Lecture Notes in Mathematics 1466

Editors:
A. Dold, Heidelberg
B. Eckmann, Ztirich
E Takens, Groningen
Wojciech Banaszczyk

Additive Subgroups
of Topological Vector
Spaces

Springer-Verlag
Berlin Heidelberg NewYork
London Paris Tokyo
Hong Kong Barcelona
Budapest
Author
Wojciech Banaszczyk
Institute of Mathematics
L6d~ University
Banacha 22
90-238 L6d~, Poland

Mathematics Subject Classification (1980): I IH06, 22-02, 22A10, 22A25, 22B05,

40J05, 43-02, 43A35, 43A40, 43A65, 46A 12, 46A25, 46B20, 47B 10, 52A43, 60B 15

ISBN 3-540-53917-4 Springer-Verlag Berlin Heidelberg New York

ISBN 0-387-53917-4 Springer-Verlag New York Berlin Heidelberg

This work is subject to copyright. All rights are reserved, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, re-use of
illustrations, recitation, broadcasting, reproduction on microfilms or in other ways, and
storage in data banks. Duplication of this publication or parts thereof is only permitted
under the provisions of the German Copyright Law of September 9, 1965, in its current
version, and a copyright fee must always be paid. Violations fall under the prosecution
act of the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 199 I
Printed in Germany
Printing and binding: Druckhaus Beltz, Hemsbach/Bergstr.
2146/3140-543210 - Printed on acid-free paper
 I~REFACE

In the commutative harmonic analysis there are at least two im-

portant theorems that make sense without the assumption of the local
compactness of the group and the existence of the Haar measure: the
Pontryagin-van Kampen duality theorem and the Bochner theorem on posi-
tive-definite functions. The Pontryagin-van Kampen theorem is known to
be true e.g. for Banach spaces, products of locally compact groups or
additive subgroups and quotients of nuclear Frechet spaces. The Boch-
ner theorem remains valid for locally convex spaces over p-adic fields,
for nuclear locally convex spaces (the Minlos theorem), their sub-
groups and quotients. These lecture notes are an attempt of clearing up
the existing material and of determining the "natural" limits of the
appl~cability of the theory. Pontryagin duality is discussed in chap-
ter 5 and the Bochner theorem in chapter 4.
Our exposition is based on the notion of a nuclear group. Roughly
speaking, nuclear groups form the smallest class of abelian topological
groups which contains locally compact groups and nuclear spaces and is
closed with respect to the operations of taking subgroups, Hausdorff
quotients and arbitrary products. The definition and basic properties
of nuclear groups are gathered in chapter 3. It turns out that, from
the point of view of continuous characters, nuclear groups inherit many
properties of locally compact groups.
In chapter 2 we show that the assumption of nuclearity is essen-
tial: if a separable Frechet space E is not nuclear, it contains a
discrete additive subgroup K such that the quotient group E/K does
not admit any non-trivial continuous unitary representations.
In section i0 we apply nuclear groups to obtain answer to an old
problem of S. Ulam on rearrangement of series in topological groups.
From the point of view of convergence of series and sequences, nuclear
groups inherit many properties of finite dimensional and nuclear spaces.
The characteristic feature of our considerations is their geome-
trical complexion. The heart of the monograph is section 3 on relations
between lattices and n-dimensional ellipsoids in R n. The main t o o ~
used here are the Minkowski convex body theorem and the Korkin-Zolota-
rev bases. To derive the results of chapters 3-5 from those of section
 VI

3, we need only some, rather elementary, t o p o l o g y and t o p o l o g i c a l al-

gebra. The m a i n result of c h a p t e r 2 is a c o n s e q u e n c e of the M i n k o w s k i -
-Hlawka t h e o r e m and c e r t a i n p r o p e r t i e s of e l l i p s o i d s of inertia of con-
vex bodies. The a n a l y t i c a p p a r a t u s is made use of to a slight d e g r e e
only. In that sense, our a p p r o a c h to d u a l i t y is kept in the spirit of
the o r i g i n a l g e o m e t r i c a l idea of Pontryagin.
This m o n o g r a p h lies on the line of several b r a n c h e s of mathematics,
sometimes e v e n quite distant, and the author wishes to thank m a n y per-
sons for their remarks and advice w h i c h e n a b l e d him to present each
p a r t i c u l a r b r a n c h in c o n f o r m i t y w i t h the c u r r e n t state of knowledge:
S. K w a p i e ~ and A. P e l c z y ~ s k i for their help in functional analysis,
e s p e c i a l l y the local t h e o r y of B a n a c h spaces; H.W. Lenstra, Jr. for
c o m m e n t s on the g e o m e t r y of numbers; V.M. Kadets for i n f o r m a t i o n con-
c e r n i n g r e a r r a n g e m e n t of series, and m a n y others.
P a r t i c u l a r thanks are d i r e c t e d to W. W o j t y ~ s k i for his encourag-
ing suggestions; this w o r k is a d e v e l o p m e n t of his ideas.

Ldd~, July 1990

 COII'£1f,II'I'S

Preface
Chapter 1, PRELIMINARIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
1. Topo]ogical groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
2. Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3. Geometry of numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

Chapter 2, EXOT]C GROUPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4. R e p r e s e n t a t i o n s of a b e l l a n t o p o l o g i c a ] groups . . . . . . . 45
5. Q u o t i e n t s of named spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6. Q u o t i e n t s of n o n - n u c l e a r spaces . . . . . . . . . . . . . . . . . . . . . 60

Chapter 3, NUCLEAR GROUPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

7, Nuclear groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8. Characters of nuclear groups ........................ 80
9. Nuclear v e c t o r groups ............................... 86
10. The L e v y - S t e i n i t z theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Chapter 4, THE BOCHNERTHEOREM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

II. Preliminaries ....................................... 110
12. The Bochner theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
13. The SNAG theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

Chapter 5, PONTRYAGIN DUALITY . . . . . . . . . . III I I I I 132

14. Preliminaries ............... I I I I i l 132
15. L o c a l l y convex v e c t o r groups I I I i l I 138
16, N u c l e a r i t y of dual groups , , , I I I I i I 144
17. Strong r e f l e x i v i t y .......... I I 1 | | | 151
18. Groups w i t h boundedness . . . . . I I I l l I 162

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Index of symbo]s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
 Chapter 1

~ E S

In this c h a p t e r we e s t a b l i s h n o t a t i o n and terminology. We also

state some s t a n d a r d facts in a form c o n v e n i e n t to us. S e c t i o n 1 is de-
v o t e d to a b e l i a n t o p o l o g i c a l groups and s e c t i o n 2 to topological vector
spaces. In s e c t i o n 3 we give some m o r e or less k n o w n facts about addi-
tive s u b g r o u p s of R n.

i. T o p o l o g i c a l groups

The groups under c o n s i d e r a t i o n will be m o s t l y a d d i t i v e subgroups

or q u o t i e n t groups of vector spaces. T h e r e f o r e we shall apply the addi-
tive n o t a t i o n mainly, d e n o t i n g the n e u t r a l e l e m e n t b y 0. Naturally,
we shall keep the m u l t i p l i c a t i v e n o t a t i o n for groups of, say, n o n - z e r o
c o m p l e x numbers or linear operators. The a d d i t i v e groups of integers
and of real and c o m p l e x numbers will be d e n o t e d by Z, R and C, re-
spectively. The m u l t i p l i c a t i v e g r o u p of c o m p l e x numbers with modulus
1 will be d e n o t e d by S.
By a c h a r a c t e r of a group G we m e a n a h o m o m o r p h i s m of G into
the g r o u p T: = R/Z. We shall f r e q u e n t l y identify T w i t h the i n t e r v a l
(- ½, ½]. The c a n o n i c a l p r o j e c t i o n of R onto T w i l l be d e n o t e d by
p. Thus p(x) = x for x = (- ,~]. The value of c h a r a c t e r X at
a point g will be d e n o t e d by x(g) or, sometimes, by <x,g>.
Now, let G be an a b e l i a n t o p o l o g i c a l g r o u p (we do not assume
t o p o l o g i c a l groups to be Hausdorff). The set of all c o n t i n u o u s charac-
ters of G, with addition defined pointwise, is an a b e l i a n group
again. We call it the dual g r o u p or the c h a r a c t e r g r o u p of G and de-
note by G-.
C h a r a c t e r s are u s u a l l y d e f i n e d as h o m o m o r p h i s m s into S. Such a
d e f i n i t i o n is c o n v e n i e n t in h a r m o n i c analysis, when we c o n s i d e r com-
plex-valued functions " s y n t h e s i z e d " of c o n t i n u o u s c h a r a c t e r s (such a
s i t u a t i o n w i l l take place in c h a p t e r 4). However, it leads to the mul-
t i p l i c a t i v e n o t a t i o n on G- w h i c h is i n c o n v e n i e n t in duality theory
w h e n we try to m a i n t a i n s y m m e t r y b e t w e e n G and G- (especially w h e n
we c o n s i d e r t o p o l o g i c a l vector spaces). There are also c e r t a i n tech-
nical reasons for w h i c h we have c h o s e n T i n s t e a d of S.
We shall have to c o n s i d e r various t o p o l o g i e s on G . The dual
group e n d o w e d w i t h a given topology ~ w i l l be d e n o t e d G T. By G~,
G- and G^ we shall denote the dual g r o u p endowed, respectively,
c pc
w i t h the t o p o l o g y of u n i f o r m convergence on finite, compact and pre-
compact subsets of G (i.e. with the t o p o l o g y of pointwise, compact
and p r e c o m p a c t convergence). The second one is u s u a l l y called the com-
pact-open topology.
Now, let A be a subset of G. If X is a c h a r a c t e r of G,- then
we w r i t e

I x ( A ) I = sup {Ix(g)l: g ~ A}.

The set

{X • G : Ix(A) I ~ }

is c a l l e d the p o l a r of A in G - ; we d e n o t e it by o
A G. If the m e a n -

ing of G is clear, we s i m p l y w r i t e A° instead of o

A G. By A p'
° A°
c
and A° we d e n o t e the set A° endowed with the t o p o l o g y of point-
pc
wise, c o m p a c t and p r e c o m p a c t convergence, respectively.
If A is a s u b g r o u p of G, then

A° = {X • G^ : XIA ~ 0};

this follows, for instance, from (1.2). Thus A° is a c l o s e d subgroup

of G~; we call it the a n n i h i l a t o r of A.
A subset A of G is said to be q u a s i - c o n v e x if to each g• G \ A
there corresponds some X • A° with Ix(g)l > {. The set

1
N {g • G : l×(g) l ~ ~}
X~A °

is e v i d e n t l y the s m a l l e s t quasi-convex subset of G containing A; we

call it the q u a s i - c o n v e x hull of A. We say that G is a locally
quasi-convex group if it admits a base at zero c o n s i s t i n g of quasi-
-convex sets. Observe that if G is a H a u s d o r f f locally quasi-convex
group, then it admits sufficiently many continuous characters (i.e.
continuous characters separate the p o i n t s of G). Observe also that the
polar of any subset of G is a q u a s i - c o n v e x subset of each of the
groups G , Gc and Gpc ; therefore all the three groups are locally
quasi-convex.

(l.lJ ra,~wA~ Let g,h be two e l e m e n t s of an a b e l i a n group G. If

X is a c h a r a c t e r of G such that Ix(g)l, Ix(h) l and Ix(g + h)l
1
are less than 7' then x(g + h) = x(g) + x(h).
 ~ - One has

x ( g + h) - x(g) + x(h) (mod Z),

i.e.

(i) x ( g + h) - x(g) - ×(h) g Z.

From our assumption we obtain

(2) Ix(g + h) - x(g) - x(h) l

1 1 1
< Ix(g + h)l + Ix(g) l + Ix(h)l < -~ + ~" + ~" = i .

Now (i) and (2) imply that x ( g + h) - x(g) - x(h) = 0. •

~.2) ;Jmmua. Let X be a character of a n a b e l i a n group G. Let

m be a positive integer and g an element of G, such that
1
x(kg) < ~ for k = l,...,m. Then ×(mg) = mx(g).

Proof. By the preceding lemma, for each k = l,...,m-1, we have

x ( ( k + l)g) = x(kg) + x(g)- Thus

m-I m-I m-i

Z x ( ( k + l)g) = >q x(kg) + ~ x(g),
k=l k=! k=l

which means that x(mg) = mx(g). •

Let A be a subset of a n a b e l i a n group G. By gp A we denote

the subgroup of G generated by A. For each m = 1,2,..., we de-
note

A m = {a I + ... + am : a l , . . . , a m E A}.

~1.3) PIam~sr~Ium. Let G be an abelian topological group. The

polars of c o m p a c t (resp. finite, precompact) subsets of G forma base
at zero in Gc (resp. in G , Gpc).

Proof. Let U be a neighbourhood of zero in Gc (rasp. in G ,

Gpc). There exist an ~ > 0 and a compact (resp. finite, precompact)
subset Y of G, such that the set W = {X E G : IX(Y) I < £} is
contained in U. Choose an integer m > (4E) -I. The set A = ym is
compact (resp. f i n i t e , p r e c o m p a c t ) . B y (1.2), f o r e a c h X ~ A°, we
1
have Ix(Y) I ~ ~ IX(A) I ~ 4--m < ~" Thus A O c W. •

By No(G ) we denote the family of a l l neighbourhoods of zero in

an a b e l i a n topological group G (we d o n o t assume neighbourhoods to
be open).

~I.4J r~mm~. A character X of a n a b e l i a n topological group G is

continuous if a n d o n l y if X E U° for a certain U ~ No(G).

Proof. The necessity of t h e c o n d i t i o n is t r i v i a l . To prove the

sufficiency, choose any £ > 0. We can find an integer m > (4£) -1

and then some W E No(G ) with W TM c U. By (1.2), we have iX(W) I ~

[X(U) I ~ 4 ~ < £" This means that X is c o n t i n u o u s at zero. •

(I.5) PR~OSITI~. The polars of n e i g h b o u r h o o d s of zero in a n a b e -

lien topological group G are compact subsets of Gpc.

Proof. Choose any U ~ No(G). The group G~ is c o m p a c t because

we may identify it w i t h a closed subgroup of t h e p r o d u c t TG (see a l s o

(1.8)). Since U °p is a c l o s e d subset of Gp, it is enough to show

that the identity mapping U° ~ U° is c o n t i n u o u s .

p pc
Choose any < ~ U° and let W be a n e i g h b o u r h o o d of < in U°
pc"
By (1.3), there is s o m e p r e c o m p a c t subset A of G such that

W': = (K + A °) n U ° c W.

Next, we can find some V E No(G ) with V 3 c U. Since A is p r e -

compact, there exist some gl,..0,gn ~ A such that

(i) A c { g i } ~ = 1 + V.

The set

W" = {X E U O : Ix(gi) i
- <(gi )I ~ i-~ for i = i ' . "" ,n}

is a n e i g h b o u r h o o d of K in U °. It r e m a i n s to show that W" c W'.

P
So, choose any X E W". We have to s h o w that X - m ~ A°- Take
any g e A. In v i e w of (i), w e m a y w r i t e g = gi + h for some i =
l,...,n and some h ~ V. Now, from (1.2) we obtain

Ix(v) 5_-<. Ix(u)l =< and I,<(v) ls I,<(u) l s y-~.

Hence

I(x - K)(g)l <= I x ( g i ) - K(gi)l + Ix(h) l + I < ( h ) l =< 3 = !4 "

12

An abelian group G is c a l l e d divisible if to e a c h g E G and

each n = 1,2 .... there corresponds some h E G with n h = g.
 (1.6) PRof~s/1,x~. Let H be a s u b g r o u p of an a b e l i a n group G.
Every homomorphism of H into a d i v i s i b l e group can be e x t e n d e d to a
homomorphism of G.

For the proof, see e.g. [38], Theorem (A.7).

Let G,H be a b e l i a n topological groups. An i s o m o r p h i s m ~ of G

onto H is c a l l e d a topological isomorphism if ¢ and ¢-i are con-
tinuous. If there is a t o p o l o g i c a l isomorphism of G onto H, then we
say that G and H are t o p o l o g i c a l l y isomorphic and w r i t e G - H. An
injective homomorphism ¢ : G + H is c a l l e d a topological embedding if
is a t o p o l o g i c a l isomorphism of G onto the g r o u p ¢(G) endowed
with the t o p o l o g y induced from H.

(1.7) PI~DI~SIYI~I. Let GI,...,G n be a b e l i a n topological groups.

There is a c a n o n i c a l topological isomorphism between (G 1 × ... × Gn) c

and (GI) c x ... × (Gn) c •

This is a s t a n d a r d fact. For the proof, we refer the reader to

[38], (23.18) - the a s s u m p t i o n there that GI,...,G n are locally com-
pacb- is inessential. For i n f i n i t e products, see (14.11) below.

Let H be a s u b g r o u p of an a b e l i a n topological group G. We say

that H is d u a l l y closed in G if to each g EG \ H there corresponds
some X E H° with x(g) # 0 (this is e q u i v a l e n t to the assertion that
H is a q u a s i - c o n v e x subset of G). Next, we say that H dually
embedded in G if each c o n t i n u o u s character of H can be e x t e n d e d to
a continuous character of G. Observe that d u a l l y closed subgroups are
closed. Observe also that each c o n t i n u o u s character of H can be ex-
tended in a u n i q u e w a y to a c o n t i n u o u s character of H.

Let us recall shortly basic facts concerning the P o n t r y a g i n - van

Kampen duality theorem. The p r o o f s can be f o u n d e.g. in [38], §24. By
a compact (rasp. locally compact) g r o u p we shall m e a n a g r o u p w h i c h is
compact (rasp. locally compact) and separated. Locally compact abelian
groups are c a l l e d LCA groups.

(~.8) PROFOSZS~Z~. Let G be an LCA group. Then G = G is an

c pc
LCA group, too, and the e v a l u a t i o n map is a t o p o l o g i c a l i s o m o r p h i s m of
G onto (G)c" If H is a c l o s e d subgroup of G, then H is dual-
ly c l o s e d and d u a l l y embedded. Moreover, the canonical mappings
Gc/H~ + Hc and (G/H) c- + H°c are b o t h topological isomorphism. If G
is compact, G is discrete. If G is d i s c r e t e , Gc is compact.

There are canonical topological isomorphisms R ~ R, Z ~ T, T ~ Z.

c c c

~1.gJ1~ol~ssrfx~. Let G be an LCA group. Then there exist an

n = 0,1,2,..., a compact group K and a discrete group D, such that
G is topologically isomorphic to a closed subgroup of Rn × K x D.

Pmm~f. Being an LCA group, G contains an open subgroup A ~

C

Rn x H for some n = 0,1,2,... and some compact group H ([69], Theo-

rem 25 or [38], (9.14)). Let ~ : G + G /A be the natural projection.
Every (abelian) group is a quotient of a free one. So, we can find a
free abelian group F and a homomorphism ~ of F onto G /A. Let

{fi}i~i be a system of free generators of F. For each i E I, choose

some Xi e G with ~ ( X i) = ~ ( f i ). Let ~ : F ~ G be the homo-

morphism given by a ( f i) = Xi for i ~ I. We obtain the following

commutative diagram:

G- ~ >G-/A
The formula

p(a,f) = a + a(f) (a ~ A, f ~ F)

defines a homomorphism p : A x F ~ G . We shall prove that

^

(I) p(A x F) = G .

To this end, choose an arbitrary × ~ G . Since ~(F) = G /A, we can

find some f e F with ~(f) = ~(X). Then ~ o(f)) = ~(f) = ~(X),
which means that a : = X - a(f) e ker ¢ = A. Thus X = a+ o(f) =
p(a,f) e p(A × F), which proves (i).
Let us endow F with the discrete topology. Since A × {0} is an
open subgroup of A × F and p is a topological isomorphism (in fact,
an identity) of A x {0} onto the open subgroup A of G~, it follows

that p : A × F ~ Gc is b o t h continuous and open. Consequently, Gc

(A × F ) / k e r p. So, in virtue of (1.8 , we have

G (Gc) c ((A x F)/ker p)c - (ker p)~.

In other words, G is topologically zsomorphic to a closed subgroup of

(A × F) c. From (1.7) we get (A × F) - Ac × F c ~ Rn × Hc × F and
c c

it r e m a i n s to o b s e r v e that F is c o m p a c t and H discrete. •

c c

The completion of an a b e l i a n topological group G will be d e n o t e d

by G. We shall identify G with a dense subgroup of G. The clo-
sures in G of e l e m e n t s of a n y g i v e n base at z e r o in G form a base
at zero in G ([23], Ch. III, §3, Proposition 7).

(I.IO) P R O B O S ~ . Let Go be a d e n s e subgroup of an a b e l i a n to-

pological group G. Let H be the closure in G of a closed sub-
group H° of Go and let ~ : G ~ G/H be the canonical projection.
Then the c a n o n i c a l bijection G o / H ° ~ ¢(G o) is a t o p o l o g i c a l isomor-
phism of Go/H o onto a dense subgroup of G/H.

This is P r o p o s i t i o n 21 of [23], Ch. III, §2.

(1.~1) P ~ o s / T ~ . Let G be an a b e l i a n topological group. If G

is a k-space, then G is a c o m p l e t e group.

Pz,x~f. The space TG of c o n t i n u o u s mappings from G to T is

complete in the compact-open topology ([52], Ch. 7, Theorem 12). It
remains to o b s e r v e that G- is a c l o s e d subset of T G. •

(I.12) P B D P O S I T I ~ . Let G be an a b e l i a n group and B a family of

subsets of G satisfying the following conditions:

(a) every member of B contains zero;

(b) to e a c h U ~ B there corresponds some V • B with -V c U;
(c) to e a c h U e B there corresponds some V • B with V + V c U.

Then there exists a unique topology T on G compatible with the

group structure, such that B is a b a s e at zero for ~.

For the proof, see [23], Ch. III, §1.2.

Let {Gi}i~ I be a f a m i l y of a b e l i a n topological groups indexed

b y a set I. The product of these groups is d e f i n e d in the usual way;
we d e n o t e it b y II G.. It is e v i d e n t that the p r o d u c t of a family of
ieI 1
locally quasi-convex groups is locally quasi-convex.
Let {Pij : Gi + Gj; i,j E I, i ~ j} be an i n v e r s e system of
topological groups, that is to say, I is a d i r e c t e d set and, for
each pair i,j E I with i ~ j, a continuous homomorphism Pij : Gi ~
Gj is given, such that Pij'Pjk = Pik if i ~ j ~ k. We d e f i n e the
limit of this s y s t e m in the usual way, i d e n t i f y i n g it w i t h the appro-
priate s u b g r o u p of the p r o d u c t
I-[ G . . If I is the set of p o s i t i v e
i~I 1
integers, then we speak of an inverse sequence. Naturally, the limit
of an inverse system of l o c a l l y q u a s i - c o n v e x groups is l o c a l l y quasi-
-convex. The p r o d u c t
II G i is c a n o n i c a l l y t o p o l o g i c a l l y isomorphic
i•I
to the limit of the inverse system II G i w h e r e K runs t h r o u g h all
i•K
finite subsets of I and the p r o j e c t i o n s PKL for K o L are defined
in the usual way.

The limit G of the inverse system {Pij : Gi ~ Gj} m a y be equal

to zero. If, however, I is at most c o u n t a b l e and all Pij are onto,
then also all p r o j e c t i o n s Pi : G + Gi are onto. Kaplan [50], Lemma
4.6, p r o v e d that if {Pij : Gi ~ Gj} is an inverse sequence of LCA
groups such that Pij(Gi) is dense in Gj for all pairs i,j with
i ~ j, then also Pi(G) is dense in Gi for e v e r y i.
Again, let {Gi}i• I be a family of a b e l i a n t o p o l o g i c a l groups.
Their d i r e c t sum, d e n o t e d by ~ Gi, is a l g e b r a i c a l l y the s u b g r o u p of
i•I
the p r o d u c t
II Gi, c o n s i s t i n g of finite s e q u e n c e s (that is, an ele-
i•I
ment (gi)i•i of II G i b e l o n g s to ~ G i if and only if gi = 0
iEI iEI
for all but a finite number of indices i). We shall consider on Z G i

the a s t e r i s k and the r e c t a n g u l a r topologies. To d e s c r i b e them, we have

to i n t r o d u c e some a d d i t i o n a l notions.
Let U be a subset of an a b e l i a n g r o u p G. For each g • U, we
define

n u = sup {n : kg • U for k = i, .... n}

and g/U = (nu)-I. This means, in particular, that g/U = 0 if and

and o n l y if kg • U for e v e r y k.
Let us suppose that, for each i • I, we are given some Ui •
No(Gi). We d e n o t e

Z U i = { (gi)i•i • ~ Gi : gi • Ui for all i • I},

i•I ieI

Z*U i = {(gi)i•i • ~ U. : Z (gi/Ui) < I}.

i•I i•I 1 i•I

Let B be the f a m i l y of all sets of the form Z U i w h e r e U i • No(G i)

i•I
for e v e r y i. Similarly, let ~* be the family of all sets of the
form Z*U i where U i E No(Gi) for e v e r y i. It follows from (1.12)
i~I
that there is a u n i q u e t o p o l o g y on ~ G i c o m p a t i b l e w i t h the g r o u p
i~I
structure, for w h i c h B is a base at zero; we call it the r e c t a n g u l a r
topology. Conditions (a) - ( c ) of (1.12) are s a t i s f i e d trivially. Simi-
larly, it follows from (1.12) that there is a unique t o p o l o g y on ~ Gi
i~I
c o m p a t i b l e w i t h the g r o u p structure, for w h i c h B* is a base at zero;
we call it the a s t e r i s k topology. The o n l y n o n - t r i v i a l thing here is
to v e r i f y (1.12) (c) w i t h B r e p l a c e d by B*:

(1.13) ~m~A. Let {Gi}ie I be a family of abelian topological

groups and let B* be d e f i n e d as above. T h e n to each U ~ B* there
c o r r e s p o n d s some V ~ B* with V + V c U.

To prove (1.13), we need the f o l l o w i n g simple proposition whose

v e r i f i c a t i o n is left to the reader:

(1.14) TJmmfA. Let U be a zero-containing subset of an abelian

group G. Then g/(U + U) ~ ½(g/U) for e a c h g ~ U. If V is another
z e r o - c o n t a i n i n g subset of G and V + V c U, then (g + h) / U
max (g/V,h/V) for all g,h ~ V.

of (1.13). Choose an a r b i t r a r y U ~ B*. We have U= ~ * Ui

i~I
for some U i ~ No(Gi), i ~ I. For e~ch i ~ I, we can find some

Vi E N o ( G i ) with V~I c Ui . Set V = Z* Vi .

Now, take any s e q u e n c e s (gi)iEl and (hi)i~ I belonging to V.

From (1.14) we g e t

[(gi + hi)/Ui] ~ ½ ~ [(gi + h i ) / ( V i + Vi)]

ieI i~I

< ! Z max (gi/Vi, hi/V i) ~ ~1 iEI

Z [gi/Vi + h i / V i] < i.
= 2 i~I

Thus (gi + hi)ieI ~ Z * U~,± w h i c h means that V + V c U. •

iEI

The a s t e r i s k t o p o l o g y is, by definition, finer than the r e c t a n g u -

lar one. For c o u n t a b l e d i r e c t sums, these two t o p o l o g i e s are identi-
cal:

C2-]L~J ~ Let (G n)n=l be a s e q u e n c e of a b e l i a n topolo-

co
gical groups. Then the a s t e r i s k t o p o l o g y on ~ Gn is equal to the
n=l
 10

rectangular one.

Proof. Let U n E No(Gn) for n = 1,2, . . . . We h a v e to s h o w that

* U contains a rectangular neighbourhood of zero. For each n =

n=l n
1,2,..., we can find some V n • No(Gn) with V 2n c U n. From (1.14) it
follows by induction that

(g/U n) ~ 2 - n ( g / V n) for all n = 1,2,... and all g • V n.

So, if (gn)n=l • Z Vn, then

n=l

Z (gn/Un) & Z 2 - n ( g n / V n ) < Z 2 -n = i.

n=l n=l n=l

Thus ~ V n c ~* U . •
n= 1 n= 1 n

In g e n e r a l , the rectangular topology is not equivalent to the a s t e -

risk one (consider, for instance, an u n c o u n t a b l e direct sum of real
lines). In the sequel, speaking of d i r e c t sums of topological groups,
we shall always assume that they are e n d o w e d with the asterisk topology,
unless it is e x p l i c i t l y stated otherwise. Notice that if {Gi}i• I is a
family of locally convex spaces, then the g r o u p Z Gi is t o p o l o g i c a l -
i•I
lY i s o m o r p h i c to t h e i r locally convex direct sum.

(I.161PRQF~SrfI~. The direct sum of an a r b i t r a r y family of locally

quasi-convex groups is l o c a l l y quasi-convex.

An e a s y proof is left to the reader.

(1.17) l~gPOs~I~m. Let G be the d i r e c t s u m of a f a m i l y {Gi}i• I of

Hausdorff abelian groups. For each i • I, let ~i : G ~ Gi be the
canonical projection. If P is a p r e c o m p a c t subset of G relative to
the asterisk or r e c t a n g u l a r topology, then ~ (P) = {0} for all but
1
finitely many indices i.

Proof. Suppose that P is p r e c o m p a c t in the rectangular topology.

Set J = {i E I : ~i(P) # {0}}. We h a v e to s h o w that J is finite.
Suppose the contrary. To e a c h i • J there corresponds some gi • P
with ~ i ( g i ) # 0. Next, there is some U i ~ N o ( G i) with ~i(gi ) ~ U i
because Gi is s e p a r a t e d . The set U = Z Ui is a r e c t a n g u l a r neigh-
i•I
 11

b o u r h o o d of zero in G. So, there is a finite subset A of P such

that P c A + U because P is precompact. Since A is finite and
c o n s i s t s of finite s e q u e n c e s w h i l e J is infinite, it follows that
there is an index j ~ J such that ~j(A) = {0}. Then

~j(p) c ~j(A + U) = ~j(A) + ~j(U) = {0} + Uj = Uj.

On the other hand, we have gj E p and ~j(gj) ~ Uj, w h i c h is a con-

tradiction. •

Let {Pij : Gi + Gj; i,j E I, i ~ j} be a direct system of

a b e l i a n t o p o l o g i c a l groups, that is to say, I as a directed set and,
for each pair i,j ~ I with i & j, a continuous homomorphism Pij :
G i + Gj is defined, such that P i j ' P j k = Pik if i S j ~ k. Let G
be the direct sum of the f a m i l y {Gi}i~ I and let GO be the s u b g r o u p
of G g e n e r a t e d by all e l e m e n t s of the form

gi - Pij(gi ) (i,j ~ I; i & j; gi ~ Gi)

(we treat Gi and Gj as s u b g r o u p s of G). We d e f i n e the limit of

the system c o n s i d e r e d as the q u o t i e n t g r o u p G/G o . When all groups
Gi are locally c o n v e x spaces, we o b t a i n the usual d e f i n i t i o n of the
inductive limit.
Kaplan [50] d e f i n e d the limit of the d i r e c t system as the group
G/G o . He p r o v e d that if I is countable, the groups Gi are local-
ly compact and all m a p p i n g s Pij are injective, then GO is closed
([50], T h e o r e m 8, p. 433).
It is not h a r d to see that if J is a c o f i n a l subset of I and
all groups Gi are locally q u a s i - c o n v e x , then the limit of the s y s t e m
{Pij : Gi + Gj; i,j ~ I, i ~ j} may be i d e n t i f i e d w i t h the limit
of the s u b s y s t e m {Pij : Gi ~ Gj; i,j E J, i ~ j}. The a s s u m p t i o n
of local q u a s i - c o n v e x i t y is essential.
If I is the set of p o s i t i v e integers, then we speak of direct
sequences. In v i e w of the last remark, when considering limits of
c o u n t a b l e direct systems we may r e s t r i c t o u r s e l v e s to limits of d i r e c t
sequences.
The d i r e c t sum of a family {Gi}ie I of locally quasi-convex
groups is e a s i l y seen to be t o p o l o g i c a l l y i s o m o r p h i c to the limit of
the direct system {PKL : Z G i ~ Z Gi} where K,L run through
ieK iEL
finite subsets of I and the e m b e d d i n g s PKL are defined in the
usual way.
 12

t1.18J ~XTX~. Let G be the limit of a direct sequence

{Pn : G n ~ Gn+l} of a b e l i a n topological groups, in w h i c h all m a p p i n g s
Pn are topological embeddings. Then the topology of G induces orig-

inal topologies on the groups G n. Consequently, if all g r o u p s Gn are

separated, so is G.

Pmsaf. We m a y a s s u m e t_hat (G n ) n = 1 is an i n c r e a s i n g sequence of sub-

groups of G. Let B be the family of all sets of the from

U1 + U2 + ... := U (U 1 + ... + U n)
n=l
where Un ~ N o ( G n) for e v e r y n. It f o l l o w s directly from (i.15) that
B is a b a s e at zero in G.
Fix an i n d e x m and c h o o s e an a r b i t r a r y V ~ No(Gm). We are to
find some U ~ No(G) with U N G m c V. Naturally, we m a y assume that
m = I. There is some U 1 E No(G1) with U 1 + U 1 c V, and a simple
inductive argument allows us to find, for e a c h n ~ 2, some Un,W n E

N o ( G n) with W n N Gn_ 1 c Un_ 1 and Un + Un c W n" It r e m a i n s to s h o w

that G 1 N (U 1 + U 2 + ...) c V. Set Yk = U1 + "'" + U k + Uk for

k = 1,2,... . It is e n o u g h to s h o w that G 1 N Yk c V for e v e r y k.
For k = i, this is o b v i o u s . For k > i, we u s e induction:

G 1 n Y k c G 1 N Gk_ 1 N (U 1 + ... + Uk_ 1 + W k)

= G 1 N (U 1 + ... + Uk_ 1 + (W k n Gk_l))

c G 1 N (u I + ... + Uk_ 1 + Uk_l) = G 1 N Yk-l'

which is c o n t a i n e d in V due to the inductive assumption. •

A topological vector space is locally convex if and o n l y if it is

a Hausdorff locally quasi-convex group (see (2.4)). Komura [55] showed
that the limit of an u n c o u n t a b l e direct system {Pij : Ei ~ Ej} of
locally convex spaces in w h i c h all m a p p i n g s Pij are topological em-
beddings need not be locally convex, and even if it is, it need not
induce original topologies on the spaces E i. If, in (1.18), all
groups Gn are locally convex spaces, G is a l o c a l l y convex space,
too ([80], Ch. II, (6.4)). If we assume only that all G "s are Haus-
n
dorff locally quasi-convex groups, then probably G n e e d not be lo-
cally quasi-convex. See, however, (7.9). Vilenkin [99] considered an-
other topology on the limit of a d i r e c t system. Under his definition,
the limit of a n y d i r e c t system of a b e l i a n topological groups is a io-
 13

cally q u a s i - c o n v e x group.
Let G,H be abelian t o p o l o g i c a l groups and let # : G ~ H be a
continuous homomorphism. Then the formula

<~(x),g> = <X,#(g)> (X E H ; g ~ G)

defines a homomorphism ~ : H- ~ G-. We call it the dual h o m o m o r p h i s m

and denote by ¢ It is clear that # : HT ~ G is continuous when
T is the t o p o l o g y of pointwise (rasp. compact, precompact) con-
vergence. If {Pij : Gi ~ Gj; i,j 6 I; i ~ j} is an inverse system
of abelian topological groups, then {Pij : G i ~ Gi} is a direct sys-
tem, and vice versa.
Let {Gi}iE I be a family of abelian t o p o l o g i c a l groups indexed
by a set I. Suppose that, for each i ~ I, a closed subgroup Hi of
Gi has been chosen. Let G be the subgroup of the product II G.
i~I 1
consisting of all sequences (gi)i~i such that gi 6 Hi for all but
finitely many indices i. We topologize G by i d e n t i f y i n g it with the
limit of the inverse system

~KL : Z Gi x ~ (Gi/H i) ~ X Gi × ~ (Gi/H i)

i~K i~K i~L i~L

where K,L are finite subsets of I with K D L, and ~KL is the

canonical projection. Endowed with this topology, G will be called
the reduced product of groups Gi relative to subgroups Hi and de-
noted by ~ (G i : Hi). It is clear that the topology of G induces
iEI
original topologies on the subgroups G i. Notice that if Hi = Gi for
almost all i, then II Gi; G =
if H i = {0} for almost all i, then
i~I
G = Z G i. More precisely, if H is the subgroup of G consisting
ieI
of all (gi)i~i such that gi ~ Hi for every i, then H has the
usual product topology and G/H is topologically isomorphic to the
direct sum Z (Gi/Hi). Since inverse limits and direct sums of lo-
ieI
cally q u a s i - c o n v e x groups are locally quasi-convex, G is locally
quasi-convex provided that so are all groups Gi and Gi/H i-

f1.19J PRU~SSITI~. For each i ~ I, let ~i : G ~ Gi be the ca-

nonical projection. A subset X of G is p r e c o m p a c t if and only if
~i(X) is p r e c o m p a c t in Gi for all i, and $i(X) c Hi for almost
all i.
 14

This is a d i r e c t consequence of (1.17) and the d e f i n i t i o n of G.

(I.20) PR O ~ S X ~ [ ~ . If I is at most countable and G locally

quasi-convex, then G m a y be i d e n t i f i e d with the limit of the d i r e c t
system

~KL : II G i x II H i ~ II G i × II H i
ieK i~K ieL i~L

wher e K,L are finite subsets of I with K c L and ~KL is the

canonical embedding.

The proof is left to the reader as an exercise.

