 ; : : : ; s , based upon the fol-
lowing observation:

Proposition 9.12. Assume that . ; V / 2 Rep.kŒFs / satisfies hypothesis (H). Then

for N WD PNc . / we have:
(a) rk.i  n / D rk.Ni  m / for 1 i s,
(b) rk.c1  n / D rk.N1  cm /.

Proof. For the proof of (a) we show that

i W im.Ni  m / ! im.i  m /; wi D .0; : : : ; wi ; : : : ; 0/t C X C Y 7! .i  n /wi ;

(9.9)

is an isomorphism with i ı Ni D ci ı i . For this let  WD Pc . /. Then Y

ker.i  sn / by Proposition 9.1. Thus i is well-defined and injective. By (H) we
have Ui .c/ D V , so i is moreover surjective and hence an isomorphism of vector
spaces. The compatibility with the group operations is now obtained as

i .i .i  sn //.wi C X C Y / D .i  n /ci wi D ci i .i  sn /.wi C X C Y /:

For the proof of (b) we show that the map

1 W im.N1  cm / ! im.c1  n /;

w D .w1 ; : : : ; ws /t C X C Y 7! .c1  n /wi ; (9.10)

is an isomorphism with 1 ı .N1  m / D .c1  n / ı 1 . According to (9.5) we

have
Y D f.2    s .v/; : : : ; s .v/; v/t j v 2 ker.c1  n /g;
which immediately gives 1 .Y / D 0. Further by (9.3) we also have
t
.1  csn /X D .1  csn / ker.1  n /; : : : ; ker.s  n / D 0:

Thus, 1 is well-defined and both injective as well as surjective, hence an isomor-

phism. The compatibility with the group action follows since

1 .1  csn /2 .w C X C Y / D .c1  n /2 ws

D .1  n /1 .1  csn /.w C X C Y /:

t
u

The above rank formulas lead to an explicit description of the Jordan normal
forms of the images under convolution. They also form a crucial prerequisite in the
explicit examples computed in the subsequent paragraphs.
9 The Katz Algorithm 265

Corollary 9.13. Let k D kN be algebraically closed, . ; V / 2 Rep.kŒFs / absolutely

irreducible with s > 1, c 2 k  and  D PNc . /.
(a) Let diag.Ji;1 : : : ; Ji;ki / be the Jordan normal form of i with Jordan blocks
Ji;j of lengths nij and eigenvalues aij . Then the Jordan normal form
diag.JQi;1 ; : : : ; JQi;kQ / of i is obtained as follows: Every Jordan block Ji;j of i leads
i
to a Jordan block JQi;j of i of length mij with eigenvalues aQ ij D caij , where
8
ˆ
<nij if aij ¤ 1; 1c ;
mij D nij  1 if aij D 1;
:̂
nij C 1 if aij D 1c :

The other Jordan blocks of i have length 1 and eigenvalue 1.

(b) Let diag.J1;1 ; : : : ; J1;k1 / be the Jordan normal form of 1 D 1    s with
Jordan blocks J1;j of lengths n1j and with eigenvalues a1j . Then the Jordan nor-
mal form diag.JQ1;1 ; : : : ; JQ1;kQ1 / of 1 is obtained as follows: every Jordan block
J1;j leads to a Jordan block JQ1;j of length m1j and eigenvalue aQ 1j D ca1j ,
where 8
ˆ
<n1j if a1j ¤ 1; 1c ;
m1j D n1j C 1 if a1j D 1;
:̂
n1j  1 if a1j D 1c :
All other Jordan blocks of 1 have length 1 and eigenvalue c.
Proof. The assertion of Corollary 9.13 follows almost immediately from Proposi-
tion 9.12. For (a) one needs to observe that the lengths of Jordan blocks of i with
eigenvalue aij ¤ 1; 1c is preserved, while those for eigenvalues aij D 1 get shorter
by one due to the factorization by the eigenspace of i for the eigenvalue 1 (as part
of X ). Using PN1=c ı PNc D Id this implies that the lengths of Jordan blocks of i for
the eigenvalue aij D 1c are increased by 1.
In (b) again the lengths of Jordan blocks of 1 for eigenvalues a1j ¤ 1; 1c are
preserved. But here because of the factorization by the eigenspace of 1 for 1c
according to (9.5) the lengths of Jordan blocks for the eigenvalue 1c are decreased
by 1, and hence correspondingly increase by 1 for the eigenvalue 1. t
u
From Corollary 9.13 it is now possible relatively easily to derive the invariance
theorem of Katz (1996), Ch. 6.0, for the rigidity defect.
Theorem 9.14 (Katz). Let . ; V / 2 Rep.kŒFs / satisfy (H). Then for all c 2 kN  the
rigidity defect is invariant under PNc , that is, rid.PNc . // D rid. /.
Proof. For i 2 S WD f1; : : : ; s; 1g let aij 2 kN denote the eigenvalues of i 2 GLn .k/.
N
.`/
Further let ei .aij / be the number of Jordan blocks of length at least `. Then
CW .i / with W D k nn has dimension
X X .`/
dimk .CW .i // D ei .aij /2 ; (9.11)
`1 j
266 III Action of Braids

as is easily read off from the Jordan normal form. Correspondingly we denote by
eQi.`/ .aQ ij / the number of Jordan blocks of length at least ` of i with eigenvalue aQ ij
for i 2 S . According to Corollary 9.13 these satisfy
8 .`/
ˆ
<ei .aij / for aij ¤ 1; 1c and `  1;
.`/
eQi .caij / D ei.`C1/ .aij / for aij D 1 and `  1;
:̂ .`1/
ei .aij / for aij D 1c and `  2;

for i D 1; : : : ; s, respectively
8 .`/
ˆ
<e1 .a1j / for a1j ¤ 1; 1c and `  1;
.`1/
eQ1.`/
.ca1j / D e1 .a1j / for a1j D 1 and `  2;
:̂ .`C1/
e1 .a1j / for a1j D 1c and `  1:

From this we obtain

X .`/ .1/
X .`/ .1/
ei .aij /2  ei .1/2 D eQi .caij /2  eQi .1/2 DW pi ;
` `
X X
1 1 2
.`/
e1 .a1j /2  e1 . / D eQ1
.`/
.ca1j /2  eQ1
.1/
.c/2 DW p1 ;
c
` `

with the eigenspace dimensions ei.`/ .1/, eQi.1/ .1/, e1

.1/ 1 .1/
. c /, eQ1 .c/. So pi , p1 and
P
also p WD i 2S pi are invariant under the convolution PNc . From Proposition 9.12
we obtain moreover that
.1/ .1/
n  ei .1/ D m  eQi .1/ DW ri for i D 1; : : : ; s;
.1/ 1
n  e1 . / D m  eQ1
.1/
.c/ DW r1
c
P P
and from this the invariance of r WD i 2S ri and q WD i 2S ri2 . This translates to
the rigidity defect as
X X .`/
rid. /  2 D.s  1/n2  ei .aij /2
i 2S j;`

X .1/ s
.1/ 1 2
D.s  1/n2  p  e1 . /  ei .1/2
c
i D1
Xs
D.s  1/n2  p  .n  r1 /2  .n  ri /2
i D1
D  2n2  p C 2nr  q D 2n.r  n/  p  q:

This expression is clearly PNc -invariant, since on the one hand n.r  n/ D .m  r/m
by Proposition 9.2, and on the other hand the convolution PN1=c inverse to PNc inter-
changes the roles of 1c and c (at 1 respectively 1 ). t
u
9 The Katz Algorithm 267

For the Katz algorithm we now need one further operation on Rep.kŒFs /, the
multiplication with c D .c1 ; : : : ; cs / 2 .k  /s . This is defined by

Mc W Rep.kŒFs / ! Rep.kŒFs /; . ; V / 7! .Mc . /; V /; (9.12)

with Mc . / WD .c1 1 ; : : : ; cs s /.

Remark. Obviously Mc preserves irreducibility and absolute irreducibility of . ; V /.

The next result contains the core of the algorithm presented by Katz (1996),
Ch. 6.4, for the existence of linearly rigid s-tuples.

Theorem 9.15 (Katz). Let k D kN be an algebraically closed field. Then every irre-
N s with 1    s D n is connected
ducible linearly rigid s-tuple .1 ; : : : ; s / 2 GLn .k/
N N 
to an s-tuple in GL1 .k/ D k via iterative application of multiplication and convo-
lution.

Proof. Let V D k n and W D k nn with n  2. By assumption we have 1 D n .

Further we use the notation

ni WD minfrk.ci i  n / j ci 2 k  g:

If  D .1 ; : : : ; s / is linearly rigid then by Theorem 9.11 and (9.11) we have

X
s
sn2  2.n2  1/ D dimk .CW .i //
i D1
X s X
s X
s
.n  ni /2 C ni .n  ni / D n.n  ni / D sn2  n ni ;
i D1 i D1 i D1

from which it follows that

X
s X
s
n ni 2.n2  1/ and ni < 2n:
i D1 i D1

As . ; V / is irreducible, so is .Mc . /; V /, so by Corollary 9.10 the product c1    cs

must be different from 1. Hence by Proposition 9.2 the dimension m of the image
of the operator PN1=c ı Mc, for c D .c1 ; : : : ; cs / and c WD c1    cs , satisfies

X
s
1
mD rk.ci i  n / C rk. c1 1    s  n /  n
c
i D1
X
s
D ni  n < n:
i D1

Since moreover
rk.c N1  m / D rk.1  n / D 0
268 III Action of Braids

we also get c N1 D m for the next step in the induction. By decreasing dimension
this stops after finitely many steps with n D 1. t
u
Remark. Corollary 9.13, Theorem 9.14 and Theorem 9.15 also hold for not nec-
essarily algebraically closed fields k, as long as these contain all eigenvalues of
N
1 ; : : : ; s in k.

9.5 Braid Compatibility

In the next result we show that convolution commutes with the action of the Artin
braid group on the generators i of Fs in Theorem 1.2 and hence also with the action
on its images . ; V / 2 Rep.kŒFs / up to isomorphism.
Theorem 9.16. Let . ; V / 2 Rep.kŒFs / and ˇ 2 BQ s an element of the full Artin
braid group. Then for c 2 k  we have:
(a) Pc . ˇ ; V / Š .Pc . /ˇ ; V s /,
(b) PNc . ˇ ; V / Š .PNc . /ˇ ; W /.
Proof. The proof relies on a generalized Burau representation D W BQ s ! GLns .k/
of the braid group BQ s , which is given by
 
0 i
D.ˇi / D diag.n ; : : : ; n ; Di ; n ; : : : : ; n / with Di D
1 1  i C1

on the generators ˇi . Indeed, direct calculation shows that

Pc . /ˇi D D.ˇi /1 Pc . ˇi /D.ˇi /;

and thus (a) holds. Part (b) follows from this by observing that D.ˇi /X D X and
D.ˇi /Y D Y . t
u
In the case i    s D 1 by Theorem 1.7 we have moreover an action of the full
Hurwitz braid group HQ s on the generators of the factor group Gs D Fs =h1    s i
modulo simultaneous conjugation. Obviously this satisfies:
Proposition 9.17. The Pochhammer transformation Pc and the convolution PNc both
commute with simultaneous conjugation in GLn .k/ and in GLsn .k/ respectively
GLm .k/.
With the proof of Theorem 9.16 one thus sees that in the case 1 D n both Pc
and PNc commute with the action of the full Hurwitz braid group HQ s . From this we
conclude:
Corollary 9.18. Let  D .1 ; : : : ; s / 2 GLn .k/s with 1    s D n and ˇ 2 HQ s an
element of the full Hurwitz braid group. Then in GLsn .k/ respectively GLm .k/ we
have, modulo simultaneous conjugation:

ŒPc . ˇ / D ŒPc . /ˇ respectively ŒPNc . ˇ / D ŒPNc . /ˇ :

9 The Katz Algorithm 269

Thus convolution PNc sends braid orbits of s-tuple classes Πin GLn .k/ to such
in GLm .k/. If G GLn .k/ is a subgroup, then under suitable additional stabil-
ity hypotheses this will also send braid orbits of classes of generating systems in
˙s .G/= Inn.G/ (see Proposition 5.1) to corresponding braid orbits of subgroups of
GLm .k/. We will see stringent applications for this in Sections 10.3 and 10.4.
To close this section we formulate two further properties of the convolution PNc
in the form of exercises, since they will not be needed here.

Exercise 9.19. Let . ; V / 2 Rep.kŒFs / with  2 GLn .k/s , and c 2 k  .

(a) The dual representation .  ; V  / of . ; V / satisfies

PNc .  ; V  / Š PNc . ; V / :

(b) The representation . 1 ; V / with the inverse group action satisfies

PNc . 1 ; V / Š .PN1=c . 1 /; V /:

(See Dettweiler and Reiter (2000), Thms. 5.4 and 5.5.)

270 III Action of Braids

10 Applications of the Katz Algorithm

In this paragraph we present several applications of Katz’s algorithm. In particular

we will find very general G-realizations over Q for many classical groups over IFq
even for higher prime powers q. The results presented here originate mainly in the
papers of Dettweiler and Reiter (1999, 2000), and partly also go back to Völklein
(1993, 1998) and Thompson and Völklein (1998) but who did not use the Katz
algorithm.

10.1 Jordan–Pochhammer Tuples

The simplest non-trivial linearly rigid s-tuples in GLn .k/ are the classical Jordan–
Pochhammer tuples. They are obtained for n  3 by a single application of the con-
volution to an n-tuple a D .a1 ; : : : ; an / in GL1 .k/ Š k  . As here we are interested
in finite groups we will use the base field k D QIFq for a prime power q.1 We choose
1 ¤ ai 2 k  for i D 1; : : : ; n and set a1 WD niD1 ai , so that a1    an a1 D 1. Let
c 2 k  . Then

 D .1 ; : : : ; n / WD Pc .a1 ; : : : ; an / 2 GLn .k/n (10.1)

1
together with 1 D 1    n and 0 D 1 is called a (normalized) Jordan–Poch-
hammer tuple, or JP-tuple for short. For c … f1; a0 g we call  an irreducible Jordan–
Pochhammer tuple. (In Völklein (1998) this is also called a Thompson tuple.)

Remark. (a) The elements 1 ; : : : ; n are pseudo-reflections with characteristic poly-

nomial fi .X / D .X  1/n1 .X  cai / for i D 1; : : : ; n. The product 1 is a c-fold
pseudo-reflection with characteristic polynomial f1 .X / D .X  c/n1 .X  ca1 /.
(b) For 1 i n, in the case cai ¤ 1; 1 the element i is a non-involutory
homology, in the case cai D 1 it is a transvection, and in the case cai D 1 a
reflection.
Up to the form of f1 .X / these claims follow immediately from the definition
in (10.1), for f1 .X / one can use in addition Corollary 9.13(b).

Proposition 10.1. Let .1 ; : : : ; n / D Pc .a1 ; : : : ; an / 2 GLn .k/n , n  2, be an irre-

ducible Jordan–Pochhammer tuple and G WD h1 ; : : : ; n i the subgroup generated
by it. Then:
(a) G is irreducible and  D .0 ; : : : ; n / is linearly rigid.
(b) In the case n > 3 the group G is moreover primitive.
1
Proof. Clearly .a0 ; : : : ; an / 2 GL1 .k/nC1 with a0 D a1 is linearly rigid, and the
subgroup ha1 ; : : : ; an i GL1 .k/ is absolutely irreducible. By Proposition 9.2 we
have Pc D PNc as c … f1; a0 g. Since n  2 hypothesis (H) holds by Proposition 9.4(b).
Thus  D .0 ; : : : ; n / is linearly rigid by Theorem 9.14 and . ; V / is irreducible by
Corollary 9.7 where V D k n .
10 Applications of the Katz Algorithm 271
L
For the proof of (b) we assume that V D m i D1 Vi is a direct sum of subspaces
Vi of dimension l that are transitively permuted by G. In the case l > 1 we then
have V1 \ ker.i  n / ¤ 0 for all i , from which we get i .V1 / V1 and by the
irreducibility of G then m D 1. In case l D 1 we have m D n and the action of G
on the set of Vi ’s induces a permutation representation n W G ! Sn . If the image
n .i / is non-trivial for some generator i , then after renumbering we may assume
that i .V1 / D V2 , say. But as V1 ˚ ker.i  n / D V D V2 ˚ ker.i  n / there exist
subspaces Wi ker.i  n / with V1 ˚ V2 D V1 ˚ W1 D V2 ˚ W2 . Then

i .V1 ˚ V2 / i .V1 / C i .W1 / D V2 C W1 V1 ˚ V2

shows the i -invariance of V1 ˚ V2 . Thus i .V1 / D V2 , i .V2 / D V1 and i .Vj / D

Vj for j  3. In particular n .i / D .1 2/ is a transposition. The same holds for
1 using ker.c1  n /. So n .G/ Sn is a transitive group generated by n C 1
transpositions with product the identity. In the case n > 3 this contradicts the
Hurwitz relative genus formula (see Section I.5.2). So G is primitive. t
u
The next result of Völklein (1998) classifies the finite groups generated by irre-
ducible JP-tuples for n > 8.
Theorem 10.2. Let .1 ; : : : ; n / D Pc .a1 ; : : : ; an / be an irreducible JP-tuple over
IFq D IFp .a1 ; : : : ; an ; c/ and G D h1 ; : : : ; n i GLn .q/. Then for n > 8 we have:
(a) G leaves invariant a non-trivial bilinear form on . ; IFnq / if and only if q D p
is an odd prime, n D 2m is even and a1 D : : : D an D c D 1. In this case G Š
Sp2m .p/.
(b) G leaves invariant a non-trivial Hermitian form on . ; IFnq / if and only if q D
qQ is a square and the norms satisfy NIFq =IFqQ .ai / D NIFq =IFqQ .c/ D 1 for 1 i n.
2

In this case SUn .q/ Q G GUn .q/.Q

(c) If we are neither in case (a) nor (b) then SLn .q/ G GLn .q/.
Proof. We start with two general remarks: an automorphism ˛ W a 7! aN of k D IFq
of order 1 or 2 can be extended to an automorphism of GLn .k/ via

˛  W  7!   WD .N 1 /t : (10.2)

It transforms a JP-tuple  D .1 ; : : : ; n / WD Pc .a1 ; : : : ; an / to another JP-tuple

˛  . / D .1 ; : : : ; n / D PcN 1 .aN 11 ; : : : ; aN n1 / with 1

D .1 / . Note that any non-
trivial G-invariant bilinear form on . ; V /, with V D IFnq , is non-degenerate due to
the irreducibility of . ; V /.
Now assume that . ; V / carries a non-degenerate bilinear form. Then from the
form of 1 ; : : : ; n we obtain on the one hand that cai D ˙1, and from the form of
1 also c D ˙1. Since c ¤ 1 this shows that c D 1 and ai D 1. So 1 ; : : : ; n
are transvections in GLn .p/ with p ¤ 2. As det.i / D 1 we also have det.1 / D 1
and hence n D 2m is even. By the Theorem II.2.2 of Kantor this forces that G D
Sp2n .p/, showing (a).
Now assume that . ; V / carries a non-degenerate Hermitian form. Then q D qQ 2
is a square and G is a subgroup of GUn .q/. Q By Huppert (1967), II, 10.5, the latter
272 III Action of Braids

is equivalent to i D i for i D 1; : : : ; n (up to simultaneous conjugation in GLn .q/).

As above this implies that cN aN i D .cai /1 as well as cN D c 1 and so aN i D ai1 . This
leads to the necessary and sufficient condition NIFq =IFqQ .ai / D 1 for i D 1; : : : ; n and
NIFq =IFqQ .c/ D 1, which shows the first part of (b).
For the last step we assume that . ; V / has no non-trivial invariant bilinear form,
but possibly an invariant Hermitian form. If 1 ; : : : ; n are reflections, which can
only happen when p ¤ 2, then cai D 1 for i D 1; : : : n. Then the Theorem II.2.4
of Wagner shows that for n > 8 we either have SUn .q/ Q G GUn .q/ Q or SLn .q/
G GLn .q/. Indeed, the remaining case (d) in the cited result can not occur here, as
SnC1 and SnC2 cannot be generated by nC1 transpositions with product 1 (compare
Proposition II.2.8).
If not all of the i are reflections then there is either a transvection or a non-
involutory homology among the i . Now let N denote the normal subgroup of G
generated by the conjugates in G of this element. Then .N; V / is irreducible, as
G is primitive, hence also primitive according to the proof of Proposition 10.1(b).
So assertions (b) and (c) hold for N in place of G. In the case of a non-involutory
homology this follows from the Theorem II.2.3 of Wagner, and in the case of a
transvection from the Theorem II.2.2 of Kantor, where in characteristic 2 the groups
SnC1 and SnC2 can be excluded as above. As N E G we finally deduce (b) and (c)
also for G (using the characterization of unitary groups as the fixed points under
˛  ). t
u
Since there are no irreducible JP-tuples over k D IF2 , the theorems of Kantor and
Wagner allow us to give the following more precise result:
Remark. (a) If G is not generated by reflections then Theorem 10.2 holds when
n  4.
(b) If G is generated by non-involutory homologies, then the assertion of Theo-
rem 10.2 even holds for n  3.
We reformulate the special case (a) in Theorem 10.2 as an existence result (com-
pare also Prop. III.10.5 in the first edition Malle and Matzat (1999)):
Corollary 10.3. For odd primes p 2 IP and m  2 the groups S2m .p/ possess G-
p p1
realizations over Q. p  /, with p  D .1/ 2 p, with respect to suitable rigid gen-
erating systems.
Proof. By the proof of Theorem 10.2(a) the elements 1 ; : : : ; n are transvections
generating Sp2m .p/. Moreover  D .0 ; : : : ; n / is linearly rigid, where 0 is also
a transvection according to the part (b) of the Remark preceding Proposition 10.1.
It is wellpknown that Sp2m .p/ contains two classes of transvections with charac-
ter field Q. p  /. Both of them are GL-stable and are interchanged by CSp2m .p/.
Consequently  is rigid by Proposition 9.8 and thus with the Basic Rigidity Theo-
remp I.4.8 leads to a G-realization of S2m .p/ D Sp2m .p/=Z .Sp2m .p// over
Q. p  /. t
u
This result will be extended considerably to certain prime powers in Theo-
rem 10.9.
10 Applications of the Katz Algorithm 273

Remark. The Classification Theorem 10.2 for irreducible JP-tuples with its applica-
tions also holds with m D n T1 in place of n for the reducible case 1 ¤ c D a0 , since
then W D IFnq =Y with Y D niD1 ker.i  n / is irreducible in dimension n  1.

10.2 Linear and Unitary Groups

In this section we start from modified Jordan–Pochhammer tuples to construct G-

realizations over Q for linear and unitary groups. These results were first obtained
by Völklein (1993) and Dettweiler and Reiter (1999) but with different proofs.

Theorem 10.4. Let p 2 IP be an odd prime, q D p e ¤ 3 and n > '.q  1/. Then
GLn .q/ and PGLn .q/ possess G-realizations over Q.

Proof. We start with the case that n D 2m is even. Then from n D 2m > '.q  1/ D
2l with l  1 we first get m  2 and thus n  4. Now let a 2 IFq be a generator of IF q
and a1 D a; : : : ; a2l the primitive powers of a. We let a D .a1 ; : : : ; a2l ; 1; : : : ; 1/ 2
IF2mC1
q . Then a1    a2l D 1 and a1 D 1. With Proposition 9.2 the convolution PN1
yields
PN1 .a1 ; : : : : ; a2mC1 / D .1 ; : : : ; 2mC1 / 2 GL2m .q/2mC1
with 1 D 2m . According to Corollary 9.13 the 1 ; : : : ; 2l are homologies with
non-trivial eigenvalues a1 ; : : : ; a2l respectively, and 2lC1 ; : : : ; 2mC1 are trans-
vections. Furthermore, the group G D h1 ; : : : ; 2mC1 i is irreducible by Corol-
1
lary 9.7 and Theorem 9.14, and  D .0 ; : : : ; 2mC1 / with 0 D 1 D 2m
is linearly rigid. From Theorem 10.2(c) with the subsequent Remarks concern-
ing homologies and the case c D a0 we thus obtain that G D GLn .q/ since
det.1 / D a generates IF q . The trivially satisfied stability condition in Proposi-
tion 9.8(b) implies the rigidity of  . The conjugacy classes Ci of i for i D 1; : : : ; 2l
are permuted by the cyclotomic character according to Corollary I.2.7, while those
for i D 2l C 1; : : : ; 2m C 1 are rational. Let V S2l denote the corresponding per-
mutation group, then ˙V .C0 ; : : : ; C2mC1 / is V -symmetric. Consequently by Theo-
rem 3.11 the field of definition KV of NN  from Theorem 3.4 (Hurwitz classification)
is regular over Q and rational. Thus we have KV D Q.v; t/ in the notation of Theo-
rem 3.11. So first PGLn .q/ D GLn .q/=Z .GLn .q// has a G-realization over Q. Now
by Theorem II.1.4 the center Z .GLn .q// has a complement in NGLn .q/ .h2mC1 i/.
This yields the normalizer condition (N) for the Galois extension NN  Q.v/=Q.v/.t/
of algebraic function fields in one variable over Q.v/ with the Galois group GLn .q/
and the field of definition KV D Q.v; t/. Thus by Theorem I.3.9 also GLn .q/ has a
G-realization over Q.
Now we consider the case when n D 4m C 1. Here we extend the 2m C 1-tuple
.1 ; : : : ; 2mC1 / 2 GL2m .q/2mC1 introduced in the first part to .Q 1 ; : : : ; Q 2mC3 / via
Q 2mC2 D Q 2mC3 WD 2m . Then we again have Q 0 D Q 1 D 2m , and
274 III Action of Braids

.Q 0 ; : : : ; Q 2mC3 / is linearly rigid. From Proposition 9.2 we get with the convolution
PN1
PN1 .Q 1 ; : : : ; Q 2mC3 / D .1 ; : : : ; 2mC3 / 2 GL4mC1 .q/2mC3
with 1 D 4mC1 D 0 . Here according to Corollary 9.13 the elements 1 ; : : : ; 2l
are non-involutory homologies with non-trivial eigenvalues a1 ; : : : ; a2l respectively,
the elements 2lC1 ; : : : ; 2mC1 are reflections, and the Jordan normal forms of
2mC2 and 2mC3 consist of 2m Jordan blocks of length 2 and one of length 1,
all with eigenvalue 1. By Theorem 9.14  D .0 ; : : : ; 2mC3 / is linearly rigid and
H WD hi is irreducible by Corollary 9.7, hence also primitive by the proof of
Proposition 10.1(b). Now let N E H be the normal subgroup generated by the
non-involutory homologies. Then N is itself primitive and does not leave invariant
any non-trivial Hermitian form. So N D GLn .q/, with n D 4m C 1, by the The-
orem II.2.3 of Wagner, using that det.1 / D a generates IF q . Thus we also have
H D GLn .q/. Since both the stabilizer condition (S) (by Proposition 9.8) and the
normalizer condition (N) for 2mC1 (by Theorem II.1.4) are satisfied, a symmetriza-
tion as above yields G-realizations of PGLn .q/ and GLn .q/ over Q for n D 4m C 1.
It remains to consider the case n D 4m C 3 with n  7. Here we start with
.a1 ; : : : ; a2l ; 1; : : : ; 1/ 2 IF2mC3
q , where a1 ; : : : ; a2l are chosen as in the first case,
whence a1 D 1. The convolution PN1 now leads to

PN1 .a1 ; : : : ; a2mC1 / D .1 ; : : : ; 2mC3 / 2 GL2mC2 .q/2mC3

with 1 D 2mC3 D 0 . Again, 1 ; : : : ; 2l are non-involutory homologies, while

2lC1 ; : : : ; 2mC3 are now transvections. Here we set Q 2mC2 D 2mC2 , Q 2mC3 D
2mC3 and replace .1 ; : : : ; 2mC3 / by .1 ; : : : ; 2mC1 ; Q 2mC2 ; Q 2mC3 /. Then again
Q 1 D 2mC2 . Now according to Proposition 9.2 application of PN1 yields

PN1 .Q 1 ; : : : ; Q 2mC3 / D .1 ; : : : ; 2mC3 / 2 GL4mC3 .q/2mC3

with 1 D 4mC3 . Here, as above, 1 ; : : : ; 2l are non-involutory homologies

and 2lC1 ; : : : ; 2mC1 are reflections. With the arguments from the previous two
cases one obtains again that the group H D h1 ; : : : ; 2mC3 i is the whole group
GL4mC3 .q/ and that  D .0 ; : : : ; 2mC3 / is linearly rigid as well as rigid. From
this, the existence proof for G-realizations of PGL4mC3 .q/ and GL4mC3 .q/ can be
completed as above using symmetrization. t
u

Remark. For odd n Theorem 10.4 continues to hold for q D 3 with l D 0; for even
n and l D 0 the above construction leads to the group SLn .3/ instead of GLn .3/.

Theorem 10.5. Let p 2 IP be odd, q D p e and n > '.q C 1/. Then GUn .q/ and
PGUn .q/ possess G-realizations over Q.

Proof. As in the proof of Theorem 10.4 we start with the case that n D 2m is
even, with n > '.q C 1/ D 2l, where l  1 and thus n  4. Now let a 2 IF q2
be
a generator of the cyclic subgroup of order q C 1, and a1 D a; : : : ; a2l its primitive
powers. Then NIFq2 =IFq .ai / D NIFq2 =IFq .1/ D 1 for all i . Application of PN1 to
10 Applications of the Katz Algorithm 275

.a1 ; : : : ; a2l ; 1; : : : ; 1/ 2 IF2mC1

q2
with a1 D 1 then yields

PN1 .a1 ; : : : ; a2mC1 / D .1 ; : : : ; 2mC1 / 2 GL2m .q 2 /2mC1

with homologies 1 ; : : : ; 2l , transvections 2lC1 ; : : : ; 2mC1 , and 1 D 2m . By

the Theorem of Katz the tuple  D .0 ; : : : ; 2mC1 / is linearly rigid and generates
an irreducible subgroup G of GL2m .q 2 /. By Theorem 10.2(b) in the case c D a0 we
thus have G D GU2m .q/. The stability condition NGLn .q 2 / .G/ D Z .GLn .q 2 // 
G in Proposition 9.8(b) is satisfied according to the proof of Theorem II.3.2.
Thus  D .0 ; : : : ; 2mC1 / is rigid in G. The cyclotomic character permutes the
conjugacy classes Ci D Œi for i D 1; : : : ; 2l, and fixes C0 ; C2lC1 ; : : : ; C2mC1 .
So ˙V .C0 ; : : : ; C2mC1 / becomes V -symmetric for the corresponding permuta-
tion group V S2l , whence the field of definition KV of NN  is regular over
Q and rational, say KV D Q.v; t/. Consequently PGU2m .q/ D G=Z .G/ pos-
sesses a G-realization over Q. According to the proof of Theorem II.3.2 we have
NG .h2mC1 i/ D Z .G/h2mC1 i and hence the normalizer condition (N) for
NN  Q.v/=Q.v/.t/. So GU2m .q/ also has a G-realization over Q (compare the proof
of Theorem 10.4).
Now let n D 4m C 1 and thus n  5. In analogy to what we did in the proof of
Theorem 10.4 we extend .1 ; : : : ; 2mC1 / from the first case to .Q 1 ; : : : ; Q 2mC3 / with
Q 2mC2 D Q 2mC3 D 2m and thus Q 1 D 2m (and Q i D i else). Application of
PN1 yields

PN1 .Q 1 ; : : : ; Q 2mC3 / D .1 ; : : : ; 2mC3 / 2 GL4mC1 .q 2 /2mC3

with homologies 1 ; : : : ; 2l with non-trivial eigenvalues a1 ; : : : ; a2l respectively,

reflections 2lC1 ; : : : ; 2mC1 , and unipotent elements 2mC2 ; 2mC3 with Jordan nor-
mal form containing one block of size 1 and all others of size two, as in the corre-
sponding case of Theorem 10.4. Here 1 D 4mC1 D 0 . Then  D .0 ; : : : ; 2mC3 /
is linearly rigid and H WD hi is an irreducible subgroup of GL4mC1 .q 2 /. As i D i
for all i by the choice of the ai and of c, H leaves invariant a Hermitian form, and
since a generates IFq 2 over IFp , we have H D GU4mC1 .q/. As above  is rigid in H
since it moreover satisfies the stability condition. So we can complete the proof as
in the first case to obtain G-realizations over Q for PGU4mC1 .q/ and GU4mC1 .q/.
In the last case n D 4m C 3 we have n  7. Here we start with the tuple
.a1 ; : : : ; a2l ; 1; : : : ; 1/ 2 IF2mC3
q2
with ai as in the first case and a1 D 1. From
this we obtain

PN1 .a1 ; : : : ; a2mC3 / D .1 ; : : : ; 2mC3 / 2 GL2mC2 .q 2 /2mC3

with 1 D 2mC2 . As in the corresponding part of the proof of Theorem 10.4 we

modify the i to .Q 1 ; : : : ; Q 2mC3 / with Q 2mC2 D 2mC2 , Q 2mC3 D 2mC3 and
Q i D i else. A second application of PN1 then yields

PN1 .Q 1 ; : : : ; Q 2mC3 / D .1 ; : : : ; 2mC3 / 2 GL4mC3 .q 2 /2mC3

276 III Action of Braids

with 1 D 4mC3 D 0 . As above we may conclude that  D .0 ; : : : ; 2mC3 / is

linearly rigid and generates H D GU4mC3 .q 2 /. The stability condition already ver-
ified above then yields the rigidity of  in H . From here on the remaining assertions
for n D 4m C 3 can be deduced as in the previous two cases. t
u

10.3 Symplectic Groups

In this section we present examples of G-realizations over Q from braid rigid gen-
erating systems which are not rigid. Here the braid compatibility of the convolution
plays a crucial role. The results presented here essentially go back to Thompson and
Völklein (1998) and Dettweiler and Reiter (2000). We start with some technical
preparations.

Proposition 10.6. Let Q 2 k nn be a matrix and h 2 GLn .k/ be such that ht
Qh D Q for h D 1; : : : ; s. Then for all c 2 k  we have

Pc 1 .h /t QPc .h / D Q for Q D .Qij /si;j D1 2 k snsn

with 8
ˆ
<c
1=2
Q.i1  n /.j  n / i <j
1=2
Qij D c Q.i1  cn /.j  n / for i D j
:̂ 1=2
c Q.i1  n /.j  n / i > j:

Proof. The stated formulas are easily verified by solving the associated linear sys-
tem of equations. t
u

Corollary 10.7. If in Proposition 10.6 we additionally have that Qt D Q and

c D 1, then
Qt D ˙Q for Q D .Qij /:
So if h1 ; : : : ; s i is an orthogonal (respectively symplectic) group, then P1 . / and
PN1 . / generate subgroups of a symplectic (respectively orthogonal) group.

Proof. In the special case c D 1 with Qt D Q Proposition 10.6 first yields

t
Qij D ˙Qij and from this

Qt D .Qij /t D ˙.Qij / D ˙Q: t

u

Proposition 10.8. Let G D D2m < GL2 .C/ be a dihedral group of order 4m in its
natural reflection representation, C D .C1 ; : : : ; Cs / a class vector of G with classes
C1 ; C2 of reflections, C3 ; : : : ; ClC3 of non-involutory bihomologies and ClC4 ; : : : ; Cs
the class of the central involution 2 , satisfying ˙.C/ ¤ ;. Then we have:
(a) j˙.C/= Inn.G/j D 2l .
(b) The pure Hurwitz braid group Hs acts sharply transitively on ˙.C/= Inn.G/
with im.Hs / Š Z2l . This is generated by the images ˇN24 ; : : : ; ˇN2;lC3 of ˇij 2 Hs .
10 Applications of the Katz Algorithm 277

(c) The fixed field LN s in MN s =Q.t

N 1 ; : : : ; ts / of the kernel of the above action is
generated by
tj  t1 t3  t1
uj for j D 4; : : : ; l C 3 with u2j D W :
tj  t2 t3  t2

In particular LN s =Q
N is a rational function field.

Proof. The group G D D2m is the semidirect product G D T Z2 of a diagonal

1
torus T of order 2m containing elements of the  diag.a; a / by a cyclic
 form
01
group Z2 generated by a reflection , say  D . If now there exists some
10
.1 ; : : : ; s / 2 ˙.C/ then due to 1    s D 2 all of 1 2 ; 3 ; : : : ; s lie in T and
1 2 is a square in G if 3    s is a square in T . So the latter is a necessary and
sufficient condition for ˙.C/ ¤ ;.
The conjugacy classes of a non-involutory bihomology  D diag.a; a1 / consists
of  and its inverse. So for the computation of j˙.C/= Inn.G/j we may arrange by
conjugation that 1 D  and 3 2 C3 . Then 2 D 11 3    s is a reflection, and
we have exactly two choices for i 2 Ci for each 4 i l C 3. This yields 2l
possibilities, giving j˙.C/= Inn.G/j D 2l as claimed.
The action of Hs on ˙.C/= Inn.G/ is obtained from Proposition 5.1. Let’s first
observe that ˇ12 D ˇ12 acts on all i by conjugation from the left with 1 2 2 T , i.e.,
ˇ12 acts trivially on ˙.C/= Inn.G/, so that ˇN12 D id. Since T is abelian, the braids
ˇij for 3 i < j s fix all classes of generating systems Œ 2 ˙.C/= Inn.G/,
whence ˇNij D id for these i; j as well. The braids ˇij for i D 1; 2 and 3 j s
all act on j by conjugation with i . So the action is transitive and we have
ˇN1j D ˇN2j as well as ˇN2j 2
D id for j l C 3 and ˇNij D id else. Finally, the
sphere relation ˇN12 ˇN23 ˇN24    ˇN2;lC3 D id from Section 1.2 yields the product rela-
tion ˇN23 ˇN24    ˇN2;lC3 D id. So the image Hs acts by

HN s D hˇN23 ; : : : ; ˇN2;lC3 j ˇN2j

2
D id; ˇN23    ˇN2;lC3 D idiab Š .ZZ=2ZZ/l ;

since all generators ˇN2j commute.

For the proof of (c) we may assume that s D l C 3. Then HN s D 1 for s D 3 and
N N 1 ; t2 ; t3 /. For s D 4 the sphere relation gives that
L D Q.t

HN 4 D hˇN14 ; ˇN24 j ˇNi24 D id; ˇN14 ˇN24 D idiab Š ZZ=2ZZ;

that is, LN 4 =LN 3 .t4 / has degree 2 and ramifies in c tt44 t N

t2 with c 2 L3 . From the sphere
1

relations we further deduce that ˇN13 D ˇN23 D ˇN24 ¤ id with ˇN13 ˇN23 D id. So by
symmetry LN 4 =Q.t N 1 ; : : : ; t4 / also ramifies in t3 t1 . This shows that
t3 t2

t4  t1 t3  t1
LN 4 D LN 3 .t4 ; u4 / with u24 D W
t4  t2 t3  t2
278 III Action of Braids

after a normalization according to the sphere relation ˇN23 ˇN24 D id. Since we have
N 1 ; t2 ; t3 ; u4 / the extension LN 4 =Q
t4 2 Q.t N is moreover rational. The claim for s > 4
now follows by induction over s and permutation of the variables. t
u

After these preparations we now come to the main result of this section:

Theorem 10.9. Let p 2 IP be odd, q D p e and 2m > q. Then:

(a) If e is even then the symplectic group S2m .q/ has a p
G-realization over Q.
(b) If e is odd then S2m .q/ has a G-realization over Q. p  /.

Proof. We start with the case that q 1 .mod 4/, which implies in particular that
q  5. Then the dihedral group Dq1 of order 2.q  1/ is isomorphic to the gen-
eral orthogonal group GOC 2 .q/. Further, '.q  1/ D 2l is even with l  1. Let
a 2 IF q be a generator and a1 D a; : : : ; a2l its primitive powers, with alCi D ai
1

for 1 i l. Since 2m > q and '.q  1/ q  2 we have m  l C 2. Now choose

mC2
.1 ; : : : ; mC2 / 2 Dq1 with reflections 1 ; 2 , a bihomology 3 of order 4, repre-
sentatives 4 ; : : : ; lC3 of the different classes of bihomologies diag.ai ; ai1 /, and
lC4 D : : : D mC2 D 2 . Then after a suitable choice of the parity of the reflec-
tions according to Proposition 10.8 we may attain 1 D 2 , and the corresponding
class Œ D Œ1 ; : : : ; mC2 ; 1 of generating systems of Dq1 lies in the unique
Hs -braid orbit of length 2l for s D m C 3.
With the convolution PN1 according to Proposition 9.2 we obtain

PN1 .1 ; : : : ; mC2 / D .1 ; : : : ; mC2 / 2 GL2m .q/mC2 :

Here by Corollary 9.13 the elements 1 ; 2 are transvections, 3 is a bihomology of

order 4, 4 ; : : : ; lC3 are bihomologies of order q  1, lC4 ; : : : ; mC2 are bitransvec-
tions and 1 D 2m . By Corollaries 9.7 and 10.7 the group H generated by the
i is an irreducible subgroup of Sp2m .q/. Now let N E H be the normal sub-
group generated by the transvections in H . Then H itself is also irreducible and
by the Theorem II.2.2(a) of Kantor isomorphic to Sp2m .q/ Q for some qjq. Q But then
H D Sp2m .q/, since for example the traces ai C ai1 generate IF q.
Now denote by Ci WD Œi for i D 1; : : : ; m C 2 the conjugacy class of i in H ,
1
and Cs D Œ1 D Œ2m . Then C D .C1 ; : : : ; Cs / is GL-stable, since CSp2m .q/
interchanges the two classes of transvections and leaves invariant all of the other
conjugacy classes. By Corollary 9.18 the convolution PN1 then transfers the Hs -
braid orbit of Œ to a corresponding Hs -braid orbit in ˙.C/= Inn.H /, and due to
the invertibility of PN1 we moreover have j˙.C/= Inn.H /j D 2l . Now choose  2
˙.C/. Then by Proposition 10.8

K D QC .t1 ; t2 ; t3 ; u4 ; : : : ; ulC3 ; tlC4 ; : : : ; tmC2 ; t/ DW QC .t_ ; u/

1
with QC D QC1 .wq1 C wq1 /, where wq1 denotes a primitive .q  1/st root of
p
unity and QC1 D Q, respectively QC1 D Q. p  /, according to whether C1 is ratio-
nal (if e is even) or non-rational (when e is odd) (see e.g. Thompson and Völklein
(1998), 1.4 and 1.5). The cyclotomic character permutes the conjugacy classes
10 Applications of the Katz Algorithm 279

C4 ; : : : ; ClC3 . Let the corresponding permutation group V Sl act on t4 ; : : : ; tlC3

as well as on u4 ; : : : ; ulC3 in the same way and let U be the QC .t_ /-vector space

U D QC .t_ /hu4 ; : : : ; ulC3 i

with the lifted action of the cyclotomic character on the ui . Then by Propo-
L
sition 3.10 (Lemma of Speiser) U has a QC1 .t_ /-form UQ D lC3 _
i D4 QC1 .t /vi ,
and the V -symmetrized fixed field KV WD QC1 .t_ ; v/ of NN  is regular and ratio-
nal over QC1 . So by the Rigid Braid Orbit Theorem (Corollary 5.8), S2m .q/ D
Sp2mp .q/=Z .Sp2m .q// has a G-realization over QC1 , with QC1 D Q or QC1 D
Q. p  / according to the parity of e.
In the second case q 3 .mod 4/ with q  3 we proceed accordingly, now start-
ing from the dihedral group DqC1 D GO 2 .q/ GL2 .q/ and a generating element a
of a cyclic subgroup of order q C 1 in IF
q2
and using l D 12 '.q C 1/ and m  l C 1.
Observe that in this case again the subgroup H GL2m .q/ generated by  is iso-
morphic to Sp2m .q/. t
u

Remark. Observe that the second symmetrization p ! used in Dettweiler and Reiter
(2000) in order to get rid of the irrationality p  which interchanges t1 and t2 is
not independent of the V -symmetrization used here, as !.uj / D u1 j , so that the
Lemma of Speiser cannot be applied.

10.4 Orthogonal Groups

Using the transformation rules for the convolution PN1 in Corollary 10.7 it is possi-
ble in suitable circumstances to construct G-realizations of orthogonal groups start-
ing from G-realizations of symplectic groups. Examples for this, resulting from The-
orem 10.9, are collected in the present section. These have their origin in Dettweiler
and Reiter (2000). We start with the case of odd-dimensional orthogonal groups.

Theorem 10.10. Let n D 2m C 1 and q D p e be odd with m > q. Then SOn .q/
possesses a G-realization over Q.

Proof. As in the proof of Theorem 10.9 we first treat the case when q 1 .mod 4/.
For this we start with the variation

.Q1 ; : : : ; QmC2 / WD .1 ; 2 ; 3 ; : : : ; mC1 ; mC2 / 2 Sp2m .q/mC2

of the .m C 2/-tuple .1 ; : : : ; mC2 / constructed there, with Q1 D 1 D 2m . Then
the convolution PN1 by Proposition 9.2 gives

PN1 .Q1 ; : : : ; Q2mC2 / DW .1 ; : : : ; mC2 / 2 GL4m1 .q/mC2 :

By Corollary 9.13 here 1 is a reflection, 3 a rational bihomology of order 4,

4 ; : : : ; lC3 are bihomologies of order q  1, lC4 ; : : : ; mC1 are bireflections, 2
280 III Action of Braids

and mC2 are unipotent with Jordan normal form as in Corollary 9.13(a), and 1 D
4m1 . The group H generated by the i is an irreducible subgroup of GO4m1 .q/
by Corollaries 9.7 and 10.7. According to the proof of Proposition 10.1(b) then H
as well as its normal subgroup generated by its reflections are both primitive. But
then by the Theorem II.2.4 of Wagner et al. we have

GO4m1 .q/  H  N  ˝4m1 .q/

Q Q
with qjq:

Since the traces of 4 ; : : : ; lC3 generate IFq over IFp and det.1 / D 1, spin.4 / …
.IFq / , we hence have H D GO4m1 .q/. Now let Ci WD Œi for i D 1; : : : ; m C 2
2
1
denote the conjugacy class of i in H , and Cs WD Œ1 D Œ4m1 for s D m C 3.
Then by Table 1.1 in Chapter II the class vector C D .C1 ; : : : ; Cs / is GL-stable.
According to Corollary 9.18 the Hs -action on the braid orbit of Œ1 ; : : : ; s is trans-
ported by PN1 isomorphically onto ˙.C/= Inn.H /. For  2 ˙.C/ we obtain the
field of definition

K D QC .t_ ; u/ with 1
QC D Q.wq1 C wq1 /

N
of NN  =Q.t/, in the notation of the proof of Theorem 10.9, since each of the classes
C1 ; C2 ; C3 ; ClC4 ; : : : ; Cs is rational. V -symmetrization exactly as in the symplectic
case then leads to the rational field of definition KV D Q.t_ ; v/ from the proof of
Theorem 10.9. This shows the existence of a G-realization NQ =KV of SO4m1 .q/ D
GO4m1 .q/=Z .GO4m1 .q// over Q.
For dimension 4m C 1 we use a second variation of .1 ; : : : ; mC2 /, namely

.Q1 ; : : : ; QmC3 / WD .1 ; 2 ; : : : ; mC2 ; 2m / with Q1 D 2m

and analogously put

.1 ; : : : ; mC3 / WD PN1 .Q1 ; : : : ; QmC3 / 2 GL4mC1 .q/mC3 :

Then H WD hi becomes the group GO4mC1 .q/. As above V -symmetrization then
yields a G-realization over Q of SO4mC1 .q/ when 2m > q.
In the case q 1 .mod 4/ we proceed completely analogously with the start-
ing tuple .1 ; : : : ; mC2 / 2 Sp2m .q/mC2 from the corresponding part of the proof of
Theorem 9.10. The variants of  and of  as above and V -symmetrization then yield
the assertion of the theorem in this case as well. t
u
Corollary 10.11. If in Theorem 10.10 we moreover have q ˙3 .mod 8/ then also
the simple groups On .q/ possess G-realizations over Q.
Proof. We start again with the case q 1 .mod 4/, for which we vary the genera-
tors of D2m D GOC 2 .q/ in the proof of Theorem 10.9 by letting

 0 WD .1 ; 2 ; 3 ; 40 ; : : : ; lC3

0
; 2 ; : : : ; 2 / 2 GOC
2 .q/
s

with pairwise non-conjugate bihomologies j0 of order .q  1/=2. (Then 3 j0 has
order q  1 for j D 4; : : : ; l C 3.) As in the proofs of Theorems 10.9 and 10.10 from
10 Applications of the Katz Algorithm 281

this we obtain successively

 0 WD PN1 . 0 /; Q 0 ; 0 WD PN1 .Q 0 / and Q 0 WD .10 ; 20 ; : : : ; mC2

0
/:

As in those cases the group HQ 0 WD hQ 0 i is isomorphic to SO4m1 .q/. But now we
have spin.Qi0 / 2 .IF 0  2
q / for i D 4; : : : ; m C 2, while spin.Q3 / … .IFq / as q
2
5
0
.mod 8/. Because of Q1 D 4m1 there hence exists a unique further i 2 f2; 3g
with spin.Qi0 / … .IF 0
q / , say Q2 . We denote the V -symmetrized G-realization over
2

Q belonging to Q 0 by NQ 0 =KQ 0 , with KQ 0 D KVQ 0 D Q.t_ ; v/ as above. The fixed field

LQ 0 of O4m1 .q/ herein has degree 2 over KQ 0 and is only ramified in the numerator
divisors of t  t2 and t  t3 . Thus LQ 0 is rational over KQ 0 and over Q, and NQ 0 =KQ 0
yields a G-realization of O4m1 .q/ over Q. Starting instead with the second variant
of  in the proof of Theorem 10.10, applied to  0 we then also obtain a G-realization
of O4mC1 .q/ over Q (when 2m > q).
The case q 1 .mod 4/ and so q 3 .mod 8/ is proved analogously, now
C
starting from the group GO 2 .q/ in place of GO2 .q/, and with l D 2 '.q C1/ instead
1

of 2 '.q C 1/.
1
t
u

For orthogonal groups in even dimension we show:

Theorem 10.12. Let n D 2m be even, q D p e odd and m > q.

(a) If q 1 .mod 4/ then PGOC C
n .q/ and PSOn .q/ possess G-realizations over Q.
(b) If q 1 .mod 4/ then PGOn .q/ and PSO

n .q/ possess G-realizations
over Q.

Proof. In (a), with q 1 .mod 4/, we use as starting point the variation

.Q1 ; : : : ; QmC3 / WD .1 ; : : : ; mC1 ; mC2 ; 2m / 2 Sp2m .q/mC3

of the tuple .1 ; : : : ; mC2 / from the proof of Theorem 10.9. Application of the con-
volution PN1 yields

PN1 .Q1 ; : : : ; QmC3 / DW .1 ; : : : ; mC3 / 2 GL4m .q/mC3 :

Here, according to Corollary 9.13 the elements 1 ; 2 are reflections, 3 ; : : : ; lC3

are bihomologies of order 4, q  1 respectively, lC4 ; : : : ; mC1 are bireflections,
mC2 ; mC3 are unipotent and 1 D 4m . By Corollaries 9.7 and 10.7 the group
H D hi generated by the i is an irreducible subgroup of GOC 4m .q/. As H is
C
moreover primitive by the proof of Proposition 10.1(b), we have H  ˝4m Q for
.q/
Q by the Theorem II.2.4 of Wagner. As above, the traces of 4 ; : : : ; lC3
some qjq
C
generate IFq and we have det.1 / D 1 and spin.4 / … .IF q / . Thus, H D GO4m .q/.
2
1
Now let Ci WD Œi in H for i D 1; : : : ; m C 3 and Cs WD Œ1 for s D m C 4.
Then the components C1 ; C2 ; C3 ; ClC4 ; : : : ; Cs of C D .C1 ; : : : ; Cs / are rational, and
1
we have QC D Q.wq1 C wq1 / for a primitive .q  1/st root of unity wq1 . Since
the two classes of reflections in GOC 4m .q/ are conjugate in the conformal group
COC4m .q/, the stabilizer F1 WD f˛ 2 CO C
4m .q/ j C1 D C1 g has index 2 in CO4m .q/.
˛ C
282 III Action of Braids

But then F1 D CGL4m .q/ .H /  H and the class vector C of H is GL-stable. Conse-
quently the Hs -action on the braid orbit of Œ1 ; : : : ; mC3 is transported isomorphi-
cally to ˙.C/= Inn.H /. As above V -symmetrization then leads to a G-realization
NQ =KQ over Q of PGOC Q C
4m .q/. The fixed field L of PSO4m .q/ herein has degree 2 over
Q
K and is only ramified at the two reflection classes C1 ; C2 , so in the numerator divi-
Q KQ and then also L=Q
sors of t  t1 and t  t2 . Thus L= Q are rational, and PSOC
4m .q/
also has a G-realization over Q.
In dimension 4m C 2 we use the variation

.Q1 ; : : : ; QmC4 / WD .1 ; : : : ; mC2 ; 2m ; 2m / 2 Sp2m .q/mC4

of .1 ; : : : ; mC2 /. This yields

PN1 .Q1 ; : : : ; QmC4 / DW .1 ; : : : ; mC4 / 2 GL4mC2 .q/mC4 :

Then H WD hi is a primitive subgroup of GO4mC2 .q/, whence H D GO4mC2 .q/

according to the theorem of Wagner. Copying the proof in the previously treated
case then furnishes G-realizations over Q for PGO4mC2 .q/ and PSOC
4mC2 .q/.
For the proof of (b) we start with the variation

.Q1 ; : : : ; QmC3 / WD .1 ; : : : ; mC1 ; mC2 ; 2m / 2 Sp2m .q/mC3

of the .m C 2/-tuple .1 ; : : : ; mC2 / used in the corresponding case of Theorem 10.9.
An application of PN1 leads to

PN1 .Q1 ; : : : ; QmC3 / DW .1 ; : : : ; mC3 / 2 GL4m .q 2 /mC3 :

The group H WD hi generated by the i is an irreducible and primitive subgroup of

GOC 2
4m .q /. Stability under the Frobenius automorphism of IFq 2 =IFq yields that H
GO4m .q/, which by the theorem of Wagner again shows equality H D GO

4m .q/.
As above the class vector C D .C1 ; : : : ; Cs / formed of the Ci D Œi is GL-stable
1
with QC D Q.wq1 C wq1 /. Thus V -symmetrization leads to a G-realization of
PGO4m .q/ and then of PSO
 N
4m .q/ over Q. Application of the convolution P1 to
the second variation

.Q1 ; : : : ; QmC4 / WD .1 ; : : : ; mC2 ; 2m ; 2m / 2 Sp2m .q/mC4

of .1 ; : : : ; mC2 / gives

PN1 .Q1 ; : : : ; QmC4 / DW .1 ; : : : ; mC4 / 2 GL4mC2 .q 2 /mC4 ;

which then leads to the corresponding result for the case n D 4m C 2 with m > q.
t
u
Remark. Using 3 D diag.y; y 1 / 2 D2m GL2 .q/ with a primitive third (respec-
tively sixth) root of unity y (in place of 3 D diag.x; x 1 / with a primitive fourth
root of unity x) yields the corresponding assertions of Theorem 10.12 in the case
that q ˙1 .mod 6/ instead of q ˙1 .mod 4/.
10 Applications of the Katz Algorithm 283

10.5 Results for Groups in Characteristic Two

In the case of groups over fields of even order we only work out the example case
of linear and unitary groups from Dettweiler and Reiter (2000).

Theorem 10.13. Let q D 2e and n D 4m > 2q.

(a) If e is even then PGLn .q/ has a G-realization over Q.
(b) If e is odd then PGUn .q/ has a G-realization over Q.

Proof. In case (a) we have q 1 .mod 3/, so IF q contains a primitive third root of
unity y. Further let a 2 IFq be a generator of IF
q and a1 D a; a2 ; : : : ; a2l the primitive
powers of a, where again we set l D 2 '.q  1/. Starting from the tuple
1

a D .a1 ; : : : ; a2l ; y; y 1 ; : : : ; y; y 1 / 2 .IF

q/
2m

with product 1 we form

PNy .a1 ; : : : ; a2m ; y 1 / DW .1 ; : : : ; 2mC1 / 2 GL2m .q/2mC1

with 1 D y2m , and then

PNy 1 .1 ; : : : ; 2m ; y2mC1 ; y 1 2m / DW .1 ; : : : ; 2mC2 / 2 GL4m .q/2mC2 :

Then 1 ; : : : ; 2m are homologies with eigenvalues a1 ; : : : ; a2l ; y ˙1 respectively,

2mC1 is unipotent, 2mC2 is conjugate to diag.y; y 1 ; : : : ; y; y 1 /, and 1 D
y 1 4m . The associated tuple Q D .1 ; : : : ; 2mC1 ; y2mC2 / thus has product Q1 D
4m and is linearly rigid. It generates a group H which by Corollary 9.7 and the
proof of Proposition 10.1(b) is a primitive subgroup of GL4m .q/. Due to the pres-
ence of homologies H is not contained in GU4m .q/, so by Theorem II.2.3 we
have SL4m .q/ Q H for some qjq.Q As det.Q1 / D a generates IF q we thus have
H D GL4m .q/, and Q is a rigid generating system of H by Proposition 9.8(b). The
cyclotomic character permutes the conjugacy classes C1 ; : : : ; C2m of the Qi , gen-
erating a subgroup V of S2l  Z2ml of order 2l D '.q  1/. By the multi-variate
variant of the Strong Rigidity Theorem 3.11 the V -symmetrized field of definition
N
KVQ of NN Q =Q.t/ is a rational function field over Q and so leads to a G-realization of
PGL4m .q/ over Q.
In part (b) we have that q 1 .mod 3/. Here, we use a generator a of a cyclic
subgroup of order q C 1 of IF q2
and a primitive third root of unity y 2 IFq 2 and then
proceed as above. t
u

Remark. For n D 4m C 2 by Völklein (1993) there also exist G-realizations of

PGLn .q/ and PGUn .q/ over Q for n > '.q  1/ or n > '.q C 1/ respectively. For
these no simplified approach via the Katz algorithm seems to be known (see also
Section III.9.4 of the first edition Malle and Matzat (1999)).
284 III Action of Braids

Corollary 10.14. Under the assumptions of Theorem 10.13 we also have:

(a) If e is even and gcd.n; q  1/ D 1 then Ln .q/ has a G-realization over Q.
(b) If e is odd and gcd.n; q C 1/ D 1 then Un .q/ has a G-realization over Q.

Proof. Under the stated conditions Ln .q/ agrees with PGLn .q/, respectively Un .q/
agrees with PGUn .q/, so that the claim is immediate from Theorem 10.13. t
u

Further G-realizations of groups in characteristic 2 can be obtained starting from

suitable tuples in GL2 .q/ and GO˙2 .q/, see Dettweiler and Reiter (2000). In order
to obtain G-realizations over Q one needs to take into account the Remark at the end
of Section 10.3.
IV Embedding Problems

The question of whether a given Galois extension can be embedded into a larger one
is called an embedding problem. These occur in a natural way if, starting from the
chief factors, one tries to realize composite groups as Galois groups over a given
field. In the number theoretical context this question already has a long history
which beginning with the work of Scholz (1929) and Reichardt (1937) finally
led to the realization of all solvable groups as Galois groups over Q by Šafarevič
(1954a,c,d, 1989). In this chapter we mainly study embedding problems for geomet-
ric field extensions, for which interest arose only much later, starting with Saltman
(1982).
The first part contains the elementary theory. After several simple reduction the-
orems in Paragraph 1 we present in Paragraphs 2 and 3 the two presently known
basic constructions for proper solutions of embedding problems. These are the solu-
tion of split embedding problems with abelian kernel and the solution of centerless
embedding problems with GAR-kernel. These are presented here in the form suit-
able to regular solutions as introduced by Matzat (1995) as parametric solutions.
By Fried and Völklein (1992) the second of these constructions also leads to the
proof reproduced in Section 3.3 that the absolute Galois group of a Hilbertian PAC
field is free. Paragraph 4 contains methods for the verification, and an overview of
the presently known GAR-realizations of simple groups. In Paragraph 5 we study
embedding problems with abelian kernel and give the criterion from Matzat (1991b)
for the solvability of geometric Frattini embedding problems. This is applied to the
An -polynomials constructed by Mestre (1990) to realize all central extensions of
An as Galois groups over Q.t/.
In the second part we employ cohomological methods. The principal result in
Paragraph 6 is the criterion of Serre (1984) for the solvability of central embedding
problems with kernel Z2 , which relies on the computation of the cohomological
embedding obstruction from invariants of the quadratic trace form. As examples we
treat the central extensions of the symmetric groups Sn following Sonn (1991).
Paragraph 7 contains local-global principles for more general Brauer embedding
problems found by Sonn (1990, 1994a,b) with applications to the realization of

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 285

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_4
286 IV Embedding Problems

central extensions of almost simple groups over Qab .t/. In Paragraph 8 we study
the question when the solvability of all accompanying Brauer embedding problems,
this is the concordance (compatibility) condition introduced by Delone and Fad-
deev (1944), implies the solvability of the originally given embedding problem.
The remaining second embedding obstruction is here called Hasse obstruction. For
this, a cohomological description is derived in Paragraph 9. Finally in Paragraph 10
we prove the Theorem of Scholz (1937) and Reichardt (1937) on the realizability of
nilpotent groups as Galois groups over arbitrary global fields. Here we use a variant
of proof given by Rzedowski-Calderon (1989) and Madan, Rzedowski-Calderon
and Villa-Salvador (1996).
1 Geometric Embedding Problems 287

1 Geometric Embedding Problems

This first paragraph of the chapter serves as introduction and for the comparison
of different types of embedding problems over Hilbertian fields. Also, the two first
elementary reduction theorems for embedding problems are proved.

1.1 Hilbertian Fields

For the convenience of the reader and to fix notation we briefly recall the most
important definitions and results on Hilbertian fields. All proofs can be found in the
monograph of Fried and Jarden (1986), Ch. 11 and Ch. 12, for example.
First let K be an arbitrary field and fi .t; X / 2 K.t/ŒX , t D .t1 ; : : : ; tr /, for
i D 1; : : : ; m irreducible separable polynomials over the rational function field K.t/.
Then the set of all a 2 K r for which fi .a; X / 2 KŒX is defined and irreducible is
called a Hilbertian set of K in K r :

HK .f1 ; : : : ; fm / WD fa 2 K r j fi .a; X / 2 KŒX def. and irred.g: (1.1)

(This corresponds to the notion of separable Hilbert set in loc. cit.) K is called a
Hilbertian field, if every Hilbertian set of K is nonempty.
The Hilbertian fields occurring in the sequel are all obtained from the follow-
ing three theorems. The first was already proved by Hilbert (1892) in the case of
number fields, in the case of function fields over infinite fields of constants by Franz
(1931) and for congruence function fields by Inaba (1944) (see also Fried and Jar-
den (1986), Cor. 12.8 and Thm. 12.9):

Theorem 1.1. Fields with product formula are Hilbertian.

From this, one obtains the so-called classical Hilbertian fields by the following
theorem:

Theorem 1.2. If K is a Hilbertian field, then so is any finitely generated separable

extension field L of K. Moreover, every Hilbert set of L also contains elements
from K.

This theorem follows from Cor. 11.7 and Remark 11.8(b) in loc. cit. The next
result is equivalent to a theorem of Weissauer (1982), which originally was proved
with model theoretic methods. A classical proof goes back to Fried and can be
found in Fried and Jarden (1986), Prop. 12.14, respectively also in Matzat (1987),
Ch. IV.A.

Theorem 1.3. Let K be Hilbertian, N=K an (infinite) Galois extension, and L a

proper finite separable extension of K with L \ N D K. Then the composite M D
LN in a separable algebraic closure KN of K is Hilbertian. Moreover, any Hilbert
set of M contains elements of L.
288 IV Embedding Problems

Note that a Hilbert set of M does not necessarily contain elements from K. From
this result the theorem of Weissauer can easily be reobtained in its original formu-
lation (compare Fried and Jarden (1986), Cor. 12.15, or Matzat (1987), Ch. IV.A,
Folgerung 1):

Corollary 1.4 (Weissauer (1982)). If K is a Hilbertian field, L=K an (infinite) sep-

arable algebraic field extension, and N the Galois hull of L=K, then every finite
separable extension field M of L not contained in N is Hilbertian.

From this theorem respectively its corollary, it follows for example that Qab is
Hilbertian. In fact, Qab has degree 2 over its maximal totally real subfield, and this
is Galois over Q. Since also the maximal totally real intermediate tr N
p field Q of Q=Q is
Galois over Q, all proper finite extension fields of Q like Q . 1/ are Hilbertian,
tr tr

although Qtr itself is not.

1.2 Solutions of Embedding Problems

N
Let K continue to be an arbitrary field and K WD Gal.K=K/ the Galois group of a
separable algebraic closure KN of K. A Galois extension N=K with group G is then
determined by the restriction ' W K ! G D Gal.N=K/ with ker.'/ D Gal.K=N N /.
Now given a group extension GQ D H  G of a group H as normal subgroup with G
and corresponding canonical epimorphism  W GQ ! G, we are led to the question of
whether there exists a homomorphism 'Q W K ! G, Q which extends ' via  such that
the following diagram commutes:

j˜ j

1 H G̃ G 1
ι k (1.2)

This is called the embedding problem E .'; / given by ' and . The homomorphism
'Q is called a solution of the embedding problem E .'; / and the corresponding field
NQ WD KN ker.'/
Q
a solution field of E .'; /. Note that 'Q determines NQ but not vice
versa. If here 'Q is an epimorphism, then 'Q respectively NQ is called a proper solution
(field) of the embedding problem. In the latter case we have Gal.NQ =K/ Š G. Q
For a better description of embedding problems we employ the group theoretic
terminology: H is called the kernel of the embedding problem E .'; /. The embed-
ding problem is called finite if GQ is a finite group. It is called split respectively non-
split if the corresponding group extension GQ D H  G splits respectively does not
split. Further E .'; / is called a central respectively Frattini embedding problem, if
the kernel H lies in the center of GQ resp. in the Frattini subgroup of G. Q
1 Geometric Embedding Problems 289

If more particularly K=k is a function field with field of constants k, an embed-

ding problem E .'; / with epimorphisms ' W K ! G and  W GQ ! G is called a
geometric embedding problem if N WD KN ker.'/ is geometric over K or equivalently
regular over k. Correspondingly, a (proper) solution 'Q of such an embedding prob-
lem is called a geometric (proper) solution if also NQ WD KN ker.'/ Q
is geometric over
K. Since only geometric embedding problems can possess geometric solutions, this
hypothesis will sometimes be assumed implicitly.
In the following the deformable solutions of embedding problems will be of par-
ticular interest. This notion can be made more precise as follows: From a Galois
extension N=K with group G and a system of r  1 variables t D .t1 ; : : : ; tr / over
K one obtains a Galois extension N  WD N.t/ over K  WD K.t/ with the group
G  Š G. Thus every embedding problem E .'; / over K with ' W K ! G and
 W GQ ! G can be lifted to a uniquely defined embedding problem E .'  ;   / over
K  with the extensions '  W K  ! G  of ' and the composition   W GQ ! G 
of  with the above isomorphism. A solution 'Q  of E .'  ;   / is now called a reg-
ular solution of the embedding problem E .'; / (in r variables), if the field N is
algebraically closed in NQ  WD .KN  /ker.'Q / , i.e., NQ  =N is a regular field extension


(by Nagata (1977), Thm. 3.5.2). The existence of a regular solution implies the
existence of solutions:
Theorem 1.5. (a) If a finite embedding problem E .'; / over a Hilbertian field K
possesses a regular (proper) solution, then it also possesses a (proper) solution.
(b) If a finite geometric embedding problem over an algebraic function field K=k
possesses a regular (proper) solution, then it also possesses a geometric (proper)
solution.
Proof. Let K be a Hilbertian field, K  D K.t/ and 'Q a regular solution of E .'; /
with solution field NQ  D K  ŒX =.f .t; X // whose existence is assumed in (a).
Then by definition there exists a specialization t 7! a 2 K r such that NQ a WD
KŒX =.f .a; X // is Galois over K with Gal.NQ a =K/ Š Gal.NQ  =K  /. Now let }a be
the place of K  =K defined by t 7! a with prime ideal pa (compare Section III.6.1),
}O a an extension of }a to the separable algebraic closure KN  of K  with prime ideal
pO a and Ka the decomposition field of pO a =pa (see Nagata (1977), Ch. VI, 3). Then
we have
N  N  
Ka WD Gal.K =Ka / Š Gal.kK =K / Š K :

Furthermore, as Ka \ NQ  D K  the homomorphisms '  , 'Q  by restriction to Ka

define epimorphisms

'a W Ka ! Gal.Ka N  =Ka / Š Gal.N  =K  / D G  ;

'Qa W Ka ! Gal.Ka NQ  =Ka / Š Gal.NQ  =K  / GQ  ;

with   ı 'Qa D 'a . Moreover, by construction the first of these also satisfies
a ı 'a D '  ı a with the isomorphism a W Ka ! K (respectively for the corre-
sponding factor groups) given by the residue class map. Thus 'Qa WD a ı 'Q a ı a1 is
an epimorphism of K onto Gal.NQ a =K/ with  ı 'Q a D '. This proves (a).
By Theorems 1.1 and 1.2 the algebraic function field K=k in (b) is always Hilber-
tian, so the existence of a regular solution by (a) implies at least the existence of
290 IV Embedding Problems

solutions. It remains to show that among these are even geometric ones. By assump-
tion k respectively N is algebraically closed in N resp. NQ  , and in particular NQ  =k
is regular. If kN is the algebraic closure of k in a separable algebraic closure of NQ  , the
polynomial f .t; X / 2 K  ŒX remains irreducible in kK N  ŒX . We now apply Theo-
0
rem 1.3 to an intermediate field K of K=k different from K over which K is finite
separable. Using the notation N 0 WD kK N 0 , L0 WD K and M 0 WD kK N in analogy to
Theorem 1.3 we obtain that every Hilbertian set of the Hilbertian field kK N contains
elements from K. Thus there exists an a 2 K for which f .a; X / remains irreducible
r
N
even in kKŒX . The splitting field NQ a of f .a; X / with Gal.NQ a =K/ Š Gal.NQ  =K  /
is then geometric over K, and in the same way as in part (a) it yields a geometric
solution 'Qa of the given embedding problem E .'; /. If here Gal.NQ  =K / Š G, Q
Q Q
then also Gal.Na =K/ Š G, thus regular proper solutions of the embedding problem
always furnish geometric proper solutions. t
u

Remark. If in Theorem 1.5(a) resp. (b) the given solution field NQ  is different from
N  , then there even exist infinitely many different (geometric) solutions of the given
embedding problem.

In the next section the different behavior of the notions introduced so far are
illustrated on an elementary reduction theorem.

1.3 Direct Decomposition of the Kernel

If the kernel H of a finite embedding problem Q E .'; / is a direct product of

normal subgroups of GQ D H  G, say H D riD1 Hi , then in a natural way one
obtains embedding problems E .'; i / for the factor groups GQ i WD G=H Q _
i with
Hi WD H1    Hi 1 Hi C1    Hr , with kernels H=Hi Š Hi . The following result,
_ _

which in the case of Galois algebras goes back to Kochendörffer (1953), answers
the question on the connection between the solvability respectively the proper solv-
ability of E .'; / and of E .'; i /.

Theorem 1.6. Assume that the kernel of the finite embedding Q problem E .'; / over
K (resp. K=k) decomposes into a direct product H D riD1 Hi of normal sub-
groups Hi of GQ D H  G. Then for the corresponding embedding problems E .'; i /
with the fixed field N of ker.'/ we have:
(a) E .'; / possesses a (proper) solution if and only if E .'; i / for i D 1; : : : ; r
possess (proper) solutions (linearly disjoint over N ).
(b) If E .'; / possesses a geometric (proper) solution then E .'; i / for i D
N ). If E .'; i /
1; : : : ; r possess geometric (proper) solutions (linearly disjoint over kN
N then
for i D 1; : : : ; r possess geometric (proper) solutions linearly disjoint over kN
E .'; / possesses a geometric (proper) solution.
(c) E .'; / possesses a regular (proper) solution if and only if E .'; i / for i D
1; : : : ; r possess regular (proper) solutions.
1 Geometric Embedding Problems 291

Proof. First it is clear that a solution 'Q W K ! GQ leads to solutions 'Qi W K ! GQ i

of E .'; i / by composition with the canonical epimorphisms i W GQ ! GQ i . In this
way obviously a geometric resp. regular solution of E .'; / yields geometric resp.
regular solutions of E .'; i /. If moreover 'Q is proper with solution field NQ , then
Q
Qr fixed fields Ni of Hi are proper solution fields of E .'; i /, which since H D
_
the
i D1 Hi are linearly disjoint over N , and even after extension of constants with
kN over kNN in the geometric case. It remains to show that the reverse implications
in (a), (b) and (c) hold.
So assume that the embedding problems E .'; i / in (a) have solutions 'Qi W K !
GQ i with solution fields NQ i . Then 'Qi can be decomposed into an epimorphism Q i W
Q Q Q Q
K ! Ei WD Gal.Ni =K/, followed by a monomorphism "Qi W Ei ! Gi , so that i ı
"Qi ı i D '. By assumption G is the subdirect product (pull-back) of the GQ i over G:
Q Q

GQ D GQ 1 G    G GQ r D f.1 ; : : : ; r / j i .i / D j .j / for i; j D 1; : : : ; rg

with  W GQ ! G, .1 ; : : : ; r / 7! 1 .1 /. The composite NQ of the fields NQ i (in K)

N
is Galois over K and contains the field N . With the canonical mapping Q i from
EQ WD Gal.NQ =K/ to EQ i D Gal.NQ i =K/ we get i ı "Q i ı Q i D Q where Q denotes the
canonical epimorphism from EQ onto G D Gal.N=K/. Thus

"Q W EQ ! G;
Q  7! ..Q"i ı Q i /.//i D1;:::;r ;

defines an injective homomorphism with  ı "Q D .

Q By composition with the restric-
tion Q W K ! EQ we obtain a homomorphism 'Q WD "Q ı Q W K ! GQ with

 ı 'Q D  ı "Q ı Q D Q ı Q D '

und hence a solution of E .'; /. If moreover the 'i are proper solutions and the
fields NQ i are linearly disjoint over N , 'Q is surjective due to

Y
r Y
r
ŒNQ W N D ŒNQ i W N D jHi j D jH j;
i D1 i D1

and hence gives a proper solution. This shows (a), compare Fig. 1.1.
If the E .'; i / possess geometric solutions, which even after extension of con-
stants to the algebraic closure remain linearly disjoint, k is algebraically closed
in the solution field NQ , and the construction above yields a geometric solution of
E .'; /. With the same argument as above it also follows that this solution is proper
if and only if the solutions of the E .'; i / are.
In (c) we assume the existence of regular solutions 'Qi of E .'; i /. Let NQ i =K.ti /
be the solution fields, where without loss of generality the parameter systems ti D
.ti1 ; : : : ; ti;ri / form an algebraically independent system of variables. By assumption
N is algebraically closed in each of the NQ i , hence also in the composite NQ  of these
fields in an algebraic closure of K  WD K.t1 ; : : : ; tr /. As above the canonical homo-
morphism Q  W K  ! EQ  WD Gal.NQ  =K  / can be extended to a homomorphism
'Q  W K  ! GQ with   ı 'Q  D '  , which is a regular solution of the embedding
292 IV Embedding Problems

GK
ỹ
ỹ i
p̃i
Ẽ Ẽi
k̃
ẽ j

1 H G̃ G 1
κ

ẽ i
1 Hi G̃i G 1
ki
Fig. 1.1 Subdirect product

problem E .'; /. If all the solutions 'Qi are proper, 'Q  is also proper by (a), since
the fields K  NQ i are linearly disjoint over K  N . t
u
This theorem shows that in contrast to the case for ordinary solutions of embed-
ding problems in the regular case the proper solvability of E .'; / is equivalent to
the proper solvability of the embedding problems E .'; i /. Sometimes the addi-
tional hypothesis in Theorem 1.6(a) for the general case can already be assured by
group theoretic conditions, like the following easy observation.
Remark. If the normal subgroups Hi in Theorem 1.6 possess pairwise coprime
orders, the solution fields of the E .'; i / are necessarily linearly disjoint over N .
A trivial application of Theorem 1.6(c) now yields:
Corollary 1.7. If the groups Hi for i D 1;Q
: : : ; r possess G-realizations over a field
K, then so does their direct product H D riD1 Hi .
Proof. This follows immediately from Theorem 1.6(c), since G-realizations are reg-
ular proper solutions of embedding problems with trivial cokernel. t
u

1.4 From Improper to Proper Solutions

In several important cases the solvability of embedding problems implies the proper
solvability. The first is the following special result on Frattini embedding problems
(see Dentzer (1995b)).
Proposition 1.8. Every solution of a finite (geometric) Frattini embedding problem
is a proper (geometric) solution.
1 Geometric Embedding Problems 293

Proof. First let E .'; / be an ordinary Frattini embedding problem with kernel H
over K, 'Q a solution of E .'; / with solution field NQ and Gal.NQ =K/ D E, Q W
Q
K ! E the canonical epimorphism and " W E ! G D H  G the homomorphism
Q
with " ı D '. Q Since . ı "/.E/ D G we conclude that GQ D hH; ".E/i, which
Q
implies G D ".E/ since H ˚.G/. Q Hence " and so also 'Q are epimorphisms.
Now let E .'; / be a geometric Frattini embedding problem over the algebraic
function field K=k with the proper solution 'Q and the corresponding field NQ , so that
NQ H D N and NQ G D K. With the algebraic closure kQ of k in NQ the composites kK Q
Q

Q (in NQ ) are Galois over K. In particular the subgroup U WD Gal.NQ =kN

and kN Q / of
H is normal in G,Q and with HQ WD H=U we have

Q
G=U Q
Š Gal.kN=K/ Q
D Gal.kN=N Q kK/
/  Gal.kN= Q Š HQ  G:

Since by assumption H ˚.G/ Q we also have HQ Q

˚.G=U /. Thus HQ D 1 and
kQ D k, NQ =k is regular and hence NQ =K geometric. t
u

A general but conditional result is already due to Ikeda (1960) and Nobusawa
(1961).

Theorem 1.9 (Ikeda (1960)). Assume that every finite split (geometric) embedding
problem over a field K (resp. K=k) with kernel H is (geometrically) properly solv-
able. Then any finite (geometric) embedding problem E .'; / with kernel H over K
having a (geometric) solution also possesses a (geometric) proper solution.

Proof. Let N WD KN ker.'/ and GQ D H  G the inverse image of . By assumption there

exists a homomorphism 'Q from K D Gal.K=K/ N to GQ with  ı 'Q D '. The kernel
of 'Q is then a subgroup 1 of K whose fixed field N1 contains the field N and
is Galois over K with group G1 D Gal.N1 =K/. In particular there exist canonical
epimorphisms '1 W K ! G1 , " W G1 ! G with " ı '1 D ' and a monomorphism
"Q W G1 ! GQ with "Q ı '1 D '.
Q In the subdirect product

GQ 1 WD GQ G G1 D f.; Q O 2 G1 ; .Q / D "./g

Q O / j Q 2 G; O

the kernel of the projection p2 from GQ 1 onto the second factor is isomorphic to
ker./ D H , and C1 WD f.Q".O /; O / j O 2 G1 g forms a complement to ker.p2 / in GQ 1
isomorphic to G1 . Hence GQ 1 is the semidirect product of H with C1 Š G1 .
Now by assumption the embedding problem E .'1 ; p2 / has a proper solution, i.e.,
there exists an epimorphism 'Q1 W K ! GQ 1 with p2 ı 'Q1 D '1 . This can be com-
posed with the projection p1 W GQ 1 ! GQ to an epimorphism 'Q 0 WD p1 ı 'Q1 W K ! GQ
satisfying
 ı 'Q 0 D  ı p1 ı 'Q 1 D " ı p2 ı 'Q 1 D " ı '1 D ':
Consequently 'Q 0 is a proper solution of E .'; /.
If E .'; / is a geometric embedding problem over the function field K=k with
geometric solution ', Q then E .'1 ; p2 / also becomes a split geometric embedding
problem over K=k. By assumption this possesses a proper geometric solution 'Q1 ,
294 IV Embedding Problems

GK

j̃ 1
j1
p2
1 H G̃1 G1 1
p1 e j
ẽ

1 H G̃ G 1
ι k
Fig. 1.2 Proper solution

i.e., the corresponding solution field NQ 1 WD KN ker.'Q1 / is regular over k. Being a sub-
field of NQ 1 , NQ WD KN ker.'Q / is also regular over k and hence a geometric extension
0

field of K with Gal.NQ =K/ Š G. Q t

u

Corollary 1.10. If the embedding problem E .'; / in Theorem 1.9 possesses a reg-
ular solution over K in r variables and if over K.t1 ; : : : ; tr / every split embedding
problem with kernel H is regularly properly solvable, then E .'; / also has a regu-
lar proper solution.

Proof. By assumption N is algebraically closed in the fixed field N1 of ker.'Q  /

with respect to the lifted embedding problem E .'  ;   / over K  WD K.t/, t D
.t1 ; : : : ; tr /, extending E .'; /. If 'Q1 is a regular proper solution of the embedding
problem over K  corresponding to E .'1 ; p2 /, then again N1 is algebraically closed
in the corresponding solution field NQ 1 . Hence N is even algebraically closed in the
fixed field NQ  NQ 1 of the kernel of 'Q 0 WD p1 ı 'Q 1 , and 'Q 0 yields a regular proper
solution of E .'; /. t
u

In the last section of this paragraph we give some fields over which every finite
embedding problem is solvable.

1.5 Fields with Projective Galois Group

A profinite group is called projective profinite group, if in the diagram (1.2) with
K replaced by every (pro-) finite embedding problem has a solution, equiva-
lently if the cohomological dimension of is at most 1. The fields of importance
here with projective absolute Galois group are collected in the following theorem.

Theorem 1.11. The following fields have projective absolute Galois group:
(a) the fields of transcendence degree 1 over an algebraically closed field,
(b) the finite fields and their algebraic extension fields,
1 Geometric Embedding Problems 295

(c) the p-adic number fields with algebraically closed residue class field,
(d) the algebraic number fields containing all roots of unity,
(e) all PAC-fields (see Section III.6.4 for the definition).

Theorem 1.11(a) is the Theorem of Tsen (see for example Shatz (1972), Ch. IV,
Thm. 24), part (b) is a theorem of Chevalley (loc. cit., Thm. 25), part (c) is a theorem
of Lang (loc. cit., Thm. 27), part (d) is a theorem of Serre (1964), Ch. II, Prop. 9
(see also Ribes (1970), Ch. V, Thm. 8.8), and part (e) is a theorem of Ax (see Fried
and Jarden (1986), Thm. 10.17). As an immediate consequence we obtain:

Corollary 1.12. Over the fields listed in Theorem 1.11, every finite embedding prob-
lem is solvable.

A corresponding characterization theorem for fields over which all finite embed-
ding problems are properly solvable is obtained from the following:

Theorem 1.13 (Freiheitssatz of Iwasawa (1953)). A profinite group of countably

infinite rank is free profinite precisely when every finite embedding problem for
has a proper solution.

A proof for this result can be found for example in Ribes (1970), Ch. I, Thm. 9.3,
or in Fried and Jarden (1986), Cor. 24.2.
296 IV Embedding Problems

2 Split Embedding Problems with Abelian Kernel

In this paragraph we solve split embedding problems with abelian kernel over a
Hilbertian field K using a wreath product construction. Over fields with projective
Galois group this leads to the proper solvability of finite embedding problems with
solvable kernel. Next we characterize those finite groups which can hereby be con-
structed inductively as geometric Galois groups over every Hilbertian field.

2.1 Wreath Products

We prove that every Galois extension N =G K over a Hilbertian field K can be

embedded into a Galois extension with a wreath product H o G as Galois group,
if only H possesses a G-realization over K. Since K.t/ is Hilbertian we can assume
without loss of generality that the involved G-realization is a G-realization of one
variable. In the next proposition, tG denotes a generating system of K.t/G =K
according to III.(3.21).

