Graduale Texts in Mathematics 36

Editorial Board: F. W. Gehring

P. R. Halmos (Managing Editor)
C. C. Moore
Linear Topological
Spaces

lohn L. Kelley Isaac Namioka

and
W. F. Donoghue, Jr. G. Baley Price
Kenneth R. Lucas Wendy Robertson
B. J. Pettis W. R. Scott
Ebbe Thue Poulsen Kennan T. Smith

Springer-Verlag Berlin Heidelberg GmbH

J ohn L. Kelley Isaac Namioka
Department of Mathematics Department of Mathematics
University of California University of Washington
Berkeley, California 94720 Seattle, Washington 98195

Editorial Board
P. R. Halmos
Indiana University
Department of Mathematics
Swain Hali East
Bloomington, Indiana 47401

F. W. Gehring C. C. Moore
University of Michigan University of California at Berkeley
Department of Mathematics Department of Mathematics
Ann Arbor, Michigan 48104 Berkeley, California 94720

AMS Subject Classifications

46AXX

Library of Congress Cataloging in Publication Data

Kelley, John L.
Linear topologica! spaces.
(Graduate texts in mathematics; 36)
Reprint of the ed. published by Van Nostrand, Princeton, N.J., in series: The
University series in higher mathematics.
Bibliography: p.
lncludes index.
1. Linear topological spaces. I. Namioka, Isaac, joint author. IT. Title. ITI. Series.
QA322.K44 1976 514'.3 75-41498
Second corrected printing
AII rights reserved.
No part of this book may be translated or reproduced in any form without written
permission from Springer-Verlag Berlin Heidelberg GmbH
~ 1963 by J. L. Kelley and G. B. Price.
Originally published by Springer-Verlag New York Heidelberg Berlin in 1963
Softcover reprint of the hardcover 1st edition 1963
Originally published in the University Series in Higher Mathematics (D. Van
Nostrand Company); edited by M. H. Stone, L. Nirenberg and S. S. Chem.

ISBN 978-3-662-41768-3 ISBN 978-3-662-41914-4 (eBook)

DOI 10.1007/978-3-662-41914-4
 FOREWORD

THIS BOOK ISA STUDY OF LINEAR TOPOLOGICAL SPACES. EXPLICITLY, WE

are concerned with a linear space endowed with a topology such that scalar
multiplication and addition are continuous, and we seek invariants relative
to the dass of all topological isomorphisms. Thus, from our point of view,
it is incidental that the evaluation map of a normed linear space into its
second adjoint space is an isometry; it is pertinent that this map is relatively
open. W e study the geometry of a linear topological space for its own sake,
and not as an incidental to the study of mathematical objects which are
endowed with a more elaborate structure. This is not because the relation
of this theory to other notions is of no importance. On the contrary, any
discipline worthy of study must illuminate neighboring areas, and motiva-
tion for the study of a new concept may, in great part, lie in the clarification
and simplification of more familiar notions. As it turns out, the theory of
linear topological spaces provides a remarkable economy in discussion of
many classical mathematical problems, so that this theory may properly be
considered to be both a synthesis and an extension of older ideas.*
The textbegins with an investigation of linear spaces (not endowed with
a topology). The structure here is simple, and complete invariants for a
space, a subspace, a linear function, and so on, are given in terms of cardinal
numbers. The geometry of convex sets is the first topic which is peculiar to
the theory of linear topological spaces. The fundamental propositions here
( the Hahn-Banach theorem, and the relation between orderings and convex
cones) yield one of the three general methods which are available for attack
on linear topological space problems.
A few remarks on methodology will clarify this assertion. Our results
depend primarily on convexity arguments, on compactness arguments (for
example, Smulian's compactness criterion and the Banach-Alaoglu theorem),
and on category results. The chief use of scalar multiplication is made in
convexity arguments; these serve to differentiate this theory from that of
• I am not enough of 2 scholar either to affirm or deny that a11 mathematics is both
a synthesis and an extension of older mathematics.
V
VI FOREWORD

topological groups. Compactness arguments-primarily applications of the

Tychonoff product theorem-are important, but these follow a pattern
which is routine. Category arguments are used for the most spectacular of
the results of the theory. It is noteworthy that these results depend essen-
tially on the Baire theorem for complete metric spaces and for compact
spaces. There are non-trivial extensions of certain theorems (notably the
Banach-Steinbaus theorem) to wider classes of spaces, but these extensions
are made essentially by observing that the desired property is preserved by
products, direct sums, and quotients. No form of the Baire theorem is
available save for the classical cases. In this respect, the role played by
completeness in the general theory is quite disappointing.
After establishing the geometric theorems on convexity we develop the
elementary theory of a linear topological space in Chapter 2. With the
exceptions of a few results, such as the criterion for normability, the
theorems of this chapter are specializations of well-known theorems on
topological groups, or even more generally, of uniform spaces. In other
words, little use is made of scalar multiplication. The material is included
in order that the exposition be self-contained.
A brief chapter is devoted to the fundamental category theorems. The
simplicity and the power of these results justify this special treatment,
although full use of the category theorems occurs later.
The fourth chapter details results on convex subsets of linear topological
spaces and the closely related question of existence of continuous linear func-
tionals, the last material being essentially a preparation for the later chapter
on duality. The most powerful result of the chapter is the Krein-Milman
theorem on the existence of extreme points of a compact convex set. This
theorem is one of the strongest of those propositions which depend on con-
vexity-compactness arguments, and it has far reaching consequences-for
example, the existence of sufficiently many irreducible unitary representa-
tions for an arbitrary locally compact group.
The fifth chapter is devoted to a study of the duality which is the central
part of the theory of linear topological spaces. The existence of a duality
depends on the existence of enough continuous linear functionals-a fact
which illuminates the role played by local convexity. Locally convex spaces
possess a large supply of continuous linear functionals, and locally convex
topologies are precisely those which may be conveniently described in terms
of the adjoint space. Consequently, the duality theory, and in substance the
entire theory of linear topological spaces, applies primarily to locally convex
spaces. The pattern of the duality study is simple. We attempt to study
a space in terms of its adjoint, and we construct part of a "dictionary" of
 FOREWORD Vll

translations of concepts defined for a space, to concepts involving the adjoint.

For example, completeness of a space E is equivalent to the proposition that
each hyperplane in the adjoint E* is weak* dosed whenever its intersection
with every equicontinuous set A is weak* dosed in A, and the topology of
E is the strongest possible having E* as the dass of continuous linear func-
tionals provided each weak* compact convex subset of E is equicontinuous.
The situation is very definitely more complicated than in the case of a
Banach space. Three "pleasant" properties of a space can be used to dassify
the type of structure. In order of increasing strength, these are: the topol-
ogy for Eis the strongest having E* as adjoint (E is a Mackey space), the
evaluation map of E into E** is continuous (E is evaluable), and a form of
the Banach-Steinhaus theorem holds for E (E is a barrelled space, or ton-
nele). A complete metrizable locally convex space possesses all of these
properties, but an arbitrary linear topological space may fail to possess any
one of them. The dass of all spaces possessing any one of these useful
properties is dosed under formation of direct sums, products, and quotients.
However, the properties are not hereditary, in the sense that a dosed sub-
space of a space with the property may fail to have the property. Complete-
ness, on the other hand, is preserved by the formation of direct sums and
products, and obviously is hereditary, but the quotient space derived from a
complete space may fail to be complete. The situation with respect to
semi-reflexiveness ( the evaluation map carries E onto E**) is similar. Thus
there is a dichotomy, and each of the useful properties of linear topological
spaces follows one of two dissimilar patterns with respect to "permanence"
properties.
Another type of duality suggests itself. A subset of a linear topological
space is called bounded if it is absorbed by each neighborhood of 0 ( that
is, sufficiently large scalar multiples of any neighborhood of 0, contain the
set). We may consider dually a family f!l of sets which are to be con-
sidered as bounded, and construct the family (fiJ of all convex cirded sets
which absorb members of the family f!l. The family Yll defines a topology,
and this scheme sets up a duality ( called an internal duality) between
possible topologies for E and possible families of bounded sets. This in-
ternal duality is related in a simple fashion to the dual space theory.
The chapter on duality condudes with a discussion of metrizable spaces.
As might be expected, the theory of a metrizable locally convex space is
more nearly perfect than that of an arbitrary space and, in fact, most of
the major propositions concerning the internal structure of the dual of a
Banach space hold for the adjoint of a complete metrizable space. Count-
ability requirements are essential for many of these results. However, the
Vlll FOREWORD

structure of the second adjoint and the relation of this space to the first
adjoint is still complex, and many features appear pathological compared
to the classical Banach space theory.
The Appendix is intended as a bridge between the theory of linear topo-
logical spaces and that of ordered linear spaces. The elegant theorems of
Kakutani characterizing Banach lattices which are of functional type, and
those which are of L 1-type, are the principal results.
A final note on the preparation of this text: By fortuitous circumstance
the authors were able to spend the summer of 1953 together, and a complete
manuscript was prepared. We feit that this manuscript had many faults,
not the least being those inferred from the old adage that a camel is a horse
which was designed by a committee. Consequently, in the interest of a
more uniform style, the text was revised by two of us, I. Namioka and
myself. The problern lists were revised and drastically enlarged by Wendy
Robertson, who, by great good fortune, was able to join in our enterprise
two years ago.
J. L. K.
Berkeley, California, 1961

Note on notation: The end of each proof is marked by the symbol

II I.
 ACKNOWLEDGMENTS

WE GRATEFULLY ACKNOWLEDGE A GRANT FROM THE GENERAL RESEARCH

funds of the University of Kansas which made the writing of this book
possible. Several federal agencies have long been important patrons of the
sciences, but the sponsorship by a university of a large-scale project in mathe-
matics is a significant development.
Revision of the original manuscript was made possible by grants from the
Office of Naval Research and from the National Science Foundation. We
are grateful for this support.
We are pleased to acknowledge the assistance of Tulane University which
made Professor B. J. Pettis available to the University of Kansas during the
writing of this book, and of the University of California which made Pro-
fessor J. L. Kelley available during the revision.
W e wish to thank several colleagues who have read all or part of our
manuscript and made valuable suggestions. In particular, we are indebted
to Professor John W. Brace, Mr. D. 0. Etter, Dr. A. H. Kruse, Professor
V. L. Klee, and Professor A. Wilansky. We also wish to express our
appreciation to Professor A. Robertson and Miss Eva Kallin for their help
in arranging some of the problems.
Finally, Mrs. Donna Merrill typed the original manuscript and Miss
Sophia Glogovac typed the revision. W e extend our thanks for their expert
servrce.

IX
 CONTENTS

CHAPTER 1 LINEAR SP ACES

PAGE
1 LINEAR SPACES . 1
Bases, dimension, linear functions, products, direct sums,
projective and inductive Iimits.
PROBLEMS • 11
A Cardinal numbers; B Quotients and subspaces; C Di-
rect sums and products; D Space of bounded functions;
E Extension of linear functionals; F Null spaces and
ranges; G Algebraic adjoint of a linear mapping; H Set
functions; I Inductive limits
2 CoNVEXITY AND ORDER 13
Convex sets, Minkowski functionals, cones and partial or-
derings.
PROBLRMS • 17
A Midpoint convexity; B Disjoint convex sets; C Min-
kowski functionals; D Convex extensions of subsets of finite
dimensional spaces; E Convex functionals; F Families of
cones; G V ector orderings of R 2 ; H Radial sets; I Z - A
dictionary ordering; J Helly's theorem
3 SEPARATION AND EXTENSION THEOREMS 18
Separation of convex sets by hyperplanes, extension of linear
functionals preserving positivity or preserving a bound.
PROBLEMS • 23
A Separation of a linear manifold from a cone; B Alterna-
tive proof of lemma 3.1 ; C Extension of theorem 3.2;
D Example; E Generalized Hahn-Banach theorem; F Gen-
eralized Hahn-Banach theorem (variant); G Example on
non-separation; H Extension of invariant linear functionals

CHAPTER 2 LINEAR TOPOLOGICAL SPACES

4 TOPOLOGICAL SPACES 27
Brief review of topological notions, products, etc.
X
 CONTENTS Xl
PAGE
PROBLEMS • 32
A Compact and locally compact spaces; B Separability;
C Complete metric spaces; D Hausdorff metric on a space
of subsets; E Contraction mapping
5 LINEAR TOPOLOGICAL SPACES, LINEAR FUNCTIONALS, QUO-
TIENT AND PRODUCTS 33
Local bases, continuity of linear functions, product and quo-
tient spaces.
PROBLEMS • 41
A Exercises; B Natural, non-vector topologies; C Pro-
jective topology; D Attempt at a strongest vector topology;
E Strongest vector topology I; F Box topology; G Alge-
braic closure of convex sets I; H Linearly closed convex
sets I ; I Locally convex sets
6 NoRMABILITY, METRIZABILITY, AND EMBEDDING; LOCAL
CONVEXITY • 43
Embedding in normed spaces, metrizable spaces, and in
products of pseudo-normed spaces.
PROBLEMS • 51
A Exercises; B Mappings in pseudo-normed spaces I;
C Topologies determined by pseudo-metrics; D Products
and normed spaces; E Positive linear functionals; F Locally
convex, metrizable, non-normable spaces; G Topology of
pointwise convergence; H Bounded sets and functionals;
I Strongest locally convex topology I ; J Inner products;
K Spaces of integrable functions I; L Spaces of measurable
functions I ; M Locally bounded spaces; N Spaces of in-
tegrable functions II
7 CoMPLETENEss • 56
Completeness and total boundedness, characterization of
finite dimensional spaces, completion.
PROBLEMS • 64
A Finite dimensional subspaces; B Completion of a pseudo-
metrizable, pseudo-normable, or locally convex space; C Com-
pleteness for stronger topologies; D Extension of a one-to-one
mapping; E Complementary subspaces; F Totally bounded
sets; G Topologies on a direct sum; H Hilbert spaces;
I Hilbert spaces: Projection; J Hilbert spaces: Orthogonal
complements; K Hilbert space: Summability; L Hilbert
spaces: Orthonormal bases; M Spaces of integrable func-
tions III; N Spaces of measurable functions II; 0 The sum
of closed subspaces
Xll CONTENTS
PAGE
8 FuNCTION SPACES 68
Uniform convergence on the members of a family, com-
pleteness, equicontinuity, compactness and countable com-
pactness.
PROBLEMS • 79
A Converse of 8.1; B Mappings in pseudo-normed spaces li;
C Pointwise Cauchy nets; D Product of ff.91 and ff111
E Functional completion; F Additive set functions; G
Boundedness in B.91; H Compactness of sets of functions;
I Spaces of continuous functions I; J Distribution spaces I

CHAPTER 3 THE CATEGORY THEOREMS

9 CATEGORY IN TOPOLOGICAL SPACES . 84

Condensation theorem, Baire category theorem, Osgood's
theorem on point of equicontinuity.
PROBLEMS • 87
A Exercise on category; B Preservation of category;
C Lower semi-continuous functions; D Generalized Baire
theorem; E Embedding of a finite dimensional compact metric
space into an Euclidean space; F Linear space of dimension
No ; G Image of a pseudo-metrizable linear space; H Ad-
ditive set functions; I Sequential convergence in L 1 (X,p,)
10 THE ABSORPTION THEOREM AND THE DIFFERENCE THEOREM 90
PROBLEMS • 95
A Continuity of additive mappings; B Subspaces of the
second category; C Linear spaces with pseudo-metrizable
topology; D Midpoint convex neighborhoods; E Sets of
sequential convergence; F Problems in topological complete-
ness and metric completion
11 THE CLOSED GRAPH THEOREM • 97
Closed graph theorem and open mapping theorem.
PROBLEMS • 100
A Comparison of topologies; B Subspace of LP n Lq;
C Symmetrie Operators ; D An open mapping theorem;
E Closed relation theorem; F Continuously differentiable
functions; G M appings into the space L 1 ; H Condition
for a closed graph; I Closed graph theorem for metrizable
spaces?; J Continuity of positive linear functionals
12 EQUICONTINUITY AND BOUNDEDNESS 102
Elementary properties, uniform boundedness, Banach-Stein-
baus theorem.
PROBLEMS • 105
A Boundedness of norms of transformations; B The prin-
ciple of condensation of singularities; C Banach-Steinbaus
 CONTENTS
...
Xlll
PAGE
theorem ; D Strongest locally convex topology II ; E Closed
graph theorem I; F Continuous functions non-differentiahte
on sets of positive measure; G Bilinear IIIappings

CHAPTER 4 CONVEXITY IN LINEAR TOPOLOGICAL

SPACES

13 CONVEX SUBSETS OF LINEAR TOPOLOGICAL SPACES 110

lnterior, closure, linear combinations of convex sets, closed
convex extensions of totally bounded sets, continuous func-
tionals on convex sets.
PROBLEMS . 114
A Midpoint convexity; B Condensation corollary; C Con-
vex extension of bounded and totally bounded sets; D Trans-
lates of convex sets; E Extension of open convex sets; F
Hypercomplete spaces; G Closed graph theorem II
14 CONTINUOUS LINEAR FUNCTIONALS 117
Existence and extension of continuous linear ftinctionals,
adjoint of subsJ)aces, quotient spaces, products and direct
sums.
PROBLEMS . 123
A Exercises; B Further separation theorems; C A fixed
point theorem; D Strongest locally convex topology III;
E Strongest vector topology II ; F Algebraic closure of
convex sets II; G Linearly closed convex sets II; H A
fundamental theorem of game theory; I Camplex measures;
J Spaces of continuous functions II; K Space of convergent
sequences; L Hilbert spaces II; M Spaces of integrable
functions IV
15 ExTREME POINTS 130
The Krein-Milman theorem.
PROBLEMS . 132
A A bounded set with no extreme point; B Existence of
extreme points; C Extreme image points; D Maximum of
a linear functional; E Subsets of a compact convex set;
F Two counter-examples; G Extreme half lines; H Limits
and extreme points; I Extreme points in Ll and L 00 ; J Ex-
treme points in C(X) and its adjoint

CHAPTER 5 DUALITY

16 PAIRINGS 137
Paired spaces, weak topologies, polars, compactness criteria,
completeness relative to uniform convergence on the mem-
bers of a family, subspaces, quotients, direct sums, and
products.
XlV CONTENTS
PAGe
PROBLEMS • 148
A Duality between totally bounded sets; B Polar of a sum;
C Inductive Iimits II; D Projective limits; E Duality be-
tween inductive and projective Iimits; F Sequential con-
vergence in V (X,p.) II; G Dense subspaces; H Helly's
condition; I Tensor products I
17 THE WEAK TOPOLOGIES 153
W eak and weak* topologies, weak compactness, subspaces,
quotients, products, and direct sums.
PROBLEMS • 161
A Exercises; B Total subsets; C Uniformly convex spaces
I; D Vector-valued analytic functions; E Stone-Weierstrass
theorem; F Completeness of a direct sum; G Inductive
Iimits III; H Integration proof of theorem 17.11 ; I Weakly
compact convex extensions; J W eak* separability; K Helly's
choice principle; L Existence of weakly convergent sequence
18 ToPOLOGIEs FOR E AND E* 165
Admissible topologies, strong topology, equicontinuous,
weak* compact, strong bounded and weak* bounded sets,
barrelled spaces, topologies yielding a given dual, Mackey
spaces, products, sums, etc.
PROBLEMS • 176
A Exercises; B Characterization of barrelled spaces; C
Extension of the Banach-Steinhaus theorem; D Topologies
admissible for the same pairing; E Extension of the Banach-
Alaoglu theorem ; F Counter-example on weak* compact sets;
G Krein-Smulian theorem; H Example and counter-example
on hypercomplete spaces; I Fully complete spaces; J Closed
graph theorem III; K Spaces of bilinear mappings; L Tensor
products II
19 BouNDEDNEss 180
Bound topologies, products and quotients.
PROBLEMS • 188
A Inductive Iimits IV; B Closed graph theorem IV; C
Completeness of the adjoint
20 THE EVALUATION MAP INTO THE SECOND ADJOINT 189
Semi-reflexive, evaluable and reflexive spaces.
PROBLEMS • 195
A Example of a non-evaluable space; B An evaluable prod-
uct; C Converse of 20.7(i); D Counter-example on quo-
tients and subspaces; E Problem; F Montel spaces; G
Strongest locally convex topology; H Spaces of analytic func-
tions I; I Distribution spaces II; J Closed graph theorem
for a reflexive Banach space; K Evaluation of a normed
 CONTENTS XV
PAGE
space; L Uniformly convex spaces II; M A nearly reflexive
Banach space
21 DUAL TRANSFORMATIONS • 199
Existence and uniqueness of duals, continuity and openness
relative to admissible topologies, adjoint transformations,
continuity and openness.
PROBLEMS . 206
A Completely continuous mappings; B Riesz theory; C
Complete continuity of the adjoint; D Schauder's theorem;
E Closable mappings; F Stone-Cech compactification
22 PsEUDO-METRIZABLE SPACES 209
Boundedness properties, weak* closed convex sets, structure
of adjoint space.
PROBLEMS • 217
A Condition for completeness: B Embedding theories; C
lnductive Iimits V; D Spaces of analytic functions II; E
Spaces /P ( w) ; F Köthe spaces; G Counter-example on
Frechet spaces; H An adjoint with a bound topology; I
Example on Montel spaces; J Spaces of analytic functions 111

APPENDIX ORDERED LINEAR SPACES

23 ÜRDERED LINEAR SPACES • 224

Order dual, conditions that a continuous functional belong
to order dual, elementary properties of vector lattices, lattice
pseudo-norms.

24 L AND M SP ACES 236

Kakutani's characterizatien of Banach lattices which are of
functional type or of L 1 type.
ßiBLIOGRAPHY 248
LIST OF SYMBOLS 249
INDEX • 251
 Chapter 1

LINEAR SPACES

This chapter is devoted to the algebra and the geometry of linear

spaces; no topology for the space is assumed. It is shown that a
linear space is determined, to a linear isomorphism, by a single
cardinal number, and that subspaces and linear functions can be
described in equally simple terms. The structure theorems for
linear spaces are valid for spaces over an arbitrary field; however, we
are concerned only with real and complex linear spaces, and this
restriction makes the notion of convexity meaningful. This notion is
fundamental to the theory, and almost all of our results depend upon
propositions about convex sets. In this chapter, after establi3hing
Connections between the geometry of convex sets and certain analytic
objects, the basic separation theorems are proved. These theorems
provide the foundation for linear analysis; their importance cannot
be overemphasized.

1 LINEAR SPACES
Each linear space is characterized, to a linear isomorphism, by a
cardinal number called its dimension. A subspace is characterized by
its dimension and its co-dimension. After these results have been
established, certain technical propositions on linear functions are
proved (for example, the induced map theorem, and the theorem
giving the relation between the linear functionals on a complex linear
space and the functionals on its real restriction). The section ends
with a number of definitions, each giving a method of constructing
new linear spaces from old.

Areal (complex) linear space (also called a vector space or

a linear space over the real (respectively, complex) field) is a
1
2 CH. 1 LINEAR SPACES

non-void set E and two operations called addition and scalar mul-
tiplication. Addition is an operation EB which satisfies the following
axwms:
(i) For every pair of elements x and y in E, x EB y, called the sum
of x and y, is an element of E;
(ii) addition is commutative: x EB y = y EB x;
(iii) addition is associative: x EB (y EB z) = (x EB y) EB z;
(iv) there exists in E a unique element, 8, called the origin or the
(additive) zero element, suchthat for all x in E, .x EB 8 = x;
and
(v) to every x in E there corresponds a unique element, denoted
by -x, suchthat X EB ( -x) = e.
Scalar multiplication is an operation · which satisfies the following
axwms:
(vi) For every pair consisting of a real (complex) number a and
an element x in E, a · x, called the product of a and x, is an
element of E;
(vii) multiplication is distributive with respect to addition in E:
a·(x EB y) = a·x EB a·y;
(viii) multiplication is distributive with respect to the addition of
real (complex) numbers: (a + b)·x = a·x EB b·x;
(ix) multiplication is associative: a·(b·x) = (ab)·x;
(x) 1 · x = x for all x in E.
From the axioms it follows that the set E with the operation
addition is an abelian group and that multiplication by a fixed scalar
is an endomorphism of this group.
In the axioms, + and juxtaposition denote respectively addition
and multiplication of real (complex) numbers. Because of the rela-
tions between the two kinds of addition and the two kinds of multi-
plication, no confusion results from the practice, to be followed
henceforth, of denoting both kinds of addition by + and both kinds
of multiplication by juxtaposition. Also, henceforth 0 denotes,
ambiguously, either zero or the additive zero element 8 of the abelian
group formed by the elements of E and addition. Furthermore, it is
customary to say simply " the linear space E" without reference to
the operations. The elements of a linear space E are called vectors.
The scalar field K of a real (complex) linear space is the field of real
(complex) numbers, and its elements are frequently called scalars.
The real (complex) field is itself a linear space under the convention
that vector addition is ordinary addition, and that scalar multiplication
 SEC. 1 LINEAR SPACES 3
is ordinary rnultiplication in the field. If it is said that a linear space
Eis the real (cornplex) field, it will always be understood that Eis a
linear space in this sense.
Two linear spaces E and F are identical if and only if E = F and
also the operations of addition and scalar rnultiplication are the sarne.
In particular, the real linear space obtained frorn a cornplex linear
space by restricting the dornain of scalar rnultiplication to the real
nurnbers is distinct frorn the latter space. It is called the real
restriction of the cornplex linear space. It rnust be ernphasized
that the real restriction of a cornplex linear space has the sarne set of
elernents and the same operation of addition; rnoreover, scalar
rnultiplication in the cornplex space and its real restriction coincide
when both are defined. The only difference-but it is an irnportant
difference-is that the dornain of the scalar rnultiplication of the real
restriction is a proper subset of the dornain of the original scalar
rnultiplication. The real restriction of the cornplex field is the two-
dirnensional Euclidean space. (By definition, real (complex)
Euclidean. n-space is the space of all n-tuples of real (complex,
respectively) nurnbers, with addition and scalar rnultiplication
defined coordinatewise.) It rnay be observed that not every real
linear space is the real restriction of a cornplex linear space (for
exarnple, one-dirnensional real Euclidean space).
A subset A of a linear space Eis (finitely) linearly independent
if and only if a finite linear cornbination .L {a 1x1 : i = 1, · · ·, n}, where
x1 E A for each i and x1 # xj for i # j, is 0 only when each a1 is zero.
This is equivalent to requiring that each rnernber of E which can be
written as a linear cornbination, with non-zero coefficients, of distinct
rnernbers of A have a unique such representation (the difference of
two distinct representations exhibitslinear dependence of A). A sub-
set B of E is a Hamel base for E if and only if each non-zero elernent
of E is representable in a unique way as a finite linear cornbination
of distinct rnernbers of B, with non-zero coefficients. A Harne! base
is necessarily linearly independent, and the next theorern shows that
any linearly independent set can be expanded to give a Harne! base.
1.1 THEOREM Let E be a linear space. Then:
(i) Each linearly independent subset of E is contained in a maximal
linearly independent subset.
(ii) Each maximallinearly independent subset is a Hamel base, and
conversely.
(iii) Any two Hamel bases have the same cardinal number.
4 CH. 1 LINEAR SPACES

PROOF The first propos1t10n is an immediate consequence of the

maximal principle, and the elementary proof of (ii) is omitted. The
proof of (iii) is made in two parts. First, suppose that there is a
finite Hamel base B for E, and that A is an arbitrary linearly in-
dependent set. It will be shown that there are at most as many
members in A as there are in B, by setting up a one by one replace-
ment process. First, a member x1 of A is a linear combination of
members of B, and hence some member of B is a linear combination
of x1 and other members of B. Hence x1 tagether with B with one
member deleted is a Hamel base. This process is continued; at the
r-th stage we observe that a member Xr of A is a linear combin-
ation of x 1 , · · ·, Xr _ 1 and of the non-deleted members of B, that Xr
is not a linear combination of x 1 , · · ·, Xr _ 1 and that therefore one
of the remaining members of B can be replaced by Xr to yield a
Harne! base. This process can be continued until A is exhausted, in
which case it is clear that A contains at most as many members as
B, or until all members of B have been deleted. In this case
there can be no remaining members of A, for every x is a linear
combination of the members of A which have been selected. The
proof of (iii) is then reduced to the case where each Hamel base is
infinite.
Suppose that B and C are two infinite Hamel bases for E. For
each member x of BIet F(x) be the finite subset of C such that x is a
linear combination with non-zero coefficients of the members of F(x).
Since the finite linear combinations of members of U {F(x): x E B}
include every member of B and therefore every member of E, C =
U {F(x): x E B}. Let k(A) denote the cardinal number of A; then
k(C) = k(U {F(x): x E B}) ~ N0 ·k(B) = k(B) because B is an in-
finite set (see problern lA). A similar argument shows that k(B) ~
k(C), whence k(B) = k(C).III
The dimension of a linear space is the cardinal number of a
Harne! base for the space.
A linear space F is a subspace (linear subspace) of a linear space
E if and only if F is a subset of E, F and E have the same scalar field
K, and the operations of addition and scalar multiplication in F
coincide with the corresponding operations in E. A necessary and
sufficient condition that F be a subspace of Eis that the set F be a non-
empty subset of E, that F be closed under addition and scalar multi-
plication in E, and that addition and scalar multiplication in F
coincide with the corresponding operations in E. If A is an arbitrary
subset of E, then the set of alllinear combinations of members of A is
 SEc. 1 LINEAR SP ACES 5
a linear subspace of E which is called the linear extension (span,
hull) of A or the subspace generated by A.
It is convenient to define addition and scalar multiplication of
subsets of a linear space. If A and B are subsets of a linear space E,
then A + B is defined to be the set of all sums x + y for x in A and
y in B. If x is a member of E, then the set {x} + A is abbreviated
to x + A, and x + A is called the translation or translate of A
by x. If a is a scalar, then aA denotes the set of all elements ax for
x in A, and - A is an abbreviation for (- 1)A; this coincides with the
set of all points - x with x in A.
Using this terminology, it is clear that a non-void set F of E is a
linear subspace of E if and only if aF + bF c F for all scalars a
and b. lf F and G are linear subspaces, then F + G is a linear sub-
space, but the translate x + F of a linear subspace is not a linear
subspace unless x E F. A set of the form x + F, where F is a linear
subspace, is called a linear manifold or linear variety, or a flat.
Two linear subspaces F and G of E are complementary if and
only if each member of E can be written in one and only one way as
the sum of a member of F and a member of G. Observe that if a
vector x has two representations as the sum of a member of F and
a member of G, then (taking the difference) the zero vector has a
representation other than 0 + 0. It follows that linear subspaces F
and G are complementary if and only if F + G = E and F n G =
{0}. It is true that there is always at least one subspace G com-
plementary to a subspace F of E, for one may choose a Hamel base
B for F, adjoin a set C of vectors to get a base for E, and let G be the
linear extension of C. It can be shown that, if both G and H are
complementary to F in E, then the dimension of G is identical with
the dimension of H (see problern lB). The co-dimension (de-
ficiency, rank) of FinE is defined tobe the dimension of a subspace
of E, which is complementary to F in E.
Let E and F be two linear spaces over the same scalar field, and let
T be a mapping of E into F. Then T is a linear function 1 from E
to F if and only if for all x and y in E and all scalars a and b in K,
T(ax + by) = aT(x) + bT(y). A linear function T is a group
homomorphism of E, under addition, into F, under addition, with the
additional property that T(ax) = aT(x) for each scalar a and each x
in E. The range of a linear function T is always a subspace of F.
Notice that there exists a linear function T on E with arbitrarily
1 'Function', 'map', 'mapping', and 'transformation' are all synonymous, and
they are used interchangeably throughout the text.
6 CH. 1 LINEAR SPACES
prescribed values on the elements of a Hamel base for E, and that
every linear function T is completely determined by its values on
the elements of the Hamel base. Consequently linear functions exist
in some profusion.
The null space (kernel) of a linear function T is the set of all x
suchthat T(x) = 0; that is, the null space ofT is T - 1 [0]. It is easy
to see that T i.s one-to-one if and only if the null space is {0}. A
one-to-one linear map of E onto Fis called a linear isomorphism
of E onto F. The inverse of a linear isomorphism is a linear iso-
morphism, and the composition of two linear isomorphisms is again
a linear isomorphism. Consequently the dass of linear spaces is
divided into equivalence classes of mutually linearly isomorphic
spaces. A property of one linear space which is shared by every
linear isomorph is called a linear invariant. The dimension of a
linear space is evidently a linear invariant, and, moreover, since it is
easy to see that two spaces of the same dimension and over the same
scalar field are linearly isomorphic, the dimension is a complete
linear invariant. That is, two linear spaces over the same scalar
field are isomorphic if and only if they have the same dimension.
If S is a linear map of E into a linear space G and U is a linear map
of G into a space F, then the composition U o S is a linear map of E
into F. It is clear that the null space of U o S contains that of S.
There is a useful converse to this proposition.
1.2 INDUCED MAP THEOREM Let T be a linear transformation from
E into F, and Iet S be a linear transformation from E onto G. If the
null space of T contains that of S, then there is a unique linear trans-
formation U from G into F such that T = U o S. The function U is
one-to-one if and only if the null spaces ofT and S coincide.
PROOF If x is any element of G, and S - 1 [ x] is the set of all elements
y in E for which S(y) = x, then S - 1 [x] is a translate of S - 1 [0].
Consequently (since S - 1 [0] c T - 1 [0]) T has a constant value, say z,
on S - 1 [x]. It now follows easily that T = U o S if and only if
U(x) is defined to be z, and that the function U is one-to-one if and
only if the null spaces coincide.jjj
The scalar field is itself a linear space, if scalar multiplication is
defined to be the multiplication in the field. A linear functional on
a linear space E is a linear function with values in the scalar field.
The null space N of a linear furrctional f which is not identically zero
is of co-dimension one, as the following reasoning demonstrates. If
x f/= N, then f(x) #- 0; and if y is an arbitrary member of the linear
 SEc. 1 LINEAR SPACES 7
space E, then f(y - [f(y)lf(x)]x) = 0. That is, y - xf(y)lf(x) is
a member of N and hence each member of E is the sum of a member
of N and a multiple of x. A subspace of Co-dimension one can
clearly be described as a maximal (proper) linear subspace of E.
Moreover, if N is any maximal linear subspace of E and x is any
vector which does not belong to N, then each member y of E can be
written uniquely as a linear combination f(y)x + z, where z E N.
The function f is a linear functional, and its null space is precisely N,
and hence each maximal subspace is precisely the null space of a linear
functional which is not identically zero. Thus the following are
equivalent: null space of a linear functional which is not identically
zero, maximal linear subspace, and linear subspace of Co-dimension
one.
It also follows from the foregoing discussion that if g is a linear
functional whose null space includes that of a linear functional f
which is not identically zero, and if f(x) # 0, then g(y) = (g(x)f
f(x))f(y). That is, g is a constant multiple of f. This result is a
special case of the following theorem.
1.3 THEOREM ON LINEAR DEPENDENCE A linear functional fo is a
linear combination of a finite set / 1 , · · · ,fn of linear functionals if and
only if the null space of / 0 contains the intersection of the null spaces of
/1, · · ·,fn·
PROOF If / 0 is a linear combination of / 1 , · · ·, fm then the null space
of / 0 obviously contains the intersection of the null spaces of / 1 , / 2 , · · ·,
fn· The converse is proved by induction, and the case n = 1 was
established in the paragraph pr.eceding the statement of the theorem.
Suppose that the null space N 0 of / 0 contains the intersection of the
null spaces N 1 , · · ·, Nlc+l ofj1 , · · ·,A+l· If each of the functionals
/ 0 , /11 ···,Ais restricted to the subspace N~c+ 1 , then, for x in N~c+ 1 ,
it is true that f 0 (x) = 0 whenever / 1 {x) = · · · = f~c(x) = 0. Hence,
by the induction hypothesis, there are scalars a 11 · · ·, ak such that
/ 0 (x) = 2 {a 1ft(x): i = 1, · · ·, k} whenever x E Nk+l· Consequently
/ 0 - 2 {atf1 : i = 1, · · ·, k} vanishes on the null space of A+ 1 , and is
therefore a scalar multiple of A+1·111
U p to this point the scalar field K has been either the real or
complex numbers, but in the next theorem it will be assumed that K
is complex. Suppose then that E is a complex linear space, and that
f is a linear functional on E. For each x in E let r(x) be the real part
of f(x). It is a Straightforward matter to see that r is a linear func-
tional on the real restriction of E; the fact that f(x) = r(x) - ir(ix)
8 CH. 1 LINEAR SPACES

for every x in E is perhaps a little less expected, but it is also Straight-

forward. On the other hand, if r is a linear functional on the real
restriction of E and f(x) = r(x) - ir(ix) for every x in E, then it is
easy to see thatfis linear on E. These remarks establish the following
result.
1.4 CoRRESPONDENCE BETWEEN REAL AND CaMPLEX LINEAR FuNc-
TIONALS The eorrespondenee defined by assigning to eaeh linear
funetional on E its real part is a one-to-one eorrespondenee between all
the linear funetionals on. E and all the linear funetionals on the real
restrietion of E. lf fandrare paired under this eo"espondenee, where
f is a linear funetional on E and r is a linear funetional on the real
restrietion of E, thenf(x) = r(x) - ir(ix)for every x in E.
This section is concluded with a few definitions on the construction
of new linear spaces from old. If F is a linear space and X is a set,
then the family of all functions on X to F is a linear space, if addition
and scalar multiplication are defined pointwise (that is, (f + g)(x) =
f(x) + g(x) and (af)(x) = af(x)). Many, if not most, linear spaces
which are studied are subspaces of a function space of this sort. For
example: if F is the scalar field and X is the unit interval, then each
of the following families is a linear subspace of the space of all func-
tions on [0: 1] to K: all bounded functions, all continuous functions,
all n-times differentiahte functions, and all Borel functions. The
family of all analytic functions on an open subset of the plane is
another interesting linear space. If X is a a-ring of sets, then each
of the families of all additive, bounded and additive, and countably
additive complex functions on X is a linear space.
If E and F are linear spaces, the set of alllinear functions on E to F
is a subspace of the space of all functions on E to F. If F is the
scalar field, then this subspace is simply the family of all linear
functionals on E. This space is called the algebraic dual of E,
and is denoted by E '.
The product X {E: x EX} is the space of all functions on a set X
to a linear space E. More generally, if for each member t of a non-
void set A there is given a linear space Et over a fixed scalar field,
then the product X {Et: t E A} is the set of all functions x on A such
that x(t) E Et for each t in A. This product is a linear space under
pointwise (coordinate-wise) addition and scalar multiplication. The
subspace ~ {Et: t E A} of X {Et: t E A} consisting of all functions
which are zero except at a finite number of points of A is called the
direct sum. The projection Ps of the product X {Et: t E A} onto
 SEC. 1 LINEAR SPACES 9
the coordinate space Es is defined by Ps(x) = x(s). There is also a
natural map ls of the space Es into the direct sum, defined for each
x in Es by letting ls(x)(t) be zero if t i' s and letting ls(x)(s) = x.
This map is called the injection of Es into the direct sum .L {Et: t E A}.
The following simple proposition is recorded for future reference.
1.5 PROJECTIONS AND lNJECTIONS The projection Ps is a linear map
of the product X {Et: t E A} onto the coordinate space Es. The injection
ls is a linear isomorphism of the coordinate space Es onto the subspace of
the direct sum .L {Et: t E A} which consists of all vectors x such that
x(t) = 0 for t i' s.
If F is a linear subspace of a linear space E, the quotient space
(factor space, coset space, difference space) EfF is defined as
follows. The elements of EfF are sets of the form x + F, where x is
an element of E; evidently two sets of this form are either disjoint
or identical. Addition EB and scalar multiplication · in EfF are
defined by the following equations:
(x + F) EB (y + F) = (x + F) + (y + F)
a·(x + F) = ax + F.
It can be verified that the dass E/F with the operations thus defined
is a linear space. The map Q which carries a member x of E into
the member x + F of EjF is called the quotient mapping; alter-
natively, Q may be described as the map which carries a point x of E
into the unique member of EjF to which x belongs. It is Straight-
forward to see that Q is a linear mapping of E onto EjF, and that F
is the null space of Q. It follows that an arbitrary linear function T
can be represented as the composition of a quotient map and a linear
isomorphism. Explicitly, if T is a linear map of E into G and Fis
the null space of T, then Fis also the null space of the quotient map Q
of E onto EjF, and hence there isalinear isomorphism U of EfF into
G such that T = U o Q by the induced map theorem 1.2.
There is a standard construction for new linear spaces which is
based on the direct sum and the quotient construction. W e will
begin with an example. Consider the dass C of all complex func-
tions J, each of which is defined and analytic on some neighborhood
of a subset A of the complex plane. The domain of definition of a
member f of the dass C depends onf, and the problern is (roughly) to
make a linear space of C. One possible method: define an equiva-
lence relation, by agreeing that f is equivalent to g if and only if f - g
is zero on some neighborhood of A. It is then possible to define
10 CH. 1 LINEAR SPACES

addition and scalar multiplication of equivalence classes so that a

linear space results. An alternative method of defining the linear
space is the following.
For each neighborhood U of A Iet Eu be the linear space of all
analytic functions on U, and for a neighborhood V of A such that
V c U Iet Pv.u be the map of Eu into Ev which carries each member
of Eu into its restriction to V. Let ~ be the family of all neighbor-
boods of A, and Iet N be the subspace of .L {Eu: U E ~} consisting of
all members ~ of this direct sum such that the sum of the non-zero
values of ~ vanishes on some neighborhood of A; equivalently, N is
the set of all members ~ of the direct sum such that for some U in ~
it is true that, if ~(T) =I 0, then T;::) U, and .L {PT,u(~(T)):
U c T} = 0. Each member of .L {Eu: U E ~}/N contains elements
~ with a single non-vanishing coordinate ~( U) = f, and, if g is the
unique non-vanishing coordinate of another member fJ of the sum,
then ~ and fJ belong to the same member of .L {Eu: U E ~}/N if and
only if f = g on some neighborhood of A. The quotient .L {Eu:
U E ~}/N is therefore a linear space which represents, in a reasonable
way, our intuitive notion of the space of all functions analytic on a
neighborhood of the set A. The advantage of using this rather
complicated procedure, rather than the equivalence dass procedure
outlined earlier, is that there are standard ways of topologizing each
Eu, the direct sum and the quotient space, so that a suitable topology
for the space of functions analytic on a neighborhood of A is more or
less self-evident.
The notion of inductive limit of spaces is a formalization of the
process described in the preceding. An inductive system (direct
system) consists of the following: an index set A, directed by a
partial ordering ~ ; a linear space Et for each t in A ; and for every
pair of indices t and s, with t ~ s, a canonical linear map Qts of Es
into Et such that: Qts o Qsr = Qtro for all r, s, and t such that t ~
s ~ r, and Qit is the identity map of Et for all t. The kernel of an
inductive system is the subspace N of the direct sum .L {Et: t E A}
consisting of those f for which there is an index t in A such that
s ~ t whenever f(s) =I 0 and such that .L {Qts(f(s)): s ~ t} is the
zero of Et. The inductive Iimit, lim ind {Et: t E A}, is defined to
be the quotient space (_L {Et: t E A})/N. The term "inductive
Iimit" is justified by the fact that if B is a cofinal subset of A, then
lim ind {Et: t E B} is linearly isomorphic in a natural way to
lim ind {Et: t E A} (see problern ll).
There is a construction which is dual, in a certain sense, to that of
 SEC. 1 PROBLEMS 11
the inductive limit. A projective system (inverse system) of
linear spaces consists of the following: an index set A directed by a
partial ordering ~ : a linear space E1 for each t in A ; for each pair of
indices s and t with t ~ s, a canonicallinear transformation Pst from
E 1 into Es with the property that if t ~ s ~ r, then Prt = Prs o Ps 11
and the property that Pss is the identity transformation for all s in A.
The spaces E1 are to be thought of intuitively as approximations to a
limit space, the accuracy of the approximation increasing as the in-
dices increase. The canonical maps Pst relate the various approxima-
tions. The projective Iimit (inverse Iimit) of the system is the
subspace of the product X {E1 : t E A} which consists of all x such that
for every pair of indices t and s with t ~ s, Ps1(x(t)) = x(s). The
projective limit of the system is denoted by lim proj {E1: t E A}.
A simple example of a projective limit is the following. Suppose
that A is an index set for each of whose members t there is defined a
linear space E1 in such a way that the intersection of each pair of the
spaces contains a third. Then A can be directed by agreeing that
t ~ s if and only if E1 c Es. The resulting system is a projective
system if each Pst is taken to be the identity transformation. There
is clearly an algebraic isomorphism between the projective limit of
the system and the intersection of the spaces Et. The union of a
family of subspaces which is dirccted by :::) is isomorphic, in a dual
fashion, to an inductive limit.

PROBLEMS
A CARDINAL NUMBERS
If :F is the family of all finite subsets of an infinite set A, then k(:F) =
k(A). What is the cardinal number of the family of countable subsets
of A?

ß' QUOTIENTSAND SUBSPACES

(a) If F is a subspace of a linear space E, and G is any subspace of E
complementary to F, then G is isomorphic to EJF and hence dim G =
dim (EJF).
(b) If F and G are subspaces of a linear space E, then dim (F + G) +
dim (F n G) = dim F + dim G.

C DIRECT SUMS AND PRODUCTS

(a) Any linear sp~ce E is isomorphic to the direct sum ~ {K1: t E A}
where K 1 is the field K for each t and A is a Hamel base for E.
(b) Let E be the product space X {K": 1, 2, · · ·} where K" is the field K
for every n. For each a in K let x(a) be the element (a, a 2 , ••• , a", ... ),
and Iet Q = {x(a): 0 < a < 1}. Then Q is a linearly independent set of
12 CH. 1 LINEAR SPACES

cardinal 2Ko. Hence any space which is the product of infinitely many
non-trivial factor spaces necessarily has dimension at least 2Ko; thus there
are linear spaces which are not isomorphic to any product of copies of K.
(c) If E is a direct sum L: {Et: t E A} of linear spaces, then the algebraic
dual E' of Eis isomorphic to the product X{Et': t E A}. For any linear
space E, E' is isomorphic to X {Kt: t E A} where Kt is the field K for each
t and A is a Hamel base for E. In particular, if Eis finite dimensional then
E' and E are isomorphic. (Cf. 14.7.)

D SPACE OF BOUNDED FUNCTIONS

Let B( S) be the linear space of all bounded scalar valued functions on an
infinite set S and let a1 , a2 , • • • ES. For each a "# 0 such that Iai ; ; ; 1
define Ua by ua(x) = an for x = an and ua(x) = 0 otherwise. The family
{ua} is linearly independent. If s is the cardinal of Sand if.c is the cardinal
of the scalars, then the dimension of B( S) is c•.

E EXTENSION OF LINEAR FUNCTIONALS

If F is a subspace of a linear space E and f is a linear functional on F,
then there is a linear functional/ on E which coincides with f on F.

F NULLSPACESAND RANGES
Let T be a linear mapping of one linear space E into another linear space
F and let N be the null space of T. Then EjN is isomorphic to the
subspace T[E] of F.

G ALGEBRAIC ADJOINT OF A LINEAR MAPPING

If T is a linear mapping of a linear space E into another linear space F,
then the mapping T' that assigns to each element g of F' the element
g o T of E' is linear on F' to E'; it is called the algebraic adjoint of T. The
null space of T' consists of those elements of F' which vanish on T[E],
and T'[F'] consists of those elements of E' which vanish on the null space
ofT.

H SET FUNCTIONS
Let .sd be a family of sets in a linear space E, and assume that for each
subset A of E there exists a smallest member cf>(A) of .sd containing A.
Let A, Al> · · ·,An, and At for t in B be arbitrary subsets of E. Then
c/>(U {At: t E B}) = c/>(U {cf>(At): t E B}). If .sd is closed under scalar
multiplication, then, cf>(aA) = acf>(A). If .sd is closed under translation,
then cf>(L:{A 1 :i= 1,2, .. ·,n}) :::::> L:{cf>(A 1):i= 1,2, .. ·,n}. If .sd is
closed under addition, then cf>(L: {A 1 : i = 1, 2, · · · , n}) c L: {cf>( A 1): i =
1, 2, · · ·, n}.

I INDUCTIVE LIMITS (see 16C, 17G, 19A, 22C)

If B is a cofinal subset of A, then lim ind {Et: t E B} is isomorphic to
lim ind {Et: t E A}. Write F = L: {Et: t E A}, G = L: {Et: t E B}, and Iet
M and N be the respective kernels. Let J be the natural injection of G
 SEC. 2 CONVEXITY AND ÜRDER 13
into F, and Iet QM and QN be the quotient maps. Put I = QM o J o QN - 1 •
Then I is an isomorphism of GfN onto FfM. (First, J(N) c M, and this
shows that I is well-defined: I(x) is independent of the member of x chosen
in QN - 1 (x). Next, J(y) E M implies y E N, and hence I is one-one.
Finally, if z E FfM, choose w E QM -l(z). Take t so large that w(s) =F 0
implies s ~ t and so that t E B; define y E G by taking y(s) zero for all
s E B "' {t} and y(t) equal to the sum in Et of the images of all the values
of w. Then if x = QN(y), I(x) = z.)

2 CONVEXITY AND ORDER

This section begins with a few elementary propositions about convex
sets and circled sets. The two principal theorems of the section
establish a correspondence between certain convex sets and Minkowski
functionals (subadditive, non-negatively homogeneaus functionals),
and a correspondence between cones and partial orderings.
The line segment joining two points x and y of a linear space is
the set of all points of the form ax + by with a and b non-negative
real numbers such that a + b = 1, or equivalently, the set of all
points ax + (1 - a)y with a real and 0 ~ a ~ 1. This set is
denoted by [x:y]. The open line segment joining x and y, denoted
by (x:y), is the set of all points of the form ax + (1 - a)y with a real
and 0 < a < 1. The set (x:y) with the point x adjoined is denoted by
[x:y), and the set (x:y) with the point y adjoined is denoted by (x:y].
Asetin a linear space Eis convex if, whenever it contains x and y,
it also contains [x:y]. Clearly any subspace is convex, and so is any
translate of a convex set; a simple computation shows also that any
finite linear combination of convex sets is again convex. From the
definition it is obvious that the intersection of the members of any
family of convex sets is convex, and the union of the members of a
family of convex sets which is directed by ::> (the union of any two
members of the family is contained in some third member) is convex.
Since Eis convex, the family of all convex sets containing any particu-
lar set A in E is non-void, and the intersection of the members of this
family is the smallest convex set containing A. This intersection is
the convex extension (hull, envelope, cover) of A and will be
denoted by <A). It is easy to verify that the set of all finite linear
combinations .L {a 1x 1 : i = 1, · · ·, n}, where n is any positive integer,
each x 1 is in A, each a1 is real and non-negative, and .L {a1 : i = 1, · · ·, n}
= 1, is a convex set containing A. On the other hand, it follows
by a finite induction that, if A is convex, it contains all such combina-
tions. It is then clear that <A> is the set of all such combinations.
The following theorem states these and a few other simple facts.
14 CH. 1 LINEAR SPACES

2.1 CoMPUTATION RuLES FOR CoNVEX SETS Let E be a linear space.

Then:
(i) For any subset A of E the smallest convex set (A) which contains
A consists of allfinite linear combinations of the form 2 {aixi:
i = 1, · · ·, n}, where xi E A, ai is real and non-negative, and the
sum of the coefficients ai is one.
(ii) lf A and Bare non-void subsets of E and a and b are scalars,
then (aA + bB) = a (A) + b (B).
(iii) The intersection of convex sets t"s convex.
(iv) lf a family of convex sets is directed by :::J, then the union of the
members of the family is convex.
(v) lf A and Bare non-void subsets of E, then (A U B) = U {[x:y]:
x E (A), y E (B)}.

The union of all line segments with one end point in a set A and
the other in a set B is sometimes called the join of A and B. The
last proposition in the preceding Iist then implies that the join of two
non-void convex sets is convex, and is the convex extension of the
unwn.
A set A is circled if and only if aA c A whenever !a! ~ 1. A
circled set A has the property that aA = A whenever !a! = 1. In
particular, a circled set A is always symmetric, in the sense that
A = - A. It is easy to see that A is circled and convex if and only
if A contains ax + by whenever x and y are in A and Iai + !bl ~ 1.
In a real linear space, a set is convex and circled if and only if it is
convex and symmetric. The smallest linear subspace containing a
non-void convex circled set Ais simply the set U {nA: n = 1, 2, · · · }.
The smallest circled set containing a set A is called the circled
extension of A and is denoted by (A). Clearly (A) = U {aA:
!a! ~ 1}. The convex extension of a circled set is circled. The
smallest convex circled set containing A is ((A)), and this is precisely
the set of alllinear combinations 2 {a;x;: i = 1, · · ·, n} where X; E A
and 2 {!a;!: i = 1, · · ·, n} ~ 1.
A subset A of a linear space is called radial at a point x if and
only if A contains a line segment through x in each direction. More
precisely stated, A is radial at x if and only if for each vector y different
from x there is z, z -1= x, such that [x:z] c [x:y] n A. The radial
kernel of a set A is the set of all points at which A is radial. It
should be observed that, even in two-dimensional real Euclidean
space, a set•may be radial at a single point (see problern ZH). The
radial kernel of the radial kernel of A is usually quite different from
 SEC. 2 CONVEXITY AND ÜRDER 15
the radial kerne! of A. However, if A is convex, then the situation is
simpler.
2.2 THEOREM Let A be a convex subset of a linear space E. If x is
a member of the radial kerne! of A and y E A, then the half-open interval
[x:y) is contained in the radial kerne! of A.
Consequently the radial kerne! of a convex set is convex and is its own
radial kerne/.
PROOF Let v = ax + (1 - a)y, with 0 < a ~ 1. For any fixed z
in E there is a positive number r such that x + bz E A whenever
0 < b < r. Since A is convex, a(x + bz) + (1 - a)y = v + abz E
A for all b such that 0 < b < r. Hence v + cz E A if 0 < c < ar,
and it follows that v is in the radial kerne! of A-I II
It is worth observing that if f is a linear map of E onto a vector
space F, and if a subset A of E is radial at x, then the set f[A] is
radial at f(x). In particular, if f is a linear functional which is not
identically zero and A is radial at x, then f[A] is radial at f(x); and
if A is convex and radial at each of its points, then f [A] is open.
Certain convex sets which are radial at 0 can be described by
means of real valued functions. If U is a set which is radial at 0,
the Minkowski functional for U is defined to be the real valued
function p defined on E by p(x) = inf {a: ~ x EU, a > 0}. A
a
simple computation shows that p(ax) = ap(x) whenever a ~ 0; that
is, p is non-negatively homogeneous. If U is convex, then p is
subadditive; that is, p(x + y) ~ p(x) + p(y). If U is circled, then
p(ax) = lalp(x); that is, p is absolutely homogeneous. A non-
negative functional on E which is absolutely homogeneaus and sub-
additive is called a pseudo-norm.

2.3 CONVEX SETS AND MINKOWSKI FUNCTIONALS

(i) If U is a convex set which is radial at 0, and if p is the Minkowski
functional for U, then {x: p(x) < 1} is the radial kerne! of U and
U C {X: p( X) ~ 1}.
(ii) If p is a non-negative, non-negatively homogeneaus subadditive
functional on E, and if V = {x: p(x) ~ 1}, then Visa convex set
which is radial at 0, and p is the Minkowski functional for V.
Moreover, V is circled zf and only zf p is a pseudo-norm.
The proof is a Straightforward verification and is omitted. If p is a
pseudo-norm, then the set of all vectors x such that p(x) = 0 is a
linear subspace. In case this subspace consists simply of 0, the
16 CH. 1 LINEAR SPACES

pseudo-norm is called a norm. Formally, a norm is an absolutely

homogeneous, subadditive, non-negative functional p such that
p(x) = 0 only if x = 0. When only one norm p is being considered,
the notation JJxJJ is frequently used for p(x).
The final theorem of this section relates the notion of partial
ordering of a linear space E to a geometrical object in E. Suppose
that ~ is a partial ordering of a reallinear space E; that is, ~ is a
relation such that x ~ x for each vector x of E and such that x ~ z
whenever x ~ y and y ~ z. Let us suppose further that the ordering
is translation invariant, in thesensethat x + z ~ y + z whenever
x ~ y. In this case the ordering is determined by the set P of all
vectors x such that x ~ 0, for then x ~ y if and only if x - y ~ 0.
Conversely, if Pis any subset of E suchthat 0 E P and P + P c P,
then a partial ordering which is translation invariant is defined by
agreeing that x ~ y if and only if x - y E P; moreover, Pis precisely
the set of vectors which are greater than or equal to 0 relative to this
ordering. Finally suppose that ~ is a translation invariant partial
ordering which also has the property that if x ~ y, then ax ~ ay for
all non-negative scalars a. Then the set P of non-negative vectors
(that is, {x: x ~ 0}) has the properties: P + P c P and aP c P for
each non-negative scalar a. It is evident that, conversely, a set P
with these two properties is precisely the set of all non-negative
vectors relative to the ordering: x ~ y if and only if x - y E P.
lt remains to formalize the discussion of the preceding paragraph.
A vector ordering of E is a partial ordering ~ such that x + z ;;;
y + z whenever x ~ y and ax ~ ay whenever x ~ y and a is a non-
negative scalar. A cone in E is a non-void subset P such that
P + P c P and aP c P whenever a is a non-negative scalar. If ~
is a vector ordering, then {x: x ~ 0} is a cone, and this cone is called
the positive cone of the ordering. If P is a cone in E, then the
corresponding order is defined by setting x ~ y if and only if
x - y E P. The discussion above can be summarized, in this
terminology, as follows.
2.4 CoNES AND ÜRDERINGS Each vector ordering is precisely the
ordering corresponding to the cone of non-negative vectors. Each cone
Pis the positive cone of the vector ordering corresponding to P.
The section is concluded with a few remarks about cones. A cone
is always convex, and in fact a non-void set P is a cone if and only if
it is convex and aP c P for each non-negative scalar a. Each linear
subspace is a cone, and if P is a cone in a real linear space, then
 SEc. 2 PROBLEMS 17

P - Pis a linear subspace. If Pis an arbitrary cone in a reallinear

space, then it is not generally true that x = y if x ~ y and y ~ x.
This condition is equivalent to the requirement that P n (- P) =
{0}. The set P n (- P) is always a reallinear subspace. If A is an
arbitrary non-void subset of E, then there is a smallest cone which
contains A; this cone is generated by A, and is called the conical
extension of A.

PROBLEMS
A MIDFOINT CONVEXITY
A subset A of a linear space E is midpoint convex iff -!(x + y) is in A
whenever x and y are in A.
(a) The following conditions are equivalent:
(i) A is midpoint convex;
(ii) A + A = 2A;
(iii) a1 A + a 2 A + · · · + anA = A whenever a 1 , a 2 , · · ·, an is any
finite collection of non-negative dyadic rationals whose sum is 1.
(b) Amidpoint convex set of real numbers is convex if it is either open
or closed.

ß DISJOINT CONVEX SETS

(a) (Kakutani's Iemma) If A and Bare disjoint convex sets and x is a
point not in their union, then either <A U {x}) and B are disjoint or else
<B U {x}) and A are disjoint.
(b) (Stone's theorem) If A and B are disjoint convex sets in a linear
space E, there are disjoint convex sets C and D such that A c C, B c D,
and E =CU D.

C MINKOWSKI FUNCTIONALS
(a) Let u be a convex setradial at 0 and assume that u = n{ut: t E A}
where each U1 is convex. If p and Pt are Minkowski functionals for U
and ut, then p(x) = sup {Pt(x): t E A}.
(b) If U and V are radial at 0 and if p and q are the Minkowski functionals
for U and V respectively, then inf {p(v) + q(v- x): v E V} is the Min-
kowski functional for the join of U and V.
(c) Let p be the Minkowski functional for a convex set U radial at 0.
Then {x: p(x) > 1} is the radial kerne! of the complement of U.

D CONVEX EXTENSIONS OF SUBSETS OF FINITE DIMENSIONAL SPACES

(a) Areallinear space E has finite dimension ~n if and only if for each
subset A of E and for each point x of <A), there is a subset {x 1 , · · ·, xn+ 1 }
of A such that x belongs to the convex extension of {x 1 , · · ·, Xn+l}.
(b) Each point of the convex extension of a connected subset A of
Euclidean n-space is in the convex extension of a subset {x 1 , · · ·, xn} of A.
18 CH. 1 LINEAR SPACES
E CONVEX FUNCTIONALS
For a real-valued function p on a linear space E with p(O) = 0, any two
of the following statements imply the third:
(i) p is non-negatively homogeneous;
(ii) p is subadditive;
(iii) p is convex, that is, p(L, {a1x1 : 1 ;;;i i ;;;i n}) ;;;i L, {a1p(x1): 1 ;;;i i ;;;i n}
for any finite combination with each a 1 non-negative and L, {a1 : 1 ~
i ;;;i n} = 1.
F FAMILIES OF CONES
The intersection of an arbitrary family of cones is again a cone; if the
family is directed by c, then the union of the members is also a cone.
Any finite linear combination of cones is a cone. If, for any set A, C(A)
temporarily denotes the conical extension of A, then C(A) = U {a<A):
a ~ 0}; for any two sets A and B and scalars a and b, C(aA + bB) c
aC(A) + bC(B); and C(U {Ai: t E B}) = C(U {C{A1): t E B}) for any
family {A 1 : t E B} of sets.
G VECTOR ORDERINGS OF R2
There are ten essentially different vector orderings of R 2 •
ß RADIAL SETS
In R 2 there exists a set which is radial at only one point.

I Z - A DICTIONARY ORDERING
Let B be a Harne! base for an infinite dimensional real linear space E
and Iet B be weil ordered without the maximum element. Define an
ordering of E by: x ~ 0 iff x = 0 or, on writing x as a linear combination
of members of B, its last (relative to the ordering of B) non-zero coefficient
is positive. Then E is linearly ordered by ~ , and ~ is a vector ordering.
Moreover, for each x in E there is y in E such that y ~ 0 and x + ay ~ 0
for every positive number a.
J HELLY'S THEOREM
In Rn Iet {A 1 : i =-= 1, · · ·, r}, for r > n + 1, be convex setssuchthat for
n n
{A,: i =F k} is non-void. Then {A: i = 1, · · ·, r} is also non-
each k,
n
void. (Let xk E {A;: i =F k}. It is possible to choose r numbers IX;, not
all zero, such that L, {1X1 : 1 ~ i ~ r} = 0 and L, {1X1x1 : 1 ~ i ~ r} = 0.
Separate the terms having IX; ~ 0 and those for which IX; < 0.)
Let C(f be a family of compact (see 4) convex subsets of Rn such that
every n + 1 sets in C(f have a non-void intersection. Then the intersection
of all the members of C(f is non-void.

3 SEPARATION AND EXTENSION THEOREMS

This section contains the fundamental separation and extension
theorems. These theorems are essential to the study of duality.
Sufficient conditions are given for the separation of convex sets by a
hyperplane, and for the extension of a linear functional from a sub-
space to the entire space, preserving positivity or preserving a bound.
 SEC. 3 SEPARATION AND EXTENSION THEOREMS 19
There are a number of theorems on separation of convex sets and
extension of linear functionals which, although superficially different,
are more or less mutually equivalent. The presentation given here
begins with a theorem about cones and derives the other results as
corollaries, but this arrangement is primarily a matter of taste. Most
of the theorems of the section concern real linear spaces, and it will
be assumed that the linear space under discussion is real unless
explicitly stated otherwise.
If f is a non-identically-zero linear functional on a reallinear space
E and t is any real number, each of the sets {x: f(x) ;?; t} and
{x:f(x) ~ t} is a half-space, and the pair are complementary half-
spaces. Their intersection, the set f -l[t], is a linear manifold. It
is the translate of the null space ofjby any vector x such thatf(x) = t.
The null space of f is evidently a maximal proper subspace of E; each
translate of a maximal proper subspace of a linear space is called a
hyperplane. Each hyperplane is a linear manifold, and is evidently
contained in no other linear manifold except the space E itself. Each
hyperplane is the intersection of two complementary half-spaces.
A proper cone in Eis a cone which is non-empty and is not identical
with E. Recall (theorem 2.2) that the radial kernel of a convex set is
convex, and that the half-open line segment (x:y] joining a point x of
a convex set to a point y of its radial kernel is always contained in the
radial kernel.
A half-space which is a cone is called conical. Evidently P is a
conical half-space if and only if for some linear functional J, not
identically zero, P = {x: f(x) ;?; 0}.
3.1 LEMMA Let E be a real linear space. A proper cone P in E is a
half-space ij and only ij it has a non-void radial kerne/ P 0 and the union
of the sets P and - P 0 is the entire space.
PROOF If P is a half-space, then clearly the radial kernel P0 of P is
non-void and E = P U (- P0 ). Conversely, suppose P0 satisfies
the latter conditions. The sets - P0 and P are disjoint, for if - x E P 0
and x t: P, then 0 belongs to the half-open line segment [- x:x) and
is comequently in the radial kernel of P; but then P is not a proper
cone. It follows that P 0 , - P0 , and P n (- P) are mutually disjoint,
and that the union of these three sets is the entire space.
The intersection F = P n (- P) is a linear subspace, and the
proof will be completed by showing that Fis of Co-dimension one.
lf this is shown, then it will follow that there is a non-identically-zero
linear functional f whose null space is F, f will map P 0 onto a convex
20 CH. 1 LINEAR SPACES
subset of the non-zero real numbers, and P must then be identical
with one of the conical half-spaces {x:f(x) ~ 0} or {x:f(x);;;; 0}.
The proof then reduces to showing that, if x is a fixed point of P 0 and
y an arbitrary point which is distinct from x, then y is a linear com-
bination of x and some member of F. If y E - P 0 , then the line
segment [x :y] must intersect F, for each of P 0 and - P 0 is radial at
every one of its points, and the non-void disjoint open sets {t: tx +
(1- t)yEP0 } and {t: tx + (1- t)yE -P0 } cannot cover the unit
interval. Hence for each member y of - P0 there isanurober a such
thaty + ax E F. Finally, if y EP0 , then -y E -P0 and -y + ax E F
for some scalar a; and if y belongs to neither P 0 nor - P 0 , then
yEF.JJJ
3.2 THEOREM In a real linear space E each proper cone which is
radial at some point is contained in some conical half-space.
PROOF Let C be a proper cone which is radial at a point x. Then the
vector - x does not belong to C, for in this case 0 would be in the
radial kernel of C and C would not be a proper cone. The cone C is
contained in a cone P which is maximal with respect to the property
of not containing - x, in view of the maximal principle. It will be
shown that P is a conical half-space.
Suppose y tf: P. The set of all points ay + p, for a ;;;; 0 and p in P,
is a cone which properly contains P, and the maximality property of
P implies that ay + p = - x, for some a and some p. Clearly
a # 0, and since (- aj2)y belongs to the line segment (p:x], it
follows that P is radial at (- aj2)y and hence at - y. Hence, if
y tf: P, then - y is a member of the radial kernel P 0 of P, and the
preceding lemma shows that·P is a conical half-space.JJI
The principal theorem on the extension of positive functionals is
easily deduced from the preceding. As a preliminary, notice that if
P is a cone and f is a non-identically-zero linear functional which
is non-negative on P, then f(x) > 0 for each point x at which P is
radial, for f[P] is radial at f(x).
3.3 EXTENSION OF PosiTIVE FuNCTIONALS Let P be a cone in a real
linear space E, and Iet F be a linear subspace which intersects the radial
kerne/ of P. Then each linear functional f on F which is non-negative
on P n F can be extended to a linear functional on E which is non-
negative on P.
PROOF If f is identically zero, the extension is clearly possible. If
not, let x be a fixed point of F which belongs to the radial kernel of P,
 SEC. 3 SEPARATION AND EXTENSION THEOREMS 21
and observe that f(x) > 0. If A = {y:f(y) ~ 0}, then P + A is
evidently a cone; and since f (- x) < 0, it follows that - x f/= P + A
and consequently P + A is a proper cone which is radial at x. Hence
by theorem 3.2 there is a conical half-space containing P + A; that
is, there is a non-identically-zero functional g which is non-negative
on P + A. If f(y) = 0, then both y and - y belang to A and hence
g (y) = 0. Consequently g (y) = af(y) for some scalar a and for all
y in F. Finally, both f(x) and g(x) are positive; therefore a > 0,
and (1/a)g is the required extension of J.lll
The following theorem is the first, historically, of the theorems of
the section; it is probably the most immediately applicable form of
the extension-separation principle.
3.4 HAHN-BANACH THEOREM Let E be a real linear space, Iet F be
a subspace, Iet p be a subadditive non-negatively homogeneous functional
on E, and Iet f be a linear functional on F suchthat f(x) ;;;:; p(x) for all
x in F. Then there is a linear functional f- on E, an extension of j,
suchthat f -(x) ;;;:; p(x) for all x in E.
PROOF Consider the linear space E x K, where K is the real field,
and let P be the set of all vectors (x,t) which are "above" the graph
of p; explicitly, P = {(x,t): t ~ p(x)}. It is easy to see that P is a
cone and that P is radial at (0,1). For (x,t) in F x K let g(x,t) =
t - f(x). Then g is nor.-negative on P n (F x K), the extension
theorem 3.3 can be applied, and there is consequently an extension g-
of g which is non-negative on P. Then for t ~ p(x) it is true that
g-(x,t) = g-(x,O) + g-(O,t) = g-(x,O) + t ~ 0, and in particular,
g-(x,O) + p(x) ~ 0 for all x. It follows that f-(x) = -g-(x,O) is
the required extension of J.lll
There is a geometric form of the preceding theorem which is
frequently useful. The connection between the preceding and the
geometric statement is based on this fact: if A is a convex subset of a
reallinear space and A is radial at 0, then a linear functional is less
than or equal to one on A if and only if f(x) ;;;:; p(x) for all x, where p
is the Minkowski functional for A. (Recall that p(x) = inf {t: t > 0
and (xft) E A}, so that f(x) ;;;:; p(x) for all x if and only if f(x) ;;;:; t for
all positive t such that (xft) E A; that is, f(xft) ;;;:; 1 for xft in A.)
Theorem 3.4 can then be rewritten in the following form.
3.5 CoROLLARY Let E be a real linear space, F a subspace, A a
convex subset of E which is radial at 0, and Iet I be a linear functional
on F which is at most one on F n A. Then there is a linear lunctional
I- on E, an extension ol j, such that I- (x) ;;;:; 1 lor all x in A.
22 CH. 1 LINEAR SPACES
The preceding result irnplies a theorern concerning the extension
of linear functionals on cornplex linear spaces. To obtain the
following, notice that if.f is a linear functional, then the suprernurn of
lf(x)l for x in a convex circled set is the sarne as the suprernurn of the
real part of f(x). The following proposition is a consequence of this
rernark and of theorern 1.4.
3.6 CoROLLARY Let E be a real or complex linear space, F a linear
subspace, A a convex circled set which is radial at 0, and Iet I be a linear
lunctional on F such that ll(x)l ~ 1 lor x in A n F. Then there is a
linear lunctionalf- on E, an extension ofl, suchthat 11-(x)l ~ 1 on A.
The Minkowski functional of a convex circled set which is radial at
0 is a pseudo-norrn. The preceding result can then be restated as
follows.
3.7 COROLLARY Let E be a real or complex linear space, F a linear
subspace, p a pseudo-norm, and I a linear lunctional on F such that
ll(x)l ~ p(x) lor x in F. Then there is a linear lunctional 1- on E,
an extension of1. such that II -(x) I ~ p( x) lor all x.
Two subsets, A and B, of a real linear space can be separated if
there are cornplernentary half-spaces which contain A and B respec-
tively. A linear functional I is said to separate A and B if and only
if I is not identically zero and sup {f(x): x E A} ~ inf {f(x): x E B}.
Clearly I separates A and B if and only if -I separates B and A.
The linear functional I strongly separates A and B if the inequality
above is strong; that is, if sup {f(x): x E A} < inf {f(x): x E B}. A
linear functional g on a cornplex linear space is said to separate
(strongly separate) two sets if and only if the real part of g separates
(strongly separates, respectively) the sets.
The problern of separating (or strongly separating) two sets can
always be reduced to the problern of separating (strongly) a point
and a set. Explicitly, a linear functional I separates A and B if and
only if I separates {0} and B - A, and I separates A and B strongly if
and only if I strongly separates {0} and B - A.
3.8 SEPARATION THEOREM Let A and B be non-void convex subsets
ol a complex or real linear space E and suppose that A is radial at some
point. Then there is a linear lunctional I separating A and B if and
only if Bisdisjoint lrom the radial kerne[ ol A.
PROOF There is no loss in generality in assurning E is a real linear
space. If f separates A and B and A 0 is the radial kernel of A, then
 SEC. 3 PROBLEMS 23
f[A 0 ] is an open subset of the space of allreal numbers and is disjoint
from f[B], and consequently A 0 is disjoint from B. On the other
band, suppose that A 0 is disjoint from B. In view of theorem 3.2
there will be a linear functional f separating A and B provided the
cone consisting of all non-negative multiples of B - A is a proper
subset of the entire space E. Assuming that U {t(B - A): t ~ 0} = E,
choose a member x of A 0 and a member y of B, and then
choose u in A, v in B, and t ~ 0 so that -(y- x) = t(v - u).
Then x + tu= y + tv and hence [1/(1 + t)]x + [t/(1 + t)]u =
[1/(1 + t )]y + [t/(1 + t )]v. That is, the line segment [x:u) inter-
sects the segment [y:v). This is a contradiction, for [x:u) c A 0 ,
and A 0 was supposed to be disjoint from B.lll
3.9 THEOREM ON STRONG SEPARATION Two convex non-void subsets,
A and B, of a complex or reallinear space E can be strongly separated by a
linear functional if and only if there is a con'vex set U which is radial at
0 suchthat (A + U) n Bis void.
PROOF It may be assumed that E is a reallinear space. If f strongly
separates A and B, and inf {f(y): y E B} - sup {f(x): x E A} =
d > 0, then the inverse under f of the open interval (- d/2 :d/2) is the
required set U. To prove the converse, assume that U is convex
and radial at 0 and that (A + U) n Bis void. Then 0 f/; B - A - U,
and B - A - U has a non-void radial kernel. The preceding
separation theorem applies, and there is a non-identically-zero linear
functional f such that f is non-negative on B - A - U. Then
f(y) - f(x) ~ f(z) for y in B, x in A, and z in U. The set J[U] is
a neighborhood of 0 in the space of real numbers, and it follows that
inf {f(y): y E B} - sup {f(x): XE A} > 0.111

PROBLEMS
A SEPARATION OF A LINEAR MANIFOLD FROM A CONE
If F is a linear manifold in a real linear space E and P is a cone in E,
the radial kernel of which is non-void and disjoint from F, then there
exists a linear functional f on E and a non-positive constant a such that
f(x) = a for every x in F,f(x) ~ 0 for each x in P, andf(x) > 0 for each x
in the radial kernel of P.

ß ALTERNATIVE PROOF OF LEMMA 3.1

Suppose Pisa proper cone in areallinear space E, the radial kernel P 0
of P is non-void, and P U ( -P0 ) = E. Let F = P n (-P), and con-
sider the quotient space EfF. There is in EfF an ordering that is induced
by the ordering in E generated by P; this ordering in EfF is linear and
24 CH. 1 LINEAR SPACES

Archimedean, and hence EfF has dimension 1, so that Fis the null space
of a linear functional that does not vanish identically.

C EXTENSION OF THEOREM 3.2

In theorem 3.2 the hypothesis can be weakened by assuming that the
given cone is radial at some point with respect to its linear extension.

D EXAMPLE
Not every proper cone is contained in a half-space. (See 2I.)

E GENERALIZED HAHN-BANACH THEOREM

Let A be a convex set in a real linear space E and suppose that p is a
convex functional on A (that is, p(_L {a;X;: 1 ;;::;; i ;;::;; n}) ; : ; L {a;p( X;): 1 ;;::;;
i;;::;; n} whenever X;EA, a; ~ 0, and _L{a,: 1;;::;; i ~ n} = 1). Let F be a
subspace of E and f a linear functional of F such that f ; : ; p on A n F.
Then if A c F - C where C ;s the conical extension of A, there isalinear
functional f- on E such that f- is an extension of fand f- ; : ; p on A.
This proposition obviously implies a dual statement, in which p is a
concave functional and p ; : ; f on A n F by hypothesis; then if A c F - C
there isalinear functionalf- which extendsf and which dominates p on A.

F GENERALIZED HAHN-BANACH THEOREM (VARIANT)

Let A be a circled convex subset of a (real or complex) linear space E,
and Iet F be a subspace of E. Let f be a linear functional of F and Iet p
be an absolutely convex functional on A (that is, p(_L {aixi: 1 ;;::;; i ; : ; n}) ; : ;
L {iadp(xi): 1 ;;::;; i ; : ; n} whenever xi E A and .L {lad: 1 ;;::;; i ;;::;; n} = 1). If
lf(x)J ;;::;; p(x) whenever x is in A n F, and if A c F - C, where Cis the
conical extension of A, then there exists a linear functional f- on E which
is an extension of f and is such that If- (x) I ;:;:; p( x) for every x in A.
NoTE: The condition that A be contained in F- C used in problems
3E and 3F is satisfied in case A = E or, more generally, in case Cis radial
at some point in F. This, combined with the remark that any constant
function is convex and concave, and if non-negative is also absolutely
convex, yields a variety of corollaries to the above two theorems, including
propositions 3.3 through 3. 7 of the text.

G EXAMPLE ON NON-SEPARATION
In R 3 consider the closed cone defined by the equations x ~ 0, y ~ 0,
z 2 ;;::;; xy, and the line having the equations x = 0 and z = 1. There is no
plane in R 3 which contains the line and has the cone lying on one side of it;
yet the line is disjoint from the cone.

H EXTENSION OF INVARIANT LINEAR FUNCTIONALS

Let E be a linear space, F a subspace of E, f a linear functional on F,
and p a non-negatively homogeneous, non-negative, subadditive real
functional on E such that f ; : ; p. In addition, Iet 2 1 be a farnily of linear
operators in E.
 SEC. 3 PROBLEMS 25
(a) Foreach x in E Iet g(x) = inf {p(x + L: {R1(y1): 1 ~ i ~ k}): y 1 E E,
R1 E .fE 11 k a positive integer}. Then g is a non-negatively homogeneous,
non-negative, subadditive functional on E and g ~ p.
(b) The inequality f(x) ~ g(x) holds for each x in F if and only if there
is an extension f of f such that f ~ p on E and fo R = 0 for each R in .fE1 •
(Apply the Hahn-Banachtheorem to fand g.)
For present purposes a family .fE of linear operations in E is admissible
if S [ F] c F and f oS = f for every S in .fE; a linear extension 1 of f is .fE-
admissible if/ ~ p on E andfoS = /for every S in !E.
(c) Suppose .fE is an admissible family of linear operators; then the
following are equivalent:
(i) there exists a linear extension 1of f that is .fE -admissible,
(ii) for each finite subset !E' of .fE there exists a linear !L'' -admissible
extension of f,
(iii) f(x) ~ p(x -r L: {Si(Y1) - y 1 : 1 ~ i ~ k}) for x in F, y 1 in E, and
S1 in !L'.
 Chapter 2

LINEAR TOPOLOGICAL SPACES

This chapter is largely preliminary in nature; it consists of a brief

review of some of the terminology and the elementary theorems of
general topology, an examination of the new concept "linear topo-
logical space" in terms of more familiar notions, and a comparison
of this new concept with the mathematical objects of which it is an
abstraction. After an introductory section on topology, we consider
linear topological spaces, subspaces, quotient spaces, product spaces,
and linear functions. With the exception of a few simple proposi-
tions relating to circled sets, these theorems are specializations of
familiar results on topological groups (in other words, little use is
made of scalar multiplication).
In the third section we compare an arbitrary linear topological
space with its antecedents: normed spaces and metrizable spaces.
This is an example of a classical procedure. Each new mathematical
construct is compared with the examples from which it derives (in
this case, normed spaces, metrizable spaces, and products of these) in
order to determine the extent to which the central ideas of the ex-
amples have been isolated. As frequently happens, this comparison
is made by representing the new object in terms of the old, and
embedding theorems for arbitrary linear topological spaces in products
of metrizable (or in some cases, products of normable spaces) result.
lt is noteworthy that two new notions, that of bounded set and that
of locally convex topology, arise. These notions lie deeper than the
theorems of this chapter indicate. They signal the beginning of the
theory of linear topological spaces, in distinction to the theory of
topological groups (in other words, we are beginning to utilize scalar
multiplication ).
26
 SEc. 4 ToPOLOGICAL SPACES 27
The last two sections, on completeness and on function spaces
respectively, are essentially fragments of elementary analysis (that is,
general topology). With a few exceptions, the results listed there
are specializations of theorems concerning topological groups or, even
more generally, uniform spaces.