Let G be a t o p o l o g i c a l group (abelian or not). By a r e p r e s e n -

tation of G in a real or c o m p l e x Banach space E we m e a n a homo-
morphism of G into the g r o u p GL(E) of a u t o m o r p h i s m s of E. We say
that E is the space of the r e p r e s e n t a t i o n . The operator being the
value of a r e p r e s e n t a t i o n ¢ at a p o i nt g E G will be denoted by
¢(g) or ~g. We say that ¢ is f a i t h f u l if ~g # id for g # I. A

representation ¢ : G ~ GL(E) is c a l l e d w e a k l y (resp. strongly) con-

tinuous if, for each fixed u ~ E, the m a p p i n g g ~ CgU is con-
tinuous in the w e a k (resp. strong) topology on E. We say that
is u n i f o r m l y continuous if the m a p p i n g g + Cg is c o n t i n u o u s in the
norm t o p o l o g y on GL(E).
We say that ¢ is a c y c l i c representation if there exists a vec-
tor u e E such that the linear subspace spanned over the vectors
CgU, g ~ G, is d e n s e in E; then u is c a l l e d a cyclic vector of
¢. A subspace M of E is c a l l e d invariant (relative to ¢) if
Cg(M) c M for each g ~ G. If the only invariant subspaces of E
are E itself and {0}, we say that ¢ is an i r r e d u c i b l e represen-
tation.
Let H be a c o m p l e x Hilbert space. A representation ~ : G ~GL(H)
is c a l l e d unitary if all o p e r a t o r s Cg, g E G, are unitary. A uni-
tary r e p r e s e n t a t i o n is w e a k l y continuous if and only if it is strong-
ly c o n t i n u o u s ; then we call it s i m p l y a continuous unitary represen-
tation.
if ¢ is an i r r e d u c i b l e unitary representation of an abelian
grou p in a c o m p l e x Hilbert space H, then d i m H = I. Since the
multiplicative g r o u p of n o n - z e r o complex numbers is topologically iso-
morphic to R x T, it follows that the study of irreducible unitary
representations of a b e l i a n groups can be r e d u c e d to the study of their
characters.
 15

Let ¢1,¢2 be two r e p r e s e n t a t i o n s of a group G in Banach

spaces EI,E2, respectively. We say that these representations are
equivalent if there is a topological isomorphism 0 : E1 + E2 such
that ¢i 0 = 8¢ 2 . If EI,E 2 are complex Hilbert spaces and the repre-
sentations ~ and ~ are unitary, then we say that ¢i is unit-
arily e q u i v a l e n t to ¢2 if there is a u n i t a r y isomorphism G : El+ E 2
with ¢18 = 8¢ 2.
Let {Hi}is I be a family of c o m p l e x Hilbert spaces. Their Hilbert
sum, denoted by ~ Hi, is the subspace of the product II Hi, con-
iEI 2 i~I
sisting of all sequences (ui)ie I with ~ lluill < ~. The. inner prod-
i~I
uct in ~ Hi is given by the formula
i~I

((ui),(vi)) = ~ (ui,vi).
ieI

Let G be a topological group and suppose that, for each i E I,

a continuous unitary representation ¢i of G in Hi is given. Then
the formula

¢(g).(ui)iE I = (¢i(g).ui)i~ I

defines a continuous unitary representation ¢ of G in the space

Hi; we call ¢ the Hilbert sum of the r e p r e s e n t a t i o n s ¢i and
iEI
denote it by ~ ¢..
i~I 1

(I°21) ~ / T / ~ . Let ¢ be a u n i t a r y representation of a group

G in a c o m p l e x Hilbert space H. Then there exists a family {Hi)iE I
of closed, invariant, pairwise orthogonal subspaces of H such that
U Hi is linearly dense in H and, for each i ~ I, the represen-
iEI
tation ¢i of G in Hi given by

el(g) .u i .¢(g).u. I . (g ~ G; ul ~ Hi )

is cyclic . In other words, every unitary representation is unitarily

equivalent to the Hilbert sum of cyclic representations.

This is Theorem (21.14) of [38].

A complex-valued function ~ on a group G is said to be posi-

tive-definite (shortly, p.d.) if, for each n = 1,2,..., it satisfies
the c o n d i t i o n
 16

n n
5q. 5q li~ ~(g~igj) > 0
i=l j =i J

for all Xl,...,Xn E C and g l , . . . , g n ~ G. E v e r y linear c o m b i n a t i o n

of characters, with non-negative coefficients, is a p.d. f u n c t i o n on G
(see [38], (32.9)).

(1.22) PI~laO~[~#. Let ~ be a p.d. f u n c t i o n on a g r o u p G. T h e n

(a) ¢(0) ~ 0;

(b) l¢(g)l S ~(0) for all g E G;

(c) l@(g) - ¢(h)l 2 & 2~(0)[¢(0) - Re ¢ ( g - h)] for all g,h ~ G.

This is a d i r e c t c o n s e q u e n c e of the d e f i n i t i o n of a p.d. f u n c t i o n

(cf. [38], (32.4)).

(I.23) I~ZH,O S l Y I ~ . Let ¢ be a u n i t a r y r e p r e s e n t a t i o n of a g r o u p

G in a H i l b e r t space H. For e a c h u E H, the function #(g) =
(~gU,U) is p o s i t i v e - d e f i n i t e . If G is a t o p o l o g i c a l g r o u p and
is continuous, then so is ¢. If ¢ is continuous and u is a
cyclic vector of ¢, then ¢ is continuous.

For the proof, see [38], (32.8) (a), (b) and (f).

(I.24) PI~UI~OSlTIOm. Let ~ be a c o n t i n u o u s p.d. f u n c t i o n on a to-

p o l o g i c a l group G. T h e n there exists a continuous cyclic unitary
representation ¢ of G w i t h a cyclic vector u such that (¢gU,U) =
%(g) for all g ~ G.

This follows from [38], (32.3) and (32.8) (f).

( 1 . 2 5 ) la~olao61Y~,~M. Let ¢,~ be two c o n t i n u o u s unitary r e p r e s e n t ~

tions of a t o p o l o g i c a l group G w i t h cyclic v e c t o r s u,w, respec-
tively. If (¢gU,U) = (?gW,W) for all g ~ G, then ¢ and ~ are
u n i t a r i l y equivalent.

For the proof, see [18], P r o p o s i t i o n 3, p. 146 or [38], (32.8)

(b).

(1.26) I~r~. The m a t e r i a l of this section is standard.

2. V e c t o r spaces

All v e c t o r spaces o c c u r r i n g are a s s u m e d to be real. The o n l y ex-

 17

c e p t i o n s are c o m p l e x spaces c o n n e c t e d w i t h r e p r e s e n t a t i o n s of groups

and v e c t o r spaces over u l t r a m e t r i c fields, considered in s e c t i o n 7.
For the t e r m i n o l o g y (but not for the notation) concerning topological
vector spaces we refer the reader to S c h a e f e r ' s book [80]. Locally
c o n v e x spaces (in particular, all n u c l e a r spaces) are meant to be
separated.
Let E be a v e c t o r space. By E# we d e n o t e the space of all
linear f u n c t i o n a l s on E. If E is a t o p o l o g i c a l v e c t o r space, the
space of all c o n t i n u o u s linear f u n c t i o n a l s on E is d e n o t e d by E*;
e n d o w e d w i t h the t o p o l o g y of u n i f o r m c o n v e r g e n c e on bounded, compact
and p r e c o m p a c t sets, it is d e n o t e d by Eb,E c and Epc, respectively.
The d i m e n s i o n of E is d e n o t e d by dim E. If A is a subset of E,
then span A d e n o t e s the linear s u b s p a c e of E s p a n n e d over A and
conv A is the c o n v e x hull of A.

~2.11 ~ . It is c o n v e n i e n t to set span @ = {0}. Consequently,

if Ul,U2,... is a s e q u e n c e in E, the symbol span {ui}i< k for
k = 1 s h o u l d be m e a n t as {0}.

E v e r y v e c t o r space m a y be t r e a t e d as an a d d i t i v e group. Similar-

ly e v e r y t o p o l o g i c a l v e c t o r space m a y be t r e a t e d as a topological group.

~2.2J n ~ . Let U be a symmetric, radial and a b s o r b i n g subset

of a v e c t o r space E and let X be a character of E such that
Ix(U) I < ½. T h e n there exists e x a c t l y one f e E# with Pf = X. For

each u ~ U, one has f(u) = X(U); consequently,

1
sup {If(u) : u e U} = Ix(U)[ < ~.

Px.6,~ The u n l q u e n e s s of f is obvious. We shall p r o v e the ex-

u
istence. Given u ~ E, we can find a positive integer n with ~ e U.
Set f(u) = nx(~). This d e f i n i t i o n does not d e p e n d on the choice of n.
Indeed, suppose that m is some other positive integer with ~ ~ U.
m
u u
By ( 1 . 2 ) , we h a v e xlm u)=mX(m-~-ff) a n d x C n ~--~) = n x ( ~ - - ~ ) . Hence

mx(u) = mnX(~n) = nx(U).

It remains to show that f is linear. Take any u,w E E. There

is some m such that ~u , ~
w and -u+w
~- all b e l o n g to U. Then, by (i.i),
 18

,U+W, U (~
X~--m--~ = x(~) + x ),

1 1 1
i.e. ±f(Um + w) = m±f(u) + ~(w), which proves that f is an a d d i t i v e

mapping. Now, take any scalar I ~ R. Let [x] and {x} denote the
integer and the fractional parts of x, respectively. Since f is
additive, for e a c h positive integer p, we have

1 1
f(lu) = f(~plu) = ~f(plu)

= ~f([pl]u + {pl}u) = ~[pl]f(u) + !f({pl}U).p

The latter expression tends to If(u) as p ~ ~. Thus f(lu)= f(u). •

If f is a c o n t i n u o u s linear functional on a t o p o l o g i c a l vector

space E, then pf is e v i d e n t l y a continuous character of E. The
mapping f + pf is a h o m o m o r p h i s m of E* into E ; we shall denote
it b y PE"

(2.3) PR~,osrrxom. Let E be a topological vector space. Then

PE is an a l g e b r a i c isomorphism of E* onto E ; it is a t o p o l o g i c -

al i s o m o r p h i s m of Ec onto Ec.

Pmm~f. That PE maps E* onto E follows directly from (2.2).

The continuity is t r i v i a l . To p r o v e that PE is an o p e n mapping of

Ec onto E , take any U ~ No(Ec). There is a c o m p a c t subset K of

E such that

K 0 : = {f ~ E* : If(u) l ~ 1 for all u ~ K} c U.

The set

H = {tu ~ E : t ~ [-i,i] and u ~ K}

is c o m p a c t , so t h a t H° ~ No(E~)'-v It r e m a i n s to p r o v e that P E ( K 0) D

HO .
SO, take any X ~ H°. There is s o m e f ~ E* with pE(f) = X.
Due to (2.2), we have

sup {If(u) l : u ~ H} = Ix(H)I <= !4 '

which means that f ~ K0 •

 19

~2.4; ~ r r l u m . A topological vector space E is l o c a l l y con-

vex if and only if it is a H a u s d o r f f locally quasi-convex group.

Proof. Let E be a l o c a l l y convex space. Take any symmetric,

closed, convex U E N o ( E ). Next, take any v ~ E \ U. B y the H a h n -
-Banach theorem, there is s o m e f 6 E* with sup {f(u) : u ~ U } < f(v).
Therefore we can find a positive number c such that

sup {f(u) : u ~ U} < c < f(v) < 3c.

f
Set h = 4-~" Since U is s y m m e t r i c , we have

sup { lh(u) l : u e u} < < h(v) < 4"

Then ph E E and

1
sup {Iph(u) l : u ~ U} < ~ < Iph(v) l,

which means that v does not belong to the quasi-convex hull of U.

Since v ~ E \ U was arbitrary, it f o l l o w s that U is q u a s i - c o n v e x •
This proves that E is a l o c a l l y q u a s i - c o n v e x group because symmetric,
closed, convex neighbourhoods of zero form a base at zero in E.
Now, suppose that a topological vector space E is a Hausdorff
locally quasi-convex group• For each subset A of E*, denote

AO = {u E E : f(u) E 1 1
[-~,~] + Z for all f ~ A},

1 1
AO = u ~ E : f(u) ~ [-~,~] for all f e A}.

It f o l l o w s from (2.3) that every quasi-convex subset of E has the

form A° for some A c E*. So, there exists a family A of s u b s e t s
of E* such that that the family {A ° : A ~ A} constitutes a base
at z e r o in E. To p r o v e that E is a l o c a l l y convex space, it suf-
fices to s h o w that {Ao : A ~ A} is s u c h a base, too.
Choose any A E A • We h a v e to s h o w that A ° ~ No(E). Since E
is a t o p o l o g i c a l vector space, there is s o m e radial U E No(E ) con-
tained in A °. Then U c A . •
o

~2.5) PROPQSITI~. Let K be a s u b g r o u p of a topological vector

space E. Then the set n{x-l(0) : X e K°} is e q u a l to the weak
closure of K. Hence K is d u a l l y closed in E if a n d o n l y if it
is w e a k l y closed; it is w e a k l y dense if a n d o n l y if (E/K) = {0}.
 20

Pmmm[. Let ~w denote the w e a k closure of K. Take any u ~ ~w

and X E K °. By (2.3), we have X = Pf for some f ~ E*. Then
f(K) c ker p = Z. This implies that f(u) ~ Z because u 6 ~w. Hence
X(U) = ~f(u) = 0 and, consequently, ~w c n{x-l(0) : X 6 K°} •

To p r o v e the opposite inclusion, choose any u E E \ ~w. There is

a weak neighbourhood of u in E disjoint from K. In o t h e r words,
there are some fl ..... fn ~ E* and e > 0 such that if J f i ( v - u)J< e
for i = l,...,n, then v ~ K. Consider the continuous linear opera-
tor F = (fl,...,fn) : E ~ R n. The set

{(x I ..... Xn) E Rn : Jx i - fi(u)J < ~ for i = 1 ..... n}

is a n e i g h b o u r h o o d of F(u) in Rn disjoint from F(K). Thus F(u)

F(K). According to (1.8), we can find some < ~ (Rn) - with <(F(K)) =
{0} and K(F(u)) # 0. Then KF ~ K O and (KF)(u) # 0. •

A subset X of a v e c t o r space is c a l l e d radial if tu ~ X for

all u ~ X and t ~ (0,i). Let X,Y be two radial subsets of a v e c -
tor space E. We write X ~ Y if X is a b s o r b e d by Y, i.e. if
X c tY for some t > 0. Suppose that X ~ Y. For each linear sub-
space L of E, we denote

d(X,Y;L) = inf {t > 0 : X c tY + L}.

Next, for e a c h k = 1,2,..., we d e n o t e

dk(X,Y;E) = inf d ( X , Y ; L )
L
where the infimum is t a k e n over all linear subspaces L of E with
d i m L < k. If E is c o n t a i n e d in s o m e other vector space F, then
dk(X,Y;F) = dk(X,Y;E). Therefore we may simply write dk(X,Y) in-
stead of dk(X,Y;E). The numbers dk(X,Y) are called Kolmo~orov
diameters of X with respect to Y. We take the following natural
convention: for t > 0, the expression "dk(X,Y) < t" should be read
as "X - < Y and dk(X,Y) < t".

(2.6) u mma. Let X,Y,Z be three radial subsets of some vector

space.

(a) If Z is c o n v e x and X,Y ~ Z, then

dk+l_l(X + Y,Z) ~ dk(X,Z) + dI(Y,Z) (k,l = 1 , 2 , . . . ) .

 21

(b) If X~ Y< Z, then

dk+l_l(X,Z) ~ dk(X,Y)dl(Y,Z).

For (b), see the p r o o f of L e m m a 7.1.2 in [79]. The proof of (a)

is s i m i l a r a n d we l e a v e it to the reader.

~2.7) ~ . Let (m n ) n = l be a s e q u e n c e of v e c t o r spaces and let

m be a p o s i t i v e integer. Suppose that, for e a c h n = 1,2,..., we are
given radial subsets Xn,Y n of En such that

(I) d k ( X n , Y n) < 2 - m n k -m (k = 1,2 .... ).

Then X : = II X and Y : = II Yn are radial subsets of the s p a c e

n=l n n= 1

E : = IIEn, and dk(X,Y) ~ k -m for e v e r y k. If EO is the sub-

n=l
s p a c e of E consisting of f i n i t e sequences (that is, Eo = ~ En),
n=l
then dk(X N Eo, Y N E o) ~ k -m for e v e r y k.

Pm~. Let [x] denote the i n t e g e r p a r t of x. F i x an a r b i t r a r y

k = 1,2 . . . . . If 1 = [k2-n], then, by (i),

d l + l ( X n , Y n) ~ 2 - m n ( l + i) -m < 2 - m n ( k 2 - n ) -m = k -m.

So, for e a c h n = 1,2,..., we can find a subspace Ln of En with

dim Ln ~ [k2 -n] and Xn c k - m y n + L n . Then L : = ~ Ln is a sub-

n=l
s p a c e of E and w e h a v e
o

dim L = ~ dim L ~ ~ [k2 -n] < X k2 -n = k.

n=l n n= 1 n= 1

It r e m a i n s to o b s e r v e that X c k-my+ L and X N Eo'C k-m(Y N E o) + L. •

~2.S2 I . Let E,F be s o m e v e c t o r spaces a n d let X,Y be ra-

dial subsets of E with X ~ Y.

Ca) For each linear operator ~ : E ~ F, one has

dk(~(X),#(Y)) ~ dk(X,Y) (k = 1,2 .... ).

(b) For each linear operator ~ : F ~ E with ~(F) = E , one has

d k ( ~ - l ( x ),¥-l(Y)} = d k ( X , y ) (k = 1,2 .... ).

 22

This is a d i r e c t consequence of the d e f i n i t i o n of K o l m o g o r o v dia-

meters.

Let E be a n o r m e d space. The distance of a p o i n t u ~ E to a

subset A of E is d e n o t e d by d(u,A). The closed unit ball in E
is d e n o t e d by BE or, sometimes, by B(E). We say that E is a u n i t -
ary space if its norm is d e f i n e d by an i n n e r product. The inner prod-
u c t of v e c t o r s u,w E E is d e n o t e d by u,w).
Let ~ : E + F be a b o u n d e d linear operator acting between norm-
ed spaces. For each k = 1,2 .... , we w r i t e

dk(~ : E ~ F) = d k ( ~ ( B E ) , B F ) -

The numbers dk(~ : E ~ F) are called the K o l m o g o r o v numbers of ~. In

general, they depend not only on ~, but also on F. For example,
= 1
dk(id : 11 ~ c o ) = i, while dk(id : 11 ~ 1 ~) ~ for k = 2,3 ....

(see [76], 11.11.9 and Ii. Ii.i0). If the m e a n i n g of F is clear from

the c o n t e x t , we simply write dk(~) instead of dk(~ : E ~ F) (see
also (2.10)). The following lemma is an i m m e d i a t e consequence of d e f i -
nitions.

~2.9J r ~ . Let X : E" ~ E, ~ : E ~ F and ~ : F ~ F~ be b o u n d -

ed o p e r a t o r s acting between normed spaces. Then

(a) II~II = d l ( ~ : E + F) ~ d2(~ : E ~ F) ~ ... ~ 0;

(b) dk(~# X : E" ~ F') & El~Jlllxl!dk(~ : E ~ F) (k = 1 , 2 , . . . ) .

~2.~oJ ~ . Let ~ : E ~ F be a b o u n d e d linear operator. If F

is a s u b s p a c e of some unitary space F" , then

dk(~ : E ~ F) = dk(~ : E ~ F ") (k = 1 , 2 , . . . ) .

An e a s y proof is left to the reader.

Let X be a c o n v e x , absorbing subset of a v e c t o r space E. The

mapping u ~ inf {t > 0 : u ~ tX} is c a l l e d the Minkowski functional
of X; it is a s e m i n o r m if a n d only if X is s y m m e t r i c .
Now, let p be a s e m i n o r m on E. We denote B = {u ~ E : p(u)&l}.
P
Sometimes we shall write B(p) instead of Bp. The quotient space
E/p-l(0) will be d e n o t e d by Ep and the natural projection E ~ Ep
by ~p. We shall always consider on Ep the canonical norm given by
 23

II Sp(U)ll = p(u) for u E E. Thus ~p(Bp) = B(Ep). We say that p is

a pre-Hilbert seminorm if

p2(u + w) + p2(u - w) = 2p2(u) + 2p2(w)

for all u,w E E. This holds if and only if the norm on satisfies E
P
the p a r a l l e l o g r a m identity, i.e. if and only if ~D is a u n i t a r y space.
The following lemma is a direct c o n s e q u e n c e of our definitions.

(2.11) ~ . Let ¢ : E ~ F be a linear operator between vector

spaces. If p is a p r e - H i l b e r t seminorm on E, then ¢(Bp) is an ab-
sorbing subset of the space ¢(E) and the Minkowski functional of
¢(Bp) is a p r e - H i l b e r t seminorm on ¢(E).

Let us suppose that p,q are two seminorms on a vector space E,

such that Bp~ Bq, i.e. such that q £ cp for a certain c > 0. The
canonical operator from Ep to Eq will be d e n o t e d by Apq. We have
have the following c o m m u t a t i v e diagram:

id
E > E

Ep Pq > Eq.

~2.12)~. Let p,q be two seminorms on a vector space E, with

Bp Bq. Then dk(Bp,B q) = dk(Apq : Ep + Eq) for each k = 1,2 . . . . .

The proof is quite easy and we leave it to the reader.

(2.73) T.~aA. Let p be a seminorm on a vector space E and M

a linear subspace of E. If q is a p r e - H i l b e r t seminorm on E and
Bp ~ Bq, then

dk(M A Bp, M A Bq) ~ dk(Bp,B q) (k = 1,2,...).

Pm~of. Let r,s be the r e s t r i c t i o n s to M of p,q, respective

lyly. We may identify Mr and Ms with a subspace of Ep and Eq,
respectively. Accordingly, we may treat Ars : Mr + Ms as a restric-
tion of Apq : Ep ~ Eq. Applying (2.12), (2.10) and (2.9) (b), for
each k = 1,2 .... , we have
 24

dk(M N Bp, M N Bq) = dk(Br,Bs) = dk(Ars : Mr ~ Ms )

= dk(Ars : M r ~ Eq) & dk(Apq : Ep + Eq)

= dk(Bp,Bq). •

The a s s u m p t i o n that q is a p r e - H i l b e r t seminorm is essential.

For example, dk(B(ll),B(l~)) = ½ for k = 2,3 .... , while

dk(C o N B(ll),c ° N B(I~)) = dk(B(ll),B(Co)) = 1

for k = 1,2 .... , (see [76], 11.11.9 and ii.ii.i0).

( 2 . 1 4 ) T.l~mlJ. TO each m = 2,3,... there corresponds a constant

cm > 0 with the following property: if X,Y are two symmetric, con-
vex subsets of some vector space, with dk(X,Y) & k -m for every k,
then there are p r e - H i l b e r t seminorms p,q on span X with X c Bp
Bq c y and dk(Bp,Bq) =< Cm.k -m+2 for every k.

The proof can be obtained by standard methods used in the theory

of bounded operators in Banach spaces. It is rather long, but presents
no serious difficulties. One may apply, for instance, theorem 8.4.2 of
[75].

(2.15) n]~im~. Let p,q be two p r e - H i l b e r t seminorms on a vector

space E, with dk(Bp,B q) ~ 0. Let (ak)k= 1 and (bk)k= 1 be two
non-increasing sequences of positive numbers, such that dk(Bp,B q)
ak.b k for every k. Then there exists a pre-Hilbert seminorm r on E
such that dk(Bp,B r) & a k and dk(Br,B q) & b k for every k.

This is an easy c o n s e q u e n c e of the spectral theorem for compact

operators. The details of the proof are left to the reader.

( 2 . 1 6 ) r.Jmlu. Let p,q be two p r e - H i l b e r t seminorms on a vector

space E, with Bp ~ Bq. Let us denote

B p0 = {f ~ E # : If(u) I "< 1 for all u ~ Bp} p

Bq
0 = {f ~ E # : If(u)l & 1 for all u ~ Bq}.

Then dk(B0,B0)q
p = dk(B p,Bq) for every k.

Proof. Consider the linear m a p p i n g F : Ep ~ E given by

 25

F(f)(u) = f(~p(U)) (f ~ Ep; u ~ E).

It is not hard to verify that F-I,B0, *

~ p; = B(Ep) and F-I(B ~ ) = Apq(B(Ep)).
* *

Naturally, we may treat F as a m a p p i n g onto F( E~) = span B P0 . Then,

by (2.8) (b),

P = dk(F-l( B0%F-I,B0,.
dk(B0,B0)q q,, ~ pJ)

= dk(Apq(B(Eq),B(Ep))) = dk(Apq : Eq + Ep)

for every k. It is well known that dk(Apq : E q ~ E p ) =dk(Apq : Ep~Eq)

bee e . g . [76], 11.7.8). Hence, by ( 2 . 1 2 ) , f o r eaeh k = 1,2 . . . . . we
have

dk(B0q,B 0"p; = dk(Apq : Ep ~ Eq) = d k ( B p , B q ) . •

A locally convex space E is said to be nuclear if to each con-

vex U ~ No(E) there corresponds some convex W ~ No(E) such that

dk(W,U) ~ ~ for every k.

(2.XT) I m m ~ F Q S I ~ . Let E be a nuclear space. Choose any c > 0

and m = 1,2,... . Then to each convex U E No(E) there corresponds
some symmetric and convex W E No(E ) such that dk(W,U) ~ ck -m for
every k.

This is an easy c o n s e q u e n c e of (2.6) (b) (cf. the proof of Propo-

sition 7.1.1. in [79]).

f2.18) ~ . The m a t e r i a l of this section is standard.

3. Geometry of numbers

By Rn we denote the n - d i m e n s i o n a l euclidean space with the usu-

al norm. The closed unit ball in Rn is d e n o t e d by Bn; there will
be no p o s s i b i l i t y of c o n f u s i o n with the symbol Bp where p is a
seminorm. The n - d i m e n s i o n a l Lebesgue measure on Rn is denoted by
VOln, and the measure of Bn by ~n" We have

~n = V°in (Bn) = ~n/2[F(l + n/2)]-i - (2~___ee)n/2.

Throughout the section, D is an n - d i m e n s i o n a l ellipsoid in Rn

with centre at zero and principal semiaxes E1 ~ "'" ~ ~n" Thus
 26

dk(Bn,D) = ${I for k = l,...,n.

f3.1) P H o P o s r l ~ . Let K be a closed subgroup of R n. Denote by

K the maximal linear subspace contained in K. If K # K o, then
o
there exist linearly independent vectors al,...,am, all orthogonal
to Ko, such that

K = K o + {kla I + ... k m a m : k I ..... k m e Z}.

The proof c a n be found, for instance, in [23], Ch. VII, §i, n ° 2.

Let K be a c l o s e d subgroup of Rn and K° the m a x i m a l linear

subspace contained in K. It follows from (3.1) that every component
of K has the form u + K for a c e r t a i n u ~ K. The subspace K
o o
will be called the zero c o m p o n e n t of K.
By a lattice in Rn we mean an additive subgroup generated by n
linearly independent vectors. According to (3.1), lattices may be de-
fined as n - d i m e n s i o n a l discrete subgroups of R n. Let L be a lat-
n
tice generated by vectors Ul,...,Un; we say that the system (Uk)k= 1
is a b a s i s of L. Let (ek)~= 1 be an o r t h o n o r m a l basis in Rn and

let ~ : Rn + Rn be the linear operator given by ~e k = u k. The q u a ~

tity Id e t ~I is c a l l e d the d e t e r m i n a n t of L and denoted by d(L);
it d o e s not depend on t h e choice of a b a s i s . The set

{u ~ R n : (u,w) e Z for all w E L}

is a l a t t i c e , too. We call it t h e p o l a r lattice and denote by L*. One

has d(L*) = [d(L)] -I and L** = L. All these facts are standard.
The proofs can be found, for e x a m p l e , in [28] or [33].

~3.2) ~ . Let L be a lattice in Rn with L N D : {0}. Then

there exists some u E L* with

(i) 0 < l}ull ~ n($ I ... <n ) - I / n

Proof. Suppose the contrary. Then L* N (rB n) = {0} for a cer-

tain r > n(~ 1 ... <n ) - I / n So, by virtue of t h e fundamental Minkow-

ski t h e o r e m (see e.g. [33], Theorem 1 o n p. 123), we have

d(L*) = 2 -n v o l n
> (rB n) > r n n-n/2
: 2-nrn~ n =

O n the o t h e r hand, from the Minkowski theorem and the assumption that
L N D : {0} we get
 27

d(L) ~ 2 -n vol n (D) = 2 -n ~n<l ... ~n ~ n - I / 2 ~ l "'" ~n"

Hence d(L) d(L*) ~ rnn-n~l ... ~n" Since d(L)d(L*) = i, it f o l l o w s

that r $ n(~l ... ~n )-I/n, which is a c o n t r a d i c t i o n . •

(3.3) ~ , ~ A . Let M be an (n - l ) ~ d i m e n s i o n a l subspace of R n,

n ~ 2. If ~i & "'" ~ Dn-i are the p r i n c i p a l semiaxes of D N M, then
<k ~ nk ~ ~k+l for k = l,...,n-l. Let ~ : Rn ~ M be the orthogonal

projection. If ~i & "'" ~ ~n-i are the p r i n c i p a l semiaxes of ~(D),

then ~k ~ %k & ~k+l for k = l,...,n-l.

This is a w e l l - k n o w n geometrical fact.

(3.4) ~ . ~ A . Let P be an a r b i t r a r y n-dimensional rectangular

parallelepiped in Rn circumscribed on D. Then

diam

~£. Let ~ : Rn ~ Rn be a linear o p e r a t o r such that ~(B n) = D.

We can choose an o r t h o n o r m a l basis (e k ) kn= l in Rn such that P has

the form
n
{ ~ tke ~ : Itkl ~ s~~ for k = l ..... n}
k=l

for some coefficients S l , . . . , s n > 0. For each k = 1 ..... n, one has

sk = sup (¢u,e k) = sup (u,¢*e k) = l]¢*ekH.

u~B u~B n
n
Hence

n 2 1/2 n 2)1/2.
!2 d i a m P = (k=l
~ sk) = (k:iZ H¢*ektl

Now, it r e m a i n s to o b s e r v e that the r i g h t side is equal to the H i l b e r t -

2 2 1/2
-Schmidt norm of ~, i.e. to (~i + "'" + ~n ) " "

(3.5) n w ~ a . Let M be an (n - l ) - d i m e n s i o n a l affine subspace of

Rn with

(i) M n B # 0.
n

Let us s u p p o s e that
 28

(2) ~i 2 + ... Sn 2 < I.

Then D n M is a n (n - l ) - d i m e n s i o n a l ellipsoid; denoting its princi-

pal semiaxes by ~l'''''nn-l' one has

-2 -2 < 1
(3) N1 + "'" + D n - i "

~rG~f. That D N M is an (n - l ) - d i m e n s i o n a l ellipsoid and not

an empty set follows from (i) b e c a u s e (2) implies that int D contains
B . Let u be t h e c e n t r e of D N M and M the linear subspace
n o
M - u. We may assume that D1 & "'" S ~n-l" Then

(4) ~i = dk(Bn,D) (k = l , . . . , n ) ,

-i
(5) nk = dk(B n N Mo,(D n M) - u) (k = l , . . . , n - l ) .

Let # : Rn ~ Rn be a linear operator with ¢(D) = B n. Denote E =

¢( B n ), N = ~(M), N O = ¢ ( M o) and w = Cu. Let 41% "'" ~ ~n be the

principal semiaxes of E. Then ~k = ~ I for k = l,...,n and, by

(2),

2<1.
(6) ~ + "'" + ~n "

Let P be some n-dimensional rectangular parallelepiped circum-

scribed on E with the property that one of its (n- l)-dimensional
faces is p a r a l l e l to N. Let ~ : Rn + N be the orthogonal projec-
O
tion. From (6) a n d (3.4) we get p c Bn, and (i) m e a n s that N N E # ~.
Hence it e a s i l y follows that

(7) ~(p) c (B n n N) - w.

It is c l e a r that n(P) is a n (n - l ) - d i m e n s i o n a l rectangular paral-

lelepiped circumscribed on the (n - l ) - d i m e n s i o n a l ellipsoid n(E). We
may assume that N = R n-l. Hence, by (7) and (3.4), we may write
O

n-i
X d ~n ( ~ (~E ) , ( B A N) - w) < i.
k=l

This implies that

n-i
(8) Z d2(E A N o , ( B n A N) -w) < 1
k=l
 29

because E N N O c w(E). Since, evidently,

dk(E N N o , ( B n N N) - w) = dk(B n N Mo,(D N M) - u)

for k = l,...,n-l, from (8) and (5) we obtain (3). •

t3.6) z.umm. Let M be an (n - l ) - d i m e n s i o n a l subspace of R n,

n ~ 2. Suppose that the p r i n c i p a l semiaxes of D satisfy the condi-
tion gn2 + ... + ~ 2 ~ I. Let r be a f i x e d number belonging to

[0,i] and let E be the set of all those u ~ M for w h i c h the inter-
section of D and the straight line passing through u and perpen-
dicular to M is a s e g m e n t with length not less than 2r. Then E is

an (n - l ) - d i m e n s i o n a l ellipsoid in M. If ~ l , . . . , D n _ 1 are its p r i n -

cipal semiaxes, then ~[2 + ... -2
nn_l ~ 1.

The proof is s i m i l a r to the p r e c e d i n g one. It is also based on

(3.4). We leave the d e t a i l s to the reader.

~3.7) z asiA. Let L be a l a t t i c e in R n with L N D = {0}. Then

n
we c a n find a basis (Uk)k= 1 of L such that

d(Uk, span {ui)i<k) ~ k - l ( ~ l ... ~k )I/k

for k = 1 ..... n (see (2.1)).

Proof. By (3.2), there is s o m e w ~ L* such that

0 < IIwll & n ( ~ l ... ~n )-I/n.

The set {(u,w) : u ~ L} is a n o n - z e r o subgroup of Z, therefore it

has the form pZ for a certain p = 1,2, . . . . Choose some un 6 L
with (Un,W) = p and let M be the o r t h o g o n a l complement of w. It
is c l e a r that L = Zu n ~ (L A M), and that

d(Un,M) = pllwll-I ~ n - l ( ~ 1 ... ~n )I/n

We m a y assume that M = R n-l. Then Ln_ 1 : = L n R n-I is a lattice in

R n-I and Dn_ 1 : = D n R n-I is an (n - l ) - d i m e n s i o n a l ellipsoid in

R n-I with Ln_ 1 n Dn_ 1 = {0}. Let D1 ~ "'" & nn-i be the p r i n c i p a l

semiaxes of Dn_ I. By repeating the above argument, we can find some

Un_ 1 E Ln_ 1 and some (n - 2 ) - d i m e n s i o n a l subspace N of R n-l, such

 30

that Ln_ 1 = ZUn_ 1 ~ (Ln_ 1 n N) and

d(Un_l,N) ~ (n - l ) - l ( q l ... qn_l )I/(n-l)

From (3.3) we get qk ~ ~k for k = 1 ..... n-l. Hence

d(Un_!,N) ~ (n - l ) - l ( ~ l ... ~n_l )I/(n-l)

Next, we m a y assume that N = R n-2, and so on. After n steps we shall

find generators U n , U n _ 1 ..... u I of L, with the d e s i r e d properties. •

(3.8) ~ . Let a e Rn and let K be a s u b g r o u p of Rn with

K N D = {0} and d(a,K) > 1 Then there exists a linear functional f

= 4"
on Rn such that f(K) 1 1
c [~,~] + Z and

n k2 -2/k]i/2
(1) llfll S 1 + [ Z (El ... ~k )
k=l

Proof. By a d d i n g , if n e c e s s a r y , several sufficiently distant gen-

erators, we c a n find a lattice L in Rn with K c L, L N D = {0}
and d(a,L) = d(a,K). So, we m a y assume that K is a l a t t i c e itself.

Due to (3.7), we c a n find a basis n 1

(Uk)k= of K such that

k -I ~k )I/k
d(Uk, span {ui}i< k) (~I "'"

for k = l,...,n. Let Wl,...,w n be the Gram-Schmidt orthonormaliza-

tion of the system U l , . . . , u n. We m a y w r i t e

u k = CklW 1 + ... + CkkW k (k = l , . . . , n )

for s o m e coefficients Ckl. Since Ckk > 0 for e v e r y k, we h a v e

(2) Ckk ~ k-l(~l ... ~k )I/k (k = l , . . . , n ) .

We m a y w r i t e a = alw I + ... + anW n for some coefficients a k. Take

Pn ~ Z such that

la n - P n C n n I ~ {Cnn.

Next, take Pn-I ~ Z such that

 31

1
lan_l - PnCn,n_l - Pn_iCn_l,n_l i --< ~ C n _ l , n _ I.

Having found pn,Pn_l,...,pk, take Pk-i g Z such that

n
lak_ 1 - PiCi,k_l I <= l2 c k - l , k - l '
i=k-i

and so on. After n steps we shall obtain some integers pn,Pn_l,...,

Pl • Set a o = PlUl + ... + PnUn and write a" = a - a o = a { w I + ... +

a n W n. Then, clearly,

(3) lakl -< i C k k (k : 1 ..... n).

Since a° K and d(a,K) > ~1 we have d(a',K) > --

1 whence
= ' = 4'

tla'lJ > 1 Therefore we can find a functional h on Rn with h(a') •

= 4"
1 3
[~,~] and JihJ} < i. Set h k = h ( w k) for k = l,...,n. We shall con-

struct inductively a sequence fl'''''fn of real coefficients such

that

(4) if k - hkl < ckl (k = 1 .... ,n),

(5) Cklf I + ... + Ckkf k ~ Z (k : l , . . . , n ) ,

1 3
(6) a{f I + ... + anf n ~ [~,~].

Put k : 1 in (3). Then it is n o t hard to see that we can fi~d:a c o e f -

ficient fl such that C l l f I • Z, ill hll S Cl~ and

n 1
a{f I + 7 akh k g [~,-~].
k=2

Having constructed fl .... 'fk-l' we can find, by (3), a coefficient

fk such that Ifk - hkl -< c kk'

-I Cklfl + "'" + Ckkfk e Z and

n
1 3
qfl + "'" + a{fk + Z aih i E
[~,~] -
i=k+l

After n steps we shall obtain fl,...,fn satisfying (4) - (6).

Consider the linear functional f on Rn given by f(wk) = fk'

k = l,...,n. By (5), we have f ( u k) ~ Z f o r every k, whence f(K) c Z.

 32

1 3
From (6) we get f(a') E [~,~], so t h a t

1 3
f(a) = f(a" + a O) = f(a') + f(a O) E [~,~] + Z.