Proposition 2.1. Let K be a field, H a finite group with G-realization of one vari-
able over K, and G a permutation group of degree r. Then there exists a geometric
Galois extension M over the unirational field K.tG / with

Gal.M=K.tG // Š H o G and Gal.M=K.t// Š H r : (2.1)

Proof. Let N =H K.t/ be a geometric Galois extension, x 2 N a primitive element

of N=K.t/ and f .t; X / the minimal polynomial of x over K.t/. If t1 ; : : : ; tr denote
algebraically independent transcendentals over K, K.t/ WD K.t1 ; : : : ; tr / and xi a
zero of f .ti ; X / for i D 1; : : : ; r in an algebraic closure of K.t/, then M WD K.t; x/
with x D .x1 ; : : : ; xr / is a geometric Galois extension of K.t/ with

Y
r
Gal.M=K.t// D Hi Š H r ;
i D1

where Hi D Gal.M=K.t; x_ i // Š H and xi D .x1 ; : : : ; xi 1 ; xi C1 ; : : : ; xr /. Any  2

_

G permutes the transcendentals ti via ti WD t.i / and thus induces an automorphism



of K.t/=K. The set of these automorphisms forms a subgroup of Aut.K.t/=K/ with

the unirational fixed field K.tG /. Since every extension Q of  2 Gal.K.t/=K.tG //
to M permutes the zeroes of the polynomials
Qr f .ti ; X /, we have Q 2 Aut.M=K.tG //
and M=K.t / is Galois. Obviously i D1 Hi is normal in GQ WD Gal.M=K.tG //.
G
Q
The groupQ of those Q 2 G permuting x1 ; : : : ; xr forms a complement isomorphic
to G of riD1 Hi in GQ which permutes the factors Hi by conjugation inside GQ
according to HiQ D H.i / . Consequently we have GQ Š H o G. t
u
2 Split Embedding Problems with Abelian Kernel 297

Since K.tG / is not in general a rational function field, Proposition 2.1 and the
Hilbert irreducibility theorem alone do not suffice to obtain Galois extensions over
K with wreath products as Galois groups.
Theorem 2.2. Let K be a field, H a finite group with G-realization in one variable
over K, G a transitive subgroup of Sr , GQ Š H o G and  W GQ ! G the canonical
epimorphism. Then every embedding problem E .'; / with epimorphism ' W K !
G possesses a regular proper solution.
Proof. Let N be the fixed field of ker.'/ with Gal.N=K/ D G and t1 ; : : : ; tr alge-
braically independent transcendentals over K. Further let K.tG / and M be the same
fields as those in the proof of Proposition 2.1 with Gal.M=K.tG // Š H o G. Then
the composite MQ WD NM is also Galois over K.tG / with the group

D WD Gal.MQ =K.tG // Š Gal.MQ =N.tG //  Gal.MQ =M / Š .H o G/  G:

Denoting by N respectively K.t/ the representation of  2 G as element of

Gal.N=K/ resp. of Gal.K.t/=K.tG //, we see that

GQ WD fı 2 D j ıjN D N ; ıjK.t/ D K.t/ for some  2 Gg

is a subgroup of D isomorphic to H o G, since any ıN 2 Gal.M=K.tG // extends

to a unique ı 2 G. Q By definition of GQ the fixed field MQ GQ is regular over K with
Q
N MQ G D N.t/. Since moreover the N -vector space U WD ˚riD1 N ti is G-invariant,
Q
Q
MQ G is a purely transcendental function field over K by Speiser’s Lemma (Proposi-
Q
tion III.3.10), say MQ G D K.v/.
Now set K WD K.v/, N  WD N.v/ D N.t/ and E .'  ;   / the embedding prob-


lem lifted from E .'; / to N  =K  . Then the canonical epimorphism 'Q  W K  !

Gal.MQ =K  / Š GQ satisfies   ı 'Q  D '  . Thereby 'Q  is a proper solution of
E .'  ;   / and hence a regular proper solution of the original embedding problem
E .'; /. t
u
Special cases of this theorem can already be found in Kuyk (1970), Prop. 1, and
Saltman (1982), Thm. 3.12(d).

2.2 Split Extensions with Abelian Kernel

From the embedding theorem for wreath products we can immediately deduce an
embedding theorem for semidirect products with abelian kernel, using the following
group theoretical result (see for example Suzuki (1982), Ch. 2, Thm. 10.10):
Proposition 2.3. Let GQ D H G be a semidirect product of a finite abelian group
H with a finite group G of order r. Then GQ is isomorphic to the factor group of the
regular wreath product H o G by a normal subgroup U contained in the base group
Hr:
GQ Š .H o G/=U with U H r : (2.2)
298 IV Embedding Problems

This fact together with Theorem 2.2 and the existence of G-realizations for finite
abelian groups allows to deduce the following embedding theorem, which in the
case of number fields already goes back to Scholz (1929), in the case of Hilbertian
fields to Uchida (1980) and in the geometric case to Saltman (1982):
Theorem 2.4. Over any field every finite split embedding problem with abelian ker-
nel has a regular proper solution.
In particular every finite split (geometric) embedding problem with abelian ker-
nel over a Hilbertian field K (function field K=k) has a (geometric) proper solution.
Proof. Let N =G K be a Galois extension with epimorphism ' W K ! G and
GQ D H G a semidirect product of H with G and the projection  W GQ ! G, so
that we have to solve the embedding problem E .'; /. For this, let W WD H o G
be the regular wreath product with the canonical epimorphism W W ! G. Since
by Theorem I.5.1 respectively Theorem III.4.5 and Corollary III.4.8 with Corol-
lary 1.7 every finite abelian group has a G-realization over Q as well as over IFp , by
extension of constants it has a G-realization over K. So all assumptions for E .'; /
in Theorem 2.2 are satisfied. Thus E .'; / possesses a regular proper solution 
with  ı  D '  , where E .'  ;  / denotes the corresponding lifted embedding
problem.
Now let K  be the fixed field of im.'  / and M  the solution field corresponding
to  . Then
Gal.M  =K  / Š W D H o G:
By Proposition 2.3 the wreath product W has a normal subgroup U H r with
W=U Š G.Q The fixed field NQ  WD .M  /U then contains N  D NK  and is Galois

over K with group
Gal.NQ  =K  / Š GQ D H G:
We denote by Q  the restriction map from Gal.M  =K  / onto Gal.NQ  =K  /. Then
with 'Q  WD Q  ı  we have

  ı 'Q  D 
ı 
D ' ;

and 'Q  is a proper solution of the embedding problem E .'  ;   /. Hence 'Q  is a
regular proper solution of the original embedding problem E .'; /.
The application to ordinary and geometric solutions of embedding problems over
Hilbertian fields follows immediately from Theorem 1.5. t
u
Using the reduction theorem of Ikeda (Theorem 1.9 and Corollary 1.10) in addi-
tion to the above result we obtain:
Corollary 2.5. If a finite embedding problem with abelian kernel over a Hilbertian
field K possesses an ordinary (geometric/regular) solution, then it also possesses a
corresponding proper solution.
Since over fields with projective Galois group (improper) solutions of finite
embedding problems always exist, induction over a chief series yields from Corol-
lary 2.5:
2 Split Embedding Problems with Abelian Kernel 299

Theorem 2.6 (Iwasawa (1953)). Let K be a Hilbertian field with projective absolute
Galois group. Then every finite embedding problem over K with solvable kernel has
a proper solution.
Proof. Let N=K be a finite Galois extension with epimorphism ' W K ! G D
Gal.N=K/ and GQ D H  G a group extension with finite solvable kernel and canon-
ical epimorphism  W GQ ! G. Further let H1 be maximal with respect to inclusion
among the normal subgroups of GQ properly contained in H and GQ 1 WD G=HQ 1 . Then
H=H1 is a chief factor of GQ and hence a finite abelian group. Since K is projective
by assumption, it follows from Corollary 2.5 that the embedding problem E .'; 1 /
with canonical epimorphism 1 W GQ 1 ! G has a proper solution 'Q1 . Replacing G by
GQ 1 and H by H1 we may inductively continue until GQ s D G.Q Then  D 1 ı    ı s ,
and 'Q WD 'Qs is a proper solution of E .'; /. t
u
In the next section we shall give a group theoretic characterization of those
groups which can be realized as Galois groups by successive solution of split embed-
ding problems with abelian kernel and passage to Galois intermediate fields.

2.3 Semiabelian Groups

A group G is called a semiabelian group if it is generated by a finite set of abelian

subgroups Ai

G D hA1 ; : : : ; An i with Ai NG .Aj / for i j: (2.3)

The following theorem contains the characterizations of semiabelian groups which

are of interest to us (see Stoll (1995) for a further characterization):
Theorem 2.7 (Dentzer (1995b)). For a finite group G ¤ 1 the following are equiv-
alent:
(a) G is semiabelian.
(b) G possesses an abelian normal subgroup A and a semiabelian proper sub-
group U with G D AU .
(c) G can be constructed in finitely many steps, starting from the trivial group, of
semidirect products with a finite abelian kernel and taking factor groups.
Proof. If G D hA1 ; : : : ; An i, where without loss of generality we may assume An 6
hA1 ; : : : ; An1 i, is a semiabelian group, then A WD An is an abelian normal subgroup
and U WD hA1 ; : : : ; An1 i by (2.3) is a semiabelian proper subgroup of G.
If (b) holds, the action of U on A defines a homomorphism ' W U ! Aut.A/ and
thus a semidirect product AU with finite abelian kernel. Now via the epimorphism

W AU ! G; .; / 7! ;

G is isomorphic to a factor group of AU . Thus (c) follows by descending induction

on the order of U .
300 IV Embedding Problems

Now let G be constructed according to (c) in a finite number of steps taking

factor groups of semidirect products with abelian kernel, say as the last member of
a finite sequence of quotients Gi WD .Ai Gi 1 /=Ni with Ai abelian and G0 D 1. Let
G D .An Gn1 /=Nn with minimal n. Then G cannot be a factor group of Gn1 ,
since otherwise G could also be obtained as factor of An1 Gn2 in contradiction
to the minimality of n. Hence ANn WD An Nn =Nn is an abelian normal subgroup and
GN n1 WD Gn1 Nn =Nn a proper subgroup of G with G D GN n1 ANn D hGN n1 ; ANn i.
Since herein by construction GN n1 can be obtained in at most n  1 steps as factor
group of a semidirect product with abelian kernel, (2.3) follows with ANn in place of
An by descending induction on n. t
u
From this characterization we immediately obtain from Theorem 2.4:
Corollary 2.8. Finite semiabelian groups possess G-realizations over any field and
therefore are Galois groups over every Hilbertian field.
In view of this result it is natural to ask which groups lie in the class of
semiabelian groups. A partial answer is given by:
Proposition 2.9 (Thompson (1986)). (a) Finite groups G of nilpotency class 2, i.e.,
with G 0 Z .G/, are semiabelian.
(b) Finite solvable groups G with only abelian Sylow subgroups are semiabelian.
Proof. Since the properties of G in (a) and (b) also hold for every subgroup, the
proof can in both cases be done by induction over the group order.
For (a), let U denote a maximal and hence semiabelian subgroup of G. Then
the group A generated by a  2 G n U and Z .G/ is an abelian subgroup of G
with AU D G. As G=Z .G/ is abelian by assumption, A is even normal. Thus G is
semiabelian by Theorem 2.7(b).
For (b), let F be the Fitting subgroup of G, i.e., the maximal nilpotent normal
subgroup. Being nilpotent, F is the direct product of its Sylow subgroups and hence
by assumption abelian. Since for G ¤ 1 the Frattini subgroup ˚.G/ is a proper
subgroup of F (Huppert (1967), Ch. III, Satz 4.2(c)), G possesses a maximal and
by induction semiabelian subgroup U with F U D G. This proves part (b). t
u
In the particular case of p-groups this leads to:
Corollary 2.10 (Dentzer (1995b)). Let p be a prime.
(a) All groups of order p n with n 4 are semiabelian.
(b) The groups of order 25 are semiabelian.
Proof. Groups of order p or p 2 are abelian, those of order p 3 of nilpotency class
at most 2. If jGj D p 4 , then G has a maximal abelian normal subgroup A of order
jAj D p a and 8 a.a C 1/ and thus a  3 (Huppert (1967), Ch. III, Satz 7.3(b)).
If G is non-abelian, it possesses a maximal subgroup U ¤ A with AU D G which
proves (a).
Part (b) of the corollary follows by explicit verification of the condition in Theo-
rem 2.7(b) for all 51 isomorphism types of groups of order 32. t
u
2 Split Embedding Problems with Abelian Kernel 301

Unfortunately among the groups of order p 5 for p ¤ 2 and among those

of order 26 there exist some which are not semiabelian (see Dentzer (1995b),
Prop. 2.10 and App. A).
302 IV Embedding Problems

3 Embedding Problems with Centerless Kernel

We study embedding problems over Hilbertian fields K in which the kernel H has
trivial center. If here H possesses a GA-realization over K with an additional ratio-
nality condition, then every embedding problem over K with this kernel and even
with kernel H r is solvable. The latter can be used via induction over a chief series
for the realization of groups as Galois groups whose composition factors possess a
GAR-realization over K as defined below. Finally the methods presented here allow
us to prove that the absolute Galois groups of Hilbertian PAC-fields are free.

3.1 The Notion of GAR-Realization

The notion of a GA-realization was already introduced in Section I.5.2: Gal.N=K/

is called a GA-realization of H over k, if N=K is a geometric Galois extension
with group H over a rational function field K=k, which is part of a geometric (with
respect to k) Galois extension with group A WD Aut.H / (under identification of H
with Inn.H /, which is only possible in case Z .H / D 1). Here N A =k does not
necessarily have to be rational. Such a GA-realization is called a GAR-realization
of H over k if in addition the following rationality condition is satisfied:
N D kK
(R): Every finite extension field R=N A with kR N is a rational function field
over the algebraic closure of k in R.
(Here kN denotes the algebraic closure of k in an algebraic closure of K.) Sufficient
criteria for the validity of (R) will be discussed in the next paragraph.
The advantage of this notion of GAR-realization lies in the following easy group
theoretic fact (see for example Suzuki (1982), Ch. 2, Thm. 7.11):

Proposition 3.1. Let GQ D H  G be a group extension of a centerless finite group H

with a finite group G. Then GQ is isomorphic to a subgroup U of Aut.H /  G with

U \ Aut.H / D Inn.H / and p2 .U / D G (3.1)

(where p2 denotes the projection onto the second factor.)

This proposition leads to the following embedding theorem:

Theorem 3.2. Let K be a field and H a finite group with GAR-realization over K.
Then every finite embedding problem E .'; / over K with kernel H possesses a
regular proper solution.
In particular, such ordinary (geometric) embedding problems over a Hilbertian
field always possess (geometric) proper solutions.

Proof. Let N =G K be a finite Galois extension with epimorphism ' W K ! G and

GQ D G  H a group extension with canonical epimorphism  W GQ ! G. According to
3 Embedding Problems with Centerless Kernel 303

the automorphism condition (A) there exists a geometric Galois extension M=K.t/
with Aut.M=K/  A Š Aut.H /, where moreover K.t/ is the fixed field of the sub-
group Inn.H / Š H of A. Denote the composite of N with M respectively with M A
in an algebraic closure of M by MQ resp. MQ A . Then MQ =M A is Galois with

Gal.MQ =M A / D Gal.MQ =MQ A/  Gal.MQ =M / Š Aut.H /  G: (3.2)

By Proposition 3.1 the group GQ is isomorphic to a subgroup U of Aut.H /  G with

property (3.1). Consequently the fixed field R WD MQ U is a geometric extension field
of M A with NR D N.t/. The rationality condition (R) now implies that R=K is
a rational function field. Since N is algebraically closed inside MQ , the canonical
epimorphism 'Q  W R ! Gal.MQ =R/ yields a regular proper solution of E .'; /.
The application to ordinary and geometric embedding problems as usual follows
immediately from Theorem 1.5. t
u
Furthermore, by induction over a normal series Theorem 3.2 yields:
Corollary 3.3. Let K be a Hilbertian field and G a finite group which has a normal
series
G F G0 F G1 F    F Gn D 1
with the following properties:
(1) G=G0 is isomorphic to the Galois group of a Galois extension N0 =K.
(2) Gi 1 =Gi has a GAR-realization over K for i D 1; : : : ; n.
Then there exists a Galois extension N=K with Gal.N=K/ Š G.
If here K=k is an algebraic function field and N0 =K a geometric Galois exten-
sion, then there exists a geometric Galois extension N=K with Gal.N=K/ Š G.
Proof. Let G=Gi 1 be isomorphic to the Galois group of a (geometric) Galois
extension Ni 1 =K and H WD Gi 1 =Gi . Then by Theorem 3.2 there exists a (geo-
metric) Galois extension Ni =K with Gal.Ni =K/ Š G=Gi . Thus Corollary 3.3 is
obtained from Theorem 3.2 by induction on i . t
u
Remark. In the proofs of Theorem 3.2 and Corollary 3.3 the rationality condition
(R) was only used to guarantee that every geometric extension field R=N A with
N D kK
kR N is a rational function field over k (compare to the definition in Matzat
(1985b) resp. Matzat (1987), Ch. IV, 4).

3.2 Embedding Problems with Characteristically Simple Kernel

To obtain an assertion in the form of Corollary 3.3, in which the assumption on the
factors of a normal series are replaced by the corresponding, but much more easily
verifiable assumptions on the composition factors, we first have to solve embedding
problems with characteristically simple non-abelian kernel. The first step is given
by:
304 IV Embedding Problems

Proposition 3.4. Let K be a field and H a finite non-abelian simple group with GA-
realization over K. Then the r-fold direct product H r possesses a GA-realization
over K.

Proof. In the proof let A WD Aut.H / and AN WD Out.H /. By assumption H has

a GA-realization over K, i.e., there exists a tower of geometric Galois exten-
N
sions N  K.t/  K.t/A , t D .t1 ; : : : ; ts / a system of independent transcendentals,
N
with Gal.N=K.t// Š H and Gal.N=K.t/A / Š A. Further let f .t; X / 2 K.t/ŒX
resp. g.u; X / 2 K.u/ŒX be minimal polynomials of primitive elements of N=K.t/
resp. N=K.u/, where u D .u1 ; : : : ; uq / with uj D hj .t/ 2 K.t/ is a generating
N
system of the unirational function field K.t/A =K. If now t WD .t1 ; : : : ; tr / with
ti D .ti1 ; : : : ; ti s / is a system of rs over K independent transcendentals, then the
zeroes of f .ti ; X / 2 K.ti /ŒX generate fields Ni =K.ti / in an algebraic closure of
K.t1 ; : : : ; tr /, with Hi WD Gal.Ni =K.ti // Š H and

Y
r
Gal.M=K.t// Š Hi Š H r (3.3)
i D1

A
for the composite M WD N1    Nr . For the fixed field Ni i of Ni under the action
N
induced by A we write K.ui / WD K.ti /Ai with uij D hj .ti / 2 K.ti /. Then with
u WD .u1 ; : : : ; ur / we further have

Y
r
Gal.M=K.u// Š Ai Š Ar : (3.4)
i D1

Since H is non-abelian simple the automorphism group of H r is the wreath

product of A with the symmetric group Sr :

Aut.H r / Š A o Sr Š Ar Sr : (3.5)

An action of  2 Sr on K.u/ is obtained in a natural way by restriction to K.u/ of

the automorphisms of K.t/ given by the action tij WD t.i /;j . Since the restriction N
satisfies
uijN D hj .ti / D hj .t.i / / D u.i /;j (3.6)
and hence permutes the fields K.ui / according to K.ui /N D K.u.i / /, this yields a
faithful representation of Sr into Aut.K.u/=K/, whose fixed field will be denoted
by F .
By (3.6) all extensions Q of N 2 Gal.K.u/=F / onto M only interchange the
zeroes of g.ui ; X / 2 K.ui /ŒX for i D 1; : : : ; r, so M=F is Galois. By fixing exten-
sions of N which permute a given system of representatives of the zeroes of g.ui ; X /
we thus obtain a complement to Ar in Gal.M=F /, and easily verify

Gal.M=F / Š Ar Sr Š A o Sr Š Aut.H r /: (3.7)

3 Embedding Problems with Centerless Kernel 305

Since here the purely transcendental function field K.t/ occurs as the fixed field of
Inn.H r / Š H r , Gal.M=K.t// yields a GA-realization of H r . t
u
After these preparations we come to the embedding theorem for characteristically
simple kernel:
Q
Theorem 3.5. Let K be a field and riD1 Hi Š H r a characteristically simple finite
group whose simple factor H is non-abelian and has a GAR-realization over K.
Then every finite embedding problem E .'; / over K with kernel H r has a regular
proper solution.
In particular, such ordinary (geometric) embedding problems over a Hilbertian
field always possess proper (geometric) solutions.
Proof. Let N =G K be a finite Galois extension with epimorphism ' W K ! G and
GQ D H r  G a group extension with canonical epimorphism  W GQ ! G. By Theo-
rem 1.6(c) we can in addition assume without loss of generality that H r is a mini-
mal normal subgroup of GQ and hence G acts transitively on the set fH1 ; : : : ; Hr g of
direct factors of H r .
The notations M , K.t/, K.u/ and F , as well as Ni , Hi and Ai , are used as in the
proof of Proposition 3.4. Further let MQ WD NM be the composite of N and M in an
algebraic closure of M . Then MQ =F is Galois with

Gal.MQ =F / D Gal.MQ =NF /  Gal.MQ =M / Š Aut.H r /  G: (3.8)

By Proposition 3.1 there exists a subgroup U of Gal.MQ =F / isomorphic to G, Q whose

fixed field R WD MQ U is a geometric extension field of F=K with NR D N.t/. It
remains to establish that R=K is a rational function field.
In what follows we identify U with GQ and U=H r with G. Moreover the nor-
malizer of Hi in GQ is denoted by Ci . First we prove that the fixed field MQ D of
D WD \riD1 Ci is a rational function field over N D , where N D denotes the fixed
field of the group D restricted to N . From the assumptions it follows that the Galois
group of the field extension K1 N1 =K1 .t1 / obtained from N1 =K.t1 / by extension
of constants with K1 WD N C1 is a GAR-realization of H1 over K1 , where K1 .u1 /
is the unirational fixed field of A1 Š Aut.H1 /. GQ permutes the generating systems
u1 ; : : : ; ur of K.ui /=K according to (3.6). This implies that u11 ; : : : ; u1q are ele-
ments of MQ C1 , and the subfield N.t1 / of N.t/ remains invariant under C1 . The
fixed field N.t1 /C1 is thus a geometric extension field of K1 .u1 /, whose compos-
ite with N yields the field N.t1 /. From the rationality condition (R) for H1 now
follows that N.t1 /C1 is a rational function field over K1 , say N.t1 /C1 D K1 .v1 /
with a transcendence basis v1 D .v11 ; : : : ; v1s /. As the permutation representation
of GQ on fH1 ; : : : ; Hr g is transitive, there exist elements i 2 G for i D 1; : : : ; r with

.1/i D i . Then the transcendence basis fvi WD v1i j i D 1; : : : ; rg is Ci -invariant,
r Q
and as D D \i D1 Ci we see that vij 2 M for 1 i r, 1 j s. Moreover
D

N.vi / D N.ti /, so v WD .v1 ; : : : ; vr / forms a system of rs independent transcenden-

tals over N D . In addition to the inclusion N D .v/ MQ D N.t/ we have the degree
estimates
ŒN.t/ W MQ D  ŒN W N D D ŒN.v/ W N D .v/ ;
306 IV Embedding Problems

which due to N.t/ D N.v/ implies MQ D D N D .v/. By construction every  2 G D

Gal.N.v/=R/ permutes the transcendence basis vij according to vij D v.i /;j . In
particular the N -vector space ˚riD1 ˚sj D1 N vij is G-invariant. Since moreover by
the above R=K is geometric with NR D N.v/ D N.t/, it follows from Speiser’s
Lemma (Proposition III.3.10) that R=K is a rational function field. Now N is alge-
braically closed in MQ , so 'Q  W R ! Gal.MQ =R/ gives a regular proper solution of
E .'; /.
The assertion about ordinary and geometric embedding problems again follows
by application of Theorem 1.5. t
u
Theorem 3.5 is suited for induction proofs along a chief series. A first result is
given in:
Theorem 3.6 (Embedding Theorem for Centerless Kernel). Let K be a Hilbertian
field and H a finite group whose composition factors have GAR-realizations over
K. Then every finite embedding problem E .'; / over K with kernel H possesses a
proper solution.
If here the embedding problem is geometric, it even has a geometric proper solu-
tion.
Proof. Let N =G K be a finite (geometric) Galois extension with restriction ' W K !
G and GQ D H  G a group extension with the canonical epimorphism  W GQ ! G.
Furthermore let H1 be maximal with respect to inclusion among the normal sub-
groups of GQ properly contained in H and GQ 1 WD G=H Q 1 . Then H=H1 is a chief
Q
factor of G and hence a characteristically simple group. By assumption the simple
factors of H=H1 are non-abelian and possess a GAR-realization over K. It now
follows from Theorem 3.5 that the embedding problem E .'; 1 / with the canonical
epimorphism 1 W GQ 1 ! G has a (geometric) proper solution 'Q1 . Replacing G by
GQ 1 and H by H1 we may inductively continue until GQ s D G.Q Then  D 1 ı    ı s ,
and 'Q WD 'Qs is a (geometric) proper solution of E .'; /. t
u
In the special case G D 1 Theorem 3.6 yields:
Corollary 3.7. Let K be a Hilbertian field and H a finite group whose composition
factors have GAR-realizations over K. Then H possesses a G-realization over K.
In the next paragraph we hence study the question of which simple groups can
be shown to have GAR-realizations over Q or at least over Qab . But first we apply
the methods developed so far to determine the Galois groups of Hilbertian PAC-
fields. Clearly, for these fields the verification of the rationality condition (R) is not
necessary.

3.3 Galois Groups of Hilbertian PAC-Fields

To be able to apply the Theorem of Conway and Parker (Thm. III.6.10) also to the
solution of embedding problems, these have to be reduced to the case where the
3 Embedding Problems with Centerless Kernel 307

Schur multiplier of the kernel is generated by commutators. The decisive induction

step is achieved in the next proposition.
Proposition 3.8. Let K be a field. Then every finite split embedding problem E .'; /
over K with non-abelian characteristically simple kernel H can be reduced to a
Q whose kernel HQ has trivial center as well as triv-
split embedding problem E .'; /,
ial Schur multiplier.
Proof. In the proof we follow the outline of the proof of Proposition III.6.12. As
usual let G WD im.'/ and GQ be the inverse image of , so that GQ D H G by
assumption. Obviously the action of G on H defined by the semidirect product
can be lifted in a unique way to an action on the universal central extension R of H .
In this way we obtain a semidirect product E WD RG with canonical epimorphism
W E ! G.Q As in the proof of Proposition III.6.12 let S be a finite non-abelian sim-
ple group with trivial Schur multiplier (say S D SL2 .8/), and EQ WD S o E D S e E
with e WD jEj the regular wreath product with canonical epimorphism Q W EQ ! E.
Then E acts on S e by permutation of the factors in the regular permutation repre-
sentation and we have

EQ D S e E D S e .RG/ D .S e R/G D HQ G

with HQ WD S e R and Z .HQ / D 1. Since both S and R are perfect groups with
trivial Schur multiplier, the Schur multiplier M.HQ / is also trivial, as can easily be
seen.

GK

j̃ E j

1 H̃ Ẽ G 1
κ̃
ỹ

1 R E G 1

1 H G̃ G 1
k
Fig. 3.1 Trivializing the Schur multiplier

The embedding problem E .'; / Q with Q WD  ı ı Q is hence split with ker-

nel H , where moreover Z .H / D M.HQ / D 1. But this proves the assertion, since
Q Q
Q yields a (proper) solution 'Q WD ı Q ı 'QE
every (proper) solution 'QE of E .'; /
308 IV Embedding Problems

of E .'; /. Furthermore geometric respectively regular solutions of E .'; /

Q lead to
geometric resp. regular solutions of E .'; /. t
u
With Proposition 3.8 we are able to prove a pendant to Theorem 3.5 for PAC-
fields.
Proposition 3.9. Let K be a PAC-field of characteristic 0. Then every finite split
embedding problem E .'; / over K with characteristically simple non-abelian ker-
nel has a regular proper solution.
Proof. By Proposition 3.8 we can replace E .'; / by a finite split embedding prob-
Q whose kernel HQ has trivial center and trivial Schur multiplier. For this
lem E .'; /,
let G D im.'/, N WD KN ker.'/ and GQ the inverse image of .
Q
By the Theorem of Conway and Parker (more precisely, Cor. III.6.11) there
exists, for s 2 IN large enough, a class vector C 2 Cl.HQ /s in which every conju-
gacy class of HQ occurs the same number of times and for which ˙.CSs /= Inn.HQ /
forms a single orbit under the full Hurwitz braid group HQ s . Let V Ss denote the
symmetry group of C extended by the permutations on the components of C induced
by A WD Aut.HQ /. Then B WD ˙.CV /= Inn.HQ / is a single orbit under HsV and A acts
on B. By Theorem III.5.10 there exists a geometric Galois extension NA =KVA with

Gal.NA =KVA / Š A and Gal.NA =KV / Š HQ :

Here KVA and KV are regular over Q due to .C /V D CV , and we have NA D N
since Z .HQ / D 1. Consequently N N =KKVA is a Galois extension with group

Gal.N N =KKVA / Š A  G:

According to Proposition 3.1 this Galois group contains a subgroup U isomorphic to

Q whose fixed field R is regular over K with NR D NKV . As intermediate field of
G,
NKV =KKVA , R is also an intermediate field of MN s .t/=Q.tV /. Since in this Galois
extension t D tsC1 remains invariant (as V Ss ), R D R.t/ with R WD MN s \ R, so
that
Gal.N N =R.t// Š GQ and Gal.N N =N R.t// Š HQ :
Now R is regular over the PAC-field K, so R=K possesses a rational place } for
which we can moreover assume that its extension to N R=N.t/ is unramified in the
sense of Section III.6.2. Denote by }Q an extension of } on N N and by pQ the cor-
responding valuation ideal. Then by passage to the residue class fields N N pQ =Rp
with p WD pQ \ R we obtain by Theorem III.6.4 a Galois extension NQ =K.t/ with

Gal.NQ =K.t// Š GQ and NQ  N.t/:

Since here N is algebraically closed inside NQ , the canonical epimorphism 'Q  W

Q
K.t / ! Gal.N =K.t// yields a regular proper solution of E.'; /.
Q u
t
After these preparations we are in a position to prove the main result of this
section.
3 Embedding Problems with Centerless Kernel 309

Theorem 3.10 (Fried and Völklein (1992)). The absolute Galois group of a count-
able Hilbertian PAC-field of characteristic 0 is free profinite of countable infinite
rank.

Proof. The absolute Galois group K of a countable Hilbertian PAC-field K has

at most countable rank, since K and hence also KN contain only countably many
elements. (Here rank is defined as the cardinality of a minimal set of topological
generators, compare Fried and Jarden (1986), Ch. 15.1.) Since split embedding
problems with elementary abelian kernel Zpn over a Hilbertian field are always prop-
erly solvable by Theorem 2.2, the rank of K cannot be finite. By the Freiheitssatz
of Iwasawa (Theorem 1.13) it hence suffices to show that every finite embedding
problem over K has a proper solution.
Obviously we may assume that the kernel H of the embedding problem is char-
acteristically simple, since the general case can be reduced to this by induction over
a chief series. Since K is a PAC-field, the theorem of Ax (Theorem 1.11(e)) implies
that K is projective profinite and therefore every finite embedding problem over K
with kernel H has at least an improper solution. From this the existence of a proper
solution can be concluded with the Theorem of Ikeda (Theorem 1.9), if every finite
split embedding problem over K with kernel H has a proper solution. The latter
now follows from the Theorem of Iwasawa (Theorem 2.6) in the case of abelian
kernel, and from Proposition 3.9 and using Theorem 1.5(a) for non-abelian kernel.
t
u

Remark. In Theorem VI.4.10 we will prove the generalization by Pop (1996) of

Theorem 3.10 to positive characteristic using rigid analytic methods.
p
Since the PAC field Qtr . 1/ already introduced in Section III.6.4 is Hilber-
tian by the Theorem of Weissauer (Corollary 1.4) from the Theorem of Fried and
Völklein we may conclude:
p
Corollary 3.11. The absolute Galois group of Qtr . 1/ is free profinite of count-
able infinite rank.

Corollary 3.11 states a result of the type of the following conjecture due to
Šafarevič:

Conjecture of Šafarevič: The absolute Galois group of Qab is free profinite of count-
able infinite rank.

Unfortunately Qab is not a PAC field by a result of Frey (see Fried and Jarden
(1986), Cor. 10.15). The proof of the conjecture would follow with the results of
this section, if every finite non-abelian simple group possessed a GAR-realization
over Qab . A list of the results obtained so far on this question is contained in the next
paragraph.
310 IV Embedding Problems

4 Verification of the GAR-Property

We present sufficient criteria under which a GA-realization of a group is even a

GAR-realization. These are then applied to check the rationality condition for the
GA-realizations in one and more variables over Q or over Qab constructed so far.

4.1 GAR-Realizations in One Variable

In the first section we give criteria under which a GA-realization in one variable of
G over k, i.e., the Galois group Gal.N=k.t// of a geometric field extension N=k.t/
as constructed in Chapters I and II, is in fact a GAR-realization over k. The simplest
sufficient condition is furnished by an oddness condition for prime divisors in the
ramification locus. In the following let as usually kN denote the algebraic closure of
N NN WD kN
k in a separable algebraic closure of K WD k.t/ and KN WD kK, N , and so on.

Proposition 4.1. Let K D k.t/ be a rational function field in one variable over a
perfect field of constants k and Gal.N=K/ a GA-realization of a finite group G over
k with fixed field F D N Aut.G/ . If F=k contains a prime divisor Q of degree 1 whose
inertia group in K=N FN has odd index in Gal.K= N FN / Š Out.G/, then Gal.N=K/ is a
GAR-realization of G over k.
N is separable, we
Proof. Let R be a field as in condition (R) in Section 3.1. Since k=k
have g.R=k/ Q D g.K=N k/N D 0, where kQ denotes the algebraic closure of k in R. By
assumption the prime divisor Q 2 IP.F=k/ splits into an odd number of prime divi-
sors in IP.K= N which hence form an orbit of odd length under Gal.K=F
N k/, N /. Since
N
Gal.K=R/ N /, IP.K=
Gal.K=F N contains an orbit of odd length under Gal.K=R/,
N k/ N
which yields a prime divisor of odd degree in R=k. Q Thus R=kQ is a rational function
field. t
u

For some of the possible groups of outer automorphisms determined in Theo-

rem I.6.2, the assumption formulated in Proposition 4.1 is always satisfied.

Corollary 4.2. Let Gal.N=K/ be as in Proposition 4.1. Further assume that all
prime divisors Q 2 IP.F=k/ ramified in K=F have residue degree 1. If either
Out.G/ D Zn for some n 2 IN or Out.G/ D Dn for odd n 2 IN, then Gal.N=K/
is a GAR-realization of G over k.

Proof. In the first case the index of the inertia group of the prime divisors ramified
of degree n in K=N FN equals 1, in the second case it equals n for those ramified of
order 2. Hence the corollary is an immediate consequence of Proposition 4.1. t
u

With these observations our results on GA-realizations in Theorem II.10.3 imply

the following:
4 Verification of the GAR-Property 311

Theorem 4.3. The following simple groups possess GAR-realizations in one vari-
able over Q (here p always denotes a prime number):
(a) The non-abelian simple alternating groups An with n ¤ 6.

(b) The linear groups L2 .p/ for pa D 1, where a 2 f2; 3; 5; 7g.
(c) The linear groups L2nC1 .p/ for gcd.2n C 1; p  1/ D 1, p > 3 and p 6 1
.mod 12/.
(d) The unitary groups U2nC1 .p/ for gcd.2n C 1; p C 1/ D 1, p > 2 and p 6 1
.mod 12/.
(e) The symplectic groups S2n .p/ for odd primes p 6 ˙1 .mod 24/, p6 jn, or for
p D 2.
(f) The orthogonal groups O2nC1 .p/ for n  1 and odd primes p 6 ˙1
.mod 24/.
(g) The orthogonal groups OC
2n .2/ for n  5.
(h) The orthogonal groups O
2n .2/ for n  3.
(i) The groups G2 .p/.
(j) The groups F4 .p/ for p  5.
(k) The groups E6 .p/ for p 5; 17; 23; 35; 44; 47 .mod 57/.
(l) The groups 2E6 .p/ for p 3; 10; 13; 22; 34; 40; 52 .mod 57/, p > 3.
(m) The groups E8 .p/ for p  7.
(n) The sporadic simple groups with the possible exception of M23 .

Proof. GA-realizations in one variable over Q for the simple groups listed in the
theorem were constructed in Corollary I.5.4 for An , in Theorem I.8.9 for L2 .p/, and
in Theorem III.7.12 for M24 . For the remaining groups this assertion is contained in
Theorem II.10.3. Since in all cases Out.G/ is either trivial or cyclic of order 2, the
assertion follows immediately from Corollary 4.2. t
u

4.2 Fields of Constants with Trivial Brauer Group

If the Brauer group Br.k/ of the field of constants, or more precisely its 2-torsion
group Br2 .k/ is trivial, the rationality condition (R) is automatically satisfied for
every GA-realization in one variable over k. This is a consequence of the following
fact, which is a particular case of the Application in Serre (1979), Ch. X, 7 (see
also Serre (1964), Ch. III, 2, Ex. 2).

Proposition 4.4. Let K D k.t/ be a rational function field in one variable over a
perfect field of constants k of characteristic different from 2 and with Br2 .k/ D 0.
Then every conic over k has a k-rational point.

As a consequence of this we obtain:

312 IV Embedding Problems

Corollary 4.5. Under the assumptions of Proposition 4.4 every GA-realization of a

finite group G in one variable over k is a GAR-realization.

Proof. Since the field R=kQ in the proof of Proposition 4.1 has genus 0, it is either
rational or the function field of a conic without rational points (see for example Artin
(1967), Ch. XVI, Thms. 6 and 8). Proposition 4.4 rules out the second possibility,
so that in fact R=kQ is rational. t
u

Remark. Corollary 4.5 can in particular be applied to fields k of characteristic dif-

ferent from 2 with projective Galois group like k D Qab , since for such fields we
always have Br2 .k/ D 0 by Serre (1964), Ch. II, 2, Prop. 5 (see also Ribes (1970),
Ch. V, Cor. 3.7).

With Corollary 4.5 from Theorem II.10.2 we get immediately:

Theorem 4.6. The following simple groups possess GAR-realizations in one vari-
able over Qab :
(a) The non-abelian simple alternating groups An .
(b) The groups of Lie type G.p/ for 2 < p 2 IP with the possible exception
3
D4 .p/.
(c) The groups S2n .2/, OC 
2n .2/, O2n .2/.
(d) The sporadic simple groups.

Obviously not all non-abelian finite simple groups G can possess GAR-reali-
zations in one variable over k D Q respectively k D Qab , since in general Out.G/
does not embed into Aut.k.t/=k/ Š PGL2 .k/. This necessitates the investigation of
GAR-realizations in several variables.

4.3 GAR-Realizations in Several Variables

The following proposition contains a sufficient condition for GA-realizations in sev-

eral variables constructed from Theorem III.5.10 to satisfy the rationality condition
(R) and hence to be a GAR-realization.

Proposition 4.7. Let G be a group with Z .G/ D 1, A WD Aut.G/ and N =KVA the
Galois extension constructed in Theorem III.5.10 with

Gal.N =KVA / Š Aut.G/ and Gal.N =KV / Š G;

and k the algebraic closure of Q in KV . Further assume that KV =k.t/ is ratio-
nal and has a transcendence basis u1 ; : : : ; us such that the k-vector space U WD
˚siD1 kui remains invariant under Gal.KV =KVA /. Then Gal.N =KVA / is a GAR-
realization of G over k.
4 Verification of the GAR-Property 313

Proof. Let K WD KV , F WD KVA and KN the composite of K with an algebraic

closure kN of k in an algebraic closure of K. Further let R=F be a finite field exten-
N D KN D k.u/
sion with kR N and kQ WD kN \ R. Then with Gal.K=F
N / the group  WD
N N N N
Gal.K=R/ also leaves the k-vector space U WD ˚i D1 kui invariant. By Speiser’s
s

Lemma (Proposition III.3.10) the fixed point set of  in UN hence forms an r-

Q
dimensional k-vector space in R, whose bases are algebraically independent gener-
ating systems of R=k. Q t
u

Remark. The assumption of linearity of Gal.KV =KVA / on U in the above proposi-

tion coincides with the transcendence condition (T) in Völklein (1994) respectively
(L) in Völklein (1996); this can also already be found in Matzat (1992), Anm. 2.

The above proposition together with Theorem III.5.10 yields the existence theo-
rem for GAR-realizations in several variables.

Theorem 4.8. Let G be a finite group with Z .G/ D 1, C 2 Cl.G/s with s  3,

V a symmetry group of C and B V . / the V -symmetrized braid orbit of Π2
˙.CV /= Inn.G/. If B V . / is rigid in ˙.CV /= Inn.G/ and Aut.G/ acts on B V . /,
then there exists a geometric Galois extension N =KV , regular over QVC , with

Gal.N =KVA / Š Aut.G/ and Gal.N =KV / Š G: (4.1)

If here KV =QVC is a rational function field and Gal.KV =KVA / acts linearly on the
k-vector space generated by a transcendence basis of KV =QVC , then Gal.N =KV /
is a GAR-realization of G over QVC .

Remark. In the case V D 1 the rationality of KV =QVC can be checked using the
Braid Orbit Theorem III.5.7.

As an example for the application of this theorem we obtain GAR-realizations

over Q from the G-realizations in Corollary III.10.14 of the simple groups Ln .q/
and Un .q/.

Corollary 4.9 (Völklein (1994)). Assume q D 2e and n D 4m > 2q.

(a) For e even and gcd.n; q  1/ D 1 the groups Ln .q/ possess GAR-realizations
over Q.
(b) For e odd and gcd.n; q C 1/ D 1 the groups Un .q/ possess GAR-realizations
over Q.

Proof. To verify (b) we show that the Galois extensions NQ =KVQ in Theorem III.
10.13 are GAR-realizations of PGUn .q/ over Q. Here by Theorem III.3.11 the field
KVQ is a rational function field over Q of the form KVQ D Q.v1 ; : : : ; v2m ; t/ where

N 1 ˚ : : : ˚ Qv
Qv N 2m D Qt
N 1 ˚ : : : ˚ Qt
N 2m DW UN :

Since the class vector C D .C1 ; : : : ; C2m / is V -symmetric with respect to the sym-
metry group V from Theorem III.10.13, and ˙.CV /= Inn.G/ forms a single HsV -
314 IV Embedding Problems

orbit B for V D Sym.C/, the outer automorphism group

Out.Un .q// Š Gal.IFq 2 =IF2 / Š Z2e

acts on B. Further by construction the vector space UN is mapped to itself by

Gal.Q.v; t/=KVQ / Š Out.Un .q//. Thus all hypotheses of Theorem 4.8 are satisfied
and the assertion follows.
For part (a) we can argue in a similar manner, where here due to the non-trivial
graph automorphism an additional consideration is required. This can be found in
Völklein (1994), Cor. 2. t
u

Frequently it is useful to employ the specializations to two variables discussed in

Chapter III.7, since for these the GAR-property reduces to the symmetry property
that Aut.G/ acts on BSV0 . /.

4.4 Specialization to GAR-Realizations in Two Variables

The basis for the existence proof of GAR-realization in two variables is given by:

Q tQ/ be the Galois extension from the Twisted Braid

Theorem 4.10. Let N =QVC .v;
Orbit Theorem III.7.10 with

Q tQ// Š G:
Gal.N =QVC .v; (4.2)

If Aut.G/ acts on the V -symmetrized braid orbit BSV0 . /, then (4.2) is a GAR-
realization of G over QVC .

Proof. The fact that (4.2) is a GA-realization over QVC follows directly from the
proof of Theorem III.5.10 and hence need not be repeated. It remains to verify the
rationality condition (R). For this let KVA D QVC .v; Q tQ/Out.G/ , R be an intermediate
N v;
field of Q. Q tQ/=KVA with QRN D Q. Q tQ/ D KN V and kQ the algebraic closure of Q in
N v;
R. Then for R WD R \ Ms we have QR D KNV (in the notation of Section III.7.2),
N 0 N
and hence g.R=k/ Q D 0. The group Gal.KN V =R/ permutes the ramified prime divi-

sors Qij =QQ i of KN V =Q.
N u/,
Q and the ramification indices eij correspond to the cycle

lengths of .ˇQi /. Thus the oddness condition (O0 ) entails the rationality of R=k.Q By
assumption we also have

g.R=R/ D g.KV =KV / D gsV . / D 0:

According to the proof of Theorem III.7.3 the numerator divisor Ps of .t  u/ is the

only non-algebraic prime divisor in S0 which therefore does not split in MN s0 .t/=KVA
and thus possesses the residue degree 1 even in R=R. Consequently, R=R and
hence also R=kQ are rational function fields. t
u
4 Verification of the GAR-Property 315

If the field of constants has trivial Brauer group, then according to Proposition 4.4
and Corollary 4.5 the rationality condition is void for all function fields in one vari-
able. As Br.Qab / D 0 we thus obtain from Theorem 4.10, by dropping the oddness
condition (O0 ):

Corollary 4.11. Under the assumptions in the Twisted Braid Orbit Theorem III.7.10
but without the oddness condition (O0 ), if Aut.G/ acts on the V -symmetrized braid
orbit BSV0 . /, then G possesses a GAR-realization in two variables over Qab .

Finally we demonstrate the applicability of Theorem 4.10 on the example of the

groups L3 .3/ and L3 .4/. Note that since Out.L3 .4// D Z2  S3 , the group L3 .4/
cannot possess a GAR-realization in one variable over Q.

Example 4.1. Let G D L3 .3/, C D .3A; 3A; 3B; 2A/ in Atlas notation and V D
h.12/i. Being a rational class vector, C is also V -symmetric. Since the conju-
gacy classes 3A, 3B and 2A of G are invariant under all outer automorphisms,
Out.G/ Š Z2 acts on ˙.CV /= Inn.G/ D ˙.C/= Inn.G/. According to Przywara
(1991), Satz 3(h), ˙.C/= Inn.G/ forms a single H4V -orbit B of length 12, on
which the ˇQi have the permutation types B .ˇQ1 / D .42 ; 22 /, B .ˇQ2 / D .26 / and
B .ˇQ3 / D .6; 4; 12 /. Thus for Œ 2 B we have g4V . / D 0 as well as the oddness
condition (O0 ). Hence by Theorem 4.10 the group L3 .3/ has a GAR-realization in
two variables over Q. t
u

Example 4.2. Let G D L3 .4/, GQ D PL3 .4/ D G  S3 , C Q D .2A; 2C; 6E; 3B/ a class
Q Q
vector of G and V D 1. Then C is rational and hence V -symmetric, and Out.G/ Q Š
Z2 acts on ˙.C /= Inn.G/ D ˙.C/= Inn.G/. By Przywara (1991), Satz 1(h), H4V
V

has a single orbit B on ˙.C/= Inn.G/ of length 54, on which the ˇQi have the per-
mutation types .54 ; 38 ; 25 /, .410 ; 27 / and .312 ; 28 ; 12 /. For Π2 B we thus have
g4V . / D 0 and (O0 ). Consequently GQ has a GAR-realization Gal.N=Q.v; Q tQ// in two
variables over Q.
To obtain a GAR-realization even for the subgroup G D L3 .4/ of G, Q a further
consideration is necessary. Let F be the fixed field in N of Aut.G/ Q D GQ  Z2 , K,
L D Q.v; Q tQ/ the intermediate fields corresponding to G and GQ respectively, and NN ,
FN , KN and LN the fields obtained from the former by extension of constants with Q. N In
particular the field extension NN =F is then Galois. Here K= N LN is a Galois extension of
algebraic function fields in one variable over the field of constants Q. N v/
Q with group
S3 . In this the prime divisors corresponding to the three classes 2C , 6E and 3B
are ramified of orders 2, 2 and 3. Consequently KN is a rational function field over
N v/.
Q. Q Furthermore NN =KN is Galois with group G and corresponding class vector
C D .6  2A; 3  3A/ of length 9.
Now let R be an intermediate field of K=F N of finite degree over F satisfying
N D KN and with field of constants k.
QR Q Then by the proof of Theorem 4.10, the field
RQ WD R \ LN is a rational function field in one variable over the field of constants
RQ WD RQ \ MN 40 , which in turn by the property (R) of Gal.N=L/ is a rational function
field in one variable over k, Q say RQ D k. Q u/
Q with Q. N u/Q D Q. N v/.
Q As MN 40 \ R D R, Q RQ
N
is also algebraically closed in R. So extension of constants with Q of R=R yields Q
316 IV Embedding Problems

N Q.
K= N v/,
Q proving g.R=R/ Q D 0. Since NN =R is Galois, the automorphisms of K=RN
can only permute the three prime divisors of N =K ramified of order 3. Thus R=RQ
N N
possesses at least one prime divisor of odd degree and hence is a rational function
field.
This proves that R=kQ and also K=Q are rational function fields, and so Gal.N=K/
is a GAR-realization in two variables of L3 .4/ over Q. t
u

Some further examples can be found in Matzat (1992) and in Przywara (1991).
5 Frattini Embedding Problems 317

5 Frattini Embedding Problems

Via a reduction theorem of Nobusawa (1961) every finite embedding problem may
be decomposed into a Frattini embedding problem followed by a split embedding
problem. Since until now only split embedding problems have been treated, we here
collect some results on Frattini embedding problems. The embedding theorem pre-
sented in Section 5.2 contains the theorems of Feit (1989) and Völklein (1992a) as
special cases. This is applied to realize the covering groups of several simple groups
S as geometric Galois groups over Q.t/, so for example S D An with kernel Z2
using the An -polynomials constructed by Mestre (1990), and some sporadic simple
groups S with Schur multiplier Z3 .

5.1 A Decomposition Theorem

We recall from Section 1.1 that an embedding problem E .'; / with G D im./
is called a Frattini embedding problem if the kernel H is contained in the Frattini
group ˚.G/ Q of GQ D H  G. The relevance of Frattini embedding problems is illumi-
nated by the following result:

Theorem 5.1 ((Nobusawa (1961)). Every finite (geometric) embedding problem

over a field K can be decomposed into a (geometric) Frattini embedding problem,
followed by a split (geometric) embedding problem, both over K.

Proof. Let E .'; / be a finite embedding problem over K and

1 ! H ! GQ ! G ! 1
(5.1)
 

be the corresponding group extension. Also let U denote a minimal subgroup of GQ

Q Then this satisfies H \ U ˚.U / and U=.H \ U / Š G, hence
with hH; U i D G.

1 ! H \U ! U ! G ! 1
(5.2)
1 1

is a Frattini extension. The action of U on H inside of GQ defines a split group

extension
1 ! H ! E ! U !1 (5.3)
2 2
so that E D H U is a semidirect product of H with U with respect to the given
action. Obviously
 W E D H U ! G; Q .;  / 7! ;
is an epimorphism of E on GQ with kernel isomorphic to H \ U . If the embedding
problems E .'; 1 / and E .'1 ; 2 / belonging to (5.2) and (5.3) are solvable over K,
318 IV Embedding Problems

i.e., if there exist homomorphisms '1 W K ! U and '2 W K ! E with 1 ı '1 D '
and 2 ı '2 D '1 , then 'Q WD  ı '2 W K ! GQ is a homomorphism with

 ı 'Q D  ı  ı '2 D 1 ı 2 ı '2 D 1 ı '1 D '

and hence a solution of E .'; /. If E .'1 ; 2 / is properly solvable, i.e., '1 (by Propo-
sition 1.8) and '2 are epimorphisms, then 'Q is an epimorphism as well, and therefore
a proper solution of E .'; /.
If E .'; / is assumed to be a geometric embedding problem and hence the
geometric embedding problems E .'; 1 / and E .'1 ; 2 / possess geometric (proper)
solutions, then 'Q is itself a geometric (proper) solution of E .'; /. t
u
Corollary 5.2. If in Theorem 5.1 the Frattini embedding problem E .'; 1 / has a
regular solution '1 and moreover the lifted split embedding problem E .'1 ; 2 / has
a regular (proper) solution '2 , then E .'; / has a regular (proper) solution '  .
Proof. This follows immediately from the proof of Theorem 5.1, since under the
above assumptions the field N WD KN ker.'/ is algebraically closed in the solution
 
field N1 WD .KN 1 /ker.'1 / and N1 is algebraically closed in N2 WD .KN 2 /ker.'2 / , where
 
K1 WD K.t1 / and K2 WD K.t1 ; t2 /. t
u

5.2 The Frattini Embedding Theorem

From now on let K=k denote a rational function field over a subfield k of Q N and
N=K a finite geometric Galois extension with group G. By using the Hilbert irre-
ducibility theorem we may then assume that K=k has transcendence degree 1, say
K D k.t/. Then NN WD QN N is a geometric Galois extension of Q.t/ N with group G
and hence may be classified by the Hurwitz classification for function fields in one
variable (Theorem I.4.1): NN D NN  with  2 ˙s .G/.
Now let GQ D H  G be a finite group extension with structure homomorphisms
;  as in (5.1). Then the class of generating systems Œ 2 ˙s .G/= Inn.G/ is called
uniquely liftable to a class of generating systems ŒQ 2 ˙s .G/= Q Inn.G/ Q if there exists
a class vector C Q D .CQ 1 ; : : : ; CQ s / of GQ with .C/
Q D C WD .Œ1 ; : : : ; Œs /, such that
Œ has precisely one preimage ŒQ 2 ˙.C/= Q Inn.G/Q under . (By this definition for
example the class of generating systems Œ; ; 1 respectively Œ; 1;  of Z2 D hi
is uniquely liftable to a class of generating systems ŒQ 2 .2A; .n  1/A; nA/ of Sn ,
see Proposition I.5.2).
For Frattini extensions the unique liftability can be read off from the decomposi-
tion of conjugacy classes.
Proposition 5.3. Let GQ D H  G be a Frattini extension, C D .C1 ; : : : ; Cs / a class
vector of G such that for each i D 1; : : : ; s  1 there exist conjugacy classes CQ i of GQ
with .CQ i / D Ci and
Y jCQ i j
s1 Q
j Inn.G/j
D : (5.4)
jCi j j Inn.G/j
i D1
5 Frattini Embedding Problems 319

Then for each class of generating systems Œ 2 ˙.C/= Inn.G/ of G there exists
Q with .CQ s / D Cs such that ˙.C/=
precisely one CQ s 2 Cl.G/ Q Inn.G/
Q contains one
and hence precisely one preimage ŒQ of Œ under .

Proof. Let  D .1 ; : : : ; s / be an element of Π. Then any preimage Q 2 GQ s

of  under  is a generating system of G. Q The number of such generating sys-
tems whose first s  1 components fall into the respective classes CQ i is equal to
Q
j Inn.G/j=j Inn.G/j by (5.4). Since the last component Q s of Q is uniquely deter-
mined by Q 1    Q s D 1, all these generating systems are contained in a single orbit
under Inn.G/. Q Hence Œ is uniquely liftable to ŒQ 2 ˙.C/= Q Inn.G/,
Q and in partic-
ular the conjugacy class CQ s WD ŒQ s is independent of the choice of  inside Œ and
of Q i inside CQ i for i D 1; : : : ; s  1. t
u

The next proposition facilitates the determination of jCQ i j=jCi j in special cases:

Proposition 5.4. Assume the hypotheses of Proposition 5.3. Then with

fi WD jCQ i j=jCi j we have:
(a) If H CGQ .Q i / and gcd.jH j; o.Q i // D 1, then fi D 1.
(b) If H has trivial intersection with CGQ .Q i /, then fi D jH j.

In the following we call C a k-rational class vector respectively a k-symmetric

class vector if all ı 2  WD Gal.K=K/N satisfy Cc.ı/ D C resp. Cı D C with the
notations from Section I.4.2. (With this definition the Q-rational class vectors are the
rational class vectors in the sense of Section I.4.2.) We can now state the embedding
theorem announced in the introduction which mainly applies to Frattini embedding
problems.

Theorem 5.5 (Frattini Embedding Theorem). Let K D k.t/ be a rational function

field in one variable over a (not necessarily finite) number field k, N =G K a geomet-
ric Galois extension with QNN D NN  2 N N S .G/ and ' W K ! G the corresponding
epimorphism. Further let GQ D H  G be a finite group extension with epimorphism
 W GQ ! G satisfying the two properties:
(1) The class Œ of generating systems of G is uniquely liftable to a class ŒQ of
generating systems of GQ in a k-symmetric class vector C. Q
(2) There exist a prime divisor P 2 IP.K=k/ of degree 1 and an extension P0 of
P onto N 0 WD N Z .G/ such that Z .G/ Q has a complement in the preimage in GQ of
the decomposition group D.P0 =P/.
Then the embedding problem E .' 0 ;  0 /, where ' 0 ;  0 denote the composition of
';  with the canonical epimorphism from G to G 0 WD G=Z .G/, has a geometric
proper solution. In particular there exists a geometric Galois extension NQ =K with
NQ  N 0 and
Gal.NQ =K/ Š GQ and Q N NQ D NN Q : (5.5)

Proof. Let GN WD Gal.NN  =Q.t//

N Š G, WD Gal.NN  =K/,  WD Gal.Q.t/=K/.
N First
from D GN  Gal.NN  =N / it follows that Πis a fixed point under the action of 
on ˙.G/= Inn.G/ given in I.(4.9). By assumption (1) the same holds for the class of
320 IV Embedding Problems

generating systems ŒQ 2 ˙.C/= Q Inn.G/.

Q Hence K  KQ , and by Proposition I.4.6
the field NQ 2 NS .G/ is Galois over K. Now we identify GQ with Gal.NN Q =Q.t//
N N Q N and
Q N N Q N
write H WD Gal.NQ =N / and WD Gal.NQ =K/. Since then by Proposition I.4.6
every element of Q acts as an inner automorphism on G, Q the centralizer C Q .G/

Q is
a supplement to GQ in Q .
In the case Z .G/ Q is even a complement to GQ in Q which
Q D 1 the group C Q .G/

Q Q
we denote by . Its fixed field N is a geometric Galois extension of K with corre-
sponding restriction
'Q W K ! Gal.NQ =K/ Š G: Q (5.6)
Now NQ contains the fixed field N ı of Q  H , and Gal.NN  =N ı / centralizes the
group G, N so N 0 , being the fixed field of C .G/,N is a subfield of N ı and thus
also of N . Hence in particular there exists an epimorphism W GQ ! G 0 with
ı 'Q D ' 0 . Under extension of constants with Q, N the tower of fields NQ =N 0 =K
N N 0 N
goes over into NQ =N =Q.t/. Since .ŒQ / D Œ by construction, and for a suitable
choice of representatives even .Q / D  and hence  ı Q D  , the restriction
'QN W K ! Gal.NN  =Q.t//,
N 'N 0 W K ! Gal.NN 0 =Q.t//
N satisfy  0 ı 'QN D 'N 0 and more-
Q 0 0
over ı 'N D 'N . Hence we have D  and 'Q is a geometric proper solution of
E .' 0 ;  0 /.
In the case Z .G/ Q ¤ 1 we denote the fixed field of C Q .G/ Q by L. As L  N 0

this contains the fixed field K of D.P =P/. Then LN WD QL
0 0 N and KN 0 WD QK N 0 satisfy
Gal.NN Q =L/ N Š Z .G/
Q and Gal.NN Q =KN / Š GQ , where GQ D  .D.P =P// is the
0 0 0 1 0

preimage of D.P0 =P/ in G. Q By the Remark in Section I.3.2, GQ possesses a com-

Q
plement  in inside the group Gal.NN Q =K 0 /, with fixed field M WD NN Q say. The
Q Q
Galois closure MQ of M=K is then the fixed field of \ Q C Q .G/

Q and thus contains L.

N̄σ̃
M
L̄
M̃
Ñ
L̃
L K̄ 
N Q̄(t)
K̃ 
K
K̃
K
Fig. 5.1 Field tower for the Frattini Embedding Theorem

Denote the composites of K; K 0 ; L with the algebraic closure kQ of k in MQ by

Q KQ 0 resp. L,
K; Q so the corresponding Galois groups are Gal.MQ =K/ Q Š G,
Q Gal.MQ =KQ 0 /
Š GQ and Gal.MQ =L/
0 Q Š Z .G/.Q Moreover Gal.MQ =K/ is isomorphic to a subgroup
of the holomorph Hol.G/. Q Here the elements of the complement Gal.MQ =M / act on
Gal.MQ =K/ Q since Gal.NN Q =K 0 / D hGQ 0 ; C Q .G/i

Q as elements of Gal.MQ =KQ 0 / by con-
jugation. Now the centralizer Gal.MQ =L/ of Gal.MQ =K/ Q in Gal.MQ =K/ is isomor-
5 Frattini Embedding Problems 321

phic to a subgroup of Gal.MQ =KQ 0 / containing Gal.MQ =L/Q Š Z .G/ Q (compare with
the proof of Theorem I.3.9). By assumption (2), the group Gal.MQ =L/ Q possesses
Q
a complement in this subgroup and hence also in Gal.M =L/. If we now denote
N NQ D NN Q and Gal.NN Q =NQ /
the fixed field of this complement by NQ , then since Q
Q
CQ .G/ we have

Gal.NN Q =K/ D Gal.NN Q =Q.t//

N  Gal.NN Q =NQ /:

Thus NQ =K is a geometric Galois extension with Gal.NQ =K/ Š GQ and NQ  N 0 .

From this we conclude as in the case Z .G/ Q D 1 that the corresponding restriction
'Q as in (5.6) yields a geometric proper solution of E .' 0 ;  0 /. t
u

The example given above of an extension of Z2 to an Sn shows that in gen-

eral one cannot expect NQ to contain the field N , since otherwise the Sn -extension
constructed in Theorem I.5.3 would contain every geometric Z2 -extension of Q.t/
ramified in two given points.

Remark. In the case Z .G/Q D 1 the condition (2) of Theorem 5.5 is always satis-
fied. According to Theorem I.3.9 it can hence be weakened to the normalizer con-
dition (N).

If condition (2) of Theorem 5.5 is not satisfied, we obtain at least the following:

Corollary 5.6. If only condition (1) of Theorem 5.5 is satisfied, then the embedding
problem given by 'N W K ! Gal.N 0 =K/ Š G=Z .G/ and the residue class map
Q
N W G=Z Q ! G=Z .G/ induced by  possesses a geometric proper solution.
.G/

Proof. The field L in the second part of the proof furnishes a solution field. t
u

5.3 Centerless Frattini Extensions

In the special case of Frattini extensions with trivial center we obtain from Theo-
rem 5.5 the following version of Lemma 2.7 in Feit (1989):

Proposition 5.7. Assume a Frattini extension GQ D H  G with Z .G/ Q D Z .G/ D 1

satisfies the following: The conjugacy classes Ci D Œi of G possess k-rational
classes CQ i as preimages in GQ for i D 1; : : : ; s with fi D 1 for i D 1; : : : ; s  2 and
fs1 D jH j and with Sym.C/ Q D Sym.C/. Then the conditions (1) and (2) of the
Frattini Embedding Theorem are satisfied.

Proof. From the k-rationality of the classes CQ i and Sym.C/ Q D Sym.C/ it follows
Q
that the class vector C is k-symmetric. Thus the assertion follows immediately from
Proposition 5.3 and the above Remark. t
u

As an application we prove a result of Feit where the idea of the proof is to obtain
central Frattini extensions as subgroups of centerless Frattini extensions.
322 IV Embedding Problems

Theorem 5.8 (Feit (1989)). (a) The central group extensions of types 3  A6 and
3  A7 possess G-realizations over Q.
(b) The covering groups 3  S with S 2 fM22 ; McL; Suz; ON; Fi22 ; Fi024 g possess
G-realizations over Q.

Proof. For the proof of (a) we can restrict ourselves to non-split group extensions.
Starting points are the rational rigid class vectors of the symmetric groups S6 and S7
from Proposition I.5.2, hence .2B; 5A; 6B/ for S6 and .2B; 6C; 7A/ for S7 in Atlas
notation. In the case n D 6, both classes 2B and 6B possess a single preimage class
in 3  S6 of the same order, say 2BQ and 6B,Q and 5A splits into three preimage classes,
with one, 5A, Q containing elements of order 5 and two with elements of order 15.
Hence C Q WD .2B; Q 5A;
Q 6B/
Q is a rational class vector of 3  S6, and we have f2 D 1
and f1 D f3 D 3. By Proposition 5.7, all assumptions of Theorem 5.5 are hence
satisfied and there exists a geometric Galois extension over Q.t/ with the group
Q As 5AQ  3  A6 , only two prime divisors of Q.t/=Q of
3  S6 for the class vector C.
degree 1 are ramified in the fixed field K of 3  A6. Thus K=Q is a rational function
field and 3  A6 possesses a G-realization over Q. The same proof applies to the case
n D 7 with f3 D 1 and f1 D f2 D 3 and leads to a G-realization of 3  A7 over Q.
The results collected in (b) are obtained by the same method using the following
rationally rigid class vectors of Aut.S / from Propositions II.9.1, II.9.2, II.9.4, II.9.7,
and the corresponding class indices fi :

Aut.M22 / W .2B; 4C; 11A/; f3 D 1; f1 D f2 D 3;

Aut.McL/ W .3A; 4B; 10B/; f2 D 1; f1 D f3 D 3;
Aut.Suz/ W .2C; 8D; 13A/; f3 D 1; f1 D f2 D 3;
Aut.ON/ W .2B; 4A; 22A/; f2 D 1; f1 D f3 D 3;
Aut.Fi22 / W .2D; 5A; 42A/; f2 D 1; f1 D f3 D 3;
Aut.Fi024 / W .2C; 8D; 29A/; f3 D 1; f1 D f2 D 3:

This completes the proof of Theorem 5.8. t

u

For another realization of the group 3  A6 see also Example III.8.1.

Remark. With the exception of J3 , Theorem 5.8(b) contains all sporadic groups
whose Schur multiplier has order divisible by 3. The rational class vector C D
.2B; 3B; 8B/ of Aut.J3 / in Proposition II.9.8(a), which at present yields the only
known G-realization for J3 over Q, has class indices f1 D f2 D f3 D 3 in 3  Aut.J3 /,
so that the classes of generating systems contained in ˙.C/ do not lift uniquely to
3  Aut.J3 / !

Finally we demonstrate with an example that with the centerless version of the
Frattini Embedding Theorem also centers different from Z3 can be treated.

Example 5.1. Let S WD L3 .4/ and G WD PGL3 .4/. The Schur multiplier of S has
order 48 and possesses precisely one elementary abelian factor group E4 of order 4.
5 Frattini Embedding Problems 323

This defines a central non-split group extension

1 ! E4 ! SQ ! L3 .4/ ! 1:

By a theorem of Thompson (1973) there exists an automorphism of the full covering

group of L3 .4/ cyclically permuting the three non-trivial elements of E4 . Obviously
this is already induced by PGL3 .4/, hence the above exact sequence may be lifted
to a centerless Frattini extension:

1 ! E4 ! GQ ! PGL3 .4/ ! 1:

According to Example 4.2 the group PL3 .4/ possesses a G-realization in two
and hence also in one variable over Q with the class vector .2A; 2C; 6E; 3B/, say
N=Q.tQ/. As 2A; 3B  PGL3 .4/ the fixed field of G D PGL3 .4/ in this field exten-
sion is rational, say N G D Q.t/, and the PGL3 .4/-extension N=Q.t/ has the class
vector .2A; 2A; 3A; 3B; 3B/. The preimage classes 2AQ and 3AQ in GQ of 2A and 3A
are uniquely determined by the conditions f1 D f2 D f3 D 1, and by Proposi-
tion 5.4(b) we have f4 D f5 D 4. Hence the preimage vector C Q of C is rational
Q
with Sym.C/ D Sym.C/ and by the Frattini Embedding Theorem together with
Proposition 5.7 there exists a Galois extension NQ =Q.t/ containing N=Q.t/ with
Gal.NQ =Q.t// Š G. Q Here the fixed field of SQ , which coincides with the fixed field of
Q
L3 .4/ in N=Q.t/, is rational, say NQ S D Q.u/. Hence there exists a geometric Galois
extension NQ =Q.u/ with Gal.NQ =Q.u// Š SQ and Z .SQ / D E4 . t
u
In the next section we treat central Frattini embedding problems without the
detour via a centerless Frattini embedding problem.

5.4 Central Frattini Extensions and 2  An

For central Frattini extensions the assumptions of Theorem 5.5 can for example be
verified by use of the following version of a theorem of Völklein (1992a):
Proposition 5.9. Assume a central Frattini extension GQ D H  G with
G D Gal.N=K/ and Z .G/ D 1 satisfies:
(1’) We have gcd.o.i /; jH j/ D 1 for i D 1; : : : ; s.
(2’) The field N=k possesses a prime divisor of degree 1.
Then the conditions (1) and (2) of Theorem 5.5 are satisfied.
Proof. By Proposition 5.4(a) it follows from (1’) that fi D 1 for i D 1; : : : ; s. Now
let CQ i for i D 1; : : : ; s  1 denote the preimage classes of Ci in GQ whose elements
have the same order as those of Ci . Then Πmay by Proposition 5.3 be uniquely
lifted to a class of generating systems ŒQ of GQ with Q i 2 CQ i for i D 1; : : : ; s  1,
which uniquely determines a conjugacy class CQ s D ŒQ s . If now C Q D .CQ 1 ; : : : ; CQ s /
N
were not k-symmetric, then there would exist a ı 2 Gal.Q.t/=k.t// with CQ sc.ı/ ¤ CQ s
and hence ŒQ ¤ ŒQ , in contradiction to the unique liftability of Œ D Œ ı . Thus
ı

(1) is satisfied.
324 IV Embedding Problems

Now let P0 be the prime divisor of degree 1 in N=k whose existence is assumed
in (2’). Then the decomposition group D.P0 =P/ in Gal.N=K/ equals the inertia
group and thus has order coprime to H . The group H D Z .G/ Q hence has a com-
0 Q
plement in the preimage of D.P =P/ in G, which proves (2). t
u
Using the above version of the Frattini embedding theorem we now construct
G-realizations over Q for the groups 2  An . Since the An -extensions exhibited in
Theorem I.5.3 do in general not embed into extensions with the covering groups
(see Vila (1985)), we first present families of An -polynomials found by Mestre
(1990).
Lemma 5.10. For the general polynomial of odd degree n  3 over Q

g.X / D X n  s1 X n1 C : : : ˙ sn 2 Q.s1 ; : : : ; sn /ŒX

there exists a coprime polynomial h 2 Q.s/ŒX and a separable polynomial q 2

Q.s/ŒX , both of degree n  1, with
gh0  g 0 h D q 2 ; (5.7)
where 0 denotes formal differentiation with respect to X.
Proof. Let K WD Q.s/ and KQ WD Q.t/ where t1 ; : : : ; tn are the zeroes of g.X / in
an algebraic closure KN of K. Then K=K Q is Galois with group Sn , which acts on
t1 ; : : : ; tn in its natural permutation representation: ti D t.i / . The equation (5.7) in
KŒX is obviously equivalent to the identity
.h=g/0 D .q=g/2 (5.8)
in the quotient field K.X /. As the degree of the numerator is smaller than the degree
of the denominator, both sides possess partial fraction expansions in K.X N / of the
form
h X ui q X vi
n n
D ; D :
g X  ti g X  ti
i D1 i D1

Thus the solvability of (5.8) is equivalent to the solvability of the following system
of algebraic equations
0 1
Xn
v
ui D vi2 and vi @ A D 0 for i D 1; : : : ; n:
j
(5.9)
ti  tj
i ¤j D1

If here vi and hence also ui vanished, then ti would be a zero of h in contradiction

to the assumed coprimeness of g and h. Thus (5.9) reduces to the linear system for
the vj
X n
vj
D 0 for i D 1; : : : ; n: (5.10)
ti  tj
i ¤j D1

Since the coefficient matrix is skew symmetric of odd dimension n, it possesses an

N / WD X n  X we deduce that the
eigenvalue 0. Specializing g for example to g.X
5 Frattini Embedding Problems 325

rank of this matrix is actually equal to n  1. Consequently there exists an eigenvec-

tor v 2 KQ n to the eigenvalue 0 uniquely determined up to scalar multiples. Since the
rows and columns of the coefficient matrix of the system of linear equations (5.10)
are permuted by  2 Sn D Aut.K=K/ Q like the variables ti , there exist x./ 2 K
with
vi D x./v.i / and x./ D x./ x./ for ;  2 Sn :
By Hilbert’s Theorem 90 there then also exists an eigenvector v 2 KQ n satisfying
vi D v.i / for all  2 Sn . This yields solution polynomials h; q 2 KŒX for (5.8)
and thus (5.7) with deg.h/ n  1.
N As is easily verified,
To prove the remaining condition we again specialize g to g.
the polynomials

hN WD n2 X n1  .n  2/2 ; qN WD nX n1 C n  2 2 QŒX ; (5.11)

then constitute solutions of equation (5.7) with g replaced by g. N In particular the

solutions satisfy in this case that hN is prime to gN of degree n  1 and qN is separable.
Hence this is also true for the generic solutions h; q of the original non-specialized
problem. t
u

Proposition 5.11. The polynomials g; h in Lemma 5.10 satisfy:

.a/ f .t; X / WD g.X /  t h.X / 2 Q.s; t/ŒX (5.12)

has the Galois group Sn over Q.s; t/ and group An over Q.t; t/ respectively, where
Q.t/ is the splitting field of g.X / over Q.s/.

g.u/h.X /  g.X /h.u/

.b/ fQ.u; X / WD 2 Q.s; u/ŒX (5.13)
X u
has the Galois group Sn1 over Q.s; u/ respectively An1 over Q.t; u/.

Proof. By the coprimeness of g and h in KŒX asserted in Lemma 5.10 the polyno-
mial f .t; X / 2 K.t/ŒX is certainly irreducible. Moreover its discriminant D.f / is
a polynomial in t, whose degree is bounded by 2.n  1/ according to the expansion
theorem for determinants applied to the resultant R.f; f 0 /. If bi is one of the n  1
N then h.bi / ¤ 0 due to the coprimeness
distinct zeroes b1 ; : : : ; bn1 of q 2 KŒX in K,
of g and h. Setting ci WD g.bi /= h.bi /, the polynomials fi .X / WD f .ci ; X / satisfy
fi .bi / D 0. Thus from

h.bi /fi0 .bi / D h.bi /g 0 .bi /  g.bi /h0 .bi / D q 2 .bi /;

it follows moreover that fi0 .bi / D fi00 .bi / D 0, i.e., that bi is even a threefold zero
of fi .X /. By the Dedekind discriminant theorem at least .t  ci /2 is a divisor of
D.f /. Computation of the discriminant of the specialized polynomial fN D gN  t hN
with gN D X n  X and hN from (5.11)

D.fN/ D D.g/.n
N n .n  2/n2 t n1 C 1/2 (5.14)
326 IV Embedding Problems

shows that the ci are all distinct. Thus the above estimate for the degree of the
discriminant as a polynomial in t forces

Y
n1
D.f .t; X // D c0 .t  ci /2 with c0 D2 D.g/ in Q.s/ ; (5.15)
i D1

since D.f .0; X // D D.g/. The above shows that f .t; X / decomposes modulo
.t  ci / in the form

fi .X / D f .ci ; X / D .X  bi /3 gi .X /

with a separable polynomial gi .X / 2 KŒX of degree n  3 and prime to X  bi .

Since t  ci cannot be an inessential discriminant divisor the decomposition of
f .t; X / over the residue class field K of K.t/ modulo .t  ci / corresponds to the
decomposition of the numerator divisor Pi of .t  ci / in a root field L=K.t/ of
f .t; X / (see Narkiewicz (1990), IV, 3, Thm. 4.12). Hence Pi splits in L=K.t/
as Pi D Q3i Ri with deg.Qi / D 1 and a divisor Ri prime to Qi . If N=K.t/ is the
Galois closure of L=K.t/, then for each extension P Q i of Pi onto N the inertia
Q
group I.Pi =Pi / is a cyclic group of order 3, which by Theorem I.9.1 in the natural
permutation representation of Sn is generated by a 3-cycle. Thus by the Hurwitz
classification the Galois group Gal.f .t; X // contains a transitive normal subgroup
generated by 3-cycles. Consideration of the possible blocks of imprimitivity shows
that such a group is necessarily also primitive, so it follows from a theorem of Jor-
dan (see Huppert (1967), Kap. II, Satz 4.5) that An Gal.f .t; X //. Therefore
(a) follows by the discriminant formula (5.15) since Gal.g/ Š Sn over Q.s/ and
Gal.g/ D 1 over Q.t/.
With exactly the same proof as for Corollary I.9.11 it follows from this that
Gal.fQ.u; X // is the one point stabilizer of Sn and An respectively, which implies (b).
t
u

Theorem 5.12 (Mestre (1990)). For n  4 the covering groups 2  An of An possess

G-realizations over Q.

Proof. For odd n we obtain from the polynomial f .t; X / 2 Q.t; t/ŒX in (5.12) by
specialization of the ti to ai 2 Q with the Hilbert irreducibility theorem polynomials

fa .t; X / 2 Q.t/ŒX with Gal.fa / Š An (5.16)

and a 2 Qn with pairwise distinct components ai . The splitting field Na =Q.t/ of

such a polynomial is then geometric over Q.t/ and belongs in the Hurwitz classifi-
cation to the class vector C consisting of n  1 copies of the class of 3-cycles. Now
gcd.2; 3/ D 1, so condition (1’) in Proposition 5.9 is satisfied. As

Y
n
fa .0; X / D .X  ai /;
i D1
5 Frattini Embedding Problems 327

the numerator divisor P0 of t splits completely in the root field La =Q.t/. Thus
the image of the decomposition group D.P Q 0 of P0 to
Q 0 =P0 / of any extension P
Na in the natural permutation representation of An is trivial. So we even have
D.P Q 0 =P0 / D 1, and P0 splits also in Na =Q.t/ into prime divisors of degree 1.
In particular, the condition (2’) in Proposition 5.9 is satisfied and by the Frattini
Embedding Theorem Na =Q.t/ may be embedded into a Galois extension NQ a =Q.t/
with
Gal.NQ a =Q.t// Š 2  An:
The fixed field L of An1 inside Na =Q.t/ has genus g.L=Q/ D 0 by the Hurwitz
genus formula and is rational since deg.Qi / D 1 for the divisors Qi of Pi defined
in the proof of Proposition 5.11. So we have L D Q.u/ and

Gal.NQ a =Q.u// Š 2  An1;

i.e., also the suitably specialized An1 -extensions in (5.13) can be embedded into
geometric Galois extensions over Q.u/ with group 2  An1. t
u
Explicit generators for the Galois extensions with group 2  An are calculated
in the papers of Schneps (1992) and Swallow (1994), based on work of Crespo
(1989).
Remark 1. For n ¤ 6; 7 the group 2  An in Theorem 5.12 is the full covering group
of An . In the two exceptional cases n D 6 and n D 7 the Schur multiplier of An is
cyclic of order 6 (compare Theorem 5.8(a)).

Remark 2. By application of the Hurwitz classification for arbitrary algebraically

closed intermediate fields of C=Q (see for example Matzat (1987), Kap. II, 1.1) in
Theorem 5.5, the specialization of t to a 2 Qn with pairwise distinct components can
be avoided and one thus obtains G-realizations of 2  An in n C 1 variables over Q.
In the last section of this paragraph we will see that for the An -extensions of
Mestre every central embedding problem can be solved.

5.5 Central Extensions of An

The starting point is given by the following easy group theoretic fact.
Proposition 5.13. Let G be a perfect group with universal central extension group
R and GQ D H  G a central group extension with canonical epimorphisms W R !
G and  W GQ ! G. Then there exists a unique homomorphism W R ! GQ with
ı D .
Proof. The central Frattini extension U from (5.2) contained in GQ D H  G is a
factor group of R. By composition of the canonical map from R onto U with the
embedding of U into GQ we obtain the stated homomorphism W R ! G, Q which is
uniquely determined by the equation  ı D . t
u
328 IV Embedding Problems

This yields a reduction theorem for the corresponding embedding problems.

Theorem 5.14. Let N=K be a finite Galois extension with perfect Galois group G
and canonical epimorphism ' W K ! G. Further let W R ! G,  W GQ ! G be as
in Proposition 5.13. Then we have:
(a) If 'Q is an ordinary (geometric/regular) solution of E .'; /, then ı 'Q is an
ordinary (geometric/regular) solution of E .'; /.
(b) If K is a Hilbertian field, then the existence of an ordinary (geometric/regular)
solution of E .'; / implies the existence of an ordinary (geometric/regular) proper
solution of E .'; /.

Proof. Part (a) of the assertion is an immediate consequence of Proposition 5.13.

Part (b) follows from (a) using Corollary 2.5. t
u

A Galois extension N=K for which the embedding problem for each central
extension GQ of G D Gal.N=K/ is solvable is called a universally central embed-
dable Galois extension. Examples for these are given in the following:

Corollary 5.15. Let K be a Hilbertian field of characteristic zero. Then we have:

(a) The geometric Galois extensions N =An Q.t/ in Theorem 5.12 for n  5 and
n ¤ 6; 7 are universally central embeddable.
(b) The geometric Galois extensions N =S Q.t/ with S 2 fMcL; Suz; ON; Fi22 ;
Fi024 g in Theorem 5.8 are universally central embeddable.

Proof. The proof follows from Theorem 5.14 and the cited Theorems, since in (a)
the groups 2  An for n D 5 and n  8 and in (b) the groups 3  S are the full covering
groups of An respectively S (compare the remarks in Sections 5.4 and 5.3). t
u

Remark. The result of Corollary 5.15(a) remains true for n D 6 and n D 7 if one
uses the G-realizations over Q for 6  A6 and 6  A7 constructed by Mestre (1998)
(compare also the end of Section 6.4).

In the next paragraph we treat central embedding problems which cannot be

solved by unique liftability of classes of generating systems.
6 The Quadratic Trace Form 329

6 The Quadratic Trace Form

In this paragraph we study central embedding problems with kernel Z2 on a coho-

mological basis. The main ingredient here is a criterion going back to Serre (1984),
with which the cohomological embedding obstruction can be expressed in terms
of the Hasse-Witt-invariant of the quadratic trace form of the corresponding field
extension. Thus it is possible to realize all central extensions of the symmetric
groups Sn for n  4 as geometric Galois groups over Q.t/.

6.1 The Cohomological Embedding Obstruction

As before let N =G K be a finite Galois extension with epimorphism ' W K ! G

and GQ D H  G a finite group extension with epimorphism  W GQ ! G. If here the
kernel H is abelian, then the equivalence class of the group extension H  G with
given action of G on H is described by a cohomology class h 2 H 2 .G; H / (see for
example Huppert (1967), Kap. I, Satz 17.2, or Suzuki (1982), Ch. 2, 7). Via the
inflation

'  W H 2 .G; H / ! H 2 . K ; H /; h 7! '  .h/ DW h.'; /; (6.1)

h is mapped to an element h.'; / 2 H 2 . K ; H /. With this the solvability of E .'; /

may be decided.
Theorem 6.1. Let N =G K be a finite Galois extension with canonical epimorphism
' W K ! G and GQ D H  G a finite group extension with abelian kernel defined
by the cocycle b D .b; / 2 Z 2 .G; H / in the cohomology class h 2 H 2 .G; H /, and
 W GQ ! G the canonical epimorphism.
(a) The embedding problem E .'; / is solvable if and only if h.'; / D 0 in
H 2 . K ; H /.
(b) Every cochain c D .c / 2 C 1 . K ; H / with @.c/ D '  .b/ defines a solution
'Qc of E .'; / with kernel c WD f 2 N j c D 1g and vice versa.
Proof. We employ the following commutative diagram with the fiber product (sub-
direct product) Q WD GQ G K and the projections p1 W Q ! GQ and p2 W Q ! K :
"
1 ! H ! GQ G K

! ! 1
? p2 K?
? ?
p1 ?
y '?
y
(6.2)
 
1 ! H ! GQ ! G ! 1

Since the group extension in the top row of (6.2) belongs to the cohomology class
h.'; /, it splits precisely if h.'; / D 0. But then there exists a homomorphism
" W K ! Q with p2 ı " D IdK , and 'Q WD p1 ı " is a homomorphism from K to
GQ with  ı 'Q D ' and hence yields a solution of E .'; /. Conversely, if the embed-
ding problem E .'; / possesses a solution ',Q then the homomorphism " W K ! Q ,
330 IV Embedding Problems

 7! .'.
Q /;  / splits the group extension in the top row of (6.2). But this implies
h.'; / D 0.
For the explicit description of solutions we go back to the defining factor set
'  .b/. Since this splits, there exists a c D .c / 2 C 1 . K ; H / with @.c/ D '  .b/,
i.e., with
c'.
1
2/
c2 D c1 2 b'.1 /;'.2 / : (6.3)
From this we may define a homomorphism

'Qc W K ! GQ D f.; / j  2 H;  2 Gg;  7! .c ; '. //; (6.4)

from K to GQ D H  G with the usual composition

.1 ; 1 /.2 ; 2 / D .b1 2 2 ; 1 2 /;

1 ;2 1

with  ı 'Qc D '. Since conversely every homomorphism 'Q W K ! GQ with  ı 'Q D '
via (6.4) yields a 1-cochain c D .c / satisfying (6.3), and thus leads to @.c/ D '  .b/,
the above procedure in fact produces all solutions of E .'; /. t
u

Remark. If in Theorem 6.1 we replace K by the Galois group E of a Galois exten-

sion M=K containing N , and ' by the restriction W E ! G, then E .'; / pos-
sesses a solution inside M=K if and only if  .h/ D 0 in H 2 .E; H /.