4 TOPOLOGICAL SPACES
This section reviews 1 a number of topological definitions and results,
the most noteworthy being the Tychonoff product theorem.
A topology for a set X is a family !7 of subsets of X, to be called
open sets, suchthat the void (empty) set, the set X itself, the union
of arbitrarily many open sets, and the intersection of finitely many
open sets, are open. If several topologies for X are being con-
sidered, the members of !7 will be called !7-open, or open relative
to !7, and a similar convention will hold for other topological objects
to be defined. The set X with the topology !7 is a topological
space; it is denoted by (X,!/). A subset A of Xis closed if and
only if the complement X "' A of A is open. The intersection of all
closed sets containing a set A is a closed set called the closure of A ;
the closure of A is denoted by Ä or A-. A subset B of A whose
closure contains A is dense in A. The union of all open sets con-
tained in A is an open set called the interior of A and denoted by A 1•
A set U is a neighborhood of a point x if and only if x belongs to the
interior of U. The family IJI/x of neighborhoods of x is the neighbor-
hood system of x, and a subfamily fJIJ of IJI/x suchthat each member
of IJI/x contains a member of fJIJ is a base for the neighborhood
system of x. A topological space is regular if and only if for each
point x, the family of closed neighborhoods of x is a base for the
neighborhood system of x. Thus, X is regular if and only if for each
x in X and each neighborhood U of x there is a closed neighborhood
V of x such that V c U. A topological space X is separable if
and only if there exists a countable subset A of X which is dense in X.
If X and Y are topological spaces, then Y is a topological sub-
space of X if and only if Y c X and the sets which are open in Y
are precisely the intersections of Y with open subsets of X. Equiva-
lently, a subset of Y is closed in Y if and only if it is the intersection
of Y and a elosed subset of X. If Z is a subset of X, then Z may be
given a topology such that it becomes a topological subspace of X.
The relative (induced) topology on Z, or the relativization of the
1 For more complete treatment of the topics in this section, see Kelley [5].
28 CH. 2 LINEAR ToPOLOGICAL SPACES

topology of X to Z, is the family of all subsets U of Z such that for

some open subset V of X, U = Z n V. Clearly Z with the relative
topology is a topological subspace of X. A subset of Z which is
open with respect to the relative topology is open in Z.
If (X,iT) and ( Y, cff) are topological spaces, and if f is a map of X
into Y, then f is continuous at a point x of X if and only if the
inverse image of each cft-neighborhood of f(x) is a 3 -neighborhood
of x. The function f is continuous if and only if f is continuous at
each point of X, or, equivalently, if and only if the inverse image
of each 0lt-open set of Y is 3-open. If A is a subset of X, then the
function f is continuous on A if and only if the restriction of f to A
is continuous as a function on the topological subspace A. The
function f is open (interior) if and only if the image under f of each
open set is open. The function f is relatively open if and only if
the image of each open set is open inj[X], that is, open in the relative
topology for j[X]. Finally, the function f is topological, or a
homeomorphism, if and only if it is continuous, one-to-one (no
two points have the same image) and the inverse map, f- I, is a
continuous map of j[X] onto X; equivalently, f is topological if and
only if it is one-to-one, continuous, and relatively open.
A generalization of sequential convergence is necessary; Moore-
Smith convergence will be employed. A relation ~ directs a set A
if it is transitive on A (if a ~ ß and ß ~ y, then a ~ y, for a, ß, and y
in A), reflexive on A (a ~ a for a in A), and has the property: for a
in A and ß in A there is y in A such that y ~ a and y ~ ß. A net
isapair {x, ~}such that x is a function and ;;; directs the domain of x.
More generally, {xw a E A, ~ }, or simply {xa, a E A}, is called a net
if x is a function whose domain contains A and ~ directs A. A net
{xw a E A} is eventually in a set B if and only if, for some a in A,
x 8 E B whenever ß ~ a. A net in a topological space converges to
a point (the point is a Iimit of the net) if and only if the net is even-
tually in each neighborhood of the point. The fact that the net {xa,
a E A} converges to y in X will be denoted by Xa --+ y or y = lim Xa.
A point is a duster point of a net if and only if the net is not eventu-
ally in the complement of any neighborhood of the point (in the usual
terminology, the net is frequently or repeatedly in each neighbor-
hood of the point). A net {y 8 , ß E B, ~} is a subnet of {xa, a E A, ~}
if and only if there exists a function n on B to A such that n is order
preserving, y 8 = Xncß) for all ß in B, and for each a in A there is aß
in B such that n(ß) ~ a. If a net converges to a point z, each subnet
also converges to z; a point z is a Iimit of a subnet if and only if z is
 SEc. 4 ToPOLOGICAL SPACES 29
a duster point of the net. Note that a subnet of a sequence (a
sequence is a function on the set of positive integers with the usual
ordering) need not be a sequence, although a subsequence of a
sequence is a subnet of this sequence. A function f is continuous at
a point x' of the domain space if and only if for each net {xa, cx E A}
converging to x', the net {f(xa), cx E A} converges to f(x') in the
range space.
A topological space is Hausdorff (separated, T2 ) if and only if
for distinct points x and y there are disjoint neighborhoods U of x
and V of y. Equivalently, a space is Hausdorff if and only if no net
in the space converges to more than one point. A topological space
is compact (bicompact) if and only if it has the Borel-Lebesgue
property: each cover by open sets has a finite subcover; equivalently,
a topological space is compact if and only if each net has a duster
point. A space is countably compact if and only if each countable
cover by open sets has a finite subcover. A space is countably
compact if and only if each sequence has a duster point. A space is
sequentially compact (quasi-compact) if and only if each sequence
has a subsequence which converges to a point in the space. Although
countable compactness is implied by the two other kinds of compact-
ness, it implies neither of these, nor does either of these imply the
other. A space is locally compact if and only if each point has a
compact neighborhood. Each locally compact Hausdorff space is
regular.
A pseudo-metric (semi-metric, ecart) on a set X is a non-
negative function d, defined for each pair of points of X, such that
d(x,y) = d(y,x), d(x,x) = 0, and d(x,z) ;;:;; d(x,y) + d(y,z) for all x, y,
and z in X. A metric on Xis a pseudo-metric d suchthat if d(x,y) =
0, then x = y. A set X with a pseudo-metric (metric) is called a
pseudo-metric (respectively, metric) space. In a pseudo-metric
space the open (respectively, closed) sphere of radius r about a
point x is the set of all points y such that d(x,y) < r (respectively,
d(x,y) ;;:;; r). Each pseudo-metric (metric) d on X generates a topol-
ogy for X in the following fashion. A set U in X is open relative to
the pseudo-metric (respectively, metric) topology for x if and only
if for each point x in U there is a positive number r such that the open
sphere of radius r about x is contained in U. A topological space X
is pseudo-metrizable (respectively, metrizable) if and only if there
is a pseudo-metric (metric) such that the pseudo-metric (metric)
topology is the topology of X. If X is pseudo-metrizable, there are,
in general, many pseudo-metrics on X such that the pseudo-metric
30 CH. 2 LINEAR ToPOLOGICAL SPACES

topology is the topology of X; consequently, the pseudo-metric

topology does not determine the pseudo-metric. Clearly, a pseudo-
metrizable topological space is Hausdorff if and only if the space is
metrizable.
lf Xis the set of real or complex numbers and d(x,y) = lx - Yl for
x and y in X, then d is a metric on X. U nless specifically stated
otherwise, it will be assumed that the field of complex numbers and
the field of real numbers are assigned the topology of the metric d.
It is called the usual topology for the complex (respectively, real)
numbers.
The comparison of topologies is of interest. If .r and 0/1 are two.
topologies for a set X such that each 0/1-open set is .r-open, then the
topology .r is stronger (larger, finer) than o/1, and 0/1 is weaker
(smaller, coarser) than ff. Equivalently, .r is stronger than 0/1 if
and only if ff :::> 0//. More precise expressions would be ".r is at
least as strong as 0/1" and "0/1 is at least as weak as ff," but one
usually foregoes such grammatical exactness. (Some authors give
the terms " stronger " and " weaker " a significance exactly the
reverse of that defined here, but the definitions given are those which
are most common in the Iiterature of linear topological spaces.) lf a
net converges relative to a topology, it also converges relative to
weaker topologies, and a set compact relative to a topology is compact
relative to weaker topologies. Topologies for a set are in general not
comparable, but the collection of all topologies for a set X is a lattice
under the ordering c ; the strongest topology is the discrete topol-
ogy, for which all sets are open, and the weakest topology is the
trivial (indiscrete) topology, for which only X and the void set
are open.
There is a natural way to topologize the cartesian product X {Xt:
t E A} of topological spaces xt. Recall that for each t in A, the pro-
jection Pt of the product into the t-th coordinate space Xt is defined
by Pt(x) = x(t) for each x in the product, and that Pt(x) is the t-th
coordinate of x. For convenience, a cylinder will be d~fined to be a
subset of the product of the form Pt - 1 [ U], where U is open in Xt; that
is, a set C is a cylinder if and only if for some t in A and some open
subset U of X 11 Cis the set of all x in the product such that x(t) E U.
The product topology for X {Xt: t E A} is the family of all arbitrary
unions of finite intersections of cylinders; the product topology is
sometimes called the product of the topologies for Xt. A base for
the neighborhood system of a point x, relative to the product topology,
is the family of all finite intersections of cylinders containing x; that
 SEc. 4 ToPOLOGICAL SPACES 31
is, the family of all sets of the following form: the set of all y such that
y(t) E U 1 for each t in a finite Set of indices, where U 1 is an open
neighborhood of x(t) in X 1• A net {xa, cx E B, ;:;; } in the product
converges tu a point x' of the product if and only if for each t in A
the net {xa(t), cx E B, ~} converges to x'(t). For this reason the
product topology is often called the topology of coordinatewise
convergence. It is easy to verify that the projection of the product
into each coordinate space is open and continuous, and that the
product topology is the weakest topology such that each projection is
continuous. A map into the product is continuous if and only if the
map followed by the projection P 1 is continuous for each t in A.
Finally, for a fixed point Bin the product, and for each t in A, the map
lt, defined as follows, is a homeomorphism of xt into the product:
for each s in A, and each point z in X 1, 11(z)(s), the s-th coordinate of
11(z), is either z or B(s), depending on whether s is or is not equal to t.
Thus a product (of non-void factors) contains topological copies of
the factors.
4.1 TYCHONOFF THEOREM The product of compact spaces is, with
the product topology, a compact space.
A particular case of a product is that in which all coordinate spaces
are identical. Then the product X {X: t E A} is simply the set of all
functions on A to X. In this case the product topology is frequently
called the topology of pointwise convergence (the simple topol-
ogy, or the topology of simple convergence).
The following corollary to the Tychonoff theorem is stated for
reference.
4.2 CoROLLARY The set of all functions on a set A to a compact
space Xis compact relative to the topology of pointwise convergence.
There is a useful extension of the method whereby the product
topology is defined. Suppose Fis a dass of functions, each member
f of F being on a set X to a topological space Y 1 . In general there
will be many topologies for X which will make each member of F
continuous-the discrete topology always has this property. Among
these topologies there is a weakest, namely, the topology having for a
subbase the family of all inverses under arbitrary members f of F of
open subsets of Y1 . This topology is called the projective topology
for X (or the F-projective topology). The projective topology can
be described alternatively by specifying that a subset U of Xis open
relative to the projective topology if and only if U is the inverse
32 CH. 2 LINEAR TüPOLOGICAL SPACES

under P of an open subset of X { Y1 : I E F}, where P is the map

which sends a point x of X into the point with 1-th coordinate l(x)
(that is, P(x)(f) = l(x)). An arbitrary function g on a topological
space Z into X is continuous relative to the projective topology for X
if and only if the composition I o g is a continuous map of Z into Yr
for each I in F. The product topology for a cartesian product is itself
the F-projective topology, where Fis the dass of all projections into
coordinate spaces.
There is a procedure which is dual to the foregoing. Suppose G
is a family of functions, each member g of G being on a topological
space X 9 to a fixed set Y. Then there is a strongest topology for Y
which makes each member of G continuous, namely the family of all
subsets U of Y such thatg- 1 [U] is open in X 9 for eachg in G. This
topology is called the induced topology for Y (the G-induced
topology). A map h of Y into an arbitrary topological space Z is
continuous relative to the induced topology if and only if h o g is a
continuous map of X 9 into Z for each g in G. Finally, if q is an
arbitrary continuous open map of a topological space X onto a topo-
logical space Z, then Z necessarily has the {q}-induced topology.

PROBLEMS
A COMPACT AND LOCALLY COMPACT SPACES
(a) If (X,.r) is compact and (X,Olf) is Hausdorff, and if o/1 is weaker than
.r, then o/1 = .r.
A compact subset of a regular space has a compact closure.
(b) Let X be a locally compact Hausdorff space. If A is a compact
subset of X and U is an open set containing A, then there is a compact set
B with A c B 1 c B c U.
If also X is the union of a sequence of compact sets, then there is a
sequence {Bn: n = 1, 2, · · ·} of compact sets whose union is X such that
Bn c Bn+li for each n.

ß SEPARABILITY
(a) A compact metric space is separable.
(b) If k(A) ~ 2No and if, for each t E A, X 1 is a separable space, then
X {X1: t E A} is separable.
C COMPLETE METRIC SPACES
A sequence {xn: n = 1, 2, · · ·}in a metric space is Cauchy iff d(xm,xn)-+ 0
as m,n-+ oo; a metric space is complete iff every Cauchy sequence is
convergent.
A compact metric space is complete; the converse is not true.
Completeness is not a topological property: there are metrics d1 and d2 on
 SEc. 5 LINEAR ToPOLOGICAL SPACES AND PRonucTs 33
I = {x: 0 < x < 1}, both giving the usual topology, such that (l,d1 ) is
complete but (l,d2 ) is not.

D HAUSDORFF METRIC ON A SPACE OF SUBSETS

Let E be a metric space", with metric d, and let G be the set of closed
non-void subsets of E. For each x E E and each A E G let d(x,A) = inf
{d(x,y):yEA}, and for AEC and BEGlet d(A,B) = max{sup{d(a,B):
a E A}, sup {d(A,b): bEB}}. The function d(A,B) is a metric on G,
except that it may take infinite values. (Infinite values may always be
avoided by using an equivalent bounded metric d' (e.g., d'(x,y) = min
{d(x,y), 1} or d'(x,y) = d(x,y)/(1 + d(x,y)) ). This is called the Hausdorff
metric on the space of closed non-void sets.
When E is complete, G is also complete. (Suggestion for proof: Suppose
that {An: n = 1, 2, · · ·} is a Cauchy sequence in G; let A be the set of Iimits
of all Cauchysequences{xn}inEwithxn EAn for eachn. Showd(A,An)~O.)

E CONTRACTION MAPPING
Let (X,d) be a complete metric space and f a contraction mapping of X
into itself: that is, d(f(x),j(y)) :;;:; rd(x,y) for some constant r < 1 and all
x and y in X. Then there is a unique point z in X with f(z) = z.
It is an irresistible temptation to describe here the elegant classical
application of this result, which proves Picard's theorem on the existence
of solutions of a differential equation. Consider the differential equation
dyfdx = cp(x,y). Suppose that there is a neighborhood U of the point
(a,b) in which cp is continuous, and a constant l such that lcf>(x,y 1 ) -
cp(x,y 2 )1 :;;:; lly 1 - y 2 1 whenever (x,y 1 ) EU and (x,y 2 ) E U. Then there is
one and only one solution y = y(x) of the differential equation with
y(a) = b.
(First, let m = sup {lcf>(x,y)l: (x,y) E U}. Choose h and k so that {(x,y):
lx - al :;;:; h, Iy - bl :;;:; k} lies in U and so that hl < 1 and hm :;;:; k.
Next, let X be the set of continuous functions on the interval [a - h:
a + h], to the interval [b- k: b + k], with the metric d(y,z) = sup
{I y(x) - z(x)l: lx - al :;;:; h}. Now define a mapping f of X into itself
by f(y)(x) = b + gcp(t,y(t))dt. Then f is a contraction mapping, and the
unique fixed point is the required solution.)

5 LINEAR TOPOLOGICAL SPACES, LINEAR FUNCTIONS,

QUOTIENTS, AND PRODUCTS
It is shown that the topology of a linear topological space can be
localized, in the sense that the topology is entirely determined by the
family of neighborhoods of 0. The notion of continuity of a linear
function can also be localized; a linear function is continuous if and
only if it is continuous at 0, and this is the case if and only if the func-
tion is uniformly continuous. A theorem on the extension of linear
functions is proved. The quotient topology is defined, and a topo-
logical form of the induced map theorem proved. Elementary
propositions on product spaces are demonstrated.
34 CH. 2 LINEAR ToPOLOGICAL SPACES

A linear topological space is a linear space E with a topology such

that addition and scalar multiplication are each continuous simul-
taneously in both variables; more precisely, such that each of the
following maps is continuous: (a) the map of the product, E x E,
with the product topology, into E, which is given by (x,y)-+ x + y
for x and y in E; (b) the map of the product, K x E, of the scalar
field and E, with the product topology, into E, which is given by
(a,x)-+ ax for a in K and x in E. The topology of a linear topo-
logical space is called a vector topology. Thus a topology for a
linear space is a vector topology if and only if addition and scalar
multiplication are continuous, and a linear topological space is a
linear space with a vector topology.
Suppose Eis a linear topological space. Because of the continuity
of addition, translation by a member x of E, where the translation by
x of a vector y is x + y, is continuous. Since translation by x has a
continuous inverse, namely translation by - x, it is a homeomorphism.
Similarly, multiplication by a non-zero scalar is a homeomorphism.
It follows that, if a set A is open (closed), so are x + A, and aA, for
each x in E and each non-zero scalar a. Hence, a set U is a neighbor-
hood of a point x if and only if - x + U is a neighborhood of 0; in
other words, the neighborhood system at x is simply the family
of translates by x of members of the neighborhood system at 0. A
base for the neighborhood system of 0 is called a local base or a
local base for the topology (or a system ofnuclei). The topology
is completely determined by any local base Cf/, for a set V is open if
and only if for each x in V there is a U in Cf/ such that x + U c V.
lt is, therefore, of some interest to describe those families of subsets
of a linear space E which are a local base for some vector topology.
5.1 THEOREM ON LocAL BASES Let E be a linear topological space,
and IetCf/ be a local base. Then
(i)for U and V in Cf/ there is a W in Cf/ suchthat W c U n V;
(ii)for U in Oll there is a member V of Cf/ such that V + V c U;
(iii)Jor U in Cf/ there is a member V of Oll such that a V c U for
each scalar a with Iai ; ; 1;
(iv) for x in E and U in Cf/ there is a scalar a suchthat x E aU; and
(v) for U in Cf/ there is a member V of Cf/ and a circled set W such
that V c W c U.
(vi) If Eisa Hausdorff space, then n u
{U: E Oll} = {0}.
Conversely, Iet E be a linear space and Cf/ a non-void family of subsets
which satisfy (i) through (iv), and Iet .'T be the family of all sets W such
 SEC. 5 LINEAR TOPOLOGICAL SPACES AND PRODUCTS 35
that, for each x in W, there is U in Oll with x + U c W. Then .r is a
vector topology for E, and Oll is a local base for this topology. lf,further,
(vi) holds, then .r
is a Hausdorff topology.

The proof of this proposition is Straightforward and will be omitted.

If Oll and "/'" are families satisfying the conditions (i) to (iv) of
theorem 5.1, and if each member of Oll contains a member of "/'", then
the topology determined by "/'" is stronger than the topology deter-
mined by Oll. The converse of this proposition is also true. If each
member of Oll contains a member of "/'", and if each member of "/'"
contains a member of Oll, then Oll and "/'" determine the same topology.
In this case the families Oll and "/'" are said to be equivalent.
Part (iii) of the preceding theorem implies that for each linear
topological space the family of circled neighborhoods of 0 is a local
base.
There are several elementary propositions about linear topological
spaces which will be used frequently, sometimes without explicit
mention. The following facts are among those most commonly used;
the proofs are given as samples of computations in linear topological
spaces. Recall that A 1 is the interior of A and A- is the closure
of A.

5.2 ELEMENTARY COMPUTATIONS Let E be a linear topologica[

space, and Iet Oll be the family of neighborhoods of 0. Then
(i) the closure of x + A is x + A- for each vector x and each
subset A of E; the closure of aA is aA- for each non-zero
scalar a and each set A;
(ii) if A and Baresubsets of E, then A- + B- c (A + B)-;
(iii) A + U is open if U is open, and hence A + B 1 c (A + B)1;
(iv) C + Dis compact if C and D are compact;
(v) the closure A- of the set Ais n {A + V: V E Oll};
(vi) if Cis compact and U open, and if C c U, then there is V in
Oll suchthat C + V c U;
(vii) C + Fis closed if Cis compact and Fis closed;
(viii) the closure of a subspace is a subspace;
(ix) the closure of a circled set is circled;
(x) if the interior of a circled set contains 0, it is also circled;
(xi) the family of all closed circled neighborhoods ofO is a local base;
(xii) each convex neighborhood of 0 contains the closure of a convex
circled neighborhood ofO; and
(xiii) the closure of a convex set is convex.
36 CH. 2 LINEAR ToPOLOGICAL SPACES
PROOF (i) This is a direct consequence of the fact that translation
and multiplication by non-zero scalars are homeomorphisms.
(ii) If xEA- andyEB-, then (x,y}E(A x B)- in the product;
and since addition is continuous, x + y E (A + B)-.
(iii) Observe that A + U = U {x + U: x E A}, and x + U is open
for every x.
(iv) This follows from the fact that C + D is the image of the
compact set C x D under the continuous map, addition.
(v) A point x belongs to A- if and only if for each neighborhood
V of 0, x + V intersects A; that is, if and only if x E A - V. Hence
A- = n{ A - V: V E 0//}, but V is a neighborhood of 0 if and only
if - V is a neighborhood of 0. The required result follows.
(vi) Suppose that for each neighborhood V in the neighborhood
system Oll of 0 there is an element xv in C suchthat Xv + V intersects
the complement of U. Consider the net {xv, V E 0//, c }. Since C
is compact, this net has a duster point x 0 in C. Thus every neighbor-
hood of x 0 intersects the complement of U. This situation is im-
possible since x 0 is an element in the open set U.
(vii) If x does not belong to C + F, then the compact set x - Cis
disjoint from F. Hence, by (vi), there is a neighborhood V of 0 such
that V + x - C is disjoint from F. Consequently, x + V is dis-
joint from C + F, and x does not belong to the closure of C + F.
Hence, C + F is closed.
(viii) Suppose that F is a subspace of E, and that b and c are
scalars. ThenbF- + cF- = (bF)- + (cF)- c (bF + cF)- c F-,
in view of (i) and (ii), and F- is therefore a subspace.
(ix) If Ais circled, xEA-, and 0 < Iai;;:;; 1, then axEaA- =
(aA)- c A- and A- is therefore circled.
(x) If B is the interior of a circled set A, then aB c aA c A
whenever 0 < a ;;:;; 1, and, since aB is open, aB c B.
(xi) For each neighborhood V of 0 there is a circled neighborhood
U such that U + U c V. Then U - is circled, by (ix), and is
contained in U + U, by (v).
(xii) If u is a convex neighborhood ofO and V= n {aU: Iai = 1},
then V is a circled neighborhood of 0 because U contains a circled
neighborhood of 0. Clearly V is a convex subset of U. The set
(1J2)V is a convex circled neighborhood of 0, and its closure is con-
tained in (1/2)V + (1j2)V. The latter set is a subset of V because
V is convex.
(xiii) The proof is similar to that of (viii) and is omitted.lll
A particular consequence of part (xi) of the foregoing theorem is
 SEc. 5 LINEAR ToPOLOGICAL SPACES AND PRonucTs 37
that the family of closed neighborhoods of a p6int is a base for the
neighborhood system of the point. In other words, the topology is
regular.
The notion of continuity of a linear function can be localized, as
might have been expected in view of the localization of the topology
of a linear topological space. Recall that a function Ton a topological
space E to a topological space F is continuous at a point x if for each
neighborhood W of T(x) there is a neighborhood of x whose image
under T is a subset of W. If T is linear, if E is a linear topological
space with local base il/1, and if F is a linear topological space with
local base '"Y, then T is continuous at x if and only if for each V in '"Y
there is U in il/1 suchthat T[U + x] c V+ T(x); that is, if and only
if T[U] c V. Thus a linear function is continuous at some point x
of its domain if and only if it is continuous at 0, and this is the case if
and only if it is continuous at every point. This proves the following
proposition.

5.3 LocALIZATION OF CoNTINUITY A linear function on a linear

topological space to a linear topological space is continuous at some point
of its domain if and only if it is continuous at every point of its domain.
If f is a linear functional, then continuity can be deduced from
premises which are even weaker than continuity at some point.

5.4 CoNTINUITY OF LINEAR FuNCTIONALS If f is a linear functional

on the linear topological space E and f is not identically zero, then the
following conditions on f are equivalent:
(i) f is continuous;
(ii) the null space of f is closed,·
(iii) the null space oj fisnot densein E;
(iv) fis bounded on some neighborhood oJO;
(v) the image under f of some non-void open setisaproper subsei of
the scalar field; and
(vi) the function r, where r(x) is the real part of f(x) for x in E, is
continuous.
PROOF If f is continuous, the null space of J, being the inverse
image of a closed set, is closed; and if the null space of f is closed,
then, since f is not identically zero, the null space of f is not dense in
E. Hence, (i) implies (ii), and (ii) implies (iii). If the null space N
of f is not dense in E, there is a point x of E and a circled neighbor-
hood U of 0 such that x + U is disjoint from N. Then f must be
38 CH. 2 LINEAR ToPOLOGICAL SPACES
bounded on U, for otherwise, since U is circled· and radial, J[U] is
the entire scalar field, and for a suitably chosen u in U,J(x + u) = 0.
This conclusion contradicts the fact that x + U is disjoint from N.
Consequently, (iii) implies (iv). Iff is bounded by Mon a neighbor-
hood U of 0, thenf- 1 [{t: ltl ~ e}] contains (e/M)U and hencefis
continuous at 0. Thus (iv) implies (i). It is clear that condition (v)
is implied by continuity of J. Conversely, if for some scalar a the
setf- 1 [a] is disjoint from some open set, thenf- 1 [a] is notdensein
E and hence, by a translation argument, f- 1 [0] is not dense in E.
Consequently f is continuous by (iii). Finally (vi) is obvious in view
of the definition of r and the fact that f(x) = r(x) - ir(ix) for all x.111
Continuous linear functions have an important extension property.
This property is a direct consequence of the fact that a continuous
linear function T is automatically uniformly continuous, in the
following sense: for each neighborhood V of 0 in the range space
there is a neighborhood U of 0 in the domain such that for every x
and y it is true that T(x) - T(y) E V whenever x - y E U. (ln-
tuitively, two points are close together if their difference is near 0,
so that there is a "uniform" notion of nearness.) The next proposi-
tionisaform of a well-known principle applying to arbitrary uniformly
continuous functions.
The graph of a function T with domain D is the subset of E x F
consisting of all pairs (x,T(x)) for x in D. (Many authors define a
function T to be what is here called the graph of T.) Let E and H
be linear spaces with the same scalar field K. Then a subset G of
E x H is the graph of a linear transformation whose domain and
range are linear subspaces of E and H respectively if and only if G is a
linear subspace of E x H such that (O,y) E G implies y = 0.
In the next theorem the following notation is used: if T is a func-
tion, Gr is the graph of T.
5.5 EXTENSION BY CoNTINUITY lf T is a continuous linear function
on a subspace F of a linear topological space E to a Hausdorff linear
topological space H, then the closure Gr of Gr in the product E x His
the graph of a continuous linear extension of T.
PROOF It must be shown first that Gr is the graph of a function;
that is, if (O,y) E Gr, then y = 0. If (O,y) E Gr. then there is a net
{(x(nYa), a E A} in Gr such that (xa,Ya)--+ (O,y). Then Xa--+ 0 and
Ya = T(xa)--+ y, and y = 0 because H is Hausdorff and T is con-
tinuous. Since the closure of a subspace is a subspace, Gr is the
graph of a certain linear transformation T, which will be shown to be
 SEC. 5 LINEAR TOPOLOGICAL SPACES AND PRODUCTS 39
continuous. For this purpose it is sufficient to prove the continuity
of T at 0, and this will follow if it is shown that T [ U] c V whenever
V is a closed neighborhood of 0 in H and U is an open neighborhood
of 0 in E such that T [ U] c V. If x is a point of U which belongs
to the domain of T, then there is x" in the domain of T such that
Xa --7 x and T(xa) --7 T(x). Since U is open, Xa is eventually in U;
hence T(xa) is eventually in V, and T(x) E V because V is closed.JJJ

5.6 REMARK It is always true that the domain Dr of the exten.sion

T of a continuous linear T is contained in the closure Dr of the
domain of T. Under certain circumstances Dr = Dr. Using
the definitions of "Cauchy net" and "complete" which are given in
the first paragraph of Section 7, it can be shown that Dr = Dr if H
is complete. For, if x E Dr, there exists a net {xa, a E A} in Dr
such that Xa --7 x. Then {x", a E A} is a Cauchy net, and therefore
{T(xa), a E A} is a Cauchy net. When His complete, there exists a
z in H such that T(xa) --7 z; hence, (x,z) E Gr = Gr and x E Dr.
From this fact it follows that Dr c Dr, and hence that Dr = Dr.
If F is a linear subspace of a linear topological space E, then F
itself, with the relativized topology of E, is a linear topological space.
The space F is called a linear topological subspace of E if and only
if Fis both a linear subspace and a topological subspace of E.
lf F is a linear subspace of a linear topological space E, the linear
topological quotient space is the quotient space ElF with the
topology such that a set U in EI F is open if and only if Q- 1 [ U] is
open in E, where Q is the quotient map; that is, Q(x) = x + F.
This topology for ElF is the quotient topology, and this topology is
a vector topology. Unless there is a statement to the contrary, the
quotient space always has the quotient topology.

5.7 THEOREM Let F be a linear topological subspace of a linear

topological space E, Iet ElF be the quotient space, and Iet Q be the
quotient map. Then the map Q is linear, continuous, and open, and
the linear topological space EI F is H ausdorff if and only 1j F is closed.
Moreover, a function T on ElF is continuous (open) if and only if the
composition T o Q is continuous (open).
PROOF It is not difficult to see that ElF is a Hausdorff space if and
only if Fis closed (see part (v) of theorem 5.2 and part (vi) of 5.1).
If U is open in ElF, then by the definition of the quotient topology
Q- 1 [U] is open in E and Q is continuous. To show that Q[V] is
open in ElF when V is open in E, it is sufficient to show that
40 CH. 2 LINEAR ToPOLOGICAL SPACES

Q- 1 [Q[V]] V+ Fisopen in E. But the sum of the open set V

and any set is open. If T is a map of EjF into G and W is a subset
of G, then r- 1 [W] is open in EjF if and only if Q- 1 [T- 1 [W]] =
( T o Q):: 1 [ W] is open in E; this fact proves that T is continuous if
and only if Tc Q is continuous. In a similar manner it follows,
from the fact that Q is both open and continuous, that T is open if
and only if T o Q is open.[[[

The following is a convenient form of the induced map theorem

for linear topological spaces; it follows immediately from theorem 5. 7.
A tQpological isomorphism is a linear isomorphism which is also
a homeomorphism.
5.8 INDUCED MAP THEOREM Let T be a linear function on a linear
topological space E to a linear topological space G, let F be the null
space of T, and let S be the induced map of EjF into G. Then S is
continuous (open) if and only if T is continuous (open). The map T is
continuous and open zf and only if S is a topological isomorphism from
EjF onto G.
If, for each member t of an index set A, E1 is a linear topological
space, then the product X {E1 : t E A} is a linear space and may be
topologized by the product topology. Recall that for t in A, P 1 is
the projection of the product onto the coordinate space E1, and 11 is
the injection of E1 into the product (see Section 1). The proof of the
following theorem requires only a direct application of the definitions
and the preceding results.

5.9 THEOREM The product X {E1 : t E A} of the linear topological

spaces E 1 is a linear space and the product topology is a vector topology.
The projections P 1 are continuous open linear functions, and the injections
11 are topological isomorphisms.
There is a Iemma on products which is frequently useful. A
projection (idempotent operator) on a linear space E is a linear
function Ton E to E such that T o T = T. If T is a projection, then
for each vector x the vector x - T(x) belongs to the null space of T,
and hence each vector x is the sum of the member T(x) of the range
of T and the member x - T(x) of the null space. Moreover, if x
belongs to the range of T and also to the null space of T, then x =
T(y) for some y, and 0 = T(x) = T o T(y) = T(y) = x. Con-
sequently Eis the direct sum of the range of T and the null space of T.
In case E is a linear topological space and T is continuous, this
 SEC. 5 PROBLEMS 41
decomposition of E is a decomposition into a topological product, in
a sense made precise by the following proposition.
5.10 LEMMA ON PROJECTIONS Let E be a linear topological space,
Iet T be a continuous projection on E, and Iet N and R be respectively the
null space and the range of T. Then E is topologically isomorphic
to R x Nunder the map which carries a vector x into (T(x), x - T(x)).
The Straightforward proof of this lemma is omitted.
The lemma above can be applied to show that an arbitrary linear
topological space differs but little from a Hausdorff space. We
begin by observing that because a linear topological space is neces-
sarily regular, it is Hausdorff if and only if {0} is closed. Let E be an
arbitrary linear topological space and let F be {0}- ; then F is a linear
subspace of E, and there is a complementary subspace G. The
relative topology for F is the trivial topology (only F and the void set
are open), and G, with the relative topology, is Hausdorff because
{0} = G n {0}- is closed in G. Finally, the linear function whose
null space is G, and which projects E onto F, is continuous, because
every linear function to F is continuous. It follows from the lemma
above that E is topologically isomorphic to F x G.
5.11 THEOREM ON NoN-HAUSDORFF SPACES Let E be a linear
topological space, Iet F be the closure of {0}, and Iet G be an arbitrary
subspace of E which is complementary to F. Then the relative topology
for G is H ausdorff, and E is topologically isomorphic to G x F, where
F has the tri·vial topology.

PROBLEMS
A EXERCISES
(1) Any hyperplane in a linear topological space is either dense or closed.
(2) If A is a closed set of scalars not containing 0 and B is a closed sub-
set of a linear topological space not containing 0, then U {aB: a E A} is
closed.
(3) If §' is a topology foralinear space E suchthat the mappings (x,y)-
x + y and (a,x)- ax are continuous. in each variable separately, and
continuous at (0,0) in E x E and at (0,0) in K x E respectively, then §' is
a vector topology.

ß NATURAL, NON-VECTOR TOPOLOGIES

In no linear space containing more than one point is the discrete topology
a vector topology.
Let Y be a set with the topology induced by a family G of mappings,
each member g of G mapping a topological space X 9 into Y (see the end
42 CH. 2 LINEAR ToPOLOGICAL SPACES

of Section 4). Then the relative topology for Y"' U {g(X9 ): g E G} is

discrete. Consequently if Y is a linear space, each X 9 a linear topological
space and the members of G linear, the G-induced topology on Y is not a
vector topology if Y =I U {g(X9 ): g E G}.
In particular, if Y is the direct sum L: {X9 : g E G} and each g is the
injection of X 9 into Y, the G-induced topology is a vector topology only if
every X 9 except one is zero dimensional.

C PROJECTIVE TOPOLOGY
Let X be a linear space and F a dass of linear functions, each member f
of F mapping X into a linear topological space Y1 • Then the projective
topology for X, the weakest making each member of F continuous (see
Section 4) is a vector topology for X. (Cf. 16D.)

D ATTEMPT AT A STRONGEST VECTOR TOPOLOGY

If E is an infinite dimensional linear space, the family of all circled sets
radial at 0 is not a local base for a vector topology for E.
(Let {e1 : i = 1, 2, · · ·} be a linearly independent set, An= L: {a 1e1 : la11~
n-1, i = 1, · · ·, n} and A = U {An: n = 1, 2, · · ·}. If Bis a subspace
complementary to that generated by A, then there is no circled set C radial
at zero with C + C c A + B.)

E STRONGEST VECTOR TOPOLOGY I (see 14E; also 6C, 6I)

On any linear space there exists a strongest vector topology, the upper
bound of all vector topologies.

F BOX TOPOLOGY
Let {Ea: cx E A} be an infinite family of non-zero Hausdorff linear topo-
logical spaces and Iet F = X{Ea: cx E A}. Then the topology which has
for a local base all sets of the form X{Va: cx E A}, where Va belongs to a
local base in Ea, is not a vector topology on F. With this topology, Fis not
connected; the component of 0 is the direct sum L: {Ea: cx E A}.

G ALGEBitAle CLOSURE OF CONVEX SETS I (see 14F)

If A is a convex subset of a linear topological space and 0 E A, then
n {r A: r > 1} c A-. If the origin is an interior point of A, then {rA: n
r > 1} = A-. (Use the continuity ofscalar multiplication and 5.2(v).)

H LINEARLY CLOSED CONVEX SETS I (see 14G; also 18H)

A convex subset of Rn or cn is closed if its intersection with every
straight line is closed.
A convex subset of a linear topological space with a non-void interior has
the same property. (Use the previous problem.)

I LOCALL Y CONVEX SETS

A subset A of a linear topological space is locally convex iff for each
x E A there is a neighborhood V of x stich that V n A is convex.
 SEC. 6 NORMABILITY, METRIZABILITY, AND EMBEDDING 43
A closed, connected, locally convex subset A of a Hausdorff linear
topological space is convex.
((i) If x E A and y E A, then there are points z1(1 ~ i ~ n) of A with
x = z 1 , y = Zn and [z1 : z1 + 1 ] c A for 1 ;:;;: i < n.
(ii) 1f xEA andyEA and if zEA with [x:z] c A and [z:y] c A,
then [x: y] c A.)

6 NORMABILITY, METRIZABILITY, AND EMBEDDING;

LOCAL CONVEXITY
This section gives necessary and sufficient conditions that a linear
topological space be normable, or metrizable, or topologically iso-
morphic to a subspace of a product of normable or metrizable spaces.
The solution of these problems requires two new notions-that of a
bounded set in a linear topological space and that of a locally convex
linear topological space. A few very elementary propositions about
these notions are proved.
The simplest sort of linear topological space is that in which the
topology is defined by means of a norm. Recall that a norm p for a
linear space E is a subadditive (p(x + y) ~ p(x) + p(y)), absolutely
homogeneaus (p(tx) = ltl p(x)), non-negative functional on E, such
that p(x) = 0 if and only if x = 0. If p is a norm for E, then the
metric associated with p is defined by d(x,y) = p(x - y). There is
no difficulty in seeing that d is actually a metric-the triangle in-
equality is a direct consequence of subadditivity of p. The metric
topology .r associated with d is defined by calling a set U .r-open
if and only if for each member x of U there is a positive number e
suchthat the open sphere of radius e about x (that is, {y: d(x,y) < e})
is contained-in U. It is easy to verify that the linear space E, with
the topology .r, is a linear topological Hausdorff space, and .r is called
the norm topology. An arbitrary linear topological space is called
normable if and only if there is a norm whose topology is that of
the space. Of course there may be many norms which give the same
topology for E. Two norms, p and q, for E have the same norm
topology if and only if they assign the same neighborhoods to 0,
since a local base determines the topology of a linear topological space.
Evidently p and q have the same norm topology if and only if q(x) ~
ap(x) ~ bq(x) for some real numbers a and b, and for all x. Such
norms are said to be equivalent.
A pseudo-norm p has all of the properties of a norm except that p
may vanish for non-zero vectors. If p is a pseudo-norm for E and
d(x,y) = p(x - y), then d is a pseudo-metric for E, the linear space
E with the pseudo-metric topology .r isalinear topological space (not
 CH. 2 LINEAR ToPOLOGICAL SPACES
necessarily Hausdorff) and .'Y is called the pseudo-norm topology.
A linear topological space is pseudo-normable if and only if there
is a pseudo-norm whose topology is that of the space, and two pseudo-
norms are equivalent if and only if they have the same topologies.
The characterization of pseudo-normable linear topological spaces
depends on the observation: If p is a pseudo-norm for E and V is the
open unit sphere about 0 (that is, V= {x: p(x) < 1}), then V is a
convex, circled neighborhood of 0, and the family of all multiples a V
of V, with a > 0, is a base for the neighborhood system of 0. Re-
phrasing this last requirement, V has the property that if U is a
neighborhood of 0, then for some real nurober a it is true that V c a U.
A subset A of a linear topological space is called bounded if and only
if for each neighborhood U of 0 there is a real nurober a such that
A c a U. It is then clear that in each pseudo-normable linear
topological space there is a convex bounded neighborhood of 0.
The converse of this proposition is the following theorem.
6.1 NoRMABILITY THEOREM A linear topological space is pseudo-
normable if and only if there is a bounded convex neighborhood of 0. It
is normable if and only if it is pseudo-normable and H ausdorff.
PROOF If there is a bounded convex neighborhood of 0 in a linear
topological space E, then there is a bounded circled convex neighbor-
hood U of 0, because each convex neighborhood of 0 contains a
convex circled neighborhood of 0. Let p be the Minkowski func-
tional of U; explicitly, p(x) = inf {t: t > 0 and x E tU}. Then p is
a pseudo-norm, and the open unit p-sphere about 0 contains (lf2)U
and is contained in U. If G is an arbitrary neighborhood of 0, then
for some positive nurober a it is true that aU c G. Hence the
positive multiples of U are a local base for the topology of E, and
since this family is also a local base for the pseudo-norm topology,
the two tQpologies coincide and E is pseudo-normable.
The converse of this result has already been observed, and it is
evident that a space is normable if and only if it is Hausdorff and
pseudo-normable.ill
A subspace F of a normable space E is also normable because a
norm for E restricted to F is a norm for F. If F is a closed sub-
space of a pseudo-normable space, then EfF is normable; and even if
Fis not closed, EfF is pseudo-normable. Moreover, a finite product
of normable spaces is normable. However, an infinite product of
non-trivial normable spaces is never normable.
We digress from the principal topic of the section to discuss briefly
 SEC. 6 NüRMABILITY, METRIZABILITY, AND EMBEDDING 45

the notion of boundedness. This notion is highly important in the

sequel, and the elementary facts about boundedness will be used
frequently 'without explicit mention. It is easy to see that a subset
A of a pseudo-normed space (E,p) is bounded if and only if it is of
finite diameter (that is, sup {p(x - y): x, y E A} is finite). However,
a set may be of finite diameter relative to a pseudo-metric and still
fail to be bounded (if d is a pseudo-metric, then e(x,y) =
min [l,d(x,y)] defines a pseudo-metric such that the diameter of the
entire space is at most one ). A linear function is called bounded if
and only if it carries each bounded set into a bounded set. Several
elementary facts about bounded sets and bounded functions are
summarized in the following proposition, whose proof is omitted.

6.2 ELEMENTARY FAcTs ON BouNDEDNEss

(i) lf A and Bare bounded subsets of a linear topological space and

a is a scalar, then each of the sets aA, A-, A + B, and A U B
is bounded.
(ii) A sufficient condition that a linear function be continuous is
that the image of some neighborhood of 0 be bounded. This
condition is also necessary if the range space is pseudo-normable.
(iii) Each continuous linear function is bounded. If the domain
space is pseudo-normable, then a linear function is bounded if and
only if it is continuous.

lt is possible to characterize those linear topological spaces which

are topologically isomorphic to subspaces of a product of normable
spaces, or to a product of pseudo-normable spaces. The key to this
characterization is the following observation. If E1 is a linear space
with pseudo-norm p1, for each member t of an index set A, and
E = X {E1: t E A} is the product, then a subbase for the neighbor-
hoods of 0 in E is the family of all sets of the form {x: p 1(x) < a},
where a is an arbitrary positive number and t is an arbitrary member
of A. Each of these sets is convex, and since the family of finite
intersections of these is a local base for E, it follows that the family
of convex neighborhoods of 0 is a local base. A linear topological
space is called locally convex, and its topology is called a locally
convex topology, if and only if the family of convex neighborhoods
of 0 is a local base. Clearly each subspace of a locally convex space
is locally convex, and the preceding discussion may be summarized
as follows: each subspace of a product of pseudo-normable linear
topological spaces is locally convex. The converse of this proposition
46 CH. 2 LINEAR ToPOLOGICAL SPACES
is also true; it will be demonstrated after the proof of a preliminary
lemma. Recall that a map R of E into F is called relatively open if
the image of each open set U is open in R[E], and that R is a topo-
logical map of E into F if R is one-to-one, continuous, and relatively
open.
6.3 EMBEDDING LEMMA For each t in an index set A Iet Rt be a
linear map of a linear topological space E into a linear topological space
Ft. Let R be the map of E into the product F = X {Ft: t E A} which
is defined (coordinatewise) by R(x)t = Rt(x) for x in E and t in A.
Then R is continuous ij each Rt is continuous, and R is relatively open ij
it is true that for each neighborhood U of 0 in E there is t in A and a
neighborhood VofO in Ft suchthat Rt- 1 [V] c U. The map R is a
topological isomorphism of E onto a subspace of F ij it is continuous,
relatively open, and one-to-one.
PROOF A linear map into a product is continuous if and only if the
map followed by projection into each coordinate space is continuous,
and, in view of the definition of R, R followed by projection into Ft
is simply Rt. Consequently R is continuous if each Rt is continuous.
To prove the assertion concerning relative openness, observe that if
U is a neighborhood of 0 in E and V is an open neighborhood of 0
in Ft such that Rt -l[V] c U, then R[U] contains {y: y E F and
Yt E V} n R[E]; this set is open in R[E], and the lemma follows.lll
6.4 EMBEDDING OF LOCALLY CONVEX SPACES A linear topo/ogica/
space E is locally convex if and only if it is topologically isomorphic to a
subspace of a product of pseudo-normable spaces.
PROOF It has already been observed that each subspace of a product
of pseudo-normable spaces is locally convex. Suppose that E is
locally convex. Each convex neighborhood V of 0 contains a convex
circled neighborhood of 0 (namely n
{aV: Iai = 1}), and hence the
family Oll of convex circled neighborhoods of 0 is a local base. For
each U in Olllet Pu be the Minkowski functional of U; then Pu is a
pseudo-norm. Let F u be E with the Pu pseudo-norm topology,
and let Ru be the identity map. Then the embedding lemma 6.3
applies, and the map of E into X {F u: U E Oll} is a linear isomorphism
onto a subspace of this product.lll
If a linear topological space E is locally convex and Hausdorff,
then the foregoing theorem can be sharpened to state that E is topo-
logically isomorphic to a subspace of a product of normed spaces
(problem 6D). A topological isomorphism of a space E onto a sub-
 SEC. 6 NORMABILITY, METRIZABILITY, AND EMBEDDING 47
space of a space F is sometimes called an embedding of E in F.
The result above can then be stated as: each locally convex linear
topological space can be embedded in a product of pseudo-normable
spaces.
The class of locally convex spaces has many important properties
other than that of the preceding theorem, and, in fact, almost all of
the theory of linear topological spaces concerns locally convex spaces.
It is evident that each subspace of a locally convex space is locally
convex, and it is not hard to see that quotients and products of locally
convex spaces are of the same sort. There are, for localiy convex
spaces, certain simplifications which can be made in the specification
of the topology. The propositions having to do with this simplifica-
tion are given here for later reference; the Straightforward proofs are
omitted.

6.5 LocAL BASES FOR LocALLY CoNVEX SPACES In each locally

convex linear topological space E there is a local base Oll such that
(i) the sets in Ollare convex circled sets each of which is radial at 0;
(ii) the intersection of two sets in Oll contains a set in Oll;
(iii) for each U in Oll and each scalar a # 0, aU E Oll; and
(iv) each set in Oll is closed.
Conversely, if E is a linear space, then any non-void family Oll of
subsets which satisfies (i), (ii), and (iii) is a local base of a unique locally
convex topology for E.

Notice that the family of convex neighborhoods of 0 in any linear

topological space is a local base for some locally convex topology .r,
in view of the foregoing proposition. Clearly ff is the strongest
locally convex topology which is weaker Ü!an the original topology
of the space.

6.6 CoNTINUOUS PsEUDO-NORMS Let E be a linear space with a

locally convex topology .r, and Iet P be the family of all continuous
pseudo-norms. Then
(i) the family of p-unit spheres about 0, for p in P, is a local base
for ff; that is, a set U is a ff-neighborhood of 0 if and only if
there is a member p of P and a positive number r such that
{x: p(x) < r} c U;
(ii) a net {xa, a E D} in E converges to a point x if and only if
p(xa - x) converges to zero for each p in P;
48 CH. 2 LINEAR TOPOLOGICAL SPACES
(iii) a linear mapping T of a linear topological space F into Eis con-
tinuous if and only if the composition p o T is continuous for each
p in P; and
(iv) a subset of Eis bounded if and on(v if it is of finite p-diameter for
eachp in P.

There are, of course, many linear topological spaces which are not
locally convex. The theory of such spaces is meager; however, it is
possible to describe in a natural way those spaces which are metrizable.
A consequence of this description will be the fact that an arbitrary
linear topological space can be embedded in a product of pseudo-
metrizable spaces.
Notice that if p is a pseudo-norm for E and d is the associated
pseudo-metric, then disinvariant in thesensethat d(x + z, y + z) =
d(x, y) for all x, y, and z in E. Moreover, d is absolutely homo-
geneous in the sense that d(tx,ty) = itid(x,y) for all x and y in E
and for every scalar t. Conversely, if d is an invariant, absolutely
homogeneaus pseudo-metric, then d is the pseudo-metric associated
with the pseudo-norm p, where p(x) = d(O,x). It follows that if a
pseudo-metric is to be constructed for a linear topological space
which is not locally convex, then one cannot hope to obtain both
invariance and absolute homogeneity.

6.7 METRIZATION THEOREM A linear topological space E is pseudo-

metrizab/e if and only if there is a countable local base for the topology.
If E is pseudo-metrizable, there is a pseudo-metrz'c d for E, whose
topology is that of E, such that d is invariant, and such that each sphere
about 0 is circled.

PROOF lt is clear that the topology of a pseudo-metrizable space

has a countable local base. Conversely, assume that {Un} is a count-
able local base for the topology. By theorem 5.1, it may be supposed
that each Un is circled, that U0 = E, and that Un + Un + Un c
Un-1 foreachinteger n, n ~ 1. Defineg(x) tobe z-n if XE un-1 ~
um and to be 0 if X belongs to every Un. The proof proceeds by
treating g(x - y) as a first approximation to the desired pseudo-
metric, which is then constructed by a chaining argument. Observe
that for each positive number a the set {x: g(x) < a} is circled, since
this set is one of the sets Un.
For each x and y in E, define d(x,y) to be the infimum, over all
finite subsets x 0 = x, .:\\, · · ·, Xn+ 1 = y, of the sum L: {g(xi- xi+ 1 ):
 SEC. 6 NORMABILITY, METRIZABILITY, AND EMBEDDING 49

i = 0, · · ·, n}. It is not difficult to see that d is an invariant pseudo-

metric, and that each sphere about 0 is circled. To prove that the
pseudo-metric topology is the topology of E, it is sufficient to show,
in view of the definition of g, that d(x,y) :;;;; g(x - y) :;;;; 2d (x,y), for
all x and y in E. The first of these inequalities is clear, and the
second will be proved by induction on the number n of links in the chain.
Foreach n it must be shown that g(x 0 - Xn+ 1) :;;;; 2 .L {g(x 1 - xi+ 1):
i = 0, · · ·, n}. For convenience, call the number .L {g(xi- X;+l):
i = r, · · ·, s} the length of the chain from r tos + 1. Let a be the
length of the chain from 0 to n + 1 ; one can suppose that a < 2- 2
( otherwise the assertion is trivial). Assurne that the requjred in-
equality is proved for integers smaller than n. Clearly one may
suppose n ;;;;; 1. Break the chain 0, 1, · · ·, n + 1 into three parts,
from 0 to k, k to k + 1, and k + 1 to n + 1 (here k may be equal to
0 or to n), in such a way that the lengths of the first and third parts
are at most a/2 each. By the induction hypothesis, g(x 0 - xk) is at
most 2(a/2) = a, and g(xk+ 1 - Xn+l) is at most a. If m is the
integer ;;;;; 2 suchthat 2-m:;;;; a < 2-m+l, then Xo - Xk, Xk- Xk+1>
and xk +1 - Xn +1 all belong to Um _1 ; hence, x 0 - Xn +1 E Um_ 2 , and
g(xo - Xn+l) :;;;; 2-m+1 :;;;; 2a.lll
Since a space is metrizable if and only if it is pseudo-metrizable
and Hausdorff, a linear topological space is metrizable if and only if
it is Hausdorff and there is a countable local base. Each metrizable
linear topological space can be assigned an invariant metric such that
the spheres about 0 are circled.
6.8 CoROLLARY Let E be a linear space, and let {Vn} be a sequence
of circled sets which are radial at 0 and such that V n + V n c V n_ 1 for
n ;::; 2. Then there exists an invariant pseudo-metric d on E such that
E with the pseudo-metric topology of d is a linear topological space in
which {Vn} is a local base.
PROOF It follows from theorem 5.1 that the family {V n} is a local
base for some topology on E with which E is a linear topological space.
The conclusion of the theorem is now a consequence of the metrization
theorem.lll
The metrization theorem may be applied to show that an arbitrary
linear topological space can be embedded in a product of pseudo-
metrizable spaces, in much the same way that the existence of a
"large" collection of continuous pseudo-norms was used to show
that a locally convex space is embeddable in a product of pseudo-
normed spaces. The two essential facts are given in the foregoing
50 CH. 2 LINEAR ToPOLOGICAL SPACES
corollary 6.8, and the emhedding Iemma 6.3. The Straightforward
proof of the following proposition is omitted.
6.9 EMBEDDING THEOREM FOR LINEAR ToPOLOGICAL SPACES Each
linear topological space can be embedded in a product of pseudo-metrizable
linear topological spaces.
lt is shown in the prohlems at the end of this section that the
continuous image of a metrizahle linear topological space may fail to
he metrizahle, although the image under a continuous open map of a
metrizable space is always metrizahle. The same statement holds
with "metrizahle" replaced hy "pseudo-metrizahle ". Suhspaces
and quotient spaces of pseudo-metrizahle spaces are always of the
same type. The question of whether the product of pseudo-
metrizahle spaces is pseudo-metrizahle is of some interest, and it can
he answered easily. In order to avoid complications in the statement
of the theorem, a linear topological space E is called non-trivial (not
indiscrete) if and only if there are open sets other than E and the
void set.
6.10 PRODUCTS OF PsEUDO-METRIZABLE SPACES The product X {E1 :
t E A} of non-triviallinear topological spaces is pseudo-metrizable if and
only if each coordinate space E1 is pseudo-metrizable and A is countable.
PROOF Ohserve that, if the product is pseudo-metrizahle, then,
since each E1 is isomorphic to a subspace of the product, each Co-
ordinate space is pseudo-metrizahle. Suppose that A is uncountable.
Each neighborhood of 0 in the product contains a finite intersection
of neighborhoods of the form {x: x(t) E U}, where U is a neighbor-
hood of 0 in a coordinate space E 1• Hence, if N is the intersection of
countahly many neighborhoods of 0 in the product, then there is a
Countahle set B of indices and a neighhorhood ut of 0 in Et for each
t in B suchthat N contains {x: x(t) E Utfor all t in B}. If s E A - B,
and U8 is a neighborhood of 0 in E8 which is different from E8 , then N
is not contained in the neighhorhood {x: x(s) E Us} of 0. Therefore,
the product cannot have a countahle local hase, and is not pseudo-
metrizahle.
Finally, suppose that the index set A is the set of positive integers,
and that, for each n in A, dn is a pseudo-metric for En. By replacing
each dn hy min [l,dn], it can he assumed that dn is everywhere less
than or equal to one. It is then easy to verify that d(x,y) =
2: {2-ndn(x(n), y(n)): n E A} is a pseudo-metric for the product and
has the required properties.JJJ
 SEC. 6 PROBLEMS 51

PROBLEMS
A EXERCISES
(1) A family o/1 of subsets of a linear space E is a local base for a locally
convex vector topology for E if the following conditions are satisfied.
(i) Each U E 0lt is midpoint convex ( U + U c 2 U ), circled and
radial at 0;
(ii) the intersection of any two members of o/t contains a member of o/t;
(iii) for each U E o/t there is some V E o/t with 2 V c U.
(2) If E is a real linear space and .r is a topology for E such that the
family of all convex neighborhoods of 0 is a base for the neighborhood
system of 0, and if x + y and ax are continuous in each variable separately,
then .r is a vector topology.
(3) A linear topological space E is pseudo-metrizable if and only if
there exists a nested local base (that is, a base linearly ordered by in-
clusion).
(4) If E and F arereallinear topological spaces with F Hausdorffand T
is an additive bounded mapping of E into F, then T is linear.
(5) Let E be a separable pseudo-metrizable linear topological space of
dimension greater than one. Then E contains a countable dense set A no
three points of which lie on any straight line. The complement of A is
radial at each of its points, but has void interior. (Let {xn} be dense, and
put Sn = {x: XE E, d(x,xn) < 1/n}. Letyl = xl; choose Y2 in s2 ,..., {xl};
choose y 3 in S3 but not on the line determined by y 1 and y 2 and so on.
Let A = {yn: n = 1, 2, · · ·}.)
(6) The following four conditions on a subset A of a linear topological
space are equivalent:
(i) A is bounded;
(ii) given any neighborhood U of 0 there is some e > 0 such that
aA c U for all a with Iai < e;
(iii) every countable subset of A is bounded;
(iv) if {xn} is any sequence in A, then !im xnfn = 0.

ß MAPPINGS IN PSEUDO-NORMED SPACES I (see 8ß)

Let E and F be pseudo-normed spaces, with pseudo-norms p and q,
respectively. A linear mapping T of E into F is continuous if and only if
there is some constant k with q(T(x)) ~ kp(x) for all x in E. The smallest
such constant, r(T) say, is equal to sup {q(T(x)): p(:"(;) ~ 1}, and q(T(x)) ~
r(T)p(x) for all x in E.
The function r is a pseudo-norm on the space of all continuous linear
mappings of E into F, and is a norm if and only if q is a norm provided
dim E > 0. In particular, the set of all continuous linear functionals on
a pseudo-normed space Eis a normed space, called the adjoint (conjugate,
dual) of E and denoted by E*; the norm in E* is given by II/II =
sup {lf(x)l: !lxll ~ 1}. Then llxll = sup {lf(x)l- II/II ~ 1}. (Use 3.7.)

C TOPOLOGIES DETERMINED BY PSEUDO-METRICS

(a) Let E be a pseudo-metrizable linear topological space. Jf d is an
52 CH. 2 LINEAR TOPOLOGICAL SPACES

invariant pseudo-metric such that each sphere about 0 is circled (c.f.

theorem 6.7) and q(x) = d(x,O), then q has the following properties:
(i) q(x) ii;; 0 for all x in E and q(O) = 0;
(ii) q(x + y) ~ q(x) + q(y) for all x,y in E;
(iii) q(ax) ~ q(x) whenever Iai ~ 1; and
(iv) lim q(xfn) = 0 for all x in E.
n-oo
Conversely, if q is a functional on a linear space E satisfying (i) through (iv),
then the function d given by d(x,y) = q(x - y) is a pseudo-metric for E
defining a vector topology.
(b) Given any family {q": a E A} of functions on a linear space E each
satisfying (i) through (iv), the finite intersections of the sets of the type
{x: qa(x) < e}, with a E A and e > 0, form a local base for a vector topology
on E, the topology determined by the family {qa}· Conversely, every
linear topological space has a topology which can be determined in this way.
(c) The strongest vector topology on a linear space is determined by
the family of all functions q satisfying (i) through (iv).

D PRODUCTS AND NORMED SPACES

(a) A locally convex Hausdorff space is topologically isomorphic to a
subspace of a product of normed spaces. (For each continuous pseudo-
norm p, Efp- 1 (0) is a normed space.)
(b) A product of linear topological spaces is pseudo-normable if and
only if a finite nurober of the factor spaces are pseudo-normable and the
rest have the trivial topology.

E POSITIVE LINEAR FUNCTIONALS

Let S be any set and B(S) the space of bounded real-valued functions
on S, with 11-~11 = sup {jx(s)l: s ES} for each .~ in B(S). If rp is a real-
valued linear functional on B(S) such that rp(.~) ii;; 0 whenever x is non-
negative on S, then rp is continuous on B(S) and ,P(l) = sup{j,P(x)J:
x E B(S), llx!l ~ 1}. The same Statement holds if S has a topology and
B( S) is replaced by the space of all continuous bounded real-valued
functions.

F LOCALLY CONVEX, METRIZABLE, NON-NORMABLE SPACES

Let S be a topological space containing a countable sequence {Sn} of
compact proper subsets such that sl c s2 c ... and U{ Sn: n = 1' 2,
· · ·} = S. Let E be any linear family of scalar-valued functions con-
tinuous on S. Foreach n and each x in E Iet llxlln = sup {jx(s)J: s E Sn},
and Iet un = {x: XE E, llxlln ~ 1/n}. Then {Un: n = 1, 2, ... } is a local
base for a locally convex metrizable vector topology for E. If for each
pair of positive integers m and n and each point s0 in S ""' Sn there is some
.~in Un suchthat Jx(s0 )j ii;; m, then this topology is not pseudo-normable.
For particular cases, take S to be Rn and E to be all the scalar-valued
continuous functions, or S to be an open set in R 2 and E to be all complex
functions analytic in S. (Cf. 81, 20H.)
 SEC. 6 PROBLEMS 53
G TOPOLOGY OF POINTWISE CONVERGENCE
Let S be the interval [0, 1] and E the set of continuous real valued
functions on S. For each finite subset T of S, and each e > 0 Iet U T,e =
{x: lx(s)l < e for all s E T}. The sets {UT.e: e > 0, Tc S, T finite}
form a local base for a locally convex Hausdorff topology on E, called
the topology of pointwise convergence. With this topology, E is not
metrizable. (Suppose that {U T, •• J is a countable local base, and choose
sES with siU{T;:i= 1,2, .. ·}. Show that no UT,,e, is contained
in U{S),l•)
This topology is weaker than the topology on E defined by the norm
llxll = sup {lx(s)l: s ES}. The identity mapping of E with the norm
topology onto E with the topology of pointwise convergence provides an
illustration promised in the text of a continuous image of a metrizable
linear topological space which is not metrizable. (Cf. 9G.)

H BOUNDED SETSAND FUNCTIONALS

(a) lf E is a linear topological space containing a bounded set A which
is a subset of no finite dimensional subspace of E, then there is a linear
functional on E that is not bounded on A. Such a set exists in every
infinite dimensional pseudo-metric space.
(b) Let E be a Hausdorff linear topological space. Then each bounded
subset of E is a subset of a finite dimensional subspace of E if and only if
every linear functional on E is bounded.

I STRONGEST LOCALLY CONVEX TOPOLOGY I (see 12D, 14D, 20G)

If E is any linear space, the set of all convex circled subsets of E radial
at 0 forms a local base for a Hausdorff locally convex topology, which is
the strongest locally convex topology on E. Every linear mapping of E
with this topology into any locally convex space is continuous; in particular
every linear functional is continuous. With this strongest locally convex
topology, an infinite dimensional linear space E is not metrizable and
every bounded set is contained in a finite dimensional subspace (see the
previous problem). The strongest locally convex topology coincides with
the strongest vector topology (SE) if and only if the dimension of Eis finite
or N0 • (If {ea: a E A} is a non-countable basis of E, the set {x: x =
L: Aaea, L: IAal 112 ~ 1} is a neighborhood of zero in the strongest vector
topology and cannot contain a convex circled set radial at 0.)

j INNER PRODUCTS (see also 7H et seq.)

Let E be a complex linear space. A functional f from E x E to the
complex numbers is called Hermitean bilinear, or sesquilinear iff for each
y in E the mapping x ~ f(x,y) is linear and for each x E E the mapping
y ~ f(x,y) is conjugate linear (that is, f(x,ay + bz) = äf(x,y) + bf(x,z)).
It is called (Hermitean) symmetric iffj(x,y) = f(y,x) for all x,y E E; then
f(x,x) is real for all x E E. Finally the symmetric Hermitean bilinear
functional f is called strictly positive, or positive definite, if f(x,x) > 0 for
all x # 0. A strictly positive symmetric Hermitean bilinear functional f
54 CH. 2 LINEAR TüPOLOGICAL SPACES

on E x E is called an inner product, or a scalar product, for E. A complex

linear space E with a fixed inner product f is called an inner product space
and f(x,y) is usually simplified to (x,y).
(a) (The CBS inequality.) For all x,y in E, /(x,y)l2 ;;;; (x,x)(y,y), with
equality only if x and y are linearly dependent. (Note that the inner
product of (y,y)x - (x,y)y with itself is non-negative.)
(b) Put Jlxll = V(x,x); this defines a norm for Ecalled the inner product
norm. The inner product can be recovered from the norm by means of
the polarization identity
(x,y) = Hllx + Y// 2 - llx - y// 2 + illx + iy// 2 - illx - iyll 2 )
(c) (The parallelogram identity)
llx + Yll 2 + llx - Yll 2 = 2llxll 2 + 2IIYII 2 •
This identity characterizes inner product norms; any norm satisfying it
corresponds to an inner product on E.
(d) If E has the inner product norm topology, the inner product is a
continuous function on the topological product E x E.

K SPACES OF INTEGRABLE FUNCTIONS I (see 6N, 7M, 14M; also 9I, 16F,
22E)
Let (X, .'/, fL) be a measure space. For each p > 0, LP(X, fL), or more
shortly LP(fL), denotes the space of all real or complex valued measurable
functions f on X such that IJIP is integrable. (For measure theory, we
generally follow the definitions and notations used in the book by Halmos
[4].) Then LP(fL) isalinear space and if p;;;; 1 the mappingf~ 11/l!v =
(f IJIPdfL) 11 P is a pseudo-norm on LP(fL) suchthat 11/llv = 0 if and only if
f(x) = 0 almost everywhere. It is usually called the LP-norm and con-
vergence relative to it is called convergence in mean of order p. In case Y:'
is the set of all subsets of X and fL(A) is the number of elements of A (with
fL(A) = + 00 when A is an infinite set) the space LP(fL) is usually denoted
by fP(X). The function f belongs to fP(X) if and only if f(x) = 0 for all
but a countable number of values of x and 11/IIP = (.2 {lf(x)/P: x EX,
f(x) i= 0}) 11 P < oo. In particular, when X is the set of positive integers,
we write simply fP; it is the space of sequences x = {xn} with !Jxl!v =
(2: {/xn/P: n = 1, 2, })1/P < oo.
The proofthat II· !IP is a pseudo-norm for p > 1 requires the Hölder
0. 0

and Minkowski inequalities. One pattern of proof may be sketched as

follows. Suppose p > 1 and 1/p + 1/q = 1.
(a) If a > 0, b > 0, a > 0, ß > 0 and a + ß = 1, then aabß ;;;; aa + ßb.
Hence, if c > 0 and d > 0, then cd;;;; cPjp + dqjq.
(b) (Hölder) IfjELP(fL) andgEU(fL), then
I JjgdfLI;;;; 11/!lv·llgllq·
(c) (Minkowski) lfj, g E LP(fL), then
II! + g/lp;;;; /1/llp + llgllp·
(Note that lf + giP;;;; lfl/f + giP- + lgl/f + giP-1, and it is true that
1

I!+ glp-1 E U(fL).)

 SEC. 6 PROBLEMS 55
L SPACES OF MEASURABLE FUNCTIONS l (see 7N)
Let (X, Y, p.) be a finite measure space, and Iet S(X,p.) be the space of
all real valued measurable functions. For each positive integer n, Iet Un
be the subset of S(X,p.) defined by Un = {f: p.({x: lf(x)l > 1/n}) < 1/n}.
(a) The family of sets {Un: n = 1, 2, · · ·} is a local base for a vector
topology .'Y, say, and (S(X,p.), .'Y) is a pseudo-metrizable linear topological
space.
(b) A sequence Un} in S(X,p.) converges to 0 relative to the topology .'Y
iff p.({x: lfn(x)l > r}) ~ 0 for all r > 0. The topology .'Y is, therefore,
referred to as the topology of convergence in measure.
(c) Let X be the closed interval [0,1] and Iet p. be Lebesgue measure on
X. Then the convex extension of Unis S(X,p.) for each n. Consequently
the only continuous linear functional on S(X,p.) is the functional which is
identically zero. Moreover, there is no non-negative linear functional
which is not identically zero. (For suppose that f(x) = 1. Define by
induction an increasing sequence {xn(t)} and a decreasing sequence {In} of
intervals, I 71 + 1 beingonehalf of In, so that Xn+ 1 (t) = Xn(t) except on In,
Xn+l(t) = 4xn(t) on In> andf(xnxn) ~ 2n-I, where Xn is the characteristic
function of I n• Then f (!im Xn) ~ 2n- 1 for all n.)

M LOCALLY BOUNDED SPACES

A linear topological space Eis locally bounded if and only if for each x 0 in
E and each open set G about x 0 there is a bounded open set B such that
x 0 E B c G.
(a) E is locally bounded if and only if it contains a non-void bounded
open set; any locally bounded space is pseudo-metrizable. With the
strongest locally convex topology weaker than the given topology, E is
pseudo-normable.
(b) If E is locally bounded there is a sequence {Bn} of bounded sets such
that E = U {Bn}·
(c) Parts (ii) and (iii) of theorem 6.2 remain true if " pseudo-normable"
is replaced by " locally bounded".
(d) The space K'", consisting of all scalar sequences with the usual
product topology, is not locally bounded.

N SPACES OF INTEGRABLE FUNCTIONS II (see 6K, 71\1, 14M)

When 0 < p < 1, the mappingf~ 11/!lv = (f lf1Pdp.) 11 P (see 6K) isno
Ionger a pseudo-norm, but for n = 1, 2, · · ·, the sets {f: 11/llv;;::: 1/n} form
a local base for a pseudo-metrizable vector topology on LP(X,p.). In fact,
an invariant pseudo-metric for the topology is d(f,g) = f I/- glvdp..
(a) For each p with 0 < p < 1, the space U(X,p.) is locally bounded
(see the previous problem).
(b) If 0 < p < 1, the space U(X,p.) is locally convex if and only if the
set of values taken by p. is finite. (Otherwise there are disjoint measurable
sets An with p.(An) > 0. Given U = {f: 11/llv ;;::: e} there are elements
fn EU with 11/nllv = e and fn(x) = 0 for X$ An. If gn = n- 1 "J, {!,:
1 ;;::: r ;;::: n}, I gn II v = en< 1 - vllv, and gn belongs to the convex extension
of U.)
56 CH. 2 LINEAR ToPOLOGICAL SPACES

(c) A measurable subset A of X is called an atom iff 0 < JL(A) < oo

and, for every measurable subset B of .A, either JL(B) = 0 or JL(B) = JL(A).
If A is an atom, the mapping f ~ fAfdJL is a continuous linear functional
on LP, for 0 < p < 1. On the other hand, if X contains no atoms and
0 < p < 1, the only continuous linear functional on LP is the zero func-
tional. (If </> is a non-zero linear functional on LP, there is a bounded
function .fo vanishing outside a set A 0 of finite measure with </J{f0 ) = 1.
There are disjoint subsetsBand C of A 0 with JL(B) = JL(C) = -!JL(A 0 ); Iet
g and h be equal to 2f0 on B and C, respectively, and zero elsewhere.
Then either 1</>(g)l ;?; 1 or 1</J(h)l ;?; 1. Suppose the former and put
A 1 = B,f1 = g. Continue to obtain a sequence {/n} with !lfniiP ~ 0 but
I<PUn)l ;?; 1 for all n.)
(d) If 0 < p < 1 and X is an infinite set, the strongest locally convex
topology on /P(X) weaker than its non-locally convex topology is the
relativization to /P(X) of the norm topology of l 1 (X), and for this locally
convex topology /P(X) is dense in P(X). (For each x EX Iet ex be the
function which takes the value 1 at the point x and is. zero elsewhere. If
11/11 1 ~ 1 and F is a finite subset of X, L: {f(x)ex: x E F} belongs to
the convex circled extension of {/: 11/IIP ~ 1} and converges to f as F
increases.)

7 COMPLETENESS
A product space is complete if and only if the factors are complete.
For a subset, compactness is equivalent to completeness tagether with
total boundedness. For a Hausdorff linear topological space, the
space being finite dimensional is equivalent to the space being topolog-
ically isomorphic to Euclidean space, and to the existence of a totally
bounded non-void open set. Finally, each space can be embedded
densely in a complete space, called a completion, and the completion
is essentially unique.

There is a natural extension of the notion of a Cauchy sequence.

A net {xa, a E A, ;?; } in a linear topological space Eis a Cauchy net
if and only if Xa - xß converges to zero; stated precisely, the net is a
Cauchy net if and only if for each neighborhood U of 0 there is a y
in A such that if both a and ß follow y in the order ;?; , then Xa -
xß E U. A Cauchy sequence is a sequence which is a Cauchy net.
Just as for sequences, it follows immediately that each net which con-
verges to a point of the space is a Cauchy net. A subset A of the
space E which has the property that each Cauchy net in A converges
to some point in A, is said to be complete. The image of a Cauchy
net under a continuous linear function is a Cauchy net; hence, the
image under a topological isomorphism of a complete space is com-
plete. A closed subset of a complete set is complete, and if the space
is a Hausdorff space, so that each net converges to at most one point,
 SEC. 7 COMPLETENESS 57
then a complete subset is necessarily closed. This simple Statement
contains one of the most important and useful properties of com-
pleteness.
The relation between completeness of a linear topological space
and completeriess relative to a metric is simple, but a little delicate.
Recall that, if d is a pseudo-metric for a set X, then X is complete
~elative to d if and only if each sequence {xn} which is a Cauchy
sequence relative to d converges in the pseudo-metric topology to a
member of X. It is not hard to see that X is complete relative to d
if and only if each net {xa, a E A} which is a Cauchy net relative to d
(that is, a net such that d(xa,xß) converges to 0) converges to a point
of X. However, it is possible to have a linear space E and a metric d
for E such that E is not complete relative to d, but E with the topology
of d is a complete linear topological space. As an example, consider
the set of real numbers, which (with the usual topology) is a complete,
one dimensional space over the reals, but which is not complete
relative to the metric d(x,y) = [arctan x - arctan y[, although that
metric generates the usual topology. If the pseudo-metric d is
invariant (that is, d(x,y) = d(x + z, y + z) for all x, y, and z), then
this situation cannot occur, for if d(xa,xß) converges to 0, then
d(xa - xß, 0) converges to 0, and completeness implies completeness
relative to d. We record this fact for future reference. 1

7.1 THEOREM Let E be a linear topological space, and let d be an

invariant pseudo-metric whose topology is that of E. Then E is complete
if and only if E is complete relative to d.
Cauchy nets in a product X .{E1 : t E A} of linear topological spaces,
and completeness of such a product, can be characterized in a simple
manner.

7.2 CoMPLETENESS IN PRooucTs A net in a product of linear to-

pological spaces is a Cauchy net if and only if its projection into each
coordinate space is a Cauchy net.
Consequently, the product of a non-void family of spaces is complete tf
and only if each of the factors is complete.
1 A further note of explanation may be helpful. Completeness is always
defined, either explicitly or implicitly, relative to a uniform structure. Given
a metric, it is possible to define a metric uniform structure and a metric topol-
ogy. If E is a linear topological space, it is possible to define a uniform
structure by using the topology and the algebraic structure. This is the
structure which is of primary concern in this study. It is, in general, quite
distinct from the metric uniform structure.
58 CH. 2 LINEAR ToPOLOGICAL SPACES
PROOF If {X1., a E B, ~} is a net in the product X {Et: t E A}, then
Xa - Xp converges to 0 relative to the product topology if and only if
the projection into each coordinate space Et converges to 0; that is,
if and only if {xa(t), a E B, ~} is a Cauchy net in Et for each t. It
follows that, if each Et is complete, then each Cauchy net in the
product converges to a point. The converse is easily established
by using the injection map.lll
If E is a complete linear topological space, then each Cauchy
sequence in E, being a Cauchy net, converges to some point of E.
It is not true, however, that a space in which each Cauchy sequence
converges to some point is necessarily complete. In fact, Iet E be
the space of all real-valued functions on the real interval (0:1), topol-
ogized by pointwise convergence. Then E is simply the product
space X {Kt: t E (0:1)}, where each K 1 is the space of real numbers;
with the usual product topology, E is a complete linear topological
space. Let F be the subspace of E which consists of all functions
which vanish except at a countable nurober of points. Then if a
sequence Un} in F converges to a function f', the function f' vanishes
outside a countable set, and hence f' belongs to F. Thus each
Cauchy sequence in F converges to a point in F. However, since
every f in E can be approximated by an element in F on any finite set
of points, it follows that F is dense in E relative to the product topol-
ogy, and accordingly there are Cauchy nets in F which converge to
members of E which are not in F; therefore, since Eis a Hausdorff
space, Fis not complete.
A subset of a linear topological space is sequentially complete if
and only if each Cauchy sequence in the set converges to a point in
the set. In this terminology, the example above shows that a sequen-
tially complete space may fail tobe complete. For pseudo-metrizable
spaces the concepts of completeness and sequential completeness
actually coincide.
A locally convex space which is metrizable and complete is called a
Frechet space, and a linear space with a norm which is complete
relative to the norm topology is called a Banach space. The
quotient of a complete space may fail to be complete, but the quotient
of a complete pseudo-metrizable space is complete (see problern 20D
and 11.3).
As a first application of the notion of completeness it will be shown
that a finite dimensional Hausdorff linear topological space must
necessarily be topologically isomorphic to a product of a finite nurober
of copies of the scalar field.
 SEc. 7 CoMPLETENEss 59
7.3 UNIQUENESS THEOREM FOR FINITE DIMENSIONAL SPACES A
finite dimensional subspace F of a linear topological H ausdorff space E
is topologically isomorphic to a finite dimensional Euclidean space, and is
therefore complete and closed in E.

PROOF Let xi, i = 1, · · ·, n, be a (Hamel) base for F, and let T be the

linear transformation from the product of n scalar fields X {K:
i= 1, .. ·,n} onto F defined by T(a 1 ,a2 , . . ·,an) = ,L{aix 1 :i=
1, · · ·, n}. Then T is clearly one-to-one, and since E is a linear
topological space, T is continuous. It will be shown by induction on
n that T - 1 is continuous and that T is consequently a topological
isomorphism. If the dimension of F is one, and {x1 } is a base for F
(that is, x 1 =f 0), then the function T, where T(a) = ax 1 , is a con-
tinuous one-to-one map of the scalar field onto F. Since T - 1 is a
linear functional with the closed null space {0}, it follows that T - 1
is continuous. If the dimension of F is n + 1, then, by the induction
hypothesis, every maximal linear subspace H in F is topologically
isomorphic to a Euclidean space, is therefore complete and closed in
E, and hence every linear functional is continuous. In particular the
functions x--+ ai(x) are continuous; thus T - 1 is continuous.lll
There is a useful corollary to the foregoing theorem.

7.4 CoROLLARY If Fis a closed subspace of a linear topological space

E and if Gis a finite dimensional subspace, then F + G is closed.