Finally, (4) a n d (2) y i e l d the estimate

n 211/2
llfll ~ {lhll + llf - hll ~ llhll + [ ~ If k - hkl
k=l

n
~ 2 1/2 n k2 -2/k~I/2
1 + [ X Ckk~ 1 + [ ~ (~I " .. ~k )
k=l k=l

~3.9J ~ . Let a ~ Rn and let K be a s u b g r o u p of Rn with

K A (a + D) = 0. Suppose that ~[2 + ... + ~ 2 & i. Tnenthere exists

an o r t h o g o n a l projection w : Rn ~ Rn with 1 ~ dim ~(R n) ~ n, such ~

that w(K) A 2B = {0} and d(~(a),~(K)) ~ i.

n

Proof. We shall apply induction on n. For n = i, the lemma is

trivial. Suppose that it is t r u e for the space R n-l.
If K A 2B = {0}, we take ~ = id. So, assume that there is
n
some u ~ K with 0 < llull ~ 2. Let M be the orthogonal complement
of u and ~M the orthogonal projection onto M. Let E be the set
of all those points v E M for w h i c h the intersection of a + D with
the straight line passing through v and orthogonal to M is a seg-
ment with length not less than llull. It is not hard to see that
WM(K) N E = 0- From (3.6) it f o l l o w s that E is an e l l i p s o i d with

principal semiaxes ql' .... ~n-i such that q~2 + " " + q n --2
i =< 1 " There-

fore, b y our inductive assumption, there is an orthogonal projection

o : M ~ M with 1 ~ dim o(M) ~ n - i, s u c h that ~(~M(K)) A 2(M N B n) ={0}

and d(O~M(a),o~M(K)) Z i. So, we m a y take ~ = o~ M. •

(3.10) ~ I ~ D [ ~ . If al,a2,... is a s e q u e n c e of non-negative

n u m b e r s not all zero, then

Z (a I a 2 ... an)i/n < e ~ an .

n=l n=l

For the p r o o f , see [35J, 9.12.

f3.111 TJmm~. Let a e Rn and let K be a s u b g r o u p of Rn with

K 0 (a + D) = 0. Suppose that
 33

n
(1) Z k ~ .I- < I.
2%
k=l

Then there exists a linear functional f on Rn with llfli < 4, f(K) c Z

1 3
and f(a) E [~,~] + Z.

Proof. We m a y a s s u m e that
n
D = { ( X l , . . . , x n) ~ R n : k=l~k~
"-2X2k "
< I}.

Denote ~k = ( k ~ k ) i / 2 for k = l,...,n, and

n
E = {(Xl,...,Xn) g Rn : ~ nk-2x2k ~ i}
k=l
n
From (i) w e get ~ S I. Thus, by (3.9), there exists an ortho-
k=l
gonal projection ~ : Rn ~ Rn with m : = d i m ~ ( R n) = l , . . . , n , such
that ~(K) 0 2E = {0} and (n(a) + E) n ~(K) = ~. Hence, by (i),

d(~(a),~(K)) > nl
: i~ /2 > 1 > 4"
i

Let 4 1 .< ... < ~m be the p r i n c i p a l semiaxes of the ellipsoid

~ ( R n) n 2E. By (3.8), there exists a linear functional h on Rn

1 3
with h(IT(K)) c Z, h(~(a)) e [~,~] + Z and

m
llhll < 1 + [ ~ k2(~l ... ~ k ) - 2 / k ] I/2
k=l

Set f = h~. From (3.3) it f o l l o w s that ~k ~ 2~k for k = l,...,m.

Hence, applying (i), (3.10) a n d the i n e q u a l i t y k(k!) - I / k < e, we get

1 m
ilfll = llh~ll < iihll.iIwit = llhli < 1 + ~ [ )q (n I ... q k ) - 2 / k ] I/2
k=l

m k2 -2/k( 2 -2 ... k 2 D k 2 ) i / k ] I/2

= 1 + [ ~ (k!) 1 Dl
k=l

1 e 3 / 2 [ m k21]k2]i/2 1 m kl]i/2
<1+ 5 ~ = 1 + [ }= k~
k=l k=l

< 1 + 1 e3/2 L
r ~=
n k~ k l ] i / 2 < 1 + 1 e3/2 < 4. •
k=l
 34

{3.12J T ~ . Let Wl,...,w n be the Gram-Schmidt orthogonaliza-

n 2 n 2
tion of some system u I ..... u n ~ D. Then ~ llWkll ~ ~ ~k"
k=l k=l

This is a d i r e c t consequence of (3.4).

(3.13J ra,mmqa. Let E be an e l l i p s o i d in Rn with centre at zero

and p r i n c i p a l semiaxis ql ,...,q n such that

(1) q12 + . . . 1
+ q2 < 4"

Let K be a s u b g r o u p of Rn and × a character of K, such that

1
(2) Ix(K n Bn) I < ~.

T h e n we can find a closed subgroup K" of Rn with K c K" and a

character X" of K" with X'IK = X, such that all non-zero com-

ponents of K" are d i s j o i n t from E, and

1
(3) IX'(K" n ~Bn) I < I×(K n Bn) I.

Pms~f. By (2) and (1.2), we h a v e

1 1 1 1
IX(K N ~Bn) n < ~ I x ( K N Bn) I < ~ < ~.

Therefore, X is a c o n t i n u o u s character of K due to (1.4). The con-

tinuous extension ~ of X onto K is a c o n t i n u o u s character of K.
Denote B ° = int B . From (2) we obtain
n n

1
(4) I~(K A B°)In --< I×(K N Bn) I <

because B° is open. If all non-zero components of K are disjoint

n
from E, we may take K" = K and X" = ~. So, assume that some non-
-zero component M of K has common points with E. Let Mo be the
zero component of K.

Let wI g M be the v e c t o r orthogonal to M o. We m a y assume that

twI ~ K for t g (0,i). It is e a s y to v e r i f y that the formula

×l(U + tw I) = p[~(U) + t~(wl)] (u E K; t ~ R)

defines a character X1 of the group K 1 : = K + R w I. Notice that

K1 is c l o s e d , being the v e c t o r s u m of the c l o s e d group K a n d of the
 35

compact interval [0,Wl]. Obviously, XIj K = X. Denote

A 1 = {u E B °n : u + W l , U - w I E B~}

It is c l e a r that A1 is s y m m e t r i c and c o n v e x . We s h a l l p r o v e that

(5) JXI(K l A AI) j ~ J~(m A B°)l.

n

So, take any v g K 1 A A I. We m a y w r i t e v = (i - t)u I + tu 2 for

some U l , U 2 ~ K A B°n with u2 - u 1 = w1 and for some t E [0,1). It

is not d i f f i c u l t to see t h a t Ul,U 2 and wI all b e l o n g to K 0 B°

n"
Hence

(61 17(ul)l • l~lu2)l ' 171Wlll ~ I~cK n B°)l

n .

In v i r t u e of (4) a n d (i.I), we h a v e

[(u 2) = ~(u 1) + ~(Wl).

Hence, by (6),

JXI(V) j = J×l(Ul + tWl) j = Jp[~(u I) + t ~ ( w l ) ] j

= JP[(I - t ) ~ ( u I) + t ~ ( u 2 ) ] j ~ J(l - t ) ~ ( u I) + t~(u2) j

J~(K A BO)J.
n

This proves (5).

If all n o n - z e r o components of K1 are d i s j o i n t from E, then we
stop. In the o t h e r case, we c a n r e p e a t the a b o v e argument. We s h a l l
f i n d some v e c t o r w2 orthogonal to M ° + Rw I such that w2+ M ° + Rw 1
is a n o n - z e r o component of K1 h a v i n g c o m m o n p o i n t s with E. We s h a l l
also obtain an e x t e n s i o n ×2 of ×i onto the closed subgrbup

K 2 = K 1 + Rw2, such that JX2(K 2 A A2) j ~ JXI(K 1 A AI) j where

A 2 = {u E A 1 : u + w2, u - w 2 ~ AI}.

Since A1 was symmetric and c o n v e x , so is A 2.

Then we proceed by induction. After several s t e p s we s h a l l o b t a i n

some vectors Wl,W2,...,w p s u c h t h a t the g r o u p Kp= K+Rw I + ... + RWp
is c l o s e d a n d all its n o n - z e r o components are d i s j o i n t from E (it m a y
happen that K p = Rn). The s y s t e m w I .... ,Wp is o b t a i n e d b y the G r a m -
 36

-Schmidt orthogonalization of s o m e s y s t e m of v e c t o r s belonging to E.

We s h a l l also obtain a sequence A 1 D A 2 D ... o A p of symmetric,
convex subsets of Rn such that

A k + 1 = {u ~ A k : u + W k + l , U - W k + 1 ~ Ak} (k = 1 ..... p-l)

and a c h a r a c t e r Xp of Kp with Xpl K = X and IXp(Kp n Ap) I

I~(K n B n)
O l.

Set K" = Kp and X" = Xp. To v e r i f y (3), it is e n o u g h to show

that ½ B n c Ap. So, take an a r b i t r a r y u ~ ½Bn. In o r d e r to prove

that u ~ Ap, w e h a v e to s h o w t h a t u + EpWp ~ Ap_ 1 for Ep = -- i.

+
Next, to p r o v e that u + e p W p E Ap_l, we h a v e to s h o w t h a t u + EpWp +

E p _ l W p _ 1 ~ Ap_ 2 for ep-i = ± i, a n d so on. Thus, we s h o u l d p r o v e

that u + epWp + ... + ~lWl ~ B°n for all s y s t e m s of s i g n s ep,...,£ 1 =

±i. From (3.12) we h a v e

IIw 1112 + . . . + IlWpll2 ~ rl 2 + . . . + T]2.

Hence, by (i),

llu + ~ p W p + ... + £1Wlll ~ ilull + llEpWp + ... + ElWlii

-<5 1+ (,Wp,,2+"'" +,,wi,2)1/2<½+½__i •

~3.~4J rJmm~. Let K be a s u b g r o u p of Rn with K N D = {0} and

let × be a c h a r a c t e r of K. T h e n we c a n f i n d a l i n e a r functional f
on Rn with PflK = X and

(i) iJfll ~ ~ 2 - 3 / 2 [ Xn k2~2]i/2

k=l

ImmDaf. We m a y a s s u m e K to be a l a t t i c e in R n. According to
(3.7), w e c a n f i n d a b a s i s Ul, .... u n of K such that

k-i 1 ~k)i/k
d(Uk, span {ui)i=l) Z k- (~I "'"

for k = 1 .... ,n. Let el, .... e n be the G r a m - S c h m i d t orthogonaliza-

t i o n of the s y s t e m U l , . . . , u n. Then

u k = a k l e I + ... + a k k e k (k = l , . . . , n )
 37

for some coefficients akl with

lakkl Z k-l(~ 1 ... ~k )I/k (k = l,...,n).

Now, we can find, in succession, coefficients fn,fn_l, .... fl such

that

P(aklfl + ... + akkf k) = X(U k) (k = 1 ..... n).

Ifkl ~ 12akk I-I (k = l,...,n).

By taking f(ek) = fk for k = l,...,n, we obtain some linear func-

tional f on Rn with pf(u k) = X(U k) for k = l,...,n. Hence

PfJK = X because (uk)n was a basis of K. Finally, by (3.10)

k=l
and the inequality k(k!) -I/k < e, we get

n n n
)ifl12 = ~ f~ ~ ~ (2akk)-2 < l k2(~l -2/k
k=l k=l " ~ k=iX ... ~k )

= ! ~n k2(12~2.22~2 . . . . k2~2)i/k(k!)_2/k
4 k=l

< e2 ~n (12~i 2 ..... k 2 ~ 2 ) I/k < ~1 e 3 ~n k2~k 2. •

k=l k=l

(3.15J ~mm~a. Let K be a subgroup of Rn and X a character

of K with Jx(K n D) I ~ ¼. Then there exists a linear functional f
n 1
on Rn with PflK = X and llfll ~ 5 ~ k~k
k=l

Proof. Without loss of generality we may assume that

n 1
(1) 7. k ~k : i.
k=l
Next, we may assume that

n 2 -2 ~ i}.
D = {(Xl,...,x n) E R n : ~ Xk~ k
k=l
Define
n
E = {(Xl,...,x n) g R n : Z k-lx2~k I < 2}
k=l

where % = 0.48. By (i), we have

 38

7 d (E,D) = %2
n21
3q k~ < ~.
k=l k=l

So, by (3.13), there exist a closed subgroup K" of Rn such that

all its n o n - z e r o components are d i s j o i n t from E and a c h a r a c t e r X"
of K" with X'IK = X and

(2) Ix'(K N ½D) I S Ix(K N D) I .

Let K° be the z e r o c o m p o n e n t of K'. By (2.3), there is a l i n e a r

functional h on K° with ph = × IK o" From (2) it follows that

Iph(u) l < ¼ for u ~ KO A ½D. Hence lh(u)l <

From (i) we get B n c D, which implies that

(3) llhll < I

= 2"

Let M be the o r t h o g o n a l complement of Ko in Rn and let r =

d i m M. Denote N = K" N M. According to (3.i), we h a v e K" = K o ~N.

Let EM be the o r t h o g o n a l projection of E onto M. Let ql & "'" ~

Dn and ~i ~ "'" ~ ~r be the p r i n c i p a l semiaxes of E and EM, re-

spectively. Applying (3.3) s e v e r a l times, we see t h a t ~k ~ ~k for

~i.1/2~i/2
k = i, .... r. We h a v e qk = ~ %k for k = 1 .... ,n; therefore

1/2 1/2
(4) ~k ~ %k ~k (k = i .... ,r).

Since non-zero components of K" are d i s j o i n t from E, it fol-

lows t h a t N A E M = {0}. In v i r t u e of (3.14), there is a linear f u n c -

tional hi on M with p h l l N = ×'IN and

llhlll & ½ e3/2[ ~r k2 ~k-2 ] 1/2

k=l

Applying (4) a n d (i), we o b t a i n

(5) llhlll £ ~ % e 3/2.

Let w and I be the o r t h o g o n a l projections of Rn onto

K"
o
and M, respectively. Set f = h~ + hl~ I. An easy verification
shows that P f i K = X. From (3) and (5) we o b t a i n
 39

llfll ~ (llhll2 + llhll12)I/2 1 + ~ 1 e 3 ) 1/2

~ (4 < 5. •
4% 2

(3.16) L~mul. Let U be a s y m m e t r i c , radial subset of Rn and let

L be a lattice in Rn such that gp(L N U) = L. Then we can find

generators u I .... ,u n of L such that uk ~ U + span {ui}i< k for

k = 1 .... ,n (see (2.1)).

Pmsof. Choose some u ~ L N U. Let uI be one of the generators

of t h e g r o u p L N Ru. We have U l = tu for a certain t~[-l,l], which
implies that uI e U because U is s y m m e t r i c and radial.
Now, suppose that, for a certain m= l,...,n-l, we have found some
linearly independent vectors u I ..... ,u m E L such that

u k ~ U + span {ui}i< k (k = 1 .... ,m),

m m
g p ( { U k } k = I) = L N s p a n { U k } k = I.

m
L e t us w r i t e M = span { U k } k = I. If L N (U + M) c M, then

L = gp (L N U) c gp (L N (U + M)) c gp (M) = M,

whence dim s p a n L ~ d i m M = m < n, which is i m p o s s i b l e . So, we can

find some v ~ L N (U + M) with v ~ M. Let Um+ 1 be o n e of t h e ele-
ments of t h e set [L n (M + Rv)] \ M which are nearest to M. We have
U m + 1 ~ L N (U + M) because U is r a d i a l and symmetric. It is n o t d i f -

ficult to o b s e r v e that gp ( { U k ~,m+l,

k=l) = L N (M + R U m + l ) . If m + 1 = n,

we are through. If not, then we can repeat the above argument to o b t a i n

some vector Um+2, and so on. After a finite number of steps we shall

obtain vectors Ul,...,u n with the desired properties. •

(3.17) ~ . Let L be a l a t t i c e in Rn wi~h gp (L N D) = L.

Suppose that 512 + "'" + ~ n2 =

< i. Then we can find a rectangular paral-

lelepiped P c ½B n such that {u + P } u E L is a disjoint covering of

Rn .

J~f. Due to (3.16), w e c a n find g e n e r a t o r s Ul, .... u n of L such

that u k ~ D + span {ui}i< k for k = 1 ..... n. So, there are some

Vl,...,v n E D such that u k E v k + span {ui}i< k for k = 1 .... ,n. Let

Wl,...,w n be the Gram-Schmidt o r t h o g o n a l i z a t i o n of t h e s y s t e m Ul, .... u n.
 40

Then Wl, .... w n is the G r a m - S c h m i d t orthogonalization of v I ..... Vn;

so, b y (3.12),

(i) }lWl.12 + ... + llWnl,2 _< ~I ... ~ n~ < I.

Set

p = {tlw I + ... + t n W n : _ !2 <

"
tl t'°"
,t n < ½}

It is n o t h a r d to see t h a t {u + P } u E L is a d i s j o i n t covering of R n.
From (i) we get p c ½ B n. •

(3.]8) c~I~rJ.~Y. Let p,q be two p r e - H i l b e r t seminorms on a v e c t o r

space E, with

(I) Z dd(Bp,Bq)~ & i.

k=l

Let K be a s u b g r o u p of E such that

(2) K = gp (K N Bp).

Then s p a n K c K + iBq.

Proof. Choose any u g s p a n K. In v i e w of (2), we m a y write u =

~lUl + ... + a n U n for some U l , . . . , u n E K N Bp and some c o e f f i c i e n t s

~l,...,~n. We m a y a s s u m e Ul, .... u n to be l i n e a r l y independent.

u n u n
Denote M = span { k}k= 1 and L = gp { k)k=l . Then L is a lat-
t i c e in M. From (i) and (2.13) we o b t a i n

n
~q d 2 ( B p N M,Bq N M) < i.
k=l

So, according to (3.17), the f a m i l y {v + l ( B q N M)}VE L is a c o v e r i n g

of M. Hence u g M c L + l(Bq N M) c K + iBq. •

(3.19) ~J. Let K be a s u b g r o u p of Rn with span K = R n and

2 2<1 1
gp (K N D) = K. Suppose that ~i "'" ~n " 4" Then BnC c o n v (KqBn).

~ . Without loss of g e n e r a l i t y we may assume that K is a lat-

t i c e in R n. It f o l l o w s from (3.18) that

(i) K + Bn = Rn .
 41

Suppose that there is some u ~ (½B n) \ c o n v (K n Bn). Since K is a

lattice, cony (K N Bn) is compact. So, there is some f ~ (Rn) * w i t h
f(u) > i, such that f < 1 on K n B n. We have llfH > 2 because
llull ~ ½. Choose w ~ Rn such that llwll = ~ and f(w) = llfll.Uwll. By
(i), there is some v ~ K with ilw - vll< ! Then llvll < llw - vJl +
l)wll ~ i, i.e. v ~ K N Bn, whence f(v) < i. Thus

JJfll'lJwlt - f(v) = if(w) - f(v) J ~ JlfJl.lJw - VJl

and, consequently,

!
4 < liwll - l~fll-l'f(v) <z llw - vll<" !4 '

which is i m p o s s i b l e . •

(3.20) CORUTJmRT. Let p,q be two p r e - H i l b e r t seminorms on a vec-

tor space E, with

n ~(Bp,Bq) 1
(i) X d ~ 4"
k=l

Then, for any s u b g r o u p K of E, we h a v e

dk(COnV (K n Bp), conv (K N Bq)) ~ 2 d k ( B p , B q) (k = 1,2 .... ).

Proof. Set M = span (K N Bp). We s h a l l p r o v e that

(2) M N ½ Bq c c o n v (K n Bq).

So, choose any u ~ M n ½Bq. We shall find some Wl,...,w n ~ K n Bp

n n
such that u ~ N : span {wi)i= I. Let L = gp {wi}i= 1 and let r,s be

the r e s t r i c t i o n s to N of p,q, respectively. We h a v e Wl,...,w n ~B r

and u e ½ B s. From (I) and (2.13) it f o l l o w s that

Hence u ~ cony (L n Bs) according to (3.19). This p r o v e s (2). From

(2) and (2.13) we get

d k (conv (K N Bp), conv (K n Bq)) ~ dk(M n Bp, M D ½Bq)

dk(Bp,~q) = 2 d k ( B p , B q) (k = 1,2 .... ). •

 42

Let us f o r m u l a t e the results of section 3 in the language of geo-

metry of n u m b e r s . To this aim, we h a v e to i n t r o d u c e some notions. Let
L be a lattice in Rn and let U be a symmetric, convex body in R n.
The successive minima of L with respect to U are d e f i n e d in the
following way:

ki(L,U) = inf {r > 0 : d i m span (L n rU) ~ i} (i = l , . . . , n ) .

The quantity

~(L,U) = inf {r > 0 : L + rU = R n}

is c a l l e d the c o v e r i n g radius of L with respect to U.

~3.21J r m ~ a . Let U be a s y m m e t r i c convex body in R n. For each

lattice L in Rn one has

[~(L,U)] n ~ d(L)/VOln(U).

This is a s t a n d a r d fact; see e.g. inequality (8) in [28], Ch. XI,

§I, n ° 3.

Now, (3.2) c a n be w r i t t e n in the following way: for e a c h lattice

L in Rn , one has

II(L*,Bn)II(L,D) < n(~l ... ~n )-I/n.

The p r o o f of (3.2) also implies that

< 4 -2/n 2n
I I ( L * , B n ) I I ( L , B n)

On the other hand, Conway and T h o m p s o n proved that to e a c h n = 1,2 ....

there corresponds a lattice L n = L n* in Rn such that

1 2 ( L n , B n ) > (5 n l ) 2 / n - 1 = n
2 2~e

(see [66], Ch. II, Theorem 9.5).

In c o n n e c t i o n with these inequalities, it is w o r t h , perhaps, no-

ticing the following thing. Let U be a s y m m e t r i c convex body in Rn

and U0 the p o l a r body, that is,

U 0 = {u E R n : (u,v) ~ 1 for all v ~ U}.

The Bourgain-Milman inequality

 43

VOln(U) VOln(U0) ~ cn~ 2

n

(see [24]) implies that

XI(L,U)XI(L*,U0) & cln

for each lattice L in Rn; here c and cI are some universal con-
stants. On the other hand, from Siegel's mean value theorem one can de-
duce that there exists a universal constant c2 such that to each sym-
metric convex body U in Rn there c o r r e s p o n d s a lattice L with

kI(L,U)XI(L*,U0) ~ c2n;

the proof will be given somewhere else.

Let L be a lattice in Rn and a I .... ,a n some fixed basis of

L. For each x = (Xl,...,Xn) ~ R n, let Fx be the set of all linear
functionals f on Rn such that f(a i) ~ Z + x i for i = l,...,n.
Then it is clear that

(i) ~(L*,B n) = max min llfll.

x~R n feF
x

Lemma (3.14) says that if L n D = {0}, then to each x ~ Rn there

corresponds some f ~ F with
x

½ 3/2 [ ~n k 2 ~ 2 ] 1/2
llfll ~ -e
k=l

Then, by (I),

~(L*,Bn) < 1 e3/2 n k2~k2] i/2

[ ~q
k=l

Consequently, for each lattice L in R n, one has

n
~IT*,Bn)klCL,D) S ~3/2 [ X k2~ 211/2.
k=l

In particular, if D = Bn, then

n 1/2
~ ( L * , B n ) X I ( L , B n) ~ ½e 3/2 [ Z k 2] < 2.25n 3/2
k=l

A somewhat better inequality

~ ( L * , B n ) X I ( L , B n) & ½n 3/2
 44

was o b t a i n e d in [58], Theorem (2.14). On the other hand, from (3.21)

and the M i n k o w s k i - H l a w k a theorem it follows that, for every n, there
exists a lattice L in Rn such that

(2) ~(L,,Bn)XI(L,Bn) > ~2/n = n

2~e"

The methods applied in section 3 also allow to prove the following

result: for every lattice L in R n, one has

n
X i ( L , D ) X n _ i + I ( L * , B n) & 6 ~ k ~i (i = l,...,n).
k=l

The d e t a i l e d proof is given in [ii]. In particular, taking ~i = "'" =

~n = I, we obtain

X i ( L , B n ) X n _ i + I ( L * , B n) ~ 3n(n + I) (i = l,...,n).

This differs only by a constant factor from the bounds

Xi(L,Bn)Xn_i+I(L*,Bn) ~ ~n 2 (i = l,...,n)

which were o b t a i n e d in [58] for n ~ 7.

Let L be a lattice in R n. From (3.11) it follows immediately

that to each a ~ Rn there corresponds some v E L* such that

(3) llvll-lp((v,a)) ~ [8n(n + l)]-id(a,L).

This result was i n d e p e n d e n t l y o b t a i n e d by Hastad [36], with [8n(n + I)] -I

replaced by [6n 2 + i] -I. It follows from (2) that the right side of
(3) cannot be replaced by cn-ld(a,L) with c greater than ~e.
Papers [58] and [36] are based on the notion of the so-called Kor-
kin-Zolotarev bases (see [58]). The proofs of (3.11) and (3.14) given
above are, in fact, similar to those given in [36] and [58], respecti-
vely; K o r k i n - Z o l o t a r e v bases occur in the proofs of (3~8) and (3.14).

~3.22) ~ . The m a t e r i a l of this section is taken from [5], [7]

and [8]. The idea of applying (3.4) in the proofs of (3.5), (3.6) and
(3.12) comes from [i0]. Lemma (3.15) is a s t r e n g t h e n i n g of Lemma 1.5
of [8]. Lemma (3.17) and C o r o l l a r y (3.18) can be found in [15]; (3.19)
and (3.20) are new.
 Chapter 2

EKOTICGRO~PS

It is not hard to find abelian topological groups without (non-

-trivial) continuous characters; perhaps the simplest examples are the
spaces L p, 0 ~ p < 1 (see (2.3) or [38], (23.32)). It is much more
difficult to find abelian groups without non-trivial continuous unit-
ary representations (the so-called exotic groups). The first example of
such a group was obtained only in 1974 by W. Herer and J.P.R. Chris-
tensen [37]. In this chapter we present various constructions of abe-
lian groups without continuous characters or unitary representations.
Section 4 wears a preliminary complexion; we gather here several more
special technical results on continuous representations of abelian to-
pological groups.

4. Representations of abelian topological groups

Let X be a measure space with a positive measure ~. By L~(X,~)

we denote the complex Hilbert space of all (classes of) square-inte-
grable functions on X, with the usual norm. By L~(X,p) we denote
the complex Banach space of all (classes of) essentially bounded func-
tions on X, with the ass sup norm. We may treat Lc(X,~) as an al-
gebra of operators in L~(X,~), identifying a function belonging to
Lc(X,~)) with the corresponding operator of pointwise multiplication.
If ~(X) < ~, we define L~(X,~) as the complex space of all
(classes of) measurable functions on X, with the topology of conver-
gence in measure. This topology can be defined by the F-norm

Ifl = f min (l,lf(x)l)d~(x).

X

If A is a subset of C and p = 0, 2, ~, then by L~(X,~) we de-

note the subset of L~(X,~) consisting of A-valued functions. If
X = (0,i), and p is the Lebesgue measure, we write L~(0,1) instead
of L~(X,~).
By an Ls(X,~)-representation of a group G we mean a representa-
tion of G in the space L~(X,~) by operators belonging to L~(X,~).
 46

Naturally, such representations are u n i t a r y .

If @ is a l i n e a r operator from a vector space E to the space
L~(X,~), then the formula

(~uf)(x) = f(x).exp [2~i(0u)(x)] ( u ~ E; f EL~(X,~); x~X)

2~i@
defines an Ls(X,~)-representation ¢ of E; we write ¢ = e

~4.1J ~ m ~ . A linear operator 0 from a topological vector

space E to L~(X,~)- is c o n t i n u o u s if a n d o n l y if e 2~i8 is a con-
tinuous representation of E.

In the p r o o f we need the following fact:

~4.2) ~ . Let X be a m e a s u r e s p a c e with a finite, positive

0
measure ~. If f g LR(X,~) and If(x) l ~ 1 a.e., t h e n we can find
some t g (0,i) with

~({x E X : tf(x) g [~,~] + Z}) > ~(X).

proof. Let 1 be the L e b e s g u e measure on (0,i). If x E X and

If(x)l ~ i, then an e a s y argument shows that

1 3 2
X({t ~ (0,I) : tf(x) E [~,~] + Z}) _>- ~.

Hence, b y the F u b i n i theorem, we obtain

1
1 3
f ~({x ~ X : tf(x) ~ [~,~] + Z})dt
0

= f l({t ~ (0,i) : tf(x) ~ [~,~3 + Z})d~(x) > ~(X).

X

This implies that the function under the first integral must assume a
value not less than ~(X) for a certain t E 0,i). •

P m a o f a f ~4.1). Denote ¢ = e 2~i@ Suppose first that @ is c o n -

tinuous. To p r o v e the continuity of ¢, choose any f ~ L~(0,1) and

E > 0. There is s o m e 6 > 0 such that if Y is a m e a s u r a b l e subset

of X with ~(Y) < 6, then

f If(x)12d~(x) < e.
Y
 47

Since 0 is c o n t i n u o u s , there is some U ~ No(E) such that

~({x ~ x : l(0u)(x)l > E}) ~ 6

for u ~ U. Now, choose any w,v ~ E with w - v ~ U and d e n o t e

Y = {x ~ X : l(@w)(x) (0v)(x)] > ~}.

Then

l]¢wf - ¢vfll

= f ]@(x).exp [2~i(Sw)(x)] - f(x)'exp [2~i(0v)(x)]]2d~(x)

X

= f + f ]exp [2~i(0w)(x)3
Y X\Y
- exp [2~i(Sv)(x)]]2.]f(x)]2d~(x)

4 f ]f(x)]2d~(x) + (2~E) 2 f ]f(x) 12d~(x)

Y X\Y

< 4E + 4~2E2]Ifll 2.

Since ~ > 0 was arbitrary, this proves that ¢ is c o n t i n u o u s .

Now, suppose that @ is n o t c o n t i n u o u s . T h e n we c a n f i n d a con-

stant c > 0 such that each U ~ No(E) contains a vector u with

~({x ~ x : l(ou)(x)[ ~ e}) ~ c.

Hence, by ( 4 . 2 ) , each U E No(E ) contains a vector u with

(i) p({x ~ X : (0u)(x)~ [¼,~] + Z}) ~ b

(we use t w i c e the f a c t t h a t E has a b a s e at z e r o c o n s i s t i n g of radial

sets). Set fo ~ i. From (i) we get

ll~ufo - ~0foll 2 = f ]I - e x p [2~i(0u)(x)]12d~(x) ~ ~c

X
13
becasue ]I - e x p [2~is]] 2 ~ 2 whenever s e [~,~] + Z. This means
that ¢ is not c o n t i n u o u s . •

Observe t h a t the r e p r e s e n t a t i o n e 2~i0 is u n i f o r m l y continuous if

and o n l y if @ is a c o n t i n u o u s operator from E to LR(0,1).
 48

(4.3)~l, Fx~I_ Let X be a m e a s u r e space w i t h a p o s i t i v e mea-

sure ~ and let ¢ be a c o n t i n u o u s Ls(X,~) - r e p r e s e n t a t i o n of the
group R. T h e n there is a u n i q u e f u n c t i o n 0 E L~(X,~) with

(0sf)(x) = f ( x ) . e x p [2~is (x)] (s ~ R ; f ~L~(X,~); xEX).

This is a c o n s e q u e n c e of Stone's t h e o r e m on c o n t i n u o u s one-para-

meter groups of u n i t a r y o p e r a t o r s (see e.g. [47], T h e o r e m 5.6.36).

(4.4)~5~DI'QSlTIQg_ Let X be a m e a s u r e space w i t h a finite, posi-

tive m e a s u r e ~ and let ¢ be a c o n t i n u o u s Ls(X,~) - representation
of a t o p o l o g i c a l vector space E. T h e n there exists a unique c o n t i n u -
ous linear o p e r a t o r 8 : E ~ L~(X,~) with ¢ = e2~i8

l~f. For each fixed u E E, the m a p p i n g s + #(su) is a con-

tinuous Ls(X,~) - r e p r e s e n t a t i o n of R. Thus, by (4.3), there is a
0
unique function % u ~ LR(X'D) with

[¢(su)f](x) = f(x). exp [2~iS%u(X)] (s ~ R ; f ~ L C2(X,~); x E X )

From the u n i q u e n e s s of % in (4.3) it follows e a s i l y that %su = S%u

and % u + v = %u + %v for all u,v ~ E and s E R. So, the mapping
0 : E ~ L~(X,~) g i v e n by u ~ %u is linear. The c o n t i n u i t y of 8 fol-
lows from (4.1). •

(4.5) P~Di~srrx(1_ Let K be a s u b g r o u p of a s e p a r a b l e t o p o l o g i c a l

vector space E. If the q u o t i e n t g r o u p E/K admits a non-trivial con-
tinuous u n i t a r y r e p r e s e n t a t i o n , then there exists a non-zero continu-
ous linear o p e r a t o r 8 : E ~ L~(0,1) with 8(K) c L~(0,1).

Proof. Let ~ be a n o n - t r i v i a l c o n t i n u o u s u n i t a r y r e p r e s e n t a t i o n
of E/K in a H i l b e r t space H. In v i e w of (1.21), we m a y assume # to
be cyclic. This implies that H is separable. Let A be the algebra
of o p e r a t o r s in H g e n e r a t e d by o p e r a t o r s ~g, g ~ E/K. Then A is
an a b e l i a n s e l f - a d j o i n t algebra in H c o n t a i n i n g the identity operator.
The c l o s u r e A of A in the strong o p e r a t o r t o p o l o g y is an abelian
yon N e u m a n n algebra in H. T h e r e f o r e we can d e c o m p o s e H into an at
most c o u n t a b l e H i l b e r t sum of A-invariant subspaces Hn such that,
for each n, either dim H n = 1 or the r e s t r i c t i o n of A to Hn is
u n i t a r i l y i s o m o r p h i c to the a l g e b r a LC(0,1) in the Hilbert space
 49

9
L~(0,1). The last s e n t e n c e follows from the s t a n d a r d results on the
s t r u c t u r e of a b e l i a n von N e u m a n n algebras; perhaps the best r e f e r e n c e
here will be [47], S e c t i o n 9.4.
The s u b s p a c e s Hn, b e i n g i n v a r i a n t for A, are i n v a r i a n t for ~.
So, ~ can be d e c o m p o s e d into a H i l b e r t sum of some r e p r e s e n t a t i o n s
~n w h i c h are either o n e - d i m e n s i o n a l or unitarilyequivalent to LS(0,1)-
-representations. One of these r e p r e s e n t a t i o n s is non-trivial, there-
fore we m a y assume that ~ itself is o n e - d i m e n s i o n a l or an LS(0,1)-
-representation. In the first case, E/K admits a non-trivial con-
tinuous character; let us d e n o t e it by ×. Let ~ : E ~ E/K be the
natural projection. Then X~ e E . By (2.3), we have X~ = Pf for
some f E E . Since X~ # 0 and x~(K) = {0}, it follows that f # 0
and f(K) c Z. Now, we m a y d e f i n e 8 by the formula (0u)(x) = f(u)
for u ~ E and x ~ (0,i).
It remains to c o n s i d e r the case w h e n ¢ is an LS(0,1) - repre-
sentation. Then ~ is an LS(0,1) - r e p r e s e n t a t i o n of E and, by
(4.4), there exists a c o n t i n u o u s linear o p e r a t o r 8 : E + L~(0,1) with
e 2~i8 = ~. It is clear that 8 # 0 and 0
8(K) c LZ(0,1 ). •

A n o n - t r i v i a l H a u s d o r f f a b e l i a n g r o u p is c a l l e d exotic if it does
not admit any n o n - t r i v i a l c o n t i n u o u s u n i t a r y r e p r e s e n t a t i o n s . We say
that G is s t r o n g l y exotic if it does not admit any w e a k l y c o n t i n u o u s
r e p r e s e n t a t i o n in H i l b e r t spaces.
In c o n n e c t i o n w i t h the above definition, let us notice that e v e r y
topological group (abelian or not) admits a f a i t h f u l s t r o n g l y c o n t i n u -
ous r e p r e s e n t a t i o n by b o u n d e d o p e r a t o r s in a suitably chosen Banach
space. It suffices, for instance, to c o n s i d e r the r e p r e s e n t a t i o n by
shift o p e r a t o r s in the space of b o u n d e d and u n i f o r m l y c o n t i n u o u s func-
tions on the group.
An a b e l i a n t o p o l o g i c a l g r o u p G is said to be b o u n d e d if to each
U E No(G ) there c o r r e s p o n d a p o s i t i v e integer n and a finite subset
A of G, such that A + Un = G (this d e f i n i t i o n m a k e s sense also for
n o n - a b e l i a n groups). For instance, all c o m p a c t or c o n n e c t e d groups are
bounded. If K is a s u b g r o u p of a n o r m e d space E, the quotient g r o u p
E/K is b o u n d e d if and o n l y if there exists a n u m b e r r > 0 such that
E = K + rB (cf. the proof of (5.1) (b) below).

( 4 . 6 ; J OrE. In s e c t i o n 18, the e x p r e s s i o n " b o u n d e d group" w i l l be

used in a c o m p l e t e l y d i f f e r e n t meaning.
 50

A representation ¢ of a g r o u p G in a B a n a c h space is called

bounded if sup {11¢(g)II : g • G} < ~.

(4.7) r . ~ J . Every weakly continuous representation of a bounded

and metrizable group in a B a n a c h space is b o u n d e d .

~ . Let ¢ be a weakly continuous representation of a b o u n d e d

and metrizable group G in a B a n a c h space E. For simplicity, let us
assume that G is a b e l i a n . Let (gn) be an arbitrary null-sequence in

G. For each u • E, the sequence (¢(gn)U)n=l converges weakly to u

and is t h e r e f o r e bounded. Hence, by the Banach-Steinhaus theorem, we
have

(i) sup {II¢(gn)II : n = 1,2 .... } < ~.

Since (gn) was an arbitrary null-sequence and G is metrizable, from

(i) it f o l l o w s that there is s o m e U • No(G) such that

C = sup {II¢(g) : g • U} < ~.

Since G is a b o u n d e d group, there are a positive integer n and a

finite subset A of G, such that A + U n = G. Then

sup {Ji¢(g) : g E G} < Cn max {ll¢(g)ll : g • A} < ~. •

(4.8) [ ~ m ~ . Every bounded representation of a n a b e l i a n group in a

Hilbert space is e q u i v a l e n t to a u n i t a r y representation.

Pz~f. There is an i n v a r i a n t mean o n the space of all bounded,

real-valued functions on an abelian group [[38~, (17.5)), and it is
enough to repeat the standard argument for compact groups (cf. [53],
Exercise 1 in sect. 9.3). •

(4.9) P ~ P o s l T i ~ . Every non-trivial abelian group of automorphisms

of a B a n a c h space admits a non-trivial character continuous in the u n i -
form topology.