Corollary 6.2. If in Theorem 6.1 in addition the field K is Hilbertian, then h.'; / D
0 is even equivalent to the proper solvability of E .'; /.

Proof. This statement follows from the above Theorem by the Theorem of Ikeda
(resp. Corollary 2.5). t
u

The following proposition contains an illustrating example for Theorem 6.1

which will be needed later on. For this denote by .a; b/ the class of the quaternion
algebra
Aa;b D KŒi; j with i 2 D a; j 2 D b; ij D j i
in the Brauer group Br2 .K/, which is isomorphic to H 2 . K ; Z2 / (see for example
Serre (1979), Ch. XIV, 2).
p
Proposition 6.3. Let K be a field of characteristic different from 2, N WD K. x/
a quadratic extension of K with group G D Z2 and epimorphism ' W K ! Z2 .
Further let  be the canonical epimorphism of Z4 onto Z2 . Then the following are
equivalent:
(1) The embedding problem E .'; / has a (proper) solution.
(2) The element x lies in the norm group of N=K.
(3) In H 2 . K ; Z2 / Š Br2 .K/ we have .1; x/ D 0.

Proof. Since Z4 D Z2  Z2 is a Frattini extension, every solution 'Q of E .'; / is

already proper by Proposition
p 1.8. The solution field NQ then has the form NQ D
p
N. y/ with yDa Cb x and a; b 2 K. A generating element  of Gal.NQ =K/ Š Z4
6 The Quadratic Trace Form 331

then acts via

p p p p p p
W y 7!y 0 7!  y 7!  y 0 ; x 7!  x;
p p
where y 0 WD a  b x. Consequently c WD yy 0 x 1 is -invariant and thus an ele-
ment of K  with c 2 x D yy 0 D a2  b 2 x. In particular we get that

a2  b 2 x x
xD 2
D NN=K (6.5)
c c
is a norm of N=K. Conversely a solution of (6.5) leads
p p to a solution 'Q W K !
Gal.NQ =K/ of E .'; / with NQ WD N. y/ for y WD a C b x. This proves the equiv-
alence of (1) and (2).
In the case b 2 C c 2 ¤ 0 it follows from (6.5) that x may be written as the sum of
two squares:
 2  2
ab ac
xD C DW u2 C v 2 : (6.6)
b2 C c2 b2 C c2
A corresponding representation also exists in the case b 2 C c 2 D 0 since then 1 is
a square in K. Thus we have

.1; x/ D .1; u2 C v 2 / D .1; w 2 C 1/ D .w 2 ; w 2 C 1/ D 0

for w WD uv . Conversely from .1; x/ D 0 there follows the existence of nontrivial

a; b; c 2 K with a2  b 2 x  c 2 x D 0 (see for example Jacobson (1980), Ch. 9.15),
and therefore (6.5). This shows the equivalence of (2) and (3). t
u
In the next section we introduce the second ingredient of Serre’s formulae.

6.2 The Trace Form

Let K be a field with characteristic different from 2 and L=K a separable field
extension of degree n. Then

qL=K W L ! K; x 7! traceL=K .x 2 /; (6.7)

is a non-degenerate quadratic form on L with corresponding bilinear form

1
bL=K W L  L ! K; .x; y/ 7! .qL=K .x C y/  qL=K .x/  qL=K .y//: (6.8)
2
P
Denote by f´1 ; : : : ; ´n g a basis of L=K and write x 2 L as x D niD1 xi ´i with
xi 2 K. Then

X
n
qL=K .x/ D bij xi xj with bij D bL=K .´i ; ´j / D traceL=K .´i ´j /; (6.9)
i;j D1
332 IV Embedding Problems

and this defines a non-degenerate quadratic form over K:

X
n
qL=K .X/ D bij Xi Xj 2 KŒX1 ; : : : ; Xn : (6.10)
i;j D1

If further
X
n
qO L=K .X/ WD ai Xi2 (6.11)
i D1

is a diagonal form equivalent to qL=K .X/, then the element

X
w2 .qL=K / WD .ai ; aj / 2 Br2 .K/; (6.12)
1i <j n

which is independent of the choice of the diagonal form qO L=K , is called the Hasse-
Witt-invariant of qL=K .X/ (see for example Jacobson (1980), Prop. 9.9 or also Serre
(1992), Ch. 9). By the above, w2 .qL=K / only depends on the field extension L=K.
Furthermore the element of the square class group K  =.K  /2

Y
n
ai .K  /2 D dL=K .K  /2 (6.13)
i D1

is the discriminant of the quadratic form qL=K .X/. It contains the discriminant dL=K
of the field extension L=K.
The next theorem furnishes an important tool for the computation of the Hasse-
Witt-invariants of quadratic forms over rational function fields K D k.t/:
Theorem 6.4 (Harder). Let k be a field of characteristic different from 2 and
.M; b/ a regular bilinear space over the polynomial ring kŒX consisting of a
finitely generated projective kŒX -module M and a non-degenerate bilinear form
b W M  M ! kŒX . Then .M; b/ is extended from k, i.e., there exists a vector space
V over k such that
.M; b/ D kŒX ˝k .V; b/: (6.14)
The proof for this result can be found in Scharlau (1985), Ch. 6, Thm. 3.3.
Now call a non-degenerate quadratic form over a ring R a strictly non-degenerate
quadratic form if its discriminant is a unit in R. Then we obtain the following variant
of the Theorem of Harder (see also Serre (1992), Thm. 9.1.3):
Corollary 6.5. A strictly non-degenerate quadratic form over kŒX is extended from
k, i.e., it is equivalent over kŒX to a quadratic form with coefficients in k.
From this we can easily deduce:
Theorem 6.6 (Serre (1992)). Let k be a field of characteristic different from 2, K WD
k.t/ and L=K a finite separable field extension with Galois closure N=K. Further
Q of N=K not dividing the numerator divisor of t
assume that for all prime divisors P
Q
the order of the inertia group I.P=P/ in N=K is odd. Then qL=K .X/ is equivalent
over K to a quadratic form with coefficients in k.
6 The Quadratic Trace Form 333

Proof. Let R WD kŒt , S the algebraic closure of R in L and Dt the t-different of

L=K. By the assumption on the orders of the inertia groups in N=K the inverse of
the different D1
t is the square of a fractional ideal, D1
t D A say. The R-module
2

A is free of rank n WD ŒL W K and by construction self-adjoint under the quadratic

form qL=K .X/. Consequently dL=K is a unit in R and thus an element of k. This
proves the assertion using Corollary 6.5. t
u

6.3 A Criterion of Serre

For the understanding of Serre’s criterion one needs some prerequisites from the
cohomology theory of the symmetric groups Sn : It is well known that

H 1 .Sn ; Z2 / D hsn i Š Z2 for n  2; (6.15)

H 2 .Sn ; Z2 / D hQsnC ; sQn i Š Z2 ˚ Z2 for n  4; (6.16)

(see for example Huppert (1967), Kap. V, Satz 25.8), where

sn W Sn ! f˙1g

is the signature and sQn˙ are the cohomology classes of the following two non-split
central extensions SQn˙ of Sn :

SQnC D h; 1 ; : : : ; n1 j  2 D 1; i2 D 1;

.i i C1 /3 D 1; i D i ; i j D j i for j  i C 2i; (6.17)

SQn D h; 1 ; : : : ; n1 j  2 D 1; i2 D ;

.i i C1 /3 D ; i j D j i for j  i C 2i: (6.18)

Further in H 2 .Sn ; Z2 / with the cup product the following holds:

sQnC C sQn D sn  sn DW sQn (6.19)

where sQn is the cohomology class of the central extension SQn D Sn Z2 Z4 .
Now as in the previous section let K be a field of characteristic different from 2
and L D K.x/ a separable field extension of degree n with minimal polynomial
f 2 KŒX . Then any numbering x D x1 ; : : : ; xn of the roots of f defines an embed-
ding of Gal.f / into Sn and thus with the natural homomorphism K ! Gal.f / a
homomorphism ' W K ! Sn . This induces further homomorphisms

'1 W H 1 .Sn ; Z2 / ! H 1 . K ; Z2 / Š K  =.K  /2 ; (6.20)

'2 W H 2 .Sn ; Z2 / ! H 2 . K ; Z2 / Š Br2 .K/: (6.21)

334 IV Embedding Problems

The images of '1 and '2 can be computed explicitly by formulae discovered by
Serre (1984).

Proposition 6.7 (Serre’s formulae). With the notations introduced above we have

.a/ '1 .sn / D dL=K  .K  /2 ; (6.22)

.b/ '2 .QsnC / C .2; dL=K / D w2 .qL=K / D '2 .Qsn / C .2; dL=K /: (6.23)

Part (a) of Proposition 6.7 follows directly by combining the map ' of K to Sn
with the sign sn . The proof of part (b) uses non-abelian cohomology and will not
be given here. The reader can find it in the original paper of Serre (1984) or also in
Fröhlich (1985). In Serre’s article, only the first part of equation (6.23) is shown,
but the second follows easily with (6.19) and part (a) since

'2 .QsnC / C .2;dL=K / D '2 .Qsn / C '2 .sn  sn / C .2; dL=K /

D '2 .Qsn / C .dL=K ; dL=K / C .2; dL=K /
D '2 .Qsn / C .1; dL=K / C .2; dL=K / D '2 .Qsn / C .2; dL=K /:

An elementary proof of Proposition 6.7(b) using orthogonal representations

instead of non-abelian cohomology has been discovered by Ledet (2000) (see also
Ledet (2005), Thm. 6.2.2).
Now let N=K be the Galois hull of L=K with Galois group G WD Gal.N=K/ Š
Gal.f / and
1 ! Z2 ! GQ ˙ ! G ! 1 (6.24)
˙ ˙
the central group extension induced by

1 ! Z2 ! SQn˙ ! Sn ! 1:

Then by Theorem 6.1 the corresponding embedding problems E .';  ˙ / are solvable
precisely when '2 .Qsn˙ / D 0. With these interpretations Serre’s formulae (6.23) read
as follows:

Theorem 6.8 (Serre (1984)). Let K be a field of characteristic different from 2,

L WD K.x/ a separable extension of degree n  4 and G D Gal.f / Sn the Galois
group of the minimal polynomial f of x over K. Further let N be the Galois closure
of L=K and ' W K ! G the canonical epimorphism. Then the following assertions
are equivalent for the central Galois extension GQ ˙ D Z2  G:
(1) The embedding problem E .';  ˙ / has a solution.
(2) In Br2 .K/ we have w2 .qL=K / D .˙2; dL=K /.
If these are satisfied and either K is Hilbertian or E .'; / does not split, then
E .';  ˙ / even possesses a proper solution.

Remark. If the Galois group G D Gal.f / in Theorem 6.8 is already a subgroup of

An , then dL=K 2 .K  /2 and hence .˙2; dL=K / D 0.
6 The Quadratic Trace Form 335

With Theorem 6.8 one obtains the original proof of Theorem 5.12 of Mestre
(1990). For this one essentially has to show that the Hasse-Witt-invariant of the
quadratic trace form qLa =Q.t / of the root field La =Q.t/ of fa .t; X / is trivial. By
Theorem
Qn 6.6 it suffices to check this at t D 0. The latter is obvious since fa .0; X / D
i D1 .X  ai / with ai 2 Q. To deduce finally that the so obtained solution of the
embedding problem in AQn D 2  An is proper geometric, one can, for example, use
Proposition 1.8.
In the next section we present the application of Serre’s criterion to central exten-
sions of Sn .

6.4 Central Extensions of Sn

The following reduction step for central embedding problems was outlined to us by
H. Völklein.

Theorem 6.9. For a Galois extension N =G K the following are equivalent:

(1) Every central (geometric) embedding problem has a (geometric) solution.
(2) Every (geometric) embedding problem of Gal.N=K/ into subdirect products
.M  G/ G=G 0 P , where P is a central Frattini extension of G=G 0 and M  G is a
representation group of G, has a (geometric) solution.

Proof. We reduce (2) to (1). For G WD Gal.N=K/ let

1 ! H ! GQ ! G ! 1
 
be a central group extension and E .'; / the corresponding embedding problem
with the canonical epimorphism ' W K ! G. By the Theorem 5.1 of Nobusawa
this can be decomposed into a central Frattini embedding problem followed by a
split embedding problem with abelian kernel. Since the latter is always solvable
by Theorem 2.4, we may assume without loss that H is contained in the Frattini
subgroup of GQ so that
H ˚.G/ Q \ Z .G/ Q :

Let G D F=R be a presentation of G as factor group of the free group F on n

generators modulo the relations R. Since H ˚.G/, Q the extension GQ also has a
Q Q
presentation on n generators, say G D F=R, such that H is the preimage of R
under the canonical epimorphism from F=RQ to F=R. As moreover H is central in
Q we see that RQ  ŒR; F , and GQ is a factor group of F=ŒR; F . By Huppert (1967),
G,
Kap. V, Satz 23.5, FN D F=ŒR; F fits into an exact sequence

1 ! A  M ! F=ŒR; F ! G ! 1;

where M D .F 0 \ R/=ŒR; F is the Schur multiplier of G and A is free abelian

of rank n. Finite factor groups of FN which project onto G correspond to normal
subgroups U of A  M . Without loss we may restrict to the case where U D A.m/
336 IV Embedding Problems

A  M is the free subgroup of m-th powers of the first factor, for some m 2 IN, since
arbitrary finite epimorphic images of FN are clearly factors of such with kernel U as
above. Then A=U is a central normal subgroup of GQ D FN =U , with G=.A=U Q /Š
M  G a representation group of G. Also, the commutator group G of GQ is normal
Q 0

in GQ and disjoint from A=U since M D .F 0 \ R/=ŒR; F Š G 0 \ H . Now, we have

ı ı
Q GQ 0 Š F=RQ F 0 =.RQ \ F 0 / Š F=RQ F 0 R=
G= Q RQ Š F=F 0 RQ and G=G 0 Š F=F 0 R:

Q GQ 0 is an extension
This shows that G=

1 ! F 0 R=F 0 RQ ! G=
Q GQ 0 ! G=G 0 ! 1

with kernel
ı
Q Š R=RQ .R \ F 0 R/=
F 0 R=F 0 RQ Š R=.R \ F 0 R/ Q RQ Š H=M Š Zm
n
;

so that
n
1 ! Zm Q GQ 0 ! G=G 0 ! 1:
! G= (6.25)
This is a Frattini extension since A=U H Q so
˚.G/, n
Zm Q G /. Thus GQ
˚.G= Q0
may be obtained as a subdirect product
Q GQ 0 :
GQ D .M  G/ G=G 0 G= t
u

Remark. If K is a Hilbertian field, then the existence of a solution to a central

embedding problem always implies the existence of a proper solution (compare the
proof of Theorem 5.14(b)).

If G is a group with cyclic commutator factor group of order p 2 IP the criterion

takes an especially simple form:

Corollary 6.10. For a Galois extension N =G K with jG=G 0 j D p 2 IP the following

are equivalent:
(1) Every central (geometric) embedding problem has a (geometric) solution.
(2) Every (geometric) embedding problem of Gal.N=K/ into a group of type
GQ Zp Zpm for m  1 has a (geometric) solution, where GQ is an arbitrary repre-
sentation group of G.

We can now prove the main result of this section.

Theorem 6.11 (Sonn (1991)). For n  4 every central group extension of Sn pos-
sesses a G-realization over Q.

Proof. According to Corollary 6.10 it suffices to show that for suitable G-realiza-
tions of Sn over Q the embedding problems into SQn˙ Z2 Z2m possess geometric
solutions, which are then automatically proper. For this let n  5 be an odd natural
number and m 2 IN fixed, where without loss we may take m  2 since SQn˙ are
epimorphic images of SQn˙ Z2 Z4 .
6 The Quadratic Trace Form 337

By Theorem I.5.1 the cyclic group Z2m has a G-realization over Q, say

Gal.M=Q.v// Š Z2m :

Let the quadratic intermediate field herein be generated by the square rootpof
x 2 Q.v/. Let w be transcendental over Q.v/ and set K WD Q.v; w/. Then x
also generates the quadratic intermediate field of the Galois extension KM=K with
Gal.KM=K/ Š Z2m . The zeroes t1 ; t2 of the polynomial
p p
g2 .X / WD .X  w  x/.X  w C x/ D X 2  2wX C w 2  x 2 KŒX

are algebraically independent over Q since 2w and w 2 x are algebraically indepen-

dent. Completing these to a system of independent transcendental elements t1 ; : : : ; tn
over Q, we see that
Y
n
g.X / WD g2 .X / .X  ti / D X n  s1 X n1 C : : : C .1/n sn (6.26)
i D3

is a polynomial with coefficients algebraically independent over Q. From these

according to Lemma 5.10 we form the polynomial
f .t; X / WD g.X /  th.X / 2 K.t; t/ŒX with t D .t3 ; : : : ; tn /; (6.27)
which by Proposition 5.11(a) has Galois group Sn since D.g/ D D.g2 / D2 x. Since
Q is Hilbertian, t may be specialized to a D .a3 ; : : : ; an / 2 Qn2 with pairwise dis-
tinct ai such that the residue class polynomial fa .t; X / also has group
p Sn over K.t/.
Now the splitting field Na of fa over K.t/ contains the field K.t; x/, so we get as
Galois group of the composite MNa over K.t/

Gal.MNa =K.t// Š Sn Z2 Z2m : (6.28)

Let La be a root field of fa over K.t/. For the determination of the quadratic
trace form qLa =K.t / .X/ of La =K.t/ we may specialize
p the variable t to 0 by The-
orem 6.6, since the inertia groups of Na =K.t; x/ all have order 3 by the proof
of Proposition
p 5.11. Thus we obtain with (6.27) and (6.26) from trace.12 / D 2 and
trace. x / D 2x
2

qLa =K.t / 2X12 C 2xX22 C X32 C : : : C Xn2

and hence the Hasse-Witt-invariant

w2 .qLa =K.t / / D .2; 2x/ D .2; x/: (6.29)

With Serre’s formulae (Proposition 6.7(b)) and using dLa =K.t / D2 x the embedding
obstructions thus equal

'  .QsnC / D w2 .qLa =K.t / / C .2; dLa =K.t / / D .2; x/ C .2; x/ D 0;

'  .Qsn / D w2 .qLa =K.t / / C .2; dLa =K.t / / D .2; x/ C .2; x/ D .1; x/ D 0
338 IV Embedding Problems
p
by Proposition 6.3, since by construction K.t; x/=K.t/ can be embedded into
a Z4 -extension. Consequently for G D Sn both embedding problems E .';  ˙ / in
Theorem 6.8 possess a proper solution. Their solution fields NQ a˙ satisfy

Gal.NQ a =K.t// Š SQn˙ : (6.30)

p
Now NQ a˙ =K.t/ is not geometric with respect to K, since x is algebraic over K.
But as NQ a˙ =Q is regular, extension of constants with QN furnishes a Galois extension
N NQ a =QK.t/
Q ˙ N with isomorphic Galois group. If fQa .v; w; t; X / 2 K.t/ŒX is a gen-
˙

erating normal polynomial of this Galois extension, then by Theorem 1.3 the Hilbert
set HQK.t Q˙ 3
N / .fa .v; w; t; X // contains elements in Q . In particular w; t may be spe-
cialized to elements a1 ; a2 2 Q such that fQa˙ .v; x/ WD fQa˙ .v; a1 ; a2 ; X / generates a
geometric Galois extension NQ a˙ =Q.v/ with

Gal.NQ a˙ =Q.v// Š SQn˙ : (6.31)

p
Since its quadratic subextension is Q.v; x/, we further see that

GQ ˙ WD Gal.M NQ a˙ =Q.v// Š SQn˙ Z2 Z2m : (6.32)

It remains to show that M NQ a˙ =Q.v/ is geometric. This follows from the fact that
p
UQ WD Gal.M NQ a˙ =Q.v; x// Š AQn Z2 Z2m

is the only maximal normal subgroup of GQ ˙ . This completes the proof for odd n.
As in the proof of Theorem 5.12 of Mestre one notes that the root field La of fa
(obtained from fa by specializing w; t to a1 ; a2 as above) is a rational function field
over Q, say La D Q.u/. Thus NQ a˙ =Q.u/ is a geometric Galois extension with

Gal.NQ a˙ =Q.u// Š SQn1 Z2 Z2m :

This yields the desired result also in the case of even n. t

u
The above proof even shows the following about the splitting field Na =Q.v/ of
fa .v; t/:
Corollary 6.12. For n  4 the Galois extensions Na =Sn Q.v/ for n odd respectively
Na =Sn Q.u/ for n even are universally central embeddable.
Remark. By Theorems 5.12 and 6.11 the three groups AQ4 Š SL2 .3/, SQ4C Š GL2 .3/
and SQ4 possess G-realizations over Q. According to Dentzer (1995b) these are the
only non-semiabelian groups G of order jGj < 64.
Apart from the above examples the criterion of Serre was also applied to prove
the existence of G-realizations over Q for the groups 6  A6 , 6  A7 , SL2 .7/ and 2  M12
by Mestre (1994, 1998) and for 2  S6 .2/ by Häfner (1992). In particular the G-
realizations of the simple groups A6 ; A7 ; etc. are universally central embeddable.
Further applications are collected in Bayer-Fluckiger (1994), 5.
7 Brauer Embedding Problems 339

7 Brauer Embedding Problems

Brauer embedding problems are the natural generalization of the central embedding
problems with kernel Z2 considered in the previous paragraph. Here we first prove
that the solvability of Brauer embedding problems implies the existence of proper
regular solutions. Then we show that the ordinary solvability satisfies a local-global
principle in the horizontal case (localization with respect to prime divisors of the
field of constants) as well as in the vertical case (localization with respect to prime
divisors over the field of constants). The latter finally allows to obtain G-realizations
for a number of central extensions of simple groups over Qab .

7.1 Regular Solutions of Brauer Embedding Problems

A Brauer embedding problem over K is an embedding problem E .'; / with finite

abelian kernel, whose kernel H is isomorphic to a subgroup of the group of units
UN of N D KN ker.'/ as G-operator group (with G D im.'/). Thus H is generated by
a root of unity n of order n lying in N , and the embedding obstruction h.'; / for
h 2 H 2 .G; H / lies in H 2 . K ; h n i/. The name is explained by the following result:

Proposition 7.1. Let K be a field of characteristic prime to n. Then we have

H 2. K ; h n i/ Š Brn .K/; (7.1)

where Brn .K/ denotes the n-torsion of the Brauer group Br.K/ of K.

Proof. From the short exact sequence

n
1 ! h n i ! KN  !KN  ! 1

follows, using Hilbert’s Theorem 90, the exactness of

n
0 D H 1. K;K
N  / ! H 2 . K ; h n i/ ! H 2 . N  /!H
K;K
2
. N /
K;K

and hence the assertion due to

Br.K/ Š H 2 . K;K
N / (7.2)

(see for example Serre (1979), Ch. XIV, 2). t

u

By Proposition 7.1 the solvability of Brauer embedding problems may be decided

inside Brn .K/ and hence inside the Brauer group Br.K/. From one solution of such
a Brauer embedding problem, further ones can be obtained in an easy way, as is
shown in the next theorem:
340 IV Embedding Problems
p
Theorem 7.2. Let NQ WD N. n x/ be a solution field of the Brauer embedding prob-
lem E .'; / with kernel H Š Zn over K. Then all solution fields of E .'; / are of
the form p
NQ a WD N. n ax/ with a 2 K  : (7.3)
Proof. According to Theorem 6.1(b) there exists c 2 C 1 . K ; H / such thatp'Qc W
Q Q
K ! G D H  G in (6.4) is a solution of E .'; / with solution field N WD N. x/.
n


By assumption H is isomorphic as G-module to h n i N . For a fixed zero y of
X n  x 2 N ŒX we get apart from c a further homomorphism

dW N ! h n i;  7! y  1 ;

with the same kernel c D Gal.K= N NQ /. Since c and d differ only by an automor-
phism of h n i and as n 2 N , the cochain c defining NQ can be chosen so that we
even have d. / D c and hence c D y  1 for all  2 N .
If now N.y/ Q with yQ n D xQ 2 N is any solution field of E .'; /, a suitable defining
cochain cQ satisfies correspondingly cQ D yQ  1 and ı.cc Q 1 / D 1. In particular we
have ccQ 1
2 Z . K ; h n i/ Z . K ; K /, which since H 1 . K ; KN  / D 1 implies
1 1 N 

Q
cc 1
2 B . K ; KN  /. Hence there exists an aQ 2 KN  with cQ c1 D aQ  1 for all  2
1

K . Since .aQ n / 1 D 1 this shows that aQ n DW a 2 K. As

yQ  1 D cQ D aQ  1 c D .ay/
Q  1 ;
p
Q D N.ay/
it follows that N.y/ Q pD N. n ax/.
Conversely, any field N. n ax/ with a 2 K clearly gives a solution field for
p  1
E .'; /, with the cochain . n a c / 2 C 1 . K ; h n i/, for example. u
t
The preceding theorem has the following easy consequence:
Corollary 7.3. If a Brauer embedding problem over a field K is solvable, then it
also possesses a proper regular solution.
p
Proof. Let NQ D N. n x/ be a solution field of the given Brauerp embedding problem
E .'; / over K with N D KN ker.'/ . Then NQ .t/ D N.t; n x/ is a solution field of
the Brauer embedding problem E .'  ;   / lifted to the rational function
p field K.t/
(compare Section 1.2). By Theorem 7.2 the field NQ  WD N.t; n tx/ also gives a
solution field of E .'  ;   /. As N is algebraically closed in NQ  and moreover ŒNQ  W
N.t/ D n, this belongs to a proper regular solution of E .'; /. t
u
It thus remains to study the question of solvability of Brauer embedding prob-
lems. Here local-global principles prove to be effective.

7.2 The Horizontal Local-Global Principle

The basis for the subsequent proofs is given by the following structure theorem for
the Brauer group of rational function fields, which goes back in characteristic zero to
Faddeev (1951), and for non-zero characteristic to Auslander and Brumer (1968):
7 Brauer Embedding Problems 341

Theorem 7.4. Let K D k.t/ be a rational function field in one variable, and p a
prime different from the characteristic of k. Then the sequence of p-primary com-
ponents of Brauer groups
M
0 ! Brp1 .k/ ! Brp1 .K/ ! Brp1 .KP /= Brp1 .KP/ ! 0 (7.4)
P2IP0 .K=k/

is exact and splits, where IP0 .K=k/ is the set of primes of K=k different from P1 ,
KP denotes the completion and KP the residue class field of K at P.
A simple proof of this result can be found in Scharlau (1969), Kor. 2.2, for
example. With the following exact sequence going back to Witt (see Serre (1979),
Ch. XII, Thm. 2)

0 ! Br.KP/ ! Br.KP / ! KP ! 0; (7.5)

where KP denotes the group of continuous homomorphisms from the absolute
Galois group KP to Q=ZZ, we obtain from (7.4) the Theorem of Auslander, Brumer
and Faddeev in the following form:
Corollary 7.5. Under the assumptions of Theorem 7.4 we have
M

Brp1 .K/ Š Brp1 .k/ ˚ . KP /p 1 : (7.6)
P2IP0 .K=k/

By construction, this result is compatible with extension of constants, i.e., for a

field extension l=k linearly disjoint from K=k and L WD l.t/ the following diagram
becomes commutative
L 
Brp1? .K/ ! Brp .k/ ˚ P2IP ? 0 .K=k/ . KP /p
1 1

? ?
y y (7.7)
L 
Brp1 .L/ ! Brp1 .l/ ˚ Q2IP0 .L= l/ . LQ /p1 ;

where the horizontal arrows denote the isomorphisms in Corollary 7.5 and the verti-

cal arrows are the restriction maps, such that on the right hand side KP is mapped

component-wise onto ˚QjP LQ .
With this statement the following result can be derived without difficulty:
Proposition 7.6. Let K D k.t/ be a rational function field in one variable over a
global field and p a prime different from the characteristic of k. Then the product
of the restriction maps
Y
 W Brp1 .K/ ! Brp1 .Kp / (7.8)
p2IP.k/

is injective, where p runs over all prime divisors (places) of k including the infinite
ones, and Kp WD kp .t/ denotes the rational function field over the completion of k
with respect to p.
342 IV Embedding Problems

Proof. For the proof we employ (7.7) with l D kp and L D Kp . By Corollary 7.5 it
thus remains to show that the map
M

 W Brp1 .k/˚ . KP /p1 !
P2IP0 .K=k/
Y  M

Brp1 .kp / ˚ . Kp Q /p 1 (7.9)
p2IP.k/ Q2IP0 .Kp =kp /

is injective. The injectivity on Brp1 .k/ is assured by a classical theorem of Brauer,

Hasse and Noether (see for example Weil (1974), Ch. XI, Thm. 2). It thus suffices
to show the injectivity of
Y M

P W . KP /p 1 ! . Kp Q /p1 :
p2IP.k/ QjP

Therefore let f 2 ker.P / and N=KP the cyclic extension defined by the fixed
field of ker.f /. Then we have NKp Q D Kp Q for all p 2 IP.k/ and Q 2 IP.Kp =kp /
with QjP, i.e., every prime divisor of KP splits completely in N=KP. This implies
N D KP and hence f D 0, as desired. t
u
Remark. The assertion of Proposition 7.6 continues to hold for rational function
fields in several variables over global fields (see Sonn (1990), Thm. 2).
Now let k be a global field and E .'; / a finite embedding problem over K WD
k.t/ with ' W K ! G D Gal.N=K/ and  W GQ ! G. Then to each p 2 IP.k/ is
associated in a natural way a restriction map

'p W Kp ! Gp D Gal.NKp =Kp /; (7.10)

and after identification of Gp with the decomposition group of some pQ =p in G also

an epimorphism
p W GQ p WD  1 .Gp / ! Gp : (7.11)
These two epimorphisms define an embedding problem E .'p ; p / over Kp with the
same kernel as E .'; /. In the case of Brauer embedding problems Proposition 7.6
furnishes the following local-global-principle.
Theorem 7.7 (Sonn (1990)). Let K D k.t/ be a rational function field over a global
field and E .'; / a Brauer embedding problem over K, whose kernel has order
prime to the characteristic of k. Then E .'; / is solvable over K if and only if
for all p 2 IP.k/ the local Brauer embedding problems E .'p ; p / are solvable over
Kp D kp .t/.
Proof. By Theorem 1.6 we may assume that the order n of the kernel is a prime
power. Let h.'; / 2 H 2 . K ; h n i/ Š Brn .K/ be the cohomological obstruction to
the embedding problem E .'; /. Then the images h.'; /p 2 Brn .Kp / of h.'; /
under the map  in (7.8) constitute the cohomological obstructions of the local
Brauer embedding problems E .'p ; p /. If all these are solvable, then h.'; /p D 0
7 Brauer Embedding Problems 343

for all p 2 IP.k/ by Theorem 6.1. The solvability of E .'; / then follows from the
injectivity of . t
u
With the proof of Theorem 7.7 we immediately obtain the following variant of
the Theorem of Brauer, Hasse and Noether used in the proof:
Corollary 7.8. Over global fields the local-global principle for Brauer embedding
problems is valid.
With Theorem 7.7 Brauer embedding problems over rational function fields with
global field of constants can be reduced to Brauer embedding problems over rational
function fields with complete field of constants. Such field extensions will be the
subject of the following chapter.

7.3 The Vertical Local-Global Principle

The basis of the vertical local-global principle for Brauer embedding problems is
given by the following corollary to Theorem 7.4:
Proposition 7.9. Let K D k.t/ be a rational function field in one variable and p a
prime different from the characteristic of k. Then the product of the restriction maps
Y
 W Brp1 .K/ ! Brp1 .KP / (7.12)
P2IP0 .K=k/

is injective, where P runs over all prime divisors of K=k different from P1 and
KP denotes the completion of K with respect to P.
Remark. Proposition 7.9 also holds in the case that p equals the characteristic of K
(see for example Kucera (1994), Lemma 8).
As in the previous section this implies a local-global-principle for Brauer embed-
ding problems:
Theorem 7.10 (Sonn (1994a)). Let K D k.t/ be a rational function field in one
variable and E .'; / a Brauer embedding problem over K, whose kernel has order
prime to the characteristic of k. Then E .'; / is solvable over K if and only if for
all P 2 IP0 .K=k/ the local Brauer embedding problems E .'P ; P / are solvable
over KP .
The vertical local-global-principle becomes particularly easy to use if the field
of constants has a projective absolute Galois group, since then the local embedding
problems have to be solved only for the finitely many ramified P.
Theorem 7.11 (Sonn (1994b)). Let K D k.t/ be a rational function field in one
variable over a field k with projective Galois group and N=K a finite Galois exten-
sion ramified at the prime divisors P1 ; : : : ; Pr of K=k and possibly at the infinite
344 IV Embedding Problems

prime P1 . Also let E .'; / with ' W K ! Gal.N=K/ be a Brauer embedding

problem whose kernel H has order prime to the characteristic of k. Then E .'; / is
solvable if and only if the finitely many local embedding problems E .'Pi ; Pi / for
i D 1; : : : ; r are solvable.

Proof. Note that for prime divisors P=PQ not ramified in N=K, the local field
extensions NPQ =KP are extensions of constants of power series fields, since KP D
Q
KP..t// and NPQ D N P..t//. Thus solutions of E .'P ; P / can be obtained by lift-
ing solutions of a finite embedding problem with kernel H over the field KP with
projective absolute Galois group, which is always solvable by Corollary 1.12. t
u

From the above we may derive a particularly simple sufficient criterion for the
solvability of local embedding problems in the ramified case.

Corollary 7.12. Assume that K contains the exp.G/-th Q roots of unity, where GQ D
H  G. For the solvability of E .'; / it suffices that for each Pi 2 IP.K=k/ ramified
in N=K different from P1 the inertia group Ii D I.P Q i =Pi / satisfies either

gcd.jIi j; jH j/ D 1 or CG .Ii / D Ii : (7.13)

Proof. The Galois extension NPQ i =KPi is composed of the extension of constants
NP Q i ..t//=KPi ..t// followed by a cyclic extension N Q =N P Q i ..t// of degree ei D
Pi
Q i . Thus the first of these two extensions is Galois and the second
jIi j regular over N P
a Kummer extension with the group Ii Z .Gi / for Gi D Gal.NPQ i =KPi /.
In the case gcd.jIi j; jH j/ D 1 the preimage of Ii in GQ i D H  Gi splits into a
direct product H  IQi with IQi Š Ii and Pi .IQi / D Ii . By the assumption on the
projectivity of K the embedding problem belonging to the group extension

1 ! H ! GQ i =IQi ! Gi =Ii ! 1

is solvable (by extension of constants), with solution 'Qi W K ! GQ i =IQi , say. Accord-
ing to the universal mapping property of the fiber product there also exists a homo-
morphism
'QPi W K ! GQ i D Gi Gi =Ii GQ i =IQi
with Pi ı 'Q Pi D 'Pi and thus a solution to E .'Pi ; Pi /.
In the second case Ii Š CG .Ii / Š NG .Ii / we have Ii D Gi since Gi E NG .Ii /.
Hence NPQ i =KPi is a Kummer extension generated by the ei -th root of some u 2
KP
. Denoting by eQi the exponent of GQ i D H  Gi , the field KPi .v/ with v eQi D u
yields a solution field and thus a (not necessarily proper) solution of E .'Pi ; Pi /.
t
u

Remark. Using the reduction theorem of Kochendörffer in Section 8.1 the condi-
tions for the solvability of E .'Pi ; Pi / on the inertia groups Ii can be weakened
to
gcd..Gi W Ii /; .H W Ci // D 1; (7.14)
7 Brauer Embedding Problems 345

where Gi D CG .Ii / and Ci is a complement of the Frattini extension inside the

group extension H  Ii D  1 .Ii / (as described in Theorem 5.1).

In the next section we collect some examples for the application of the vertical
local-global-principle.

7.4 Covering Groups of Simple Groups over Qab .t /

With the help of the vertical local-global-principle a number of covering groups of

classical groups which do not belong to the matrix groups investigated in Chapter II
can be realized as geometric Galois groups over Qab .t/.

Theorem 7.13 (Sonn (1994b), Malle and Sonn (1996)). The universal central exten-
sions of the following almost simple groups have G-realizations over Qab .t/:
(a) PGLn .q/ for n  3 or q  4,
(b) PGUn .q/ for n  3, .n; q/ ¤ .3; 2/,
(c) PCSp2n .q/ for n  2,
(d) SO2nC1 .q/ for n  3, q odd,
(e) E7 .q/ and E7 .q/ad for q D p  with p  3.

Proof. First note that if G is such that S G Aut.S / for a non-abelian sim-
ple group S , then the Schur multiplier of G is a subgroup of that of S (see Sonn
(1994b), Lemma 6). In case (d), the Schur multiplier of O2nC1 .q/ is cyclic of order 2
except for O7 .3/, where it has order 6. In the general case, the criterion in Corol-
lary 7.12 applies to the Galois realizations of SO2nC1 .q/ in Theorem II.3.7 with the
class vector containing one class of involutions, a class of elements of odd order, and
a class of self-centralizing semisimple elements, since the orders of the elements in
the first two classes are coprime. This also gives the result for the exceptional case
O7 .3/.
Similarly in cases (b), (c) we can start with the G-realization of GUn .q/ respec-
tively CSp2n .q/ in Theorems II.3.2 and II.3.4, since again these have the properties
required in Corollary 7.12. Some care has to be taken due to several exceptional
multipliers, see Sonn (1994b) and Malle and Sonn (1996).
The Schur multiplier of E7 .q/, q odd, has order 2. So for (e), the assertion may be
deduced from Theorem II.5.17, where we obtained a Galois realization of E7 .q/ad
with respect to a class vector containing a class Cp of elements of order p and a
class CTı containing elements generating their proper centralizer.
The proof for case (a) is similar, starting from the G-realizations of the groups
PGLn .q/ given by Walter (1984) (see also the Remark after Theorem II.3.2; the
G-realizations of Belyi presented in Section II.1.3 are not suited for an application
of Corollary 7.12.) See Sonn (1994b) for details. t
u

For the sporadic groups, we have an almost complete result (see Malle and Sonn
(1996)):
346 IV Embedding Problems

Theorem 7.14. The full covering groups of all groups G with S G Aut.S /,
where S is sporadic simple with the possible exception of M22 , possess
G-realizations over Qab .t/.

Proof. By the Remark after Theorem 5.8, the threefold covering groups of G as in
the Theorem, for S ¤ J3 , are already realized as Galois groups over Q.t/. We treat
the remaining groups S with non-trivial Schur multiplier.
In Propositions. II.9.1–II.9.7 we proved G-realizations for the automorphism
groups of sporadic groups S with class vector C listed in Table 7.1. In all cases,
elements from the first two classes have coprime orders, while those in the third
class generate their full centralizer. Choosing one ramification point at infinity, it
follows that Corollary 7.12 applies regardless of the order of the p-primary part of
the Schur multiplier we are looking at.

Table 7.1 Class vectors for some sporadic groups

Aut.S / C
M12 W 2 .2C; 3A; 12A/
J2 W 2 .3A; 8C; 14A/
HSW 2 .2C; 5C; 30A/
Co1 .3A; 5C; 13A/
Fi22 W 2 .2D; 5A; 42A/
B .2C; 3A; 55A/
Ru .2A; 5A; 13A/

This only leaves the groups Suz and J3 to consider. Hunt (1986) gave a G-
realization for Aut.Suz/ with the rationally rigid class vector .2C; 3B; 28A/. The
Schur multiplier here has order 6, and elements from class 28A are self-centralizing.
For J3 it is easy to verify that .2B; 3A; 34A/ provides a semirational rigid class
vector of Aut.J3 /. As J3 has Schur multiplier of order 3 we can apply the criterion.
t
u

Further applications of Brauer embedding problems in particular for groups of

small order are collected in the monograph of Ledet (2005).
8 Concordant Embedding Problems 347

8 Concordant Embedding Problems

Every embedding problem is accompanied by embedding problems belonging to

sections of the group extension. At least a necessary condition for the solvability of
an embedding problem is then the solvability of all accompanying Brauer embed-
ding problems. This condition was introduced by Delone and Faddeev (1944) and
is nowadays called concordance condition. In order to avoid special cases in posi-
tive characteristic in the characterization, we first solve embedding problems whose
kernel has order a power of the characteristic of the ground field. For this we employ
among others the reduction theorem of Kochendörffer, which will be proved more
generally for regular solutions. After that we show that the standard reduction theo-
rems including the theorem of Kochendörffer continue to hold for the concordance
condition. Finally as the main result of this paragraph we show, using a duality the-
orem of Tate, that abelian embedding problems over global fields are concordant if
and only if all induced local embedding problems are solvable.

8.1 The Reduction Theorem of Kochendörffer

As a preparation for the proof of the theorem of Kochendörffer for regular embed-
ding problems we need to introduce a generalization of the wreath product construc-
tion. Let G be group with subgroup U and let

1 ! H ! UQ ! U ! 1

be an exact sequence of groups. We choose and fix a system of right coset represen-
tatives of U in G and denote by N the coset representative of  2 G. Let H oU G be
the set of pairs

H oU G WD f.; / j  2 G; W G=U ! UQ such that . .// D  1  N 1 g:

We define a multiplication on H oU G by
2
.1 ; 1 /.2 ; 2/ WD .1 2 ; 1 2/

where  ./ WD . 1 / for all  2 G. This multiplication endows H oU G with the
structure of a group, an extension of H .GWU / by G (see Ishkhanov, Lure and Faddeev
(1997), Ch. 3, 7, for example). This extension is called here the wreath extension
(transference in the Russian literature) of H with G over the group extension UQ D
H  U of H with the subgroup U of G. It is the natural generalization of the twisted
wreath product H oU G (defined for split extensions H U in Suzuki (1982), Ch. 2,
10) to the non-split case.
We can now state an embedding theorem generalizing Theorem 2.2.
348 IV Embedding Problems

Theorem 8.1. Let K be a field and N =G K a finite Galois extension. Further let
U G be a subgroup with fixed field L WD N U , UQ WD H  U a finite group extension
with kernel H and H oU G the wreath extension over UQ . Assume that the embedding
problem E . ; / with the restriction W L ! U and the canonical epimorphism
W UQ ! U has a (proper) regular solution. Then the embedding problem E .'; /
defined by the restriction ' W K ! G and the canonical epimorphism  W H oU G !
G possesses a (proper) regular solution.
Proof. Let NQ 1 denote the solution field of a (proper) regular solution of E . ; /,
where without loss we may assume that this solution is 1-regular and the Galois
group
UQ 1 WD Gal.NQ 1 =L.t1 // Š UQ 1 UQ D H  U
is isomorphic to a subgroup UQ 1 D H1  U of UQ . Then in particular NQ 1 =N is regular
with
H1 WD Gal.NQ 1 =N.t1 // Š H1 H
(where H1 D H in the case of a proper solution). Further let x be a primitive element
of L=K and xi WD x i 2 N with a system fi j i D 1; : : : ; rg of coset representatives
of U in G (and 1 D 1). Let y be a primitive element of NQ 1 =L.t1 / with minimal
polynomial f .Y; t1 / 2 L.t1 /ŒY . For a given system t D .t1 ; : : : ; tr / of independent
variables over K denote by NQ i the splitting field of f i .Y; ti / 2 K.xi ; ti /ŒY inside
an algebraic closure of K.t/. Then the composite NQ  WD NQ 1    NQ r is regular over N .

Via its action on the cosets of U , the group G also acts on ft1 ; : : : ; tr g via ti D t1 i
and in this way extends to a group G  Š G of automorphisms of N.t/=K.tG /. By

Speiser’s Lemma (Proposition III.3.10) the fixed field of this action K  WD N.t/G
is rational over K and can be generated explicitly in the following form:

X
r
K  D K.v1 ; : : : ; vr / with vj WD xij 1 ti : (8.1)
i D1

By Cramer’s rule we have the inverse formula

Vi .x; v/
ti D ; (8.2)
V .x/

where V .x/ denotes the Vandermonde determinant and Vi .x; v/ the determinant
obtained from V .x/ by replacing the i -th row by .v1 ; : : : ; vr /.
Equation (8.2) shows that ti 2 K  .xi /. Consequently the zeroes of the polynomi-
als f i .Y; ti / also generate the field extensions NQ i .v/=K .xi /. Since G  permutes
these polynomials, the composite NQ  of the NQ i .v/ D NQ i .t/ for i D 1; : : : ; r is Galois
over the fixed field K  of G  , and by construction the Galois group equals

Gal.NQ  =K  / Š H1 oU G H oU G; (8.3)

with H1 D H in the proper case. In particular the restriction map 'Q  W K !

Gal.NQ  =K  / gives a (proper) regular solution of E .'; / in r variables. t
u
8 Concordant Embedding Problems 349

Remark. The assertion ti 2 K  .xi / is easier to verify if instead of vj we use the

generators vj WD Vj .x; t/=V .x/ from Matzat (1987) where Vj .x; t/ denotes the
determinant obtained from V .x/ by replacing the j -th column by the transposed of
P j 1
.t1 ; : : : ; tr /. Then the inverse formula takes the form ti D rj D1 xi vj .

With the help of this theorem we arrive at the following generalization of the
classical reduction theorem of Kochendörffer.

Theorem 8.2 (Kochendörffer (1953)). Let E .'; / be a finite embedding problem

with abelian kernel H over K and U a subgroup of G D '. K / of index .G W U /
prime to jH j. Further let L denote the fixed field of U , the restriction of ' to
L K and the restriction of  to UQ D H  U H  G D G. Q Then E .'; /
possesses a (regular) solution if and only if E . ; / possesses a (regular) solution.

Proof. Evidently every solution field NQ of the embedding problem E .'; / is also
a solution field of E . ; /, so that the (regular) solvability of E .'; / implies the
(regular) solvability of E . ; /. It thus remains to prove the reverse assertion.
Let h 2 H 2 .G; H / be the cohomology class describing the group extension GQ D
H  G. Then UQ D H  U is a group extension belonging to the restriction U G
.h/ 2
2
H .U; H /. We consider the following commutative diagram
'
H 2?.G; H / ! H 2 .? K ; H /
? G ? K
yU yL

H 2 .U; H / ! H 2 . L; H /

with inflations '  and  . As jH j and .G W U / are coprime, the multiplication by

.G W U / on H 2 . K ; H / is injective. Since the composition of restriction and core-
striction acts as multiplication by .G W U / on H 2 . K ; H /, the restriction L
K
is injec-
tive (see for example Serre (1964), Ch. I, Cor. to Prop. 9, or Shatz (1972), Ch. II,
Prop. 10). In particular from the solvability of E . ; /, i.e., from  .U G
.h// D 0,

we immediately obtain ' .h/ D 0 and thus by Theorem 6.1(a) the solvability of
E .'; /. According to the Remark following Theorem 6.1 a solution field can
already be found inside the Galois closure of MQ =K, where MQ denotes the solu-
tion field of E . ; /.
In the case of regular solvability an additional argument is necessary. By the
proof of Theorem 8.1 the field NQ 1 .v/ is a solution of the embedding problem
E .  ;  / lifted from E . ; / to L.v/. By the first part of the proof the correspond-
ing embedding problem E .'  ;   / then also possesses a solution 'Q  with solution
field contained in the Galois closure of NQ 1 .v/=K  , hence inside NQ  . Since by con-
struction N is algebraically closed in NQ  , any such solution is automatically a reg-
ular solution of E .'; /. t
u

Remark. By Corollary 2.5 the above theorem of Kochendörffer also holds for proper
(regular) solutions of embedding problems, but in general not for geometric solu-
tions (proper or not).
350 IV Embedding Problems

As a simple application of the reduction theorem of Kochendörffer we obtain the

very useful

Theorem 8.3. Let K be a field of characteristic p and H a finite p-group. Then we

have:
(a) Every finite embedding problem over K with kernel H is solvable.
(b) If K is a Hilbertian field then every finite embedding problem over K with
kernel H has a proper solution.

Proof. Let first H be abelian. By the reduction theorem of Kochendörffer we may

assume by passage to a Sylow p-subgroup that G D Gal.N=K/ is a finite p-group.
Then part (a) of the theorem follows from the fact that the maximal pro-p factor
group of K is a free pro-p group (see Shatz (1972), Ch. III, Prop. 30, or also Serre
(1964), Ch. I, Prop. 16(iii) together with loc. cit., Ch. II, Prop. 3). In the case of a
Hilbertian field K the maximal pro-p factor group of K is not finitely generated.
Consequently every finite embedding problem over K with abelian p-kernel even
possesses a proper solution, which proves (b).
The assertion for a general p-group H now follows by induction along a GQ D
H  G-normal series of H with abelian factors. t
u

Remark. In the case of embedding problems with abelian kernel H Theorem 8.3
will allow to assume, using Theorem 1.6, that the characteristic of K does not divide
jH j.

This observation will prove to be very useful in the study of the concordance
condition.

8.2 The Concordance Condition

First we give a precise definition of the notion of an accompanying embedding prob-

lem. For this let E .'; / be a finite embedding problem over a field K with epimor-
phisms ' W K ! G D Gal.N=K/ and  W GQ D H  G ! G. For a subgroup B G
let L WD N B be the fixed field and U WD  1 .B/. Further let V be a normal subgroup
of U contained in H , A WD H=V and BQ the group extension A  B D U=V . Then the
embedding problem E . ; / defined by the canonical epimorphisms W L ! B,
W BQ ! B is called an accompanying embedding problem of E .'; /. If more-
over E . ; / is a Brauer embedding problem, it is called an accompanying Brauer
embedding problem. Clearly, if E .'; / is solvable, then all accompanying embed-
ding problems are solvable, see Fig. 8.1.
For an embedding problem E .'; / to possess sufficiently many accompanying
Brauer embedding problems, the fixed field N of ker.'/ has to contain enough roots
of unity. So let n be the part of exp.H / prime to the characteristic of K, Nn WD N. n /
with a primitive n-th root of unity n , Gn WD Gal.Nn =K/ with restriction 'n W K !
Gn , and GQ n WD GQ G Gn the subdirect product of GQ with Gn over G with projection
8 Concordant Embedding Problems 351

V
G̃
H
B̃ A
B
K L N
G
Fig. 8.1 Accompanying Brauer embedding problem

n onto Gn . Then Theorem 1.6(a) implies that the embedding problem E .'n ; n / is
solvable if and only if E .'; / is solvable, and moreover that E .'n ; n / is properly
solvable if and only if E .'; / is properly solvable with solution field NQ linearly
disjoint from Nn over N .
In the following E .'; / is called a concordant embedding problem, if all accom-
panying Brauer embedding problems of E .'n ; n / are solvable. The concordance
condition is hence a necessary prerequisite for the solvability of E .'; /. In the
investigation of concordant embedding problems by definition we may assume with-
out loss of generality that N contains a primitive n-th root of unity.
Now let H be an abelian group and HO .p/ WD Hom.H; KN  / the part of the charac-
ter group HO of H prime to the characteristic p of K, with exp.HO .p/ / D n. By taking
the action of K on H given by the restriction map ' W K ! G and the action of
G on H given by the group extension G, Q HO .p/ can be made into a K -module via

1
 ./ WD ..  // for  2 HO .p/ ;  2 K;  2 H: (8.4)

Let be the stabilizer of  2 HO .p/ in K and K its fixed field, so that  becomes
a -homomorphism. Obviously K is contained in N . In addition to the inflation
'  W H 2 .G; H / ! H 2 . K ; H / from (6.1) we obtain a homomorphism
 
 W H 2 . K;H / ! H 2 . ; H / ! Br.K / (8.5)

by composition of the restriction  with the homomorphism  W H 2 . ; H /!

H 2 . ; KN  / induced by .

Theorem 8.4. Let E .'; / be a finite embedding problem over K with abelian ker-
nel H and with cohomological obstruction h.'; / 2 H 2 . K ; H /. Then we have:
(a) E .'; / is concordant if and only if

 .h.'; // D 0 for all  2 HO .p/ : (8.6)

352 IV Embedding Problems

(b) E .'; / is solvable if in addition the following map is injective:

Y Y

H WD  W H 2 . K ; H / ! Br.K /: (8.7)
2HO .p/ 2HO .p/

Proof. By the above we may assume that the fixed field N of ker.'/ contains a
primitive n-th root of unity for n D exp.HO .p/ /. Via restriction every  2 HO .p/ defines
epimorphisms

' W ! G WD Gal.N=K /;  W GQ WD H  G ! G ; (8.8)

and thus an embedding problem E .' ;  / accompanying E .'; /. This is even a
Brauer embedding problem, since  is a -homomorphism and moreover .H /
N  . From the concordance condition follows the solvability of E .' ;  / and there-
fore the vanishing of the embedding obstruction which by (8.5) equals  .h.'; // 2
Br.K /.
Now assume (8.6) and let E . ; / be an accompanying Brauer embedding prob-
lem of E .'; /. By definition, if we denote by L the fixed field of ker. /, there
exists a L -homomorphism  W H ! N  with .H / D ker. / by identifying the
kernel of E . ; / with a group of roots of unity in N . Thus L contains the field K
and E . ; / is an accompanying embedding problem of E .' ;  /, which is hence
solvable. This proves (a).
Part (b) of the theorem follows immediately from (a) together with Theo-
rem 6.1(a). t
u

Remark. From Theorem 8.4(a) it follows that the concordance condition does not
change if we adjoin further roots of unity to N .

Fortunately the basic reduction theorems for embedding problems continue to

hold for concordant embedding problems. So in addition to Theorem 1.6 we obtain:

Corollary 8.5. Let E .'; / be

Qa finite embedding problem with abelian kernel H .
Assume that moreover H D riD1 Hi decomposes into a direct product of normal
subgroups Hi of GQ D H  G. Then E .'; / is concordant if and only if the accompa-
nying embedding problems E .'; i / in Theorem 1.6 for i D 1; : : : ; r are concordant.

Proof. The assertion follows

Q immediately from Theorem 8.4(a) with Remark due
to the equality HO .p/ D riD1 HO i;.p/ . t
u

Further we obtain the following supplement to the Theorem 8.2 of Kochen-

dörffer.

Corollary 8.6. Assume the hypotheses of Theorem 8.2. Then the embedding problem
E .'; / is concordant if and only if E . ; / is concordant.

Proof. With the notations introduced in the proof of Theorem 8.2,  from (8.5)
and the analogously defined map   W H 2 . L ; H / ! Br.L / in the Brauer group of
8 Concordant Embedding Problems 353

L D LK we obtain the following commutative diagram with vertical restriction

maps:

' 
H 2?.G; H / ! H 2 .? K ; H / ! Br.K
? /
? G ? K ? K
yU yL yˇL
 

H 2 .U; H / ! H 2 . L; H / ! Br.L /
By Theorem 8.4(a) the embedding problem E .'; / is concordant if and only if the
cohomology class h 2 H 2 .G; H / defining GQ D H  G satisfies

. ı '  /.h/ D  .h.'; // D 0 for all  2 HO .p/ :

Since the cohomology class UG

.h/ defines UQ D H  U , this immediately implies the
concordance condition for E . ; /:

.  ı  G
/.U .h// D 0 for all  2 HO .p/ : (8.9)

Conversely the concordance condition (8.9) for E . ; / first yields

K
.ˇL ı  ı '  /.h/ D 0 for all  2 HO .p/ :

But since gcd.ŒL W K ; jH j/ D 1, the restriction ˇLK

is injective on the exp.HO .p/ /-
torsion of the Brauer group Br.K / Š H . ; KN / (see for example Serre (1964),
2 

Cor. to Prop. 9, or Shatz (1972), Ch.II, Prop. 10). So the previous equation allows
to recover the concordance condition for E .'; /. t
u

With the first of these two reduction theorems for the concordance condition we
get the following easy solvability criterion:

Theorem 8.7. Let E .'; / be a finite embedding problem over K with abelian ker-
nel H . Assume that K acts trivially on HO .p/ . Then the concordance condition
implies the solvability of E .'; /.

Proof. Let p be the characteristic of K. Then the concordance condition is vacu-

ously satisfied for the Sylow p-subgroup Hp of H , and the corresponding embed-
ding problem is solvable by Theorem 8.3(a). Thus by Theorem 1.6(a) we may
assume that p does not divide jH j and hence we have HO .p/ D HO . By the structure
Q
theorem for finite abelian groups, HO is a direct product HO D riD1 hi i of cyclic
groups. Since K is supposed to act trivially on hi i, we have i D K and hence
' i D ' for i D 1; : : : ; r. Let

\
r
Hi WD ker.j / Š hi i
i ¤j D1

for the kernels of the accompanying Brauer embedding problems E .';  i / in the
proof of Theorem 8.4. Since H Š HO these lead to a direct decomposition of the
354 IV Embedding Problems

kernel H of E .'; /:

Y
r
HD Hi with H1    Hi 1 Hi C1    Hr D ker.i /:
i D1

Thus the embedding problems E .';  i / coincide with the accompanying embed-
ding problems E .'; i / in Corollary 8.5. The concordance condition for E .'; /
then implies the solvability of E .'; i / for i D 1; : : : ; r and with Corollary 8.5 also
the solvability of E .'; /. t
u

In the following two sections we study the concordance condition in the case of
local and global ground fields.

8.3 Concordance over Local Fields

The main tool for the investigation of concordant embedding problems over local
fields, i.e., over fields complete with respect to a discrete valuation and with finite
residue field, is a duality theorem of Tate, which we cite without proof (see also
Serre (1964), Ch. II, Thm. 2 with Rem. 2, or Ishkhanov, Lure and Faddeev (1997),
Thm. A.3.1).

Theorem 8.8 (Duality Theorem of Tate (1962)). Let K be a local field and H
a finite K -module of order prime to the characteristic of K. Then the groups
H i . K ; H / and H 2i . K ; HO / are in duality by virtue of the cup product

 W Hi. K;H /  H
2i
. O ! H 2.
K;H / K;K
N  / Š Br.K/ (8.10)

for i D 0; 1; 2.

This allows to give a very easy proof (found by Hoechsmann (1968)) of the
following important result:

Theorem 8.9 (Demuškin and Šafarevič (1959)). Concordant finite embedding prob-
lems with abelian kernel over a local field are always solvable.

Proof. Let E .'; / be a concordant embedding problem with kernel H over a local
field K. By Theorems 8.3 and 1.6 we may assume without loss of generality that
the characteristic of K is prime to jH j and thus we have HO .p/ D HO . Moreover,
the fixed field N of ker.'/ can be assumed to contain a primitive exp.H /-th root

of unity. By Theorem 8.4(b) we have to show the injectivity of the map H . Let
O
X WD H . K ; H / be the group of characters of H invariant under the K -action
0

defined in (8.4). Then for each g 2 H 2 . K ; H / and  2 X we have using the cup-
product
g   D  .g/ 2 H 2 . K ; KN  / Š Br.K/ (8.11)
8 Concordant Embedding Problems 355

(see for example Shatz (1972), Ch. II, Remark 1 following Thm. 9). By the Duality
Theorem of Tate for g ¤ 0 there exist  2 X with  .g/ ¤ 0. Thus
Y Y
 W H 2 . K;H / ! Br.K/ (8.12)
2X 2X


is injective, and a fortiori the same is true for H . t
u

Proposition 8.10. The conclusion of Theorem 8.9 also holds for the fields IR and C.

Proof. It suffices to show the injectivity of (8.12) for K D IR and K D C. For K D C

this is trivial due to H 2 . K ; H / D 0.
So let K D IR. Then the only non-trivial element of K is the involution  map-
ping every root of unity to its inverse. The  2 H 0 . K ; HO / are hence characterized
by .  / D ./1 for all  2 H . Now let

T WD f    j  2 H g H:

In the case T D 1 we always have   D  1 , so every  2 HO is K -invariant and

the assertion follows from Theorem 8.7. Otherwise, the factor group H1 WD H=T
has smaller order than H , and we can prove (8.12) by induction on jH j. In the long
cohomology sequence

!2
: : : ! H 2 . K;T / ! H 2 . K;H / ! H 2 . K ; H1 / ! : : :
!
induced by 1 ! T ! H ! H1 ! 1, with K Š Z2 , we have that H 2 . K ; T /
maps to 0 by the definition of T (see Huppert (1967), Kap. I, Satz 16.10) and hence
!2 is injective. Thus for g 2 H 2 . K ; H / with g ¤ 0 there exists by induction a
1 2 H 0 . K ; HO 1 / with 1 .!2 .g// ¤ 0. Denoting by  the character obtained by
inflation of 1 to H , we obtain  2 H 0 . K ; HO / with  .g/ D 1 .!2 .g// ¤ 0. u
t

Remark. If in Theorems 8.3 and 8.9, respectively in Proposition 8.10 the ground
field K is not Hilbertian, then the investigated solvable embedding problems will in
general not have a proper solution!

8.4 Concordance over Global Fields

With the help of Theorem 8.9 of Demuškin and Šafarevič one obtains a character-
ization of the concordance condition for embedding problems over global fields K
by local solvability, where global fields are the finite algebraic number fields over
Q, and algebraic function fields in one variable over finite fields. This relies on a
local-global principle for the concordance condition. In the sequel as in Section 7.2
we denote by IP.K/ the set of all prime divisors of K (including the infinite ones, if
they exist).
356 IV Embedding Problems

Proposition 8.11. A finite embedding problem E .'; / with abelian kernel over a
global field is concordant if and only if for all p 2 IP.K/ the local embedding prob-
lems E .'p ; p / are concordant.

Proof. The local embedding problems were defined in Section 7.2. For the proof as
above we may assume without loss of generality that jH j is prime to the character-
istic of K and that the fixed field N of ker.'/ contains the exp.H /-th roots of unity.
We consider the following commutative square

H Q
H 2 . ?K ; H / !Br.K / 2HO?
? ?
y˛ yˇ
Q H Q

Q
p2IP.K/ H . Kp ; H / !
2
p2IP.K/ 2HO Br.Kp; /

 
Q 
with H from (8.7), the product H D p2IP.K/ p;H of the analogously defined
local maps and the products of restriction maps
Y Y Y
˛D ˛p ; ˇ D ˇ ;p :
p2IP.K/ 2HO p2IP.K/

Let first E .'; / be a concordant embedding problem with embedding obstruction


h.'; / 2 H 2 . K ; H /. Then by Theorem 8.4(a) we have H .h.'; // D 0. This

implies H .h.'; /p / D 0 for p 2 IP.K/ and the embedding obstruction h.'; /p D
˛p .h.'; // of the local embedding problem E .'p ; p /. So E .'p ; p / is concordant.
Conversely, if all local embedding problems E .'p ; p / are concordant, the
embedding obstruction h.'; / of E .'; / satisfies
 
.ˇ ı H /.h.'; // D .H ı ˛/.h.'; // D 0:

For the concordance of E .'; / it remains to show

Q that ˇQ is injective on the image

of H . For this we compare ˇ with the map ˇQ D 2HO
ˇ , with components
Y Y
ˇQ D ˇQ p
;Q W Br.K / ! Br.K p/
;Q
Q 2IP.K /
p Q 2IP.K /
p

the injective maps obtained from the Theorem of Brauer, Hasse and Noether (see
for example Weil (1974), Ch. XI, Thm. 2). By construction for b D .b / 2HO 2
H
.H 2 . K ; H // and pQ jp the images ˇQ ;Qp .b / are embedding obstructions of Brauer
embedding problems conjugate over Kp to the localization E .' ;p ;  ;p / at p of
E .' ;  /. Thus in particular ˇQ ;Qp .b / vanishes for pQ jp if and only if ˇ ;p .b / D 0.

This proves injectivity of ˇ on the image of H and hence the assertion. t
u

Remark. Obviously Proposition 8.11 remains valid more generally if the Brauer
group satisfies a local-global principle, as for example in the case of rational func-
tion fields over global fields (compare Sections 7.2 and 7.3).
8 Concordant Embedding Problems 357

Using the results on local fields the previous proposition immediately gives:

Theorem 8.12. Let E .'; / be a finite embedding problem with abelian kernel H
over a global field K. Then we have:
(a) E .'; / is concordant if and only if for all p 2 IP.K/ the local embedding
problems E .'p ; p / are solvable.
(b) E .'; / is solvable, if in addition the following product of restriction maps is
injective: Y
˛ W H 2 . K ; H / ! H 2 . Kp ; H /: (8.13)
p2IP.K/

Proof. Part (a) follows with the Theorem 8.9 of Demuškin and Šafarevič and Propo-
sition 8.10 immediately from the local-global principle for the concordance condi-
tion in Proposition 8.11.

Under the assumption (a) in the proof of Proposition 8.11 we have H .h.'; // D
0 for the embedding obstruction h.'; / 2 H 2 . K ; H / of E .'; / and hence also
  
.H ı ˛/.h.'; // D .ˇ ı H /.h.'; // D 0. Since H is injective by Theorem 8.9
and Proposition 8.10 respectively, and ˛ is so by assumption, we conclude that
h.'; / D 0 and hence the solvability of E .'; /. This proves part (b). t
u
358 IV Embedding Problems

9 The Hasse Embedding Obstruction

In this paragraph we study the embedding obstruction of concordant embedding

problems. This will be called the Hasse embedding obstruction, since it was dis-
covered by Hasse (1948) and constitutes an obstruction to the validity of the local-
global-principle for embedding problems. We then present cohomological descrip-
tions of the Hasse obstruction for arbitrary and for global fields, whose coincidence
is proved using a comparison theorem of Tate for the cohomology of global fields.

9.1 Kummer Extensions

As in the previous paragraph let E .'; / be a finite embedding problem over a field
K with abelian kernel H . Moreover we will always assume that the characteristic
of K does not divide the order jH j and that the fixed field N of ker.'/ contains a
primitive exp.H /-th root of unity. This is possible by Sections 8.1 and 8.2 without
loss of generality. Then by (8.4) the character group HO of H becomes a K -module
Q
and hence also a G-module via
1
 ./ D ..  // for all  2 HO ;  2 G;
Q  2 H: (9.1)

Every solution field NQ of E .'; / is a Kummer extension of N and as such is

described by an exact sequence
"
1 ! N  ! XQ ! HO ! 1 (9.2)

of H -modules, where XQ denotes the elements of NQ  whose exp.H /-th power lies
in N , " is the inclusion and  the homomorphism

x
 W XQ ! HO ; x 7! .x/ with .x/./ D . 2 H /:
x
Therefore the action of H on XQ is given by

x D x  "..x/.// for all x 2 XQ ;  2 H (9.3)

(see for example Jacobson (1980), Thm. 8.23). Since NQ is Galois over K, this is
Q
even an exact sequence of G-modules. This yields the following criterion for the
solvability of E .'; /:

Theorem 9.1. Let E .'; / be a finite embedding problem over K with abelian ker-
nel H , whose order is prime to the characteristic of K. Then E .'; / possesses a
Q
solution if and only if there exist a G-module XQ and G-homomorphisms
Q " and 
satisfying (9.2) and (9.3).
9 The Hasse Embedding Obstruction 359

Proof. It remains to deduce the solvability of E .'; / from the existence of XQ sat-
isfying (9.2) and (9.3) (which are necessary by what precedes). With a system of
representatives fx j  2 HO g of HO in XQ we construct from XQ an N -algebra
M
AQ WD Nx
2HO

with the product of representatives

x xQ D c ; Qx Q with c D .c ; Q/ 2 Z 2 .HO ; N  /

defined in XQ . Obviously AQ is Galois over N with group H and also Galois over K
with group G.Q Therefore it splits into a direct sum of K-isomorphic extension fields
Q
Ni of N Galois over K with

Gal.NQ i =K/ Š f 2 GQ j NQ i D NQ i g

(see for example Wolf (1956), 9, Satz 1, with comments in 11). The embedding
NQ of one of these into a given separable algebraic closure KN of K yields a solution
field generated by elements from X. Q Therefore the restriction of K to Gal.NQ =N /
defines a solution 'Q of E .'; /. This is proper precisely when AQ is already a field.
t
u
Q
We now want to check to what extent the G-module XQ is determined as H -
module by N ; H and the action (9.3) of H on X . The H -action on XQ , which is
 O Q
trivial on HO D XQ =N  by (9.1) and also on N  XQ , can be extended to a ZZŒH -
action on these modules in a natural way. Then (9.2) becomes an exact sequence of
ZZŒH -modules, where for better distinction XQ as ZZŒH -module is denoted by X . By
trivial extension of the H -action from HO to ZZŒHO and extension to a ZZŒH -action,
the free ZZ-module ZZŒHO becomes a ZZŒH -module. The fiber product XH of X with
ZZŒHO over HO in the category of ZZŒH -modules with respect to the homomorphisms
X Y
 W X ! HO and  W ZZŒHO ! HO ; m  7! m ; (9.4)
2HO 2HO

thus decomposes as ZZ-module into the direct sum of N  and ZZŒHO :

XH D X HO ZZŒHO D N  ˚ .ZZŒHO / (9.5)

with the ZZ-section W ZZŒHO ! XH . According to (9.3) we have the well-defined

ZZŒH -action

x D x  . ı H .x//./ for x 2 XH ;  2 H;
360 IV Embedding Problems
P
on XH . Using the direct decomposition x D .u; . 2HO m // and substituting
ZZŒHO for .ZZŒHO / in (9.5) we obtain the formula
X Y X
.u; m / D .u ./m ; m /: (9.6)
2HO 2HO 2HO

Thus XH is determined up to equivalence as the direct sum of the ZZ-modules N 

and ZZŒHO with the ZZŒH -action (9.6) not depending on X , and therefore can be
constructed even without the knowledge of X . This proves:
Q
Corollary 9.2. Denote by X the G-module XQ in Theorem 9.1 considered as ZZŒH -
module and ZZŒH the trivial ZZŒH -module over HO . Then as a ZZ-module the fiber
O
product in the category of ZZŒH -modules XH WD X HO ZZŒHO decomposes into a
direct sum N  ˚ ZZŒHO with the ZZŒH -action given in (9.6). In particular, there
exists a ZZŒH -epimorphism from XH onto X .

The assertion of Corollary 9.2 can also be visualized by the following commuta-
tive diagram with exact rows:
"H
XH ! ZZŒ?HO ! 1
eH W 1 ! N  ! ?
H

x ? ?
?  jj y X y (9.7)
1
"
e W 1 ! N  ! X ! HO ! 1

Since here the two exact sequences are determined up to equivalence, they define
elements eH D eH .'; / 2 Ext1H .ZZŒHO ; N  / and e D e.'; / 2 Ext1H .HO ; N  / (with
the convention ExtnH .; / WD ExtnZZŒH .; / for groups H ) which depend formally on
' and  via N WD KN ker.'/ and H WD ker./. These satisfy 1 .e/ D eH with the map
1 induced on Ext1H .HO ; N  / by .

Remark. For each  2 G acting on ZZŒHO by (9.1) and on N  via the Galois action
we obtain an exact sequence equivalent to the first line of (9.7) because of
Y X Y X
.u ./m ; m / D .u  .  /; m  /:
2HO 2HO 2HO 2HO

Therefore, the element eH 2 Ext1H .ZZŒHO ; N  / even lies in Ext1H .ZZŒHO ; N  /G .

9.2 Definition of the Hasse Obstruction

Q
Starting from the G-module XQ in Theorem 9.1 we may form the fiber product of
X and ZZŒH over H as ZZŒGQ -module. This fits into the following commutative
Q O O
9 The Hasse Embedding Obstruction 361

diagram with exact rows

"Q Q
XQ H ! ZZŒ?HO ! 1
eQH W 1 ! N  ! ?
H H

x ?Q ?Q
?Q jj y X y (9.8)
1
"
eQ W 1 ! N  ! XQ ! HO ! 1

with eQH D eQH .'; / 2 Ext1GQ .ZZŒHO ; N  / and eQ D e.';

Q / 2 Ext1GQ .HO ; N  /. As a first
condition for the solvability of the embedding problem E .'; / we have to answer
the question whether the H -action on XH may be extended to a G-action Q on XQ H
compatible with the given G-action.
The following lemma contains an argument needed to prove the subsequent two-
step solvability criterion and will itself be used again later on. Here a G-module M
is called cohomologically trivial in dimension i , if H i .U; M / D 0 for all subgroups
U of G.

Lemma 9.3. Let M be a G-module cohomologically trivial in dimension i. Then we

have
H i .G; Hom.ZZŒHO ; M // D 0: (9.9)

Proof. Let R be a system of representatives of the finitely many G-orbits on HO and

G the stabilizer of  2 R. Then ZZŒHO possesses the following direct decomposition
as G-module Y
ZZŒHO D ZZŒG=G :
2R

From this the Lemma of Shapiro (see for example Serre (1964), Ch. I, Prop. 10, or
Shatz (1972), Ch. II, Thm. 8) yields
Y
H i .G; Hom.ZZŒHO ; M // Š H i .G; Hom.ZZŒG=G ; M //
2R

Y Y
Š H i .G ; Hom.ZZ; M // Š H i .G ; M / D 0: t
u
2R 2R

Theorem 9.4 (Yakovlev (1964)). Let E .'; / be a finite embedding problem with
abelian kernel H of order prime to the characteristic of K and N be the fixed
field of ker.'/. Further denote by eH D eH .'; / 2 Ext1H .ZZŒHO ; N  /G the uniquely
determined equivalence class of extensions of ZZŒH -modules in (9.7) according
to Corollary 9.2 with Remark. Then we have:
(a) E .'; / is concordant if and only if there exists an element eQH D eQH .'; / 2
Ext1GQ .ZZŒHO ; N  / with restriction 1 .eQH / D eH . This is then uniquely determined
as the preimage of eH .'; /.
(b) E .'; / has a solution, if in addition there exists an element eQ D e.';Q / 2
Ext1GQ .HO ; N  / with Q1 .e/
Q D eQH .
362 IV Embedding Problems

Proof. By Theorem 9.1 and the above we only have to prove part (a). The group
extension
 
1 ! H ! GQ ! G ! 1
with G D '. K / and H D ker./ corresponding to the cohomology class h 2
H 2 .G; H / gives rise to the Hochschild–Serre sequence (see for example Serre
(1964), Ch. I, 2.6, or Shatz (1972), Ch. II, 4, (25))
1  #2
Q M / ! H 1 .H; M /G ! H 2 .G; M H /
0 ! H 1 .G; M H / ! H 1 .G;
1

Q
with the inflation 1 and the restriction 1 and the transgression #2 for the G-module

M WD N ŒH 
Š Map.HO ; N  / Š Hom.ZZŒHO ; N  /; (9.10)

where the canonical isomorphisms in (9.10) are used for the identification. Using
that H acts trivially on HO and N  , and with the canonical isomorphism

H i .F; N ŒH  / D H i .F; Hom.ZZŒHO ; N  // Š ExtiF .ZZŒHO ; N  / (9.11)

Q H g (see for example Benson (1991), Prop. 3.1.8(iii)), as well as the
for F 2 fG; G;
vanishing of Ext1G .ZZŒHO ; N  / by Hilbert’s Satz 90 and Lemma 9.3 we obtain the
two isomorphic exact sequences
1  #2
0 ! H 1 .G;Q N ŒH  / ! H 1 .H; N ŒH  /G ! H 2 .G; N ŒH  /
1
? ? ?
yŠ yŠ yŠ (9.12)
1 O  1 O  G 2 O 
0 ! ExtGQ .ZZŒH ; N / ! ExtH .ZZŒH ; N / ! ExtG .ZZŒH ; N /:

Here the element eH 2 Ext1H .ZZŒHO ; N  /G is given as an element of the group

H 1 .H; N ŒH  /G Š Hom.H; Hom.ZZŒHO ; N  //G by the homomorphism
X Y
ıH W H ! Hom.ZZŒHO ; N  /;  7! . m  7! ./m /:
2HO 2HO

Thus an element eQH 2 Ext1GQ .ZZŒHO ; N  / Š H 1 .G;

Q N ŒH  / with  .eQH / D eH
1
exists precisely if the homomorphism ıH can be extended to a crossed homo-
morphism ıQH from GQ to Hom.ZZŒHO ; N  / (for this see for example Hilton and
Stammbach (1971), Ch. VI.5). Furthermore, eQH is then uniquely determined by eH
because of the injectivity of 1 .
To determine the extendability of ıH we let fQ j  2 Gg be a system of rep-
resentatives of G in GQ with 1Q G D 1GQ and with cocycle b D .b; / 2 h (compare
Theorem 6.1). Then for each

c WD .c .// 2HO 2 Map.HO ; N  / Š N ŒH 

9 The Hasse Embedding Obstruction 363

we obtain by multiplicative extension of ıH from

X Y
ıQH W GQ ! Hom.ZZŒHO ; N  /; Q 7! . m  7! c ./m /
2HO 2HO

a well-defined map from GQ to N ŒH  . It can easily be checked that this is a crossed

homomorphism with respect to the given action of G on HO and N if and only if

c c D c b; with b; WD ..b; // 2HO 2 Map.HO ; N  / Š N ŒH  ;

for all ;  2 G, i.e., if the factor system b splits over N ŒH  . Therefore, if we

denote by 2 the cohomological map

2 W H 2 .G; H / ! H 2 .G; N ŒH  / (9.13)

induced by the embedding W H ! N ŒH  , then #Q H becomes a crossed homomor-

phism if and only if 2 .h/ D 0.
Finally we prove that 2 .h/ D 0 if and only if E .'; / is concordant. For this we
use the final part of the Hochschild–Serre sequence for the group extension K D
N G

Q 'Q 
 G #2  
H 1. N ; N ŒH / ! H 2 .G; N ŒH / ! H 2 . K ; N ŒH /: (9.14)

Now let R denote a system of representatives of the orbits in HO under the action of
G respectively K and ΠG the G-orbit of  2 R in HO . Then we have
Y Y
H 2 .G; N ŒH  / Š H 2 .G; Map.Œ G ; N  // Š H 2 .G ; N  /
2R 2R

by the Lemma of Shapiro since Map.ΠG ; N  / is the G-module induced from the
G -module N  . With the corresponding decomposition of the K -module N ŒH 
from (9.14) we arrive at the following diagram

'

H Q
H 2 .G;
? H/ ! H 2 . ?K ; H / ! 2HO
Br.K /
?  ?
y2 y
x
?
'Q  ? (9.15)
H 2 .G; N ŒH  / ! H 2 . K ; N ŒH  / ?
?
? ?
? ?
yŠ yŠ
Q 
Q Q
2R H 2
.G ; N / ! 2R H 2
. ; N  / ! 2R Br.K /

Here the homomorphisms in the lower row are injective on the exp.H /-torsion part,
since their composite by Proposition 7.1 gives the embedding of the exp.H /-torsion
of the relative Brauer groups H 2 .G ; N  / Š Br.K ; N / into the absolute Brauer
groups Br.K / (see Jacobson (1980), Ch. IV.7). Since the preceding considerations
364 IV Embedding Problems

are independent of the choice of the system of representatives R  HO the charac-


terization of the concordance condition H ı '  .h/ D 0 from Theorem 8.4(a) by

2 .h/ D 0 follows immediately from the diagram (9.15). t
u
Remark. By the explicit description of the transgression #2 (see for example Suzuki
(1982), Ch. 2, 7) it follows from

ıQH .Q /
Q c c
  D b; for Q ; Q 2 GQ with .Q / D ; .Q / D  2 G;
ıQH .Q / ıQH ./
Q c

that #2 .eH / D 2 .h/. Therefore,

p.'; / WD #2 .eH / D 
2 .h/ 2 H 2 .G; N ŒH 
/ Š Ext2G .ZZŒHO ; N  / (9.16)

constitutes the concordance obstruction (or first embedding obstruction) of E .'; /.
By Theorem 9.4(b) the Hasse embedding obstruction (or second embedding
obstruction) for the solvability of a concordant embedding problem is the obstruc-
Q / 2 Ext1GQ .HO ; N  / with Q1 .e/
tion for the existence of an eQ D e.'; Q D eQH . This is
obtained in a natural way as the image of eQH under the map

Q 1 W Ext1GQ .ZZŒHO ; N  / ! Ext1GQ .Y .HO /; N  / (9.17)

in the long sequence of Ext-groups induced by the embedding Q in the exact

sequence of ZZŒG -modules

Q Q
1 ! Y .HO / ! ZZŒHO ! HO ! 1 (9.18)
Q for Q in (9.8). Thus (9.17) with (9.11) leads to the follow-
with Y .HO / WD ker./
ing consequence of Theorem 9.4 (see also Ishkhanov, Lure and Faddeev (1997),
Thm. 3.11):
Corollary 9.5. The Hasse embedding obstruction q.'; / for the solvability of a
concordant finite embedding problem E .'; / with abelian kernel is the image of
eQH .'; / 2 Ext1GQ .ZZŒHO ; N  / under the map Q 1 in (9.17):

q.'; / WD Q 1 .eQH .'; // 2 Ext1GQ .Y .HO /; N  / Š H 1 .G;

Q Hom.Y .HO /; N  //: (9.19)

In the next sections under additional assumptions we give more convenient

expressions for the Hasse embedding obstruction.

9.3 Translation of the Hasse Obstruction

First we want to simplify the Hasse obstruction by reducing G to a suitable factor

group of a subgroup of index prime to jH j. The first step is achieved as follows:
9 The Hasse Embedding Obstruction 365

Proposition 9.6. Let E .'; / be a concordant finite embedding problem for which
the assumptions of Theorem 9.4 are satisfied. Let U be a normal subgroup of
G D '. K / acting trivially on the character group HO , with factor group F WD G=U
and fixed field L WD N U . Then the Hasse obstruction q.'; / 2 Ext1GQ .Y .HO /; N  /
has a representation as the preimage in Ext1F .Y .HO /; L / under an injective homo-
morphism.

Proof. We first consider the following commutative diagram, whose lower row is
the first part of the Hochschild–Serre sequence (9.12) restricted to Y .HO /, and with
1 , 1 , Q 1 in their previous meaning:


O  1
O 
? H ; N / ! ExtH .ZZŒ
Ext1GQ .ZZΠ1
? H ;N /
?  ? 
yQ1 y1 (9.20)
Q 1 Q
0 ! Ext1G .Y .HO /; N  / ! Ext1GQ .Y .HO /; N  / ! Ext1H .Y .HO /; N  /:
1

The maps here satisfy .1 ı 1 /.eQH / D 1 .eH / D 0, since eH has preimages under
1 in Ext1H .HO ; N  / by Corollary 9.2. Consequently the Hasse obstruction q.'; /
possesses precisely one preimage

qG D qG .'; / 2 Ext1G .Y .HO /; N  / with Q 1 .qG / D q.'; /: (9.21)

For the second step we exploit the exact sequence

! 
1 ! U ! G ! F ! 1: (9.22)

The first part of the Hochschild–Serre sequence for this group extension yields an
exact sequence
 !
0 ! Ext1F .Y .HO /; L / ! Ext1G .Y .HO /; N  / ! Ext1U .Y .HO /; N  /F :
1 1
(9.23)

Since U acts trivially on ZZŒHO and hence on Y .HO /, and Y .HO / is finitely generated
as subgroup of ZZŒHO and torsion free, Y .HO /Lsplits as U -module into a direct sum
of finitely many copies of ZZ, say Y .HO / Š riD1 ZZ. Using Hilbert’s Satz 90 this
yields

Y
r Y
r
Ext1U .Y .HO /; N  / D Ext1U .ZZ; N  / D H 1 .U; N  / D 0:
i D1 i D1

Thus the map 1 in (9.23) is even bijective, and there exists a unique element

qF D qF .'; / 2 Ext1F .Y .HO /; L / with 1 .qF / D qG .'; /: (9.24)

t
u
366 IV Embedding Problems

In the special case G D U we may recover Theorem 8.7 from Proposition 9.6.
With the reduction principle of Kochendörffer for concordant and solvable embed-
ding problems we obtain the final form of Theorem 9.4:
Theorem 9.7 (Yakovlev (1964)). Let E .'; / be a finite embedding problem with
abelian kernel H of order prime to the characteristic of K, N the fixed field of
ker.'/ and G D Gal.N=K/. Further let V be a subgroup of G with
gcd..G W V /; jH j/ D 1 containing a normal subgroup U of G with trivial action
on HO . Then the Hasse obstruction q.'; / may be represented by an element of
Ext1F .Y .HO /; L / with F WD V =U and L WD N U .
Remark. From now on we will distinguish between the different representations of
the Hasse obstruction, namely q.'; / in Ext1GQ .Y .HO /; N  / from Equation (9.19),
in Ext1G .Y .HO /; N  / from (9.21) respectively in Ext1F .Y .HO /; L / from (9.24), only
by indicating the relevant group.
In the next section we further study the Hasse embedding obstruction in the case
of a global ground field K.