PROOF Let Q be the quotient map of E onto the Hausdorff space

EfF (see theorem 5.7). Then Q[G] is a finite dimensional subspace
of E/F and is consequently closed by theorem 7.3. Since Q is con-
tinuous, G + F = Q - 1 [Q[ G]] is a closed subspace of E.lll
The notion of completeness can be reformulated in a fashion which
is suggestive of compactness. Compactness of a set A is equivalent
to the following property: if d' is a family of closed subsets of A with
the finite intersection property (that is, each intersection of a
finite nurober of members of d' is not void), then the intersection of
all of the members of d' is not void. A family d' of subsets of a linear
topological space E contains small sets if and only if for each
neighborhood U of 0 there is a member B in d' and a member x of 'E
suchthat B c x + U; or, equivalently, if and only if, given U, there
is Bind' suchthat B - B c U, or suchthat B- - B- c U.

7.5 SMALL SET CHARACTERIZATION OF CoMPLETENESS A subset A

if each family d' of
of a linear topological space Eis complete zf and only
60 CH. 2 LINEAR TOPOLOGICAL SPACES
closed subsets of A such that d has the finite intersection property and
contains small sets has a non-void intersection.

PROOF Let A be complete, and let d be a family of closed subsets of

A such that d contains small sets and has the finite intersection
property. Then the family f!J, consisting of those sets which are
the intersection of a finite nurober of members of d, has the finite
intersection property and is directed by c. Form a net {xB, BE
f!J, c} by selecting an element xB from each B in f!J. Since d con-
tains small sets, so does f!J; and it follows that this net is a Cauchy net
which converges to a point x' of A since A is complete by hypothesis.
Since for each member C of f!J the net {xB, BE f!J} is eventually in C
(in fact, if D follows C in the order, Xv E D c C) and Cis closed, x'
belongs to every Bin f!J and hence to their intersection. Consequently
the intersection of all the sets in d is non-void. Thus, the condition
is necessary.
To prove the converse, observe that, if {xa, a E A, ;:;;; } is a Cauchy
net, it is possible to construct a family containing small sets as follows.
For each a in A let Ba be the set of all points x8 with ß ;:; ; a, and let
f!J be the family of all such sets. Since the net is a Cauchy net, the
family f!J contains small sets; and since A is directed, the inter-
section of finitely many members of f!J is non-void. The family rJ
of closures of sets in f!J is a family of closed sets which has the proper-
ties proved above for f!J. If it is assumed that such a family has a
non-void intersection, then there is a point x' which belongs to the
closure of every Ba. lf U is a neighborhood of 0 and Ba- - Ba- c
U, then Ba c x' + U; hence, the net converges to x' -II I
The connection between completeness and compactness is simple.
A compact subset of a linear topological space is automatically com-
plete, for each Cauchy net has a duster point, and a Cauchy net
converges to each of its duster points. Of course a set may be
complete but not compact-for example, the space of real numbers;
however, there is a condition which, combined with completeness,
implies compactness. A subset B of a linear topological space E is
totally bounded (sometimes called precompact) if and only if for
each neighborhood U of 0 there exists a finite set N such that B c
N + U. It is clear that each compact set is totally bounded, and
each totally bounded set is bounded. It is pointed out in the prob-
lems at the end of this section that if A is totally bounded, then so are
A-, the smallest closed circled set containing A, and the product aA
for each scalar a.
 SEc. 7 CoMPLETENESS 61
7.6 HAUSDORFF's THEOREM ON ToTAL BouNDEDNESS A subset of a
linear topological space is compact if and only if it is totally bounded and
complete.
PROOF It is easy to see that a compact set is totally bounded and
complete. Conversely, if A is a totally bounded complete subset of
a linear topological space, the proof that A is compact proceeds as
follows. Let d be a family of closed subsets of A, and suppose that
d has the finite intersection property. Then d is contained in a
maximal family of this sort; for convenience, let d itself be such a
maximal family. It must now be shown that the intersection of all
the members of /2/ is non-void. Since A is complete, the desired
result will follow if it is proved that d contains small sets. Since A
can be covered by a finite number of translates of an arbitrarily small
open set, it is sufficient to show that if Bi> i = 1, · · ·, n, is a finite
covering of A by closed subsets of A, then for some i, B 1 E d. Since
d is maximal, a closed subset B 1 of A can fail to belang to d only
because its adjunction to d destroys the finite intersection property;
that is, there is a finite subfamily f!Ji of d whose intersection with Bi is
void. If this is the case for each i, then there is a finite subfamily
(namely U {f!J 1 : i = 1, · · ·, n}) of d whose intersection is disjoint
from every Bi and is therefore void. This contradiction establishes
the desired result.lll
There is a weakened form of compactness which still implies total
boundedness. Recall that a set A is countably compact if and only if
every covenng of A by enumerably many open sets has a finite
subcovering.
7.7 THEOREM lf A is a subset of a linear topological space E such
that every sequence of points in A has a cluster point in E, then A is
totally bounded.
In particular, each countably compact subset of E is totally bounded,
and hence each countably compact complete subset of Eis compact.
PROOF Let U be a circled neighborhood of 0. Choose a subset B
of A which is maximal with respect to the following property: if x
and y are in B, then xisnot in y + U. If Bis finite, A c B + U;
otherwise B is an infinite subset of A, and there is a sequence {xn} of
distinct points of B. By hypothesis, there is a duster point y of the
sequence. Furthermore, if V is a neighborhood of 0 such that
V - V c U, and if Xm and Xn belang to V + y, then Xm - Xn E
V - V c U; hence, Xm E Xn + U. This contradiction establishes
the theorem.lll
62 CH. 2 LINEAR ToPoLoGICAL SPACES
The notion of total boundedness yields an interesting characteriza-
tion of finite dimensional spaces and hence, to a topological isomor-
phism, of Euclidean n-space.

7.8 CHARACTERIZATION OF FINITE DIMENSIONAL SPACES A Haus-

dorf! linear topological space E is finite dimensional if and only if there
is a totally bounded neighborhood of 0.
PROOF One half of the theorem has been established. The con-
verse will be proved with the aid of the following lemma: if Y is a
proper closed subspace of E, if U is a bounded neighborhood of 0,
and if V is a circled neighborhood of 0 such that V + V c U, then
it is possible to find a point x in U suchthat (x + V) n Y is empty.
For, otherwise, V + V c U c Y + V. By induction it follows
that, foreachpositive integer m, mV c Y + Vor V c Y + (1/m)V.
Since V is bounded and Y is closed, V c n {Y + (1/m)V: m =
1, 2, · · ·} = Y, which implies that Y is not a proper subspace.
Now suppose that U is a totally bounded neighborhood of 0, and
that V is a circled neighborhood of 0 such that V + V c U. Choose
a non-zero member x 1 of U, and let Y 1 be the subspace generated
by x 1 • Since Eis Hausdorff, Y 1 is closed; hence, by the lemma, an
element x2 can be chosen from U such that (x 2 + V) n Y 1 is
empty. In general, choose Xn+l from u such that (xn+l + V) n Yn
is empty, where Yn is the subspace generated by x 1 , · • ·, Xn. If
some Yn is E, the theorem is proved. Otherwise the sequence {xn}
in U has the property that Xm f/: Xn + V for distinct m and n. Finally,
the set of all Xn is not totally bounded, but this set is a subset of U
and hence U is not totally bounded.jjj
In particular, each locally compact Hausdorff linear topological
space is finite dimensional.
The section is concluded with a proof that each linear topological
space can be completed, andin an essentially unique way.
If E is a linear topological space, if E 11 is a complete linear topo-
logical space, and if T is a topological isomorphism of E into E 11 such
that the image of E under T is dense in E 11 , then E 11 with the map T
is a completion of E.

7.9 LEMMA lf E is pseudo-metrizable, then there is a pseudo-

metrizable completion E 11 of E.
PROOF Let d be an invariant pseudo-metric, which gives the topology
of E, and let E 11 be the set of all Cauchy sequences in E. For {xn}
 SEC. 7 COMPLETENESS 63

and {yn} in E'\ let d'"({xn}, {Yn}) = lim d(xn,Yn)· With the defini-
tions {xn} + {yn} = {xn + Yn} and a{xn} = {axn}, E 11 is a linear
topological space with invariant pseudo-metric d 11 • For an x in E
let T(x) be the sequence each of whose terms is x. Clearly T pre-
serves distance, and the problern reduces to showing that E 11 is
complete relative to d 11 • Observe that T[E] is densein E 11 , because
T(xn) is near {xn} if n is large. It follows that for each Cauchy
sequence in E 11 there is a Cauchy sequence in T [ E] so that the
distance between the n-th term of the former and the n-th term of
the latter converges to zero. Consequently E 11 will be proved
complete relative to d 11 if it is shown that each Cauchy sequence in
T[E] converges to some member of E 11 • But if {T(xn)} is a Cauchy
sequence, then {xn} is a Cauchy sequence and is therefore a member
of E'\ and T(xn) converges to {xn}· Since d 11 is an invariant pseudo-
metric, E 11 is complete by 7.1.111

7.10 CaMPLETION THEOREM Each linear topological space E can be

mapped by a topological isomorphism onto a dense subspace of a complete
linear topological space E 11 •

PROOF According to theorem 6. 9 there is a topological isomorphism

S which maps E into a product X {Et: t E A} of pseudo-metrizable
linear spaces. Let E / be a pseudo-metrizable completion of Et, and
let Tt embed Et in E/. If T is the transformation such that for x
in E, T(x)(t) = Tt(S(x)(t)), then T is a topological isomorphism of E
into X {E t 11 : t E A}. Since X {E/: t E A} is complete by theorem
7.2, the closure E 11 of T [E] is complete and T [E] is a dense su bspace
of E 11 ·111
It was shown earlier (theorem 5.11) that an arbitrary non-Hausdorff
space E 11 is topologically isomorphic to the product of two subspaces,
Fand G, where Fis the closure of {0} and G is an arbitrary subspace
which is complementary to F. Moreover, G with the relative topol-
ogy is Hausdorff. It is clear that if E 11 is complete, then G is neces-
sarily complete. Using these facts and the preceding theorem the
following proposition becomes obvious.

7.11 COROLLARY Each linear topological Hausdorff space has a

H ausdorff completion.

Finally, the completion of a linear topological Hausdorff space is,

in a sense which is made precise by the following theorem, unique.
64 CH. 2 LINEAR ToPOLOGICAL SPACES
7.12 UNIQUENESS OF CaMPLETION lf E 11 and E- are linear topo-
logical H ausdorff spaces which are completions of a space E, and if T
and S ttre the embedding mappings, then T o S - 1 has a unique continuous
extension which is a topological isomorphism of E- onto E 11 •
PROOF By 5.5 both ToS - 1 and So T - 1 have unique continuous
extensions to all of E- and E 11 respectively. Since (To s- 1 ) o(So T- 1 )
is the identity, the composition of the extensions of these two
maps is the identity map on a dense subspace of E\ and it is
therefore the identity on E 11 • It follows that the continuous exten-
sion of T o S - 1 is a topological isomorphism of E- onto E 11 .111
In making use of the results of this section it is frequently con-
venient, when confusion is unlikely, to refer to E as a subspace of E 11
without reference to the embedding map. A more precise statement
is made if the occasion demands it.

PROBLEMS
A FINITE DIMENSIONAL SUBSPACES
If F is a finite dimensional subspace of a linear topological space E, then
F is topologically isomorphic to G x H where G is a Euclidean space and
H is a finite dimensional space with the trivial topology.

ß COMPLETION OF A PSEUDO-METRIZABLE, PSEUDO-NORMABLE, OR LOCALLY

CONVEX SPACE
(a) A completion of a pseudo-metrizable, pseudo-normable, or locally
convex space is a space of the same sort. In particular, the completion of
a locally convex metrizable spa.ce is a Frechet space, and the completion
of a normed space is a Banach space.
(b) Let E be a pseudo-normed linear space with a pseudo-norm p, and
let E 11 be a completion of E relative to the p-topology. Then p can be
uniquely extended to a continuous pseudo-norm pll on E II and the pll-
topology is the topology for the completion E II,

C COMPLETENESS FOR STRONGER TOPOLOGIES

Let E be a Hausdorff linear topological space with topology .r, and let
o/1 be a stronger vector topology having a local base of .r-closed o/1-neighbor-
hoods of 0. Then a subset of E complete relative to .r is also complete
relative to o/1. (Cf. 18.3 and 18D.)

D EXTENSION OF A ONE-TO-ONE MAPPING

Let E and F be a Hausdorff linear topological space and T a one-
to-one continuous linear mapping of E into F. Let T 11 be the continuous
extension of T, mapping E 11 into F 11 • A necessary and sufficient condition
 SEC. 7 PROBLEMS 65
for T 11 tobe one-to-one isthat whenever {xy: y E T} is a Cauchy net in E and
T(xy)--+ T(x 0 ), then Xy--+ x0 •
Let E be a Hausdorff linear topological space with topology ff, and Iet
Cilt be a stronger vector topology. Then (E,Cil/) 11 can be embedded in
(E,ff) 11 if and only if every Cil/-Cauchy net which is ff-convergent is Cil/-
convergent to the same Iimit.
This latter condition is satisfied if there is a local base of ff -closed
Cil/-neighborhoods of 0.

E COMPLEMENTARY SUBSPACES
(a) Let E be a Hausdorff linear topological space and Iet F be a closed
subspace with finite co-dimension. Then any subspace complementary
to Fis isomorphic to EfF.
(b) Let E be a Hausdorff locally convex space and Iet F be a finite
dimensional subspace. Then there exists a (closed) complementary
subspace isomorphic to EfF.
(c) Let E be a Hausdorff linear topological space on which there are no
continuous linear functionals not identically zero. Then no finite-
dimensional subspace has a closed complement.

F TOTALLY BOUNDED SETS

A subset B of a linear topological space is totally bounded iff
(*) for each neighborhood U of 0, there exists a finite set N such that
B c N+ U.
(a) The set N may be supposed tobe a subset of B.
(b) A subset of a totally bounded set is totally bounded.
( c) The closure of a totally bounded set is totally bounded.
(d) The image of a totally bounded set under a continuous linear map is
totally bounded.
(e) A subset of a product is totally bounded if and only if each of its
projections is totally bounded.
(f) A scalar multiple, and the closed circled extension, of a totally
bounded set are totally bounded.

G TOPOLOGIES ON A DIRECT SUM

Let {Ea: cx E A} be an infinite family of non-zero Hausdorff linear
topological spaces, Iet F = X {Ea: cx E A}, and Iet E be the direct sum
L {Ea: cx E A}. Letff 1 be the usual product topology on Fand §"2 the box
topology, defined in SF. The relativization to E of ff 2 is a vector topology
stronger than the relativization of ff 1 ; E is closed in F for ff 2 and dense
in F for ff 1 • If each Ea is complete, then so is E for ff 2 • (The direct sum
topology (see section 14) is stronger than the topology induced by ff 2 , and
E is also complete under this direct sum topology whenever each Ea is
complete (14.6).)

H HILBERT SPACES (see 14L)

An inner product space is called a Hilbert space if and only if it is com-
plete with respect to the inner product norm. Let E be an inner product
66 CH. 2 LINEAR TOPOLOGICAL SPACES

space and Iet H be its completion. Then the inner product on E can be
uniquely extended to H x H so that the extended function is an inner
product for the completion H. Therefore, any inner product space has a
completion which is a Hilbert space.

I HILBERT SPACES: PROJECTION

Let F be a clos~d convex set in a Hilbert space H.
(a) If x is any point in H, then there is a unique element P(x) in F
satisfying llx - P(x)ll = inf {llx - zll: z E F}, the mapping P is called
the projection of H on F. (The parallelogram law can be used to show that
an arbitrary minimizing sequence converges.)
(b) Re (x - P(x), y - P(x)) ~ 0 for each x in H and y in F, where
Re z denotes the real part of the complex number z. In particular, if F
is a cone, then Re (x - P(x),y) ~ 0 and Re (x - P(x), P(x)) = 0 for each
x in H and each y in F; if F is a linear subspace, then (x - P(x),y) = 0
for each x in H and each y in F.
(c) IIPx - Pyll ~ llx - Yll for all x, y EH.
(d) If Fis a linear subspace, then Pisa continuous linear transformation
with the properties
(i) Pa P = P (that is, Pis idempotent).
(ii) (P(x),y) = (x,P(y)) for all x and y in H.
(iii) F = {x: P(x) = x}.
(e) If P is any continuous linear transformation from H into itself for
which (i) and (ii) hold, then there is a closed linear subspace F of H such
that P is the projection on F.

J HILBERT SPACES: ORTHOGONAL COMPLEMENTS

Let A be a subset of a Hilbert space H. A point x in H is said to be
orthogonal to A, written x _l A, if (x,z) = 0 for every z in A. The
orthogonal complement of A in H, denoted by A .L, is the set {x: x _l A}. In
the following propositions A,B are subsets of H.
(a) A .L is a closed linear subspace of H.
(b) A c B implies A.L :J ß.L.
(c) A.L is identical with the orthogonal complement of the closed linear
extension of A.
(d) A.L.L is the closed linear extension of A.
(e) Let F be a closed linear subspace of H, and Iet P be the projection of
H on F.L. Then the projection of H on Fis I - P, where I is the identity
mapping of H onto itself. Hence each element of H can be uniquely
expressed as the sum of an element in F and an element in F.L; that is, F
and F.L are complementary in the sense of Section 1.

K HILBERT SPACE: SUMMABILITY

Let {x": a E A} be a subset of a I:Iilbert space H indexed by A. Let d
be the family of all finite subsets of A; then d is directed by :J. For
FEd Iet xF = L:{xa: aEF}; then {xF, FEd, :J} is a net in H. If this
net is convergent to a point x in H, the set {x": a E A} is summable and the
Iimit x is denoted by L: {xa: a E A}.
 SEC. 7 PROBLEMS 67
(a) A subset {xa: IX E A} of His summableifand only if foreachpositive
nurober e there is a finite subset B of A such that, whenever C is a finite
subset of A ,..., B, 112 {xa: IX E C}ll < e.
(b) If a subset {xa: IX E A} is summable, then the set of indices IX for
which Xa :F 0 is countable. Furthermore, if those indices IX for which
Xa :F 0 are enumerated arbitrarily as IX 1 , a 2 , • • ·, then 2 {xa: IX E A} =
lim2{xa,: 1 ~ r ~ n}.
n-oo
(c) If a subset {xa: IX E A} is orthogonal, that is, any two distinct elements
of the set are orthogonal, it is summableifand only if the subset {llxall 2 :
IX E A} of the real line is summable. If x = 2 {xa: IX E A}, then llxll 2 =
2 {llxall 2 : IX E A}.
L HILBERT SPACES: ORTHONORMAL BASES
A subset A of a Hilbert space H is orthonormal iff A consists of elements
of norm one and (x,y) = 0 whenever x,y E A and x :F y.
(a) (Bessel's inequality) If {x1 , x 2 , • • ·, xn} is a finite orthonormal
subset of H then, for each x in H, 2 {l(x,x,)[2: 1 ~ r ~ n} ~ llxll 2 • (The
projection of x on the subspace generated by xl> · · ·, Xn is 2 {(x,x,)x,:
1 ~ r ~ n}.)
(b) Let A be an orthonormal subset of H. Then, for each x in H,
{(x,y)y: y E A} is summable and, if P is the projection of H on the closed
linear extension of A, then P(x) = 2 {(x,y)y: y E A}.
(c) An orthonormal basis for His an orthonormal subset A of H whose
closed linear extension is H. An orthonormal subset A is a basis if and
only if A.L = {0}. Given any orthonormal subset B of H, there exists an
orthonormal basis for H containing B; in particular any Hilbert space
admits an orthonormal basis.
(d) Let A be an orthonormal basis for H. Then for all x,y in H,
x = 2 {(x,y)y: y E A}, (x,y) = 2 {(x,z)(z,y): z E A} and llxll 2 = 2 {l(x,y)l 2 :
yEA}.
(e) Any two orthonormal bases of a Hilbert space have the same cardinal
number. Two Hilbert spaces are isomorphic, in the sense that there is an
algebraic isomorphism between them which preserves the norm and hence
also the inner product, if and only if the cardinal numbers of orthonormal
bases for the two spaces are equal.
(f) Every separable Hilbert space H has a countable orthonormal basis
{en: n = 1, 2, · · ·} and the mapping x-+ {(x,en): n = 1, 2, · · ·} is an iso-
morphism of H onto [2 • In general, for any Hilbert space, if Xis any set
with the cardinal nurober of an orthonormal basis for H, then H is iso-
morphic to l 2 (X). For each cardinal number, there is a Hilbert space
whose orthonormal basis has this cardinal number.
(g) Is the cardinal nurober of a Harne! base the same as that of an
orthonormal basis ?

M SPACES OF INTEGRABLE FUNCTIONS Ill (see 6K, 6N, 14M)

Foreach p > 0 the space U(X,fL) is complete with respect to the topol-
ogy of convergence in mean of order p. (Any Cauchy sequence contains
a subsequence {/n} with 2 {II /n + 1 - fn II P: n = 1, 2, · · ·} convergent, and
68 CH. 2 LINEAR ToPOLOGICAL SPACES

then Un(x)} converges a.e., to f(x), say. Fatou's Iemma now shows that
Je V' and 11/n- /IIP- 0.)
If J, g E L 2 , then Jg is integrable and {f,g) = Jjgdp. is an inner product
defining the norm. Thus U(X,p.) is a Hilbert space (see the previous
problems) and so every space L 2 (X,p.) is isomorphic to a suitable space
[2(Y).

N SPACES OF MEASURABLE FUNCTIONS Il (see 6L)

Let (X,Y',p.) be a finite measure space; then the space S(X,p.) is complete
relative to the topology of convergence in measure.

0 THE SUM OF CLOSED SUBSPACES

Let F and G be closed subspaces of a linear topological space E. The
result 7.4 states that if G is finite dimensional, then F + G is closed.
The sum of two closed subspaces generally fails to be closed, even if E is a
Hilbert space. In fact, if E is a Hilbert space and F n G = {0}, then
F + Gis closed if and only if the angle between Fand Gis zero (that is,
sup {l{x,y)l: x E F, y E G, llxll = IIYII = 1} < 1).

8 FUNCTION SPACES
This section is concerned with completeness and compactness proper-
ties of various families of functions on a set S to a linear topological
space E, where the family is given the topology ~ of uniform con-
vergence on members of a collection d of subsets of S. In particular,
we consider the family of functions which are bounded or totally
bounded on members of d, and we also study families of linear
functions and families of continuous functions. A criterion for
compactness of the latter (equicontinuity) and a surprising relation-
ship to countable compactness_ are obtained.
Let S be any set, and Iet E be a linear topological space. The set
F(S,E) of all functions on S to E, with addition and scalar multiplica-
tion defined pointwise, is a linear space. For each subset A of S,
and for each n<:;ighborhood U of 0 in E, Iet N(A,U) be the family of
all membersf of F(S,E) with the property thatf[A] c U. Observe
that N(A, U) is circled if U is circled. A subset Gof F(S,E) is open
relative to the topology of uniform convergence on A if and only
if for each f in G there is a neighborhood U of 0 in E such that
f + N(A, U) c G. It is easy to verify that this definition gives a
topology for F(S,E) such that the family of sets of the form f +
N(A, U) is a base for the neighborhood system of f. Howe':er, this
topology need not be a vector topology for F( S,E) because the
neighborhoods of 0 need not be radial at 0. In fact, if U is circled,
then N(A, U) is radial at 0 if and only if for each f there is a scalar s
such that f E sN(A, U), that is, f[A] c sU. It follows that if G is
 SEC. 8 FUNCTION SPACES 69
a subspace of F(S,E) such that G with the relativized topology of
uniform convergence on A is a linear topological space, then J[ A] is
a bounded subset of E for each member f of G. Weshall say that a
function f is bounded on a set A if and only if f[A] is a bounded
subset of E.
A net {Jy, y Er} in F(S,E) converges to f uniformly on a set A
if and only if the net converges to f relative to the topology of uniform
convergence on A. Clearly this is the case if and only if /y - f is
eventually in N(A,U), for each neighborhood U of 0 in E.
lf .91 is an arbitrary non-void family of subsets of S, then the topol-
ogy !/;; of uniform convergence on members of .91 is defined
to be the weakest topology which is stronger than that of uniform
convergence on A, for every A in d. Convergence relative to .9d
can be described in a somewhat less esoteric fashion: a net converges
to f relative to ffd if and only if the net converges to f uniformly on
each member of .91. A base for the neighborhood system of 0 relative
to .9d is the family of all finite intersections of sets of the form
N(A,U), where Ais a member of .91 and U is a neighborhood of 0 in
E. Since N(A,U) n N(B,U) = N(A u B,U), the topology of
uniform convergence on members of .91 is identical with the topology
of uniform convergence on members of ifß, where fJß is the family of
all finite unions of members of .91. lf the union of two members of
.91 is always contained in some member of .91 (that is, if .91 is directed
by :::> ), then the intersection of two sets of the form N(A, U) contains
a set of the same form, and consequently the family of such sets is a
base for the neighborhood system of 0.
The pointwise topology, or the topology of pointwise convergence,
is the topology ~ where .91 is the family of all sets {t} for all t in S.
The topology of uniform convergence on S is often called the uniform
topology. It is identical with the topology ~"' where .91 = {S}.
Many of the theorems of this section concern the topology of uniform
convergence on S. Most of these theorems have corollaries which
have to do with the topology .'Td of uniform convergence on each
member of a family .91 of subsets of the domain set S. The proofs of
these corollaries are, for the most part, omitted; they are Straight-
forward extensions of the theorems and of the foregoing remarks.
Throughout, E is a fixed linear topological space, and .91 is a fixed
family of subsets of S.

8.1 FUNCTIONS BOUNDED ON S The jamily B(S,E) of a/l junctions

from S to E which are bounded on S is a subspace of F(S,E) which is
70 CH. 2 LINEAR ToPOLOGICAL SPACES

closed relative to the uniform topology, and the uniform topology is a

vector topology for B( S,E ).
PROOF That B(S,E) is a subspace of F(S,E) is a C{)nsequence of
the relations (f + g)[S] c f[S] + g[S] and (tf)[S] = tf[S] and of
elementary computations with bounded sets. Let Ur, y E r} be a
net in B(S,E) which converges uniformly to f; it must be shown that
f E B( S,E ). Let U be a neighborhood of 0 in E, and let V be a
circled neighborhood of 0 such that V + V c U. Choose y such
thatf(t) - fy(t) E V for all t inS, and let a besuch that fr[S] c aV.
Then f[S] c V+ aV c max (l,lai)U. It follows that f[S] is
bounded. The definition of B(S,E) and the fact that E is a linear
topological space show that B(S,E) with the uniform topology is a
linear topological space.lll
It follows from the foregoing theorem that if a net of functions,
each of which is bounded on a subset A of S, converges uniformly on
A to a functionj, thenf is also bounded on A. The next proposition
is then an easy corollary.
8.2 CoROLLARY The class B .",( S,E) of all functions f on S to a linear
topological space E such that f is bounded on each member of d is closed
in F(S,E) relative to the topology :Y;; of uniform convergence on members
ofd.
Moreover, :Y;; is a vector topology for B.",(S,E).
A function f on S to a linear topological space Eis totally bounded
on a subset A of S if and only if f [A] is totally bounded; that is, a
function f is totally bounded on S if and only if the range of f is
a totally bounded subset of E.
8.3 FuNCTIONS WITH TüTALLY BouNDED RANGES The class T(S,E)
of all functions from S to E, each of which has a totally bounded range, is
a subspace of B(S,E) which is closed in the uniform topology.
PROOF If C is totally bounded, then C is bounded; hence, T( S,E) c
B( S,E ). If C and D are totally bounded, then so are C + D and aC
for all scalars a. It follows that T(S,E) is a subspace of B(S,E).
Suppose that the net Ur, y E F} converges uniformly to J, and that
each fr has a totally bounded range. Fora given neighborhood U of
0 in E, choose a neighborhood V of 0 such that V + V c U. Then
choose y so that f(t) - fr(t) E V for all t in S, and choose a finite
set C such thatfr[S] c C + V. Thenf[S] c C + U, andf[S] is
totally bounded.lll
There is an obvious corollary.
 SEC. 8 FUNCTION SPACES 71
8.4 CoROLLARY The class T sJ( S,E) of all functions on S to E, each
of which is totally bounded on each member of d, is closed relative
to ~ in the space B sJ( S,E) of all functions bounded on members of d.

The next theorem shows that certain completeness properties of E

are inherited by B(S,E) and T(S,E). The critical fact needed for
the demonstration is contained in the following Iemma.

8.5 LEMMA If Ur, y E T} is a net of functions on S to E such that

Uit), y E T} converges to f(t) for each t in a subset A of S, and if
Ur, y E T} is a Cauchy net relative to the topology of uniform con-
vergence on A, then Ur, y E T} converges to f umformly on A.

PROOF Let U be a closed neighborhood of 0 in E. If a is such that

fa(t) - f 8 (t) E U for all t in A and all ß ~ a, then fa(t) - f(t) E U
for all t in A, because U is closed. Hence Ur, y E T} converges to f
uniformly on A.lll
If Ur, y E T} is a net which is a Cauchy net relative to the uniform
topology, then Uit), y E T} is a Cauchy net in E for each t in S, and
if E is complete, it is then possible to choose a Iimit fU) of Ur(t),
y E T} for each t. Then Ur, y E T} converges to f uniformly, by the
preceding Iemma. Of course, the same argument applies to
sequences, if E is sequentially complete. The following proposition
is then clear.

8.6 CoMPLETENESS RELATIVE TO THE UNIFORM ToPOLOGY If Eis

complete or sequentially complete, then so is the space B( S,E) of bounded
functions on S to E, with the uniform topology.
A particular consequence of this theorem is that T(S,E) is com-
plete or sequentially complete if E is complete or sequentially com-
plete, for T(S,E) is closed in B(S,E).
The argument used in the proof of the foregoing theorem may be
extended to the case of uniform convergence on each member of a
family d of sets. If Ur(t), y E T} converges to f(t) for each t in
U {A: A E d}, and if Ur, y E T} is a Cauchy net relative to ~' then
this net converges to f uniformly on members of d. In view of this
fact it is easy to demonstrate the following proposition.

8.7 CoMPLETENESS RELATIVE TO ~ If Eis complete or sequentially

complete, then BsJ(S,E) has the same property relative to the topology
~.
72 CH. 2 LINEAR TOPOLOGICAL SPACES
There is another useful completeness property which is inherited
by the function space Bd( S,E ). The key to the proposition is the
observation that, if t belongs to a member A of .s#, then the map
which carries f in Bs1 (S,E) into f(t) is a continuous linear function
on B.o~(S,E), with the topology .?Jd, into E. (That is, uniform con-
vergence on A implies convergence at each point of A.) Con-
sequently, if Gis a subset of Bd(S,E) which is bounded (or totally
bounded) relative to .?Jd, then the set of all f(t) for f in G is
also bounded (totally bounded, respectively). It follows that if
{fy, y E T} is a Cauchy net in a bounded set G, and if E has the prop-
erty that closed bounded sets are complete, then {fy{ t), y E T} converges
to a Iimit f(t) for each t in U {A: A E d}. The argument given
before then demonstrates the following theorem.
8.8 BouNDED CoMPLETENESS If E has the property that each closed
bounded (or totally bounded) set is complete, then Bs11(S,E), with the
topology .?Jd, has the same property.
Of course Ts11(S,E), being a closed subspace of Bs11(S,E), shares
the property stated in the foregoing theorem.
If S has a topology, it is possible to consider continuous functions
from S to E. Although the topology of uniform convergence is
independent of any topology for S, the dass of continuous functions
on S has special properties relative to uniform convergence.
8.9 CoNTINUOUS FuNCTIONS The class C(S,E) of continuous
functions from a topological space S to a linear topological space E is a
closed linear subspace of F(S,E) relative to the uniform topology.
PROOF It is clear that C(S,E) is a linear subspace of F(S,E). To
show that C( S,E) is closed, assume that {fy, y E T} converges uni-
formly to j, and that all jy are continuous. The proof will be com-
pleted by showing that f is continuous. In order to show that f is
continuous at t 0 , Iet U be a neighborhood of 0 in E and V a neighbor-
hood of 0 such that - V + V + V c U. Choose y such that
fy{t) - f(t) E V for all t in S. Let W be a neighborhood of t 0 such
that jy(t) - jy(t 0 ) E V for all t in W; then f(t) - j(t 0 ) = f(t) -
fy(t) + jy(t) - j(t 0 ) + jy(t 0 ) - f(t 0 ) E {-V + V + V) c U for all t
in W. Thus f is continuous.lll
8.10 CoROLLARY The class of bounded continuous functions on S to
E is a closed linear subspace of B(S,E) with respect to the uniform
topology.
 SEC. 8 FUNCTION SPACES 73
As in the case of the earlier theorems, there is a corollary which
involves the topology .r,w. A function f is continuous on a set A if
and only if the function J, restricted to A, is continuous with respect
to the relativized topology. (The reader is reminded that a function
may be continuous on A but fail tobe continuous at any point of A.)
The foregoing theorem implies that if a net of functions continuous
on A converges uniformly on A to a function J, then f is continuous
on A. The following proposition is now manifest.
8.11 FUNCTIONS CoNTINUOUS ON MEMBERS OF d The jamily
Ca(S,E) of all Junctions which are continuous on each member of a
Jamily d of subsets of a topological space S is closed relative to Y;; in the
space F(S,E) of allfunctions on S to E.
A consequence of the preceding theorem is the fact that the family
of all functions which are bounded and continuous on each member
of d is closed relative to .'Y;1 in Bs1 (S,E), and is complete if E is
complete.
The notion of equicontinuity of a family of functions on a topo-
logical space S to a linear topological space E is a natural extension of
the usual definition for real valued functions, and the fundamental
theorems about equicontinuity are established by means of familiar
arguments. A family M of functions on S to E is equicontinuous
at a point s of S if and only if for each neighborhood U of 0 in E
there exists a neighborhood V of s in S such that f(t) - f(s) E U for
all t in V and for allfin M; that is,J[V] c f(s) + U for allfin M.
The family M of functions is equicontinuous if it is equicontinuous
at each point of S. If the functions in a net in F(S,E) form an
equicontinuous family, the net is called an equicontinuous net.
8.12 POINTWISE CLOSURE OF AN EQUICONTINUOUS F AMIL Y The
closure relative to the pointwise topology of a Jamily M which is equi-
continuous at a point t is equicontinuous at t.
PROOF Let t 0 in S and a member U of the neighborhood system of 0
in E be given. Choose a circled neighborhood V of 0 in E such that
V+ V+ V c U and a neighborhood W of t 0 such that f(t) -
f(t 0 ) E V for all t in Wand all f in M. It will be shown that, for all
f in the pointwise closure of M, f(t) - f(t 0 ) E U whenever t E W.
Let {fy, y E T} be a net in M which converges pointwise to J0 • For
t in W choose y such that f 0 (t) - jy(t) E V and jy(t 0 ) - f 0 (t 0 ) E V.
Then fo(t) - fo(to) = fo(t) - fy(t) + jy(t) - fy{to) + jy(to) - /o(to),
which is an element in V + V + V E V. III
74 CH. 2 LINEAR ToPOLOGICAL SPACES
The foregoing theorem implies that the pointwise closure of an
equicontinuous family of functions consists of continuous functions.
This is an important fact.

8.13 ToPOLOGIES FOR AN EQUICONTINuous F AMIL Y 11 M is an

equicontinuous lamily ol lunctions on S to a linear topological space E,
then the topology lor M ol pointwise convergence on a subset A ol S is
identical with the topology ol pointwise convergence on the closure A-
oiA.
PROOF If U is a circled neighborhood of 0 in the range space E and
s is a point of A- then, by equicontinuity, there is a neighborhood V
of s suchthat g(t) E g(s) + U for all members g of M and all t in V.
If t is a member of A n V, and if I and g are members of M such that
g(t) - l(t) E U, then a simple computation shows that g(s) - l(s) E
U + U + U. Hence {g: g(t) El(t) + U} c {g: g(s) El(s) + U +
U + U}. It follows that each of the neighborhoods of I relative to
the topology of pointwise convergence on A- contains a neighbor-
hood of I relative to the topology of pointwise convergence on A.jjj

8.14 JoiNT CoNTINUITY Let M be an equicontinuous lamily ol

lunctions on S to E, and Iet M have the topology ol pointwise con-
vergence on S. Then l(s) is jointly continuous in I andin s, in the sense
that the map ol M x S into E defined by (f,s)-+ l(s) is continuous
relative to the product topology.
PROOF Suppose that (f,s) E M x Sand that U is a neighborhood of
0 in E. If V is a neighborhood of s such that g(t) E g(s) + U for all
g in M and all t in V, and if h is a member of M suchthat h(s) El(s) +
U, then h(t) E h(s) + U c l(s) + U + U. That is, if (h,t) E {g:
g(s) El(s) + U} x V, then h(t) el(s) + U + U. Joint continuity
follows.jjj
The results which are most useful in the study of linear topological
spaces concern families of linear functions which are continuous or
equicontinuous. When E and F are linear spaces, the linear space of
alllinear functions on E into Fis denoted by L(E,F).

8.15 THEOREM Let E and F be linear topological spaces, Iet F be

complete, and Iet d be a lamily ol bounded subsets ol E. Then, with the
topology .r.,;, the space L(E,F) n B"'(E,F) (the lamily ol linear lunc-
tions on E to F which are bounded on each member ol d) is a complete
linear topological space.
 SEC. 8 FUNCTION SPACES 75
PROOF Consider first the particular case where F is Hausdorff and E
is the linear extension of the union of the members of .91. Then .rsr
convergence of a net of functions implies pointwise convergence, and
since the pointwise Iimit of linear functions is evidently linear, the
space L(E,F) n Bd(E,F) is a closed subspace of B 91 (E,F) and hence
complete.
The proof in the general case requires a little more manipulation.
One first shows that L(E,F) is complete relative to the topology of
pointwise convergence by decomposing F into the product of a
Hausdorff space G and a trivial space H by means of theorem 5.11
on non-Hausdorff spaces, and using the fact that Gis complete. A
~-Cauchy net of linear functions will be necessarily a Cauchy net
relative to the topology of pointwise convergence on A 0 = U {A:
A E .91} and will hence converge pointwise on the linear extension E 0
of A 0 , to some linear function f on E0 into F. This convergence will
be uniform on members of d in view of 8.5, and the net will then
converge relative to ~ to any linear function on E to F which is an
extension of j. The details of this proof are left to the reader.lll
The notion of an equicontinuous family M is particulady simple in
case the functions are linear. In this case the definition can be
phrased: M is equicontinuous at a point s of E if and only if for each
neighborhood U of 0 in F there is a neighborhood V of 0 in E such
that, if t Es + V, then f(t) Ej(s) + U for all f in M. Rephrased,
for U there is V such that f (r) E U whenever r E V and f E M. The
following proposition is then straightforward.
8.16 EQUICONTINUITY AND UNIFORM EQUICONTINUITY THEOREM
Let M be a family of linear functions on a linear topological space E to
a linear topological space F. Then the following conditions are equiv-
alent:
(i) M is an equicontinuous family;
(ii) M is equicontinuous at some point;
(iii) for each neighborhood U of 0 in F there is a neighborhood V of 0
in E such that j[V] c U for all f in M; and
(iv) if U is a neighborhood ofO in F, then n {j- 1 [U]:jE M} is a
neighborhood of 0 in E.
The discussion of equicontinuity is concluded with a proposition
concerning equicontinuity and uniform convergence on totally
bounded sets. The result relies on the fact, noted in the preceding
theorem, that an equicontinuous family of linear functions is "uni-
formly " equicontinuous.
76 CH. 2 LINEAR TüPOLOGICAL SPACES
8.17 THEOREM lf M is an equicontinuous family of linear functions
from a linear topological space E to a linear topological space F, then the
pointwise topology and the topology of uniform convergence on totally
bounded sets coincide on M.
PROOF I t is clear that the pointwise topology is weaker than the
topology of uniform convergence on totally bounded sets. In order
to prove the reverse, it is necessary to prove that, if a net {Tr, y E T}
in M converges pointwise to T 0 in M, then {Tr, y E T} converges
uniformly to T 0 on totally bounded sets. Let U be a neighborhood
of 0 in F; Iet V be a circled neighborhood of 0 in F such that V +
V + V c U; and Iet W be a neighborhood of Q in E such that Tin M
and x in W imply that T(x) E V. If Bis a totally bounded subset of
E, then choose Xv x 2 , • • ·, Xn such that B c U {x1 + W: i = 1, 2,
· · ·, n}. Next, choose an a such that Ty{x1) - T 0 (x1) E V for all
i = 1, 2, · · ·, n, and for all y ~ a. Then any x in B belongs to some
x1 + W, and Ty(x) - T 0 (x) = Ty(x) - T y(x1) + Ty(x1) - T 0 (x 1) +
T 0 (x 1) - T 0 (x) E V + V + V c U. It follows that T converges
uniformly to T 0 on B, and the proof is complete.lll
The final propositions of the section concern compactness, relative
to the topology of pointwise convergence, of a family of continuous
functions on a compact topological space. The most surprising
result is that countable compactness is a sufficient condition for
compactness for certain families of functions-a theorem which is of
considerable importance in the later study of the weak topology for a
linear topological space.

8.18 CoNTINUITY OF POINTWISE LIMITS Let S be a compact topo-

logical space, and let G be the space of all continuous functions on
Stoa compact metric space (Z,d) with the topology of pointwise
convergence. T hen for each subset F of G, the following conditions
are equivalent.
(i) Each sequence in F has a duster point in G.
(ii) For all sequences {Sm} in S and ftn} in F, it is true that
lim limfn(sm) = lim limfn(sm) whenever each of the Iimits
m n n m
exists.
(iii) The closure of F in G is compact (equivalently, each Iimit,
relati~·e to the topology of pointwise convergence, of members of F
is continuous).
PROOF Assurne (i), and let {sm J and { /n} be sequences inS and F,
 SEC. 8 FUNCTION SPACES 77
respectively, for which lim limfn(sm) and lim limfn(sm) exist. Let s
m n n m
in S and f in F be duster points of {sm} and Un}; then lim lim fn( sm) =
m n
limj(sm) = j(s) = limfn(s) = lim limfn(sm)·
m n n m
Next assume (ii), and Iet H be the space of all functions on Sinto Z
with the topology of pointwise convergence. By the theorem of
Tychonoff (see 4.1 or 4.2), the space H is compact. Therefore, in
order to prove (iii), it suffices to show that the closure of F in H
consists of continuous functions. Let f be an element of H be-
longing to the closure of F, and assume that f is not continuous at a
point s in S. Then there is a neighborhood U of f(s) in Z such that
each neighborhood of s contains a point t of S with f(t) not be-
longing to U. Take any / 1 in F; then there is an s1 in S such that
d(/1 (s),f1 (s 1 )) < 1 and f(s 1 ) rt U. Take / 2 in F so that d(/2 (s 1 ),
f(s 1 )) < 1 and d(fis),f(s)) < 1. Now choose s2 in S such that
d(/1(s),f1(s 2 )) < 1/~ (i = 1, 2) and f(s 2 ) rt U. Then take / 3 in F so
that d(f(si),f3 (sJ) < 1/2 and d(f(s),f3 (s)) < 1/2. Proceeding in
this way, one obtains sequences Un} and {sm} in F and S such that,
for each n, d(/1(s),t(s")) < 1/n (i = 1, 2, · · ·, n), d(f(s1 ),fn+l(sJ) <
1/n (j = 1, 2, · · ·, n), d(f(s),fn+l (s)) < 1/n, andf(sn)rt U. Then
lim limfn(sm) = limfn(s) = f(s), and limfn(sm) = f(sm) rt U. Since it
n m n n
is possible to take a subsequence of {sm} so that the corresponding
subsequence of {!(sm)} converges to a pointoutside of U, the assump-
tion that f is not continuous contradicts the iterated Iimit condition
of (ii).
That (iii) implies (i) is evident from the net characterization of
compact spaces (see Section 4 ).111
The next two propositions are essentially lemmas needed for the
proof of compactness of certain countably compact families of func-
tions. However, the results are not without interest in themselves.

8.19 LEMMA Let F be a family of continuous functions on a compact

topological space S to a metric space Z, and Iet f be a continuous function
on S to Z which belongs to the closure of F relative to the topology of
pointwise convergence. Then f is a duster point of some sequence in F.

PROOF Foreach positive integer n Iet sn be the space of n-tuples of

members of S, with the product topology. For each member
(sl, ... , Sn) of sn there is a member g of F such that the distance
d(g(s1),f(s1)) < 1/n for i = 1, · · ·, n, becausefis in the closure of F.
78 CH. 2 LINEAR ToPOLOGICAL SPACES
Because hoth g and f are continuous, it is true that max d(g (ti),
i
f(ti)) < lfn for all points (tu · · ·, tn) in a sufficiently small neighhor-
hood of (s 1 , · · ·, sn)· But sn is compact, and it follows that there is
a finite suhfamily Gn ofFsuchthat for every (s1 , • · ·, sn) in sn, for
some g in Gm it is true that max d(g(si),f(s,)) < lfn. If {gn} is a
i
sequence containing the union of a sequence of sm:h finite sets Gn,
then clearly f is a duster point of {gn}·lll
As a preliminary to the next theorem we recall that if Fis a countahle
family of functions on a set S to a metric space Z, then the weakest
topology :Y for S which makes each memher of F continuous is
pseudo-metrizahle. (Convergence of a net relative to :Y is equiv-
alent to convergence of its natural image in the Countahle product
X {Z: f E F} of metric spaces.) If each memher of F is continuous
relative to some compact topology IJlt for S then :Y is weaker than
IJlt and hence S, with :Y, is compact.

8.20 THEOREM Let G be the family of all continuous functions on

a compact space S to a compact metric space Z, Iet G have the topology
of pointwise convergence, and Iet F be a subfamily of G such that each
sequence in F has a cluster point in G. Then each function f which
belongs to the closure of F is a cluster point of a sequence in F, and each
cluster point of a sequence in F is the Iimit of a subsequence.

PROOF The closure F - of F in the space of all functions on S into Z

consists of continuous functions, by virtue of theorem 8.18, and it is
necessary to show that each memher f of this closure is a Iimit of a
sequence in F. Because of theorem 8.19 it may he assumed that F
is countahle, and, hy 8.18 again, the set F satisfies the iterated Iimit
condition: lim limfn(sm) = lim limfn(sm) whenever {sm} and Un} are
m n n m
sequences in S and F, respectively, such that each of these Iimits
exist. Let :Y he the weakest topology for S such that each memher
of F is continuous. In view of the remarks preceding the theorem,
the set S with the topology :Y is compact and pseudo-metric. In
brief, it may he supposed that S is compact and pseudo-metric, and
that the closure F- of F consists of continuous functions (this last
hecause of the iterated Iimit condition). There is a countahle dense
suhset S0 of S, and we will show that the topology for F - of point-
wise convergence on S 0 is identical with the topology of pointwise
convergence on S. This fact will establish the theorem, for the
space F- with the topology of pointwise convergence on S0 is pseudo-
 SEC. 8 PROBLEMS 79
metric (it is the weakest topology which makes evaluation at each of
the countable number of members of S0 continuous), and there is
consequently a sequence in F which converges to f. Finally, if a net
in F - converges to g relative to the topology of pointwise convergence
on S 0 , and h is a member of F - which is a duster point of the net
relative to the topology of pointwise convergence on S, then g = h on
S 0 and therefore g = h on S, because both g and h are continuous.
It follows without difficulty that pointwise convergence on S0 implies
pointwise convergence on S.JJJ
The final result of the section is an immediate consequence of the
two preceding theorems.

8.21 THEOREM ON CoUNTABLE CoMPACTNESS Let G be the Jamily of

all continuous functions on a compact space S to a compact metric space
Z, Iet G have the topology of pointwise convergence, and Iet F be a sub-
family of G suchthat each sequence in F has a duster point in G. Then
the closure of F in G is compact, and each member of this closure is the
Iimit of a sequenc.: in F.
In particular, each countably compact family of continuous functions
on S to Z is compact and sequentially compact.

PROBLEMS
A CONVERSE OF 8.1
Let G be a subspace of the space F ( S,E) of all functions on S to E. If
the uniform topology for G is a vector topology, then each function of G
is bounded.

ß MAPPINGS IN PSEUDO-NORMED SPACES II (see 6B)

Let E and F be pseudo-normed linear spaces and let d = { S }, where S
is the unit sphere in E. Then L(E,F) () Bs1(E,F) is the space, G say, of
all continuous linear mappings of E into F and :Ts1 is the pseudo-norm
topology for G(6B). Hence if Fis complete, Gis complete relative to the
pseudo-norm topology, and if F is a Banach space, so is G. In particular,
the space of continuous linear functionals on a pseudo-normed space is a
Banach space.
Suppose that E and F are Banach spaces. A linear mapping of E into
F is completely continuous iff it maps the unit sphere of E into a relatively
compact subset of F. The set L(E,F) () T s1(E,F) = H, say, is the set of
all completely continuous mappings of E into F, and is a closed subspace
of G. If E = F, then His an ideal in G (that is, if TE G and SE H, then
ToSEHand So TE H). (Cf. 21A-D.)

C POINTWISE CAUCHY NETS

Let E and F be linear topological spaces and Ua: a E A} an equicontinuous
80 CH. 2 LINEAR ToPOLOGICAL SPACES

net of linear mappings of E into F. If {fa(x): a E A} is a Cauchy net for

each x in a dense subset of E, then it is a Cauchy net for every x in E.

D PRODUCT OF :7;,; and ff91

Let E and F be linear spaces, Iet G be a linear topological space, and Iet
.91 and PA be families of non-void subsets of E and F, respectively. Then
the natural algebraic isomorphism on L(E x F,G) onto L(E,G) x L(F,G)
is a homeomorphism relative to .r'(l and ff",; x :T91 where t:t' = {(A u {0}) x
(B u {0}): A E d, BE PA}.
If .91 is a family of non-void subsets of E and if F and G are linear
topological spaces, there is a similar result for L(E,F x G).

E FUNCTIONAL COMPLETION
Let E be a linear topological space which is composed of real valued
functions on a set S, addition and scalar multiplication being defined as
they are usually defined in function spaces. Assurne that E has the property
(P) If Ua} is a net in E converging to J,
then fa(t) ~ f(t) for each t in S.
(In other words, the topology of E is stronger than that of pointwise
convergence.) Is it always possible to embed E in a complete linear
topological space which is composed of functions on S and which has the
property (P)?

F ADDITIVE SET FUNCTIONS

Let .91 be a ring of sets in the sense of Halmas [4] ; a function 4> on .91 to
the complex numbers is additive if </>(A U B) = </>(A) + </>(B) whenever A
and B in .91 are disjoint.
(a) Let BA(J?!) be the set of all bounded additive functions; then BA(J?!)
is a closed subspace of the space of all bounded complex-valued functions
on .91 with the uniform topology.
(b) When .91 is a a-ring, Iet CA(S?I) denote the set of all countably additive
functions on .91 to the complex numbers; 4> is countably additive iff
L: {</>(Ak)} = </>(U {Ak}) whenever {Ak} is a sequence of pairwise disjoint
elements of .91. Then CA(J?!) is a closed subspace of BA(J?!).

G BOUNDEDNESS IN B.s;(S,E)
Let S be a set, E be a linear topological space, and .91 be a family of sub-
sets of S. Then a subset Gof B",;(S,E) is ff",; bounded if and only if it is
uniformly bounded on each A in .91; this is the case if and only if {g(s):
g E G, s E A} is bounded for each A in .91.

H COMPACTNESS OF SETS OF FUNCTIONS

(a) Let S be a set and Iet E be a Hausdorff linear topological space.
Then a subset Gof the space F(S,E) of all functions on S to Eis compact
relative to the topology of pointwise convergence if and only if G is closed
and for each t in S, {g(t): g E G}- is compact.
 SEc. 8 PROBLEMS 81
(b) Let S be a compact space and let E be a Hausdorff linear topological
space. Then a subset Gof the space C(S,E) of all continuous functions
on S to E is compact relative to the uniform topology if and only if G is
closed relative to the uniform topology, G is equicontinuous and
{g(t): g E G}- is compact for each t in S. (On an equicontinuous family
of functions the uniform topology agrees with the topology of pointwise
convergence.)
(c) A topological space S is called a k-space if and only if a subset A of S
is closed whenever A n B is closed in B for all compact subsets B of S.
A Hausdorff space S which is either locally compact or satisfies the first
countability axiom (that is, at each point x of S there is a countable base
for the neighborhood system of x) is a k-space. A function f on a k-space
Sinto a topological space is continuous if and only ifJIB is continuous for
each compact subset B of S.
(d) A family G of continuous functio.ns on a topological space S to a
linear topological space E is called equicontinuous on a set A if and only if
the family of restrictions of members of G to A is equicontinuous. The
following theorem is a generalization of the classical Ascoli theorem.
Theorem Let S be a k-space, Iet E be a Hausdorff linear topological
space, and let d be the family of all compact subsets of S. Then a subset
G of C (S,E) is compact relative to the topology ffs~, the topology of uniform
convergence on compact sets, if and only if Gis ~-closed, Gis equicontinuous
on each compact subset of S, and {g(t): g E G}- is compact foreacht inS.

I SPACES OF CONTINUOUS FUNCTIONS l (see 141)

If X is a Hausdorff space, the space of all real or complex valued con-
tinuous functions on Xis denoted by C(X). Those functions of C(X)
which are bounded on X form a subspace denoted by B(X). A scalar
valued function on X is said to vanish at infinity iff to each e > 0 corres-
ponds a compact subset A of X suchthat lf(x)l < e whenever x EX "' A.
The continuous functions which vanish at infinity form a subspace of B(X)
denoted by C0 (X). The support or carrier of a scalar valued function f
on X is the smallest closed subset of X outside which f vanishes. The
continuous functions of compact support on X form a subspace of C0 (X),
denoted by K(X).
(a) With the topology of uniform convergence on X, B(X) is a Banach
space, C0 (X) is a closed subspace, and K(X) is a dense subspace of C0 (X).
The norm in B(X), defined by II/II = sup {1/(x)l: x EX}, is often referred
to as the supremum norm.
(b) With the topology .r of uniform convergence on compact subsets of
X, C(X) is a Hausdorff locally convex linear topological space. If X has
a sequence {Bn} of compact subsets, suchthat every compact subset of Xis
contained in some Bm the topology Y is metrizable. The converse
assertion holds when X is completely regular and Hausdorff. If X is a k-
space (see 8H(c)), C(X) is complete relative to the topology .r. If X is
locally compact, K(X) is densein C(X).
(c) Foreach compact subset B of X, let K 8 (X) be the subspace of K(X)
consisting of those functions whose supports are contained in B, and
82 CH. 2 LINEAR ToPOLOGICAL SPACES

consider on K 8 (X) the topology of uniform convergence on B. The con-

vex circled subsets of K(X) which intersect each K 8 (X) in a neighborhood
of the origin form a local base for a Hausdorff locally convex topology on
K(X) (called the inductive Iimit topology-see 16C(j)). It is the strongest
topology coinciding on each K 8 (X) with the topology of uniform conver-
gence.

j DISTRIBUTION SPACES l (see 201; also 16C(k), 16D(k))

Let x = (x 1 , x2 , · · ·, xn) be a point of Rn and Iet p = {p 1 , p2, · · · fn} be
a set of n non-negative integers; put IPI = P1 + P2 + · · · + Pn· For
each p, define the differential operator DP = 81Pijax1P18X2p2 ... axnPn.
(a) Let ~ff<ml be the space of real or complex valued functions on Rn
which are at least m times continuously differentiable. The topology of
compact convergence for all derivatives is defined by the family of pseudo-
norms {qK<ml: K compact, K c Rn}, where
qK<ml(j) = sup {IDPj(x)l: x E K, 0;;;; IPI ;;;; m}.
The space ~ff<ml is locally convex and metrizable; if K(r) = {x: lxd ;;;; r,
1 ;;;; i.;;;; n} for each positive integer r, the corresponding pseudo-norms
determine a countable local base. This topology is the weakest that
makes continuous each mapping DP(O ;;;; IPI;;;; m) of g<ml into C(Rn), the
space of continuous functions on Rn with the topology of compact con-
vergence (8I(b )).
(b) Let lff be the space of repeatedly differentiable real or complex
valued functions on Rn (i.e., functions which are infinitely many times
differentiable). The topology of compact convergence for all derivatives
is defined by the family of all pseudo-norms of the form qK<ml above, for
each compact subset K of Rn and for each m = 0, 1, 2, · · ·. The space lff
is locally convex and metrizable. The topology is the weakest that makes
each DP(I PI = 0, 1, · · ·) a continuous mapping of C into C(Rn). Foreach
p, the differential operator DP is a continuous mapping of C into itself.
(c) The spaces g<ml and C are complete. (If Ua: IX E A} is a Cauchy net
in g<m>, then each {DPfa: IX E A} is Cauchy and so DPfa ___,. gP for each p.
Show that g 0 E gcml and DPg 0 = gv.)
(d) Let [» K be the space of repeatedly differentiable real or complex
valued functions on Rn whose supports lie in a compact subset K of Rn.
The relativization to [»K of the topology of C has a countable base {qK<ml:
m = 0, 1, · · ·} of pseudo-norms. With this topology, [» K is a Frechet
space (a complete metrizable locally convex space).
(e) Let [» be the space of repeatedly differentiable real or complex
valued functions on Rn having compact support. It is a dense subspace of
lff. A more useful stronger topology for [» is defined by taking for a local
base the family of all convex circled subsets U of [»such that U n [»K is a
neighborhood of 0 in [»K for each compact subset K of Rn. This is the
strongest locally convex topology with the property that its relativization
to each [»K coincides with the original Frechet space topology for [»K·
For each p, DP is a continuous mapping of !!fi into itself.
 Chapter 3

THE CATEGORY THEOREMS

This short chapter is concerned with the concept of category and

with its application to the theory of linear topological spaces. The
results include some of the most profound and most useful theorems
of the subject of linear topological spaces, and are among the most
important of the applications of category.
In the first section, which treats category in topological spaces,
three primary tools are established. The first two are weil known
-Banach's condensation theorem and a classical result of Baire.
The third is a result of Osgood that in essence is a double Iimit
theorem.
The major part of the chapter is devoted to four theorems on
category in linear topological spaces. The first two of the theorems
concern subsets of a linear topological space, the third deals with con-
tinuity (and dually, openness) of a linear function, and the fourth
treats equicontinuity and uniform boundedness of a family of linear
functions. It may be remarked that, although certain of the results
of the chapter hold in a more general situation than that prescribed
here, the proofs of the generalizations depend on the theorems of this
chapter.
A remark on classification may be of interest. Two of the four
theorems (the difference theorem and the closed graph theorem) are
essentially propositions about topological groups. The other two
results depend on scalar multiplication (via convexity) in a very
simple fashion. Intensive use of scalar multiplication is not made
until the succeeding chapter, where convexity and a closely related
topic, continuous linear functionals, are investigated.
83
84 CH. 3 THE CATEGORY THEOREMS

9 CATEGORY IN TOPOLOGICAL SPACES

This section contains two theorems on category which are essential to
the discussion of linear topological spaces. The first of these is the
condensation theorem; from the fact that a set A is of the second
category it is inferred that there is a non-void open set of points at
which A is of the second category. The second theorem is the dassie
result of Baire on the category of a complete metric space (or a locally
compact Hausdorff space). The section concludes with Osgood's
theorem on the points of equicontinuity of a convergent sequence of
functions.

A subset A of a topological space Xis nowhere densein X (rare)

if and only if the interior of the closure of A is void. Equivalently,
A is nowhere dense in X if and only if the interior of the complement
of Aisdensein X. The set Ais of the first category in X (meager,
exhaustible) if and only if A is the union of a countable family of
sets each of which is nowhere dense in X, and A is of the second
category (inexhaustible) in X if and only if it is not of the first
category. A set A is said to be of the first or second category if
and only if A is, respectively, of the first or second ·category in itself
(that is, in A).
The following theorem implies that if A is of the second category
in X, then there is a non-void open set W, such that every non-void
open subset of W intersects A in a set which is of the second category
in X.

9.1 THEOREM Let A be a subset of a topological space X, and let W

be the union of alt open sets which intersect A in a set which is of the first
category in X. Then the intersection of A with the closure of W,
A n W-, is of the first category in X.

PROOF Choose a disjoint family UU of open sets such that UU is

maximal with respect to the following property: for V in UU, V n A
is of the first category in X. Then-and this is the critical point of
the proof-the intersection of A with the union U of the members
of UU is of the first category in X. For if, for each V in UU, V n A is
the union of a sequence {Bv.n} of nowhere dense sets, then the union
Bn = U {Bv.n: V E UU} is, foreachpositive integer n, nowhere dense.
It follows that A n U is of the first category in X, and since U - "' U
is nowhere dense, A n U- is of the first category in X. Finally,
because UU is maximal, U - contains every open set which intersects
A in a set of the first category in X; therefore U - = W- .JJJ
 SEC. 9 CATEGORY IN TOPOLOGICAL SPACES 85
The theorem above is sometimes phrased in a slightly different
fashion. A subset A of a topological space is of the first category
at a point x if and only if there is a neighborhood W of x such that
W n A is of the first category in X; if A is not of the first category at
x, then it is, by definition, of the second category at x. In this case,
for every neighborhood U of x, U n A is of the second category in X.
It is clear from the definition that the set of points at which A is of
the first category is open; it is precisely the set W of theorem 9.1, and
in view of this theorem, W is identical with the interior of W-.
Consequently A is of the second category at the points of the comple-
ment of the interior of W-, which is the closure of X '""' W -. The
following proposition is then clear.

9.2 CoNDENSATION THEOREM The set of points at which a subsei A

of a topological space X is of the second category is the closure V - of an
open set V. Moreover, the intersection of A with the complement of V
is of the first category in X.
The existence of sets which are of the second category is a problern
of the first importance. In spite of a great deal of work on the sub-
ject, there is essentially only one method known for showing that a
set is of the second category. Some refinement of the following
results is possible, but the next two propositions are the basic existence
theorems.

9.3 THEOREM If X is a topological space which is either a H ausdorff

or a regular space, and if A is a compact subset with non-void interior,
then A is of the second category in X.
The proof of theorem 9.3 is omitted; it is an obvious variant of the
proof of the Baire theorem, 9.4.
A particular consequence of theorems 9.3 is that a locally compact
Hausdorff space is of the second category in itself, and each non-void
open subset of a compact, regular (or Hausdorff) space is of the
second category in the space.
The other result required depends on a non-topological concept.
Let X be a set and d a pseudo-metric on X. A sequence {xn} is a
Cauchy sequence relative to d if and only if d(xm,xn) converges to
zero; that is, if and only if for every e > 0 there is an integer p such
that d(xm,xn) < e whenever m ~ p and n ~ p. A subset A of Xis
complete relative to d if and only if each sequence in A which is a
Cauchy sequence relative to d converges to some point of A.
86 CH. 3 THE CATEGORY THEOREMS
9.4 BAIRE THEOREM If X is a space with a pseudo-metric d, and if A
is a complete subset of X with non-void interior, then A is of the second
category in X.

PROOF Suppose A is the union of a countable nurober of nowhere

dense sets: A = U {Fn: n = 1, 2, · · ·}. Since F 1 is nowhere dense,
the open set Ai "' F 1 - is non-void, and it follows that there is a
dosed sphere V1 of radius at most one such that V1 c Ai "' F 1 -.
Recursively, select a dosed sphere Vn of radius at mostz-nsuch that
Vn c (Vn_ 1 )i ,..., Fn-. This selection is possible because each Fn is
nowhere dense. The intersection of the sets Vn is non-void, because
A is complete, and a point of this intersection is dearly a member of
A which belongs to no Fn- This is a contradiction.lll
The statement of the Baire theorem has a rather curious form,
since a topological condusion (that a set A is of second category in X)
is deduced from a non-topological premise (that the set A is com-
plete relative to a pseudo-metric d). It is dear that a formally
stronger result of the following form can be stated: if X is a topo-
logical space for which there exists a pseudo-metric d such that X has
the pseudo-metric topology, then a subset A of X which has a non-
void interior and is complete relative to d is of the second category in
X. This question is examined further in a problern on metric
topological completeness at the end of the next section.
The fin'al theorem of this section concerns the limit, relative to the
topology of pointwise convergence, of a sequence of continuous
functions. This proposition has to do with the set of all points of
equicontinuity of a sequence, although the original statement of the
theorem was in terms of "almost uniform convergence." We recall
that a dass F of functions on a topological space X to a pseudo-metric
space ( Y,d) is equicontinuous at a point x if and only if for each
e > 0 there is a neighborhood U of x such that d (f(x),f(y)) < e for
all y in U and all f in F. The dosure, relative to the topology of point-
wise convergence, of a dass which is equicontinuous at x is equi-
continuous at x, and, in particular, each member of the dosure is
continuous at x. A subset Z of a topological space X is residual in
X if and only if X "' Z is of the first category in X.

9.5 Oscooo THEOREM If Un} is a sequence of continuous functions

on a topological space X into a pseudo-metric space ( Y,d), and zf the
sequence Un} converges pointwise to a function f on X to Y, then Un} is
equicontinuous at the points of a residual subsei of X.
 SEC. 9 PROBLEMS 87
In particular,j is continuous at the pointsofaresidual subset of X.
PROOF Because the sequence Un(x)} is convergent, the requirement
of equicontinuity of Un} at x ~an be rephrased: for e > 0 there are a
neighborhood U of x and a positive integer k such that d (fp(y),
A(y)) < e for y in U and for p ;;:; k. Let Kn,k be the closed set
defined by Kn,k = {x: d(JP(x), A(x)) ~ 1/n for p ;;:; k}, and let Kn =
U {Kn,ki: k = 1, 2, · · · }, the union of the interiors. Then the set of
points of equicontinuity is the intersection n{ Kn: n = 1, 2, · · ·} of
the sets Kn. On the other hand, U {Kn,k: k = 1, 2, · · ·} = X, since
the sequence Un} is convergent. Hence X "" Kn c U {Kn,k "" Kn,ki:
k = 1, 2, · · · }, which is obviously of the first category in X, and
hence X"" n {Kn: n = 1, 2, · · ·} is of the first category.lll

PROBLEMS
A EXERCISE ON CATEGORY
The set of rational numbers is not the intersection of a countable number
of open subsets of the space of all real numbers.

ß PRESERVATION OF CATEGORY
Let X and Y be topological spaces, and Iet f be a continuous mapping of
X into Y such that the image of each open subset of X is a somewhere
dense subset of Y. Let A be a subset of X and B a subset of Y.
(a) If Bis nowhere densein Y,f- 1 [B] is nowhere densein X.
(b) If B is of the first category in Y, then f- 1 [ B] is of the first category
in X.
(c) If A is of the second category in X, f [A] is of the second category
m Y.

C LOWER SEMI-CONTINUOUS FUNCTIONS

Let E be a topological space and Iet f/> be a family of lower semi-
continuous real functions on E, that is, the set {x: x E E, 4>(x) ~ a} is closed
in E for each real number a and each 4> in f/>. Suppose that the set of x in
E for which sup {4>(x): 4> E f/>} is finite is of the second category in E.
Then there is a non-void open subset G of E and a positive integer k such
that sup {4>(x): 4> E fl>, x E G} ~ k.

D GENERALIZED BAIRE THEOREM

A subset A of a topological space X is of the second category in X if and
only if Ais somewhere densein X and there exists a sequence {ßn: n =
1, 2, · · ·} offamilies of open subsets of X with these properties: (i) for any
closed nowhere dense set N and any non-void open set G in X such that
G c A- and G n N is void, there exists for each positive integer n an
element En of @"n such that N n En is void and G n En is not void;
88 CH. 3 THE CATEGORY THEOREMS

(ii) if {En: n = 1, 2, · · ·} is a sequence with En in Cn and A n Pn not

void for each n, where Pn = n {E1 : i = 1, 2, · · ·, n}, then A n Pis not
void, where P = n {E1 : i = 1, 2, · · ·}.

E EMBEDDING OF A FINITE DIMENSIONAL COMPACT METRIC SPACE INTO AN

EUCLIDEAN SPACE
A most elegant application of the Baire category theorem is the proof by
W. Hurewicz of the "embedding theorem" for finite dimensional separable
metric spaces. This theorem does not have a direct connection with
linear space theory, but the beauty of the proof more than justifies the in-
clusion of the theorem here. Let X be a set, then a finite covering of X is
a finite family of subsets of X whose union is X. If X is a topological
space and each member of a finite covering is open, then the covering is
called open. The order of a finite covering of X is the smallest non-
negative integer n such that any n + 2 distinct members of the covering
have an empty intersection. Equivalently, the order of a finite covering is
the largest integer n such that some n + 1 distinct members of the ·covering
have a non~empty intersection.
In order to illustrate the method, we restriet attention to compact metric
spaces. A finite covering of a metric space is called an e-covering if and
only if the diameter of each member of the covering is less than e. A
compact metric space Xis said to be of dimension r, where r is an integer,
if and only if (i) for each positive number e, there is a finite open e-covering
of X of order less than or equal to r, and (ii) there is a positive number e
such that each finite open e-covering of X is of order greater than r - 1.
Let En denote the Euclidean n-space, i.e., the n-dimensional real linear
topological space in which a topology is given by the norm defined by
llxll = (L; {x,2 : i = 1, 2, · · ·, nW 12 where x1 is the i-th component of the
vector x in En. A map on a metric space X into a space Y is said to be
an e-map if and only if the diameter of f - 1 [{y}] is less than e for each y
in Y.
Lemma Let X be a compact metric space of dimension less than or
equal to n, and Iet f be a continuous map of X into E 2 n + 1 • Then, for any
two positive numbers e and d, there is a continuous e-map g on X into
E 2 n+ 1 suchthat sup {llf(x) - g(x)ll: x EX} < d.
(Since f is uniformly continuous there is a positive number e1 , such that
e1 < e and llf(x) - f(y)ll < df2 whenever d(x,y);;;;; e1 (d is the metric for
X). Let {U 1 , U 2 , · · • , Uk} be a finite open e1 -covering of X of order less
than or equal to n. Choose vectors Pi> i = 1, 2, · · ·, k, in E2 n + 1 so that, for
each i, dist (p1,f[ U,]) < df2 and {p 1 , p 2 , · · ·, pk} are in general positipn in
E 2 n+l• that is, if t;'s are real numbers such that t 1 p11 + · · · + t1Pu = 0,
j ~ 2n + 2, and t 1 + · · · + t1 = 0, then t 1 = · · · = t1 = 0. [This is
equivalent to: for m ;;;;; 2n + 2, no m points of p 11 • • • , pk are contained in a
translate of a linear subspace of dimension m - 2.] For each i, Iet t/11 =
d(x, -U1), and Iet c/J1 = (L;{tfo1 :i= 1,···,k})- 1 tfo1 ; then L:{c/J1 :i= 1,2,
· · ·, k} = 1, c/J1 G; 0, and c/J 1(x) > 0 if and only if x E U1• Now Iet g(x) =
L: {c/J1p1 : i = 1, 2, · · ·, k}; then g is a continuous e-map such that
sup{llf(x)- g(x)ll x EX} < d.)
 SEC. 9 PROBLEMS 89
Theorem Let X be a compact metric space of dimension less than or
equal to n. Then X can be homeomorphically mapped into E 2 n+l·
(Let E be the space of all continuous maps on X into E 2 n+l> with the
topology of uniform convergence. Let B. be the set of all continuous e-
maps on X into E 2 n+l· Then B. is an open dense subset of E.)

F LINEAR SPACE OF DIMENSION N0

An infinite dimensional Hausdorff linear topological space that is of the
second category cannot be the union of a countable number of finite-
dimensional subspaces. In particular, an infinite dimensional complete
metrizable space cannot be of linear dimension N0 •

G IMAGE OF A PSEUDO-METRIZABLE LINEAR SPACE

The example in 6G shows that a continuous image of a pseudo-metrizable
linear topological space need not be pseudo-metrizable.
(a) Let T be an additive mapping of a linear topological space E into a
linear topological space F such that the image of each neighborhood of 0
in E is a somewhere dense set in F. Then the closure of the image of each
neighborhood of 0 in E is a neighborhood of 0 in F. If, in addition, T is
continuous and E is pseudo-metrizable, then F is pseudo-metrizable.
(b) If E is a pseudo-metrizable linear topological space, then the image
of B under a continuous linear mapping is either pseudo-metrizable or of
the first category.
(c) If Eis a pseudo-metrizable subspace of a linear topological space F,
then the closure of E is pseudo-metrizable. If F is not pseudo-metrizable,
then E is nowhere dense in F.

H ADDITIVE SET FUNCTIONS

Let Y be a u-ring of subsets of a set X and {4in} a sequence of finitely
additive scalar-valued functions on Y; suppose that cf>n(A) --? cf>(A) for
each A E Y.
(a) If each cfon is absolutely continuous with respect to a measure p, on
Y, then {cfon} is equiabsolutely continuous [called uniformly absolutely con-
tinuous by Halmos] and cf> is absolutely continuous with respect top,.
(b) If each cfon is countably additive, so is cf>.
(c) Let .9 0 be a subset of Y suchthat (i) if A, BE .9 0 , then A U BE .90
and (ii) if A E .90 and B c A, then B E .90 , and Iet .90 be directed by c.
(This will be applied to the set of subsets of a measure space which have
finite measure.) If each net {cf>n(A): A E .90} converges, then the con-
vergence is uniform in n and cf> also converges.
(Let Yv. be the set of subsets of Y of finite measure; put d(A,B) =
1-'(A ....., B) + p,(B- A) for A, BE Yv.. Then Yv. is a complete metric space.
A finitely additive function on Y is absolutely continuous if and only if it is
a continuous function on Y v.; for this, it is sufficient that it be continuous at
one point of Y v.· A set C/> of such functions is equiabsolutely continuous
if and only if it is an equicontinuous set of mappings on Yv.. Osgood's
theorem 9.6 gives (a).
To prove (b ), it is sufficient to construct a finite measure v so that each
90 CH. 3 THE CATEGORY THEOREMS

cfon is absolutely continuous with respect to v. For then applying (a) with
this rneasure v, it follows that cfo is countably additive, v being finite. Now if
!Ln is the total variation of cfon (Halrnos [4] p. 122; also 141), then Cn =
sup{,un(A):AE9'} < 00 andv = 2;{2-"(1 + cn)- 1 p.n:n = 1,2,···} will
suffice.
For (c), apply (a) again with the rneasure v to deduce that cfon(B) is srnall
uniforrnly in n if P.n(B) is srnall enough for a finite nurnber of n's, say
1 ~ n ~ m; use the convergence along 9'0 of cfon for 1 ~ n ~ m to find
A E 9' 0 so that B n A = cfo and BE 9'0 irnply that P.n(B) is srnall for 1 ~
n ~ m.)

I SEQUENTIAL CONVERGENCE IN Ll(X,p.) (see 16F)

Let {/n} be a sequence of elernents of V(X,p.) such that {JAfnd,u} con-
verges for each rneasurable subset A of X.
(a) There exists f E V(X,,u) with lirn fAfndp. = fAfdp. for each rneasur-
able A c X. (Use (a) and (b) of the previous problern with cfon(A) =
fAfnd,u and the Radon-Nikodyrn theorern.)
(b) Suppose als.o that Un} is convergent in rneasure on every subset of
finite rneasure. Then Un} is convergent in norrn.
(We rnay suppose, in view of (a), that fAfndp.--+ 0 for each A. Apply
(c) of the previous problern with 9' 0 = 9'v., the set of subsets of finite
rneasure, to show the existence of a set A E 9'v. with fx-A lfnld,u srnall
uniforrnly in n. It follows frorn (a) ofthat problernthat Ja 1/nldp. is srnall
uniforrnly in n if p.(B) is srnall.)

10 THE ABSORPTION THEOREM AND THE DIFFERENCE

THEOREM
The results of this section have to do with the deduction of bounded-
ness and continuity properties frorn prernises about category. The
two principal theorerns are the following. lf A is a closed convex
syrnrnetric set in a linear topological space E such that A absorbs each
point of a convex set B, and if B is bounded and of the second category
in itself, then A absorbs B (the absorption theorern). lf A is a subset
of E that is of the second category in E and that satisfies the condition
of Baire, then A - Ais a neighborhood of 0 (the difference theorern).
There are corollaries concerning the absorption of cornpact convex
sets and the continuity of linear functions.
A subset A of a linear space absorbs a set B, and Bis absorbed by
A, if and only if there is a positive number r suchthat B c sA for all
numbers s such that s G; r. This definition is clearly an extension of
the notion whereby boundedness was defined; a set is bounded if
and only if it is absorbed by each Qeighborhood of 0.

10.1 ABSORPTION THEOREM Let A and B be convex sets in a linear

topological space E such that A is closed and absorbs each point of
 SEC. 10 ABSORPTION THEOREM AND DIFFERENCE THEOREM 91
B U (- B). lf B is bounded and of the second category in itself, then A
absorbs B.
PROOF By the hypotheses, there is some positive integer N such that
n {nA n B: n ~ N} is somewhere dense in B. Consequently there
is a point y of B and a neighborhood U of 0 in E such that nA =
(nA)- :::> ( U + y) n B for n ~ N. Since B is bounded, so is
B - B; thus there is a real number b, 0 < b < 1, such that
U :::> b(B - B) and hence nA :::> (y + b(B - B)) n B for
n ~ N. The set y(1 - b) + bB is contained in y + b(B - B) and,
since Bis convex, it is also contained in B; therefore, nA :::> y(1 - b)
+ bB for n ~ N. The set A absorbs each point of - B, and
consequently there is an integer m, which may be supposed to be
larger than N, such that -y(1 - b) E rA for r ~ m. Then rA
contains (1/2)[y(1 - b) + bB] + (1/2)[ -y(1 - b)] = (1f2)bB, and
(2rfb)A :::> B, for all r ~ m.JJJ
There is a very useful corollary to the absorption theorem; it is
obtained by recalling from theorem 9.3 that each compact subset of a
regular topological space is of the second category in itself.

10.2 CoROLLARY A closed convex set A absorbs a compact convex

set B if A absorbs each point of B U (- B).
The following is also a useful consequence of the absorption
theorem. A set A is called sequentially closed if A contains the
Iimit points of each sequence in A.

10.3 CoROLLARY Let A be a closed convex set and B be a sequentially

closed, convex, circled and bounded set which is sequentially complete.
Then A absorbs B if A absorbs each point of B.
PROOF Without loss of generality, one can assume that Bis radial at
0. In view of the absorption theorem it is sufficient to exhibit a
linear topology which is stronger than the original one :Y, and with
respect to which the set B is closed and of the second category in
itself. Let p be the Minkowski functional of B; then p is a pseudo-
norm, and the pseudo-norm topology is stronger than :Y since B is
bounded. Clearly B is closed relative to the new topology. If B is
shown ta be complete with respect to the pseudo-metric associated
with p, then Bis of the second category in itself relative to the pseudo-
norm topology (Baire theorem 9.4). Let {xn} be a Cauchy sequence
relative to p in B; then it is a :Y-Cauchy sequence. Since B is
sequentially complete, the sequence {xn} converges to a point y in B
92 CH. 3 THE CATEGORY THEOREMS
relative to !T. Given a positive number e, there is an integer k such
that p(xm - xn) ~ e whenever m,n ~ k, or equivalently xm - Xn E eB
whenever m,n ~ k. Because B is sequentially closed, xm - y E eB,
that is p(xm - y) ~ e for all m ~ k. Therefore, the sequence {xn}
converges to y relative to the pseudo-norm topology.lll
A subset A of a topological space X satisfies the condition of
Baire if and only if there is an open set U such that both U ,..., A and
A ,..., U are of the first category in X. Roughly speaking, a set
satisfies the condition of Baire if it is open-give and take a set of the
first category. Each closed set satisfies the condition of Baire
because it is equal to the union of its interior and its boundary, and
the latter is nowhere dense. More generally, the complement of any
set satisfying the condition of Baire also satisfies the condition. It is
evident that the union of a countable number of sets satisfying the
condition of Baire is a set of the same sort, and it follows that the
family of all sets satisfying the condition is closed under complements,
countable unions, and countable intersections. In particular, every
Borel set satisfies the condition of Baire.
The proof of the ditference theorem will require one elementary
fact about category in linear topological spaces. If a subset of such a
space Eis of the second category in E, then Eis of the second category.
Each set A which is radial at 0 is then of the second category in E
because: E c U {nA: n = 1, 2, · · ·}; hence nA is of the second
category in E for some n, and therefore A itself is of the second
category in E. It follows by a translation argument that each set
which is radial at some point, and .in particular every non-void open
set, is of the second category in E whenever Eis of the second category.
10.4 DIFFERENCE THEOREM fj a subset A oj a linear topological
space Eis of the second category in E and satisfies the condition of Baire,
then A - A is a neighborhood of 0.
PROOF For convenience, for this proof only, let us agree that a set A
fills an open set U if U ,..., A is of the first category. Each set which
is of the second category in E and satisfies the condition of Baire fills
some non-void open set. If A fills an open set U and B fills an open
set V, then A n B fills U n V; in particular, if U n V is non-void,
then A n B is non-void, because each open set is of the second
category. Consequently, if A fills U and (x + U) n U is non-
void, then (x + A) n A is non-void. But the set of all points x
suchthat (x + U) n U is non-void is precisely the set U - U, by an
elementary calculation (x = u - vif and only if x + v = u). Hence,
 SEC. 10 ABSORPTION THEOREM AND DIFFERENCE THEOREM 93
A - A contains the open neighborhood U - U of 0 whenever A fills
the non-void open set U, and the theorem follows.iil
There are a number of important results which are Straightforward
corollaries to the difference theorem.