~f. Let G be a n o n - t r i v i a l abelian group of automorphisms of

a Banach space and let A be t h e complex Banach algebra spanned over
G. Let m be the set of a l l m u l t i p l i c a t i v e linear functionals of A.
Suppose first that there are some g E G and f • ~ such that
f(g) # I. The multiplicative group C \ {0} admits a continuous cha-
racter X such that k(f(g)) # i. Then kf is a non-trivial continu-
ous c h a r a c t e r of G.
 51

So, w e m a y assume that f(g) = 1 for all g ~ G and f em. This

implies that G c e + rad A where e is t h e u n i t of A and rad A
the radical of A. Since e + r a d A c e x p A, we obtain G c e x p A. Thus
we can find an e l e m e n t a ~ A such that exp a E G and e x p a # i.
Then a ~ 2~iZe and the Hahn-Banach theorem implies the existence of
a continuous R-linear functional f : A ~ R such that f(2~iZe) c Z
and f(a) ~ Z.
Since span G is d e n s e in A and all functionals in m are tri-
vial on G, it f o l l o w s that m consists of o n l y one element. There-
fore the exponential mapping is s i m p l y periodic, i.e. exp x = e im-
plies that x ~ 2~iZe (see e.g. [39], Sect. 5.6). Hence 0f(x)= 0 if
e x p x = e. Consequently, there exists a continuous homomorphism × of
the multiplicative group exp A into T such that the diagram

f
A > R

X
exp A - - > T

commutes. Moreover, x(exp a) = p ( f ( a ) ) # 0. Thus Xj G is a c o n t i n u -

w

ous non-trivial character of G. •

From (4.9) it f o l l o w s that if an abelian topological group does

not admit non-trivial continuous characters, t h e n it d o e s n o t admit a n y
non-trivial uniformly continuous representations in B a n a c h spaces.

(4.1o) ~;~n~LE. Let H be a complex Hilbert space with an ortho-

normal basis ( e n ) n = I. L e t us s e t

ele n = ene I = e n for n = 1,2,...;

e2e n = ene 2 = 0 for n = 2,3,...;

ene m = eme n = 0 for m,n = 3,4,...; m # n;

2
e n = e 2 + en for n = 3,4,...

A direct verification shows that these formulae define on H a struc-

t u r e of a c o m m u t a t i v e Banach a l g e b r a with unit e I. The multiplicative
group G = exp H may be identified with H / e x p -I (el). A direct calcu-
 52

lation shows that exp-l(el) is g e n e r a t e d by the e l e m e n t s 2~ie I and

-i
2~i(e 2 + an), n = 3,4, . . . . So, exp (e I ) is not w e a k l y c l o s e d in
H and (2.5) implies that c o n t i n u o u s c h a r a c t e r s do not s e p a r a t e p o i n t s
of G (more precisely, x(exp 2~ie 2) = 0 for all X ~ G-).
Nevertheless, G admits a continuous, f a i t h f u l u n i t a r y represen-
tation in a s e p a r a b l e H i l b e r t space. The proof is similar to that of
(5.1) (e) and c o n s i s t s in constructing a continuous linear operator
0 : H ~ L~(0,1) w i t h the p r o p e r t y that 0u ~ L~(0,1) if and only if
-I
u E exp (el).

(4.11) ~ . The m a t e r i a l of this s e c t i o n is standard, with the ex-

c e p t i o n of (4.7), (4.9) and (4.10) w h i c h are taken from [9] and [6].

5. Q u o t i e n t s of n o r m e d spaces

By co and i p, 1 ~ p < =, we d e n o t e the c l a s s i c a l Banach se-

quence spaces w i t h their usual norms. By an, n = 1,2,..., we d e n o t e
the s e q u e n c e (0, .... 0,i,0,...) with 1 in the nth place.

(5.1) ~ . Let (an)n= 1 be a s e q u e n c e in 11 such that

(I) a n ~ span {ei}i< n for each n = 1,2,...

co
(see (2.1)). Suppose that {an}n= 1 is a dense subset of 11 and let

K = gp {a n + an}n= I. Then

(a) K is a d i s c r e t e s u b g r o u p in each of the spaces co and ip ,

i < p < ~;

(b) the groups Co/K and IP/K, 1 < p < =, are bounded;

(c) the g r o u p s Co/K and IP/K, p > i, do not a d m i t any non-

-trivial c o n t i n u o u s characters;

(d) the groups Co/K and IP/K, p > 2, are s t r o n g l y exotic;

(e) the groups IP/K, 1 < p < 2, admit faithful continuous

unitary representations in s e p a r a b l e H i l b e r t spaces;

(f) II/K admits s u f f i c i e n t l y m a n y c o n t i n u o u s characters.

Pm~. T h r o u g h o u t the proof, E d e n o t e s one of the spaces co

and 1 p, 1 & p < =, and $ : E ~ E/K is the n a t u r a l projection.
 53

(a) Choose any u E K \ {0}. We m a y w r i t e

u = kl(a I + e I ) + ... + k m ( a m + e m)

for some kl,...,k m ~ Z with k m # 0. From (1) it f o l l o w s that the

mth coordinate of u is e q u a l to k m. Hence, according to the d e f i -

nition of the norm in E, we h a v e llull ~ JkmJ ~ i.

(b) First, we shall prove that

(2) K + 2B E = E.

Choose any u ~ E. Since {an} is d e n s e in 1 I, it is d e n s e in E.

So, there is s o m e n = 1,2,... with u ~ an + BE . Then

u ~ (a n + e n) - e n + B E c K + B E + B E = K + 2BE,

which proves (2). Now, choose any U ~ No(E/K). We have ~(rBE) c U

for s o m e r > 0. Take an i n t e g e r m such that mr ~ 2. Then, by (2),

Um D ~(rBE)m = ~((rBE)m ) = ~(mrBE) D ~(2BE) = ~ ( 2 B E + K)

= ~(E) = E/K.

(c) Let E = co or E = ip with p > i. Since {an} is d e n s e

in E and (e n) converges weakly to zero, it f o l l o w s that K is
weakly dense in E. Now, it r e m a i n s to a p p l y (2.5).

(d) Let E = co or E = Ip with p > 2. Suppose to the c o n t r a -

ry t h a t E/K is n o t strongly exotic. Then, in v i r t u e of (b), (4.7),
(4.8) and (4.5), there exists a non-zero continuous linear operator
8 : E ~ L~(0,1) with 8(K) c L~(0,1). Since 8 # 0, we c a n find s o m e

u ~ E such that the set

23
Q = {t ~ (0,I) : (su)(t) ~ z + [~,~])

has a positive measure. Let (akn) be a s u b s e q u e n c e of (a n ) converg-

ing to u. The sequence (Sa k ) converges in m e a s u r e to 8u. Apply-

n
i n g the Egorov theorem, we m a y assume that 8akn ~ 8u uniformly on Q.
Hence

(Oakn)(t) ~ Z + [~,~]
 54

for t ~ Q and n sufficiently large (say, for n > no). All func-
tions 0(a k + ek ) assume integer values only, therefore
n n
1 3
(0ekn)(t) E Z + [7,7]~
~ for t E Q and n > no •

This implies that

1
(3) l@ek )(t) I ~ for t ~ Q and n > n .
O
n

Set fn = n-1/2@ek for n = 1,2, .... Each s u b s e r i e s of the series

n
n -I/2 is c o n v e r g e n t in E. So, by the continuity of 0, each
n=l ekn
subseries of ~ f is c o n v e r g e n t in L~(0,1). Hence it f o l l o w s that
n=l n

f~(t) < ~ for almost all t E (0,i) (see [73], Lemma on p. 29).
n=l ~
On the other hand, (3) implies that Z f2(t) = ~ for t ~ Q.
n=l n
The contradiction obtained completes the proof of (d).

(e) Let E = ip where 1 & p & 2. Choose any u = (Xn)n=l ~ E\K.

Being discrete, K is c l o s e d in E. Moreover, we have

m
~. X n e n ~ u as m ~ ~.
n=l

Therefore we can find an i n d e x m such that

m
m
W : ----- 7. X n e n ~ L : : gp {a n + e n } n = l ,
n=l

!
(4) Z x~ < 9"
n=m+l

m
Let M = span {a n + e n } n = I. Then L is a l a t t i c e in M and w ~ L,

so t h a t w e c a n find some f ~ M* with f(L) c Z and f(w) ~ Z. Mul-

tiplying f, if n e e d be, by a suitable integer, we may assume that

(5) f(w) e Z + 12
[~,~].

We shall construct a bounded linear operator e : E ~ L~(0,1) with

0(K) c L (0,I) and 0u ¢ L Z ( 0 , 1 ) . We may write

n-i
(6) an = k=l
Z ~knek (n = 2 , 3 , . . . )
 55

for some coefficients ~kn" We shall construct step - functions

hl,h2,.., on (0,i) such that lhm+ll, lhm+21 .... < I, the functions

hm+l,hm+2,.., are pairwise orthogonal and

n-i
(7) hn + ~q ~ k n h k E LZ2(0,1)
k=l

for n = 2,3 . . . . . Set h n --- f ( e n) for n = l,...,m. The functions

hm+l,hm+2,.., will be constructed inductively.

Suppose we have constructed functions hl,...,hn_ 1 for a certain

n > m + i. The interval (0,I) decomposes into a finite union of s o m e

intervals Ii such that, for each i, the functions hl,...,hn_ 1 are
constant on I i. Fix an arbitrary index i a n d let p be t h e v a l u e
n-i
of t h e function Z eknhk on I i. L e t us w r i t e I i = (a,b). If p c Z,
k=l
we set hn - 0 on I i. If p ~ Z, then

c : = b + (b - a ) ( [ p ] - p) ~ (a,b)

and we define

l
ip] - p for t ~ (a,c)

hn(t) =
[p] - p + 1 for t ~ (c,b).

Here [p] denotes the integer part of p. Then lhn(t)l < 1 and

n-I
hn(t) + ~.. a k n h k ( t ) = hn(t ) + p e Z
k=l

for t ~ I.. Moreover, we have

1
b
f hn(t)dt = 0.
a

In t h e same way we define hn on other intervals. Then (7) is satis-

fied and it is c l e a r that hn is o r t h o g o n a l to h l , . . . , h n _ I.

The sequence (h n ) n = l has the desired p r o p e r t i e s . We only have to

verify (7) for n = 2,3 ..... m. But, by (6), for any such n, we have

n-i n-i n-i

hn + >2 ~ k n h k - f(e n) + Z a k n f(e k) = f(e + Z eknek)
k=l k=l n k=l

= f(e n + a n ) e Z.
 56

Since lhnl ~ 1 for n > m and the functions h m + l , h m + 2 .... are

pairwise orthogonal, it follows that the conditions 8e n = hn,

n = 1,2,..., d e f i n e a b o u n d e d linear o p e r a t o r G : E ~ L~(0,1). From
2
(7) we see that 8(a n + e n) ~ LZ(0,1) for each n = 1,2,...; conse-

quently, @(K) c L~(0,1). On the other hand, from (5) we d e r i v e

m m
@w = 8 n=l~ Xnen ~ n=l~ Xn f(en) = f(w) ~ Z + [~,~]i
2

and (4) yields

lle(u - w)ll = lle Z Xnenll = ( Z x2~i/2 < 1

n~ ~,
n=m+l n=m+l

w h i c h implies that 0u is not an i n t e g e r - v a l u e d function.

We m a y treat e as a continuous linear o p e r a t o r from E into

L~(0,1).- Then, by (4.1), e 2~i0 is a c o n t i n u o u s unitary representa-
tion of E in the space L~(0,1).~ Since e 2~ie is trivial on K, the

formula ~$ = e 2~i0 defines a c o n t i n u o u s u n i t a r y r e p r e s e n t a t i o n ¢ of

E/K in L~(0,1), with ~(~(g)) # i.

So, to each g ~ X : = (E/K) \ {0} there c o r r e s p o n d s a continuous

unitary representation Cg of E/K in L~(0,1), with Cg(g) # i. Be-
ing continuous, Cg remains n o n - t r i v i a l on some n e i g h b o u r h o o d Ug of
g. But X is a L i n d e l o f space, t h e r e f o r e we can find a c o u n t a b l e sub-
set A of X such that {Ug}gEA is a c o v e r i n g of X. Then ~
.~ _ geA g
is a c o n t i n u o u s faithful u n i t a r y r e p r e s e n t a t i o n of E/K i~, a s e p a r a b l e
H i l b e r t space.

(f) In view of (2.5), it is enough to show that K is w e a k l y

c l o s e d in 1 I. The proof is similar to that of (e) and we leave out the
details, the more so that that we shall not use (f) in the sequel. •

(5.2) B~m~RZ. A t o p o l o g i c a l g r o u p is c a l l e d m o n o t h e t i c if it con-

tains a cyclic dense s u b g r o u p ([38], (9.2)). The groups Co/K and IP/K
from (5.1) are monothetic. In fact, the f o l l o w i n g s t a t e m e n t is true: if
K is a l i n e a r l y dense s u b g r o u p of a metrizable and c o m p l e t e vector
space E, then E/K is a m o n o t h e t i c group. The proof, b e i n g not dif-
ficult, is omitted.

It follows from (5.1) (c) and (f) that the group K o c c u r r i n g there
is w e a k l y dense in ip for p > i, but w e a k l y c l o s e d in 1 I. This
 57

gives rise to the q u e s t i o n w h e t h e r each d i s c r e t e (resp. closed) subgroup

of 11 is w e a k l y c l o s e d (cf. [82], P r o b l e m 2). The answer is negative:

(5.3) ~ . E v e r y infinite d i m e n s i o n a l n o r m e d space E contains

a free and d i s c r e t e s u b g r o u p K such that E/K is s t r o n g l y exotic.

The proof is given in [6]. We do not repeat it here b e c a u s e it is

similar to that of (6.1), given below, and even a bit simpler.

(s.4) P l ~ s s r f x ~ . Let K be a free and d i s c r e t e s u b g r o u p of a non-

-zero t o p o l o g i c a l vector space E. If K is w e a k l y dense in E, then
K cannot be d u a l l y embedded.

Proof. Let (ei)i~ I be a system of free g e n e r a t o r s of K. Let X

be the c h a r a c t e r of K g i v e n by x(e i) = ~ for i E I. Suppose that
× can be e x t e n d e d to a c o n t i n u o u s c h a r a c t e r X of E. Then, due to
(2.3), there is a c o n t i n u o u s linear f u n c t i o n a l f on E with Pf = X-

For each i ~ I, we have Pf(ei) = ~(ei)

- = x(ei) = z
~' i.e.

(i) f(ei) E ½ + Z.

This means that (2f)(K) c Z. So, if K is w e a k l y dense in E, then

2f = 0, which contradicts (1). •

It follows from (5.4) that the group K from (5.3) is not dually
embedded. Similarly, the group K from (5.1) is neither dually e m b e d d e d
in co nor in I p, p > i; it is clear, however, that K is d u a l l y
e m b e d d e d in 11 .

(5.5) ~ z s . The first example of a c l o s e d s u b g r o u p of a Banach

space w h i c h is not w e a k l y closed was given by Hooper [41], p. 254. Sid-
ney [823 p r o v e d that if a B a n a c h space has a s e p a r a b l e infinite dimen-
sional q u o t i e n t space, then it c o n t a i n s a w e a k l y dense proper closed
subgroup. He also p r o v e d that if a B a n a c h space X has a normalized
basis (en)~= 1 such that en tends w e a k l y to zero, then X contains

a w e a k l y d e n s e d i s c r e t e subgroup. His m e t h o d is d i f f e r e n t from ours u s e d

in (5.1).

(5.6) B ~ K S . The fact that the groups K in (5. i) and (5.3) are
free is not accidental; Sidney [82], p. 983, p r o v e d that a countable and
 58

discrete subgroup of a normed space must be free. Countability is es-

sential here: LZ(0,1) is a discrete but not free subgroup of LR(0,1).
Conversely, a closed and free subgroup of a Banach space must be discrete.
These facts were pointed out to the author by W. Wojtyfiski.

The example of an exotic group given by Herer and Christensen [37]

was a separable, metrizable and complete topological vector space (the
space of m e a s u r a b l e functions relative to a certain p a t h o l o g i c a l sub-
measure). Naturally, the space L~(0,1) is not an exotic group. We
shall see, however, that it has an exotic quotient space.
We shall have to d i s t i n g u i s h between a m e a s u r a b l e function and its
class of equivalence; the space of all r e a l - v a l u e d m e a s u r a b l e functions
on (0,1) will be denoted by t0 and the class of equivalence of a
function f ~ i0 by {f}. A sequence (fn)n=l~ of functions belonging
to i0 is said to be pointwise linearly independent if to each system
tl,...,t r ~ (0,i) with t I < ... < t r there c o r r e s p o n d some indices

nl, • ..,n r such that det ]fni ( t j ) l [ , j = l # 0.

~5.7) r ~ . Let (f n)n=l be a pointwise linearly independent se-

quence in L0 If 0
0 : LR(0,1) ~ L 10,i) is a continuous linear operator
with 8{f n} = 0 for every n, then 8 = 0.

l~f. Let us write L 0 = L~(0,1) and denote the Lebesgue measure

on (0,I) by I. Let (~k)k=l be a sequence of functions from i 0.
Suppose that the m e a s u r e of the set

X = {t : #k(t) = 0 for almost all k}

is equal to I. Let then (~k)~=l be a sequence of m e a s u r a b l e m a p p i n g s

from (0,i) into itself, such that if A is a m e a s u r a b l e subset of
(0,1) with l(A) = 0, then ~(~kl(A) n {t : ~k(t) # 0}) = 0 for every
k. Given a function f ~ i 0, we set

~ ~k(t)f(~k(t)) if t ~ X
=i
fF(t) =
if t ¢ X.

Then fF s L0 and the m a p p i n g f ~ fF induces, as can e a s i l y be seen,

a continuous linear operator F : L 0 ~ L 0. It was proved by S. Kwapiefi
 59

[56] that, conversely, each c o n t i n u o u s linear o p e r a t o r from L0 into

itself can be r e p r e s e n t e d in such a form. So, to prove our lemma, we
have to show that F = 0 p r o v i d e d that F{f n} = 0 for e v e r y n.
Let us suppose that the latter c o n d i t i o n is satisfied. T h e n there
is a subset Y of X with A(Y) = I, such that, for each t E y,
uo

(i) 5q %k(t)fn(~k(t)) = 0 (n = 1,2 .... ).

k=l

Let us fix an a r b i t r a r y t e Y. Since t E X, there exists an index

m such that

m
(2) ~k(t)f(~k(t)) = ~ #k(t)f(~k(t))
k=l k=l

for any f u n c t i o n f on (0,I). Hence, by (i),

m
(3) 5q %k(t)fn(~k(t) ) = 0 (n = 1,2 .... ).
k=l

Denote r = card { ~ ( t ) }k=l

m and choose indices kl, .... k r e { i , . . . , m }

such that ~kl(t) < "'" < ~k (t). T h e n we can find some coefficients
r
El' .... ~r such that

m r
(4) k=iX ~k(t)f(~k(t)) = j=IZ ~jf(~kj(t))

for any f u n c t i o n f on (0,I). Hence, by (3),

r
(5) j=IX ~jfn(~kj(t)) = 0 (n = 1,2 .... ).

Since the s e q u e n c e (fn) is p o i n t w i s e linearly independent, we can

find indices nl, "''' n r such that det Ifni(#kj(t))Iri,j=l # 0. In

view of (5), this implies that E1 .... = ~r = 0. Hence, by (2) and

(4), we have fF(t) = 0 for each f e L0 Since t was an a r b i t r a r y
element of Y and A(Y) = i, it follows that F{f} = 0 for each
0
f ~ i , i.e. that F = 0. •
In v i e w of (5.7) and (4.5), to o b t a i n an exotic q u o t i e n t space of
L~(0,1) it suffices to find a p o i n t w i s e l i n e a r l y i n d e p e n d e n t s e q u e n c e
(fn) of m e a s u r a b l e f u n c t i o n s such that their classes {fn } are not
l i n e a r l y dense in L~(0,1). There are m a n y such sequences. Perhaps
 GO

nk
the s i m p l e s t one is the sequence of functions t ~ t where (nk)k= 1

is a s e q u e n c e of p o s i t i v e numbers such that nk+I/n k ~ 1 + a and a is

a s o l u t i o n of the i n e q u a l i t y 2a > (i + a) l+I/a (a > 3.403...). That

the s e q u e n c e (t nk) is p o i n t w i s e l i n e a r l y i n d e p e n d e n t follows from the

w e l l - k n o w n fact that if n I < ... < n r and t I < ... < t r. then
n.
d e t l t . l3 l ri,j=l > 0 (see e.g. [77], Part 5, P r o b l e m 76). That the func-

nk
tions t are not l i n e a r l y dense in L (0,i) is a c o n s e q u e n c e of a
result of S. M a z u r [64].

(5.8) ~ . Let (nk)k= 1 be a sequence of positive integers such

that nk+i/n k ~ q for all k, where q is some fixed number larger

than i. From the theory of t r i g o n o m e t r i c series it follows that the

closed linear s u b s p a c e of L~(0,2~) spanned over all functions
sin (nkt) and cos (nkt) , k = 1,2 .... , c o n s i s t s of all f u n c t i o n s of
the form

~q (a k cos (nkt) + b k sin (nkt))

k=l

where Z (ad + b~) < ~, so that it is not the w h o l e space L~(0,2~)

k=l
(see, e.g. [105], Lemma (6.5), p. 203). It seems quite possible that,
at least for c e r t a i n s e q u e n c e s (nk) with nk+i/n k ~ q, the system
c o n s i s t i n g of functions sin (nkt) and cos nkt) is p o i n t w i s e line-
arly i n d e p e n d e n t on (0,2~).

(5.9) ~ . The m a t e r i a l of this section is taken from [9]. Propo-

sition (5.4), Lemma (5.7) and the e x a m p l e of an exotic quotient of L 0,
s u b s e q u e n t to (5.7), are new.

6. Q u o t i e n t s of n o n - n u c l e a r spaces

This section is d e v o t e d to the proof of the f o l l o w i n g fact:

(6.1) ~ . Let E be a m e t r i z a b l e locally convex space. If E

is not nuclear, then it c o n t a i n s a d i s c r e t e s u b g r o u p K such that the
q u o t i e n t group (span K ) / K is exotic.

The proof will be p r e c e d e d by several lemmas. We have to i n t r o d u c e

 61

some n e w n o t a t i o n . Let U,W be two c o n v e x b o d i e s (compact convex sets

with non-empty interiors) in an n-dimensional vector space N. Their
volume ratio (relative to a n y t r a n s l a t i o n - i n v a r i a n t m e a s u r e on N) w i l l
be d e n o t e d by U~. That is,
lWl

~ = vol ~(U)
vol ~(W)

where vol is the L e b e s g u e measure on Rn and ~ : N ~ Rn is any

linear isomorphism.
Let ¢ : E + F be a b o u n d e d linear operator acting b e t w e e n normed
spaces. For each k = 1,2,..., let us d e n o t e

I@(B E A N) I i/k
Vk(¢ : E ~ F) = sup [
N IB F N ~(N)] ~

where the s u p r e m u m is t a k e n o v e r all l i n e a r subspaces N of E with

d i m N = d i m ~(N) = k. If d i m ~(E) < k, we set Vk(¢ : E ~ F) = 0.
If F is a s u b s p a c e of some n o r m e d space F', then obviously,
Vk(~ : E ~ F') = Vk(~ : E ~ F) for e v e r y k. Therefore we m a y s i m p l y
write Vk(~) instead of Vk(~ : E ~ F) (cf. the r e m a r k s b e f o r e (2.9)).
It is c l e a r that Vk(#) ~ II~II for e v e r y k.

(6.21 ;K;FB. The numbers Vk(~) satisfy all conditions in P i e t s c h ' s

definition of s-numbers ([76], Ii.i.i) except monotonicity; it may
happen that Vk+l(~) > Vk(~) (see [59] a n d [3]).

(6.3)/~uw~. Let ~l,...,~s be b o u n d e d operators for which the

composition ~i "'" ~s is d e f i n e d . Then

V k ( ¢ 1 ... Cs ) ~ V k ( ~ I) ... V k ( ~ s) (k = 1 , 2 , . . . ) .

I~,m~f. It is e n o u g h to c o n s i d e r the c a s e s = 2. So, suppose that

we are g i v e n o p e r a t o r s ~i : E2 ~ E1 and ~2 : E3 ~ E2" Let N be an

arbitrary subspace of E3 with dim N = dim (¢I¢2)(N) = k. Then

d i m ~2(N) = dim ~I(~2(N)) = k and we m a y w r i t e

J(~I~2(B(E3 ) A N) I i/k
]
IB(E 1 N (~i~2)(N) I
 62

= [I~I(@2(B(E 3) N N)) I J¢l(B(E2) O ¢2(N))J]I/k

@I(B(E2) n ¢2(N))J JB(E I) n ~I(@2(N))J

I~2(B(E 3) N N)I i/k • [I¢I(B(E 2) n ~2(N))I i/k

= [ ]
IB(E 2) O ¢2(N) I IB(E I) O ¢I(~2(N))I ]

& Vk(¢2)'Vk(~l).

Since N was arbitrary, it follows that Vk(¢l¢ 2) ~ Vk(¢l).Vk(¢2). •

Let ¢ : E ~ F be a bounded operator acting between normed spaces.

By hk(~ : E ~ F), k = 1,2 .... , we denote the Hilbert numbers of
(see [76], 11.4).

(6.4J z~mfA. For every bounded operator ¢ : E + F, one has

Vk(¢ : E ~ F) ~ hk(¢ : E ~ F) (k = 1,2,...).

Pz~Qf. Take any operators X,~ with lixli,ll~ll ~ 1 such that the
composition X¢~ is defined and acts between unitary spaces. Fix an ar-
bitrary k = 1,2, . . . . We have to show that Vk(¢) ~ dk(X¢~). Denote
e = X~. Since e acts between unitary spaces, we have Vk(e) =
[dl(8) ... dk(e)]i/k Hence, applying (6.3) with s = 3, we derive

dk(e) ~ [dl(O) ... dk(e)]i/k = vk(X¢~) ~ Vk(X)Vk(~)Vk(~)

< {IXlllJWllVk(¢) < Vk(¢ ). •

{6.5) z~IMa. Let E be a locally convex space. Suppose that there

exists an E > 0 such that to each continuous seminorm p on E there
corresponds another seminorm p" ~ p with Vk(Ap-p) = o(k-£). Then E
is a nuclear space.

l~m~f. Choose an arbitrary continuous seminorm Po on E. Next,

take an integer s > 5E -I. Due to our assumptions, we can find contin-
uous seminorms Ps ~ "'" ~ Pl ~ Po such that

Vk(ApiPi_l)) = o(k -~) (i = l,...,s).

Hence, by (6.3),
 63

(I) Vk(ApsPo) ~ Vk(A ) ... o(k-~S).

PsPs_l Vk(AplP o) =

Let us denote A = ApsPo : Eps ~ Epo. From the inequality

k k
II dj(A) < ekk! II hj(A)
j :i j :i

([76], 11.12.3) and from (6.4) we derive

k ]I/k k i/k k ]I/k

dk(A) < [ II dj(A) < ekkI[ II hj(A)] < ek[ II vj(A)
j =I j =i j =i
(k = 1,2 .... ).

In view of (i) and the inequality cs > 5, this implies that dk(A) =
o(k-4). Consequently, A is a nuclear operator ([79], Proposition
7.2.2). We have thus shown that, for each continuous seminorm on E,
there exists another seminorm such that the corresponding operator is
nuclear. This means that E is a nuclear space ([79], Theorem 7.2.7). •

g6.6J ramm~. Let U c W be two symmetric convex bodies in some n-

-dimensional vector space N. Let M be some m-dimensional subspace
of N. Then

Iu n MI ~ m_i . Iul
Iw n HI n! Iwl

~ . Suppose first that m = n - I. Then we may assume that

N = R n and

M = R n-I : = {(X 1 .... ,X n) ~ R n : x n = 0}.

Set

h(U) = sup {X n : (X 1 ..... X n) 6 U},

h(W) = sup {x n : (Xl,...,x n) ~ W}.

It is clear that

vol n (U) ~ h(U) VOln_ 1 (U 0 Rn-l),

vol n (W) E ~h(W) VOln_ 1 (W N Rn-l).

Hence
 64

Iu N M I = v o l (U N M) ~ h(W) v o l (U) ~ 1 vol (U) = (n-l): IUl

IW N M I vol (W N M) n h ( U ) v o l (W) n vol (W) n! IWl

Now, suppose that m < n - i. Then we can find linear subspaces

M = Mm c Mm+ 1 c ... c M n = N with d i m M i + 1 = 1 + d i m M.1 for i = m,

m+l,...,n and, due to t h e above, we have

lu n MI z m!
tUN Mm+iI
lw n M i (m + 1 ) ! t w n Mm+ll

> m! (m + i), lulv N M m + 2 ,i

= " ~ .o.

( m + i)! (m + 2)! IW N Mm+21

m! (m + I)~ (n- 1)~ Iu n Mnl

(m + i)! (m + 2)!
. o o

n! Iw n Mnl
= m! IU n NI ,
n~ Iw n NI

g6.TJ r ~ . Let M be a finite dimensional subspace of a n infinite

dimensional normed space E. Then there exists a subspace N of E w i t h
codim N < ~, such that

Ilu + vii > ½11vii for all u E M and v ~ N.

The proof is s t a n d a r d .

(6.~J r ~ . Let E,F be normed spaces and let ¢ : E ~ F be an

injective bounded linear operator such that

(i) lim sup kl/2vk(¢) = ~.

k~

If N is a s u b s p a c e of E with codim N < ~, then

(2) l i m s u p k l / 2 v k ( ¢ i N) =
k~

~f- Denote s = codim N. If L is a k-dimensional subspace

of E, then 1 : = dim (L n N) ~ k - s and (6.6) yields

A : = 11/2 [,I¢(BE N L N N)I ] i/i = 11/2 [ I~¢(BE N L) N N I ] I/i

IBF n ¢(L n N) I IB F N ~(L) N NI
 65

> (~)I/2 k l / 2 i! I¢(BE n L) I i/i

- [~.m IB F n ¢(L)I ]
Let us d e n o t e
I¢(BE D L) 1
Q =
IB F D ~(L)

T h e n we m a y w r i t e
i/(k-s)
A ~ (k - s) I/2 kl/2 [(k - s)!] Ql/(k-s)
k k!

= (k - s, I/2 (k - s ) ! ] i / ( k - S ) k - S / 2 ( k - S ) [ k l / 2 Q l / k ] k / ( k - s )
T ' [ k!

Thus, if k and kl/2 Ql/k are b o t h large, so is A (here s is

fixed). In v i e w of (I), this implies (2). •

(6.9) ZEI~A. Let U be a s y m m e t r i c c o n v e x b o d y in R n. If f is

a linear functional on Rn such that

vol
n ({u ~ u : tf(u) I ~ i}) ~ ½ vol n (u),
then

vol (U) -I f f(u)2du < 7.

n
U

This is an e a s y c o n s e q u e n c e of the B r u n n - M i n k o w s k i inequality (see

[31], Corollary to S t a t e m e n t 2.1).

(6.1o) ~ . Let U be a s y m m e t r i c convex body in R n. Let L be

a lattice in Rn such that

(i) d(L) < ~ n 2 ( n + 2) -n/2 v o l n (U)

and let al,...,a n be some f i x e d b a s i s of L. Then there exist real

numbers Cl,...,c n s u c h that if f is a l i n e a r functional on Rn w i t h

(2) f(a i) ~ c i + Z for all i = 1 ..... n,

then

(3) vol n ({u E U : If(u)l > ~}) > vol n (U).

Proof. Let )J il be the n o r m on Rn defined by

 66

llvll2 = 1 f (v,u)2du, v ~ R n,
vol n (U) U

a n d let D be the c o r r e s p o n d i n g u n i t ball. Then D is a n e l l i p s o i d

and vol n (U)-I/2D is the s o - c a l l e d Binet ellipsoid of U (see [65]).
Let C be the L e g e n d r e ellipsoid of U, defined b y the c o n d i t i o n

0c ( v , u ) 2 d u = oc ( v , u ) 2 d u for all v ~ R n.
U C

The connection between Legendre and Binet ellipsoids as w e l l k n o w n in

mechanics (see [42]); one has

(4) vol (U)-I/2 D = ( n + 2 )I/2c0

n vol (C)
n
where

C O = {u E R n : (u,w) ~ 1 for all w ~ C}.

Blaschke [22] p r o v e d that vol n (C) ~ v o l n (U) (see a l s o [42]). From

this and (4) we get

vol (U) n/2

vol n (D) = (n + 2) n/2 (vol (C)) V°in (C0)
n

(n + 2) n/2 v o l n (C 0)

v o l n (U) 1
= (n + 2) n/2 vol n (C) vol n (C 0) V ° i n (C) v o l n (U)

(n + 2) n/2 2 1
n vol (U)
n
Hence, by (i),

vol n (D) < d(L) -I = d ( L * ) .

This implies that L* + D # R n (see (3.21)). Choose some w ~ L* + D

a n d set c i = (w,a i) for i = l,...,n.
Now, let f be a l i n e a r functional on Rn satisfying (2). Let
v be the v e c t o r defined by f(u) = (v,u) for all u ~ R n. Then, by
(2),

(v - w , a i) = (v,ai) - (w,ai) = f(ai) - ci ~ Z

 67

for i = l,...,n, w h i c h means that v - w e L*. Consequently, v g D,

so that

1 f f2(u)d u = 1 S (v,u)2du = llvll2 > I.

vol n (U) U vol n (U) U

Inequality (3) follows now from (6.9).

~faf t6.1). Let Po ~ Pl & "'" be a sequence of seminorms de-

fining the topology of E. More precisely, we assume that {B(Pn)}n=0
is a base at zero in E. Due to (6.5), we may assume that

(1) lim sup kl/2vk( + Epo ) =

k+~ ApnP o : Epn

for n = 1,2, . . . . Suppose first that all Pn s are norms. We shall

construct inductively a sequence Mo,MI,M2,... of finite d i m e n s i o n a l
subspaces of E.
Set M o = {0}. Next, suppose that Mo,MI,...,Mn_ 1 have been con-
structed. Due to (6.7), there is a subspace Nn of E with codim Nn<~ ,
such that

(2) Po(X + Y) ~ ½Po(y) for all x ~ M + ... + Mn_ 1 and y ~ N n-

From (i) and (6.8) it follows that

limk~sup kl/2vk(ApnPolNn) = ~.

So, we can find a finite d i m e n s i o n a l subspace Mn of Nn such that

(3) k(n) : = dim M n ~ n,

I/k(n)
(4) k ( n ) I / 2 [ I M n A B(Pn)I] > 2~e,
IMn N BlPo) I

and we may continue the induction.

After c o n s t r u c t i n g the subspaces Mn we proceed to the construc-
tion of generators of the subgroup K. Let us fix an arbitrary
n = 1,2, . . . . We may identify Mn with R k(n) ; then Uo: = M n A B ( P o )
and U
n : = M n N B(Pn) are bounded, symmetric and convex subsets of
Rk(n), and (4) says that

(5) V°ik(n) (Un) > (2~e)k(n)k(n) -k(n)/2

V°ik(n) (U o)
 68

According to the M i n k o w s k i - H l a w k a theorem ([33], p. 202 "or. [28], Ch.

VI, §3), we c a n f i n d a l a t t i c e Ln in M n = R k(n) such that

(6) L n N U ° = {0},

(7) d ( L n) < VOlk(n) (Uo).

Let (ani)k(n) be some f i x e d b a s i s of Ln After easy calculations,

i=l
from (7) and (5) we o b t a i n

d(Ln) < ~k(n)-2[k(n) + 2] - k ( n ) / 2 VOlk(n) (U n)

So, b y v i r t u e of (6.10), there exist real coefficients C • w.ith i =

nl
l,...,k(n), such that

(*) if f is a l i n e a r functional on M with f ( a n i ) E Cni + Z

n
for i = 1 ..... k(n), then

1
(8) yOlk(n) ({u a U n : If(u)l > }) ->_ ~ V O l k ( n ) (Un).

Now, an e a s y i n d u c t i v e argument allows us to c o n s t r u c t vectors

(9) W n i a M ° + ... + Mn_ 1 (i = l , . . . , k ( n ) ; n = 1,2,...)

satisfying the f o l l o w i n g condition:

(**) given arbitrary integers n o ~ 0, p # 0, m ~ 1 and j = i,

...,k(n), there is an i n d e x n > no such that

1
(i0) Wni = - ~Cni(Wmj + amj) (i = 1 ..... k(n)).

Let K be the s u b g r o u p of E generated by all vectors W n i + ani

where n = 1,2,... and i = l,...,k(n). We s h a l l p r o v e that K is
discrete. Choose any u ~ K \ {0}. For a c e r t a i n n 3 i, we m a y write

k(n)
u = v + ~ ri(Wni + ani)
i=l

where v E M O + ... + M n _ l , and ri are i n t e g e r s not all equal to

zero. By (2) and (9), we h a v e

k(n) k(n) 1 k(n)

Po(U) = Po(V + ~ riWni + Z riani) ~ 2--Po( i ~ 1 riani)"
i=l i=l
 69

k(n) 1
From (6) we get ~ riani ~ B(Po) , which implies that Pu (u) > 7"
i=l
Finally, denote F = span K and s u p p o s e that F/K admits a non-
-trivial continuous unitary representation. Then, d u e to (4.5), there
exists a continuous non-zero linear operator @ : F + L~(0,1) with
0
0(K) c L Z ( 0 , 1 ). For e a c h n = 1,2 ..... we h a v e

(ii) 0 ( W n i + ani ) E L~(0,1) (i = l , . . . , k ( n ) ) .

For e a c h p a i r n,i with n = 1,2,... and i = l,...,k(n), choose a

measurable function ~ni on (0,i), the c l a s s of which is e q u a l to

0(Wni + ani). By (ii), we m a y a s s u m e that

(12) #ni(t) E Z (i = l , . . . , k ( n ) ; t E (0,i))

for n = 1,2, . . . . It f o l l o w s f r o m our c o n s t r u c t i o n that the vectors

W n i + ani form a Hamel basis in F. Consequently, each vector u ~ F
c a n be w r i t t e n in the f o r m

k(n)
(13) u = Z Z Xni(Wni + ani)
n=l i=l

w i t h all b u t f i n i t e l y m a n y c o e f f i c i e n t s x. e q u a l to zero. N a t u r a l l y ,
1
such a representation is u n i q u e . For t ~ (0,i), let us w r i t e

k(n)
ft(u) = Z Z Xni~ni(t).
n=l i=l

It is c l e a r that ft is a l i n e a r functional on F. If u is given

by (13), then

k(n)
e(u) =
Xni0(Wni + ani).
n=l i=l

This means that, for e a c h f i x e d u e F, the c l a s s of the f u n c t i o n

t ~ ft(u) is e q u a l to 0(u). Observe that, for e v e r y n,

(14) ft(Wni + ani) = ~ni(t) (i = l , . . . , k ( n ) ; t e (0,i)).