9.4 The Hasse Obstruction for Global Fields

In this section let N =G K be a finite Galois extension of global fields. For such
an extension we have the following two fundamental short exact sequences of G-
modules
˛ ˇ
1 ! N  ! J.N / ! C.N / ! 1 (9.25)
with the idele group J.N / and the idele class group C.N /, and

˛Q ˇQ
1 ! Dı .N / ! D.N / ! ZZ ! 1 (9.26)

with the free abelian group D.N / (written multiplicatively) over IP.N / and the obvi-
ous extended G-action from IP.N / to D.N /, and the degree map
Y X
ˇQ W D.N / ! ZZ; pmp 7! mp  deg.p/;
p2IP.N / p2IP.N /

into the trivial G-module ZZ.

Now we tensor with a torsion free G-module M and extend the G-action on the
G-module A onto M ˝ A via

.x ˝ a/ WD x  ˝ a for x 2 M; a 2 A;  2 G:

By this process, (9.25) and (9.26) again yield short exact sequences of G-modules
and then long cohomology sequences for the modified cohomology groups of Tate.
These satisfy:
9 The Hasse Embedding Obstruction 367

Theorem 9.8 (Comparison Theorem of Tate (1966)). Let N =G K be a finite Galois

extension of global fields and M a torsion free G-module. Then the long exact
sequences of Tate cohomology groups originating from (9.25) and (9.26) are con-
nected for all i 2 ZZ by the commutative diagram

@i ˛Q i ˇQi @i C1
! H i .G; M
? ˝ D .N // ! H .G; M
0 i
? ˝ D.N // ! H i .G;?M ˝ ZZ/ !
? N ? J ? C
yi yi yi
@j ˛j ˇj @j C1
! H j .G; M ˝ N  / ! H j .G; M ˝ J.N // ! H j .G; M ˝ C.N // !

with isomorphisms iN , iJ and iC , where j WD i C 2.

For the proof we refer the reader to the original paper Tate (1966).
Now let M be a G-module, finitely generated and free as ZZ-module. Then
Hom.M; ZZ/ is torsion free and satisfies the assumptions of Theorem 9.8. The valid-
ity of

Hom.M; ZZ/ ˝ A Š Hom.M; ZZ/ ˝ Hom.ZZ; A/ Š Hom.M; A/ (9.27)

for all G-modules A then implies, together with Theorem 9.8, the exactness of

: : : ! H i .G; Hom.M; D.N /// ! H i .G; Hom.M; ZZ// !

! H i C3 .G; Hom.M; N  // ! H i C1 .G; Hom.M; D.N /// ! : : : (9.28)

for all i 2 ZZ. The decomposition

Y Y
D.N / D Dp with Dp WD pQ ZZ
p2IP.K/ Q 2IP.N /;Q
p pjp

of D.N / as G-module first implies

M
Hom.M; D.N // Š Hom.M; Dp /:
p2IP.K/

Denoting by Gp the decomposition group of an extension pQ of p 2 IP.K/ in IP.N /

we obtain that Dp respectively Hom.M; Dp / are the G-modules induced from the
Gp -modules pQ ZZ respectively Hom.M; ZZ/. The Lemma of Shapiro then gives
M
H i .G; Hom.M; D.N /// Š H i .Gp ; Hom.M; ZZ//: (9.29)
p2IP.K/

Equations (9.28) and (9.29) now yield the following exact sequence as a conse-
quence of the Comparison Theorem of Tate:

Corollary 9.9. Let N =G K be as in Theorem 9.8 and M a G-module which is finitely

generated and free as ZZ-module. Then the following sequence of Tate cohomology
368 IV Embedding Problems

groups is exact for all i 2 ZZ:

L ˇQi @i C3 ıiC
! .Gp ; Hom.M; ZZ// ! H i .G; Hom.M; ZZ// !
p2IP.K/ H
i

.iJC1 /1 ı˛i C3 L

H i C3 .G; Hom.M; N  // ! p2IP.K/ H
i C1
.Gp ; Hom.M; ZZ// !(9.30)

A further step for the main result of this section will be given by the following
short exact sequence.

Proposition 9.10. Let A be a finite abelian group and A WD Hom.A; Q=ZZ/ its dual
group. Then the following sequence is exact:

0 ! Hom.ZZŒA ; ZZ/ ! Hom.Y .A/; ZZ/ ! A ! 0: (9.31)

Proof. From the exact injective resolution

0 ! ZZ ! Q ! Q=ZZ ! 0

of ZZ we obtain the following piece of the long exact sequence of Hom- and Ext-
groups

0 D Hom.A; Q/ ! Hom.A; Q=ZZ/ ! Ext1 .A; ZZ/ ! Ext1 .A; Q/ D 0

with Ext1 D Ext1ZZ and thus

Ext1 .A; ZZ/ Š Hom.A; Q=ZZ/ D A : (9.32)

Further the exact sequence (9.18) with A in place of HO yields:

0 ! Hom.A; ZZ/ ! Hom.ZZŒA ; ZZ/ ! Hom.Y .A/; ZZ/ !

Ext1 .A; ZZ/ ! Ext1 .ZZŒA ; ZZ/ :

Here we have Hom.A; ZZ/ D 0 due to the finiteness of A and Ext1 .ZZŒA ; ZZ/ D 0 be-
cause ZZŒA is free as ZZ-module. Thus the above, together with (9.32), yields (9.31).
t
u
The commutative diagram in Figure 9.1 of Tate cohomology groups is patched
together from the following ingredients. The upper two rows come from the exact
sequence (9.30) of Tate for F; L in place of G; N and the F -modules ZZŒHO and
Y .HO /. The first two columns are the long cohomology sequences arising from the
short exact sequence (9.31), and we have used the abbreviations

V WD Hom.ZZŒHO ; ZZ/; W WD Hom.Y .HO /; ZZ/;

P WD H .F; Hom.ZZŒHO ; L //; Q WD H 1 .F; Hom.Y .HO /; L //:
1

The third row is completed with the cokernel T of the homomorphism  defined
by the vanishing of the cohomology groups in the last row which will be proved
9 The Hasse Embedding Obstruction 369

together with Theorem 9.11. The sums are all taken over p 2 IP.K/.
 −2  −1
H ⏐(F ,V ) −→ H−2 †
⏐ (F,V ) −→ P⏐ −→ H ⏐(F ,V )
   
 −2 
H ⏐(F ,W ) −→ H−2 Q −→ H−1 (F ,W )†
⏐ (F,W ) −→ ⏐
  
 −2
H ⏐ (F , Ĥ ∗) −→ H−2 ∗
⏐(F, Ĥ ) −→ T −→ 0
 
 −1 −1
H (F ,V ) −→ H (F,V )†
†

Fig. 9.1 Hasse obstruction for global fields

From this we can deduce the following result (see Ishkhanov, Lure and Faddeev
(1997), Thm. 3.14.2, for the version dualized via loc. cit. Thm. A.6).

Theorem 9.11 (Yakovlev (1967)). Assume the hypotheses of Theorem 9.7 and more-
over let K be a global field. Then there exists an isomorphism
M
W Q D Ext1F .Y .HO /; L / ! T D H 2 .F; HO  /=. H 2 .Fp ; HO  //:
p2IP.K/
(9.33)

In particular the Hasse embedding obstruction from Theorem 9.7 possesses a rep-
resentation q.'; / 2 T .

Proof. We first prove that the four cohomology groups marked by $ in Figure 9.1
vanish. Obviously
P D H 1 .F; Hom.ZZŒHO ; L // D 0 (9.34)
by Lemma 9.3 since L is cohomologically trivial in dimension 1. Since the trivial
F -module ZZ is cohomologically trivial in dimension 1, the same lemma gives

H 1 .F; V / D H 1 .F; Hom.ZZŒHO ; ZZ// D 0; (9.35)

H 1 .Fp ; V / D H 1 .Fp ; Hom.ZZŒHO ; ZZ// D 0: (9.36)

1
It remains to show the vanishing of H .Fp ; W /. Since by Lemma 9.12 below

H 1 .Fp ; W / D H 1 .Fp ; Hom.Y .HO /; ZZ// Š H 1 .Fp ; Y .HO // (9.37)

it suffices to prove the vanishing of H 1 .Fp ; Y .HO //. For this we employ the follow-
ing long cohomology sequence derived from (9.18):

Q0 @1
0 !H 0 .Fp ; Y .HO // ! H 0 .Fp ; ZZŒHO / ! H 0 .Fp ; HO / !
Q 1
H 1 .Fp ; Y .HO // ! H 1 .Fp ; ZZŒHO /:
370 IV Embedding Problems

Let Rp be a system of representatives of the Fp -orbits in HO and F the stabilizer in

Fp of  2 Rp . Then analogously to the proof of Lemma 9.3 we have
Y
H 1 .Fp ; ZZŒHO / Š H 1 .Fp ; ZZŒFp =F /
2Rp
Y Y
Š H 1 .F ; ZZ/ Š Hom.F ; ZZ/ D 0:
2Rp 2Rp

since F is finite. Further, Q0 is surjective since every Fp -invariant element of HO

is the image of an Fp invariant element of ZZŒHO , and hence Q 1 is injective. This
implies H 1 .Fp ; Y .HO // D 0, which together with (9.37) finally gives

H 1 .Fp ; W / D H 1 .Fp ; Hom.Y .HO /; ZZ// D 0: (9.38)

Substituting (9.34), (9.35), (9.36) and (9.38) in Figure 9.1 we obtain from the lower
three rows the existence of a monomorphism W Q ! T making the diagram com-
mutative. By a simple diagram chasing in the upper three rows it follows that is
also surjective. This completes the proof of Theorem 9.11. t
u

Lemma 9.12. Let G be a finite group and M a G-module, which as ZZ-module is

finitely generated and free. Then we have

H i .G; Hom.M; ZZ// Š H i .G; M / for i 2 ZZ: (9.39)

Proof. We consider ZZ; Q and Q=ZZ as trivial G-modules. Since M is a projective

ZZ-module,

0 ! Hom.M; ZZ/ ! Hom.M; Q/ ! Hom.M; Q=ZZ/ ! 0 (9.40)

becomes an exact sequence of ZZ- as well as ZZŒG -modules. Now the multiplication
by jGj on Hom.M; Q/ is an isomorphism. Consequently, Hom.M; Q/ is relative
(weakly) projective (see Cartan and Eilenberg (1956), Ch. XII, Ex. 1). This implies
(by loc. cit., Ch. XII, Prop. 2.2) that

H i .G; Hom.M; Q// D 0 for i 2 ZZ: (9.41)

Thus in the long cohomology sequence derived from (9.40) we get

@i
0 ! H i 1 .G; Hom.M; Q=ZZ// ! H i .G; Hom.M; ZZ// ! 0;

so the connecting homomorphisms @i become isomorphisms. Together with the

isomorphisms

i 1;i W H i 1 .G; Hom.M; Q=ZZ// ! H i .G; M / (9.42)

9 The Hasse Embedding Obstruction 371

coming from the duality theorem (loc. cit., Ch. XII, Thm. 6.4, (4’), resp. Ishkhanov,
Lure and Faddeev (1997), Thm. A.6) this yields the required isomorphisms i 1;i ı
@1
i for (9.39). t
u

The representation of the Hasse obstruction in Theorem 9.11 is particularly suited

for the shrinking process which was used by Šafarevič (1954d, 1958) for the real-
ization of solvable groups as Galois groups and for the solution of split embedding
problems with nilpotent kernel over number fields (compare the dualized version in
Ishkhanov, Lure and Faddeev (1997), Prop. 5.5.4). A complete proof of the theo-
rem of Šafarevič has been executed in some detail for example in the monograph
Neukirch, Schmidt and Wingberg (2000), Ch. X. Here instead of giving a short
report on this proof we prefer to explain the easier special case of nilpotent groups
but with complete proofs. This again constitutes a bridge between embedding prob-
lems and the results on G-realizations over fields in positive characteristic presented
in the next chapter.
372 IV Embedding Problems

10 Nilpotent Galois Groups over Global Fields

To close this chapter we use the method of Scholz (1937) and Reichardt (1937)
to construct Galois extensions with nilpotent Galois group over global fields K,
where we assume moreover that the group order is coprime to the order of the group
of roots of unity contained in the prime field of K. In the case of global fields of
positive characteristic this technical condition could be removed by a method of
proof presented by Madan, Rzedowski-Calderon and Villa-Salvador (1996). This
then yields in particular G-realizations for all nilpotent groups over finite fields.

10.1 Scholz Extensions

In what follows K will usually denote a global field of characteristic p. Otherwise

we will speak of general fields.
Nilpotent groups are direct products of groups of prime power order. By Theo-
rem 1.6(a) it hence suffices to realize finite `-groups G as Galois groups, for ` 2 IP.
These possess an upper central series with cyclic factors, say

G D Gn ; Gn1 D Gn =Z` ; Gn2 D Gn1 =Z` ; : : : ; G1 D Z` ;

where the cyclic normal subgroup Z` is contained in the center of Gi for each i .
Thus `-extensions over K may be built by solving central embedding problems with
cyclic `-kernel. Such embedding problems are always solvable for ` D p and even
properly solvable over global fields by Theorem 8.3(b), so we may assume from
now on that ` ¤ p.
The following result shows that the Hasse obstruction for such embedding prob-
lems is always trivial:
Proposition 10.1. Let K be an arbitrary field and E .'; / a finite central embed-
ding problem over K with elementary abelian kernel H . If E .'; / is concordant
then it is also solvable.
Proof. Let ` 2 IP be the exponent of H . By our above remark we may assume that
` is different from the characteristic of K. Let ` denote an `-th root of unity. Since
gcd.ŒK. ` / W K ; `/ D 1 the restriction
K
K.`/
W H 2. K;H / ! H 2 . K.` / ; H / (10.1)

is injective (compare the proof of Theorem 8.2). Thus we may assume that K con-
tains ` . Now E .'; / is a central embedding problem over K D K. ` /, so K acts
trivially on H via '. K /, and hence because of ` 2 K also trivially on HO by (8.4).
Thus by Theorem 8.7 the solvability of E .'; / follows from its concordance. u t
Remark. In the case of a global field K with ` 2 K and a cyclic kernel H Proposi-
tion 10.1 also follows immediately from the Theorem of Brauer, Hasse and Noether
10 Nilpotent Galois Groups over Global Fields 373

(see for example Weil (1974), Ch. XI, Thm. 2, resp. the proof of Prop. 7.6), which
yields the injectivity of the product of the restriction maps
Y
 W H 2 . K ; H / ! H 2 . Kp ; H /
p2IP.K/

(see Corollary 7.8 resp. Theorem 8.12(b)).

With Theorem 8.12(a) the preceding Proposition gives:

Corollary 10.2. For a finite central embedding problem E .'; / with elementary
abelian kernel over a global field K we have: E .'; / is solvable if and only if for
all p 2 IP.K/ the local embedding problems E .'p ; p / are solvable.

Remark. If in Proposition 10.1 and Corollary 10.2 the embedding problem even has
a cyclic kernel, then the lifted embedding problem over K. ` / becomes a Brauer
embedding problem. Because of the product formula for Hasse-invariants (see for
example Weil (1974), Ch. XIII, Thm. 2) it then suffices to prove local solvability
for all but one p 2 IP.K. ` //.

We now formulate a sufficient condition for the local and hence global solvability
of central embedding problems with cyclic kernel. For this, let Ram.N=K/ denote
the set of prime divisors of K ramified in N=K. Here for simplicity in the case of
number fields we add those infinite primes of K which yield a non-trivial extension
of the completion Kp .

Proposition 10.3. Let E .'; / be a finite central embedding problem with cyclic
kernel H Š Z` for ` 2 IP over a global field K and N the fixed field of ker.'/ with
Gal.N=K/ D G. Furthermore, let

S0 WD fp 2 IP.K/ j pj` or pj1g (10.2)

in the case that K has characteristic 0, and S0 WD ; otherwise. Assume that all
p 2 Ram.N=K/ and pQ 2 IP.N / with pQ jp satisfy the following three conditions:

.0/ p … S0 ; .1/ N .p/ 1 .mod `e /; .2/ D.pQ =p/ D I.pQ =p/;

where `e D exp` .jH:Gj/ and N denotes the absolute norm. Then E .'; / is solv-
able. In the case ` 2 K one exception for p 2 Ram.N=K/ is admissible.

Proof. For p 2 IP.K/ we consider the local embedding problems E .'p ; p / induced
by E .'; /:

GK
j̃ j

1 Z G̃ G 1
k
374 IV Embedding Problems

with Gp D Gal.NpQ =Kp / Š D.pQ =p/ (compare (7.10) and (7.11)). If E .'p ; p / is split,
it is trivially solvable. This case occurs in characteristic 0 for example for all prime
divisors p of 1, since then Gp D 1 by the convention introduced before Proposi-
tion 10.3. By Zassenhaus’ Splitting Theorem (Huppert (1967), Ch. I, Thm. 18.1)
the same holds for all prime divisors of jGj with p6 j` since there gcd.`; jGp j/ D 1
by (2). Hence we may now assume that E .'p ; p / is a Frattini embedding problem
and moreover that p is only tamely ramified in N=K since the prime divisors of `
are unramified in N=K by (0).
In the unramified case Gp is the Galois group of a residue field extension of
global fields, hence cyclic. As E .'p ; p / is a Frattini embedding problem, this
implies that GQ p is again cyclic. Since the maximal unramified Galois extension of
Kp has procyclic Galois group, 'p can be lifted to an unramified solution 'Qp of
E .'p ; p /.
In the tamely ramified case again Gp Š D.pQ =p/ Š I.pQ =p/ is cyclic of order `f for
some f e 1, say. Hence the condition `e 2 Kp ensures that NpQ =Kp is a Kummer
extension, generated by an `f -th root of a prime element of Kp . As `f C1 2 Kp
the `f C1 -th roots of this prime element generate a cyclic Galois extension of Kp
with group GQ p D Z`f C1 in the non-split case, and there exists an epimorphism
'Qp W Kp ! GQ p with p ı 'Q p D 'p .
Thus all local embedding problems E .'p ; p / are solvable and hence by Propo-
sition 10.1 also the global embedding problem E .'; /. In the case ` 2 K we may
admit one exception by the Remark following Corollary 10.2. t
u

In generalization of conditions (0)–(2) in Proposition 10.3 we call a finite Galois

extension N=K an n-Scholz extension relative to S1  IP.K/ if all p 2 Ram.N=K/ n
S1 and pQ 2 IP.N / with pQ jp satisfy

(0) p … S0 ; (1) n 2 Kp ; (2) D.pQ =p/ D I.pQ =p/: (10.3)

(In the general definition of relative Scholz extensions it is usually assumed that
moreover S0  S1 , so that condition (0) becomes superfluous.)
Furthermore, we call an embedding problem E .'; / over K an n-Scholz embed-
ding problem (relative to S1 ) if the fixed field N of ker.'/ is an n-Scholz extension
of K (relative to S1 ), and a solution 'Q of E .'; / is called an n-Scholz solution (rel-
ative to S1 ) if the solution field NQ of 'Q is a n-Scholz extension of K (relative to
S1 ).

Corollary 10.4. Let K be a global field and S1  IP.K/ either empty or consisting
of a prime divisor which is nonsplit in K. ` /=K. Then every finite central `m -Scholz
embedding problem relative to S1 with kernel Z` and exp` .jGj/ < `m possesses a
proper solution over K.

Proof. If E .'; / is split, then it possesses a proper solution by Theorem 2.4. In the
non-split case it at least has a solution by Proposition 10.3 and using the injectivity
K
of the restriction K. `/
in (10.1). But by Proposition 1.8 this is a proper solution,
since in this case we have a Frattini embedding problem. t
u
10 Nilpotent Galois Groups over Global Fields 375

For an inductive proof it remains to determine when Scholz embedding problems

even possess (proper) Scholz solutions.

10.2 Scholz Embedding Problems

A Galois extension N=K over a global field K with `-Galois group G can be
obtained for example by tensoring a Galois extension N0 =K0 with Gal.N0 =K0 / Š
G over the prime field K0 D Q respectively K0 D IFp .t/, p ¤ `, with K over K0 ,
at least in the case that K=K0 and N0 =K0 are linearly disjoint. The latter can be
guaranteed by requiring that Ram.K=K0 / \ Ram.N0 =K0 / D ;. Thus we will now
construct Galois extensions with `-Galois groups over K0 whose socle is ramified
outside a given finite set of prime divisors of K0 . Here the socle of an `-Galois
extension N=K with group G is the maximal elementary abelian intermediate field,
i.e., the fixed field of ˚.G/ D G ` G 0 .

Lemma 10.5. Let K D Q respectively K D IFp .t/, ` ¤ p a prime with ` … K,

S1 WD f.`/g in the case K D Q, resp. S1 D ; otherwise, and S  IP.K/ a finite
subset containing S0 . Then we have:
(a) Every split central (geometric) `m -Scholz embedding problem E .'; / over K
relative to S1 with kernel Z` and exp` .'. K // < `m possesses a proper (geometric)
`m -Scholz solution relative to S1 .
(b) If '. K / is an `-group and the socle of N=K of the fixed field N of ker.'/
is ramified outside of S only, then E .'; / possesses also such a solution. Moreover,
its solution field NQ satisfies

Ram.NQ =K/  Ram.N=K/ [ fqg for some q 2 IP.K/ n S: (10.4)

Proof. For r 2 Ram.N=K/ let r denote the associated prime number in the case of
Q respectively the associated monic prime polynomial r 2 IFp Œt with numerator
divisor r in the case of IFp .t/ (with r D 1=t in the case r D p1 ). Further let N 
denotep̀the field generated over N by a primitive `m -th root of unity `m and the
roots r for r 2 Ram.N=K/. By the Dirichlet Density Theorem (see for example
Narkiewicz (1990), Cor. 7 to Prop. 7.9) there exists a prime ideal q 2 IP.K/ n S
with associated prime element q which splits completely in N  =K. In particular it
satisfies N .q/ 1 .mod `m /.
.q/
In the case K D Q let KC denote the maximal real subfield of the cyclotomic
field Q. q / of degree .q  1/=2. Correspondingly in the case of K D IFp .t/ let
.q/
KC denote the maximal subfield of the cyclotomic extension K .q/ =K in which
the denominator divisor p1 of .t/ splits completely. This field coincides with the
ray class field modulo q and has degree .p d  1/=.p  1/ over K, with d D deg.q/
.q/
(see for example Goss (1996), Ch. 7.5). Then in both cases KC =K is unramified
outside q.
376 IV Embedding Problems

Since ` ¤ 2 in the case K D Q respectively `6 j.p  1/ in the case K D IFp .t/

.q/
the extension KC contains a cyclic subfield L of degree `. The divisor q is
=K p̀
completely split in K. r/=K for r 2 Ram.N=K/, so there exists an x 2 K with
x ` r .mod q/. By the decomposition law r is also completely split in L=K.
The composite NQ WD LN thus has the Galois group Gal.NQ =K/ D G  Z` with
Ram.NQ =K/ D Ram.N=K/ [ fqg. Since we have q … S and by the decomposition
properties of q and r 2 Ram.N=K/ in N=K respectively L=K the extension NQ =K
is `m -Scholz relative to S1 . Obviously ' may now be extended via the canonical
epimorphism W K ! Gal.L=K/ Š ker./ to an `m -Scholz solution (relative to
S1 )
'Q D '  W K ! G  Z` D GQ with  ı 'Q D ':
In the case of positive characteristic L=K is a geometric field extension. Thus NQ =K
is also a geometric field extension. This proves (a).
If N=K is an `-extension with socle S , then the socle SQ of the solution field NQ
of 'Q satisfies
Ram.SQ =K/  Ram.S=K/ [ fqg:
In particular the socle of NQ =K is unramified in S, which proves (b). t
u

Remark. If in Lemma 10.5 the denominator divisor p1 of .t/ splits completely in

N=K, then it also splits completely in NQ =K.

Lemma 10.6. Let K and S1 be as in Lemma 10.5. Then every (geometric) non-split
central `m -Scholz embedding problem E .'; / over K relative to S1 with kernel Z`
and exp` .'. K // < `m possesses a proper (geometric) solution where the solution
field NQ satisfies
Ram.NQ =K/  Ram.N=K/ [ S1 : (10.5)

Proof. The existence of a proper solution 'Q with solution field NQ already follows
from Corollary 10.4. This will now be modified so that also (10.5) holds.
For this let q 2 Ram.NQ =K/ n Ram.N=K/ with q … S1 , i.e., q ¤ .`/, and Q 2
IP.N / a prime divisor of q. Then there exists a Q Q 2 IP.NQ / with Q Q ` D Q. Since
Q
the inertia group I.Q=q/ is a subgroup of the multiplicative group of the residue
field we have N .Q/ Q D N .Q/ 1 .mod `/. On the other hand we have N .Q/ D
lk
N .q/ with k 2 IN, which implies N .q/ 1 .mod `/. Thus as in the proof of
.q/
Lemma 10.5 the field KC possesses a subfield L of degree ` linearly disjoint to
N=K which is unramified outside q. The composite LNQ =K thus has Galois group
Gal.LNQ =K/ Š GQ  Z` , and the inertia group I of the extensions Q QQ 2 IP.LNQ / of q
lies in a central subgroup of type Z`  Z` . Since q is tamely ramified, I is cyclic.
The fixed field NQ 0 of I then has the Galois group

Gal.NQ 0 =K/ Š .GQ  Z` /=I Š G;

Q

and we have Ram.NQ 0 =K/ D Ram.NQ =K/ n fqg.

10 Nilpotent Galois Groups over Global Fields 377

Now as in Lemma 10.5 the given solution 'Q of E .'; / can be extended via the
canonical epimorphism W K ! Gal.L=K/ first to

'QQ D 'Q  W K ! GQ  Z` D Gal.LNQ =K/:

Then by composing with the canonical map LNQ =NQ 0 W Gal.LNQ =K/ ! Gal.NQ 0 =K/
we obtain an epimorphism 'Q 0 onto Gal.NQ 0 =K/ Š GQ with 'Q 0 ı  D '. Therefore,
'Q 0 defines a proper solution of E .'; / with solution field NQ 0 . Since in the case of
positive characteristic E .'; / is a geometric Frattini embedding problem, by Propo-
sition 1.8 also 'Q 0 is a geometric solution. The assertion hence follows by induction.
t
u
Remark. By using a theorem of Tate on the solvability of global embedding prob-
lems with local prescription (see for example Serre (1992), Prop. 2.1.7) the excep-
tional position of ` in Lemma 10.6 can be removed.
Lemma 10.7. Let K, S1 and S be as in Lemma 10.5. Then every (geometric) non-
split central `m -Scholz embedding problem E .'; / over K relative to S1 with ker-
nel Z` and j'. K /j D `n < `m possesses a proper (geometric) `m -Scholz solution
relative to S1 , where the solution field NQ satisfies

Ram.NQ =K/  Ram.N=K/ [ S1 [ fqg for some q 2 IP.K/ n S: (10.6)

Proof. By Lemma 10.6 the embedding problem E .'; / has a proper solution 'Q
whose solution field NQ satisfies (10.5). We are going to modify this solution to
obtain a Scholz solution, such that at most one additional ramified place appears.
For r 2 T WD Ram.NQ =K/ n S1 with associated prime element r fix prime divisors
R 2 IP.N / and R Q 2 IP.NQ /. Since I.R=r/ D D.R=r/ and ker./ Š Z` the decom-
position group D.R=r/Q is contained in the preimage IQ of type Z`  I.R=r/ in G.
Q If IQ
Q
is cyclic, then IQ D I.R=r/ Q
D D.R=r/. In the non-cyclic case the above shows that
Q
either also D.R=r/ Q
D I.R=r/ Q
or D.R=r/ Q
D I.R=r/  Z` . Now let
Q
T0 WD fr 2 T j D.R=r/ Q
D I.R=r/g;
T1 WD T n T0 D fr1 ; : : : ; rk g;
Q
and ri the prime elements associated to ri . For ri 2 T1 let i WD NR=N 2 ker.'/ Š
i
Z` be the Artin symbol of Ri (see for example Narkiewicz (1990), Ch. 7, 3). The
i are independent of the choice of the extension Ri =ri because ker.'/ Z .G/. Q
Since R Q 1 =R1 is inert, h1 i D Gal.NQ =N /, so there exists ei 2 IN with i D  ei .
1
0
The field K. `m / is the composite of K. ` /=K with a cyclic p̀ field K =K of degree
`m1 . Thus NQ K 0 =K is an `-extension and N 0 WD K. ` [ f r j r 2 Tg/ is a Kum-
mer extension with group Z`s over K. ` / for some s 2 IN. As ` … K the group
Gal.N 0 =K/ Š Z`s Zf with 1 ¤ f j.`  1/ has no quotient of order `. In particular,
NQ K 0 and N 0 are linearly disjoint over K.
By the Dirichlet Density Theorem there exists a q 2 IP.K/ p̀ n S with associated
prime element q which splits completely in NQ . `m [ f r j r 2 T0 g/=K and in
378 IV Embedding Problems
q p̀
e
K. ` ; `
ri =r1 i /=K for i D 2; : : : ; k and which is inert in K. ` ; r 1 /=K. ` /. As
.q/
N .q/ 1 .mod `/ the field KC as in the proof of Lemma 10.5 possesses a cyclic
subfield L of degree ` over K which is unramified outside of q and hencep̀is lin-
early disjoint with NQ =K. Since the prime divisor q splits completely in K. r/=K
for r 2 T0 there exists an x 2 K with x ` r .mod q/. By thedecomposition law
r splits completely in L=K, so the Artin symbol evaluates to L=K r
D 1. Corre-

spondingly r1 is inert in L=K with Artin symbol 1 WD L=K r1 ¤ 1. Since q splits
q
e
completely in K. ` ; ` ri =r1 i /=K there exist xi 2 K with

e
r1 i xi` ri .mod q/ for i D 2; : : : ; k;

L=K e
which implies i WD ri
D 1i .
Now let D Š Z` be the decomposition group in Gal.LNQ =N / Š Z`  Z` of an
extension of R1 and NQ 0 WD .LNQ /D . Then the extensions of prime divisors r 2 T0
split completely in LNQ =N and hence a fortiori in NQ 0 =N , in particular r satisfies the
Scholz condition (1). By construction
 Q of NQ 0 the same holds for r1 . For the remaining
ri 2 T1 the projections of LN Ri
=N
onto Gal.LN=N / Š Gal.L=K/ respectively
Q e e
Gal.N =N / are given by the pairs .i ; i / D .1i ; 1 i /. But the automorphism of
Q Q 0
LN =N belonging to .1 ; 1 / is trivial on N , so the extensions of the ri for i D
2; : : : ; k split in NQ 0 =N and thus satisfy the Scholz condition (1). Finally it remains
to state that the remaining ramification divisor q satisfies the Scholz conditions (0),
(1) and (2) since q … S, N .q/ 1 .mod `m / and q is unramified in NQ =K.
As in the proof of Lemma 10.6 there now exists an epimorphism 'Q 0 W K !
Gal.NQ 0 =K/ which on N coincides with 'Q and hence with '. This gives a proper `m -
Scholz solution of E .'; / with respect to S1 , unramified outside of Ram.NQ =K/ [
fqg. If E .'; / is geometric, then this solution is geometric as well, being the solu-
tion to a Frattini embedding problem (by Proposition 1.8). t
u
Remark. If in Lemma 10.6 the denominator divisor p1 of .t/ splits completely in
N=K, then by construction of L this also holds in NQ =K.
In conclusion we obtain (see also Serre (1992), Thm. 2.1.3):
Theorem 10.8. Let K D Q respectively K D IFp .t/ with p ¤ ` and ` … K, S1 D
f.`/g for K D Q resp. S1 D ; otherwise, and S a finite subset of IP.K/ containing
S0 . Then we have:
(a) Every central (geometric) `m -Scholz embedding problem E .'; / relative to
S1 over K with j'. K /j D `n < `m with kernel Z` possesses a proper (geometric)
`m -Scholz solution relative to S1 , with a solution field NQ satisfying

Ram.NQ =K/  Ram.N=K/ [ S1 [ fqg for some q 2 IP.K/ n S:

(b) If the socle of N=K is ramified outside of S, then E .'; / also possesses a
solution with this additional property.
10 Nilpotent Galois Groups over Global Fields 379

Proof. Assertion (a) follows from Lemma 10.5(a) and Lemma 10.7. In the case of
a split embedding problem the second part is a consequence of Lemma 10.5(b), and
in the non-split case it follows from the fact that in a Frattini embedding problem
the socle remains the same. Indeed, otherwise NQ 0 =K would contain a Z` -extension
Q
L=K with LQ 6 N , in contradiction to the fact that Gal.NQ 0 =N / is contained in the
Frattini subgroup ˚.G/ Q of G.Q t
u

10.3 The Theorem of Scholz and Reichardt

With the preparations in the previous sections it is now easy to prove the general ver-
sion of the Theorem of Scholz-Reichardt for global fields (compare Scholz (1937),
Reichardt (1937) for number fields and Rzedowski-Calderon (1989), Geyer and
Jarden (1998) for function fields).

Theorem 10.9. Let K be a global field of characteristic p  0. Then every finite

nilpotent group G with 26 jjGj in the case p D 0 resp. gcd.jGj; p  1/ D 1 in the
case p > 0 occurs as Galois group over K.

Proof. A nilpotent group is the direct product of its `-Sylow subgroups, so by The-
orem 1.6(a) it suffices to realize finite `-groups for `jjGj as Galois groups over K.
Moreover, by Theorem 8.3(b) we may assume that ` ¤ p.
Now let K0 D Q respectively K0 D IFp .t/ for some t 2 K and S be the set of
prime divisors of K0 ramified in K=K0 . By our assumptions on ` we have ` … K0 .
By induction it then follows from Theorem 10.8 that for every finite `-group G`
there exists a Galois extension N` =K0 with Galois group G` whose socle S.N` /
is unramified outside S, and which is geometric in the function field case. Now,
S.K \N` / K \S.N` / so the socle S.K \N` / is unramified over K0 . This implies
that S.K \ N` / D K0 , and hence also K \ N` D K0 since N` =K is an `-extension.
So K and N` are linearly disjoint over K0 , and the composite N` K=K has Galois
group isomorphic to G` (and is geometric in the function field case). t
u

In the case of global fields of characteristic 0 Šafarevič (1954a, 1989) was able
to remove the assumption 26 jjGj in Theorem 10.9 in a complicated induction pro-
cess by stepwise shrinking the embedding obstructions. With a generalization of the
method of proof he then succeeded to solve split embedding problems with nilpo-
tent kernel and thus realize all finite solvable groups as Galois groups over finite
number fields (see Šafarevič (1954b,c,d, 1958)). Revised proofs can also be found
in the monographs Ishkhanov, Lure and Faddeev (1997) and Neukirch, Schmidt
and Wingberg (2000).
In contrast to this it is possible in the case of global fields of positive character-
istic p to remove the assumption gcd.jGj; p  1/ D 1 by an elementary method of
Madan, Rzedowski-Calderon and Villa-Salvador (1996) which proceeds by varia-
tion of the base field.
380 IV Embedding Problems

10.4 Nilpotent Galois Groups over Global Function Fields

In this section let k be a finite field with p r elements, K WD k.t/ and ` 2 IP with
ord` .p r  1/ D e  1. Furthermore, let G be an `-group of order `n with Frattini
subgroup ˚.G/ D G 0 G ` of order `m and Frattini factor group G=˚.G/ Š Z`s with
s D n  m. The ground field K will now first be replaced by F WD k.u/ with u D
t ` . After extension of constants with kQ WD k. `m / the extension of KQ WD kK
Q over
m

Q Q Q
F WD kF is cyclic of order ` and K=F is Galois with group
m

Q / Š Z`m Z`f
Gal.K=F with f < m:

As in the previous section we denote by F .q/ the cyclotomic extension for a prime
polynomial q 2 kŒu with numerator divisor q 2 IP.F=k/. With these notations we
have:

Lemma 10.10. Let S be a finite subset of IP.F=k/. Then there exist infinitely many
families T WD fq1 ; : : : ; qs g of pairwise distinct prime divisors qi 2 IP.F=k/ n S with
(1) deg.qi / D di with d1 < d2 : : : < ds ,
(2) `nCe j.p rdi  1/ respectively `nCe 2 Fqi for 1 i s,
(3) qi is inert in K=F for 1 i s,
(4) qi splits completely in Fi1 =F for 1 < i s, where Fi1 is the composite of
p̀
the F .qj / . q j / for 1 j i  1.

Proof. We first note that KQ and FQi1 WD Fi1 . `nCe / are Galois over F and linearly
disjoint over FQ . Thus by the Dirichlet Density Theorem there exist infinitely many
qi 2 IP.F=k/ which split completely in FQi1 =F and stay inert in K= Q FQ . For these
we obtain (2) and (4). Now let q 2 IP.F=k/ be a prime divisor whose extensions
qQ 2 IP.K= Q are inert in K=
Q k/ Q FQ . Then the decomposition group in Gal.K=FQ / satisfies
D.Qq=q/ D Gal.K= Q FQ / because f < m, so (3) follows. It is now easy to obtain from
these infinitely many qi 2 IP.F=k/ infinitely many families T D fq1 ; : : : ; qs g with
pairwise distinct qi 2 IP.F=k/ n S. t
u

After these preparations we come to the main point of the proof:

Proposition 10.11. Let k be a finite field of characteristic p with ` 2 k, K WD k.t/

and G a finite `-group of rank s. Then for each T as in Lemma 10.10 there exists a
geometric Galois extension N=K with

Gal.N=K/ Š G and Ram.N=K/ D fQq1 ; : : : ; qQ s g; (10.7)

where qQ i 2 IP.K=k/ are the uniquely determined extensions of qi 2 T.

Proof. For the proof we have to perform the three steps in Lemmata 10.5–10.7 in
suitably modified form. First we construct an `nCe -Scholz-extension F0 =F with
group Gal.F0 =F / Š G=˚.G/. For qi 2 T with associated prime element qi let
.q / .q /
FC i =F denote the corresponding ray class field modulo qi . Since ŒFC i W F D
10 Nilpotent Galois Groups over Global Fields 381

.q /
.p rdi  1/=.p r  1/ the extension FC i =F possesses a cyclic intermediate field
Li =F of degree `. The composite F0 of the fields L1 ; : : : ; Ls is then a geometric
Galois extension over F with

Gal.F0 =F / Š Z`s Š G=˚.G/:

By the choice of T each qi splits completely in Lj =F for j < i and ramifies in

p̀
Li =F . For j > i the divisor qj splits completely in Li . qi /=F , hence by the
proof of Lemma 10.5 it follows that qi splits completely in Lj =F . Thus F0 =F is a
geometric `nCe -Scholz extension.
The group G possesses a lower central series

G F G0 D ˚.G/ F G1 F : : : F Gm D 1

with Gi 1 =Gi Š Z` for 1 i m. By Proposition 10.3 the central Frattini embed-

ding problem

1 ! Z` D G0 =G1 ! G=G1 ! G=G0 D Gal.F=F0 / ! 1

with kernel Z` has a proper geometric solution with solution field FQ0 , say. As in the
proof of Lemma 10.6 any divisor q 2 Ram.FQ0 =F / n T different from p1 satisfies
N .q/ 1 .mod `/. Hence the full cyclotomic extension F .q/ =F (the ray class field
modulo qp1 ) possesses a cyclic intermediate field L of degree ` unramified outside
q and p1 . But L=F and FQ0 =F are linearly disjoint, so Gal.LFQ0 =F / Š G0  Z` .
Now let F1 be the inertia field of an extension Q 2 IP.LFQ0 =F / of q. Then as in the
proof of Lemma 10.6 we obtain Gal.F1 =F / Š G=G1 and Ram.F1 =F / n fp1 g 
Ram.FQ0 =F / n fqg. After finitely many induction steps we thus arrive at a Galois
extension F1 =F with

Gal.F1 =F / Š G=G1 and Ram.F1 =F /  T [ fp1 g:

Now there might still exist qi 2 T inert in F1 =F0 , or p1 might ramify. These
problems can be solved in the function field case by a translation technique. For this
p̀ m1
let v WD u D t ` . Then F .v/ and F1 are linearly disjoint over F and hence we
have
Gal.F1 .v/=F / Š G=G1  Z` :
By construction qi 2 T is inert in F .v/=F and satisfies the Scholz conditions (10.3)
in F0 =F . Thus the extensions Qi 2 IP.F0 =k/ of qi are inert in F0 .v/=F0 . Since in
the case of global fields the decomposition groups modulo inertia groups are cyclic,
the Qi have residue degree 1 in F1 .v/=F0 .v/. Hence the uniquely determined exten-
sions qQ i 2 IP.F .v/=k/ of qi satisfy the Scholz conditions (10.3) for F1 .v/=F .v/.
Finally, p1 is ramified in F .v/=F but unramified in F0 =F . Thus the extensions
of p1 are ramified in F0 .v/=F0 and unramified in F0 .v/=F .v/. Since the inertia
groups are cyclic, these extensions are also unramified in F1 .v/=F0 .v/. Thus the
unique extension pQ 1 2 IP.F .v/=k/ of p1 is unramified in F1 .v/=F .v/. This proves
that F1 .v/=F .v/ is an `nCe -Scholz extension with group G=G1 .
382 IV Embedding Problems

After m such induction steps we finally obtain a geometric Galois extension

Fm =K with Galois group G=Gm Š G. t
u

From Theorem 10.9 and Proposition 10.11 we now obtain:

Theorem 10.12 (Madan et al. (1996)). Over a global field K=k of positive charac-
teristic every finite nilpotent group occurs as Galois group of a geometric Galois
extension.

Proof. We decompose the nilpotent group G into the direct product of its `-Sylow
subgroups G` . By Theorem 10.9 for each `6 jjk  j there exists a geometric Galois
extension N` =K with Gal.N` =K/ Š G` . For ` dividing jk  j we choose a rational
subfield k.t/ of K and construct by Proposition 10.11 a Galois extension N`0 =k.t/
with Gal.N`0 =k.t// Š G` and unramified in S WD Ram.K=k.t//. Since then N`0 and
K are linearly disjoint over k.t/, the composite N` WD N`0 K is a geometric Galois
extension of K with group Gal.N` =K/ Š G` . t
u

Remark. In Madan, Rzedowski-Calderon and Villa-Salvador (1996) the following

extension of Theorem 10.12 is shown: If s is the maximal rank of the `-Sylow sub-
groups G` of the nilpotent group G then there exists a geometric Galois extension
N=K with Gal.N=K/ Š G and j Ram.N=K/j s. This is obtained by a general-
ization of Proposition 10.11 to the case ` … k together with the construction of an
element t 2 K and a geometric Galois extension N=k.t/ with Gal.N=k.t// Š G,
for which all q 2 Ram.N=k.t// are inert in K=k.t/.

As an immediate consequence of Theorem 10.12 we obtain:

Corollary 10.13. Over a finite field every finite nilpotent group possesses a G-
realization.

The theorem of Šafarevič implies the corresponding result for all finite solvable
groups.
V Additive Polynomials

In this chapter we construct polynomials with prescribed Galois group in positive

characterisic. The new underlying phenomenon here is the fact that all finite Galois
extensions can be generated by finite vector spaces of roots of suitable polynomials.
Polynomials having such a vector space of zeroes over IFq are called q-additive.
Their Galois groups possess in a natural way a matrix representation into a gen-
eral linear group over IFq . The Galois theory of q-additive polynomials can thus be
called modular Galois theory.
Finite linear groups also occur as Galois groups in the theory of difference mod-
ules with respect to the Frobenius operator, which will here be called Frobenius
modules for short. This connection provides a non-trivial upper bound for the Galois
group of a finite Galois extension. Good lower bounds can then be obtained in anal-
ogy to the Dedekind Criterion. If the two bounds agree — and this will usually hap-
pen in our applications — the cumbersome algorithmic computation of the Galois
group can be avoided. This will be explained in Paragraphs 1 and 2.
In Paragraphs 3 and 4 we will start from a Steinberg cross section matrix to con-
struct Frobenius modules and additive polynomials with Galois group over IFq .t/ a
given finite group of Lie type G.IFq /, in particular for all infinite series of classical
groups (sometimes only over fields of odd characteristic). In the final Paragraph 5
we then demonstrate how to use field restriction to obtain Galois extensions with
the same Galois group over smaller fields of constants and in particular over IFp .t/.
The contents of this chapter was developped, on the basis of Matzat (2003) and
Malle (2003) in the diploma theses of Garcia Lopez, Albert, Maier and Stichel
and first published in Garcia Lopez (2010), Albert and Maier (2011) and Stichel
(2014).
A non-constructive proof for the existence of G-realizations over IFq of the IFq -
points of semisimple simply connected linear algebraic groups defined over IFq had
earlier been found by Nori (1994), see also Gille (2000). Some more special poly-
nomials for various series of nearly simple classical groups had been constructed

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 383

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_5
384 V Additive Polynomials

before by Abhyankar and collaborators. These are mainly collected in a series of

papers on “nice polynomials for nice groups” (see among others Abhyankar (1994,
1996a,b), Abhyankar and Loomis (1998, 1999), and Abhyankar and Inglis (2001)).
We give hints on these at the end of the corresponding sections.
1 Frobenius Modules 385

1 Frobenius Modules

This first paragraph contains a short introduction to the theory of (ordinary) Frobe-
nius modules and their connection to Galois extensions in positive characteristic as
well as to additive polynomials. The presentation follows the first parts of Matzat
(2003). As Frobenius modules are special types of difference modules, farther reach-
ing information can for example be found in the monograph of van der Put and
Singer (1997).

1.1 Ordinary Frobenius Modules

Throughout this chapter K will denote a field of positive characteristic char.K/ D

p 2 IP, and q D p e denotes a power of p. Then the map

 WD q W K ! K; a 7! aq ;

is called the (ordinary) Frobenius endomorphism, and the pair .K; q / a Frobenius
field or F-field for short. The fixed field K  of  is then a subfield of IFq .
A pair .M; ˚/ consisting of a K-vector space M and a -semilinear map ˚ W
M ! M , that is, a map with the properties

˚.x C y/ D ˚.x/ C ˚.y/ and ˚.ax/ D .a/˚.x/ (1.1)

for all x; y 2 M and a 2 K is called an (ordinary) Frobenius module, or F-module

for short, over .K; /. It is called a trivial F-module if M has a ˚-invariant basis.
Now let L=K be an F-field extension with Frobenius endomorphism L , and
ML WD L ˝ M with the Frobenius endomorphism ˚L WD L ˝ ˚ the extended F-
module. Then we call

L .M / WD fx 2 ML j ˚L .x/ D xg
Sol˚

the solution space of M over L. If .ML ; ˚L / is trivial in the above sense, then
.L; L / is called a solution field of M . We will usually suppress the index and write
; ˚ in place of L and ˚L .
Now assume that M has finite dimension n over K and let B D fb1 ; : : : ; bn g
denote an ordered K-basis. Then there exists a representing matrix D D DB .˚/ 2
K nn of ˚ with ˚.B/ D B  D (where B is considered as a row vector). In the case
that D 2 GLn .K/ we call .M; ˚/ dualizable. Then the representing matrix of the
dual Frobenius endomorphism ˚  on the dual space M  with respect to the dual
basis B  is given by
DB  .˚  / D .D 1 /t : (1.2)
Finally, .M; ˚/ is called a cyclic F-module if there exists a ˚-cyclic element x in
M , i.e., if the set f˚ i .x/ j i 2 INg contains a basis of M .
386 V Additive Polynomials

Remark. If .M; ˚/ is an F-module over .K; / with basis B D fb1 ; : : : ; bn g and if

BQ WD B  C with C 2 GLn .K/ is a further basis of M , then the representing matrices
DB .˚/ and DBQ .˚/ satisfy

DBQ .˚/ D C 1 DB .˚/.C /: (1.3)

Then DBQ .˚/ is called a -equivalent matrix to DB .˚/.

Proposition 1.1. Let .M; ˚/ be an F-module over an F-field .K; / with
dimK .M / D n. Then for all F-extension fields .L; L /  .K; / with fixed field
k WD LL the solution space Sol˚ L .M / is a k-vector space with
dimk .Sol˚
L .M // n.
Proof. Obviously W WD Sol˚ L .M / is a k-vector space. Assume dimk .W / > n. Then
there exists a minimal set of vectors S D fx1 ; : : : ; xm g  ML of cardinality m > 0
which is linearly independent over k and linearly dependent over L. By minimality
x1 ; : : : ; xm1 are linearly independent over L. Thus

X
m1
xm D ai xi with ai 2 L
i D1

and not all ai in k. But then the ˚-image satisfies

X
m1
xm D .ai /xi D 0:
i D1

Subtracting the second equation from the first shows that the L-linearly independent
subset fx1 ; : : : ; xm1 g of S is linearly dependent over L, leading to a contradiction.
t
u
The next result shows that every dualizable F-module over K possesses a mini-
mal solution field which defines a finite Galois extension of K.
Theorem 1.2. Let .M; ˚/ be a dualizable F-module over an F-field .K; /. Then:
(a) There exists an F-extension field .N; /  .K; / of .K; / with dimension
dimk .Sol˚N .M // D n D dimK .M /, where k D N .


(b) If N in (a) is minimal then N=K is a finite Galois extension (and determined
uniquely inside an algebraic closure KN of K).
Proof. Let  D q . Then we have L IFq for every F-extension field .L; / 
.K; / and hence .y/ D y for y 2 L . Now let B D fb1 ; : : : ; bn g be a basis of M
over K and D WD DB .˚/ 2 GL Pn .K/ the representing matrix of ˚ with respect to
B. Then every solution vector niD1 bi yi 2 Sol˚
L .M / with yi 2 L satisfies

By D ˚.By/ D ˚.B/.y/ D B  D  .y/;

where y D .y1 ; : : : ; yn /t and .y/ D .y1q ; : : : ; ynq /t DW yq . This yields an algebraic

system of equations y  Dyq D 0 for y1 ; : : : ; yn .
1 Frobenius Modules 387

Now let KN be an algebraic closure of K with the Frobenius endomorphism N D

q . Then the solution set S in KN n of the above system of equations y  Dyq D 0 is
N
finite by Proposition 1.1. As the projective variant of this algebraic system of equa-
tions has no solutions on the infinite hyperplane, the theorem of Bezout in IPn .K/ N
n
shows that, counting multiplicities, there exist exactly q solutions in K (see N n

Šafarevič (1994), Ch. IV.2.1). The Jacobian matrix of y  Dyq D 0 is the identity
matrix In . Thus all solutions are regular points on S and have multiplicity one (see
Iitaka (1982), 2.6). Hence S consists of exactly q n simple solutions y1 ; : : : ; yq n ,
which by Proposition 1.1 form an IFq -vector space. The field N WD K.yij / gener-
ated by the components yij of all vectors yi D .yi1 ; : : : ; yi n /t 2 KN n , 1 i q n ,
together with the restriction N WD N q jN is then a solution field of .M; ˚/, and we
have IFq N .
Now obviously N is minimal inside KN with these properties and according to
the Remark before Proposition 1.1 independent of the choice of basis. Since all
solutions of the algebraic system of equations y  Dyq D 0 in KN n are simple, N=K
is moreover separable and normal, hence Galois. t
u
A linearly independent system of solutions y1 ; : : : ; yn 2 N n of Dyq D y is called
a fundamental system of solutions, and the matrix Y WD .y1 ; : : : ; yn / is a fundamental
solution matrix. It is characterized by the property that N .Y / D D 1 Y . With this
we obtain the following characterization of solutions fields:
Corollary 1.3. Let .K; / be an F-field with K  D IFq . Then .N; / is a minimal
solution field of a dualizable F-module .M; ˚/ over .K; / if and only if:
(1) there exists a matrix Y D .yij /ni;j D1 2 GLn .N / with N .Y / D D 1 Y , and
(2) we have N D K.Y /, i.e., N=K is generated by the entries yij of Y .
Such a characterization of minimal solution fields is well-known from the Galois
theory of linear differential and difference equations (see e.g. van der Put and Singer
(1997)).

1.2 Cyclic Frobenius Modules

If .K; / is a Frobenius field with  D q then the matrix

0 1
´1 : : : ´n
B ´q1 : : : ´qn C
B C
q .´1 ; : : : ; ´n / WD B :: :: C with ´1 ; : : : ; ´n 2 K;
@ : : A
n1 n1
´q1 : : : ´qn

is called Moore matrix, and its determinant det.q .´1 ; : : : ; ´n // the Moore deter-
minant of ´1 ; : : : ; ´n . It is well-known that ´1 ; : : : ; ´n are linearly independent over
k WD K  if and only if det.q .´1 ; : : : ; ´n // ¤ 0 (see Goss (1996), Cor. 1.3.4). In
the theory of Frobenius modules the Moore matrix hence plays the role of the Van-
dermonde matrix. In particular we have (see Goss (1996), Lemma 1.3.6):
388 V Additive Polynomials

Proposition 1.4. Let .K; / be an F-field with  D q , and ´1 ; : : : ; ´n 2 K linearly

independent over k WD K  . Then the polynomial

f .X / WD det.q .´1 ; : : : ; ´n ; X //= det.q .´1 ; : : : ; ´n // 2 KŒX

satisfies Y
f .X / D .X  ´/ with S D IFq h´1 ; : : : ; ´n i:
´2S
Polynomials of this kind with an IFq -vector space as solution set are called q-
additive polynomials (resp. IFq -linear polynomials in Goss (1996), or q-polynomials
in Smith (1995)). They have the special form

X
n
i
f .X / D ai X q
i D0

(compare Goss (1996), Cor. 1.2.2). Here, f .X / is separable if and only if a0 ¤ 0,

i.e., if the solution space S has dimension n over IFq .
Now we can answer the question on sufficient conditions to guarantee that an
F-module possesses cyclic elements.
Theorem 1.5. Let .M; ˚/ be a dualizable F-module over an F-field .K; / with
n
 D q such that dimK .M / D n. If jKj > q2 then M is a cyclic F-module.
Proof. In a first step we choose a basis B D fb1 ; : : : ; bn g of M and consider the
corresponding representing matrix D WD DB .˚/. Further we let N=K be a minimal
solution field of .M; ˚/ with fundamental matrix Y D .yij /ni;j D1 D .y1 ; : : : ; yn / 2
GLn .N /. Then we have .Y / D D 1 Y by Corollary 1.3. Now let Q denote the
q q
ideal in KŒX1 ; : : : ; Xn generated by .X1 ; : : : ; Xn /t  D 1 .X1 ; : : : ; Xn /t . Then by
Theorem 1.2 the factor ring R WD KŒX1 ; : : : ; Xn =Q has dimension dimK .R/ D q n .
n
If K has more than q2 elements, by Kreuzer and P Robbiano (2000), Prop. 3.7.22,
there exist elements ci 2 K such that X WD X1  niD2 ci Xi is in normal 1-position,
i.e., so that the first components of all solutions are pairwise distinct (see Kreuzer
and Robbiano (2000), Def. 3.7.21). This transformation corresponds to the base
change
X
n
B 7! BQ WD .b1 C ci bi ; b2 ; : : : ; bn / WD B  C;
i D2

and the corresponding fundamental solution matrix is given by YQ WD .Qy1 ; : : : ; yQ n / D

C 1 Y .
In a second step we apply the Buchberger algorithm with respect to the anti-
lexicographical ordering to the system of equations in X; X2 ; : : : ; Xn given by Q.
Since the first components of the solutions are pairwise distinct we then obtain a
Gröbner basis of the form
n
X
q
f .X / D aQ i X i and Xj D fj .X / 2 KŒX for j D 2; : : : ; n
i D0
1 Frobenius Modules 389

with suitable aQ i 2 K. By Theorem 1.2 the solution space of f .X / is given by the

IFq -vector space SQ of the first components of the solutions yQ i D .Q yQi1 ; : : : ; yQi n /t , 1
Q 1 Q
i n, of yQ D D yQ with D D DBQ .˚/. Thus we obtain f .X / D Q́2SQ .X  Q́ /, and
q
P i
moreover f .X / is q-additive and separable, say f .X / D niD0 ai X q with a0 ¤ 0.
Every basis f´1 ; : : : ; ´n g of SQ can now be completed via Q́ ij WD fj .´i / to a basis
Q́ 1 D .´1 ; Q́ 12 ; : : : ; Q́ 1n /; : : : ; Q́ n D .´n ; Q́ n2 ; : : : ; Q́ nn / of Sol˚
N .M /. Thus Q́ 1 ; : : : ; Q́ n
are linear combinations of yQ1 ; : : : ; yQn . So there exists a matrix C 0 2 GLn .IFq / with
ZQ WD . Q́ 1 ; : : : ; Q́ n / D YQ C 0 , and ZQ is by definition also a fundamental solution matrix
of .M; ˚/, with respect to B. Q
In the third step we show first that there exists a matrix CQ 2 GLn .K/ with Z WD
q .´1 ; : : : ; ´n / D CQ  Z. Q For the first row of this matrix the choice cQ1j D ı1j will
obviously do. Since
Q D .CQ /DQ 1 ZQ
.Z/ D .CQ /.Z/

we can take the first row of DQ 1 for the second row of CQ . By induction we see that
the i th row of CQ is given by the first row of  i 1 .DQ 1 /    DQ 1 . This proves the
existence of a matrix CQ 2 K nn with Z D CQ Z. Q Moreover this is invertible due to
Q N Q Q
det.Z/ ¤ 0. By the base change B 7! B WD B C 1 the Moore matrix Z becomes the
fundamental solution matrix of .M; ˚/ with respect to the basis BN D .bN1 ; : : : ; bNn /.
In particular we have

.Z/ D DQ 1 Z and N D BN  DN
˚.B/

with
0 1 0 1
0 1  aa10 : : :  an1
a0
 a10
B :: C B 1 C
B C B C
DN 1 D B : C respectively DN D B :: C:
@ 1 A @ : A
a0 a1 : : : an1 1 0

From this we obtain

a1 N an1 N
˚.bN1 / D bN2  b1 ; : : : ; ˚.bNn1 / D bNn  b1
a0 an

and thus the required ˚-cyclicity of bN1 . t

u
By Theorem 1.5 in particular all F-modules over an algebraic function field K
are cyclic since there jKj D 1. The monic q-additive polynomial

det.q .´1 ; : : : ; ´n ; X // Y
f .X / D D .X  Q́ /
det.q .´1 ; : : : ; ´n //
Q́2SQ

constructed in the proof, whose zeroes are precisely the vector space consisting of
the first components of the solution space, will be called the characteristic polyno-
mial of the F-module. According to the proof of Theorem 1.5 it satisfies:
390 V Additive Polynomials

Corollary 1.6. Let .M; ˚/ be a dualizable F-module over .K; / and f .X / the
characteristic polynomial of .M; ˚/ (with respect to a basis in normal 1-position).
Then the splitting field N=K of f .X / is the minimal solution field of .M; ˚/ (inside
N
K).
Using the representing matrix DN D DBN .˚/ in the proof of Theorem 1.5 one
sees that every q-additive polynomial f .X / occurs as the characteristic polynomial
of a Frobenius module. In this situation we also write Df in place of D, N and the
corresponding F-module .M; ˚/ is called the F-module associated to f .X / and
denoted .Mf ; ˚/. The inverse Af WD Df1 of DN D Df is the companion matrix of
f .X /.

1.3 Galois Groups

In this subsection we attach to every dualizable F-module a linear group as Galois

group and derive an important upper bound for these.
Proposition 1.7. Let .M; ˚/ be a dualizable F-module over .K; / with solution
field N and Y; YQ 2 GLn .N / two fundamental solution matrices of .M; ˚/ with
respect to the same basis. Then there exists a matrix C 2 GLn .N  / with YQ D Y  C .
Proof. Let  D q , so N  D IFq . Since the columns of Y as well as of YQ consist
of the coordinate vectors of a basis of the solution space Sol˚ N .M /, which is an
IFq -vector space, the matrix C is just the corresponding base change matrix on the
coordinate vector space. t
u
Corollary 1.8. Let .K; / be an F-field and .N; / a minimal solution field of an
F-module over K as in Proposition 1.7. Then the Galois group Gal.N=K/ is iso-
morphic to a subgroup of GLn .N  /.
Proof. Let Y 2 GLn .N / be a fundamental solution matrix of the F-module .M; ˚/
(with respect to a basis B of M ). Then for  2 Gal.N=K/, .Y / is again a fun-
damental solution matrix. By Proposition 1.7 there hence exists a matrix C 2
GLn .N  / with .Y / D Y  C . This induces a faithful representation

W Gal.N=K/ ! GLn .N  /;  7! C : t
u

The representation of Gal.N=K/ into GLn .q/ is also called the Galois group
of the F-module .M; ˚/ (with respect to a given basis of M ) and denoted by
Gal.M; ˚/.
Remark. The Galois group of .M; ˚/ restricts, via

.´1 ; : : : ; ´n / WD .´1 ; : : : ; ´n /C for ´i D y1i ;

to the solution space V of the first solution components and so directly furnishes a
faithful permutation representation on the zeroes of the characteristic polynomial.
1 Frobenius Modules 391

In the next theorem we use the language of linear algebraic groups, see e.g.
Springer (1998), Ch. 2.1. It gives an extremely useful upper bound for the Galois
group Gal.M; ˚/ of a Frobenius module.

Theorem 1.9 (Upper Bound Theorem). Let .M; ˚/ be a dualizable F-module over
the F-field .K; / with  D q and K  D IFq . Further let G be a connected linear
algebraic group defined over IFq . If there exists a basis B of M with DB .˚/ 2
G.K/, then the Galois group of the F-module satisfies

Gal.M; ˚/ G.IFq /:

Proof. In the proof we utilize the modified Lang isogeny

 W G.K sep/ ! G.K sep/; .xij / 7! q ..xij //.xij /1 ;

over the separable closure K sep of K. As G is a connected linear algebraic group, 

is surjective by the Theorem II.1.1 of Lang–Steinberg. So there exists a matrix Y D
.yij / 2 G.K sep/ with .Y / D D 1 for D D DB .˚/ 2 G.K/, respectively Y q D
D 1  Y . That is, Y is a fundamental solution matrix of .M; ˚/. By Corollary 1.3
the field N D K.Y / is a minimal solution field of .M; ˚/. Since .Y / D Y  C for
 2 Gal.N=K/ we have that C not only lies in GLn .q/, but also in G.N /. Thus
C 2 G.IFq / for all  2 Gal.N=K/. t
u

The existence of such a non-trivial upper bound for the Galois group is special to
positive characteristic and does not have an analogue in characteristic zero.
In the following examples we assume that K D IFp .t/ is a rational function field
in one variable and  D p .

Example 1.1. The simplest example of a connected linear algebraic group is the
multiplicative group Gm with Gm .K/ D K  . We let .M; ˚/ be a 1-dimensional F-
module with DB .˚/ D .t 1 / 2 Gm .K/. Then by Corollary 1.8 the Galois group of
.M; ˚/ is a subgroup of Gm .IFp / D IF
p . We obtain the characteristic polynomial
from X p D .X / D tX as

f .X / D X p  tX D X.X p1  t/ 2 KŒX ;

where the second factor X p1  t 2 KŒX is irreducible. Thus the minimal solution
field N has degree p  1 over K, and Gal.M; ˚/ D IFp. t
u

Example 1.2. Now consider the additive

 group Ga and let .M; ˚/ be a 2-dimensio-
1 t
nal F-module over K with DB .˚/ D 2 Ga .K/. Then by Theorem 1.9 the
0 1
Galois group of .M; ˚/ is a subgroup of Ga .IFp / Š .IFp ; C/. From
!   !
X1 1t X1
˚ D
X2 0 1 X2
392 V Additive Polynomials

we obtain that X1p D X1 C tX2 and X2p D X2 , so for X D X1 the characteristic

polynomial is
2
f .X / D X p  .t p1 C 1/X p C t p1 X:
This factorises as Y
f .X / D .X p  X  ct/
c2IFp

into a product of irreducible Artin–Schreier polynomials of degree p. So the Galois

group of .M; ˚/ contains at least p elements, and we deduce that Gal.M; ˚/ D IFp .
t
u
G D SL
Example 1.3. As final example we consider the group  2 . For this let .M; ˚/
t 1
be a 2-dimensional F-module over K with DB .˚/ D 2 SL2 .K/. Thus the
1 0
Galois group G of .M; ˚/ is a subgroup of SL2 .p/. With X D X1 we obtain the
characteristic polynomial
2 2 1
f .X / D X p  tX p C X D X.X p  tX p1 C 1/ 2 KŒX :

As the second factor is irreducible, p 2 1 must divide jGj. Now we specialize t 7! 2

and X p1 7! Y . Then the second factor becomes

Y pC1  2Y C 1 D .Y  1/.Y p C : : : C Y  1/;

where again the second factor is irreducible. By the Dedekind Criterion (Corol-
lary I.9.3) the solution field N then contains an intermediate field L=K of degree p.
So jGj is even divisible by p.p 2  1/, which implies that G D SL2 .p/. t
u

1.4 Effective Frobenius Modules

In this section we are concerned with the question under which conditions on a
Frobenius module .M; ˚/ we are in the favorable situation that the containment
of DB .˚/ in a certain linear algebraic group G.K/ already determines the Galois
group, i.e., the upper bound in Theorem 1.9 is attained.
For simplification we introduce the following definition: a dualizable F-module
.M; ˚/ over .K; / is called effective if there exists a linear algebraic group G
defined over K  and a basis B of M such that

DB .˚/ 2 G.K/ and Gal.M; ˚/ Š G.K  /:

Obviously the three F-modules given in the examples at the end of the previous
section are effective F-modules.
Proposition 1.10. Let .K; / be an F-field with K  IFp and .M; ˚/ a dualizable
F-module with basis B. Further let G be a linear algebraic group over K with
1 Frobenius Modules 393

DB .˚/ 2 G.K/. Then there exists a matrix C 2 GLn .IFp / such that with BQ D BC
the matrix DBQ .˚/ lies in the connected component Gı .K/ of G.

Proof. By a theorem of Borel and Serre the connected component Gı of G has a

finite supplement H in G, i.e., H is a finite linear algebraic group over IFp (see
for example Wehrfritz (1973), Lemma 10.10). So D D DB .˚/ can be factorized
into the product D D D0  C0 with D0 2 Gı .K/ and C0 2 H .IFp /. As C0 defines a
trivial F-module over IFp , there exists C 2 GLn .IFp / with C01 D .C /C 1 . Then
for the basis BQ WD B  C of M we obtain by (1.3)

DBQ .˚/ D C 1 DB .˚/.C / D C 1 D0 C:

Thus DBQ .˚/ is contained in the connected component Gı .K/ of G. t

u

Corollary 1.11. If a Frobenius module .M; ˚/ with DB .˚/ 2 G.K/ is effective,

then the linear algebraic group G is connected.

Proof. Let .M; ˚/ be an effective F-module over .K; / with DB .˚/ 2 G.K/
and Gal.M; ˚/ D G.IFq /. By Proposition 1.10 we may assume without loss of
generality that DB .˚/ lies in the connected component Gı .IFq K/. So we have
Gal.M; ˚/ Gı .IFq / by Theorem 1.9, so by effectivity that G D Gı . t
u

There remains the question under which conditions F-modules are effective,
whose Galois group is the group of IFq -rational points of a connected linear alge-
braic group G? Does this always hold, for example, if K=IFp is an algebraic function
field in one variable? The answer seems to be unknown.
394 V Additive Polynomials

2 Computation of the Galois Group

In the first four sections we present an algorithm for the computation of the Galois
group of a q-additive polynomial in finitely many steps. Unfortunately the neces-
sary calculations can become quite cumbersome for larger Galois groups. For this
reason in the fifth section we derive lower bounds for the Galois group, which are
similar in flavor to the Dedekind Criterion (see Corollary I.9.3). Agreement of the
upper bound from Theorem 1.9 with the composite of all lower bounds then renders
the algorithmic determination of the Galois group superfluous. In the subsequent
sections this method will usually prove successful. The first four sections follow the
thesis of Garcia Lopez (2010), while the last one is based upon Matzat (2003) with
the multivariate version presented in Albert and Maier (2011).

2.1 An Invariant Theoretic Criterion

Let K be a (at first) arbitrary field and V a K-vector space with basis B D
fb1 ; : : : ; bn g. Then the symmetric algebra SymK .V / will be denoted by KŒV , and
its field of fractions by K.V /. It is well-known that KŒV is isomorphic to a poly-
nomial ring KŒt1 ; : : : ; tn in linearly independent forms t1 ; : : : ; tn on the dual vector
space V  , and we have K.V / Š K.t1 ; : : : ; tn /. In the following t1 ; : : : ; tn usually
denote the bidual basis of a given basis B in V  Š V  KŒV . In case L is a field
containing K, we write LŒV or LŒt1 ; : : : ; tn for L ˝K KŒV .
Now let G be a subgroup of the general linear group P GL.V / Š GLn .K/. Then
 2 G acts on the coordinates v D .v1 ; : : : ; vn /t of v D niD1 vi bi 2 V via v 7! C v
with a matrix C 2 GLn .K/. Consequently,  2 G also acts on the polynomi-
als f .t1 ; : : : ; tn / in KŒV Š KŒt1 ; : : : ; tn via .f .t1 ; : : : ; tn // D f ..t1 ; : : : ; tn /C /.
The K-algebra of G-invariant polynomials in KŒV is called the ring of invariants
KŒV G , and the field of G-invariant rational functions in K.V / the field of invari-
ants K.V /G .
In the standard literature on invariant theory (e.g. Smith (1995), Derksen and
Kemper (2002) etc.) usually KŒV stands for SymK .V  / with the inverse action
of G on V  . The use of KŒV WD SymK .V / in this monograph has the advantage
that the action of G on a basis b1 ; : : : ; bn of V and its bidual basis t1 ; : : : ; tn in
V  Š V  KŒV agree. Since statements on invariants do not depend on whether
the action of G is ordinary or inverse, the results from invariant theory cited in the
sequel hold for KŒV Š SymK .V / Š SymK .V  / as well as for SymK .V  /.
The first statements needed from invariant theory are collected in the following
Remarks. (1) If H is a subgroup of G GL.V / then we clearly have

KŒV H
 KŒV G
and K.V /H  K.V /G :
2 Computation of the Galois Group 395

(2) If G GL.V / is a finite group, then

K.V /G D Frac.KŒV G
/:

This is shown for example in Smith (1995), Prop. 1.2.4.

The best known example of an invariant ring is obtained from the natural per-
mutation representation of the symmetric group Sn as the group G GLn .K/ of
permutation matrices. Here KŒV G is the ring of symmetric polynomials over K.
This is itself a polynomial ring, generated over K for example by the elementary
symmetric polynomials

X Y
i
si WD tjk for 1 i n:
1j1 <<ji n kD1

An analogue for q-additive polynomials is given by the Dickson algebra. For this
let V be an n-dimensional vector space over K  IFq and G D GLn .q/. Then the
polynomial
Y
fq;n .X / WD .X  t/ with W WD IFq ht1 ; : : : ; tn i
t 2W

is q-additive by our observations in Paragraph 1. So it has the form

X
n
ni
fq;n .X / D .i /
dq;n Xq with .0/
dq;n D 1:
i D0

.i /
The non-trivial coefficients di WD dq;n , 1 i n, of fq;n .X / are G-invariant homo-
geneous polynomials in t1 ; : : : ; tn of degree q n  q ni . They are called Dickson
invariants. Obviously K.V / D K.t1 ; : : : ; tn / is the splitting field of fq;n .X / over
K.d/ D K.d1 ; : : : ; dn / K.V /G . So the d1 ; : : : ; dn are algebraically independent
over K and K.V /=K.d/ is Galois with Galois group H a subgroup of G. Since

Y
n
ŒK.V / W K.d/  .q n  q ni / D jGj
i D1

we even have H D G and the invariant field K.V /G is generated over K by the
Dickson invariants:
K.V /G D K.d1 ; : : : ; dn /:
Thus the roots of the Dickson polynomial

X
n
ni
fq;n .X / D di X q with d0 D 1 (2.1)
i D0
396 V Additive Polynomials

generate the Galois extension K.t/=K.d/ with group GLn .q/. (This indeed is a
generic polynomial in the sense of Jensen, Ledet and Yui (2002), compare Exam-
ple (1.1.2).)
By the Theorem of Dickson (see Smith (1995), Thm. 8.1.5) the polynomial
subring KŒd of KŒV G is already the full ring of invariants of G, i.e., we have

KŒV G
D KŒd1 ; : : : ; dn :

This is also called the Dickson algebra of degree n over K in this context. It consti-
tutes the starting point for the algorithm for the computation of the Galois group of
a q-additive polynomial that we will present here.
Now let K again be an arbitrary field, V an n-dimensional K-vector space and
H G GLn .K/ linear groups with finite index .G W H / D m. Then FH 2 K.V /H
is called a G-relative H -invariant if FH is a primitive element of the extension
K.V /H =K.V /G . If 1 ; : : : ; m is a system of representatives for the cosets of H in
G then
Y
m
ResF H
G=H .X / WD X  j .FH /
j D1

is called a G-relative resolvent for H . This will now be used in the case of q-additive
polynomials.
Proposition 2.1. Let K  IFq be a field, f .X / 2 KŒX a separable q-additive poly-
nomial and z WD .´1 ; : : : ; ´n / an ordered IFq -basis of the zero space V  K sep of
f .X /. Further let Gz WD Galz .f / be the representation of the Galois group of f .X /
with respect to the basis z with Gz G GLn .q/. If then H G is a subgroup of
G, FH 2 KŒV H a G-relative H -invariant and z W KŒV ! KN the specialization
homomorphism determined by ti 7! ´i , then

ResFH .z/ FH
G=H .X / WD z .ResG=H .X // 2 KŒX :

Proof. We write

X
m
ResF
G=H .X / D
H
ai .t/X i 2 KŒV G
ŒX :
i D0

Then for  2 Gal.f / with corresponding representing matrix C 2 Galz .f / we have

X
m X
m X
m
.ResFH .z/
G=H
.X // D .z .ai .t///X i D z .ai .t/C /X i D z .ai .t//X i :
i D0 i D0 i D0

So the elements z .ai .t// all lie in the fixed field N Gal.f / under Gal.f / of the split-
ting field N of f .X / and hence in K. t
u
The next theorem is a q-additive analogue of the Theorem of Stauduhar, which
plays a central role in the computation of Galois groups in characteristic zero (com-
pare also Stroth (1998), 17).
2 Computation of the Galois Group 397

Theorem 2.2. Assume the notations in Proposition 2.1. Furthermore assume that
the specialized G-relative H -resolvent ResF H .z/
G=H
.X / is separable. Then
FH .z/
ResG=H .X / has a zero in K if and only if Galz .f / is conjugate in G to a sub-
group of H .
Proof. We first assume that Galz .f / is G-conjugate to a subgroup of H , that is,
there exists  2 G with Galz .f / H 1 . Clearly .FH / is a G-relative H 1 -
invariant. Thus we have for all  2 Gal.f / that

.z .FH .t/// D z .FH .tC // D z ..FH //:

So, a WD z ..FH // is invariant under Gal.f / und thus is a zero of ResF H .z/
G=H
.X /
in K.
FH .z/
For the converse assume that ResG=H .X / has a zero a 2 K. Then there exists a
coset H 2 G=H with z ..FH // D a 2 K. We now assume that Galz .f / is not a
subgroup of H 1 . Then there exists C 2 Galz .f / n H 1 with ..FH // ¤
.FH /. Due to

z ...FH /// D .z ..FH /// D .a/ D a

F .z/
H
this gives a contradiction to the separability of ResG=H .X /. t
u
Our proof of Theorem 2.2 also leads to the following observation:
Remark. The assertion of Theorem 2.2 remains true in the inseparable case if
FH .z/
ResG=H .X / either has no zero, or has at least one simple zero in K.
We are now faced with the task to determine suitable G-relative H -invariants
FH 2 KŒV H .

2.2 Computation of Homogeneous Invariants

We first show how to reduce the computation of G-relative H -invariants FH 2

KŒV H to the case K D IFq .
Proposition 2.3. Let K  IFq be a field and H < G GLn .q/ subgroups of GLn .q/.
Then we have:
(a) There exists a primitive element FH of K.V /H =K.V /G with FH 2 IFq ŒV D
IFq Œt1 ; : : : ; tn .
(b) If H < G is maximal then there also exists a G-relative H -invariant FH 2
IFq ŒV that is homogeneous in t1 ; : : : ; tn .
Proof. Part (a) follows immediately from the translation theorem of Galois the-
ory starting from the corresponding field extension IFq .V /H =IFq .V /G . Part (b) is
obtained from that by decomposition into homogeneous components. t
u
398 V Additive Polynomials

According to Proposition 2.3 we may now restrict ourselves to the computation

of homogeneous G-relative H -invariants over IFq . By the Remark (1) in Section 2.1
we have IFq ŒV H  IFq ŒV G and thus IFq ŒV H d
 IFq ŒV Gd
for the corresponding
homogeneous components in degree d . From this we obtain immediately:

Corollary 2.4. Let G GLn .q/ and H < G a maximal subgroup. Then there exists
a homogeneous G-relative H -invariant FH of degree d if and only if

dimIFq .IFq ŒV H
d / ¤ dimIFq .IFq ŒV G
d /:

The computation of bases of IFq ŒV H G

d and IFq ŒV d is always possible with
algorithms from constructive invariant theory, for example with the linear algebra
method of Kemper described in Derksen and Kemper (2002), Ch. 3.3.1. If the index
.G W H / is prime to the characteristic of the underlying fields, then the homogeneous
G-relative H -invariants of degree d are just given by the elements which are not
fixed under the relative Reynolds operator
1 X
H G
RG=H W IFq ŒV ! IFq ŒV d; f 7! .f /:
d
.G W H /
2G=H

Starting with a basis of IFq ŒV H

d
these can also be determined in this simple fashion
(for this see Derksen and Kemper (2002), Algorithm 3.1.1).

2.3 Specialization of Relative Resolvents

In this section we describe an algorithm for the computation of the Galois group of a
q-additive polynomial. This works inductively along a chain of maximal subgroups.
For this we make the following induction hypothesis: Let K  IFq be a field, f .X / 2
KŒX a separable q-additive polynomial, V  K sep the vector space of zeroes of
f .X / and z D .´1 ; : : : ; ´n / an IFq -basis of V . We assume that Galz .f / G and
that H < G is a maximal subgroup of G of index .G W H / D m. Further let FH 2
IFq ŒV H be a G-relative H -invariant with (without loss of generality) z .FH / 2 K
(respectively Gz H ) and ResF H .z/
G=H .X / 2 KŒX .
With the symmetric bilinear form
X
ˇ W IFq .V /H  IFq .V /H ! IFq .V /G ; .g; h/ 7! .gh/;
2G=H

over IFq .V /G we introduce the symmetric matrix

0 1
ˇ.FH0 ; FH0 / : : : ˇ.FH0 ; FHm1 /
B :: :: C
ColFG=H WD @
H
: : A;
m1 0 m1 m1
ˇ.FH ; FH / : : : ˇ.FH ; FH /
2 Computation of the Galois Group 399

which we call the G-relative Colin matrix of FH in view of the precursor work of
Colin (1995) in characteristic zero. This enjoys the following property:

Proposition 2.5. Let FH 2 IFq ŒV H be a G-relative H -invariant and g 2 IFq ŒV H

arbitrary. Then there exist xj 2 IFq .V /G with

X
m1
j
gD xj FH :
j D0

Here, x0 ; : : : ; xm1 are solutions of the linear system of equations

.ˇ.FH0 ; g/; : : : ; ˇ.FHm1 ; g//t D ColFH

.x ; : : : ; xm1 /t :
G=H 0

Proof. The first part of the assertion follows from the fact that FH is a primitive ele-
ment of the extension IFq .V /H =IFq .V /G , the second from the linearity of ˇ.FHi ; g/
in the second argument:

X
m1
j
ˇ.FHi ; g/ D ˇ.FHi ; FH / xj :
j D0

t
u

After this preparation we come to the principal result of this section:

Theorem 2.6. Under our inductive assumption we have:

(a) The determinant of the G-relative Colin matrix of FH is different from zero:
det.ColF H
G=H
/ ¤ 0.
(b) The specialized resolvent ResFH .z/
G=H .X / is separable if and only if the deter-
F .z/
H
minant of the specialized Colin matrix satisfies det.ColG=H / ¤ 0.

Proof. Let 1 ; : : : ; m be a system of cosets representatives of H in G. Then

X
m X
m
ˇ.FHi ; FHj / D l .FHi FHj / D l .FHi /l .FHj /:
lD1 lD1

This leads to the following factorization of the Colin matrix:

0 1
1 ::: 1 0 m1
1
B 1 .FH / : : : m .FH / C 1 : : : 1 .FH /
B C B :: :: C
ColFG=H D B
H
:: :: C@: : A:
@ : : A m1
1 : : : m .FH /
1 .FHm1 / : : : m .FHm1 /

So det.ColF H
G=H / is the square of the determinant of the Vandermonde matrix in
1 .FH /; : : : ; m .FH / and hence different from zero.
400 V Additive Polynomials

H F .z/
If moreover the polynomial ResG=H .X / is separable, then its discriminant
Y 2
D.ResFH .z/
G=H / D z .i .FH //  z .j .FH //
i <j

does not vanish and by the above coincides with the determinant of the specialized
Colin matrix det.ColFH
G=H /. t
u

As an immediate consequence we obtain:

Corollary 2.7. If ResFH .z/

G=H .X / is separable the specializations of the invariants g 2
IFq ŒV H are computable successively using Proposition 2.5.

As a means to facilitate the computation of the specialized Colin matrix we

observe the following:
H F .z/
Proposition 2.8. The specialized Colin matrix ColG=H can be computed directly
from the coefficients of the specialized relative resolvent ResFH .z/
G=H
.X / using the
Newton identities.

Proof. The specialized G-relative resolvent ResFH .z/

G=H
.X / has the form

F .z/
Y
m X
m
H
ResG=H .X / D X  i .z .FH // D X m C .1/i ai X mi 2 KŒX
i D1 i D1

with the elementary symmetric polynomials

ai D si 1 .z .FH //; : : : ; m .z .FH // ;

whose values we assume to be given. The entries of the specialized Colin matrix
have the form
X
m
ˇ.z .FHi ; FHj // D l .z .FHi Cj //:
lD1

So they are power sums in the l .z .FH //. The latter can be computed recursively
from the a1 ; : : : ; am using the Newton identities. t
u

It now remains to understand how the assumptions on separability of the special-

ized resolvent ResFH .z/
G=H .X / can be guaranteed. This is the aim of the next section.