10.5 BANACH'S SUBGROUP THEOREM Jj G is a subset of a linear

topological space E such that G- G c G (that is, Gis a subgroup of E),
and if G contains a set A that is of the second category in E and satisfies
the condition of Baire, then G = E. Hence if Gis any proper subgroup
of E, then either G is of the first category in E or eise G fails to satisfy
the condition of Baire.
PROOF Let N = (A - A)i; by the difference theorem N is a neighbor-
hood of 0. Since G => G - G => N and 0 E N, we have G => G -
G => G - N => G, and hence G = G - N. Since N is open, so is
- N; it follows that G - N is open and that G- c G - N. Hence
G = Qi = G-. From the connectedness of E it follows that
G = E.lll
The first assertion in the following corollary has been proved
earlier; the second is a direct consequence of the difference theorem.

10.6 CoROLLARY If E is a linear topological space of the second

category and A is a subset of E that is radial at some point, then A is of
the second category in E. If A also satisfies the condition of Baire,
then A - A is a neighborhood of 0.

10.7 CoNTINUITY AND ÜPENNESS Let E and F be linear topological

spaces and T an additive map of E into F. If for each neighborhood V
of 0 in F the set T -l[V] contains a set that is of the second category in
E and that satisfies the condition of Baire, then T is continuous.
Dually, ~f for each neighborhood U of 0 in E the set T[U] contains
some set that is of the second category in F and that satisfies the condition
of Baire, then T is an open mapping onto F.
PROOF Let V be any neighborhood of 0 in F, and choose a neighbor-
hood W of 0 in F such that W- W c V. Then T -l[V] =>
T - 1 [W] - T -l[W], where T -l[W] contains some set A that is of
the second category in E and satisfies the condition of Baire. Then
A - A is a neighborhood of 0 that is contained in T - 1 [V], and
hence T is continuous. The remaining half of the theorem is proved
dually.lll
Proofs of the following two corollaries are quite easy to construct
and will be omitted. Both use the simple observation that any
94 CH. 3 THE CATEGORY THEOREMS
additive operation T is partially homogeneous, in the sense that
T (~) = T~x) for every non-zerointeger n and every x in the domain
ofT.

10.8 CoROLLARY Let E and F be linear topological spaces and T an

additive map of E into F. If E is of the second category, then T is
continuous if and only if T - 1 [V] satisfies the condition of Baire for every
open neighborhood V of 0 in F.
Dually, if Fis of the second category, then T is an open mapping onto
F if and only if T [ U] is radial at 0 and satisfies the condition of Baire
for every open neighborhood U of 0 in E.

10.9 CoROLLARY Let E and F be linear topological space$ and T an

additive map of E into F. lf T [A] is bounded in F for some set A of the
second category in E that satisfies the condition of Baire, then T is
continuous.
Dually, if T [A ] is of the second category in F and satisfies the condition
of Baire for some bounded set A in E, then T is an open mapping.
The assumption on T in the first part of the last corollary can be
varied in case F is the scalar field K. Recall that a real valued
function r is upper semi-continuous if and only if {x: r(x) ~ a} is
closed for each real number a. A function r is lower semi-
continuous if - r is upper semi-continuous.

10.10 CoNTINUITY OF LINEAR FuNCTIONALS Let f be a linear

functional on a linear topological space E, Iet r be the real part of f, and
Iet A be a set which is of the second category in E and satisfies the condi-
tion of Baire. Then f is continuous if
(i) r is bounded from above (or below) on A, or
(ii) r is upper (or lower) semi-continuous on A.
PROOF Suppose that r is bounded from above on A. Because A is
of the second category and satisfies the condition of Baire, there is a
non-void open set U such that (E ,. . , A) n U is of the first category.
Making an obvious translation argument, it may be supposed that U
is a symmetric neighborhood of 0. Then [- (E ,. . , A)] n (- U) =
[ E ,...., (- A)] n U is of the first category, and it follows that V = A n
(- A) is of the second category and satisfies the condition of Baire.
But V is symmetric, and r is therefore bounded from both above and
below on V. Then r is also bounded on V - V, which, by the
 SEC. 10 PROBLEMS 95
difference theorem, is a neighborhood of 0. Hence r is continuous
and therefore f is continuous. This establishes the first statement of
the theorem.
To establish the second statement, suppose that r is upper semi-
continuous on A. Foreach positive integer n let An = {x: x E A and
r(x) < n}. Then An is open in A, and therefore, being the inter-
section of A and an open set, An satisfies the condition of Baire.
Finally, since A is of the second category, it is true that An is of the
second category for some n, and the result of the preceding paragraph
implies that f is continuous.lll

PROBLEMS
A CONTINUITY OF ADDITIVE MAPPINGS
(a) Let T be an additive function on one linear topological space E to
another, F. Suppose that there is a set A of the second category in E and
satisfying the condition of Baire such that T considered as a function only
on A is continuous. Then T is continuous as a function on E. (By a
translation argument and the condensation theorem, A can be assumed
to be of the second category at 0.)
(b) Any pseudo-norm defined on a linear topological space that is
bounded on some non-void open set, or on some set of the second category
that satisfies the condition of Baire, is continuous.
(c) Let T be an additive function on a linear topological space E to a
pseudo-normed space F with pseudo-norm p. Then T is continuous if
either of the following two conditions holds:
(i) there is a real function 4> defined on some set A of the second category
in E such that tf>(x) ~ p(T(x)) for each x in A, and the set An =
{x: x E A, tf>(x) ;;;; n} satisfies the condition of Baire in E for each
positive integer n;
(ii) E is pseudo-metrizable and of the second category, and
lim sup p(T(xn)) ~ p(T(x))
whenever lim Xn = x in E.

ß SUBSPACES OF THE SECOND CATEGORY

Let E be any infinite dimensional linear topological space of the second
category. Then E contains a maximal proper linear subspace F that is of
the second category in E. Moreover:
(a) F does not satisfy the condition of Baire.
(b) There exists a discontinuous linear functional on E. Thus if E is
an infinite dimensional complete metric (or normed) space it must contain
a subspace that is infinite dimensional, of the second category and metric
(or normed) but not complete.
(There is a sequence {en} of linearly independent elements of E. Let
B U {en} be a Hamel base of E and Iet Hn be the (maximal proper) linear
subspacespanned by B U {em: m =F n}. Then E = U {Hn: n = 1, 2, · · · }.)
96 CH. 3 THE CATEGORY THEOREMS

C LINEAR SPACES WITH PSEUDO-METRIZABLE TOPOLOGY

Let E be a linear space and .r a pseudo-metrizable topology for E such
that x + y and ax are continuous in each variable and E is of the second
category relative to .?/. Then .r is a vector topology for E. (By means of
Osgood's theorem, show that x + y is continuous in x and y simultaneously,
and similarly for ax.)

D MIDPOINT CONVEX NEIGHBORHOODS

(a) In a linear topological space E any subset of the second category in
E which satisfies the condition of Baire must contain a translate of a sym-
metric set which also satisfies the conditions.
(b) Any set A which is midpoint convex (A + A = 2A) in a linear
topological space E and which contains a symmetric set that is of the
second category in E and satisfies the condition of Baire is necessarily a
neighborhood of 0.
(c) Any midpoint convex set that contains a set which is of the second
category in E and which satisfies the condition of Baire has a non-void
interior.

E SETS OF SEQUENTIAL CONVERGENCE

(a) If {Tn} is a sequence of continuous linear maps of a linear topological
space E into a pseudo-metrizable linear topological space, then the set of
points x for which {Tn(x)} is Cauchy is either of the first category in E or
identical with E.
(b) Let {T m.n: m,n = 1, 2, · · ·} be a double sequence of continuous
linear maps of a linear topological space E into a pseudo-metrizable linear
topological space, and suppose that for each m there is an Xm in E such that
the sequence {T m,n(xm): n = 1, 2, · · ·} is not Cauchy. Then the set of all
points x of E, for which {T m,n(x): n = 1, · · ·} is not Cauchy for all m, is
residual.

F PROBLEMS IN TOPOLOGICAL COMPLETENESS AND METRIC COMPLETION

A metrizable space (X,Y) is called topologically complete if and only if it
is homeomorphic to a complete metric space. A topological space (X,.?/)
is an absolute G" if and only if it is metrizable, and is a G" (a countable
intersection of open sets) in every metric space in which it is topologically
embedded. In the following, (a), (b), and (c) are Iemmas for (d) and (e).
(a) Let (X,d) be a complete metric space, Iet U be an open subset of X,
and Iet .'Y be the induced metric topology on U. Then ( U,.'Y) is topo-
logically complete. (For x in U, Iet f(x) = ( dist (x,X ~ U)) -l, and Iet
d*(x,y) = d(x,y) + lf(x)- f(y)i. Then (U,d*) is a complete metric
space.)
(b) AG" subset of a complete metric space with the relative topology is
topologically complete. (If U = n{ Un: n = 1, 2, · · · }, consider the map
of U into the product of the complete metric spaces ( Un,dn *), where dn * is
constructed from d and un as in (a).)
(c) If a topological subspace Y of a Hausdorff space Xis homeomorphic
to a complete metric space (Z,d), and if Y is densein X, then Y is a G6 in
 SEC. 11 THE CLOSED GRAPH THEOREM 97
X. (One can assume that Y = Z. Let Un be the set of all points x in X
such that, for some neighborhood V of x, the diameter of V n Y is less
than lfn. It is easy to show that Y = n{Un: n = 1, 2, · · ·}.)
(d) A metrizable space (X,d) is topologically complete if and only if it
is an absolute Gö.
(e) If a metrizable linear topological space is topologically complete,
then it is complete.

11 THE CLOSED GRAPH THEOREM

The closed graph theorem and the open mapping theorem, the two
principal theorems of this section, are among the most celebrated
results of linear space theory. In these theorems continuity and
openness of mappings are deduced from formally weaker conditions
by means of arguments involving category. There is an important
corollary on the continuity of the inverse of a continuous map.
As a preliminary to the proof of the closed graph theorem we
recall that if A is a subset of a linear topological space E, and if E
is of the second category in itself, and if A is radial at 0 (or at any
other point), then A is of the second category in E. In particular,
if T is an arbitrary linear map of a space E which is of the second
category in itself into a linear topological space F, and if V is a
neighborhood of 0 in F, then T -l[V] is of the second category in E.
Then the interior of T- 1 [V]- is non-empty and r- 1 [V]- -
T - 1 [V]- is a neighborhood of 0. But (T -l[V- V])- =
(T- 1 [V]- r- 1 [V])- => r- 1 [V]-- T- 1 [V]-, and itfollowsthat
T -l[V- V] is dense in some neighborhood of 0. Consequently
such a map has the property that the inverse of each neighborhood
of 0 in F is dense in some neighborhood of 0 in E.
11.1 CLOSED GRAPH THEOREM A linear Iransformation T of a
linear topological space E into a complete pseudo-metrizable linear
topological space F is continuous provided
(i) the graph of T is closed in E x F, and
(ii) for each neighborhood V of 0 in F the closure ofT -l[V] ts a
neighborhood of 0 in E.
Condition (ii) is automatically satisfied if Eis of the second category.
PROOF By the metrization theorem 6.7, there is an invariant pseudo-
metric d for F, whose topology is that of F. For a positive number e,
Iet S(e) = {y: d(O,y) :;;:; e}. We prove the theorem by showing that
r- 1 [S(e)]- c r-
1 [S(2e)]. Let XE r-
1 [S(e)]-; then we construct
98 CH. 3 THE CATEGORY THEOREMS
inductively a sequence {yn: n = 0, 1, · · ·} in F such that (a) y 0 =
0, d(YmYn+ 1) ~ ef2n (n = 1, 2, · .. ), and (b) XE T -l[Yn + S(ef2n)]-
(n = 0, 1, 2, · · · ). Clearly y 0 ( =0) satisfies (a) and (b). Next
assume that y 0 , • • ·, Yn have been chosen. Then by (b) and (ii),
XE T- 1 [Yn + S(ef2n)] + r- 1 [S(e/2n+l)]-. Hence there is an ele-
ment x' of E such that T(x') E Yn + S(ef2n) and x Ex' +
T- 1 [S(ef2n+l)]- c T- 1 [T(x') + S(ef2n+ 1 )]-. Let Yn+l = T(x');
then (a) and (b) are clearly satisfied. F or n = 0, 1, 2, · · ·, let Wn =
Yn + S(e/2a- 1 ). Then by (a), Wn =:J Wn+l· Since Fis complete,
there is an element y in n {Wn: n = 0, 1, · · ·} by theorem 7.5.
Obviously y E W0 = S(2e).
In view of (i), the proof is complete if it is shown that the point
(x,y) is in the closure of the graph of T. Let U and V be arbitrary
neighborhoods of x and y respectively. Then there is an n such that
Wn c V. lt follows from (b) that XE r- 1 [Wn]- c r- 1 [V]-.
Hence u n r- 1 [V] -# 4> or equivalently u X V intersects the graph
of T.JJJ
There are some notable consequences of the closed graph theorem
-the following proposition is one of the most useful.

11.2 CoNTINUITY OF INVERSE MAPS A one-to-one continuous linear

map of a complete pseudo-metrizable linear topological space onto a
Haus darf! linear topological space of the second category is necessarily
a topological isomorphism.

PROOF The inverse of such a transformation has a closed graph, the

closed graph theorem applies, and the inverse is therefore
continuous.JJI
A particular case of the foregoing result concerns two Hausdorff
topologies, .'T 1 and .'T2 , for a single linear space E. lf .'T1 c .'T 2 ,
then the identity map of E with .'T 2 into E with .'T1 is continuous,
and hence its inverse (also the identity) has a closed graph. Con-
sequently, if E with .'T1 is a linear topological space of the second
category, and if E with .'T 2 is a complete metrizable linear topological
space, then the identity map is a topological isomorphism, and
.'Tl = .'T2.
The argument used in the preceding paragraph may be extended to
discuss the relation between two distinct topologies .'T 1 and .'T2 on a
linear space E by considering the topology .'T 1 n .'T 2 and the topol-
ogy whose subbase is .'T 1 U .'T 2 • This line of investigation is
 SEC. 11 THE CLOSED GRAPH THEOREM 99
pursued in a problern at the end of this section, the most striking
result being: if E with each of f7 1 and f7 2 is a complete metrizable
linear topological space, and if f7 1 =f f7 2 , then there is a non-zero
vector x which belongs to U + V whenever U is a .?7 1 -neighborhood
of 0 and V is a f7 2 -neighborhood of 0.
The open mapping theorem is dual to the closed graph theorem;
actually, either theorem may be derived fairly easily from the other.
As a preliminary to the open mapping theorem we prove a Iemma
which is of some interest in itself.

11.3 LEMMA The image under an open, continuous linear map T of a

complete pseudo-metrizable linear topological space E is complete and
pseudo-metrizable.

PROOF lt is easy to see that the family of images of the members of a

local base for Eis a local base for the range space F, and consequently
the range space is pseudo-metrizable. Suppose {yk} is a Cauchy
sequence in F which has no Iimit point, and suppose that {Un} is a
local baseforEsuch that U 1 = E, and Un+ 1 + Un+ 1 c Un for each
n. Since the family of images of the sets Un is a local base for F, it
may be supposed (by choosing a subsequence of {yk} if necessary)
thatyk- Yk- 1 E T[Uk] for all k. Choose z 1 suchthat T(z 1) = y 1 ,
and for k > 1 choose a member zk of Uk such that T(zk) = Yk -
yk_ 1, and let xk = L {zi:j = 1, · · ·, k}. It is clear that T(xk) = Yk
for all k, and {xk} is a Cauchy sequence because xk - xk _ 1 E U k and
hencexkH>- XkE uk+1 + uk+2 + ... + uk+p c uk. Butthen{xk}
converges to a point x of E, and therefore T(xk) = Yk converges
to a point of F. This is a contradiction.lll

11.4 ÜPEN MAPPING THEOREM Let T be a continuous linear map of

a complete pseudo-metrizable linear topological space E into a H ausdorff
linear topological space F. If the rangeofT is of the second category in
F, then T maps E onto F, Fis complete and metrizable, and T is an open
mapping.

PROOF The null space N of T is closed in E because {0} is closed in

F. Consequently the quotient space EfN is Hausdorff, and since
the quotient map Q of E onto EfN is open and continuous, the pre-
ceding Iemma implies that EfN is complete and metrizable. Finally,
there is a continuous induced map S of EfN into F suchthat So Q =
T (see theorem 5.8). Since the range T [E] of T is of the second
category in F, it is dense in F and of the second category in itself.
100 CH. 3 THE CATEGORY THEOREMS

Hence, by theorem 11.2, it follows that S is a topological isomorphism

of EfN onto T[E]. In particular, T[E] is complete, and hence
T[E] = F. It follows that T is open, and the space Fis complete
and metrizable.lll

PROBLEMS
A COMPARISON OF TOPOLOGIES
Let f/ 1 and f7 2 be topologies for a linear space E such that (E,f/ 1 ) and
(E,f/ 2 ) are complete metrizable linear topological spaces. Then the
following are equivalent:
(i) f/1 =!= f7 2·
(ii) The vector topology, whose subbase is f/ 1 U f/ 2 is not complete.
(iii) The topology f/ 1 n f/ 2 is not Hausdorff.
(iv) There is a non-zero vector x such that x E U + V for each f/ 1-
neighborhood U of 0 and for each f7 2 -neighborhood V of 0.
(v) There is a sequence {xn} in E suchthat Xn ~ 0 relative to f/ 1 and
Xn ~ y =!= 0 relative to f7 2 •
(Prove in the order (i) => (ii) => (iii) => (iv) => (v) => (i).)

B suBsPAcE oF LP nu
Let E be a closed linear subspace of each of D'(f') and Lq(!-')(0 < p,
q ~ oo ). Then the two topologies induced on E coincide. (Show that
(v) of the previous problern cannot hold by extracting a subsequence
convergent almost everywhere.)

C SYMMETRIC OPERATORS
A linear operator T (defined everywhere) on a Hilbert space His sym-
metric if and only if (Tx,y) = (x,Ty) for all x and y in H.
A symmetric linear operator is continuous.
Part (e) of 7I can now be stated as follows. An idempotent symmetric
linear operator (defined everywhere) on a Hilbert space His the projection
of H on some closed linear subspace.

0 AN OPEN MAPPING THEOREM

Let En be a complete metrizable linear topological space for n = 1, 2, · · ·
and Iet un be a continuous linear mapping of E,. into a linear topological
Hausdorff space E. lf E = U {un[En]: n = 1, 2, · · · }, then every con-
tinuous linear mapping of E onto a complete metrizable space is open.
(Let f be a continuous linear mapping of E onto a complete metrizable
space F; then, for some n, the range of f o un is of the second category in
F.) Cf. 19B.
E CLOSED RELATION THEOREM
A relation between two sets X and Y is a subset of X x Y. Let R be
 SEC. 11 PROBLEMS 101
a relation between X and Y and Iet A be a subset of X; then R[ A] denotes
the set {y: y E Y and, for some x in A, (x,y) ER}. If E and F are linear
spaces, then a linear relation between E and F is a linear subspace of
EX F.
Let E and F be complete metrizable linear topological spaces and Iet R
be a linear relation between E and F such that R is closed in E x F and
R[E] is of the second category in F. Then R[U] is open in F whenever
U is open in E, and R[E] = F. (Notice that R is a complete metrizable
space, and consider the projection of R into F.)

F CONTINUOUSL Y DIFFERENTIABLE FUNCTIONS

Let C be the space of all continuous real valued functions on [0,1],
with the supremum norm, and Iet E be a closed subspace of C consisting
of functions which have continuous derivatives. Then E is finite dimen-
sional. (Show that differentiation is continuous on E, and hence that the
unit sphere of Eis equicontinuous.)

G MAPPINGS INTO THE SPACE L 1

Let E be a metrizable linear space and Iet T be a linear mapping of E
into the space V(X,fL). If, for each measurable subset A of X the linear
functional x --+ f A T(x)d!L on Eis continuous, then T is continuous. (First
extend T to the completion of E, using 91.)

H CONDITION FOR A CLOSED GRAPH

Let T be a linear mapping of a linear topological space E into a linear
topological space F. Then the graph of T in E x F is closed if and only
if n{T[U] + V:
in E and F.
uE Oll, V E 1"} = {0}, where 0/( and "rare local bases

I CLOSED GRAPH THEOREM FOR METRIZABLE SPACES?

It is an immediate consequence of the closed graph theorem 11.1 and
the Baire category theorem that a linear mapping of one complete metrizable
linear topological space into another is continuous if (and only if) its graph
is closed. This is no langer true if the hypothesis of completeness on
either one of the two spaces involved is dropped out (but cf. 12E). One
way of seeing this is to notice first that, on any infinite dimensional normed
space, there is a strictly stronger and a strictly weaker norm topology.
(Let {Xn: n = 1, 2, · · ·} be a linearly independent set, with II Xn II = 1 for
each n, and Iet {ya: a E A} be a set, with IIYa II = 1 for each a, which,
tagether with {xn: n = 1, 2, · · ·}, makes up a Harne! base. Take for
the strictly smaller unit sphere the convex circled extension of the set
{n- 1 xn: n = 1, 2, · · ·} U {ya: a E A}. A strictly larger unit sphere may
be reached by carrying out this first construction in the adjoint space.)
102 CH. 3 THE CATEGORY THEOREMS

j CONTINUITY OF POSITIVE LINEAR FUNCTIONALS

With certain restrictions on a linear topological space with a vector
ordering, every positive linear functional is continuous. Let E be a
metrizable linear topological space of the second category and Ca complete
cone with C - C = E. If f(x) ;:::; 0 on C, then f is continuous. (Let
{Un: n = 1, 2, · · ·} be a local base with Un+l + Un+l c Un. Use the
open mapping theorem 11.4 to show that {Un n C - Un n C: n =
1, 2, ... } is also a local base. If f is not bounded by 1 on un n c, there
is a sequence of points Xn E un n c so that Yn = L Xr ~ y and
l~r~n
f(y) ;:::; f(Yn) ;:::; n for all n.)

12 EQUICONTINUITY AND BOUNDEDNESS

Suppose that F is a family of continuous linear functions on a linear
topological space E to a linear topological space H. It is not hard to
see that, if F is equicontinuous, then F is bounded relative to the
topology of uniform convergence on boumied subsets of E, and that
boundedness relative to this topology implies boundedness relative to
the topology of pointwise convergence. The surprising fact (and this
is the import of this section) isthat the converses of these propositions
also hold provided suitable conditions are imposed.

Throughout, F will be a family of linear functions, each on a linear

topological space E to a linear topological space H. Recall that such
a family F is equicontinuous if and only if for each neighborhood V
of 0 in H there is a neighborhood U of 0 in E suchthat f[ U] c V for
all f in F. Restated, the family Fis equicontinuous if and only if, for
each neighborhood V of 0 in H, it is true that the set n
(! - 1 [V]:
f E F} is a neighborhood of 0 in E.
The family F is uniformly bounded on a subset A of E if and
only if the set of all points of the form f(x), for f in F and x in A, is
a bounded subset of H. The set of all such points f(x) will be
denoted by·F [A]; in this terminology, Fis uniformly bounded on A
if and only if F [ A] is a bounded subset of H. Restating again, F is
uniformly bounded on A if and only if each neighborhood V of 0 in
H absorbs F [A]. The family of sets of the form {f: f linear on E to
H, f[A] c V}, where Visa n·eighborhood of 0 in H, is a local base
for the topology of uniform convergence on A, and clearly V absorbs
F [ A] if and only if {f: f [A] c V} absorbs F. lt follows that F is
uniformly bounded on A if and only if F is bounded relative to the
topology of uniform convergence on A. Finally, a neighborhood V
of 0 in H absorbs F [A] if and only if the set n
{f - 1 [V]: f E F}
absorbs A. It follows that if Fis equicontinuous and A is bounded,
then Fis necessarily uniformly bounded on A.
 SEc. 12 EQmcoNTINUITY AND BouNDEDNEss 103
The family F is said to be pointwise bounded on a set A if and
only if for each point x of A it is true that the set F[x] of all points
f(x), for f in F, is a bounded subset of H. Equivalently, Fis point-
wise bounded if and only if Fis uniformly bounded on {x} whenever
x E A. It is clear that if F is uniformly bounded on A, then F is
pointwise bounded on A. Finally, the family of all sets of the form
{!: f(x) E V}, where x E A and V is a neighborhood of 0 in H, is a
subbase for the neighborhood system of 0 relative to the topology of
pointwise convergence on A, and it follows without difficulty that F
is pointwise bounded if and only if it is bounded relative to the topol-
ogy of pointwise convergence on A. The foregoing facts may be
summarized as follows.

12.1 ELEMENTARY RELATIONS BETWEEN EQUICON-

TINUITY AND BOUNDEDNESS
Let F be a family of linear functions, each on a linear topological space
E to a linear topological space H. Then the equivalences listed in (ii),
(iii), and (iv) hold. Moreover, the condition (i) implies each of the
conditions listed in (ii ), and each of the conditions in (iii) implies each
of the conditions of (iv).
(i) The family Fis equicontinuous.
(ii) The family F is uniformly bounded on each bounded set.
<'>- F is bounded relative to the topology of uniform convergence
on bounded sets.
(iii) The family Fis uniformly bounded on a set A.
<'>- The set F[A] is bounded.
<'>- F is bounded relative to the topology of uniform convergence
on A.
<'>- For each neighborhood V of 0 in H the set n {f- 1 [V]:
f E F} absorbs A.
(iv) The family Fis pointwise bounded on A.
<'>- F[x] is bounded for each x in A.
<'>- F is bounded relative to the topology of pointwise convergence
on A.
<'>- Foreach neighborhood V of 0 in H the set n {j - 1[V]: f E F}
absorbs each point of A.
The major theorem of this section states that uniform boundedness
on each bounded set, andin fact equicontinuity, may be deduced from
pointwise boundedness under suitable circumstances.
104 CH. 3 THE CATEGORY THEOREMS
12.2 BANACH-STEINHAUS THEOREM If F is a family of continuous
linear functions, each on a linear topological space E to a linear topo-
logical space G, and if Fis pointwise bounded on some set which is of the
second category in E, then F is equicontinuous.
PROOF Let U be a neighborhood of 0 in G, let V be a closed neighbor-
hood such that V - V c U, and let A be a subset of E which is of
the second category in E and such that Fis pointwise bounded on A.
The set w n
= {f -l[V]: f E F} absorbs each point of A in view of
12.1 (iv); hence some integer multiple of W is of the second category
in E; and hence W is of the second category in E. But the set W is
closed, and consequently the interior of W is non-void and W - W
isaneighborhoodofO. Finally,n {f- 1 [U):jEF} => n {f- 1 [V)-
f-1[V]: jEF} => W- W, and the family F is therefore equi-
continuous.lll
A very useful formulation of the basic principle is geometric in
nature. A barrel (or disk) in a linear topological space Eis a closed,
circled, convex set D which is radial at 0. If E is of the second
category, then each barrel D is of the second category (because D is
radial at 0), and hence the interior of D is non-void (because D is
closed). Consequently D - D is a neighborhood of 0, and it is true
that D - D = D + D = 2D because D is convex and circled. It
follows that each barrel in a space E which is of the second category
in itself is necessarily a neighborhood of 0. This simple fact is note-
worthy.
A locally convex space with the property that each barre! is a
neighborhood of 0 is said to be barrelled. There are linear topo-
logical spaces which are of the first category which, nevertheless, have
this barrel property (see problern 12D). The following theorem
applies to these.

12.3 BARREL THEOREM Let E be a linear topological space such that

each barre/ is a neighborhood of 0, and Iet F be a pointwise bounded
family of continuous linear functions on E to a locally convex space H.
Then the family Fis equicontinuous.
Consequently, in this case, F is umformly bounded on each bounded
subset of E.
If V is a closed convex circled neighborhood of 0 in H, then
n
PROOF
f - 1 [V] is a barrel, for each f in F. The intersection {f - 1 [V]:
f E F} is evidently closed, convex and circled, and because Fis point-
wise bounded, this intersection absorbs each point of E (part (iv) of
 SEC. 12 PROBLEMS 105
proposltlOn 12.1). In brief, nu-1 [V]:jEF} is a barre!, hence a

neighborhood of 0 in E, and F is proved to be equicontinuous.lll

The following theorem, which yields uniform boundedness in
certain cases where equicontinuity does not hold, is essentially a
corollary to the absorption theorem.

12.4 UNIFORM BouNDEDNESS THEOREM Let F be a family of

continuous linear functions, each on a linear topological space E to a
locally convex space H. If A is a subset of E which is convex, circled,
bounded, sequentially closed and sequentially complete, and if F is point-
wise bounded on A, then Fis uniformly bounded on A.
PROOF Let V be a closed, convex and circled neighborhood of 0 in H.
Then the set =w n{!- 1 [V] :f E F} is a closed, convex and circled
set which absorbs each point of A (theorem 12.1(iv)). In view of
corollary 10.3, the set W absorbs A; hence Fis uniformly bounded
on A (theorem 12.1(iii)).lll

PROBLEMS
A BOUNDEDNESS OF NORMS OF TRANSFORMATIONS
Let {Ta: a E A} be a family of continuous linear mappings of a pseudo-
normable space E of the second category into a pseudo-normed space
(F,p). If sup{p(Ta(x)): aEA} < oo for each x in E, then sup{IITall:
a E A} < 00.

ß THE PRINCIPLE OF CONDENSATION OF SINGULARITIES

Let {T m,n: m,n = 1, 2, · · ·} be a double sequence of continuous linear
maps of a pseudo-normable space E of the second category into a pseudo-
normed space (F,p) such that, for each m, sup {II Tm.nll: n = 1, 2, · · ·} = oo.
Then there is a residual subset A of E such that, for each x in A,
sup {p(Tm.n(x)): n = 1, 2, · · ·} = oo for all m. (See 10E.)
C BANACH-STEINHAUS THEOREM
Let E and F be linear topological spaces and suppose that {Tn: n =
1, 2, · · ·} is a sequence of continuous linear mappings of E into F such that
T(x) = !im Tn(x) exists for all x in a subset of E of the second category.
If F is complete, then T(x) = lim Tn(x) exists everywhere, and T is
continuous and the convergence is uniform on every totally bounded set.
(Use 12.2, 8.13, and 8.17.)
If E and F are locally convex spaces, the same conclusion holds if
lim Tn(x) exists for all x in E and Eis barrelled (by 12.3).

D STRONGEST LOCALLY CONVEX TOPOLOGY li (see 61, 14D, 20G)

A linear space E is barrelled when it has its strongest locally convex
topology. It is then of the second category if and only if it is finite dimen-
sional. (Use lOB and 7.3.)
106 CH. 3 THE CATEGORY THEOREMS

E CLOSED GRAPH THEOREM I (see 13G, 18J, 19B; also 11I, 201)
In the Statement of the closed graph theorem 11.1, the condition (ii) is
also automatically satisfied when the spaces are locally convex and E is
barrelled. Thus any linear mapping, with a closed graph, of a barrelled
space into a complete locally convex pseudo-metrizable space is con-
tinuous. It is natural to try to generalize these conditions on the two
spaces E and F. In the case of F, this may be clone (see 13G, 181). On
the other hand, the condition that E be barrelled is, in a certain sense,
necessary: suppose that E is a locally convex space with the property that
every linear mapping, with a closed graph, of E into any Banach space is
continuous. Then Eis barrelled. (Let B be a barre! and N = n {(1/n)B:
n = 1, 2, · · · }. Let F be the completion of EfN, normed by taking Q[B]
as its unit sphere, Q being the quotient mapping. It is sufficient to prove
that the graph of Q is closed; one way of doing this is to use the condition
in llH, and the fact that Q[B] is a closed subset of EfN.)
The conditions on the two spaces E and F cannot be reversed: even if E
is a Banach space and F a complete barrelled space, the theorem fails.
This situation may be realized by anticipating a result of 14D, stating that
any space is complete for its strongest locally convex topology (see the
previous problem).
F CONTINUOUS FUNCTIONS NON-DIFFERENTIABLE ON SETS OF POSITIVE
MEASURE
Let E be the space of continuous real-valued functions on [0,1], with the
norm given by llxll = sup {jx(t)i: 0 ~ t ~ 1}, and let F be the space of
Lebesgue measurable functions on [0, 1] with the topology of convergence
in measure (6L). Suppose the functions x extended beyond t = 1, for
example by periodicity, and for each n = 1, 2, · · · Iet Tn be the linear
mapping of E into F defined by Tn(x)(t) = n(x(t + 1/n) - x(t)).
(a) Each Tn is continuous.
(b) There are functions of E which are non-differentiahte on sets of
positive Lebesgue measure. (If each x were differentiable almost every-
where, then lim Tn(x) would exist for each x: use 12C.)
G BILINEAR MAPPINGS
Let E, F, and G be three linear topological spaces. A mapping f of
E x F into Gis called bilinear iff, for each y E F, the mapping x ~ f(x,y)
isalinear mapping of E into G and, for each x E E, the mapping y ~ f(x,y)
is a linear mapping of F into G. When G is the scalar field, f is called a
bilinear functional.
(a) A bilinear mapping f is continuous if and only if it is continuous at
the origin of E x F. The continuous bilinear mappings of E x F into
G form a linear space B. If E, F, and G are pseudo-normed spaces, with
pseudo-norms p, q, and r, f is continuous iff there is a constant k with
r(f(x,y)) ~ kp(x)q(y). The smallest such constant is sup {r(f(x,y)):
p(x) ~ 1, q(y) ~ 1}; denoting it by s(f), s is a pseudo-norm for B. If r is
a norm, so is s.
(b) If E and F are barrelled metrizable spaces and G is locally convex,
every pointwise bounded family of continuous bilinear mappings of E x F
 SEC. 12 PROBLEMS 107
into Gis equicontinuous. (Let {fa: cx E A} be the family. Foreach closed
convex circled neighborhood W of 0 in G, and each x E E, {y:fa(x,y) E
W for all cx} is a barrel in F. Hence if Yn----'>- 0, {fa(x,yn): cx E A, n =
1, 2, · · ·} is bounded in G. Thus {x:fa(x,yn) E W for all cx and all n} is a
barrel in E, so that, if Xn----'>- 0, {fa(xmYn): cx E A, n = 1, 2, · · ·} is bounded
in G.)
(c) If E, F, and G are locally convex, E and F are metrizable and E is
barrelled, every separately continuous bilinear mapping f of E x F into G
is continuous (f is called separately continuous iff all of the linear mappings
y----'>- f(x,y) and x----'>- f(x,y) are continuous). (The proof is similar to that
of (b).)
(d) Let E, F, and G be locally convex spaces and f a separately con-
tinuous bilinear mapping of E x F into G. Also Iet A be a bounded sub-
set of E and B a complete, sequentially closed, convex, circled bounded
subset of F. Then /[ A x B] is bounded in G. (If W is a closed convex
circled neighborhood of 0 in G, {y: f(x,y) E W for all x E A} is a barrel in F.
Now use 10.3.)
 Chapter 4
CONVEXITY IN LINEAR TOPOLOGICAL
SPACES

This chapter, which begins our intensive use of scalar multiplica-

tion in the theory of linear topological spaces, marks the definite
separation of this theory from that of topological groups. The
results obtained here do not have generalizations or even analogues in
the theory of groups.
The first section is devoted to technical matters concerning convex
subsets of linear topological spaces and of locally convex linear topo-
logical spaces. W e find, for example, that the convex extension of a
bounded or totally bounded subset of a locally convex space is of the
same sort.
The second section is devoted primarily to the existence of con-
tinuQus linear functionals; the principal theorems of the section are
simple consequences of the separation and extension theorems of
section 3. However, these propositions signal the beginning of the
study of the duality which is the central feature of the theory of linear
topological spaces. The power of linear space methods lies in the
fact that many natural and important problems can be attacked by
directing primary attention to the family of all continuous linear
functionals on the space. The dass of all continuous linear func-
tionals is itself a linear space, and in many important cases this dass,
with a suitable topology, determines the original space to a topological
isomorphism. In such cases a powerful duality exists, and each
problern concerning the linear topological space has an equivalent
formulation which is a statement about the space of continuous linear
functionals. It may happen that an apparently difficult problern may
have a formulation which is dual in this sense and which is much
108
 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES 109
easier to investigate. (As an example of the usefulness of this duality,
recall that one of the simplest ways of showing that the square of a
measurable function is integrable is to show that its product with
each square-integrable function is integrable.) But in order to
construct a duality between a linear topological space and the space of
its continuous linear functionals it is necessary that continuous linear
functionals exist, and exist in some abundance. Because of this fact
the propositions of Section 14 play a critical role in the development.
It is also the duality theory which lends importance to the concept
of local convexity. There are linear topological spaces such that the
only continuous linear functional is the functional which is identically
zero. The space of Borel functions on the unit interval, with the
topology of convergence in Lebesgue measure, is an example of such
a space. The theorems of section 14 ensure that this cannot happen
for locally convex spaces (unless the topology is trivial); local con-
vexity is sufficient for the existence of continuous linear functionals.
It is also true that local convexity is, in a weakened sense, necessary
for the existence of such functionals. If .r is an arbitrary vector
topology for a linear space E, then the family of all convex cirded .r-
neighborhoods of 0 is a local base for a locally convex topology Oll,
which might be called the locally convex topology derived from .r.
Alternatively, the topology Oll can be described as the strongest
locally convex topology which is weaker than the topology .r. Each
linear functional which is Ol/-continuous is also .9""-continuous,
because Oll c .r. But conversely, if f is a linear functional which is
.r-continuous, then f is bounded on a convex cirded .r-neighborhood
ofO (namely,f- 1 [{t: lti < l}D, and consequentlyfis 01!-continuous.
Thus the topology .r and the topology Oll yield exactly the same dass
of continuous linear functionals.
There are several consequences of the remarks of the preceding
paragraph. A dass C of linear functionals on a linear space E is said
to distinguish points of E if and only if for each pair x and y of
distinct points of E it is true that there is a member f of the dass C
such thatf(x) # f(y). ·If .r is a vector topology for E and the dass
of .r-continuous linear functionals distinguishes points of E, then, in
view of the foregoing paragraph, the dass of linear functionals which
are continuous relative to the derived topology Oll has the same
property. In this case it is dear that Oll is a Hausdorff topology.
The theorem of this chapter will show that if Oll is a Hausdorff topol-
ogy, then there are enough .9""-continuous functionals to distinguish
points. Consequently the dass of .r-continuous linear functionals
110 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES

distinguishes points if and only if the class of o/1-continuous linear

functionals distinguishes points, and this is the case if and only if the
derived locally convex topology o/1 is Hausdorff. There are many
cases in which these facts can be used to extend a proposition which
is established for locally convex linear topological spaces to a result
which holds for any linear topological space for which there are
enough continuous linear functionals to distinguish points. A case
in point is the extreme point theorem of Section 15 (see the problems
which follow that section).
Section 15, the last section of this chapter, exhibits the first con-
sequences of the theorems on the existence of continuous linear
functionals. The principal theorem of the section asserts that,
under suitable conditions, a convex set has extreme points (a point is
an extreme point of a convex set A if it is not an interior point of any
line segment contained in A). This theorem has been used to
demoostrate the existence of irreducible representations of locally
compact groups and, more generally, the existence of such representa-
tions for certain Banach algebras. The theorem also occurs as an
essential element in various calculations of a purely Banach space
nature; in brief, it has taken its place with the Hahn-Banach theorem
and the Tychonoff theorem as an indispensable theorem of algebraic
analysis.

13 CONVEX SUBSETS OF LINEAR TOPOLOGICAL SPACES

The section is devoted to a number of technical propositions con-
cerning convex sets. In particular, we consider the closures and the
interiors of convex sets, and the convex extensions of sets which are
compact, or bounded, or totally bounded. Continuity of the restric-
tion of a linear functional to a convex set is characterized in several
equivalent ways.
The first theorem of the section concerns the construction of
convex sets from arbitrary sets by means of combined topological and
algebraic operations. A convex body is a convex se't with non-
void interior.
13.1 CoNvEx ExTENSIONS AND CoMBINATIONS OF CoNvEx SETS.
Let E be a linear topological space, let A and B be subsets of E, and let
a and b be scalars. Then:
(i) If A is convex, so are A- and Ai; if A is a convex body,
then tA- + (1- t)Ai c Ai for 0 ~ t < 1, Ai is the radial
kerne! of A, Ai- = A-, and A-i = Ai.
 SEc. 13 CoNvEx SuBSETS oF LINEAR ToPOLOGICAL SPACES 111
(ii) The smallest closed convex set containing B is the closure of the
convex extension <B) of B.
(iii) lf <A)- and <B>- are compact, then <A U B)- and <aA +
bB)- are compact.
(iv) If <A) is compact and <B> is closed, then <aA + bB) is closed;
if, further, Eis Hausdorffand <B) is bounded, then <A U B) is
closed.

PROOF (i) If A is convex, then by theorem 5.2(xiii) A- is convex.

If Ai is void, clearly it is convex. Now assume that Ai is non-void.
If it is shown that tA- + (1 - t )Ai c Ai for 0 ~ t < 1, the con-
vexity of Ai also follows. Because tA- + (1 - t)Ai is open, it is
sufficient to prove that tA- + (1 - t )Ai c A. Let x E Ai; then
(1 - t )(Ai - x) is an open neighborhood of 0. Therefore, tA- c
(tA)- c tA + (1 - t )(Ai - x) = tA + (1 - t )Ai - (1 - t )x c
A - (1 - t)x, and tA- + (1 - t)Ai c A follows.
Let A be a convex body. Since an open set is radial at each of its
points, Ai is contained in the radial kernel of A. Conversely, if z is
in the radial kernel and x is any point in Ai, there is y in A such that
z E (y:x]. But (y:x] c Ai by what has just been shown; hence,
zEAi.
The inclusions Ai- c A- and A-i => Ai are obvious. Let A be
a convex body. To see that Ai- => A-, it is sufficient to show that
any element x of A- "' Ai is in Ai-. If y is in Ai, clearly the set
(x:y] is non-void and contained in Ai by what was proved earlier, and
the closure of (x:y] contains the end point x. Hence x is in Ai-.
To establish A-i c Ai, again choose y in Ai. Any point x in A-i is
in. the radial kernel of A-, so that there is an element z in A- such
that x E [y :z). Since y E Ai and z E A-, the earlier result implies
[y:z) c Ai; therefore, xEAi.
(ii) This is an obvious corollary to part (i).
(iii) If <A)- and <B)- are compact, then their join (the union of
allline segments with one end point in <A)- and the other in <B)-)
is also compact because it is the image of [0:1] x <A)- x <B)-
under the continuous map: (t,x,y)-+ tx + (1 - t)y. But it is easy
to see that this join is closed and is precisely <A U B)-. The
proposition concerning aA + bB is proved in similar fashion.
(iv) It is true that <aA + bB) = <aA) + <bB), and if <A) is
compact and <B) is closed, then <aA) + <bB) is the sum of a com-
pact set and a closed set and is closed, in view of proposition 5.2(vii).
Assurne that <B) is bounded and that Eis a Hausdorff space. The
112 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES
set (A U B) is the image of [0:1] x (A) x (B) under the con-
tinuous mapping (a,x,y)--+ ax + (1 - a)y. If {za, a E C} is a net in
(A U B) which converges to the point z in E, and if Za = aaXa +
(1 - aa)Ya• then by the compactness of [0:1] and (A) there are sub-
nets {aß, ß E D} and {xß, ß E D} which converge to a in [0:1] and x in
(A), respectively. If a = 1, then (1 - aß)Yß--+ 0 since (B) is
bounded; hence, since E is a Hausdorff space, z = x E (A) c
(A U B). If a < 1, thenyß--+ (z- ax)/(1 - a), whichisamember
y of (B) since (B) is closed. Then z = ax + (1 - a)y E (A U B)
is closed.lll
The Minkowski functional of a convex body having 0 in its interior
has special properties, which are listed in the following proposition.
It should be mentioned that this proposition, the proof of which is
omitted, can be used to derive some of the results on convexity which
have been deduced here by other methods.
13.2 MINKOWSKI FuNCTIONAL OF A CoNVEX BonY Let A be a
convex body suchthat 0 E A 1, and Iet p be the Minkowskifunctional of A.
Then:
(i) p is uniformly continuous;
(ii) the interior of A is {x:p(x) < 1} and the closure of A is
{x:p(x) ~ 1}.
If A and B are bounded (totally bounded) subsets of a linear
topological space E and a is a scalar, then, as has already been shown,
each of aA, A + B, and the closed circled extension of Aare bounded
(totally bounded). It is not in general true that the convex extension
of a bounded or of a totally bounded set is of the same sort (see
problern C of this section). However, this proposition does hold for
locally convex spaces.
13.3 CoNvEx ExTENSIONS OF BouNnEn AND ToTALLY BouNnEn SETs
lf the linear topological space Eis locally convex, then the closed convex
circled extension of a bounded (totally bounded) set is again bounded
(totally bounded).
PROOF Let B be the closed convex circled extension of A, and Iet U
be a neighborhood of 0. Then there exists a closed convex circled
neighborhood V which is contained in U. Hence, if tA c V, then
tB c V c U. It follows that B is bounded if A is bounded. As-
sume now that A is totally bounded, and Iet C be the circled extension
of A; then C is totally bounded also. Corresponding to the given
neighborhood U of 0 there exists a convex neighborhood V of 0 such
 SEc. 13 CoNVEX SuBSETS OF LINEAR ToPOLOGICAL SPACES 113
that V + V c U. If R is a finite set in E such that C c V + R,
then <C) c <V+ R) = <V) + <R) = V + <R). Since R is
finite, <R) is compact, and hence there exists a finite set S in E such
that <R) c V + S. Then <C) c V + V + S c U + S; that is,
< C) is totally bounded. Hence, ( C)- is totally bounded, and the
closed convex circled extension of A is totally bounded.lll
The following theorem is an immediate consequence of the pre-
ceding result.

13.4 CoNVEX ExTENSION OF A COMPACT SET In a complete locally

convex space the closed circled convex extension of any totally bounded
set is compact.
PROOF Recall that a set A is compact if and only if A is totally
bounded and complete, and apply the preceding theorem.lll
A subfamily f!J of a family d is called a co-base (or dual base)
for d if and only if each member of d is contained in some member
of f!J. In this terminology, the family of closed bounded (totally
bounded) convex circled subsets of a locally convex linear topological
space is a co-base for the dass of bounded sets (totally bounded sets).
The remaining theorem of this section concerns linear functions
which are continuous on a convex set. The motivation for the
theorem is the fact that the space of all linear functions which are
bounded and continuous on a convex set A is complete relative to
uniform convergence on A, whereas the space of linear functions
which are continuous on E and bounded on A may fail to be complete
relative to the same topology. Recall that f IA is the function f
restricted to A, so that f is continuous on A if and only if JIA is
continuous.
13.5 FuNCTIONS CoNTINUOUS ON A CoNVEX SET Let A be a con·vex
subset of a linear topological space E and Iet f be a linear functional.
Then:
(i) if Eis a real linear space, then JIA is continuous if and only if
f- 1 [a] n Ais closed in Afor each scalar a;
(ii) ~f E is a real linear space and f[A] is symmetric, then JIA is
continuous if and only if f- 1 [0] n A is closed in A;
(iii) if E is a complex linear space and A is circled as well as convex,
thenf!A is continuous if rmd only iff- 1 [0] n Ais closed in A;
(iv) if g is an arbztrary linear functwn and A is circled and convex,
then giA is uniformly continuous if and only if giA is continuous
at 0.
114 CH. 4 CoNVEXITY IN LINEAR ToPoLoGICAL SPAcEs

PROOF (i) The necessity of the condition is obvious. For the

sufficiency, it is enough to prove that a set of the form {x:f(x);;:;
a} n A or {x:f(x) S; a} n A is closed in A. Put B = {x:f(x);;:;
a} n A and let y be a point in A "' B. Then f(y) > a and there is
a circled neighborhood U of 0 such that ( y + U) n f- 1 [ a] n A is
empty. Then (y + U) n Bis also empty since, if x E y + U, then
[x :y] c y + U. Hence Bis closed in A. Similarly a set of th'e form
{x:f(x) ~ a} n Ais closed in A.
(ii) Let a be a real number for which f- 1 [a] n A is not empty.
Then there is y in A such that f(y) = - a. Consider the map h on
A into A defined by h(x) = (x + y)/2. The map h is continuous
and the set f- 1 [ a] n A is the inverse image of f- 1 l0] n A under h.
Assertion (ii) now follows in view of (i).
(iv) This is a . consequence of the equation g(x) - g(y) =
2g((x - y)/2) and the fact that (x - y)/2 belongs to A whenever x
and y belang to A.
(iii) Assurne that f - 1 [0} n A is closed in A. Let r be the real
part of f, and let B be the convex symmetric subset {x:f(x) is real}
ofE. Thenf- 1 [0] n An B=r- 1 [0] n An B. Henceby(ii)r
is continuous on A n B. It follows that, for a given positive number
e, there is a circled neighborhood U of 0 suchthat Jr(x)! < e for all x
in A n B n U; hence, lf(x)l < e whenever XE A n U. Therefore,
f i~ continuous at 0 on A, and by (iv) f is continuous on A. The
converse is obvious.J!I

PROBLEMS
A MIDPOINT CONVEXITY
Let E be a linear topological space. A subset A of Eis called midpoint
convex iff 1 (x + y) E A wherever x E A and y E A.
(a) If A is a subset of E such that A + A c 2A-, then A- is convex;
if Ai + A' c 2A, then Ai is convex. (First show that A- and A' are mid-
point convex. For x andy in A- and 0 ;;:; t ;;:; 1, Iet cp(t) = tx + (1 - t)y,
and use 2A.)
(b) The closure A- and interior A' of a midpoint convex set A are con-
vex. If A is also circled, then A- and Ai are circled.

ß CONDENSATION COROLLARY (cf. 9.2)

If A is a convex set of the second category in a linear topological space E,
then A is of the second category at each point of A-. (Let x be a point
of A at which A is of the second category and Iet y be a point of A-. If V
is an open neighborhood of y, choose t ·so that 0 < t ;;:; 1 and tx + (1 -
t)y E V and Iet cp(z) = tz + (1 - t )y. Then cp is a homeomorphism of E
 SEC. 13 PROBLEMS 115
into itself, and there is a neighborhood U of x suchthat .P[U] c V. Then
.p[A n U] c A n V, .p[A n U] is of the second category m E and
therefore A n V is of the second category in E.)

C CONVEX EXTF.NSION 0" BOUNDED AND TOTALLY BOUNDED SETS

In a linear topological space which is not locally convex~ the convex
extension of a bounded or totally bounded set may fail to have the same
property.
In the space P 12 (6N), the set {x: llxlll/2 < 1} is bounded but its convex
extension is not bounded. Let xll = (1 ,0,0, · · · ), x 12 = (0,-!,0, · · · ),
x 22 = (0,0,-!,0, · · · ), x 13 = (0,0,0,!,0, · · ·) and in general Iet Xmn(1 ~
m ~ n) be the sequence with all terms zero except the (-!n(n - 1) + m)-th,
which is 1/n. Then A = {xmn: 1 ~ m ~ n, n = 1, 2, · · ·} is totally
bounded, but (A), which contains all the points Yn = (1/n) 2: {xmn=
1 ~ m ~ n}, is not totally bounded.

D TRANSLATES OF CONVEX SETS

Let B be a non-empty subset of a Hausdorff linear topological space E,
x 1 and x 2 be two points in E, and t 1 and t 2 be two non-negative real numbers.
+
(a) If B is convex, t 1 ~ t 2 , and x 1 - x 2 E (t 2 - t 1 )B, then x 1 t1B c
x2 + t 2 B.
(b) If B is bounded and sequentially closed, t 1 ~ t 2 , and x 1 + t 1 B c
x2 + t 2 B, then x 1 - x 2 E (t 2 - t 1 )B.
(c) If Bis bounded and contains more than one point, and if x 1 + t 1 B c
x 2 + t 2 B, then t 1 ~ t 2 •
(d) If Bis bounded and convex and contains more than one point, then
some translate of t 1 B is contained in t 2 B if and only if t 1 ~ t 2 •

E EXTENSION OF OPEN CONVEX SETS

Let E be a locally convex linear topological space and Iet F be a subspace
of E.
(a) Let A be a relatively open ccmvex non-void subset of F. Then for
each convex open set B in E such that B => A there is a convex open set
C in E such that B => C and C n F = A. lf A is circled, C may be
chosen to be circled. (By a translation argument it can be supposed that
0 E A. Choose a convex circled open neighborhood U of 0 in B, and take
C to be the convex extension of A U U.)
. (b) , Any convex subset A of F that is open relative to F is the inter-
secti(>n of F with some open convex set C in E; if A is circled, C may be
taken' to be circled.
(c) If Fis closed and Ais any convex set in F that is openrelative to F,
then for any x 0 in E not in A there is an open convex set C in E such that
C n F = A and x 0 is not in C. lf A is circled, C may be taken to be
circled.
(d) Let p be a pseudo-norm defined in F and continuous; then there
exists a pseudo-norm p- defined and continuous on E that is an extension
of p. If Fis closed and x 0 is any point of E not in F, p- can be chosen so
that p-(x0 ) ~ 1.
116 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACEs

F HYPERCOMPLETE SPACES
The set of non-void subsets of a linear topological space E can be made
into a uniform space in the following way. Let o/1 be a local base and for
each U E o/1 put
Vu = {(A,B): Ac B + U) and B c A + U}.
Then {V u: U E 0//} is the base of a uniformity, called the H ausdorff uni-
formity, because it is the uniformity of the Hausdorff metric (see 4D) when
E is a metrizable space.
With this uniformity, the space of subsets of E does not satisfy Haus-
dorff's separation axiom, because the uniformity fails to distinguish a set
from its closure. To avoid this superßuity, we res"trict attention to the set
C of non-void closed subsets of E, with the-Hausdorff uniformity.
A net {Ar: y E T} in Cis convergent to A E C if for each U E o/1 there is
some y(U) with Ar c A + U and A c Ay + U for all y ~ y(U); it is
Cauchy if for each U E Oll there is some y( U) with Aa c Aß + U for all
a,ß ~ y(U).
When Eis complete and metrizable, then (by 4D) ~' is complete. How-
ever, without the hypothesis of metrizability, 11' may fail to be complete,
even when E .is a locally convex Hausdorff space; numerous examples will
soon transpire.
A locally convex Hausdorff space E is called hypercomplete if the set Cf}
of convex circled non-void closed subsets of E is complete under the
Hausdorff uniformity. A hypercomplete space is complete. Since Cf} is
always a closed subset of C, a complete metrizable locally convex space is
hypercomplete.
To ensure that Cf} is complete, it is sufficient that every decreasing
Cauchy net should converge. (For if {Ar: y E T} is any Cauchy net, Iet
Cr be the closed convex extension of U {Aa: a ~ y}. Then {Cy: y E T}
is a decreasing Cauchy net; it therefore converges, to c = n{Cy: y E T}.
Now for each convex circled neighborhood U of 0, there is some y( U) with
Cy c C + U, C c Cy + U and Aa c Ap + U for all a, ß, y ~ y(U).
Hence Ar c Cy c C + U; also U {Aa: a ~ y} c Ar+ U and so Cy c
(Ar+ U)- c Ar+ 2U Thus C c Ar+ 3U and so {Ar: y E T} con-
verges to C. The device used here for dealing with the closure is often
useful: if Ais any set and o/1 a local base, then A- =n {A + U: U E o/1}.)
The image by an open continuous linear mapping of a hypercomplete
space is hypercomplete (cf. 11.3), and therefore so is the quotient by a
closed subspace. Hypercompleteness is also inherited by closed sub-
spaces. An example of a non-metrizahle hypercomplete space is given
in 18H.

G CLOSED GRAPH THEOREM Il (see 12E, 18J, 19ß)

A linear mapping T of a locally convex Hausdorff space E into a hyper-
complete space F (see the previous problem) is continuous, provided that
(i) the graph of Tin E x Fis closed
(ii) for each neighborhood V of 0 in F, the closure of T -l[V] ts a
neighborhood of 0 in E. (Cf. 11.1)
 SEC. 14 CoNTINUOUS LINEAR FUNCTIONALS 117
(Scheme for proof: Iet o/.1 and "Y be -local bases of convex circled neighbor-
hoods of 0 in E and F.
{a) The set {(T[U])-: U E o/.1} is Cauchy. For if V E "Y there is some
UE o/.1 with ZU c (T- 1 (V])-. lt follows that if U1 c U and U2 c U,
(T[U1 ])- c (T[U2 ])- + V.
(b) Since Fis hypercomplete, the set converges toB c n {T[U] + V:
U E o/.1, V E "Y}. For if V E "Y, there is some U v E o/.1 with B c (T [ U])- +
1-V for all U c Uv. Hence B c n {(T[U])- + 1-V: U E o/1} c
n {T[U] + V: u E o/.1}, and this holds for all V E "Y.
(c) Since the graph is closed, B = {0} by llH. Hence T[U] c V.
Condition (ii) is satisfied whenever Eis barrelled.) (Cf. 12E.)

14 CONTINUOUS LINEAR FUNCTIONALS

Here the fundamental theorems on existence and extension of con-
tinuous linear functionals are presented as simple consequences of the
separation and extension theorems of Section 3. The continuous
linear functionals on subspaces, quotient spaces, products, and direct
sums are described. The problems at the end of the section contain
other representation theorems for continuous linear functionals on
various linear topological spaces.
This section begins with theorems on the existence and extension
of continuous linear functionals. All of these theorems are simple
consequences of the separation and extension theorems of Section 3
and of proposition 5.4 on the continuity of linear functionals. The
theorems are collected partly for convenient reference, but mainly to
emphasize the fact that for locally convex spaces there always exists a
rich supply of continuous linear functionals. The section ends with
an explicit description of the continuous linear functionals on a prod-
uct or direct sum. A number of other results concerning the form
of continuous linear functionals for various spaces are outlined in
the problern set.
Recall that if p .is a pseudo-norm for a linear space E and I is a
linear functional, then I is continuous if and only if p*(f) =
sup {ll(x)l: p(x) ~ 1} is finite. The number p*(f) is the norm ofl
on E, and p* is the conjugate norm to p.
14.1 ExTENSION THEOREM Let E be a lz"near space, Fa subspace, and
I a linear lunctional on F. Then
(i) ij E is a pseudo-normed space and I is continuous on F, there is a
linear lunctional I- on E, an extension ofJ, such that the norm of
f- on E is equal to the norm of f on F;
(ii) if K is a convex circled set which is radial at 0, and ij f isalinear
lunctional on F suchthat IJ(x)l ~ 1 for x in K n F, then there
118 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES
isalinear functionalf- on E, anextension ofj, suchthat 11- (x)l ~
1 for x in K; and
(iii) Ij ff is a locally convex topology for E, and f is ff -continuous on
F, then there is a ff -continuous linear functional f - on E which
is an extension of j.
PROOF Part (i) is a direct cansequence af the Hahn-Banach thearem
3.4, part (ii) is the geametric farmulatian 3.6 af the same result, and
(iii) fallaws fram (ii) in view af the fact that there is a canvex circled
neighbarhaad K af 0 in E such that I/I is baunded an K n F,
because f is cantinuaus an F.lll
Recall that a linear functianal f an a linear space E separates the
sets A and B if f is nan-identically-zera and r(x) ;2; r(y) far x in A
and y in B, where r is the real part af j. Mareaver, f is cantinuaus if
and anly if r is. cantinuaus.

14.2 SEPARATION THEOREM Suppose that A and B are non-void

convex subsets of a linear topological space E and that the interior of A
is non-void. Then there is a continuous linear functional f on E sepa-
rating A and B if and only if Bisdisjoint from the interior of A.
PROOF The interiar af A is identical with the radial kernel by 13.1,
and the extensian thearem 3.8 applies. Any functianal separating A
and B is necessarily cantinuaus because its real part is baunded fram
abave an the interiar af A, which is a nan-vaid apen set.lll
There is a fact cancerning the faregaing thearem which is frequently
useful. A nan-zera reallinear functional f maps the radial kernel af a
canvex set A inta the radial kernel af f[ A]. It follaws that if x belangs
ta the interior af A and f separates A and B, then f(x) < inf {f(y):
y EB}.
Recall that a linear functianal f an a linear space E strangly sepa-
rates A and B if and anly if sup {r(x): x E A} < inf {r(y): y E B},
where r is the real part af j.

14.3 THEOREM ON STRONG SEPARATION Let A and B be non-void

disjoint convex subsets of a locally convex linear topological space E.
Then:
(i) there is a continuous linear functional f strongly separating A and
B if and only if 0 is not a member of the closure of B - A; and
(ii) if B is circled, then there is a continuous linear functional f such
that sup {lf(x)l: x E B} < inf {lf(y)l: y E A} if and only if 0 is
not a member of the closure of B - A.
 SEC. 14 CoNTINUOUS LINEAR FUNCTIONALS 119
PROOF This is clearly a variant of theorem 3.9 on strong separation.
If f strongly separates A and B, then f is bounded away from 0 on the
set B - A because sup {r(x): x E A} < inf {r(y): y E B} where r is
the real part of J. Hence, if f is continuous, then 0 does not belong
to (B - A)-. To show the converse, suppose that 0 rf= (B - A)-
and that E is locally convex. Then there is a convex circled open
neighborhood U of 0 such tliat 0 rf= B - A + U and the preceding
theorem shows that there is a continuous linear functionalf separating
B-A+ Uand{O}. Ifristherealpartofj,thensup{r(x):xEA}~
inf {r(y): y E B} + inf {r(z): z E U}. It follows that f 1s a con-
tinuous functional which strongly separates A and B. A simple
argument serves to establish part (ii).lll
It has already been shown that if A is compact and B closed, then
B - A is closed (theorem 5.2, part (vii)). The foregoing theorem
then yields the following as a corollary.
14.4 CoROLLARY Let A and B be non-void disjoint convex subsets of
a locally convex linear topological space, and suppose that A is compact
and B is closed. Then there is a continuous linear functional strongly
separating A and B, and iJ B is circled, there is a continuous linear
functional f such that sup {IJ(x)l: x E B} < inf {1/(y)l: y E A}.
In particular, a point which does not belong to a closed convex
subset of a locally convex space can always be strongly separated from
the set.
The preceding theorems show the existence of a rich supply of
continuous linear functionals for a locally convex space. In applica-
tions one frequently needs to know not only the existence, but some
sort of explicit description, of the continuous linear functionals. A
number of such descriptions (representation theorems) are given in
the problems at the end of the section: for the present we shall
describe the continuous linear functionals on subspaces, quotient
spaces, product spaces, and on direct sums.
The adjoint (conjugate, dual) of a linear topological space E is
the space E* of all continuous linear functionals on E. If several
topologies for E are being considered, then (E,ff)* will denote the
.r-continuous linear functionals. If F is a subspace of E, then the
set of all members f of E* which vanish identically on F is the anni-
hilator of F, or the space orthogonal to F, and is denoted by Fj_.
Each member f of Fj_ induces a continuous linear functional on the
quotient space EfF, in view of the induced map theorem 5.8, and it is
evident that each continuous linear functional g on EfF is induced by
120 CHo 4 CoNvExiTY IN LINEAR ToPOLOGICAL SPACES
a memb€r of p.1 (namely, by the composition of g with the quotient
map)o Consequently (ElF)* is isomorphic to F.L; this isomorphism
will be called, with remarkable lack of originality, a canonical
isomorphismo The space p.1 may also be described as the null space
of the map which restricts each member of E* to Fo Consequently
there is an induced isomorphism of E* IF .l into F* In general this
0

is not an isomorphism of E*IF.L onto F*-for example, F may be a

one-dimensional subspace of a linear topological space which has no
non-triviallinear functionalso However, if Eis locally convex, then,
by the extension theorem 1401 (iii), each continuous linear functional
on F is the restriction of some continuous linear functional on Eo
Thus:

14.5 ADJOINT OF A SuBSPACE AND A QuoTIENT SPACE Let E be a

linear topological space, Iet F be a subspace, and Iet p.1 be the annihilator
of F in E*o Then (ElF)* is canonically isomorphic to F.L, and, if E
is locally convex, E* IF .l is canonically isomorphic to F* 0

14.6 ADJOINT OF A PRODUCT A linear functional cp on a product

X {E1 : t E A} of linear topological spaces is continuous relative to the
product topology if and only if cp can be represented in the form cp(x) =
.L {g1(x1): t E A} where g is a member of the direct sum .L {E1*: t E A}
of the adjointso

PROOF Recall that a local base for the product topology is the family
of sets U of the following form: U is the set of all x in the product
such that, for each t in a fixed finite set B of indices, x 1 belongs to a
given neighborhood U 1 of zero in E 1 If f is a continuous linear
0

functional which is bounded on such a neighborhood U, and if x is a

point of the product such that x 1 is 0 for t in B, then every scalar
multiple of x belongs to U, and it follows that f(x) = 00 Con-
sequently, for each continuous linear functional f there is a finite set
B of indices such that, if x 1 = 0 for t in B, then f(x) = Oo Finally,
if 11 is the injection of E 1 into the product X {E1 : t E A}, and if y = x -
.L Ut(x1): t E B}, then y 1 is 0 fort in B; hence,f(y) = 0, and therefore
f(x) = .L {f(11(x 1)): t E B}o But, for each t, f o 11 is a continuous
linear functional on E 1 : that is, a member of E1*o I t follows that each
continuous linear functional f an the product X {E 1: t E A} can be
represented in the form:f(x) = .L {g 1(x1): t E A} whereg is a member
of the direct sum of the spaces E 1*0 It is evident that, conversely,
any functional ofthisform is an element of the adjoint of the producto III
 SEC. 14 CoNTINUOUS LINEAR FUNCTIONALS 121
There is a natural topologization of the direct sum 2 {E1 : t E A} of
locally convex spaces E1• Recall that for each t in A there is an
algebraic isomorphism, called an injection, of E1 into the direct sum.
The injection 11 is defined by letting 11(z), where z E E1, be the
member of the direct sum which is 0 for s =I= t and z for s = t. The
direct sum may be given· the topology induced by the mappings 11
(see Section 4); explicitly, a set is open relative to this topology if and
only if its inverse under each / 1 is open. Unfortunately, this topology
is usually not a vector topology for the direct sum (see problern SB),
so a variant of the induced topology is desirable. · Let ~ be the
family of all convex circled subsets U of the direct sum such that
11-l[U] is a neighborhood of 0 for each t. By the convexity, each
member of ~ is radial at 0. Therefore, in view of theorem 6.5,
the family ~ is a local base of a unique locally convex topology for the
direct sum; this topology is called the direct sum topology. If,
for the moment, we overlook the distinction between E1 and its iso-
morphic image 11[E1J, then a convex circled subset U ofthe direct sum
is a neighborhood of 0 if and only if its intersection with each E1 is 3
neighborhood of 0. For each t, let U1 be a neighborhood of 0 in E1 ;
then the convex extension <U {U1 : t E A}) is a neighborhood of 0 in
the direct sum, and each neighborhood of 0 relative to the direct sum
topology contains a neighborhood of this form.
Before discussing the elementary properties of the direct sum
topology we digress to observe that the construction of this topology
is a special case of a more general method. If F is a family of linear
functions, each member f of F being on a linear topological space E1
to a fixed linear space H, then the inductive topology for H (the F-
inductive topology) has as a local base the family of all convex
circled subsets U of H, which areradial at 0, such thatj- 1 [U] is a
neighborhood of 0 in Er for each f in F. It is evident that the F-
inductive topology is the strongest locally convex topology which
makes each member f of F continuous, and that a linear function g
on H to a locally convex space is continuous relative to the inductive
topology if and only if g o f is continuous for each f in F. An im-
portant special case is that in which each Er is a subspace of H and
each f is the identity map. Then the inductive topology can be
described as the strongest locally convex topology whose relativization
to each Er is weaker than the topology of Er.
The elementary properties of the direct sum topology are easy to
establish; in particular, it is almost self-evident that, dually to 14.6,
the adjoint of a direct sum is the product of the adjoints. Part (iv)
122 CH. 4 CONVEXITY IN LINEAR TüPOLOGICAL SPACES
can also be proved using the completeness criterion of 17.7 (see
problern 17F).

14.7 ELEMENTARY PROPERTIES oF THE DIRECT SuM ToPOLOGY

(i) The direct sum topology is the strongest locally convex topology
with the property that each injection is continuous.
(ii) Each injection is a topological isomorphism of a factor onto a
subspace of the direct sum.
(iii) A linear function on a direct sum to a locally convex space is
continuous if and only if its composition with each injection
is continuous.
(iv) The direct sum of complete locally convex spaces is complete
relative to the direct sum topology.
(v) A linear functional ~ on a direct sum .L {E1 : t E A} of locally
convex spaces is continuous relative to the direct sum topology if
and only if there is a member f of X {E1*: t E A} such that
~(x) = .L {ft(x1): t E A} for all x in the direct sum.

PROOF We prove only part (iv) of the proposition. In this proof we

write x( t) for x1• If { Xa: a E D} is a Cauchy net in the direct sum
E = .L {E1 : t E A}, then, because projection into each coordinate
space E1 is continuous, {xa(t): a E D} is a Cauchy net in E1 and hence
converges to one or more points of E 1• Select x(t) to be one of the
points to which {xa(t)} converges, taking care to select 0 if possible.
Foreach t suchthat x(t) -=1- 0, let U 1 be a closed convex neighborhood
of 0 in E 1 such that x(t) rf- U1, and, for t for which x(t) = 0, let
U 1 = E 1• Since {xa} is Cauchy, there is ß in D such that y s ß
implies xr - xß E <U {U1 : t E A}). Then, for each t, xr(t) -
xß(t) E Ut. Since ut is closed, x(t) - xß(t) E ut. Hence, if x(t) -=1- 0
then xß(t) -=f. 0. It follows that x belongs to the direct sum and the
net {xa} converges to x relative to the topology 9 of coordinate-wise
convergence. In order to prove that {xa} converges to x relative to
the direct sum topology, it is sufficient to show that the net Xa - x is
eventually in the closure of each neighborhood (of 0) of the form
U = <U {U1 : t E A}), where U1 is a convex neighborhood of 0 in E1•
Because the net {xa - xß} is eventually in U, the proof is complete if
it is shown that the closure U - relative to the direct sum topology is
also 9-closed.
The rest of the proof is devoted to showing that U - is 9-closed.
Let y be a point of E belanging to the 9-closure of U. Let B =
 SEC. 14 PROBLEMS 123

{t: y(t) :/= 0}; theny belongs to the 9-closure of u n CL {Et: t E B}).
(This is a consequence of the fact that, if z E U, then a point obtained
from z by replacing one or more coordinates of z by 0 is still in U.)
The set B is finite, and the direct sum topology relativized to L {Et:
t E B} is identical to the relativized 9; hence y E U-. Since the
direct sum topology is stronger than 9, U - is 9-closed.lll
Finally, the following description of the bounded and the totally
bounded subsets of a product and of a direct sum is given for ref-
erence; its proof is left to the reader.
14.8 BouNDED AND ToTALLY BouNDED SuBSETS OF PRonucTs AND
SUMS
(i) The family of all products of totally bounded (or bounded) subsets
of the factors is a co-base for the family of totally bounded (or
bounded) subsets of the product of linear topological spaces.
Equivalently, a subset of a product is totally bounded (or
bounded) if and only if its projection into each coordinate space is
totally bounded (or bounded).
(ii) The family of all convex extensions of finite unions of images
under injection of totally bounded (or bounded) subsets of the
factors is a co-base for the family of totally bounded (or bounded)
subsets of the direct sum of locally convex Hausdorff spaces.

PROBLEMS
A EXERCISES
(1) There is a non-trivial continuous linear functional on a linear topo-
logical space E if and only if E properly contains a convex body.
(2) If for each x in a linear topological space Ethereis a convex neighbor-
hood of zero not containing x, then there are enough continuous linear
functionals on E to distinguish points.
(3) A linear manifold in a linear topological space is contained in a
closed hyperplane if and only if its complement contains a convex body.
(4) If A is a closed convex subset of a locally convex space E and Bis a
subset of E, then B c A if and only if f[B] c f[A] for every continuous
linear functional f on E.
(5) A closed linear manifold in a locally convex space is the intersection
of all the closed hyperplanes containing it.

ß FURTHER SEPARATION THEOREMS

Let E be a linear topological space and A and B non-void subsets of E.
(a) If A and B are strongly separated by a continuous linear functional
on the real restriction of E, there is a continuous linear functional on E
such that the distance between f [A] and f [B] is positive.
124 CH. 4 CONVEXITY IN LINEAR TOPOLOGICAL SPACES

(b) If B - A is midpoint convex (see 13A) and somewhere dense, the

following are equivalent:
(i) A and B are strongly separated by a continuous linear functional on
the real restriction of E;
(ii) for some continuous linear functional I on E the distance between
I[A] andi[B] is positive;
(iii) 0 is not in the closure of B - A.
(c) Suppose that A and Bare disjoint and convex and that Eis locally
convex. Then there is a continuous linear functional I such that the
distance betweeni[A] andi[B] is positive if and only if 0 is not a member
of the closure of B - A.

C A FIXED POINT THEOREM

Lemma Let A be a compact convex subset of a linear topological space
E. lf x 0 is a point such that for each y in A there is a continuous linear
functional strongly separating x 0 and y, then there is a continuous linear
functional on the real restriction of E strongly separating x 0 and A.
Theorem Let A be a non-void compact convex subset of a linear topolog-
ical space E and suppose that for each non-zero member x of A - A there is
a continuous linear functionall on E withl(x) =I 0. Let T be a continuous
mapping of A into itself such that T(L, {a,x1 : 1 ~ i ~ n}) = L, (a,T(x,):
1 ~ i ~ n} wherever x, E A and a, ~ 0 for 1 ~ i ~ n and L, {a1: 1 ~ i ~ n}
= 1. Then there is a point x 0 in A such that T(x 0 ) = x 0 • (Suppose
not, and put S (x) = x - T(x). Then S [A] is convex and compact and
0 rf: S [ A]. By the Iemma there is a continuous linear functional I on the
real restriction of E positive (say) on S[A]. Consider the point at whichl
attains its minimum.)

0 STRONGEST LOCALLY CONVEX TOPOLOGY 111 (see 61, 12D, 20G)

Let E be a linear space of dimension cx with its strongest locally convex
topology. Then E is the direct sum of cx copies of the scalar field and is
complete (see 14.7). Every linear manifold in Eis closed (see 61) and the
relative topology for any linear subspace F of E is the strongest locally
convex topology for F.

E STRONGEST VECTOR TOPOLOGY Il (see SE)

Let E be a vector space with its strongest vector topology. Regarding E
as a direct sum of copies of the scalar field, relative to some fixed Harne!
base {e1 : t E A}, Iet x(t) denote the t-th coordinate of x, and / 1 the injection
of !he t-th coordinate space into E.
(a) For convenience, call x and y disjoint iff they take their non-zero
values on disjoint subsets of A. Then E has a local base 0l! of neighbor-
hoods U for which x and y disjoint and x + y E U implies x E U and
y E U. (The topology of E can be defined by the family of all functions q
with the properties (i) through (iv) of 6C. For such q, Iet
q'(x) = L, {q(l1(x(t))): t E A}.
 SEC. 14 PROBLEMS 125
Then q' satisfies the conditions (i) through (iv) of 6C and q'(x + y) =
q'(x) + q'(y) whenever x and y are disjoint.)
(b) The space E is complete.
(Let {xy: y E T} be a Cauchy net in E. Then {xy} converges in the
strongest locally convex topology on E, to x 0 say. Put Yr = Xy - x 0 , Iet
U E Olt and choose a neighborhood V of the origin with V + V c: U.
There is an cx with Yn - Yr E V wherever ß,y ;;; cx. Let B be the finite set
of those t with Ya(t) =? 0. There is aß;;; cx with .L {I1(Yn(t)): t E B} E V;
if Zp = L {I1(yn(t)): t E A ,..., B}, then Ya -Zn EU and so Ya EU, and
Xy ---J> x 0 .)

F li (see SG)
n
ALGEBRAIC CLOSURE OF CONVEX SETS
If A is a convex circled subset of a linear space E, then {r A: r > 1} is
the closure of A in the strongest locally convex topology on E. (In the
linear subspace generated by A, with its strongest locally convex topology,
Ais a neighborhood of zero; use SG.)

G LINEARLY CLOSED CONVEX SETS li (see SH)

A convex circled subset of a linear space with its strongest locally convex
topology is closed if its intersection with every straight line is closed.
(Use the previous problem.)
If the linear space has a countable base, convex subsets have the same
property (see also 18H).

H A FUNDAMENTAL THEOREM OF GAME THEORY

Let A and B be non-void compact convex subsets of the linear topo-
logical spaces E and F, respectively, and Iet 4> be a real-valued function on
A x B with the following properties:
(i) for each y E B the mapping x ___,. cf>(x,y) is continuous on A,
(ii) if a; ;;; 0 for 1 ~ i ~ n and .L {a;: 1 ~ i ~ n} = 1, then
cf>(.L {a;x;: I ~ i ~ n},y) = .L {a1cp(x;,y): 1 ~ i ~ n}
for all x 1 E A and
cf>(x, .L {a;y;: 1 ~ i ~ n}) = .L {a;cp(x,y;): 1 ~ i ~ n}
for all Yt E B.
Then sup inf cf>(x,y) = inf sup cf>(x,y). (In this equality, ~ is
xeA yeB . yeB xeA
immediate; to prove the reverse inequality denote the right side by c.
Then c may be assumed finite and it is enough to prove that, if Ay =
{x: XE A, cf>(x,y) ;;; c}, thenn{Ay: y E B} is non-void. Since each Ay is
closed and A is compact, it is sufficient to show that any finite intersection
of sets Ay is non-void. Suppose n {AYI: 1 ~ i ~ n} = 4> and map A
into Rn by f where f(x) = (cp(x,y 1 ) - c, cp(x,y 2 ) - c, · · ·, cp(x,yn) - c). By
(i) and (ii) f[A] is a compact convex subset of Rn and f[A] n P = cp,
where Pis the positive cone in Rn, that is, the set of points all of whose CO-
ordinates are non-negative. There is a linear functional g on Rn strongly
separatingf[A] and P. All the coordinatesg; ofg have the same sign; we
126 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES

may suppose them non-negative and that L: {g1: 1 ~ i ~ n} = 1. Then if

Yo = _L{gty;: 1 ~ i ~ n},
sup cf>(x,y0 ) = c + sup L: {g;(cf>(x,y1) - c): 1 ~ i ~ n} < c,
xeA xeA

which contradicts the definition of c.)