Since 0 ~ 0, and {Wni + ani} is a H a m e l b a s i s in F, there

are some m = 1,2,... and j = l,...,k(m) with 0(Wmj + amj) # 0.
Let k be the L e b e s g u e measure on (0,i). Replacing n by m in
(12), we see t h a t t h e r e e x i s t a measurable subset X of (0,1) with
X(X) > 0 and an i n t e g e r p # 0, such that
 7O

(15) ~mj(t) = p for all t e X.

T a k e an a r b i t r a r y integer n > 0. B y v i r t u e of (**), there is

o
an i n d e x n > no s u c h that (I0) is s a t i s f i e d . L e t us w r i t e k = k(n).
Next, fix an a r b i t r a r y t E X. From (i0) w e h a v e

f t ( W n i ) = - l-cnift(w
p m3. + amj ) (i -- I, . . . .,k)

Hence, replacing n by m in (14) a n d u s i n g (15), we get

(16) ft(Wni) = - pl-cni#mj(t) = -Cni (i = l , . . . , k ) .

From (14) a n d (12) we d e r i v e

ft(Wni + a~i) = ¢ni(t) ~ Z (i = l , . . . , k ) .

Thus, by (16),

(17) ft(ani) ~ - ft(Wni) + Z c Cni + Z (i = l , . . . , k ) .

Let us t r e a t ft as a l i n e a r functional on M n. From (*) a n d (17) it

follows that

volk ({u g U n , Ift(u)l > i}) > 1 VOlk (Un) "

Since this h o l d s for e a c h t E X, f r o m the Fubini t h e o r e m w e infer t h a t

there exists a vector un g Un such that

k({t ~ (0,i) : Ift(Un) I ~ I}) > IX(X) "

Consequently, we h a v e

(18) 18(u n) I = f min (i, Ift(Un) l)dt > I X ( X ) .

X

We h a v e thus s h o w n t h a t to e a c h n = 1,2,... there corresponds

o
some v e c t o r u n g B(Pn ) c B(p n ) s u c h t h a t (18) is satisfied. Since
o
{F N B ( P n ) } n = l is a b a s e at z e r o in F, it f o l l o w s that 0 cannot

be c o n t i n u o u s . The contradiction obtained completes the p r o o f .

If Pn s are n o t norms, the a b o v e argument requires only small
technical modifications. •
 71

(6.~u~) R m ~ l ~ . The a s s u m p t i o n of m e t r i z a b i l i t y in (6.1) is essen-

tial. Let E be an i n f i n i t e d i m e n s i o n a l vector space w i t h a countable
Hamel basis. The t o p o l o g y of E is d e f i n e d by some norm and by the
family of all s e m i n o r m s u ~ If(u)[ where f ~ E #. Naturally, E is
a non-nuclear locally c o n v e x space. An e a s y a r g u m e n t shows that each
c l o s e d s u b g r o u p of E is d u a l l y c l o s e d and d u a l l y e m b e d d e d (see [4],
p. 132).

(6.12) BOTM. Theorem (6.1) is new. The m a i n idea of the proof, as

w e l l as Lemmas (6.3) - (6.7), are taken from [4].
 Chapter 3

NUCLEAR GROUPS

7. N u c l e a r groups

(7-I)~ITI~. A Hausdorff abelian group G is c a l l e d nuclear if

it s a t i s f i e s the following condition:

given an arbitrary U ~ No(G) , c > 0 and m = 1,2 .... , there

exist: a vector space E, two symmetric and convex subsets X,Y of
E with

(i) dk(X,Y) ~ ck -m (k = 1,2 .... ),

a subgroup K of E and a homomorphism ¢ : K ~ G, such that

#(K n X) ~ No(G) and ¢(K n Y) c U.

In o t h e r words, G is n u c l e a r if e a c h n e i g h b o u r h o o d of zero con-

tains another neighbourhood which is " s u f f i c i e n t l y small" with respect
to t h e original one. From the proof of (9.6) it f o l l o w s that (i) m a y be
replaced by the condition

-m
d k ( X , y ) < Co k o (k = 1,2 .... )

where co and m° are some universal constants. One may take, for in-

stance, c o = 10 -2 and m ° = 4. The author does not know whether it is

sufficient t o take, say, co = 1 and mo = 1 (cf. (9.2) and (9.3)).

The above definition is r a t h e r complicated. However, it a p p e a r s in

a very natural way when one tries to d e f i n e "intrinsically" a reasonable
class of a b e l i a n topological groups which would include subgroups and
Hausdorff quotients of n u c l e a r spaces. For a simpler definition, see (9.9).

(7.2) PR~SSI~IQm_ A Hausdorff abelian group is nuclear if a n d

only if it s a t i s f i e s the following condition:

given arbitrary U E No(G), c > 0 and m = 1,2, .... there exist:

a vector space E, two pre-Hilbert seminorms p,q on E with

d k ( B p , B q) ~ c k -m (k = 1,2 .... ),
 73

a subgroup K of E and a homomorphism # : K ~ G, such that

~(K n Bp) ~ No(G ) and ~(K n Bq) c U.

Pmaof. The "if" part is t r i v i a l . The "only if " part follows di-
rectly from (2.14). •

(7.3) P ~ o s r T x ~ . Discrete abelian groups are nuclear.

Proof. Set E = {0} in (7.1). •

(7.4)]PJD~6x~om. Every nuclear locally convex space is a nuclear

group.

Px~f. Let E be a n u c l e a r space. Choose arbitrary c > 0, m =

1,2,... and U E No(E ) . By (2.17), we c a n find symmetric and convex

sets W,V E No(E ) such that W c V c U and

dk(W,V) ~ ck -m (k = 1 , 2 , . . . ) .

Now, we m a y take X = W, Y = V, K = E and ~ = id E in (7.1). •

(7.SJPRO~POS~51[~. Subgroups and Hausdorff quotient groups of nu-

clear groups are nuclear.

Pmm~f. Let H be a s u b g r o u p of a n u c l e a r group G. Choose ar-

bitrary c > 0, m = 1,2,... and U ~ No(G). Next, take E,X,Y,K and
as in (7.1).

Set K" = ~-I(H) and ~" = ~i K- Then ~'(K n Y) c U n H and

~'(K" n X) = H n #(K n X) ~ No(H). Since {U n H : U E No(G)} = No(H),

it f o l l o w s that H is a n u c l e a r group.

Suppose now that H is c l o s e d and let ~ : G ~ G/H be the n a t u r a l

projection. Denote ~" = ~ . Then ~"(K n Y) c ~(U) and ¢"(K n X) E
No(G/H) because ~ is open. Since {$(U) : U e No(G)} = No(G/H), it
follows that G/H is a n u c l e a r group. •

( 7 . 6 ) PHG~S/T3[~I~o The product of an a r b i t r a r y family of nuclear

groups is a n u c l e a r group.

Proof. Let G be the p r o d u c t of a family {Gi}i~ I of nuclear

groups. Take arbitrary c > 0, m = 1,2 .... and U ~ No(G). We can

find a finite subset J of I and, for e a c h i E j, scme U i e No(Gi)

 74

such that II Ui × lq G i c U. We may write J = { i n } np= l . By (7.1),

i~J i~J
for e a c h n = l,...,p, we can find a vector space En, two symmetric
and c o n v e x subsets Xn,Y n of En with

d k ( X n , Y n) < c 2 - m n k -m (k = 1 , 2 , . . . ) ,

a subgroup Kn of En and a homomorphism ~n : K n ~ Gi n such that

~n(Kn A X n) ~ N ° ( G i n ) and ~n(Kn N Y n ) ~ Uln. . For each i ~ I \ J,

there exist a vector space Ei, a subgroup Ki of Ei and a surjec-

tire homomorphism #i : Ki ~ Gi (every group is a q u o t i e n t of a free
one; a free group is a d i r e c t sum of Z's which, in turn, is a sub-
group of the corresponding direct s u m of R's). Set

P P P
E = II E x II Ei, X = II X n × II Ei, Y = II Y n x If E..
n= 1 n i~J n=l i~J n=l i~J i

Then X,Y are symmetric and convex subsets of the vector space E.
From (2.7) we get

dk(X,Y) ~ ck -m (k = 1,2 .... ).

Next, define

P
K = II K n × 1[ Ki and ( P x (~i) : K ~ G.
= ~n)n=l i~J
n=l i~J

It is c l e a r that ~(K A Y) c U. On the o t h e r hand, we have

P
IT ~ n ( K n A Xn) × l'I G i c ~(K A X),
n=l i~J

whence ~(K n X) ~ No(G). •

~7.7).lm~Gl,osn~. The limit of an i n v e r s e system of n u c l e a r groups

is a n u c l e a r group.

This is a d i r e c t consequence of (7.5) and (7.6).

d'7.~J namsFoszarxaI. The direct s u m of a c o u n t a b l e family of n u c l e a r

groups is a n u c l e a r group.
 75

~EKR~f° Let (G n ) n = l be a s e q u e n c e of n u c l e a r groups and let G =

Z G . Take arbitrary c > 0, m = 1,2 .... and U e No(G ) . There

n=l n
are some U n e No(Gn) , n = 1,2,..., with U n c U. By (7.1), for
n=l
every n, we can find a vector space En, two symmetric and convex
subsets Xn,Y n of En with

d k ( X n , Y n) < c 2 - m n k -m (k = 1 , 2 , . . . ) ,

a subgroup Kn of En and a homomorphism #n : Kn ~ Gn' such that

#n(Kn A X n) e No(Gn) and %n(Kn A Y n ) c U n. L e t us d e f i n e

E = ~ En, X = ~ Xn, Y = ~ Yn"

n=l n=l n=l

From (2.7) we get dk(X,Y) ~ ck -m for e v e r y k. Setting K = ~ K

n=l n

and ~ = (~n)~=l : K ~ G, we have ~ ( K A Y) c U and Z ~ n l K n N X n) c

n=l
~(K A X), whence ~(K n x) e No(G ) . •

17.9) P R a R ~ r / ~ . The limit of a c o u n t a b l e direct system of n u c l e a r

groups is a n u c l e a r group provided that it is s e p a r a t e d .

This follows directly from (7.5) and (7.8).

The limit of a d i r e c t system of n u c l e a r groups (and e v e n of LCA
groups) need not be separated. See, however, (1.18) and the remarks
following the d e f i n i t i o n of the d i r e c t system of g r o u p s in s e c t i o n i.
From (7.5) - (7.9) we see that the p e r m a n e n c e properties of nu-
clear groups are similar to t h o s e of n u c l e a r spaces.

(7.1o) ~ u m . LCA groups are nuclear.

~ . Let G be an L C A g r o u p s . According to (1.9), there exist

a compact group K, a discrete group D and some m= 0,1,2,..., such
that G is t o p o l o g i c a l l y isomorphic to a s u b g r o u p of R m × K × D. Ap-
plying (7.3) - (7.6), w e see that it is e n o u g h to p r o v e that K is
nuclear. Now, K m a y be identified with a subgroup of s o m e product of
circles. A circle, in turn, is a q u o t i e n t of the nuclear space R. T h a t
K is n u c l e a r follows now from (7.4) - (7.6). •

Let G be a t o p o l o g i c a l group and N a set of indices. By C o ( N , G )

we shall denote the group consisting of all functions f : N ~ G with
 76

the following property: to e a c h U e No(G) there corresponds a finite

subset M of N such that F(v) ~ U for all ~ ~ N \ M. We m a k e
Co(N,G) into a topological group by taking as a b a s e at z e r o the family
of sets of the form {f : f(9) E U for all ~} where U E No(G). If
G is a f i e l d of s c a l a r s , then Co(N,G) is a t o p o l o g i c a l vector space
over G.

~Y.11) PRaB~slrIam. Locally convex spaces over loca!lycompact ultra-

metric fields are nuclear groups.

For the terminology concerning vector spaces over ultrametric

fields, we refer the reader to [68].

Pr~mf. Let K be a locally compact ultrametric field. Every lo-

cally convex space over K is an i n v e r s e limit of B a n a c h spaces over
K. Thus, in v i e w of (7.7), we h a v e to s h o w only that Banach spaces
over K are nuclear groups.
Let E be a B a n a c h space over K. We m a y identify E with the
space Co(N,K) for some set N ([68], Corollary 2, p. 44). From
standard results on the structure of locally compact fields it f o l l o w s
that K, as an a d d i t i v e topological group, m a y be identified with an
inverse limit of d i s c r e t e groups (see e.g. [103], Ch. I, §2). In o t h e r
words, we m a y assume that K is a s u b g r o u p of s o m e product I I D i of
i~I
discrete groups. Then E may be i d e n t i f i e d with a subgroup of the
group H = Co(N , II Di). In v i e w of (7.5), it r e m a i n s to s h o w that H
i~I
is a n u c l e a r group.

We may treat H as a s u b g r o u p of the product

II ( II Di) -- II D
~N i~I (v,i)6NxI ~,i

where D~, i = D i. For each finite subset J of I, let

wj : H + Dj : = II D
(~,i)ENxI ~,i

be the n a t u r a l projection. It is c l e a r that the family {ker ~j},

where J ranges over finite subsets of I, forms a b a s e at z e r o in H.
Thus we m a y identify H with the inverse limit of the d i s c r e t e groups
Dj. That H is n u c l e a r follows now from (7.3) and (7.7). •
 77

(7.12) R J ~ : . L o c a l l y c o n v e x spaces over local fields s a t i s f y Gro-

t h e n d i e c k ' s d e f i n i t i o n of n u c l e a r spaces, b a s e d on tensor products. On
the other hand, a real (or complex) t o p o l o g i c a l v e c t o r space is a nu-
clear g r o u p if and o n l y if it is a nuclear l o c a l l y c o n v e x space (see
(7.4) and (8.9)). T h e r e f o r e it w o u l d be i n t e r e s t i n g to give a charac-
t e r i z a t i o n of nuclear groups similar to some c h a r a c t e r i z a t i o n of nu-
clear spaces. Naturally, tensor p r o d u c t s and b i l i n e a r m a p p i n g s do not
make m u c h sense for t o p o l o g i c a l groups. Nevertheless, one may speak of
s u m m a b l e and a b s o l u t e l y s u m m a b l e families of e l e m e n t s of a b e l i a n topo-
logical groups. In this connection, see (10.16).

{7.13) rJmmu. Let p,q be two p r e - H i l b e r t s e m i n o r m s on a vector

space E, such that k=l~ d~(Bp,Bq) ~ i. Take any U l , . . . , u m ~ Bp with

m ~ 2. If a vector y e E b e l o n g s to the set

{tlu I + ... + tmU m : 0 & t l , . . . , t m ~ i},

then there exists a subset I of {l,...,m} such that 1 & card I ~ m - i

and y - Z u i E Bq.
igI

This is Lemma 4 of [i0].

(7.14) r ~ Let p,q be two p r e - H i l b e r t s e m i n o r m s on a vector

space E, such that ~ d ~ ( B _ , B ) ~ I. Let L be a subgroup of E

k=l ~
and ~ a m a p p i n g from K A 2Bq into some g r o u p G, such that ~(u+w)=
$(u) + ~(w) for all u,w ~ K N 2Bq with u + w g 2Bq. Then

(a) if U l , . . . , u n g Bp and u I + ... + u n g 2Bq, then

~(u I + ... + u n) = ~(u I) + ... + ~(Un);

(b) the formula

n n
¢( Z u i) = Z ~(u I) (Ul,...,u n ~ K N Bp)
i=l i=l

defines a h o m o m o r p h i s m ¢ : gp (K A Bp) ~ G w h i c h is identi-

cal w i t h ~ on the set 2Bq N gp (K N Bp).

Proof. To prove (a), we a p p l y i n d u c t i o n on n. For n = 2, the

 78

validity of (a) is o b v i o u s . So, assume that (a) is t r u e for n less

than some fixed integer m = 3,4, . . . . We shall prove that (a) is t r u e
also for n = m.

Take any u I ..... u m E K A Bp with

(i) uI + ... + u m m 2Bq.

Due to (7.13), there is a s u b s e t I of {l,...,m} with 1 ~ card I

m - i, such that

(2) Z ui -
1
-- i ~ u i E Bq.
iEI 2 1 •

Then, by (i), both ~ u i and ~ ui belong to 2Bq. Now, f r o m our

ieI i~I
inductive a s s u m p t i o n we o b t a i n

m
4( X u i) = %( + Z u i) = ~( ~ u i) + 4( X u i)
i=l iel i~I i~I i~I

m
= Z ~(ui) + X ~ ( u i) = X $(ui).
i~I i¢I i=l

This completes the p r o o f of (a). To p r o v e (b), we o n l y h a v e to show

n m
that # is w e l l - d e f i n e d , i.e. that if 5q u i = 5q wj for s o m e Ul,
i=l j=l
n m
.... u n ~ K n B p and w I ..... w m E K A Bp, then 5q ~(ui) = Z @(uj).
i=l j =i
O n e has ~(0) = 4(0 + 0) = ~(0) + ~(0), whence ~(0) = 0. Next, for
each j = l,...,m, one has

4(wj) + 4(-wj) = ~(wj + (-wj)) = ~(0) = 0,

i.e. ~(-wj) =-$(wj). Thus, by (a),

n m n m
5q. %(ui) - 5q. ~(wj) = >q ~(ui) - 51 ~(-wj)
i=l j =i i=l j =I

n m
= ~( ~ u i + m )) = %(0) = 0.
i=l j=l (-wj

(7.15) RRUPO61TI~. Let G,H be locally isomorphic abelian groups.

If G is nuclear, so is H.
 79

For a d e f i n i t i o n of l o c a l l y isomorphic groups, see [23], Ch. III,

§i, n ° i.

Proof. Choose arbitrary U ~ No(H), m = 1,2,... and c > 0. We

h a v e to f i n d a v e c t o r space E, two pre-Hilbert seminorms p,q on E
with d k ( B p , B q) < ck -m for k = 1,2,..., a subgroup K of E and a
homomorphism ~ : K~H, s u c h that ~(K N Bp) 6 No(H) and %(K N Bq) c U.
We m a y a s s u m e that c < i.
Let y be a l o c a l isomorphism of G into H. Thus y is a ho-
meomorphism of some W ~ No(G) onto some V E No(H ), such that
y(u + w) = y(u) + y(w) for a n y u,w ~ W with u + w ~W. W e m a y as-
sume that U c V. Then U" = y-I(u) E N o ( G ). Since G is a nuclear
group, from (7.2j it f o l l o w s that there exist a vector space E, two
pre-Hilbert seminorms p,r on E such that

1 . -m-i
d k ( B p , B r) <. ~ c K (k = 1 , 2 , . . . ) ,

a subgroup K of E and a homomorphism ~ : K ~ G, such that

$ ( K N Bp) ~ No(G) and ~ ( K N 2Br) c U'. By (2.15), there exists a
pre-Hilbert seminorm q on ~E such that d k ( B p , B q) ~ ck -m and

d k ( B q , B r) & "~k-I for k = 1,2, . . . . We h a v e B p c Bg c B r since

c ~ i. Observe that

d k2( B q, B r) & ¼ Z k -2 < 1

k=l k=l

Let #" be the r e s t r i c t i o n of ~ to the set K N 2B r. For a n y

u , w ~ K N 2B r with u + w s 2Br, we h a v e

~'(u + w) = ~(u + w) = ¥(~(u) + ~(w)) = ¥~(u) + ¥~(w)

= ~'(u) + $'(w).

So, according to (7.14), there is a h o m o m o r p h i s m ~ : K ~ H which is

identical with ~" on t h e set 2B r N gp (K N Bq). From our assump-
tions it n o w f o l l o w s that

~(K N Bq) = ~ ' ( K N Bq) c y ~ ( K N B r) c y ~ ( K N 2B r) c y(U') = U.

O n the o t h e r hand, we have

~ ( K N Bp) = ~ ' ( K N Bp) = 7 ~ ( K N Bp) E No(H )

 80

because ¢(K n Bp) • No(G) and y is a h o m e o m o r p h i s m . •

(7.16) ~ . The m a t e r i a l of this section is new.

8. C h a r a c t e r s of n u c l e a r groups

~8.1) l~FOSlPlq[~. Let p,q be p r e - H i l b e r t seminorms on a vector

space E. Let K be a s u b g r o u p of E and X a character of K with
IX(K D Bq) I ~ ¼. Then there is an f • E# with PfIK = X, such that

sup {If(u)I : u • Bp} & 5 Z kdk(Bp,Bq) ,

k=l

provided that the r i g h t side is finite.

i~m~f. We m a y a s s u m e that

(1) 7 k d k ( B p , B q) = i.
k=l

Consider the c a n o n i c a l diagram

id
E > E

E > E
P q

Take any u,w • K with ~q(U) = ~q(W). Then Cq(m(u-w)) = m~q(U-W)

= 0 for e v e r y m. Hence m(u - w) ~ K n B for e v e r y m. So, due
q
to (1.2), we h a v e

1 1
]X(U - w)] < ~Ix(K N Bq) I < 4--m

for e v e r y m, which implies that X(U) = X(W). This p r o v e s that the

formula <(~q(U)) = K(u) defines a character K of ~q(K). Observe
that

1
(2) I<(¢q(K} N B{Eq}) I = ]x(K n Bq) I < 4"

For each u • K, define

A u = {h e Ep : ph~p(U) = X(u)}.
 81

n
Now, choose any Ul,...,u n ~ K. Let M = span {~q(Uk)}k= 1 c Eq
and m = dim M. It follows from (2.11) that ~q(Bp) is an absorbing
subset of Eq and its Minkowski functional r is a pre-Hilbert semi-
norm on Eq. Therefore M n ~q(Bp) and M n B(Eq) are n-dimensional
ellipsoids in M. So, by (2) and (3.15), there exists a linear func-
tional g on M such that pg = < on M A ~q(K) and
m
(3) sup {Ig(~ I : u E M N ~q(Bp)} ~ 5 ~ kdk(M N ~q(Bp),M N B(Eq)).
k=l
Applying (2.13) and (2.8) (a), we see that

(4) dk(M N ~q(Bp),M n B(Eq)) = dk(M N Br,M n B(Eq)) ~ dk(Br,B(Eq))

= dk(~q(Bp),~q(Bq)) ~ dk(Bp,B q)

for every k. Now, h" = gApq is a linear functional on

A-I(M)
pq , and
from (i), (3) and (4) we get llh'll ~ 5. Let h ~ E be any extension
P
of h" with llhll ~ 5. For each k = l,...,n, we have

ph~p(U k) = ph'~p(Uk) = pgApqCp(U k) = pg~q(U k) = <¢q(U k)

= X(Uk).
, n
Thus h ~ 5B(Ep) N k=iN Auk.

Since the sets Au are all weakly closed and 5B(Ep) is weakly
compact, it follows that there exists some h ~ 5B(Ep) n N A u. It
u~K
remains now to take f = h~p. •

~8.2J ~ . Let H be a subgroup of a nuclear group G. Each

equicontinuous subset of H is the canonical image of an equicontin-
uous subset of G-.

Proof. Let X be an equicontinuous subset of H . There is some

W ~ No(H) with X c W °. Next, there is some U ~ No(G) with U N HcW.
By (7.2), we can find a vector space E, two pre-Hilbert seminorms p,
q on E with

(i) dk(Bp,Bq) ~ (33)-ik -3 (k = 1,2 .... ),

a subgroup K of E and a homomorphism # : K ~ G, such that

~(K N Bp) e No(G ) and #(K N Bq) c U. Define
 82

- 1
Y = {K ~ G : IK($(K n Bp)) I ~ 4"

It is c l e a r that Y is an e q u i c o n t i n u o u s subset of G . It r e m a i n s to
show that each × ~ X is a r e s t r i c t i o n of s o m e X 6 Y.
So, take any X 6 X. let K" = $-I(H). Then X# is a character

of K" and Ix~(K" N Bq) I ~ IX(U N H) I & ¼. By (8.1), there is some

f ~ E# such that Pf = X~ on K" and

sup {If(u) I : u ~ Bp} & 5 X kdk(Bp,Bq).

k=l
Hence, in v i e w of (i), we h a v e

sup {If(u) i : u ~ Bp} ~ 5 (33) -1 1

X k -2 < 7"
k=l

It is c l e a r t h a t the f o r m u l a K($(U)) = pf(u) for u E K defines a

character K of the g r o u p ~(K), with

1
JK(~(K n Bp)) I < s u p {If(u) I : u 6 Bp} < 4"

Next, the f o r m u l a

~($(u) + h) = X(~(u)) + x(h) (u E K, h E H)

defines a character ~ of the g r o u p ~(K) + H. We have ~I~(K) = K

and ~IH = X- Let us e x t e n d ~ in a n y w a y to a c h a r a c t e r X of G (see

(1.6)). Then X ~ Y and XIH = X. "

(s.3) ~ = J m ~ . E a c h s u b g r o u p of a nuclear g r o u p G is d u a l l y embed-

d e d in G. •

(8.4) PRomQsrFi~m. Let p,q be p r e - H i l b e r t seminorms on a vector

space E, with 59 k d k ( B p , B q ) ~ i. Let a E E a n d let K be a sub-

k=l
g r o u p of E, such that a ~ K + Bq. Then there exists some f ~ E#

with f(K) c Z, f(a) 1 3

~ [~,7] + z and sup { If(u)l : u 6 Bp} < 4.

~c~ Define

Q = {h ~ E p : h~p(a) E [ , ] + Z}

and, for e a c h u ~ K,
 83

A u = {h 6 E p : h~p(U) ~ Z}.

we h a v e ~q(a) ~ $q(K) + B(Eq). We may now repeat the p r o o f of (8.1),

applying (3.11) i n s t e a d of (3.15), to s h o w t h a t 4B(Ep)DQ n n Au #
unF
for a n y f i n i t e subset F of K. Since Q and Au are all w e a k l y
closed and 4B(Ep) is w e a k l y compact, there is some h 6 4B(Ep) n Q n
n Au and it r e m a i n s to t a k e f = h~p. •
uEK

iS.5) ~ . Nuclear groups are l o c a l l y q u a s i - c o n v e x .

Proof. Let G be a n u c l e a r group. Take any U ~ N (G). B y (7.2),

o
we c a n f i n d a v e c t o r space E, two p r e - H i l b e r t seminorms p,q on E
with

1
(i) X kdk(Bp,Bq) 5'
k=l

a subgroup K of E and a homomorphism # : K ~ G, s u c h that

#(K N Bp) ~ No(G) and ~ ( K N Bq) c U. To c o m p l e t e the p r o o f , it suf-
f i c e s to s h o w t h a t the q u a s i - c o n v e x h u l l of ~(K n Bq) is contained
in U.
So, take any g ~ G \ U. We are to find some X E G with
JX(%(K n Bp))J < 41- and Jx(g)J > i. Suppose first that g ~ #(K).

Discrete groups admit sufficiently many characters, therefore we can

find a character K of G with Kj~(K ) = 0 and K(g) # 0. Then

JnK(g)J > 1 for a c e r t a i n n a n d w e m a y take X = nK.

Now, suppose that g ~ #(K). Let K" = ker %. Then g = ~(u)

for s o m e u ~ Bq + K'. B y (i) a n d (8.4), t h e r e is s o m e f ~ E with
1 3 1
f(K') c Z, f(u) ~ [~,~] + Z a n d sup {Jf(w) J : w E Bp} < . The for-
mula K($(W)) = pf(w) for W ~ K defines a character K of %(K) with
JK(~(K n Bp))J < and JK(g)J = JK(~(U))J > ~. Let X = < if

JK(g)J > a n d let X = 2K if jx(g) J = • Then JX(#(K n Bp))J <

and j×(g)j > i. •

Closed subgroups of nuclear groups are dually

closed.

Proof° Let H be a c l o s e d subgroup of a nuclear g r o u p G. Choose

any u ~ G \ H. W e are to f i n d s o m e X e G with XjH - 0 and X(U) # 0.
 84

Let ~ : G ~ G/H be the c a n o n i c a l projection. Then ~(u) # 0. Since

G/H is Hausdorff, there is some U E No(G ) with ~(u) ~ U. By (7.5),
the g r o u p G/H is nuclear. Therefore, by (8.5), we can find some quasi-
-convex set W e No(G ) with W c U. So, there is some ~ e (G/H)
with J<~(u) J > ~ and we may set X = <~- "

Closed subgroups of nuclear spaces are weakly

closed.

This follows i m m e d i a t e l y from (8.6), (7.4) and (2.5).

(8.8) ~ . Let K be a d i s c r e t e s u b g r o u p of a nuclear space

E. Then K is an at most c o u n t a b l y g e n e r a t e d a b e l i a n free g r o u p (see
(5.6)). Since E/K admits s u f f i c i e n t l y m a n y c o n t i n u o u s characters, it
admits a c o n t i n u o u s f a i t h f u l u n i t a r y r e p r e s e n t a t i o n (the Hilbert sum
of these characters). It was p r o v e d in [5] that if the t o p o l o g y of E
can be d e f i n e d by a family of norms, then E/K admits a f a i t h f u l uni-
formly c o n t i n u o u s u n i t a r y representation. Simple examples show that it
is e s s e n t i a l to assume that K is discrete.
The c o u n t a b l e p r o d u c t Re of real lines does not admit a faithful
uniformly continuous unitary representations. Indeed, from the proof
of (4.5) it follows e a s i l y that such a r e p r e s e n t a t i o n w o u l d induce an
injective c o n t i n u o u s linear o p e r a t o r ~ : R e ~ LR(0,1), which is im-
possible.

(8.9) l ~ J ~ s l - r ~ . Let F be a t o p o l o g i c a l vector space. If F

is a n u c l e a r group, then it is a n u c l e a r locally c o n v e x space.

l~Qf. From (8.5) and (2.4) we infer that F is a l o c a l l y c o n v e x

space. Choose any c o n t i n u o u s s e m i n o r m s on F. By (7.2), we can find
a vector space E, two p r e - H i l b e r t s e m i n o r m s p,q on E with

(i) d k ( B p , B q) ~ ~k -I (k = 1,2 .... ),

a subgroup K of E and a ,bomomorphism ~ : K ~ F, such that

%(K n Bp) ~ No(F) and ~(K N Bq) c B s. In v i e w of (2.13), we m a y as-
sume that span K = E. The formula

n n
Z tkU k = Z t k C s # ( U k) (t k ~ R, u k E K)
k=l k=l

defines a linear o p e r a t o r ~ : E ~ F s. To prove this, we o n l y have to

 85

show that ¢ is w e l l - d e f i n e d , i.e. that ~ t k C s ¢ ( U k) = 0 whenever

n
tkU k = 0. So, choose any U l , . . . , u n ~ K. Set M = {Uk}k= 1 and

H = K N M. The formula ~(u) = $s¢(U) defines a homomorphism n :H

F s. We h a v e n ( H n Bp) = C s ¢ ( H N Bp) c Cs(Bs) = B(Fs) , which implies
that ~ is c o n t i n u o u s in the u s u a l topology on the f i n i t e d i m e n s i o n a l
space M. Let ~ : H + F be the c o n t i n u o u s extension of ~. A c c o r d -
s
ing to (3.1), we m a y w r i t e H = Ho ~ ~" where Ho is a closed l i n e a r
subspace of M and H" is a f r e e d i s c r e t e group. We can therefore
extend ~ to a c o n t i n u o u s homomorphism ~ : M ~ F s. Being additive
and c o n t i n u o u s , ~ is a l i n e a r m a p p i n g . Thus

n n n
5q t k ¢ s ¢ ( U k ) = Z t k O ( U k) = a( Z tkU k) = 0(0) = 0,
k=l k=l k=l

which proves that ¢ is w e l l - d e f i n e d . Set X = cony (K n Bp) and Y =

cony (K N Bq). From (i) w e g e t

¢o
,1- o0
2
k=l k=l = ~-~ < 4"

Hence, in v i r t u e of (3.20),

(2) dk(X,Y) & 2dk(Bp,Bq) (k = 1 , 2 , . . . ) .

It is not d i f f i c u l t to see t h a t ~sl(¢(Y)) c B s and ~s1(¢(X)) D ¢(KABI0) '

whence ~sI(C(X)) ~ No(F). Finally, from (i), (2) a n d (2.8) w e obtain

dk(~sl(C(X)),~sl(C(Y))) =< d k ( X , Y ) < k -I (k = 1,2 .... ) .

This proves that F is a n u c l e a r space because s was an a r b i t r a r y

continuous seminorm on F. •

(8.10) I ~ % ¢ S . Let E,F be t w o n o r m e d spaces a n d let ~(E,F) be

the f a m i l y of all b o u n d e d linear operators ~ : E ~ F which satisfy
the f o l l o w i n g condition: if K is a closed subgroup of F, then
~-I(K) is a w e a k l y closed subgroup of E (or, w h i c h is the same, if
K is a s u b g r o u p of E, then ~ maps the w e a k c l o s u r e of K in E
into the c l o s u r e of ~(K) in F). Proposition (8.4) says t h a t if E,

F are u n i t a r y and Z kdk(~) < ~, then ~ ~ ~(E,F). On the other

k=l
 86

hand, from the proof of Lemma 6 in [4] it follows that if E,F are
a r b i t r a r y n o r m e d spaces and ¢ E G(E,F), then kVk(~) remains b o u n d e d
as k ~ ~.

Let ~ : E" ~ E and X : F ~ F" be a r b i t r a r y b o u n d e d linear op-

erators. It is clear that if ~ ~ G(E,F), then ¢~ ~ (E',F) and
×~ E ~(E,F-). However, it is not known w h e t h e r the sum of two opera-
tors from ~(E,F) belongs to G(E,F).
The s i t u a t i o n d e s c r i b e d above is typical in the f o l l o w i n g sense.
A t h e o r e m on a d d i t i v e subgroups of n u c l e a r spaces leads to a certain
s t a t e m e n t on linear o p e r a t o r s b e t w e e n u n i t a r y spaces which, in turn,
is r e d u c e d to a s t a t e m e n t on lattices and e l l i p s o i d s in Rn (cf. the
remarks at the end of section 3). This gives rise to a c e r t a i n class A
of o p e r a t o r s acting in B a n a c h spaces; A is u s u a l l y c l o s e d with re-
spect to c o m p o s i t i o n w i t h b o u n d e d o p e r a t o r s but p r o b a b l y not with re-
spect to a d d i t i o n of operators. -If dk(#) are small enough (say,
dk(~) -- k -I or --k-2), then ~ 6 A. On the other hand, if # E A,
then Vk(~) c a n n o t be too large.

Most i m p o r t a n t e x a m p l e s of such classes of o p e r a t o r s are connect-

ed w i t h (8.4) (the class 8 c o n s i d e r e d above), w i t h (8.1) and with
(12.2). V e r y i n t e r e s t i n g (and p r o b a b l y difficult) problems of this
appear in c o n n e c t i o n w i t h the c o n s i d e r a t i o n s of section 6. See also
(10.18).

(8.11) mm~E. The m a t e r i a l of this section is m o s t l y new. For sub-

groups and q u o t i e n t groups of nuclear spaces, the results of this sec-
tion were p r o v e d in [7] and [8]. P r o p o s i t i o n (8.1) is a s t r e n g t h e n i n g
of L e m m a 1.5 of [8].

9. N u c l e a r v e c t o r groups

(9*l)/21Elr'~lnr'F~a~. Let F be a v e c t o r space and ~ a t o p o l o g y on

F such that FT is an a d d i t i v e t o p o l o g i c a l group. We say that FT is
a locally c o n v e x v e c t o r g r o u p if it is s e p a r a t e d and has a base at zero
c o n s i s i t i n g of symmetric, c o n v e x sets.

This n o t i o n was i n t r o d u c e d by D.A. R a i k o v in [78], p. 301. Ob-

that, for e a c h I ~ R, the m a p p i n g u ~ ~u of F into itself is
T
continuous. O b s e r v e also that FT is a t o p o l o g i c a l vector space if
and o n l y if it has a base at zero c o n s i s t i n g of a b s o r b i n g sets; then
it is a locally c o n v e x space.
 87

(9.2JaEFImITIGm. A locally c o n v e x vector g r o u p F is called a

nuclear v e c t o r group if to e a c h symmetric, convex U e No(F) there
corresponds a symmetric, convex W ~ No(F) such that

dk(W,U) ~ k -I (k = 1,2 .... ).

[9.3JPRumDsrFx¢~. Let F be a nuclear vector group. Take any m =

1,2,... and c > 0. To e a c h radial U e No(F) there c o r r e s p o n d s some
symmetric, convex W e No(F) with

dk(W,U) ~ ck -m (k = 1,2 .... ).

The proof is similar to that of (2.17).

(9.4) PRUPUSlqrlUL E v e r y n u c l e a r vector g r o u p is a nuclear group.

This is an i m m e d i a t e c o n s e q u e n c e of (7.1) and (9.3) (see the proof

of (7.4)).

fg.5) ReUmOSlTI~. The c o m p l e t i o n of a n u c l e a r vector g r o u p is a

nuclear vector group.

Pmm~. Let F be a n u c l e a r vector group. We i n t r o d u c e m u l t i p l i -

cation by real numbers in F in the f o l l o w i n g way. Let f ~ F and
t ~ R. T h e r e is a g e n e r a l i z e d s e q u e n c e (f~) in F, converging to
f. Then (tf c) is a C a u c h y s e q u e n c e in F. We set tf = lim tf . It
is e a s y to see that F w i t h the m u l t i p l i c a t i o n thus d e f i n e d is a vec-
tor space.
Let {Ui}i~ I be a base at zero in F, consisting of symmetric,
convex sets. T h e n {Ui)i~ I is a base at zero in F. It is e a s y to see
that the sets Ui are s y m m e t r i c and convex. Choose any i ~ I. There
is some j 6 I such that

d k ( U j , U i) < k -I (k = 1,2,...).

For any fixed k, we can find a linear subspace L of F with

dim L < k and Uj c k-iUi + L. For each E > O, we have Uj c vj +
E~j, whence

Uj c k-Iu i + EUj + L c k - i U i + L = (k -I + E)U i + L.

Therefore
 88

d k ( U j , U i) & k -I (k = 1,2 .... ),

which proves that F is a n u c l e a r vector group. •

Let G be a topological group. By Wo(G) we shall denote the

weight of G at zero, i.e. the least cardinal number m such that G
has a base at zero w i t h cardinality m.

t9.6) l"RJfl~. Let G be a nuclear group. Then there exist a nu-

clear vector group F with Wo(F) = Wo(G) , a subgroup H of F and
a closed subgroup Q of H, such that G is topologically isomorphic
to H/Q.