2.4 Linear Tschirnhaus Transformations

Let K  IFq be a field, f 2 KŒX a separable q-additive polynomial and z D

.´1 ; : : : ; ´n / a basis of the IFq -vector space of zeroes V  K sep . If 0 ¤ g 2 KŒX
2 Computation of the Galois Group 401

is a further q-additive polynomial then

Y  X
n 
g .f /.X / WD X ai g.´i / 2 K sep ŒX
.a1 ;:::;an /2IFn
q i D1

is called the linear Tschirnhaus transform of f with respect to g. Clearly g .f /.X /

is again a q-additive polynomial and it satisfies g .f /.X / 2 KŒX because of the
invariance of its coefficients under Gal.f /.
Proposition 2.9. Let f; g 2 KŒX be as above. If the Tschirnhaus transform
g .f /.X / is separable then the splitting fields and hence the Galois groups of f
and g .f / over K agree.
Proof. Let N be the splitting field of f .X / over K, so ´i 2 N . Let fQ WD g .f /
denote the linear Tschirnhaus transform of f with respect to g. Then g.z/ WD
.g.´1 /; : : : ; g.´n // is a generating system of VQ in the splitting field NQ of fQ.X /
over K. Because fQ is separable, VQ is an n-dimensional IFq -vector space and
g.´1 /; : : : ; g.´n / are linearly independent elements thereof. As N=K is Galois and
NQ N , the extension N=NQ is also Galois. Now let Q 2 Gal.N=NQ /. Then there
exists C D .cij /ni;j D1 2 GLn .q/ with

X
n
Q .´j / D cij ´i for j D 1; : : : ; n:
i D1

As g.´j / 2 NQ it is invariant under Q . Hence we see that

X
n
g.´j / D .g.´
Q j // D cij g.´i /:
i D1

Consequently, C D In , and so Q D idN , whence N D NQ and Gal.f / D Gal.fQ/. u

t
The next result shows the existence of a linear Tschirnhaus transform which pos-
sesses a separable specialized resolvent and thus solves the problem posed at the
end of the previous subsection.
Theorem 2.10. Let K  IFq be an infinite field, f .X / 2 KŒX a q-additive poly-
nomial, V  K sep the vector space of zeroes of f .X / and z D .´1 ; : : : ; ´n / an
IFq -basis of V . Further let Galz .f / G GLn .q/, H G a subgroup of G and
FH 2 KŒV H a G-relative H -invariant. Then there exists a q-additive polynomial
g.X / 2 KŒX with respect to which the Tschirnhaus transform of f .X / preserves
the Galois group and whose specialized resolvent is separable, i.e., we have
F .g.z//
Galg.z/ .g .f // Š Galz .f / H
and D.ResG=H / ¤ 0:

Proof. Let again N denote the splitting field of f .X / over K and q .´1 ; : : : ; ´n / 2
GLn .N / the Moore matrix with basis elements ´i of the vector space of zeroes V .
402 V Additive Polynomials

With these for N ŒX WD N ŒX1 ; : : : ; Xn we define the ring automorphism

X
n
j 1
W N ŒX ! N ŒX ; Xi 7! Ti WD ´qi Xj :
j D1

As det.q .´1 ; : : : ; ´n // ¤ 0 the T1 ; : : : ; Tn are also algebraically independent over

N and
Y Xn 
F .X / WD X ci Ti
c2IFn
q i D1

is a (general) q-additive polynomial of degree q n . Consequently the discriminant

D.F / of F .X / is a non-zero polynomial in X1 ; : : : ; Xn , say D.F / D h.X1 ; : : : ; Xn /
2 N ŒX . As T1 ; : : : ; Tn are algebraically independent over N , ResF H .T1 ;:::;Tn /
G=H
.X / is
Q Q
separable. So there exists a polynomial h.X1 ; : : : ; Xn / 2 N ŒX with h.X1 ; : : : ; Xn / WD
F .T ;:::;Tn /
D.Res H 1 Q
/ ¤ 0. But then also the product d.X/ WD h.X/  h.X/ 2 N ŒX is
G=H
non-zero. As jKj D 1 by assumption there exist ai 2 K with d.a0 ; : : : ; an1 / ¤ 0.
Pn1 k
If we now set g.X / WD kD0 ak X q then the discriminants D.g .f // D h.a0 ; : : : ;
FH .g.z//
an1 / and D.ResG=H Q 0 ; : : : ; an1 / are non-zero. Thus the claim follows
/ D h.a
by Proposition 2.9. t
u

Remark. The conclusion of Theorem 2.10 remains correct for finite fields K when-
ever we have jKj > m.m  1/, where m D .G W H / (see Garcia Lopez (2010),
Ch. 5).

Obviously by successive specialization Xi 7! ai 1 we can find in finitely many

steps elements a0 ; : : : ; an1 2 K with d.a0 ; : : : ; an1 / ¤ 0. From this we finally
obtain:

Corollary 2.11. Let K  IFq and f .X / 2 KŒX a q-additive polynomial, then the
Galois group Gal.f / of f .X / over K can be computed in finitely many steps.

Proof. This follows from Theorem 2.2 by induction along a chain of maximal sub-
groups. To improve the separability condition step by step one has to use the auxil-
iary results developped in Sections 2.3 and 2.4, thus possibly moving to Tschirnhaus
transforms with better separability properties. This can be achieved with finitely
many calculations. t
u

2.5 The Modular Dedekind Criterion

In this last section we derive also lower bounds for the Galois group of a q-additive
polynomial. In the case of agreement with the upper bound from Theorem 1.9 these
render superfluous the sometimes rather cumbersome application of the algorithm
for the computation of Galois groups described above.
2 Computation of the Galois Group 403

We assume here that .K; q / is an F-field with IFq K. Further let v be a non-
trivial Krull valuation on K with valuation ring Ov , valuation ideal Pv G Ov and
residue field kv WD Ov =Pv (see Engler and Prestel (2005), Ch. 2.1). If K=IFq is
finitely generated and the transcendence degree of K=IFq agrees with the rank of
the valuation v, then kv =IFq is a finite extension (see Engler and Prestel (2005),
Cor. 3.4.4) and thus kv is a finite field. The canonical homomorphism Ov ! kv will
usually be denoted by v .
Theorem 2.12 (Modular Dedekind Criterion). Let .K; q / be an F-field with val-
uation v, valuation ring Ov , valuation ideal Pv and finite residue field k with
Œk W IFq D m. Further let .M; ˚/ be a dualizable F-module over K with basis
B D fb1 ; : : : ; bn g and D WD DB .˚/ 2 GLn .O/. Let DP WD v .D/ 2 GLn .K/ denote
the image of D under the canonical morphism v . Then we have:
(a) The Galois group Gal.M; ˚/ GLn .q/ contains elements that are GLn .IFq /-
conjugate to
DO WD DP q .DP /    qm1 .DP /:
(b) If G is a connected linear algebraic group defined over IFq and if D lies in
G.O/ then Gal.M; ˚/ contains an element which is G.IFq /-conjugate to D. O
Proof. Let N=K be a minimal solution field of .M; ˚/ and Y WD .yij /ni;j D1 2
GLn .N / a fundamental solution matrix of .M; ˚/, so that q .Y / D D 1 Y . Fur-
ther let vQ be an extension of the valuation v to N with valuation ring O, Q valuation
Q Q
ideal P and residue field k (according to Engler and Prestel (2005), Thm. 3.1.2). As
D 1 2 GLn .O/ Q it follows from the ultrametric triangle inequality for vQ that the yij
Q So the reduction Y Q WD vQ .Y / 2 GLn .k/
as well as det.Y / and det.Y 1 / lie in O. Q
P
is well-defined and we have YPQ D DP q .YPQ /. By Corollary 1.3 the field NPQ WD
k.YPQ / kQ is a minimal solution field of the reduced F-module .VP ; v ı ˚/ with
the reduced vector space VP Š k n .
By assumption the residue field k has finite degree m over IFq . Hence Gal.NPQ =k/
is cyclic and generated by the Frobenius automorphism P D qm . From this we ob-
tain
YPQ D DPQ P .YPQ / with DPQ D DP q .DP /    qm1 .DP /:
Setting CPQ WD YPQ1 DP
1
YPQ this gives

P .YPQ / D YPQ CPQ ;

whence CPQ is the representing matrix of P , and CPQ 2 GLn .q/.

Now let Kh be the decomposition field of v in N=K. Then the residue class
map from Gal.N=Kh/ to Gal.k=k/ Q is surjective (by Engler and Prestel (2005),
Lemma 5.2.6). Hence the extension Q P WD qm of P to Gal.k=k/Q has a preimage
 2 Gal.N=Kh / Gal.N=K/. For this there exists a matrix C 2 GLn .q/ with
.Y / D Y  C . Via the natural epimorphism vQ we obtain from this

Q P .YPQ / D P .YPQ / D YPQ C ;

which implies that C D CPQ . Part (a) of the theorem follows.

404 V Additive Polynomials

Under the assumptions in part (b) we have Y 2 G.K sep/ by the Theorem II.1.1
of Lang–Steinberg. Application of the canonical map then gives YPQ 2 G.IFq / and
1
P .YPQ / 2 G.IFq /, since G is defined over IFq . As CPQ D YPQ DP YPQ this shows that
1
C is conjugate to DPQ 2 G.IFq / inside G.k/ Q G.IFq /. t
u

Here we are particularly interested in those representing matrices D with coef-

ficients in IFq Œt1 ; : : : ; tr DW R and in specializations ti 7! ai 2 IFq . For this let
A WD .t1 a1 ; : : : ; tr ar /GR be the corresponding maximal ideal and RA the local-
ization of R at A with maximal ideal ARA . Then by a theorem of Chevalley there
exist valuations v on K D IFq .t1 ; : : : ; tr / with Ov RA and Pv \ RA D ARA D
.t1  a1 ; : : : ; tr  ar / (see e.g. Engler and Prestel (2005), Thm. 3.1.1), for example
the map

v W IFq .t1 ; : : : ; tr / ! ZZr ; f .t1 ; : : : ; tr / 7! ord.t1 a1 / .f /; : : : ; ord.tr ar / .f / ;

of IFq .t/ into the lexicographically ordered abelian group ZZr given by the multi-
plicities of .ti  ai / in f .t/. For this the residue field kv coincides with IFq , so we
have

Ov =Pv D kv Š IFq Š RA =ARA Š IFq Œt1 ; : : : ; tr =.t1  a1 ; : : : ; tr  ar /:

In this special case we obtain from Theorem 2.12:

Corollary 2.13. Let K D IFq .t1 ; : : : ; tr / be a rational function field in r variables

over IFq with Frobenius endomorphism  D q . Further let .M; ˚/ be an n-
dimensional dualizable F-module over K with basis B and representing matrix D D
DB .˚/ 2 GLn .IFq Œt1 ; : : : ; tr /. Then for all a D .a1 ; : : : ; ar / 2 IFrq with det.D.a1 ; : : : ;
ar // ¤ 0, Gal.M; ˚/ contains elements Ca 2 GLn .q/ which are GLn .IFq /-conjugate
to D.a/.

The group generated by the elements Ca in Corollary 2.13 usually already gives
quite good lower bounds for Gal.M; ˚/, as will become apparent in the following
paragraphs.
As first applications we continue our Examples 1.1 to 1.3 from Paragraph 1, so
again K D IFp .t/ and  D p .

Example 2.1. First let .M; ˚/ be the 1-dimensional F-module over K with D WD
DB .˚/ D .t 1 /. Then a primitive element w 2 IF
p also satisfies hD
1
.w/i D IF
p.
So Gal.M; ˚/ contains the group Gm .IFp / by Theorem 2.12. t
u

 2.2. Now let .M; ˚/ be a 2-dimensional F-module over K with DB .˚/ D

Example

1 t
. Then Gal.M; ˚/ contains by Corollary 2.13 an element conjugate to
0 1
 
11
, of order p, so it contains a copy of the group Ga .IFp /. t
u
01
2 Computation of the Galois Group 405

Example
 1 2.3. Finally let .M; ˚/ be the F-module over K with D WD DB .˚/ D
t 1
. So D lies in the Borel subgroup B2 .IFp .t// of upper triangular matrices
0 t
in SL2 .IFp .t//. According to Theorem 1.9 the Galois group Gal.M; ˚/ is then a
subgroup of B2 .IFp / with jB2 .IFp /j D p.p  1/. As D.1/ 2 GL2 .IFp / has order p
and D.w/ has order p  1, we obtain that j Gal.M; ˚/j  jB2 .IFp /j and hence
Gal.M; ˚/ Š B2 .IFp /. t
u
406 V Additive Polynomials

3 Polynomials for Split Groups of Lie Type

In this paragraph we construct polynomials over IFq .t/ D IFq .t1 ; : : : ; tn / with Galois
group G.IFq / a split group of Lie type, where G is one of the linear groups SLnC1 ,
Sp2n ; SO2nC1 ; SO2n or G2 respectively. For this we utilize Frobenius modules
whose representing matrix lies in the natural representation of G over IFq .t/ and
which is sufficiently general. The results presented here are mostly taken from the
thesis of Albert and form part of the joint publication Albert and Maier (2011).

3.1 Linear Groups SLnC1

We start with some preparatory observations. For the construction of sufficiently

general representing matrices D D DB .˚/ in G.IFq .t1 ; : : : ; tn // we use so-called
Steinberg cross sections. Their importance is demonstrated by the following result,
which we will not prove here as it is not used in the sequel (see Steinberg (1965),
Thm. 1.4).
Theorem 3.1 (Steinberg cross section). Let G be a (connected) semisimple linear
algebraic group of simply connected type of rank n over a field k, T G a maximal
torus and ˛1 ; : : : ; ˛n a system of simple roots of G with respect to T with !1 ; : : : ; !n 2
NG .T/ mapping to the corresponding simple reflections in the Weyl group NG .T/=T,
and with corresponding root subgroups X1 ; : : : ; Xn G. Then

Y
n
S WD Xi !i WD X1 !1    Xn !n
i D1

contains representatives of all regular conjugacy classes of G.

The set S in Theorem 3.1 will be called a Steinberg cross section of G. As an
algebraic variety it is isomorphic to an n-dimensional k-vector space (Steinberg
(1965), Thm. 7.1). In the sequel we will only use its construction. The property of
being a section then offers a good chance to find sufficiently general specializations.
This will be verified explicitly in each application.
Remark. According to Theorem 3.1 the set S is an n-parameter family S.t1 ; : : : ; tn /
in G.IFq /, and so can be considered as an element of G.IFq .t//.
As an easy example we first compute a Steinberg cross section of SLnC1 .
Proposition 3.2. The group GAn WD SLnC1 possesses a Steinberg cross section of
the form 0 1
t1 : : : tn 1
B 1 0C
B C
SAn .t1 ; : : : ; tn / D B : :: C
@ :: :A
1 0
3 Polynomials for Split Groups of Lie Type 407

with characteristic polynomial

X
n
hAn .t1 ; : : : ; tn / D X nC1 C .1/i 1 ti X nC1i C .1/nC1 :
i D1

Proof. According to Section II.1.2, the i  i 1 , 1 i n, form a system of sim-

ple roots for a root system of type An . The corresponding root subgroups have
the form Xi D Xi;i C1 D fId Cti Ii;i C1 j ti 2 IFq g with Weyl group representatives
0 1
!i D diag.1; : : : ; 1; ; 1; : : : ; 1/, where the block of size 2 occurs in positions
1 0
.i; i C 1/. With this we get
 
t1 1
Xi .ti /!i D diag.1; : : : ; 1; ; 1; : : : ; 1/
1 0

again with the 2-block at positions .i; i C 1/. From this induction over j yields
0 1
t1 : : : tj 1
B 1 0 C
B C
B : : C
Y j B : : :
: C
B C
Sj WD B
Xi .ti /!i D B 1 0 C
C
i D1 B 1 0C
B C
B :: C
@ : A
0 1

and thus for j D n the expression for SAn .t1 ; : : : ; tn / as claimed. The computation
of the characteristic polynomial is straightforward. t
u
To be able to successfully apply the Modular Dedekind Criterion we need gen-
erating systems .1 ; : : : ; s / of G.IFq / with the property that for all GLn .IFq /-
conjugates Qi 2 G.IFq / of i the tuple .Q1 ; : : : ; Qs / also generates G.IFq /. Such
generating systems will be called robust (with respect to GLn .IFq /). Such a robust
generating system .1 ; : : : ; s / 2 G.IFq /s occurs for example if there exists no
proper subgroup of G.IFq / containing elements with all element orders occurring
for 1 ; : : : ; n . For the groups GAn .IFq / robust systems are provided by the next
result which essentially goes back to Malle, Saxl and Weigel (1994).
Proposition 3.3. Let n  1 and q be a prime power.
(a) The group GAn .IFq / D SLnC1 .q/ possesses cyclic maximal tori T1 and T2 of
orders .q nC1  1/=.q  1/ and q n  1 respectively.
(b) For .n; q/ … f.1; 2/; .1; 3/; .1; 4/; .1; 7/; .1; 9/g pairs .1 ; 2 / of generators of
T1 ; T2 form robust generating systems of GAn .IFq /.
(c) In the cases excluded in (b), a robust system .1 ; 2 ; 3 / is obtained by adding
an element 3 of order gcd.2; p  1/p.
Proof. We first consider the case n D 1, that is, the groups SL2 .q/. Then asser-
tions (a) and (b) easily follow from the Theorem of Dickson (see Huppert (1967),
408 V Additive Polynomials

Kap. II, Satz 8.27), since by order considerations no proper subgroups can contain
conjugates of both 1 ; 2 . In the excluded cases the same holds using a third element
3 of order gcd.2; p  1/p.
In the general case n > 1 part (a) follows from Malle, Saxl and Weigel (1994),
Tab. III, and part (b) from Malle, Saxl and Weigel (1994), proof of Thm. 3.1 (second
paragraph). In the special case of SL4 .2/ Š A8 the elements 1 ; 2 have orders 7,
15 respectively, which again cannot both lie in a proper subgroup. t
u
This allows us to prove the following result:
Theorem 3.4. Let n  1.
(a) The Frobenius module .MAn ; ˚/ over K D IFq .t1 ; : : : ; tn / with representing
matrix DB .˚/ D SAn .t1 ; : : : ; tn / is dualizable.
(b) The Galois group of MAn ; ˚/ is SLnC1 .q/.
(c) The corresponding Galois extension N=IFq .t/ is geometric and generated by
the zeroes of the characteristic polynomial of .MAn ; ˚/

nC1
X
n
i
fAn .X / D X q C .1/i ti X q C .1/nC1 X:
i D1

Proof. The F-module .MAn ; ˚/ is dualizable as det.SAn .t// ¤ 0 by Proposition 3.2.

By Theorem 1.9 its Galois group G WD Gal.MAn ; ˚/ is a subgroup of SLnC1 .q/
since SAn .t/ 2 GAn .IFq .t//.
The generators 1 ; 2 from Proposition 3.3 possess the Jordan normal forms
n 
diag.v; v q ; : : : ; v q / over IFq with an element v 2 IFq of order .q nC1  1/=.q  1/,
q n 1
respectively diag.w; w q ; : : : ; w q ; w  q1 / with w of order q n  1. Consequently
n1

the conjugacy classes of 1 and 2 in G.IFq / are both determined by their charac-
teristic polynomials hj .X /, j D 1; 2. Write

X
n
hj .X / D X nC1 C .1/i C1 cij X nC1i C .1/nC1
i D1

with coefficient vector cj D .c1j ; : : : ; cnj / 2 IFnq . Specializing t D .t1 ; : : : ; tn / in

hAn .t/ to cj we obtain that SAn .cj / D D.cj / is a GLn .IFq /-conjugate of j , hence
also a generator of a torus conjugate to Tj . Apart from the listed exceptions these
form a robust generating system for SLnC1 .q/ by Proposition 3.3.
In the special cases we have n D 1.Then the specialization SA1 .˙2/ D D.˙2/
˙2 1
leads to an element 3 conjugate to . Since its characteristic polynomial
1 0
equals h3 .X / D X 2 ˙ 2X C 1 and 32 ¤ I2 in case of odd characteristic, 3 has
order 2p, respectively p in characteristic 2. Thus here (b) follows from Proposi-
tion 3.3(c).
It remains to calculate the characteristic polynomial of .MAn ; ˚/. This is obtained
from the non-linear system of equations

SAn .t/.X1q ; : : : ; XnC1

q
/t D .X1 ; : : : ; XnC1 /t
3 Polynomials for Split Groups of Lie Type 409

by solving for X WD X1 . We have

X
n
 ti Xiq C XnC1
q
D X1 D X and
i D1

q i
Xi D Xi C1 resp. Xi C1 D .1/i X q for i D 1; : : : ; n:
From this we conclude that
X
n
i nC1
 ti .1/i 1 X q C .1/n X q D X;
i D1

which is the polynomial given in (b).

The geometricity of N=IFq .t/ finally follows from the fact that all specializations
used in the proof are compatible with extensions of constants and thus the elements
generate the same group after extension of constants. t
u
Remark. By the Hilbert irreducibility theorem (Theorem IV.1.1) Theorem 3.4 also
yields Galois extensions over IFq .t/ with Galois group SLnC1 .q/. In the non-
exceptional cases of Proposition 3.3 such extensions can very easily be given explic-
itly, for example by the specialization

ti 7! ci1 C t.ci 2  ci1 /

(in the notation of the previous proof). Indeed, after the further specialization t 7! 0,
respectively t 7! 1 we arrive at the robust generators 1 ; 2 .
Corollary 3.5. The fixed field of the center Z .SLnC1 .q// in the Galois extension
N=K in Theorem 3.4 is a Galois extension over IFq .t/ with group LnC1 .q/. It is
generated over K by the roots of

q nC1 1 X
n
q i 1
fNAn .Y / D Y q1 C .1/i ti Y q1 C .1/nC1 :
i D1

Proof. The irreducible polynomial fNAn .Y / arises by dividing fAn .Y / by X and

then substituting X q1 by Y . Let L be the intermediate field generated by the roots
of fNAn .Y /. Since Z .SLnC1 .q// acts on the root space V of fAn .Y / by scalar mul-
tiplication with suitable elements c 2 IF N
q , it acts trivially on the roots of fAn .Y /.
Hence L is contained in the fixed field of Z .SLnC1 .q//, and its non-trivial Galois
group is a factor group of SLnC1 .q/=Z .SLnC1 .q//, so Gal.L=K/ Š LnC1 .q/. u t
The polynomial fAn .X / in Theorem 3.4(c) is obtained from the Dickson polyno-
mial (2.1) by specializing the Dickson invariant dnC1 to .1/nC1 . (The specializa-
q1
tion dnC1 7! .1/nC1 tnC1 would lead to a generic polynomial for SLnC1 .q/. This
follows from the Theorem of Dickson (Smith (1995), Thm. 8.1.1) that
SLnC1 .q/ q1
IFq ŒV D IFq Œd1 ; : : : ; dn ; gnC1 with gnC1 D dnC1
410 V Additive Polynomials

in combination with a result of Kemper and Mattig on generic polynomials explained

in Jensen, Ledet and Yui (2002), Prop. 1.1.3.)
Further specializations ti 7! 0 for i D 1; : : : ; n  1 in fAn .X / and fNAn .Y / respec-
tively give the nice SLnC1 .q/- and LnC1 .q/-polynomials of Abhyankar (1994).

3.2 Symplectic Groups

We next turn our attention to the symplectic groups Sp2n .q/, n  2, with q any
prime power. First we construct a Steinberg cross section.

Proposition 3.6. The groups GCn .IFq / WD Sp2n .q/ possess a Steinberg cross section
of the form
0 1
t1 : : : tn1 tn 1
B 1 C
B C
B : C
B :: C
B C
B 1 0 C
B
SCn .t1 ; : : : ; tn / D B C:
tn1 0 1 C
B C
B :: :: :: C
B : : : C
B C
@ t1 1A
1 0

Its characteristic polynomial is given by the symmetric polynomial

X
n1 X
n1
hCn .X / D X 2n C ti X 2ni C tn X n C ti X i C 1:
i D1 i D1

Proof. In the groups GCn D Sp2n the root subgroups and corresponding simple
reflections for i D 1; : : : ; n  1 have the form
   
1 ti 1 ti
Xi .ti / D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/;
0 1 01
   
01 01
!i D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/;
10 10
where the 2-blocks are at position .i; i C1/, and symmetrically at .2ni 1; 2ni /.
For i D n we have
 
1 tn
Xn .tn / D diag.1; : : : ; 1; ; 1; : : : ; 1/;
0 1
 
01
!n D diag.1; : : : ; 1; ; 1; : : : ; 1/;
10
3 Polynomials for Split Groups of Lie Type 411

with the 2-block in middle position .n; nC1/ (see Digne and Michel (1991), 15.2).
From this we obtain
   
ti 1 t 1
Xi .ti /!i D diag.1; : : : ; 1; ; 1; : : : ; 1; i ; 1 : : : ; 1/; for i < n;
1 0 10

respectively  
tn 1
Xn .tn /!n D diag.1; : : : ; 1; ; 1; : : : ; 1/:
1 0
Induction then easily gives the stated expression for SCn .t1 ; : : : ; tn /.
To calculate the characteristic polynomial of SCn .t1 ; : : : ; tn / we first eliminate the
strictly lower triangular entries in the upper left hand and the lower right hand block
to obtain a matrix 0 1
g1 t2 : : : tn 1
B g2 C
B C
B :: : : C
B : :: :: C
B C
CQ D BB g n 1 C
C
B h X n C
B n1 C
B :: : C
@ : : : A
h0 X
Pi P
with t0 WD 1, gi WD j D0 tj X i j and hi WD ij D0 tj X j . Under consideration of
the factors used for the twofold triangulation this leads to
  X
n X
n1
gn 1 n
hCn .X / D det n DX ti X n1 C ti X i
hn1 X
i D0 i D1

which is the claimed expression. t

u
A robust generating system for Sp2n .q/ can be deduced from Guralnick and
Malle (2012):
Proposition 3.7. Let n  3.
(a) The group GCn .IFq / D Sp2n .q/ possesses maximal tori T1 ; T2 of orders
q n C 1 and .q n1 C 1/.q C 1/ respectively.
(b) Any pair .1 ; 2 / of regular elements of T1 ; T2 of maximal order forms a
robust generating pair.
Proof. The existence of maximal tori of the given orders is immediate by Malle
and Testerman (2011), Prop. 25.3, for example. The assertion in (b) follows from
Guralnick and Malle (2012), Cor. 3.4. Indeed, let H be generated by 1 ; 2
as in the statement. As n  3 any two elements of the given orders generate
an irreducible subgroup of GL2n .q/. Easy order considerations show that the
only possibility for H is as in Guralnick and Malle (2012), Cor. 3.4(1). As by
assumption H Sp2n .q/, we either have H D Sp2n .q/ as claimed, or q is even
C
and H GO˙ 2n .q/. But the order of GO2n .q/ is not divisible by q C 1, while
n
412 V Additive Polynomials

GO 2n .q/ does not contain tori of order .q

n1
C 1/.q C 1/ (see again Malle and
Testerman (2011), Prop. 25.3), hence no regular elements with that centralizer.
n1 n1
Note that 1 has Jordan normal form diag.v; v q ; : : : ; v q ; v q ; : : : ; v q ; v 1 /

over IFq for some element v 2 IFq of orders q n C 1, while 2 has Jordan nor-
n2 n2
mal form diag.v; v q ; : : : ; v q ; w; w 1 ; v q ; : : : ; v q ; v 1 / for suitable elements

v; w 2 IFq of orders q n1 C 1; q C 1 respectively. t
u

With these preparations we are in a position to prove the Galois realizations:

Theorem 3.8. Let n  3.

(a) The Frobenius module .MCn ; ˚/ over K D IFq .t1 ; : : : ; tn / with representing
matrix DB .˚/ D SCn .t1 ; : : : ; tn / is dualizable.
(b) The Galois group of .MCn ; ˚/ is Sp2n .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial of .MCn ; ˚/

2n
X
n1
i
q nCi n
X
n1
i
fCn .X / D X q C tni Xq C tn X q C ti X q C X:
i D1 i D1

Proof. From Proposition 3.6 we see that .MCn ; ˚/ is dualizable and Sp2n .q/ is an
upper bound for its Galois group. The elements 1 ; 2 from Proposition 3.7(b) yield
a robust generating system of Sp2n .q/. According to the proof of Proposition 3.7
not only hCn but also the characteristic polynomials of 1 and 2 possess symmet-
ric coefficients; so the ti can be specialized so that hCn becomes the characteristic
polynomial of 1 or 2 . This yields elements that are GLn .IFq /-conjugate to the i .
It remains to compute the characteristic polynomial of .MCn ; ˚/. It is obtained
from the algebraic system of equations
q q
SCn .X1 ; : : : ; X2n /t D .X1 ; : : : ; X2n /t

by solving for X WD X1 . The individual equations yield

X
n
 ti Xiq  tn Xnq C XnC1
q
D X1 D X;
i D1

i 1
Xiq1 D Xi resp. Xi D X q for i D 1; : : : ; n;
q
tni Xnq C XnCi C1 D XnCi for i D 1; : : : ; n  1;
n
Xnq D X2n resp. X2n D X q :
3 Polynomials for Split Groups of Lie Type 413

From this we get by induction

X
j 1
i nCi nCj
X2nj D  tjqi X q  Xq for j D 0; : : : ; n  1;
i D0
X
n1
qi 1 nCi 1 2n1
XnC1 D  tni Xq  Xq :
i D0

Substitution into the first equation then yields the stated characteristic polynomial.
The geometricity of the Galois extension follows as in Theorem 3.4. t
u

Corollary 3.9. The fixed field of the center Z .Sp2n .q// in the Galois extension
N=K in Theorem 3.8 gives a Galois extension over IFq .t/ with group S2n .q/. It is
generated over K by the roots of the projective version fNCn .Y / of fCn .X / which
arises by dividing fCn .X / by X and substituting Y for X q1 .

The nice polynomials of Abhyankar (1996b) for Sp2n .q/ and S2n .q/ over
IFq .t; u/ can be obtained from fCn .X / respectively fNCn .Y / by specializing ti 7! 0
for i D 1; : : : ; n  2 (see also Abhyankar and Loomis (1998, 1999)).
We close by pointing out that Elkies (1997), using a completely different method,
obtained essentially the same polynomial for Sp2n .q/.

3.3 Odd-Dimensional Orthogonal Groups

Here we consider the groups SO2nC1 .q/. As SO2nC1 .2l / Š Sp2n .2l / we may
restrict ourselves to the case of odd q. We start again with the construction of a
Steinberg cross section. For this we introduce an additional parameter s which will
allow us later to find specializations into suitable classes modulo squares.

Proposition 3.10. The groups GBn .IFq / D SO2nC1 .q/ with q odd possess a Stein-
berg cross section of the form
0 1
t1 st2 : : : stn1 2s tn2 stn s
B 1 C
B C
B 1 C
B C
B :: C
B : C
B C
B 1 0 C
SBn .t1 ; : : : ; tn ; s/ D B
B
C:
C
B tn 1 C
B tn1 0 0 1 C
B C
B : :: C
B :: : C
B C
@ 1 t
1A
s
 1s 0
414 V Additive Polynomials

Its characteristic polynomial hBn .X / has the anti-symmetric form

1 X n X n
hBn .X / D .si 1 C si / X 2nC1i  .si 1 C si / X i
s
i D0 i D0

t1
with s1 D 0, s0 D 1s , s1 D s , si D ti for i D 2; : : : ; n  1 and sn D 12 tn2 .

Proof. For the computation of the Steinberg cross section we use the root subgroups
   
1 ti 1 ti
Xi .ti / D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/; for i D 1; : : : ; n  1;
0 1 01
0 1
1 tn  12 tn2
Xn .tn / D diag.1; : : : ; 1; @0 1 tn A ; 1; : : : ; 1/
0 0 1
with the 2-blocks at positions .i; i C 1/ and symmetrically at .2n C 1  i; 2n C 2 
i /, and the 3-block at position .n; n C 1; n C 2/. As representatives for the simple
reflections we choose
   
0s 01
!1 D diag. ; 1 : : : ; 1; 1 /;
10 s
0
   
01 01
!i D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/; for i D 2; : : : ; n  1;
10 10
and 0 1
0 0 1
!n D diag.1; : : : ; 1; @ 0 1 0 A ; 1; : : : ; 1/:
1 0 0
Here, the 2-blocks are in positions .i; i C 1/ and .2n C 1  i; 2n C 2  i /, and the
3-block is in the central position .n; n C 1; n C 2/.From this an easy induction yields
the claimed expression for the Steinberg cross section.
The characteristic polynomial of SBn .t1 ; : : : ; tn ; s/ can be computed as shown in
the proof of Proposition 3.7. After suitable transformations we obtain
0 1
gn stn s
hBn .X / D det @ tn X C 1 0 A
hn1 0 X n
Dgn .X /.X C 1/X n  stn2 X n  s.X C 1/hn1 .X /
X n  X
n1 
D ssi X ni .X C 1/X n  2ssn X n  s.X C 1/ si X i
i D0 i D0

with the corresponding gn ; hn1 . t

u

Next, we need a robust system of generators.

3 Polynomials for Split Groups of Lie Type 415

Proposition 3.11. Let q be odd and n  3.

(a) The groups GBn .IFq / D SO2nC1 .q/ possess cyclic maximal tori T1 ; T2 of
orders q n C 1, q n  1 respectively.
(b) Any pair of generators .1 ; 2 / of T1 and T2 is a robust generating system of
GBn .IFq /.
Proof. Part (a) follows from Malle, Saxl and Weigel (1994), Tab. III. Generators i
n1 n1
of Ti have Jordan normal form diag.v; v q ; : : : ; v q ; 1; v q ; : : : ; v q ; v 1 / over

IFq for elements v 2 IFq of orders q n C 1, q n  1 respectively. Part (b) then follows
from Malle, Saxl and Weigel (1994), proof of Thm. 3.1 (fourth paragraph). t
u
After these preparations the proof of the next result proceeds as usually.
Theorem 3.12. Let q be odd and n  3.
(a) The F-module .MBn ; ˚/ over K D IFq .t1 ; : : : ; tn ; s/ with the representing
matrix DB .˚/ D SBn .t1 ; : : : ; tn ; s/ is dualizable.
(b) The Galois group of .MBn ; ˚/ is SO2nC1 .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial fBn .X / of .MBn ; ˚/, given by

X
n
i 1 nC1
s q fBn .X / D  .siq1 C tnq1 si / X q C .tn1
q
tnq1  tn2q / X q
2
i D0

X
nC1
qi i nCi
q1 q
C .sni C1 C tn sni / X q
i D2

with the si from Proposition 3.10.

Proof. Part (a) and the assertion that the group SO2nC1 .q/ is an upper bound for
the Galois group follow directly from Proposition 3.11 and Theorem 1.9.
From the shape of the Jordan normal forms of 1 and 2 given in the proof of
Proposition 3.11 it is immediate that the characteristic polynomials of 1 and 2
are anti-symmetric. As hBn .X / is not generically anti-symmetric, it still remains to
verify that it specializes to both of these characteristic polynomials. For this let

X
n1 X
n1
h.X / D X 2nC1 C ai X 2ni C an X n  ai X i  1
i D0 i D0

be the characteristic polynomial of 1 or 2 . Then specialization of the coefficients

of hBn .X / leads to the system of equations

a1 D 1 C t1 ; a2 D 1 C st2 ;
1
ai D s.ti 1 C ti / for i D 3; : : : ; n  1 and an D stn1 C stn2 :
2
The first n  1 of these equations can be solved inductively for the ti , and the last
one is of the form 2s tn2 D u 2 IFq . By a suitable choice of the class of s modulo
squares, we can also guarantee this latter condition.
416 V Additive Polynomials

The geometricity of N=K follows as in the proof of Theorem 3.4. The computa-
tion of the characteristic polynomial fBn .X / from

SBn .X1q ; : : : ; X2nC1

q
/t D .X1 ; : : : ; X2nC1 /t

is achieved according to the example in the proof of Theorem 3.8. t

u

3.4 Even-Dimensional Orthogonal Groups SOC

2n .q/

For the even-dimensional orthogonal groups SOC 2n .q/ for simplicity we will only
consider the case when q is odd. Moreover, whenever convenient we will assume
that n  4, since in smaller rank the groups are either not simple or isomorphic to
other groups treated before. As in the odd-dimensional case we introduce an addi-
tional parameter for the Steinberg cross section, which originates from the freedom
of choice of Weyl group representatives.

Proposition 3.13. For odd q the group GDn .IFq / D SOC

2n .q/ possesses a Steinberg
cross section SDn .t1 ; : : : ; tn ; s/ of the form
0 1
t1 st2 : : : stn2 stn1 tn stn stn1 s
B 1 C
B C
B 1 C
B C
B :: C
B : C
B C
B 1 C
B C
B t 0 1 :C
B n C
B t 1 0 0 C
B n1 C
B t 0 1 C
B n2 C
B :: : : C
B : : : : C
:
B C
@  ts1 0 1 A
 1s 0

Its characteristic polynomial hDn .X / has the symmetric form

1 X n1 X n1
hDn .X / D .si  si 2 /X 2ni  .2tn2 C tn1
2
C tn2 /X n C .si  si 2 /X i
s
i D0 i D0

t1
with s2 D s1 D 0, s0 D 1s , s1 D s
, si D ti for i D 2; : : : ; n  2 and sn1 D tn1 tn .

Proof. Proceeding as for Proposition 3.10 we first choose root subgroups

   
1 ti 1 ti
Xi .ti / D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/; for i D 1; : : : ; n  1;
0 1 01
3 Polynomials for Split Groups of Lie Type 417
0 1
10 tn 0
B0 1 0 tn C
B
Xn .tn / D diag.1; : : : ; 1; @ C ; 1; : : : ; 1/;
00 1 0A
00 0 1
as well as    
0s 01
!1 D diag. ; 1 : : : ; 1; 1 /;
10 s
0
   
01 01
!i D diag.1; : : : ; 1; ; 1; : : : ; 1; ; 1 : : : ; 1/; for i D 2; : : : ; n  1;
10 10
and 0 1
0 0 1 0
B0 0 0 1C
!n D diag.1; : : : ; 1; B
@1 0
C ; 1; : : : ; 1/
0 0A
0 1 0 0
with the 2-blocks at positions .i; i C 1/ and symmetrically at .2n  i; 2n C 1  i /,
respectively with a 4-block in the middle position (see Malle and Testerman (2011),
Ex. 11.7). From this an easy induction yields Sj WD X1 .t1 /!1    Xj .tj /!j for j D
1; : : : ; n  1 and thus finally Sn D SDn .t1 ; : : : ; tn ; s/ as given above.
Performing the transformation described in Proposition 3.6 one obtains the char-
acteristic polynomial
0 1
gn1 stn stn1 s
B tn X 1 0 C
hDn .X / D det B
@ tn1 1 X
C
0 A
hn2 0 0 X n1

of SDn .X / with

X
n1 X
n2
gn1 .X / D s si X n1i and hn2 .X / D si X i ;
i D0 i D0

which equals

.X nC1  X n1 /gn1 .X /  s.tn2 C tn1

2
/X n C 2stn tn1 X n1  s.X 2  1/hn2 .X /:

Expanding gives the claimed formula. t

u
Robust generating systems for GDn are given as follows:
Proposition 3.14. Let n  4 and q be odd.
(a) The group GDn .IFq / D SOC2n .q/ has maximal tori T1 ; T2 of orders .q
n1
C 1/
.q C 1/ and q  1 when n is odd, respectively of orders .q
n n1
C 1/.q C 1/ and
n n
.q 2 C .1/ 2 /2 when n is even.
(b) Any two elements 1 of T1 and 2 of T2 of maximal order form a robust
generating system of GDn .
418 V Additive Polynomials

Proof. The existence of the Ti follows from Malle, Saxl and Weigel (1994),
Tab. III. The first torus, which we consider for all n, is generated by elements 1
n2 n2
with Jordan normal form diag.v; v q ; : : : ; v q ; w; w 1 ; v q ; : : : ; v 1 / over IFq ,

where v; w 2 IFq have orders q n1 C 1 and q C 1 respectively. (In particular T1 is
not cyclic.) For odd n the second torus T2 is cyclic, generated by elements 2 with
n1 n1 
Jordan normal form diag.v; : : : ; v q ; v q ; : : : ; v 1 / over IFq , where v 2 IFq
has order q n  1. For even n, T2 is generated by elements 2 with Jordan normal
form
n 1 n 1 n 1 n 1
diag.v; v q ; : : : ; v q 2 ; w; : : : ; w q 2 ; w q 2 ; : : : ; w 1 ; v q 2 ; : : : ; v 1 /
 n n
over IFq . Here both v; w are elements of IFq of order q 2 C .1/ 2 . (Hence in this
case T2 is not cyclic either.)
Part (b) of the assertion follows from Malle, Saxl and Weigel (1994), proof
of Thm. 3.1 (Paragraphs 8 and 9). (The only exceptional case occurring there is
SOC4 .2/, but we assume here that q is odd.) t
u
Theorem 3.15. Assume that n  4 and q is odd.
(a) The F-module .MDn ; ˚/ over K D IFq .t1 ; : : : ; tn ; s/ with representing matrix
DB .˚/ D SDn .t1 ; : : : ; tn ; s/ is dualizable.
(b) The Galois group of .MDn ; ˚/ is SOC 2n .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial fDn .X / of .MDn ; ˚/, which is given by

1 X
n1
q 2 2 i
f .X / D
2 Dn
.d q1 tnq si C .tn1  d q1 tn1 /siq1 C tnq siq2 / X q
sq i D0
q q2 q 2
qC1 q n
C ..d q1
tn1  tn1 /sn1 C.sn2  2d q1 tn1  d q  d q1 sn2 /tnq / X q
q 2 2 Cq q 2 q 2 q q 2 2 nC1
 .2tn1 tnq C tnq sn1 C.tn1  d q1 tn1 /sn2  d q1 tnq sn3 /Xq
X
n
q i C1 q2 q 2 C1 qC1 q q i C1 nCi
 .tnq sni Ctn1  d q1 tn1 /sni 1  d tn sni 2 / X q
i D2

with the si as given in Proposition 3.13 and d WD tnqC1  tn1

qC1
.
Proof. The assertion in (a) as well as the fact that SOC 2n .q/ is an upper bound for
Gal.MDn ; ˚/ follow directly from Proposition 3.13 and Theorem 1.9.
To show that SOC 2n .q/ is also a lower bound for Gal.MDn ; ˚/ it suffices to find
specializations of SDn .t1 ; : : : ; tn ; s/ to elements Qi 2 SOC 2n .q/ such that suitable pow-
ers of the Qi are GLn .IFq /-conjugate to the i in Proposition 3.14(b), since then
.Qi ; Q2 / form a robust generating system of SOC 2n .q/. Obviously the characteristic
polynomials of both 1 and 2 are symmetric, that is, of the form

X
n1 X
n1 Y
n
1
h.X / D X 2n C ai X 2ni C an X n C ai X i C 1 D .X  ci / X 
ci
i D1 i D1 i D1
3 Polynomials for Split Groups of Lie Type 419

with suitable ai 2 IFq and ci 2 IFq . We claim that hDn .X / can be specialized to
any such a symmetric polynomial h.X /. This leads to the system of equations

a0 D 1; a1 D t1 ; a2 D st2  1; a3 D st3  t1 ;
ai D s.ti  ti 2 / for i D 4; : : : ; n  2;
an1 D s.tn tn1  tn3 /; an D s.tn2 C tn1
2
 2tn2 /:

Elimination of t1 ; : : : ; tn1 yields the following biquadratic equation for tn

u 2 v 2
tn4 C t C D 0;
s n s
where
n
b2c b n1
2 c
X X
uD an2i and vD an12i :
i D0 i D0

If the discriminant s12 .u2  4v 2 / of the underlying quadratic polynomial is a square

in IFq , say u2  4v 2 D c 2 , we find tn2 D u˙c
2s . With a suitable choice of s 2 IFq such
a tn can hence be found in IFq . It remains to show that u2  4v 2 is in fact a square
in IFq . For this we first note the following identities:

u C 2v D .1/nC1 h.1/; u  2v D h.1/:

Multiplying these we obtain

Y
n
.1 C ci /2 .1  ci /2
u2  4v 2 D .1/n h.1/h.1/ D ;
i D1
ci2

which visibly is a square in IFq .

The specialization just constructed maps the Steinberg section SDn .t1 ; : : : ; tn ; s/
to an element Q1 , resp. Q2 2 SOC 2n .q/ which has the same characteristic polyno-
mial as 1 ; 2 respectively. According to the multiplicative Jordan decomposition
the order of Qi is then the order of i multiplied by a power of the characteristic p.
As the order of i is prime to p, there is some power of Qi of order prime to p
and with the same characteristic polynomial as i and which is hence GLn .IFq /-
conjugate to i .
The geometricity of N=K is obtained as in the proof of Theorem 3.4. Finally the
characteristic polynomial of .MDn ; ˚/ is computed from the algebraic system of
equations
q q
SDn .t1 ; : : : ; tn ; s/.X1 ; : : : ; X2n /t D .X1 ; : : : ; X2n /t
by solving for X WD X1 . t
u

Remark. The projective variant fNDn .Y / of fDn .X / generates a geometric Galois

extension with group PSOC
2n .q/ over K (compare Corollary 3.5).
420 V Additive Polynomials

3.5 The Dickson Groups G2 .q/

For the Dickson group G2 .q/ with q odd we use its smallest 7-dimensional repre-
sentation over IFq . Its image is contained inside the orthogonal group SOC
8 .q/ (see
Malle (2003), and Albert and Maier (2011), Ch. 6.2). With this one obtains:

Proposition 3.16. For q odd the group G2 .q/ possesses a Steinberg cross section of
the form 0 1
t u 1 : : : :
B1 : : : : : : C
B C
B : t 2 : t u 1 : C
B C
SG2 .t; u/ D B
B : 2t : 1 : : : C :
C
B : 1 : : : : : C
B C
@ : : : : t : 1A
: : : : 1 : :
Its characteristic polynomial is given by

hG2 .X / D .X  1/ X 6 C .t C 2/X 5 C .2 C 2t  u/X 4 C .2 C 2t  2u  t 2 /X 3

C .2 C 2t  u/X 2 C .t C 2/X C 1 :

Proof. We use the following root subgroups and representatives of simple reflec-
tions for the computation of the Steinberg cross section (see Malle (2003)):
0 1
  1 t t 2  
1t @ 1 t
X1 .t/ D diag. ; 0 1 2t A ; /;
01 0 1
00 1
   
1t 1 t
X2 .t/ D diag.1; ; 1; ; 1/;
01 0 1
0 1
  0 0 1  
0 1 @ 0 1
!1 D diag. ; 0 1 0 A ; /;
1 0 1 0
1 0 0
   
0 1 0 1
!2 D diag.1; ; 1; ; 1/:
1 0 1 0

The Steinberg cross section in the statement is obtained by expanding the product

SG2 .t; u/ D X1 .t/!1 X2 .u/!2 :

Its characteristic polynomial can be determined by elementary calculations. t

u

Proposition 3.17. (a) The group G2 .q/ with q odd contains elements 1 ; 2 of
orders q 2 ˙ q C 1.
(b) For q ¤ 3 any two such elements form a robust generating pair.
(c) For q D 3 there exists an element 3 2 G2 .3/ of order 8 such that .1 ; 2 ; 3 /
is a robust generating system.
3 Polynomials for Split Groups of Lie Type 421

Proof. For (a) let v 2 IFq be an element of order q 2 C q C 1, q 2  q C 1 respec-
2 2
tively. Then the characteristic polynomial gv .x/ of diag.v; v q ; v q ; v q ; v q ; v 1 /
is symmetric in IFq ŒX , i.e., it has the form

gv .X / D X 6  a1 X 5 C a2 X 4  a3 X 3 C a2 X 2  a1 X C 1;

with
2 2
a1 Dv C v q C v q C v q C v q C v 1 ;
2 2 2 2 2 2 q
a2 Dv 1q C v 1Cq C v 1q C v 1Cq C v qq C v qCq C v qq C v q
2 2 1
C v 1q C v q C v 1q C v q1 C 3;
2 2
a3 Dv 2 C v 2q C v 2q C v 2q C v 2q C v 2 C 2  2a1 D 2a2  2 C a12  2a1 :

Since here a3 satisfies the same relations as in the second factor of hG2 .X /, we
have that hv .X / WD .X  1/gv .X / can be obtained as a specialization over IFq from
hG2 .X /. Consequently G2 .q/ contains an element v with characteristic polynomial
2 2
hv .X /. This has Jordan normal form diag.v; v q ; v q ; 1; v q ; v q ; v 1 / over IFq and
hence order q C q C 1, respectively q  q C 1.
2 2

For the proof of (b) as in Section II.4.3, we use the list of maximal subgroups
compiled by Kleidman (1988b). According to this for q > 3 there is no maximal
subgroup containing elements of both orders q 2 C q C 1 and q 2  q C 1. For q D 3
there is no such subgroup which additionally contains a third element of order 8
(see Conway et al. (1985), p. 60). Such an element is obtained via the specialization
.t; u/ 7! .0; 1/. t
u

Theorem 3.18. Let q be odd.

(a) The F-module .MG2 ; ˚/ over K D IFq .t; u/ with representing matrix
DB .˚/ D SG2 .t; u/ is dualizable.
(b) The Galois group of .MG2 ; ˚/ is G2 .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial of .MG2 ; ˚/
7 2 q 3 6 2 q 2 5 2 q 2 4
fG2 .X / D X q C .t q C t q /X q C .t 2q  uq /X q  .uq t q C t 2q /X q
2 Cq 3 2 q 2 2 qC1 2 q
C .t q C uq /X q C .ut q  t q /X q  .t q C 1/X q  t q X:

Proof. The dualizability of .MG2 ; ˚/ follows from Proposition 3.16. Application of

Theorem 1.9 shows that G2 .q/ is an upper bound for Gal.MG2 ; ˚/. Proposition 3.17
with Corollary 2.13 then gives that G2 .q/ is also a lower bound for the Galois group,
whence Gal.MG2 ; ˚/ Š G2 .q/.
422 V Additive Polynomials

The geometricity of N=K follows as in Theorem 3.4. The characteristic poly-

nomial fG2 .X / finally can be computed with the Buchberger algorithm (in anti-
lexicographical ordering) from the system of algebraic equations

SG2 .t; u/.X1q ; : : : ; X7q /t D .X1 ; : : : ; X7 /t

with X WD X1 . t
u

For the easier case of even prime powers q one may use the 6-dimensional
representation of G2 .q/ over IFq embedding it into Sp6 .q/ (see Malle (2003),
resp. Albert and Maier (2011), Ch. 6.2). With this one obtains in a similar way
a Steinberg cross section of the form
0 1
t u 1 : : :
B1 : : : : : C
B 2 C
B: t : u 1 :C
B
SG2 .t; u/ D B
even C:
C
B: 1 : : : :C
@ : : : t : 1A
: : : 1 : :

This satisfies:

Proposition 3.19. Let q be even.

(a) The F-module .MG2 ; ˚/ over K D IFq .t; u/ with DB .˚/ D SGeven
2
.t; u/ is
dualizable.
(b) The Galois group of .MG2 ; ˚/ is G2 .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial of .MG2 ; ˚/
6 2 5 4 3 2
fG2 .X / D X q C t q X q C uq X q C t 2q X q C u X q C t X q C X:

The proof is left as an exercise to the interested reader; it can also be found in
Albert and Maier (2011), Ch. 6.2.
Somewhat more complicated polynomials for the groups G2 .q/ over IFq .t; u/
had already been constructed in a similar fashion by Malle (2003), Thms. 4.1
and 4.3.
Due to the large degrees of their minimal faithful representations, similar com-
putations for the large exceptional groups of types F4 ; E6 ; E7 and E8 become rather
unwieldy and thus corresponding polynomials have not (yet) been constructed.
4 Polynomials for Twisted Groups of Lie Type 423

4 Polynomials for Twisted Groups of Lie Type

In order to realize twisted groups of Lie type as Galois groups over IFq .t/ we
need a twisted variant of the upper bound from Theorem 1.9 and a corresponding
replacement for the Steinberg cross section, which we will call a pseudo Steinberg
cross section. Using this we will realize the unitary groups 2An .q/ D SUnC1 .q/, the
orthogonal groups 2Dn .q/ D SO 2 2
2n .q/ for q odd, the Suzuki groups B2 .q /, the Ree
2 2 3
groups G2 .q / and the Steinberg triality groups D4 .q/ for q odd as Galois groups.
Moreover we compute generating polynomials for the corresponding Galois exten-
sions. These results are all taken from the thesis of Maier and form part of Albert
and Maier (2011).

4.1 The Special Unitary Groups SUn .q/

Twisted groups of Lie type are defined as fixed point groups of simple algebraic
groups under Steinberg endomorphisms. Let G be a linear algebraic group defined
over IFq and ' W G ! G an endomorphism such that some power ' r is the Frobenius
endomorphism q on G. Then ' is a Steinberg endomorphism of G. Note that then
the group of fixed points G' under ' satisfies G' Gq D G.IFq /.
For the realization of twisted groups of Lie type we need a replacement for the
Steinberg cross section which takes into account Steinberg morphisms. The basis
for this is yielded by Maier’s variant of the upper bound in Theorem 1.9 (see Albert
and Maier (2011), Thm. 3.3):

Theorem 4.1 (Twisted Upper Bound Theorem). Let G be a connected linear alge-
braic group over IFq with a Steinberg endomorphism ' W G ! G such that ' r D q
for some r 2 IN. If .M; ˚/ is an F-module over K  IFq with basis B and if the
representing matrix DB .˚/ has the form

DB .˚/ D D0 '.D0 /    ' r1 .D0 /

for some D0 2 G.K/, then Gal.M; ˚/ G' .

Proof. The starting point is again the Theorem II.1.1 of Lang–Steinberg which guar-
antees the existence of a matrix Y 2 G.IFq / with '.Y / D D01 Y . This satisfies

q .Y / D ' r .Y / D' r1 .D01 Y / D ' r1 .D01 /' r1 .Y /

D' r1 .D01 /    '.D01 /D01 Y D DB .˚/1 Y:

By Theorem 1.9 there is an embedding

Gal.M; ˚/ ! G.IFq /;  7! C WD Y 1 .Y /:

424 V Additive Polynomials

Since ' and  commute and D0 2 G.K/ is  -invariant we obtain

'.C / D '.Y 1 .Y // D'.Y 1 /'..Y // D '.Y 1 /.'.Y //

DY 1 D0 .D01 Y / D Y 1 .Y / D C :

So C 2 G.IFq / is '-invariant for every  2 Gal.M; ˚/ and thus lies in G' . t

u

The special unitary groups can be obtained as fixed points SUn .q/ WD SL'n under
the Steinberg endomorphism

' W SLn ! SLn ; C 7! J  .q .C /t /1  J;

where J WD antidiag.1; : : : ; 1/. Here clearly we have ' 2 D q 2 . We now construct

a replacement SUn .t1 ; : : : ; tn / in SUn .q/ for the Steinberg cross section of the form
D0 '.D0 /, which will contain m WD b n2 c root subgroups of SLn .q 2 /. It is called
a pseudo Steinberg cross section. An easy calculation shows that in the nota-
q
Qm of Proposition 3.2 we have '.!i / D !ni and '.Xi .ti // D Xni .ti /. Thus
tion
i D1 Xi .ti /!i seems like a good choice for D0 .

Proposition 4.2. (a) For n D 2m C 1  3 the group G'Un D SUn .q/, .n; q/ ¤ .3; 2/,
possesses a pseudo Steinberg cross section of the form
0 q q q 1
t1 : : : tm1 tm .1/m
B 1 C
B C
B :: C
B : C
B C
B 1 0 C
B C
SUn .t1 ; : : : ; tm / D B
B .1/ m
tm 1 C:
C
B m
.1/ tm1 0 1 C
B C
B : : : C
B :: :: :: C
B C
@ .1/m t1 0 1A
.1/m 0

Its characteristic polynomial is given by

X
m X
m
hUn .X / D X 2mC1 C tiq X 2mC1i  ti X i  1:
i D1 i D1
4 Polynomials for Twisted Groups of Lie Type 425

(b) For n D 2m  4 the group G'Un D SUn .q/ possesses a pseudo Steinberg cross
section of the form
0 q q q q 1
t1 : : : tm2 tm1 tm .1/m
B 1 C
B C
B : C
B :: C
B C
B 1 C
B C
B
SUn .t1 ; : : : ; tm / D B m
.1/ tm 1 C:
C
B m
.1/ tm1 0 0 1 C
B C
B :: :: :: C
B : : : C
B C
@ m
.1/ t1 1A
.1/m 0

Its characteristic polynomial is given by

X
m1
q q q
hUn .X / D X 2m C .ti  ti 1 / X 2mi C  tm1  tm1 C .1/m tm
qC1
Xm
i D1

X
m1
C .ti  ti 1 / X i C 1 with t0 D 1:
i D1

Proof. In the case n D 2m C 1 in order to simplify our calculations we prefer to use

0 1
Im 0 q q 1
B m C t1 : : : tm .1/m
B .1/ 1 C B 1 C
B .1/ m
t 01 C B C
B 1 C B : C
D0 D B :: :: C C with '.D 0 / D B : : C
B : : B C
B C @ 1 0 A
@ .1/m tm1 1A
Im
.1/m tm 0
 
S Am 0
instead of X1 .t1 /!1    Xm .tm /!m D . Then we obtain SUn .t1 ; : : : ; tm /
0 Im
in (a) as the product D0  '.D0 /. Its characteristic polynomial is most easily calcu-
lated with the methods from Propositions 3.6 and 3.10.
In the case n D 2m we vary the above D0 by replacing the identity matrix Im
therein by Im1 . Setting SUn .t1 ; : : : ; tm / WD D0 '.D0 / we obtain the pseudo Stein-
berg cross section of SUn .q/ given in (b) with the characteristic polynomial hUn .X /
as stated. t
u

Proposition 4.3. For n > 2 we have:

(a) The groups GUn D SUn .q/ possess cyclic maximal tori T1 ; T2 of orders
q n .1/n
qC1
and q n1  .1/n1 .
(b) For .n; q/ … f.3; 2/; .4; 3/; .6; 2/g any two generators .1 ; 2 / of T1 ; T2 form
a robust generating system of GUn .
426 V Additive Polynomials

(c) Adding an element 3 of order 8 respectively 9 in the case .n; q/ D .4; 3/,
respectively .n; q/ D .6; 2/, the triple .1 ; 2 ; 3 / is a robust generating system
of GUn .

Proof. Assertion (a) follows from Malle, Saxl and Weigel (1994), Tab. III. Part (b)
for .n; q/ 2 f.3; 3/; .3; 5/; .5; 2/g can be read off immediately from Conway et al.
(1985), pp. 14, 34 and 72. For the other .n; q/ in (b) the proof is given in Malle,
Saxl and Weigel (1994), Thm. 3.1 (third paragraph). In the remaining cases .n; q/ 2
f.4; 3/; .6; 2/g in (c) a robust system as claimed is obtained using Conway et al.
(1985), pp. 54 and 115. t
u
For the proof of our main result the following observation will be useful:

Lemma 4.4. Let f .X / D X n Can1 X n1 C: : : Ca1 X C.1/n 2 IFq 2 ŒX be a sep-

arable polynomial with the property that x 2 IFq is a zero of f if and only if x q is.
q
Then we have ai D ani for all 1 i n  1.

Proof. The symmetry relation among the coefficients follows directly from the sym-
metry of the zeroes x1 ; : : : ; xn 2 IFq using the product relation x1    xn D 1. t
u

This completes our preparations for the proof of the main result of this section.
Theorem 4.5. Let n > 2 and q be a prime power, with .n; q/ ¤ .3; 2/. Then with
m WD b n2 c we have:
(a) The F-module .MUn ; ˚/ over K D IFq 2 .t1 ; : : : ; tm / with the representing
matrix DB .˚/ D SUn .t1 ; : : : ; tm / is dualizable.
(b) The Galois group of .MUn ; ˚/ is SUn .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial fUn .X / of .MUn ; ˚/, given by

X
m
.q 2 /mC1i
X
m
.q 2 /2mC1 .q 2 /2mC1i q 2 /i
fUn .X / D X C ti X  ti X .q X
i D1 i D1
for n D 2m C 1;
2 /2m
X
m1
2 /mC1i 3 q 2 mCi 1 2 /2mC1i
fUn .X / D X .q C ti.q  tm
q
ti.q1/ X .q
i D1
q2 q 3 q q3 q 2 .qC1/ 2 m
C  tm1 tm  tm1 C .1/m tm X .q /
X
m1
3
q 3 q q 2 i q 3 q
C  tiq1 C tm ti X .q / C tm X
i D1
for n D 2m

where t0 D 1.

Proof. Part (a) again follows directly from the shape of DB .˚/ in Proposition 4.2,
as does the claim that SUn .q/ is an upper bound for Gal.MUn ; ˚/ by Theorem 4.1.
4 Polynomials for Twisted Groups of Lie Type 427

For (b) it remains to show that SUn .t1 ; : : : ; tm / can be specialized to a robust
generating system of SUn .q/. For this let first n D 2m C 1 be odd. Then hUn .X /
is by Proposition 4.2(a) a general polynomial with the symmetry property from
Lemma 4.4. The elements 1 ; 2 from Proposition 4.3(b) have Jordan normal forms
2 n1
diag.v1 ; v1q ; v1q ; : : : ; v1.q/ /;
.q/n2 2 C:::.q/n2
diag.v2 ; v2q ; : : : ; v2 ; v21Cqq /

with elements vi 2 IFq of the same order as i (see Malle, Saxl and Weigel (1994)).
Thus the characteristic polynomials of 1 and 2 also enjoy the symmetry property
from Lemma 4.4. Consequently hUn .X / possesses specializations to the character-
istic polynomials of 1 and 2 . The same specializations applied to SUn .t1 ; : : : ; tm /
then yield a robust generating system of SUn .q/. In the case n D 2m the character-
istic polynomials of 1 ; 2 both have the shape

X
m1
q
X
m1
Xn C ai X ni C am X m C ai X i C 1
i D1 i D1

with ai 2 IFq 2 and am 2 IFq . Such a polynomial is obtained from hUn .X / by the
specialization ti 7! ui with

X
i
ui D 1 C aj for 1 i m1 and uqC1
m D .1/m .am C um1 C uqm1 /:
j D1

This is possible since the right hand side of the last equation is an element of IFq and
hence a q C 1st power in IFq 2 . The corresponding statement also holds for 3 in the
two exceptional cases in Proposition 4.3(c). Thus these specializations ti 7! ui of
SUn .t1 ; : : : ; tm / lead to a robust generating system .1 ; 2 / respectively .1 ; 2 ; 3 /
of SUn .q/. Hence, in both cases, n odd and n even, SUn .q/ is also a lower bound
for Gal.VUn ; ˚/ by Corollary 2.13.
The generating polynomial for the corresponding Galois extension is finally
obtained by solving the algebraic system of equations
2 2
SUn .X1q ; : : : ; Xm / D .X1 ; : : : ; Xm /t
q t

for X WD X1 . Its geometricity ensues as in Theorem 3.4. t

u

Remark. With the methods of the subsequent section it is also possible to con-
struct Frobenius modules .MQ Un ; ˚/ of dimension 2n with Galois group SUn .q/
over IFq .t1 ; : : : ; tn /. For this the computation of a generating additive polynomial
turns out to be difficult (see Albert and Maier (2011), Ch. 5.1). A general method
for the descent of ground fields will be presented in Paragraph 5.
428 V Additive Polynomials

Corollary 4.6. The fixed field of the center Z .SUn .q// in the Galois extension
N=K in Theorem 4.5 gives a Galois extension over IFq 2 .t/ with the simple group
Un .q/. It is generated by the roots of the projective variant fNUn .Y / of fUn .X /.

The nice polynomials for SUn .q/ and Un .q/ presented by Abhyankar (1996a)
for n odd and by Abhyankar and Inglis (2001) for n even can be obtained from the
polynomials fUn .X / in Theorem 4.5 and fNUn .Y / in Corollary 4.6 by specializing
ti 7! 0 for i D 1; : : : ; m  1 (and possibly changing the sign at X or Y , respectively).

4.2 The Orthogonal Groups SO

2n .q/

In order to obtain polynomials which can easily be calculated, in the case of

orthogonal groups SO 2n .q/ (with q odd) we replace the natural 2n-dimensional rep-
resentation of GDn by the one of the linear algebraic group GO2n obtained by field
restriction. (Thus this section can also be regarded as an explicit precursor exam-
ple for the method of field restriction presented in the next paragraph.) As we have
GO2n .IFq / D SO 
2n .q/ this will yield a Galois extension with the group SO2n .q/
over IFq .t/ (instead of over IFq 2 .t/).
The orthogonal group SO 2n .q/ is defined as the fixed point subgroup in SO2n
under the Steinberg endomorphism

' W SO2n ! SO2n ; C 7! N 1 q .C /N;

 
01
where N D diag.1; : : : ; 1 ; 1; : : : ; 1/ with a central 2-block (see Malle and
10
Testerman (2011), Ex. 22.9(2)). This clearly satisfies ' 2 D q 2 . In order to con-
struct a linear group GO2n with GO2n .IFq / D SOC 
2n .q / D SO2n .q/ we choose an
2 '

element y 2 IFq 2 with y D y. This satisfies IFq 2 D IFq .y/. As y 2 2 IFq we then
q

obtain an IFq -linear embedding

 
a.t/ y 2 b.t/
W IFq 2 .t/ ! IFq .t/22 ; a.t/ C yb.t/ 7! 2 IFq .t/22
b.t/ a.t/

and from this a homomorphic embedding

 W GL2n .IFq 2 .t// ! GL4n .IFq .t//; .cij / 7! .cij / :

Here the image H .IFq .t// WD .GL2n .IFq 2 .t/// is a linear algebraic group defined
over IFq , although  is not a morphism in the sense of algebraic geometry.
The non-trivial generating element 2 Gal.IFq 2 =IFq / satisfies .y/ D y q D y.
This extends to .GL2n .IFq 2 .t/// via
   
aij .t/ y 2 bij .t/ aij .t/ y 2 bij .t/
 W H .IFq .t// ! H .IFq .t//; 7! :
bij .t/ aij .t/ bij .t/ aij .t/
4 Polynomials for Twisted Groups of Lie Type 429

With this the defining relation C D N 1 q .C /N of SO 2n .q/ can be transferred

inside SOC 2n .q 2
/ to H .IFq .t// as .C / D .N / 1
 .C /.N /. For C D aij .t/ C
ybij .t/ 2 GL2n .IFq 2 .t// we now write

det.C / D f .aij .t/; bij .t// C yg.aij .t/; bij .t// DW f .C / C yg.C /

with polynomials f; g 2 IFq ŒXij ; Yij . With this let

GO2n .IFq .t// WD fC 2 H .IFq .t// j f .C / D 1; g.C / D 0; CQ .J /C D .J /;

C D .N /1  .C /.N /gI

here CQ is obtained from C by blockwise transposition of all of its 2  2-blocks. This

is a linear algebraic group defined over IFq .

Proposition 4.7. For n  4 and odd q the linear algebraic group GO2n satisfies:
(a) GO2n .IFq / Š SO 
2n .q/ and GO2n .IFq .t// Š SO2n .IFq 2 .t//,
(b) GO2n .IFq .t// Š SO2n .IFq .t//; in particular GO2n is connected.

Proof. The two claims in (a) follow directly from the definition of GO2n . Clearly
there exists a natural isomorphism over IFq .t/ of the form

GO2n Š f.C1 ; C2 / 2 SO2n  SO2n j C1 D N 1 C2 N g:

This shows that GO2n .IFq .t// Š SO2n .IFq .t// is connected. t
u

We can now easily obtain a pseudo Steinberg cross section for SO2n .q/ from the
one for SOC
2n .q/ in Proposition 3.13 by replacing tn1 and tn by tn1 C ytn and
tn1 C ytn respectively.

Corollary 4.8. Let n  4. The group SO

2n .q/ with q odd possesses a pseudo Stein-
berg cross section of the form

SO2n .t1 ; : : : ; tn ; s/ WD SDn .t1 ; : : : ; tn2 ; tn1 C ytn ; tn1 C ytn ; s/;

with characteristic polynomial

hO2n .X / D hDn .t1 ; : : : ; tn2 ; tn1 C ytn ; tn1 C ytn ; s/.X /:

Proof. By definition
˚ C
SO
2n .IFq 2 .t// D C 2 SO2n .IFq 2 .t// j N
1
q .C /N D C :
430 V Additive Polynomials

These matrices have the form

0 1
a1 .a1 /
B :: :: C
B C1 : : C2 C
B C
B a .a n1 /
C
B n1 C Ci 2 IFq .t/n1n1 ;
B b1 : : : bn c ::: c1 C
C DB B .b1 / : : : .cn /
n C with
B .bn / ::: .c1 /C
C
B C ai ; bi ; ci ; di 2 IFq 2 .t/:
B dn1 .dn1 / C
B :: :: C
@ C 3 : : C4 A
d1 .d1 /

So according to Proposition 3.13

SDn .t1 ; : : : ; tn2 ; tn1 C ytn ; tn1 C ytn ; s/

is an element of SO
2n .IFq 2 .t; s// and hence also a pseudo Steinberg cross section of
SO 2n .q/. t
u
Proposition 4.9. Let n  4 and q be odd. Then:
(a) The group SO2n .q/ possesses maximal tori T1 and T2 of orders q C 1 and
n

.q n1
C 1/.q  1/.
(b) Any two elements 1 ; 2 of maximal order in T1 , T2 respectively form a robust
generating pair of SO2n .q/.
Proof. Assertion (a) can be found in Malle, Saxl and Weigel (1994), Tab. III. The
n1 n1
elements in T1 ; T2 have Jordan normal forms diag.v; v q ; : : : ; v q ; v q ; : : : ; v 1 /
q n C1 q n2 1 q n2
with v D 1, resp. diag.v; v ; : : : ; v
q
; w; w ; v ; : : : ; v 1 / with
n1
v q C1 D 1 and w q1 D 1.
Assertion (b) follows from Malle, Saxl and Weigel (1994), proof of Thm. 3.1
(Paragraph 7). (Note that we assume q to be odd here.) t
u
Theorem 4.10. Let n  4 and q be odd.
(a) The F-module .MO2n ; ˚/ over K D IFq .t1 ; : : : ; tn ; s/ with representing matrix
DB .˚/ D .SO2n .t1 ; : : : ; tn ; s// is dualizable.
(b) The Galois group of .MO2n ; ˚/ over K is SO 2n .q/.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial of .MO2n ; ˚/.
Proof. As in the proof of Theorem 4.5 it only remains to show that .SO2n .t; s//
can be specialized to elements of maximal order in the two tori T1 and T2 from
Proposition 4.9. These then form a robust system of generators for GO2n .IFq / Š
SO2n .q/. Since the upper bound from Theorem 1.9 and the lower bound from Corol-
lary 2.13 do agree we then conclude that Gal.MO2n ; ˚/ Š SO 2n .q/.
For simplicity we first work with the 2n-dimensional natural representation
SO2n .t; s/ instead of the 4n-dimensional .SO2n .t; s//. According to Proposition 4.9
the characteristic polynomials h .X / of  2 f1 ; 2 g are separable polynomials in
4 Polynomials for Twisted Groups of Lie Type 431

IFq ŒX and symmetric, since with x 2 IFq also x 1 is a zero of h .X /. In particular

it suffices to identify the h .X / as specializations of hO2n .X /. For this let

X
2n
h .X / D ai X i with a2ni D ai 2 IFq :
i D0

By Proposition 3.13 and Corollary 4.8 we have

1 X n1
hO2n .X / D .Qsi  sQi 2 / .X 2ni C X i /  .2tQn2 C tQn1
2
C tQn2 / X n
s
i D0

with tQn1 D tn1 C ytn , tQn D tn1 C ytn , sQ2 D sQ1 D 0, sQ0 D 1s , sQ1 D ts1 , sQi D ti
for i D 2; : : : ; n  2, and sQn1 D tQn1 tQn D y 2 tn2  tn1
2
. From this we reach h .X /
by specializing the ti to ui and s to r 2 IFq such that

u1 D a1 ; rui D .ai C ai 2 C : : :/ for i D 2; : : : ; n  2;

r.y 2 u2n  u2n1 / D run1 C an1 D an1 C an3 C : : : ;
1 1 1
r.y 2 u2n C u2n1 / D un2  an D  an  .an2 C an4 C : : :/:
2 2s s
Since
h .1/ D 4ru2n1 and h .1/ D .1/n1 4ry 2 u2n
this is possible if and only if

Y
n
.1  xi /2 .1 C xi /2 Y
n
.1/n h .1/h .1/y 2 D y 2 D y2 .xi1  xi /2
i D1
xi2 i D1
Qn
is a square in IFq , that is, if ´ WD y i D1 .xi1 xi / is an element of IFq . But the latter
is easily verified using the zeroes of h1 .X / and h2 .X / given in Proposition 4.9.
Since the ui and r are elements of IFq ,  commutes with the specialization ti 7!
ui . So .1 / and .2 / also form a robust generating system of .SO 2n .q//. t
u
In fact, the characteristic polynomials of h. /,  2 f1 ; 2 g can easily be com-
puted. As will be shown in Corollary 5.3 they can be obtained as the product

h. / .X / D h .X /h .X / D h .X /2 :

Remark. The generating additive polynomial in Theorem 4.10(c) is computable

explicitly from

.SO2n .t1 ; : : : ; tn ; s//.X1q ; : : : ; X4n

q t
/ D .X1 ; : : : ; X4n /t

with X WD X1 , but it extends over more than half a page. The interested reader
can find it in Albert and Maier (2011), Thm. 5.3. Its projective variant generates a
geometric Galois extension with group PSO 2n .q/ over K.
432 V Additive Polynomials

4.3 The Suzuki Groups 2B2 .q 2 /

The Suzuki groups are best defined as fixed point subgroups inside the 4-dimensional
symplectic group GC WD GC2 D Sp4 .k/ over an algebraically closed field k of char-
acteristic 2. According to a result of Chevalley (see Geck (2003), Thm. 3.3.6) there
exists a uniquely determined isomorphism of algebraic groups
0 1
1 t1 t3 C t1 t2 t4 C t1 t3
B0 1 t2 t3 C
' W GC ! GC ; A.t1 ; : : : ; t4 / WD B
@0 0
C
1 t4 A
0 0 0 t1
0 1
1 t2 t4 t3 C t2 t4
2
B 0 1 t12 t4 C t12 t2 C
7! A0 WD B @0 0 1
C
t2 A
0 0 0 1

with '.At / D .A0 /t . This satisfies

' 2 .A.t1 ; t2 ; t3 ; t4 // D A.t12 ; t22 ; t32 ; t42 /;

hence ' 2 D 2 . Setting

'q WD ' ı 2l for q 2 D 22lC1

(see Section II.4.5), then in analogy to the previous section we obtain 'q2 D q 2 , i.e.,
'q is a (twisted) Steinberg morphism of GC . Its group of fixed points
'
2
B2 .q 2 / WD GCq with q 2 D 22lC1

is the Suzuki group as in Section II.4.5 of order q 4 .q 2  1/.q 4 C 1/.

Proposition 4.11. The Suzuki group 2B2 .q 2 / for q 2 D 22lC1 and l  1 has a pseudo
Steinberg cross section 0 1
lC1
t t2 1 :
B1 : : : C
SSuz .t/ WD B
@ : t : 1A
C

: 1 : :
of the form SSuz .t/ D D0  'q .D0 /. Its characteristic polynomial is given by

hSuz .X / D X 4 C t X 3 C t 2lC1 X 2 C t X C 1:

Proof. We use the root subgroups X1 ; X2 and the Weyl group representatives !1 ; !2
for Sp4 .q 2 / from the proof of Proposition 3.6 for n D 2. These satisfy
lC1
'q .X1 .t// DX2 .t 2 /; 'q .!1 / D !2 ;
2l
'q .X2 .t// DX1 .t /; 'q .!2 / D !1 :
4 Polynomials for Twisted Groups of Lie Type 433

So if we set D0 WD X1 .t/!1 then with

lC1
SSuz .t/ WD D0  'q .D0 / D X1 .t/!1 X2 .t 2 /!2

we obtain the stated pseudo Steinberg cross section as well as its characteristic poly-
nomial. t
u
Proposition 4.12. For q 2 D 22lC1 with l  1 we have:
(a) The Suzuki group 2B2 .q 2 / has cyclic maximal tori T1 ; T2 of respective orders
q ˙ 2lC1 C 1.
2

(b) Any two generators of T1 and T2 form a robust generating system of 2B2 .q 2 /.
Proof. Both claims follows almost immediately from the table of maximal sub-
groups of 2B2 .q 2 / determined by Suzuki (1962). t
u
After these preparations we can now formulate the main result of this section.
Theorem 4.13. For q 2 D 22lC1 with l  1 we have:
(a) The F-module .MSuz ; ˚/ over K D IFq 2 .t/ with representing matrix DB .˚/ D
SSuz .t/ is dualizable.
(b) The Galois group of .MSuz ; ˚/ over K is 2B2 .q 2 /.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial of .MSuz ; ˚/
8 2 6 lC1 4 2
fSuz .X / D X q C t q X q C t 2 X q C t X q C X:

Proof. Again for parts (a) and (b) we only need to verify that 2B2 .q 2 / is a lower
bound for Gal.MSuz ; ˚/. For this let v1 ; v2 2 IF2 be elements of orders
q 2 ˙ 2lC1 C 1. As .q 2 C 2lC1 C 1/.q 2  2lC1 C 1/ D q 4 C 1, their minimal poly-
nomials hi .X / over IFq 2 have the form
2 4 6
hi .X / D .X  vi /.X  viq /.X  viq /.X  viq /
2 2
D .X  vi /.X  viq /.X  vi1 /.X  viq /:

Thus, these are symmetric polynomials, i.e.,

hi .X / D X 4 C ai X 3 C bi X 2 C ai X C 1
2 2 2 2 2
with ai D vi C vi1 C viq C viq 2 IFq 2 and bi D v11Cq C vi1q C vi1Cq C
2
vi1q 2 IFq 2 . Since
2 2 4 Cq 2 4 q 2
D vi1Cq C vi1q C viq C viq
lC1
ai2 D bi

both hi .X / are specializations of hSuz .X / for t 7! ai . According to Proposition 4.11

the same specializations, applied to the matrix SSuz .t/, yield a robust generating
system of 2B2 .q 2 /. So 2B2 .q 2 / is a lower bound for Gal.MSuz ; ˚/ by Corollary 2.13.
434 V Additive Polynomials

Again a generating polynomial is obtained by elimination from the algebraic

system of equations
q2 q2
SSuz .t/.X1 ; : : : ; X4 /t D .X1 ; : : : ; X4 /t

with X WD X1 . t
u

Remark. The assertion of Theorem 4.13 continues to hold in the solvable case q 2 D
2. This was checked in Garcia Lopez (2010) with the algorithm for the computation
of Galois groups of additive polynomials presented in Paragraph 2.

4.4 The Ree Groups 2 G2 .q 2 /

Here we start from the simple algebraic group GG WD GG2 .k/ over an algebraically
closed field k of characteristic 3. Then by Section II.4.2, there exists a Steinberg
morphism

'q W GG ! GG ; with 'q2 D q 2 for q 2 D 32lC1

(compare also Kemper, Lübeck and Magaard (2001)). Its fixed point subgroup is
called the Ree group
'
2
G2 .q 2 / WD GGq with q 2 D 32lC1 ;

of order q 6 .q 2  1/.q 6 C 1/. For this a pseudo Steinberg cross section can be deter-
mined completely analogously to the case of the Suzuki groups in the previous sec-
tion:

Proposition 4.14. For the Ree groups 2 G2 .q 2 / with q 2 D 32lC1 , l  0,

lC1
SRee .t/ WD SG2 .t; t 3 /

is a pseudo Steinberg cross section of the form SRee .t/ D D0  'q .D0 /. Its charac-
teristic polynomial is given by
lC1
hRee .X / D hG2 .t; t 3 /.X /:

Proof. We use the root subgroups X1 ; X2 and Weyl group representatives of G2 .q 2 /

in the proof of Proposition 3.16. They satisfy
lC1
'q .X1 .t// DX2 .t 3 /; 'q .!1 / D !2 ;
3l
'q .X2 .t// DX1 .t /; 'q .!2 / D !1
4 Polynomials for Twisted Groups of Lie Type 435

(see Steinberg (1967), p. 178). With D0 WD X1 .t/!1 then

lC1
SRee .t/ WD D0  'q .D0 / D X1 .t/!1 X2 .t 3 /!2

gives the stated pseudo Steinberg cross section as well as its characteristic polyno-
mial. t
u

Proposition 4.15. For q 2 D 32lC1 with l  0 we have:

(a) The Ree group 2 G2 .q 2 / has cyclic maximal tori T1 and T2 of orders
q ˙ 3lC1 C 1.
2

(b) For l > 0 any two generators of T1 and T2 form a robust generating system
of 2 G2 .q 2 /.
(c) For q 2 D 3 every triple .1 ; 2 ; 3 / of elements of 2 G2 .3/ of orders 6,7, and 9
is a robust generating system.

Proof. Assertions (a) and (b) follow directly from the list of maximal subgroups
given in Kleidman (1988b). The claim in (c) is easily checked for example using
Conway et al. (1985), p. 6. t
u

Theorem 4.16. For q 2 D 32lC1 with l  0 we have:

(a) The F-module .MRee ; ˚/ over K D IFq 2 .t/ with representing matrix DB .˚/ D
SRee .t/ is dualizable.
(b) The Galois group of .MRee ; ˚/ over K is 2 G2 .q 2 /.
(c) The corresponding Galois extension N=K is geometric and is generated by
the zeroes of the characteristic polynomial fRee .X / of .MRee ; ˚/, which is given by
lC1
fRee .X / D fG2 .t; t 3 /.X /:

Proof. By our previous considerations it only remains to show that SRee .t/ special-
izes to a robust generating system of 2 G2 .q 2 /. In the case when l D 0, so q 2 D 3,
the specializations t 7! 0; 1; 1 yield elements of orders 6,9,7 respectively and thus
by Proposition 4.16(c) a robust generating system. For l  1 we choose primitive
roots of unity v1 ; v2 in IFq of orders q 2 ˙ 3lC1 C 1 and then follow the lines of the
argument in the proof of Theorem 4.13. t
u

4.5 The Steinberg Triality Groups 3D4 .q/

We next consider the triality groups 3D4 .q/ for odd q. Here our starting point is the
group G WD GD4 D SO8 .k/ over an algebraically closed field k of odd characteristic,
N WD G=Z .G/ Š PSO8 .k/. Denoting the images
respectively its projective version G
N
of root subgroups Xi of G in G by X N i , there exists an automorphism ' of algebraic
436 V Additive Polynomials

groups uniquely determined up to inner automorphisms such that

N ! G;
'WG N with '.XN 2 / D XN 2 and ' W XN 1 ! XN 3 ! XN 4 ! XN 1

(see Deriziotis and Michler (1987)). The Steinberg triality groups are then obtained
as fixed point subgroups of the endomorphisms 'q WD ' ı q :
3 N 'q
D4 .q/ WD G with 'q3 D q 3 :

By construction these are subgroups of PSOC 3 3

8 .q /. The full preimage of D4 .q/
C 3
in SO8 .q / will be denoted by H .q/. This is a direct product of D4 .q/ with
3

Z .SOC 3
8 .q //. In contrast to Section 3.4 here we choose

!i WD Xi .1/Xi .1/Xi .1/

as representatives for the simple reflections for SO8 , where Xi denotes the negative
root subgroup of Xi . Then for the projective image DN 0 of

D0 WD X1 .t/!1 X2 .u/!2 2 SOC

8 .IFq 3 .t; u//

we obtain
'q .DN 0 / DXN 3 .t q /!N 3 XN 2 .uq /!N 2 ;
'q2 .DN 0 / DXN 4 .t q /!N 4 XN 2 .uq /!N 2 :
2 2

Thus, SNTri .t; u/ WD DN 0 'q .DN 0 /'q2 .DN 0 / is a pseudo Steinberg cross section for 3D4 .q/
and
2 2
STri .t; u/ WD X1 .t/!1 X2 .u/!2 X3 .t q /!3 X2 .uq /!2 X4 .t q /!4 X2 .uq /!2

is a preimage of SNTri .t; u/ in SOC

8 .IFq 3 .t; u//. Explicit calculation shows that
0 2 2
1
t t q C uqC1 t q u uq 1 :
B1 : C
B : : : : : : C
B 2 2 C
B : t q  uq Cq uq : 1 : : : C
B C
B uq
2
: C
STri .t; u/ D B : 1 : : : : C
B : t q
: u q
: 1 : : C
B C
B : u : 1 : : : : C
B C
@ : t : : : : : 1A
: 1 : : : : : :
2 2 2 CqC1
with D .t q u C t q uq  uq /: This achieves the proof of the first part of
the next statement:

Proposition 4.17. The image SNTri .t; u/ of STri .t; u/ in PSOC 3

8 .q / is a pseudo Stein-
3
berg cross section for D4 .q/. Its characteristic polynomial is symmetric and has
4 Polynomials for Twisted Groups of Lie Type 437

the form

hTri .X / D X 8 C .t C uq /.X 7 C X /

2 2 2
C .t q u  tuq  t q uq C u1CqCq /.X 6 C X 2 /
2 2 2 2
C .t  t qCq  u1Cq  uq C t q u C t q uq /.X 5 C X 3 /
2 2 2
C .2 C 2tuq C u2 C u2q  2u1CqCq C t 2q C t 2q / X 4 C 1:

Proof. During the computation of the coefficientsPof hTri .X / from STri .t; u/ it turns
out that these are symmetric, i.e., that hTri .X / D 8iD0 ai X i with a8i D ai . t
u

Proposition 4.18. (a) The group 3D4 .q/ has a cyclic maximal torus TN 1 of order
q 4  q 2 C 1 and a bicyclic maximal torus TN 2 of order .q 2 C q C 1/2 .
(b) Every pair of elements N1 ; N2 of TN 1 ; TN 2 of maximal order is a robust generat-
ing system of 3D4 .q/.