I COMPLEX MEASURES
The adjoints of spaces of continuous functions can be identified with
spaces of measures. When the functions are complex valued, so are the
measures. Some of the properties of complex valued measures which are
not immediate extensions of those of real valued measures are collected in
this problem; they are sometimes omitted from texts on measure theory.
If /L 1 and /L 2 are two signed measures defined on the same a-ring !/' of sub-
sets of a set X, the set function /L = /L 1 + il-' 2 will be called a complex
measure on X.
(a) For each A in !/', Iet I~LI(A) denote the supremum, over all finite
families {Br: 1 ~ r ~ n} of disjoint subsets of A, of the sums 2: {I/L(Br)l:
1 ~ r ~ n}. Then I~LI is a measure on X, called the total variation of /L·
For a signed measure ~-'• I~LI = /L + + /L-, where /L + and /L- are the upper
and lower variations of ~-'• at least one of which is finite, and /L = /L + - /L-.
The set of all finite complex measures on X, defined on the a-ring !/' of
subsets of Xis a Banachspace with the norm II~LII = sup {I~LI(A): A E !/'}.
(b) A complex valued function I = 11 + if2 on X is integrable with
respect to the complex measure /L = /L 1 t il-' 2 iff both 11 and 12 are in-
tegrable with respect to the upper and lower variations of both /L 1 and /L 2 ;
when it is, f ld/L is defined to be the obvious linear combination of eight
terms. If I is measurable with respect to !/',I is integrable with respect
to /L iff II I is integrable with respect to I~LI, and then I ffd /L I ~ f I I Id I~LI·
(c) If Xis a topological space, the set belonging to the a-ring generated
by the compact subsets of X are called Bore/sets. A (positive, signed or
complex) measure /L on X defined on the Bore! sets is called a Bore/ measure
iff I~LI(A) is finite for every compact subset A of X. The Bore! measure
/L = /L 1 + i/L 2 is said to be regular iff, for each Bore! set A,
I~LI(A) = sup {I/LI (B): B c A and B compact}
= inf {I~LI (C): C :::> A and C open Bore/},
or equivalently, iff the upper and lower variations of /L1 and /L 2 are all
regular. For a finite regular Bore! measure, the total variation, II~LII as
defined in (a) is the supremum, over all finite families {Br: 1 ~ r ~ n} of
disjoint compact subsets of X, of the sums 2: {I!L(Br)l: 1 ~ r ~ n}. Every
continuous function vanishing at infinity on X is Bore! measurable.

J SPACES OF CONTINUOUS FUNCTIONS Il (see 81)

Let X be a locally compact Hausdorff space and K (X) the space of
continuous scalar valued functions of compact support on X. The
starting point of this representation theory is the theorem that, when the
scalars .are real, every positive linear functional cp on K(X) can be ex-
pressed in the form cp(f) = f l(x)d/L(x), where /L is a regular Bore! measure
 SEC. 14 PROBLEMS 127
on X, uniquely determined by cp. (See, for example, Halmas [4] pp. 247-
248.)
(a) Assurne that the scalars are real; then each positive linear functional
4> on K(X) is continuous with respect to the inductive Iimit topology :!7
(see 81). Conversely, if 4> E (K{X), :!7)*, then 4> can be written as 4> = 4> +
- 4>-, where both 4> + and 4>- are positive linear functionals on K(X).
Therefore there are two regular Bore! measures fL + and fL- on X such that
cfo(f) = ffdfL + - J.fdfL- for all f in K(X). (Warning: if neither fL + nor fL- is
finite fL + - fL- is not a signed measure.) If the scalars are complex, then
there is a similar representation of a continuous linear functional on K(X)
using four regular Bore! measures on X.
(Suppose first that the scalars are real and Iet 4> E (K(X))*. For each
f ~ 0, put cfo+(J) = sup {cfo(g): 0 ~ g ~ f}; then ifjE K(X), cfo+(J) < oo,
by the continuity of cp. Both 4> + and 4> + - 4> can be extended to give
positive linear functionals on K (X). For complex scalars, consider
the real and imaginary parts of 4> acting on the real subspace of K(X)
consisting of the real valued functions.)
(b) Riesz's Theorem The adjoint of the Banach space C 0 (X), with the
uniform norm, is the space of all finite regular Bore! signed or complex
measures on X, with the total variation as norm. (Since the inductive
Iimit topology on K(X) is stronger than the uniform topology, to each
4> E (C0 (X))* corresponds a unique regular Bore! measure fL with cfo(f) =
f fdfL for each f E K (X). If {Er: 1 ~ r ~ n} are disjoint compact subsets
of X and e > 0 is given, there is an open relatively compact set C con-
taining the sets Er and suchthat fL(C ~ U {E,: 1 ~ r ~ n}) < e, since fL
is regular. Then there is a function f with support contained in C such
that I f(x) I ~ 1 for all x andf(x)fL(Er) = lfL(Er) I for all x in Er(1 ~ r ~ n).
Since 2 {lfL(Er) I: 1 ~ r ~ n} ~ lf fdfLI + e, fL is finite and 11~-LII ~ 114>11·
The formula q,(f) = ffdfL now extends to C0 (X) and the inequality of
14I(b) gives 114>11 ~ 11~-LII·)
(c) The support (or carrier) of a regular Bore! measure fL is the smallest
closed set A such that fL(E) = 0 for all Bore! sets E disjoint from A.
(Such a set exists; it is the intersection of all sets with the above property.)
The adjoint of the space C(X) of all continuous functions on X with the
topology of uniform convergence on compact subsets of X is the space of
regular Bore! measures with compact support on X. (lf 4> E (C(X))*,
there is a compact set A and a constant k suchthat lc/>(f)l ~ k sup {lf(x)l:
x E A}. The restrictions to A of the functions of C (X) form the space
C(A) and so cp defines a member cfoA of (C(A))*. By (b), there is a regular
Bore! measure fL on A, which can be extended to X by putting fL(E) =
fL(E n A), with cfo(f) = f fdp,.)

K SPACE OF CONVERGENT SEQUENCES

Let X be any set. When X is assigned the discrete topology, the
corresponding space C0 (X) (see the previous problem) is usually denoted
by c0 (X). In particular, when Xis the set of positive integers we write
simply c0 ; thus c0 is the space of all scalar-valued sequences convergent to
zero. Let Y be the one-point compactification of X. Then the space C ( Y)
is isomorphic to the space of all functions on X convergent at infinity, and
128 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES

c0 (X) is then a closed hyperplane in C(Y) under the uniform norm. If X

is the set of positive integers, C ( Y) is denoted simply by c; it is the space
of all convergent sequences.
If p. is a regular Borel measure on the discrete space X, there is a function
g on X such that p.(A) = L: {g(x): x E A} for each subset A of X. Thus
the previous problern shows that the adjoint of c0 (X) is Jl(X) and the
adjoint of c( Y) is Jl( Y ), the general form for continuous linear functionals
being f ~ L: {f(x)g(x): x E A}. Naturally, this result is more easily
obtained directly. In particular the continuous linear functionals on c are
of the form
{xn: n= 1, 2, · · ·}~ L;{xnyn: n = 1, 2, · · ·} + (lim Xn) · Yo
n-+ oo
with {yn: n = 0, 1, 2, ···}in Jl.

L HILBERT SPACES II (see 7H et seq.)

Every Hilbert space H is self-adjoint in the following sense. To each
continuous linear functional f on H corresponds a unique element y such
that f(x) = (x,y) for all x in H, and the mapping f ~ y is a conju-
gate isomorphism of H* onto H. (By a conjugate isomorphism T is
meant a one-to-one mapping satisfying T(ax + by) = äT(x) + bT(y) and
I T(x)ll = llxll· Given f =F 0, Iet N = {x:f(x) = 0} and choose z E NL
with z =F 0. Then y is a scalar multiple of z.)

M SPACES OF INTEGRABLE FUNCTIONS IV (see 6K, 6N, 7M)

Let (X, .9, p.) be a measure space. Foreach p in the range 0 < p < oo,
U(X,p.) is the space of measurable functions f on X such that I f IP is in-
tegrable, with the topology determined by the local base of sets {f: 1/IIP <
1/n} for n = 1, 2, · · ·, where 11/IIP = (f lf1Pdp.) 11 P. In U(X,p.), 11/IIP = 0
if and only if f(x) = 0 almost everywhere. The subset of .9 consisting of
measurable sets of finite measure is denoted below by .90 •
(a) If 1 < p < oo, any continuous linear functional cf> on LP(X,p.) has
the form cf>(f) = f fgdp. with g E Lq(X,p.), where q is the index conjugate to
p, so that q = pf(p - 1). Conversely any linear functional of the above
form is continuous, with l c/>11 = llgllq; two functions of U(X,p.) define the
same linear functional if and only if they are equal almost everywhere.
Thus the conjugate of LP(X,p.) is isometrically isomorphic to the quotient
of Lq(X,p.) modulo, the subspace of functions equal to zero almost every-
where. It is usually more convenient to regard the spaces LP(X,p.) as
consisting of (equivalence classes of) functions specified only up to a set of
measure zero; this also makes the pseudo-norm f ~ I f I P a true norm.
With this slight gloss, the adjoint of the space U(X,p.) becomes the space
Lq(X,p.) ·for 1 < p < oo.
(Hölder's inequality shows that any c/> of the above form is continuous
and that l c/>11 ~ I gllq· To show that any c/> E (U)* has the required form,
first assume that XE .90• Then, if v(A) = c/>(xA), where XA is the charac-
teristic function of A, v is a signed or complex measure on .9. By the
Radon-Nikodym theorem there is a function gELl with v(A) = JA gdp.
and therefore, for all f EU, c/>(f) = f fdv = f fgdp.. If now {gn} is a
 SEC. 14 PROBLEMS 129
sequence of bounded non-negative J.L-measurable functions increasing to lgl,
llgnllqq ~ f gnq-llgldJ.L = rfo(gnq-l sgng) ~ llrfollllgnllqqfp,
and so
llgllq = sup {llgnllq: n = 1, 2, · · ·} ~ llrfoll·
In the general case wh~n (X,Y,J.L) is not totally finite, there is, for each
A E .70 , a function gA E Lq vanishing off A with rfo(fxA) = fgAdJ.L. If k =
sup {II gAIIq: A E .7 0 }, k ~ llrfoll and, for any increasing sequence {An} of sets
of .70 for which llgAJq__,..k, {gA.} converges in Lq. The Iimit g is in-
dependent of the particular sequence chosen and is the required function.)
(b) The case p = 1 requires more delicate treatment. A scalar valued
function f on X is called locally measurable if f ·XA is measurable for each
A E !7 0 • A property is said to hold local(v almost everywhere iff the set on
which it fails intersects each set of !7 0 in a set of measure zero. The func-
tions on X which are locally measurable and bounded locally almost
everywhere form a linear space, denoted by L 00 (X,J.L). For such a function
f, if
II/ I oo = inf {k: lf(x)l ~ k locally almost everywhere},
then lf(x)l ~ 11/lloo locally almost everywhere. The mapping/__,..11/lloo
is a pseudo-norm on L 00 (X,J.L) with respect to which L 00 (X,J.L) is complete.
When X is totally finite or totally a-finite, the word "locally" may be
omitted without changing the meaning above.
The measure space (X,Y,J.L) is called localizable iff to each family {gA:
A E !7 0 } of scalar valued functions, such that gA is defined and measurable
on A and gA(x) = g 8 (x) for almost all x in A n B, corresponds a locally
measurable function g on X coinciding with gA on A except for a set of
measure zero for each A in Y 0 • Every totally finite, or totally a-finite,
space is localizable.
(c) If (X,Y,J.L) is localizable" any continuous linear functional rfo on
U(X,J.L) has the form rfo(f) = f fgdJ.L with g E L 00 (X,J.L), and any linear
functional of the above form is continuous, with llrfoll = llgll oo· Two
functions of L 00 ( X,J.L) define the same linear functional if and only if they
are equal locally almost everywhere. Thus, if we agree to identify func-
tions of L "'(X,J.L) which are equallocally almost everywhere, the adjoint of
U(X,J.L) is L 00 (X,J.L), provided that (X,Y,J.L) is localizable.
( A linear functional of the above form is continuous and I rfo I ~ II g II oo.
If rfo E (U)* and XE .70 , a proof similar tothat in (a) shows that there is a
function g E U with rfo(f) = f fgdJ.L for all f in U. If B is the set on which
lg(x)l ~ k, kJ.L(B) ~ fa lgldJ.L = f (xa sgn g)gdJ.L = rfo(xa sgng) ~ llrfoiiJ.L(B).
Hence J.L(B) = 0 for k > llrfoll, so that gELoo and llglloo ~ llrfoll· In the
general case, for each A E !7 0 there is a function g A defined and measurable
on A with lgA{x)l ~ 114>11 almost everywhere and rfo(fxA) = fAfgAdJ.L.
Since Xis localizable there is a function g E L 00 with g(x) = gA(x) almost
everywhere in A and this function g is the one required.)
(d) Conversely, if the adjoint of U(X,J.L) is L 00 (X,J.L), the space (X,Y,J.L)
is localizable.
(First consider a uniformly bounded family {gA: A E .7 0 }. If Y 1 is the
subset of Y consisting of sets of a-finite measure, there is, for each B in
130 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES

Y\, a g8 extending gA for each A E !/ 0 with Ac B. Now each jE U

vanishes off some B E !/ 1 ; put ,P(f) = f fg 8 d11-. There is a g E L"' with
cfo(f) = f jgd11- and g is the required locally measurable function. The
case of a general family {gA} reduces to this special one.)
(e) The adjoint of /P(X) is lq(X), where, for 1 < p < oo, q = pf(p- 1)
and, for p = I, q = oo. Here the space Xis always localizable. But if X
is uncountable, !/ can be taken to be the set of all finite or countable sub-
sets of X and then the functions of L"' are not all measurable, but only
locally measurable. An example of a non-localizable space is given by
Halmos ([4] §31, (9)).
(f) If 0 < p < 1, the adjoint of fP(X) is /"'(X). (Use 6N(d).) On the
other hand, if X is an interval of the real line and 11- is Lebesgue measure,
the adjoint of U(X,P-) for 0 < p < 1 consists of the zero functional only.
(6N(c)).

15 EXTREME POINTS
The principal result of this section is the powerful theorem of Krein
and Milman: each convex compact subset of a locally convex linear
topological Hausdorff space has extreme points, and is in fact the
closed convex extension of the set of its extreme points.
A point x of a convex set A is an extreme point of A if and only if
x is not an interior point of any line segment whose endpoints belang
to A. Thus, in the Euclidean plane, the "corner" points of the set
{(x,y): max {lxl, IYl} ;;;; 1} are extreme points, and every point of the
circumference of {(x,y): x 2 + y 2 ;;;; 1} is an extreme point of this set.
An open convex set clearly has no extreme points; however, the
existence of extreme points of a bounded closed convex set is a likely
sounding conjecture. Unfortunately it is false even for Banach
spaces (see problern A of this section). However, if a convex set is
compact, and the containing linear topological space is locally convex
and Hausdorff, then extreme points da exist. This existence theorem,
which has far reaching consequences, is the principal result of this
section.
A set A is a support of a convex set B in a linear space if and only
if A satisfies the following conditions:
(i) A is a non-void convex subset of B;
(ii) if an interior point of a segment in B belongs to A, then the
segment is a subset of A; in other words, if x E B, y E B, and
tx + (1 - t)y E A for some t such that 0 < t < 1, then
x E A and y E A.
The supports enjoy the following properties. If A is a support of
a convex setBandBis a support of a convex set C, then A is a support
of C. If, for each t in a set S, B 1 is a support of a convex set C, then
 SEc. 15 ExTREME PoiNTS 131
the intersection n {Bt: t ES} is either void or a support of c also.
Finally, suppose f is a linear transformation of a linear space E into F.
If Bis a support ofj[C], where Cis a convex set in E, thenf- 1 [B] n C
is a support of C. A point x is an extreme point of a convex set A in
a linear space E if and only if {x} is a support of A.
The following theorem is the fundamental result of the section.

15.1 EXISTENCE OF EXTREME POINTS (KREIN-MILMAN THEOREM)

Let A be a convex compact subset of a locally convex linear topological
H ausdorff space E. Then each closed support of A contains an extreme
point of A, and A is the closed convex extension of the set of all its
extreme points.

PROOF Let R be any closed (hence compact) support of A; then by

the maximal principle there exists a family d of closed supports of A
which is maximal with respect to the following properties: (i) R E d;
and (ii) .!21 has the finite intersection property. Let S = n {B:
BE .!21}. Then S is a non-void closed support of A since A is com-
pact. Furthermore, S is a minimal closed support of A in the sense
that it contains as a proper subset no closed support of A. Hence
each closed support of A contains a minimal closed support of A.
Let S be any minimal closed support of A, and assume that a line
segment [x:y], where x i= y, is contained in S. Then there exists a
continuous linear functional f on the real restriction of E such that
f(x) "1- f(y). Since S is compact, f is bounded on S, and C =
{z: z ES, f(z) = sup {f(y): y ES}} is a non-void closed proper sub-
set of S. Then C is a support of S because it is the inverse under f
of a support of f[S], and hence Cis a support of A. But S was
presumed to be minimal, and this is a contradiction. Therefore, S
consists of only one point, which is an extreme point of A.
The foregoing paragraphs show that each closed support of A con-
tains an extreme point of A. Since A itself is a closed support of A,
A has at least one extreme point.
Let D be the closed convex extension of the set of all extreme points
of A. Then clearly D c A. Assurne that A ~ D is non-void,
and take x 0 in A ~ D. Then, by 14.4, there exists a continuous
linear functional g on the real restriction of E such that g(x 0 ) >
sup {g(y): y E D}. Let B = {x: x E A, g(x) = sup {g(z): z E A}}; as
before, Bis a closed support of A, and by the choice of g, B c A ~ D.
Then, as shown in the first paragraph of the proof, B contains an
extreme point of A contrary to the definition of D.lll
132 CH. 4 CONVEXITY IN LINEAR TOPOLOGICAL SPACES
The following theorem gives information on the set of extreme
points of certain convex compact sets. In general, little is known of
the structure of the set of extreme points.
15.2 THEOREM Let A be a compact subsei of a locally convex Haus-
dorf! space E, and suppose that the closed convex extension (A)- of A
is compact. Then all of the extreme points of (A)- belong to A.
PROOF Let U be a convex circled closed neighborhood of 0 in E.
Then, since A is compact, there exists a finite nurober of points
x 1 , · · ·, Xn in A such that U {(U + xi): i = 1, · · ·, n} :::::> A. Let
Ai = A n (U + x 1); then (A)- is contained in the convex extension
of U {(Ai)-: i = 1, · · ·, n}, since the latter is closed by theorem 13.1.
Then each point y of ( A)- can be expressed as y = L {a1y 1 : i =
1, · · ·, n}, where the at's are non-negative real numbers such that
.L {a1 : i = 1, · · ·, n} = 1, and Yi E (Ai)- c U + xi. If y is an ex-
treme point of (A)-, then for some i, y = Yi· Therefore, y E U +
xi c A + U; since U is an arbitrary convex circled closed neighbor-
hood of 0 and A is closed, it follows that y E A.JJI

PROBLEMS
A A BOUNDED SET WITH NO EXTREME POINT
The unit sphere in the space c0 (see 14K) has no extreme point.

ß EXISTENCE OF EXTREME POINTS

The requirement in 15.1 that the space E be locally convex may be
replaced by the hypothesis that any two points in A may be separated by
a linear functional continuous on E.

C EXTREME IMAGE POINTS

Let f be a continuous linear map of E into F where E and F are locally
convex and Hausdorff. Let A be a compact convex subset of E. Then
every extreme point of /JA] is the image of an extreme point of A.
D MAXIMUM OF A LINEAR FUNCTIONAL
Each real continuous linear functional on a convex compact subset of a
Hausdorff locally convex space assumes its maximum at an extreme point.

E SUBSETS OF A COMPACT CONVEX SET

If c; is a compact convex subset of a locally convex Hausdorff space E,
then for every subset S of C the following are equivalent.
(i) sup {f(x): x E S} = sup {f(x): x E C} for all continuous linear
functionals f on the real restriction of E;
(ii) C = (S)-;
(iii) S - includes all extreme points of C.
 SEc. 15 PROBLEMS 133
F TWO COUNTER-EXAMPLES
(1) In a finite dimensional space, a convex compact set is the convex
extension of the set of its extreme points, this extension being closed.
This fails in general, even in a Banach space. In the space ["' (6K), Iet en
be the sequence with n-th term 1 and other terms 0, and Iet A be the closed
convex extension of the set {enfn: n = 1, 2, · · ·} tagether with 0. Then A
is compact but is not the convex extension of the set of its extreme points.
(2) In theorem 15.2, the hypothesis that the closed convex extension of
the compact set A should be compact is essential: failing this, there may
be an extreme point of (A)- which is not in A. Let C(I) be the space
of real valued continuous functions on the interval I = [0, 1] and Iet E be
the space of real valued functions on C(I) with the topology of pointwise
convergence. Let e be the evaluation mapping of I into E: for each x EI,
e(x) is the element of E defined by e(x){f) = f(x) for allf E C(I).
(a) e[l] is a <;ompact subset of E.
(b) If g{f) = föf(x)dx for jE C(I), then g E (e[I])- butg ~ e[I].
(c) If F is the subspace of E generated by e[I] U {g}, then g is an
extreme point of F n (e[I])-.

G EXTREME HALF LINES

Let C be a convex cone such that C n (- C) = {0}. The only extreme
point of C is the vertex, zero. A half-line L from zero is an extreme half-
line of C if it is contained in C and if every open line segment contained in
C and intersecting L is contained in L. Then a half-line L is extreme if
and only if for every hyperplane H not containing 0 and intersecting C the
one point of H n L is an extreme point of H n C. Let ;;;; be the order
induced by C. Then the point x in Cis in an extreme half-line if and only
if x ;;;; y for y E C implies y = tx for some t.

H LIMITS AND EXTREME POINTS

Let {Aa} be a directed family of subsets of a topological space. Then
lim sup Aa is defined to be the set of points x such that every neighborhood
of x intersects some Aa for arbitrarily large a.
(a) If A = lim sup Aa, then A is closed and contains n {Aa -}. If the
family is decreasing, that is, Aa => A 8 whenever ß ;;;; a, then A = n {Aa -}.
If Ba= (U {A 8 : ß;;;; a})- then lim sup Aa = n {Ba}·
(b) Let {Ca} be a decreasing family of compact convex subsets of a
locally convex Hausdorff space E and let Aa be the set of extreme points
of Ca. Then lim sup Ca = (lim sup Aa)-.
(Clearly lim sup Aa c lim sup Ca, and lim sup Ca is a convex compact
set. Let f be a continuous linear functional on the real restriction of E;
then, for each a, there is Xa in Aa such that f(xa) = sup {f(x): x E Ca}·
Let x 0 be a duster point of the net {xa}; then x 0 E lim sup Aa and f(x 0 ) ;;;;
sup {f(x): x E lim sup Ca}·
(c) Let {Ca} be a directed family of compact convex subsets of a locally
convex Hausdorff space E, whose union is contained in a compact convex
subset of E, and let Aa be the set of extreme points of Ca. Then
lim sup Ca c (lim sup Aa)-; if lim sup Ca is convex, then lim sup Ca =
134 CH. 4 CoNVEXITY IN LINEAR ToPOLOGICAL SPACES

(!im sup Aa), that is, !im sup Aa contains all the extreme points of
!im sup Ca.
(Let Ba= (U {Ap: ß;?; a})- and use (b).)

I EXTREME POINTS IN V AND L 00 (see 6K and 14M)

Let X be the real line and Iet p, be the Lebesgue measure on X.
(a) The unit sphere of Ll(X,p,) has no extreme points.
(b) A member f of L 00 (X,p,) is an extreme point of the unit sphere if
and only if f 2 = 1.
(In these problems regardf = g to mean thatf(x) = g(x) almost every-
where.)

J EXTREME POINTS IN C(X) AND ITS ADJOINT

Let C(X) be the space of all real valued continuous functions on a
compact Hausdorff space X, with the supremum norm. Then f is an
extreme point of the unit sphere of C (X) if and only if j2 = 1.
A member F of the unit sphere of the adjoint of C (X) is an extreme
point if and only if F is represented by " ± a point measure "; that is, if
and only if there is x in X and F(f) = f(x) for allf in C(X), or F(f) =
-f(x) for allfin C(X). (See 14J.)
 Chapter 5
DUALITY

This chapter is devoted to the duality which is the central part of

the theory of linear topological spaces. The pattern of investigation
is simple: we seek to find, for each proposition about a linear topo-
logical space E, an equivalent proposition which is stated in terms of
the adjoint space E*. Of course, it is necessary that E*, in some
sense, describe E rather closely, and consequently our results, with
minor exceptions, are for locally convex spaces.
The first section is devoted to a number of propositions, primarily
geometric in character, which are intended for application both to a
linear topological space E and to its adjoint E*. These results are
consequently framed in terms which are immediately applicable to
both cases: we consider two arbitrary linear spaces E and F and a
bilinear functional (the pairing {unctional) on their product E x F.
The weak topologies for E and for F are the weakest topologies which
make the pairing functional continuous in each variable separately.
The pairing is (essentially) completely determined by either one of
the spaces with its weak topology, since the weakly continuous linear
functionals on E are precisely those functionals which are represented
by members of F. The geometry of a pairing is investigated by
means of polars, where the polar of a subset A of E is the set of all f
in F such that l<x,f)l ~ 1 for all x in A. Among the geometric
propositions which have important linear space consequences we
note: a weakly closed convex circled subset A of E is weakly compact
if and only if each linear functional which is bounded on the polar of
A is represented by a member of E, and each linear functional ~hich
is weakly continuous on a convex circled subset A of E can be approxi-
mated, uniformly on A, by a functional which is weakly continuous
on E.
135
136 CH. 5 DUALITY

The results on weak topologies have immediate consequences for the

natural pairing of a space and its adjoint. In particular, closed convex
sets are weakly closed, weakly bounded sets are bounded, and the last
result mentioned in the preceding paragraph yields a characterization of
completeness. The most striking result concerning the weak topology
foralinear topological space is the Eberlein-Krein-Smulian theorem:
the weakly closed convex extension of a weakly countably compact
subset of a complete locally convex space is weakly compact.
The structure of an arbitrary locally convex space E is considerably
more complicated than the theory of Banach spaces would indicate.
We use four families of subsets of E* to classify useful properties of E.
Consider the classes ~. ~. !7, and iY of all convex subsets of E*
which are, respectively, equicontinuous, with weak* compact closure,
strongly bounded, and weak* bounded. If E is a Banach space,
these classes are identical, but in general we have ~ c ~ c !7 c iY,
and each inclusion may be proper. The importance of these classes,
as a gauge of the properties of E, is the following: ~ = ~ if and only
if the topology for ~ is the maximal locally convex topology which
yields E* as adjoint, ~ = !7 if and only if the evaluation map of E
into the second adjoint E** is continuous (this map is always relatively
open), and ~ = iY if and only if the Banach-Steinhaus theorem for
Eis true (the space is then called a barrelled space, or tonnele). The
complexity in structure is further displayed by a Iack of permanence
properties for "good" characteristics. Thus, although quotients,
products, and direct sums of spaces for which the Banach-Steinhaus
theorem is true again have this property, closed subspaces and
adjoints may fail to retain the property. On the other hand, the
property of being semi-reflexive (the evaluation map carries the space
onto the second adjoint) is retained by products, direct sums, and
closed subspaces, but not by quotient spaces or by adjoints. In
general, little can be said about the adjoint of a linear topological
space, and even the adjoint of a complete metrizable space exhibits
features which are quite pathological from the point of view of
classical Banach space theory.
The theory of dual mappings isanatural part of the study of duality.
We characterize continuity and openness of a Iinear-transformation T
in terms of the adjoint map T*. Finally, the last section of the
chapter is devoted to the theory of metrizable spaces. As might have
been expected, there are substantial results here which apparently
have no generalization; however, even in this case, the adjoint space
may have a relatively complex structure.
 SEC. 16 PAIRINGS 137

16 PAIRINGS
This section is concerned with a bilinear pairing of spaces E and F.
The relationship between the spaces is studied by means of the topol-
ogy w(E,F) that F induces on E, and by means of polars. A number
of properties of a subset Qf E, most notably w(E,F)-compactness, can
be described in terms of polars. The completeness of E relative to
the topology of uniform conyergence on members of a family of sub-
sets of Fis characterized. There arealso a number of" tool" theorems
concerning subspaces, quotients, etc.
The study of a linear topological space and its adjoint is facilitated
by a further abstraction of certain essentials. If E is a vector space
and F is a linear space of linear functionals on E, then the relation in
which E stands to F is not dissimilar to that in which F stands to E.
In particular, each point x of E corresponds to a linear functional 4>
on F by means of the definition: 4>(!) = f(x) for all f in F. It is
convenient to choose a setting in which the roles of the two spaces are
entirely symmetric since theorems are obtained thereby which have
consequences both for the space E and for the space F of functionals
on E. It turns out that the abstraction which is made also yields a
highly efficient mechanism for many linear space computations.
Let E and F be two linear spaces over the same scalar field. A
bilinear functional on the product E x F is a functional B such
that
B(ax + by,f) = aB(x,f) + bB(y,f)
and
B(x,af + bg) = aB(x,f) + bB(x,g)
for all x and y in E, allfand g in F, and all scalars a and b. A pairing
isanordered pair <E,F) of linear spaces together with a fixed bilinear
functional on their product. The fixed bilinear functional in a
pairing is usually written in the common inner product notation; that
is, B(x,y) is written <x,y). Mention of the functional is sometimes
omitted, and it is said that E and F are paired spaces. In contexts
where an ordered pair of spaces has already been specified, the word
" pairing ", or the words "pairing of the spaces", may be used to
signify the bilinear functional alone.
Each pairing of two linear spaces E and F defines a mapping from
either of the two spaces into the space of all scalar functions on the
other. The canonical map T on F carries a member f of F into
the function T(f) on E such that T(f)(x) = <x,f) for all x in E.
The space P of all scalar functions on E is simply the product X {K:
x E E}, where K is the scalar field, obviously, P is a linear space.
138 CH. 5 DuALITY

Because the pairing functional is bilinear, the canonical map T is

linear, and the image of each member ofFisalinear functional on E;
consequently T[F] isalinear subspace of the algebraic dual E of E, 1

which in turn is a linear subspace of the product P. On the other

hand, if F is an arbitrary linear subspace of the algebraic dual of E,
then the natural pairing of E and F is the bilinear functional on
E x F defined by (x,f) = f(x) for all x in E andfin F. In this case
the canonical map of F into E 1 is the identity. In the general case,
when Fis not a subspace of the algebraic dual E it is not unusual to
1
,

identify (that is, to fail to distinguish between) a point f of Fand its

canonical image T (f), and to say that f is a linear functional on E.
Nevertheless, it must be remernbered that T(f) and T(g) may be
equal for distinct members fand g of F. It is obvious that T (f) and
T(g) are distinct for f -1= g if and only if T(h) -1= 0 for every non-zero
member h of F; and this condition is satisfied if and only if for each
member f of Fother than 0, there is some x in E such that (x,f) -1= 0.
This condition is described by saying that E distinguishes points
inF.
Clearly the roles of E and F are interchangeable here. There is a
canonical linear mapping of E into the subspace F 1 of X {K: f E F}
such that the image of a point x of E is the functional which carries f
of F into (x,j), and the mapping is one-to-one if and only if F dis-
tinguishes points in E. When E distinguishes points in F and F
distinguishes points in E, the pairing is said to be separated. For
the natural pairing of a linear space E and its algebraic dual E 1 it is
obvious that E distinguishes points of E and that E distinguishes
1 1

points of E. Hence this pairing is separated.

By making use of the canonical map a very useful topology can be
defined for a linear space E which is paired with a space F. The
canonical map T carries E into the space P = X {K: f E F} of all
functionals on F, and the product P may be assigned the product
topology. This topology is also called the topology of pointwise
convergence. The space Pis complete, locally convex and Hausdorff,
and the algebraic dual of Eis a closed subspace. The space Eis now
topologized by defining a set U to be w(E,F)-open if and only if U
is the inverse image under T of a set which is open in P relative to the
product topology. Formally, the weak (E,F) topology, denoted by
w(E,F), is the family of all subsets U of E suchthat for some subset V
of P, V is open relative to the topoiogy of pointwise convergence and
T - 1 [V] = U. Equivalently, a subset B of Eis w(E,F)-closed if and
only if B is the inverse image under T of a closed subset of P. A
 SEc. 16 PAIRINGS 139

word of caution is necessary: if B is w( E,F )-closed in E, it does not

follow that T [B] is closed in P relative to the topology of pointwise
convergence. It is true, however, that if B is w(E,F)-closed, then
T[B] is closed in T[E], where T[E] has the relativized product
topology. The space P is locally convex and Hausdorff, and it
follows that E with the topology w(E,F) is locally convex, and that it
is Hausdorff if and only if T is one-to-one; equivalently, E with the
topology w(E,F) is Hausdorff if and only if F distinguishes members
of E.
The situation with respect to boundedness, total boundedness, and
compactness relative to the w(E,F) topology is quite simple. If B
is a bounded subset of P =X {K:jE F}, then the projection of B
into each coordinate space is bounded, and hence at = sup {l<x,J)I:
x E B} < oo for each f in F. Then B is a subset of the set X {At:
f E F}, where At = {t: t E K and lt I ~ at}· By the Tychonoff
theorem 4.1, this product is compact. It follows that each bounded
closed subset of Pis compact, and that each bounded subset is totally
bounded (it is also possible to derive total boundedness directly from
boundedness without appealing to the Tychonoff theorem). In
view of the definition of the topology w(E,F), it is clear that a subset
of Eis w(E,F)-bounded or totally bounded if and only if the image
under the canonical map T of the subset has the same property.
Consequently a subset B of E is bounded if and only if it is totally
bounded; Bis totally bounded if and only if sup {l(x,J)I: x E B} < oo
for each f in F.
The following theorem summarizes the preceding results, and
states a few other simple propositions about the topology w(E,F)
which follow from known properties of the product topology (see
Section 4).

16.1 ELEMENTARY PROPERTIES OF THE w(E,F)-ToPOLOGY Let E and

F be paired linear spaces. Then:
(i) the space E with the topology w(E,F) is a locally convex linear
topological space, and E is a Hausdorff space if and only if F
distinguishes points of E;
(ii) the family of all sets of the form {x: I(x,f)l ~ 1 for i = 1, · · ·, n},
where {!1> / 2 , • • ·, fn} is an arbitrary finite subset of F, is a local
base for w(E,F);
(iii) a net {xa, a E A} in E converges to an element x in E relative to
the topology w(E,F) if and only if {(xa,f), a E A} converges to
(x,f) for each f in F;
140 CH. 5 DUALITY
(iv) a lunction R on a topological space G to Eis continuous relative
to the topology w(E,F) if and only if g1 is continuous on G lor
each I in F, where g1(z) = <R(z),l> lor every z in G;
(v) 11 ff is a topology lor E such that the map x ~ <x,l> is ff-
continuous lor each I in F, then ff is stronger than w(E,F); and
(vi) a subsei B ol E is w(E,F)-totally bounded if and only if it is
w(E,F)-bounded, and this is the case if and only if sup {l<x,l>l:
x E B} < oo lor each I in F.
It is clear that there is a complete duality betwaen the roles of E
and F in the foregoing discussion and that the argument which Ieads
to theorem 16.1 can be carried out with the roles of the two spaces
E and F interchanged. Alternatively, a pairing of F and E can be
defined by setting <<1, x)) = <x, 1>, and the results on the topology
w(E,F) which have just been obtained yield theorems on a topology
for F, the weak (F,E) topology for F, denoted by w(F,E). Ex-
plicitly, a subset U of Fis a member of the topology w(F,E) if and
only if U is the inverse image, under the canonical map, of a subset V
of the space of all functionals on E, such that V is open under the
topology of pointwise convergence. Clearly each theorem on the
topology w(E,F) has a dual which applies to w(F,E). These dual
results will not be stated.
If E and F are paired and g is a linear functional on E, then g is
represented by an element I of F if and only if g(x) = <x,l> for all
x in E. The w(E,F)-topology yields a precise description of those
linear functionals which can be represented by members of F.

16.2 REPRESENTATION OF w(E,F)-CONTINUOUS LINEAR FUNCTIONALS

A lunctional g on E is represented by some member ol F if and only if g
is a w( E,F )-continuous linear lunctional. The member ol F which
represents a linear lunctional g is unique if and only if E distinguishes
the members ol F.
PROOF If there is some element I in F such that g(x) = <x,l> for
every x in E, it is obvious from the properties of the bilinear functional
thatg is linear, and from (iii) of theorem 16.1 thatg is continuous with
respect to the topology w(E,F). If, conversely, g is linear and con-
tinuous on E, then from (ii) of theorem 16.1 there is a finite set
/ 1 , · · · ,ln in F suchthat lg(x)l ~ 1 whenever x is in E and l<x,f;)l ~ 1
for i = 1, · · ·, n. In particular, if x is in E and <x,ft> = 0 for
i = 1, · · ·, n, then li(ax) = aft(x) = 0 for each scalar a and all i.
Thus, lg(ax)l ~ 1 for every a, and hence g(x) = 0. Thus the null
 SEc. 16 PAIRINGs 141

space of g contains the intersection of the null spaces of T (!1 ), · · ·,

TUn), where T is the canonical map of F into the algebraic dual of E,
and hence g isafinite linear combination of T(f1 ), · · ·, T(fn), by 1.3.
Since T is linear, it follows that g is the canonical image of an element
f of F. The final statement of the theorem is obvious.lll
The most economical device for defining locally convex topologies
in linear spaces, and for performing calculations concerning these
topologies, isthat of polars. Let (E,F) be a pairing. The polar in
F of a subset A of E, written A 0 , is the set of all f in F such that for
each x in A, l<x,f)l ~ 1; the polar in E of a subset B of F, written
B 0 , is the set of all x in E such that for each f in B, I(x,j) I ~ 1. In
the next section several topologies will be described by means of
polars. Here it suffices to remark that the family of all polars of
finite subsets of F is a local base for w(E,F). Since polars will be
used extensively throughout the rest of the book, several rules for
computation with them are collected in the following theorem. Of
course, each statement in the following list has a dual form, obtained
by interchanging the roles of E and F.
16.3 CoMPUTATION RuLES FOR PoLARS Let (E,F)' be a pairing.
Then:
(i) for each subsei A of E, A 0 is convex, circled, and w(F,E)-closed;
(ii) if A c B c E, then A 0 ::::> B 0 ;
(iii) if A c E and if a is a non-zero scalar, then (aA) 0 = (1fa)A 0 ;
(iv) if A is a non-void subset of E, then (A 0 ) 0 , which will be written
A 0 0 henceforth, is the smallest convex circled w(E,F)-closed set in
E which contains A;
(v) if A is a non-void subset of E, then A 0 0 ° = A 0 ;
(vi) if {At: t E C} is afamily of subsets of E, then (U {At: t E C}) 0 =
n {At 0 : t E C}; and
(vii) if {At: t E C} is a family of convex circled w(E,F)-closed sets in
E, then <n {At: t E C}) 0 is the smallest convex circled w(F,E)-
closed set in F which contains At0 for every t in C.
PROOF Statements (i), (ii), and (iii) are clear from the definition of
polar. As to (iv), it is evident that if Cis the smallest convex circled
w(E,F)-closed set which contains A, then C c A 0 0 ; on the other hand,
if x is a point not in C, then by virtue of 14.4 there is a w(E,F)-
continuous linear functional g on E which in absolute value is at most
1 on C and is greater than 1 at x. By the representation theorem
there is an f in F such that g(y) = (y,j) for all y in E. Clearly
f E A 0 ; hence, x i A 0 0 • Conclusion (v) follows immediately from (i)
142 CH. 5 DUALITY

I
and the dual of (iv). As to (vi), it is obvious that (x,f) ~ 1 for each I
x in the union of the members of a family of sets if and only if it holds
for each x in each member. From (iv) and the dual of (vi) it follows
that
(n {At: t E C}) 0 = (n {At o: t E C})
0 0 = (U {At
t E C})0 °,
0 :

which yields (vii).lll

There are several direct consequences of the preceding proposition.
First, recall (16.1) that a subset A of E is w(E,F)-bounded if and
only if for each f in Fit is true that sup {I (x,f) I: x E A} is finite. But
this supremum is finite if and only if l<x,af)l ~ 1 whenever a is a
scalar which is sufficiently small in absolute value~that is, if and only
if af E A 0 for all sufficiently small a. This establishes the first state-
ment of the following theorem, and the second follows in view of
16.3(iv).

16.4 PoLARS OF BouNDED SETS Let E and F be paired linear spaces.

Then
(i) a subset A of Eis w(E,F)-bounded if and only if its polar A 0 is
radial at 0; and
(ii) if A is a subset of E, then A 0 is w(F,E )-bounded if and only if the
w(E,F)-closed convex circled extension A 0 0 of A is radial at 0.
The computation rules 16.3 yield an easy proof for a result con-
cerning w(E,F)-dense subspaces of E. If Gis a linear subspace of E,
then G 0 0 is the w(E,F)-closure of G, in view of part (iv) of 16.3, and it
follows from part (v) that G 0 0 = E if and only if G 0 = E 0 • This last
condition can be restated: if (x,f) = 0 for all x in G, then (x,f) = 0
for all x in E. The following proposition is a trivial consequence.

16.5 w(E,F)-DENSE SUBSPACES The linear extension of a subset A of

Eis w(E,F)-dense in E if and only if each member of F which vanishes
on A vanishes on E (more precisely, if and only if whenever (x,f) = 0
for all x in A, then (x,f) = 0 for all x in E).
All of the compactness results of this chapter are derived from the
following fundamental theorem.

16.6 SMULIAN's CRITERION FüR w(E,F)-CoMPACTNESS Let E and F

be paired linear spaces and Iet B be a w(E,F)-closed convex circled subset
of E. Then B is w(E,F)-compact if and only if B 0 is radial at 0 and
each linear functional on F which is bounded on B 0 is represented by some
member of E.
 SEC. 16 PAIRINGS 143
PROOF Let T be the canonical map of E into the algebraic dual F' of
F, let E have the topology w(E,F), and let F' have the topology
w(F',F). The setBis compact if and only if T[B] is compac"t, in
view of the definition of w(E,F). But bounded subsets of F' are
totally bounded by 16.1(vi), and F' is complete because it is a closed
subset of a product of complete spaces. Consequently B is compact
if and only if T [B] is bounded and closed in F '.
The set B 0 is radial at 0 if and only if Bis bounded, by 16.4, and B
is bounded if and only if T [B] is bounded because of the definition
of w(E,F).
Finally, each linear functional bounded on B 0 is represented by a
member of E if and only if the polar of B 0 in F' is contained in T [ E].
But B 0 is identical with the polar of T [B] in F, by a direct verification,
and the second polar of T[B] is the closure of T[B] in F' by 16.3(iv).
Hence each linear functional bounded on B 0 is representable by a
member of E if and only if the closure of T[B] is contained in T[E].
But B was supposed tobe closed; hence T[B]- c T[E] if and only
if T [B] is· closed. The results of the last three paragraphs yield the
theorem.JJI
The topology w(E,F) is seldom metrizable; however, there is a
result on metrizability which is frequently useful. Assurne that B is
a w(E,F)-compact subset of E, and that A is a subset of F which
distinguishes points of B in the sense that, if x and y are distinct
members of B, then there is a member f of A suchthat (x,f) =F (y,f).
This assumption implies that the topology of pointwise convergence
on A is Hausdorffon B; it is weaker than the topology w(E,F), with
respect to which Bis compact. It follows that the topology w(F,E)
and the topology of pointwise convergence on A coincide on B (the
topological theorem used here is given in problern 4A). If A is
countable, then the topology of pointwise convergence on A is the
relativized product topology, where B is considered as embedded in
X {K: f E A}, where K is the scalar field, and the product is countable.
Hence B is metrizable. These considerations prove the following
proposition.
16.7 THEOREM lf Bis a w(E,F)-compact subset of E, and if Ais a
subset of F which distinguishes members of B, then the topology w(E,F)
for Bis identical with the topology of pointwise convergence on A; if A
is iountable, the topology w(E,F) for Bis metrizable.
The last theorems of the section concern completeness of E relative
to the topology of uniform convergence on members of a family of
144 CH. 5 DUALITY

subsets of F, where E and F are paired linear spaces. Suppose that

d is a family of w(F,E)-bounded subsets of F and that T is the
canonical map of E into the algebraic dual F' of F. We define the
topology of uniform convergence on members of d for E
(again denoted by ~) as follows: a subset of E is open relative to
~ if and only if the set is the inverse image under T of a subset of F'
which is open relative to the topology of uniform convergence on
members of d. It is easy to verify that (E,~) is a locally convex
space and non-zero scalar multiples of finite intersections of polars of
members of d form a local base. A net in E converges to a Iimit
relative to ~ if and only if its image under T converges to the
image of the Iimit uniformly on members of d.
It was shown eariier (in Section 8 on function spaces) that if d is a
family of subsets of a topological space X and E is the family of all
those functions f on X to a complete linear topological space such that
f is bounded and continuous on each member of d, then E is com-
plete relative to the topology ~ of uniform convergence on members
of d. The principal remaining theorem can be regarded as a sort of
converse to this proposition. Under certain circumstances com-
pleteness relative to ~ of a family G of linear functionals implies
that G is the family of alllinear functionals which are bounded and
continuous on each member of d. The critical step in the proof of
this converse is furnished by the following approximation theorem.
16.8 APPROXIMATION THEOREM Let E and F be paired linear spaces,
and Iet A be a w(E,F)-closed convex circled subset of E. If f is any
linear functional on E which is w(E,F)-continuous an A, then for each
e > 0 there is a point g in F such that if(x) - (x,g)l ;;::; e for all x
in A.
PROOF lt is enough to prove the tlieorem for the case in which Fis a
subspace of the algebraic dual E' of E; for, in general, if T is the
canonical map of F into E', then the topologies w(E,T[F]) and
w(E,F) are identical.
If f is a linear functional on E which is w(E,F)-continuous on A,
then for each positive real number e there is a w(E,F)-neighborhood
U of 0 suchthat I/I is at moste on U n A; that is, (/je) E (A n U) 0
where the polar is taken in E'. The set U may be supposed tobe the
polar of a finite subset B of F, and A is the polar of A 0 by 16.3.
Hence, using 16.3 again, (A n U) 0 = (A 0 0 n B0 ) 0 = (A 0 U B) 0 °,
and the latter set is the w(E',E)-closed convex circled extension of
A 0 U B. But the w(E',E)-closed convex circled extension B 0 ° of
 SEC. 16 PAIRINGS 145
the finite set B is w(E',E)-compact, and hence A 0 + B 0 °is w(E',E)-
closed by 5.2(vii). Consequently (A 0 U B)0 ° c A 0 + B 0 °. Then
1/e is a member of A 0 + B 0 °, and therefore there is a member g of F
(in fact a member of eB 0 °) such that I - g E eA 0 ; restated, there is a
w(E,F)-continuous linear functional g on E which approximates I
within e on A.lll .
The completeness theorem is a direct consequence of the fore-
going. Recall that a family .91 of sets is directed by :::> if and only if
for A and B in .91 there is C in .91 such that C => A U B.
16.9 GROTHENDIECK's CoMPLETENESS THEOREM Let E and F be
paired linear spaces, and Iet .91 be a non-void lamily ol w(F,E)-bounded,
w(F,E )-closed, convex and circled subsets ol F such "that .91 is directed
by :::>
(i) The canonical image T[E] ol Ein F' is dense relative to ~in
the space ol all linear lunctionals on F which are w(F,E)-
continuous on each member ol .91.
(ii) 11 the linear extension ol the union ol the members ol .91 is F, then
E is complete relative to ~ if and only if each linear lunctional
on F which is w(F,E)-continuous on each member ol .91 is w(F,E)-
continuous on F; equivalently, this is the case if and only if each
hyperplane which intersects every member ol .91 in a w(F,E)-
closed set is w(F,E)-closed.
PROOF Let G be the space of all linear functionals on F which are
w(F,E )-continuous on each member of .91. First observe that each
I in G is bounded on each A in .91, because I is w(F,E)-uniformly
continuous on A and Ais totally bounded relative to w(F,E). There-
fore the space G is a locally convex space with the topology ~" and
the family of all the sets of the form{/: ll(x)l < e for all x in A} is a
local base for ~- In view of the approximation theorem, part (i) is
clear. If the linear extension of the union of .91 is F, then ( G,~)
is a complete Hausdorff space. Since E is complete relative to ,Y:,;
if and only if the canonical image of E is a complete subspace of G,
E is ~-complete if and only if T[E] = G, which is exactly the
content of part (ii).lll
The section is concluded with a computation of polars in sums and
products and a discussion of subspaces and quotient spaces. If E1
is paired to F 1 for each t in an index set A, then the natural pairing
of .L {E1: t E A} and X {F1: t E A} is defined by letting (x,f) =
.L {(x(t),f(t)): t E A} for each x in the direct sum and each f in the
product. It will be convenient in the statement and proof of the
146 CH. 5 DuALITY
following proposition to consider the spaces Et as subspaces of the
direct sum, thus avoiding explicit mention of the injection maps.

16.10 LEMMA ON SUMS AND PRODUCTS Let Et and Ft be paired linear

spacesfor each t in an index set A and Iet 2 {Et: t E A} and X {Ft: t E A}
have the natural pairing. Then:
(i) if Bt is a subset of Et for
each t in A, then the polar of the union
of the sets Bt is the product X {Bt 0 : t E A} of the polars Bt0 in
Ft of Bt; and
(ii) if Ct is a circled subset of Ft for each t in A, then the polar of the
product X {Ct: t E A} contains the convex extension B of the
union of the polars ( Ct)o of Ct in Et and is contained in aB for
each real number a, a > 1.

PROOF To prove (i), observe that if f EX {Bt 0 : t E A} and x belongs

to B 11 then l<x,f>l = l<x(t),j(t))i, which is at most one because
f(t) E Bt0 • On the other hand, if f belongs to the polar of the union
of the sets B 11 then for each t and each x in Bt, l<x(t),j(t))i =
l<x,J>I ~ 1; hence, f(t) E Bt0 • To prove (ii), note that if x E (Ct)o,
then l<x,J>I = l<x(t),j(t))i ~ 1 for each f in X {Ct: t E A} and
hence U {(Ct) 0 : t E A} c (X {Ct: t E A}) 0 • Since the polar of X {Ct:
t E A} is convex, it contains B. To prove the other inclusion, Iet x
belong to the polar in E of the product X {Ct: t E A}, and Iet at be
the supremum of l<x(t),j(t)>i for f in X {Ct: t E A}. Because ct is
circled, if j(t) E Ct and l<x(t),j(t))i = a, then <x(t), bj(t)) = a
for some suitable b with lbl = 1; that is, with bf(t) E Ct. Since
l<x,J>I = 12 {<x(t),J(t)): t E A}l ~ 1 for f in X {Ct: t E A}, it
follows that 2 {at: t E A} ~ 1. By the definition of at, x(t)fat belongs
to the polar of C 1 when at =/: 0, and hence, if y = 2 {at(x(t)fat):
at =1: 0}, y belongs to the convex extension of the union of the sets
(Ct) 0 • Finally, x = (x - y) + y, and <x - y, /) = 0 for all f in
X {Ct: t E A}; hence, b(x - y) belongs to the convex extension of the
union of (Ct)o for all scalars b. Therefore, for a > 1,

x = a [!a y + (1 - !)a ·(a(x -- y)]

1)
E aB. III

If E and F are paired linear spaces and His a subspace of E, then

there is an induced pairing of EfH and H 0 which is defined by
letting <A,j) = <x,j), where A is an arbitrary member of EfH, x is
an arbitrary member of A, and f belongs to H 0 • (Notice that the
polar of H coincides with the annihilator H j_.) The next proposition
 SEC. 16 PAIRINGS 147
identifies the topology w(E(H,H 0 ), and the topology w(H,F(H 0 )
which is derived from the induced pairings of Hand F(H 0 •
16.11 ToPOLOGIES FüR SuBSPACES AND QuoTIENTS Let E and F be
paired linear spaces, Iet H be a subspace of E, and Iet Hand F(H 0 and
E(H and H 0 have the induced pairings. Then
(i) the topology w(E,F) relativized to His identical with the topology
w(H,F(H 0 ), and
(ii) the quotient topology for E/H which is derived from w(E,F) is
identical with w(E(H,H 0 ).
PROOF The family of polars in Hof finite subsets f 1 + H 0 , • • · , fn +
H 0 of F(H 0 is a local base for the topology w(H,F(H 0 ). But the polar
in Hof such a subset is simply the set {x: x EH and l<xJi>l ~ 1 for
i = 1, · · ·, n}, and this is the intersection with H of the polar of
f 1 , · · ·,fn in E. It follows that w(H,F(H 0 ) is the relativization of
w(E,F), and (i) is established.
To establish (ii), observe that the quotient topology for E(H has a
local base consisting of all collections of the form {x + H: x E X 0 },
where X is a finite subset of F and X 0 is the polar of X in E. On
the other hand, the topology w( E/H,H 0 ) has a local base consisting of
collections {y + H: y E Y0 } where Y isafinite subset of H 0 • Clearly
the latter topology is weaker, and identity of the two is established if
it is shown that for each finite subset X of F there is a finite subset Y
of H 0 such that Y 0 + H c X 0 + H. To prove this fact first note
that given a finite subset X of F there is a finite subset Y of H 0
suchthat X 0 ° n H 0 c Y 0 °. (In case Fis a Hausdorff space the closed
convex circled extension X 0 °of X is a finite dimensional compact set,
and there is clearly a finite subset Y of H 0 whose closed convex exten-
sion contains X 0 ° n H 0 • If Fis not Hausdorff, then a simple argument
using the description 5.11 of non-Hausdorff spaces serves to establish
the same result.) Since X 0 ° n H 0 c Y 0 °, it follows from the
computation rules 16.3 and from 5.2(v) that
Y 0 c (Xo u H) 0 0 c (Xo + H)- c X 0 + H + X 0 c (tX)o + H.
Hence Y 0 + H c (!X) 0 + Hand the theorem follows.lll
There is an asymmetry in the preceding result which stems from
the fact that H 0 is necessarily w(F,E)-closed. This difference is
exhibited in the following proposition, which is an immediate corollary.
16.12 CoROLLARY If E and F are paired linear spaces and G is a
subspace of F the quotient topology for E(G0 is the topology w(EfGmG- ),
where EfGo and the closure c- of G have the induced pairing.
148 CH. 5 DUALITY

A particular consequence of this corollary is that the quotient

topology for EfG0 is certainly not identical to the topology w(EfG0 ,G)
if E separates points of F and G is not closed, for in this case the two
topologies yield distinct classes of continuous linear functionals.

PROBLEMS
A DUALITY BETWEEN TOTALLY BOUNDED SETS
Suppose that E and F are paired linear spaces, that A is a w(E,F)-
bounded subset of E and that B is a w(F,E )-bounded subset of F. Then
A is totally bounded in the topology on E of uniform convergence on B if
and only if B is totally bounded in the topology on F of uniform con-
vergence on A. (Use 8.17 and 16.1(vi).)
If d and f!J are families of w(E,F)- and w(F,E)-bounded subsets of E
and F, then each A E d is totally bounded in the topology of uniform con-
vergence on members of f!J if and only if each B E f!J is totally bounded in
the topology of uniform convergence on members of d.

ß POLAR OF A SUM
Let (E,F) be a pairing and Iet A and B be non-void subsets of E. If A
and B are convex and circled, then 1(A0 n B 0 ) c (A + B) 0 c A 0 n B 0 ,
To obtain a precise expression for (A + B) 0 , valid more generally, define
the following operation 6. : if X and Y are subsets of F, Iet X 6 Y =
U {r X n s Y: r ~ 0, s ~ 0, r + s ~ 1}. Then X n Y c 2(X 6. Y) and if
X and Y are circled, X 6. Y c X n Y. Now if A and B are circled, and e
is a number suchthat O<e<l, then e(A + B) 0 c A 0 6.B 0 c (A+Bt.
C INDUCTIVE LIMITS li (see 11, 17G, 19A, 22C)
Let E be a linear space and {E1,f1 : t E A} be a family of locally convex
spaces E1 and linear mappings ft of E1 into E. The strongest locally
convex topology for E which makes each / 1 continuous is called the in-
ductive topology for E determined by {E 1,f1 : t E A}.
(a) Let E 0 be the linear span of U {f1[E1]: t E A}. If M is a complement
of E 0 in E, then the inductive topology is the direct sum of its relativization
to E 0 and the strongest locally convex topology for M.
This means, roughly speaking, that any parts of E outside E0 have a
topology independent of the determining family {E1,f1 : t E A}: this con-
tributes very little to the utility of the notion of inductive topology. It is
therefore reasonable to assume, whenever this is convenient, that the linear
span of U {/1[ E1]: t E A} is the whole given linear space E.
(b) Let {E1 : t E A} be an inductive system (see Section 1) and Iet E =
lim ind {E1 : t E A}. The space Eis (algebraically) isomorphic to a quotient
of the direct sum 2: {E1: t E A}. If each E 1 is locally convex, there is a
natural topology for E, namely, the quotient of the direct sum topology.
If Q denotes the quotient mapping and / 1 the injection of E 1 into the direct
sum, then this topology for E is the strongest locally convex topology
which makes each Q o / 1 continuous. Thus the natural topology of an
inductive Iimit is an inductive topology.
 SEC. 16 PROBLEMS 149
Suppose also that, for t ~ s, the canonical mappings Q18 of E. into E1 are
all continuous. The natural topology of E is then the same as the natural
topology defined by any cofinal subset B of A (cf. 11).
In these circumstances, this topology will be called the inductive Iimit
topology for lim ind {E1 : t E A}.
Suppose now that the space E has the inductive topology determined by
{Et>ft: t E A}, and Iet E0 be the linear span of U {ft[E1]: t E A}.
(c) A local base is formed by the family of all convex circled subsets U
of E radial at 0 for which ft -l[ U] is a neighborhood of zero in E1 for each
t E A. Let o//1 be a local base in E1 for each t E A and assume that E 0 = E.
Then the family of all convex circled extensions of sets U {f1[U1]: t E A},
as U1 runs through o//1 for each t E A, is also a local base.
(d) A linear mapping T of E into a locally convex space is continuous if
and only if each T o f 1 is continuous.
(e) The quotient topology of any locally convex space by a subspace is a
special case of an inductive topology; so also is a direct sum topology.
(f) When E 0 = E, the inductive topology determined by {Et>ft: t E A}
is a quotient of the direct sum topology for L: {E1: t E A}. (The kernel of
the quotient mapping is the null space of the mapping .L {x(t): t E A}--+
L: {ft(x(t)): t E A}.)
(g) An inductive topology may fail to be Hausdorff, even if each E 1 is a
Hausdorff space; it may fail to be complete, even if each E1 is complete.
(For the last part, see the counter-example in 20D.)
(h) Let F be a locally convex space such that, for each t, any linear
mapping T of E1 into F is continuous provided that its graph is closed.
Then any linear mapping of E into F has the same property. (Use (e),
and exhibit the graph T o f 1 as an inverse image of the graph of T.)
(i) Let K(X) be the set of continuous real or complex valued functions
of compact support on a topological space X. Foreach compact subset B
of X, Ka(X) is the subspace of K(X) of functions whose supports lie in B,
with the topology of uniform convergence on B; the topology of K(X) is
defined by the local base consisting of all convex circled sets U radial at 0
such that, for each compact subset B of X, U n K 8 (X) is a neighborhood
of 0 in KR( X) (see 8I). This is the inductive topology determined by the
spaces K 8 (X) and their injections into K(X). For, ordering the compact
subsets of X by inclusion, the spaces K 8 (X) form an inductive system, and
if B c C, the injection of K 8 (X) into Kc(X) is continuous (in fact a
homeomorphism). Then K(X) = lim ind {K8 (X): B compact, B c X}
and the inductive topology for K(X), being the strongest locally convex
topology making the injection of each K 8 (X) into K(X) continuous, has
the local base described. In view of (b ), it is sufficient to determine the
topology of K(X) by means of any co-base for the compact subsets of X;
that is, any family of compact subsets {Ba} of X with the property that each
compact subset of Xis contained in some {Ba}·
U) Let !7) be the set of infinitely differentiable real or complex valued
functions on Rn of compact support. For each compact subset K of Rn,
!l)K is the subspace of !7) offunctions with supports in K, with the metrizable
topology defined by the sequence {qm} of pseudonorms, where qm(f) =
sup {IDPj(x)l: x E K, 0 ~ IPI ~ m} (see 8J). A local base for the topology
150 CH. 5 DUALITY

of P} is formed by all convex circled sets U radial at 0 such that, for each
compact subset K of Rn, U n P}K is a neighborhood of 0 in P}K· This is
the inductive topology determined by the spaces P}K and their injections
into !1), For each positive integer r, let K (r) = {x: lxd ;;:; r, 1 ;;:; i ;;:; n};
then {K (r): r = 1, 2, · · ·} is a co-base for the compact subsets of Rn, and,
again by (b ), the family {!1)K<r>: r = 1, 2, · · ·} is sufficient to determine the
inductive topology for !1).

0 PROJECTIVE LIMITS
Let F be a linear space and {Ft>g1 : t E A} a family of locally convex
spaces F1 and linear mappings g1 of F into F1• The weakest topology for F
which makes each g1 continuous is called the projective topology for F
determined by {Ft>g1: t E A}.
(a) Let N = n {g1 - 1(0): t E A}. Then the projective topology for F is
the product of its quotient for FfN and the trivial topology for N.
Thus the assumption that N = {0} involves no loss; clearly it is essential
if F with the projective topology is to be a Hausdorff space.
(b) A projective limit (see Section 1) has a natural topology, the relativi-
zation of the product; this is a projective topology.
(c) If Fhasthe projectivetopology determined by{Ft>g1: t E A} and if N =
{0}, thenFmay be exhibitedas lim proj {g1[F]: t E A}withits natural topology.
Suppose now that F has the projective topology determined by {Ft>g 1 :
t E A}, and let N = n {gt -l(O): t E A}.
(d) The projective topology is locally convex; a local subbase is formed
by the family of all sets g1 -l[U1], as U1 runs through a local base in F 1•
(e) A linear mapping T of any linear topological space into F is con-
tinuous if and only if each g1 a T is continuous.
(f) Relativization and product are special cases of projective topologies.
(g) If N = {0}, the projecti~ topology is a relativization of the product
topolQgy for X {F1 : t E A}.
(h) Any locally convex topology is the projective topology determined
by a family of normed spaces or Banach spaces. (For each convex circled
neighborhood U of 0 in F, let Fu = Ffn {eU: e > 0}. Compare (g) and
(h) with 6.4.)
(i) lf N = {0} and each F1 is a Hausdorff space, then so is F; even if
each F1 is complete, F may fail to be complete.
(j) A subset B of F is [totally] bounded if and only if each g1[B] 1s
[totally] bounded.
(k) The space t! is the projective limit of the spaces c<m> (see 81).

E DUALITY BETWEEN INDUCTiVE AND PROJECTIVE LIMITS

Let {E1: t E A, ~, Q1.(t ~ s)} be an inductive system. Suppose also
that each E1 is a locally convex Hausdorff space and that each Q1.(t ~ s) is
continuous. Let F1 be the adjoint of E1 and define P 81 for t ~ s by
<x.,P.1(y1)) = <Q 1.(x.),y1) for all X 8 E E. and y 1 E F 1• Then {F1 : t E A, ~,
P81(t ~ s)} is a projective system. The adjoint of lim ind {E1: t E A} with
the inductive topology is lim proj {F1: t E A}.
There is an exactly similar dual result.
Topological conjectures generally fail: see 20D and 22G.
 SEC. 16 PROBLEMS 151
F SEQUENTIAL CONVERGENCE IN V(X,p.) li (see 91)
(a) The space V(X,p.) is weakly sequentially complete: that is, if {fn} is
a sequence of elements of V(X,p.) Cauchy in the weak topology on V(X,p.),
then there is an element f in V(X,p.) such that fn---'?- f weakly. (Without
loss of generality, X may be supposed to be totally a-finite, so that 14M
applies. Use 9I(a) to show the existence ofj E V(X,p.) with !im fAfndp. =
fAfdp. for each measurable A c X. lt follows that f f.gdp.---'?- f fgdp. for
every simple function g. The uniform boundedness theorem shows that
{fn} is bounded in norm; since the set of simple functions is dense in the
adjoint of V(X,p.), it follows that fn---'?- f weakly.)
(b) If {fn} is a sequence in V(X,p.) weakly convergent and pointwise
convergent a.e., then it is convergent in norm. (Use 9I(b).)
(c) In the space 11, a weakly convergent sequence is convergent in norm.
(U se (b): weak convergence implies pointwise convergence on a discrete
measure space.)

G DENSE SUBSPACES
Let E and F be paired linear spaces, and Iet .91 be a family of w(F,E)-
bounded, w(F,E )-closed, convex and circled subsets of F, directed by ::>,
whose union spans F. If G is a subspace of E and ~~ the topology of
uniform convergence on the members of .91, the following are equivalent:
(i) Gis ~;-dense in E;
(ii) w(F,G) and w(F,E) agree on members of .91;
(iii) w(F,E)-closed convex subsets of members of .91 are w(F,G)-closed;
(iv) whenever B is a w(F,E)-closed convex subset of some member of
.91 and x rf: B, there is an element of G which strongly separates B
and x.
(The implications (i) => (ii) => (iii) and the equivalence of (iii) and (iv)
are Straightforward; to show that (iii) implies (i) use Grothendieck's
completeness theorem 16.9.)

H HELLY'S CONDITION
(a) Let E be a linear space and B a convex circled subset of E. Sup-
pose that {fi: 1 ;:;; i ;:;; n} is a finite set of linear functionals on E and
{ci: I ;:;; i ;:;; n} a set of scalars. Then for each e > 0 there is a point
x. E (1 + e)B with fi(x.) = ci if and only if
(*)I L {aici: 1 ;:;; i;:;; n}l ;:;; sup {I.L {aij;(x): 1 ;:;; i;:;; n}l: x EB}
for all choices of the scalars {ai: 1 ;:;; i;:;; n}. (Map each z in E into the
point (/1 (x), · · · .fn(x)) of Kn, and use a Hahn-Banachtheorem in Kn.)
(b) It can be deduced from (a) that if <E,F) is a pairing and A a non-
void subset of E, the w(E,F)-closed convex circled extension of A is A 0 0
(16.3(iv)). (Let B be the convex circled extension of A; then B 0 0 = A 0 0 •
If z E A 0 0 but z rf: B-, there arelinear functionalsft E F with l.t;(z - x)l > 1
for all x E B. Put c, = J,(z) and use (a).)
(c) Let E be a linear topological space and B a convex circled compact
subset of E. Suppose that F is a family of continuous linear functionals
on E and {c1 : f E F} a family of scalars. Then there is a point x 0 E B with
152 CH. 5 DUALITY

= c1 for all f E F if and only if the condition (*) holds for every choice
j (x 0 )
of/1 , /2 , · · ·,fn in Fand c1 = c11 for 1 ~ i ~ n.
(d) It follows from (c) that if (E,F) is a pairing and Ba convex circled
w(E,F)-compact subset of E, then any linear functional on F bounded on
B 0 is represented by a member of E (cf. Smulian's weak compactness
criterion 16.6).
[When E is the adjoint of a Banach space, or a reflexive Banach space,
and Bis its unit sphere, (c) becomes a theorem of Hahn on the existence of
solutions of an infinite system of linear equations.]

I TENSOR PRODUCTS l
Let E and F be linear spaces, with algebraic duals E and F For each
1 1
•

x in E and y in F, define the bilinear form x ® y on E 1 x F 1 by

(x ® y)(x 1 ,y 1 ) = (x,X 1 )(y,y 1 ) .

Let k be the canonical mapping of E x F into the space B(E ,F of bi- 1 1
)

linear forms on E x F defined by k(x,y) = x 0 y, and let E 0 F, the

1 1

tensor product of E and F, be the linear subspace of B(E ,F spanned by1 1

)

k(E,F).
(a) Any element z of E ® F may be written in the form :L {x 1 ® y 1 :
x 1 E E, y 1 E F, 1 ~ i ~ n}; z is the bilinear form defined by z(x 1 ,y1 ) =
L {(xh:~~><Y~>Y'>= 1 ~ i ~ n}.
(b) The set of linear forms on E ® F is isomorphic to the set of bilinear
forms on E x F: to g E (E ® F)' corresponds the bilinear form g o k on
E x F. (To show that each bilinear form b is the image of some g, put
g(x 0 y) = b(x,y) and g(:L x1 ® y 1) = :L b(xhy1). Forthistobe meaning-
ful, it is necessary that :L x1 ® y 1 = 0 should imply :L b(xhy1) = 0. To
prove this, express each x, and y 1 in terms of linearly independent elements,
and choose X E EI and y' E Fl suitably.)
1

(c) If Gis any third linear space, the set of linear mappings of E ® F
into G is isomorphic to the set of bilinear mappings of E x F into G.
Suppose that E and F are locally convex spaces.
(d) There is one and only one locally convex topology !T for E ® F
such that, for every locally convex space G, the set of continuous linear
mappings of E ® F into G corresponds to the set of continuous bilinear
mappings of E x F imo G, that is, the isomorphism in (c) preserves
continuity. lf Oll and "Y are local bases in E and F, a local base for !T is
formed by the convex circled extensions of all the sets U 0 V = {x ® y:
x E U, y E V}, as U and V run through Oll and "f/". (Show that the topology
with this local base produces the correct continuous linear mappings; to
prove uniqueness, take G = E ® F and consider the identity mapping.)
(e) lf E and F are Hausdorff spaces, so is E ® F with the topology .r.
(lf z 0 =f 0, z 0 may be written :L {x1 ® y 1 : 1 ~ i ~ n} with the x 1 and the y 1
linearly independent. There are continuous linear forms f and g with
f(x 1 ) > 1, g(y 1 ) > 1, and f(x 1) = g(y1) = 0 for 2 ~ i ~ n. If U =
{x: if(x)i ~ 1} and V = {y: lg(y)i ~ 1}, then iz{f,g)l ~ 1 for z E U 0 V
and z 0 {f,g) > 1.)
 SEC. 17 THE WEAK TOPOLOGIES 153
The completion of E ® F with the projective tensor product topology Y is
~

denoted by E ® F. ,,
(f) lf E and F are metrizable, E ® F is a Fn!chet space. lf also E and
F are barrelled, so is E ® F. (See 12G(b).)

17 THE WEAK TOPOLOGIES

The results on pairing are applied here to the case of a space E with
locally convex topology Y and adjoint (E,Y)*. Many of the standard
theorems on weak and strong closure, weak boundedness, weak*
compactness, and continuity of functionals follow easily from earlier
work. The most profound theorem of the section gives conditions
ensuring the weak compactness of the closed convex extension of a set.
Finally, we identify the weak and weak* topologies for subspaces,
quotients, and so on.

Throughout this section E will be a linear space with a vector

topology .r, and E* (or (E,ff)*) will be the space of allff -continuous
linear functionals. This section is devoted primarily to application
of earlier results to the natural pairing of E and E*. Other applica-
tions-notably to the relationship between w(E*,E)-compact,
w(E*,E)-bounded, and ff-equicontinuous subsets of E*-are
postponed to the following section.
The weak topology w for a linear space E with vector topology .r
is the topology w( E,E*) of the natural pairing of E and E*. We
review a few of the properties of this topology. (See 16.1.) A net
{xw a E A} in E converges weakly to a point x if and only if {f(xa),
a E A} converges to f(x) for each continuous linear functionalf. The
weak topology is locally convex, and the family of all polars of finite
subsets of E* is a local base. The topology w is Hausdorff if and only
if E* distinguishes points of E, and this is always the case if the
topology .r is locally convex and Hausdorff. A function R on a
topological space to E is continuous relative to the weak topology if
and only if the composition f o R is continuous for every f in E*. A
subset A of E is weakly bounded if and only if it is weakly totally
bounded, and this is the case if and only if each member of E* is
bounded on A.
Each w(E,E*)-continuous linear functional is a member of E*,
and consequently the dass of weakly continuous linear functionals is
identical with the dass of .r-continuous linear functionals. This last
proposition may be phrased as a statement about hyperplanes as: a
154 CH. 5 DUALITY

hyperplane is :Y-closed if and only if it is weakly closed. This last

formulation is also a direct corollary to the following theorem.

17.1 WEAKLY CLOSED CONVEX SETS A convex subset of a locally

convex space (E,:Y) is :Y-closed if and only if it is weakly closed.

PROOF Let A be convex and :Y-closed, and let x be an element

which is not in A. Then there is by theorem 14.4 a :Y -continuous
linear functional J, which strongly separates A and x. Since f is
weakly continuous, no net in A can converge weakly to x; hence, A is
weakly closed. The converse is obvious since :Y is stronger than w.111

17.2 CoROLLARY The weak closure of a subset A of Eis a subset of

the :Y-closure of the convex extension of A. If E is metrizable, each
point of the weak closure of A is the :Y -Iimit of a sequence {xn}; each Xn
is of the form :L {t1y 1 : i = 1, · · ·, n}, where the t1 are positive real
numbers suchthat L {t1 : i = 1, · · ·, n} = 1, and y 1 E A.

17.3 CoROLLARY If A is a closed convex subset of E, and if f is a

linear functional on E, then f is weakly continuous on A if and only if f
is :Y-continuous on A.

PROOF It may be supposed without loss of generality that f is a real

lin-ear functional on a real space, for f is continuous if and only if its
real part is continuous. By theorem 13.5, f is w-continuous on A
if and only if f - 1 [a] n A is w-closed relative to A for all scalars a.
Since f - 1 [ a] and A both are convex, and since A is w-closed and
:Y-closed,J - 1 [a] n Ais w-closed relative to A if and only ifj - 1 [a] n
Ais w-closed; that is, if and only if f- 1 [a] n Ais :Y-closed relative
to A. Thus, f is w-continuous on A if and only if f is :Y -continuous
onA.III
It.may be noted that the proof of theorem 17.1, on weakly closed
convex sets, depends on the Hahn-Banach theorem, via the proposi-
tion on strong separation. The relationship between :Y-bounded
sets and weakly bounded sets will be established by means of a
category argument (via the absorption theorem 10.1) and a compact-
ness result which is the earliest of the compactness theorems of linear
space theory. If U is a neighborhood of 0 in a linear topological
space (E,:Y), then each linear functional which is bounded on U is
:Y-continuous and is hence a member of E*. The Smulian compact-
ness criterion 16.6 then shows that the polar of U in E* is w(E*,E)-
 SEC. 17 THE WEAK TüPOLOGIES 155

compact. This topology w(E*,E) is called the weak* topology and

is denoted w*.

17.4 BANACH-ALAOGLU THEOREM Jj U is a neighborhood oj 0 in a

linear topological space E, then the polar of U in E* is weak* compact.

17.5 WEAKLY BouNDED SETS A subset of a locally convex space

(E,.'Y) is weakly bounded if and only zf it is .'Y-bounded.
PROOF Each .'Y-bounded set is weakly bounded because .'Y is
stronger than w. Suppose that B is a weakly bounded set and that
U is a .'Y -neighborhood of 0; it can be assumed that U is convex,
circled and .'Y-closed, and hence that U is weakly closed. It follows
that U is the "polar of U 0 , where U 0 is the polar of U in E* (see
computation rules 16.3). The Banach-Alaoglu theorem 17.4 shows
that the polar U 0 is w( E* ,E )-compact. The polar of B in E* is
w(E*,E)-closed and radial at 0, and it follows from the absorption
theorem 10.1 (or its corollary 10.2) that B 0 absorbs U 0 • Finally,
taking polars in E, it is clear that U absorbs B 0 0 and therefore absorbs
the subset B of B 0 o·lll
The elementary properties of the weak* topology are similar to
those of the weak topology (see the paragraph prior to 17.1, or theorem
16.1); we shall not list these in detail, but merely note a few special
properties. The space E* is a linear subspace of the algebraic dual
E' of E, and the weak* topology is the relativized topology of point-
wise convergence. Consequently w* is always a Hausdorff, as well
as a locally convex, topology. A linear functional on E* is w*-
continuous, according to 16.2, if and only if it is represented by a
member of E. This yields the following proposition.

17.6 WEAK*-CoNTINUOUS LINEAR FuNCTIONALS A linear functional

r/> on the adjoint E* of a linear topological space E is weak* continuous if
and only if it is the evaluation at some point of E; that is, if and only zf
for some member x of E and all f in E* it is true that rf>(f) = f(x).
Weshall need, in the discussion of weak compactness, a description
of a locally convex linear topological space in terms of its adjoint. If
Eis a linear space with locally convex topology .'Y, then the family Olf
of all convex, circled .'Y -closed neighborhoods of 0 is a local base for
the topology. But each member of Olf is also weakly closed, by 17.1,
and is therefore the polar of a subset of the adjoint E*. If d is the
family of all polars of members of Olf, then .'Y can be described as
156 CH. 5 DUALITY

the topology Y:o~ of un.iform convergence on members of d, since a

local base for :Y;.1 is the family of positive scalar multiples of polars of
members of d, and this is precisely the family o/1. Finally, it is easy
to see (using theorem 8.16) that the members of d are equicontinuous
sets, and in fact d is a co-base for the family of all equicontinuous
sets. These facts and the completeness theorem 16.9 yield:

17.7 UNIFORM CoNVERGENCE ON EQmcoNTINuous SETS, AND

CoMPLETENESS Each locally convex topology is the topology of uniform
convergence on equicontinuous subsets of the adjoint.
Consequently, a locally convex linear topological space E is complete
if and only if each linear functional on E* which is weak* continuous on
every equicontinuous set is weak* continuous on E* (equivalently, if each
such linear functional is evaluation at some member of E ).

A few remarks may further darify the situation. The adjoint of a

non-complete locally convex space E and the adjoint of the completion
E 11 of E are (roughly speaking) the same, because each continuous
linear functional on E is uniformly continuous and thus has a unique
continuous extension to E 11 • But the topologies w( E* ,E) and
w( E* ,E 11 ) are surely not identical, since the dass of continuous linear
functionals consists in one case of evaluation at points of E and in the
other case of evaluations at points of E 11 • Nevertheless, the topol-
ogies w(E*,E) and w(E*,E 11 ) agree on each equicontinuous subset
of E*, for each such subset is contained in the polar of a neighborhood
of 0, this polar is w( E* ,E 11 )-compact by 17.4, and the identity map of
the polar, being a w(E*,E 11 )-w(E*,E)-continuous map of a compact
space onto a Hausdorff space, is topological.
There is a result on weak compactness which is an immediate
corollary to the preceding theorem.

17.8 CoROLLARY Each weakly compact subset of a locally convex

space is complete.

PROOF Let {xm a E A} be a Cauchy net in a weakly compact set B,

and let x be a weak duster point. Let cf> be a linear functional on E*
such that the functionals on E* corresponding to Xa converge uni-
formly on equicontinuous sets to cf>, and note that cf>(f) = f(x) for
each f in E* because x is a weak duster point of the net {xa, a E A}.
Thus the functionals corresponding to Xa converge uniformly on
equicontinuous sets to the functional corresponding to x, and in
view of 17.7, Xa converges to x.111
 SEc. 17 THE WEAK ToPOLOCIES 157
The next three propositions are essentially preparation for the proof
of the principal result on weak compactness. A non-void family d
of sets is called an algebra if the union I of memhers of d helongs to
d and, whenever A and Bare memhers of d, A U Band I ,..., A (and
hence A n B and the null set) helong to d. A real valued function
m on d is called additive if m(A U B) = m(A) + m(B) for any
disjoint memhers A and B of d. Given a countahle family of suh-
sets of I, the smallest algehra containing the family and I is countahle
(see Halmos [4] p. 23).