Proof. Choose a base B at zero in G with card B = Wo(G). Due

to (8.5), we m a y assume that B consists of quasi-convex sets. Let
A

RG be the v e c t o r space of all real-valued functions on G-. For each

U ~ B, let

g~U EU o ~ ( X ) = x(g)}

and let Y U = c o n v X U. If V ~ B and V + V c U, then it is clear

that Y V + Y V c YU" It is a l s o clear that Y-U = -Yu because both U
and Y are symmetric sets. Thus, according to (1.12), there is a u n i q u e

topology ~ on RG such that F : = (R G ,T) is a t o p o l o g i c a l group

for w h i c h {Yu}u~B is a b a s e at zero.

Consider the h o m o m o r p h i s m o : G ~ TG given b y the formula

c(g)(x) = x(g) (g ~ G, X ~ G-).

^

It f o l l o w s from (8.5) that G separates points of G. Therefore

is i n j e c t i v e . Let PG : RG- ~ TG~ be the canonical projection given

by the formula

PG(~)(X) = P(~(X)) (~ E RG X ~ G-)

Set H pGI(~(G)) and Q = H fl pGl(0). We shall prove that H/G with

the topology induced from F is t o p o l o g i c a l l y isomorphic to G. Con-

sider the canonical diagram
 89

> o(G) < H/Q

Here ~ and ~ are b o t h algebraical isomorphisms. We shall prove

-i
that a ~ is a t o p o l o g i c a l isomorphism.

Choose any U E 8. We are going to s h o w that

(i) PG(H O YU ) = o(U).

We b e g i n with the inclusion

(2) PG(H n Yu ) o o(U).

So, take any ~ E o(U). We h a v e ~ = o(g) for some g e U. We m a y

treat ~ as an e l e m e n t of RG . Then ~(X) = a(g)(x) = x(g) for e a c h

X ~ G , which implies that ~ e X u. Next, we h a v e pG(~) = ~ = a(g) E

a(G), whence ~ ~ H. Thus ~ E H D YU and ~ = pG(~ ) e PG(H A YU ),
which proves (2).
To p r o v e the o p p o s i t e inclusion, choose any ~ ~ H D YU" Since

YU = c o n y XU, and XU is s y m m e t r i c , we m a y w r i t e ~ = tl~ 1 + ... + tn~ n

for some ~l,...,~n ~ XU and some tl,...,t n > 0 with tl+ ...+ t n = i.

According to the d e f i n i t i o n of XU, for e a c h k = l,...,n, there is

1
some gk E U with ~k(X) = x(g k) for all X • U °. Hence l~k(X) I £
for all x ~ U° and k = l,...,n. From this we get

(3) I~(X)I ~ tll~l(X)I + ..- + tnl~n(X)] £ ¼ (X • U°) •

Now, since ~ e H, we h a v e QG(~) = d(g) for some g ~ G. By (3),

for e a c h X ~ U °, we h a v e

1
J×(g)l = lo(g)(×)J = IPG(~)(×)} = IP(¢(X))I =< } & ( × ) l --<-~-
This implies that g ~ U because we h a v e assumed U to be q u a s i - c o n -
vex. Hence pG(~) = a(g) ~ a(U), which yields PG(H N YU) c a(U) since
~ H N YU was arbitrary. From this and (2) we o b t a i n (i).
 90

Now, (I) may be w r i t t e n as o-l~(H A YU ) = U. Since this holds

for all U E B, we get {o-l~(H A YU)}U~B = ~. But {~(H A YU)}UE~
-I
is a base at zero in H/Q, and B is such a base in G. Hence ~
is a topological isomorphism.
To complete the proof, it remalns to show that F is a nuclear
vector group. So, take any U ~ No(F). ~here is some V ~ 8 with
YV c U. By (7.2), there is a vector space E, two pre-Hilbert semi-
norms p,q on E with

dk(Bp,Bq) & 10-2k -4 (k = 1,2,...),

a subgroup K of E and a homomorphism ~ : K ~ G, such that

O(K A Bp) ~ No(G) and ~(K A Bp) c ½V. Due to (2.15), we can find a
pre-Hilbert seminorm r on E with p ~ r ~ q,

(4) dk(Br,Bq) < 3 ~ k -3 (k = 1,2,...),

(5) dk(Bp,B r) < ~k -I (k = 1,2,...).

Now, we shall construct a linear operator ~ : E ~ RG Take any

X E V °. Then, by (1.2),

Ix(½v)l < ½1x(v) l < 1 i 1

I
so Ix,(K N Bp) I < 9" In virtue of (8.1), there is some f ~E # with
X
Pfx = X~ on K and

sup {Ifxlu) I : u ~ B r) S 5 ~ kdk(Br,Bq).

k=l
Hence, by (4), we get

16) sup {[fx(U) l: u 6 Br} < ~ k-2 < 4"

k=l

For each u ~ E, define

) if X ~ V°,

(¢u)(x)
=Ifx(u if X ~ V°.
 91

Let P : RG ~ RG be the p r o j e c t i o n g i v e n by the f o r m u l a

X E VO ,
(P~)(X) =I~0 (X) if

if X C V °-

We shall p r o v e that

(7) P(H) n ¢(B r) c P(Xv).

So, take a n y K ~ P(H) n ~(Br). Then K = P~ for some ~ ~ H and

= #u for some u ~ B r. Since ~ E H, there is some g E G with
pG(~) = o(g). For e a c h X e V °, we have

(8) x(g) = ~(g)(x) = pG(~)(x) = p(~(x)) = ~(P~(x))=p(5(x))

= p(~u(x)) = pfx(U).

Since u ~ Br, from (6) it f o l l o w s that

1 (X ~ V°).
Ix(g) l = IPfx(U)I --< Ifx(U) < ~"

Thus ~ ~ XV and K = P~ ~ P(Xv) , which proves (7).

Now, s i n c e ~(K n Sp) ~ No(G), there is some W ~ 8 with

w c #(K n Bp). We s h a l l prove that

(9) P(Xw) c P(H) N ~(Bp).

Choose any ~ ~ P(Xw). There is some ~ ~ Xw with K = P~. Next,

there is some g E W with ~(X) = x(g) for all X ~ W°- Finally, we
have g = ~(u) for some u e K fl Bp. Now, take any X e V°. Since
1
Bp c Br, we have u E Br, and from (6) we o b t a i n Ifx(U) I < ~. Then

K(X) = ~(X) = x(g) = X~(U) = pfx(U) = fx(U) = (@u)(x)

(observe that V° c W° because W c V). For X ¢ V°, we h a v e ~(X) =

(~u)(x) = 0. Hence ~ = ~u e ~(Bp), which implies that

(i0) P(Xw) c ~(Bp).

 92

Let D ~ RG be given by the formula n(X) = x(g) for X E G-.

Then PG(D)(X) = x(g) = o(g)(×) for every X , i.e. pG(q) = o(g).
Thus q E H. Moreover, P~ = ~, which implies that P(Xw) c P(H).
From this and (i0) we obtain (9).
Set M = ¢(E). The Minkowski functionals of ¢(Bp) and ¢(B r)
are pre-Hilbert seminorms on M (see (2.11)). From (2.8) (a) it fol-
lows that

(ii) dk(~(Bp),~(Br)) ~ dk(Bp,B r) (k = 1,2,...).

hence, by (5),

1 Zk-2<~.
X d (¢(Bp),¢(Br)) & ~ k=l
k=l

Now, (3.20) implies that

(12) dk(COnV (P(H) N ¢(Bp), conv (P(H) D ¢(Br)))

2dk(¢(Bp),¢(Br)) (k = 1,2 .... ).

We have YW c p-l(P(yw) ) and YV = P-I(P(Yv))" Finally, by (2.8) (b),

(7), (9), (ii), (12) and (5), we obtain

dk(Yw,Y V) ~ dk(P-I(P(Yw)),P-I(P(Yv)) = dk(P(Yw),P(Yv))

= dk(P(conv Xw),P(conv XV)) = dk(COnV P(Xw),Conv P(Xv))

£ dk(COnV (P(H) n ¢(Bp)),conv (P(H) A ¢(Br))) & ~ k - l < k -I

for every k. Thus F is a nuclear vector group.

Since {Wv}v~ B is a base at zero for F, we have

Wo(F) £ card {Yv}vE B £ card B = Wo(G).

On the other hand, Wo(G) = Wo(H/Q) ~ Wo(H) & Wo(F). •

~9.7) PROPOSITI~. Let G be a metrizable nuclear group. Then there

exist a metrizable and complete nuclear vector group F, a closed sub-
group H of F and a closed subgroup Q of H, such that the com-
pletion of G is topologically isomorpic to H/Q.
 g3

~£. By (9.6), there exist a m e t r i z a b l e nuclear vector group

F', a subgroup H" of F" and a closed s u b g r o u p Q" of H', such
that G is t o p o l o g i c a l l y i s o m o r p h i c to H'/Q'. Let F be the com-
p l e t i o n of F" and let H and Q be the c l o s u r e s in F of H" and
Q', respectively. From (I.i0) it now follows that the c o m p l e t i o n of
G is t o p o l o g i c a l l y i s o m o r p h i c to H/Q. Finally, F is a nuclear
vector group, due to (9.5). •

~9.sJ ~ . The c o m p l e t i o n of a m e t r i z a b l e nuclear group is a

nuclear group.

This follows i m m e d i a t e l y from (9.7), (9.4) and (7.5).

~9.9) ~m~m~. It follows from (9.6), (9.4) and (7.5) that we might
define nuclear groups as H a u s d o r f f q u o t i e n t s of subgroups of nuclear
vector groups. Such a d e f i n i t i o n is m u c h simpler and allows us to omit
the long and c o m p l i c a t e d proof of (9.6). However, it has also some dis-
advantages. Firstly, it is not "intrinsic". Secondly, a p p l y i n g this de-
finition, we w o u l d e n c o u n t e r d i f f i c u l t i e s in s e c t i o n 7, with proving
that groups l o c a l l y i s o m o r p h i c to nuclear groups are nuclear. Also,
which is m o r e important, we w o u l d have t r o u b l e s in section 16 with show-
ing that the dual group of a m e t r i z a b l e nuclear group is nuclear. In
fact, we w o u l d have to repeat there the a r g u m e n t from the proof of
(9.6).

~9.10) ~ . The m a t e r i a l of this section is new.

10. The L e v y - S t e i n i t z theorem

co

Let Z gi be a c o n v e r g e n t series in a H a u s d o r f f abelian group

i=l ~o

G. Its sum will be d e n o t e d s i m p l y by ~ gi' that is,

i=l
j
Z gi = lira ~q gi"
i=l j+= i=l

To s i m p l i f y the notation, we shall often write Z gi instead of Z gi

i=l
to denote both the series and its sum; this should not lead to misun-
derstandings.
The set of sums of the series Zgi, d e n o t e d by ~(Tgi;G) is
d e f i n e d in the f o l l o w i n g way: a point s ~ G belongs to ~(Zgi;G)
 94

if there is a p e r m u t a t i o n ~ of indices such that the series Zg~(i)

c o n v e r g e s to s.

[Io.12 J~mQRK. It m a y h a p p e n that S (~gi;G) # S(~gi;G). For ex-

ample, d e n o t i n g the g r o u p of r a t i o n a l numbers by Q, we have

S( ~ (-l)i ;Q)
= Q and S ( X (-l)i ;R) = R.
i=l i i=l i

If the m e a n i n g of G is clear, we shall w r i t e s i m p l y S ( ~ g i ) instead

of S(Zgi;G).

The L e v y - S t e i n i t z t h e o r e m says that the set of sums of a conver-

gent series in Rn is an affine subspace (cf. also (I0.Ii)). This
t h e o r e m can be g e n e r a l i z e d to c e r t a i n i n f i n i t e d i m e n s i o n a l spaces; see
the remarks f o l l o w i n g (10.2). For series in B a n a c h spaces, see (10.12)
and (10.13).

On p. 61 of his book [89], S. U l a m p o s e d the following problem:

in c o n n e c t i o n w i t h the L e v y - S t e i n i t z theorem, he and G a r r e t t B i r k h o f f
had n o t i c e d that the f o l l o w i n g a s s e r t i o n is true:

(*) the set of sums of a c o n v e r g e n t series in a c o m p a c t g r o u p G

is a coset m o d u l o a c e r t a i n s u b g r o u p of G;

does a similar result h o l d for m o r e general, non-compact topological

groups? This s e c t i o n is an attempt to answer the question; the m a i n
result is (10.3).

tlO.2J m01~. U l a m and B i r k h o f f did not leave, as it seems, any in-

d i c a t i o n s and one doe~ ,~t ~ e r y w e l l k n o w how they had shown (*). For com-
pact m e t r i z a b l e groups, (*) follows from (10.3). In this particular case,
the proof of (10.3) can be m u c h simplified, for it suffices to apply
(10.9) and to m a k e use of the fact that the set of sums of a c o n v e r g e n t
series in the c o u n t a b l e p r o d u c t of real lines is a linear manifold
(see below). W i t h o u t the m e t r i z a b i l i t y assumption, (*) is false; see
(i0.14).

Let E be a real t o p o l o g i c a l vector space. By a linear manifold

in E we m e a n a set of the form u + M where u ~ E and M is a
linear s u b s p a c e of E. et us c o n s i d e r the f o l l o w i n g assertion:

(**) the set of sums of e v e r y c o n v e r g e n t series in E is a c l o s e d

linear manifold.
 g5

If E is the locally c o n v e x d i r e c t sum of an a r b i t r a r y number of real

lines, the v a l i d i t y of (**)follows i m m e d i a t e l y from the L e v y - S t e i n i t z
t h e o r e m for Rn b e c a u s e e v e r y c o n v e r g e n t series in E lies in a fi-
nite d i m e n s i o n a l subspace.
Troyanski [88] p r o v e d that (**) is true if E is the countable
p r o d u c t of real lines. His result had been o b t a i n e d earlier by Wald
[I00] and was r e d i s c o v e r e d later by K a t z n e l s o n and M c G e h e e [51]. See
also H a l p e r i n [34]. The v a l i d i t y of (**) for some other nuclear spaces
can be d e d u c e d from the results of B a r a n y [16], as w e l l as from those
of Beck [19] and P e c h e r s k i i [74]; see (10.18). By modifying Barany's
proof, the author showed in [i0] that (**) is s a t i s f i e d for e v e r y met-
rizable nuclear space E; cf. also (10.8), (i0.Ii) and (10.14).
The aim of this section is to prove the f o l l o w i n g fact:

~I@.3J TR~URIN. Let ~ gi be a c o n v e r g e n t series in a metrizable

nuclear g r o u p G. Then P : = S( ~ g i ) Zgi is s u b g r o u p of G (not
n e c e s s a r i l y closed). If G is complete, P is a continuous homo-
m o r p h i c image of a n u c l e a r F r e c h e t space.

The a s s u m p t i o n of m e t r i z a b i l i t y is essential; see (10.14).

Let us b e g i n w i t h some definitions. Let ~ gi be a convergent

series in a H a u s d o r f f a b e l i a n g r o u p G. By T(~gi;G) we shall de-
note the set d e f i n e d in the f o l l o w i n g way: a point s E G b e l o n g s to
6(Zgi;G) if there e x i s t a p e r m u t a t i o n ~ of i n d i c e s and a sequence

Jl < J2 < "''' such that

Jn
s = lim ~ g~(i)"
n~ i=l

For each m = 1,2,..., let Am be the c l o s u r e in G of the set of a ~

points of the form ~ gi where I is a finite subset of {m,m+l,...}.
i~I
We d e f i n e A(Zgi;G) = n A m-
m=l

R e m a r k s a n a l o g o u s to (I0.i) are applied to C(~gi;G) and A(Zgi;G) ,

as well. W h e n there is no r i s k of m i s u n d e r s t a n d i n g , we shall briefly
write C( Z gi ) and A( ~ g i ) .

~I0.4J R~SPUSlTI~. Let 5qg i be a c o n v e r g e n t series in a H a u s d o r f f

abelian group G. Then A( ~ g i ) is a closed s u b g r o u p G. Moreover,
 96

A(>q.g i) + ~lg i = ~(7~.gi).

This is a w e l l - k n o w n fact; see e.g. [23], Ch. III, §5, Exercise 3.

For the first time, it a p p e a r s in Wald's paper [I01]. For series in

R n, it w a s known to Steinitz [87].

(zo.5) rJm~a. Let D c E c F be three o-symmetric n-dimensional

closed ellipsoids in Rn with

n
~q d (D,E) < 1 and 71 d (E,F) < ~.
k=l k=l

m
Let Ul,...,u m ~ D and a e E be such that a + Z u~ E E. Then
i=l
there exists a permutation o of {i, .... m} such that

J
a + 7 uo(i) ~ F (j = 1 ..... m).
i=l

This is L e m m a 6 of [i0].

(~0.6) r~A. If u. is a convergent series in a metrizable nu-

1

clear vector group, then S( Z u i ) = Z(Zui).

We only have to show that C( Z u i) c S(~ui), the o p p o s i t e

inclusion being trivial. So, take any s • C(Zui). There are a p e r m u -

tation z of indices and a sequence J l < J2 < "''' such that

Jn
s = lim Z u (i)-
n+= i=l

Let us denote our nuclear vector group by F. It follows easily from

(9.3) and (2.14) that we can find a fundamental sequence U 1 m U 2 m ...

of c o n v e x neighbourhoods of zero in F, such that, for every n, the

Minkowski functional Pn of Un is a p r e - H i l b e r t seminorm on Mn : =
span Un, and

(i) Un = {u • M n : Pn(U) < I},

co

(2) ~q. d (Un+l,U n N Mn+l) < ~.

k=l
 97

Increasing the i n d i c e s in' if n e e d be, we m a y a s s u m e that, for e v e r y

n,
Jn
(3) s - Z u(i)
i=l Un+2'

(4) u~(i) g Un+ 2 for all i > Jn"

Let us n o w fix an a r b i t r a r y index n. Replacing n by n + 1 in

(2), we get

(5) z 2 'Un+l n M n + 2 I =< ¼ < i

k=l

Similarly, replacing n by n+ 1 in (3), w e get

3n+l
(6) s - Z u(i)
i=l Un+2

because U n + 3 c Un+2. Let L be the l i n e a r subspace of F spanned

Jn
over the v e c t o r s - ~ u (i) and the v e c t o r s ui for i = in+l,...,
i=l
Jn+l" Set 1 = d i m L. From (3) and (4 we get L c M n + 2. Therefore,

replacing n by n + 1 and n + 2 in (i), we see t h a t U n N L,Un+IAL

and Un+ 2 N L are o-symmetric closed 1-dimensional ellipsoids in L.

In v i r t u e of (2.13), from (5) a n d (2) we o b t a i n

1 1
7 d ~ ( U n + 2 N L , U n + 1 N L) < I, 5q. d 2 1
k=l k=l k ( U n + l q L'Un n L) < 4"

Thus, by (4), (3) and (6), from (10.5) we i n f e r that there exists a
permutation ~n of the set {~(Jn+l),...,~(Jn+l)}, such that

Jn j
(7) s - Z u (i) Z u E U (Jn + 1 < j < Jn+l )
i=l i = J n + l On~(i) n = = "

Let p be the p e r m u t a t i o n of p o s i t i v e integers given by p(i) =

~n~(i) when Jn + 1 ~ 1 ~ Jn+l and b y p(i) = i when i < 31" Since
(7) h o l d s for e v e r y n, we m a y w r i t e
 98

J
S - Z g U (J > Jn; n = 1 2,...)
i= 1 U p ( i ) n ' "

This means that the series Up(i) converges to s. Thus

s g S( Z ui). •

Let F be a n u c l e a r vector group. The set n{span U : Ue No(F)}

is a c l o s e d linear subspace of F; we shall denote it b y F o. Notice

that F° is a n u c l e a r space.

glo.TJ ~mmma. Let Z ui be a convergent series in a n u c l e a r vector

group F. Then A( Z u i) is a c l o s e d linear subspace of F contained
in F
o

Pmmmf. For each m = 1,2, .... let Am be the closure in F of

t h e s e t of a l l p o i n t s of t h e form ~ ui where I is a finite subset
ieI
of {m,m+l,...}. Denote A = A( Z ui); then A = N A . Since
m= 1 m

u i + 0, to e a c h U g No(F) there corresponds an index m such that

ui E U for = m;
i > hence A m c s p a n U = s p a n U. This implies that

A c Fo. Since A is, by (10.4), a closed subgroup of F, we only

have to p r o v e that A is r a d i a l .
Choose arbitrary s g A and % g (0,i). It is to be shown that
%s ~ A. So, take any m = 1,2 .... and any U ~ No(F). Due to (9.3)

and (2.14), we can find some W ~ No(F) and two pre-Hilbert seminorms
p,q on s p a n W, such that W N B n B Q U and
P q
a0

(1) ~ d2(Bp,Bq) < i.

k=l

Since Bp ~ No(F) and u i + 0, there is a n i n d e x n ~ m such that

(2) u i ~ Bp for all i a n.

Next, since U ~ No(F) and s ~ An, there is a finite subset I

of { n , n + l .... } such that

(3) s ~ 7 u i + U.
i~l

From (i), (2) a n d (7.13) it f o l l o w s that there is a s u b s e t J of I

 99

such that

% Z u i E B q + Z u..
iEI i~J 1

F r o m this and (3) w e d e r i v e

%s E %U + % ~ u i c (i + %) + Z u i.
iEI iEJ

Since U E No(F ) was arbitrary, it f o l l o w s that % s ~ An . But n ~ m,

therefore A n c Am and, consequently, % s ~ Am . This proves that

%s E A because m was arbitrary. •

(Io.8; ~ . Let Zgi be a convergent series in a m e t r i z a b l e

nuclear vector group F. Then S ( ~gi ) - Zgi is a c l o s e d linear sub-

space of F contained in F .
o

This is a d i r e c t consequence of (10.4), (10.6) and (10.7). T h e as-

sumption of m e t r i z a b i l i t y is e s s e n t i a l ; see (10.14).

~ o . 9 ) ;ammx. Let X be a set of indices and let Rx denote the

space of a l l real-valued functions on X, endowed with the topology of
pointwise convergence. Similarly, let TX denote t h e g r o u p of all f u n c -
tions g : X + T endowed with the topology of pointwise convergence and

let ~ : Rx + Tx be the natural projection given by

~(u)(x) = p(u(x)) (u ~ RX; x s X).

For each convergent series ~u i in R x, one has

S(~(ui);TX) = ~(S(Xui;RX)).

Pmm~f. The inclusion ~(S(Zui)) c S(~$(ui)) is trivial. To prove

the opposite one, take any g E S(Z$(ui)). There is a p e r m u t a t i o n
of p o s i t i v e integers such that

J
(i) ~ ~(U~(i)) ~ g as j ~ ~.
i=l

Fix an arbitrary x ~ X. From (i) w e g e t

J
Z ~(u )(x) ~ g(x) as j ~ ~.
i~ 1 (i)
 100

which c a n be w r i t t e n as

J
(2) p( ~q U (X)) ~ g(x) as j ~ co.
i= 1 (i)

Since (u i) is a n u l l sequence in R X, we have ui(x) ~ 0 as i ~ ~.

From this and (2) it f o l l o w s that there is a n i n t e g e r k(x) such that

J
(3) ~q. U (i)(X) ~ g(x) + k(x) as j ~ ~.
i=l

Since (3) h o l d s for e a c h x e X, the series Zu (i) converges to t h e

function g + k in R x. Thus g + k ~ S(Zui). We have

¢(g + k) = ¢(g) + ~(k) = g + 0 = g,

which proves that g ~ ~(S(~ui)). •

The proof of (10.3), given below, is, in fact, a modification of

the proof of (10.9). The differences may bear a technical character on-
ly.

Proof of (10.3). It is c l e a r that

~(Zgi;G) = G n S(Zgi;G),

therefore we may assume that G is c o m p l e t e . By (9.7), we may write

G = H/K where H is a c l o s e d subgroup of some metrizable and complete
nuclear vector group F and K is a c l o s e d subgroup of S. Since
H/K may be identified with a closed subgroup of F/K, we may simply
assume that G = F/K. Let # : F ~ F/K be the natural projection.
We may restrict ourselves to t h e c a s e w h e n K d o e s not contain any
lines. Indeed, let L be the maximal linear subspace of L contained
in K (i.e. the union of a l l linear subspaces contained in K). W e h a v e
c K because K is c l o s e d . Since F is a locally c o n v e x vector group)
it f o l l o w s that ~ is a v e c t o r subspace of F, the proof being the
same as for topological vector spaces. Thus ~ = L. Let ~ : F + F/L
be the canonical projection. The group F/L has a natural structure of
a nuclear vector group; it is m e t r i z a b l e and complete. Evidently, ¢(K)
is a c l o s e d subgroup of F/L containing no lines. It r e m a i n s to be ob-
served that F/K is c a n o n i c a l l y topologically isomorphic to (F/L)/~(K).
Let U1 o U2 D ... be a fundamental sequence of n e i g h b o u r h o o d s of
zero in F, consisting of symmetric convex sets. In view of (9.3) and
(2.14), we may assume that, for every n, the Minkowski functional
 101

Pn of Un is a pre-Hilbert seminorm on Mn : = span Un with

B ( p n) = U n. We may also assume that

(I) Z d i ( U n + l , U n N M n + I) & 1
k=l

for e v e r y n. Denote L n = span (K N U n) for n = 1,2, . . . .

For every n, let £ be the family of all linear functionals f
n
on Ln+ 1 satisfying the c o n d i t i o n s

(2) f(K N Un+l) c Z,

(3) sup {f(u) : u ~ U n A Ln+l} < ~.

We shall prove that F is at m o s t c o u n t a b l e . Denote E = Ln+ 1 and

n
let r,s be the r e s t r i c t i o n s to Ln+ 1 of Pn and Pn+l' respec-
tively. We h a v e B r = U n n Ln+ 1 and B s = Un+ 1 N L n + I. From (i) and
(2.13) we obtain

oo

(4) 57 2
k=l d k ( B s ' B r ) --< i.

As usual, we h a v e the canonical diagram

id
E > E

Asr t
E > E
s r

We s h a l l prove that the m a p p i n g h ~ h¢r is a b i j e c t i o n of the set

~ = {h ~ E : h~r(K N Un+l) c Z}
r

onto r .
n

If h E ~, then it f o l l o w s immediately f r o m our definitions that

h~ r ~ F n. So, take an a r b i t r a r y f ~ F n. We h a v e f(u) = 0 for all
u E ker ~r; this follows easily from (3). Consequently, there is some

h E E#r with h~ r = f. From (3) we see that h is bounded and (2)

yields hCr(K N Un+l) c Z° Thus h e ~. Finally, if h I # h2, then

 102

hl~ r # h 2 ~ r because ~r is s u r j e c t i v e .

From (4) a n d (2.12) it f o l l o w s that dk(Asr) ~ 0 as k+~, which

means that Asr is a c o m p a c t operator. Therefore Er is separable

and, consequently, E is a s e p a r a b l e Hilbert space. It is obvious
r
that ~ is an a d d i t i v e subgroup of E r. Furthermore, ~ A int B ( E r) =
{0}. Indeed, if h ~ ~ \ {0}, then h(u) # 0 for a certain u
~r(K NUn+ I) because E = span (K N Un+l). But h(u) e Z and

~r(K n Un+l) c ~r(E n u n) = B(Ur),

which implies that llhlJ ~ i. Being a discrete subgroup of a separable

space, ~ is at m o s t countable. Consequently, so is F n.
Let B be the family of all sets of the form

Un, f = U n + {u e L m + 1 : f(u) = 0}

where m,n = 1,2,... and f E F . Observe that B is at m o s t count-

m
able. It is e v i d e n t that B satisfies conditions (a) - (c) of (1.12).
So, there is a u n i q u e topology ~ on F such that F is a t o p o l o g i c -

al g r o u p for w h i c h B is a b a s e at zero. Notice that if U E B, then

cUE No(FT) for e a c h c > 0. The original topology on F is finer
than ~; we shall denote it by o. Thus we m a y w r i t e G = F /K.

We shall prove that F is a m e t r i z a b l e nuclear vector group. To

prove that FT is s e p a r a t e d , take any w ~ F \ {0}. We h a v e to find
some U ~ No(F T) with w ~ U. Suppose first that w ~ K. Since K is
a closed subgroup of the n u c l e a r vector group Fo, from (9.4) and (8.6)

it f o l l o w s that there is some × ~ F° with XI K ~ 0 and X(W) # 0.

By (1.4), there is an i n d e x n such that < !4"

Ix(Un) I ~ Next, accord-

ing to (2.2) there is some f ~ M# with pf = X M and

' n
n
1
(5) sup {If(u)I : u ~ u n} _-< 4"

We h a v e f(K N M n) c Z because XI K ~ 0. Therefore f" := f i L n + 1 ~ F n.

Since pf(w) = X(W) # 0, it f o l l o w s that c : = If(w) I # o. By (5),

for e a c h v ~ Un, we h a v e

If(w - cv) I => If(w) l - clf(v) I Z c - ¼c = ~c

0 , >3
 103

which means that w ~ c U n + f-l(0). Hence w ~ CUn, f. , either.

Now, suppose that w ~ K. Since w # 0 and K d o e s not c o n t a i n

a n y lines, there is s o m e t ~ (0,i) with tw ~ K. According to the
above, there is s o m e r a d i a l U ~ No(F Y) with tw ~ U. Hence w ~ U,
either.
Being separated, F is m e t r i z a b l e . Moreover, F is a locally
convex vector group because all sets Un, f in 8 are symmetric and
convex. Take any m , n = 1,2 .... and any f E F m. Since F is a nu-

clear vector group, t h e r e is an i n d e x 1 > n such that d k ( U l , U n) < k -I

for e v e r y k. Set N = {u E L m + 1 : f(u) = 0}. According to (2.6) (a),
for e v e r y k, we h a v e

d k ( U l , f , U n , f) = d k ( U 1 + N , U n + N) ~ dk(Ui,U n+N) + dI(N,U n+N)

d k ( U I , U n) + d I ( N , N ) < k -I

because dI(N,N) = 0. This proves that FT is a nuclear v e c t o r group.

Our next goal is to s h o w t h a t the m a p p i n g # : F T ~ Fc/K is con-

tinuous. Take an a r b i t r a r y index n. We h a v e to s h o w t h a t

(6) U n + K E No(F).

Let r and s be d e f i n e d as b e f o r e . From (4) and (3.18) we h a v e

Ln+ 1 : = span (K D U n + I) = s p a n (K n B s) c ½B r + gp (K n B s)

c ½ U n + K.

Hence ½ U n + L n + 1 c U n + K, which proves (6) because, evidently,

U n + Ln+ 1 ~ No(FT).

The c o m p l e t i o n F
of F is a n u c l e a r v e c t o r group dueGo (9.5).
T T
Let ~ : F ~ G be the c a n o n i c a l e x t e n s i o n of ~.
Y
Without loss of g e n e r a l i t y we m a y a s s u m e that Z gi = 0; then

P = S(Zgi;G). Since G = F /K, we c a n f i n d a n u l l sequence (si)i= 1

J
in F with ~(sj) = Z gi for j = 1,2, . . . . Set uI = sI and
i=l
u i = s i - si_ 1 for i = 2,3, . . . . Then
J
(7) i=iZ U i = sj j~ > 0 in F
 104

and ¢(u i) = gi for i = 1,2, . . . . From (10.8) it f o l l o w s that Q :=

$(Zui;F T) is a c l o s e d linear subspace of the n u c l e a r Frechet space

(mT)o-
It r e m a i n s to s h o w that ~(Q) = p. The inclusion ~(Q) c p is
trivial. To p r o v e the o p p o s i t e one, choose any a ~ P. There is a p e r -
mutation ~ of p o s i t i v e integers such that the series ~g~(i) con-
verges to a. Choose some w • F with #(w) = a. We have

9 J
~(W - i=l~ U (i)) = a - i=iZ g~(i) j~= > 0.

Consequently, there is a s e q u e n c e (zj)j= 1 in K such that

J
zj + Z . > w in F .
i=l u~(i) 3~ o

Deflne v I = zI and v i = z i - zi_ 1 for i = 2,3, . . . . Then vi e K

for e v e r y i, and

J
(8) ~ [U (i) + Vi] j+~ > W in F
i=l

From (7) and (8) it f o l l o w s that (ui)i= 1 and (u (i) + v i ) i = 1 are

null sequences in F o. Hence (vi)i= 1 is a n u l l sequence in Fo, too.

We shall prove that the series Zv i satisfies Cauchy's criterion

in F t. Take any m,n = 1,2 .... and any f • F m. There is an i n d e x

31 such that v i • Um+ 1 for i > Jl; then v i ~ Lm+ 1 because v i ~ K.

Consequently, we h a v e f(vi) e Z for i > Jl" Since f E Fm and

vi ~ 0 in Fo, replacing n by m in (3), w e see t h a t f(v i) ~ 0.

Hence there is an i n d e x J2 > Jl such that f(v i) = 0 for i > J2"

J
This means that Z v i • Un, f for e a c h J > J2"
i=j 2

Since FT is c o m p l e t e , the series Zv i converges in F to a

certain point y. Let K be the c l o s u r e of K in FT; then y e
because all vi were in K. Consequently,

$(y) • $(K) c ~(K) = $(K) : {0} = {0}.

From (8) we infer that the series Zu (i) converges to w- y in F t.

Thus w - y • Q. Finally,
 105

~(w - y) = ~(w) - ~(y) = #(w) - 0 = a,

which proves that p c ~(Q). •

(Io.Io) ~ . Let ~ : R2 ~ T2 be the c a n o n i c a l projection. Take

some u = (x,y) ~ R2 with y/x irrational and denote L = span {u}.

Next, set u.I = (-l)i

i u and gi = ~(ui) for e v e r y i. It is not h a r d

to see that S (Zui;R2) = L and P : = S (Zgi;T2) = @(L). Thus P is

a dense, but non-closed subgroup of T 2. Notice that ~(~ gi ) =

A( Z gi ) = T 2.

(1o.ll) R ~ s . Let ui be a c o n v e r g e n t series in a real topo-

logical vector space E. Let us d e n o t e

F ( Z u i) = {f ~ E* : Z ]f(ui) ] < ~}.

i=l

Then the set

F0(Zu i) = {u ~ E : f(u) = 0 for all f E F( ~ u i ) }

is a c l o s e d linear subspace of E.

The Levy-Steinitz theorem is o f t e n formulated in the following,

somewhat stronger version: for e a c h convergent series Zu i in R n,
one has

(i) S( Z u i) = Zu i + F0(Zui).

This result was obtained by E. Steinitz [87]; it is sometimes called

the S t e i n i t z theorem. However, in the literature there is no conse-
quence, and the e x p r e s s i o n s "Steinitz theorem" and "L~vy-Steinitz the-
orem" are u s e d exchangeably. It w a s proved in [i0] that (i) holds for
each convergent series Zu i in a m e t r i z a b l e nuclear space. Example
(i0.i0) shows that there is no r e a s o n a b l e way of e x t e n d i n g this result
to s e r i e s in n u c l e a r groups.
It has been proved in [14] that if a metrizable locally convex
space is not nuclear, then it c o n t a i n s a convergent series ~u i such
that A(Zui) is not a linear subspace. Then (i) d o e s not hold, for,
in v i e w of (10.4) and the o b v i o u s inclusion A( Z u i) c F 0 ( Z u i ) , we
have

S( Z u i) c C( ~ u i) = ~ u i + A( Z u i) c Z u i + T0(~ui).
 106

It is not k n o w n w h e t h e r the f o l l o w i n g s en t e n c e is true: if a m e t r i z a b l e

locally convex space is not nuclear, then it c o n t a i n s a convergent se-
ries Zu i such that ~( Z u i) is not a linear manifold.

(lO.12) la~o~zs. In i n f i n i t e dimensional Banach spaces, the Levy-

-Steinitz theorem fails to h o l d (Problem 106 from the Scottish Book; see
[81], p. 188). The solution of this p r o b l e m has an i n t e r e s t i n g history;
see [34] or [43], pp. 44-45. V.M. Kadets [45] p r o v e d that e v e r y infinite
dimensional Banach space contains a convergent series w i t h a n o n - c o n v e x
set of sums. See also [46], especially Theorem i0.
Some years ago M.I. Kadets found an i n t e r e s t i n g example of a series
in a H i l b e r t space, making the c o n j e c t u r e that the set of sums of this
series consists of e x a c t l y two points. The c o n j e c t u r e was p r o v e d inde-
pendently by K. W o ~ n i a k o w s k i and P.A. Kornilov; see [44]. Making use of
this e x a m p l e and a p p l y i n g a standard technique, one can c o n s t r u c t a se-
ries w i t h n-point set of sums in every infinite dimensional normed
space; see [46], Theorem i0. A n o t h e r example of a series w i t h two-point
set of sums was o b t a i n e d by P. Enflo.

(I0.13) ~ . There are several analogous of the Levy-Steinitz

theorem which are v a l i d in i n f i n i t e dimensional Banach spaces. T h e y as-
sert that if ~u i is a c o n v e r g e n t series in a B a n a c h space E, then

condition (i) of (i0.Ii) is s a t i s f i e d under various additional assump-

tions on E and ~ u i. A typical example is the s i t u a t i o n when E is

a Hilbert space and ~lluill2 < ~. The best source of i n f o r m a t i o n here

is [46], pp. 158-159.

(10.14)~. Let R (0'I) be the space of all r e a l - v a l u e d func-

tions on the unit interval, endowed with the t o p o l o g y of pointwise
convergence (i.e. the p r o d u c t of c o n t i n u u m real lines). The e x a m p l e of
M.I. Kadets mentioned in (10.12) allows one to construct a series in
R (0'I) with two-point set of sums (see [43], Theorem 6.4.3, p. 172).

Applying (10.9), we see that the p r o d u c t T (0'I) of c o n t i n u u m circles

contains a convergent series such that its set of sums c o n s i s t s of ex-
actly two p o i n t s and is not a c o s e t m o d u l o any s u b g r o u p of T (0'I) (cf.
(i0.2)).

(lO.15) IaDOIRK. From the results of this section it follows easily

that if (gi) is an a r b i t r a r y null s e q u e n c e in a metrizable and complete
nuclear group, then there exist a permutation ~ of indices and a se-
 107

quence of signs E i = ±I such that the series ~£ig~(i) is c o n v e r g -

ent. On the other hand, it can be shown that if a metrizable locally
c o n v e x space (and p r o b a b l y even a locally q u a s i - c o n v e x group) is not
nuclear, then it c o n t a i n s a null s e q u e n c e (u i) such that the series
Zeiu (i) is d i v e r g e n t for each p e r m u t a t i o n ~ and each sequence

£. = ±i.
1

(I0.16)~. Let (gi)iEi be a s y s t e m of e l e m e n t s of an abe-

lian t o p o l o g i c a l g r o u p G. We say that (gi)i~i satisfies the C a u c h y

c r i t e r i o n of u n c o n d i t i o n a l c o n v e r g e n c e if to each U E No(G ) there

c o r r e s p o n d s a finite subset J of I such that gi e U for each
i~K
subset K of I\J. Next, we say that the system (gi)i~i is absolu-

tely s u m m a b l e if Z (gi/U) < ~ for each U ~ No(G). It turns out

iEI
that if a s y s t e m of e l e m e n t s of a n u c l e a r g r o u p s a t i s f i e s the Cauchy
c r i t e r i o n of u n c o n d i t i o n a l convergence, then it is a b s o l u t e l y summable.
Hence, e v e r y u n c o n d i t i o n l y c o n v e r g e n t series in a complete nuclear
group is a b s o l u t e l y convergent. The proof will be g i v e n in a separate
paper.