Proof. Both assertions are immediate from the list of maximal subgroups of 3D4 .q/
determined in Kleidman (1988a). Here elements of TN 1 have preimages in H .q/
with Jordan normal form
3 3 Cq 2 Cq1 2 q 2 Cq 3 q 2 qC1 3
diag.v1 ; v1q ; v1q ; v1q ; v1q ; v1q ; v1q ; v11 /;

respectively

diag.v2 ; v21 v3 ; v21 v31Cq ; v22q1 v3q ; v22qC1 v3q ; v2 v3q1 ; v2 v31 ; v21 /;

where v1 is a primitive root of unity of order q 4  q 2 C 1, and v2 ; v3 are roots of


unity of order q 2 C q C 1 in IFq (see Deriziotis and Michler (1987), Table 1.1). u t

In order to recognize elements of maximal (odd) order in the preimages Ti in

H .q/ of TN i as specializations of STri .t; u/ the following observation of P. Müller is
very useful:

Lemma 4.19. For every y; ´ 2 IFq with q odd there exists x 2 IFq 3 n IFq with

trIFq3 =IFq .x.x C y// C NIFq3 =IFq .x/ D ´:

Proof. The minimal polynomial of an element x 2 IFq 3 n IFq over IFq has the form
2
gx .X / D .X  x/.X  x q /.X  x q / D X 3  a.x/X 2 C b.x/X  c.x/
2 2 2 C1 2
with a.x/ D x C x q C x q , b.x/ D x 1Cq C x qCq C x q and c.x/ D x 1CqCq .
So for y 2 IFq we have

tr.x.x C y// D a.x/2  2b.x/ C ya.x/:

438 V Additive Polynomials

We hence need to find an element x 2 IFq 3 n IFq with c.x/ D a.x/2  2b.x/ C
ya.x/  ´. For this it suffices to prove the existence of elements a; b 2 IFq such that

fa;b .X / WD X 3  aX 2 C bX C .a2  2b C ya  ´/

is irreducible in IFq ŒX . Indeed, then every zero x of fa;b .X / is a solution to our

problem. We first observe that there certainly is a 2 IFq with

fa;b .2/ D 8  4a C a2 C ya  ´ ¤ 0

(independent of b 2 IFq ). Furthermore, for every v 2 IFq n f2g there exists a unique
b 2 IFq satisfying fa;b .v/ D 0. So there are at most q  1 elements b 2 IFq such that
fa;b .X / (with a as chosen above) has a zero in IFq n f2g. Thus, we can find b 2 IFq
for which fa;b .X / has no zero in IFq and hence is irreducible. t
u

Corollary 4.20. Let q be odd. Then for i D 1; 2 the characteristic polynomials

hi .X / WD hi .X / are specializations of hTri .X /.

Proof. According to Proposition 4.18 the characteristic polynomials hi .X / of

preimages i 2 H .q/ of Ni of odd order are symmetric for i D 1; 2, i.e., they are of
the form

hi .X / D X 8 C ai .X 7 C X / C bi .X 6 C X 2 / C ci .X 5 C X 3 / C di X 4 C 1

with coefficients in IFq 3 . Moreover we have bi 2 IFq , and

2 Cq 2
ci D aiq  ai and di D ai2q C ai2q  2bi  2;

as can be seen by elementary calculations. Now let xi 2 IFq 3 n IFq be the solutions
q2
from Lemma 4.19 for .y; ´/ D .ai ; bi /. Then the specializations

i W IFq 3 Œt; u ! IFq 3 ; t 7! ai  xiq ; u 7! xi ;

satisfy
i .a/ Di .t C uq / D ai C xiq  xiq D ai ;
2
i .b/ D  tr.xi .xi C aiq //  N .xi / D bi :
Using the above formulas for ci and di this yields

i .c/ D ci and i .d / D di

for the corresponding coefficients a; b in hTri .X /. Hence we have i .hTri .X // D

hi .X /. Consequently STri .t; u/ can be specialized to elements Qi 2 H .q/, i D 1; 2,
with characteristic polynomial hi .X /. t
u
4 Polynomials for Twisted Groups of Lie Type 439

For the construction of a Frobenius module for 3D4 .q/ we use the 64-dimensional
natural linear representation

W PSO8 .IFq .t; u// ! GL64 .IFq .t; u//

given by conjugation.

Theorem 4.21. Let q be odd.

(a) The F-module .MTri ; ˚/ with DB .˚/ D .SNTri / over K D IFq 3 .t; u/ is dual-
izable.
(b) The Galois group of .MTri ; ˚/ is 3D4 .q/.
(c) The corresponding Galois extension N=K is geometric and can be generated
by the zeroes of the characteristic polynomial of .MTri ; ˚/.

Proof. It suffices to show that .SNTri .t; u// can be specialized to a robust gen-
erating system of .3D4 .q//. By Corollary 4.20, STri .t; u/ has specializations
Q1 ; Q2 2 H .q/ with characteristic polynomial hi .X /, for i D 1; 2. Their images
Ni in 3D4 .q/ have order divisible by q 4  q 2 C 1, q 2 C q C 1 respectively and thus
by Proposition 4.18 form a robust generating pair of 3D4 .q/. As commutes with
the specializations i , this property is preserved by passage to the 64-dimensional
representation of 3D4 .q/. t
u

The calculation of the generating polynomial of N=K of degree .q 3 /64 from the
equation
.SNTri /.X1q ; : : : ; X64
3 q3 t
/ D .X1 ; : : : ; X64 /t
is possible in principle but at present seems out of reach computationally.
440 V Additive Polynomials

5 Field Restriction in Modular Galois Theory

Modular Galois theory has not only the advantage of the existence of non-trivial
upper bounds for Galois groups, which has no analogue in characteristic zero, but
also possesses the method of field restriction. Here one starts from a Galois exten-
sion N=K with group G.k/ over a field of constants k D IFq , which is given by an
effective Frobenius module .M; ˚/ over K, that is, the representing matrix of the
Frobenius operator ˚ is an element of G.K/. Then via field restriction the under-
lying representation of the linear algebraic group G over IFq can be transformed to
a representation over IFp . Since this transformation is compatible with specializa-
tions used to derive lower bounds, this results for example in the fact that all Galois
realizations from the previous two sections over the natural field of definition IFq of
G can even be realized over the prime field IFp ; the corresponding generating poly-
nomials are much harder to compute then, though. This method also seems to have
been in the intention of Abhyankar in his paper Abhyankar and Keskar (2001). Our
presentation here follows Stichel (2014).

5.1 Base Field Reduction

We start with an overview on some known results concerning field restriction. For
this let K be a field, say K D IFp .t1 ; : : : ; tn /, and L=K a finite field extension as for
example L D IFq .t1 ; : : : ; tn / with q D p l . Then the following fact is known from the
theory of linear algebraic groups:

Theorem 5.1. Let K be a field, L=K a finite Galois extension and G D GL a linear
algebraic group over L. Then we have:
(a) There exists a linear algebraic group G D GK over K and an L-homomor-
phism  W GK ! GL with the following universal mapping property: for all affine
varieties YK over K with an L-morphism % W YK ! GL there exists exactly one
K-morphism  W YK ! GK with % D  ı .

D
YK GK∗

° Y

GL

(b) If % in (a) is an L-homomorphism, then  is an L-homomorphism.

(c) If GL in (a) is connected then GK is also connected.

Proof. The proof is composed of several result in Springer (1998), namely of

Thm. 11.4.16 with Exerc. 11.4.20(1) for (a), of Prop. 12.4.2 for (b) and of
Exerc. 12.4.7(3) for (c). t
u
5 Field Restriction in Modular Galois Theory 441

Remark. From the construction of G , with G D Gal.L=K/ we obtain the relation

O G
KŒG Š LŒG
2G

between the coordinate rings.

The algebraic group GK given by Theorem 5.1 is said to be obtained by field
restriction with respect to L=K from GL . By the universal property it is determined
uniquely up to K-isomorphisms. More precisely we may also denote it as

K .GL / WD GK :
ResL

It enjoys the following properties:

Corollary 5.2. Let L=K be a finite Galois extension with group G D Gal.L=K/, G
a linear algebraic group over L and G D ResL K .G/. Then:

(a) G .K / Š
sep
QG.K ˝K L/.
sep

(b) G .L/ Š 2G G .L/ and  is the projection onto the factor indexed by
the identity element of G.

Proof. Again the results are taken from Springer (1998), Ch. 12.4.4 for (a), and
Prop 11.4.22 and Ch. 12.4.5 for (b). t
u

In order to make field restriction somewhat more concrete we choose a basis

y1 ; : : : ; yl of L=K. Then there exists a K-linear isomorphism

M
l X
l
W L ! Kyi ; ai yi 7! a1 y1 C : : : C al yl ;
i D1 i D1

inducing for all m  1 a K-linear embedding

 W GLm .L/ ! GLml .K/;  2 Aut.Lm / 7! . / 2 Aut.K ml /;

with matrices  with respect to a basis x1 ; : : : ; xm and . / with respect to the basis
fxi yj g. With these notations we obtain:

Corollary 5.3. Let L=K be a finite Galois extension with group G and G a linear
algebraic group over L. Then:
(a) G .K/ Š .G.L// as groups.
(b) For any A 2 G.L/ the characteristic polynomial of .A/ splits as
Y
h.A/ .X / D hA .X /:
2G

Proof. Let HK WD .G.L// be the finite linear algebraic group defined over K and
% W HK ! G the embedding defined over L. Then by Theorem 5.1 there exists a
442 V Additive Polynomials

K-homomorphism  W HK ! G with  ı  D % . From this we obtain a sequence

of homomorphisms
 Y
HK ! G .K/ ,! G .L/ ! G .L/;
2G

where , according to Corollary 5.2(b), is the projection onto the factor with  D 1.
Due to  ı D % we hence have .G.L// Š G.L/, and  is injective. As moreover
is injective on G .K/ by Springer (1998), Prop. 12.4.6, we have G.L/ Š G .K/.
This shows (a).
Assertion (b) follows immediately from Corollary 5.2(b), as taking the charac-
teristic polynomial is compatible with extension of constants. t
u

5.2 Application to Groups of Lie Type

We start with a typical example, namely the Galois extension N=L with group
GAn .IFq / D SLnC1 .q/ constructed in Section 3.1. This originated from a Frobe-
nius module .MAn ; ˚/ with DB .˚/ D SAn .t1 ; : : : ; tn / over L D IFq .t1 ; : : : ; tn /.
Now we choose a basis y1 ; : : : ; yl of IFq =IFp , which is also a basis of L=K for
P
K D IFp .t1 ; : : : ; tn /, and replace the transcendentals ti by ti WD lj D1 tij yj with
new transcendentals tij . Then from Proposition 3.2 we obtain the characteristic
polynomial

X
n
hAn .t1 ; : : : ; tn / D X nC1 C .1/i 1 ti X nC1i C .1/nC1 :
i D1

Let us write t D .t1 ; : : : ; tn / and t D .t11 ; : : : ; tnl /. The -image of the Steinberg
cross section SAn .t1 ; : : : ; tn / will be denoted by

SAn .t/ WD .SAn .t // 2 G .IFp .t//

with G D ResL
K .SLnC1 /, and its characteristic polynomial by
Y
hAn .t/ WD hAn .t /
2G

where G D Gal.IFq =IFp /. Here, the Galois action is given by

P
.ti / D lj D1 tij .yj /.
Now let .MAn ; ˚/ be the l.n C 1/-dimensional F-module over K  D IFp .t/ with
DB .˚/ D SAn .t/ 2 .SLnC1 .IFq .t ///. This is obviously dualizable. Now SLnC1

is connected, hence so is ResL K .SLnC1 / by Theorem 5.1(c). Thus Gal.MAn ; ˚/
5 Field Restriction in Modular Galois Theory 443

by Theorem 1.9 becomes a subgroup of G .IFp / Š .SLnC1 .q// Š SLnC1 .q/.

We now want to compare the lower bounds. For this let .1 ; 2 /, respectively
.1 ; 2 ; 3 / be the robust generating system of SLnC1 .q/ from Proposition 3.3.
Due to .G.IFq // Š G .IFp / Š G.IFq / their images .i / form a robust gen-
erating system of .SLnC1 .q//. Now let ti 7! ci 2 IFq denote the specialization
in Theorem 3.4, which maps SAn .t/ to an element conjugate to 1 , and write
P
ci D lj D1 cij yj with cij 2 IFp . Then the specialization tij 7! cij 2 IFp maps SAn .t/
to an element that is .GLnC1 .q//-conjugate to .1 /. The analogous observation
applies to 2 and 3 . So by Corollary 2.13 the group .SLnC1 .q// is also a lower
bound for Gal.MAn ; ˚/. Consequently, we have

Gal.MAn ; ˚/ D SLnC1 .q/;

and SLnC1 .q/ occurs as the Galois group of a geometric Galois extension over
K  D IFp .t11 ; : : : ; tnl /.
Completely analogous considerations lead to the following statement:

Theorem 5.4. Let G be one of the groups of Lie type

SLnC1 .q/; Sp2n .q/; SO2nC1 .q/; SOC

2n .q/; G2 .q/; SUn .q/;
SO 2 2 2 2 3
2n .q/; B2 .q /; G2 .q /; D4 .q/

Q as considered in Paragraphs 3
with their corresponding representations in GLm .q/
and 4 (where qQ D q or qQ D q 2 according to the cases). Then writing qQ D p l we
have:
(a) The Frobenius module .MG ; ˚/ with DB .˚/ D SG .t/ 2 .G.IFqQ .t /// over

K D IFp .t/ is dualizable.
(b) The Galois group Gal.MG ; ˚/ over K  is G.IFqQ /.
(c) The corresponding Galois extension N  =K  is generated by a characteristic
polynomial of .MG ; ˚/. It can be computed from the algebraic system

SG .t/.X1p ; : : : ; Xml

p t
/ D .X1 ; : : : ; Xml /t :

Proof. The proof can in all cases be achieved in complete analogy to the above
P
example of SLnC1 .q/. Here s should — if it occurs — be split up as s D lj D1 sj yj
into the additional transcendentals s1 ; : : : ; sl . t
u

5.3 Explicit Polynomials for SLnC1 .q/

In this section we present polynomials for some special Galois extensions obtained
by field restriction.
444 V Additive Polynomials

We start with the group GAn .IFq / D SLnC1 .q/ for q D 4. For this let IF4 D IF2 .x/
with x 2 C x D 1, and set ti D ti C ui x. Then we have
0 1
t1 : : : tn 1 u1 ::: un 0
B1 0 C
B C
B :: C
B : C
B C
B 1 0 C
SAn .t; u/ D .SAn .t // D B
 
Bu1
C:
B : : : un 0 t1 C u1 : : : tn C un 1C
C
B 1 0C
B C
B :: C
@ : A
1 0

From the corresponding system of algebraic equations

SAn .t; u/.X12 ; : : : ; XnC1

2
; Y12 ; : : : ; YnC1
2
/t D .X1 ; : : : ; YnC1 /t

one obtains the equations

X
n X
n
ti Xi2 C XnC1
2
C ui Yi2 D X1 ; Xi2 D Xi C1 for i D 1; : : : ; n;
i D1 i D1
X
n X
n
ui Xi2 C .ti C ui /Yi2 C YnC1
2
D Y1 ; Yi2 D Yi C1 for i D 1; : : : ; n:
i D1 i D1

i 1 i 1
This first yields Xi D X 2 , Yi D Y 2 for X D X1 , Y D Y1 , and then we obtain
from Theorem 5.4:

Proposition 5.5. Let f .X / respectively g.Y / be the polynomial which is obtained

by elimination of Y resp. of X from the system

X
n X
n
2i 2nC1 i
ti X CX C ui Y 2 D X;
i D1 i D1
X
n X
n
2i i nC1
ui X C .ti C ui /Y 2 C Y 2 D Y:
i D1 i D1

Then the field NAn generated over IF2 Œt; u by the zeroes of f , respectively g, has
Galois group
Gal.NAn =IF2 .t; u// Š SLnC1 .4/:

Example 5.1. For n D 1, t1 D t and u1 D u the equations in Proposition 5.5 read

tX 2 C X 4 C uY 2 D X; uX 2 C .t C u/Y 2 C Y 4 D Y;
5 Field Restriction in Modular Galois Theory 445

from which one first obtains

1 4
Y2 D .X C tX 2 C X / and Y 2 D u2 X 4 C .t C u/2 Y 4 C Y 8
u
and from this by replacing Y 2 in the second equation the polynomial

X 16 C .t 4 C t 2 u2 C u4 /X 8 C .t 4 u2 C u4 t 2 C u6 C u3 C 1/X 4
C.u2 t 2 C u3 t C u4 /X 2 C u3 X D 0:

By Proposition 5.5 this has Galois group SL2 .4/ over IF2 .t; u/. t
u
The second example concerns the groups GAn .IFq / D SLnC1 .q/ with q D p 2 ¤
4. Here let IFq D IFp .x/ with x 2 D y 2 IFp and ti D ti C xui . Then according to
Proposition 3.2 we get
0 1
t1 : : : tn 1 u1 y : : : un y 0
B 1 0 C
B C
B : C
B :: C
B C
B 1 0 C
 B
SAn .t; u/ D B C:
C
Bu1 : : : un 0 t1 : : : tn 1C
B 1 0C
B C
B :: C
@ : A
1 0

The system of algebraic equations

SAn .t; u/.X1p ; : : : ; XnC1

p
; Y1p ; : : : ; YnC1
p
/t D .X1 ; : : : ; YnC1 /t

then leads to the equations

X
n
p p
X
n
p p
 ti Xi C XnC1  ui yYi D X1 ; Xi D Xi C1 for i D 1; : : : ; n;
i D1 i D1
X
n X
n
 ui Xip  ti Yip C YnC1
p
D Y1 ; Yip D Yi C1 for i D 1; : : : ; n:
i D1 i D1

i 1 i 1
From this one finds Xi D .1/i 1 X p for X D X1 and Yi D .1/i 1 Y p for
Y D Y1 . This shows:
Proposition 5.6. Assume p > 2 and let f .X / respectively g.Y / be the polynomial
which is obtained by elimination of Y resp. of X from the system

X
n
i nC1
X
n
i
XC .1/i 1 ti X p C .1/n X p C .1/i 1 ui yY p D 0;
i D1 i D1
446 V Additive Polynomials

X
n
i
X
n
i nC1
YC .1/i 1 ui X p C .1/i 1 ti Y p C .1/n Y p D 0:
i D1 i D1

Then the field NAn generated over IFp Œt; u by the zeroes of f , respectively of g, has
Galois group
Gal.NAn =IFp .t; u// Š SLnC1 .p 2 /:

Example 5.2. We again give explicit polynomials for SL2 .p 2 /, p > 2. For this let
n D 1, t D t1 and u D u1 . Then the equations in Proposition 5.6 become
2 2
X C tX p  X p C uyY p D 0; Y C uX p C tY p  Y p D 0:

From these by elimination of X one obtains the polynomial

4 2 2 p 3 2 p 2 Cp 2 1 2
Y p  .t p C t p up / Y p C .t 2p up  yup  up  1/ Y p
2 p 2 1 2 1
C .t p up C tup / Y p C up Y

with Galois group SL2 .p 2 / over IFp .t; u/ for p ¤ 2. t

u
VI Rigid Analytic Methods

The solution of the inverse problem of Galois theory over the field C.t/ was achieved
by a blend of topological and analytical methods. A similar approach is possible
for any ground field complete with respect to a non-archimedean valuation. The
suitable analytic structures are provided by the so-called rigid analytic spaces. They
satisfy a GAGA-principle, which makes it possible to recover algebraic structures
from analytic constructions. This replaces the Riemann Existence Theorem in the
complex case. In the first paragraph we collect some definitions and results on rigid
analytic geometry and sketch a proof of the GAGA-principle for covers of the rigid
analytic projective line.
With these prerequisites it is then possible to solve the inverse problem for the
rational function field over a complete ultrametric field. This includes in particu-
lar the rational function field over the field Qp of p-adic numbers. The solution is
achieved by a cut and paste process mimicking the complex case. Further we solve
the inverse problem over IFp .t/ for any prime p by specializing Galois extensions of
IFp ..u//.t/. The result is due to Harbater (1984), who used formal geometry instead
of rigid analytic geometry. Harbater (1995a) and Pop (1995) later showed how to
solve sufficiently many embedding problems over IFp .t/ in order to conclude the
freeness of the absolute Galois group of this field from the Freiheitssatz of Iwasawa.
In the third paragraph we show how to construct large free quotients of fun-
damental groups of function fields over complete ultrametric fields with explicit
information on the action of the Galois group, following Pop (1994). Using the
gluing procedure, one first obtains free products of finite groups as Galois groups,
and then, taking a suitable projective limit, also free groups. This is used in the
fourth paragraph to solve embedding problems over function fields over large fields
as introduced by Pop (1996). This allows to prove that the absolute Galois group of
a countable Hilbertian PAC-field is free profinite, generalizing Theorem IV.3.8.
In the final paragraph, we first prove the projectivity of the fundamental group of
an affine curve. We then state the conjecture of Abhyankar (1957) concerning the
finite quotients of the fundamental group in positive characteristic with fixed number
of ramification points, and give some hints on the recent proof by Raynaud (1994)

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 447

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3_6
448 VI Rigid Analytic Methods

for the case of the affine line of this conjecture (see also Saı̈di (2000)). Finally we
prove the reduction of the general case to the 1-point case, due to Harbater (1994a),
following Pop (1995).
1 Results from Rigid Analytic Geometry 449

1 Results from Rigid Analytic Geometry

Rigid analytic spaces are best defined locally via their algebras of holomorphic func-
tions, the Tate algebras. We then introduce coherent sheaves on rigid analytic spaces
and sketch a proof of the 1-dimensional GAGA-theorem. The definitions and results
in this paragraph, unless otherwise stated, are taken from Fresnel and van der Put
(1981) or Bosch, Güntzer and Remmert (1984).

1.1 Tate Algebras

In the whole paragraph, k denotes a field complete with respect to a valuation with
corresponding ultrametric absolute value j  j W k ! IR. The valuation ring and valu-
ation ideal are

k 0 WD f´ 2 k j j´j 1g; k 00 WD f´ 2 k j j´j < 1g; (1.1)

with residue class field k 0 =k 00 . For an integer n  1 we define

X 
Tn WD Tn .k/ WD khZ1 ; : : : ; Zn i WD aj Z j aj 2 k; P lim
j
.aj / D 0 ; (1.2)
i ji !1
j2INn
0

the ring of convergent power series in the n indeterminates Z D .Z1 ; : : : ; Zn / on the

unit disc with respect to the ultrametric absolute value j  j, where Zj WD Z1j1    Znjn ,
and j D .j1 ; : : : ; jn /. Now
X
jj aj Zj jj WD maxfjaj j j j 2 INn0 g
j2INn
0

defines a norm on the algebra Tn . As for k above, we set

Tn .k/0 WD ff 2 Tn .k/ j jjf jj 1g; Tn .k/00 WD ff 2 Tn .k/ j jjf jj < 1g:

Then the residue ring Tn .k/0 =Tn .k/00 is canonically isomorphic to the polynomial
ring in n indeterminates over the residue class field k 0 =k 00 .
A Tate algebra A over k is a finite extension of Tn .k/ for some n. Here finite
extension means that A is finitely generated as a Tn -module. The following proper-
ties of Tate algebras can be proved by elementary calculations, using an analogue
of the Weierstraß division and preparation theorem. For a proof see Fresnel and van
der Put (1981), Thm. II.3, or Bosch, Güntzer and Remmert (1984), 5.2.6, 6.1.2,
6.1.3 and 6.2.4.
Theorem 1.1. (a) Tate algebras are Noetherian rings.
(b) Tn is a unique factorization domain.
(c) For any maximal ideal m of Tn , Tn =m is a finite extension of k.
450 VI Rigid Analytic Methods

(d) Every Tate algebra has the form A D Tn =a for some n and some ideal a G Tn .
(e) Algebra homomorphisms between Tate algebras are continuous; all Banach
norms on a Tate algebra are equivalent.

Let A be a Tate algebra over k. By Theorem 1.1(d) there exist n and a G Tn with
A D Tn =a, where Tn D khZ1 ; : : : ; Zn i. We can then define

AhY1 ; : : : ; Ym i WD khZ1 ; : : : ; Zn ; Y1 ; : : : ; Ym i=.a  khZ1; : : : ; Zn ; Y1 ; : : : ; Ym i/

for any finite set fY1 ; : : : ; Ym g of further indeterminates. This is independent of the
choice of presentation for A. By Sp.A/ we denote the set of maximal ideals of
A. For f 2 A and m 2 Sp.A/ we define f .m/ to be the image of f in A=m. By
Theorem 1.1(c) and (d), A=m is a finite field extension of k, so the absolute value
j  j has a unique extension to A=m and it makes sense to speak of jf .m/j. We define
a topology on X D Sp.A/, to be generated by the sets of the form Uf WD fm 2 X j
jf .m/j 1g, where f runs over A. A subset R  Sp.A/ is called a rational subset
(in X ) if there exist f0 ; : : : ; fn 2 A with

R D fm 2 Sp.A/ j jfi .m/j jf0 .m/j for 1 i ng (1.3)

and A is generated by f0 ; : : : ; fn as an ideal: A D .f0 ; : : : ; fn /.

To each rational subset R  Sp.A/ we associate the Tate algebra

B WD AhX1 ; : : : ; Xn i=.f1  X1 f0 ; : : : ; fn  Xn f0 /: (1.4)

The canonical homomorphism ' W A ! B induces a homeomorphism Sp.'/ W

Sp.B/ ! R  Sp.A/, such that B satisfies the following universal mapping prop-
erty: For any homomorphism W A ! C of Tate algebras with Sp. /.Sp.C //  R
there exists a unique homomorphism  W B ! C such that  ı ' D . In particular,
a rational subset R is of the form Sp.B/ for some Tate algebra B which is indepen-
dent of the choice of the elements f0 ; : : : ; fn used to present R (see Fresnel and van
der Put (1981), Lemme III.1.2). We write OX .R/ WD B, where X D Sp.A/, and
call it the ring of holomorphic functions on R. The following properties of rational
sets are elementary (see Fresnel and van der Put (1981), Lemme III.1.3, or Bosch,
Güntzer and Remmert (1984), 7.2.3, 7.2.4).

Proposition 1.2. (a) If R1 ; R2 are rational in X D Sp.A/, then so is R1 \ R2 , and

we have OX .R1 \ R2 / D OX .R1 / Ő A OX .R2 /.
(b) If R1  R2  X , and R1 is rational in R2 , while R2 is rational in X , then
R1 is also rational in X .

Here Ő A denotes the complete tensor product, see Bosch, Güntzer and Remmert
(1984), 2.1.7.
1 Results from Rigid Analytic Geometry 451

1.2 Rigid Analytic Spaces

Let A be a Tate algebra. In order to obtain the right notion of sheaves on the space
X D Sp.A/ of maximal ideals of A we introduce a Grothendieck topology (see
for example Bosch, Güntzer and Remmert (1984), 9.1). For this let the admissible
subsets of X be the finite unions of rational subsets. The admissible coverings
are the finite coverings by admissible subsets. It can be verified that this defines a
Grothendieck topology T on X (see Bosch, Güntzer and Remmert (1984), 9.1.4).
Then by Proposition 1.2 a (pre-) sheaf of rings on T is given by R 7! OX .R/ for
rational subsets R  X . For any finitely generated A-module M define MQ to be the
presheaf with R 7! MQ .R/ WD M ˝A OX .R/. The non-trivial fact that these define
sheaves was first proved by Tate (see Fresnel and van der Put (1981), Thm. III.2.2,
or Bosch, Güntzer and Remmert (1984), 8.2.1):

Theorem 1.3 (Tate). Let A be a Tate algebra, M a finitely generated A-module.

Then every admissible covering of X D Sp.A/ is acyclic for the (pre-) sheaf MQ . In
particular, OX and MQ are sheaves.

The ringed space .Sp.A/; T; OSp.A/ / is called an affinoid analytic space, with
structure sheaf (or sheaf of holomorphic functions) OSp.A/ .
More generally, a ringed space .X ; T; OX / consisting of a space X with a
Grothendieck topology T and a sheaf OX on X is called a rigid analytic space, if
there exists an admissible covering fXi j i 2 I g of X such that .Xi ; TjXi ; OX jXi /
is an affinoid analytic space for all i 2 I . A morphism of rigid analytic spaces is by
definition a morphism of locally ringed spaces with Grothendieck topology.
An important method for constructing rigid analytic spaces is by gluing. For this
we first define a gluing datum. This consists of a family fXi j i 2 I g of affinoid ana-
lytic spaces, with rational subsets Xij  Xi and isomorphisms of analytic spaces

'j i W Xij ! Xj i for all i; j 2 I , satisfying the following properties:
(1) Xi i D Xi , 'i i D IdXi for all i 2 I ,
(2) 'ij is inverse to 'j i for all i; j 2 I ,
(3) 'kj ı 'j i D 'ki on Xij \ Xi k for all i; j; k 2 I .
Note that Xij D Xj i D ; is allowed. Gluing data define rigid analytic spaces:

Proposition 1.4 (Gluing of Spaces). Let fXi j i 2 I g, fXij ; 'ij j i; j 2 I g be a glu-

ing datum. Then there exists a rigid analytic space .X ; T; OX / unique up to iso-
morphism with the following properties:
(a) there exist homeomorphisms i W Xi ! i .Xi /  X onto open subsets
of X , S
(b) X D i 2I i .Xi / is an admissible covering,
(c) i .Xij / D j .Xj i / D i .Xi / \ j .Xj /, and 'j i D j1 ı i on Xij for
all i; j 2 I ,
(d) i induces an isomorphism between the structure sheaf OXi and OX jXi .

For a proof see Bosch, Güntzer and Remmert (1984), 9.3.2. Similarly, mor-
phisms of rigid analytic spaces may be glued (see loc. cit., 9.3.3).
452 VI Rigid Analytic Methods

Proposition 1.5 (Gluing of Morphisms). Let X and Y be analytic spaces with

admissible coverings fXi j i 2 I g, fYi j i 2 I g. Let f'i W Yi ! Xi j i 2 I g be
analytic morphisms such that 'i and 'j have same restriction to Yi \ Yj for all
i; j 2 I . Then there exists a unique analytic morphism ' W Y ! X with 'jYi D 'i
for all i 2 I .

A rigid analytic space .X ; T; OX / is called connected if the only idempotent

global sections of OX are the constants 0 and 1. The following criteria for connect-
edness will be used in subsequent constructions:

Lemma 1.6. Let X D X1 [X2 be an admissible covering of a rigid analytic space

by connected analytic spaces Xi such that X1 \ X2 is non-empty. Then X is con-
nected.

Proof. Let f 2 OX .X / be an idempotent global section. Then for i D 1; 2 the

restriction f jXi is an idempotent global section on Xi , hence constant since Xi is
connected. Restriction to the non-empty intersection X1 \ X2 shows that these two
constants must in fact coincide. Thus f is constant, proving the connectedness of
X. t
u

This will be particularly interesting for the gluing of spaces:

Corollary 1.7. In the setting of Proposition 1.4 assume that all the Xi are con-
nected, and for each pair i; j 2 I there exists a sequence i D i1 ; i2 ; : : : ; in D j of
indices in I such that Xim ;imC1 ¤ ; for m D 1; : : : ; n  1. Then X is connected.

Proof. Induction on the length of a chain of indices as in the assumption, with

Lemma 1.6 as induction base, shows the following: if f 2 OX .X / is idempo-
tent then for any pair i; j 2 I the two constant functions f jXi , f jXj agree. Thus
any idempotent global section on X is constant. t
u

Lemma 1.8. Let X D [niD0 Xi be an admissible covering of a connected analytic

space such that X0 \ Xi is connected for i D 1; : : : ; n and Xi \ Xj D ; for 1
i < j n. Then X0 is connected.

Proof. Let f0 2 OX0 .X0 / be an idempotent global section. Since X0 \ Xi is con-

nected, the restriction f0 jX0 \Xi D ai is constant. Let fi 2 OXi .Xi / be the con-
stant function ai for 1 i n. Then by Proposition 1.4 (Gluing of Spaces) the fi ,
i D 0; : : : ; n, may be glued to a function f 2 OX .X / which is idempotent since
the fi are. But X is connected, so f is constant, and hence f0 D f jX0 is constant
as well. t
u

We conclude this section with some standard examples of affinoid analytic

spaces.

Example 1.1 (Discs). Let k be algebraically closed. By Theorem 1.1(c) we then

have T1 =m Š k for every m 2 X D Sp.khZi/ D Sp.T1 /. For ´ 2 k with j´j > 1 the
element Z  ´ is invertible in T1 , so it does not generate a proper ideal of T1 . We
1 Results from Rigid Analytic Geometry 453

may hence identify the topological space X D Sp.T1 / with k 0 D f´ 2 k j j´j 1g

via the map .Z  ´/T1 7! ´. In particular, T1 is the ring of holomorphic functions
on the unit disc k 0 . By suitable translation and scaling, this gives the holomorphic
functions on discs centered at arbitrary points.
More generally we thus obtain that the topological space X D Sp.Tn / can be
identified with the polydisc

fz D .´1 ; : : : ; ´n / 2 k n j j´i j 1 for 1 i ng : t

u

Example 1.2 (Circles). We write

( 1 )
X
1
khZ; Z i D ai Z j ai 2 k; lim .ai / D 0
i
(1.5)
ji j!1
i D1

for the Tate algebra khZ1 ; Z2 i=.Z1 Z2  1/. If again k is assumed to be alge-
braically closed, Sp.khZ; Z 1 i/ can be naturally identified with the unit circle
f´ 2 k j j´j D 1g by the same recipe as in Example 1.1. t
u

Example 1.3 (Annuli). A third important type of affinoid analytic spaces is given by
annuli. They are obtained from Tate algebras

A WD khZ1 ; Z2 i=.Z1 Z2  ´0 / where 0 < j´0 j < 1:

If k is algebraically closed, A can be naturally identified with the annulus f´ 2 k j

j´0 j j´j 1g as in the previous examples. t
u

1.3 Analytification of Algebraic Varieties

We have seen above how a rigid analytic space can be obtained by gluing. The other
important mechanism for the construction of analytic spaces is by analytification of
algebraic varieties. We start with the case of projective space.
Let IPn .k/ be the n-dimensional projective space over k. By Example 1.1, the
polydiscs
Ui WD f.´0 ; : : : ; ´n / j j´i j  j´j j for 0 j ng
have associated Tate algebras Tn . These are clearly rational and cover the projective
space. The Ui with structure sheaves OUi together with Uij D Uj i WD Ui \ Uj
and the identity maps 'ij D IdUij form a gluing datum. By the Gluing of Spaces
(Proposition
Sn 1.4) we obtain the structure of a rigid analytic space X WD IPn .k/an WD
i D0 Ui on IP .k/ with structure sheaf OX such that OX .Ui / WD OUi .
n

More generally, let k be algebraically closed and X =k an algebraic variety.

Then there exists a canonical analytification of X . By the Gluing Proposition 1.4,
it suffices to construct this in the case that X is affine. Then X can be identified
with a closed subset of k n for some n, and there exists an ideal a D .f1 ; : : : ; fs / of
454 VI Rigid Analytic Methods

kŒZ1 ; : : : ; Zn with

X D fz 2 k n j f1 .z/ D : : : D fs .z/ D 0g :

Let q 2 k with 0 < jqj < 1. Then

Xm WD fz 2 X  An .k/ j j´i j jqjm for 1 i ng

can be identified with the affinoid analytic space defined by the Tate algebra
ı
Am WD khZ1 ; : : : ; Zn i .f1 .q m Z/; : : : ; fs .q m Z//

via
Xm ! Sp.Am /; z 7! .Z1  q m ´1 ; : : : ; Zn  q m ´n / :
The Xm form an inductive system andSmay be glued together since Xm is ratio-
nal inside XmC1 . We obtain X an WD m Xm and O.X an / D lim O.Xm /. It can

be proved that this analytification of X is independent of the chosen embedding
X ,! k n (see Bosch, Güntzer and Remmert (1984), 9.3.4, or Fresnel and van der
Put (1981), III.4.5). Via the preceding construction we may and will from now on
identify the points P 2 X of algebraic varieties X and their rigid analytic versions
X an .
Let X be a rigid analytic space. A sheaf F of OX -modules is called a coherent
sheaf, if there exists an admissible covering fXi j i 2 I g of X by affinoid analytic
spaces Xi such that F .Xi / is a finitely generated OX .Xi /-module and

B
F jXi D F .Xi / for all i 2 I :

As in the algebraic case there is a simple characterization of coherent sheaves on

affinoid spaces (see Fresnel and van der Put (1981), Thm. III.6.2, or Bosch, Güntzer
and Remmert (1984), 9.4.3):

Theorem 1.9. Let X be an affinoid analytic space and F a coherent sheaf. Then
F .X / is a finitely generated OX -module and F Š F .X /. B
The GAGA-principle compares coherent algebraic and analytic sheaves. We first
explain how coherent sheaves on an algebraic variety can be analytified. It suffices
to do this in the affine case.
Let k be algebraically closed, X D Spec.R/ an affine algebraic variety over
k and F an algebraic coherent sheaf on X . Let fSp.Bi / j i 2 I g be a basis of
the Grothendieck topology on the analytification X an . We define a presheaf F an
on X an by setting F an .Sp.Bi // WD F .X / ˝R Bi for i 2 I . Then this defines a
coherent analytic sheaf on X an (see Köpf (1974), Bem. 3.2). Furthermore, if ' W
F ! G is a morphism of coherent sheaves on X , then
an
'Sp.Bi/
WD ' ˝ Bi W F .X / ˝ Bi ! G .X / ˝ Bi for i 2 I
1 Results from Rigid Analytic Geometry 455

defines a morphism ' an W F an ! G an of analytic coherent sheaves (see Köpf (1974),

Satz 3.1). This analytification satisfies (see loc. cit. Satz 3.6):

Proposition 1.10. Let 0 ! F1 ! F2 ! F3 ! 0 be an exact sequence of coherent

X -sheaves. Then the associated sequence of coherent X an -sheaves 0 ! F1an !
F2an ! F3an ! 0 is also exact.

Proof. Since exactness of sequences of sheaves is a local property, it suffices to con-

sider the case where X D Spec.R/ is affine. The Fi are coherent, so the sequence
0 ! F1 .X / ! F2 .X / ! F3 .X / ! 0 of R-modules is also exact. The flatness
of R ! B for rational subsets Sp.B/  X an (see Fresnel and van der Put (1981),
Thm. III.7.2) now shows that the sequence 0 ! F1 .X / ˝R B ! F2 .X / ˝R B !
F3 .X / ˝R B ! 0 is also exact, hence the result follows by the construction of the
analytification. t
u

1.4 The GAGA-Principle for IP1 .k/an

For our purposes the most important fact on rigid analytic spaces is the GAGA-
principle. Over C this was proved by Serre (1956). This was adapted to the rigid
analytic setting by Köpf (1974). We will only need the statement for subspaces of
IP1 .k/. For this let IP1 .k/an denote the analytification of the 1-dimensional (alge-
braic) projective line IP1 .k/ over k. We fix the covering

IP1 .k/an D X1 [ X2 ; with X1 D f´ 2 k j j´j 1g; X2 D f´ 2 k j j´j  1g;

and let X12 WD X1 \ X2 D f´ 2 k j j´j D 1g. With respect to this covering for any
integer n we define the twisted structure sheaf OX .n/ by

U 7! f.f1 ; f2 / 2 OX .X1 \ U / ˚ OX .X2 \ U / j f1 D Z n f2 on X12 \ U g

(1.6)
for admissible open subsets U  IP1 .k/an . (See Fresnel and van der Put (1981),
III.8.5, for the fact that this is an invertible coherent sheaf, and any invertible coher-
ent sheaf is isomorphic to one of this type.)
The following result was shown to us by M. van der Put:

Proposition 1.11. Let F be a coherent sheaf on IP1 .k/an . Then there exists n > 0
such that F .n/ WD F ˝ OX .n/ is generated by global sections.

Proof. Since Xi , i D 1; 2, is an affinoid analytic space, the restriction of F to

Xi is isomorphic to MQ i for some finitely generated O.Xi /-module Mi by Theo-
rem 1.9. Let F .Xi / D hai1 ; : : : ; ai r i be a generating system of Mi . Denote by i the
restriction maps i W F .Xi / ! F .X12 /. Then by the coherence of F we see that
456 VI Rigid Analytic Methods

F .X12 / D hi .ai1 /; : : : ; i .ai r /i for i D 1; 2. Let U; V 2 O.X12 /rr be matrices

with
V  .1 .a11 /; : : : ; 1 .a1r //t D.2 .a21 /; : : : ; 2 .a2r //t ;
U  .2 .a21 /; : : : ; 2 .a2r //t D.1 .a11 /; : : : ; 1 .a1r //t ;
so that V U.2 .a21 /; : : : ; 2 .a2r //t D .2 .a21 /; :P
: : ; 2 .a2r //t . By Example 1.2, the
Tate algebra of X12 is given by O.X12 / D f 1 i D1 ci Z j limji j!1 .ci / D 0g.
i

Thus clearly for any  > 0 there exists n WD n./ 2 IN and

1
X
V 0 2 GLr .f ci Z i j lim .ci / D 0g/
i !1
i Dn

such that jj.V  V 0 /U jj < . By Fresnel and van der Put (1981), Lemme III.6.3,
(a special case of the Lemma of Cartan) for  suitably small there then exist Vi 2
GLr .O.Xi //, i D 1; 2, with

Id .V  V 0 /U D 1 .V1 /1 2 .V2 / :

Hence we get 1 .V1 /V 0 .1 .a11 /; : : : ; 1 .a1r //t D 2 .V2 /.2 .a21 /; : : : ; 2 .a2r //t .
By construction we have Z n V 0 D 1 .VQ / for some matrix VQ 2 GLr .O.X1 //. In
particular we see that

1 .V1 VQ .a11 ; : : : ; a1r /t / D Z n 2 .V2 .a21 ; : : : ; a2r /t /:

Thus V1 VQ .a11 ; : : : ; a1r /t 2 F .X1 / and V2 .a21 ; : : : ; a2r /t 2 F .X2 / satisfy the
compatibility condition (1.6) for F .n/; since F .n/ is a sheaf, there exists a global
section .b1 ; : : : ; br / having these two elements as projections. Finally, V1 VQ and V2
are invertible, and the result follows. t
u

We now formulate the GAGA-principle for the 1-dimensional projective analytic

space IP1 .k/an , and sketch a proof closely following the one given by Serre (1956)
for the general complex analytic case (see also Fresnel and van der Put (2004),
Ch. 9.3.1).

Theorem 1.12 (GAGA for IP1 ). Let k be a field complete with respect to a non-
archimedean valuation.
(a) Let G be an algebraic coherent sheaf on the 1-dimensional projective space
IP1 .k/. Then analytification yields an isomorphism of the global sections G .IP1 .k//
and G an .IP1 .k/an /.
(b) Let F ; G be algebraic coherent sheaves on IP1 .k/. Then every analytic homo-
morphism F an ! G an is induced by a unique algebraic homomorphism F ! G .
(c) For every coherent analytic sheaf F on IP1 .k/an there exists an algebraic
coherent sheaf G on IP1 .k/ unique up to isomorphism with G an Š F .

Outline of proof. Let G be an algebraic coherent sheaf on X WD IP1 .k/ and G an its
analytification on X an WD IP1 .k/an . Then there exist canonical homomorphisms  i W
H i .X ; G / ! H i .X an ; G an / of cohomology groups for all i  0 compatible with
1 Results from Rigid Analytic Geometry 457

short exact sequences, that is, for any short exact sequence 0 ! G1 ! G2 ! G3 ! 0
of coherent X -sheaves all squares in the diagram

H i .X ; G1 / ! H i .X ; G2 / ! H i .X ; G3 / ! H i C1 .X ; G1 /
? i ? i ? i ? i C1
y y y y (1.7)
1 2 3 1
i C1
i
H .X an
; G1an / ! H .X
i an
; G2an / ! H .X
i an
; G3an / !H .X an
; G1an /

commute (see Köpf (1974), Satz 3.10). To prove (a) one shows more generally that
 i is an isomorphism for all i  0, the assertion being the special case i D 0. For the
structure sheaf O D O.X an / the cohomology groups can be computed explicitly
on both sides: they vanish for i > 0, and the global sections are equal to k (see
Köpf (1974), Thm. 4.6). Now the machinery of Serre’s proof can be applied in this
setting: First, one considers the twisted structure sheaves G D O.n/, n 2 ZZ. To any
point P of X an is associated an exact sequence

0 ! O.n  1/ ! O.n/ ! OP .n/ ! 0

where OP is supported by the single point P , and similarly for the analytifications.
Taking cohomology the associated diagrams (1.7) allow to use induction on n by
applying the snake lemma, see Köpf (1974), Hilfssatz 4.10.
Now let G be an algebraic coherent sheaf on X . By Hartshorne (1977),
Cor. II.5.18, there exist n; r 2 IN and a surjective morphism O r ! G .n/ of coherent
sheaves. By loc. cit., Prop. II.5.7, the kernel is again coherent, so we have an exact
sequence
0 ! N ! O r ! G .n/ ! 0
of coherent X -sheaves. Taking global sections of this sequence and its analytifica-
tion yields long exact sequences of cohomology connected by diagram (1.7). The
snake lemma now allows to apply descending induction on i to show the bijectivity
of  i . Part (a) follows.
For (b) we consider the coherent sheaf A D Hom.F ; G / of homomorphisms
from F to G , and its analytic version B D Hom.F an ; G an /. Analytification yields
an injection  W A an ! B. If AP D HomOP .FP ; GP / denotes the stalk of A at a
point P 2 IP1 .k/, then by definition

AP
an
D HomOP .FP ; GP / ˝OP OP
an
:

Since F an is coherent, the stalk of B at P is given by

BP D HomOP
an .FP ˝O
P
OP
an
; GP ˝OP OP
an
/:

The ring OP is local and Noetherian, so its completion OOP is faithfully flat over
OP (see Matsumura (1980), Thm. 56). Since the completions of OP and OP an

coincide, the descent property of faithful flatness shows that OP is faithfully flat
an

over OP . This implies the bijectivity of P . It follows that first , and then the
458 VI Rigid Analytic Methods

composition

A .IP1 .k// ! A an .IP1 .k/an / ! B.IP1 .k/an /
with the isomorphism from part (a), applied to the coherent sheaf A , are also bijec-
tive. Now the global sections A .IP1 .k// of A are the homomorphisms from F to
G , and similarly for B, showing part (b).
For part (c) let F be an analytic coherent sheaf on IP1 .k/an . By Proposition 1.11
there exists n 2 IN such that F .n/ is generated by global sections. Thus for a suit-
able r we obtain a surjective morphism .O an /r ! F .n/ of coherent sheaves. By
Bosch, Güntzer and Remmert (1984), 9.4.3, Prop. 2, the kernel of a morphism of
coherent sheaves is again coherent, so there exists a coherent sheaf N completing
the exact sequence

0 ! N ! .O an /r ! F .n/ ! 0: (1.8)

Applying Proposition 1.11 again to the coherent sheaf N we get a surjection

.O an /s ! N .m/ for suitable s; m 2 IN, which together with (1.8) yields an exact
sequence
.O an /s ! .O an /r .m/ ! F .n C m/ ! 0: (1.9)
By part (b) the morphism ' W .O an /s ! .O an /r .m/ comes from a uniquely deter-
mined algebraic morphism . Thus (1.9) induces a sequence

O s ! O r .m/ ! G ! 0 (1.10)

of algebraic sheaves, with coherent G . In particular with Proposition 1.10 we obtain

that G an D F .n C m/, hence F D G .n  m/an as claimed. To prove uniqueness,
assume that G1 ; G2 are two algebraic coherent sheaves with analytification F . Then
there exists an analytic isomorphism ' W G1an ! G2an , which by part (b) is induced
by an algebraic morphism W G1 ! G2 with an D '. With the kernel A and the
cokernel B we obtain an exact sequence of coherent sheaves

0 ! A ! G1 ! G2 ! B ! 0:

By analytification Proposition 1.10 shows that A an D B an D 0. Looking at the

global sections this forces A D B D 0, hence & W G1 ! G2 is an isomorphism. ut
A morphism ' W Y ! X of rigid analytic spaces is called finite if there exists
an admissible affinoid covering fXi j i 2 I g of X , such that f' 1 .Xi / j i 2 I g

is an admissible covering of Y and the associated morphisms 'X i
W OX .Xi / !
1 
OY .' .Xi // of Tate algebras are finite. Here a morphism ' W A ! B is called
finite, if B is a finitely generated A-module via '  .

Corollary 1.13. Let ' W X ! IP1 .k/an be a finite morphism of analytic spaces.
(a) There exists a finite algebraic morphism of algebraic varieties W XQ !
IP1 .k/ whose analytification is isomorphic to ' W X ! IP1 .k/an .
1 Results from Rigid Analytic Geometry 459

(b) Let 'Q W Y ! X be finite. Then there exists a finite morphism of algebraic
varieties Q W YQ ! XQ with Q an D '.
Q

Proof. The direct image F WD ' .OX / of the structure sheaf on X is a coherent
sheaf on IP1 .k/an (see Bosch, Güntzer and Remmert (1984), 9.4.4, Prop. 3). Thus
by Theorem 1.12(c) there exists an algebraic sheaf G on IP1 .k/ with G an D F .
By construction F is a sheaf of rings. The multiplication defines an analytic mor-
phism W F ˝ F ! F of coherent analytic sheaves. By Theorem 1.12(b) this
is induced by an algebraic morphism & W G ˝ G ! G with &an D . Since, as in
the proof of Theorem 1.12, all ring extensions are faithfully flat, the morphism &
has the properties of a multiplication, thus making G into a sheaf of rings. Now
XQ1 D Spec.G .IP1 .k/ n f1g// and XQ2 D Spec.G .IP1 .k/ n f0g// may be glued to an
algebraic covering XQ WD Spec.G / D XQ1 [ XQ2 of IP1 .k/ (see Hartshorne (1977),
Ex. III.5.17) with analytification X ! IP1 .k/an , proving (a).
In part (b) the morphisms ' and ' ı 'Q W Y ! IP1 .k/an are analytifications of
algebraic morphisms by part (a). Moreover 'Q induces a morphism 'Q W ' .OX / !
.' ı '/
Q  .OY / commuting with the multiplication on these two sheaves of algebras.
By Theorem 1.12(b) this is induced by an algebraic morphism
' .OXQ / ! .' ı '/Q  .OYQ / of sheaves of algebras (see part (a)). This defines an alge-
braic morphism Q W YQ ! XQ with analytification 'Q W Y ! X . t
u
460 VI Rigid Analytic Methods

2 The Inverse Problem over Qp .t / and IFp .t /

In this paragraph we mimic the cut and paste approach for the solution of the inverse
problem over C.t/ via topological methods to obtain a similar result for any field
complete with respect to an ultrametric valuation. The complex analytic structure
is replaced by the rigid analytic structure introduced in Paragraph 1, but otherwise
the proofs closely follow the complex example. The main results are the solution of
the inverse problem over Qp .t/ and over IFp .t/ and the freeness of the fundamental
group of IFp .t/ for any prime p.

2.1 Induced Covers

Let k be complete with respect to an ultrametric valuation. Let ' W Y ! X be a

morphism of rigid analytic spaces over k. The group Deck.'/ is defined to be the
group of analytic automorphisms  of Y with ' D ' ı . The cover ' W Y ! X
is called a Galois cover if the quotient Y = Deck.'/ exists and is isomorphic to
X . For any subgroup G Deck.'/ we say that ' W Y ! X is G-equivariant; if
moreover G acts regularly on the generic fibres of ' we say for brevity that ' W Y !
X is Galois with group G. We will be interested primarily in connected Galois
covers (with respect to the Grothendieck topology, or, equivalently, with respect to
the Zariski topology, see Bosch, Güntzer and Remmert (1984), 9.1.4), since these
lead to Galois extensions of function fields.
Proposition 2.1. Let G be a finite group, H G a subgroup, and ' W Y ! X a
Galois cover with group H . Then there exists a unique Galois cover

H .'/ W IndH .Y / ! X
IndG G
(2.1)

with group G with an H -morphism  W Y ! IndG G

H .Y / such that ' D IndH .'/ ı .
If H1 H2 G is a chain of subgroups, then as G-covers
H2
IndG G
H1 .Y / Š IndH2 .IndH1 .Y //:


Proof. Let IndGH .Y / WD G=H  Y D [ i 2G=H .i ; Y / be the disjoint union of
copies of Y indexed by a system of coset representatives i of H in G, where the
representative for H is chosen to be 1. Then G acts on IndG H .Y / in a natural way
via
 W IndG G
H .Y / ! IndH .Y /; .i ; y/ 7! .j ; .y//;
for  2 G with i D j  for  2 H . This defines a G-cover IndG G
H .'/ W IndH .Y / !
X , .i ; y/ 7! y. Further the embedding

 W Y ! Y1 ,! IndG
H .Y /

H .'/ ı .
yields an H -morphism  with ' D IndG
2 The Inverse Problem over Qp .t / and IFp .t / 461

For the uniqueness, let 'Z W Z ! X be another Galois cover with group G and
with H -morphism Z W Y ! Z . We define

W IndG
H .Y / ! Z ; .i ; y/ 7! .y/ WD i .Z .y//;

a G-equivariant morphism with Z D ı . Since the roles of Z and IndGH .Y / are

interchangeable, we obtain a G-equivariant isomorphism W IndGH .Y / ! Z . The
remaining statements in the proposition are now easy consequences of the above
construction. t
u

The cover IndGH .'/ W IndH .Y / ! X is called the induced cover of ' W Y ! X
G

from H to G. Note that such covers are not connected whenever H is a proper sub-
group of G. It is easy to see that on the side of Tate algebras, induction corresponds
to taking direct products of isomorphic copies. Thus induced covers do not in gen-
eral yield extensions of function fields on the algebraic side. The induced cover
IndG1 .X / ! X is called the trivial cover of X of group G.

Proposition 2.2. Let K=k be a function field in one variable. Let X ! IP1 .k/ be
a projective normal geometrically integral model of K=k.t/ and X an ! IP1 .k/an
the covering of analytic spaces obtained from it by analytification. Then X an is a
trivial cover above some affinoid subset of IP1 .k/an if and only if there exists a prime
divisor P 2 IP.k.t/=k/ of degree 1 which splits completely in K=k.t/.

Proof. First assume that the prime divisor P 2 IP.k.t/=k/ of degree 1 splits com-
pletely in K=k.t/. Without loss of generality we may take P to be the numer-
ator divisor of .t/. Then X ! IP1 .k/ is unramified in a small neighborhood of
0 2 IP1 .k/, hence locally is defined by an equation f .t; X / D 0 such that f .0; X / D
Q n
i D1 .X  ´i / splits into linear factors over k. By the analytification procedure
described in Example 1.3, the cover X an ! IP1 .k/an is given locally near 0 by
Q algebras khZi ! khZ; X i=f .Z; X /. By Hensel’s Lemma we
the morphism of Tate
have f .Z; X / D niD1 .X  fi .Z// for certain fi .Z/ 2 khZi in a small neighbor-
hood of 0. But
Yn Mn
khZ; X i= .X  fi .Z// Š khZi;
i D1 i D1
 n
hence X is equal to
an
[ i D1 Sp.khZi/
close to 0, which is a trivial cover of
Sp.khZi/.
Conversely if X an ! IP1 .k/anLis a trivial cover above f´ 2 k j j´j 1g, the
local Tate algebra is a direct sum niD1 khZi, thus equal to the analytification of a
function algebra defined by a polynomial with k-rational zeroes at some point of the
unit disc. The corresponding extension of function fields K=k.t/ then has a prime
divisor which splits completely. t
u
462 VI Rigid Analytic Methods

2.2 The Inverse Problem over Complete Ultrametric Fields

For a; b 2 k we denote by

D.a; b/ WD f´ 2 k j j´  aj jbjg and D.a; b/ı WD f´ 2 k j j´  aj < jbjg

the closed respectively open ultrametric discs. (Note that both are open as well as
closed for the topology induced by the ultrametric valuation of k.)
The fundamental building blocks for arbitrary Galois covers are cyclic covers
with a base point.

Proposition 2.3. Let k be a field complete with respect to an ultrametric valuation

and n 2 IN. Then there exists a connected Galois cover X ! IP1 .k/an of ana-
lytic spaces with cyclic Galois group Zn of order n, which is a trivial cover above
IP1 .k/an n D.0; 1/ı .

Proof. We just have to translate the results of Chapter III.4 on cyclic extensions to
the present setting. It is clearly enough to prove the result for n the power of a prime
p. If p is different from the characteristic of k, then by Theorem III.4.5 there exists
a cyclic geometric Galois extension K=k.t/ in which the denominator divisor P1
of .t/ splits completely. If the characteristic of k equals p, Corollary III.4.8 gives
the same result. Application of Proposition 2.2 completes the proof. t
u

The main result is now achieved by gluing the cyclic Galois covers.

Proposition 2.4. Let k be a field complete with respect to an ultrametric valuation.

Then for every finite group G there exists a connected Galois cover X ! IP1 .k/an
with group G, which is a trivial cover above IP1 .k/an n D.0; 1/ı .

Proof. We follow the proof given by Liu (1995). For cyclic groups, the assertion
is just Proposition 2.3. By induction it then suffices to show, if G D hH1 ; H2 i is
generated by two subgroups for both of which the assertion holds, then it also holds
for G. Let 'i W Xi ! IP1 .k/an , i D 1; 2, be the coverings with group Hi whose
existence is assumed. For a fixed q 2 k, 0 < jqj < 1, the discs

D1 WD D.0; q 2 /; D2 WD D.q; q 2 /

are disjoint. We write D1ı WD D.0; q 2 /ı and D2ı WD D.q; q 2 /ı . By change of coor-
dinates we may and will assume that 'i is a trivial cover above IP1 .k/an n Diı .
Let D0 WD IP1 .k/an n .D1ı [ D2ı / and XQi WD 'i1 .Di /. Above Di , i D 1; 2, let
Yi WD IndG Q G
Hi .Xi / be induced covers and above D0 let Y0 WD Ind1 .D0 / be the
induced covering of the identical covering. For i D 1; 2 both coverings Y0 ! D0 and
Yi ! Di are trivial on the intersection D0 \ Di D Di n Diı and therefore isomorphic
to each other on this intersection. We may thus apply the Gluing Propositions 1.4
and 1.5 to deduce the existence of a G-equivariant analytic cover Y ! IP1 .k/an .
By construction, Y0 is a disjoint union of connected subspaces Y0; for  2 G,
while Yi , i D 1; 2, is a disjoint union of connected spaces Yi; for  2 G=Hi . But if
2 The Inverse Problem over Qp .t / and IFp .t / 463

 2 Hi then Y0; \ Yi; ¤ ;. Since H1 and H2 generate G, Corollary 1.7 shows
that Y is connected, and Y ! IP1 .k/an is Galois with group G. By construction,
its restriction to IP1 .k/an n D.0; 1/ı is a trivial cover since D1 [ D2  D.0; 1/ı . u
t
Theorem 2.5 (Harbater (1987)). Let k be a field complete with respect to an ultra-
metric valuation. Then the inverse Galois problem over k.t/ has a positive solution.
Proof. Let G be a finite group. Then by Proposition 2.4 there exists an analytic
connected Galois cover X ! IP1 .k/an with group G. By the GAGA-principle
Corollary 1.13 this is the analytification of an algebraic connected Galois cover
XQ ! IP1 .k/ with Galois group G. The function field N D k.XQ / of XQ then yields
a Galois extension of k.t/ D k.IP1 .k// with Galois group G. t
u
Corollary 2.6. The inverse Galois problem over Qp .t/ has a positive solution.
Remark. Together with Corollary I.1.7 the above Corollary shows that the inverse
problem of Galois theory is solved for rational function fields over all completions
of Q. By Theorem 2.5 the corresponding result holds more generally for rational
function fields over the completions of arbitrary global fields.
When choosing rational ramification points for the cyclic extensions glued in
Proposition 2.4, Corollary 2.6 yields Galois extensions defined over the field Q0p WD
Qp \ QN of algebraic p-adic numbers, thus solving the inverse problem over Q0p .t/
(see Harbater (1987), Cor. 2.11).

2.3 The Inverse Problem over IFp .t /

By a standard specialization argument it is possible to derive a solution of the inverse

Galois problem over IFp .t/ from the above result for complete non-archimedean
fields.
Theorem 2.7 (Harbater (1984)). The inverse Galois problem over IFp .t/ has a pos-
itive solution.
Proof. The power series field k WD IFp ..u// is complete ultrametric, so by The-
orem 2.5 for any finite group G there exists a connected algebraic Galois cover
XQ ! IP1 .k/ with group G. This cover is already defined over a finitely gener-
ated extension ring R D IFp Œx1 ; : : : ; xs of IFp , yielding a cover ' W XQR ! IP1 .R/
with smooth generic fiber which becomes isomorphic to XQ ! IP1 .k/ after base
change to k. Since IFp is algebraically closed, there exist infinitely many IFp -
rational points on XQR . By the Bertini–Noether Theorem (see Fried and Jarden
(1986), Prop. 8.8) infinitely many of these give specializations such that XQR remains
irreducible, i.e., there exists a maximal ideal m G R, with R=m Š IFp , such that
XQR IP1 .R/ IP1 .IFp / ! IP1 .IFp / is connected with Galois group G. The function
field of this cover yields a Galois extension of IFp .t/ with the desired Galois group.
t
u
464 VI Rigid Analytic Methods

A refinement of the above argument makes it possible to solve split embedding

problems over K D IFp .t/ and thus to show that the absolute Galois group of such
fields K is free profinite of countable rank.

2.4 The Conjecture of Šafarevič for IFp .t /

The conjecture of Šafarevič (see IV, 3.3) states that the absolute Galois group of
Qab is free profinite of countable rank. In the usual analogy between number fields
and function fields over finite fields, one is led to ask a similar question for IFp .t/,
the maximal cyclotomic extension of IFp .t/. In this case the rigid analytic methods
yield a positive answer, found by Harbater (1995a) and Pop (1995). The simplified
proof presented here, which only uses gluing arguments from Paragraph 1, was
found independently by Haran and Völklein (1996) and van der Put.
We start by solving certain split geometric embedding problems E .'; / (see
Ch. IV.1 for the notation).

Proposition 2.8. Let k be a complete ultrametric field and E .'; / a finite split geo-
metric embedding problem over k.t/ with fixed field N of ker.'/. If there exists a
prime divisor P 2 IP.k.t/=k/ which splits completely in N=k.t/, then E .'; / has
a proper geometric solution.

Proof. Let E .'; / a finite split embedding problem over k.t/ belonging to the split
exact sequence

1 ! H D ker./ ! GQ D H G ! G D Gal.N=k.t// ! 1:

By Proposition 2.2 for u 2 k with juj < 1 the analytification X of a suitable model
of N=k.t/ may be taken such that its restriction X1 ! D1 WD D.0; u2 / is connected
with group G and trivial above D1 n D1ı . Further, by Proposition 2.4 there exists a
connected cover X2 ! D2 WD D.u; u2 / with group H D ker./ which is a trivial
Q Q Q
cover above D2 n D2ı . Let Y1 WD IndG G .X1 /, Y2 WD IndH .X2 /, and Y0 WD Ind1 .D0 /
G G
ı ı
for D0 WD IP .k/ n .D1 [ D2 /. This is a special case of the setup in the proof of
1 an

Proposition 2.4. In particular, Yi , i D 0; 1; 2, may be glued to a connected cover

Y ! IP1 .k/an with group GQ D hH; Gi.
Since H is a system of coset representatives of G in G, Q by the construction

in the proof of Proposition 2.1 we have Y1 D H  X1 D [ 2H X1; with H -
action .X1; / WD X1; . Thus the canonical map X1 D X1;1 ! Y1 =H is an

isomorphism. Similarly we have Y2 D [ 2G X2; , with .; x/ D .; t .x// for


.; x/ 2 X2; and  2 H . Thus

[
: [
: [
:
X2; =H D .X2; =H / D D2; D IndG
1 .D2 /;
2G 2G 2G
2 The Inverse Problem over Qp .t / and IFp .t / 465

where D2; D X2; =H Š D2 . Moreover we trivially have Y0 =H D IndG 1 .D0 /. This

shows that Y =H coincides with the trivial cover IndG1 .D 0 [ D 2 / above D0 [ D2 ,
while it equals X1 ! D1 above D1 . It follows that Y =H Š X .
Via the GAGA-principle (Corollary 1.13), Y ! IP1 .k/an corresponds to a con-
nected Galois cover YQ ! IP1 .k/ with group G,Q such that the extension of function
Q
fields k.Y /=k.t/ contains N=k.t/ as a Galois subextension. Since by construction
Q 2 GQ acts via .Q / on the quotient Y =H Š X this yields a proper solution of
E .'; /. t
u

Extension of constants now allows to obtain the following consequence of Propo-

sition 2.8 and Theorem 2.7:

Corollary 2.9. Every split embedding problem over IFp .t/ has a proper solution.

Proof. Let E .'; / be an embedding problem over IFp .t/ and N the fixed field of
ker.'/ with Galois group G D '. /. Without loss of generality we may assume that
N=IFp .t/ is unramified at the denominator divisor P1 of t, and hence P1 splits
completely. Extension of constants with the complete field k WD IFp ..u// trans-
forms E .'; / to a geometric embedding problem E .'  ;   / with N  D N k and
Gal.N  =k.t// Š G with a prime divisor P1 which splits completely. Since the
power series field k is complete ultrametric, by Proposition 2.8 the split embedding
problem E .'  ;   / possesses a proper geometric solution. Let NQ  =k.t/ be a solu-
Q and YQ  ! IP1 .k/ a corresponding normal
tion field, such that Gal.NQ  =k.t// Š G,
model. By the argument in the proof of Theorem 2.7 there exists a specialization of
YQ  ! IP1 .k/ to a connected cover YN ! IP1 .IFp /, i.e., with Gal.NQ =IFp .t// Š G,Q
where NQ WD IFp .YN /. Since N  was obtained from N by extension of constants, any
such specialization of YN will be such that NQ contains N and is a solution field of
the original embedding problem E .'; /. It gives a solution 'Q by the reasoning in
the proof of Theorem IV.1.5(a). t
u

Theorem 2.10 (Harbater (1995a), Pop (1995)). The absolute Galois group of IFp .t/
is free profinite of countable rank.

Proof. By the Theorem of Tsen (Theorem IV.1.11(a)) the absolute Galois group
WD IFp .t / of IFp .t/ is projective. Since IFp .t/ is countable, can have at most
countable rank. Thus by the Freiheitssatz of Iwasawa (Theorem IV.1.13) the result
is proved if every finite embedding problem for has a proper solution. By the
Theorem of Ikeda (Theorem IV.1.9) it even suffices to properly solve split embed-
ding problems. But these always have proper solutions by Corollary 2.9. t
u

Corollary 2.11. Let k be an algebraically closed countable field, K=k a function

field in one variable. Then K is a free profinite group (of countable rank).

Proof. In characteristic 0 this is Theorem I.2.2, while in positive characteristic it

follows from the previous Theorem 2.10 since subgroups of finite index of free
profinite groups of countable rank are again free of countable rank. t
u
466 VI Rigid Analytic Methods

Remark. The above corollary even holds without restriction on the cardinality of
k. Then the rank of K is equal to the cardinality of k. This can be proved by
showing that the number of solutions to the embedding problems considered in the
proof of Theorem 2.10 is equal to the cardinality of k and then invoking a result of
Melnicov (1980) (see Pop (1995), Thm A and Cor. on p. 556, as well as Harbater
(1995a), Lemma 4.3 and Thms. 3.5, 3.6 and 4.4, for the details, and for a more
general assertion where the ramification can be controlled).
3 Free Quotients of the Fundamental Group 467

3 Free Quotients of the Fundamental Group

The structure of the fundamental group of C.t/ is well known by the Riemann Exis-
tence Theorem. If C is replaced by a non-algebraically closed field which is com-
plete with respect to a non-archimedean valuation, no comparable result is known
at present. In this section we give at least partial information by constructing a large
free factor of the fundamental group, using rigid analytic methods, on which the
action of the Galois group of the field of constants can be described explicitly.

3.1 Free Composites of Galois Extensions

Let k be complete with respect to an ultrametric valuation and K D k.t/. We con-

sider a finite family M D fNi =K j i 2 I g, jI j < 1, of finite Galois extensions Ni
of K inside a fixed algebraic closure KN of K, regular over k. Let YQi be the normal-
ization of IP1 .k/ in Ni , and 'i W YQi ! IP1 .k/ the corresponding covering of curves.
Denote by X WD IP1 .k/an , Yi WD YQi for i 2 I , the rigid analytifications. Now
an

M is called a compatible family if moreover there exists an admissible covering

fU g [ fUi j i 2 I g of X by affinoid spaces such that
(1) the Ui , i 2 I , are non-empty and pairwise disjoint,
(2) 'i has no ramification above U and Uj for i ¤ j ,
(3) the preimage Vi WD 'i1 .Ui /  Yi of Ui is connected,
(4) the preimage @Vi WD 'i1 .@Ui / of @Ui WD U \Ui contains an admissible subset
@Ui;1 which is isomorphic to @Ui via 'i .
The above conditions can be interpreted as controlling the ramification locus of the
geometric field extensions Ni =k.t/: in a certain sense, they have to be sufficiently
separated.
The conditions (1)–(4) have some obvious consequences. First, the @Ui are non-
empty since IP1 .k/an is connected, and pairwise disjoint since the Ui are. Thus the

union @U WD [ @Ui is disjoint. The analytic projective line X can be recovered by

gluing [ Ui with U along @U . Further, assumption (2) implies that the Ni =K are
linearly disjoint. Indeed, the extension .Ni \ Nj /=K is unramified for i ¤ j , hence
trivial since there exist no unramified field extension of K D k.t/.

Proposition 3.1. Let k be complete with respect to an ultrametric valuation,

Q KD
k.t/, M D fNi =K j i 2 I g a compatible family with composite N D i 2I Ni and
Galois groups Gi D Gal.Ni =K/. Then there exists a geometric Galois extension
M=K with Galois group ˘ WD .X —i 2I Gi /O the profinite completion of the free prod-
uct of the Gi , having N=K as a subextension.

Proof. Since M is compatible, there exists an admissible covering fU g [ fUi j

i 2 I g of X satisfying (1) to (4). The automorphism group Gi WD Deck.'i / is
anti-isomorphic to the Galois group Gi of Ni =K. For  2 Gi we define @Ui; WD
468 VI Rigid Analytic Methods


S the @Ui; for  2 Gi are connected and pairwise
.@Ui;1 /. Then by our assumptions
disjoint subsets of @Vi , and 2G  @Ui; is a Gi -invariant admissible covering
i
of @Vi . Furthermore, for ;  2 Gi we have .@Ui; / D @Ui;  . In other words,
G
@Vi Š Ind1 i .@Ui /.
Let now  be the kernel in the natural exact sequence
Y
1 !  ! ˘ ! Gi D Gal.N=K/ ! 1 : (3.1)
i 2I

Let   be an open normal subgroup of ˘ and H the quotient group


1 !  ! ˘ ! H ! 1:

The Gi are naturally embedded into ˘ and their images generate ˘Q topologically.
This implies H D h.Gi / j i 2 I i, and since   and ˘= Š Gi we have
.Gi / Š Gi . We write H  for the opposite group of H . Let UQ WD IndH

1 .U / and
VQi WD IndH

G
.Vi /. So by definition
i

[
: [
:
UQ D H   U D U ; VQi D H  =Gi  Vi D Vi; ; (3.2)
2H  2H  =Gi

and according to Proposition 2.1 the elements  2 H  act on UQ and VQ via

.; U / D .; U/; .; Vi / D . 0 ;  0 .Vi //;

where  D  0  0 for some  0 2 Gi . We now glue UQ and VQi , i 2 I , over

[
: [
:
@UQ WD IndH

1 .@U / D @Ui; (3.3)
2H  i 2I

and
[
: [
: [
:
@VQi WD IndH

G
.@Vi / D @Vi; D @Ui;; (3.4)
i
2H  =Gi 2H  =Gi 2Gi

by identifying @Ui;i ; i in (3.4) with @Ui; i i in (3.3). By the Gluing Proposi-

tions 1.4 and 1.5 this yields an H  -equivariant analytic cover YH ! X . By the
same reasoning as in the proof of Proposition 2.4 this is connected since H is gen-
erated by the .Gi /. Further, YH =H  Š X , so the cover is Galois with group H  .
By the GAGA-principle (Corollary 1.13(b)) this is the analytification of a connected
algebraic cover YQH ! IP1 .k/ with Galois group H  and factoring through the YQi .
The corresponding extension NH =K of function fields is geometric by Proposi-
tion 2.2 since YH ! X is the trivial cover above U .
For a sequence of epimorphisms
 0
˘ ! H ! H 0 ;
3 Free Quotients of the Fundamental Group 469

with corresponding analytic spaces YH ; YH 0 , by construction YH = ker. 0 / is canon-

ically isomorphic to YH 0 . Thus the above construction yields a projective system
fYH ! X gH of analytic covers and fYQH ! IP1 .k/gH of connected algebraic cov-
ers. For the corresponding geometric function field extensions NH =K this translates
to an injective system, with projective system of Galois groups H D Gal.NH =K/.
Taking the direct limit yields a geometric field extension M WD [H NH , with Galois
group ˘ over K, which contains the composite N of the Ni . t
u

3.2 Galois Action

We now refine the procedure of the first section to accommodate an additional

Galois action. Let k0 be a complete ultrametric field, k=k0 a finite Galois exten-
sion with group G, K0 D k0 .t/, K D kK0 D k.t/ and X D IP1 .k/an . A compatible
family M D fNi =K j i 2 I g with respect to K=k is called a G-compatible family
if moreover
(5) M is K0 WD Gal.KN 0 =K0 /-invariant,
(6) the admissible covering fU g [ fUi j i 2 I g of X is G-invariant.
Q
Since the composite N WD Ni is Galois over K0 by assumption (3), we have an
exact sequence

1 ! Gal.N=K/ ! Gal.N=K0 / ! G ! 1

of Galois groups, which splits since IP1 .k/ has k-rational points. A group theoretic
section W G ! Gal.N=K0 / of this sequence of Galois groups is called an M -
section, if for all i 2 I the connected component @Ui;1 of @Vi (which is isomorphic
to @Ui via 'i ) is invariant under (the opposite group of) the stabilizer of Ni in .G/.

Lemma 3.2. Let W G ! Gal.N=K0 / be an M -section for the G-compatible family

M . Then there exists a labeling @Ui; ,  2 Gi , of the connected components of @Vi
such that for all  2 G  we have

.@Ui; / D @Uj;  ; where j is defined by   WD 1 2 Gj :

Proof. It is clearly sufficient to prove this for a G-orbit on M . For such an orbit
we choose a fixed Ni with stabilizer Hi in G. Then we may label the components
of @Vi by  2 Gi , starting from the component @Ui;1 of @Vi fixed by Hi , whose
existence is guaranteed by the fact that is an M -section, such that

 0 .@Ui; / D @Ui; 0 

for all ;  0 2 Gi . Next, for  2 G  and j with Nj D Ni , we set @Uj; WD .@Ui; /,
where  D   2 Gj for  2 Gi . To show that this labeling is well-defined it is clearly
sufficient to consider the case i D j , and  D 1. But then  lies in Hi and stabilizes
470 VI Rigid Analytic Methods

@Ui;1 . This yields

.@Ui; / D ..@Ui;1 // D  ..@Ui;1 // D .@Uj;1 / D @Uj; ;

proving the assertion. t

u

Q Any section W G ! Gal.N=K0 / defines an action of G on Gal.N=K/ Š

i 2I Gi and thus also on ˘ D .X i 2I Gi /O. With respect to this we can speak of
—
the semidirect product ˘ G.

Theorem 3.3. Let M be a G-compatible family and an M -section. Then the field
extension M=K constructed in Proposition 3.1 is Galois over K0 and can be
extended to M W G ! Gal.M=K0 / such that

1 ! Gal.M=K/ ! Gal.M=K0 / ! G ! 1

is canonically isomorphic to

1 ! ˘ ! ˘ G ! G ! 1;

where ˘ G is defined with respect to the action induced by .

Q
Proof. Let  be the kernel of ˘ ! i 2I Gi as in (3.1). We show that the construc-
tion in Proposition 3.1 of the universal field extension M of K attached to M is
compatible with the action of G. Let   be an open G-invariant normal sub-
group of ˘ and H D ˘= the quotient group. We identify H G Š .˘ G/=. It
now suffices to prove that the fixed field NH of  in M is normal over K0 and the
section W G ! Gal.N=K0 / may be prolonged to a section H W G ! Gal.NH =K0 /
such that the sequence

1 ! Gal.NH =K/ ! Gal.NH =K0 / D Gal.NH =K/: H .G/ ! Gal.K=K0/ ! 1

becomes canonically isomorphic to

1 ! H ! H G ! G ! 1: (3.5)
 Q
S: define a natural action ofG as group of analytic automorphisms on U
We may
and i 2I VQi as in (3.2): For  2 G let

.; x/ D . ; .x// for .; x/ 2 H   U D UQ ;

.; x/ D . ; .x// 2 Hj  Vj for .; x/ 2 Hi  Vi and   2 Gj ; (3.6)

where we have identified H  =Gi with the kernel Hi of the canonical epimorphism
Q
H  ! Gi . In this action H  is normalized by G  , extending the action on i 2I Gi
via and thus defining a semidirect product H G with section H extending .
Furthermore, by the construction in Lemma 3.2 the action (3.6) of G  is compatible
with the gluing as in the proof of Proposition 3.1 used to define YH . So we obtain an
action of H  G  as a group of analytic automorphisms of YH with YH =H  Š X .
3 Free Quotients of the Fundamental Group 471

This shows that the exact sequence

1 ! Deck.Y ! X / ! Deck.Y ! X0 / ! Deck.X ! X0 / ! 1

is canonically isomorphic to (3.5). Let NH denote the function field of the cor-
responding cover YQH ! IP1 .k/ of algebraic curves existing by Corollary 1.13(b).
Then by the above NH is Galois over K0 with group
Gal.NH =K0 / D Gal.NH =K/ H .G/ Š H G:
This proves that the procedure in the proof of Proposition 3.1 can be made compat-
ible with the Galois action by G. Going to the projective limit of the covers YH ,
respectively the injective limit of the field extensions NH , we obtain the assertion
of the theorem. t
u

3.3 A Free Quotient of the Algebraic Fundamental Group

We now show that G-compatible families with good sections exist in one important
case. In this section we assume that the characteristic of the residue class field of k0
with respect to the ultrametric valuation equals the characteristic of k0 . The more
general case of mixed characteristic is studied in Pop (1994).
Let k0 be a complete ultrametric field. A finite set
S D fP1 ; Q1 ; : : : ; Pr ; Qr g  IP1 .kN0 /
of 2r points of IP1 .kN0 / stable under k0 with corresponding set of prime divisors
S D fP1 ; Q1 ; : : : ; Pr ; Qr g  IP.kN0 .t/=kN0 / is called pairwise adjusted (with respect
to k0 ) if there exists a finite Galois extension k=k0 with group G D Gal.k=k0 / such
that
(7) the set of pairs f.Pi ; Qi / j i D 1; : : : ; rg forms a system of imprimitivity for the
action of k0 ,
(8) there exists an admissible covering fU g [ fUi j i 2 I g of X D IP1 .k/an with
Pi ; Qi 2 Ui satisfying (1) of Section 3.1 and (6) of Section 3.2, and a G-
invariant family of GPi -equivariant analytic isomorphisms

fi W Ui ! D.0; 1/ j i 2 I g

with i .Pi / D 0, i .Qi / 2 D.0; 1/ı and i .Ui \ U / D D.0; 1/ n D.0; 1/ı,
where GPi denotes the decomposition group of Pi in G.
Lemma 3.4. Let S D fP1 ; Q1 ; : : : ; Pr ; Qr g be pairwise adjusted. Furthermore,
let n D .n1 ; : : : ; nr / 2 INr with ni D nj if Pi and Pj are G-conjugate. Then for all
suitably large k=k0 there exists a G-compatible family M D fNi =k.t/ j 1 i rg
such that for all i 2 I the extension Ni =k.t/ is only ramified in fPi ; Qi g and has
Galois group Gal.Ni =k.t// Š Zni .
472 VI Rigid Analytic Methods

Proof. We fix n D .n1 ; : : : ; nr / 2 INr with ni D nj if Pi and Pj are G-conjugate.

Denote by m the p-prime part of the least common multiple of the ni , i D 1; : : : ; r,
where p is the characteristic of k0 . Assume that k contains the m-th roots of
unity. In Proposition 2.3 we constructed a connected cyclic Galois cover YQi of
D.0; 1/ with group Zni . For the p-part of ni we replace the Galois extension
from Corollary III.4.8 by the cyclic Galois extension of k.t/ in Theorem III.4.7
described by the specialization t 7! .1=.t.t  1//; 0; : : : ; 0/ and again denote by YQi
the connected cyclic Galois cover obtained by analytification. As before, this is
a trivial cover above some open disc. Now by construction in all cases the cor-
responding cyclic Galois extension NQ i =k.t/ is totally ramified in two points and
Gal.NQ i =k.t// is generated by the inertia elements i ; i at the two ramification
points subject to the relation i i D 1. Furthermore k acts via the cyclotomic char-
acter as iı D ic.ı/ for ı 2 k . Via i1 the cover YQi ! D.0; 1/ may be transported
to a Galois cover Yi ! Ui with group Gi Š Zni trivial above @Ui WD Ui \ U .
S G
Above Ui0 WD U [ j ¤i Uj we consider the trivial cover Ind1 i .Ui0 /. Gluing above
@Ui D Ui \Ui0 gives a connected Galois cover 'i W Yi ! X with group Zni . By the
GAGA-principle this gives rise to a family M D fNi =k.t/ j i D 1; : : : ; rg of cyclic
Galois extensions of k.t/ with groups Zni . More precisely, Ni =k.t/ is generated by
a root of
t  qi
x .ni /p0  D0 (3.7)
t  pi
together with the coordinates of a solution x of

xp  x D .1=..t  pi /.t  qi //; 0; : : : ; 0/ (3.8)

in the Witt ring W .k.t// of Witt vectors of length .ni /p . Here pi (respectively qi )
is such that Pi is the numerator divisor of .t  pi / (respectively Qi is the numerator
divisor of .t  qi /).
The compatibility of the family M is clear from the definitions. It remains to
show that M is k0 .t / -invariant. But this follows since Ni D Niı0 if ı.Pi / D Pi 0
for all ı 2 k0 .t / and from property (8). t
u

After these preparations we can prove a result of Pop (1994) which may be
thought of giving one half of the Riemann Existence Theorem in the case of a base
field k0 complete with respect to a non-archimedean valuation, but with precise
information on the action of k0 on the free normal subgroup (compare with Theo-
rem I.2.6).