17.9 LEMMA Let m be a non-negative real valued additive function

defined on an algebra d, and Iet {Ak} be a sequence of members of d
such that, for some positive number e, m(Ak) > e for all k. Then there
exists a subsequence {AkJ of {Ak} such that m(n {Ak,: i = 1, · · ·, n}) is
positive for all n.

PROOF It is sufficient to prove that there is an integer s and a positive

numher d such that m(A. n Ak) ~ d for infinitely many integers k.
Application of the diagonal process, heginning with the suhsequence
of all such Ak, then yields the desired suhsequence.
Let Bn = U {Ak: k = 1, · · ·, n}; then {m(Bn)} is a hounded non-
decreasing sequence of numhers. Let a = lim m(Bn); then there is
n
a positive integer r such that a - m(Br) < ef2. Hence for any
integer k greater than r, there is an integer s suchthat 1 ~ s ~ r and
m(As n Ak) ~ ef2r. Therefore there is an s such that m(A 8 n
Ak) ~ ef2r holds for an infinite numher of k's, and the assertion of
the previous paragraph is estahlished.lll
In the following we use the notation c(A) for the numher of elements
in a finite set A.

17.10 LEMMA Let {I1} be a sequence of pairwise disjointfinite non-

void sets and Iet I= U {I1 : i = 1, 2, · · · }. If {Ak} is a sequence of
subsets of I such that,for some positive number e, lim inf c(Ak n I 1)fc(I1)
i
> e for each k; then there is a subsequence {Ak.} of {Ak} such that
n {Ak,: i = 1, 2, · · ·, n} is not void for each n.

PROOF Let Sil he the smallest algehra of sets containing the sequence
{Ak} and I. The algehra d is countahle; hence there is a sequence
{ii} of integers such that lim c(A n I 1,)fc(I11 ) exists for each A in d.
j

Denote this Iimit hy m(A). Then m is a non-negative real valued

158 CH. 5 DUALITY

additive function on d, and m(Ak) > e for each k. In view of the

previous Iemma the present one is now obvious.lll
The proof of thc next theorem for the case in which C is compact
(the only case we need later) is easy if the representation of the dual
of the space of all continuous scalar valued functions on a compact
Hausdorff space, together with Lebesgue's bounded convergence
theorem, is assumed to be known (see problern 17H). However, the
proof which is given here avoids measure theory.

17.11 THEOREM Let C be a compact (or countably compact) subset

of a linear topological space E, and Iet {/,.} be a sequence of continuous
linear functionals on E which is uniformly bounded on C. If, for each
x in C, limf,.(x) = 0, then the same equality holds for each x in the
n
closed convex extension ( C)- of C.

PROOF Without loss of generality, one can assume that the sequence
{/,.} is in the polar C 0 in E* of C. Assurne that for some x in ( C)- it
is false that limf,.(x) = 0. Then, if necessary by taking a sub-
sequence, we can " assume that, for some positive number e, IJ,.(x)l > e
for all n. F or each i, there is an x 1 in ( C) such that I/,.( x 1) I > e for
each n ~ i. It is possible to choose x1 in the form: x1 = 1fc(I1) .Z: {xa:
a E I 1}, where I 1 is a finite index set and Xa E C for each a in I 1• Let
I= U {I1 : i = 1, 2, · · ·} and Ak = {a: a EI and IA(xa)l > e/2}. If
i ~ k, then
e < 1/k(xt)l ~ 1/c(It) .Z: {lfk(xa)l: a E It}
= 1/c(It)[.Z: {IA(xa)l: a E Ak n I;}
+ .Z: {lfk(xa)l: a EI; "' Ak}]
~ 1fc(I1)[c(Ak n. I 1) + (ef2)c(I1 "' Ak)]
~ c(Ak n I 1)/c(I1) + ef2.

Hence, i ~ k implies that c(Ak n I 1)fc(I1) > ef2. In view of the

lem.ma above, there is a subsequence {AkJ of {Ak} such that, for each
n
n, the intersection {Ak1 : i = 1, · · ·, n} contains at least one element,
say a,.. Let Yn = xa.; then Yn is in C, and for each i such that i ~ n,
it holds that l/k1(y,.)l > ef2. Let Yo be a duster point in C of the
sequence {y,.}. Then, for each i, lfdYo)l ~ ef2, but this contradicts
limf,.(Yo) = 0.111
II

The next theorem is the principal result on weak compactness.

 SEC. 17 THE WEAK TOPOLOGIES 159
17.12 THEOREM ON WEAK CoMPACTNESS Let E be a Hausdorff
locally convex linear topological space, and Iet A be a bounded subset of E
such that the closure C- of the convex extension C of A is complete.
Then the following are equivalent:
(i) Each sequence in A has a weak duster point in E.
(ii) Foreach sequence {xn} in A and each equicontinuous sequence Un}
of linear functionals it is true that lim limfn(xm) = lim limfn(xm)
m n n m
whenever each of the Iimits exist.
(iii) The weak closure of A is weakly compact.
(iv) The weak closure of Cis weakly compact.

PROOF To establish that (i) implies (ii), let x 0 be a duster point in E

of {xm} and let / 0 be a weak* duster point in E* of Un}· The latter
exists because Un} being equicontinuous, is contained in a weak*
compact set (17.4). If lim limfn(xm) exists, then lim limfn(xm) =
m n m n
limf0 (xm) = f 0 (x 0 ). Similarly, if lim limfn(xm) exists, it is equal
m n m
to f 0 (x 0 ).
In order to prove (iii) from (ii), first observe that it suffices to prove
that the weak dosure A- of A in the completion E 11 is weakly com-
pact. For, the dosure C- (in E) is dosed in E 11 and hence weakly
dosed in E 11 ; therefore the weak dosures of A in E and E 11 are
identical. Consequently without a loss of generality it can be
supposed that Eis complete. Now Iet F = E* and embed Ein F'
in the canonical way; then the w(F',E*)-dosure A- of A is compact
with respect to the w(F ',E*)-topology, because A is bounded. Since
the relativization of w(F ',E*) to E is simply the weak topology, it
remains to show that A- c E. Let rp be an element of A- and let
B be an equicontinuous weak*-closed (hence weak* compact) subset of
E*. By applying theorem 8.18 to the members of A restricted toB,
one sees that the iterated limit condition of (ii) implies that rp restricted
toB is weak* continuous. Hence by 17.7, rp E E.
It is dear that (iii) irriplies (i), and that (iv) implies (iii). It remains
to demonstrate (iv), assuming the first three conditions, and this will
be done by establishing the iterated Iimit condition for an arbitrary
sequence {xm} in C and an arbitrary equicontinuous sequence Un} in
F ( = E*). The sequence Un} is contained in a weak* compact set,
and in view of theorem 8.20 there is a subsequence which converges
pointwise on the weakly compact set A- (the weak dosure of A) to
some / 0 of F. We may, therefore, assume that limfn(x) = f 0 (x) for
n
160 CH. 5 DUALITY
each x in A-. As before embed E in F and regard Un - / 0 } as a
1
,

sequence of w(F ,E*)-continuous linear functionals of F which are

1 1

uniformly bounded on A-. Since Un - / 0 } converges pointwise to zero

on A-, theorem 17.11 implies that lim <fmrf>) = (j0 ,rp) for each member
n
rp .of the w(F ,E*)-closure C- of C in F
1 Since C- is w(F ,E*)-
1
•
1

compact, there is a duster point rp 0 of {xm} in C-. Hence, if the

iterated limit lim lim fn( xm) exists, it is equal to lim <fmrf>o) =
n m n
(j0 ,rp0 ). On the other hand, lim limfn(xm) = 1imf0 (xm) = (j0 ,rp0 ),
m n m
provided that the iterated limit on the left-hand side exists. It
follows that the convex extension C of A satisfies the iterated limit
condition.!ll
W e conclude this section with a brief discuss-ion of the weak
topologies for subspaces, quotient spaces, products, and direct sums.

17.13 THE WEAK ToPOLOGY FOR SuBSPACES, QuoTIENTs, AND

PRODUCTS
(i) The weak topology for a subspace F of a locally convex space E
is the relativization of the weak topology for E.
(ii) If E is a linear topological space and F is a subspace, then the
weak topology for the quotient space E/F is the quotient topology
derived from the weak topology for E.
(iii) The weak topology for a product of linear topological spaces is the
product of the weak topologies for the coordinate spaces.
PROOF In view of the identification 14.5 of the continuous linear
functionals on a subspace F of a locally convex space E, each member
of F* is represented by a member of E* jF 0 , in the induced pairing.
Theorem 16.11 then yields proposition (i), and proposition (ii)
follows from the same two theorems in similar fashion. The result
(iii) may be derived from 14.6 by a simple argument, which we leave
to the reader.lll
The weak topology for a direct sum is not the direct sum of the
weak topologies for the factors. This may be verified easily by
consideration of the direct sum of infinitely many copies of the scalar
field.

17.14 THE WEAK* ToPOLOGY FOR ADJOINTS OF SuBSPACES, Quo-

TIENTS, AND DIRECT SuMs
(i) If Fis a closed subspace of a locally convex space E and po is its
polar in E*, then E* / F 0 , with the quotient topology derived from
 SEC. 17 PROBLEMS 161
the weak* topology for E*, is topologically isomorphic to F* with
the weak* topology.
(ii) If Fis a subspace of a linear topological space E, then the subspace
F 0 of E* with the relativized weak* topology is topologically
isomorphic to (E/F)* with the weak* topology.
(iii) The adjoint of a 'direct sum, with the weak* topology, is topo-
logically isomorphic to the product of the adjoints with the product
of the weak* topologies.
The proof of the preceding theorem is left to the reader. It may
be accomplished by identifying the adjoint via 14.5 and 14.7, then
applying 16.11 to get (i) and (ii), and making a simple direct verifica-
tion to establish (iii).

PROBLEMS
A EXERCISES
(1) An infinite dimensional pseudo-normed linear space is of the first
category in its weak topology.
(2) If U is a weak neighborhood of zero in a linear topological space E
and N = n {eU: e > 0}, then E/N is finite dimensional.
(3) The subsets of the adjoint E* of a linear topological space E which
are equicontinuous relative to the weak topology are the finite dimensional
weak* bounded subsets.

ß TOT AL SUBSETS
(a) A subset A of a linear topological space is total if and only if each
continuous linear functional vanishing on A is identically zero. If Nx =
{/:jE E*,J(x) = 0}, then A is total if and only if n {Nx: XE A} = {0}.
A subset A is total if and only if the subspace generated by A is w(E,E*)-
dense in E.
(b) A subset A of the adjoint of a Hausdorff linear topological space E is
total if and only if x = 0 whenever f(x) = 0 for all f E A. A subset A is ·
total if and only if the subspace generated by Ais w(E*,E)-dense in E*.
If there is a finite total subset of E *, then E is finite dimensional.
(c) Let E be the space of all bounded real valued functions on an
infinite set S with the topology of uniform convergence. If A is a total
subset of E, then k(A) :i;; 21ccs); there is a total subset A with k(A) =
2k(s). (k(X) denotes the cardinal of X.)

C UNIFORMLY CONVEX SPACES I (see 20L)

A normed linear space is uniformly convex iff for any e > 0 there is some
d(e) > 0 such that llx- Yll < e whenever 11-t(x + y)ll > 1 - d(e) and
llxll = IIYII = 1.
(a) In a uniformly convex normed space, a net {xa} converges to x 0 in
162 CH. 5 DUALITY

the norm topology if and only if it converges to x 0 in the weak topology

and {llxall} converges to llxoll· (First prove the result when llxall = llxoll =
1 for all a. Let f be a continuous linear functional with f(x 0 ) = 1 and
I f I = 1. Then for each e > 0, 1 - d(e) < I/(1-(xa + x0 ))1 ;;;:; 111-(xa + x0 }11
eventually.)
(b) The product of two uniformly convex spaces may be equivalently
normed to be uniformly convex.
(c) Neither uniform convexity nor the convergence property in (a) are
topological (that is, independent of the particular norm). (For uniform
convexity, the plane provides a counter-example; for the convergence
property, consider the product of a uniformly convex infinite dimensional
normed space and the real numbers.)
(d) The convergence property in (a) does not imply uniform convexity.
(Consider the space 11 of sequences x = {gn} with· llxll = .L {lgnl: n =
1, Z, · · ·} Banach [1] Ch. IX, §3, p. 140.)

D VECTOR-VALUED ANALYTIC FUNCTIONS

Let D be a domain in the complex plane and x a mapping of D into a
sequentially complete locally convex Hausdorff space E. The function x
is analytic in D iff, for each z E D, lim {[x(z + ') - x(z)]f': '-+ 0} exists.
The function x is weakly analytic in D iff it is analytic in D relative to the
weak topology, or, equivalently, iff for each f E E* the function x 1 defined
by x1(z) = f(x(z)) is an analytic function in D. Then every function
weakly analytic in D is analytic in D. (Let YW = [x(z + ') - x(z)]f'
forzE D. There is a closed circle with center z and radius Zr, say, lying
in D. Show that {yW: 1'1 ;;;:; Zr} is a bounded set, by invoking 17.5.
Deduce, using Cauchy's formula, that (yW - y(r))-+ 0 as I' - r1-+ 0,
provided that 1'1 ;;;:; r and 1''1 ;;;:; r. It follows that x'(z) = lim {yW:
'-+ 0} exists uniquely, for each z E D.)

E STONE-WEIERSTRASS THEOREM
Let X be a locally compact Hausdorff space, Iet C 0 (X) be the Banach
space of all real valued continuous functions on X vanishing at oo, and Iet
S be the unit sphere {f: II/ I ; ;:; 1}. The adjoint of C0 (X) is the set of all
finite regular Borel measures on X with total variation as the norm (141).
Lemma Let F be a subspace of C0 (X) and Iet f.L be a non-zero regular
Borel measure which is an extreme point of the weak* compact set F 0 n S 0 •
Let g be a member of C 0 (X) with the property that f fgdf.L = 0 for all f in
F. Then g is a constant function almost everywhere relative to IILI·
(By adding a constant and by multiplying by a scalar, one can assume that
g ~ 0 and f gdlf.LI = 1. lf llg I ; ;:; 1, then g is 1 a.e. If llgll > 1, put
t = llgll- 1 ; then f.L = (1- l}f.L 1 + lf.L 2 , where f.L 1 (A) = fA[(1- tg)f
(1- t)]df.Landf.L 2 (A) = fAgdf.L.)
Theorem Let F be a subspace of C0 (X) satisfying:
(a) for each x in X, there is an f in F such that f(x) # 0;
(b) for each distinct pair x andy in X, there is anfin F such thatf(x) #
f(y);
(c) Fis closed under multiplication.
 SEC. 17 PROBLEMS 163
Then Fisdensein C0 (X). (If Fis notdensein C0 (X), then by the Krein-
Milman theorem there is a measure satisfying the condition of the lemma.
Let C be the support of lt-tl. that is, C = {x: each neighborhood of x has a
positive lt-tl-measure}. Then eachfin Fis constant on C.)

F COMPLETENESS OF A DIRECT SUM

The form of Grothendieck's theorem given in 17.7 yields an alternative
proof of the completeness of a direct sum of complete spaces (14.7(iv)).
Let E = L {Et: t E A}, each Et being complete; then E* =X {Et*: t E A}.
Let 4> be a linear functional on E*, weak* continuous on every equi-
continuous set.
(a) Let lt be the injection of Et* into the product X {Et*: t E A}. Then
there is a finite subset B of A such that 4> o It = 0 for each t E A "' B.
(If (</> 0 ItJUtn) = cn -=1- 0 for n = 1, 2, ... ' consider the set {nltnUtn)fCn:
n = 1, 2, · · · }).
(b) For each jEE*, <f>(f) = _L{(<f>olt)(ft): tEB}. (If jEE*,
(</> o It)Ut) = 0 for all t E A "' B implies </>(/) = 0; use again the weak*
continuity of <f>.)
(c) There is a point x E L {Et: t E B} with </>(/) = f(x) for all jE E*.
(A finite direct sum of complete spaces is complete.)

G INDUCTIVE LIMITS 111 (see 11, 16C, 19A, 22C)

Let {(En,Yn): n = 1, 2, · · ·} be a sequence of locally convex spaces such
that whenever n ;;;; m, Ern c En and the injection (identity mapping) of
(Em,Ym) into (En,Yn) is continuous. Let E = U {En: n = 1, 2, · · ·} and
let Y be the inductive topology for E, determined by the spaces En and their
injections into E; Y is the strongest locally convex topology which makes
each injection continuous, or, equivalently, whose relativization to each
En is weaker thanYn. The space (E,Y) is the inductive limit (see 16C) of
the sequence of spaces (En,Yn)· The local bases mentioned in problern
16C(d) become especially simple in this case: the family of all convex
circled subsets U of E such that U n En is a Yn-neighborhood of 0 for
each n is a local base for Y, and the convex circled extension of all sets
U {Un: n = 1, 2, · · ·}, as Un runs through a local base for Yn, constitute
a local base for Y. Also a linear mapping of E into any locally convex space
is continuous if and only if its restriction to each (En,Y n) is continuous.
(a) Assurne also that (E,Y) is a Hausdorff space. Then (E,Y) is com-
plete if and only if for each n, the closure of En in (E,Y) is 5""-complete.
(Sketch of proof: one way is immediate. Suppose that En- is complete,
and that </> is a linear functional on E* which is weak* continuous on every
equicontinuous set. First, show that there is some n for which 4> vanishes
identically on En° = En -o. Next, let (En -)* be the adjoint of En- with
the relativization of Y. Then 4> = tjJ o Q, where Q is the quotient mapping
of E* onto (En -)* = E*/En°· With the weak* topologies, Q is con-
tinuous; if A is an equicontinuous subset of (En -)* then there is an equi-
continuous weak* compact subset B of E* such that A c Q[B]. lt
follows, from the hypothesis on </>, that tjJ is weak* continuous on Q[B].
Hence (17. 7) tjJ is weak* continuous; therefore so also is <f>.
164 CHo 5 DUALITY
(b) Assurne that, for each n, En is a closed subspace of En +1: and that ff n
is the relativization of ffn+l to Eno The space (E,ff) is then called the
strict inductive Iimit of the sequence {(En,.r n)}o Then
(i) each ffn is the relativization of ff, and En is a ff-closed subspace
of E;
(ii) if each En is a Hausdorff space, then so is E;
(iii) each bounded subset of E is contained in some Eno
(Let Un be a convex circled ffn-neighborhood of o· in En; use 13E(b) to
construct a sequence {Un+m: m = 0, 1, · · ·} of ffn+m-neighborhoods of 0,
with En+m n Un+m+l = Un+m• and put U = U {Un+m: m = 0, 1, 00 ·}o
For (ii) and (iii), use 13E(c) and similar constructionso)

H INTEGRATION PROOF OF THEOREM 17oll

Let A be a compact subset of a linear topological space Eo
(a) Letz be a point of the closed convex extension of Ao Then there is
a positive Borel measure 11- on A with 11-(A) = 1 andf(z) = fAf(x)d!1-(x) for
each continuous linear functionalf on E. (If 11- is any Bore! measure on A,
the mapping f--* f Af (x)d11-(x) is a linear functional T (!1-) on the adjoint E*
of E. Then T is a weak* continuous linear mapping of the adjoint M (A)
of C(A) (14}) into E*'. The set P of positive measures with f!(A) = 1 is
weak* compact in M(A) and so T[P] is weak* closed in E*'. Being
< >-
convex, it contains the canonical image of A 0)
(b) If Un} is a sequence of continuous linear functionals on E bounded
uniformly on A withfn(x)--* 0 on A, thenfn(x)--* 0 on <A>-. (Use (a) and
Lebesgue's theorem on bounded convergence.)

I WEAKL Y COMPACT CONVEX EXTENSIONS

The weakly closed convex extension of a weakly compact set is not in
general weakly compact: the set in 18F provides a simple counter-example.
In fact, the hypothesis of theorem 17012, that the weakly closed convex
extension be complete, in some topology giving the same adjoint, is essential
(by 17o8)o
The following example shows considerably moreo Let E be the space
of sequences x = {gn} with at most a finite number of non-zero terms, with
the norm llxll = I lgnlo Then any complete convex set in E (and so
certainly any convex weakly compact set) is finite dimensional, although
there areinfinite dimensional compact setso (Suppose that K is a complete
infinite dimensional convex seto Then there is an infinite subsequence
{n(k)} for which the set Gk = {x: x E K, gn<k> =I= 0} is not empty. Use the
convexity of K to show that each Gk is dense in K; since K is of the second
category, n {Gk: k = 1, 2, 0· ·} is not empty, and this achieves a contra-
diction. On the other hand, if Xn is the sequence with all its terms zero
except the n-th, which is 1/n, then the set consisting of the points xn
together with 0 is compact andinfinite dimensional.)

J WEAK* SEPARABILITY
Let E be a locally convex spaceo
(a) If E is separable, the equicontinuous subsets of its adjoint E* are
weak* metrizable.
 SEc. 18 ToPOLOGIES FOR E AND E* 165
(b) If Eis separable and pseudo-metrizable, then E* is weak* separable.
(c) If Eis pseudo-metrizable, the converse of (a) is true.

K HELL Y'S CHOICE PRINCIPLE

Let X be a locally compact Hausdorff space whose topology has a
countable base and Iet {p.n} be a sequence of Bore! measures on X for which
{P.n( X): n = 1, 2, · · ·} is bounded. Then there is a Bore! measure fL on
X and a subsequence {nk} with f fdp.nk ~ f fdp. for each continuous real
valued functionf on X vanishing at infinity. (Regard {p.n} as a sequence of
elements of a sphere in the adjoint of C0 (X), the normed space of con-
tinuous real valued functions on X vanishing at infinity, and use 17J(a).)

L EXISTENCE OF WEAKLY CONVERGENT SEQUENCE

Suppose that Eis metrizable and that A is a subset of E with the property
that every sequence of points of A has a weak duster point in E. Then
any point of the weak dosure of A is the weak Iimit of a sequence of points
of A. (Sketch of proof: Let {Un: n = 1, 2, · · ·} be a local base with
Un+l c Un for each n, and Iet x be a point of the weak dosure of A.
Foreach n, apply 8.21 to show the existence of a sequence in A pointwise
convergent to X on Un°. Enumerate the points of all these sequences
(triangular-wise, for example) to form a sequence {xk}· Then x isaduster
point of {xk}· The dosed subspace H spanned by {xk} is separable and
metrizable; it follows from 17J(b) that there is a countable subset D of E*
which is w(E*,H)-dense in E*. By successive refinement, there is a
subsequence {xkco} with f(xkco) ~ f(x) for all jE D. Now any weak
duster point of {xkco} must be x because of the density of D; hence xk<i> ~
x weakly.)
This theorem may be regarded as an application to a metrizable linear
topological space of a generalization of theorem 8.20 on function spaces:
suppose, in that theorem, that the functions are defined on a set S which
is the union of a sequence of compact sets and that they take their values in
a metric space Z. The proof above, which takes advantage of the proper-
ties of the weak* topology, may be carried through by using in its place
the weakest topology .r for S making the elements of the sequence con-
tinuous; any duster point of the sequence is then also continuous for .r.

18 TOPOLOGIES FOR E AND E*

There are a number of more or less reasonable ( = admissible) topol-
ogies which may be assigned to a linear space which is paired with
another linear space, and this section begins with results on complete-
ness, normability, and metrizability of these. The primary interest
is in the pairing of a linear topological space E and its adjoint E*, and
these results are applied to this case. It turns out that the dasses of
equicontinuous, strongly bounded, and weak* bounded subsets of E*
are generally distinct, but these dasses coincide in the important
special case that E is barrelled (i.e., the Banach-Steinhaus theorem
166 CH. 5 DUALITY

holds). The collection of topologies for E that yield the same adjoint
E* is explicitly described. Finally, the canonical sorts of calculation
for subspaces, quotients, and products are made.

This section is devoted to a study of certain topologies for a linear

topological space and its adjoint. Properties of the strong topology
for the adjoint E* of a locally convex space E are examined; we find
that the notions of equicontinuity, of weak* compactness, and of
strong boundedness, which are (roughly speaking) equivalent for the
adjoint of a normed space, are in general quite distinct. For a
certain dass of linear topological spaces, called barrelled spaces, these
three notions coincide. Barrelied spaces have other useful properties,
the most noteworthy being that a form of the Banach-Steinhaus
theorem holds. Finally, a complete solution is given for the problern:
what dass of locally convex topologies yields a given space of linear
functionals as adjoint? It turns out that there is a strongest and a
weakest topology in this dass, and the dass consists of all locally
convex topologies which lie between. The theorems of this section
are, for the most part, very direct consequences of earlier develop-
ments. The first two theorems in particular are a simple recasting,
in a form which is convenient for the present investigation, of some
results from Sections 8, 16, and 17.
Since several topologies for a space and its adjoint will be con-
sidered, we shall describe the dass of topologies which will be of
concern. Suppose that E and F are paired linear spaces, that d is a
family of w(F,E)-bounded subsets of F, and that ff.Ff is the topology
of uniform convergence on members of d (see Section 16). Then
the non-zero scalar multiples of finite intersections of polars of
members of d form a local base for ~1 . A topology for E is called
admissible (see Kelley [5], p. 112) (relative to the pairing) if and only
if it satisfies the three conditions: .r is a vector topology for E, (the
canonical image of) each member of Fis .r-continuous, and .r = ,r:o;
for some family d of subsets of F. The elementary facts about
admissible topologies are stated in the next theorem. Part (iii) of
this theorem has already been demonstrated (17.7), and is induded
here for reference.

18.1 ADMISSIBLE TOPOLOGIES

(i) lf E and F are paired linear spaces and d is a family of w(F,E )-

bounded subsets of F such that the linear extension of U {A :
A E d} is F, then .rd is an admissible topology for E.
 SEC. 18 TOPOLOGIES FORE AND E* 167
(ii) Let E and F be paired linear spaces. Then a topology for E is
admissible if and only if it is stronger than w(E,F) and has a
local base whose members are w(E,F)-barrels.
(iii) Ij :Y is an arbitrary locally convex topology for a linear space E,
then :Y is admissible relative to the natural pairing of E and E* ;
in fact, :Y is the topology of uniform convergence an :Y-
equicontinuous subsets of E*, or equivalently, an the family of all
polars of :Y -neighborhoods of 0.

PROOF (i) It is only necessary to establish that the canonical image

T (f) of each member f of F is ~;-continuous. But this is im-
mediate, since, if f is a member of some A in d, then the functional
T(f) is bounded on the neighborhood A 0 of 0.
(ii) Suppose :Y;.1 is an admissible topology for E. Recall from
16.1(iv) that a function R from a topological space Z to E is con-
tinuous relative to the w(E,F)-topology for E if the map z ~ (R(z),j)
is continuous on Z for each f in F. Applying this theorem to the
identity map of E, with .'Y.oi> into E, with w(E,F), shows that :Ys1 is
stronger than w(E,F). A local base for §;1 is the family of finite
intersections of non-zero scalar multiples of polars of members of d.
Since members of d are bounded, by 16.4 the polars of members of
d are w(E,F)-barrels. The local base just described therefore
consists of barrels.
Conversely, if :Y is a topology for E which has a local base @1 con-
sisting of w(E,F)-barrels, then by a Straightforward use of the com-
putation rules for polars it can be shown that :Y is the topology of
uniform convergence on polars of members of PIJ. If :Y is stronger
than w(E,F), then each member of Fis .'Y-continuous because it is
w( E,F )-continuous.lll
It is clear from the foregoing that an admissible topology :Y by no
means specifies the family d such that :Y = .'Y;1 ; many different
families may give the same topology. It is desirable to construct some
sort of normalization procedure to help in describing the situation.
Recall that a subfamily ~ of a family '1/ of sets is called a co-base for
Cf? if and only if ea.:h member of '1/ is contained in some member of
~. It is evident that if ~ is a co-base for '1/, then :YFJ = :Trc. If E
and F are paired linear spaces and @1 is a family of subsets of F, then
the family @1 is admissible if and only if the following conditions are
satisfied: each member of @1 is w(F,E)-closed w(F,E)-bounded
circled and convex, each non-zero scalar multiple of a member of @1
is a member of PIJ, the union of any two members of @1 is contained
168 CH. 5 DuALITY

in a member of f!ß, and U {B: B E f!ß} = F. Given an arbitrary

family d of w(F,E)-hounded sets which covers F, an admissihle family
may be constructed in the following fashion, let d 1 consist of all finite
unions of memhers of d, let d 2 consist of all w(F,E)-closed circled
convex extensions of memhers of d 1 , and let f!ß consist of scalar
multiples of members of d 2 • The admissible family f!ß is said to be
generated by d.
The following theorem summarizes those properties of admissihle
families which will he used in the sequel.
18.2 ADMISSIBLE FAMILIES Let E and F be paired linear spaces.
(i) If d is an admissible family of subsets of F, then Y.w is an
admissible topology for E, the family of polars of members of d
is a local base for Y.";, and d is a co-base for the jamily of
Ys~-equicontinuous sets.
(ii) If d is a family of w(F,E)-bounded subsets ofFsuch that the
linear extension of U {A: A E d} is F, and if f!ß is the ad-
missible family generated by d, then Y.w = Yga.
(iii) If d is an admissible family of subsets of F, then the topology
f/.c~ is pseudo-metrizable if and only if there is a countable
co-base for d, and Y.s1 is pseudo-normable if and only if the
family of scalar multiples of some member of d is a co-base for d.
PROOF Both (i) and (ii) result from Straightforward computations
with polars, using the results of this section and of Section 16, and
these proofs are omitted. Proposition (iii) is derived from 6.7 and
6.1, using the elementary facts on pairings and the following simple
fact: if there is a countahle local hase for a vector topology, then each
local hase contains a countahle local hase.lll
There is a noteworthy corollary to Grothendieck's completeness
theorem 16.9.
18.3 THEOREM If E and F are paired linear spaces and E is complete
relative to an admissible topology Y, then Eis complete relative to each
stronger admissible topology for E.
PROOF Each admissible topology §' is the topology .'Y;1 where d
is the family of all polars of §' -neighhorhoods of 0. If Oll is a stronger
admissihle topology and f!ß is the family of polars of 0/t-neighborhoods
of 0, then d c i?J hecause each Y -neighhorhood is a Oll -neighbor-
hood. Applying (ii) of theorem 16.9 then yields the theorem.[ll
If E and F are paired linear spaces, then there is a strongest ad-
missihle topology for E. This topology is denoted hy s(E,F). In
 SEC. 18 TüPOLOGIES FüR E AND E* 169
view of the definition of " admissible ", s( E,F) is the topology of
uniform convergence on w(F,E )-bounded subsets of F: alternatively,
s(E,F) can be described by the requirement that the family of w(E,F)-
barrels be a local base (see 18.1(ii)). The topology s(F,E) for Fis
described dually; the family of w(F,E )-barrels is a local base for
s(F,E).
If E is a space with a locally convex topology :Y and E* is the
adjoint, then the strong topology for E* is the topology of uniform
convergence on :Y-bounded sets. Since a subset of the locally
convex space E is :Y -bounded if and only if it is weakly bounded (see
17.5), the strong topology is identical with s(E*,E). It is the strongest
admissible topology for E*, and the family of weak* barreis is a base
for the strong topology. The adjoint E* is a subspace of the space Eb
of all bounded linear functionals on E (see 6.2). The term "strong"
will also be applied to s(Eb,E).

18.4 METRIZABILITY AND CoMPLETENESS OF THE AoJOINT Let E be

a locally convex linear topological space, and Iet the adjoint E* have the
strong topology. Then:
(i) the space E* is metrizable if and only if the jamily of all bounded
subsets of E has a countable co-base;
(ii) E* is dense, relative to the strong topology, in the space of all
linear functionals on E which are continuous on each bounded
subset of E; and
(iii) E* is complete relative to the strong topology zf and only if each
linear functional which is continuous on bounded subsets of E is
continuous on E.

PROOF Let E be a linear space with locally convex topology :Y, and
let d be the family of all bounded closed convex circled subsets of E.
Then d is an admissible family of subsets of E, because a convex set
is weakly closed (or weakly bounded) if and only if it is :Y-closed
(:Y -bounded, respectively). All of the assertions of the theorem are
now easy consequences of theorems 18.2, 16.9, and the fact that
continuity of a linear functional on a convex circled set is equivalent
to closure of the intersection of the set and the null space of the
functional.lll
It is worthwhile noticing that part (iii) of the foregoing theorem can
be restated: E* is strongly complete if and only if a maximal linear
subspace of E is closed whenever its intersection with every closed
bounded convex circled subset is closed (see 13.5).
170 CH. 5 DUALITY
If E is a pseudo-normed space, then the strong topology for E* is
identical with the norm topology. Moreover, in this case a subset of
E* is strongly bounded if and only if it is equicontinuous. Un-
fortunately, no such simple equivalence exists in the more general
case. The following theorem gives relations which do subsist
between strong and weak* boundedness, weak* compactness, and
equicontinuity. The proofs lean very heavily on earlier results.

18.5 BouNDEDNESS, CoMPACTNESs, AND EQUICONTINUITY Let E be

a locally convex space and Iet B be a subset of the adjoint E*.
(i) If B is equicontinuous, then the weak* closed convex circled
extension of B is both equicontinuous and weak* compact.
(ii) If Bis weak* compact and convex, then Bis strongly bounded.
(iii) If B is strongly bounded, then B is weak* bounded, and if B is
weak* bounded and E is sequentially complete, then B is strongly
bounded.
PROOF A subset of E* is equicontinuous if and only if its polar in E
is a neighborhood of 0. If Bis equicontinuous, then ß 0 °is the weak*
closed convex circled extension of B and the polar of B 0 ° is B 0 (see
the computation rules for polars 16.3). Hence ß 0 °is equicontinuous,
and ß 0 ° is weak* coP-"~pact in view of the Banach-Alaoglu theorem
17.4. This establishes (i). To establish (ii), suppose that Bis weak*
compact and convex and U is a strong neighborhood of 0 in E*.
Then one can suppose U is a (weak*) barre!, and the absorption
corollary 10.2 shows that U absorbs B. Hence Bis strongly bounded.
Finally, (iii) follows directly from the uniform boundedness theorem
12.4.111
It is noteworthy that certain of the results of the preceding theorem
cannot be improved. A subset B of E* may be weak* compact
convex and circled, and may fail to be equicontinuous ( the relation
here is clarified by the discussion of Mackey spaces later in this
section). Without the requirement of convexity, (ii) may fail: there
are weak* compact sets which are not strongly bounded (see problern
18F). Since such a set is certainly weak* bounded, the classes of
strongly and weak* bounded sets do not coincide. There are strongly
bounded weak* closed sets which are not weak* compact, so that the
converse of (ii) does not hold (problem ZOA).
If E is a pseudo-normed space, then strongly bounded sets are
equicontinuous, and the preceding theorem yields the following
corollary.
 SEc. 18 ToPOLOGIES FORE AND E* 171
18.6 CoROLLARY If E* is the adjoint of a pseudo-normed space E,
then:
(i) a subset of E* is equicontinuous if and only if it is strongly bounded;
(ii) the wea~* closed circled convex extension of each strongly bounded
subset of E* is weak* compact.
The notions of strong and weak* boundedness fail to coincide even
when E is normable. However, there is an important dass of spaces
for which the situation is very simple.
A linear topological space E is a barrelled space (or disk space,
or tonnele space) if and only if the space E is locally convex and
each barrel is a neighborhood of 0. The requirement that each
barrel be a neighborhood of 0 can be translated into a requirement
concerning the adjoint space by taking polars. Explicitly, if !Y is a
locally convex topology for E, then (E,!Y) is a barrelled space if and
only if each !Y -barrel is a !Y -neighborhood of 0, which is the case if
and only if each w(E,E*)-barrel is a !Y -neighborhood of 0, which is
true if and only if the polar of each w(E*,E)-bounded subset of E* is a
!Y -neighborhood of 0, and this condition holds if and only if each
w(E*,E)-bounded set is !Y-equicontinuous. Thus a necessary and
sufficient condition that a locally convex space be a barrelled space is
that each weak* bounded subset of the adjoint be equicontinuous.
The most important fact about barrelled spaces is that theorem
12.3, which is a form of the Banach-Steinhaus theorem, applies.
Each locally convex space which is of the second category, and in
particular each locally convex complete pseudo-metrizable space, is a
barrelled space (see the discussion preceding 12.3). It will be shown
that direct sums and products of barrelled spaces are barrelled spaces.
Since each complete locally convex Hausdorff space is topologically
isomorphic to a closed subspace of a product of complete normed
spaces and since there are examples of complete locally convex spaces
which are not barrelled (see problern 20A), a closed subspace of a
barrelled space is not necessarily barrelled. However, the comple-
tion of a barrelled space and each quotient space of a barrelled space
are of the same sort, as is proven below.
18.7 PROPERTIES OF ßARRELLED SPACES
(i) If F is a family of continuous linear functionals on a barrelled
space E to a locally convex space H, and if F is bounded relative
to the topology of pointwise convergence, then Fis equicontinuous.
(ii) A locally convex space is a barrelled space if and only if each
weak* bounded subset of the adjoint is equicontinuous.
172 CH. 5 DUALITY
(iii) The Jour following conditz'ons on a subset B of the adjoint of a
barrelled space are equivalent: B is equicontinuous, the weak*
closure of B is weak* compact, B is strongly bounded, and B is
weak* bounded.
(iv) Ij Eis a barrelled space and Fa subspace, then the quotient space
EjF is a barrelled space.
(v) If Fis a dense subspace of a linear topological space E, and if F
is a barrelled space, then Eisa barrelled space. In particular, a
completion of a barrelled space is a barrelled space.
PROOF Part (i) is theorern 12.3, and parts (ii) and (iii) have already
been established by 18.5 and the discussion preceding the theorern.
To establish (iv), let E be a barrelled space, let F be a subspace, and
let Q be the quotient rnap of E onto EjF. If A is a barrel in EjF,
then the inverse of A under Q is a barrel in E, is hence a neighborhood
of 0 in E, and its irnage in EjF is therefore a neighborhood of 0
because Q is open. Hence E/F is a barrelled space. Finally,
suppose that F is a barrelled space which is a dense subspace of E.
If A is a barrel in E, then A n F is a barrel in F and is hence a
neighborhood of 0 in F. But then the closure of A n Fis a neighbor-
hood of 0 in E, because F is dense in E, and since A contains this
closure, A is itself a neighborhood of 0 in E. Consequently, E is
a barrelled space.lll
The rernainder of this section is devoted to a discussion of the
following problern: given a linear space E and a linear space F of
linear functionals on E, for what locally convex topologies ff is it
true that F = (E,!Y)*? In other words, given E*, what are the
possible topologies for E?

18.8 MAXIMAL ToPOLOGY FüRASPACE WITH A GIVEN ADJOINT Let

E and F be paired linear spaces, and Iet !Y be a locally convex topology
for E. Then the adjoint (E,!Y)* is exactly the set of alllinear functionals
which are represented by members of F if and only if w( E,F) c !Y c ~'
where ~ is the topology of umform convergence on members of the
family ~ of all w(F,E)-compact convex circled subsets of F.
PROOF Assurne that (E,!Y)* is represented by rnernbers of F or,
equivalently, that T [F] = (E,!Y)*, where T is the canonical rnap on F
into E '. Let U be a !Y -closed convex circled neighborhood of 0;
then 17.1 irnplies that U is w(E,T[F])-closed, that is, w(E,F)-closed.
But in view of theorern 17.4, the polar U 0 in F of U is a rnernber of
~; hence U = U 0 0 is a !Y~-neighborhood of 0 and !Y c !Y~. That
 SEC. 18 TOPOLOGIES FüR E AND E* 173
w(E,F) c ff follows from the continuity of the identity map on
(E,ff) onto (E,w(E,F)) (16.1(iv)).
Suppose that w(E,F) c ff c ffw; then each member of F repre-
sents a w( E,F )-continuous (hence ff -continuous) linear functional
on E by theorem 16.2. On the other hand, if f is a ff -continuous
linear functional on E, f is bounded on some neighborhood of 0.
Since ff c ffw, one can assume that f is bounded on C0 for some C
in re'. The criterion 16.6 then implies that f is represented by a
member of F.lll
If E and F are paired linear spaces, then m(E,F), the Mackey
topology for E (also called the maximal or relatively strong
topology) has for a local base the family of all polars of w(F,E)-
compact circled convex subsets of F.' It is the topology :T.-6 of
uniform convergence on w(F,E)-compact circled convex sets. A
linear topological space (E,ff) is a Mackey space if and only if
ff = m(E,E*). Thus a Mackey space is a locally convex space such
that each weak* compact convex circled subset of E* is equicontinuous.
The foregoing theorem characterizes a Mackey space as a locally
convex space E whose topology ff is the strongest locally convex
topology yielding (E,ff)* as adjoint.
A normed space is always a Mackey space, and it will be shown
(Section 22) that each locally convex pseudo-metrizable space is a
Mackey space. It will presently be proved that the product (and
the direct sum) of Mackey spaces is again a Mackey space, and since
each locally convex space is topologically isomorphic to a subspace of
a product of pseudo-normed spaces (theorem 6.4), it follows that a
subspace of a Mackey space is generally not of the same sort. The
completion, and the image under a continuous open linear mapping,
of a Mackey space is again a Mackey space, as shown in the following.

18.9 THEOREM Let ff be a locally convex topology for a linear

space E. Then:
(i) if Eis a Mackey space and Fis a subspace, then E/F with the
quotient topology is a M ackey space;
(ii) if F is a dense subspace of E such that F with the relati~>ized
topology is a Mackey space, then E is a Mackey space-in
particular, each completion of a M ackey space is a AIackey space;
(iii) each barrelled space is a Mackey space.

PROOF (i) Let F be a subspace of a Mackey space E, and let Q be

the quotient map of E onto EfF. By theorem 17.14, the polar po
174 CH. 5 DUALITY
with the relativized topology w(E*,E) is topologically isomorphic to
the space (EJF)* with the weak* topology. If Ais a weak*-compact
convex circled subset of F 0 , the polar of A in EJF is equal to the
image Q[A 0 ] of the polar A 0 of A in E. Because Eisa Mackey space,
A 0 is a neighborhood of 0 in E; hence Q[A 0 ] is a neighborhood of 0 in
EJF since Q is open. Therefore EJF is a Mackey space.
(ii) Suppose E is a locally convex space, F is a dense subspace
which with the relativized topology is a Mackey space, and C is a
w(E*,E)-compact convex circled subset of E*. It must be shown that
C0 is a neighborhood of 0 in E, and since F is a dense subspace, it is
sufficient to show C0 n F is a neighborhood of 0 in F. Bvt if R
is the map of E* into F* which carries a linear functional f into its
restriction to F, then R is continuous with respect to the weak*
topologies, as may be seen directly. Hence R[C] is w(F*,F)-compact
convex circled, and R[C] 0 = C0 n Fis a neighborhood ·of 0 in F.
(iii) This is a direct consequence of 18.7.111
The section is concluded with a description of the strong topologies
for the adjoints of direct sums and products of linear topological
spaces, and a verification that a direct sum or a product of barrelled
spaces (or of Mackey spaces) is a space of the same sort. (It should
be pointed out that the property of being a strong adjoint is not
necessarily inherited by quotients or subspaces (see problems 22G
and 20D).)
18.10 THE STRONG ToPOLOGY FOR THE ADJOINT OF A DIRECT SUM
OR A PRODUCT Let E1, for each member t of an index set A, be a locally
convex Hausdorff space. Then (2: {E1: t E A})*, with the strong to-
pology, is topologically isomorphic to X {E1*: t E A}, where eachfactor has
the strong topology; dually, (X {E1: t E A})*, with the strong topology,
is topologically isomorphic to 2: {E1*: t E A}, where each summand has
the strong topology.
PROOF It will be convenient to treat each E1 as a subspace of 2: {E1 :
t E A}, thus omitting specific mention of the injection isomorphism
of E1 into the direct sum. The adjoint of the direct sum is, by 14.7,
canonically isomorphic to X {E1*: t E A}, and the proof of (i) depends
on identifying the family .91 of polars in X {E1*: t E A} of bounded
subsets of 2: {E1 : t E A}. According to 14.8, the family of convex
extensions of finite unions of bounded subsets of the summands E1 is
a co-base for the family of bounded subsets of a direct sum; such a
finite union may be described as U {B1 : t E A}, where each B 1 is
bounded and B 1 = {0} except forafinite number of members t of A.
 SEC. 18 TOPOLOGIES FORE AND E* 175
The computation 16.10 of the polar of such a union then shows that
each member of S/1 contains a member of S/1 which is of the form
X {Bt0 : t E A}, where Bt is a bounded subset of Et for each t and
(since Bt = {0} except for finitely many t) Bt0 = Et except for finitely
many t. In brief, S/1 is a local base for the product of the strong
topologies, and (i) is established. The proof of (ii) is omitted; it
proceeds along similar lines, via identification of the adjoint 14.6,
description of bounded subsets 14.8, and the computation on polars,
16.10.111

18.11 DIRECT SUMS AND PRODUCTS OF BARRELLED SPACES AND

MACKEY SPACES
(i) The direct sum of barrelled spaces is a barrelled space, and the
direct sum of Mackey spaces is a Mackey space.
(ii) The product of barrelled spaces is a barrelled space, and the
product of Mackey spaces is a Mackey space.

PROOF It is again convenient to treat each space Et as a subspace of

L {Et: t E A}, ignoring the injection map of Et into the sum. A
barrel in L {Et: t E A} intersects each subspace Et in a barrel in Eto and
if each summand is a barrelled space, then it follows that L {Et:
t E A} is, with the direct sum topology, a barrelled space. To
establish the fact that the direct sum of Mackey spaces is a Mackey
space, we recall (theorem 17.14) that the adjoint of L {Et: t E A},
with the weak* topology, is topologically isomorphic to X {Et*: t E A}
with the product of the weak* topologies for the factors. Each weak*
compact subset of the product is consequently contained in a product
of weak* compact subsets of the factors (explicitly, a weak* compact
subset C is contained in the product of the projections of C on the
spaces Et *), and it follows that the family of polars in L {Et: t E A} of
such products is a base for the Mackey topology for the direct sum.
Finally, the polar of a product X {Ct: t E A}, by 16.10, contains the
convex extension of the union of the polars of Ct in Eto and, if each
Et is a Mackey space, then this convex extension is a neighborhood of
0 relative to the direct sum topology. Consequently the direct sum
of Mackey spaces is a Mackey space.
The adjoint of a product P = X {Et: t E A} is canonically iso-
morphic to the direct sum S = L {Et*: t E A} according to 14.6. As
a preliminary to the proof of (ii), the weak* ( = w(S,P))-bounded
subsets of S will be identified. If wt* denotes the weak* topology for
Et*• then Et is isomorphic to the adjoint of (Et*,wt*) by 17.6, and if S
176 CH. 5 DUALITY

is given the direct sum of the topologies wt*• then P is canonically

isomorphic to the adjoint of S by 14.7. The w(S,P)-bounded sets
are therefore precisely the subsets of S which are bounded relative
to the direct sum of the topologies Wt *, by 17.5. Finally, the bounded
subsets relative to a direct sum topology are identified in 14.8 and
thus: the family of convex extensions of finite unions of weak* bounded
subsets of the summands is a co-base for the family of all weak*
bounded subsets of .L {Et*: t E A}.
The proof of (ii) is now a routine verification, using the facts noted
above and the computation 16.10 on polars; the details are omitted.jjj

PROBLEMS
A EXERCISES
(1) lf (E,F) is a pairing and .'Y an admissible topology for E, then every
w(E,F)-compact convex subset of Eis .'Y-bounded.
(2) On a normed space, the norm is weakly lower semicontinuous.
(3) The adjoint (E, .'Y)* with the strong topology is norrnable if and only
if E with ffb (see §19) is pseudo-norrnable.
(4) In the adjoint of a barrelled space, the weak* closed convex extension
of a weak* compact set is weak* compact.

ß CHARACTERIZATION OF BARRELLED SPACES

A locally convex space with topology .'Y is barrelled if and only if the only
locally convex topologies which have local bases consisting of ..;1-closed
sets are those weaker than .'Y.

C EXTENSION OF THE BANACH-STEINHAUS THEOREM (12.2)

Let E be a linear topological space with topology .'Y, such that
(P): the only vector topologies which have local bases consisting of .'Y-
closed sets are those weaker than .'Y (that is, .'Y satisfies the condition of the
previous problern with the hypothesis of local convexity removed). Then
any pointwise bounded set of continuous linear rnappings on E to any linear
topological space G is equicontinuous. (If F is the set of rnappings and "f/
a local base in Gof closed sets, then the sets n {f -l[ V]: f E F}, as V runs
through "f/, form a local base of Y-closed sets for a vector topology on E.)
When Eis of the second category, it has the property (P). There are
linear topological spaces possessing (P) but not of the second category
(e.g., an N0 -dirnensionallinear space with its strongest vector topology). A
locally convex space with the property (P) is barrelled, but not every
barrelled space possesses (P) (e.g., the space [1 12 with the relativization of
the norrn topology of P ).

D TOPOLOGIES ADMISSIBLE FOR THE SAME PAIRING

The locally convex topologies Y and 0// are adrnissible relative to the
sarne pairing if and only if there is a local base of each consisting of sets
 SEc. 18 PROBLEMS 177
closed in the other. (This, together with 7C, yields an alternative proof
of 18.3.)

E EXTENSION OF THE BANACH-ALAOGLU THEOREM (17.4)

If U is any neighborhood of 0 in a linear topological space E, then the
polar of U in E* is compact relative to the topology of uniform con-
vergence on the totally .bounded subsets of E. This is the strongest
admissible topology having this property. (Use theorem 8.17 and 16A.)

F COUNTER-EXAMPLE ON WEAK* COMPACT SETS

Let E be the set of sequences x = {gn} with at most a finite nurober of
f\On-zero terms, with the norm llxll = _Lignl· Let fn(x) = gn· Then
the set consisting of the points 22 nfn together with 0 is weak* compact but
is not strongly bounded. Thus its weak* closed convex extension is not
weak* compact. In fact, if Cn = 2 {2'/r: 1 ;;; r ;;; n}, then each Cn lies in
this extension, and {gn} is a weak* Cauchy sequence which is not weak*
convergent. The weak* topology coincides with the Mackey topology
m( E* ,E) (see 171). Thus E* with the Mackey topology is not even
sequentially complete.

G KREIN-SMULIAN THEOREM
In 17.7 it is proved that a locally convex space E is complete if and
only if every hyperplane in its adjoint is weak* closed whenever its inter-
section with every equicontinuous set Cis weak* closed in C. The Krein-
Smulian theorem asserts that the same is true with "convex set" in place
of "hyperplane" if E is metrizable (see 22.6). Hypercompleteness (see
13F) is equivalent to the same property, this time for convex circled sets:
a locally convex space E is hypercomplete if and only if every convex
circled subset in its adjoint is weak* closed whenever its intersection with
every equicontinuous set Cis weak* closed in C. (Sketch of proof: Let
o/1 be a base of convex circled neighborhoods of 0.
(a) Suppose that Eis hypercomplete and that A is a convex circled sub-
setofE*withU0 n Aweak*closedforeach UEiill. Thenet{(U 0 n A)o:
U E o/1} is Cauchy. (For if V c U and W c U,
(VO n A)o + 2U ::::J {{V 0 n A)0 + U)- = ({VO n A)o + U) 0 0 ::::J
((Von A)o U U) 0 o = ((V 0 n A) 0 °n U 0 ) 0 =(V 0 n A n U 0 ) 0 ::::J {W 0 n A) 0 ).
Hence the net, being decreasing, converges to B = n {(
U 0 n A) 0 :
u E o/1}. Then A = B 0 • ForA c B0 is immediate; if u E o/1 and r > 1,
there is V E o/1 with (V 0 n A)0 c B + (r - l)U c (B U U) 0 0 +
(r - l)(B u U)0 o = r(B u U)0 o. By taking polars, it follows tliat U 0 n
B 0 c rA and thus
U 0 n B0 c n {r(U 0 n A): r > 1} c (U 0 n A)- = uo n A c A.
(b) For the converse, suppose that {Ay: y Er} is a decreasing Cauchy
net (cf. 13F). Let B = U {A/: y Er} and Iet n- be the algebraic
closure n {rB: r > 1} of B (cf. 14F). Then B- is convex and circled,
since the net decreases, and U 0 n B- is weak* closed for each U E 0//,
178 CH. 5 DUALITY

For if .r > 1 there is some cx Er with Aa c Ay + (r - 1)U for all y.

Hence Aa c r(Ar U U) 0 0 ; it follows that (U 0 n B)- c rAa 0 , and hence
that U 0 n B- = n {r(U 0 n B): r > 1} = (U 0 n B)-' which is weak*
closed.
By hypothesis B- is weak* closed. Thus B- = B 0 °. Now there is
some y with Ay c Aß + U for all ß. It is enough to show (B 0 + Ut c
Ay0 , for then Ay c (Bo + U) 0 0 c B 0 + 2U, and {Ar: y E T} converges to
Bo = n {Ay: y E T}. If f E (Bo + U) 0 , Iet m = sup {lf(x)l: XE Bo} and
n = sup{lf(x)l: xE U}. Then m + n ~ 1; if m > 0, jEmB0 ° = mB-
and so for r > 1, f E mr A/ for some ß. Thus on Ar, lf(x)l ~ mr + n. If
m = 0, then for r > 1,/E (r- 1)B- and so jE r(r- 1)Ap0 for some ß, so
that on Ay, lf(x)l ~ r(r - 1) + 1.

H EXAMPLE AND COUNTER-EXAMPLE ON HYPERCOMPLETE SPACES

In the adjoint of a hypercomplete space, there may be a convex subset
which is not weak* closed but whose intersection with every weak* closed
equicontinuous set is weak* closed. That is, the word "circled" is
essential in the previous problern; thus hypercomplete spaces do not have
as strong a property in this respect as do complete metrizable spaces.
Let E be the algebraic dual F' of a vector space F, with the topology
w(E,F), so that E is isomorphic to a product of copies of the scalar field.
Then E is hypercomplete. (The equicontinuous subsets of E* = F are
finite dimensional; use the previous problern and 14G.)
For the counter-example, take F to be the space S (X,f.L), where X =
[0,1] and f.L is Lebesgue measure, of measurable functions (see 6L). Then
the positive cone satisfies the intersection condition, but cannot be closed
in any locally convex topology. (If it were, there would exist a positive
linear functional.)

I FULLY COMPLETE SPACES

A locally convex Hausdorff space is called fully complete if a subspace of
its adjoint is weak* closed whenever its intersection with every equi-
continuous set Cis weak* closed in C. (Cf. 17.7, 18G, 22.6.)
A locally convex complete metrizable space is fully complete (by 22.6).
The algebraic dual F' of any linear space F with the topology w(F',F) is
fully complete (HD). The adjoint F* of a Frechet (complete metrizable
locally convex) space Fis fully complete under any topology between the
topology of uniform convergence on the totally bounded subsets of F and
the Mackey topology m(F*,E). The image under an open continuous
linear mapping of a fully complete space is fully complete (cf. 11.3); so
also are closed subspaces and Hausdorff quotients of fully complete spaces.
(These last Statements, which can be proved directly from the definition
by using the weak* topologies for adjoints of subspaces and quotients
(17.13), can also be deduced from the following characteristization of fully
complete spaces.)
A hypercomplete space is fully complete, by 18G. Let {Ar: y E T} be
a net of convex circled closed non-void subsets of E. We call this a scalar
net if for each y E T and e > 0 there is some cx ~ y with Aa c e Ar. The
 SEc. 18 PROBLEMS 179
Iimit of a scalar net which converges in the Hausdorff uniformity is a sub-
space of E. A proof analogaus to that sketched in 18G shows that E is
fully complete if and only if every decreasing Cauchy scalar net converges.
If E is fully complete, the space of all closed subspaces is complete, with
the Hausdorff uniformity; the converse is not true.

J CLOSED GRAPH THEOREM 111 (see 12E, 13G, 19B)

A linear mapping of a locally convex Hausdorff space E into a fully
complete space F (see the previous problem) is continuous, provided that
(i) the graph of T in E x Fis closed;
(ii) for each neighborhood V of 0 in F, the closure of T- 1 [ V] is a
neighborhood of 0 in E.
(This is a generalization of the theorem in 13 G; the same proof holds,
because the Cauchy net used there is a decreasing scalar net.)
Full completeness is, in a certain sense, also a necessary property of F if,
in addition to the above closed graph theorem, the corresponding open
mapping theorem (cf. 11.4) is also to hold. To make this precise, Iet us
call T - 1 somewhere dense iff condition (ii) is satisfied, and also call a linear
mapping S of E into F somewhere dense iff the closure of S [ U] is a neigh-
borhood of 0 in F for each neighborhood U of 0 in E. Then, if F is a
locally convex Hausdorff space, the following statements are equivalent:
(a) Fis fully complete;
(b) any linear mapping of any locally convex Hausdorff space into a
Hausdorff quotient of F is continuous, provided that its graph is
closed and its inverse is somewhere dense;
(c) any continuous somewhere dense linear mapping of F onto any
locally convex Hausdorff space is open.
(If F is fully complete, (b) follows from the theorem above and the fact
that full completeness is inherited by Hausdorff quotients; this fact enables
(c) to be deduced from (b) in much the same way as the open mapping
theorem 11.4 is deduced from the closed graph theorem 11.1. Finally,
suppose that (c) holds and that M is a subspace of F* whose intersection
with each equicontinuous set Cis weak* closed in C. By applying (c) to
the quotient mapping of F onto F/M 0 , with the topology of uniform
convergence on the sets U 0 n M as U runs through a local base in F, it
can be shown that this topology coincides with the quotient topology; it
follows that their adjoints M and M 0 0 are equal.)
If (c) is replaced by
(c') Any continuous linear mapping of F onto any barrelled Hausdorff
space is open
then, if also Fis barrelled, (c') implies (a). (The image by a continuous
somewhere dense linear mapping of a barrelled space is barrelled; hence (c')
implies (c). )

K SPACES OF BILINEAR MAPPINGS

Let E, F, and G be locally cönvex Hausdorff spaces. Various spaces of
bilinear mappings of E x F into G can be topologized and a theory
developed similar to that of spaces of linear mappings. Let cr:l and :Jß be
 180 CH. 5 DUALITY

families of bounded subsets of E and F suchthat E = U d and F = U &1,

and H a linear space of bilinear mappings of E x F into G.
(a) If for each f EH, f[A x B] is bounded for all A E d and BE &1,
then H may be given the topology Y-;/JR of uniform convergence on the
sets A x B, as A and B run through d and !!J, making it a locally convex
linear topological space.
(b) If the sets of either s~ or !JlJ are convex, circled, and complete (in
particular if either E or Fis complete), the set of all separately continuous
bilinear mappings of E x F into G can be topologized in this way (sec 12G).
(c) If the condition in (b) holds, the space of all separately continuous
bilinear fu11ctio11als 011 E x F is ~;,$-complete if (and only if) the
adjoint of E is .:1~;-complete, and the adjoint of F is 5""319-complete. (If
{f,} is Cauchy, thcn fa(x,y)-+ f(x,y) for each x E E a11d y E F, a11d clearly
f is bilinear. lf y is fixed, the linear functional x-+ f(x,y) is the ~;-Iimit
of continuous linear fu11ctionals x-+ fa(x,y).)
(d) More generally, if also G is complete, then (c) holds with "func-
tionals 011 E x F" replaced by " mappings of E x F into G ", provided
that E and F are Mackey spaces. (Consider the convergence of c/>Ua(x,y))
for each continuous linear functional q, on G.)

L TENSOR PRODUCTS II
Let E and F be locally convex Hausdorff spaces with adjoints E* and
F*. For each x E E and y E F, the bilinear functional x ® y on E' x F'
defines by restriction abilinear functional on E* x F*, which is separately
continuous when E* and F* have their weak* topologies. On the space
of all separately continuous bilinear fu11ctionals on E* x F*, the topology
of uniform convergence on products of equicontinuous subsets of E* and
F* is a vector topology (sec (b) of the previous problem). The relative
topology for E 0 F is the topology of bi-equicontinuous convergence.
(a) The space E* 0 F* is a subspacc of the algebraic dual of E ® F
(sec 16I(b )). The topology of bi-cquicontinuous convergence is thc topol-
ogy of uniform convergence on the sets U 0 C29 V 0 , as U and V run through
local bases in E and F. With this topology, E ® Fis a Hausdorff space.
(b) The topology of bi-equicontinuous convergence is weaker than the
projective tensor product topology.
The completion of E C29 F w,i~h the topology of biequicontinuous con-
verge11ce will be denoted by E ® F.
(c) If E and F are complete, E@ Fis a subspace of the space of separately
continuous bilinear functionals on E* x F* (see (c) of the previous
problem).

19 BOUNDEDNESS
The family of all bounded sets characterizes thc topology of a normed
space, but in general the relationship between boundedness and the
topology is less intimate. Begin11ing with the family !!J of bounded
sets we may try to reconstruct the topology in either of two ways.
W e may make an external construction by considering the dass E b of
 SEc. 19 BouNDEDNESS 181
bounded linear functionals and from this obtain the Mackey topology
m(E,Eb) of the pairing, or we may consider the topology with local
base the class of all convex circled subsets of E which absorbs each
member of f-4. Both of these constructions yield the same topology
ffb, but ffb may be distinct from the original topology ff. Spaces for
which ffb = ff are called bound (or bornivore). We investigate the
relation between this notion and that of barrelled spaces, and discuss
the permanence properties of bound spaces.
If E is a linear space with pseudo-normable topology ff, then ff
determines and is determined by either the adjoint E* or by the family
fJß of all ff -bounded subsets of E. In this case each of ff, fJß, and
E* determines the other two, and this fact has important consequences.
In the more general situation the relations between ff, fJß, and E*
are less intimate, but those which do exist are still of considerable
importance. The relation between the topology .r
and the adjoint
E* has been studied at some length in the preceding section, and the
present one is devoted primarily to a discussion of the relations of
these two with fJß. Certain relations are clear. The adjoint space
E* is surely determined by .r,
and, if .r
is locally convex, E* deter-
mines the family fJß, for the ff-bounded sets are precisely the weakly
bounded sets. In attempting to reconstruct .r
from the family f!ß of
bounded sets, two approaches suggest themselves. Directly, the
family of all convex circled sets which absorb each member of f!ß
may be used as a local base for a topology, or, indirectly, the natural
pairing of E with the space E b of all bounded linear functionals may
be used to define a topology. It turns out that these two approaches
yield the same topology ffb for E, a topology which in general is
properly stronger than .r. This topology ffb is examined brießy in
the present section. A few of the results of the section are valid for
spaces which are not locally convex; however, for simplicity of
Statement, the development is confined to the locally convex case.
First, let us consider brießy the abstract process which Ieads to
the notion of bounded set. Let C(/ be the family of all non-void
convex circled subsets of a linear space E. For each subfamily .~
of C(! let ,91- be the family of all members of C(! which absorb each
member of .91, and dually, let .91 _ be the family of all members of
C(/ which are absorbed by each member of .91. It is easy to establish
the following proposition about .91 _ and .91- by Straightforward
computation.

19.1 CoMPUTATION RULES FOR BouNDEDNESS Let C(! be the class of

all convex circled subsets of a linear space E, Iet _ and - be the operations
182 CH. 5 DUALITY

defined above, and Iet s1 and f!.ß be subfamilies of ~. Then:

(i) if s1 ::::> f!.ß, then .s1 _ c f!.ß _ and .s;~- c f!.ß - ;
(ii) .s1 c .s1 _ - and .s1 c .s;~- _;
(iii) .s1 _- _ = .91 _ and .s;~- _- = .s;~-;
(iv) if each member of s1 is radial at 0, then s1 _ covers E; if .s1 covers
E, then each member of .s;~- is radial at 0;
(v) each member of ~ which is absorbed by a member of .s1 _, each
scalar multiple of a member of s1 _, and the convex extension of
the union of any two members of s1 _, belongs to s1 _;
(vi) each member of ~ which absorbs a member of .s;~-, each non-zero
scalar multiple o.f a member of .s;~-, and the intersection of any
two members of .s;~-, belang to .s;~-.

It is evident that, if .s1 is a local base for a topology !Y, then .s1 _ is
the family of convex circled !Y -bounded sets. If .s1 is a family of
subsets of E which covers E, then the foregoing rules show that .s;~­
is a local base for a locally convex topology. This topology, which
has a local base consisting of all convex circled sets which absorb each
member of d, is called the d-absorbing topology. A familiar
concept can be rephrased in terms of the d-absorbing topology.
Suppose that F is a family of linear functions, each on E to a locally
convex space H, and suppose that A is a subset of E. Then, according
to 12.1, the following conditions are equivalent: the family F is
uniformly bounded on A (that is, F[A] is bounded), Fis bounded
relative to the topology of uniform convergence on A, and, for each
neighborhood V of 0 in H the set n {.f -l[V]: f E F} absorbs A.
These equivalences yield the following result about a family of
subsets of E (the details of the verification are omitted).

19.2 LEMMA ON g;;-BOUNDEDNESS AND ON EQUICONTINUITY Let

.s1 be a family of convex circled sets which covers a linear space E and
Iet F be a family of linear functions on E to a locally convex space H.
Then the following are equivalent:
(i) the family Fis uniformly bounded on each member of d;
(ii) the family F is bounded relative to the topology 3d of uniform
convergence on members of s1; and
(iii) the family F is equicontinuous relative to the d-absorbing
topology for E.
In particular, a linear function f on E to His bounded on each member
of .s1 if and only if f is continuous relative to the d-absorbing topology.
 SEC. 19 BOUNDEDNESS 183
The foregoing discussion will now be applied to the family d of all
circled convex neighborhoods of 0 in a locally convex space (E,ff).
The family d _, consisting of all convex circled sets which are
absorbed by each member of d, is then a co-base for the family of
ff-bounded subsets of E. Hence the family d _- consists of all
convex circled subsets of E which absorb every ff-bounded set.
Members of d _ - are called bound absorbing (bornivore); ex-
plicitly, U is bound absorbing if and only if U is convex and circled
and absorbs each ff-bounded subset of E. lt follows from the
computation rules 19.1 that the family of all bound absorbing sets is
a local base for a topology for E; this topology is called the bound
extension of .r, or the bound topology derived from .r, and is
denoted by ffb. If .r and .'Yb are identical, then .r is called a
bound topology and E with the topology .r is called a bound space
(a bornivore space). In other words, .r is a bound topology if
and only if .r is locally convex and each convex circled set which
absorbs every .r-bounded set is a .r-neighborhood of 0. Several
properties of the bound extension of a topology are evident from 19 .1.

19.3 BouNn ExTENSION OF A ToPOLOGY Let ffb be tlze bound

extension of a locally convex topology .r for a linear spa~e E. Then:
(i) ffb is a bound topology, and the class :!ß of .r-bounded sets is
identical with the class of .'Yb-bounded sets; moreover, ffb is the
strongest locally convex topology relative to which each member
of &1 is bounded;
(ii) a family F of linear functions on E to a locally convex space His
ffb-equicontinuous if and only if F is bounded relative to the
topology .r~ of uniform convergence on bounded sets.
In particular, a linear function on E to His ffb-continuous if and only
if it is bounded.

PROOF Let d be the family of all convex circled .'T-neighborhoods

of 0. Then d _ is a co-base for the ff-bounded sets, d _- is a
local base for ffb, and d _- _ = d _ is a co-base for the ffb-bounded
sets (the notation, and the equality just cited, are from 19.1). Clearly
ffb is a bound topology, for the family of all convex circled sets
absorbing each ffb-bounded set is simply d _-. Moreover, from
the definition, ffb is stronger than any locally convex topology
relative to which each member of d _ is bounded. This establishes
(i) and (ii) is an immediate consequence of the preceding proposition,
19.2.111
184 CH. 5 DUALITY
The bound extension :Yb of a topology :Y for a linear space E is
defined internally, in the sense that the definition does not involve
the dual space E '. N evertheless, it is possible to give an external
description of :Yb.
19.4 ExTERNAL DESCRIPTION oF THE BouND ExTENSION OF A ToPOL-
OGY The bound extension :Yb of a locally convex topology :Y is the
Mackey topology m(E,Eb), where Eb is the space of all bounded linear
functionals on E.
Consequently, a locally convex topology :Y for Eis a bound topology
if and only if each bounded linear functional is continuous and E, with
the topology :Y, is a Mackey space.
PROOF Each w(Eb,E)-compact circled convex subset of Eb is strongly
bounded, by 18.5, and is hence :Y b-equicontinuous by 19.3(ii).
Consequently each m(E,Eb)-neighborhood of 0 is a :Yb-neighborhood
of 0, and hence :Y b is stronger than m( E,E b). But E b is the space
of all :Yb-continuous linear functionals, and by 18.8, m(E,Eb) is the
strongest locally convex topology which yields Eb as adjoint. Hence
yb = m(E,Eb)-111
The relation between the notion of barrelled space and that of a
bound space is of some interest. Recall that a barrelled space (a
locally convex space such that each barrel is a neighborhood of 0) can
be characterized as a locally convex space such that each weak*
bounded subset of the adjoint is equicontinuous. A space with a
bound topology has, in view of 19.3, the property that each strongly
bounded subset of the adjoint is equicontinuous. It is then clear
that if a space has a bound topology, and if weak* bounded sets are
strongly bounded, then the space is a barrelled space. In particular,
this is the case if the space is sequentially complete (see 18.5), and
consequently a sequentially complete bound space is a barrelled space.
These facts are noted for reference.
19.5 THEOREM Let E be a linear space with a bound topology.
Thm:
(i) each strongly bounded subsei of E*, and hence each weak* compact
conve x subsei, is equicontinuous;
(ii) ~f Eis sequentially complete, then Eisa barrelled space; and
(iii) the space E* with the strong topology is complete.
PROOF Because of the remarks preceding the theorem, only (iii)
requires a proof. However, this is immediate in view of Grothen-
dieck's completeness theorem (16.9) and the fact that each linear
 SEC. 19 BOUNDEDNESS 185
functional on E which is weakly continuous on bounded subsets of E
is bounded (hence a rnernber of E*).lll
It is not true that a subspace of a bound space is necessarily a
bo.und space (see problern 20D), nor is it true that the strong topology
for the adjoint of a bound space is necessarily bound (see problern
22G). However, there is no difficulty seeing that direct surns and
quotients of bound spaces are bound.

19.6 DIRECT SUMS AND QuoTIENTS OF ßOUND SPACES lj F is a

subspace of abound space E, then the quotient topology for EJF is bound.
The direct sum of Hausdorff spaces, each of which has abound topology,
is abound space.
PROOF If f is a bounded linear functional on EJF .and Q is the
quotient rnap of E onto EJF, thenf o Q is a bounded linear function
on E. If the topology of E is bound, then f o Q is continuous, and f
is therefore continuous on EJF. Finally, if the topology of E is
bound, then E is a Mackey space, by 19.4, and therefore EJF is a
Mackey space, by 18.9. But this fact, tagether with the fact that
each bounded linear functional is continuous, irnplies that the quotient
topology for EJF is bound. The fact that the direct surn of bound
spaces is a bound space follows directly frorn the definition of a
bound space, the definition of the direct sum topology and the
description 14.8 of the bounded subsets of a direct sum.lll
There is a simple proposition which exhibits a bound space as the
quotient of a direct sum of pseudo-norrned spaces. If A is a convex,
circled subset of a linear space E, then E A is the subspace generated by
A ( or the linear extension of A). The Minkowski functional PA of
A is a pseudo-norm for EA, and EA with this pseudo-norm is called
the pseudo-normed space generated by A. If d is a family of
convex circled. subsets of E, then the natural map of the direct sum
.2: {EA: A E d} into Eis the rnapping which sends each mernber of
the direct sum into the sum of its coordinates.
19.7 DIRECT SuM CHARACTERIZATION OF BouND SPACES Let .r be
a bound topology for a linear space E, let d be a co-base for the family of
convex circled bounded subsets of E, and for each A in d let EA be the
pseudo-normed space generated by A. Then .r is the quotient topology
derived from the natural map T of the direct sum .2: {EA~ A E d}
onto E.
PROOF It will be convenient to treat each EA as a subspace of the
direct sum .2: {EA: A E d}. If U is a convex circled .?7-neighborhood
186 CH. 5 DUALITY

of 0 in E, then for each member A of .91 there is a non-zero scalar a

such that aA c U, because .91 consists of bounded sets. Hence
T -l[U] contains a non-zero scalar multiple of A in EA for every A
in d', and therefore T-l[ U] is a neighborhood of 0 relative to the
direct sum topology. Hence :Y is weaker than the quotient topology.
On the other hand, if V is a neighborhood of 0 relative to the quotient
topology, then T -l[V] n EA is a neighborhood of 0 in the subspace
EA of the direct sum; hence T - 1 [V] contains a non-zero scalar
multiple of A, and it follows that V contains a non-zero scalar multiple
of A. Thus V absorbs bounded sets, for .91 was supposed to be a
dual base for the family of bounded sets, and consequently V is a :Y-
neighborhood of 0. The quotient topology is therefore weaker than
:Y, and the two topologies coincide.lll
In condusion, it will be shown that whether or not the product
X {E1: t E A} of bound spaces is abound space depends solely on the
cardinal number of A. Tobe explicit, let us agree that m isasimple
measure on A if m is a countably additive function, defined on the
dass of all subsets of A, which assumes only the values zero and one.
An Ulam measure is a simple measure m which is not identically zero
and has the property that m({t}) = 0 for each member t of A. It is
not known whether Ulam measures exist; Ulam has shown that if
~' is the smallest cardinal of a set with an Ulam measure (provided
there is such a set), then ~' is strongly inaccessible and 2K < ~'
whenever ~ < N'. We shall prove that the product of bound
spaces is a bound space if and only if there is no Ulam measure on the
index set.
The following lemma, which furnishes the most important part of
the proof of the result just cited, uses the following notation. If f is
a function on X {E1 : t E A} and B c A, then f*B is defined to be the
function on the product such that f*B(x) = f(K 8 · x), where K 8 · x
is the product of x and the characteristic function K 8 of B.

19.8 LEMMA Each bounded linear functional on a cartesian product

X {E 1: t E A} is the sum of a finite number of bounded linear functionals
f which have the property: there is a simple measure m on A such that
f*B = m(B)f for every subset B of A.
PROOF Let f be a bounded linear functional on X {E1 : t E A}, and
let .91 be the dass of all subsets B of A suchthat f*B is not identically
zero. Then each disjoint subfamily of .91 is finite, as the following
argument demonstrates. If {Bn} is a disjoint sequence in .91 such
that f*Bn =I= 0 for every n, then there is a sequence {xn} in the product
 SEC. 19 BOUNDEDNESS 187

such that Xn(t) = 0 if t i Bn and f(xn) # 0. Then the sequence

{nxnff(xn)} is a bounded subset of the product, since its projection on
each coordinate space is bounded (see 14.8), but f is clearly not
bounded on this sequence.
It follows from the fact that each disjoint subfamily of Sil is finite
that there is a disjoint finite cover B 1 , B 2 , • • ·, Bn of A such that for
each i and for every subset C of B;, either f*C = f*B; or f*C = 0
Letting /; = f*Bt> it is clear that f is the sum of the functionals /; and
that {f;)*C is either /; or 0 for every subset C of A. The proof of the
Iemma then reduces to showing that if g is a bounded linear func-
tional such that g*C is either g or 0 for each subset C of A, and if we
set m( C) = 1 in the first case and m( C) = 0 in the second, then m is
countably additive. Since m is evidently finitely additive, this
amounts to proving that if cn is an increasing sequence of subsets of
A such that g*Cn = 0, and if C is the union of the sets Cm then
g*C = ·o. But if g*C(x) # 0, then, letting Yn(t) be nx(t) for t in
C "' Cn and zero otherwise, we have g(yn) = g*Cn(Yn) + g*(C "'
Cn)(Yn) = g*(C "' Cn)(Yn) = ng*(C "' Cn)(x) = ng*C(x). This is a
contradiction because the sequence {Yn} is bounded (its projection on
each coordinate space isafinite set), and the functionalg was supposed
to be bounded.lll

19.9 PRODUCTS OF BouND SPACES The product X {Et: t E A} of

bound spaces Et is bound if and only if there is no Ulam measure on A.

PROOF If there is no Ulam measure on A, then each simple measure

m on A consists of unit mass situated at some point t of A-that is,
for some t and for all subsets B of A it is true that m(B) is one or zero
depending on whether t E B' or t i B. The preceding Iemma then
implies that each bounded functional is a finite sum of functionals of
the formf(x) = g(xt), where g isalinear functional on Et. It is clear
that g must be bounded and hence continuous on Et> and consequently
f is continuous on the product. Hence each bounded functional on
the product is continuous.
Conversely, suppose that m is an Ulam measure on A. If f is a
scalar valued function on A and An is the set of t in A such that
1/(t)l > n, then lim m(An) = 0, and, since the measure m assumes
n
only the values one and zero, there is a positive integer n such that
m(An) = 0. Thus each scalar valued functionf is essentially bounded
on A, and, since each such function is measurable, it follows that every
scalar valued function is integrable (in fact, each such function is
188 CH. 5 DUALITY

constant m-almost everywhere, but we shall not need this fact). For
each t in A choose / 1 in E1* such that f 1 is not identically zero, and for
x in X {E1: t E A} letg(x) = fAf1(x 1)dm(t). Theng is evidently linear,
and it may be seen that g is bounded by means of the following
argument. Each bounded subset of the product is contained in a
bounded set of the form B = X {B 1 : t E A}, where each B 1 is a
bounded subset of E1. Then sup {jg(x)J: x E B} ~JA sup {l/1(x1)j:
x 1 E B 1}dm(t), and this integral of a scalar valued function exists.
Hence g is bounded on bounded subsets of the product and it remains
to be proved that g is not continuous. But if g were continuous there
would be a finite subset C of A suchthat g(x) = 0 whenever x1 = 0
for all t in C, by 14.6. But if x is chosen so that x1 .= 0 fort in C and
.ft(x1) = 1 fort in A "' C, theng(x) = fA!t(x 1)dm(t) = m(A "' C) = 1,
and it follows that g is not continuous.JJJ

PROBLEMS
A INDUCTIVE LIMITS IV (see 11, 16C, 17G, 22C)
(a) Let E be a linear space with an inductive topology, determined by a
family {E1,ft: t E A}, each E 1 being a locally convex space. Then if each
E1 is barrelled, so is E; if each E1 is a bound space, so is E; if each E1 is a
Mackey space, so is E.
(b) The topology of a locally convex space E is bound if and only if it is
the inductive topology determined by a family of pseudo-normed spaces;
if also E is a Hausdorff space, " pseudo-normed" may be replaced by
"normed "; if in addition E is sequentially complete, "normed" may be
replaced by "Banach". (Compare 19.7, and use the same device.)
ß CLOSED GRAPH THEOREM IV (see 12E, 13G, 181)
Let E be a linear space with the inductive topology determined by a
family of locally convex spaces of the second category. Let F be a Haus-
dorff inductive limit of a sequence of fully complete spaces (see 17R, 181).
(a) A linear mapping of E into F is continuous if its graph is closed.
(Sketch of proof: by 16C(i) it is enough to prove the theorem when Eis a
locally convex space of the second category. Suppose that F is the in-
ductivelimitofthesequence{Fn}· ThenE = U{T- 1 [Fn]:n = 1,2, ···}
and so there is some n with H = T - 1 [Fn] of the second category in E,
and then H - = E. The graph of the restriction of T to H is closed in
H x Fn; it follows from 18J that T is continuous on H. There is a con-
tinuous extension S of T mapping E into Fn; the graph of S in E x Fn
is the closure of its graph in H x Fn. It follows that the graph of S is con-
tained in the graph of T, because the graph of T is closed. Thus S = T
and so T is continuous.) In particular, the theorem applies when E has
an inductive topology determined by Banach spaces, and so whenever E is
a sequentially complete bound space (see the previous problem).
(b) Assurne also that E is a Hausdorff space. Then a linear mapping
of F onto E is open if its graph is closed. (If the graph of T is closed, so is
 SEC. 20 EVALUATION MAP INTO THE SECOND ADJOINT 189
T - 1 (0); Fj T - 1 (0) is then an inductive Iimit of a sequence of fully complete
spaces, and, if T = So Q, (a) applies toS - 1 .)
C COMPLETENESS OF THE ADJOINT
The adjoint of a bound space is complete under its strong topology
(19.5 (iii)). This result may be sharpened considerably. Let s1' be an
admissible family of subsets of a bound space E suchthat every sequence in
E convergent to 0 is contained in some set of s1'. Then E* is complete
under the topology of uniform convergence on the sets of s1'. (It is
sufficient, by 16.9 and 19.4, to show that every linear functional continuous
on each A E s1' is bounded.)
Let E be abound space which is complete (or has the property that the
closed convex extension of every compact set is compact). Then E* is
complete relative to any admissible topology between the strong topology
and the topology of uniform convergence on convex circled compact sets.
In particular, E* with the Mackey topology m(E*,E) is then complete.
The example in 18F shows that it is not sufficient to assume only that E
is bound.