(ZO.17) R ~ u f f ~ . By the w e a k t o p o l o g y on an abelian topological

group G we m e a n the t o p o l o g y induced by the f a m i l y of all c o n t i n u o u s
c h a r a c t e r s of Go If G is a locally c o n v e x space, this t o p o l o g y is
m u c h w e a k e r than the w e a k t o p o l o g y i n d u c e d by the family of all con-
tinuous linear functionals, but defines the same class of convergent
sequences (it is e n o u g h to c o n s i d e r the case G = R).
A series ~gi in G is said to be subseries c o n v e r g e n t if the
series ~cig i is c o n v e r g e n t for each s e q u e n c e E i = 0,i. The Orlicz-
-Pettis t h e o r e m says that if a series in a locally convex space is sub-
series c o n v e r g e n t in the w e a k topology, then it is subseries c o n v e r g e n t
in the o r i g i n a l topology, too. The same is true for series in a sepa-
ble l o c a l l y q u a s i - c o n v e x group; this follows d i r e c t l y from Theorem 7
of [48]. Hence, in v i e w of (8.6), the O r l i c z - P e t t i s theorem remains
valid for series in a r b i t r a r y nuclear groups.
It seems v e r y likely that e v e r y w e a k l y c o n v e r g e n t sequence in a
nuclear group is c o n v e r g e n t in the o r i g i n a l topology.

( I O . 1 8 ) I~WIR~S. Let ~ : E ~ F be a linear operator a c t i n g be-

tween n o r m e d spaces. By ~(¢) we shall denote the smallest number
 108

r > 0 with the following property:

to e a c h finite system Ul,...,u n ~ B E there corresponds signs

el'" • "'En = ±i such that II¢(£iu I + ... + EnUn)ll ~ r.

We say that ¢ is a b a l a n c i n g operator if 8(¢) < ~.

The proof of (10.7) is b a s e d on (7.13). Lemma (7.13) also occurs

in the p r o o f of (10.5). It is not hard to see that (7.13) says, in
fact, that Hilbert-Schmidt operators are b a l a n c i n g . More precisely, if

¢ acts between unitary spaces, then ~(¢) = [ Z d ~ ( ¢ ) ] I/2 ; the p r o o f

k=l
very similar to t h a t of (7.13), is g i v e n in [13]. On the other hand,
it c a n be shown that if ¢ is an arbitrary operator acting between
normed spaces, then ~(¢) ~ C sup k I/2 Vk(¢) where C is a u n i v e r s a l
k
constant (the numbers Vk(¢) were defined in s e c t i o n 6). It is a stan-
dard fact that every finite dimensional operator ¢ is balancing,
with 8(¢) ~ 211¢11 r a n k ¢. The Beck-Fiala theorem [20] implies that
~(id : 11 ~ 1 ~) & 2. Beck and Spencer [21] proved that the diagonal
operator ¢ : 1~ + 1~ given by Ce k = k - i / 2 ( l o g k)-lek is b a l a n c i n g .

Spencer [86] showed that the factor (log k) -I can be o m i t t e d . It is

an o p e n problem whether the canonical embedding of 12 into i~ is

balancing (this is c a l l e d the K o m l o s conjecture).
By ~(¢) we d e n o t e the smallest number r > 0 w i t h the following
property:

to e a c h finite system Ul,...,u n e B E with uI + ... + un = 0

there corresponds a permutation ~ of indices, such that
J
II ~ Cu (i)ll ~ r (j = l , . . . , n ) .
i=l

We say that ¢ is a S t e i n i t z operator if ¢(¢) < ~.

Every finite dimensional operator ¢ is a S t e i n i t z operator with

%(¢) ~ I}¢II r a n k ¢ (see [34], [32] and [i0], Remark 3). Barany [16]
-3 k
proved that the d i a g o n a l operator ¢ : 1 ~ 1~ given by Ce k = 2 ek

is a S t e i n i t z operator with ~(¢) & i. Lemma (10.5) says that if ¢

acts between unitary spaces, then #(¢) ~ ~ dk(¢). It is not hard

k=l
to see t h a t ~(¢) ~ 2~(¢) for e v e r y ¢.
 109

By ~(¢) we d e n o t e the s m a l l e s t number r > 0 w i t h the f o l l o w i n g

property:

to each finite s e q u e n c e u I .... ,u n E B E there c o r r e s p o n d s a se-

quence of signs el,...,e n = ±i such that

J
II ~ Ei¢uill ~ r (j = l,...,n).
i=l

We say that ¢ is a s t r o n g l y b a l a n c i n g o p e r a t o r if o(¢) < ~.

E v e r y finite d i m e n s i o n a l o p e r a t o r ¢ is s t r o n g l y balancing with

0(¢) ~ 211ell rank ¢ (see [17]; in fact, ~(~) ~ II~II(2 rank ¢ - i)). J.
Beck [19] p r o v e d that the d i a g o n a l o p e r a t o r ~ : 1~ ~ 1~ given by

¢e k = k-(2+e) log k ek is s t r o n g l y b a l a n c i n g for each fixed c > 0. It

seems likely that the e x p o n e n t (2 + E) log k can be r e p l a c e d here by

some p o s i t i v e c o n s t a n t i n d e p e n d e n t of k. The e x i s t e n c e of such a con-
stant w o u l d imply that, for each null s e q u e n c e (u i) in a n u c l e a r Fre-
chet space, one can choose signs E i = ±i such that the series ~Eiu i
is c o n v e r g e n t (for c o u n t a b l e p r o d u c t s of real lines this fact was p r o v e d
by K a t z n e l s o n and M c G e h e e [51]). What is more, a nuclear Fr~chet space
could be r e p l a c e d here by a m e t r i z a b l e and c o m p l e t e nuclear group. On
the other hand, if a m e t r i z a b l e locally c o n v e x space is not nuclear,
then it c o n t a i n s a null s e q u e n c e (u i) such that the series ~£iui is
d i v e r g e n t for each s e q u e n c e E i = ±i (cf. (10.15)).
By definition, one has 8(¢) G ~(¢). Pecherski~ [74], Lemma i,
proved that, up to the notation, ~(¢) ~ 3a(¢). A v e r y short and simple
proof of the i n e q u a l i t y ~(¢) ~ ~(¢) was found by S. C h o b a n y a n (un-
published). Thus e v e r y s t r o n g l y b a l a n c i n g o p e r a t o r is a Steinitz opera-
tor.

(I0.192 ~ . The results of this s e c t i o n are new. The m e t h o d of

the proof of (10.8) is taken from [i0]. The a r g u m e n t a p p l i e d to obtain
(10.8) from (10.5) through (10.6) and (10.7) is standard. In various
forms, it occurs in several papers on i n f i n i t e d i m e n s i o n a l generaliza-
tions of the L e v y - S t e i n i t z theorem; see (10.13). Its main idea goes back
to Steinitz [87]. Lemma (10.7) is a s t r a i g h t f o r w a r d c o n s e q u e n c e of the
results of [20] or [21]; cf. [i0], R e m a r k 2.
 Chapter 4

THE ~ - ~ u ~ ( l ~ l

In this chapter we show that n u c l e a r groups satisfy Bochner's the-

orem on p o s i t i v e - d e f i n i t e functions. Section l! wears an i n t r o d u c t o r y
complexion. We i n t r o d u c e here some new t e r m i n o l o g y and state several
standard results in a form c o n v e n i e n t to us. Section 12 contains the
proof of the m a i n result. Finally, in s e c t i o n 13 we give some applica-
tions. We formulate here an a p p r o p r i a t e version of the SNAG theorem for
nuclear groups and prove that e a c h c o n t i n u o u s positive-definite function
(resp. continuous unitary representation) defined on a subgroup of a
nuclear group can be e x t e n d e d to the w h o l e group.

Ii. P r e l i m i n a r i e s

Bochner's classical theorem asserts that each c o n t i n u o u s positive-

-definite function on the real line is the F o u r i e r transform of some
Radon measure. Many far-reaching generalizations of this fact are known.
Roughly speaking, they say that, under certain assumptions on a topolo-
gical group G, each continuous positive-definite function on G may
be w r i t t e n as an i n t e g r a l of some m e a s u r e on the dual object G (cf.
[63], Ch. VI, §9).
We c o n f i n e ourselves to a b e l i a n groups only. By a B o c h n e r theorem
for an a b e l i a n group G we shall m e a n a statement of the following form:

(*) each continuous positive-definite function on G is the "Fou-

rier transform of a (unique) Radon measure on G .

There are at least three situations where (*) is k n o w n to be true:

(i) the W e i l - R a i k o v theorem asserts that (*) holds for any LCA
group G;

(ii) the fact that (*) is true for every nuclear locally convex
space G is k n o w n as the M i n l o s theorem (see [67], Theorem i, p. 508
or [63], Ch. IV, T h e o r e m 4.3, p. 318); it follows easily from the M i n l o s
theorem that (*) is true w h e n G is a H a u s d o r f f quotient group of a
nuclear locally convex space (see Y a n g [104]);

(iii) each locally convex space G over p-adic field satisfies

(*); this was p r o v e d by M a d r e c k i [62].
 111

We do not s p e c i f y here the t o p o l o g y on G-. The W e i l - R a i k o v the-

orem is u s u a l l y formulated for the c o m p a c t - o p e n topology, whereas the
Minlos theorem in the l a n g u a g e of vector spaces, the t o p o l o g y on the
dual space G being the t o p o l o g y of u n i f o r m convergence on finite,
bounded or, say, compact convex subsets of G (there is a canonical
isomorphism between G* and G-; see (2.3)). The result of (iii) was
proved for the w e a k * topology on G*.
We shall p r o v e that (*) h o l d s for e v e r y n u c l e a r group G (The-
orem (12.1)); this is a c o m m o n generalization of (i) - (iii).
The B o c h n e r theorm in the form of (*) characterizes nuclear spaces
among m e t r i z a b l e locally convex spaces ([70], Theorem 5, p. 75). On the
other hand, there is v e r s i o n of the B o c h n e r theorem which is v a l i d in
any l o c a l l y convex space ([30], Theorem I, p. 348). Thus, each contin-
uous p o s i t i v e - d e f i n i t e function on a l o c a l l y convex space can be in
some w a y s y n t h e s i z e d of c o n t i n u o u s characters. The situation becomes
completely different w h e n we start to c o n s i d e r quotient groups. In
section 5 we gave an e x a m p l e of a d i s c r e t e subgroup K of the space
i p, p > 2, such that the q u o t i e n t group IP/K admits non-trivial
continuous unitary representations but does not admit any non-trivial
continuous characters. Thus, there are on IP/K continuous positive-
-definite functions which cannot be s y n t h e s i z e d of c o n t i n u o u s charac-
ters (since the latter do not exist); therefore one c a n n o t speak of any
version of B o c h n e r ' s theorem in this case. It is q u i t e possible that
similar examples can be c o n s t r u c t e d in any n o n - n u c l e a r locally convex
metrizable space.
Let G be an a b e l i a n topological group. We say that • is an
admissible topology on G if the m a p p i n g s G B X ~ x(g), g~G, are

continuous and the sets o

UG, U ~ N o (G), are c o m p a c t in G . From
(1.5) it follows that the t o p o l o g i e s of pointwise, compact and p r e c o m -
pact convergence are all admissible.
Let X be a t o p o l o g i c a l space. The f a m i l y of B o r e l subsets of X
is d e n o t e d by B(X). By a B o r e l m e a s u r e on X we m e a n a o-additive
mapping of B(X) into [0,~]. A finite Borel measure ~ on X is
called a Radon measure if, for e a c h A ~ B(X) and e a c h ~ > 0, there
exists a compact subset Q of A with ~(A \ Q) < £.
Let X,Y be two t o p o l o g i c a l spaces, ~ : X ~ Y a Borel mapping
and ~ a Borel measure on X. Then the m a p p i n g B(Y) B A + ~(~-I(A))
is a Borel m e a s u r e on Y. We call it the H - i m a g e of ~ and d e n o t e by
~ . If f is a ~ - i n t e g r a b l e function on Y, then f~ is a ~-inte-
grable function on X and

f f~d~ = f fd~ .
X Y
 112

Let G be an a b e l i a n topological group and let ~ be a t o p o l o g y

on G such that all the m a p p i n g s G~ X ~ x(g), g e G, are continuous.
By the Fourier transform of a f i n i t e Borel measure U on G we m e a n
T
the function

G 3 g + J" _ exp [2~ix(g)]du(x).

G%

The Fourier transform of ~ is d e n o t e d by ~; it is sometimes called

the inverse Fourier-Stieltjes transform of ~. A Borel measure U on
G will be c a l l e d re@ular if, for e a c h A e E(G ) and each ~ > 0,
there exists a compact equicontinuous subset Q of A with U(A \ Q ) < c.

~]I.1) ~ m ~ . Let G be an a b e l i a n topological group and let

T be a t o p o l o g y on G such that the m a p p i n g s G ~ X + x(g), g ~ G,
T

are c o n t i n u o u s . If ~ is a f i n i t e Borel measure on G~, then ~ is

a p.d. function on G with ~(0) = ~(G~). If ~ is r e g u l a r , ~ is
continuous.

~f. The fact that U is a p.d. function is well known (see

e.g. [38], (33.1)). So, assume that ~ is r e g u l a r and take any e > 0.
There is an e q u i c o n t i n u o u s subset Q of G- w~th ~(\Q) < i£. Next,
T

we c a n find some U E No(G ) such that Ix(U) I < ~ for all X E Q. The~
for e a c h g e U, we h a v e

l~(g) - ~(0)I : IS (i - e x p [2~ix(g)~)dy(x) I

G-
J- Ii - e x p [2~ix(g)]Idu(x) & f 2~Ix(g)Idu(x)
G- G-

= 2~ S [xlg) Idu(x) + 2~ S l×(g) Id~(x)

Q \Q

2~cp(Q) + ~(\Q) < [2~(G-) + i]~£.

Thus ~ is c o n t i n u o u s at z e r o and the continuity at the remaining

points follows from (1.22) (c). •

(11.~ PRO~S/T~. Let ¢ : G + H be a c o n t i n u o u s homomorphism of

abelian topological groups. Suppose that the d u a l groups G- and H-
are endowed with some topologies such that the m a p p i n g s G- B X + x(g),
g ~ G, and H- B < + K(h), h ~ H, are continuous. Suppose also that
the d u a l homomorphism ~ = ~- : H A ~ G- is c o n t i n u o u s . If ~ is a
finite Borel measure on H ~, then ~ = ~¢.
 113

~ . For each g ~ G, one has

we(g) = j c exp [2~i<x,g>] dD~(X) = f exp[2~i<~(x),g>]d~(K)

G H
= ~c exp[2~i<K,#(g)>jd~(K) = ~(#(g)). •
H

fl].3)~ITx~. Let G be an a b e l i a n t o p o l o g i c a l group and let

T be a t o p o l o g y on G such that the m a p p i n g s G ~ X ~ x(g), g ~ G,
T
are continuous. If ~,~2 are two R a d o n m e a s u r e s on G# with Ul =
then W~ = ~2.

Proof. Let Gd denote the g r o u p G endowed with the discrete

topology. The i d e n t i t y h o m o m o r p h i s m # : G- ~ (Gd) ~ is continuous.
Let ~i be the #-image of ~i' i = 1,2. Then 9i is a R a d o n m e a s u r e
On (Gd) ~ and (11.2) implies that vi = Wi" Thus ~i = ~2- From the
u n i q u e n e s s of the m e a s u r e in the W e i l - R a i k o v t h e o r e m it now follows that
~i = ~2- If Q is a compact subset of G~, then $(Q) is a compact,

hence Borel, subset of (Gd)c, and ~i(Q ) = ~I(¢-I(#(Q))) = ~i(#(Q))

for i = 1,2,. Hence ~I(Q) = ~2(Q)- This c o m p l e t e s the proof because
~,~2 are b o t h R a d o n measures. •

~II.4) ~rl~. The m a t e r i a l of this s e c t i o n is standard.

12. The B o c h n e r t h e o r e m

The aim of this section is to prove the f o l l o w i n g fact:

(12.1) ~ . Let G be a nuclear g r o u p and ~ an a d m i s s i b l e

t o p o l o g y on G . T h e n the m a p p i n g ~ + ~ establishes a one-to-one
c o r r e s p o n d e n c e b e t w e e n the f a m i l y of all regular finite Borel m e a s u r e s
on G~ and the f a m i l y of all c o n t i n u o u s p.d. functions on G.

Let E be a n u c l e a r space and E* the dual space endowed with

the t o p o l o g y of u n i f o r m c o n v e r g e n c e on finite, compact or precompact
sets. It is not d i f f i c u l t to see that the t o p o l o g y induced on E by
the c a n o n i c a l h o m o m o r p h i s m PE : E* ~ E is an a d m i s s i b l e one. Thus
(12.1) implies the Minlos theorem.

The proof given b e l o w is p a t t e r n e d upon that of the Minlos t h e o r e m

The m a i n d i f f e r e n c e lies in r e p l a c i n g the Minlos lemma ([67], Lemma 4,
p. 510) by its a n a l o g u e for p.d. functions on a d d i t i v e subgroups of R n
(lemma (12.2) below). To o b t a i n (12.1) from (12.2) is a m a t t e r of tech-
 114

nique; we h a v e found it m o s t convenient to aply here the Prokhorov

theorem on i n v e r s e limits of m e a s u r e s , in the form due to Kisy~ski
[540.

(x2.2) ~ = Let N be an n - d i m e n s i o n a l vector space, n ~ i, a n d

let p,q be two p r e - H i l b e r t seminorms on N such that

59
dn i/2(Bp,Bq) < ~_~.i
k=l

Let K be a subgroup of N and ~ a Borel measure on K with ~ ( K )

= i. Suppose that there is s o m e E > 0 such that Re ~(u) > 1 - e for

all u ~ K n Bq. Then ~((K n Bp) o

K) >
. 1 - 2e .

•~f. Suppose first that both p and q are norms. In this

case, we may simply assume that N = Rn and q is the euclidean norm

on N. Then Bq = B n and Bp is s o m e ellipsoid in Rn; let ~n .... '

~n be its p r i n c i p a l semiaxes. Obviously, we have

(i) Xn ~ i/2 = n d 1 /2(Bp,Bq)

59 < T2"I
k=l k=l

We begin with the case when K is d i s c r e t e . Let L = gp (K n Bp)

and let ~ : K p ~ Lp be the natural homomorphism. Denote ~ =~ . Ac-

cording to (11.2), we have ~ = ~]L" We shall prove that

(2) ~((L n Bp)~) ~ 1 - 2E.

Let M = span L and m = d i m M. If L = {0}, then

~((L n Bp ) Lo ) . ~ ( { o.) ~ ) . ~ ( L .- ) .G(o) ~(o) > 1 - E > 1 - 2E .

So, suppose that m Z i. Let ~l,...,O m be the principal semiaxes of

the e l l i p s d i d D = M n Bp. From (i) and (2.13) it f o l l o w s that

m ~/2 1
(3) X n < ~.
k=l

We may assume that M = Rn and

m
D = { ( X l , . . . , x m) ~ Rm : Z (Xk/~k)2 < i}.
k=l

Set ~,. = q3/4 for k = 1 ..... m and denote

J%
 115

m
2 ~<i}.
E = { ( X l , . . . , x m) E Rm : ~ (Xk/~k)
k=l
We have L = gp (L D D) and, by (3),

m 2 m )2 m 2 ~2
Z d (D,E) = 5q ( n k / ~ k = ~ qk < < 4.
k=l k=l k=l

According to (3.17), we can find a rectangular parallelotope p c E

such that {u+P}u~L is a disjoint covering of R m. Set ak =

ql/4 for k = l,...,m and denote

A = { ( X l , . . . , x m) ~ Rm : -a k < x k < a k for k = l,...,m}.

From (3) w e get a2 + ... + a 2 < 1/12 which implies that A c B .

m ' q
Therefore, for each u E L n A, we have Re ~(u) = Re 6(u) >~ 1 - e,
i.e.

(4) Re f exp [2~iK(u)]d~(K) a i - g.

L

let us write IWI = c a r d (L N W) for W c M. Consider the function

L B K ~ f(K) = 1 Re exp [2~iK(U)]

IAl u~nA

and denote X = {K E L : flK) ~ }. From (4) w e obtain

f A
f(K)d~(K) ~ 1 - E,
L

whence 9(X) ~ 1 - 2~ because f < i. Thus, to prove (2), it remains

to show that X c (L n B p ) Lo"

To this end, take any K e X and w = ( W l , . . . , w m) ~ L n D. We aae

to show that

(5) l<(w) l s ¼.
Since P c E, and the sets u + P, u E L, are pairwise disjoint, we
have

IAI VOlm(P) = vol m ( U (u + P)) ~ vol m (A + E).

u~LDA

It is clear that

A + E c {(Xl, .... Xm) ~ Rm : -ak - ~k < X k < a k + ~k

for k = l,...,m},
 116

whence
m
(6) IAI v o l m (P) ~ 2m Z (a + ~ k ).
k=l

Let A= w + A and B = A n A . The set U = M \ ((M \ B) + E) is

w w
an m-dimensional rectangular parallelotope with edges equal resp. to

2a k - 2 ~ k - lWkl, k = 1 ..... m. Since p c E, and {u + P } u ~ L is a

covering of R m, it follows that U c U (u + P). Hence

uELRB
m
(7) IBI v o l m (P) ~ vol m (U) = 11 (2a k - 2~ k - lWkl).
k=l

Let V be the symmetric difference of A and Aw, i.e.

V = (A \ A w) U (A w \ A) = (A \ B) U (A w \ B).

Let us write

1 1
s -- - - exp [2~i~(u)]," t = Z exp [2~i<(u)].
lal u~Lna lat u~Lna
w

Then s = t exp [2~iK(w)]. On the other hand,

Is - t[ < ~1
= Z I exp [2~iK(u)][ ~ }V I .
IAI u~nnv " IAI

It is clear" that IVI = 2( IAI IBI), whence, by (6) and (7),

m
~q (2a k - 2 ~ k - lWkl)
k=l
I s - tl s 211 - ~ j < 211 .
m
2 m II (a k + ~k )
k=l

Hence, applying the standard inequalities

m m
1 - x - y ~ 1 - 2x - y and II (i - x k ) h 1 - Z Xk,
l+x k=l k=l

we obtain
m m
Is - t I ~ 4 Z (~k/ak) + 2 Z (lWk}/ak)-
k=l k=l

m
Since w ~ D, we have Z (Wk/Dk)2 _-< i, and thus, lWk I ~ q k for
k=l
k = l,...,m. Therefore, by (i),
 117

m i/2 m 3/4 m 1/2 1

Is - t I ~ 4 Z ~K + 2 ~ Ok ~ 6 Z qk < 3"
k=l k=l k=l

On the other hand, we have

I s - tl = It e x p [2~i<(w)] - t I = Itl" 11 - e x p [2~iK(w)]I.

But K e X, therefore Re t = f(K) ~ 1/2 and Itl ~ 1/2. Hencewe

o
obtain (5) and, in consequence, (2). It is clear that w-l((L D Bp) L) =
(K A Bp)~. Then, by (2), we get

p((K n Bp)~ = p ( ~ - l ( ( L A Bp)~) = ~ ( ( L A Bp)~) ~ 1 - 2~.

This c o m p l e t e s the proof in the case w h e n K is discrete.

Now, let K be an a r b i t r a r y s u b g r o u p of N. We m a y c l e a r l y assume

K to be closed. Then K is the direct sum of some d i s c r e t e subgroup
and some linear s u b s p a c e (see (3.1)) and, therefore, we can find an in-
c r e a s i n g sequence (Km) of d i s c r e t e s u b g r o u p s of K such that their
union is dense in K. For e v e r y m, let ~m : K- ~ be the natural

h o m o m o r p h i s m and let ~m = P~m" From (11.2) we get Pm = ~ I ~ " Since,

as we have just proved, the lemma is true for discrete subgroups, it

follows that pm((K m N Bp) °) ~ 1 - 2E for e v e r y m. The s u b g r o u p s
Km
may be c h o s e n in such a way that the c o n d i t i o n
co

n ~I((K m n Bp) °) c (K n Bp) °

m=l
be satisfied. Then
co

p((K N Bp) °) _-> p( n ~ m l ( ( K m D B )o))

m=l P

= lim P(~[l((Kmm N B )o)) = lim pm((K m N Bp) °)

m~ P m~

> i - 2~.

This c o m p l e t e s the proof in the case w h e n p and q are norms.

Finally, c o n s i d e r the case when p and q are arbitrary pre-

-Hilbert s e m i n o r m s on N. Then, clearly, one can find two decreasing
sequences (pm) and (qm) of u n i t a r y norms on N such that

n
dl/2 1
Z k (Bpm'Bqm) < i--~ (m = 1,2,...),
k=l
 118

Bqm c Bq for e v e r y m and Bp c m=iU Bpm. A p p l y i n g our lemma to the

norms Pm and qm' we get ~((K N Bpm)°) ~ 1 - 2E for every m.

Hence

~ ( ( K N Bp) °) ~ p( n (K N B ) o ) = lim ~((K N B p m ) ° ) ~ 1 - 2£. •

m=l Pm m~=

~12.3; ~ i , ~ [ ~ . Let (~,c) be a d i r e c t e d set. For each K E ~,

let XK be a H a u s d o r f f t o p o l o g i c a l space and let ~K be a Radon

m e a s u r e on XK with PK(Xk) = I. Suppose that, for each pair (K,L)

~2 with K c L, a continuous mapping ~KL of XL into XK is given,
such that

(i) ~KK = idx K for each K E ~,

(2) ~ K L ~ L M = ~KM

for each triple (K,L,M) ~ ~3 with K c L c M, and

-i
(3) ~K(A) = ~L(~KL(A))

for e a c h pair (K,L) ~ ~2 such that K c L and for each A ~ B(XK).

Suppose further that X is a H a u s d o r f f t o p o l o g i c a l space, and that,
for e a c h K ~ ~, a continuous mapping ~K of X onto XK is given,
such that

(4) if (K,L) ~ ~2 and K c L, then ~K = ~KL~L"

Finally, suppose that, for any two d i s t i n c t points x,y E X, there

exists some K E ~ such that ~K(X) # ~K(y). T h e n the following two
s t a t e m e n t s are equivalent:

(i) there is a unique R a d o n m e a s u r e ~ on X with ~(X) =i such

that ~K(A) = ~(~KI(A)) for each K E ~ and each A e B(X);

(ii) for each c > 0, there is a c o m p a c t subset Q of X such

that ~ K ( X K \ ZK(Q)) ! c for e a c h K ~ ~.

This is T h e o r e m 3.2 of [54].

~2.4; r.R~m~. Let H be a s u b g r o u p of a nuclear vector g r o u p and

a c o n t i n u o u s p.d. f u n c t i o n on H. T h e n there exists a finite regu-
A

lar B o r e l m e a s u r e ~ on H with ~ = ~.
P
 119

Pmm~f. Suppose first that ~ (0) = i. Let K be a finitely gene-

rated subgroup of H. Then there exists on K a unique Radon measure
P
~K with PK = elK" Indeed, the closure K of K in span K, being
a closed subgroup of a finite d i m e n s i o n a l vector group, is locally com-
pact. Hence, by the W e i l - R a i k o v theorem, there is a Radon measure 9 on
(K)c with ~ = ~i~. We may identify (K)c with Kpc because, evi-

dently, each compact subset of K is c o n t a i n e d in the closure of some

precompact subset of K. T h e r e f o r e we may treat ~ as a m e a s u r e on
KDC_ and write v = ~IK. The r e s t r i c t i o n ~K of ~ to Borel subsets

of K is then a Radon measure on Kp with ~K = ~ IK" The u n i q u e n e s s

of ~K is a c o n s e q u e n c e of (11.3). From (ii.i) we get ~K(K ) = ~(0) =

= i.
Let ~ be the d i r e c t e d family of all finitely generated subgroups
of H. For each K E R, put X K = Kp. For each pair (K,L) ~ ~2 such
that K c L, let ~KL : Lp ~ Kp be the natural homomorphism. Put
X = Hp- and, for each K E ~, let ~K : Hp ~ Kp be the natural homo-
morphism.
In the situation we have just described, the assumptions of (12.3)
are satisfied. The c o n t i n u i t y of the h o m o m o r p h i s m s ~K and ~KL is
obvious and their s u r j e c t i v i t y follows from (8.3). Conditions (I), (2)
and (4) of (12.3) are s a t i s f i e d trivially, as well as the fact that,
for any two d i s t i n c t XI,X 2 ~ H , there is some K ~ ~ with ~K(XI )=
~K(X2 ). Condition (3) is a c o n s e q u e n c e of (11.2). We shall prove that
(ii) is satisfied.
Fix an E > 0. Let F denote the nuclear vector group contain-
ing H. Since ~ is continuous and ~ (0) = i, there is some
U ~ No(F) such that

(I) Re %(h) > 1 - E for h ~ H n U.

Next, there are a linear subspace N of F and three pre-Hilbert

seminorms p,q,r on N, such that B c U,
r

(2) 5 ~ kdk(Bp,B q) ~ ¼,
k=l

1/2 1
(3) 5-:. d k (Bq'Br) < I--2
k=l

and Bp E No(F ) (see (9.3), (2.14) and (2.15)). T h e n W: = H D B p eNo(H),

 120

and WHO is a c o m p a c t s u b s e t of H-p (see (1.5)).

Take any K ~ ~; w e are to s h o w that

(4) ZK(K- \ ~K(W~)) ~ E.

Denote L = K N N and M : s p a n L. Let s and t be the restric-

tions to M of q and r, respectively. Then s,t are pre-Hilbert
seminorms on M and, by (2.13), we h a v e

d k ( B s , B t) ~ d k ( B q , B r) (k : 1 , 2 , . . . ) .

Hence, by (3),
co
1
7. d l / 2 ( B s , B t ) < ~.
k=l

From (i) a n d (12.2) it n o w f o l l o w s that

~L((L N B s)Lo ) ~ 1 - E,

w h i c h m a y be w r i t t e n as

-i o
(5) ~K(~LK((L n Bq)L)) ~ 1 - £.

Since K N Bq L N Bq, we h a v e

O
(6) ~ L-I
K((L N B q )~) c (K N Bq) K.

From (2), (8.1) and (2.3) it f o l l o w s that

(7) (K N Bq) Ko c ~K(W~) .

Finally, from (5) - (7) we o b t a i n (4) b e c a u s e ~K(K ) = %(0) : i.

We h a v e p r o v e d that c o n d i t i o n (ii) of (12.3) is s a t i s f i e d . So, ac-
cording to (12.3), there is a R a d o n m e a s u r e ~ on Hp with ~(H )= 1
such that ZK(A) = ~( ~K- I (A)) for e a c h K ~ q and e a c h A ~ B(Kp). For
each K e 9, the m e a s u r e ZK is the ~K-image of Z. Hence, by (11.2),
we h a v e ZI K : ZK : %IK" Thus ~ : %. It r e m a i n s to s h o w t h a t ~ is

a regular measure.
As we h a v e just seen, to e a c h n there co,rresponds some Wn~ No(H)

such t h a t ~ K ( K ^ \ ~K ( w~) ) ~ n1 for all K ~ ~" The set Qn = w°n is

 121

equicontinuous and, by (1.5), compact in Hp. The family of sets of the

form ~KI(K \ ~ K ( Q n )) where K e ~ forms an o p e n covering of Hp \ Qn"

Hence, for e a c h compact y c Hp \ Qn' we c a n find some K E ~ such that

y c ~KI(K ^ \ ~K(Qn)). Then

- - 1
P(Y) < ~ ( ~ K I(K \ ~ K ( Q n ))) = PK (K \ n K ( Q n )) < n"

- 1
This implies that ~(H \ Qn ) ~ ~ since p is a R a d o n measure. Thus p
is r e g u l a r .
It r e m a i n s to c o n s i d e r the case ~(0) # i. If ~(0) # 0, then it
suffices to c o n s i d e r the p.d. function ~/~(0) instead of ~. On the
other hand, if ~(0) = 0, then, by (1.22) (b), w e h a v e ~ £ 0 and
is the t r a n s f o r m of the zero m e a s u r e . •

~faf (12.1;. Due to (ii.i), if p is a r e g u l a r finite Borel

measure on G~, then ~ is a c o n t i n u o u s p.d. function on G. From
(11.3) it f o l l o w s that the m a p p i n g p + p is i n j e c t i v e . We shall prove
that it is s u r j e c t i v e , as well.
So, take any continuous p.d. function ¢ on G. By (9.6) there
exist a nuclear vector group F, a subgroup H of F and a closed
subgroup K of H, such that G - H/K. We m a y simply assume that G =
H/K. Let ~ : H ~ H/K be the natural projection. Then ¢~ is a con-
tinuous p.d. function on H and, due to (12.4), there exists a regular
finite Borel measure ~ on H with v = ~. Let ~ : H ~ Kp be the
A

natural homomorphism. By (11.2), we have ~ = VlK = #~IK ~ ~(0). In view

I I

of (11.3), this means that v is the Dirac measure. Then v ( K °) =

v(~-l(0)) = v ({0}) = ~(0), i.e. ~ is c o n c e n t r a t e d on K °. We may

treat v as a finite r e g u l a r B o r e l measure onK °.
P
The natural homomorphism o : K° ~ (H/K)p is c o n t i n u o u s (in fact
p
it is a t o p o l o g i c a l isomorphism; however, the m a p p i n g ~ : Kpc
o
~ (H/K)
;
need not be c o n t i n u o u s ; consider e.g. the c a n o n i c a l mapping L ° ~(D/L)c
pc
in (17.6)). Set ~ = v . Then X is a f i n i t e regular Borel measure on
o
G . For each h ~ H, one has
P

~(~(h)) = f ^ X(~(h))d~(x ) = f <(h)dv(<) = J- _ < ( h ) d ~ ( K )

G K° H
P P P
= v(h) = ¢(~(h)),

which proves that X = ¢.

 122

Since ~ is r e g u l a r and ~(H ) ¢(0), there is a s e q u e n c e Un

No(H) with u(U~) ~ ~ ( 0 ) . The s e t s Qn = (~(Un))° are equicontinuous
and c o m p a c t in G~ • Moreover
T

)o) = k ( o ( U o N K ° ) ) = ~ ( U ° N K °) = ~(U~) + ¢(0)

l(Qn) = X(~(Un n n

= X(Gp),

i.e. X ( G p \ Q n ) + 0. Let ~ be the c o m p l e t i o n of I. T a k e any-closed

subset Y of G . T h e sets Y N Qn' n = 1,2,.•., are c o m p a c t in G$,

thus in Gp. Hence U (Y N Qn ) e B(Gp). On the o t h e r hand,

n=l

X(Y \ U (Y N Qn )) ~ ~( n (G \ Qn )) = 0.
n=l n=l P

This implies that Y is T-measurable. Hence all B o r e l subsets of G are

T
T-measurable.
Let ~ be the r e s t r i c t i o n of ~ to B (G~).
[
It is c l e a r that ~ =
X. Hence ~ = ¢. It r e m a i n s to p r o v e that ~ is regular. So, t a k e a n y
A ~ B(GT) and £ > 0. Since A is T-measurable, t h e r e are some B,N
H(GD) with X(N) = 0 and B \ N c A c B U N• Next, since X is reg-
I

ular, there exists a compact equicontinuous subset Q of B \ N with

X(B \ N \ Q) < ~. Take n so large that X ( G D+ \ Qn ) < E. The set Q
is c l o s e d in GD, thus in G~. Therefore Q N Qn is compact in G~.
Finally, we h a v e

X(A \ (Q D Qn)) = X(B \ (Q 0 Qn )) ~< X(B \ Q) + k(B \ Q n )

< E + ~. •

112.5) ~ . Let G be a n u c l e a r group and ~ an admissible

topology on G . Let {~i}i~i be a f a m i l y of r e g u l a r Borel measures

on G~, with ~i(G-) = 1 for e v e r y i. T h e n the f a m i l y {~i}iei of

p.d. functions on G is e q u i c o n t i n u o u s at zero if a n d o n l y if to each

£ > 0 there corresponds a compact equicontinuous subset Q of G~
such that ~i(G \ Q) & ~ for all i ~ I.

Proof. The sufficiency of the c o n d i t i o n follows immediately from

the p r o o f of (Ii.i). We s h a l l p r o v e the n e c e s s i t y • It is not h a r d to see
that, without loss of g e n e r a l i t y , we m a y a s s u m e that T is the t o p o l o g y
of p o i n t w i s e convergence• Next, due to (9.6), we m a y a s s u m e that G =H/K
 123

where H is a subgroup of some nuclear vector group F, and K is a

closed subgroup of H. Suppose first that K = {0}.
Choose any ~ > 0. We can find some U ~ N (F) such that
o

(i) Re p i(u) ~ 1 - ~ for all u e U and i ~ I.

According to (9.3) and (2.14), we can find some linear subspace N of

F and two pre-Hilbert seminorms p,q on N, such that Bp ~ No(F),

(2) X d ~/2(Bp,Bq ) < i--2

1
k=l

and Bq c U. Take any i ~ I and any compact X c Hp \ (H DBp)~. Since

Pi is a regular measure, it is enough to show that ~i(X) ~ ~.

Let ~ be the family of all finitely generated subgroups of H.
For each L ~ ~, let ~L : Hp ~ L be the natural homomorphism. It
is clear that

B
(H D p)Ho = n ~LI((L A Bp)L),
Lg~
whence

H \ (H A Bp) = U ~LI[L - \ (L D B p)Lo ] .

Lg~

Since X is compact, it follows that there is some L E ~ with X c

-i ~ B o
~L [L \ (L A p)L ]. Let PL be the ~L-image of Pi" From (11.2)
we get PilL = ~L" Hence, by (i), (2), (12.2) and (2.13), we obtain

O
Pi(X) < pi[~Ll(L - \ Bp)L] : PL(L- \ (L N Bp) L) < E.