Theorem 3.5 (Pop (1994)). Let k0 be a field complete with respect to a non-
archimedean valuation such that the characteristic of k0 equals the character-
istic of the residue class field with respect to this valuation. Further let S D
fP1 ; Q1 ; : : : ; Pr ; Qr g  IP1 .kN0 / be pairwise adjusted. Then the arithmetic funda-
mental group 1 .IP1 .kN0 / n S / k0 has a quotient ˘  k0 such that
alg

˘ D h˛1 ; ˇ1 ; : : : ; ˛r ; ˇr j ˛i ˇi D 1 for 1 i riO; (3.9)

3 Free Quotients of the Fundamental Group 473

where the elements ˛i ; ˇi are generators of inertia groups above the ramified prime
divisors Pi ; Qi for 1 i r.
Furthermore, k0 acts on ˘ via the cyclotomic character c as

˛iı D ˛.ic.ı/
/ı
for ı 2 k0 : (3.10)

Proof. Let n D .n1 ; : : : ; nr / 2 INr be k0 -invariant, i.e., with ni D nj whenever

Pi and Pj lie in the same orbit under the action of k0 . Let kn be the cyclo-
tomic extension of k0 with those p-prime roots of unity whose order divides one
of the ni , and set Kn WD kn .t/. Then by Lemma 3.4 there exists a G-compatible
family Mn D fNn;i =Kn j 1 i rg of cyclic Galois extensions with groups
Gi WD Gal.Nn;i =Kn / D Zni .
We next show that there exists an Mn -section. By construction the Gi are gen-
erated by inertia elements i ; i subject to i i D 1, and k0 and hence Gn WD
Gal.Kn =k0 .t// act on these via

iı D .ic.ı/
/ı
; where ı 1 .Pi / D Pi.ı/ for ı 2 K0 :

In particular the extension

1 ! Gal.Nn =Kn / ! Gal.Nn =k0 .t// ! Gn ! 1

Qr
with Nn WD i D1 Nn;i splits. Let P be the denominator divisor of .t/. By the
Remark after Theorem I.3.5 together with Proposition I.3.1 this provides a sec-
tion from k0 to k0 .t / and thus a section n W Gn ! Gal.Nn =k0 .t//, after pos-
sibly extending kn by a finite step. We claim that n is an Mn -section. For this let
Hi Gn be the stabilizer of Nn;i (via n ) and write ki WD k Hi for the fixed field in k.
The analytic spaces Yi can be regarded as analytic spaces over ki and (the opposite
group of) Hi acts on Yi by analytic automorphisms. Let Yi 0 be the preimage under
'i of Ui0 , an admissible Hi -invariant subset of Yi . Thus Hi permutes the connected
components of Yi 0 , and this action is the same as the one on the formal power series
solutions of equations (3.7) and (3.8). But these clearly have solutions in the decom-
position field of P as can be seen by Hensel’s lemma. Thus n is an Mn -section.
Hence Theorem 3.3 applies to the Gn -compatible family Mn and the Mn -section
n to yield a Galois extension Mn of Kn with group

r
X Zni O:
˘n WD Gal.Mn =Kn / D — (3.11)
i D1

The system of G-invariant tuples n 2 INr is projective, and since the above con-
structions were all canonical, starting from the pairwise adjusted set S , we obtain
an injective system of field extensions

fkN0 Mn j n 2 INr is G-invariantg

of kN0 .t/ with projective system of Galois groups f˘n g from (3.11). The limit over
the ˘n is clearly given by ˘ in (3.9). The action of k0 on ˘ is obtained from the
corresponding action on the intermediate levels. t
u
474 VI Rigid Analytic Methods

4 Large Fields

In this paragraph we study a class of fields for which the inverse Galois problem has
a positive solution. This includes rational function fields over PAC-fields and over
complete fields as special cases. As an application the absolute Galois group of a
Hilbertian PAC-field of arbitrary characteristic is shown to be free profinite.

4.1 Existentially Closed Fields

Let K  =K be an extension of fields. Then K is existentially closed in K  if for every

Zariski-closed subset A of An .K/, A.K  / ¤ ; implies A.K/ ¤ ;. This notion is
transitive in the following sense: if K1 K2 K3 is a tower of fields and Ki
is existentially closed in Ki C1 for i D 1; 2, then K1 is existentially closed in K3 .
Conversely, if K1 is existentially closed in K3 , then at least K1 is also existentially
closed in K2 .

Lemma 4.1. Let K  =K be an arbitrary field extension such that K  has a K-

rational place. Then K  is regular over K.
P
Proof. Let } W K  ! K [ f1g be the K-rational place of K  . Let niD1 ai fi D 0
with ai 2 KN in an algebraic closure KN of K and fi 2 K  . Denote by }N W KN  !
KN [ f1g the extension of } with }.a/
N N Let j 2 f1; : : : ; ng be such
D a for all a 2 K.
N i =fj / 2 K and hence
that fj is the element with the largest order of pole. Then }.f

X
n X
n
N
}. ai fi =fj / D ai }.fi =fj / D 0;
i D1 i D1

but }.fj =fj / ¤ 0 which shows that a1 ; : : : ; an are linearly dependent over K. Thus
KN and K  are linearly disjoint. t
u

Thus in particular, if K is existentially closed in K  then K  is a regular exten-

sion of K.
Existentially closed field extensions behave well with respect to regular solutions
of split embedding problems:

Proposition 4.2. Let K be existentially closed in K  . Let E .'; / be a finite split

embedding problem over K and E .'  ;   / the corresponding embedding problem
over K  obtained by extension of constants with K  . Then E .'; / has a regular
proper solution if and only if E .'  ;   / has a regular proper solution.

Proof. Let E .'; / be a finite split embedding problem over K and N the fixed
field of ker.'/ with Galois group G D '. /. Since K  =K is regular, extension of
constants with K  transforms E .'; / to an embedding problem E .'  ;   / with
N  D NK  and Gal.N  =K  / Š G. Let 'Q be a regular proper solution of E .'; /,
4 Large Fields 475

with solution field NQ =K.t/ which we may assume to be linearly disjoint from
K  over K. The composition of the restriction map from Gal.K  .t/=K  .t// to
Gal.K  K.t/=K  .t// with the conjugate of 'Q by the natural isomorphism from
Gal.K.t/=K.t// to Gal.K  K.t/=K  .t// then gives a regular proper solution 'Q 
of E .'  ;   / with solution field NQ  WD NQ K  .
So now let 'Q  be a regular proper solution of E .'  ;   / with solution field
N =K  .t/ and Gal.NQ  =K  .t// Š G.
Q  Q Let y be a primitive element of NQ  =K  .t/
and f .t; Y / the minimal polynomial of y over K  .t/. Then there exists a finitely
generated subfield K.x/, x D .x1 ; : : : ; xr /, of K  such that f .t; Y / D f1 .t; x; Y / 2
K.t/Œx ŒY generates a Galois extension of K.t; x/ with group G. Q Since K  =K is
regular, so is K.x/=K, hence the algebraic variety V D Spec.KŒx / is absolutely
irreducible. By the Bertini-Noether Theorem (Fried and Jarden (1986), Prop. 8.8)
there exists a Zariski-open subset U  V such that for u 2 U the polynomial
f1 .t; u; Y / remains irreducible with group G. Q Since K is existentially closed in K 
the set U contains a K-rational point u0 . Then the splitting field of f1 .t; u0 ; Y / is
a geometric Galois extension NQ =K.t/ with group GQ since f1 .t; u; Y / is absolutely
irreducible. By the argument used in the proof of Theorem IV.1.5(a) it can be seen
to solve the embedding problem E .'; /. t
u
The following property of Henselian fields will be used in the sequel:
Proposition 4.3. Let K be a Henselian field and KO its completion. If K=K
O is a
O
separable extension, then K is existentially closed in K.
O
Proof. Let x D .x1 ; : : : ; xn / be a K-rational point of some closed affine set A. Let
y D .y1 ; : : : ; yn / be a separating transcendence basis (see Fried and Jarden (1986),
Lemma 9.5) of K.x/=K and ´ a primitive element of the finite separable extension
K.x/=K.y/ which is integral over KŒy . Thus there exists an irreducible polynomial
f .Y; Z/ 2 KŒY; Z with f .y; ´/ D 0 and f 0 .y; ´/ ¤ 0. Since K is dense in K, O there
exist points in K arbitrarily close to y and ´. As K is Henselian, there exist a; b 2 K
with f .a; b/ D 0. This gives a specialization of y; ´ to K and hence a specialization
of x to a K-rational point of A. t
u

4.2 Characterization of Large Fields

A field k is called large if every absolutely irreducible curve over k with a simple
k-rational point has infinitely many k-rational points. Note that this is equivalent to
the property: Every function field of one variable over k that has a k-rational point
has infinitely many k-rational points. The class of large fields contains a number of
interesting examples:
Proposition 4.4. The following fields are large:
(a) all PAC-fields,
(b) all real closed fields,
(c) all Henselian fields.
476 VI Rigid Analytic Methods

Proof. The first assertion is immediate from the definition of PAC-fields.

For (b) let C be an absolutely irreducible curve over the real closed field k with
a simple k-rational point. Let f .X; Y / 2 kŒX; Y be the equation for a model of an
affine part of C such that .0; 0/ becomes the simple k-rational point. Thus there
exists a small neighborhood U of 0 such that f .0; y/ takes positive as well as
negative values for y 2 U . Hence for all small enough x0 the polynomial f .x0 ; y/
also takes positive and negative values in a neighborhood of y D 0. By the mean
value theorem this implies the existence of a simple zero y0 of f .x0 ; y/, hence the
existence of infinitely many k-rational points on C , proving (b).
Let k be Henselian, f .X; Y / 2 kŒX; Y the equation for a model of an affine part
of an absolutely irreducible curve C over k and .x; y/ with f .x; y/ D 0 a simple
point. Since k is Henselian, for all x 0 sufficiently close to x in the topology induced
by the Henselian valuation of k, there exists y 0 2 k with f .x 0 ; y 0 / D 0 such that
.x 0 ; y 0 / is again simple. t
u

Remark. According to Pop (1996) the field Qtr of all totally real numbers as well
as the totally p-adic fields are also large. It is not known whether Qab is large.

Lemma 4.5. Let k be a field. Then the extension k..u//=k.u/h of the field of formal
power series over the Henselization of k.u/ with respect to the place }u W u 7! 0 is
separable.

Proof. We may assume that the characteristic p of k is positive. According to

Fried and Jarden (1986), Lemma 9.5, k..u//=k.u/h is separable if k..u// and
1
.k.u/h / p are linearly disjoint over k.u/. Since k.u/h is separable algebraic over
1
k.u/ this holds if k..u// and .k.u// p are linearly disjoint. Any finite subexten-
1 1 1 1
sion of .k.u// p =k.u/ is contained in a field L D k.a1p ; : : : ; anp /.u p / for suit-
able a1 ; : : : ; an 2 k. To prove linear disjointness of L and k..u// it suffices to
check that ŒLk..u// W k..u// D ŒL W k.u/ . Clearly the residue field extensions
of L=k.u/ and Lk..u//=k..u// with respect to } coincide. Since moreover in
both field extensions the value group is enlarged by the factor p we conclude
ŒLk..u// W k..u//  ŒL W k.u/ and hence the result. t
u

We can now give the following connection with existentially closed field exten-
sions due to Pop (1996).

Theorem 4.6. For a field k, the following are equivalent:

(1) k is large.
(2) k is existentially closed in the field of formal power series k..u//.

Proof. First assume that k is large. Since k..u//=k has the rational place }u W u 7! 0
it follows from Lemma 4.1 that k..u//=k is a regular extension.
Let k.u/h denote the Henselization of k.u/ with respect to the place }u . Then
k..u// is the completion of k.u/h , and since k..u//=k.u/h is separable by
Lemma 4.5 it suffices by Proposition 4.3 to show that k is existentially closed in
k.u/h . Let x D .x1 ; : : : ; xr / be a k.u/h -rational point of some Zariski-closed affine
4 Large Fields 477

set A . Then the field k.x/ generated by the coordinates is a finitely generated alge-
braic extension of k.u/, so is a function field of one variable over k. By construction
k.x/ has a rational point, so by assumption it has infinitely many rational points. At
all but finitely many of these, x is finite, so specialization yields infinitely many
points in A .k/.
Conversely, assume that k is existentially closed in k..u//. Let C be an abso-
lutely irreducible curve over k and f .x; y/ D 0 a model for an affine part of C
containing a simple k-rational point, with coordinates .x1 ; y1 / say. We claim that
then C has infinitely many k-rational points. Indeed, assume by induction that we
found points with coordinates .x1 ; y1 ; /; : : : ; .xn ; yn /. Consider the Zariski-closed
affine set A defined by the equations

f .X; Y / D 0; Zi .X  xi / D 1 for i D 1; : : : ; n:

By setting X WD x1 C um for m sufficiently large we see that A has a point in the

Henselian field k..u//. Since k is existentially closed in k..u// this implies that
A has a k-rational point .x; y; ´1 ; : : : ; ´n /, so .xnC1 ; ynC1 / WD .x; y/ yields a k-
rational point of C with xnC1 ¤ xi for i D 1; : : : ; n. This shows that k is large. u t

4.3 Split Embedding Problems over Large Fields

Proposition 4.7. Let k be a large field. Then every finite split embedding problem
over k..u// has a regular proper solution.

Proof. Let E .'; / be a finite split embedding problem over K WD k..u// and N
the fixed field of ker.'/ with Galois group G D '. /. It suffices to find a proper
solution of the lifted embedding problem E .'  ;   / over K  WD K.t/ with fixed
field N  D N.t/ of '  and G  D Gal.N  =K  / Š G such that N is algebraically
closed in the solution field. Let 1 ; : : : ; r be generators of the kernel H and set
Ci WD hi i. Let fCi; j 1 i r;  2 G  g be a set of copies of the Ci with isomor-

phisms i; W Ci; ! Ci . We denote by i; the preimages under i; of i . Let

˚ WD . —
X Ci; /O
1i r
2G 

be the profinite completion of the free product of the Ci; . Then G  acts on ˚ via
.i;  / D i; . We denote the semidirect product of ˚ and G  with respect to this
action by GO  . By construction there is a natural epimorphism from GO  onto G,
Q so
it suffices to solve the split embedding problem E .'  ; O  / associated to

1 ! ˚ ! GO  ! G  ! 1 (4.1)

where O  is the composite of the natural epimorphism GO  ! H G  with   .

478 VI Rigid Analytic Methods

Let fx 2 N j  2 G  g be a normal basis of N  =K  and choose ai 2 K  such

that all xi; WD ai x are distinct for 1 i r,  2 G  . Further let bi 2 K  be
different from ai . If the bi are close enough to the ai in the topology induced by the
ultrametric topology of K then the set of pairs S WD f.xi; ; yi; /g with yi; WD bi x
is pairwise adjusted in the sense of Paragraph 3.
By Theorem 3.5 there exists a quotient ˘  K  of the arithmetic fundamental
group of IP1 .KN  / n S with

˘ D h˛i; ; ˇi; j ˛i; ˇi; D 1; 1 i r;  2 G  iO;

K  acts on ˘ via the cyclotomic character c as in (3.5), and the ˛i; , ˇi; are
generators of the inertia groups above the prime divisors corresponding to xi; , yi;
respectively. Then the epimorphism
c./
W ˘ ! ˚; .˛i; / WD i; ;

satisfies
c./ c./ c.ı/c.ı 1 / c.ı/
.˛i; /ı D .i; /ı D i;ı 1  D i;ı 1  D ı
.˛i;ı 1  / D .˛i; /:

Thus is compatible with the action of G  Š Gal.N  =K  / and we obtain a com-

mutative diagram with surjective vertical arrows

1 ! ˘ ! ˘  K  ! K  ! 1
# '  # ' # (4.2)
1 ! ˚ ! ˚G ! G  ! 1:


In particular 'Q  WD '  gives a solution for the embedding problem (4.1) such
that N is algebraically closed in the solution field and thus the proof is complete.
t
u

Theorem 4.8 (Pop (1996)). Let k be a large field. Then every finite split embedding
problem E .'; / over k has a regular proper solution. In particular every finite
group occurs as the Galois group of a geometric Galois extension of the rational
function field k.t/.

Proof. Let E .'; / be a finite split embedding problem over k and E .'  ;   / the
corresponding embedding problem obtained by extension of constants with k  WD
k..u//. By Proposition 4.7 this has a proper regular solution. Since k is existentially
closed in k  by Theorem 4.6, this implies the existence of a proper regular solution
of E .'; / by Proposition 4.2. t
u

If moreover k is Hilbertian we obtain from Theorem IV.1.5(a):

Corollary 4.9. Every finite split embedding problem over a Hilbertian large field
has a proper solution.
4 Large Fields 479

4.4 Application to Hilbertian PAC-Fields

As an easy consequence we obtain the following generalization of Theorem IV.3.10

to arbitrary characteristic:

Theorem 4.10 (Pop (1996)). The absolute Galois group of a countable Hilbertian
PAC-field is free profinite of countable rank.

Proof. This follows as in the proof of Theorem IV.3.10 since by Proposition 4.4 a
PAC-field is large, and by Corollary 4.9 every finite split embedding problem over
a Hilbertian large field has a proper solution. t
u

Remark. Together with the result of Roquette (see Fried and Jarden (1986),
Prop. 24.38) this allows the following characterization of Hilbertian PAC-fields:
A countable PAC-field is Hilbertian if and only if its absolute Galois group is free
profinite of countable rank.
480 VI Rigid Analytic Methods

5 On the Fundamental Group with Restricted Ramification

In contrast to the situation in characteristic zero described in Chapter I.1, the funda-
alg
mental group 1 .X n S / of an algebraic curve X with finite ; ¤ S  X over
an algebraically closed field k of positive characteristic is not free. Here we first
show that at least the projectivity of the fundamental group in the case of restricted
ramification remains true, then we discuss the recent solution of Abhyankar’s con-
jecture.

5.1 Projectivity

.p/
In this section we show that at least the pro-p-factor groups 1 .X n S / of
alg
1 .X n S / for an algebraic curve X with finite ; ¤ S  X are free. As a
alg
consequence we obtain the projectivity of 1 .X n S /. The first partial result is an
immediate consequence of the Grothendieck Specialization Theorem:

Proposition 5.1. Let K=k be a rational function field in one variable over an alge-
braically closed field k, S a finite subset of IP.K=k/ with jSj > 1, and p 2 IP different
.p/
from the characteristic of k. Then the Galois group Gal.MS =K/ of the maximal
.p/
p-extension MS =K unramified outside S is a free pro-p-group of rank jSj  1.

Proof. This follows by an application of the Grothendieck Specialization Theo-

alg
rem I.10.6 to the free fundamental groups 1 .X n S / D Gal.MS =K/ of rank
jSj  1 D jS j  1 in Theorem I.1.3. t
u

Remark. According to Section I.1.4 the corresponding result remains true for func-
tion fields K=k of genus g > 0 and with jSj  1. In this case the rank equals
jSj C 2g  1.

For convenience we introduce the following notation. Let K=k be an algebraic

function field of one variable. Let M be a Galois extension of K with group . A
homomorphism ' from to a finite group G is called ramified in P 2 IP.K=k/ if P
is ramified in the Galois extension M ker.'/ =K, and is called unramified otherwise.
The proof of the following theorem was communicated to us by J.-P. Serre.

Theorem 5.2. Let K=k be an algebraic function field in one variable over an alge-
braically closed field k of characteristic p > 0 and S a nonempty finite subset of
IP.K=k/. Then the Galois group Gal.MS.p/ =K/ of the maximal p-extension unram-
ified outside S is a free pro-p-group of infinite rank (equal to the cardinality of
field k).
.p/ .p/
Proof. We first show the freeness of the pro-p-group K .S/ WD Gal.MS =K/.
.p/ .p/
For this it suffices to verify that H 2 . K .S/; Zp / D 0 with trivial action of K .S/
5 On the Fundamental Group with Restricted Ramification 481

on Zp (see for example Serre (1964), Ch. I, 4, Prop. 21 and Cor. 2 to Prop. 24). Let
O be the set of open normal subgroup in K.p/ .S/ and 'O for O 2 O the restriction
map from K.p/ .S/ to GO WD Gal..MS.p/ /O =K/. Then
.p/ .p/
K .S/ D lim .'O . K .S/=O//O2O D lim .GO /O2O
 

and hence also (compare Serre (1964), Ch. I, 2, Prop. 8)

.p/ 
H 2. K .S/; Zp / D lim.'O .H 2 .GO ; Zp ///O2O : (5.1)
!

Thus from Theorem IV.6.1 with the subsequent Remark the vanishing of (5.1) fol-
lows from the solvability of all finite central embedding problems with kernel Zp
.p/
inside MS =K.
So let ' W K.p/ .S/ ! G D Gal.N=K/ be an epimorphism onto a finite group G
and
 
1 ! Zp ! GQ ! G ! 1

be a central group extension with kernel Zp ,  W K.p/ ! K.p/ .S/ the restriction map
from the Galois group of the maximal p-extension M .p/ =K to the Galois group of
the maximal p-extension MS.p/ =K unramified outside S and WD ' ı . By the The-
.p/
orem of Tsen (Theorem IV.1.11(a)) K is a projective (and hence free) profinite
group. Thus the embedding problem E . ; / possesses a solution Q W K ! G. Q
.p/

In what follows we will modify this solution so that it is unramified outside S and
hence factors through K.p/ .S/. Therefor we denote by T the finite set of ramified
places of Q not belonging to S, and by NQ the solution field of Q . Then Q induces
for every P 2 T an epimorphism of local Galois groups

Q
Q P W I.P=P/ Š
.p/
! Gal.NQ PQ =KP / Š Zp :
KP

(This is uniquely determined by P up to inner automorphisms of K.p/ .) The local

Galois extensions NQ PQ =KP are generated by Artin-Schreier-equations fP .X / D
X p  X  xP with xP 2 KP . By the Strong Approximation Theorem (see for
example Hasse (1980), Ch. 24) there exists x 2 K such that ordP .x  xP / is
large for all P 2 T and ordP .x/  0 for P 2 IP.K=k/ n .S [ T/. The zeroes of
f .X / D X p  X  x 2 KŒX then generate a Galois extension L=K with group
Zp whose localizations LP for P 2 T coincide with NQ P by Krasner’s Lemma
(see Artin (1967), Ch. 2, Thm. 9), and which by Theorem III.4.7 are trivial for
P 2 IP.K=k/ n .S [ T/. Thus there exists an epimorphism
.p/
W K ! Gal.L=K/ Š Zp
482 VI Rigid Analytic Methods

whose localizations P for P 2 IP.K=k/ n S agree with Q P . Since Zp Q the

Z .G/
map
.p/
Q W K ! G; Q  7! . /1 Q . /;
is a homomorphism which is unramified outside S by construction. Hence Q factors
through K.p/ .S/ and defines a homomorphism

'Q W .p/
K ! GQ with  ı 'Q D '

since  ı  D 0. This shows the solvability of E .'; / and hence the freeness of
.p/
K .S/.
Since the Artin-Schreier-equations f .X / D X p  X  x 2 KŒX generate dif-
ferent fields for elements x in different classes of .K; C/ modulo the subgroup
fx p  x j x 2 Kg, the rank of K.p/ .S/ is at least the cardinality of K, and hence
equal to it. t
u

Remark. In the case S D ; the Galois group Gal.MS.p/ =K/ is again a free pro-p-
group. But its rank is now finite and equal to the rank of the Hasse-Witt matrix of
K=k (see Šafarevič (1947), Thm. 2).

As a consequence of the preceding theorem we obtain the result announced at

the beginning of this section:

Theorem 5.3. Let K=k be an algebraic function field in one variable over an alge-
braically closed field k and S a nonempty subset of IP.K=k/. Then the Galois group
Gal.MS =K/ of the maximal Galois extension unramified outside S is a projective
profinite group.

Proof. By definition we have to verify that every finite embedding problem E .'; /
for K .S/ WD Gal.MS =K/ has a solution 'Q with  ı 'Q D ' (see Section IV.1.5). By
the reduction theorem of Nobusawa (Theorem IV.5.1) it suffices to solve Frattini
embedding problems, and by induction over the chief factors of the nilpotent Frat-
tini group, even embedding problems with p-elementary abelian kernel H (see also
Fried and Jarden (1986), Lemma 20.9). First let p be equal to the characteristic of
K. Then by the Theorem IV.8.2 of Kochendörffer we may restrict ourselves to the
solvability of embedding problems E .'p ; p / belonging to a Sylow p-subgroup Gp
of G D '. K .S//, where 'p W L .S/ ! Gp denotes the epimorphism belonging to
the fixed field L of Gp D .GQ p /. Observe that by the proof of Theorem IV.8.2 a
solution field of E .'; / can be found inside the Galois closure over K of a solution
field of E .'p ; p / such that with E .'p ; p / also E .'; / possesses a solution unram-
ified outside S. But the embedding problems E .'p ; p / even have proper solutions
.p/
since L .S/ is free profinite of infinite rank by the above theorem. With a similar
argument we obtain solutions of the corresponding embedding problems in char-
acteristic different from p because here L.p/ .S/ is free of finite rank by Proposi-
tion 5.1 and the subsequent Remark. t
u
5 On the Fundamental Group with Restricted Ramification 483

By Proposition 5.1 and Theorem 5.2 the group Gal.MS =K/ is not a free profinite
group if K has positive characteristic, in spite of the fact that its cohomological
dimension equals 1.

5.2 Embedding Problems with p-Kernel

A further very useful result for the study of the fundamental group with restricted
ramification in positive characteristic is the following strengthening of the embed-
ding Theorem IV.8.3 to embedding problems with restricted ramification going back
to Serre.

Theorem 5.4 (Serre (1990)). Let k be an algebraically closed field of characteristic

p and E .'; / a finite geometric embedding problem over K D k.t/ whose kernel
H is a p-group. Assume that the ramification points of the Galois extension N=k.t/
for the quotient G are contained in the nonempty set S  IP.k.t/=k/. Then E .'; /
has a proper geometric solution whose solution field NQ =k.t/ is only ramified in S.

Sketch of proof. By assumption the epimorphism ' W K ! G factors through the

Galois group K .S/ WD Gal.MS =K/ of the maximal Galois extension of K unram-
ified outside S. Therefore without loss of generality we can replace K by K .S/.
By induction along a chief series we may assume that the kernel H is an elementary
abelian p-group. Since K .S/ by Theorem 5.3 is a projective profinite group the
embedding problem E .'; / is solvable. To obtain a proper solution by the Theo-
rem IV.1.9 of Ikeda it is enough to solve properly split embedding problems with the
same kernel H . The problem therefore is reduced to properly solving split embed-
ding problems with elementary abelian p-kernel H inside K .S/.
Via the semidirect product H G and ' the group H becomes a K .S/-module.
It is easy to verify that E .'; / has a proper solution if and only if the infla-
tion '  .H 1 .G; H // of H 1 .G; H / is properly contained in H 1 . K .S/; H /. There-
fore the assertion follows from the fact proved by Serre (1990), Prop. 4, that
H 1 . K .S/; H / is an infinite dimensional IFp -vector space. t
u

5.3 The Conjecture of Abhyankar for the Affine Line

Apart from the result in the previous sections, almost nothing is known about the
fundamental group in the case of restricted ramification. Its structure even depends
on the choice of the set S of ramification points, as the following example shows:
Let p > 2 and Pi 2 IP1 .k/ such that Si WD f0; 1; 1; Pi g has ordinary respectively
supersingular j -invariant for i D 1; 2 (see for example Silverman (1986), Ch. V).
alg
It is shown in Harbater (1994b), Thm. 1.8, that in fact 1 .IP1 .k/ n Si / are non-
isomorphic for i D 1; 2.
484 VI Rigid Analytic Methods

At present the only information in this situation comes from the conjecture of
Abhyankar (1957), which describes the set of finite factor groups of the funda-
mental group. This was proved by Raynaud (1994) in the case of one ramification
point and genus zero, and then by Harbater (1994a) in general. To formulate the
conjecture, we introduce the following notations. Let G be a finite group and p a
0
prime number. Then O p .G/ is the normal subgroup of G generated by the Sylow
0
p-subgroups of G, hence G=O p .G/ is the largest factor group of G of order prime
0
to p. A group G with G D O p .G/ is called a quasi-p-group. Thus quasi-p-groups
are generated by their elements of p-power order. For Gp G a Sylow p-subgroup
of a finite group G define

O.G; Gp / WD hH j H < G is quasi-p; H \ Gp is a Sylow p-subgroup of H i:

The Abhyankar conjecture for the affine line, which was proved by Raynaud, can
be stated as follows:

Theorem 5.5 (Raynaud (1994)). Let k be an algebraically closed field of character-

istic p ¤ 0 and G a finite group. Then G occurs as the Galois group of an unramified
connected cover of the affine line over k if and only if G is a quasi-p-group.

The proof of this result goes beyond the scope of the present book. Let us just
give an informal description of the steps Raynaud uses in his proof (see also Har-
bater (1995b) for an introduction). Let k be an algebraically closed field of charac-
teristic p and G a finite quasi-p-group with Sylow p-subgroup Gp . Then an induc-
tion argument together with the Theorem I.10.6 of Grothendieck easily shows that
Theorem 5.5 is implied by the following three assertions:
(a) Let H G G be a p-group. If G=H occurs as the Galois group of an unramified
cover of the affine line over k then so does G.
(b) If G D O.G; Gp / and all proper quasi-p-subgroups of G occur as the Galois
group of an unramified cover of the affine line over k, then so does G.
(c) If G ¤ O.G; Gp / and G has no non-trivial normal p-subgroup, then G occurs
as the Galois group of an unramified cover of the affine line over k.
The three alternatives enumerated above require three quite different approaches.
Part (a) is solved by Theorem 5.4 or Theorem 5.3 respectively.
The proof of case (b) is achieved by gluing suitable rigid analytic spaces. Let
H1 ; : : : ; Hr be the proper quasi-p subgroups of G such that Hi \ Gp is a Sylow p-
subgroup of Hi . By assumption, for each of these there exists an unramified Galois
cover Xi ! A1 .K/ with group Hi , where K D k..u//. Moreover it may be reached
that the inertia groups above infinity are conjugate to subgroups Qi of Gp . Thus the
restriction of Xi to an annulus Ai centered around 1 splits into a disjoint union
of copies of Qi -covers Bi ! Ai . Now choose r points P1 ; : : : ; Pr 2 A1 .K/ and
copies of the annuli Ai centered around the Pi such that the union of the discs Di
corresponding to the annuli Ai is disjoint and A1 .K/ contains a point in each com-
ponent of A1 .K/ n .[riD1Ai [ fPi g/. Then possibly after replacing K by a finite
separable extension there exists a Galois cover Y ! A1 .K/ n fP1 ; : : : ; Pr g with
group Gp whose restriction to each Ai is a disjoint union of copies of Bi ! Ai .
5 On the Fundamental Group with Restricted Ramification 485

The proof of this fact proceeds by induction on the order of Gp and uses étale coho-
mology. Induction to G and pasting each Xi to Y over Ai yields a cover with group
G D O.G; Gp /, which is connected by the assumption in case (b).
In case (c) let R be a complete discrete valuation ring with residue field k and
with field of fractions K of characteristic zero. Such a ring can be constructed using
Witt vectors (see for example Serre (1979), 5, Thm. 3). By Theorem 2.5 there
exists a geometric Galois extension of K.t/ with group G. Since G is a quasi-p-
group, so generated by elements of p-power order, the proof of Proposition 2.4
shows that we may moreover assume that the inertia groups are (cyclic) p-groups.
Using the theory of semistable reduction it is possible to choose K and R such
that the corresponding covering Y ! X D IP1 .K/ has an R-model YR ! XR
with semistable reduction and such that XR is a blow-up of IP1 .R/. Since XR is
a model of the projective line the special fiber Xk has to be a tree of projective
lines over k. The strict transform e 0 of X in IP1 .k/ will be taken as the origin of
the tree Xk . To each irreducible component c of Yk is associated its stabilizer Dc ,
the decomposition group with respect to the generic point of c, and the pointwise
stabilizer Ic , the inertia group. It is shown by Raynaud (1994), 6.3, that XR can
be chosen such that the inertia group Ic at each component c of Yk is a p-group
and Ic is non-trivial unless c lies above a terminal component of Xk . There exists
a partial ordering on the tree Xk with the terminal components maximal and the
base component e 0 minimal. This allows to construct an analogous partially ordered
subtree T of Yk , chosen such that Ge D G for the preimage e of e 0 . Here for c a
0
component of T we define Gc WD hO p .Dd / j d 2 T; d  ci.
If c 2 T is maximal with Gc D G then using that we are not in case (a) one can
show by group theoretic arguments that either Ic D 1 or G D Gc  O.G; Gp / for
some Sylow p-subgroup Gp of G. But the latter case cannot happen in (c) (it was
treated in (b)). Thus Ic D 1 and c must be a terminal component with Dc D G.
Since c is terminal, its image c 0 in Xk is a projective line which intersects the other
components in a single point. By deleting this point we obtain an unramified cover
of the affine line with group G.

5.4 The General Case of the Conjecture of Abhyankar

Abhyankar (1957) posed the following conjecture on finite quotients of the funda-
mental group of an affine smooth connected curve C over an algebraically closed
field of characteristic p: If C is obtained by removing s > 0 points from a projective
connected normal curve of genus g then G occurs as a quotient of the algebraic fun-
0
damental group of C if and only if G=O p .G/ is generated by 2g C s  1 elements.
This general form of the conjecture was proved by Harbater (1994a) with formal
geometric methods, building on Raynaud’s result for the affine line. We sketch a
proof of this given subsequently by Pop (1995), which essentially reduces it to the
case of the affine line. The following result is a special case of Cor. 4.2.6 in Raynaud
(1994); its proof will not be given here.
486 VI Rigid Analytic Methods

Theorem 5.6. Let

1 ! H ! GQ ! G ! 1
be an exact sequence of finite groups where H is a p-group. Let k be a complete
ultrametric field, ; ¤ S  IP1 .k/an a finite set, D  IP1 .k/an affinoid such that
each connected component of IP1 .k/an n D contains a point of S . Assume given a
Galois covering X ! IP1 .k/an with group G which is unramified outside S and
trivial over D, and a Galois covering Y ! D with group H . Then there exists
a connected Galois covering Z ! IP1 .k/an with group GQ unramified outside S
Q
isomorphic to IndG H .Y / over D such that the quotient by H is isomorphic to the
covering X ! IP1 .k/an .
This allows to prove a generalization of Theorem 5.4 from p-kernel to quasi-p-
kernel.
Theorem 5.7 (Pop (1995)). Let k be an algebraically closed field of characteristic
p and E .'; / a finite geometric embedding problem over k.t/ such that the rami-
fication points of the Galois extension N=k.t/ for the quotient G are contained in
the nonempty set S  IP.k.t/=k/. Assume that there exists an analytic Galois cover
W Y ! IP1 .k  /an with group H unramified outside 1 2 IP1 .k  /an such that the
decomposition groups of points above 1 are the Sylow p-subgroups of H . Then
E .'; / has a proper geometric solution whose solution field NQ =k.t/ is only rami-
fied in S.
Proof. We first reduce the assertion to embedding problems of a very special type.
By the Theorem of Ikeda (Theorem IV.1.9) since the Galois group of the maximal
extension of k.t/ unramified outside S is projective by Theorem 5.3 we may assume
that E .'; / is split. Let Hp be a Sylow p-subgroup of H . Since all Sylow p-
subgroups of H are H -conjugate, we have .NGQ .Hp / W NH .Hp // D .GQ W H / D jGj,
so we obtain an induced embedding problem E .'; 1 /

1 ! NH .Hp / ! NGQ .Hp / ! G ! 1

for the normalizers of Hp , where 1 D jNGQ .Hp / . By Theorem 5.3 the Galois group
of the maximal extension field of k.t/ unramified outside S is projective. Thus
E .'; 1 / has a solution '1 , with solution field N1 , say, such that N1 =k.t/ is unram-
ified outside S. With G1 WD Gal.N1 =k.t// we obtain a further induced embedding
problem E .'1 ;  0 /

1 ! H ! GQ G G1 ! G1 ! 1

again with kernel H by construction. Thus any (proper, geometric) solution of

E .' 0 ;  0 / ramified only in S gives rise to a (proper, geometric) solution of E .'; /
only ramified in S. Moreover E .' 0 ;  0 / is still split, and by construction G1 nor-
malizes a Sylow p-subgroup of H . We are thus reduced to solving split embedding
problems E .'; / such that G normalizes a Sylow p-subgroup Hp of H .
As in the proof of Theorem 2.10 it suffices to solve the induced embedding
problem E .'  ;   / obtained from E .'; / by extension of constants with the com-
5 On the Fundamental Group with Restricted Ramification 487

plete ultrametric field k  WD k..u//. Note that if we have found a solution field
NQ  =k..u//.t/ ramified only in S, then the same is true for the specialized solution
field NQ =k.t/.
So let now E .'; / be a split embedding problem over k  .t/ such that G normal-
izes a Sylow p-subgroup Hp of the kernel H and N  the fixed field of ker.'/. Via
analytification of suitable curves N  =k  .t/ corresponds to an analytic Galois cover
' W X ! IP1 .k  /an with group G unramified outside a set S  IP1 .k  /an .
We first construct an analytic Galois cover Z ! IP1 .k  /an with group the
semidirect product Hp  G ramified in at most jS j C 1 points. For this, the cover
W Y ! IP1 .k  /an is subdivided into several parts. Let D0 be a disc in IP1 .k  /an
around 0 and Y1 D 1 .D0 /, Y2 D 1 .IP1 .k  /an n D0ı / and
1
@Y1 WD @Y2 D Y1 \ Y2 D .@D0 / ;

where from now on for discs D.a; b/ we will write @D.a; b/ for D.a; b/ n D.a; b/ı .
Then by a suitable choice of the uniformizing parameter for IP1 .k  /an we may
assume that the following hold. Let y 2 Y2 be a preimage of 1 with decomposition
group equal to Hp and Y0 the connected component of Y2 containing y. Then Y2
is the disjoint union of .H W Hp / copies of Y0 ,
[
:
Y2 D Y0 ;
2H=Hp

Y0 ! IP1 .k  /an n D0ı is connected Galois with group Hp and @Y0 WD Y0 \ @Y1 !
@D0 is still a connected Galois cover with group Hp which moreover is unramified.
Furthermore, by Lemma 1.8 the covering Y1 ! D0 is connected and Galois with
group H .
Next for all P 2 S we choose disjoint small discs DP around P in IP1 .k  /an
such that the preimage U 0 D ' 1 .D 0 / in X of the complement
[
:
D 0 WD IP1 .k  /an n ı
DP
P2S

0 0
is the trivial covering IndG
1 .D / of D with Galois group G (which is possible after a
suitable finite separable extension of k  by Proposition 2.2). We fix a point P 2 S
and choose a point PQ ¤ P in DP ı
having only k  -rational, distinct preimages in
X (after a suitable finite separable extension of k  ). Let DQ  DP ı
be a small disc
containing PQ and not containing P. Since DQ is analytically isomorphic to D0 , from

one piece of Y we obtain a connected Galois cover Y1 ! D0 ! DQ with group H .
Its restriction to @DQ decomposes into connected components isomorphic to @Y0 ,
which gives a connected unramified Galois cover @Y0 ! @DQ with group Hp .
We have thus obtained the Galois cover X ! IP1 .k  /an with group G, trivial
over D 0 [ @D, Q and a Galois cover of D 0 [ @DQ with group Hp given by IndHp .U 0 /
1
above D 0 and by @Y0 above @D. Q By Raynaud’s Theorem 5.6 there exists a connected
488 VI Rigid Analytic Methods

Galois cover Z ! IP1 .k  /an with group GQ p ramified only in S and some further
Q Q
point in DQ ı , isomorphic to Ind p .U 0 / above D 0 and to Ind p .@Y0 / above @DQ . Let
G G
G Hp
DQ 0 D IP1 .k  /an n DQ ı and V 0 its preimage in Z . Then V 0 is a connected Galois cover
of DQ 0 with group GQ p unramified outside S  DQ 0 .
We are now ready to construct the Galois cover of IP1 .k  /an with group GQ solving
Q
E .'; /. Let XQ1 D IndG 0 Q0 Q0
Q .V / above D . Then restriction to @D yields
G
p

Q 0 Q Q
G Q
Q .@V / Š IndG
IndG Q .IndHp .@Y0 // Š IndHp .@Y0 /
G p G
G p p

with the preimage @V 0 of @DQ 0 , by the construction of Z . Further, let XQ2 D

Q Q Q Q0 Q
H .Y1 / above D Š D0 . Its restriction to @D D @D is G-equivariantly isomorphic
IndG
to
Q Q Q
IndG G H G
H .@Y1 / Š IndH .IndHp .@Y0 // Š IndHp .@Y0 /:

Hence by the Gluing Proposition 1.4 the two covers XQ1 and XQ2 may be glued
above @DQ 0 to give a Galois cover XQ ! IP1 .k  /an with group G.Q This cover is
connected since GQ is generated by GQ p and H , and it is only ramified in S since
V 0 ! DQ 0 is only ramified in S , while Y1 ! DQ is unramified by construction.
Further, the quotient of XQ by H is isomorphic to X ! IP1 .k  /an , thus XQ solves
the embedding problem E .'; /. t
u

Remark. If H is a quasi-p-group then by the Theorem 5.5 of Raynaud there exists

an analytic Galois cover W Y ! IP1 .k  /an with group H unramified outside 1 2
IP1 .k  /an . Further by Raynaud (1994), Cor. 2.2.6, this may be chosen such that
the Sylow p-subgroups of H are the decomposition groups of points above 1.
Thus, the preceding Theorem 5.7 applies if the kernel H is a quasi-p-group. On the
other hand, starting with abelian p-groups, Theorem 5.7 allows to give an alternative
proof of the Theorem 5.4 of Serre.

We can now state the version of Abhyankar’s conjecture for ramified extensions
of the projective line; the general result for function fields in one variable is given
in Harbater (1994a) and Pop (1995).

Theorem 5.8 (Raynaud (1994), Harbater (1994a)). Let k be an algebraically closed

field of characteristic p, and S  IP.k.t/=k/ a finite set of cardinality s. Then G
0
occurs as a Galois group over k.t/ unramified outside S if and only if G=O p .G/
occurs as the Galois group of an extension of C.t/ unramified outside a set of car-
0
dinality s, i.e., G=O p .G/ is generated by s  1 elements.

Proof. With the above preparations, this result is easy to obtain. In fact, first assume
that G occurs as a Galois group over k.t/ unramified outside of S. Then the p-
0 alg
prime group G=O p .G/ also is a factor group of 1 .IP.k.t/=k/ n S/, hence by the
theorem of Grothendieck (Theorem I.10.6) already occurs over C.t/ with the same
number of ramification points.
5 On the Fundamental Group with Restricted Ramification 489

Conversely, let GQ be a finite group such that G WD G=O

Q p0 Q
.G/ occurs as a Galois
group of an extension of C.t/ unramified outside a set of cardinality s. Then again
by the theorem of Grothendieck the p-prime group G also occurs as a Galois group
of a field extension N=k.t/ unramified outside a set S of cardinality s. Let E .'; /
be the embedding problem associated to the group extension GQ D O p .G/ Q  G and
0

the Galois extension N=k.t/ with Gal.N=k.t// Š G. Then by Theorem 5.7 and
the subsequent Remark E .'; / has a proper solution whose solution field is again
ramified at most in S. t
u
Appendix: Example Polynomials

In this appendix we list example polynomials whose roots generate regular exten-
sion fields of Q.t/, respectively number fields over Q with given Galois group of
small permutation degree. The first set of examples realizes most of the equiva-
lence types of transitive permutation groups of degree less than 12 as regular Galois
groups over Q.t/. (There are 301 inequivalent transitive permutation groups of
degree 12.) Most of these results are new. In the second table, we collect the known
explicit regular Galois realizations of primitive non-solvable permutation groups of
degree at most 31 over Q.t/ from the literature. For both sets of tables the results
were mainly obtained by the rigidity method described in Chapter I and descent
arguments.
Finally, we give example polynomials generating number fields over Q with
given Galois group of permutation degree at most 14. For degree less than 12, these
were either found by a random search, and then the Galois group was verified by
the Galois group recognition programs in several computer algebra systems, or they
were obtained by specializing the parametric realizations from the first set of tables.
(Such specializations tend to have larger field discriminant.) The polynomials of
degree 12 to 14 are taken from Klüners and Malle (2000, 2002). The polynomials
listed in this table were chosen so that their coefficient sum is small.

1 Regular Realizations for Transitive Groups of Degree Less

than 12

Here we give polynomials generating regular field extensions of Q.t/ with Galois
groups most of the transitive permutation groups of degree less than 12. The generic
formulas for polynomials with symmetric or alternating group of arbitrary degree
are given separately. In all other cases the groups are numbered according to the
list in Butler and McKay (1983), so that a polynomial fn;i has Galois group the
transitive permutation group of degree n denoted by Ti in loc. cit.

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 491

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3
492 Appendix: Example Polynomials

Table 1.1 Symmetric and alternating groups

(  t .nx  n C 1/
n
Sn x
fSn .x; 1  .1/n.n1/=2 nt 2 / for n 1 .mod 2/
An
fSn .x; 1=.1 C .1/n.n1/=2 .n  1/t 2 / / for n 0 .mod 2/

Table 1.2 Degree 4

f4;3 x 4  2x 2 C t
f4;2 x 4 C tx 2 C 1
f4;1 x 4 C tx 3  6x 2  tx C 1

Table 1.3 Degree 5

f5;3 x 5 C 10x 3 C 5tx 2  15x C t 2  t C 16
f5;2 x 2 .x C 1/2 .x C 2/  .x  2/2 .x  1/t
f5;1 x.x 2  25/2 C .x 4  20x 3  10x 2 C 300x  95/t 2  4.x  3/2 t 4

Table 1.4 Degree 6

f6;14 x 6  2x 5 C .5x 2  6x C 2/ t
f6;13 x 6  .3x  2/2 t
f6;12 f6;14 .x; 1  5t 2 /
f6;11 x 6  .3x 2 C 1/t =4
f6;10 f6;13 .x; 1=.t 2 C 1//
f6;9 f6;13 .x; 1  t 2 /
f6;8 f6;11 .x; 1=.1  3t 2 //
f6;7 f6;11 .x; t 2 /
f6;6 f6;11 .x; 3t 2 C 1/
f6;5 f6;13 .x; 12t 2 .3t 2 C 1//
f6;4 f6;11 .x; .t 2 C 3/2 =.t 2  3/2 /
f6;3 x 2 .x 2 C 3/2 C 4 t
f6;2 f6;3 .x; 3t 2 C 1/
f6;1 x 6 C .3x 2 C 4/2 .3t 2 C 1/

Table 1.5 Degree 7

f7;5 .x 4  3x 3  x C 4/.x 3  x C 1/  x 2 .x  1/ t
f7;4 x 7 C 28x 6 C 63x 5 C 1890x 3 C 3402x 2  5103x C 33534 C x.x 6  63x 4
3402x  5103/ t C 13122 t 2
f7;3 see Smith (1993)
f7;2 f7;4 .x; .t 3  27t 2  9t C 27/=.3.t 3 C t 2  9t  1///
1 Regular Realizations for Degree Less than 12 493

f7;1 x 7  217 .t /x 5  77 .t /.10t 3 C 5t 2  5t  3/x 4  7.15t 6 C 15t 5  20t 4  27t 3

 13t 2  6t  13/7 .t /x 3  7.12t 9 C 18t 8  30t 7  63t 6  35t 5  14t 4  35t 3
C2t 2 C 31t C 16/7 .t /x 2  7.t  1/7 .t /.5t 11 C 15t 10  5t 9  62t 8  93t 7
 91t 6  126t 5  166t 4  113t 3  30t 2  8t  12/x  .6t 15 C 15t 14  35t 13
 126t 12  63t 11 C 70t 10  91t 9  271t 8 C 131t 7 C 427t 6 C 126t 5  84t 4
C175t 3 C 189t 2  29t  97/7 .t /

Table 1.6 Degree 8

f8;48 x 4 .x  2/2 .x 2 C x C 2/  .x  1/2 .x 2 C x C 1/ t
f8;47 x 8  .4x  3/2 t
f8;46 f8;47 .x; t 2 C 1/
f8;45 f8;47 .x; 1=.1  t 2 //
f8;44 x 8 C .4x 2 C 3/ t
f8;43 x 6 .x 2  x C 7/  108.x  1/ t
f8;42 f8;47 .x; 1=.12t 2 .3t 2  1///
f8;41 .x 2  2/4  26 .2x  3/2 t =33
f8;40 x 4 .x 4  8x 2 C 18/  27 t
f8;39 f8;44 .x; 3t 2 /
f8;38 f8;44 .x; 1=.3t 2 C 1//
f8;37 x 8 C 6x 7 C 3.7x 2 C 6x C 36/.7t 2 C 144/
f8;36 .f9;32 .x; 0/  f9;32 .t; 0//=.x  t /
f8;35 f8;44 .x; 27t 2 .t  1/=4/
f8;34 f8;41 .x; 1  t 2 /
f8;33 f8;41 .x; 1=.3t 2 C 1//
f8;32 f8;40 .x; 3t 2 /
f8;31 f8;44 .x; 27.t 2  1/2 =.t 2 C 3/3 /
f8;30 x 4 .x 4 C 4x 2 C 6/  .4x 2 C 1/.3t 2 C 2/2 .3t 2  1/=4
f8;29 f8;44 .x; 27t 2 .t 2  1/2 =4/
f8;28 f8;44 .x; 27t 4 .t 2 C 1/=4/
f8;27 f8;44 .x; 27.t 2 C 4/=.4.t 2 C 3/3 //
f8;26 f8;40 .x; 27t 2 .t  1/=..3t C 1/.3t  2/2 //
f8;25 see Smith (1993)
f8;24 .x 2 C x C 1/4  .2x C 1/2 t
f8;23 x 2 .x 2 C 396/2 .x 2 C 11/  .x 2 C 4/2 .x 2 C 256/ t
f8;22 f8;41 .x; 27t 2 =.4.t 2  1/3 //
f8;21 x 8 C 2.t 2  1/x 6 C .3t 4  t 2 /x 4 C 2.t 6 C t 4 /x 2 C t 8 C t 6
f8;20 f8;41 .x; 27t 4 .t 2 C 1/=4/
494 Appendix: Example Polynomials

f8;19 f8;41 .x; 27.t 2 C 4/=.4.t 2 C 3/3 //

f8;18 f8;41 .x; 27.t 2  1/2 =.t 2 C 3/3 /
f8;17 .x 4 C 4x 3  6x 2  4x C 1/2  16x 2 .x 2  1/2 t
f8;16 f8;40 .x; 27.t 2 C 2/2 =.t 4 .4t 2 C 9///
f8;15 x 8 C 8x 6 C 4.4t  11/x 4 C 8.t  3/.t  2/x 2 C t .t  3/2
f8;14 f8;24 .x; 1  3t 2 /
f8;13 f8;24 .x; 1=.3t 2 C 1//
f8;12 f8;23 .x; t 2 /
f8;11 f8;15 .x; t 2 /
f8;10 f8;41 .x; 22 33 t 4 .t 2 C 9/.t 2 C 1/=.t 2 C 3/6 /
f8;9 f8;24 .x; 27t 2 .t  1/=4/
f8;8 f8;15 .x; .8t 2 C 3/=.2t 2 C 1//
f8;7 f8;15 .x; 4=.t 2 C 1//
f8;6 f8;15 .x; 2t 2 C 3/
f8;5 x 8  4.t 2 C 2/.t 2 C 1/x 6 C 2.3t 2 C 1/.t 2 C 1/.t 2 C 2/2 x 4
4.t 2 C 2/2 .t 2 C 1/3 x 2 t 2 C .t 2 C 2/2 .t 2 C 1/4 t 4
f8;4 .x 4  6x 2 C 1/2 C 16x 2 .x 2  1/2 t
f8;3 f8;24 .x; 27.t 2  1/2 =.t 2 C 3/3 /
f8;2 f8;24 .x; 27.t 2 C 4/=.4.t 2 C 3/3 //
f8;1 x 8  4.t 4 C 1/x 6 C 2.4t 2 C 1/.t 4 C 1/x 4  4.t 2 C 1/.t 4 C 1/t 2 x 2 C .t 4 C 1/t 4

Table 1.7 Degree 9

f9;32 x 9  3x 8 C 4x 7  28x 6 C 126x 5  266x 4 C 308x 3  t x 2 C .3t  539/x
4 t C 805
f9;31 x 4 .x C 1/3 .x C 3/2  4=27.3x C 1/3 t
f9;30 x 4 .x  3/2 .x 3  3x 2  12/ C 28 t
f9;29 f9;31 .x; t 2 /
f9;28 f9;31 .x; 1=.3t 2 C 1//
f9;27 f9;32 .x; .t 3  6t 2 C 3t C 1/=.t 3  3t C 1//
f9;26 .x 3  19x 2 C 97x  27/.x 2  4x  7/3 C 16=27x 2 .x  7/t
f9;25 f9;30 .x; 1=.3t 2 C 1//
f9;24 x 6 .x 3 C 9x C 6/  4.3x C 2/3 t
f9;23 f9;26 .x; .43923t 2 C 18225/=.3t 2 C 1//
f9;22 f9;24 .x; 1=.3t 2 C 1//
f9;21 x 4 .x C 1/2 .x C 2/2 .x C 3/  1=35 x 2 .9x 2 C 20x C 12/ t C 1=39 t 2
f9;20 f9;24 .x; 3t 2 C 1/
f9;19 .f10;35 .x; 0/.t  1/2  f10;35 .t; 0/.x  1/2 /=.x  t /
f9;18 x 6 .x C 1/2 .x  2/ C 4=27.3x C 2/3 t
f9;17 f9;21 .x; 25 =.3t 2 C 1//
1 Regular Realizations for Degree Less than 12 495

f9;16 .x 2 C x  2/4 .x  4/ C 24 33 x 3 t
f9;15
f9;14
f9;13 f9;18 .x; 1=.3t 2 C 1//
f9;12 f9;18 .x; t 2 /
f9;11 f9;18 .x; 3t 2 C 1/
f9;10 .x 6 C 3x 4 C 10x 3 C 6x C 25/.x 3 C 3x 2 C 2/ C .x  1/.x C 2/
.x 3  3x 2  6x  1/  .x 4  7x 3 C 6x 2  13x  14/t =9 C 18t 2
f9;9 f9;16 .x; t 2 C 1/
f9;8 f9;16 .x; t 2 /
f9;7 .x 3 C 27x 2  9x  27/.x 2 C 3/3  27=4.x 2  1/2 x.x 2  9/.3t 2 C 49/
f9;6 f9;21 .x; 96.t 2  9/2 t 2 =..t 4  2t 2 C 49/.3t 2 C 1///
f9;5 f9;16 .x; .t 2 C 1/2 =.t 2  1/2 /
f9;4 f9;16 .x; 1=.3t 2 C 1/2 /
f9;3 f9;10 .x; .t 3 C 6t 2 C 3t  1/=.t 3  3t  1//
f9;2 x 9  6.t 2 C 3/x 7  6x 6 t C 9.t 4 C 9t 2 C 9/x 5 C 24t .t 2 C 3/x 4
.4t 6 C 69t 4 C 213t 2 C 81/x 3  216t 3 x 2 C 12t 2 .3t 4  11t 2 C 21/x  8t 3
f9;1 x 9  279 .t /x 7  54t .t 2  1/9 .t /x 6 C 2439 .t /.2t 4 C t 3  t 2 C 1/x 5
C243t .t 2  1/9 .t /  .4t 4 C2t 3  t 2 Ct C3/x 4  81.33t 8 C33t 7  26t 6  6t 5
C69t 4 C16t 3  36t 2  3t C10/  9 .t /x 3  2187t .t 2  1/.2t 8 C2t 7  t 6  t 5
C4t 4 C3t 3  t 2 C1/9 .t /x 2 C729.2t 3 C1/.3t 9 C9t 8 C2t 7  14t 6 C17t 4
Ct 3  9t 2  t C1/9 .t /xC2439 .t /.36t 13 C18t 12  60t 11 C30t 10 C64t 9
 81t 8  9t 7 C87t 6  36t 5  54t 4 C21t 3 C15t 2  3t  1/

Table 1.8 Degree 10

f10;43 x 10  .5x  4/2 t
f10;42 f10;43 .x; 1=.t 2 C 1//
f10;41 f10;43 .x; 1  t 2 /
f10;40 f10;43 .x; 20t 2 .5t 2  1//
f10;39 x 10  55 .x 2 C 4/ t
f10;38 f10;39 .x; 1=.t 2 C 1//
f10;37 .x 2  4/5  55 x 2 t
f10;36 f10;39 .x; 1  5t 2 /
f10;35 x 10  2x 9 C 9x 8  729.x  1/2 t
f10;34 f10;39 .x; 4.t 2 C t  1/2 =.5.t 2 C 1/2 //
f10;33 .x  2/2 .x 2 C x  1/4  .380x 6  784x 5 C 300x 4 C 360x 3  315x 2
C60x C 4/ t C 4.5x  4/2 t 2
f10;32 f10;35 .x; t 2 /
f10;31 f10;35 .x; 1=.2t 2 C 1//
496 Appendix: Example Polynomials

f10;30 f10;35 .x; 1  2t 2 /

f10;29 x 10 C 10x 6  5tx 4  15x 2  t 2 C t  16
f10;28 .x  2/.x  1/.x 4 C x 3 C 6x 2  4x C 1/.x 2 C x  1/2 C .4x 5  20x 2
C15x  2/  .10x 3  10x 2 C 1/ t C .5x  4/.8x 5  40x 2 C 35x  8/ t 2
f10;27 f10;33 .x; t =.4.t 2  1///
f10;26 f10;35 .x; .t 2  2/2 =.t 2 C 2/2 /
f10;25 f10;39 .x; 28 t 2 =..t 4 C 6t 2 C 25/.t 2 C 1/4 //
f10;24 f10;29 .x; t .t  8/=.t 2  1//
f10;23 x 2 .x 4  25/2 C .x 8  20x 6  10x 4 C 300x 2  95/t  4.x 2  3/2 t 2
f10;22 x 10 C 55 .x 2 C 256/4 t
f10;21 .x 2 C 9x C 24/2 .x 2 C 4x C 64/2 .x 2  6x C 144/  55 x 4 .x C 8/2 t =4
f10;20
f10;19 .x 2 C 1/4 .x 2 C 16/  5.x 7 C 11x 5  15x 4  5x 3 C 38x 2  15x  7/ t
C.x 5 C 10x 3  15x  15x 2 C 28/ t 2
f10;18 f10;28 .x; t 2 /
f10;17
f10;16 f10;23 .x; t 2  95=36/
f10;15 f10;23 .x; 95=.t 2  36//
f10;14 f10;23 .x; t 2 / D f5;1 .x 2 ; t /
f10;13 .x 2  5/5  55 .x 2 C 5x C 6/4 t =4
f10;12 f10;22 .x; 1=.1  t 2 //
f10;11 f10;22 .x; 1  5t 2 /
f10;10 f10;21 .x; 4.3t 2 C 32/=.4t 2 C 1//
f10;9 f10;21 .x; 4.5t 2  32//
f10;8 f10;23 .x; .7t 2  24t C 7/2 =.36.t 2  1/2 //
f10;7 f10;13 .x; 1  5t 2 /
f10;6 f10;19 .x; 4=.5t 4 C 5t 2 C 1//
f10;5 f10;22 .x; 4t 5 .t  10/=.55 .t 2 C 2t C 5///
f10;4 f10;13 .x; 4t 5 .t  10/=.55 .t 2 C 2t C 5///
f10;3 f10;22 .x; .11t 2 C 4t  11/.t 2 C 4t  1/5 =.55 .t 2 C 1/2 .t 2  1/4 //
f10;2 x 10  2.t 2  125/x 8 C .t 2  125/.t 2  4t  65/x 6  4.t 2  125/2 .t  10/x 4
C4.t 2  14t C 25/.t 2  125/2 x 2  64.2t  25/.t 2  125/2
f10;1 x 10  2010 .t /x 8 C 10.7t 4  7t 3 C 17t 2  17t C 12/10 .t /x 6
25.4t 8  8t 7 C12t 6  16t 5 C25t 4  46t 3 C67t 2  38t C9/10 .t /x 4
C510 .t /.13t 12 39t 11 C18t 10 C50t 9  125t 8 C376t 7
453t 6  214t 5 C1050t 4  1125t 3 C613t 2  164t C18/x 2
10 .t /. 1  3t C32t 2  36t 3  10t 4 C34t 5 13t 6  8t 7 C4t 8 /2
1 Regular Realizations for Degree Less than 12 497

Table 1.9 Degree 11

f11;6 .fM12 .x; 0/.2t  1/2  fM12 .t; 0/.2x  1/2 /=.x  t /
f11;5 x 11  3x 10 C 7x 9  25x 8 C 46x 7  36x 6 C 60x 4  121x 3 C 140x 2  95x C 27
Cx 2 .x  1/3 t
f11;4
f11;3 x 11  11.t 2 C 11/x 9 C 44.t 2 C 11/2 x 7  77.t 2 C 11/3 x 5 C 55.t 2 C 11/4 x 3
11.t 2 C 11/5 x  2t .t 2 C 11/5
f11;2
f11;1 x 11  5511 .t /x 9  11.30t 5 C 15t 4  30t 3  25t 2  4t C 3/11 .t /x 8
11.90t 10 C90t 9 240t 8 350t 7 229t 6 97t 5 C35t 4 C13t 3 42t 2
42t 75/11 .t /x 7 11.168t 15 C252t 14 840t 13 1750t 12 1218t 11
242t 10 C880t 9 C1265t 8 C880t 7 C836t 6 C572t 5 C437t 4 C430t 3
C224t 2 C78t 36/11 .t /x 6 11.210t 20 C420t 19 1680t 18 4550t 17
2723t 16 C2118t 15 C7971t 14 C11976t 13 C9282t 12 C6555t 11 C6523t 10
C5466t 9 C6103t 8 C4089t 7 422t 6 2128t 5 1887t 4 722t 3 C355t 2
C508t C452/11 .t /x 5 11.180t 25 C450t 24 2100t 23 7000t 22
3080t 21 C10615t 20 C27060t 19 C40865t 18 C32857t 17 C10109t 16
2398t 15 10128t 14 6994t 13 882t 12 14413t 11 33099t 10 42438t 9
36861t 8 18117t 7 550t 6 C6589t 5 C2640t 4 1063t 3 958t 2 648t
C117/11 .t /x 4 11.105t 30 C315t 29 1680t 28 6650t 27 1659t 26
C20003t 25 C44905t 24 C64445t 23 C44116t 22 34353t 21 102124t 20
135499t 19 138713t 18 92626t 17 79067t 16 119189t 15 147399t 14
166843t 13 136359t 12 38237t 11 C44396t 10 C74899t 9 C52267t 8
C2031t 7 22096t 6 12051t 5 C3115t 4 C7001t 3 C1543t 2 1896t
1160/11 .t /x 3 11.40t 35 C140t 34 840t 33 3850t 32 154t 31
C19008t 30 C39600t 29 C49203t 28 C14520t 27 120945t 26 280357t 25
348952t 24 314514t 23 145540t 22 C29359t 21 C33825t 20 27126t 19
75933t 18 85096t 17 C57717t 16 C276738t 15 C420112t 14 C438965t 13
C296100t 12 C50632t 11 97383t 10 69608t 9 C16104t 8 C68277t 7
C54527t 6 3025t 5 25355t 4 7986t 3 C3117t 2 C2302t 84/11 .t /x 2
11.9t 40 C36t 39 240t 38 1250t 37 C227t 36 C9128t 35 C17905t 34
C16150t 33 12716t 32 122980t 31 290048t 30 377822t 29
311551t 28 C1083t 27 C489620t 26 C744371t 25 C662921t 24 C433805t 23
C172463t 22 C209836t 21 C561407t 20 C810964t 19 C909892t 18
C777874t 17 C289801t 16 197823t 15 430310t 14 356065t 13 11405t 12
C252280t 11 C230131t 10 C47388t 9 93665t 8 90187t 7 24467t 6
C19479t 5 C19576t 4 4165t 3 5861t 2 C1587t C999/  11 .t /x
.10t 45 C45t 44 330t 43 1925t 42 C792t 41 C19448t 40 C36036t 39
C13761t 38 83787t 37 449020t 36 1138951t 35 1569333t 34
1270152t 33 C131912t 32 C3449677t 31 C7101292t 30 C8022157t 29
498 Appendix: Example Polynomials

C6359584t 28 C 2562879t 27  1238875t 26 C 266530t 25 C 4792381t 24

C7758954t 23 C 9292575t 22 C 6341588t 21 C 107481t 20  3610200t 19
4999456t 18  3552868t 17 C 1494614t 16 C 4899972t 15 C 3760834t 14
C620191t 13  2831935t 12  4464948t 11  2695792t 10  670956t 9
36608t 8 C 325281t 7 C 187935t 6 C 13585t 5 C 170786t 4 C 81906t 3
42372t 2  19548t C 243/11 .t /
2 Regular Realizations for Nonsolvable Primitive Groups 499

2 Regular Realizations for Nonsolvable Primitive Groups

Here we collect regular realizations for primitive non-solvable permutation groups

of degree d with 12 d 31. Simple groups in this range for which no polynomial
over Q.t/ is known to date are L2 .16/, M23 , L2 .25/ and L2 .27/. The polynomials
were taken from Häfner (1992), König (2015), Malle (1987, 1988a, 1993a), Malle
and Matzat (1985), Matzat (1987), Matzat and Zeh-Marschke (1986) and Müller
(2012). We also present surprisingly small polynomials of degree 100 with groups
Aut.HS/ and HS taken from Barth and Wenz (2016). A polynomial of degree 266
for the Janko group J2 has been obtained in Barth and Wenz (2017). In addition we
give the polynomial with Galois group Z16 from Dentzer (1995a).

Table 2.1 Primitive groups

M12 x 12 C 44x 11 C 754x 10 C 6060x 9 C 18870x 8  28356x 7  272184x 6
57864x 5 C 1574445x 4  92960x 3  1214416x 2 C 1216456x
304119  492075.2x  1/2 t
PGL2 .11/ .x 3  66x  308/4  9t .11x 5  44x 4  1573x 3 C 1892x 2 C 57358x
C103763/  3t 2 .x  11/
L2 .11/ fPGL2 .11/ .x; 28 35 =.11t 2 C 1//
L3 .3/ .x 6  6x 4 C 64x 3  36x 2 C 216/.x 4 C 8x 3  108x 2 C 432x  540/
.x 3  18x 2 C 54x  108/  .3x 4  28x 3 C 108x 2  216x C 108/2
.x 4 C 8x 3 C 108/ t
PGL2 .13/ .x 3  x 2 C 35x  27/4 .x 2 C 36/  4.x 2 C 39/6 .7x 2  2x C 247/ t =27
L2 .13/ fPGL2 .13/ .x; 1=.39t 2 C 1//
PGL2 .17/ .x 3  7x 2 C 5x  2/6  .x 17  17x 15 C 34x 14 C 85x 13  408x 12 C 289x 11
C1190x 10  2907x 9 C 1462x 8 C 3281x 7  5780x 6 C 3196x 5 C 238x 4
646x 3  68x 2 C 120x  16/t C t 2
L2 .17/ fPGL2 .17/ .x; 22 33 17=.t 2  17//
PGL2 .19/ .x 5 C 26x 4 C 69x 3 C 108x 2 C 68x C 16/4  .x 19  38x 17  38x 16
C513x 15 C 1064x 14  2299x 13  9538x 12  5358x 11 C 24358x 10
C55081x 9 C 35416x 8  40204x 7  105374x 6  98496x 5  41040x 4
C3648x 3 C 11552x 2 C 4352x C 512/t C t 2
L2 .19/ fPGL2 .19/ .x; 28 19=.t 2 C 19//
PL3 .4/ .x 3  9x 2  21x C 5/5 .x C 1/5 x  t .20x 5 C 89x 4 C 68x 3  50x 2
C16x C 1/3  .x 5 C 57x 4 C 330x 3 C 914x 2 C 1509x C 1125/
L3 .4/:3 see König (2015)
L3 .4/:22 .fAut.M22 / .x; 0/.t 2  t C 3/11  .fAut.M22 / .t; 0/.x 2  x C 3/11 /=.t  x/
L3 .4/ see König (2015)
500 Appendix: Example Polynomials

Aut.M22 / .5x 4 C 34x 3  119x 2 C 212x  164/4 .19x 3  12x 2 C 28x C 32/2
222 .x 2  x C 3/11 t
M22 fAut.M22 / .x; 1=.11t 2 C 1//
M24 4.48x 10  192x 9  256x 8 C 1104x 7 C 520x 6  1276x 5  64x 4  776x 3
1117x 2 C 391x C 52/2 .x 2 C 1/ C .16x 12  96x 11  144x 10 C 928x 9
C520x 8  1744x 7  1008x 6  1712x 5  791x 4 C 2154x 3 C 1121x 2
C1098x  t /2
PGL2 .23/ .x 8 C 3x 7 C 37x 6  24x 5 C 121x 4 C 333x 3 C 429x 2 C 216x C 36/3
.2x 24 Cx 23  322x 22 C1219x 21 C1863x 20 C4094x 19 C99084x 18
C197501x 17 C877910x 16 C1337726x 15 C3132117x 14 C8697795x 13
C15394935x 12 C16590866x 11 C4182642x 10 C6982731x 9
C36934642x 8 C43085601x 7 C13510591x 6  9423054x 5
10152936x 4  4024080x 3  824688x 2  85536x  3456/t
C.x 24  7x 23 C69x 22  460x 21  1564x 20  3289x 19 C11017x 18
C19159x 17  20792x 16  269307x 15  650440x 14  547124x 13
C609937x 12 C2106294x 11 C2682306x 10 C1410682x 9  856612x 8
1557215x 7  609132x 6 C135079x 5 C225814x 4 C113436x 3
C33764x 2 C5904xC496/t 2  .x 23 C23x 20 C23x 19 C23x 18
C161x 17 C368x 16 C529x 15 C575x 14 C1610x 13 C3036x 12
C2668x 11 C2300x 10 C3542x 9 C5428x 8 C2599x 7  1748x 6
1265x 5 C345x 4  598x 2  252x  16/t 3 Ct 4
L2 .23/ fPGL2 .23/ .x; .23  33 t 2 /=.t 2 C 23//
U4 .2/:2 .x 3 C 6x 2  8/9  24 312 x 6 .x 2 C 5x C 4/4 .x  2/ t
U4 .2/ .x 3 C 6x 2  8/9  24 312 x 6 .x 2 C 5x C 4/4 .x  2/.3t 2 C 1/
S6 .2/ .x 4  10x 2  8x C 1/7  x 3 .x 2 C 3x C 1/5 t
U3 .3/:2 .x 6  6x 5  435x 4  308x 3 C 15x 2 C 66x C 19/4 .x 4 C 20x 3 C 114x 2
C68x C 13/  22 39 .x 2 C 4x C 1/12 .2x C 1/t
U3 .3/ fU3 .3/:2 .x; 1=.t 2 C 1//
PGL2 .29/ .x 5  7x 4 C 8x 3  17x 2 C 9x  6/6  t .x 29 C 29x 26  29x 25 C 29x 24
C290x 23  638x 22 C 899x 21 C 464x 20  4118x 19 C 8323x 18
9686x 17  899x 16 C 20532x 15  46197x 14 C 55477x 13  36801x 12
8584x 11 C 66874x 10  100601x 9 C 105560x 8 73602x 7 C 34017x 6
2349x 5  11745x 4 C 10962x 3  6264x 2 C 1944x  432/ C t 2
L2 .29/ fPGL2 .29/ .x; 22 33 29=.t 2  29//
PSL5 .2/ .x 5  95x 4  110x 3  150x 2  75x  3/3 .x 5 C 4x 4  38x 3 C 56x 2
C53x  4/3 .x  3/  34 t .x 2  6x  1/8 .x 2  x  1/4 .x C 2/4 x
Aut.HS/ .x 4  5/5 .x 8  20x 6 C 60x 5  70x 4 C 100x 2  100x C 25/10
t .7x 5  30x 4 C30x 3 C40x 2  95xC50/4 .2x 10  20x 9 C90x 8  240x 7
C435x 6  550x 5 C425x 4  100x 3  175x 2 C250x  125/4 .2x 10 C5x 8
40x 6 C50x 4  50x 2 C125/4
HS fAut.HS/ .x; .5t 2 C 1/=28 /
2 Regular Realizations for Nonsolvable Primitive Groups 501

Table 2.2 The cyclic group Z16

Z16 x 16  24 16 .t /x 14 C 24 .16t 6  14t 4 C 6t 2 C 5/16 .t /x 12
26 .24t 12  28t 10 C6t 8 C36t 6  31t 4 C13t 2 C2/16 .t /x 10
C25 .128t 18  120t 16  144t 14 C560t 12  488t 10 C144t 8 C164t 6
136t 4 C56t 2 C1/16 .t /x 8
28 .16t 22 C16t 20  120t 18 C208t 16  108t 14  64t 12 C164t 10
128t 8 C73t 6  20t 4 C3t 2 C2/t 2 16 .t /x 6
C28 .64t 24  192t 22 C208t 20 C80t 18  432t 16
C520t 14  316t 12 C112t 10 C18t 8  66t 6 C67t 4  26t 2 C5/t 4 16 .t /x 4
210 .32t 22  112t 20 C160t 18  72t 16  84t 14 C144t 12  86t 10 C28t 8
17t 6 C17t 4  7t 2 C1/t 6 16 .t /x 2
C28 .8t 10  16t 8 C12t 6  4t 2 C1/2 t 8 16 .t /
502 Appendix: Example Polynomials

3 Realizations over Q for Transitive Groups of Degree up to 14

This last set of tables contains polynomials generating field extensions of Q with
transitive Galois group of degree less than fifteen. The polynomials are mainly taken
from the database Klüners and Malle (2002), which contains polynomials for all but
two transitive groups up to degree 23; see also Klüners and Malle (2000).