20 THE EVALUATION MAP INTO THE SECOND ADJOINT

If E is a normed linear space, then the natural mapping, evaluation, of
E into its second adjoint E** is an isometry. In case E maps onto
E* (we say E is reflexive) the situation of E and E* is entirely sym-
metric, and linear space methods are particularly effective.
In general, the situation is more intricate. Evaluation is always
relatively open, but may fail to be continuous (if evaluation is con-
tinuous we say E is evaluable), and even in case evaluation is dis-
continuous E may map onto E** (we say E is semi-reflexive). This
section is concerned with the exploration of this situation and its
relation to the concepts introduced earlier.
The second adjoint E** of a linear topological space Eis the space
of all strongly continuous linear functionals on E*. The evaluation
mapping of the space E into E** is defined by letting I(x), for x in E,
be the linear functional on E* whose value at a member f of E* is
f(x); that is, I(x)(f) = f(x). The principal concern of this section
is to describe conditions which imply that I is a topological iso-
morphism of E onto E**, where E** has the strong topology (the
topology of uniform convergence on strongly bounded subsets of E*).
The discussion falls naturally into two parts: we first consider condi-
tions which ensure that the evaluation maps E onto E**, and then
discuss the continuity of the evaluation mapping.
A linear topological space (E,ff) is semi-reflexive if and only if
I[ E] = E**. That is, Eis semi-reflexive if and only if each strongly
continuous linear functional on E* is the evaluation at some point of
E. The requirement that a locally convex space E be semi-reflexive
190 CH. 5 DUALITY
can be restated conveniently in terms of the natural pairing of E and
E*. The family of polars in E* of bounded subsets of E is a local
base for the strong topology for E*, and hence E is semi-reflexive if
and only if for each closed convex circled bounded subset B of E it
is true that each functional on E* which is bounded in absolute value
by one on B 0 is represented by some member of E. In view of
Smulian's compactness criterion 16.6, this is the case if and only if
each such set B is weakly compact. The following proposition
follows without difficulty (it may also be obtained from theorem 18.8).
20.1 CRITERION FOR SEMI-REFLEXIVENESS A locally convex space E
is semi-reflexive if and only if each bounded weakly closed set is weakly
compact, and this is the case if and only if m(E*,E) = s(E*,E).
It should be observed that the criterion above can be rephrased in
terms of tlie weak topology only, and consequently the notion of semi-
reflexiveness of a locally convex space (E,.?T) depends only on the
pairing of E and E*. It follows from 18.8 that if (E,.?T) is semi-
reflexive, then E with any topology which is between the weak and
the Mackey topology m(E,E*) is also semi-reflexive. In view of this
fact it is not surprising that a semi-reflexive space need not be complete,
for a space is seldom complete relative to the weak topology; however,
we will see that each closed bounded subset of a semi-reflexive space
is complete. Quotient spaces and adjoints of semi-reflexive spaces
are not, in general, of the same sort (see problems 22G and 20D);
however, products and direct sums of semi-reflexive spac.es are semi-
reflexive.
20.2 PROPERTIES OF SEMI-REFLEXIVE SPACES
(i) Each closed bounded subset of a semi-rejlexive space is complete.
(ii) lf Fis a closed subspace of a locally convex semi-rejlexive space E,
then F is semi-reflexive, and F*, with the strong topology, is to-
pologically isomorphic to E*JF 0 with the quotient topology derived
from the strong topology.
(iii) Direct sums of Hausdorff semi-reflexive spaces and products of
semi-reflexive spaces are semi-reflexive.
PROOF The first statement of the theorem is a direct corollary of
theorem 17.8 and the criterion 20.1 for semi-reflexivity. If F is a
closed subspace of a locally convex semi-reflexive space E, then Fis
weakly closed by 17.1, and the weak topology for F is the relativized
weak topology for E by 17.13. The fact that Fis semi-reflexive then
follows from the criterion 20.1. If B is a bounded subset of F, B 0 is
 SEc. 20 EvALUATION MAP INTo THE SECOND ADJOINT 191
the polar of B in F*, and f E E*, then the restriction of f to F belongs
to B 0 if and only if f belongs to the polar C of B in E. That is, the
inverse of B 0 under the canonical map T of E* onto F* is the polar
of B in E*, and T is therefore continuous relative to the strong topol-
ogies. The proof of (ii) now reduces to showing that T is open, for
continuity and openness of T implies that the induced map of E* /F 0
onto F* is topological. Let B be a bounded closed convex circled sub-
set of E; then the polar C of B n F in F* is equal to T[(B n F) 0 ], since
any continuous functional on F can be extended on E (14.1). By the
computation rule 16.3, (B n F) 0 = (B 0 u F 0 ) 0 ° c (B 0 + F 0 ) 0 °. The
set (B 0 + F 0 ) 0 ° is the m(E*,E)-closure of B 0 + F 0 which is also
the strong closure because of the semi-reflexivity of E (20.1 ). Hence
(B 0 + F 0 ) 0 ° c B 0 + F 0 + B 0 c F 0 + 2B 0 , and it follows that
C c T[F 0 + 2B0 ] = 2T[B0 ], which implies that T is open.
Part (iii) of the theorem is a direct consequence of the criterion
20.1 for semi-reflexivity, the description 14.8 of bounded subsets
of a direct sum and product, the identification 17.13 of the weak
topology of a product, and the Tychonoff theorem 4.1. We omit the
details of the verification.lll
The second half of part (ii) of the previous theorem is also a direct
consequence of theorem 21.4 of the next section and of the fact that
m(E*,E) is the same as the strong topology on E* if Eis semi-reflexive.
The traditional term for a Banachspace E for which I[E] = E** is
"reflexive", not" semi-reflexive". However, Banach spaces enjoy a
property not common to locally convex spaces in general; namely the
evaluation map is a topological isomorphism. A linear topological
space Eis called reflexive if and only if I is a topological isomorphism
of E onto E** In general, I is relatively open (that is, I [ U] is open
in I[E] whenever U is open in E), but not continuous. The follow-
ing proposition gives necessary and sufficient conditions that I be
continuous.

20.3 CoNTINUITY AND ÜPENNESS OF EvALUATION Ij E is a locally

convex space, then the evaluation map I of E into E** is relatively open;
the mapping I is continuous if and only if each strongly bounded subsei
of E* is equicontinuous, and this is the case if and only if each bound
absorbing barre! in Eis a neighborhood of 0.
PROOF Let B be an arbitrary subset of E*, let B 0 be its polar in E
and let B 0 be its polar in E**. It follows directly from the definition
of polar and of I that I[B 0 ] = I[E] n B 0 • In particular, if B is
itself the polar of a closed convex circled neighborhood U of 0 in E,
192 CH. 5 DUALITY

then B is equicontinuous and hence strongly bounded, B 0 is a neigh-

borhoodofOinE**, andi[U] = I[Bo] = I[E] n B 0 • Thereforei
is relatively open. It is also evident from the definition of polar that
I - l [ B 0 ] = B 0 , and letting B be an arbitrary strongly bounded subset
of E*, it follows that I is continuous if and only if the polar of each
strongly bounded subset of E* is a neighborhood of 0 in E. Re-
phrased, I is continuous if and only if each strongly bounded set is
equicontinuous. The last assertion of the theorem follows from the
fact that a subset of Eis the polar of a strongly bounded set if and only
if A is a bound absorbing barrel. For if A is a bound absorbing
barrel, then A absorbs each bounded set, therefore A 0 is absorbed by
each polar of a bounded set, hence A 0 is strongly bounded and
A = A 0 0 (see 16.3). Conversely, if A is the polar of a strongly
bounded s~t B, then B is absorbed by the polar of each bounded sub-
set C of E, and hence B 0 = A absorbs C 0 0 • It is then clear that A
absorbs each bounded set, and A is a barrel because it is the polar of
a weak* bounded set (see 16.4).jjj
A linear topological space E is evaluable (symmetric, infra-
tonnele, quasi-tonnele, quasi-barrelled) if and only if E is locally
convex and the evaluation map I is continuous. In view of the fore-
going theorem this is the case if and only if each bound absorbing
barrel in E is a neighborhood of 0, or equivalently, if each strongly
bounded subset of E* is equicontinuous. It is evident from the pre-
ceding that abound space, or a barrelled space, is necessarily evaluable.
The precise strength of the evaluability requirement can be visualized
as follows: Let C, CC, Y, and if' be the classes of all convex circled
subsets of E* which are, respectively, equicontinuous, with weak*
compact closure, strongly bounded, weak* bounded. Then, for an
arbitrary locally convex space E it is true that C c 'C c Y c 11".
In these terms: E is a barrelled space if and only if t: = ir; if E is a
bound space, then 6' = !/'; E is evaluable if and only if t: = !/';
and E is a Mackey space if and only if {! = Cfl. (It may be noticed
that semi-reflexivity can be characterized in terms of the analogaus
families of subsets of E if E and E* are interchanged; E is semi-
reflexive if and only if ~(E) = ii~(E).)
The following theorem summarizes properties of evaluable spaces.
The converses of the assertions of part (ii) are all false (see problems
20B and 20A).

20.4 PROPERTIES OF EVALUABLE SPACES

(i) A locally convex space E is evaluable if and only zf the e'valuation

 SEC. 20 EVALUATION MAP INTO THE SECOND ADJOINT 193
map of E into E** is continuous, and this is the case if and only if
each strongly bounded subset of E* is equicontinuous, or, equi·z:alently,
if and only if each bound absorbing barre/ in Eisa neighborhood of 0.
(ii) Each barrelled space and each bound space is evaluable, and each
evaluable space is a Mackey space.
(iii) Each quotient space of an evaluable space is evaluable.
(iv) Each sequential completion of an evaluable space is a barrelled space.
(v) Products and direct sums of evaluable spaces are evaluable.

PROOF Propositions (i) and (ii) have already been noted in the
discussion preceding the theorem. To prove (iii), notice that if Q is
the quotient map of an evaluable space E onto EfF and Disabound
absorbing barrel in EfF, then Q- 1 [D] is abound absorbing barrel in
E and is hence a neighborhood of 0. But Q is open, and consequently
Dis a neighborhood of 0 in EfF.
To prove (iv), suppose that an evaluable space F is a dense sub-
space of a sequentially complete space E, and that D is a barrel in E.
In view of theorem 10.3, D absorbs each closed convex circled
bounded subset of E, that is, D is bound absorbing. Therefore
D n F is a bound absorbing barrel in F. Since F is evaluable,
D n F is a neighborhood of 0 in F; hence D is a neighborhood of 0
in E. Therefore E is a barrelled space.
If D is a bound absorbing barrel in a direct sum 'L {E1 : t E A} of
evaluable spaces, then (considering each E1 as a subspace of the sum)
the intersection D n E1 is a bound absorbing barrel in E1 and is
consequently a neighborhood of 0 in E1• It follows that D is a
neighborhood of 0 in the direct sum, and the direct sum of evaluable
spaces is therefore evaluable.
The simplest proof that a product X {E1 : t E A} of evaluable
spaces is a space of the same sort relies on proposition 18.10, which
states that the adjoint of a direct sum (a product) is, with the strong
topology, topologically isomorphic to the product (the direct sum,
respectively) of the adjoints, where each factor is given the strong
topology. From this fact it follows that (to a topological isomorphism)
the evaluation map I of X {E1 : t E A} can be described by I(x) 1 =
I 1(x1), where I 1 is the evaluation map of E1 into E1**. If each I 1 is
continuous, then I is ah.o continuous, which shows that the product
is evaluable.lll
There is a useful corollary to part (i) of the preceding theorem.
Suppose E is a linear topological space and E*, with the strong
topology, is semi-reflexive. Then each strongly bounded closed
 194 CH. 5 DUALITY
convex circled subset of E* is weakly (w(E*,E**)) compact, and hence
weak* (that is, w(E*,E)) compact. Such a subset is consequently
equicontinuous relative to the Mackey topology m(E,E*).
20.5 CoROLLARY lf Eis a linear topological space and E*, with the
strong topology, is semi-reflexive, then E, with the Mackey topology, is
evaluable.
On the other band, it may happen that E* is evaluable and yet E,
with the Mackey topology, fails to be evaluable (see problern 20A).
That the adjoint of an evaluable space need not be evaluable is
demonstrated by an example in problern 22G.
A linear topological space E is reflexive if and only if the evaluation
map of E into E** is a topological isomorphism onto. Wehave already
derived conditions under which evaluation is continuous, and condi-
tions which ensure that evaluation carries E onto E**. These facts
(propositions 20.1, 20.4, and 20.5), and the observation that evaluation
map of a locally convex space E is one-to-one if and only if E 1s
Hausdorff, yield the following theorem.
20.6 CHARACTERIZATIONS OF REFLEXIVE SPACES A locally convex
H ausdorff space is reflexive if and only if it is semi-reflexive (bounded
weakly closed sets are weakly compact) and evaluable (strongly bounded
subsets of the adjoint are equicontinuous).
Equivalently, a locally convex Hausdorff space E is reflexive if and
only if it is a Mackey space and both E and E* are semi-reflexive.
20.7 PROPERTIES OF REFLEXIVE SPACES
(i) lf Eis reflexive so is E*.
(ii) Each reflexive space is a barrelled space, and each closed bounded
subset of a reflexive space is complete.
(iii) Products and direct sums of reflexive spaces are reflexive.
PROOF Part (i) is a direct consequence of the definition of a reflexive
space. A reflexive space is evaluable, and hence by 20.4 each bound
absorbing barrel is a neighborhood of 0. But each closed convex
circled bounded set is weakly compact (Eis semi-reflexive) and hence,
by corollary 10.2 of the absorption theorem, every barrel absorbs
such a set. Consequently each barrel is a neighborhood of 0, and it
follows that each reflexive space is a barrelled space. It has already
been observed in 20.2 that a closed bounded subset of a semi-reflexive
space is complete, and part (ii) of the theorem is thereby established.
Part (iii) is an immediate consequence of 20.2 and 20.4.111
 SEC. 20 PROBLEMS 195
Neither a closed subspace nor a quotient space of a reflexive space
is necessarily reflexive. In fact, of the two parts which make up
reflexivity-being evaluable and being semi-reflexive-one may fail
for subspaces and the other for quotients (see problern 20D). Finally,
the converse of 20.7(i) does not hold in general: it may be that the
adjoint E* is reflexivebutthat Eis not reflexive, or even semi-reflexive
(with its Mackey topology, the only one for which it is sensible to ask
the question; it is necessarily evaluable by 20.5). An example, and
a restricted form of converse, are displayed in problern 20C.

PROBLEMS
A EXAMPLE OF A NON-EVALUABLE SPACE
Let F be a non-reflexive Banachspace and Iet E = F* with the topology
m(F*,F). Then E* = F and the strong topology on E* is the norm
topology (for the w(F*,F)-bounded sets are norm bounded, by 18.5(iii)).
Thus
(a) Eis a Mackey space which is not evaluable, and so neither barrelled
nor bound;
(b) in E*, there is a convex weak* closed strongly bounded set which is
not weak* compact (cf. 18.5(ii));
(c) E* is evaluable;
(d) Eis semi-reflexive and E* is not semi-reflexive;
(e) Eis complete (see 19C).
ß AN EVALUABLE PRODUCT
Let E be a barrelled space which is not bound (see t) and F a bound
space which is not barrelled (see 18F). Then E x F is evaluable, but is
neither barrelled nor bound.

C CONVERSE OF 20. 7(i)

The result 20.7(i) states that if Eis reflexive, so is its adjoint E* under
the strong topology. The converse fails: let E be a dense proper subspace
of an infinite dimensional reflexive Banach space with the relativized norm
topology. Then E* is reflexive but Eis not semi-reflexive.
The lack of completeness is the vital point here. For suppose that Eis
a Hausdorff complete Mackey space. If E*, with the strong topology, is
semi-reflexive, then Eis reflexive. (Use the fact that Eis a closed subspace
of E**.)

D COUNTER-EXAMPLE ON QUOTIENTSAND SUBSPACES

Let {E,.r) and (F,.r 1 ) be complete linear topological spaces, so that Fis
a .r-dense proper subspace of E and .r1 is stronger than the .r-induced
t Examples of barrelled spaces that are not bound are given by Nachbin
[Proc. Nat. Acad. Sei. U.S.A., 40 (1954), 471-474] and Shirota [Proc. Japan
Acad., 30 (1954), 294-298].
196 CH. 5 DUALITY

topology on F. Let X'" denote a countable product and X<"'l a countable

direct sum of copies of X. Let G = E<wl x P', Iet T be the mapping
of G into E w defined by T (x,y) = x + y, and Iet H = T - 1 (0).
(a) The space G/H is a quotient of a complete space by a closed sub-
space, but is not complete. (Show that T is a continuous open mapping
of G onto the dense subspace E<wl + pw of Ew.)
(b) Suppose also that E and F are reflexive. Then G is reflexive, but
GfH is not semi-reflexive (for (G/H)** is isomorphic to (Ew)** = Ew).
Also G* is reflexive, but the subspace H 0 of G* is not evaluable. (For it
is semi-reflexive but cannot be reflexive, since its adjoint G/ H is not
reflexive.)
(c) The strong topology S(H 0 ,GfH) does not coincide with the topology
induced on H 0 by S(G*,G) (the adjoints being different), so that H 0 is a
subspace of a strong adjoint space which is not itself a strong adjoint space.
This also illustrates the fact that the strong adjoint of an inductive Iimit
may fail to be the projective Iimit of the corresponding strong adjoints.
(d) If E and F are Frechet spaces, then Gis a countable direct sum of
Frechet spaces, but it cannot be fully complete (sec 181), since a Hausdorff
quotient of a fully complete space is fully complete. Also G* is a direct
sum ofFrechet spaces and so is bound (sec 22.3 and 19.6) but the subspace
H 0 of G* is not bound.
The above construction may be realized by taking E = /P and F = lq
with 1 < q < p < oo.

E PROBLEM
1s every reflexive space complete?
F MONTEL SPACES
A Hausdorff barrelled space m which every closed bounded set is
compact is a Montel space.
(a) A Montel space is reflexive; its adjoint with the strong topology is
also a Montel space.
(b) A normed space can only be a Montel space if it is finite dimensional.
(c) A Montel space topology coincides with its weak topology on every
bounded set.
(d) A product or a direct sum of Montel spaces is a Montel space;
neither a closed subspace nor a Hausdorff quotient of a Montel space is
necessarily a Montel space (see 221).
(e) A strict inductive Iimit (sec 17G(b)) of a sequence of Montel spaces
is a Montel space.

G STRONGEST LOCALLY CONVEX TOPOLOGY (see 61, 12D, 14D)

Let E be a linear space of dimension a with its strongest locally convex
topology. The adjoint of Eis the algebraic dualE', and the topology of E
is the Mackey topology m(E,E '). All admissible topologies on E' coincide
(use 61); E' is the product of a copies of the scalar field, is complete and
barrelled, and is metrizable if and only if a ~ ~ 0 • Both E and E' are
reflexive Montel spaces; E is a bound space, but whether E' is always a
bound space is not known (sec 19.9).
 SEc. 20 PROBLEMS 197
H SPACES OF ANALYTIC FUNCTIONS I (see 22D, 221)
Let U he an openproper suhset of the Riemann sphere (the compactifica-
tion of the complex plane hy the one point oo ). Let A( U) he the space
of complex-valued functions, analytic on U, and vanishing at oo if oo E U,
with the topology of uniform convergence on compact suhsets of U. Then
A( U) is a complete metrizahle locally convex space which is not normahle.
If oo E U, the topology of A(U) coincides with the topology of uniform
convergence on compact suhsets of U "' { oo }. Each hounded closed
suhset of A( U) is compact; thus A( U) is a Montel space and so reflexive.
(The last part can he proved hy using Cauchy's formula for a derivative to
show that each hounded suhset of A( U) is equicontinuous at each point of
U-{oo}.)
I DISTRIBUTION SPACES II (see 8J)
(a) The space tff isaMontel space. (Sketch of proof: first, it is sufficient
to prove that any hounded suhset B of t~<m+l> is totally hounded as a suhset
of c<m>, hy 16D(j) and (k). Now this will follow, from 8J(a) and 16D(j)
again, if each DP[B], for Ip I ~ m, is a totally hounded suhset of C (Rn).
Butthis is a consequence of Ascoli's theorem, DP[ß] heing equicontinuous
because Bis a subset of c<m+l>.)
(h) The space !!) is the strict inductive Iimit (17G) of the sequence
{f!)K<r>: r = 1, 2, · · ·} (see 16C(k)); it follows in particular that!!) is com-
plete and that each hounded suhset of!!) is contained in some f!)K<r>· Each
f!)K<r> is a Montel space (this follows from (a)); therefore!!) is also a Montel
space and reflexive. Since {f!)K<r>: r = 1, 2, · · ·} is a strictly increasing
sequence of Frechet spaces, !!) is a non-metrizahle LF-space: for the
definition and consequent properties, see 22C.
(c) The continuous linear functionals on f!) are called distributions.
Every regular Bore! measure is a distrihution (emhed !!) in K (Rn): see
14J); the converse is not true. The space of distributions, with its strong
topology, is a complete non-metrizahle Montel space.
( d) The support of a distrihution T is the smallest closed suhset K of Rn
such that T(f) = 0 for all functions f whose supports do not meet K.
Such a set must always exist. The adjoint of G is the space of distrihutions
of compact support. (The adjoint of Gis a subspace of the adjoint of ~.
Suppose that T is continuous but not of compact support. Choose
inductively a sequence of compact n-dimensional cubes K, and a sequence
of fun<!tions j, of compact support S" in such a way that S, n K, = cp,
T(f,) = 1 and so that S, lies in the interior of Kr+l· This can he managed
hecause ~ is densein G. Then putf = 2: {j,: r = 1, 2, · · · }.)
J CLOSED GRAPH THEOREMFORA REFLEXIVE BANACH SPAc~
The closed graph theorem can be proved especially simply in a useful
special case, by using the fact that the unit sphere of a reflexive Banach
space is weakly compact. Let T be a linear mapping of a locally convex
space E into a reflexive Banach space F. Let V be the unit sphere in F.
If the graphofT is closed and if (T - 1 [V])- is a neighborhood of 0 in E,
then T is continuous. (If x E (T -l[V])- and '11 is a local base of convex
circled neighborhoods of 0 in E, then for each U E '1/, Au = (x + U) n
 198 CH. 5 DUALITY

T - 1[V] i= if>. It follows from the weak compactness of V that there is a

point y in V with y E ( T [A 0 ]}- for each U E if/. I t follows that, since the
graph of T is closed, y = T(x).)
The condition that (T -l[V])- be a neighborhood is satisfied \Yherever
E is barrelled, so that this proof covers a !arge number of thc cases used in
apolications.

K EVALUATION OF A NORMED SPACE

The evaluation mapping of a normed spacc onto its carionical image in
thc second adjoint is an isometric isomorphism.

L UNIFORMLY CONVEX SPACES li (see 17C)

(a) Let E bc a uniformly convex normed space. Then E is reflexive.
(Let U be the unit sphere in E; suppose that x E E**, llxll = 1, and x rf: U.
Then for some e > 0, sup {llx - Yll: y E U} > e. There is f E E* with
f(x) > 1 - d(e); thus V= {y: 11 - f(y)l < d(e)} is a w(E**,E*)-
neighborhood of x. Use the uniform convexity to show that the diameter
of U () V is less than e; thus the w(E**,E*)-closurc of U () V, to which
x belongs, has diametcr e at most.)
(b) There arereflexive normed spaces which cannot be given an equiva-
lent uniformly convex norm. (See, for example, Bourbaki [3] Ch. V,
§2, ex. 13.)
M A NEARL Y REFLEXIVE BANACH SPACE
The following example, given by R. C. James, shows that a Banach
space may be isomorphic to its second adjoint without being reflexive; in
fact here its canonical image is a subspace of co-dimension one.
Let E be the space of all sequences x = {tn} of complex numbers with
tn ~ 0 and with
llx[[ 2 = sup f:Z: {ltv<2i-1)- gp(2i)[ 2: 1 2 i 2n} + ltv<2n+1)[ 2 } < oo,
the supremum (here and later) being taken over all strictly increasing finite
sequences {p(i): 1 ;;; i;;; n} and n = 1, 2, · · ·.
(a) Let en be the sequence with n-th term 1 and all others 0. Then
IIen I = 1 and if x E E, x = L: {tnen: n = 1, 2, · · · }, the series being con-
vergent in norm.
(b) Foreach n, put fn(x) = tn· Thenfn E E* and llfnll = 1; ifjE E*,
there are scalars 1Jn with f = L: {1Jnfn: n = 1, 2, · · · }, the series being con-
vergent in norm, and for each x E E, f(x) = L: {tn1Jn: n = 1, 2, · · · }. (If
f E E*, Iet 1Jn = f(en)· If L: {1Jrfr: r = 1, 2, · · ·, n} is not convergent to j,
there is some d > 0 and a sequence of elemcnts Xn of E, the terms of the
sequence {xn} being non-zero on disjoint sets of indices, with llxnll = 1,
f(xn) > d and ,L{n- 1 xn: n = 1, 2, · · ·} convergent.)
(c) For each z E E **, Iet ~n = z(fn). Then
llzll 2 = sup {_L mp(2i-1) - ~p(2i)l 2 : 1 ~ i ~ n} + l~p(2n+l>l 2 }.
(First, prove the inequality ~ by considering linear combinations of the
elements fn of E*. Next, if Zn = L: grer: 1 ~ r ~ n} E E, there is some
jE E* with II/II = 1,j(zn) = llznll andf(er) = 0 for r > n.)
 SEc. 21 DuAL TRANSFORMATIONS 199
(d) Foreach z E E** there are scalars ~n' suchthat z = ~ 0 'e 0 + gn'en:
n = 1, 2, · · · }, where e0 is the sequence withevery term 1 and the series is
convergent in norm; the canonical image of E in E** has codimension one.
(Show that ~n--+ ~ 0 ', say, and take ~n' = ~n - ~ 0 '.)
(e) With the norm given by
izl 2 = sup {L: mp;2l-1) :- ~p;2nl 2 : 1 ;;;; i;;;; n} + l~p;2n+l)l 2 },
E** is isomorphic to E. This is equivalent to the norm of E**, given by
llzll 2 = sup {L: gp;2i-1J- ~p;2ol 2 : 1 ;;;; i;;;; n} + l~v;2n+ll + ~o'l 2 }·

21 DUAL TRANSFORMATIONS
A natural continuation of the program of describing a linear topological
space in terms of its adjoint is the description of a linear transformation
in terms of the transformation induced on the adjoint. This section
is devoted to the relationship between continuity and openness of a
mapping and of its adjoint, relative to the several possible topologies.
The results for Frechet spaces are especially sharp.
This section is devoted to the study of linear transformations by
means of the fundamental duality. Each continuous linear trans-
formation T of a linear topological space E into a linear topological
space F induces, in a natural way, an adjoint transformation T* of
F* into E*. The properties of T are intimately related to those of
T*, and, in particular, it is possible to characterize continuity and
openness of T in terms of T*.
In order to preserve symmetry as far as possible, we begin with a
discussion of paired linear spaces. Let E and F, and G and H, be
paired linear spaces, let T be a linear transformation of E into G and
let T' be a linear transformation of H into F. The transformations
T and T' are dual if and only if <T(x), h) = (x,T'(h)) for all
members x of E and all members h of H. If T is dual to T ', and if
S and S 1 are duallinear transformations of E into G and H into F,
then aS + bT is dual to aS + bT for all scalars a and b. More-
1 1

over, if U is a linear transformation of G into a linear space /, and U 1

is a dual mapping of a space J which is paired to I, then T o U is 1 1

dual to U o T.
The existence of a transformation which is dual to a given linear
transformation T can be described in terms of the continuity properties
ofT.

21.1 ExiSTENCE AND UNIQUENESS OF DUAL MAPPINGS Let E and F,

and G and H, be paired linear spaces, and Iet T be a linear transformation
of E into G. Then T has a dual T if and only if ,T is w(E,F)-w(G,H)
1
 200 CH. 5 DUALITY
continuous, and this is the case if and only if the linear lunctional x-+
(T(x), h) is w(E,F)-continuous loreachhin H.
Il T has a dual T', then (since T' has a dual), the translormation T'
is necessarily w(H,G)-w(F,E) continuous. The dual T' ol T is unique
if and only if E distinguishes members ol F.
PROOF If T is w(E,F)-w(G,H) continuous, then for each I in H the
map x-+ (T(x),l) is a w(E,F)-continuous linear functional on E;
hence, by the representation theorem 16.2 for w(E,F)-continuous
linear functionals, there is an elementf* ofFsuchthat (T(x),l) =
(x,j *) for all x in E. Moreover, the element I* is unique if and only
if E distinguishes points of F. A dual T' of T may then be con-
structed as follows: for each member I of a Harne! base B for H Iet
T'(f) be an element I* ofFsuchthat (T(x),j) = (x,l*) for all x
in E, and Iet T' be the linear transformation of H into F which has
the values so determined at the members of B. It follows without
difficulty that T' is a dual for T, and that T' is unique if and only if
E distinguishes points of F.
On the other hand, if T has a dual T ', then for each h in H, the
linear functional x-+ (T(x),h) = (x,T'(h)) is w(E,F)-continuous on
E, and if each such linear functional is w(E,F)-continuous then, by
16.1, T is w(E,F)-w(G,H) continuous on E.lll

21.2 CoMPUTATIONS WITH DuAL MAPPINGS Let E and F, and G

and H, be paired linear space_s, and Iet T be a linear translormation ol E
into G with dual T '. Then:
(i) lor each subset A ol E, T[A] 0 = (T')- 1 [A 0 ] ;
(ii) zf E distinguishes points ol F, then the polar ol the range T [E] ol
T is the ,null space ol T ';
(iii) if Cis a subset ol T[E], then T'[C 0 ] = T'[H] n T - 1 [C] 0 ; and
(iv) the translormation T' is relatively open, continuous, and maps
closed sets onto relatively closed sets in T'[H] with respect to the
topologies w(H,T[E]) and w(F,E).

PROOF ClearlyiE T[A] 0 ifandonlyifi(T(x) ,J)I = l<x,T'(/))1 ~ 1

for all x in A, and this is the case if and only if T '(/) E A 0 • This
establishes (i), and part (ii) is an immediate corollary. To prove (iii),
assume that Cis a subset of T[E] and observe that C = T[T -l[C]].
Hence, applying (i), ca = (T')- 1 [T - 1 [C] 0 ], and therefore T'[C 0 ] =
T - 1 [C] 0 n T'[H]. This establishes (iii). To prove (iv), it is
necessary only to notice that a net T'(ha) in T'[H] converges to T'(h)
if and only if ha converges to h relative to w(H, T [E]).lll
 SEc. 21 DuAL TRANSFORMATIONS 201
We now prove the principal theorem of the section, which gives
necessary and sufficient conditions for a linear transformation T be
continuous or relatively open with respect to admissible topologies.
(Recall that an admissible topology for E, where E is paired to F, is
a topology ff"" of uniform convergence on members of d, where d
is a family of w(F,E)-bounded, closed, convex circled subsets of F
such that scalar multiples of members of d belang to d, the union
of any two members is contained in a member, and d covers F.
Every locally convex topology is admissible relative to the natural
pairing of the space and its adjoint.)

21.3 CoNTINUITY AND ÜPENNEss RELATIVE To AnMISSIBLE ToPoL-

OGIES Let E and F, and G and H, be paired linear spaces, Iet d and
f!J be admissible families of subsets of Fand H, respectively, and Iet T be
a linear transformation of E into G with dual T'. Then:
(i) the transformation T is :Y;;-~ continuous if and only if for
each B in f!J there is a member A of d such that A contains
T'[B];
(ii) if T is :Yd-~ relatively open, then for each member A of d
there is a member B of f!J suchthat A n T'[H]- c T'[B]-,
where closures arerelative to the topology w(F,E); ij, in addition
E distinguishes points of Fand if each member of fJß is w(H,G)-
compact (equivalently, ~ is weaker than the Mackey topology
m(G,H)), then T'[H] is w(F,E)-closed; and
(iii) if H distinguishes points of G, if each member of d is w(F,E)-
compact (equivalently, ffd is weaker than the Mackey topology
m(E,F)), and if for A in d there is B in f!J such that A n
T'[H]- c T'[B]-, then T is ffsi-fff!l relatively open.

PROOF In view of the definition of :Y:_o; and fff!l, the transformation

T is :Y;;-ffiJJ continuous if and only if for each B in f!J there is A in
d such that A 0 c T- 1 [B0 ]. According to 21.2(i), T- 1 [B 0 ] =
T'[B] 0 , and A 0 c T'[B] 0 if and only if T'[B] 0 ° = T'[B]- c A 0 ° =
A, by taking polars. Since A is w(F,E)-closed, we have that T is
:Y;rffiJJ continuous if and only if for each B in fJß there is A in d
suchthat T'[B] c A, and (i) is proved.
To prove (ii) and (iii), we first note that T is :Y;rfff!l relatively
open if and only if for A in d there is B in !Je such that T [ A 0 ] ::::>
B 0 n T [E]. But, taking inverses und er T, this is true if and only if
Ao + T - 1 [01 => T - 1 [B0 ].
Suppose T is relatively open; then, since T'[H] 0 ::::> T - 1 [0], for
202 CH. 5 DUALITY

each A in .s1 there is Bin~ suchthat A 0 + T'[H] 0 :::> T'[B] 0 , which

implies that (A 0 + T'[H] 0 ) 0 c T'[B] 0 ° = T'[B]-. The first half
of (ii) now follows, because A n T'[H]- c (Ao + T'[H] 0 ) 0 •
Under the additional hypotheses of (ii), w(F,E) is Hausdorff and, for
each B in !!1, T'[B] is w(F,E)-compact, since T' is w (H,G)-w(F,E)
continuous (21.1); hence T'[B] is w(F,E)-closed for each B in !!1.
Using the result of the first half, we see that, for each A in .s1, A n
T'[H]- c T'[H], and, since .s1 covers F, T'[H]- c T'[H].
To establish (iii), first observe that for each member A of .s1 we
have 2A 0 + T'[H] 0 :::> A 0 + <T'[H] 0 U A 0 ) , and the latter set con-
tains the ffJ01'-closure of <T'[H] 0 U A 0 ) . The topology §"d is,
under the hypothesis of the theorem, stronger than w(E,F) and
weaker than the Mackey topology m(E,F); hence the members of F
represent all of (E,ffJOt)*, by 18.8. It follows from 17.1 that the ffd-
closure of <T'[H] 0 U A 0 ) is w(E,F)-closed. Hence 2A 0 +
T'[H] 0 => <T'[H]o u A 0 ) - = (T'[H]- n A)o for each A in .s1.
Finally, if B is a member of !!I such that A n T'[H]- c T'[B]-,
then (A n T ' [H] - )0 :::> T ' [ B] 0 , and hence for each A in .s1 there is
B in !!I suchthat 2A 0 + T'[H] 0 :::> T'[B] 0 • Since H distinguishes
members of G, T'[H] 0 = T - 1 [0], and the inclusion above can be
written as 2A 0 + T - 1 [0] :::> T - 1 [B0 ]. It follows, in view of the
criterion of the second paragraph, that T is ffJOI'-ff91 relatively open.lll
There are three especially important topologies which are defined
by a pairing. Recall that, if E and F are paired linear spaces, the
Mackey topology m(E,F) for E has a local base consisting of the
polars in E of w(F,E)-closed compact convex circled subsets of F, and
the strong topology s(F,E) for F has a local base consisting of all
polars of w(E,F)-bounded subsets of E. The following proposition
is an application of the preceding theorem to these topologies. (In
the statement of the result the term " weak " refers to the topologies
w(E,F) and w(F,E).)

21.4 CONTINUITY AND ÜPENNESS OF TRANSFORMATIONS AND THEIR

DUALS Let E and F, and G and H, be paired linear spaces, and let T
be a linear transformation of E into G. Then:
(i) T is weakly continuous if and only if T is Mackey continuous;
(ii) in case E distinguishes points of Fand H distinguishes points of G,
T is weakly continuous and weakly relatively open if and only if
T has a dual T' and the range of T' is weakly closed;
(iii) in case E distinguishes points of F and the pairing <G,H) is
 SEC. 21 DUAL TRANSFORMATIONS 203
separated, T is weakly continuous and open if and only if T' is a
weakly topological isomorphism of H onto a weakly closed
subspace of F;
(iv) in case E distinguishes points of F and the pairing (G,H) is
separated, T is Mackey continuous and open if and only if T
is weakly continuous and open; and
(v) If T' is a dual of T, then T' is strongly continuous.

PROOF If T is weakly continuous, then there is a dual T' of T, T' is

automatically weakly continuous, and the image under T' of a
weakly compact subset of H is therefore weakly compact. There-
fore, by the criterion 21.3(i), T is Mackey continuous. Conversely,
if T is Mackey continuous, then for each h in H the functional x -'?-
(T(x),h) is Mackey continuous, and is hence represented by a
member of F (theorem 18.8). Therefore x--+ (T(x),h) is weakly
continuous for every h in H, and, by theorem 21.1, T is weakly con-
tinuous.
To prove part (ii), first assume that T' exists and the range of T' is
weakly closed. In view of 21.3(iii), T is weakly relatively open if for
each finite subset A of F there is a finite subset B of T ' [H] such that
A 0 ° n T'[H] c B 0 °, and the latter is easily seentobe the case (see
the proof of 16.11(ii)). The existence of T' implies the weak con-
tinuity of T (21.1 ). Conversely, if T is weakly continuous and
relatively open, then 21.3(ii) implies that the range of T' is weakly
closed.
If T is weakly continuous and open, then T[E] = G, and since
T [E] is weakly closed, T' is relatively open. The null space of T'
is T [ E] 0 = {0}, by 21.2, and hence T' is a weakly topological iso-
morphism. Conversely, if T' is a weakly topological isomorphism
onto a weakly closed subspace of F, then T' is weakly continuous and
relatively open, and T [ E] is therefore weakly closed. Then T [ E] is
identical with the polar of the null space of T ', by 21.2, and hence
T [ E] = G. Thus T is weakly continuous and open, and (iii) is
established.
If T is weakly continuous and open, then T' is a topological iso-
morphism onto a weakly closed subspace of F, and the criterion
21.3(iii) for relative openness shows that T is Mackey relativdy open
and (since T [ E] = G) hence open. Conversely, if T is Mackey
continuous and open, then by 21.3(ii) the range T'[H] is weakly
closed. It follows from part (iii) that T is weakly open, and (iv) is
established.
204 CH. 5 DuALITY

Proposition (v) is a direct application of 21.3(i) to the strong

topologies of H and F.lll
We now apply the foregoing theorems to the study of a linear
transformation T of a linear topological space E into a linear topo-
logical space F with the natural pairings (E,E*) and (F,F*). Since
E always distinguishes elements of E*, there is at most one transforma-
tion which is dual to T. This dual transformation, if it exists, is
called the adjoint ofT and is denoted by T*. The condition for dual
transformations becomes, in this case, h(T(x)) = T*(h)(x) for all x in
E and all h in F. Rewriting, this is: T*(h) = h o T for all hin F.
The uniqueness of T* is self-evident in this formulation.
The following results are immediate consequences of 21.1. We
recall that the weak topology for a linear topological space E is the
topology w(E,E*), and the weak* topology for E* is the topology
w(E*,E).
21.5 ExiSTENCE, UNIQUENEss, AND WEAK* CoNTINUITY OF An-
JOINTs; ÜNE-TO-ÜNE TRANSFORMATIONS Let E and F be locally
convex, linear topological spaces, and Iet T be a linear Iransformation of
E into F.
(i) lf T is continuous, then T is weakly continuous, and T has an
adjoint T* if and only if T is weakly continuous. The adjoint,
if it exists, is unique and is necessarily weak* continuous.
(ii) A linear transformation S of F* into E* is the adjoint of some
linear Iransformation T of E into F zf and only if S is weak*
continuous. The transformation T, if it exists, is weakly
continuous, and is unique if Fis Hausdorff.
(iii) If T has an adjoint, T*, then the polar of T[E] is the null space
of T*, and, if in addition Fis Hausdorff, the polar of T*[F*] is
the null space of T. H ence T* is one-to-one if and only if T [E]
is weakly densein F, and, if E and F are Hausdorff, T is one-to-
one if and only if T*[F*] is weak* densein E*.
Proposition 21.4 yields the following results on linear transforma-
tions of a locally convex space.
21.6 CONTINUITY AND ÜPENNESS OF TRANSFORMATIONS AND THEIR
ADJOINTS Let T be a continuous linear transformation of a locally
convex space E into a locally convex space F, and Iet T* be the adjoint
transformation of F* into E*. Then:
(i) T is continuous relative to the Mackey topologies, and T* is
strongly continuous;
(ii) if F is Hausdorff, T is weakly relatively open if and only if
 SEc. 21 DuAL TRANSFORMATIONS 205
T* [F*] is weak* closed, and T* is weak* relatively open ij and
only ij T[ E] is closed; and
(iii) ij Fis Hausdorff, T is open relative to the Mackey topologies ij
and only ij T is weakly open.
If T is a continuous linear transformation on a locally convex space
E into a locally convex Hausdorff space F and if T is relatively open,
then theorem 21.3(ii) implies that the range of T* is weak* closed,
and hence, using part (ii) of the theorem above, T is weakly relatively
open. Therefore we have the following:

21.7 CoROLLARY Let T be a continuous linear transjormation on a

locally convex space into a locally convex Hausdorff space. Then, ij T
is relatively open, T is weakly relatively open, and, ij T is open, T is
weakly open.
The adjoint T* of a linear transformation T of E into F is always
strongly continuous, in view of theorem 21.6, and consequently the
second adjoint T** (the adjoint of T*) exists. This second adjoint
is an extension of T, in a sense made precise by the following simple
proposition.

21.8 EvALUATION AND THE SECOND ADJOINT OF A TRANSFORMATION

Let T be a continuous linear transjormation oj a locally convex space E
into a locally convex space F, and Iet I E and I F be the evaluation mappings
of E into E** and F into F**, respectively. Then T** o JE = IF o T.
PROOF Using the definitions of adjoint and evaluation, we have, for
each x in E and each f in F*,
T** o IE(x)(f) = T**(IE(x))(f) = IE(x) o T*(f)
= JE(x)(fo T) =Jo T(x) = ]Fo T(x)(f).lll
The results concerning open mappings can be very much improved
in case the domain and range spaces are Frechet spaces (locally con-
vex, metrizable, and complete). We give the results here, assuming
for the moment that it is known that each metrizable space is a
Mackey space (this is proved in the next section, and the proof given
there is independent of the following theorem).
21.9 ÜPEN MAPPINGS OF FRECHET SPACES Let T be a continuous
linear transformation of a Frechet space E into a Frechet space F.
Then the jollowing are equivalent.
(i) The map T is relatively open.
(ii) The map T is weakly relatively open.
206 CH. 5 DuALITY

(iii) The range of T is closed.

(iv) The adjoint map T* is weak* relatively open.
(v) The range of T* is weak* closed.
PROOF The equivalences (i)---?- (ii) ~ (v) and (iii) ~ (iv) result
immediately from 21.6, 21.7, and the fact that each metrizable space
is a Mackey space. That (i) ---7- (iii) follows from the fact that the
image under a continuous open linear transformation of a complete
metric space is complete, (11.3), and that (iii) ---7- (i) follows from the
open mapping theorem 11.4. It remains to show that (ii) ---7- (i).
Assurne (ii) and let T[E] = G; then the weak topology w(G,G*) is
the relativization of w(F,F*) (theorem 17.13 (i)). Therefore, T is a
weakly open map of E onto G, and, in view of 21.6(iii), T is open
relative to the Mackey topologies of E and G. Finally, since both E
and G are metrizable, T is a relatively open map of E into F.lll
Cert~ün of the implications of the above theorem can obviously be
strengthened. However, the theorem is, in substance, the best of
the known results of its type on openness of mappings.

PROBLEMS
A COMPLETEL Y CONTINUOUS MAPPINGS
Let E and F be Hausdorff linear topological spaces and let T be a linear
mapping of E into F. Then T is called completely continuous iff there is
some neighborhood U of zero which is mapped by T into a compact set.
For Banach spaces, this coincides with the usual definition of a completely
continuous mapping as one which maps bounded sets into totally bounded
sets (cf. SB and 21D). A completely continuous mapping is continuous.
Let I be a topological isomorphism of E onto F and let S = I + T. Then
(a) the null space of S is finite dimensional;
(b) S is an open mapping of E onto S[E];
(c) the range of S is closed.
(Sketch of proof: let N be the null space S - 1 [0]. For (a), show that
N n U is totally bounded and use 7.8. For (b ), suppose that S is not open,
so that there is a neighborhood U1 of zero, which may be taken tobe a sub-
set of U, with S [U1 ] not a neighborhood of zeroinS [E]. If "Y' is a base
of circled neighborhoods of zero in F, for each V E "f/' there is a point y in
V with y rt S[ U1 ] and then there is some ,\ with 0 ;;;;; ,\ ;;;;; 1 so that ,\y E V n
S[2UI] but ,\y rt S[U1]. Thus for each V there is a point Xv E 2U1 but
Xv i ul + N with s (xv) E V. Now use the properties of T and I to show
that a subnet converges to a point of N, thus contradicting Xv rt U1 + N.
For (c), suppose that {S(xa): a E A} converges to y E F. It follows from
(b) that S (xa) E S [ Xp + U] for some ß and all a ;:;;: ß, and hence that
(xp + U) n S -l[y + V] f= ~ for each V E "f/'. If zv is a point of this
set, then S(zv) ~ y and, by an argument similar tothat for (b), a subnet of
{ Zv: V E "Y'} is convergent.)
 SEC. 21 PROBLEMS 207
ß RIESZ THEORY
The theorem of the previous problern forms the foundation for the Riesz
theory of completely continuous linear mappings of a linear topological
space into itself. The following- sequence of results gives an outline of
the theory.
Let E be a linear topological space and T a completely continuous linear
mapping of E into itself. Suppose throughout that Ais a non-zero scalar,
and write AI - T = S.
(a) Lemma: Let G be a closed proper subspace of a subspace F of E
suchthat S[F] c G. If U is a convex circled neighborhood of 0 mapped
by T into a compact set, then there is a point x E F n (ZU) but x rf= G + U.
(b) The sequence {S - r[O]} of subspaces increases; there is an integer
n ~ OsuchthatS-r[O] c s-<r+l>[O]properlyforO ~ r < nands-r[O] =
s-<r+l>[O] for r ~ n. (Suppose not, and apply (a) with G = S -r[O] and
F = S -<r+l>[O]. Use the sequence {xr} thus provided to contradict the
fact that T[2 U] is totally bounded.)
(c) The sequence {Sr[EJ} decreases; there is an integer m ~ 0 suchthat
sr[E] ::::> sr+ 1 [E] properly for 0 ~ r < m and sr[E] = sr+ 1 [E] for
r ~ m. (Proof similar to that of (b); each Sr[E] is closed by (c) of the
previous prob lern.)
(d) The integers n and m of (b) and (c) are equal, and Eis the direct
sum of S -n[O] and sn[E].
By (a) and (c) of 21A and by 7E, Eis then isomorphic to the product of
S -n[O] and sn[E] (in fact, it is their topological direct sum when it is
locally convex).
It follows that the existence of a completely continuous endomorphism
on a linear topological space which is not locally convex ensures the ex-
istence of continuous linear functionals, if S - 1 [0] =1= {0} (that is, if .\ is an
eigenvalue ofT): in fact the linear functionals in the subspace (Sn[E]) 0 of
the algebraic dual E' are continuous, and this subspace has the dimension
of S -n[O].
(e) The following are equivalent: (i) Ais not an eigenvalue ofT; (ii) the
range of AI - T is the whole of E; (iii) >.I - T is an isomorphism. (The
integer n above is zero: use (d) and 21A(b ).)
(f) E is isomorphic to the product of two closed subspaces M and N, N
being finite dimensional; S = AI - T maps each of M and N into itself
and is an isomorphism on M. For all positive integers r, dim S -r[O] =
dim EfSr[E]. (Use 21A(b) applied to the restriction of S to M; for
the last part, observe that since N is finite dimensional, dim S -r[O] =
dim NfSr[N].)
(g) The eigenvalues of T form a finite set or a sequence convergent to
zero. (If e > 0, there are at most a finite number of eigenvalues with
l.\1 ~ e. Suppose not; choose a sequence {xn} of linearly independent
eigenvectors, and Iet Hn be the subspace spanned by x1 , · · ·, Xn· Apply
(a) with G = Hn_ 1 , F = Hn and S = Anl- T, An being the eigenvalue
corresponding to Xn, and so again contradict the fact that T[U] is totally
bounded.)
208 CH. 5 DUALITY

C COMPLETE CONTINUITY OF THE ADJOINT

Let E and F be locally convex Hausdorff spaces and T a completely
continuous linear mapping of E into F. Then T* is completely continuous
if E* and F* have the topologies of uniform convergence on the convex
circled compact subsets of E and F, respectively. (Use 18E.)
Let E and F coincide and S = AI - T, as in the previous problem.
Then T and T* have the same eigenvalues and the spaces S -r[O] and
S*-r[O] have the same dimension, for each r. (Use the results of the
previous problern applied to T*, with the aid of the relations (Sr[E])0 =
S*-r[O] and (S*r[E*])0 = S -r[O].)

D SCHAUDER'S THEOREM
Let E and F be locally convex Hausdorff spaces and T a weakly con-
tinuous linear mapping of E into F. Let E* have the strong topology.
Then T maps bounded sets into totally bounded sets if and only if T*
maps equicontinuous sets into totally bounded sets. (Use 16A applied to
<F,F*), with .91 the family of images of bounded sets and PA the family
of equicontinuous sets, and use the identity <T(x), y*) = <x,T*(y*)).)
The linear mapping T is called totally bounded if it maps bounded sets
into totally bounded sets. If, in addition to the above hypotheses, F is
~valuable and F* has the strong topology, then T is totally bounded if and
only if T* is totally bounded. When E and F are Banach spaces, this
becomes Schauder's theorem: T is completely continuous if and only if T*
is completely continuous.

E CLOSABLE MAPPINGS
Let E and F be locally convex Hausdorff spaces. This problern is
concerned with linear mappings which are defined only on subspaces of E
and take values in F. Let T be a linear mapping with domain D c: E
and range in F; T is called closed if its graph G is a closed subspace of
E x F, and closable if it has some closed extension. This is the case if and
only if G- is the graph of a closed linear mapping which has the smallest
domain of all the closed extensions of T. The mapping T is closed if
and only if y = T(x) whenever there is a net {xy: y E T} in D with xy-+ x
and T(xy)-+ y; it is closable if and only if y = 0 whenever there is a net
{xy: y E T} in D with Xy-+ 0 and T(xy)-+ y.
Let E* and F* be the adjoints of E and F. The mapping x-+ <T(x),g)
defines a linear functional on D. Let D* be the subspace of F* consisting
of those elements g for which this linear functional is continuous on D.
This sets up a mapping of D* into E*JD 0 • Now if (and only if) Dis dense
in E, there is a unique continuous linear functional f on E with f(x) =
<T(x),g) for all x E D; in this case we may write f = T*(g), and T* is a
linear mapping of D* into E*. Suppose then that Dis densein E.
(a) With the weak* topologies on the adjoint spaces, T* is closed. In
fact, the graph of - T* in F* x E* is the polar of the graph of T.
(b) The mapping T is closable if and only if D* is densein F*. (For
if D* is densein F*, the process can be repeated to give a closed extension
T**. Conversely, if T is closable, Iet S be the closed extensionofT with
 SEc. 22 PsEuno-METRIZABLE SPACES 209
smallest domain, and let o/1 be a local base of convex circled neighborhoods
of 0 in this domain. Then D* 0 = (T*- 1 [E*]) 0 = (S*- 1 [E*]) 0 =
n
<U {S*- 1 [U 0 ]: u E o//})0 = <U {(S[U])o: u E o//})0 = {S[U]: u E o/1},
and 11 H Ieads to the required results.)

F STONE-CECH COMPACTIFICATION
Let X be a completely regular space and B(X) the set of bounded
continuous real-valued functions on X with the norm II/ I = sup {1/(x)l:
XEX}.
(a) The mapping e of X into the adjoint M(X) of B(X) with the weak*
topology, defined by (e(x))(f) = f(x), is a homeomorphism of X onto a
subset of the surface of the unit sphere of M(X). Let ß(X) = e[X]-;
ß(X) is a compact Hausdorff space, the Stone-Cech compactification of X.
(b) Every f E B(X) has a unique continuous extension to ß(X) (suppose
e[X] and X identified).
(c) If t is a continuous mapping of X into any compact Hausdorff space
Y, then t has a unique continuous extension F mapping ß(X) into Y.
(Let T be the linear mapping of B( Y) into B(X) induced by t; obtain F by
restricting T* to ß(X).)
(d) Let H be the subset of M(X) consisting of homomorphisms: that is,
H = {4>: 4> E M(X), 4>(fg)=4>(f)4>(g)for allf,g E B(X)}. Then His weak*
closed and H"' {0} is a subset of the surface of the unit sphere. (Consider
the mappings 4>-+ 4>(/g) - 4>(f)4>(g).)
(e) H = ß(X) U {0}. (Clearly e[X] c H. Suppose that X is com-
pact, and show that if 4> EH, but 4> 'f. 0, then 4> E e[X]. For this, show
that there is some x for which 4>(/) = 0 implies f(x) = 0. For if not, a
function g can be constructed (as a finite sum of squares) so that 4>(g) = 0
but g(x) > 0 for all x EX. Thus g- 1 E B(X) and it follows that 4> = 0.
For general X, use (b).)

22 PSEUDO-METRIZABLE SPACES
This section is concerned with special properties of pseudo-metrizable
spaces. Such spaces are always bound, and both the strong and the
weak* topologies for the adjoint have localizability properties. How-
ever, the adjoint may still have a rather intricate structure, and we are
able to discover very little about the structure of the second adjoint.
This section is devoted to a number of results which are peculiar
to locally convex pseudo-metrizable spaces. The topology of such a
space E is always bound, and the weak* topology for E* has a note-
worthy "localization" property. The strong topology for E* may,
however, exhibit pathological features, and we are able to establish
relatively little concerning its structure.
Webegin with a boundedness property of E.
210 CH. 5 DUALITY
22.1 SEQUENTIAL ABSORPTION THEOREM lf Eisa pseudo-metrizable
linear topological space and A is a subset of E which absorbs each sequence
converging to 0, then A is a neighborhood of 0.
PROOF Let {Un} be a local base for the topology of E, and assurne
that A is not a neighborhood of 0. Then, for each positive integer
n, there is a point Xn in (~ un) ~ A. Evidently the sequence {nxn}
converges to 0, but nxn rf= nA. It follows that A does not absorb the
sequence {nxn}, and hence each set which absorbs every sequence
which converges to 0 is necessarily a neighborhood of 0.[[[

22.2 CoROLLARY A family F of linear transformations of a pseudo-

metrizable linear topological space E into a linear topological space H is
equicontinuous if and only if F is uniformly bounded on each sequence
in E which converges to 0.
PROOF Each equicontinuous farnily is uniforrnly bounded on every
bounded set (see 12.1 ), and in particular on each sequence which
converges to 0. Conversely, Iet U be a circled neighborhood of 0 in
Hand Iet A = n {f- 1 [UJ:/E F}. Then, the assurnption that Fis
uniforrnly bounded on sequences converging to 0 irnplies that A
absorbs each such sequence, hence A is a neighborhood of 0, and Fis
therefore equicontinuous.[[[
Theorem 22.1 irnplies that every pseudo-rnetrizable locally convex
space is bound. Proposition 19.5 then yields the properties of
pseudo-rnetrizable locally convex spaces which are listed in the
following corollary.

22.3 CoROLLARY lf E is a locally convex pseudo-metrizable space,

then the topology of Eis bound and hence also Mackey.
Consequently, for such a space E, the adjoint E* is strongly complete,
and the following conditions on weak* closed convex circled subsets A of
E* are equivalent: A is equicontinuous, A is weak* compact, and A is
strongly bounded.
The next results concern the weak* topology for the adjoint of a
pseudo-rnetrizable linear topological space E. W e are prirnarily
concerned with "localizing" the weak* topology, in the sense that
we infer (in certain cases) that a set is weak* closed if and only if its
intersection with each equicontinuous subset A of E* is weak* closed
in A. The usefulness of the result:; lies in the fact that, in proving
that a particular set B is weak* closed, we rnay restriet our attention
 SEc. 22 PsEuoo-METRIZABLE SPACES 211
to an equicontinuous (and hence to a weak* compact) subset of E*.
One theorem of this general type has already been proved: theorem
17.7 implies that a hyperplane in the adjoint of a complete locally
convex space is weak* closed if and only if its intersection with each
equicontinuous subset A of E* is weak* closed in A.
The next Iemma is the critical step in the identification of the
strongest topology which agrees with the weak* topology on equi-
continuous subsets of E*.

22.4 LEMMA Let E be a pseudo-metrizable linear topological space,

and Iet W be a subset of E* such that 0 E Wand U 0 n W is weak*
open in U 0 for each neighborhood U ofO in E. Then there is a countable
set S in E such that S 0 c Wand the points of S can be arranged to form
a sequence converging to 0.
PROOF Let {Un: n = 1, 2, .. ·}, with ul = E, be a decreasing
sequence of sets forming a local base for E, and suppose that W
satisfies the hypothesis of the Iemma. The proof of the Iemma in-
volves an induction, which in turn depends on the following pre-
lemma: if U and V are neighborhoods of 0 in E, and A is a subset of
E suchthat A 0 n U 0 c W, then there isafinite subset B of U such
that (A u B) 0 n V 0 c W. This pre-lemma is established as
follows. The set U 0 is the intersection of the sets {x}0 , for x in U,
by the definition of polar. Since A 0 n uo c W, it follows that
vo n A 0 n uo = n {V 0 n A 0 n {x}0 : XE U} c vo n w. The
set V 0 is weak* compact because V is a neighborhood of 0, and
vo n W is weak* open in vo by hypothesis. We then have an
intersection of closed subsets of a compact space contained in the
open subset V 0 n W, and by a well known compactness argument,
some finite intersection is a subset of V 0 n W. That is, there is a
finite subset B of u such that n {V 0 n A 0 n {x}0 : XE B} = vo n
A 0 n B 0 c vo n w c w, and the pre-lemma is established.
To prove the Iemma, let A 1 be a finite subset of E = U1 such that
A 1 ° n U 2° c W. Such A 1 exists because W n U 2° is weak* open
in U 2°and contains 0. Then the pre-lemma yields a finite subset A 2
of U 2 such that (A 1 u A 2 ) 0 n U 3 ° c W. Repeating this process
inductively, one can choose a sequence {An} of finite sets such that
An c Un and (A 1 u A 2 U · · · u An_ 1 ) 0 n Uno c W. Let S =
U {An: n = 1, 2, · · ·}; then clearly S is countable and can be arranged
to form a sequence converging to 0. To show that so c W, first
observe that so n Uno c W for each n. Hence so = U {So n
Un°: n = 1, 2, · · ·} c W.JJJ
212 CH. 5 DuALITY
If E is a pseudo-metrizable linear topological space, .'T8 is defined
to be the locally convex topology for E* which has for a local base
the set of all polars S 0 , where S is the set of values of a sequence
which converges to 0. This topology will be described, somewhat
loosely, as the topology of uniform convergence on sequences
converging to 0. The topology might also be described as the
topology of uniform convergence on convergent sequences, for
the uniform convergence on each sequence converging to 0 implies the
uniform convergence on each convergent sequence. It is clear that
the topology .'T8 is stronger than the weak* topology and weaker than
that of uniform convergence on compact sets, or totally bounded
sets. Among other facts, the next theorem states that the locally
convex topology .'T8 is the strongest topology which agrees with the
weak* topology on equicontinuous sets. It is not hard to see that
this strongest topology which agrees with the weak* topology on equi-
continuous sets consists of all sets U such that U n A is weak*
open in A for each equicontinuous subset A of E*.

22.5 THE LOCALIZED WEAK* ToPOLOGY Let E be a pseudo-metrizable

linear topological space. Then the following Jour topologies for E*
coincide.
(i) the topology .'T8 of uniform convergence on sequences converging to 0,
(ii) the topology of uniform convergence on compact sets,
(iii) the topology of uniform convergence on totally bounded sets, and
(iv) the strongest topology which agrees with the weak* topology on
equicontinuous sets.
PROOF Clearly the topology (ii) is stronger than the topology (i);
the topology (iii) is stronger than the topology (ii). The topology of
pointwise convergence for an equicontinuous set is identical with that
of uniform convergence on totally bounded sets (theorem 8.17), and
hence the topology (iv) is stronger than the topology (iii). It remains
to show that the topology (i) is stronger than the topology (iv). Let
U be an open set relative to the topology (iv), and Iet / 0 E U. Since,
for each equicontinuous subset A of E*, it is true that U n A is
weak* open in A, the same is true for U - f 0 , that is, ( U - / 0 ) n A
is weak* open in A whenever Ais an equicontinuous subset. Hence
by Iemma 22.4, U - fo contains a .'T8 neighborhood of 0; that is, U
is a .'T8 neighborhood of f 0 • Therefore U is a .'T8 -open set.lll
22.6 KREIN-SMULIAN THEOREM ON WEAK* CLOSED CONVEX SETS
A locally convex pseudo-metrizable space E is complete if and only if each
 SEC. 22 PSEUDO-METRIZABLE SPACES 213
convex subset of E* is weak* closed whenever its intersection with every
equicontinuous subset C of E* is weak* closed in C.
PROOF Assurne that E is complete. Then since the closed convex
circled extension of each totally bounded subset of Eis compact and
hence weakly compact, the topology .'Ts is weaker than the Mackey
topology m(E*,E) and stronger than the weak* topology w(E*,E).
Hence by 18.8, E represents (E*,.'T 8 )*, and by 17.1, each convex sub-
set of E* which is .'T8 -closed is weak* closed; that is, if A is a convex
subset of E* such that A n C is weak* closed in C whenever Cis an
equicontinuous subset, then A is weak* closed. The converse is
obvious from theorem 17.7 ·I II
The foregoing theorem can be strengthened somewhat. If E is
locally convex and pseudo-metrizable and E 11 is a completion of E,
then (E 11 )* is identical with (more precisely, isomorphic to) E*.
Each equicontinuous subset A of E* is contained in a w( E* ,E 11 )-
compact subset B, and since w(E*,E) is a weaker Hausdorff topology
for B, the relativization toB of w(E*,E 11 ) and w(E*,E) must coincide.
This remark, in conjunction with the last theorem, yields the following
proposition.

22.7 CoROLLARY Let E be a locally convex pseudo-metrizable space,

and Iet E 11 be a completion of E. Then a convex subset A of E* inter-
sects each equicontinuous set B of E* in a set which is w(E*,E)-closed in
B ij and only if Ais w(E*,E 11 )-closed.
The rest of this section is devoted to a study of the strong topology
for E*, and to a few propositions about the second adjoint. We
begin with a discussion of metrizability.

22.8 METRIZABILITY OF E* Let E be a locally convex pseudo-

metrizable space which is not pseudo-normable, and Iet E* have the
strong topology. Then each bounded subset of E* is nowhere dense, E* is
of the first category in itself, and E* is not metrizable.
PROOF Suppose that B is a strongly bounded, somwhere dense sub-
set of E*. Then B is equicontinuous, and hence there is a neighbor-
hood U of 0 in E suchthat the polar U 0 contains B. The set U 0 is
weak* closed, hence strongly closed, and 0 belongs to the (strong)
interior of U 0 because this interior is convex, circled, and non-void.
Consequently there is a bounded subset A of E such that A 0 c U 0 ,
and hence U c U 0 0 c A 0 0 • The set A 0 0 is bounded because E is
locally convex, U 0 0 is therefore a bounded convex neighborhood of 0,
214 CH. 5 DUALITY
and theorem 6.1 shows that E is pseudo-normable. This contradic-
tion establishes the first of the conclusions of the theorem. If { U n} is
a local base for E, then E* is the union of the sets Uno, each of these
is nowhere dense, and hence E* is of the first category. Finally, E*
is (strongly) complete, and if E* were pseudo-metrizable, then the
Baire theorem 9.4 would imply that E* be of the second category.lll
The following theorem is fundamental for the rest of this section.

22.9 CouNTABLE INTERSECTION OF STRONG NEIGHBORHooos Let E

be a locally convex pseudo-metrizable space, and Iet {Un} be a sequence
of convex circled strong neighborhoods of 0 in E*. Then the intersection
n {Un: n = 1, 2, · · ·} is a strong neighborhood of 0 if and only if this
intersection absorbs each strongly bounded set.
PROOF Let U = n {Un: n = 1, 2, · · · }. If U is a strong neighbor-
hood of 0, then clearly U absorbs strongly bounded sets. Con-
versely, assume that U absorbs strongly bounded sets; then to show
that U is a strong neighborhood of 0, it is sufficient to exhibit a weak*
closed convex circled subset V which is radial at 0, since such a sub-
set is the polar of a bounded subset of E. Let {Bn} be a co-base for
the strongly bounded subsets of E* suchthat each Bn is weak* compact
convex circled (there is such a co-base by 22.3). Foreach n, there is
a positive number tn such that tnBn c ~ U, and there is a weak*
closed convex circled strong neighborhood Wn of 0 which is contained
in ~Un- Let Vn = (t 1 B 1 u · · · u tnBn) + Wn- Then, since
(t 1 B 1 U · · · U tnBn) is weak*. compact, Vn is a weak* closed convex
circled strong neighborhood of 0 contained in Un- Let V= n
{Vn:
n = 1, 2, · · ·}; then it is clear that V is a weak* closed convex circled
subset of U which absorbs each Bn, and consequently absorbs each
point of E* -II I
We obtain as a corollary to 22.9 the fact that E*, which may fail to
be evaluable, nevertheless satisfies a weakened form of the evaluability
condition.

22.10 CoROLLARY Let E be a locally convex pseudo-metrizable

space. If a strongly bounded subset B of E** is the union of a sequence
{Bn} of strongly equicontinuous sets, then B is strongly equicontinuous.
PROOF Since B = U {Bn: n = 1, 2, · · · }, taking polars in E*,
Ba = n {(Bn)a: n = 1, 2, · · · }. Because (Bn)o is a strong neighbor-
 SEc. 22 PsEuno-METRIZABLE SPACES 215
hood of 0 and B 0 absorbs strongly bounded sets, B 0 ts a strong
neighborhood of 0.111
The next theorem states an extremal property of the strong topol-
ogy for E* which is similar to that proved for the topology of uniform
convergence on compact sets 'in theorem 22.5. The result may be
stated: The strong topology !/' for the adjoint E* of a pseudo-
metrizable locally convex space is the strongest of the locally convex
topologies which agree with !/' on strongly bounded subsets of E*
(equivalently, on equicontinuous subsets of E*).

22.11 LocALIZATION OF STRONG ToPOLOGY Let E be a locally

convex pseudo-metrizable space. Then a convex circled subset A of E*
is a strong neighborhood of 0 if and only if A n B is a neighborhood of
0 in B for each strongly bounded convex circled subset B of E*.

PROOF If A is a strong neighborhood of 0, then A n B is surely a

neighborhood of 0 in B for each subset B of E* which contains 0.
Conversely, assume that A is a convex circled subset of E* such that,
for each strongly bounded convex circled subset B of E*, A n B is a
neighborhood of 0 in B. Let {Bn} be a co-base for strongly bounded
sets in E* such that each Bn is convex and circled and Bn c Bn + 1 •
Then by the assumption, for each n, there is a convex circled strong
neighborhood Un of 0 in E* such that A n Bn => Un n Bn. Let
Vn = (A n Bn) + Un. Then Vn n Bn c 3(A n Bn), for the fol-
lowing reason. If h E vn n Bn, then h = f + g, where f E A n Bn
andg E un> henceg = h- /E Bn + A n Bn c 2Bn, and thereforeg E
2(Bn n Un) c 2(A n Bn) and h = f + g E 3(A n Bn)· lt follows
that n
{Vn: n = 1, 2, · · ·} c 3A, and by theorem 22.9, the proof
reduces to showing that n{Vn: n = 1, 2, · · ·} absorbs strongly
bounded sets. For a fixed m, there is a positive number t such
that tBm c Um n Bm c A n Bm c A n Bn for n ~ m. Hence
n
tBm E {Vn: n ~ m}. Since each of vl, ... ' vm-l absorbs Bm, it is
evident that Bm is absorbed by n
{Vn: n = 1, 2, · · ·}.111

22.12 CoROLLARY Let E be a locally convex pseudo-metrizable space.

Then a linear transformation T of E*, with the strong topology, into a
locally convex space F is continuous if and only if T is continuous on
each equicontinuous set (or equi7-•alently, on each strongly bounded set) in
E*. Consequently, E** with the strong topology is always a locally
convex complete metrizable space.
216 CH. 5 DUALITY

PROOF The first half is evident. E** with the strong topology is
obviously locally convex and metrizable, and the completeness of E**
follows immediately from part (iii) of theorem 18.4.111
Although the strong topology for the adjoint of pseudo-metrizable
space may fail to be bound, there are boundedness properties which
are of importance. The following result is the key to these properties.

22.13 BoUND ABSORBING SuBSETS OF E* Let E be a pseudo-metrizable

locally convex spac'e, and Iet C be a convex circled subset of E* which
absorbs strongly bounded sets. Then there is a strongly closed convex
circled subset of C which absorbs strongly bounded sets.
PROOF Let {Bn} be a co-base for the strongly bounded subsets of E*
such that each Bn is weak* compact convex and circled. Thcn
there is a sequence {tn} of positive numbers such that <U {tnBn:
n = 1, 2, · · ·}) c C. Let An= (t 1 B 1 U · · · U tnBn); then Anis
strongly closed (in fact, weak* compact) convex circled, An c An+ 1 ,
andA = U{An:n = 1,2,···} c C. Ifitisshownthat~A- c A,
then the theorem is proved, because A- absorbs .strongly bounded
sets. Let / 0 tf A ; then / 0 tf An for each n, and since An is closed,
there is a strong neighborhood Vn of 0 such that (/0 + Vn) n An is
void. Let wn = vn + ~An. Then Uo + Wn) n ~An is void, and
the intersection n {Wn: n =
1, 2, · · ·} absorbs each An and hence
each Bn. By theorem 22.9, fo + n{Wn: n = 1, 2, · · ·} is a strong
1
neighborhood of / 0 , and this neighborhood is disjoint from 2A.
1
Hence f tf 2A - ·II I
Recall that if E is a linear topological space, then Eb is the space of
all bounded linear functionals (that is, functionals which are bounded
on each bounded subset of E).

22.14 CoROLLARY If E is a locally convex pseudo-metrizable space,

then each strongly bounded subset of E*b is contained in the weak* closure
(that is, w(E*b,E*)-closure) of a strongly bounded subset of E**.
PROOF Let B be a strongly bounded subset of E*b. Then Ba is a
convex circled subset of E* which absorbs strongly bounded sets.
Hence, in view of the foregoing theorem, there is a strongly closed
convex circled subset C of Ba which absorbs strongly bounded sets.
 SEC. 22 PROBLEMS 217
Let D be the polar of C in E**; then Dis strongly bounded, C = D 01
and D 0 c B 0 • It follows that the polar in E*b of D 0 , namely the
weak* closure of D in E*b, includes B.lll

22.15 STRONG ToPOLOGY FOR THE ADJOINT If E is a locally convex

pseudo-metrizable space, then the following three statements concerning
the strong topology for E* are equivalent.
(i) E* with the strong topology is evaluable.
(ii) The strong topology for E* is bound.
(iii) E* with the strong topology is barrelled.
PROOF That (i) implies (ii) is clear from theorem 22.13. Since E* is
complete relative to the strong topology (22.3), (ii) implies (iii) by
virtue of theorem 19.5. By theorem 20.4, (iii) implies (i).lll
Finally, there are two special cases in which the adjoint is a barrelled
space ( equivalently, the strong topology is bound).

22.16 REFLEXIvE METRIZABLE SPACES /j E is a reflexive /ocal/y

convex metrizable space, then the strong topology on E* is bound.
PROOF If E is reflexive, so is E*, and hence E* is evaluable (see
20.5).111
22.17 SEPARABLE ADJOINT OF A PsEuno-METRIZABLE SPACE Let E
be a locally convex pseudo-metrizable space such that E* is separable
relative to the strong topology. Then the strong topology for E* is
bound.
PROOF In view of theorem 22.15, it is sufficient to show that E* with
the strong topology is evaluable, that is, each convex circled strongly
closed subset A of E* which absorbs strongly bounded sets is a strong
neighborhood of 0. Since E* is separable and A is closed, there is a
sequence Un} which is densein E* "' A. Foreach n, Iet un be a con-
vex circled strong neighborhood of 0 such that fn f/= Un and A c Un.
Then U = n
{Un: n = 1, 2, · · ·} ::J A and, by theorem 22.9, U is a
strong neighborhood of 0. Since E* "' Ui ( Ui is the interior of U)
is a closed set contairi.ing Un}, E* "' A c E* "' Ui and A ::J Ui;
hence, A is a strong neighborhood of 0.111

PROBLEMS
A CONDITION FOR COMPLETENESS
A locally convex pseudo-metrizable space is complete if the closed
convex circled extension of every sequence convergent to zero is weakly
218 CH. 5 DUALITY

compact. In particular, it is sufficient if the closed convex circled exten-

sion of every compact, or totally bounded, set is weakly compact.

ß EMBEDDING THEORIES
(a) Let R be the space of real numbers and C (R) the space of con-
tinuous scalar-valued functions on R with the topology of compact
convergence. Any locally convex metrizable separable space E is iso-
morphic to a subspace of C (R). (The theorem can be established by
proving the following
Lemma There is a continuous mapping f of R onto the adjoint E*
with the weak* topology, such that the family of sets f[K], where K is a
compact subset of R, form a co-base for the equicontinuous subsets of E*.
To prove this, choose a co-base {Bn: n = 0, 1, · · ·} of the equicontinuous
subsets of E* so that each Bn is convex, circled, weak* compact and weak*
metrizable, and so that Bn c Bn +1 for each n. There is a continuous
function fn on the Cantor set in [0, 1] onto Bn with the weak* topology: see,
for example, Kelley [5] Ch. V, 0. Define f by f(2n + s) = fn(s) whenever
s belongs to the Cantor set, and by continuity and linearity elsewhere.)
(b) A separable normed space is isometrically isomorphic to a subspace
of the space C([O, 1]) of continuous scalar-valued functions on [0, 1] with
the topology of uniform convergence.

C INDUCTIVE LIMITS V (see 11, 16C, 17G, 19A)

Let (E,Y) be the inductive Iimit of an increasing sequence of Frechet
spaces (En,Yn)· Then Eis called a generalized LF space. If each En is a
closed subspace of (En+l,Yn+t), the inductive Iimit is called an LF space.
In this case, the open mapping theorem shows that the injection of En into
En+l is a homeomorphism ; thus Eisa strict inductive Iimit (see 17G(b)).
It follows in particular that Eis a Hausdodf space and is therefore complete
(17G(a)).
(a) A generalized LF space is a barrelled and abound space, and so also a
Mackey space and evaluable.
(b) If { En} is a strictly increasing sequence and E is an LF space, then E
cannot be of the second category and so cannot be metrizable. An LF
space has a countable co-base for bounded sets if and only if it is the in-
ductive Iimit of a sequence of Banach spaces; equivalently, the adjoint
with its strong topology is metrizable.
(c) The adjoint of a reflexive Frechet space with its strong topology is a
generalized LF space. The adjoint of a reflexive Frechet space with its
strong topology is a complete, barrelled, and bound space.
( d) A closed subspace, or a quotient by a closed subspace, of even a
reflexive LF space may fail to be an LF space. The Krein-Smulian
theorem 22.6 does not extend even to reflexive LF spaces; further, an LF
space is not necessarily fully complete. Not all generalized LF spaces
are complete. (See 20D.)
(e) A linear mapping of one Hausdorff generalized LF space into
another is continuous if its graph is closed; a linear mapping of one onto
another is open if its graph is closed. (See 19B.)
 SEC. 22 PROBLEMS 219
(f) Let X be a topological space in which there is a countable co-base
for the compact sets (for example, a space which is locally compact and is
the union of a sequence of compact sets), and Iet K(X) be the space of con-
tinuous functions on X of compact support with the inductive Iimit
topology (see 81, 16C(j)). Then K(X) is an LF space; if Xis not com-
pact, K(X) cannot be metrizable (because of (b)).

D SPACES OF ANALYTIC FUNCTIONS II (see ZOH, ZZJ)

Let C be a closed proper subset of the Riemann sphere, and consider
the family J?t of all open proper subsets containing C, directed by putting
U ; : ;: V iff U c V. For each U E Oll, Iet A( U) be the space of analytic
functions on U with the topology of compact convergence (see ZOH).
For U ; : ;: V, Iet I uv be the natural injection of A( V) into A( U); then
{A(U): U E 011,;;::;:, luv}is an inductive system. LetF(C) =!im ind{A(U):
U E %'} with the inductive Iimit topology (see 16C). The space F(C) can
be identified with the set of equivalence classes of functions analytic on
(some open set containing) C, calling fand g equivalent iff there is some
U E J/t on which fand g agree.
(a) Each of the mappings Iuv is continuous. Since J/t contains a
countable cofinal subset {Un:n=1,2,···}, F(C)=limind{A(Un):
n = 1, 2, · · ·} and the inductive Iimit topology defined by the Frechet
spaces A(Un) is the topology of F(C). Thus F(C) is a generalized LF
space. In fact, F(C) is the inductive Iimit of a sequence of Banach
spaces. (Let Bn be the space of functions analytic on Un, continuous on
Un-, and vanishing at oo in case oo E Un, with the topology of uniform
convergence on U n-.)
(b) The spaces A( U) and F ("""' U) may be paired, in the following way.
If g' E F ("""' U ), there is a member g of g' defined on an open set V ::::> """' U
(and we may suppose oo $ U n V); there is an open set W with """'V c
W c W- c U and whose oriented boundary 8 W is rectifiable. Foreach
jEA(U), put (j,g') = (1/21Ti) frwf(z)g(z)dz. The value of the integral
is independent of the member of g' chosen and of the particular choice of
V and W satisfying the above requirements. This pairing is separated.
(c) The adjoint of A(U) is (algebraically) isomorphic to F(~ U). (lf
.P E A( U )*, there is a compact subset K of U """' { oo} and a constant k with
J.P(f)l :;;;; k sup {J/(z)J: z E K}. There is a natural mapping of A(U)
into F(K) and so .p may be regarded as a linear functional on the image
of this mapping; extend .p to F(K), so that if h' E F(K), J</>(h')J :;;;;
k sup {Jh(z)J: z E K}. Foreach ~ $ K, Iet h,(z) = ~~ z; then h/E F(K).
Putg(~) = </>(h/). ThengEF(~U); show that, for some suitable open
set W,

(j,g') = -21
7Tt
.J
ow
f(~)</>(h/)d~ = .PUo') where j 0 (z) = -21 .J /W
7Tt ow .. - z
d~,
using, for example, approximating sums or vector integration. Now
j 0 (z) = f(z) in some open set containing K; thus </>(/0 ') = </>(/).)
220 CH. 5 DuALITY

(d) With its strong topology, A(V)* is isomorphic to F(~V). (Since

both are generalized LF spaces, it is sufficient, by 22C(e), to show that the
algebraic isomorphism of F ( ~ V) onto A( V)* is continuous. This can
be achieved by proving that, for each open set V containing ~ V, the
mapping of A(V) into A(U)* is continuous.)

E SPACES /P(w)
A non-negative real valued function w defined on a set A defines a
measure !Lw on the a-ring of all subsets of A by the formula fLw(B) =
L: {w(t); t E B}; w is called the weight function for the measure fLw· The
space LP(A,fLw) bears a close resemblance to the zv spaces and it will there-
fore be denoted by /P(w). Thus /P(w) is the space of (equivalence classes of)
real or complex valued functions x with
llxllv = (f lx(t)IPdfLw(t)) 11P = (L: {lx(t)!Pw(t): t E A}) 11P < oo,
two functions x andy being regarded as equivalent iff x(t) = y(t) whenever
w~t) 'f 0. ThB adjoint of /P(w) for 1 ;;:; p < oo is lq(w), where q is the
index conjugate top (q = oo if p = 1 and q = pf(p - 1) if p > 1).
(a) lf y isareal or complex valued function on A suchthat xy E P(w) for
all x E zv(w), then y E lq(w).
(b) lf w(t) > 0 for all t E A, then /P(w) is reflexive for 1 < p < oo, and
l 1 (w) and / (w) arereflexive if and only if Ais a finite set.
00

(c) In F(w) each seqcence which is Cauchy in the weak topology is

convergent in the norm topology, and each set which is compact in the
weak topology is compact in the norm topology. (The second part is a
consequence of the first, which can be proved in the same way as the
special case of [1 in 16F.)

F KÖTHE SPACES
These spaces are projective and inductive Iimits of spaces of the type
/P(w), defined in the previous problem, and their adjoints. For each
weight function w on A, denote by /P(w)+ the space of real or complex
valued functions y on A such that y(t) = 0 whenever w(t) = 0 and
L: {lx(t)y(t)i: t E A} < oo for all x E l 11(w).
(a) The spaces /P(w) and zv(w)+ are paired by the bilinear functional
<x,y) = L: {x(t)y(t): t E A}; y E /P(w)+ if and only if there is a z E (ZP(w))*
with y = wz, and the natural norm associated with the pairing is IIYII =
llzllq, where q is the index conjugate top. This pairing does not depend
explicitly on the function w and is convenient when a family of weight
functions comes simultaneously under consideration.
(b) There is a natural partial ordering among weight functions in which
w ~ a means that w(t) ~ a(t) for all t in A. Let Q be a family of weight
functions directed by this partial ordering. If w ~ a, zv(w) c /P(a) and so,
with injections for the canonical mappings, the family {ZP( w): w E Q} forms
a projective system. The projective Iimit of this system is algebraically
isomorphic to EP(Q) = n {ZP(w): w E Q}. The space EP(Q) is dense in
each zv(w). We suppose that EP(Q) has the projective Iimit topology; the
intersections of spheres in zv( w) with E P( Q) form a local base for this
topology.
 SEC. 22 PROBLEMS 221
(c) The space EP(Q) is Hausdorff iff to each t E A corresponds an w E Q
with w(t) > 0. For each p with 1 ~ p : :;:; w, EP(Q) is complete. The
adjoint of EP(Q) is algebraically isomorphic to EP(Q)+ = U {/P(w)+:
w E Q} with the canonical pairing (x,y) = 2: {x(t)y(t): t E A} for x E EP(Q)
and y E EP(Q)+.
(d) In EP(Q) each sequence which is Cauchy in the weak topology is
convergent in the projective Iimit topology.
(e) 1f w ~ a, /P(a)+ c /P(w)+ and so, with injections for the canonical
mappings, the family {/P(w)+: w E Q} forms an inductive system, for which
EP(Q)+ is the inductive Iimit. The inductive Iimit topology on EP(Q)+ is
stronger than the strong topology, but if 1 < p < w they coincide and
EP(Q) is semi-refl.exive. (Prove that any linear functional continuous
relative to the inductive Iimit topology on EP(Q)+ is the evaluation at a
member of EP(Q).)
The case of P itself shows that this last result may fail for p = 1.

G COUNTER-EXAMPLE ON FRECHET SPACES

The Frechet space E constructed in this example has the property that
the strong topology of its adjoint E* is not bound. This is equivalent, by
22.15, to saying that E* is not evaluable orthat it is not barrelled. Thus
the example shows that none of these three properties carries over to the
adjoint with its strong topology. Next, the Frechet space E may be em-
bedded as a closed subspace of a countable product F of Banach spaces;
then F* with its strong topology is a countable direct sum of Banach
spaces, and E* is algebraically isomorphic to F*/E 0 • The quotient of
s(F* ,F) is bound; thus the strong topology s(F* / E 0 ,E) is not the quotient
of s(F*,F). Finally, this provides an example of a strong adjoint of a
projective Iimit which is not the inductive Iimit of the corresponding
strong adjoints.
Let I be the set of positive integers, Iet A = I x I, and Iet Q be a
sequence {wn} with the following properties: if n > m, then wn ~ wm; for
each (i,j) E A and each n, wn(i,j) > -!-; for i : :;:; n, !imi wn(i,j) = w; for
i > n, lim supi wn(i,j) < 1.
Let fJß be the family of subsets B of A of the following type: there are a
positive integer i 0 and a sequence {k;} of positive integers such that B =
{(i,j): i ~ i 0 and j > kJ If for each BE flß, x 8 denotes the characteristic
function of the singleton {(i 0 ,k, 0 ) }, then {x 8 ,B E fJß, c} is a net in E = E 1( Q);
regard it as a net in the algebraic dual of E* = E 1 ( Q) +.
(a) A universal subnet of {x 8 } converges relative to w((E*)',E*) to a
linear functional </> on E* which is continuous relative to the inductive
Iimit topology, that is, which is bounded on E*.
(b) Each strong neighborhood of 0 in E* contains a neighborhood of the
form V = <UnenSn°)-, where the closure is taken with respect to the
w(E*,E)-topo!ogy, Sn is the unit sphere in P(wn), and en > 0.
(c) Given V of the form in (b ), choose the sequence {kn} so that, if
j ~ kn, then 1 :::;:; (en/2n)wn(n,j), and Iet y(i,j) = 1 if j ~ k;. y(i,j) = 0
otherwise. Then y E V, and </>(y) = 1. (Let Yn(n,j) = 1 if j ~ kn, and
Yn(i,j) = 0 otherwise. Then Yn E (en/2n)Sn°, while 2: {yn: n = 1, 2, · · ·}
converges to y with respect to w(E*,E).)
222 CH. 5 DUALITY

(d) E is a Fn!chet space. The strong topology of its adjoint 1s not

bound.

H AN ADJOINT WITH A BOUND TOPOLOGY

Let E be a locally convex pseudo-metrizable space and Iet sb be the
bound extension of s = s(E* ,E ).
(a) If Ais a strongly separable subset of E*, then s and sb coincide on A.
In particular, a sequence is s-convergent if and only if it is sb-convergent.
(b) If every strongly bounded subset of E* is strongly metrizable, then
E* is a bound space.
(For (a), use a method similar to 22.17; for (b), use 22.11 and (a).)

I EXAMPLE ON MONTEL SPACES

For each p, 1 ~ p < oo, there is a Frechet Montel space Ev which con-
tains a closed subspace Fv suchthat EvfFv is topologically isomorphic to lv.
With the definitions of 22F, Iet Ev = EP(.Q), where I is the set of positive
integers, A = I x I, and .Q is the sequence {wn: n = 1, 2, · · ·} with
wn(i,j) = jn if i ~ n and wn(i,j) = in if i > n.
(a) For each p, 1 ~ p < oo, Ev is a Frechet Montel space. (To show
that a closed bounded set B is compact, show that for each e > 0 and each
n there is a bounded finite dimensional set B' such that every point of B is
within distance e of B' with respect to the norm in /P(wn).)
(b) The transformation T defined by
T(x)(j) = 2;{x(i,j): i = 1, 2, · · ·}
is a continuous and open linear transformation from Ev onto zv. (Prove
that the transformation P' defined by
T*(y)(i,j) = y(j)
is a transformation from zq (where q is the conjugate of p) into Ev * =
EP(.Q)+, that T* is the adjoint of T, that T* is one-to-one, and that the
range ofT* is w(Ev*,Ev)-closed.)

J SPACES OF ANALYTIC FUNCTIONS 111 (see 20H, 22D)

This problern gives a Köthe space representation of the space A(U) of
functions analytic in the interior {z: JzJ < 1} of the unit circle in the com-
plex plane. Take any (fixed) sequence {r1.,} of positive numbers increasing
to 1. Then the topology of A( U) is determined by the sequence {Pk:
k = 1, 2, · · ·} of pseudo-norms, where pk(f) = sup {Jf(z)J: Jzl ~ rk}•
(a) Foreach k, let wk be the weight-function defined on the set I of non-
negative integers by putting wk(i) = (rk)i, and let .Q = {wk: k = 1, 2, · · · }.
Then for each f E A( U ), the sequence {an: n = 0, 1, 2, · · ·} of coefficients
in the Taylor expansion of f about the origin lies in E 1 (.Q) and the mapping
J defined by J(f) ={an} is a topological isomorphism of A(U) onto E 1 (.Q).
(b) Foreach sequence {bn: n = 0, 1, 2, ···}in E 1 (.Q)+, putl*({bn}) = J,
wheref(z)=2;{bnz-n- 1 :n=0,1,2,···}. Then 1* is a topological
isomorphism of E 1 (.Q)+ onto the space F(-U), which is the adjoint of
A(U) under its strong topology. (Use 22D(d).) Also J* is the adjoint
of J.
 SEC. 22 PROBLEMS 223
(c) Let T be a continuous linear mapping of A(U) into itself. Then
there exists a unique kerne! B suchthat T(f) = g where, for izi < 1,
g(z) = -217TZ.J ICl = f(~)B(~,z)d~.
T(2)

Here the contour integral is taken· round a circle of radius r(z) < 1, which
may depend on z, and the kerne! B has the following properties:
(i) the mapping (~,z) ~ B(~,z) is continuous in the two variables, for
1~1 ~ 1 and lzl < 1;
(ii) to each r < 1 corresponds an r' < 1 such that, whenever lzl < r,
the mapping ~ ~ B(~,z) is analytic on g: 1~1 > r'} and vanishes at
infinity;
(iii) for each ~ with 1~1 > 1 the mapping z ~ B(~,z) is analytic on U.
Conversely, if B is a kerne! satisfying the conditions (i), (ii), (iii), it defines
a unique continuous linear mapping of A( U) into itself. (Let vzW =
(~- z)- 1 ; if T is given, put B(~,z) = uz(~) where U 2 = T*(vz)· In the
converse, the main difficulty is to show that the function T (f) defined by
such a kerne! is an analytic function in U. This can be clone using (ii) and
Hartogs' lemma-s~e, for example, Bochner and Martin [2].)
 Appendix

ORDERED LINEAR SPACES

This appendix is a brief introduction to the theory of ordered linear

spaces. Many linear topological spaces have a natural order, and the
relation between order and topology is of considerable interest. For
example, we may ask whether positive linear functionals are neces-
sarily continuous, or whether each continuous linear functional is the
difference of two positive functionals. The first section of this
chapter is devoted to introductory material and to a few results on
problems of this type. The second section contains two theorems of
Kakutani which characterize two " extremal" types of Banach lattice.

23 ORDERED LINEAR SPACES

Each continuous linear functional on an ordered normed linear space
E is the difference of positive functionals if and only if there is an
equivalent monotonic norm; in particular, this is the case if the
positive cone E is normal in the sense of Krein. A linear functional
on a vector lattice is the difference of positive functionals if and only if
it is bounded on order bounded sets. Hence the Mackey topology
m(E,E*), where E* is the order dual of E, is bound. The dass of
lattice pseudo-norms for Eisabase for the dass of m(E,E*)-continuous
pseudo-norms, and the lattice operations are continuous relative to
m(E,E*). The evaluation map of E into E** is a lattice homo-
morphism.
It is assumed throughout the section that E is a real linear space;
no topology is assumed, ab initio. It has already been noted that
there is a natural correspondence between cones in E ( convex sub-
sets, closed under multiplication by non-negative scalars) and vector
orderings ~ ; that is, between cones and partial orderings ~ such
that: x ~ y if and only if x - y ~ 0; if x ~ 0 and t is a non-negative
224
 SEC. 23 ÜRDERED LINEAR SPACES 225
scalar, then tx G; 0; and if x G; 0 and y G; 0, then x + y G; 0. An
ordered linear space (partially ordered linear space) is a real
linear space E with a vector ordering. In view of the correspondence
which has just been noted, the ordering G; is completely determined
by the cone {x: x G; 0}. This cone is called the positive cone of
the ordering and members of the cone are called positive.
If C is a cone in E, then C - C is clearly a linear subspace F of E.
In the most interesting cases, the subspace F is identical with E; in
particular, if the radial kernel of C is non-void, then C - C = E.
If C - C is the entire space E, then the cone generates E (the cone
C is reproducing). This condition can be interpreted easily in
terms of the order G; determined by the cone. If C generates E,
then for each x in E there are u and v in C such that x = u - v, and
consequently there is a positive element u such that u G; x. Con-
versely, if for each x in E there is a positive element u, such that
u G; x, then x = u - (u - x) and it follows that E = C - C.
Consequently, C generates E if and only if for each x there is u such
that u G; 0 and u G; x. An equivalent form of this requirement can
be obtained by a simple translation argument: the positive cone C
generates E if and only if the ordering ~ directs E; that is, if and
only if for members x and y of E there is z in C such that z G; x and
z G; y.
An ordering G; of a linear space is studied by means of the linear
functionals which are order preserving; that is, by means of the
linear functionals which are non-negative on the positive cone C.
The set C* of all such linear functionals on E is the dual cone, and
the set C* - C* of all differences of members of C* is the order dual
E* of E. Of course, the dual cone may consist of the single element
0, as is certainly the case if C = E. The cone C* defines an order on
E* which is called the dual ordering. If C generates E, then the
dual ordering is anti-symmetric, in the sense that, if J G; g and
g ~ J, then g = J, for in this case f - g is zero on each member of C
and hence on E = C - C.
If a cone C is closed relative to a locally convex topology for E and
x rf= C, then by the separaiion theorem 14-.4, there is a linear functional
f such that f(x) < inf {J(y): y E C}. Since C is a cone f(x) < 0,
and f is non-negative on C; that is, f E C*. In particular, if C is
closed relative to the strongest locally convex topology .'Y, which has
the family of all convex circled radial sets at 0 as a local base, then
each point which does not belong to C can be separated from C by a
member of C*. The preceding discussion can be summarized:
226 APPENDIX: ÜRDERED LINEAR SPACES
23.1 THEOREM If Cis a cone which is closed in the strongest locally
convex topology for E, then for each x in E "' C there is f in the dual
cone C* such that f(x) < 0. Consequently, in this case, x E C if and
only if f(x) ~ 0 for each f in C*.

If a linear space has both a topology and an order, it is of some

importance to know under what conditions each continuous linear
functional is the difference of continuous positive functionals. This
problern will be investigated now, deferring until later certain ques-
tions of a purely order theoretic nature. The first result concerns
ordered pseudo-normed spaces.

23.2 THEOREM Let C be a cone in a pseudo-normed space E and Jet

B be the set of all positive functionals which are of norm at most one.
Then each member of E* is the difference of bounded positive linear
functionals if and only if B - Bis a neighborhood of 0 in E*.

PROOF If B - B is a neighborhood of 0 in E*, then clearly each

member of E* is the difference of continuous positive functionals, and
only the converse requires proof. First observe that B is weak*
compact, because it is a weak* closed subset of the unit sphere in E*,
and hence B - B is weak* compact. It follows that B - B is weak*
closed, convex, and circled. lf C* n E* generates E*, then for each
fin E*, some non-zero scalar multiple ofjbelongs toB- B: that is,
B - B is radial at 0. Finally, since the unit sphere in E* is weak*
compact, the absorption theorem 10.2 asserts the existence of a
scalar a, such that a(B - B) contains the unit sphere, and hence
B - B is a neighborhood of O.JJJ
Observe that if each continuous linear functional is the difference
of two continuous positive functionals, then there is a certain uni-
formity present, in the following sense. If a sphere of radius r is
contained in B - B, then any linear functional of norm at most one
is the difference of two positive functionals of norm at most 1fr.
There is a useful corollary to the foregoing theorem: if Eis a pseudo-
normed space with a cone C such that each member of E * is the
difference of positive members then the set B of all continuous
positive functionals of norm at most one may be used to construct a
new pseudo-norm of E, as follows: Iet p(x) = sup {lf(x)J :f E B}.
Clearly p(x) ~ JJxJJ, because B is contained in the unit sphere, and
since B - B contains a multiple, say lfr, of the unit sphere, JJxJJ ~
2rp(x). Hence p is equivalent to the original pseudo-norm. But
 SEC. 23 ÜRDERED LINEAR SPACES 227
notice that p is monotonic on E; that is, if x and y are members of C
and x ?; y, then p(x) ?; p(y).

23.3 CoROLLARY lf each continuou~ linear functional on an ordered,

pseudo-normed space Eis the difference of continuous positive functionals,
then there is an equivalent monotonic pseudo-norm for E.
It is true that if p is a pseudo-norm which is monotonic on a cone
C, then each continuous linear functional is the difference of con-
tinuous positive functionals; however, a slightly stronger statement
can be proved. A cone C in a pseudo-normed space E is normal
relative to the pseudo-norm if and only if there is a positive nurober e
such that if x and y are positive, llxll ?; 1, and IIYII ?; 1, then
llx + Yll ?; e. Geometrically, this amounts to requiring that the
angle between positive vectors, as computed in terms of distance, is
bounded away from 7T. This condition can be rephrased in a con-
venient way.

23.4 LEMMA A cone C in a pseudo-normed linear space is normal

if and only if (S + C) n (S - C) is bounded, where S is the unit
sphere.
PROOF If C is normal, there is a constant k (the reciprocal of the
constant e of the definition) such that, if x and y are positive and
llx + Yll < 1, then either llxll or IIYII is less than k. If z is a
member of (S + C) n (S - C ), then z = u + x = v - y for
some x and y in C and u and v in S. Then ll(x + y)/211 ~ 1, and
hence either l x/211 < k or IIY/211 < k. In either case, since both u
and v belong to S, llzll < 2k + 1, and half of the statement of the
Iemma is proved. To prove the converse, suppose that llzll, for z in
(S + C) n (S - C), is bounded by k. If x and y are positive
elements and llx + Yll < 1 then, since y = (x + y) - x, 1t 1s true
that y E (S + C) n (S - C). It follows that IIYII ~ k, and it is
clear that the cone C is normaL III

23.5 THEOREM If a cone C is normal in a pseudo-normed space, or if,

equivalently, ( S - C) n (S + C) is bounded, where S is the unit
sphere about 0, then each bounded linear functional is the difference of
two bounded positive linear functionals.
PROOF Let B be the set of positive functionals of norm at most one.
In view of theorem 23.2 above, it must be shown that B - B is a
neighborhood of 0 in the adjoint E. Since B - B is weak* compact
228 APPENDIX: ÜRDERED LINEAR SPACES
and therefore closed, and is convex and circled, B - B is the polar
of the subset (B - B) 0 of E, and the problern reduces to showing
that (B - B) 0 is bounded; it will, in fact, be shown that B 0 , which
contains (B - B)0 , is bounded. To this end, notice that if x is a
member of E such that x/2 does not belong to the convex set S + C,
which is radial at 0, then there is a non-zero linear functional f on E
such thatf(x/2) ~ inf{f(y):yES + C} by theorem 14.2. Since f
is not identically zero on S,f(x/2) is negative, and it may be supposed
that f(x/2) = - 1; since C is a cone, f must be a positive functional,
and since f is bounded below by - 1 on S, 1/f/1 ~ 1. Then f E B,
and it follows that x is not a member of B 0 • This proves that B 0 c
2(S + C), and since B 0 is circled, B 0 c 2(- S - C) = 2(S - C).
It follows that B 0 is a subset of 2[(S + C) n (S- C)], and is hence
bounded, and the proof is complete.lil
The following is an immediate consequence of 23.3 and 23.5.

23.6 THEOREM Each continuous linear functional on an ordered,

pseudo-normed space is the difference of positive continuous linear
functionals if and only if there is an equivalent pseudo-norm which is
monotonicrelative to the order.
The existence of an order in a linear space leadstoanatural defini-
tion of order bounded set. A subset A of an ordered linear space is
order bounded if and only if there are members x and y of the space
which, respectively, precede and follow each member of A in the
ordering ~; that is, such that A c {z: x ~ z ~ y}. Each linear
functional which is the difference of positive linear functionals is
necessarily bounded on order· bounded sets; under certain circum-
stances the converse is true.

23.7 THEOREM lf Cis a normal cone in a pseudo-normed space E,

then each order bounded set is bounded. lf C is a cone containing an
interior point x of E then C is normal if and only if the set {y: - x ~
y ~ x} is bounded.
PROOF If C is a normal cone then each continuous linear functional
is the difference of positive functionals and is hence bounded on order
bounded sets. That is, each order bounded set is weakly bounded,
and is therefore bounded. To prove the second statement, notice
that if C is normal, then surely the order bounded set {y: - x ~
y ~ x} is bounded, by the preceding remarks. Finally, if x is an
interior point of C, then for some positive r, x + rS c C, where S
 SEC. 23 ÜRDERED LINEAR SPACES 229
is the unit sphere. Then x + rS + C c C, hence {y: - x ~ y ~ x} =
(-x + C) n (x- C);:, (rS + C) n (-rS- C)and,if{y: -x ~
y ~ x} is bounded, then (rS + C) n (- rS - C) is bounded. It
follows that (S + C) n (S - C) is bounded.lll
The theory of ordered spaces is simplified a great deal if further
assumptions are made about the ordering. An ordered space
(E, ~) is called a vector lattice if and only if for each x and y in E
there is a unique supremum of x and y in E; that is, there is a unique
member z of E with the property that z is greater than each of x and
y, and is less than every member of E which has this property. The
supremum of x and y is called the join and is denoted x V y. A
lattice ordering is necessarily anti-symmetric (if x ~ y and y ~ x,
then x = y) because the supremum of two elements is supposed to be
unique. Geometrically, the ordering is a lattice ordering if and only
if the intersection of two translates, x + C and y + C, of the positive
cone is a translate, x V y + C, of the cone. As an example one may
consider two-dimensional real Euclidean space with the positive cone
C1 = {(r,s): r ~ 0 and s ~ 0} or the cone C2 = {(r,s): r > 0, or r = 0
and s ~ 0}. The orderings corresponding to these two cones are
intrinsically quite different. For the first, Cl> it is true that if x and
y are positive elements and tx ~ y for every positive scalar t, then
x = 0. Such a lattice is called Archimedean; it is sometimes said,
in this case, that there is no positive element which is "infinitely
small " relative to another positive element. The ordering defined by
C2 is non-Archimedean, since t(0,1) ~ (1,1) for allpositive t. Geo-
metrically, the ordering is Arehirneclean if and only if, for each
positive x, the set C n (x - C) contains no half line. It is known,
although it will not be proved here, that any finite dimensional vector
lattice can be constructed from the usual ordering of the real numbers
by taldng products, assigning either the product ordering (as for C1 )
or the lexicographic ordering (as for C2 ).
If E is any vector lattice, then scalar multiplication by - 1 is order
inverting. lt follows that there is then a greatest element which is
less than or equal to each of two given elements; that is, there is an
infimum of each pair of elements. The meet of x and y, denoted
x 1\ y, is defined to be the infimum of x and y. Using the fact that
multiplication by - 1 is order inverting, it is evident that x 1\ y =
- [(- x) V (- y)]. Either of the operations, meet or join, determine
the ordering entirely, for x ~ 0 if and only if 0 = x 1\ 0, and x ~ 0
if and only if 0 = x V 0.
A few simple computation rules are necessary. First, because the
230 APPENDIX: ÜRDERED LINEAR SPACES
ordering is invariant under translation by a member of E, the smallest
element of E which is greater than both x + z and y + z is x V y + z.
It follows that x V y + z = (x + z) V (y + z), and a similar rela-
tion holds for the meet. If, in the preceding equation, z is replaced
by - x - y, then, rearranging terms and using the definition of the
meet of x and y, one obtains the useful identity: x + y = x V y +
x 1\ y. In particular, x = x V 0 + x 1\ 0. The first of the two
latter terms, x V 0, is called the positive part of x and is denoted x+.
The element - (x 1\ 0) = (- x) V 0 is called the negative part of x
and is denoted x-. Clearly x+ ~ 0, x- ~ 0, and x = x+ - x-. A
particular consequence of this equality is that the positive cone
generates E.
There is another way of describing the positive part of an element
x of a vector lattice. Two positive elements, x and y are disjoint if
and only if x 1\ y = 0. Observe that x+ and x- are always disjoint,
for (x V 0) 1\ [- (x 1\ 0)] may be written, using the translation in-
variance, as [(x V 0 + x 1\ 0) 1\ 0] - x 1\ 0 = x 1\ 0 - x 1\ 0 = 0.
On the other hand, if x and y are arbitrary disjoint positive elements,
then (x - y) + = (x - y) V 0 = x V y - y = x - x 1\ y = x; a
similar calculation shows that (x - y)- = y. The positive and
negative part of an element z of a vector lattice can then be described
as the unique disjointpositive elements whose difference is z. There
is another simple result on disjointness which will be useful. If x
and y are arbitrary elements, then x - x 1\ y and y - x 1\ y are dis-
joint, as may easily be verified using the translation invariance of the
ordering.
The a.bsolute value of a member x of a vector lattice is denoted JxJ
and is defined to be x+ + x-. It is easy to see that JtxJ = JtJ·JxJ,
for each scalar t.
The following result will be essential.

23.8 DECOMPOSITION LEMMA Let E be a vector lattice and Iet x, y,

and z be positive elements such that x ~ y + z. If u = x 1\ y and
v = x - x 1\ y, then u and v are positive, x = u + v, u ~ y and v ~ z.

PROOF Clearly 0 ~ u ~ x, and it must be proved that 0 ~ v ~ z.

Since x 1\ y is less than x, v = x - x 1\ y is positive. To show that
z ~ x - x 1\ y observe that z ~ x - y, from the hypothesis, and
since z ~ 0, it follows that z ~ (x - y) V 0. Applying a translation
by -x, it follows that z - x ~ (x- y- x) V (-x) = -x 1\ y,
which is the desired inequality.JJJ
 SEc. 23 ÜRDERED LINEAR SPACES 231

The order dual of a vector lattice can be characterized very nicely.

It has already been remarked that each member of the order dual is
bounded on order bounded sets, and the converse is correct if the
space is a lattice.

23.9 ÜRDER DUAL OF A VECTOR LATTICE A linear junctional j on a

vector lattice E belongs to the order dual E* if and only if f is bounded on
order bounded sets. Under the dual ordering E* is a vector lattice, and,
for anfin E*,j+ = f V 0 is given by j+(x) = sup {/(u): 0;;:; u;;:; x}
for each positive element x in E.

PROOF Let f be a linear functional on E which is bounded on order

bounded sets, and foreachpositive element x letj+(x) be the value
as defined in the statement of the theorem. Clearly j+(x) is finite
and non-negative. If x and y are positive elements, then 0 ;;:; u ;;:; x
and 0 ;;:; v ;;:; y imply that 0 ;;:; u + v ;;:; x + y; hence,j(x) + f(y) =
f(x + y);;:; j+(x + y). It follows thatj+(x) + j+(y) ;;:; j+(x + y).
On the other hand, if 0 ;;:; w ;;:; x + y, then by the decomposition
lemma there are elements u and v such that w = u + v, 0 ;;:; u ;;:; x,
and 0;;:; v;;:; y. Hence, f(w) = f(u + v) = f(u) + f(v) ;;:; j+(x) +
j+(y) from which it follows that j+(x + y) ;;:; j+(x) + j+(y).
Therefore, we havej+(x + y) = j+(x) + j+(y) for positive elements
x andy. Obviously,j+(tx) = tj+(x) foranon-negative real number
t. It is not hard to see that f + defined on the positive cone can be
extended to a linear functional (positive) on E. (The extension will
also be denoted by f +.) Since f ;;:; f +, f + - f is positive, and
f = j+ - (j+ - /) E E*. In view of the remark preceding the
theorem, the first assertion is proved. The functional f + defined
above is an upper bound of fand 0, and, to see that it is the least such,
it is enough to notice that if g ;;:; 0 and g ;;:; J, then for 0 ;;:; u ;;:; x,
f(u) ;;:; g(u) ;;:; g(x). By the usual translation argument, we see that
for each fand g in E*, f V g exists.JJI
Letfand g be members of the order dual of a vector lattice E, and
let x be a positive element of E. Then it is easy to compute
(f V g)(x), (f 1\ g)(x) and I/ J(x). In fact, (f V g)(x) = (f- g) +(x) +
g(x) = sup {f(u) + g(x - u): 0 ;;:; u ;;:; x}, or (f V g)(x) = sup {f(u) +
g(v): 0 ;;:; u, v ;;:; x, u + v = x}. Similarly, (f 1\ g)(x) = inf {f(u) +
g(v): 0 ;;:; u, v ;;:; x, u + v = x}. J/J(x) = j+(x) + ( -j)+(x) =
sup {f(u - v): 0 ;;:; u, v ;;:; x}, but an element y is of the form
y = u - v for some u and v such that 0 ;;:; u, v ;;:;_ x if and only if
JyJ ;;:; x; hence, 1/J(x) = sup {/(y): JyJ ;;:; x}.
232 APPENDIX: ÜRDERED LINEAR SPACES

There are other more or less immediate consequences of the pre-

ceding theorem. Let E be a vector lattice and Iet d be the family of
all sets which, for some positive y, are of the form {x: - y ~ x ~ y}.
Then d is a co-base for the order bounded sets of E, and each
member of .9'/ is convex and circled. Let d- be the family of all
convex circled sets which absorb each member of d. Then in view
of Iemma 19.1, d- is a local base for abound locally convex topology
.r for E, and .r is the strongest locally convex topology relative to
which each member of d is bounded. Each $"-continuous linear
functional is bounded on order bounded sets and conversely, if a
linear functional f is bounded on order bounded sets, then {x: IJ(x)i
~ 1} is a mcmber of d- andf is therefore $"-continuous. Hence
the adjoint of (E,$") is simply E*. Recall (19.4) that a bound
topology is always the Mackey topology of the pairing of E and E*, and
hence $" must be the Mackey topology m(E,E*). Consequently the
family of all convex circled sets which absorb each order bounded set
is a local base for m(E,E*). The following theorem is then clear.

23.10 THE ToPOLOGY m(E,E*). Let E be a vector lattice, Iet E be

its order dual, and Iet $" be the Mackey topology m(E,E*). Then:
(i) A linear functional on E is .r -continuous if and only if it is
bounded on order bounded sets.
(ii) The topology $" is bound and is the strongest locally convex
topology with the property that each order bounded set is bounded.
(iii) A pseudo-norm p on E is .r-continuous if and only if each order
bounded set is p-bounded.
The next theorem describes the pseudo-norms which are con-
tinuous relative to the Mackey topology m(E,E*). A set P of pseudo-
norms for a linear topological space E is called a base for the family
of continuous pseudo-norms if and only if each member of P is con-
tinuous and every continuous pseudo-norm is dominated by some
member of P. This is entirely equivalent to requiring that the
family of all unit spheres about 0, constructed from the members of
P, be a local base for the topology. It is natural to suspect that the
set of all monotonic pseudo-norms for E is a base for the m(E,E*)-
continuous pseudo-norms; actually, an even stronger result can be
proved. A pseudo-norm p is a lattice pseudo-norm if and only if
p(x) ;?; p(y) whenever jxj ;?; jyj. It is easy to see that p is a lattice
pseudo-norm if and only if p is monotonic (if x ;?; y ;?; 0, then
p(x) ;?; p(y) and p(x + y) = p(x - y) for disjoint positive elements x
 SEC. 23 ÜRDERED LINEAR SPACES 233
and y. The latter condition is evidently equivalent to the require-
ment: p(x) = p(\xJ) for all x.

23.11 ÜRDER DuAL AND LATTICE PsEuno-NoRMS lf E is a vector

lattice, then the set of all lattice pseudo-norms is a base for the family of
pseudo-norms which are continuous relative to the Mackey topology
m(E,E*).
PROOF In view of the preceding theorem a pseudo-norm p is con-
tinuous relative to m(E,E*) if and only if order bounded sets are
p-bounded. If p is a lattice pseudo-norm, then an order bounded set
{z: x ~ z ~ y} is evidently p-bounded, and consequently such a
pseudo-norm is m(E,E*)-continuous. Suppose now that p is an
arbitrary m(E,E*)-continuous pseudo-norm, and Iet q(x) = sup {p(u):
\u\ ~ jxj}. This supremum is finite because the order bounded set
{u: -\x\ ~ u ~ jx\} is supposed to be p-bounded. Clearly q ~ p,
q(tx) = \t\q(x), and if jxj ~ jyj, then q(x) ~ q(y). It remains tobe
proved that q(x + y) ~ q(x) + q(y), and, since jx\ + \Y\ ~ \x + yj,
it suffices to consider only the case where x and y are positive. In
view of the definition of q, the problern then reduces to showing that
if x and y are positive and \z\ ~ x + y, then there are elements u and
v suchthat \u\ ~ x, \v\ ~ y and z = u + v. Since z+ ~ x + y and
z- ~ x + y, there are, in view of the decomposition Iemma 23.8,
positive elements u 1 , v1 , u2 , and v 2 such that z+ = u1 + v1 , u 1 ~ x,
v1 ~ y, z- = u 2 + v 2 , u 2 ~ x, and v 2 ~ y. Since z+ and z- are
disjoint, so are u1 and u2 and v1 and v 2 , respectively. Hence ju 1 -
u2 \ = u1 + u 2 = u1 V u 2 + u 1 1\ u 2 = u 1 V u 2 ~ x, and similarly
jv 1 - v 2 \ ~ y. But z = u 1 - u2 + v1 - v 2 , and the proof is
complete.\\\
W e have seen that the order dual E* of a vector lattice is a vector
lattice, and it follows that the second order dual E** is again a vector
lattice. Let e denote the evaluation map of E into its second order
dual E**; then it is easy to see that e takes a positive element into a
positive element. The fact that e preserves lattice operations (that
is, e(x V y) = e(x) V e(y) and e(x 1\ y) = e(x) 1\ e(y)) is less
obvious. In order to prove this fact we require a Iemma which is
useful in other connections too.

23.12 LEMMA Let E be a vector lattice, let u be a positive element of

E, and Iet f be a positive linear functional on E. Then there is a positive
linear functional g with the properties: (a) g ~ f, (b) g(u) = f(u), and
(c) g(x) = 0 for each positive element x such that x 1\ u = 0.
234 APPENDIX: ÜRDERED LINEAR SP ACES
PROOF For each positive eleroent x, define g(x) = sup {f(y):
0 ~ y ~ x and y ~ tu for some real number t s 0}. It will be
shown that g is additive on the positive cone. Let x and y be arbitrary
positive eleroents, and assuroe that z is an eleroent such that 0 ~ z ~
x + y and z ~ tu for soroe t s 0. Then by the decoroposition leroroa
23.8, there are eleroents a and b such that z = a + b, 0 ~ a ~ x, and
0 ~ b ~ y. Clearly a ~ tu and b ~ tu. Hence f(z) = f(a) +
f(b) ~ g(x) + g(y), and consequently g(x + y) ~ g(x) + g(y). On
the other hand, if a and b are eleroents such that 0 ~ a ~ x, 0 ~
b ~ y, a ~ tu and b ~ su for soroe t and s, then 0 ~ a + b ~ x + y
and a + b ~ (t + s)u; thereforef(a) + f(b) = f(a + b) ~ g(x + y).
It follows that g(x) + g(y) ~ g(x + y). Hence g(x + y) = g(x) +
g(y). It is easy to see that g(tx) = tg(x) for a positive x and a non-
negative real nurober t. Therefore, g can be extended to a linear
functional on E which will be again denoted by g. That g satisfies
(a) and (b) is clear. In order to see (c), let x be a positive eleroent
such that x 1\ u = 0, and let y be an eleroent such that 0 ~ y ~ x
and y ~ tu for soroe real nurober t s 0. Then for soroe positive
integer n, y ~ nu; hence 0 ~ y ~ x 1\ nu ~ n(x 1\ u) = 0, or y = 0,
and (c) is proved.lll
23.13 THE EvALUATION INTO THE SECOND ÜRDER DuAL Let E be a
vector lattice. Then the evaluation map e on E into the second order
dual E** preserves the lattice operations. (See the remarks before the
Iemma.)
PROOF By the usual arguroent it suffices to show that e(x)+ = e(x+)
or, equivalently, to show that e(x)+(j) = f(x+) for each eleroent x
of E and each positive linear functional f on E. In view of theorero
23.9, this last equation becornes f(x+) = sup {h(x): 0 ~ h ~ f},
which can be established as follows. For each linear functional h
which satisfies 0 ~ h ~ f, h(x) ~ h(x+) ~ f(x+); hence, sup {h(x):
0 ~ h ~ j} ~ f(x+). Next, by applying the lernrna to f and x+,
we see that there is a linear functional g on E such that 0 ~ g ~ j,
g(x+) = f(x+), and g(y) = 0 foreachpositive y suchthat y 1\ x+ =
0. Since x+ 1\ x- = 0, it follows that g(x) = g(x+) - g(x-) =
g(x+) = f(x+). Hence,j(x+) = sup {h(x): 0 ~ h ~ f} which proves
the theorern.lll
The evaluation rnap of a vector lattice E into its second order dual
is not necessarily one-to-one, for, if this were the case, every vector
lattice would be Arehirneclean since the order dual of a vector lattice
is always Archirnedean. However, being Arehirneclean alone does not
 SEc. 23 ÜRDERED LINEAR SP AcEs 235

ensure that the evaluation map be one-to-one. Consider, for in-

stance, the space of all real valued measurable functions on the
interval [0, 1] modulo null functions (functions which are zero almost
everywhere) under the usual ordering. This space is an Arehirneclean
vector lattice, but it can be proved that the order dual is zero dimen-
sional. In order to give some necessary and sufficient conditions for
the evaluation map of a vector lattice to be one-to-one, we require
some preliminary facts.
For arbitrary u and v in a vector lattice E, obviously (u + v)+ ~
u+ + v+ holds. Hence for any x and y, x+ ~ (x - y)-t- + y+ or
x+ - y+ ~ (x - y)+. It follows that (x+ - y+)+ ~ (x - y)+.
By interchanging x and y, we see that (x+ - y+)- ~ (x - y)-.
By adding the last two inequalities, we obtain Ix + - y + I ~ I x - y I·

23.14 CoNTINUITY OF THE LATTICE OPERATIONS Let E be a vector

lattice and Iet §' be a locally convex topology for E such that the set of
alt continuous pseudo-nornis admits a base consisting of lattice pseudo-
norms. Then the lattice operations on E are continuous relative to the
topology 3, that is, the functions M and J defined by M(x,y) = x 1\ y
and J(x,y) = x V y are continuous on E x E into E. In particular the
lattice operations are continuous relative to the Mackey topology m(E,E*).
PROOF It can be seen easily that the functions J and Mare continuous
if and only if the function which takes x to x+ is continuous. If p is a
lattice pseudo-norm, then from the remark preceding the theorem
one obtains p( x + - y +) ~ p( x - y) for all x and y; hence the lattice
operations are continuous relative to the pseudo-norm topology
determined by p. The theorem follows.lll

23.15 ÜNE-TO-ÜNENESS OF THE EvALUATION Let E be a vector

lattice. Then the foltowing statements are alt equivalent.
(a) The evaluation map on E into the second orderdual is one-to-one.
(b) There are enough positive linear functionals to distinguish points
of E.
(c) The topology w(E,E*) is Hausdorff.
(d) The topology m(E,E*) is Hausdorff.
(e) For some locally convex Hausdorff topology the map which takes
x to x+ is continuous.
(f) The positive cone is closed for some locally convex topology on E.
PROOF (a) <c> (b) ~ (c) ~ (d) is obvious. (d) implies (e) because of
the previous theorem. ( e) implies (f) because the positive cone is the
236 APPENDIX: ÜRDERED LINEAR SPACES

set {x: x = x+}. That (f) implies (b) is a consequence of theorem

23.1.111

24 L AND M SPACES
We prove two theorems of Kakutani which characterize Banach
lattices of (roughly speaking) continuous function type, and Banach
lattices of U type. The adjoint of an M space is an L space, and
dually, the adjoint of an L space is an M space.
In this section two especially important vector lattices will be
examined. The principal theorems concern vector lattices with a
lattice norm-that is, a normsuch that, if lxl ~ IYI, then llxll ~ IIYII·
Such spaces are called normed lattices and, if the space is complete
relative to the norm, they are Banach lattices.
Many, if not most, of the real Banach spaces which are commonly
considered are Banach lattices under some natural ordering. For
example, if m is a measure on a a-ring !/ of subsets of a set X and
p ~ 1, then the space LP( m) of all !/-measurable real functions f such
that IJIP is m-integrable is a Banach lattice, relative to the usual
ordering of the real functions and with the norm: II/II = (f IJIPdm) 11 P.
(Measure theoretic terminology is that of Halmos [4].) The space
L"'(m) of all m-essentially bounded !/-measurable real functions is
also a Banach lattice; it has the noteworthy features that II/ V gll =
max [11/11, llgll] for all f ~ 0 and g ~ 0, and the function which is
constantly one is interior to the positive cone. The space of all
continuous real valued functions on a compact Hausdorff space is
also a Banach lattice, if the norm is the usual supremum norm, and
this space shares the two special properties of L "'(m) noted above.
The space V(m) also possesses a special property: if f and g are
positive members of V(m), then II/ + gll = II/II + JJgJJ. If X is
a-finite relative to the measure m, then there is another description of
(an isomorph) of V(m) which is useful. Each countably additive
real function on !/ which is of bounded total variation and is absolutely
continuous with respect to m is, by the classical Radon-Nikodym
theorem, theindefinite integral of a member of V(m), and the space
of all such countably additive functions, with variation for norm, is
isomorphic and isometric to V(m). It will be shown that a Banach
lattice is essentially of this sort if it is true that II/ + gll = 11/IJ +
JJgJJ for all positive elementsfand g.
It is convenient, before proceeding, to establish a connection
between ordering and extreme points, preparatory to an application
 SEC. 24 LAND M SPACES 237

of the theorem 15.1 on the existence of extreme points. The following

lemma is given in a form somewhat more general than is necessary here.

24.1 LEMMA Let E be an ordered linear topological space, Iet f be a

positive linear functional on E, and Iet A be a subset of the positive cope
of E such that:
(i) the set A is compact and convex, and f(x) ~ 1 for each x in A.
(ii) if x E A and x ;;;; y ;;;; 0, then y E A, and
(iii) if x E A and x =I= 0 then f(x) =I= 0 and xff(x) is a member of A.
Then a member x of A is an extreme point of A if and only if for each
y such that x ;;;; y ;;;; 0 it is true that y = f(y)x.
PROOF Suppose that x is an extreme point of A and that x ;;;; y ;;;; 0.
If x = 0, then f(x) = f(y) = 0; by (iii) it follows that y = 0, and
hence y = f(y)x. If x =1= 0 then, since f(x) ~ 1 and both 0 and
xff(x) belong to A, f(x) = 1; otherwise x could not be an extreme
point. If x =I= 0 and x - y = 0, then again y = f(y)x. If neither
y nor x - y is 0, then f(y) + f(x - y) = 1, and writing x =
f(y)[yff(y)] + f(x - y)[(x - y)ff(x - y)], the fact that xisextreme
implies that yff(y) = x. To prove the converse, suppose that x is a
member of A such that y = f(y)x for each y such that x ;;;; y ;;;; 0.
If x = 0, then since 0 is the unique member of A which is a zero of
f, and since f is positive, x is an extreme point. If x =I= 0 then
f(x) = 1 because x = f(x)x. In this case, if x = ty + (1 - t )z
where 0 < t < 1 and y and z belong to A, then clearly x ;;;; y ;;;; 0;
hence y = f(y)x and similarly z = f(z)x. Finally, since f(x) =
1 = tf(y) + (1 - t )f(z), and since f(y) and f(z) are each at most
one, it follows that f(y) = f(z) = 1, and hence y = f(y)x = x and
z = x. It follows that x is an extrem~ point.lll
A linear function f on a vector lattice E to another vector lattice F
is a lattice homomorphism if and only if f(x V y) = f(x) V f(y)
for all x and y in E. A one-to-one lattice homomorphism is a lattice
isomorphism. A real lattice homomorphism is a homo-
morphism whose range is contained in the lattice of real numbers,
with the usual ordering. It follows easily from the relation between
meet and jointhat a lattice homomorphism also preserves meets, and
that a meet preserving linear function is necessarily a homomorphism.
It is clear that a real lattice homomorphism is a positive functional,
for, if x is a positive element of E, then x = x V 0 and f(x) =
f(x) V /(0) ;;;; 0. The following theorem identifies the lattice
homomorphisms geometrically.
238 APPENDIX: ÜRDERED LINEAR SPACES
24.2 CHARACTERIZATION OF LATTICE HoMOMORPHISM If E is a
vector lattice and f is a linear functional on E, then the following three
conditions are equivalent:
(i) the functional f is positive and, if x and y are disjoint positive
elements of E, then f(x) (\ f(y) = 0;
(ii) the functional f is a lattice homomorphism; and
(iii) the functional fispositive and, if g isalinear functional suchthat
0 ~ g ~ J, then g is a scalar multiple of f.

PROOF It is first shown that (i) implies (ii). Assuming (i), suppose
that x and y are arbitrary positive members of E. Then x - x (\ y
and y - x (\ y are disjoint positive elements and consequently
0 = (f(x) - f(x (\ y)) (\ (f(y) - f(x (\ y)) = f(x) 1\ f(y) -
f(x 1\ y), the latter equality being a consequence of the translation
formula for meets: (a + b) (\ (c + b) = a 1\ c + b. It follows that
f preserves meets of positive elements, and, using the translation
formula again, it is easy to see that meets of arbitrary elements are
preserved.
To show that (ii) implies (iii) assume that f is a lattice homo-
morphism, that 0 ~ g ~ J, and that x is a point of E at which f
vanishes. It will be shown that g also vanishes at x, and hence, since
the null space of g includes the null space of J, it will follow that g is a
scalar multiple of f. Write x = x+ - x and notice that, since x+
and x- are disjoint, either f(x+) or f(x-) is 0. Since f(x) = 0 also,
f(x+) = f(x-) = 0, and since 0 ~ g ~ j, it follows that g(x+) =
g(x-) = 0 and therefore g(x) = 0.
Finally, it must be proved that (iii) implies (i). Suppose that x
and y are disjoint positive elements of E and that f(x) "#- 0. The
application of lemma 23.12 to fand x ensures the existence of a linear
functioaal g such that 0 ~ g ~ J, g(x) = f(x) and g(y) = 0. The
first two properties of g and (iii) imply that f = g; hence, f(y) =
g(y) = 0. Itfollowsthatf(x) 1\J(y) = Owhetherf(x) = Oornot.lll
The two types of normed lattices which are the concern of this
section satisfy especial conditions on the norm and the ordering. A
normed lattice is of type M (respectively of type L) if and only if for
every pair x and y of positive elements it is true that llx V Yll =
llxll V IIYII (respectively, llx + Yll = llxll + IIYII)· A Banach
lattice of type M (of type L) is an M space (respectively, an L space ).
These two sorts of normed lattice are, in a sense, dual to each other.
Each is studied by means of its adjoint space, which turns out to be a
space of the dual sort, as noted in the following theorem.
 SEC. 24 LAND M SPACES 239

Before turning to the theorem it should be stated that for

any normed lattices each continuous linear functional is the dif-
ference of continuous positive functionals, by 23.6; it is not true
that, for an arbitrary normed lattice, each positive functional is norm
bounded. For Banach lattices, each positive linear functional is
bounded; however this fact is not needed for the present discussion.
The largest element in the unit sphere of a normed lattice, if such
exists, is called the unit.

24.3 ADJOINTS OF SPACES OF TYPELAND M The adjoint of a normed

lattice is, with the dual ordering, a Banach lattice. The adjoint of a
space of type M (respectively, of type L) is an L space (an M space with
unit).
PROOF If E is a normed lattice, then surely E* is a Banach space. If
f is a continuous positive linear functional on E, then 11!11 = sup {j(x):
x f; 0 and llxll ~ 1} because llxll = II lxlll f; llx+ II V llx-11 and
one of the two numbers f(x+) and f(x-) is necessarily as great as
lf(x)l. Consequently, if fand g are bounded positive functionals,
then f 1\ g is also bounded, and it follows that E* is a lattice. More-
over, since the meet of fand g in the order dual E* is identical with
the meet in the adjoint E*, it follows from the calculation made after
23.9 that, for a positive element x, lfl(x) = sup {j(y): IYI ~ x}.
Hence, 111!1 II = sup {lfl(x): x f; 0 and llxll ~ 1} = sup {j(y):
IIY II ~ 1} = II f II· Since the norm of E* is clearly monotone, the
adjoint E* is a normed lattice.
Finally, it must be shown that the adjoint of a space of type M
(type L) is an L space (an M space with unit). If Eis of type M andf
and g are positive members of E*, then there arepositive members x
and y of E of norm 1 such thatf(x) and g(y) are, respectively, approxi-
mately IIJII and llgll· Then x V y is positive and of norm 1, and
(f + g)(x V y) f; f(x) + g(y). Hence llf + gll f; 11!11 + llgll, and it
follows that the adjoint is an L space. If E is of type L, then the
norm is linear on the positive cone of E, and consequently there is a
linear functional u such that u(x) = llxll for x ;;;; 0. For x positive
and f in E* it is then true that f(x) ~ 11!11 llxll = 11!11 u(x), and
hence f ~ I!JIIu. It follows that u is a unit, and, moreover, for
positive elements f and g of E*, because of the inequality f V g ~
(11!11 V llgll)u, it is true that llf V gll ~ (11!11 V llgll)llull = !I !II V llgll·
Therefore E* is an M space with unit.lll
If E is a normed lattice, then E** is a Banach lattice. Let e be the
evaluation map on E into E**; then, for each x in E and positive
240 APPENDIX: ÜRDERED LINEAR SPACES
element f of E*, e(x)+(j) = sup {g(x): 0 ~ g ~ f and g E E*} =
sup {g(x): 0 ~ g ~ f and g E E*} = f(x+) = e(x+)(j) (the third
equality follows from 23.13), or e(x)+ = e(x+). Hence, we have
proved the following:

24.4 THE EVALUATION MAP Let E be a normed lattice: then the

evaluation map on E into its second adjoint E** is a lattice isomorphism.
If the image of the evaluation map on a normed lattice E into E**
is again denoted by E, then the closure E - in E** of E relative to
the norm topology has the property that, whenever x and y belong to
E -, x (\ y and x V y also belong to E -. This fact is a consequence
of the continuity of lattice operations in E** (see 23.14). Hence E-is
a Banach lattice in which the given normed lattice E is densely
embedded.
The structure of an M space with unit will now be described
completely. If Eis a normed space of type M which has a unit u,
the spectrum X of Eis defined to be the set of all real homomorphisms
of E which are of norm one, with the weak* topology. It is evident
that the set of all real lattice homomorphisms is weak* closed and,
since X is precisely the set of those homomorphism h such that
h(u) = 1, it follows that Xis a weak* closed subset of the unit sphere
of E*. (A homomorphism h is positive, and hence its norm is the
supremum of values at positive elements of norm at most one, and
hence //h/1 = h(u).) The spectrum is then a compact Hausdorff
space. The normed lattice E may be mapped into the lattice C(X)
of continuous real valued functions on X by nieans of the evaluation
map cp. Explicitly, for x in E and h in X, cp(x)(h) = h(x). For x
and y in E and h in X it is true that cp(x /\ y)(h) = h(x /\ y) =
h(x) (\ h(y) = (cp(x) (\ cp(y))(h), because h is a lattice homomor-
phism. Hence cp is a lattice homomorphism of E into C (X). It will
now be shown that cp is an isomorphism, and an isometry, and that
the range of cp is densein C(X).

24.5 REPRESENTATION OF SPACES OF TYPE M Let E be a normed

lattice of type M with unit, and Iet X be the spectrum of E. Then the
evaluation map cp of E into the function space C(X) is an isometric
lattice isomorphism of E onto a dense subspace of C(X).
PROOF In view of the remarks preceding the theorem there remain
just two facts to be demonstrated: that cp is an isometry, and that the
range of cp is densein C(X). In order to show that cp is an isometry
 SEc. 24 L AND M SPACES 241

it is necessary only to show that Jlxll = l <f>(x)JI for positive members

x of E, since both E and C(X) are normed lattices. Since each
member of X is of norm one, sup {J<f>(x)(h)J: hEX} = sup {Jh(x)J:
hEX} ~ llxll· On the other hand, given a positive member x, let A =
{g:g;;;; 0 and JJgJJ ~ 1} and B = {g:gEA and g(x) = llxll}; then A
is weak* compact and Bis non-void and, since B is the set of members
of A where the evaluation at x is a maximum, Bis a support of A (that
is, if B contains an interior point of a line segment in A, then B
contains the entire segment). It follows that each extreme point of B
is an extreme point of A, and, in view of theorem 15.1, B has extreme
points. Consequently there is an extreme point h of A such that
h(x) = Jlxll· Finally, lemma 24.1 may be applied (the requisite
linear functional on A is evaluation at the unit of E), and 24.2 then
shows that h is a lattice homomorphism. lt follows that llxll =
ll<f>(x)ll·
lt remains to prove that the range of 4> is densein C(X). Forthis
purpose the following prelemma, which is of interest in itself, is
useful;

PRELEMMA Let X be a compact space and Iet A be a subset of C(X)

which is closed under the lattice operations (that is, j, g E A implies
f V g E A and f 1\ g E A). Then a function h in C (X) belongs to the
uniform closure of A if for each e > 0 and for each pair of points x and
y of X there is an f in A such that lf(x) - h(x)J < e and lf(y) -
h(y)J < e.
Suppose for a moment that the prelemma is established. Notice
that, if h and g are distinct members of X, then for some x in E,
h(x) =F g(x), and hence <f>(x), assumes different values at the members
g and h of X. It is also true that the image under cfo of the unit u of
E is the function which is constantly one on X. Therefore, given a
pair of real numbers a and b, it is possible to choose a linear combina-
tion y of x and u such that <f>(y)(g) = a and <f>(y)(h) = b. Also the
range of 4> is closed under the lattice operations; hence, in view of the
prelemma, the range of 4> is densein C(X).

PROOF OF THE PRELEMMA Let a positive number e be given, and let

fx.Y be a function in A such that lfx.Y(x) - h(x)J < e and lfx.Y(y) -
h(y)J < e. If Ux,y = {z: l!x,y(z) - h(z)J < e}, then clearly Ux,y is an
open set containing x and y. Since Xis compact, for a fixed x, there
are points y 1 , · · ·,yn such that X= U {Ux,y,: i = 1, · · ·, n}. Let
gx = fx.y 1 1\ · · · 1\ fx.Yn; then gx(y) ~ h(y) + e for each y in X and
242 APPENDIX: ÜRDERED LINEAR SP ACES
gx(y) ~ h(y)- eforeachyin Vx = n
{Ux,y,: i = 1, · · ·, n}. Since
Vx is an open set containing x and Xis compact, there are Xu · · ·, Xm
suchthat X= U {Vx,: i = 1, · · ·, m}. Let g = gx 1 V ···V gxm'
Then clearly g E A and h(y) - e ~ g(y) ~ h(y) + e for eachy in X.[[[

24.6 CHARACTERIZATION OF M SPACES Each M space with unit is

isomorphic and isometric, under evaluation, to the space of all continuous
real valued functions on its spectrum.
The preceding discussion of spaces of type M gives insight into the
structure of an Arehirneclean vector lattice E whose positive cone is
radial at some point. For if the cone is radial at a point u, then one
may construct a norm for E such that E, with this norm, is a space of
type M with unit u. In fact:

24.7 THEOREM lf E is an Archimedean vector lattice and the positive

cone is radial at a point u, then E with the norm [[x[! = inf {t: t > 0,
[xj ~ tu} is a space of type M with unit u.
PROOF First observe that the infimum in the definition of the norm
is assumed, because, if one lets s = [[x[l, then, for any positive number
t, Jxj - su ~ tu or (jxj - su)+ ~ tu and it follows from the fact that
the ordering is Arehirneclean that (lxl -su)+ = 0 or Jxl ~ su. From
this remark all the assertions in the theorem can be seen easily once
it is shown that I I is a pseudo-norm. The function I I is a
pseudo-norm because it is precisely the Minkowski functional of the
set (u- C) n (C-u) which is convex, circled, and radial at 0.[1!
Spaces of type L will now be studied, utilizing the known structure
of M spaces. This mode of attack, while perhaps not the most
direct, yields a rather concrete representation theorem and has the
advantage that elementary measure theory is available for some of
the proofs. The procedure is the following. If E is a space of type
L, then E* is an M space with unit, according to 24.5, and hence is
isomorphic to the space C (X) of continuous real functions on its
spectrum X. The evaluation map of E into E** preserves the norm
and the lattice operations. But a simple concrete representation for
the adjoint of C (X) is given by the Riesz representation theorem
(14J), for, to an isomorphism, C(X)* is the space of signed regular
Borel measures on X, with variation for norm. Explicitly, the Borel
a-ring fJ9 of Xis the smallest a-ring which contains each compact set.
A signed regular Borel measure m is a countably additive real valued
function on fJ9 such that for each Borel set B and a positive number e
 SEC. 24 L AND M SPACES 243
thereisacompactsubsetofCof Bwiththeproperty lm(B)- m(C)I < e,
and the variation llmll of m is sup {m(A) -m(B): A and B disjoint
members ol &l'}. The map which carries m into the functional on
C (X) whose value at I is f ldm is an isometry, and order preserving.
(See Halmos [4] for details.) The following theorem summarizes
the foregoing remarks.
24.8 REPRESENTATION OF SPACES OF TYPE L Let E be a space ol type
L, Iet E* be its adjoint, and Iet X be the spectrum ol E. Then E is
isometric and lattice isomorphic to a sublattice ol the space ol regular
Bore/ measures on X.
The characterization of L spaces will now be completed by describ-
ing precisely the dass of regular Borel measures on X which are the
images of Ih~mbers of E. The proof of the following theorem
depends on a sequence of lemmas, which are given after the statement
of the theorem.
24.9 CHARACTERIZATION OF L SPACES Let E be an L space, and Iet
X be the spectrum ol its adjoint E*. Then E is isometric and lattice
isomorphic to the space ol all those signed regular Borel measures on X
which vanish on each Bore/set ol the first category.
Throughout the following E will be a fixed L space, E* its con-
jugate, and X will be the spectrum of E*. To avoid notation, no
distinction will be made between E* and the space C(X) of real
continuous functions on X. Jf x is a member of E, then mx will
denote the corresponding signed measure on X.
24.10 LEMMA A subset ol E or ol E* which has an upper bound has a
least upper bound. Each monotonically increasing net in E (respectively
E*) which is bounded above converges relative to the norm topology (the
w*-topology) to its least upper bound.
PROOF The second assertion will be proved first. lf {Ja, a E D} is a
monotonically increasing net in E* which is bounded above by g,
then foreachpositive x in E the net {Ja(x), a E D} is a bounded mono-
tonically increasing net of real numbers, hence converges, and there
is therefore a linear functional I such that Ua(x), a E D} converges to
l(x) for all positive x in E; hence Ia--+ I relative to the topology w*.
For each positive x, la(x) ;;;;; l(x) ;;;;; g(x), from which it follows that I
is bounded, and it is evident that I is the supremum of the functionals
Ia· To prove that a bounded monotonic net {xa, a E D} in E con-
verges relative to the norm topology it is first shown that the net is a
244 AJ>PENDIX: ÜRDERED LINEAR SPACES

Cauchy net. Observe that for a fixed, 1Jx0 - xall is monotonic for
ß~ a, and bounded, and hence converges. Since the norm is linear
on the positive cone, for y ~ ß ~ a it is true that llxr - x0 l =
llxr - xall - llx 0 - xall· From this it follows easily that the net is
Cauchy and hence converges to a member x of E. It is not hard to
verify that x is the supremum of the members Xa, by making use of
the fact that the set {z: z ~ y} is always closed. Finally, to prove the
first Statement of the Iemma, if B is a subset of E (or of E*) with an
upper bound, then Iet d be the family of finite subsets of B directed
by ::::>, and for A in d Iet xA be the supremum of the set A. Then
the net {xA, A E d} is monotonically increasing and bounded, and its
Iimit is the supremum of B.lll
The preceding Iemma has important consequences which concern
the topological structure of the spectrum X and the nature of the
measures mx.
24.11 LEMMA The closure of each open set in X is both open and
closed. For each x in E, the measure mx vanishes on Borel sets of the
first category.
PROOF Let U be an open set in X, Iet B be the family of all non-
negative continuous real functions which are 0 on X ~ U and are
bounded by 1, and Iet f be the supremum of the set B. Then f is a
continuous functirm which is 1 on U, because for each s in U there
is a member of B which assumes the value 1 at s. Qn the other hand,
if s f/3 U - then there is a function g which is an upper bound for B
such that g(s) = 0. It follows that f must be the characteristic
function of U-, and hence U - is both open and closed. To prove
the second statement of the Iemma it is necessary only to show that
mx(S) vanishes for each nowhere dense Bore! subset S of X, and it
may be assumed that x is a positive member of E. Let B be the
family of all continuous real functions on X which are bounded by 1,
are non-negative, and are zero on S. Then B is directed by ~ , and,
since S is nowhere dense, it is easy to see that the supremum of B is
the function which is identically one. The net {!, f E B} converges
w* to 1, and hence Jfdmx-+ f1dm:r. For e > 0 there is then f in B
such that e > f (1 - f)dm:r ;:;; 0, and since (1 - f)(s) = 1 for s in
S, m:r(S) < e.[[l
The second statement of the preceding Iemma shows that the L
space E maps into a sublattice of the space of all signed regular Bore!
measures which vanish on first category Bore! subsets of X. It
remains to show that E maps onto the latter dass. The following
 SEC. 24 LAND M SPACES 245
method of attack is used. Suppose n is a positive regular Borel
measure which vanishes on sets of the first category. Let B be the
set of measures of the form mx such that 0 ;:;;; mx ;:;;; n, and let p be the
supremum of B. This supremum exists, in view of lemma 24.10
applied to the L space E**, and the suprema of finite subsets of B
converge top relative to the norm topology. Since E is isometric to
a closed subspace of E** it follows that p = my for some y in E. If
p = n then n belongs to the image of E; otherwise n- pisapositive
regular Borel measure, zero on sets of the first category, such that no
non-zero positive mx is less than n - p. The following lemma then
completes the proof.

24.12 LEMMA lf n is a positive regular Bore/ measure vanishing on

Bore/ sets of the first category and llnll # 0, then for some positive
non-zero x in E it is true that mx < n.
PROOF As a preliminary, it is to be observed that, if A is a Borel
subset of X, then there is an open and closed subset A' of X suchthat
the symmetric difference (A "' A') U (A' "' A) is of category one.
This may be proved by showing that the class .91 of all sets A such
that, for some open set C, the symmetric difference of A and C is of
first category, is closed under countable union and complementation,
and contains all compact sets and hence all Borel sets (this is a well-
known lemma of set theory). Since each open set C differs from
the open and closed set C - by a nowhere dense set, the stated result
follows.
Turning to the proof of the lemma, it is first shown that if p is a
positive measure belanging to E (more precisely, if p = mx for some
positive member of E) and if A is a Borel subset of X, then the
measure PA belongs to E, where PA(B) = p(A n B) for each B.
Since p vanishes on first category sets there is, in view of the remark
above, no lass in generality in assuming A is open and closed. Let
f and g be the characteristic functions of A and X "' A, respec-
tively. Then, in view of the remark after 23.9, 0 = (f 1\ g)(p) =
inf {f(p - q) + g(q): 0 ;:;;; q ;:;;; p, q E E}. Hence, for a positive e,
there is q in E such that 0 ;:;;; q ;:;;; p, (p - q) (A) < e, and q(X "' A)
< e. Thus qA ;:;;; PA and llq -PAli = l qA -PA + q x-AII < 2e.
Since E is closed and e arbitrary, PA E E.
Suppose now that n is a positive regular Borel measure, not identi-
cally zero, which vanishes on first category Borel subsets of X. Let
U be the union of all open subsets of X with n-measure zero. Since
246 APPENDIX: ÜRDERED LINEAR SP ACES
n vanishes on each compact subset of U, n( U) = 0. It follows that
n( U -) = 0. Let A be the complement of U - in X; then A is open
and closed, and it has the properties: n(X "' A) = 0, and if B is a
non-void open subset of A, then n(B) > 0. From this fact it follows
that a Borel subset B of A is of category one if and only if n(B) = 0,
for given such a set B there is an open set B' such that (B "' B') U
(B' "' B) is of category one, hence n(B) = n(B') = 0, therefore B' is
disjoint from A, and consequently B is a subset of the first category
set B "' B'. It follows that a Borel set B in Xis of n-measure 0 if
and only if B n A is of the first category.
Finally, let f be the characteristic function of A, and choose a
positive measure p in E such that I fdp =P 0. Then, applying the
result of the second paragraph, pA E E, and clearly PA =P 0. In view
of the characterization of the null sets of n, PA is absolutely con-
tinuous with respect to n, and by virtue of the Radon-Nikodym
theorem there is a non-negative Borel function g such that PA(B) =
I 8 gdn for any Borel set B in X. For some positive integer r the set
C = {x: g(x) ~ r} has a positive PA-measure. Then PAncis a non-
zero element of E and is dominated by rn; the Iemma is proved.!ll

24.13 NoTES A few general remarks on methodology may clarify

the representation problem. The obvious method of representation,
by means of real lattice homomorphisms, is essentially equivalent to
embedding the lattice in a lattice of real functions, by virtue of the
argument given in this section. This method is completely successful
for M spaces. However, there are lattices, such as L 1 relative to
Lebesgue measure on the unit interval, for which there are no real
homomorphisms. A possible mode of attack is to seek lattice homo-
morphisms into the extended reals. Another possible attack is
suggested by the fact that many lattices have enough order-continuous
positive linear functionals to distinguish points (a functional is order-
continuous if the supremum of the values on a monotone increasing
sequence is the value at the supremum-a suggestive statement of this
requirement: the functional satisfies the Lebesgue bounded con-
vergence theorem).
The fact that the spectrum of the adjoint of an L space is totally
disconnected suggests a strong connection with the theory of Boolean
algebras, and such a connection does, in fact, exist.
The Boolean algebra of open and closed subsets of the spectrum of
the adjoint of an L space is actually of very special kind; it is always
isomorphic to the Boolean algebra of measurable sets modulo sets of
 SEC. 24 L AND M SPACES 247
measure zero for a suitably chosen measure. The adjoint of an L
space is isomorphic to the space of linear operators which preserve
absolute continuity and it also permits a representation as the essen-
tially bounded functions relative to some measure.
Much of the motivation for the study of L spaces was derived from
the applications to the theory of Hermitian operators on Hilbert space.
In the terminology of this section the essential content of the spectral
theorem may be stated: the smallest norm (respectively, strongly)
closed real algebra of operators containing a given Hermitian operator
is, under the natural order, an M space whose spectrum is homeo-
morphic to the spectrum of the operator (respectively, is the adjoint
of an L space).
 BIBLIOGRAPHY

[1] Banach, S.,

Theorie des operations lineaires, Monografje matematyczne,
W arszawa, 1932.
[2] Bochner, S. and Martin, W. T.,
Several Camplex Variables, Princeton University Press, Princeton,
N. ]., 1948.
[3] Bourbaki, N.,
Espaces vectoriels topologiques, Actualites Scientifiques et Indus-
trielles II 89 and 1229, Hermann, Paris, 1953 and 1955.
[4] Halmos, P. R.,
Measure Theory, D. Van Nostrand Co., Inc., Princeton, N. ].,
1950.
[5] Kelley, J, L.,
General Topology, D. Van Nostrand Co., Inc., Princeton, N. ].,
1955.

248
 LIST OF SYMBOLS
(Page nurober indicates where symbol first appears)

A, 27 (E,F), 137 ffb, 183

A-, 27 E ® F, 152 ffs, 212
A
A\ 66 E®F, 153 T(S,E), 70
V
(A), 14 E® F, 180 w, 153
(A), 13 c<m>, 82 w*, 155
.s;~-, 181 (E,ff)*, 119 w(E,F), 138
.91-, 181 _L{Et: tEA}, 8 w(F,E), HO
Ai, 27 X {Et: t E A}, 8 X j_ A, 66
A 0 , 141 F(S,E), 68 lxl, 230
A(U), 196 KB(X), 81 x+, 230
Bs~(S,E), 70 K(X), 81 x-, 230
B 0 , 141 L(E,F), 74 {x, ~}, 28
B(S,E), 69 lim ind {Et: t E A}, 10 {xa, A, ~ }, 28
IX E

B(X), 81 lim proj {Et: t E A}, 11 {xa, IX E A}, 28

C*, 225 LP(p.), 54 Xa--+y, 28

C(S,E), 72 LP(X,p.), 54 [x:y], 13
C(X), 81 m(E,F), 173 (x:y), 13
C0 (X), 81 N(A,U), 68 [x:y), 13
!!)K, 82 s(E,F), 168 (x:y], 13
E', 8 S(X,p.), 55 (x,y), 137
E*, 51, 119 T', 199 X@ y, 152
E**, 189 T*, 204 X 1\ y, 229

E*, 225 .9;, 69, 1# X V y, 229

c, 82 .9d,!?4• 179 (X,ff), 27

Eb, 183 T s~(S,E), 71 y = lim Xa, 28
249
 INDEX

absolute, G 5, 96 bounded (Cant.)

value, 230 function - on a set, 69
absorb, 90 linear function, 45
absorption theorem, 90 pointwise, 103
adjoint, of a linear topological space, uniformly, 102
119
of a pseudo-normed space, 51 canonical, isomorphism, 120
transformation, 204 map, 137
admissible family, 167 categorical imperative, 1-247
generated by .9/, 168 category, theorems, 83 et seq.
algebra (of sets), 157 first, 84
algebraic, closure, 42 at a point, 85
dual, 8 second, 84
annihilator, 119 at a point, 85
approximation theorem, 144 circled, 14
Ascoli theorem, 81 extension, 14
atom, 56 closable map, 208
closed, graph theorem, 97, 106, 116
Baire, condition of, 92 relation theorem, 101
theorem, 86 map, 208
Banach, space, 58 set, 27
subgroup theorem, 93 sphere, 29
Banach-Alaoglu theorem, 155 closure, 27
Banach-Steinbaus theorem, 104
duster point, 28
barrel, 104
co-base, 113
theorem, 104
Co-dimension, 5
base, for a family of pseudo-norms,
232 complementary, 5
local, 31 complete, 56
theorem, 34 metric space, 32
for a neighborhood system, 27 relative to a pseudo-metric, 85
bilinear, functional, 106, 137 sequentially, 58
mapping, 106 topologically, 96
Borel set, 126 completely continuous map, 79, 206
bornivore = bound absorbing completion, 62
bound absorbing, 182 condensation theorem, 85
bounded,44 cone, 16
elementary facts on boundedness, 45 dual, 225
251
252 INDEX

cone (Cont.) extension theorem, 21, 117

generated by a set, 17 extreme, half-line, 133
generating, 225 point, 130
normal, 227
positive, 16 factor space = quotient space
proper, 19 finite intersection property, 59
reproducing, 225 ßat = linear manifold
conical extension, 17 Frechet space, 58
conjugate = adjoint frequently, 28
contain small sets, 59 function, absolutely homogeneous, 15
contraction, 33 additive, 157
ronverge, 28 analytic, 162
convergence in mean of order p, 54 completely continuous, 79, 206
convex, 13 continuous, 28
body, 110 at a point, 28
envelope = - extension on a subset, 28, 73
cover = - extension interior = open -
extension, 13 locally measurable, 129
hull = - extension lower semi-continuous, 94
coset space = quotient space non-negatively homogeneous, 15
open, 28
deficiency = co-dimension relatively open, 28
dense, 27 subadditive, 15
difference, space = quotient space topological = homeomorphism
theorem, 92 uniformly continuous, 38
dimension, 4 upper semi-continuous, 94
direct ( verb), 28 vanishing at infinity, 81
direct, sum, 8 weakly analytic, 162
system = inductive system
graph, 38
disjoint elements (in a lattice), 230
Grothendieck's completeness theorem,
disk = barre!
145
distinguish points, 109, 138
distribu tion, 197 Hahn-Banach theorem, 21
dual, ( of a linear topological space) half-space, 19
adjoint complementary, 19
base = co-base Harne! base, 3
ordering, 225 Hausdorff uniformity, 116
transformation, 199 Hausdorff's theorem on total bounded-
ness, 61
ecart = pseudo-metric Helly's, theorem, 18
embedding, 47 condition, 151
equicontinuous, 73 Hilbert space: 65
at a point, 73 orthogonal complement, 66
Euclidean, complex, 3 orthogonal, 66
real, 3 orthonormal, basis, 67
eventually, 28 subset, 67
exhaustible = of the first category ( see projection, 66
category) summable, 66
 INDEX 253
homeomorphism, 28 linear, extension, 5
Hurewicz, W., 88 function ( = map, mapping, trans-
hypercomplete, 116 formation), 5
hyperplane, 19 functional, 6
hull = - extension
idempotent operator, 40 invariant, 6
induced map theorem, 6, 40 isomorphism = a one-to-one linear
inductive limit, 10 function, not necessarily onto, 6
strict, 164 manifold, 5
inductive system, 10 relation, 101
inexhaustible = of the second category space, 1
(see category) span = - extension
injection ( into a direct sum), 9 variety, 5
inner product, 54 linear topological space, 34
interior, 27 barrelled, 104, 171
inverse, limit = projective limit bornivore = bound -
system = projective system bound, 183
disk = barrelled -
James, R. C., 198 evaluable, 192
JOtn ( oftwo sets), 14 fully complete, 178
join (in a lattice), 229 infra-tonnele = evaluable-
locally bounded, 55
k-space, 81 locally convex, 45
Kakutani's lemma, 17 Mackey, 173
kerne! = null space Montel, 196
kerne! (of an inductive system), 10 normable, 43
pseudo-normable, 44
Köthe space, 220
quasi-barrelled = evduable -
Krein-Milman theorem, 131
quasi-tonnele = evaluable -
Krein-Smulian theorem, 177, 212
quotient, 39
reflexive, 191
L space, 238
semi-reflexive, 189
characterization, 243
symmetric = evaluable -
lattice, Archimedean, 229 tonnele = barrelled -
Banach, 236 linear topological subspace, 39
homomorphism, 237 locally convex topology, 45
real, 237 derived from f7, 109
isomorphism, 237 strongest, 53
normed, 236 locally convex set, 42
type L, 238
representation, 243 M space, 238
type M, 238 characterization, 242
representation, 240 map = function
vector, 229 mapping = map
LF space, 218 maximal linear subspace, 7
generalized, 218 meager = of the first category (see
limit, 28 category)
line segment, 13 measure, 126
254 INDEX

measure (Cont.) pairing (Cont.)

Borel, 126 natural, 138
complex, 126 separated, 138
localizable - space, 129 partial ordering, 16
regular, 126 translation invariant, 16
simple, 186 partially ordered linear space = or-
Ulam, 186 dered linear space
meet, 229 polar, 141
metric, 29 computation rule, 141
associated with a norm, 43 positive, cone, 16, 225
Hausdorff, 33 element, 225
space, 29 part, 230
complete, 32 precompact = totally bounded
topology ( see topology) projection, 8, 40
metrization theorem, 48 projective, limit, 11
midpoint convex, 17 system, 11
Minkowski functional, 15 pseudo-metric, 29
absolutely homogeneous, 48
negative part, 230 invariant, 48
net, 28 space, 29
Cauchy, 56 topology (see under topology)
relative to a pseudo-metric, 57 pseudo-norm, 15
equicontinuous, 73 equivalent, 44
scalar, 178 lattice, 232
universal (see Kelley [5], p. 81) monotonic, 227
norm, 16 pseudo-normed space = a linear space
conjugate, 117 with a pseudo-norm
equivalent, 43 generated by A, 185
of a linear functional, 117
normed space = a linear space with a quotient, space, 9
norm mapping, 9
nowhere dense, 84
null space, 6 radial, at a point, 14
kernel, 14
open, (set), 27 rank = co-dimension
mapping theorem, 99 rare = nowhere dense
relative to, 27 real restriction, 3
§.:open, 27
relation, 101
sphere, 29
somewhere dense, 179
order, bounded set, 228
represent, 140
dual, 225
ordered linear space, 225 residual, 86
orthogonal, 119 Riesz, theorem, 127
Osgood theorem, 86 theory, 207

paired spaces, 137 scalar, 2

pairing, 137 field, 2
induced, 146 product = inner product
 INDEX 255
Schauder's theorem, 208 topological space ( Cont.)
semi-metric = pseudo-metric quasi-compact = sequentially com-
separate, 22 pact-
strongly, 22 regular, 27
separated, 22 separable, 27
separation theorem, 22, 118 separated = Hausdorff-
sequence, 29 sequentially compact, 29
Cauchy, 32, 56 T 2 = Hausdorff -
relative to a pseudo-metric, 85 topological subspace, 27
sequentially closed, 91 topology, 27 (see also under vector
set function, additive, 80, 157 topology)
countably additive, 80 coarser = weaker -
Smulian's criterion ( for compactness), of compact convergence for all de-
142 rivatives, 82
space: see und er topological space, of convergence in measure, 55
linear topological space of coordinatewise convergence
spectrum, 240 product-
Stone-Cech compactification, 209 discrete, 30
Stone's theorem, 17 F-projective, 31
strong separation theorem, 118 finer = stronger-
subnet, 28 G-induced, 32
subspace, 4 indiscrete = trivial -
generated by, 4 induced, 32
linear, 4 induced = relative -
maximal, 7 larger = stronger -
linear topological, 39 metric, 29
topological, 27 of pointwise convergence, 31
support, of a convex set, 130 product, 30
of a distribution, 197 projective, 31
of a function, 81 pseudo-metric, 29
of a measure, 127 relative, 27
supremum norm, 81 relativization, 27
symmetric, operator, 100 simple = - of pointwise convergence
set, 14 of simple convergence = - of point-
system of nuclei = local base wise convergence
smaller = weaker -
stronger, 30
tensor product, 152 trivial, 30
projective - toPology, 153 uniform, 69
topological isomorphism, 40 of uniform convergence on compact
topological space, 27 sets, 81
compact, 29 of uniform convergence· on a family
countably compact, 29 of sets, 69
Hausdorff, 29 of uniform convergence on a set,
locally compact, 29 68
metrizable, 29 usual, 30
pseudo-metrizable, 29 weaker, 30
256 INDEX
total, subset, 161 vector topology (Cont.)
variation, 126 F-inductive, 121
totally bounded, 60 inductive, 121, 148
elementary facts on- sets, 65 inductive limit, 149
function - on a subset, 70 locally convex, 45
transformation (see mapping) Mackey, 173
Tychonoff theorem, 31 non-trivial, 50
norm, 43
uniformly convex, 161 projective, 150
unit, 239 pseudo-norm, 44
quotient, 39
vector, 2 relatively strong = Mackey-
ordering, 16 strong, 169
corresponding to a cone, 16 strongest, 42
space, 1 of uniform convergence on members
vector topology, 34 of a family, 144
d ... absorbing, 182 of uniform convergence on sequences
admissible, 166 converging to 0, 212
of bi-equicontinuous convergence, weak, 153
180 weak*, 155
bound, 183 weak (E,F), 138
derived from ff = bound exten- elementary properties, 139
tion of f7 weak (F,E), 140
bomd extension, 183
direct sum, 121 weight function, 220