Table 3.1 Degree 2

T1 2 x2 C x C 1

Table 3.2 Degree 3

T2 S 3 x3  x  1
T1 3 x 3  x 2  2x C 1

Table 3.3 Degree 4

T5 S4 x4  x C 1
T4 A4 x 4  2x 3 C 2x 2 C 2
T3 D4 x4  x3  x2 C x C 1
T2 V4 x4  x2 C 1
T1 4 x4 C x3 C x2 C x C 1

Table 3.4 Degree 5

T5 S5 x5  x3  x2 C x C 1
T4 A5 x 5 C x 4  2x 2  2x  2
T3 F20 x 5 C x 4 C 2x 3 C 4x 2 C x C 1
T2 D5 x 5  x 3  2x 2  2x  1
T1 5 x 5 C x 4  4x 3  3x 2 C 3x C 1

Table 3.5 Degree 6

T16 S6 x6  x4  x3 C x C 1
T15 A6 x 6  x 3  3x 2  1
T14 PGL2 .5/ x 6  2x 5 C 4x C 2
T13 32 :D4 x6 C x5  x2  x C 1
T12 L2 .5/ x 6  2x 5  5x 2  2x  1
T11 2  S4 x6  x4 C 1
T10 32 :4 x 6 C x 5 C x 4 C x 3  4x 2 C 5
T9 32 :22 x6  x3 C 2
T8 S4 =4 x 6  x 4 C 2x 2 C 2
T7 S4 =V4 x6  x2  1
T6 2  A4 x 6  3x 2 C 1
T5 3  S3 x 6  3x 3 C 3
T4 A4 x 6 C x 4  2x 2  1
T3 D6 x6  x3  1
T2 S3 x6 C 3
T1 6 x6  x3 C 1
3 Realizations over Q for Transitive Groups 503

Table 3.6 Degree 7

T7 S7 x7 C x3  x2 C 1
T6 A7 x 7  2x 6 C 2x C 2
T5 L3 .2/ x 7  7x C 3
T4 F42 x7  2
T3 F21 x 7  8x 5  2x 4 C 16x 3 C 6x 2  6x  2
T2 D7 x 7 C 7x 3  7x 2 C 7x C 1
T1 7 x 7  x 6  12x 5 C 7x 4 C 28x 3  14x 2  9x  1

Table 3.7 Degree 8

T50 S8 x8 C x4 C x C 1
T49 A8 x 8  8x 3 C 10
T48 23 : L3 .2/ x 8  2x 7 C 8x  2
T47 S4 o 2 x 8  5x  5
T46 x 8  8x 3  8x 2 C 1
T45 x 8  3x 4  2x 2  4x  1
T44 2 o S4 x8  x2  1
T43 PGL2 .7/ x 8  x 7 C 7x 6  4x C 4
T42 A4 o 2 x 8  2x 7 C 6x 4 C 4
T41 x 8 C 4x 7  2x 4  4x 2 C 2
T40 x 8 C 4x 6  9
T39 23 :S4 x8 C x2 C 1
T38 2 o A4 x 8 C 2x 6 C 2x 4 C 2
T37 L2 .7/ x 8  4x 7 C 7x 6  7x 5 C 7x 4  7x 3 C 7x 2 C 5x C 1
T36 23 :7:3 x 8 C x 7 C x 6  3x 5 C 5x 4 C 5x 3  7x C 9
T35 2o2o2 x 8 C 2x 6 C 2
T34 x 8  x 7 C 2x 6  x 5  2x 4 C 4x 3  6x C 4
T33 x 8  4x 5 C 12x 4  8x 2 C 12x C 9
T32 x 8 C x 6 C 3x 2 C 4
T31 2 o 22 x 8 C 4x 6  8x 2  1
T30 x 8  4x 6 C 4x 4  2
T29 23 :D4 x8  x6 C x2 C 1
T28 x 8 C 4x 6 C 2
T27 2o4 x 8  8x 4 C 8x 2  2
T26 x8 C x4 C 2
T25 23 :7 x 8  4x 7 C 8x 6  6x 5 C 2x 4 C 6x 3  3x 2 C x C 3
T24 S4  2 x 8  4x 2 C 4
T23 GL2 .3/ x 8  6x 4  x 2  3
T22 x8  x4 C 4
T21 x 8  2x 6 C x 4 C 5
T20 x 8  3x 6  x 4 C 3x 2 C 1
T19 x 8 C 4x 4  4x 2 C 1
T18 22 o 2 x 8  x 6 C 2x 2 C 1
T17 4o2 x 8  2x 4 C 2
T16 x 8 C 4x 4 C 2
T15 x8 C 3
T14 S4 x 8 C 4x 6 C 4x 2 C 4
T13 A4  2 x 8 C 2x 6 C 3x 4  3x 2 C 1
504 Appendix: Example Polynomials

T12 SL2 .3/ x 8 C 9x 6 C 23x 4 C 14x 2 C 1

T11 x8 C 9
T10 x 8  2x 6 C 4x 4  3x 2 C 1
T9 D4  2 x 8 C 4x 4 C 1
T8 x8  2
T7 x 8  15x 4 C 10x 2 C 5
T6 D8 x8 C 2
T5 Q4 x 8 C 12x 6 C 36x 4 C 36x 2 C 9
T4 D4 x 8 C 3x 4 C 1
T3 23 x8  x4 C 1
T2 42 x8 C 1
T1 8 x 8 C x 7  7x 6  6x 5 C 15x 4 C 10x 3  10x 2  4x C 1

Table 3.8 Degree 9

T34 S9 x9 C x5  x2 C 1
T33 A9 x 9  3x 3 C x C 2
T32 L2 .8/ x 9 C x 7 C 2x 5 C 4x 3  x 2 C x C 1
T31 S3 o S3 x 9  x 8 C 2x 2  x C 1
T30 x 9 C 2x 5  4x 4 C 4x 3  4x 2 C x  1
T29 x 9  3x 6  5x 5 C 5x 2  1
T28 S3 o 3 x 9  2x 6  4x 3 C 3x C 1
T27 L2 .8/ x 9 C x 7  4x 6  12x 4  x 3  7x 2  x  1
T26 32 : GL2 .3/ x 9  x 7 C 5x 6 C x 5  2x 4 C 4x 3 C 3x 2  x  1
T25 x 9  3x 6 C 9x 5  9x 4  27x 3 C 9x C 1
T24 x 9  2x 6  2
T23 32 : SL2 .3/ x 9  3x 8 C x 6 C 15x 5  13x 4  3x 3 C 4x  1
T22 x 9  3x 6 C 3
T21 x 9  6x 3  6
T20 3 o S3 x 9  x 6  2x 3 C 1
T19 x 9  3x 8 C 18x 5 C 18x 4  27x C 9
T18 x9  x3  1
T17 3o3 x 9 C x 8  10x 7  14x 6 C 20x 5 C 36x 4  18x 2  8x  1
T16 32 :D4 x 9  x 8  x 5  x 4 C 3x 3 C 2x 2  1
T15 32 :8 x 9  4x 8 C 8x 7  32x 5 C 80x 4  104x 3 C 80x 2  34x C 8
T14 32 :Q4 x 9  12x 5 C 132x  128
T13 x 9  3x 3  1
T12 x 9  2x 8 C x 5  3x 3 C 4x 2  12x C 8
T11 32 :6 x 9  x 6 C 5x 3 C 1
T10 9:6 x9  2
T9 32 :4 x 9 C 2x 7  3x 6 C x 5  x 4 C 64x 3  x  1
T8 S32 x 9 C 3x 3  1
T7 32 :3 x 9  3x 8  21x 7 C 78x 5 C 69x 4  21x 3  39x 2  12x  1
T6 9:3 x 9  14x 7 C 63x 5  98x 3 C 42x  7
T5 32 :2 x 9  3x 6  3x 3  1
T4 S3  3 x 9  3x 6  6x 3  1
T3 D9 x 9  9x 6 C 27x 3  3
T2 32 x 9  15x 7 C 4x 6 C 54x 5  12x 4  38x 3 C 9x 2 C 6x  1
T1 9 x 9  9x 7 C 27x 5  30x 3 C 9x  1
3 Realizations over Q for Transitive Groups 505

Table 3.9 Degree 10

T45 S10 x 10  x 3  1
T44 A10 x 10  2x 9 C 3x 5  4
T43 S5 o 2 x 10 C 3x 6  2x 5 C 1
T42 x 10 C 5x 8  5x 7 C 5x 6  7x 5  5x 4  10x 2  4
T41 x 10  2x 9  x 6 C x 4  4x 2 C 2x  1
T40 A5 o 2 x 10  x 9  x 4  4x 3 C 4x 2  x  1
T39 2 o S5 x 10  x 2 C 1
T38 x 10  3x 8 C 2x 2 C 2
T37 24 :S5 x 10  x 2  1
T36 2 o A5 x 10 C x 4  2x 2 C 3
T35 PL2 .9/ x 10  4x 9 C 6x 8 C 12x 2 C 16x C 8
T34 24 :A5 x 10 C 4x 4 C x 2  4
T33 .5:4/ o 2 x 10 C 6x 6 C 8x 5  35x 2 C 24x C 16
T32 S6 x 10  2x 9 C x 8  9x 2 C 2x  1
T31 M10 x 10  2x 9 C 9x 8  54x 2 C 108x  54
T30 PGL2 .9/ x 10  2x 9 C 9x 8  7x 2 C 14x  7
T29 2 o .5:4/ x 10 C 10x 6 C 5
T28 x 10  10x 7 C 10x 6 C 36x 5 C 50x 4  10x 3  1
T27 x 10 C 3x 6  2x 5 C x 2 C 2x C 1
T26 L2 .9/ x 10  x 9 C 3x 8  6x 7 C 3x 6  3x 5  3x 3  6x 2  8x  1
T25 x 10 C 10x 6  5
T24 24 :5:4 x 10 C 5x 6 C 5x 2  1
T23 2 o .5:2/ x 10  5x 4  3
T22 S5  2 x 10 C 4x 2 C 4
T21 D5 o 2 x 10 C x 6  2x 5  x 4 C 3x 2  2x C 1
T20 52 :Q4 x 10  10x 8 C 35x 6  4x 5  50x 4 C 20x 3 C 25x 2  20x  17
T19 52 :D4 x 10  10x 8 C 35x 6  2x 5  50x 4 C 10x 3 C 25x 2  10x C 2
T18 52 :8 x 10 C 60x 6  208x 5 C 850x 2  8000x  4672
T17 x 10 C x 5 C 2
T16 x 10  5x 4 C 15
T15 24 :5:2 x 10  5x 4  4x 2  1
T14 2o5 x 10 C x 8  4x 6  3x 4 C 3x 2 C 1
T13 S5 =D6 x 10  x 9  x 8 C 3x 6  x 5  2x 4 C 3x 3  x 2  x C 1
T12 S5 =A4 x 10 C 2x 9 C 3x 8  x 6  2x 5  x 4 C 3x 2 C 2x C 1
T11 A5  2 x 10 C x 8  4x 2 C 4
T10 52 :4 x 10  2x 5  4
T9 52 :22 x 10  x 9  5x 8 C 11x 6 C 4x 5  10x 4 C 25x 2 C 5x  5
T8 24 :5 x 10  4x 8 C 2x 6 C 5x 4  2x 2  1
T7 A5 x 10  x 8  4x 7  3x 6  2x 5 C 8x 3  2x  1
T6 5o2 x 10  x 9 C 3x 7  3x 6 C x 5 C 5x 4  x 3 C 2x 2 C 3x C 1
T5 2  5:4 x 10 C 2
T4 5:4 x 10  5
T3 D10 x 10  3x 4 C 2x 2 C 1
T2 D5 x 10 C 5x 8 C 15x 6 C 20x 4 C 25x 2 C 15
T1 10 x 10 C x 9 C x 8 C x 7 C x 6 C x 5 C x 4 C x 3 C x 2 C x C 1
506 Appendix: Example Polynomials

Table 3.10 Degree 11

T8 S11 x 11 C x 6 C x 4 C 1
T7 A11 x 11  6x 8 C 4x 5  3x 3 C 2
T6 M11 x 11  4x 10 C 60x 7  108x 6 C 72x 5  360x 4 C 3636x  1944
T5 L2 .11/ x 11  2x 10 C x 9  5x 8 C 13x 7  9x 6 C x 5  8x 4 C 9x 3  3x 2  2x C 1
T4 F110 x 11  3
T3 F55 x 11  33x 9 C 396x 7  2079x 5 C 4455x 3  2673x  243
T2 D11 x 11  x 10 C 5x 8 C 8x 5 C 6x 4  x 3 C x 2 C 3x C 1
T1 11 x 11 C x 10  10x 9  9x 8 C 36x 7 C 28x 6  56x 5  35x 4 C 35x 3
C15x 2  6x  1

Table 3.11 Degree 12

S12 x 12  x C 1
A12 x 12 C 3x 8 C 3x 4 C 4x 3 C 4
T299 x 12 C x 2  2x C 1
T298 x 12  72x 2  120x  50
T297 x 12  2x 7 C 7x 6 C x 2  2x C 1
T296 x 12  x 7  7x 6  5x 4  x 2 C x C 1
M12 x 12  375x 8  3750x 6  75000x 3 C 228750x 2  750000x C 1265625
T294 x 12 C 4x 9  6x 7 C 2
T293 x 12  x 6  x 2  1
T292 x 12  3x 11 C 5x 9  3x 8 C 3x 7 C 2x 6  6x 5  3x 4 C 1
T291 x 12  12x 9  9x 8  64x 3  144x 2  108x  27
T290 x 12 C 3x 10  x 9 C 2x 6  3x 5 C 9x 4  3x 3 C 3x C 1
T289 x 12  4x 9 C 2x 6 C 4x 4 C 1
T288 x 12  4x 11 C 4x 10  50x 4 C 120x 3  112x 2 C 48x  8
T287 x 12  3x 8 C 2x 6 C 3
T286 x 12  x 6  3x 4  1
T285 x 12  4x 2 C 4
T284 x 12  12x 9  9x 8 C 64x 3 C 144x 2 C 108x C 27
T283 x 12  8x 9 C 24x 6 C 144x 5 C 96x 3 C 144x 2 C 48
T282 x 12  x 11 C 3x 10  x 9 C 6x 8 C 6x 6  2x 5 C 7x 4 C 4x 3 C 4x 2 C x C 1
T281 x 12  x 11 C x 10  2x 8 C 3x 7  3x 5 C 3x 4  2x 3 C 3x 2  x C 1
T280 x 12  4x 11 C 6x 10  2x 9  5x 8 C 6x 7  4x 5 C 2x 4 C 2
T279 x 12  4x 11 C 4x 10 C 4x 7  6x 6  4x 5 C 36x 2 C 36x C 9
T278 x 12 C 20x 8  80x 6 C 50x 4  320x 3  912x 2 C 1280x C 800
T277 x 12 C 3x 6 C 3x 2 C 4
T276 x 12 C 192x 6  288x 5 C 108x 4 C 256x 3  576x 2 C 432x  108
T275 x 12 C x 10  9x 9 C 11x 8  11x 7 C 17x 6  7x 5 C 2x 4 C x C 1
T274 x 12  x 8 C 2x 6  4x 3 C 1
T273 x 12  12x 9 C 9x 8 C 192x 3  432x 2 C 324x  81
M11 x 12 C 6x 11 C 15x 10 C 28x 9 C 36x 8 C 6x 7  75x 6  108x 5 C 18x 4 C 82x 3
C3x 2  6x C 5
T271 x 12  135x 8  180x 7 C 399x 6 C 918x 5 C 693x 4 C 352x 3 C 216x 2 C 96x C 16
T270 x 12 C x 10 C 4x 2  1
T269 x 12  2x 10  6x 8 C 14x 6 C x 4  8x 3 C 1
T268 x 12 C 4x 9  3x 8  64x 3 C 144x 2  108x C 27
3 Realizations over Q for Transitive Groups 507

T267 x 12 C 12x 10  8x 9 C 54x 8  48x 7 C 132x 6  72x 5  33x 4  32x 3 C 8

T266 x 12 C x 10  4x 9  2x 8  3x 7 C 4x 5 C 2x 3  2x 2  x C 1
T265 x 12  8x 10 C 7x 9 C 8x 8  7x 7  15x 6 C 21x 5 C 8x 4  14x 3  8x 2 C 7x C 1
T264 x 12  4x 11 C 6x 10  3x 9  2x 8 C 3x 7  2x 5 C x 4 C x 3  x 2 C 1
T263 x 12  162x 4  432x 3  432x 2  192x  32
T262 x 12 C 18x 8  24x 7 C 8x 6  81x 4 C 216x 3  216x 2 C 96x  16
T261 x 12 C 2x 6  4x 5 C x 4 C 1
T260 x 12  3x 2 C 3
T259 x 12  12x 10 C 54x 8  110x 6 C 93x 4  4x 3  18x 2 C 12x  8
T258 x 12  x 3  3
T257 x 12 C 4x 10  4x 2 C 4
T256 x 12 C 4x 10  5x 2 C 5
T255 x 12  2x 8  6x 6 C 9x 4  1
T254 x 12 C 6x 10  12x 9  54x 7 C 24x 6 C 180x 4 C 156x 3 C 216x 2 C 72x C 18
T253 x 12  4x 9  3x 8  32x 6  48x 5  18x 4 C 64x 3 C 144x 2 C 108x C 27
T252 x 12  12x 9 C 27x 8 C 12x 6  36x 5 C 27x 4  16x 3 C 36x 2 C 9
T251 x 12 C 48x 6  72x 5 C 27x 4 C 64x 3  144x 2 C 108x  27
T250 x 12 C 3x 2 C 5
T249 x 12  12x 10 C 54x 8  108x 6 C 81x 4  8x 3 C 24x C 8
T248 x 12 C 324x 6  648x 5 C 675x 4  744x 3 C 648x 2  288x C 48
T247 x 12  8x 9 C 24x 6 C 162x 4  32x 3 C 16
T246 x 12 C 81x 4  216x 3 C 216x 2  96x C 16
T245 x 12  12x 10  54x 8  72x 7 C 96x 6 C 9x 4 C 200x 3 C 108x 2  4
T244 x 12  3x 11  6x 10 C 13x 9 C 6x 8  15x 7 C 5x 6  15x 5 C 15x 4 C 5x 3  5
T243 x 12  9x 8  12x 7  4x 6  81x 4  216x 3  216x 2  96x  16
T242 x 12  4x 9 C 18x 8  4x 6  36x 5 C 81x 4 C 16x 3 C 108x 2 C 16
T241 x 12 C x 10  3x 8  x 6 C 6x 4  3
T240 x 12 C 6x 8 C 4x 6  4
T239 x 12  12x 9 C 9x 8  32x 6 C 48x 5  18x 4  64x 3 C 144x 2  108x C 27
T238 x 12 C 6x 10 C 9x 8  8
T237 x 12 C 10x 2 C 5
T236 x 12 C x 4 C 2x 2 C 1
T235 x 12 C 3x 8  4x 6 C 2
T234 x 12 C x 9 C 3x 3 C 4
T233 x 12  4x 3  6
T232 x 12  13x 8  26x 7  11x 6 C 6x 5 C 25x 4 C 78x 3 C 114x 2 C 76x C 19
T231 x 12  x 11 C 2x 9  x 8  4x 7 C 5x 6  x 5  x 4  x 3 C 4x 2  3x C 1
T230 x 12 C x 10  3x 8 C 4x 4 C 1
T229 x 12  18x 10  22x 9 C 102x 8 C 180x 7  96x 6  90x 5 C 81x 4  30x 3
54x 2 C 3
T228 x 12 C 4x 11 C 3x 10  2x 9 C 11x 8 C 30x 7 C 14x 6  11x 5 C 12x 4 C 30x 3
Cx 2  9x  1
T227 x 12 C 2x 10 C x 8  4x 4 C 3
T226 x 12  3x 8  6x 2 C 1
T225 x 12  3x 10 C 2x 6 C 2x 4  3
T224 x 12 C 4x 8 C 6x 6  6x 2 C 2
T223 x 12  6x 10 C 12x 6  9
T222 x 12  4x 6 C 3x 2  1
T221 x 12  2x 10  x 8 C 6x 6  x 4  4x 2  1
508 Appendix: Example Polynomials

T220 x 12  4x 9  12x 8 C 34x 6  12x 5 C 45x 4 C 42x 2 C 10

T219 x 12 C 2x 6 C x 2 C 1
T218 x 12  2x 11 C 22x 9  88x 7 C 176x 5  176x 3 C 64x C 4
T217 x 12  4x 9 C 2
T216 x 12  12x 10  8x 9 C 162x 4 C 432x 3 C 432x 2 C 192x C 32
T215 x 12  3x 10 C x 9  81x 8 C 54x 7  36x 6 C 27x 5 C 72x 4  107x 3
C54x 2  12x C 1
T214 x 12  12x 9 C 18x 8  56x 6 C 138x 4  96x 3 C 72x 2 C 72
T213 x 12  x 3  1
T212 x 12  12x 10  18x 8  96x 7  132x 6  63x 4  64x 3 C 72x 2  16
T211 x 12 C 90x 8 C 120x 7 C 40x 6 C 405x 4 C 1080x 3 C 1080x 2 C 480x C 80
T210 x 12  4x 9 C 8x 6  36x 5 C 105x 4  120x 3 C 90x 2  36x C 9
T209 x 12  8x 9 C 18x 8  24x 7 C 24x 6  33x 4  16x 3  48x  8
T208 x 12  3x 10 C 3x 6 C 3x 4 C 3
T207 x 12  x 11 C x 9  x 7  x 6 C 2x 5  x 4  3x 3 C 3x 2  2x C 1
T206 x 12  12x 9 C 15x 8  12x 5 C 18x 4  64x 3 C 96x 2  36x C 9
T205 x 12  208x 6  312x 5  117x 4  832x 3  1872x 2  1404x  351
T204 x 12  6x 9 C 18x 8 C 48x 6 C 108x 4  32x 3  72x C 24
T203 x 12  2x 6 C x 4 C 1
T202 x 12  4x 6 C 9x 4 C 4
T201 x 12 C 3x 10  12x 2 C 24
T200 x 12 C 6x 10 C 9x 8  12
T199 x 12  2x 10  4x 8  x 6 C x 4 C 4
T198 x 12  2x 10  x 8 C 6x 4  4x 2 C 2
T197 x 12 C 4x 6  9x 4 C 8
T196 x 12  2x 8  4x 6 C 6x 4 C 4x 2  1
T195 x 12 C 4x 10 C 2x 8  4x 6 C 4
T194 x 12 C 2x 10 C 2x 9  x 8  2x 7 C 4x 6  12x 5 C 6x 4 C 2x 3 C 18x 2 C 27
T193 x 12 C 6x 6 C 6x 4 C 3
T192 x 12  6x 10 C x 8 C 36x 6  30x 4  28x 2 C 18
T191 x 12 C x 10 C 2x 8  x 6 C 2x 4  3x 2 C 1
T190 x 12 C 2x 10  13x 8 C 36x 6 C 15x 4  38x 2  19
T189 x 12  6x 8 C 12x 4 C 13x 2 C 5
T188 x 12  2x 10 C 5x 6 C 5x 2  1
T187 x 12 C 8x 6  9x 2 C 1
T186 x 12  x 10  x 2  1
T185 x 12  x 4  2
T184 x 12 C x 8 C 9x 6 C 9x 4 C 7x 2 C 1
T183 x 12  7x 6  10x 4  5x 2 C 1
T182 x 12  8x 9 C 6x 8 C 20x 6  24x 5 C 18x 4  16x 3 C 24x 2 C 8
T181 x 12  18x 8  36x 6  72x 5 C 54x 4  144x 3  216x 2  72
T180 x 12  2x 10 C 5x 8  8x 6 C 6x 4  4x 2 C 1
L2 .11/ x 12 C x 11  8x 10  29x 9 C 48x 8 C 51x 7  5x 6 C 275x 5 C 642x 4 C 208x 3
C308x 2 C 41x C 2
T178 x  x 9 C 4x 3  1
12

T177 x 12  4x 9 C 4x 3 C 2
T176 x 12 C 4x 6  8x 3 C 8
T175 x 12  2x 11 C 4x 10  2x 9 C 4x 7  3x 6 C 2x 5 C x 2  2x C 1
T174 x 12 C 12x 10 C 54x 8 C 20x 6  447x 4  384x 3  792x 2  1152x  368
3 Realizations over Q for Transitive Groups 509

T173 x 12  36x 8  48x 7  32x 6 C 162x 4  288x 2 C 128

T172 x 12 C 12x 10  6x 9  54x 7  157x 6 C 210x 4 C 174x 3 C 234x 2 C 252x C 118
T171 x 12  8x 9  36x 8  72x 5 C 81x 4 C 64x 3  144x 2 C 64
T170 x 12  x 9 C 2x 6 C 4x 3 C 3
T169 x 12  8x 3 C 18
T168 x 12  10x 6  12x 3  2
T167 x 12  3x 3 C 3
T166 x 12 C 18x 10 C 135x 8 C 348x 6 C 63x 4  512x 3  270x 2 C 729
T165 x 12  16x 9 C 12x 8 C 256x 3  576x 2 C 432x  108
T164 x 12 C 4x 9 C 6x 7 C 8x 6  54x 5 C 88x 3  57x 2  90x C 111
T163 x 12  x 8  2x 6 C x 4  2x 2 C 1
T162 x 12  2x 8  8x 6 C 14x 4  16x 2 C 4
T161 x 12 C 3x 10 C 18x 2 C 9
T160 x 12 C x 10 C x 8 C x 6  4x 4 C 5
T159 x 12 C 4x 10  4x 8  24x 6  x 4 C 32x 2 C 8
T158 x 12  x 8  2x 6 C 2x 2 C 1
T157 x 12  8x 9 C 24x 7 C 44x 6  51x 4 C 48x 3  72x 2 C 16
T156 x 12  2x 9 C 2
T155 x 12  2x 10  3x 8 C 2
T154 x 12  2x 6 C 12x 4  6x 2 C 7
T153 x 12 C 2x 10 C 8x 2 C 8
T152 x 12  4x 8  2x 6 C 4x 4  1
T151 x 12  3x 8  2
T150 x 12  x 6  3x 4 C 2x 2 C 2
T149 x 12  9x 4  6
T148 x 12 C 3x 10 C 3x 8 C x 6  3
T147 x 12  3x 8  8
T146 x 12  2x 10  x 8  2x 6  2x 4  8x 2 C 8
T145 x 12 C 6x 8 C 4x 6  18x 4  24x 2  8
T144 x 12 C 6x 10 C 4x 8  24x 6  21x 4 C 22x 2 C 4
T143 x 12  6x 10 C 24x 8  56x 6 C 93x 4  90x 2 C 51
T142 x 12 C 3x 8 C 4x 6 C 6x 4 C 3
T141 x 12 C 3x 8  3
T140 x 12  x 4  4
T139 x 12 C 3x 10 C 3x 2 C 1
T138 x 12  x 4 C 1
T137 x 12 C x 8  2x 6  x 4  1
T136 x 12  x 10 C 4x 2 C 1
T135 x 12  18x 8  24x 6 C 27x 4 C 36x 2  6
T134 x 12  7x 10 C 14x 8  21x 4 C 7x 2 C 7
T133 x 12  8x 9 C 162x 8  372x 7 C 20x 6 C 432x 5  63x 4  212x 3  36x 2
C24x C 56
T132 x 12  x 10  11x 9 C 99x 8  45x 7  117x 6  27x 5 C 90x 4 C 36x 3 C 9x C 18
T131 x 12  2x 11  x 10 C 9x 9  7x 8  11x 7 C 20x 6 C x 5  19x 4 C 8x 3
C6x 2  5x C 1
T130 x 12  2x 9 C x 6 C 6x 3 C 3
T129 x 12  6x 10  2x 9 C 3x 8  30x 7 C 8x 6 C 90x 5 C 36x 4  24x 3 C 6x  1
T128 x 12  12x 10  22x 9 C 57x 8  72x 6 C 30x 5 C 15x 4  30x 3 C 6x C 1
T127 x 12  16x 9 C 18x 8  72x 6 C 36x 5  36x 4  76x 3  72x  62
T126 x 12 C x 8 C x 6  2x 4  x 2 C 1
510 Appendix: Example Polynomials

T125 x 12  2x 8  2x 6 C x 4 C 2x 2  1
T124 x 12 C 4x 10 C 10x 6 C 5
T123 x 12  2x 10 C 10x 6  8x 2 C 1
T122 x 12  2x 11  3x 10  6x 9 C 21x 8  32x 7 C 37x 6  16x 5 C 11x 4 C 32x 3
x 2 C 20x C 1
T121 x  x 9 C 2x 3 C 1
12

T120 x 12  2x 9  6x 3 C 9
T119 x 12  8x 6  8x 3  2
T118 x 12 C 8x 6  8x 3 C 2
T117 x 12  2x 9 C x 6 C 5
T116 x 12  2x 9 C 4x 3 C 4
T115 x 12  2x 8 C 3x 4  4
T114 x 12  x 4  1
T113 x 12  x 4 C 4
T112 x 12  3x 8 C 9x 4 C 1
T111 x 12  6x 8 C 68x 6 C 105x 4 C 36x 2 C 12
T110 x 12 C x 8  x 6  x 4  1
T109 x 12 C x 10  4x 2 C 1
T108 x 12  3x 8  4x 6 C 6x 4 C 4
T107 x 12 C 6x 10 C 3x 8  28x 6  21x 4 C 30x 2 C 5
T106 x 12 C 3x 10  2x 8  9x 6 C 5x 2 C 1
T105 x 12  7x 10 C 7x 8 C 14x 6  16x 4  5x 2 C 5
T104 x 12 C 6x 10 C 12x 8 C 8x 6  3x 4  6x 2  1
T103 x 12 C 3x 10  x 6 C 3x 2 C 1
T102 x 12  5x 10 C 20x 8  70x 6 C 145x 4  280x 2 C 208
T101 x 12  3x 10  3x 2 C 1
T100 x 12  x 10 C x 8 C 4x 6  x 4  x 2  1
T99 x 12  76x 8 C 325x 6  380x 4 C 125
T98 x 12  64x 10  231x 8 C 740x 6  481x 4 C 37
T97 x 12 C x 8 C 9x 4 C 1
T96 x 12  3x 4  4
T95 x 12  x 10 C 3x 6  2x 4  3x 2 C 1
T94 x 12  57x 8  38x 6 C 318x 4  204x 2 C 17
T93 x 12 C 10x 10 C 28x 8 C 6x 6  43x 4 C 6x 2 C 3
T92 x 12  9x 4  9
T91 x 12 C 5x 10 C 9x 8 C 8x 6 C 2x 4  12x 2 C 16
T90 x 12 C 2x 10  x 6 C 2x 2 C 1
T89 x 12  3x 4 C 1
T88 x 12  6x 8  4x 6  3x 4  18x 2 C 3
T87 x 12 C 6x 10 C 9x 8  4x 6  12x 4 C 1
T86 x 12 C 2x 8  2
T85 x 12  3x 11  3x 10 C 15x 9  15x 8  33x 7 C 29x 6 C 15x 5  30x 4  128x 3
30x 2 C 198x C 48
T84 x 12  6x 10 C 4x 9 C 21x 8  12x 7  52x 6  16x 3 C 48x 2 C 16
T83 x 12 C 3x 6  x 3 C 3
T82 x 12  12x 10 C 54x 8  116x 6 C 129x 4  72x 2  16
T81 x 12 C x 6 C 2
3 Realizations over Q for Transitive Groups 511

T80 x 12  90x 8 C 160x 6  135x 4 C 7200x 2  80

T79 x 12 C 4x 10 C 6x 8 C 4x 6 C 2
T78 x 12  x 9 C x 3 C 1
T77 x 12  2x 6 C 5x 2 C 1
T76 x 12 C 2x 8 C 5x 4 C 6x 2 C 1
T75 x 12 C 7x 8 C 7x 4 C 8x 2 C 1
T74 x 12  x 10 C 2x 8 C 4x 6  3x 4  3x 2 C 1
T73 x 12  3x 11 C 4x 10  x 8  6x 7 C 20x 6  10x 5 C 8x 4 C 24x 3 C 3x 2 C 12x C 9
T72 x 12  6x 10  10x 9 C 36x 8  116x 6 C 720x 5 C 696x 4  2440x 3  720x 2
C1200x C 880
T71 x 12  4x 9 C 4x 6 C 3
T70 x 12 C 9x 6  18x 3 C 9
T69 x 12  3x 10  2x 8 C 9x 6  5x 2 C 1
T68 x 12 C x 10 C 6x 8 C 3x 6 C 6x 4 C x 2 C 1
T67 x 12  x 8  x 6  x 4 C 1
T66 x 12 C 6x 10 C 12x 8 C 8x 6  3
T65 x 12  3x 4 C 4
T64 x 12 C 3x 8  16
T63 x 12  6x 10 C 104x 6 C 93x 4 C 18x 2 C 4
T62 x 12  3x 10 C 3x 8  x 6 C 4x 4  4x 2 C 1
T61 x 12  3x 4  1
T60 x 12  4x 8  9x 4 C 4
T59 x 12  6x 10 C 6x 8  4x 6  3x 4 C 3
T58 x 12  12x 8  14x 6 C 9x 4 C 12x 2 C 1
T57 x 12 C 38x 10 C 533x 8 C 3474x 6 C 10574x 4 C 12740x 2 C 4225
T56 x 12  2x 10 C x 6  2x 2 C 1
T55 x 12 C 2x 10  97x 8  360x 6  345x 4  50x 2 C 25
T54 x 12  6x 8 C 9x 4 C 2
T53 x 12 C 2x 8  16x 6 C 4x 4 C 8
T52 x 12  3x 4  6
T51 x 12 C 6x 8 C 9x 4 C 3
T50 x 12  3x 4 C 6
T49 x 12 C 3x 8  4x 6  3x 4  1
T48 x 12 C 8x 4 C 1
T47 x 12  6x 10 C 20x 9  72x 7 C 128x 6  96x 5 C 45x 4  8x 3  18x 2 C 12x  2
T46 x 12  4x 11 C 6x 10 C 4x 9  21x 8 C 40x 7  28x 6  8x 5 C 25x 4  28x 3
C10x 2  4x  1
T45 x 12  3x 9  18x 8  24x 6  9x 5 C 69x 4  x 3 C 3x  1
T44 x 12  6x 6  10x 3  6
T43 x 12  6x 9 C 10x 6 C 4x 3 C 2
T42 x 12  x 6 C 7
T41 x 12  x 9  6x 6 C x 3 C 1
T40 x 12  7x 10 C 24x 8  36x 6 C 24x 4 C 13x 2 C 1
T39 x 12  4x 6 C 2
T38 x 12 C x 6  3
T37 x 12 C x 6 C 4
T36 x 12  2x 9  2x 3 C 1
T35 x 12  x 9  x 6 C x 3 C 1
512 Appendix: Example Polynomials

T34 x 12 C 12x 10 C 54x 8 C 108x 6 C 81x 4 C 16

T33 x 12 C 2x 8 C 58x 6 C 301x 4 C 174x 2 C 25
T32 x 12 C 7x 10  x 8  23x 6  x 4 C 7x 2 C 1
T31 x 12 C 6x 10  23x 8  210x 6  360x 4  50x 2 C 25
T30 x 12  7x 10  14x 8 C 115x 6  70x 4  175x 2 C 125
T29 x 12  45x 8 C 50x 6 C 225x 4  375x 2 C 125
T28 x 12 C 2
T27 x 12 C 12x 10 C 68x 8 C 220x 6 C 392x 4 C 360x 2 C 148
T26 x 12  9x 8  8x 6  9x 4 C 1
T25 x 12 C 5x 8 C 6x 4 C 1
T24 x 12  2x 8  7x 4 C 16
T23 x 12  4x 4 C 4
T22 x 12  5x 10 C 7x 8  6x 7  17x 6  6x 5 C 7x 4  5x 2 C 1
T21 x 12 C 3x 8  4x 6 C 3x 4 C 1
T20 x 12  4x 9 C 72x 8  84x 7 C 236x 6  144x 5 C 324x 4  192x 3 C 72x 2 C 8
T19 x 12 C 24x 10 C 196x 8 C 600x 6 C 452x 4 C 112x 2 C 8
T18 x 12 C 2x 6 C 4
T17 x 12 C 4x 8 C 4x 6 C 5x 4 C 12x 2 C 2
T16 x 12  x 6 C 4
T15 x 12 C 3
T14 x 12  9x 6 C 27
T13 x 12  3
T12 x 12 C x 6  27
T11 x 12  8x 6 C 8
T10 x 12 C 9
T9 x 12 C 3x 8 C 4x 6 C 3x 4 C 1
T8 x 12  6x 10  8x 9 C 9x 8 C 12x 7  20x 6 C 9x 4  24x 3  4
T7 x 12 C 4x 10  x 8  x 4 C 4x 2 C 1
T6 x 12 C 2x 10  6x 8 C 2x 6  6x 4 C 2x 2 C 1
T5 x 12  80x 10 C 1820x 8  13680x 6 C 29860x 4  2720x 2 C 32
T4 x 12 C 6x 8 C 26x 6  63x 4 C 162x 2 C 81
T3 x 12 C 36
T2 x 12  x 6 C 1
T1 x 12  x 11 C x 10  x 9 C x 8  x 7 C x 6  x 5 C x 4  x 3 C x 2  x C 1

Table 3.12 Degree 13

T9 S13 x 13  x C 1
T8 A13 x 13 C 156x  144
T7 L3 .3/ x 13 C x 12 C 40x 10 C 13x 9  99x 8 C 180x 7  468x 6  468x 5 C 1644x 4
912 C 24x C 24
T6 F156 x 13  2
T5 F78 x 13 C 3x 9  10x 8  3x 7 C 5x 6  20x 5  11x 4 C 2x 3  10x 2  10x  3
T4 F52 x 13 C 13x 10  26x 8 C 13x 7 C 52x 6  39x 4 C 26x 2 C 13x C 2
T3 F39 x 13  39x 11 C 468x 9  1989x 7  507x 6 C 2886x 5 C 1443x 4  624x 3
234x 2 C 3
T2 D13 x 13  2x 12 C 4x 10  5x 9 C x 8 C 5x 7  11x 6 C 19x 5  22x 4 C 16x 3
10x 2 C 6x  1
T1 13 x 13  x 12  24x 11 C 19x 10 C 190x 9  116x 8  601x 7 C 246x 6 C 738x 5
215x 4  291x 3 C 68x 2 C 10x  1
3 Realizations over Q for Transitive Groups 513

Table 3.13 Degree 14

S14 x 14  x  1
A14 x 14  9x 7 C 49x 5  90
T61 x 14 C x 2  2x C 1
T60 x 14  7x 8  6x 7 C 49x 2 C 84x C 36
T59 x 14  96x 7  1568x 2 C 2304
T58 x 14  7x 8 C 6x 7 C 784x 2  1344x C 576
T57 x 14  x 2 C 1
T56 x 14 C 14x 8  24
T55 x 14  x 2  1
T54 x 14 C 7x 6 C 4
T53 x 14 C 7x 8  7x 6  9
T52 x 14  x 13 C x 12  x 11 C x 10  4x 8 C 6x 7  5x 5 C 5x 4 C x 3  4x 2 C 1
T51 x 14  7x 2  3
T50 x 14  2x 8  5x 6  3x 2  4
T49 x 14  4x 6 C 4
T48 x 14 C 7x 6 C 21x 2 C 50
T47 x 14 C 2x 12  2x 10 C x 6  8x 4 C 5x 2 C 2
T46 x 14 C 5x 10  4x 8 C 2
T45 x 14  7x 12  14x 11 C 21x 10 C 84x 9 C 35x 8  69x 7 C 7x 6 C 84x 5 C 7x 4
C77x 3 C 133x 2 C 35x C 58
T44 x  8x 10  2x 8 C 16x 6 C 6x 4  6x 2  2
14

T43 x 14 C 3x 12  4x 8 C x 6  3x 2 C 1
T42 x 14 C 7x 12  7x 10  49x 8 C 7x 6 C 49x 4  49x 2 C 9
T41 x 14  2x 12  2x 10 C x 8 C 6x 6  x 2  4
T40 x 14 C 2x 12  14x 8 C 35x 6  21x 4  7x 2 C 7
PGL2 .13/ x 14  x 13  26x 10 C 65x 6 C 13x 5  52x 2  12x  1
T38 x 14  7x 8  14x 6  7
T37 x 14  28x 11  28x 9 C 196x 8  2x 7 C 392x 6 C 616x 4  392x 3
C14x 2 C 56x C 9
T36 x  35x 12  133x 11 C 469x 10 C 1239x 9 C 742x 8  3604x 7 C 47138x 6
14

85351x 5 C 168028x 4  156394x 3 C 158718x 2  72149x C 42751

T35 x 14  9x 12 C 17x 10 C 29x 8  49x 6  67x 4  21x 2  1
T34 x 14  3x 12 C 4x 8 C x 6  3x 2  1
T33 x 14 C 14x 10 C 28x 8  35x 6 C 784x 4  140x 2  4
T32 x 14  14x 12 C 77x 10  210x 8 C x 7 C 294x 6  7x 5  196x 4 C 14x 3
C49x 2  7x C 2
T31 x 14  7x 12 C 91x 8  192x 7  126x 5  1519x 4 C 1218x 3 C 8827x 2
C11046x C 5484
L2 .13/ x 14  6x 13 C 13x 12  338x 9 C 845x 8 C 17576x 4 C 70304x C 35152
T29 x 14 C 12x 12 C 41x 10 C 26x 8  59x 6  64x 4 C 9x 2 C 17
T28 x 14 C 7x 6 C 7x 4 C 7x 2  1
T27 x 14 C 7x 8  14x 6 C 7
T26 x 14  28x 11 C 280x 10 C 567x 9 C 5061x 8 C 2273x 7  735x 6 C 33908x 5
C40348x 4  3192x 3 C 36855x 2 C 119196x C 75141
514 Appendix: Example Polynomials

T25 x 14 C 42x 12  42x 11 C 525x 10  896x 9 C 2422x 8  2536x 7 C 1225x 6

C742x 5  994x 4 C 560x 3  28x 2  168x C 56
T24 x  3x 7 C 6
14

T23 x 14  14x 12 C 77x 10  210x 8  11x 7 C 294x 6 C 77x 5  196x 4  154x 3 C 49x 2
C77x C 29
T22 x 14 C 42x 12  840x 11 C 4473x 10  77728x 9 C 235648x 8  2601696x 7
C6832756x 6  48638016x 5 C 124211584x 4  490172256x 3
C802837840x 2  1497646080x C 723639232
T21 x  x 12  12x 10 C 7x 8 C 28x 6  14x 4  9x 2  1
14

T20 x 14  2x 13  4x 12 C x 11 C 6x 9 C 10x 8  x 7 C 6x 6  13x 4  15x 3  5x 2 C x  1

T19 x 14 C 10x 8 C 8x 6  4x 4 C 2
T18 x 14 C 4x 12  30x 10 C 8x 8 C 60x 6 C 8x 4  24x 2  8
T17 x 14 C 11x 12 C 53x 10 C 15x 8  149x 6 C 89x 4  x 2  3
T16 x 14  14x 10 C 14x 8 C 22x 7 C 21x 6 C 49x 4  154x 3 C 77x 2  154x C 149
T15 x 14  87x 12 C 1456x 10  256x 9  8563x 8 C 3448x 7 C 18032x 6  9890x 5
11776x 4 C 5198x 3 C 3128x 2  506x  184
T14 x  2x 7 C 8
14

T13 x 14 C 4x 13 C 10x 11 C 39x 10 C 28x 9  13x 8 C 34x 7 C 126x 6  36x 5 C 29x 4

24x 3 C 38x 2  16x C 4
T12 x C 35x 12 C 210x 11 C 735x 10 C 2849x 9 C 10150x 8 C 45655x 7 C 94570x 6
14

C98455x 5  199381x 4  344400x 3 C 647395x 2 C 4094650x C 1010645

T11 x 14  5x 12  11x 10 C 25x 8 C 27x 6  23x 4  17x 2  1
T10 x 14 C 14x 8  84x 6 C 84x 4 C 21x 2  9
T9 x 14 C 7x 12  49x 10  245x 8 C 588x 6 C 294x 4  7
T8 x 14  x 12  3x 11 C 5x 10 C 5x 9  5x 8  9x 7 C x 6 C 14x 5  2x 4  7x 3 C x 2 C 1
T7 x 14 C 2
T6 x 14 C 13x 12 C 31x 10  9x 8  54x 6  3x 4 C 23x 2  1
T5 x 14  x 7 C 2
T4 x 14 C 7
T3 x 14 C 6x 12 C 7x 10 C x 8  3x 6 C x 4 C 3x 2 C 1
T2 x 14 C 8x 12 C 22x 10 C 8x 8  55x 6  48x 4 C 64x 2 C 71
T1 x 14 C 25x 12 C 214x 10 C 767x 8 C 1194x 6 C 686x 4 C 53x 2 C 1
References

Abhyankar, S.S. (1957): Coverings of algebraic curves. Amer. J. Math. 79, 825–856
Abhyankar, S.S. (1994): Nice equations for nice groups. Israel J. Math. 88, 1–23
Abhyankar, S.S. (1996a): Again nice equations for nice groups. Proc. Amer. Math.
Soc. 124, 2967–2976
Abhyankar, S.S. (1996b): More nice equations for nice groups. Proc. Amer. Math.
Soc. 124, 2977–2991
Abhyankar, S.S., Inglis, N.J. (2001): Galois groups of some vectorial polynomials.
Trans. Amer. Math. Soc. 353, 2941–2869
Abhyankar, S.S., Keskar, P.H. (2001): Descent principle in modular Galois theory.
Proc. Indian Acad. Sci. Math. 11, 139–149
Abhyankar, S.S., Loomis, P.A. (1998): Once more nice equations for nice groups.
Proc. Amer. Math. Soc. 126, 1885–1896
Abhyankar, S.S., Loomis, P.A. (1999): Twice more nice equations for nice groups.
Pp. 63–76 in: Applications of curves over finite fields (Seattle, WA, 1997), Con-
temp. Math., 245, Amer. Math. Soc., Providence, RI, 1999
Albert, M., Maier, A. (2011): Additive polynomials for finite groups of Lie type.
Israel J. Math. 186, 125–195
Artin, E. (1925): Theorie der Zöpfe. Abh. Math. Sem. Univ. Hamburg 4, 47–72
Artin, E. (1947): Theory of braids. Ann. of Math. 48, 101–126
Artin, E. (1967): Algebraic numbers and algebraic functions. Gordon and Breach,
New York
Aschbacher, M. (1986): Finite group theory. Cambridge University Press, Cam-
bridge
Aschbacher, M. et al. (eds.) (1984): Proceedings of the Rutgers group theory year,
1983–1984. Cambridge University Press, Cambridge
Auslander, M., Brumer, A. (1968): Brauer groups of discrete valuation rings. Indag.
Math. 30, 286–296
Barth, D., Wenz, A. (2016): Explicit Belyi maps over Q having almost simple prim-
itive monodromy groups. Preprint, arXiv:1703.02848

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 515

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3
516 References

Barth, D., Wenz, A. (2017): Belyi map for the sporadic group J1 . Preprint,
arXiv:1704.06419
Bayer-Fluckiger, E. (1994): Galois cohomology and the trace form. Jahresber.
Deutsch. Math.-Verein. 96, 35–55
Beckmann, S. (1989): Ramified primes in the field of moduli of branched coverings
of curves. J. Algebra 125, 236–255
Beckmann, S. (1991): On extensions of number fields obtained by specializing
branched coverings. J. reine angew. Math. 419, 27–53
Belyi, G.V. (1979): On Galois extensions of a maximal cyclotomic field. Izv. Akad.
Nauk SSSR Ser. Mat. 43, 267–276 (Russian) [English transl.: Math. USSR Izv.
14 (1979), 247–256]
Belyi, G.V. (1983): On extensions of the maximal cyclotomic field having a given
classical Galois group. J. reine angew. Math. 341, 147–156
Benson, D.J. (1991): Representations and cohomology I. Cambridge University
Press, Cambridge
Beynon, W.M., Spaltenstein, N. (1984): Green functions of finite Chevalley groups
of type En (n D 6; 7; 8). J. Algebra 88, 584–614
Birman, J.S. (1975): Braids, links and mapping class groups. Princeton University
Press, Princeton
Borel, A. (1991): Linear algebraic groups. Springer, New York
Borel, A. et al. (1970): Seminar on algebraic groups and related finite groups. LNM
131, Springer, Berlin
Bosch, S., Güntzer, U., Remmert, R. (1984): Non-archimedean analysis. Springer,
Berlin Heidelberg New York
Breuer, T., Malle, G., O’Brien, E. (2016): Reliability and reproducibility of Atlas
information. Submitted
Brieskorn, E., Saito, K. (1972): Artin-Gruppen und Coxeter-Gruppen. Invent. Math.
17, 245–271
Bruen, A.A., Jensen, C.U., Yui, N. (1986): Polynomials with Frobenius groups of
prime degree as Galois groups. J. Number Theory 24, 305–359
Butler, G., McKay, J. (1983): The transitive groups of degree up to eleven. Comm.
Algebra 11, 863–911
Cartan, H., Eilenberg, S. (1956): Homological algebra. Princeton University Press,
Princeton
Carter, R.W. (1985): Finite groups of Lie type: Conjugacy classes and complex char-
acters. John Wiley and Sons, Chichester
Carter, R.W. (1989): Simple groups of Lie type. John Wiley and Sons, Chichester
Chang, B., Ree, R. (1974): The characters of G2 .q/. Pp. 395–413 in: Symposia
Mathematica XIII, London
Chow, W.-L. (1948): On the algebraic braid group. Ann. of Math. 49, 654–658
Cohen, A.M. et al. (1992): The local maximal subgroups of exceptional groups of
Lie type, finite and algebraic. Proc. London Math. Soc. 64, 21–48
Cohen, D.B. (1974): The Hurwitz monodromy group. J. Algebra 32, 501–517
Colin, A. (1995): Formal computation of Galois groups with relative resolvents.
Pp. 169–182 in: Lecture Notes in Comput. Sci., 948, Springer, Berlin
References 517

Conway, J.H. et al. (1985): Atlas of finite groups. Clarendon Press, Oxford
Cooperstein, B.N. (1981): Maximal subgroups of G2 .2n /. J. Algebra 70, 23–36
Crespo, T. (1989): Explicit construction of AQn type fields. J. Algebra 127, 452–461
Dèbes, P., Fried, M.D. (1990): Arithmetic variation of fibers in families of covers I.
J. reine angew. Math. 404, 106–137
Deligne, P., Lusztig, G. (1976): Representations of reductive groups over finite
fields. Ann. of Math. 103, 103–161
Delone, B.N., Faddeev, D.K. (1944): Investigations in the geometry of the Galois
theory. Math. Sb. 15(57), 243–284 (Russian with English summary)
Demuškin, S.P., Šafarevič, I.R. (1959): The embedding problem for local fields. Izv.
Akad. Nauk. Ser. Mat. 23, 823–840 (Russian) [English transl.: Amer. Math. Soc.
Transl. II 27, 267–288 (1963)]
Dentzer, R. (1989): Projektive symplektische Gruppen PSp4 .p/ als Galoisgruppen
über Q.t/. Arch. Math. 53, 337–346
Dentzer, R. (1995a): Polynomials with cyclic Galois group. Comm. Algebra 23,
1593–1603
Dentzer, R. (1995b): On geometric embedding problems and semiabelian groups.
Manuscripta Math. 86, 199–216
Deriziotis, D.I. (1983): The centralizers of semisimple elements of the Chevalley
groups E7 and E8 . Tokyo J. Math. 6, 191–216
Deriziotis, D.I., Michler, G.O. (1987): Character table and blocks of finite simple
triality groups 3D4 .q/. Trans. Amer. Math. Soc. 303, 39–70
Derksen, H., Kemper, G. (2002): Computational invariant theory. Springer Verlag,
Berlin
Dettweiler, M. (2003): Middle convolution functor and Galois realizations. Pp. 143–
158 in: K. Hashimoto et al. Eds: Galois theory and modular forms. Kluwer
Dettweiler, M. (2004): Plane curve complements and curves on Hurwitz spaces. J.
reine angew. Math. 573, 19–43
Dettweiler, M., Reiter, S. (1999): On rigid tuples in linear groups. J. Algebra 222,
550–560
Dettweiler, M., Reiter, S. (2000): An algorithm of Katz and its applications to the
inverse Galois problem. J. Symb. Comput. 30, 761–798
Dew, E. (1992): Fields of moduli of arithmetic Galois groups. PhD-thesis, Univer-
sity of Pennsylvania
Digne, F., Michel, J. (1991): Representations of finite groups of Lie type. LMS
Student Texts 21, Cambridge University Press, Cambridge
Digne, F., Michel, J. (1994): Groupes réductifs non connexes. Ann. scient. Éc.
Norm. Sup. 27, 345–406
van den Dries, L., Ribenboim, P. (1979): Application de la théorie des modèles aux
groupes de Galois de corps de fonctions. C. R. Acad. Sci. Paris 288, A789–A792
Douady, A. (1964): Détermination d’un groupe de Galois. C. R. Acad. Sci. Paris
258, 5305–5308
Elkies, N.D. (1997): Linearized algebra and finite groups of Lie type I: Linear and
symmetric groups. Contemp. Math. 245, 77–108
Engler, A.J., Prestel, A. (2005): Valued fields. Springer Verlag, Berlin
518 References

Enomoto, H. (1976): The characters of the finite Chevalley groups G2 .q/, q D 3f .

Japan. J. Math. 2, 191–248
Enomoto, H., Yamada, H. (1986): The characters of G2 .2n /. Japan. J. Math. 12,
325–377
Faddeev, D.K. (1951): Simple algebras over a field of algebraic functions of one
variable. Trudy Mat. Inst. Steklov 38, 190–243 (Russian) [English transl.: Amer.
Math. Soc. Transl. II 3, 15–38 (1956)]
Fadell, E. (1962): Homotopy groups of configuration spaces and the string problem
of Dirac. Duke Math. J. 29, 231–242
Fadell, E., Van Buskirk, J. (1962): The braid groups of E 2 and S 2 . Duke Math. J.
29, 243–257
Feit, W. (1984): Rigidity of Aut.PSL2 .p 2 //, p ˙2 .mod 5/, p ¤ 2. Pp. 351–356
in: M. Aschbacher et al. (eds.) (1984)
Feit, W. (1989): Some finite groups with nontrivial centers which are Galois groups.
Pp. 87–109 in: Group Theory, Proceedings of the 1987 Singapore Conference. W.
de Gruyter, Berlin-New York
Feit, W., Fong, P. (1984): Rational rigidity of G2 .p/ for any prime p > 5. Pp. 323–
326 in: M. Aschbacher et al. (eds.) (1984)
Fleischmann, P. , Janiszczak, I. (1993): The semisimple conjugacy classes of finite
groups of Lie type E6 and E7 . Comm. Algebra 21, 93–161
Folkers, M. (1995): Lineare Gruppen ungerader Dimension als Galoisgruppen über
Q. IWR-Preprint 95-41, Universität Heidelberg
Forster, O. (1981): Lectures on Riemann surfaces. Springer, New York
Franz, W. (1931): Untersuchungen zum Hilbertschen Irreduzibilitätssatz. Math. Z.
33, 275–293
Fresnel, J., van der Put, M. (1981): Géométrie analytique rigide et applications.
Birkhäuser, Basel
Fresnel, J., van der Put, M. (2004): Rigid analytic geometry and its applications.
Birkhäuser, Boston
Fried, M.D. (1977): Fields of definition of function fields and Hurwitz families —
Groups as Galois groups. Comm. Algebra 5, 17–82
Fried, M.D. (1984): On reduction of the inverse Galois problem to simple groups.
Pp. 289–301 in: M. Aschbacher et al. (eds.) (1984)
Fried, M.D., Biggers, R. (1982): Moduli spaces of covers and the Hurwitz mon-
odromy group. J. reine angew. Math. 335, 87–121
Fried, M.D., Dèbes, P. (1990): Rigidity and real residue class fields. Acta Arith-
metica 56, 291–323
Fried, M.D., Jarden, M. (1986): Field arithmetic. Springer, Berlin
Fried, M.D., Völklein, H. (1991): The inverse Galois problem and rational points on
moduli spaces. Math. Ann. 290, 771–800
Fried, M.D., Völklein, H. (1992): The embedding problem over a Hilbertian PAC-
field. Ann. of Math. 135, 469–481
Fried, M.D. et al. (eds.) (1995): Recent developments in the inverse Galois problem.
Contemporary Math. vol. 186, Amer. Math. Soc., Providence
References 519

Fröhlich, A. (1985): Orthogonal representations of Galois groups, Stiefel-Whitney

classes and Hasse-Witt invariants. J. reine angew. Math. 360, 84–123
Frohardt, D., Guralnick, R., Hoffman, C., Magaard, K. (2016): Exceptional low
genus actions of finite groups. Preprint
Frohardt, D., Magaard, K. (2001): Composition factors of monodromy groups. Ann.
of Math. 154, 327–345
Garcia Lopez, F. (2010): An algorithm for computing Galois groups of additive
polynomials. Preprint
Geck, M. (2003): An introduction to algebraic geometry and algebraic groups.
Oxford University Press, Oxford
Geyer, W.-D., Jarden, M. (1998): Bounded realization of `-groups over global fields.
Nagoya Math. J. 150, 13–62
Gille, P. (2000): Le groupe fondamental sauvage d’une courbe affine in character-
istic p. Pp. 217–231 in: Bost, J.-L. et al. eds.: Courbes semi-stables et groupe
fondamental en géometrie algébrique. Birkhäuser, Basel
Gillette, R., Van Buskirk, J. (1968): The word problem and consequences for the
braid groups and mapping class groups of the 2-sphere. Trans. Amer. Math. Soc.
131, 277–296
Goss, D. (1996): Basic structures of function field arithmetic. Springer, Berlin
Granboulan, L. (1996): Construction d’une extension régulière de Q.T / de groupe
de Galois M24 . Experimental Math. 5, 3–14
Grothendieck, A. (1971): Revêtements étales et groupe fondamental. Springer,
Berlin
Guralnick, R.M., Lübeck, F., Yu, J. (2016): Rational rigidity for F4 .p/. Adv. Math.
302, 48–58
Guralnick, R.M., Malle, G. (2012): Simple groups admit Beauville structures. J.
London Math. Soc. 85, 694–721
Guralnick, R.M., Malle, G. (2014): Rational rigidity for E8 .p/. Compos. Math. 150,
1679–1702
Guralnick, R.M., Thompson, J.G. (1990): Finite groups of genus zero. J. Algebra
131, 303–341
Häfner, F. (1992): Einige orthogonale und symplektische Gruppen als Galoisgrup-
pen über Q. Math. Ann. 292, 587–618
Hansen, V.L. (1989): Braids and coverings. Cambridge University Press, Cambridge
Haraoka, Y. (1994): Finite monodromy of Pochhammer equation. Ann. Inst. Fourier
44, 767–810
Haran, D., Völklein, H. (1996): Galois groups over complete valued fields. Israel J.
Math. 93, 9–27
Harbater, D. (1984): Mock covers and Galois extensions. J. Algebra 91, 281–293
Harbater, D. (1987): Galois coverings of the arithmetic line. In: Chudnovsky, D.V.
et al. (eds.): Number Theory, New York 1984–1985. Springer, Berlin
Harbater, D. (1994a): Abhyankar’s conjecture on Galois groups over curves. Invent.
Math. 117, 1–25
520 References

Harbater, D. (1994b): Galois groups with prescribed ramification. Pp. 35–60 in: N.
Childress and J. Jones (eds.): Arithmetic Geometry. Contemporary Math. vol.
174, Amer. Math. Soc., Providence
Harbater, D. (1995a): Fundamental groups and embedding problems in characteris-
tic p. Pp. 353–369 in: M. Fried et al. (eds.) (1995)
Harbater, D. (1995b): Fundamental groups of curves in characteristic p. Pp. 656–
666 in: Proc. Int. Congress of Math., Zürich 1994 , Birkhäuser
Hartshorne, R. (1977): Algebraic geometry. Springer, Berlin Heidelberg New York
Hasse, H. (1948): Existenz und Mannigfaltigkeit abelscher Algebren mit vorgege-
bener Galoisgruppe über einem Teilkörper des Grundkörpers I. Math. Nachr. 1,
40–61
Hasse, H. (1980): Number theory. Springer, Berlin
Hecke, E. (1935): Die eindeutige Bestimmung der Modulfunktionen q-ter Stufe
durch algebraische Eigenschaften. Math. Ann. 111, 293–301
Higman, D., McLaughlin, J. (1965): Rank 3 subgroups of finite symplectic and uni-
tary groups. J. reine angew. Math. 218, 174–189
Hilbert, D. (1892): Über die Irreducibilität ganzer rationaler Functionen mit ganz-
zahligen Coeffizienten. J. reine angew. Math. 110, 104–129
Hilton, P.J., Stammbach, U. (1971): A course in homological algebra. Springer, New
York
Hoechsmann, K. (1968): Zum Einbettungsproblem. J. reine angew. Math. 229,
81–106
Hoyden-Siedersleben, G. (1985): Realisierung der Jankogruppen J1 und J2 als
Galoisgruppen über Q. J. Algebra 97, 14–22
Hoyden-Siedersleben, G., Matzat, B.H. (1986): Realisierung sporadischer einfacher
Gruppen als Galoisgruppen über Kreisteilungskörpern. J. Algebra 101, 273–285
Hunt, D.C. (1986): Rational rigidity and the sporadic groups. J. Algebra 99,
577–592
Huppert, B. (1967): Endliche Gruppen I. Springer, Berlin
Huppert, B., Blackburn, N. (1982): Finite groups II. Grundlehren 242, Springer,
Berlin
Hurwitz, A. (1891): Über Riemann’sche Flächen mit gegebenen Verzweigungs-
punkten. Math. Ann. 39, 1–61
Ihara, Y. (1991): Braids, Galois groups, and some arithmetic functions. Pp. 99–120
in: Proc. Int. Congress of Math., Kyoto 1990, Springer
Ihara, Y. et al. (eds.) (1989): Galois groups over Q. Springer, New York
Iitaka, S. (1982): Algebraic geometry. Springer Verlag, Berlin
Ikeda, M. (1960): Zur Existenz eigentlicher galoisscher Körper beim Einbettungs-
problem für galoissche Algebren. Abh. Math. Sem. Univ. Hamburg 24, 126–131
Inaba, E. (1944): Über den Hilbertschen Irreduzibilitätssatz. Japan. J. Math. 19,
1–25
Ishkhanov, V.V., Lure, B.B., Faddeev, D.K. (1997): The embedding problem in
Galois theory. Amer. Math. Soc., Providence
Iwasawa, K. (1953): On solvable extensions of algebraic number fields. Ann. of
Math. 58, 548–572
References 521

Jacobson, N. (1980): Basic algebra II. W. H. Freeman, New York

Jensen, C., Ledet, A., Yui, N. (2002): Generic polynomials. Constructive aspects of
the inverse Galois problem. Cambridge University Press, Cambridge
Kantor, W.M. (1979): Subgroups of classical groups generated by long root ele-
ments. Trans. Amer. Math. Soc. 248, 347–379
Katz, N. (1996): Rigid local systems. Annals of Math. Studies 139, Princeton Uni-
versity Press, Princeton
Kemper, G., Lübeck, F., Magaard, K. (2001): Matrix generators for the Ree groups
2
G2 .q/. Comm. Algebra 29, 407–415
Kleidman, P.B. (1987): The maximal subgroups of the finite 8-dimensional orthog-
onal groups PC 8 .q/ and of their automorphism groups. J. Algebra 110, 173–242
Kleidman, P.B. (1988a): The maximal subgroups of the Steinberg triality groups
3
D4 .q/ and of their automorphism groups. J. Algebra 115, 182–199
Kleidman, P.B. (1988b): The maximal subgroups of the Chevalley groups G2 .q/
with q odd, the Ree groups 2G2 .q/ and of their automorphism groups. J. Algebra
117, 30–71
Kleidman, P.B., Liebeck, M. (1990): The subgroup structure of the finite classical
groups. LMS lecture notes 129, Cambridge University Press, Cambridge
Klein, F. (1884): Vorlesungen über das Ikosaeder. Teubner, Leipzig
Klein, F., Fricke, R. (1890): Vorlesungen über die Theorie der elliptischen Modul-
funktionen I. Teubner, Leipzig
Klüners, J., Malle, G. (2000): Explicit Galois realization of transitive groups of
degree up to 15. J. Symbolic Comput. 30, 675–716
Klüners, J., Malle, G. (2002): A Database for Number Fields. http://galoisdb.math.
upb.de/
Kochendörffer, R. (1953): Zwei Reduktionssätze zum Einbettungsproblem für abel-
sche Algebren. Math. Nachr. 10, 75–84
König, J. (2015): Computation of Hurwitz spaces and new explicit polynomials for
almost simple Galois groups. Math. Comp., to appear
König, J. (2016): On rational functions with monodromy group M11 . J. Symbolic
Comput. 79, 372–383
Köpf, U. (1974): Über eigentliche Familien algebraischer Varietäten über affinoiden
Räumen. Schriftenreihe Math. Inst. Univ. Münster, 2. Serie, Heft 7
Kreuzer, M., Robbiano, L. (2000): Computational commutative algebra I. Springer
Verlag, Berlin
Krull, W., Neukirch, J. (1971): Die Struktur der absoluten Galoisgruppe über dem
Körper IR.t/. Math. Ann. 193, 197–209
Kucera, J. (1994): Über die Brauergruppe von Laurentreihen- und rationalen Funk-
tionenkörpern und deren Dualität mit K-Gruppen. Dissertation, Universität Hei-
delberg
Kuyk, W. (1970): Extensions de corps Hilbertiens. J. Algebra 14, 112–124
Landazuri, V., Seitz, G.M. (1974): On the minimal degrees of projective representa-
tions of the finite Chevalley groups. J. Algebra 32, 418–443
Lang, S. (1982): Introduction to algebraic and abelian functions. Springer, New
York
522 References

Ledet, A. (2000): On a theorem of Serre. Proc. Amer. Math. Soc. 128, 27–29
Ledet, A. (2005): Brauer type embedding problems. Amer. Math. Soc., Providence
Liebeck, M.W., Seitz, G.M. (1999): On finite subgroups of exceptional algebraic
groups. J. Reine Angew. Math. 515, 25–72
Liu, Q. (1995): Tout groupe fini est un groupe de Galois sur Qp .t/, d’après Harbater.
Pp. 261–265 in: M. Fried et al (eds.) (1995)
Lübeck, F., Malle, G. (1998): .2; 3/-generation of exceptional groups. J. London
Math. Soc. 59, 109–122
Lusztig, G. (1977): Representations of finite Chevalley groups. CBMS Regional
Conference Series in Mathematics 39
Lusztig, G. (1980): On the unipotent characters of the exceptional groups over finite
fields. Invent. Math. 60, 173–192
Lusztig, G. (1984): Characters of reductive groups over a finite field. Annals of
Math. Studies 107, Princeton University Press, Princeton
Lyndon, R.C., Schupp, P.E. (1977): Combinatorial group theory. Springer, Berlin
Madan, M., Rzedowski-Calderon, M., Villa-Salvador, G. (1996): Galois extensions
with bounded ramification in characteristic p. Manuscripta Math. 90, 121–135
Malle, G. (1986): Exzeptionelle Gruppen vom Lie-Typ als Galoisgruppen. Disser-
tation, Universität Karlsruhe
Malle, G. (1987): Polynomials for primitive nonsolvable permutation groups of
degree d 15. J. Symb. Comp. 4, 83–92
Malle, G. (1988a): Polynomials with Galois groups Aut.M22 /, M22 , and
PSL3 .IF4 /:2 over Q. Math. Comp. 51, 761–768
Malle, G. (1988b): Exceptional groups of Lie type as Galois groups. J. reine angew.
Math. 392, 70–109
Malle, G. (1991): Genus zero translates of three point ramified Galois extensions.
Manuscripta Math. 71, 97–111
Malle, G. (1992): Disconnected groups of Lie type as Galois groups. J. reine angew.
Math. 429, 161–182
Malle, G. (1993a): Polynome mit Galoisgruppen PGL2 .p/ und PSL2 .p/ über Q.t/.
Comm. Algebra 21, 511–526
Malle, G. (1993b): Generalized Deligne-Lusztig characters. J. Algebra 159, 64–97
Malle, G. (1993c): Green functions for groups of types E6 and F4 in characteristic 2.
Comm. Algebra 21, 747–798
Malle, G. (1996): GAR-Realisierungen klassischer Gruppen. Math. Ann. 304, 581–
612
Malle, G. (2003): Explicit realization of the Dickson groups G2 .q/ as Galois groups.
Pacific J. Math. 212, 157–167
Malle, G., Matzat, B. H. (1985): Realisierung von Gruppen PSL2 .IFp / als Galois-
gruppen über Q. Math. Ann. 272, 549–565
Malle, G., Matzat, B. H. (1999): Inverse Galois theory. Springer Verlag, Berlin–
Heidelberg
Malle, G., Saxl, J., Weigel, T. (1994): Generation of classical groups. Geom. Dedi-
cata 49, 85–116
References 523

Malle, G., Sonn, J. (1996): Covering groups of almost simple groups as Galois
groups over Qab .t/. Israel J. Math. 96, 431–444.
Malle, G., Testerman, D. (2011): Linear algebraic groups and finite groups of Lie
type. Cambridge University Press, Cambridge
Matsumura, H. (1980): Commutative algebra. Benjamin/Cummings, Reading
p B.H. (1979): Konstruktion von Zahlkörpern mit der Galoisgruppe M11 über
Matzat,
Q. 11/. Manuscripta Math. 27, 103–111
Matzat, B.H. (1984): Konstruktion von Zahl- und Funktionenkörpern mit vor-
gegebener Galoisgruppe. J. reine angew. Math. 349, 179–220
Matzat, B.H. (1985a): Zwei Aspekte konstruktiver Galoistheorie. J. Algebra 96,
499–531
Matzat, B.H. (1985b): Zum Einbettungsproblem der algebraischen Zahlentheorie
mit nicht abelschem Kern. Invent. Math. 80, 365–374
Matzat, B.H. (1986): Topologische Automorphismen in der konstruktiven Galois-
theorie. J. reine angew. Math. 371, 16–45
Matzat, B.H. (1987): Konstruktive Galoistheorie. LNM 1284, Springer, Berlin
Matzat, B.H. (1989): Rationality criteria for Galois extensions. Pp. 361–383 in: Y.
Ihara et al. (eds.) (1989)
Matzat, B.H. (1991a): Zöpfe und Galoissche Gruppen. J. reine angew. Math. 420,
99–159
Matzat, B.H. (1991b): Frattini-Einbettungsprobleme über Hilbertkörpern. Manu-
scripta Math. 70, 429–439
Matzat, B.H. (1991c): Der Kenntnisstand in der konstruktiven Galoisschen Theorie.
Pp. 65–98 in G.O. Michler and C.M. Ringel (eds.) (1991)
Matzat, B.H. (1992): Zopfgruppen und Einbettungsprobleme. Manuscripta Math.
74, 217–227
Matzat, B.H. (1993): Braids and decomposition groups. Pp. 179–189 in: S. David
(ed.): Séminaire de théorie des nombres, Paris 1991–1992. Birkhäuser, Boston
Matzat, B.H. (1995): Parametric solutions of embedding problems. Pp. 33–50 in:
M. Fried et al. (eds.) (1995)
Matzat, B.H. (2003): Frobenius modules and Galois groups. Pp. 233-268 in K.
Hashimoto, K. Miyake, H. Nakamura et al. Eds.: Galois theory and modular
forms. Kluwer Acad. Publ., Boston, MA
Matzat, B.H., Zeh-Marschke, A. (1986): Realisierung der Mathieugruppen M11 und
M12 als Galoisgruppen über Q. J. Number Theory 23, 195–202
McLaughlin, J. (1969): Some subgroups of SLn .IFp /. Ill. J. Math. 13, 108–115
Melnicov, O.V. (1980): Projective limits of free profinite groups. Dokl. Akad. Nauk
SSSR 24, 968–970
Mestre, J.-F. (1990): Extensions régulières de Q.T / de groupe de Galois AQn . J.
Algebra 131, 483–495
Mestre, J.-F. (1994): Construction d’extensions régulières de Q.t/ à groupes de
Galois SL2 .F7 / et M Q 12 . C. R. Acad. Sci. 319, 781–782
Mestre, J.-F. (1998): Extensions Q-régulières de Q.t/ de groups de Galois 6:A6 et
6:A7 . Israel J. Math. 107, 333–341
524 References

Michler, G.O., Ringel, C.M. (eds.) (1991): Representation theory of finite groups
and finite-dimensional algebras. Birkhäuser, Basel
Mizuno, K. (1977): The conjugate classes of Chevalley groups of type E6 . J. Fac.
Sci. Univ. Tokyo 24, 525–563
Mizuno, K. (1980): The conjugate classes of unipotent elements of the Chevalley
groups E7 and E8 . Tokyo J. Math. 3, 391–461
Monien, H. (2017): The sporadic group J2 , Hauptmodul and Belyi map. Preprint,
arXiv:1703.05200
Müller, P. (2012): A one-parameter family of polynomials with Galois group M24
over Q.t/. Preprint, arXiv:1204.1328
Nagata, M. (1977): Field theory. Marcel Dekker, New York
Nakamura, H. (1997): Galois rigidity of profinite fundamental groups. Sugaku
Expositiones 10, 195–215
Narkiewicz, W. (1990): Elementary and analytic theory of algebraic numbers.
Springer, Berlin
Neukirch, J., Schmidt, A. and Wingberg, K. (2000): Cohomology of Number Fields.
Springer-Verlag, Berlin.
Nobusawa, N. (1961): On the embedding problem of fields and Galois algebras.
Abh. Math. Sem. Univ. Hamburg 26, 89–92
Noether, E. (1918): Gleichungen mit vorgeschriebener Gruppe. Math. Ann. 78,
221-229
Nori, M.V. (1994): Unramified coverings of the affine line in positive characteris-
tic. Pp. 209–212 in: Bajaj, C.L., ed.: Algebraic geometry and its applications.
Springer, New York
Norton, S.P., Wilson, R.A. (1989): The maximal subgroups of F4 .2n / and its auto-
morphism group. Comm. Algebra 17, 2809–2824
Pahlings, H. (1988): Some sporadic groups as Galois groups. Rend. Sem. Math.
Univ. Padova 79, 97–107
Pahlings, H. (1989): Some sporadic groups as Galois groups II. Rend. Sem. Math.
Univ. Padova 82, 163–171 [correction: ibid. 85 (1991), 309–310]
Pochhammer, L. (1870): Über die hypergeometrischen Funktionen n-ter Ordnung.
J. reine angew. Math. 71, 312–352
Pop, F. (1994): 12 Riemann existence theorem with Galois action. Pp. 193–218 in:
G. Frey and J. Ritter (eds.): Algebra and number theory. W. de Gruyter, Berlin
Pop, F. (1995): Étale Galois covers of affine smooth curves. Invent. Math. 120,
555–578
Pop, F. (1996): Embedding problems over large fields. Ann. of Math. 144, 1–34
Popp, H. (1970): Fundamentalgruppen algebraischer Mannigfaltigkeiten. Springer,
Berlin
Porsch, U. (1993): Greenfunktionen der endlichen Gruppen E6 .q/, q D 3n . Diplom-
arbeit, Universität Heidelberg
Przywara, B. (1991): Zopfbahnen und Galoisgruppen. IWR-Preprint 91-01, Univer-
sität Heidelberg
van der Put, M., Singer, M.F. (1997): Galois theory of difference equations. Springer
Verlag, Berlin
References 525

Raynaud, M. (1994): Revêtements de la droite affine en caractéristique p > 0 et

conjecture d’Abhyankar. Invent. Math. 116, 425–462
Reichardt, H. (1937): Konstruktion von Zahlkörpern mit gegebener Galoisgruppe
von Primzahlpotenzordnung. J. reine angew. Math. 177, 1–5
Reiter, S. (1999): Galoisrealisierungen klassischer Gruppen. J. reine angew. Math.
511, 193–236
Ribes, L. (1970): Introduction to profinite groups and Galois cohomology. Queen’s
University, Kingston
Rzedowski-Calderon, M. (1989): Construction of global function fields with nilpo-
tent automorphism groups. Bol. Soc. Mat. Mexicana 34, 1–10
Šafarevič, I.R. (1947): On p-extensions. Math. Sb. Nov. Ser. 20 (62), 351–363 (Rus-
sian) [English transl.: Amer. Math. Soc. Transl. II 4, 59–72 (1956)]
Šafarevič, I.R. (1954a): On the construction of fields with a given Galois group of
order `a . Izv. Akad. Nauk. SSSR 18, 216–296 (Russian) [English transl.: Amer.
Math. Soc. Transl. II 4, 107–142 (1956)]
Šafarevič, I.R. (1954b): On an existence theorem in the theory of algebraic numbers.
Izv. Akad. Nauk. SSSR 18, 327–334 (Russian) [English transl.: Amer. Math. Soc.
Transl. II 4, 143–150 (1956)]
Šafarevič, I.R. (1954c): On the problem of imbedding fields. Izv. Akad. Nauk. SSSR
18, 389–418 (Russian) [English transl.: Amer. Math. Soc. Transl. II 4, 151–183
(1956)]
Šafarevič, I.R. (1954d): Construction of fields of algebraic numbers with given
solvable Galois group. Izv. Akad. Nauk. SSSR 18, 525–578 (Russian) [English
transl.: Amer. Math. Soc. Transl. II 4, 185–237 (1956)]
Šafarevič, I.R. (1958): The embedding problem for splitting extensions. Dokl. Akad.
Nauk SSSR 120, 1217–1219 (Russian)
Šafarevič, I.R. (1989): Factors of descending central series. Math. Notes 45, 262–
264
Šafarevič, I.R. (1994): Basic algebraic geometry I. Springer Verlag
Saı̈di, M. (2000): Abhyankar’s conjecture II: the use of semi-stable curves. Pp. 249–
265 in: Bost, J.-L. et al. eds.: Courbes semi-stables et groupe fondamental en
géometrie algébrique. Birkhäuser, Basel
Saltman, D.J. (1982): Generic Galois extensions and problems in field theory. Adv.
Math. 43, 250–283
Saltman, D.J. (1984): Noether’s problem over algebraically closed fields. Invent.
Math. 77, 71–84
Scharlau, W. (1969): Über die Brauer-Gruppe eines algebraischen Funktionenkör-
pers einer Variablen. J. reine angew. Math. 239/240, 1–6
Scharlau, W. (1985): Quadratic and Hermitean forms. Springer, Berlin
Schneps, L. (1992): Explicit construction of extensions of Q.t/ of Galois group AQn
for n odd. J. Algebra 146, 117–123
Scholz, A. (1929): Über die Bildung algebraischer Zahlkörper mit auflösbarer
galoisscher Gruppe. Math. Z. 30, 332–356
Scholz, A. (1937): Konstruktion algebraischer Zahlkörper mit beliebiger Gruppe
von Primzahlpotenzordnung I. Math. Z. 42, 161-188
526 References

Scott, L. (1977): Matrices and cohomology. Ann. of Math. 105, 473–492

Seidelmann, F. (1918): Die Gesamtheit der kubischen und biquadratischen Gle-
ichungen mit Affekt bei beliebigem Rationalitätsbereich. Math.Ann 78, 230–233
Seifert, H., Threllfall, W. (1934): Lehrbuch der Topologie. Teubner, Leipzig
Serre, J.-P. (1956): Géométrie algébrique et géométrie analytique. Ann. Inst. Fourier
6, 1–42
Serre, J.-P. (1959): Groupes algébriques et corps de classes. Hermann, Paris
Serre, J.-P. (1964): Cohomologie galoisienne. Springer, Berlin
Serre, J.-P. (1979): Local fields. Springer, New York
Serre, J.-P. (1984): L’invariant de Witt de la forme Tr.x 2 /. Comment. Math. Hel-
vetici 59, 651–679
Serre, J.-P. (1988): Groupes de Galois sur Q. Astérisque 161–162, 73–85
Serre, J.-P. (1990): Construction de revêtements étales de la droite affine en car-
actéristique p. C. R. Acad. Sci. Paris 311, 341–346
Serre, J.-P. (1992): Topics in Galois theory. Jones and Bartlett, Boston
Shabat, G.B., Voevodsky, V.A. (1990): Drawing curves over number fields. Pp.
199–227 in: P. Cartier et al. (eds.): The Grothendieck Festschrift III. Birkhäuser,
Boston
Shatz, S.S. (1972): Profinite groups, arithmetic and geometry. Princeton University
Press, Princeton
Shih, K.-y. (1974): On the construction of Galois extensions of function fields and
number fields. Math. Ann. 207, 99–120
Shih, K.-y. (1978): p-division points on certain elliptic curves. Compositio Math.
36, 113–129
Shiina, Y. (2003a): Braid orbits related to PSL2 .p 2 / and some simple groups.
Tohoku Math. J. 55, 271–282
Shiina, Y. (2003b): Regular Galois realizations of PSL2 .p 2 / over Q.T /. Pp. 125–
142 in: K. Hashimoto et al. Eds: Galois theory and modular forms. Kluwer
Shinoda, K. (1974): The conjugacy classes of Chevalley groups of type .F4 / over
finite fields of characteristic 2. J. Fac. Sci. Univ. Tokyo 21, 133–159
Shoji, T. (1974): The conjugacy classes of Chevalley groups of type .F4 / over finite
fields of characteristic p ¤ 2. J. Fac. Sci. Univ. Tokyo 21, 1–17
Shoji, T. (1982): On the Green polynomials of a Chevalley group of type F4 . Comm.
Algebra 10, 505–543
Silverman, J.H. (1986): The arithmetic of elliptic curves. Springer, New York
Simpson, W., Frame, J. (1973): The character tables for SL3 .q/, SU3 .q/, PSL3 .q/,
U3 .q/. Can. J. Math. 25, 486–494
Smith, G.W. (1991): Generic cyclic polynomials of odd degree. Comm. Algebra 19,
3367–3391
Smith, G.W. (2000): Some polynomials over Q.t/ and their Galois groups. Math.
Comp. 69, 775–796
Smith, L. (1995): Polynomial invariants of finite groups. A.K. Peters, Wellesley
Sonn, J. (1990): On Brauer groups and embedding problems over function fields. J.
Algebra 131, 631–640
References 527

Sonn, J. (1991): Central extensions of Sn as Galois groups of regular extensions of

Q.T /. J. Algebra 140, 355–359
Sonn, J. (1994a): Brauer groups, embedding problems, and nilpotent groups as
Galois groups. Israel J. Math. 85, 391–405
Sonn, J. (1994b): Rigidity and embedding problems over Qab .t/. J. Number Theory
47, 398–404
Spaltenstein, N. (1982): Caractères unipotents de 3D4 .IFq /. Comment. Math. Hel-
vetici 57, 676–691
Speiser, A. (1919): Zahlentheoretische Sätze aus der Gruppentheorie. Math. Z. 5,
1–6
Springer, T.A. (1998): Linear algebraic groups. Birkhäuser, Basel
Steinberg, R. (1965): Regular elements of semi-simple algebraic groups. Publ.
Math. IHES 25, 49–80.
Steinberg, R. (1967): Lectures on Chevalley groups. Yale University
Steinberg, R. (1968): Endomorphisms of linear algebraic groups. Mem. Amer.
Math. Soc. 80, Amer. Math. Soc., Providence
Stichel, D. (2014): Finite groups of Lie type as Galois groups over IFq .t/. J. Com-
mut. Algebra 6, 587–603
Stöcker, R., Zieschang, H. (1988): Algebraische Topologie. Teubner, Stuttgart
Stoll, M. (1995): Construction of semiabelian Galois extensions. Glasgow Math. J.
37, 99–104
Strammbach, K., Völklein, H. (1999): On linearly rigid tuples. J. reine angew. Math.
510, 57–62
Stroth, G. (1998): Algebra. de Gruyter Verlag, Berlin
Suzuki, M. (1962): On a class of doubly transitive groups. Ann. of Math. 75,
105–145
Suzuki, M. (1982): Group theory. Springer, Berlin
Swallow, J.R. (1994): On constructing fields corresponding to the AQn ’s of Mestre
for odd n. Proc. Amer. Math. Soc. 122, 85–90
Swan, R.G. (1969): Invariant rational functions and a problem of Steenrod. Invent.
Math. 7, 148–158
Tate, J. (1962): Duality theorems in Galois cohomology. Pp. 288–295 in: Proc. Int.
Congress of Math., Stockholm
Tate, J. (1966): The cohomology groups of tori in finite Galois extensions of number
fields. Nagoya Math. J. 27, 709–719
Thompson, J.G. (1973): Isomorphisms induced by automorphisms. J. Austral. Math.
Soc. 16, 16–17
Thompson, J.G. (1984a): Some finite groups which appear as Gal.L=K/ where K
Q.&n /. J. Algebra 89, 437–499
Thompson, J.G. (1984b): Primitive roots and rigidity. Pp. 327–350 in: M.
Aschbacher et al. (eds.) (1984)
Thompson, J.G. (1986): Regular Galois extensions of Q.x/, Pp. 210–220 in: Tuan
Hsio-Fu (ed.): Group theory, Beijing 1984. Springer, Berlin
Thompson, J.G., Völklein, H. (1998): Symplectic groups as Galois groups. J. Group
Theory 1, 1–58
528 References

Tschebotaröw, N.G., Schwerdtfeger, H. (1950): Grundzüge der Galoisschen Theo-

rie. Noordhoff, Groningen
Uchida, K. (1980): Separably Hilbertian fields. Kodai Math. J. 3, 83–95
Vila, N. (1985): On central extensions of An as Galois groups over Q. Arch. Math.
44, 424–437
Völklein, H. (1992a): Central extensions as Galois groups. J. Algebra 146, 144–152
Völklein, H. (1992b): GLn .q/ as Galois group over the rationals. Math. Ann. 293,
163–176
Völklein, H. (1993): Braid group action via GLn .q/ and Un .q/ and Galois realiza-
tions. Israel J. Math. 82, 405–427
Völklein, H. (1994): Braid group action, embedding problems and the groups
PGLn .q/ and PUn .q 2 /. Forum Math. 6, 513–535
Völklein, H. (1996): Groups as Galois groups. Cambridge University Press, Cam-
bridge
Völklein, H. (1998): Rigid generators of classical groups. Math. Ann. 311, 421–438
Wagner, A. (1978): Collineation groups generated by homologies of order greater
than 2. Geom. Dedicata 7, 387–398
Wagner, A. (1980): Determination of the finite primitive reflection groups over an
arbitrary field of characteristic not 2, I–III. Geom. Dedicata 9, 239–253; 10, 183–
189; 10, 475–523
Walter, J.H. (1984): Classical groups as Galois groups. Pp. 357–383 in: M. Asch-
bacher et al. (eds.) (1984)
Ward, H.N. (1966): On Ree’s series of simple groups. Trans. Amer. Math. Soc. 121,
62–89
Washington, L. (1982): Introduction to cyclotomic fields. Springer, New York
Wehrfritz, B.A.F. (1973): Infinite linear groups. Springer Verlag, Berlin
Weigel, T. (1992): Generators of exceptional groups of Lie type. Geom. Dedicata
41, 63–87
Weil, A. (1956): The field of definition of a variety. Amer. J. Math. 78, 509–524
Weil, A. (1974): Basic number theory. Springer, New York
Weissauer, R. (1982): Der Hilbertsche Irreduzibilitätssatz. J. reine angew. Math.
334, 203–220
Wilson, R.A. (2017): Maximal subgroups of sporadic groups. To appear in: Finite
Simple Groups: Thirty Years of the Atlas and Beyond, AMS Contemporary Math-
ematics Series, 2017.
Wolf, P. (1956): Algebraische Theorie der galoisschen Algebren. VEB Deutscher
Verlag der Wissenschaften, Berlin
Yakovlev, A.V. (1964): The embedding problems for fields. Izv. Akad. Nauk. SSSR
28, 645–660 (Russian)
Yakovlev, A.V. (1967): The embedding problem for number fields. Izv. Akad. Nauk.
SSSR 31, 211–224 (Russian) [English transl.: Math. USSR Izv. 1, 195–208
(1967)]
Zalesskiı̌, A.E., Serežkin, V.N. (1980): Finite linear groups generated by reflec-
tions. Izv. Akad. Nauk SSSR Ser. Math. 44, 1279–1307 (Russian) [English transl.:
Math. USSR Izv. 17 (1981), 477–503]
References 529

Zassenhaus, H. (1958): Gruppentheorie. Vandenhoeck & Ruprecht, Göttingen

Zywina, D (2015): The inverse Galois problem for PSL2 .IFp /. Duke Math. J. 164,
2253–2292.
Index

-equivalent matrix, 386 conformal orthogonal group, 111

conformal symplectic group, 108
accompanying Brauer embedding problem, connected rigid analytic space, 452
350 convolution, 251
accompanying embedding problem, 350 coroot, 93
admissible covering, 451 cyclic F-module, 385
admissible subset, 451 cyclotomic character, 14
affinoid analytic space, 451 cyclotomic polynomial, 117
algebraic fundamental group, 4, 187
almost character, 127 Dedekind criterion, 72
arithmetic fundamental group, 10, 197 Dickson algebra, 396
(full) Artin braid group, 179 Dickson invariants, 395
associated F-module, 390 Dickson polynomial, 395
AV -rigid, 64 disclosed function field of one variable, 14
AV -symmetric, 64 duality theorem of Tate, 354
AV -symmetrized irrationality degree, 64 dualizable F-module, 385

effective G-module, 392

basic rigidity theorem, 30
embedding problem, 288
Belyi triple, 102
existentially closed, 474
braid cycle theorem, 246 extension theorem, 67
braid orbit theorem, 215
braid relations, 181 F-field, 385
Brauer embedding problem, 339 F-module, 385
field of definition, 19
central embedding problem, 288 field of definition with group, 19
characteristic polynomial of an F-module, 389 field of invariants, 394
clean Belyi function, 17 field of moduli, 30
closed ultrametric disc, 462 field restriction of algebraic groups, 441
coherent sheaf, 454 finite embedding problem, 288
cohomologically trivial in dimension i , 361 finite morphism, 458
companion matrix, 390 first embedding obstruction, 364
comparison theorem of Tate, 367 fixed point theorem, 53
compatible family, 467 Frattini embedding problem, 288
concordance obstruction, 364 Frattini embedding theorem, 319
concordant embedding problem, 351 Freiheitssatz of Iwasawa, 295

© Springer-Verlag GmbH Germany, part of Springer Nature 2018 531

G. Malle, B. H. Matzat, Inverse Galois Theory, Springer Monographs
in Mathematics, https://doi.org/10.1007/978-3-662-55420-3
532 Index

Frobenius endomorphism, 385 large field, 475

Frobenius field, 385 Lemma of Scott, 260
Frobenius module, 385 Lemma of Speiser, 201
full symmetry group, 31, 63 linear Tschirnhaus transform, 401
fundamental solution matrix, 387 linearly rigid tuple, 260
fundamental system of solutions, 387 Lusztig series, 126

G-compatible family, 469 M-section, 469

G-realization, 34 mapping class group, 183
G-relative H -invariant, 396 modular Dedekind criterion, 403
G-relative Colin Matrix, 399 modular Galois theory, 383
G-relative resolvent, 396 Moore determinant, 387
GA-realization, 36 Moore matrix, 387
GAGA for IP1 , 456 morphism of rigid analytic spaces, 451
Galois group of an F-module, 390 multiplication with c, 267
GAR-realization, 302
general unitary group, 107 non-split embedding problem, 288
generating s-system, 26 normalized structure constant, 36
generic polynomial, 396
geometric (proper) solution, 289 open ultrametric disc, 462
geometric embedding problem, 289 orthogonal group, 110
geometric field extension, 8 orthogonal group of minus type, 115
geometrically conjugate, 126 orthogonal group of plus type, 111
GL-stable tuple, 260
gluing datum, 451 }-stable, 86
gluing of morphisms, 452 pairwise adjusted, 471
gluing of spaces, 451 Pochhammer transform, 251
good reduction modulo p, 88 Pochhammer transformation, 251
Green function, 128 primitive linear group, 100
group of geometric automorphisms, 43 primitive prime divisor, 118
primitive translate, 55
profinite Hurwitz braid group, 189
HsV -rigid class vector, 212
profinite Riemann existence theorem, 4
Hasse embedding obstruction, 364
projective profinite group, 294
Hasse-Witt-invariant, 332
proper solution (field) of an embedding
Hilbertian field, 287
problem, 288
Hilbertian set, 287
pseudo algebraically closed, 229
homology, 100
pseudo Steinberg cross section, 424
homomorphism ramified in, 480
pseudo-reflection, 100
(full) Hurwitz braid group, 181
pure Artin braid group, 179
Hurwitz classification, 27, 198
pure Hurwitz braid group, 181
hypothesis (H), 254
q-additive polynomial, 388
induced cover, 461 quasi-central element, 132
irrationality degree, 28 quasi-determinant, 115
irreducible Jordan–Pochhammer tuple, 270 quasi-p-group, 484

j -th braid orbit genus, 213 r-fold uncomplete product, 179

Jordan–Pochhammer tuple, 270 r-fold uncomplete symmetric product, 179
rational class vector, 29
k-rational class vector, 319 rational subset, 450
k-symmetric class vector, 319 rationally rigid class vector, 29
kernel of an embedding problem, 288 reduced braid orbit genera, 245
Index 533

reflection, 100 Steinberg cross section, 406

regular solution of an embedding problem, 289 Steinberg endomorphism, 423
regularity theorem, 212 strictly non-degenerate quadratic form, 332
relative Reynolds operator, 398 strong rigidity theorem, 32
rigid analytic space, 451 symmetric algebra, 394
rigid braid cycle, 247 symmetry group, 31, 63
rigid braid cycle theorem, 247 symplectic group, 108
rigid braid orbit theorem, 216
rigid class vector, 29 Tate algebra, 449
rigid HSV -orbit, 48, 64 thick normal subgroup, 185
rigid HsV -orbit, 212 transference, 347
rigidity defect, 263 translation theorem, 58
ring of holomorphic functions, 450 transvection, 100
ring of invariants, 394 trivial cover, 461
robust generating systems, 407 trivial F-module, 385
root, 93 twisted braid orbit theorem, 239
twisted rigidity theorem, 50
s-th V -symmetrized braid orbit genus, 235 twisted structure sheaf, 455
Scholz embedding problem, 374 twisted upper bound theorem, 423
Scholz extension, 374
Scholz solution, 374 uniform function, 126
Schur multiplier, 226 unipotent character, 127
second embedding obstruction, 364 uniquely liftable, 318
semiabelian group, 299 unirational function field, 200
semirational class, 41 universally central embeddable Galois
shape function, 226 extension, 328
socle of an `-Galois extension, 375 unramified, 187
solution field of an embedding problem, 288 unramified rational place, 223
solution field of an F-module, 385 upper bound theorem, 391
solution of an embedding problem, 288
solution space of an F-module, 385 V -configuration, 48, 232
specialization theorem, 224 V -rigid class vector, 48
sphere relations, 182 V -symmetric, 31
spinor norm, 110 V -symmetrized braid orbit, 211
split embedding problem, 288 V -symmetrized irrationality degree, 31
splitting theorem, 13, 196
stability condition, 260 wreath extension, 347