It remains to consider the case when K # {0}. Let ~ : H ~ H/K

be the canonical projection and # : (H/K)p ~ H the dual homomorphism.
For every i, let ~i be the ~-image of Pi; then ~i = P i ~ accord-
ing to (11.2). Since the family {~i } = {pi¢} of p.d. functions on H
is equicontinuous at zero, the above implies that to each ~ > 0 there
corresponds a compact equicontinuous subset Q of ~ with ~i(H-\ Q) ~ c
for all i e I. Then

pi((H/K)- \ ~-I(Q)) = pi(~-l(H- \ Q)) = ~i(H- \ Q) ~ e.

Now, it remains to observe that the closure o L ~-I(Q) is a compact equi-

continuous subset of (H/K)p (cf. (1.5)). •
 124

(12.6) m~f~. The material of this section is now.

13. T h e SNAG theorem

613.1; T ~ _ Let G be a n a b e l i a n topological group and T a

topology on G such that the mappings G ~ K ~ K(g), g • G, are
T

continuous. Let ~ : G ~ (G$) be the n a t u r a l homomorphism and p a

Radon measure on G . Then span c~(G) is a dense subset of L (G ,l~).

P~f. It is e n o u g h to s h o w t h a t the characteristic function of

compact subsets of G can be approximated in L (G ,~) by elements

of span ~(G). So, take any compact subset X of G and any ~ > 0.
T

Since p is a R a d o n measure, there exists a compact subset Y of

G- \ X with
T

(i) Z(G- \ X \ Y) < c.

Denote U = G \ Y. Since Y is a c o m p a c t subset of G , it is a

compact subset of Gp, which means that U is a n o p e n s u b s e t of Gp.

So, for each 6 E X, there are a finite subset G~ of G and some

6~ > 0, such that

(2) U%: = {K E G : l<(g ) - ~(g)l < 6~ for all g ~ G~} c U.

The s e t s U~, ~ E X, are open in Gp and cover the compact set X.

So, there is a f i n i t e subset Q of X such that

(3) X c U U~.
%

Arrange all elements of the finite set U G~ in a sequence

~Q
(gk)k=l'n Let ~ : Gp- + S n be t h e continuous homomorphism given by
n - ¢-i
the formula ~(X) = K(gk)k: 1 for K • G • Denote V = (~(X)). We
shall prove that

(4) X • V c U.

The first inclusion is t r i v i a l . To prove the second one, choose any

K • V. We have %(<) : ¢(X) for some X • X. Next, by (3), we have
X • U< for some ~ • Q. This implies that < • U$. Hence, by (2), we

get K • U, which proves (4).

 125

The restriction ~ of ~ to B o r e l subsets of Gp is a Radon

measure on G . So, ~ is a R a d o n measure on Sn . Let 11 H G - and

II llsn denote the norms in t h e spaces L (G- ,~) and L c2( S n ,~$), re-

spectively. Let XA denote the characteristic function of a set A.

Since trigonometric polynomials are dense in L~(Sn,~%) (see [38], (31.4)),

we can find some complex numbers ll,...,l p and some continuous homo-
morphisms (characters) ~k : Sn + S, k = l,...,p, such that

P
IIX¢(x) k Z= l lkDkJlsn < £.

Then

P P 2d~]i/2
]bXV - Z lkDk~[l _ = [f _ [X V - Z Xk~k$ [
k=l G G k=l

P 2d~]i/2
= [• [X~(x) ~ lkq k] < e.
Sn k=l

Hence, applying (i) and (4), we obtain

P P
IIXx - ~ >k~k¢IL _ < fIXx ×Vl] _ + IIXv - Z kk~k¢Ib _
k=l G G k=l G

[~(U \ X)] I/2 + a < £i/2 + E.

Thus it r e m a i n s to show that Dk ~ e ~(G) for k = l,...,p.

Fix any k = l,...,p. We can find some s I .... ,s n ~ Z such that

sI sn
Dk(Zl,...,z n) = zI ... zn

for (z I , ...,z n) e Sn (cf. (1.7)). Then, for each ~ e G , we have

n sk n
(Dk$)(<) : qk(<(gl),...,<(gn)) : II < ( g k ) : <( Z S k g k ) ,
k=l k=l

n
which means that Dk$ : ~( Z Skgk) e e(G).
k:l

~]3.2) ~ . Let $ be a continuous cyclic unitary representa-

tion of a nuclear group G and let ~ be an admissible topology on
 126

G . Then there exists on GT a regular finite Borel measure p such

that ~ is u n i t a r i l y e q u i v a l e n t to the representation ~ in the s p a c e
2
LC( G~ ,~ ), g i v e n b y the f o r m u l a

(%'gf)(x) = f(x) exp [2~ix(g)]

for g ~ G, X ~ G and L (G ,~).

~f. Let u be a c y c l i c vector of ~. Then the formula ~(g) =

(¢gU,U), g • G, defines a continuous p.d. function ¢ on G (see
(1.23)). By (12.1), there exists a regular finite Borel measure U on
GT with ~ = ~. Let f be the function identically equal to 1 on
G . From (13.1) it f o l l o w s that f is a c y c l i c vector of ~. For
each g e G: we h a v e

(~gf,f) = jc ^ exp [ 2 ~ i x ( g ) ] d ~ ( x ) = ~(g) = ~(g) = (~gU,U).

G
This implies that ~ and ~ are u n i t a r i l y equivalent (see (1.25)). •

Since a continuous unitary representation c a n be w r i t t e n as the

Hilbert s u m of c y c l i c representation (see (1.21)), we could easily
formulate an a n a l o g u e of (13.2) for non-cyclic represe~ations (cf.[38],
(21.14)). Instead, we shall give the S N A G theorem for n u c l e a r group in
its u s u a l form (cf. [18], p. 160).

~13.3) ~ . Let ¢ be a c o n t i n u o u s unitary representation of

a nuclear group G and let • be an a d m i s s i b l e topology on G • Then
there exists on B o r e l subsets of G a unique spectral measure P
T
such that

(i) Cg = f ~ exp [2~ix(g)]dP(x) (g e G).

G
For the d e f i n i t i o n of the spectral measure, see [18], app. B. 3.

1~m~f. In v i e w of (1.21), we m a y assume that ¢ is the Hilbert

sum of s o m e continuous cyclic unitary representations ¢i' i ~ I.
Next, according to (13.2), we may assume that, for each i ~ I, there
A

exists a regular finite Borei measure ~i on G such that ¢i is

the representation in the space L{(G-'~i)'cT given b y the formula

(~i(g)f)(x) = f(x) exp [2~ix(g)]

 127

A ; 2G
for g ~ G, X 6 G and f 6 L (G ,~i ). Let H =~ LC( ~,~i ).
i~I
For e a c h A ~ B(G~), we d e f i n e a projection P(A) : H ~ H by

P(A)(fi)i~ I = (×Afi)i~i

where fi E L ~ ( G ~ , ~ i ) for e v e r y i and XA denotes the c h a r a c t e r i s -

tic f u n c t i o n of A. It is not d i f f i c u l t to v e r i f y that P is a s p e c -
tral m e a s u r e on H, satisfying (i).
To p r o v e the u n i q u e n e s s of P, suppose t h a t we are given two spec-
tral measures P1 and P2 defined on B o r e l subsets of G such that

Cg = 3 " exp [2~ix(g)]dPi(x) (g E G, i = 1,2,).

G

For each h E H, the m a p p i n g

B (G~) B A + P~(A) : = (Pi(A)h,h) (i = 1,2),

is a r e g u l a r finite Borel measure on G~, and

(¢gh,h) = f exp [2~ix(g)]dP~ (g ~ G, i = 1,2).

G

Hence, in v i e w of the u n i q u e n e s s of the m e a s u r e in (12.1), we have

P n
= P2" Since h was arbitrary, it f o l l o w s that P1 = P2" "

We s h a l l n o w d e a l w i t h the p r o b l e m of the e x t e n d i n g of c o n t i n u o u s
p.d. functions and unitary representations.

q13.4) Z n ~ A . Let H be a s u b g r o u p of a n u c l e a r vector group F.

Then each continuous unitary representation of H c a n be e x t e n d e d to
a continuous unitary representation of F.

Proof° Let ¢ be a c o n t i n u o u s unitary representation of H. Ac-

cording to (1.21), we m a y w r i t e ¢ as the H i l b e r t sum of some con-
tinuous cyclic unitary representations ¢ . Suppose that, for every
~, we h a v e found a continuous unitary representation ~ of F with
~i H = • . Then ~ = ~9 is a c o n t i n u o u s unitary representation of

F with ~ ] H = ¢" Therefore, ¢ m a y be a s s u m e d to be a c y c l i c repre-

sentation. Then, in v i r t u e of of (13.2), we may assume that there

exists on Hp a regular finite Borel measure ~ such that ¢ is the
 128

2
r e p r e s e n t a t i o n in the space Lc(H ,p), given by the formula

(~hf)(x) = f(x) exp [2gix(h)]

for h e H, X E H and f • L (H-,p).

For each n = 1,2,..., there is a compact e q u i c o n t i n u o u s subset

Qn of Hp with p(H \ Qn ) < i/n. We may assume tkaC Q n C Qn+l" Let

L n2 be the subspace of L~(H- ,p) c o n s i s t i n g of f u n c t i o n s w i t h support

2 ^
in Qn \ Qn-l" Then Lc(H ,p) is the H i l b e r t sum of the invariant

subspaces L2 and ~ is the H i l b e r t sum of the c o r r e s p o n d i n g repre-

n
sentations ~n" As before, we have to show only that e v e r y ~n can be
e x t e n d e d to a c o n t i n u o u s u n i t a r y r e p r e s e n t a t i o n ~n of F. There-
fore, we m a y simply assume that there is a compact e q u i c o n t i n u o u s sub-
set Q of Hp with p(H \ Q) = 0. Then there is some U • No(F)
with Q • (U 0 H)~. As in the proof of (12.4), we can find a linear
subspace N of F and three p r e - H i l b e r t seminorms p,q,r of N,' such
that Bp No(F)

(i) lim d k ( B p , B q) = 0,
k~

(2) 5 5: k d k ( B q , B r) < i,
k=l

and B r c U. We may assume that N = F. Indeed, if ~' is a contin-

uous u n i t a r y r e p r e s e n t a t i o n of N with ~' = # on H N N, then the
formula

~"(h + u) = ¢(h)~'(u) (h ~ H, u ~ N)

defined a unitary representation ~" of H+N. Since the group of

unitary operators is divisible, we can extend ~" to a u n i t a r y rep-
resentation ~ of F (cf. (1.6)). We have ~IH = ~"IH = ~. More-

over, ~IN = ~"IN = ~'. Since ~' is c o n t i n u o u s and N is an open

s u b g r o u p of F, it follows that ~ is continuous.

C o n s i d e r the c a n o n i c a l d i a g r a m
 129

id id
F >F >F

Apq > >

Fp Fq Fr

Let B be the c l o s e d u n i t b a l l of (Fq)*. It f o l l o w s from (i) that

A is a c o m p a c t operator, which implies that F is s e p a r a b l e . Thus
Pq q
B is a P o l i s h space.
Consider the r e l a t i o n

= = {(x,f) • Q × B : pf~q(h) = x(h) for all h E H}.

We s h a l l p r o v e that E is a m u l t i f u n c t i o n from Q to B (for the ter-

minology concerning multifunctions, cf. [40]). So, t a k e a n y X •Q. The
formula

x ' ( h + u) = x(h) (h • H, u • r-l(0))

defines a continuous character X' of the g r o u p H' = H + r-l(0). We

have Ix'(H' N Br) I ~ 1/4. By (2) and (8.1), there exists some f' • F#
with Pf'IH, = X' and sup {If'(u)l : u • Bq} ~ i. Since X' ~ 0 on

r-l(0), it f o l l o w s that f' £ 0 on q-l(0). So, there is s o m e f •B

with f' = f~q. Then (x,f) • E.

It is o b v i o u s t h a t the g r a p h of ~ is c l o s e d . Hence, by Aumann s

theorem on m e a s u r a b l e selectors (see [40], T h e o r e m 5.2), there exists
a Borel mapping ~ : Q ~ B s u c h that p~(X)~ql H = X for ~-almost
all × • Q. Consider the c o n t i n u o u s mapping ~ : B ~ Fp given by
the f o r m u l a ~(f) = pf~q for f • B. The set 8(B) is e q u i c o n t i n u o u s
A

because, evidently, 8(B) c (41 Bq)~. Next, let a : Hp ~ F P be the

Borel mapping g i v e n b y the f o r m u l a

[~(X) if X • Q
o(X)
if X ~ Q.
 130

The image of a is e q u i c o n t i n u o u s because a(H ) = a(Q) c 8(B). There-

fore the Fourier transform of the Borel m e a s u r e Pa is continuous (cf.
the proof of (ii.i)). Let ~' be the u n i t a r y representation of F in
the space Li(F- ,pa) given by the formula

(~f)(K) = f(<) exp [2~iK(u)]

for u • F, f • L~(F- ,pa) and K • F . Due to (13.1), the function

fo a 1 is a cyclic vector of ~'. We have

(~fo,fo) = J- exp [2~iK(u)]dpa(K) = ~a(U)

F

for all u E F, w h i c h means that ~' is continuous (see (1.23)).

For almost all X • Q, we have

a(X)IH = (6a)(X) [H = P~(X)~q[ H = X,

which implies that

^
(3) o(X) IH = X for almost all × • H .

2
Hence, for each f • L c(H-,p), we have

f If(<iH)12dpa(~) = f If(a(X)iH)12dp(x) = jc_if(x)12dp(x).

F H H

This means that the formula

(Ff)(<) = f(KIH) (K • F )

defines an isometric operator ~. : L2(H -,p) ~ L2(F-,po). On the other

2
hand, for each f E Lc(F ,pa) , we have

f If(o(X))12d~(x) = f If(<)12dpa(<),
H F

which means that tlhe formula

(Af) (~) = f(a(X)) (X • H )

defines an isometric operator & : L~(F-,po) ~ L~(H-,~).

Take any f • L~(H-,p). According to (3), we have

 131

AFf(x) = (Ff)(o(X)) = f(o(X)iH) = f(x)

for almost all X ~ H . Thus ~F = id, which implies that F is an

isometry of Lc(H ,~) onto L (F-,~o). Then T : = F-1T'F is a con-

tinuous unitary representation of F in the space L~(H-,~).

2 -
Take any h E H and f s Lc(H ,p). In view of (3), for almost
all X ~ H , we have

(Thf)(x) = (AT~Ff)(x) = (~'hFf)(o(X))

= (Ff)(o(X)) exp [2~io(x)(h)] = f(o(X)iH) exp [2~ix(h)]

= f(x exp [2~ix(h)] = (¢hf)(x).

This proves that ~ I H = ¢. •

~13.5) ~ . Let A be a subgroup of a nuclear group G. Then

each continuous unitary representation of A c a n be e x t e n d e d to a c o n -
tinuous unitary representation of G.

~ . Let ¢ be a continuous unitary representation of A. In

virtue of (9.6), we may assume that there exist a nuclear vector group
F, a subgroup H of F and a closed subgroup K of H, such that
G = H/K. Let ~ : H + H/K be the canonical projection. Then ¢4 is

a continuous unitary representation of ¢-I(A). By (13.4), we can ex-

tend it t o a c o n t i n u o u s unitary representation ~' of H. Since ~'
is t r i v i a l on K, it i n d u c e s a continuous unitary representation T of
H/K with ~ = ~'IH. It is o b v i o u s that T I A = ¢. •

q13.6) ~ . Let A be a subgroup of a n u c l e a r group G. Then

each continuous p.d. function on A c a n be e x t e n d e d to a continuous
p.d. function on G.

P~f. Let % be a continuous p.d. function on A. There exists

a continuous cyclic unitary representation ¢ of A with a cyclic
vector u, such that (¢gU,U) = ¢(g) for all g E A (see (1.24)). B y
(13.5), we can extend ¢ to a continuous unitary representation
of G. Then the function G 3 g + (TgU,U) is a c o n t i n u o u s p.d. ex-
tension of ¢ (see (1.23)). •

~13.7; JK;I~. The material of t h i s section is new.

 Chapter 5
P(]~rI~RYAGIN DUALI~.f

Let K be an a d d i t i v e s u b g r o u p of a real locally c o n v e x space E.

Let E* be the dual space and K* the s u b g r o u p of E* c o n s i s t i n g of
f u n c t i o n a l s w h i c h assume integer values at points b e l o n g i n g to K. We
saw in section 8 that closed subgroups of n u c l e a r spaces are weakly
closed. Thus, if K is a closed s u b g r o u p of a nuclear Frechet space,
we m a y write K** = K. The e q u a l i t y K** = K for closed subgroups of
Rn is the heart of the P o n t r y a g i n d u a l i t y for LCA groups. These facts
were the s t a r t i n g point for the i n v e s t i g a t i o n of d u a l i t y p r o p e r t i e s of
additive s u b g r o u p s and q u o t i e n t s of nuclear spaces.
S e c t i o n 14 of this chapter c o n t a i n s the basic facts on duality
for subgroups, q u o t i e n t groups, p r o d u c t s and d i r e c t sums of abelian
t o p o l o g i o a l groups. In section 15 we give an account of known results
on the P o n t r y a g i n d u a l i t y for locally c o n v e x spaces. In section 16 we
prove that c h a r a c t e r groups of m e t r i z a b l e nuclear (vector) groups are
nuclear. S e c t i o n 17 c o n t a i n s the proof of the m a i n result of this
chapter: if G is a c o u n t a b l e product of LCA groups and metrizable,
c o m p l e t e nuclear groups, then the d u a l i t y b e t w e e n G and G- induces
d u a l i t i e s b e t w e e n a p p r o p r i a t e closed subgroups and q u o t i e n t s of G and
G-. We also give here some e x a m p l e s w h i c h show that the assumptions of
c o u n t a b i l i t y and m e t r i z a b i l i t y are essential. Finally, section 18 is
d e v o t e d to V i l e n k i n ' s t h e o r y of groups w i t h boundedness.
Let H be a s u b g r o u p of an abelian t o p o l o g i c a l group G. To sim-
p l i f y the notation, in this chapter we shall assume that the groups G-
and H° are e n d o w e d w i t h the c o m p a c t - o p e n topology. In some places,
however, w h e r e the d i f f e r e n c e b e t w e e n various t o p o l o g i e s on G- is
essential, we shall apply the n o t a t i o n Gc,Gpc etc.

14. P r e l i m i n a r i e s

Let G be an abelian t o p o l o g i c a l group. The e v a l u a t i o n map from

G into G-- will be d e n o t e d by ~G" Thus <~G(g) > = <x,g> for g ~ G,
X ~ G-. Obviously, ~G is i n j e c t i v e if and only if G admits suf-
f i c i e n t l y m a n y c o n t i n u o u s characters. We say that G is a r e f l e x i v e
group if eG is a t o p o l o g i c a l i s o m o r p h i s m of G onto G--
 133

(14.1) ~mm~a. Let A be an a r b i t r a r y subset of an a b e l i a n topolo-

gical group G and let Q be the q u a s i - c o n v e x hull of A. Then eQ(Q)
= ~G(G) N A °°.

PmsQf. The inclusion eG(Q) c eG(G) A A °° is t r i v i a l . To prove

the o p p o s i t e one, choose an a r b i t r a r y ~ ~ ~G(G) A A °°. We h a v e ~ =

eG(g) for a certain g ~ G and it r e m a i n s to s h o w that g ~ Q. Sup-

pose the c o n t r a r y ; then Ix(g)l > ¼ for a certain X • A°. Hence

I~(X)I = Ix(g) l > ¼, which is i m p o s s i b l e because ~ • A °°. •

~14.2~. Let G be an a b e l i a n topological group with ~G(G)=

G-- For each dually closed subgroup H of G, one has ~G(H)= H °°.
Each dually closed subgroup of G is the annihilitor of some closed
subgroup of G.

P~f. The first assertion is a d i r e c t consequence of (14.1) be-

cause dually closed subgroups are q u a s i - c o n v e x sets. To p r o v e the sec-
ond one, take a dually closed subgroup A of G . The set

B = {g • G : x(g) = 0 for all X • A}

is a c l o s e d subgroup of G. It is o b v i o u s that A c B °. So, it re-

mains to s h o w that B ° c A.
Suppose the c o n t r a r y and let < • B ° \ A. Since A is dually
closed in G-, we h a v e I~(K)I > ¼ for a c e r t a i n ~ • A °. Next, we
have ~ = ~G(g) for a certain g • G because ~G(G) = G Since
• A °, it f o l l o w s that x(g) = ~(X) = 0 for all × e A, whence

g ~ B. But K • B° and, therefore, <(g) = 0, which is impossible

because K(g) = ~(K). "

(14.3) ~.,~mA. Let G be a locally quasi-convex group and let G

be the image of ~G endowed with the topology induced from G Then
the m a p p i n g ~G : G + G" is open.

P~f. Choose an a r b i t r a r y U E No(G ) . We have to show that

eG(U) e No(G" ). We m a y assume U to be q u a s i - c o n v e x . Then, by (14.1),

we h a v e eG(U) = G" A U °° and it r e m a i n s to o b s e r v e that U °° • N o ( G - - )

because U° is a c o m p a c t subset of G- (see (1.5)). •

(14_4) ~ . If an a b e l i a n topological group G is a Q-space,

then eG is c o n t i n u o u s .
 134

This follows immediately from the fact that, due to the A s c o l i

theorem, compact subsets of G are e q u i c o n t i n u o u s .

(14,5) ~ . Let X,Y be topological spaces. A mapping f : X~Y

is c a l l e d k-continuous if all its restrictions to c o m p a c t subsets of
X are continuous. A topological group G is c a l l e d a k-group if all
k-continuous homomorphisms of G into topological groups are continu-
ous. This notion was introduced b y N. N o b l e [721. Naturally, if G is
a k-space, it is a l s o a k-group, but there exist k-groups which are n o t
k-spaces. Noble proved that if an a b e l i a n topological group G is a k -
-group, then aG is c o n t i n u o u s ([721, Theorem 2.3).

For our p u r p o s e s , we get little benefit from the n o t i o n of a k -

-group. An abeiian topological group G will be c a l l e d a c-group if
aG is c o n t i n u o u s . It t u r n s out that the p e r m a n e n c e properties of k-
-groups, established in [72], remain valid for c - g r o u p s . Consequently,
if we are able to p r o v e that a given topological group is a k-group,
t h e n we are also in a p o s i t i o n to p r o v e that it is a c - g r o u p , without
referring to the n o t i o n of a k-group. Besides, there exist c-groups
which are not k - g r o u p s , e.g. uncountable direct sums of r e a l lines.

(14.6J r ~ . Let H be a d u a l l y closed, dually embedded subgroup

of an a b e l i a n topological group G. If gG(G) = G , then aH(H)=H

~f. Take an a r b i t r a r y ~ E H and consider the c a n o n i c a l ho-

momorphism ~ : G- ~ H-. It is o b v i o u s that ~ ~ H °°. By (14.2), we
have aG(H) = H °O. Consequently, there is some h ~ H with a G(h) = ~#
and it r e m a i n s to s h o w that aH(h) = ~. For each X a G , one has

<~H(h),hx> = <aH(h),¢(X)> = <~H(h),XIH> = <x,h>

= <~G(h),x> = <~¢,X>.

This means that ~H(h)¢ = ~¢. Hence aH(h) = ~ because ¢ maps G

onto H- (H was assumed to be d u a l l y embedded in G). •

~14.v)~. Let H be a c l o s e d subgroup of an a b e l i a n topolo-

gical group G. If aG is c o n t i n u o u s , so is aG/H"

~ . Let ~ : G + G/H be the canonical projection. A direct

verification shows that the d i a g r a m
 135

~G
G > G

~G/H
G/H > (G/H)

is commutative. •

Let H be a c l o s e d s u b g r o u p of an a b e l i a n t o p o l o g i c a l group G.
The c a n o n i c a l h o m o m o r p h i s m s G - / H ° ~ H- and (G/H)- + H °, d e f i n e d in
the obvious way, will be d e n o t e d by %H and CH, respectively. Ob-
serve that ¢H is a c o n t i n u o u s injection; it is a s u r j e c t i o n if and
only if H is d u a l l y e m b e d d e d in G. The m a p p i n g ¢H is a c o n t i n u o u s
i s o m o r p h i s m of (G/H)- onto H° .

~14.82 ~:~a. Let H be a d u a l l y c l o s e d and d u a l l y e m b e d d e d sub-

g r o u p of a H a u s d o r f f locally quasi-convex group G. If eG(G) = G
and the g r o u p G-/H ° is locally q u a s i - c o n v e x , then #H : G~/H° ~ H^ is
a t o p o l o g i c a l isomorphism.

P~x~. The H a u s d o r f f l o c a l l y q u a s i - c o n v e x group G admits suffi-

c i e n t l y m a n y c o n t i n u o u s characters, w h i c h means that ~G is injective.

From (14.3) it follows that eG is open, i.e. (x~l : G-- + G is con-

tinuous. By (14.2), we have ~G(H) = H °°. Therefore, the h o m o m o r p h i s m

: H °° ~ H, d e f i n e d as the r e s t r i c t i o n of e~l to H °°, is contin-

uous. Let Q = G-/H ° and let y : H- ~ Q be the h o m o m o r p h i s m dual
~H ° - ~H O
to : Q + H. Since ~ and are b o t h surjective, y is
injective. A direct v e r i f i c a t i o n shows that Y#H = eQ. Let Q~ = ~Q(Q);
since H is d u a l l y embedded, #H maps Q onto H , so that we ob-
tain the f o l l o w i n g c o m m u t a t i v e d i a g r a m

~Q
Q > Q

H
 136

where ¥ is a c o n t i n u o u s isomorphism. To c o m p l e t e the proof, it re-

mains to o b s e r v e that, due t o (14.3), the m a p p i n g ~Q : Q ~ Q" is
open. •

Let f be a c o n t i n u o u s mapping of a t o p o l o g i c a l space X onto a

topological space Y. We s a y that f is c o m p a c t - c o v e r i n g if to e a c h
compact subset A of Y there corresponds a compact subset B of X
such that A c f(B).

(~4.g) P]~u~s][Ti~. An o p e n m a p p i n g of a C e c h - c o m p l e t e s p a c e X on-

to a t o p o l o g i c a l space Y is c o m p a c t - c o v e r i n g .

This fact w a s p r o v e d b y A r k h a n g e l s k i [ [2].

(14.10)~~. Let H be a c l o s e d subgroup of a Cech-complete

abelian group G. Then CH : (G/H)- ~ H ° is a topological isomor-
phism. •

(14.11) PRmvOSlTIOI. Let {Gi}ie I be a f a m i l y of a b e l i a n topologi-

cal groups. There are c a n o n i c a l topological isomorphisms

( 1-[ Gi)^ -- Z *G? and ( Z * Gi)" -- II Gi.

iEI igI 1 ieI igI

If all g r o u p s G. are r e f l e x i v e , so are I-[ G. and Z *G,.

1 ieI i ieI I

Pz~=f. To s i m p l i f y the n o t a t i o n , we s h a l l w r i t e II i and Zi in-

s t e a d of I-[ and Z 4, respectively. Let X a {IIiG i) For e a c h

iEI iEI
i ~ I, let Xi denote the r e s t r i c t i o n of X to Gi (we i d e n t i f y Gi
w i t h the c o r r e s p o n d i n g subgroup of IIiGi). It is c l e a r that Xi = 0
for all but f i n i t e l y m a n y indices i. Consider the c a n o n i c a l mapping

: (IIiG i) ~ }qiGi

given by #(X) = (Xi)ie I • It is c l e a r that # is an a l g e b r a i c iso-

morphism. To p r o v e that it is c o n t i n u o u s , choose any U E No(ZiGi). We
may assume that U = ZiUi where U i ~ No(G ) for every i. For
each i ~ I, we can f i n d some c o m p a c t subset Ki of Gi such that
K~I + K~l c U..I Then K : IIiK i is a c o m p a c t subset of IIiG i and it
 137

is e n o u g h to show that @(K °) c U.

So, take any X g K °. We m a y assume that all sets Ki are sym-

metric. Then it is not h a r d to see that

1
~.ilx(Ki)] = Ix(K) I _-< ~.

Write c.l = ']x(Ki~]

' - for e v e r y i. From the c o n d i t i o n K9 + K~ c U it
l 1
follows easily that

Ir 1 7 -1
xi/U i ~ ~[4--~iJ

where [x] is the i n t e g e r part of x. Since Zici ~ ¼, after easy

calculations we get Z i ( X i / U i) < i, which means that ¢(X) =

(Xi)ig I e Z i U i = U.

To p r o v e that ~ is open, choose any W E No((~iGi) ). We m a y

assume that W = K° for some c o m p a c t subset K of ~.G.. Next, we
ll
assume that K = ~iKi where Ki is a c o m p a c t subset of Gi for
*O *
every i. Then K~l ~ No(Gi) for e v e r y i, so that XiKi g No(ZiG ~)
and it r e m a i n s to s h o w that Z~K9 c $(W).
ii
" *O
Take any X g (~iGi) with ¢(X) = (Xi)i~ I g Z i K i . We have to
1
prove that ]X(K)] £ ~. It is not h a r d to v e r i f y that xi/K ~ ~ 4[xi(K i I
for e v e r y i. Then

I~ o 1
Ix(K) 1 <-- >qilXi(Ki )I < 4 i(xi/ni ) < ~"

The p r o o f that ( ZiGi ) is c a n o n i c a l l y topologically isomorphic

to H iG 2 is similar, and even a little bit simpler (use (1.17)).
The last a s s e r t i o n of (14.11) is a c o n s e q u e n c e of the f o l l o w i n g
canonical isomorphisms:

(IIiGi) -- (SXiGi) -- (YIiG i ) -- IIiGi,

( ~ iGi
* ) -- ( IIiG ~) -- ( Z i G i ) -- Z iGi . •

(14.12) P ~ S ~ I ~ . Let G = Z (G i : Hi) be a r e d u c e d p r o d u c t of

iEI
abelian topological groups. S u p p o s e that H i is d u a l l y c l o s e d in Gi
for almost all i. Then G is c a n o n i c a l l y topologically isomorphic
 138

to the reduced product ~q (Gi : H°). Consequently, if all groups Gi

i~I
are reflexive, then G is reflexive, too.

We omit the proof because it is very similar to that of (18.2) be-

low. For details, we refer the reader to [60]. See also [38], (23.33).

Let G = Z (G i : H i ) be a reduced product of abelian topologi-

iEI
cal groups. Let H be a subgroup of G consisting of all sequences
(gi) such that gi ~ Hi for all i. Then H may be identified with
the product II H i with the usual product topology and G/H may be
iEI
identified with the direct sum Z (Gi/Hi). We may also identify H°
iEI
with the subgroup II H~ of Z (Gi : H~). It is clear that H is
i~I i i~I i
dually closed (rasp. dually embedded) in G if and only if Hi is
dually closed (rasp. d u a l l y embedded) in Gi for every i. Similarly,
G/H is locally q u a s i - c o n v e x if and only if Gi/H i is locally quasi-
-convex for every i. The m a p p i n g SH : G-/H° ~ H (rasp. cH:(G/H)- ~
H °) is a topological isomorphism if and only if SH. : Gi/H~ ~ Hi
1
Hi
(rasp. ~ : (Gi/H i) ~ H~) is a topological i s o m o r p h i s m for every i.
Detailed proofs of these facts can be found in [60].

(14.13) IOl~S. The material of this section is taken mainly from

[8]. P r o p o s i t i o n (14.11) is known as Kaplan's d u a l i t y theorem and was
proved by Kaplan [49]. For countable products, it was o b t a i n e d inde-
p e n d e n t l y by V i l e n k i n (see [97], n ° 4). P r o p o s i t i o n (14.12) is taken
from [60]. It had been obtained earlier by V i l e n k i n [97], Theorem 7. In
the case when Gi are LCA groups and Hi are open compact subgroups,
(14.12) was proved i n d e p e n d e n t l y by Braconnier [25] and V i l e n k i n [96].

15. Locally convex vector groups

(15.1) ~l~I(m. A locally convex space E satisfies the condi-

tion ~E(E) = E if and only if closed convex hulls of compact sub-
sets of E are w e a k l y compact.

Pmm~f. By (2.3), we have a topological isomorphism PE : Ec ~ E

and an algebraical isomorphism PE* : (Ec) ~ (Ec) . Let ~ : E ~ (E c)

c
be the canonical embedding. We have the following c o m m u t a t i v e diagram:
 139

id
E > E

- PE*
PE * - c * *
E > (Ec) < (Ec)

~ A

SO, the c o n d i t i o n aE(E) = E is e q u i v a l e n t to

(i) 8(E) = (E c) .

Suppose first that closed convex hulls of c o m p a c t subsets of E

are w e a k l y compact. We h a v e to p r o v e (i). Let h E (Ec) There is a
symmetric compact subset A of E such that, by denoting

A0 = {f g E : If(u) l ~ 1 for all u g A},

one has

(2) lh(f)l ~ i for all f E A 0.

Let m be the f a m i l y of all finite dimensional subspaces of E.

Fix an a r b i t r a r y M ~ m. It is c l e a r that we c a n find some vE E with

(3) f(v) = h(f) for all f ~ M.

Let us d e n o t e

M 1 = {u a E : f(u) = 0 for all f ~ M}.

We shall prove that

(4) v E M ± + cony A.

Suppose the c o n t r a r y . Since M± is w e a k l y closed and, by our as-

sumption, conv A is w e a k l y compact, it f o l l o w s that M I +conv A is
weakly closed in E. So, b y the H a h n - B a n a c h theorem, there is some

fo E* with Ifo(V) I > 1 and

(5) Ifo(U) I ~ 1 for all u e M ± + conY A.

From (5) it f o l l o w s that fo E M N A 0. Hence, by (3) and (2), we ob-

 140

tain Jfo(V) l = Jh(fo) I ~ i, which is a c o n t r a d i c t i o n .

Now, (4) implies that there is s o m e w •conv A with w - v ~ M ±.

Then, by (3), for each f • M, one has

f(w) = f(v) + f(w - v) = h(f).

We have thus shown that, for e a c h M • m , the set

PM : = cony A n {u e E : h(f) = f(u) for a l l f • M}

is n o n - e m p t y . Since conv A is w e a k l y compact, the intersection of

all P "s contains some vector v. Then f(v) = h(f) for all f •E,
m
i.e. ~(v) = h. This proves (i).

Conversely, suppose that (i) is s a t i s f i e d . Choose an arbitrary

compact subset A of E and denote

A 0 = {f • E* : if(u) I ~ 1 for a l l u E A},

B = {u • E : If(u) l ~ 1 for a l l f • A0}.

Naturally, B is a c l o s e d , convex subset of E containing A, so

that it r e m a i n s to s h o w that B is w e a k l y compact. Since A 0 •N O (E~),

the B a n a c h - A l a o g l u theorem implies that

A 00 : = {h E (Ec) : lh(f)l < 1 for all f e A 0}

is compact in the weak* topology on (Ec) . Now, (1) implies that

~(B) = A 00 and it r e m a i n s to o b s e r v e that ~ is a t o p o l o g i c a l iso-

morphism of the space E endowed with its weak topology onto the space

(E c) endowed with its w e a k * topology. -

(15.2;PRs~ssi~zom. Every metrizable and complete locally convex

vector space is a r e f l e x i v e group.

"Vector space" m a y be r e p l a c e d here by "vector group"; see (15.7).

Pz~af. Let E be a metrizable and complete locally convex space.

Then closed convex hulls of c o m p a c t subsets of E are c o m p a c t , hence
~ A

weakly compact, and (15.1) says that ~E(E) = E That ~E is o p e n

and continuous follows from (14.3) and (14.4). •

 141

~IS.3) ~ x ~ . Every reflexive locally convex space is a re-

flexive group.

Proof. Let E be a r e f l e x i v e locally convex space. Then closed

bounded subsets of E are weakly compact ([80], Ch. IV, Theorem 5.5),
and (15.1) implies that ~E(E) = E From (2.4) and (14.3) we infer

that aE is open. We s h a l l prove that it is continuous.

In view of (2.3), it is e n o u g h to show that c o m p a c t subsets of E
C

are e q u i c o n t i n u o u s . So, let A be a compact subset of Ec . Since E

is reflexive, it suffices to show that A is a bounded subset of Eb -
Suppose the contrary; then there is some f E (Eb) with sup { If(u)l:

u~ A} = ~ . However, since E is reflexive, we h a v e f ~ (E c ) and

then sup {If(u)l : u ~ A} < co because A was c o m p a c t in E c. The
contradiction obtained completes the proof. •

(15.4) RnL~K. Since there e x i s t non-reflexive Banach spaces, from

(15.2) and (15.3) it follows that, for l o c a l l y convex spaces, reflex-
ivity is an e s s e n t i a l l y stronger property than "group reflexivity". See
also (18.2). One more notion of r e f l e x i v i t y can be o b t a i n e d by consid-
erin g the s o - c a l l e d structure of c o n t i n u o u s convergence on t o p o l o g i c a l
groups. H.-P. Butzmann [271 p r o v e d that a locally convex space is re-
flexive in the sense of c o n t i n u o u s convergence if and only if it is
complete.

(15.5) ~ . Kye [57] p r o v e d that a locally convex space E is

a reflexive group if and o n l y if c l o s e d c o n v e x h u l l s of c o m p a c t subsets
of E are w e a k l y compact and each closed convex balanced set w h i c h is
a neighbourhood of zero in the k - t o p o l o g y is a n e i g h b o u r h o o d of zero
in the o r i g i n a l topology. The first condition is s a t i s f i e d if and only
if eE(E) = E (see (15.1)). The second one is e q u i v a l e n t to the con-

tinuity of ~E"

(15.6) TJmm~. Let E be a l o c a l l y convex space, L a vector space

endowed with the d i s c r e t e topology and M a linear s u b s p a c e of E × L.
For each u ~ (E × L) \ M, there exists some X ~ (E × L) with
XIM = 0 and X(u) # 0.

~f. Choose any u E (E × L) \ M. Let ~E and ~L be canon-

ical p r o j e c t i o n s of E × L onto E and L, respectively. Suppose
first that ¢L(U) ~ eL(M). Since L is discrete, there is some K E L-
 142

with K(~L(M)) = {0} and K(~L(U)) # 0; t h e n we m a y take X = K~ L-

Next, suppose that SL(U) • ~L(M). Then ~L(U) = ~L(W) for some
w g M. Let N = ~E(M n (E x {0})). Since u ~ M, it follows that
~E(U - w) ~ N. So, there is a c o n t i n u o u s linear functional f on E

such that fin = 0 and fSE(U - w) ~ Z. Then pf • E-, pflN = 0 and

pf~E(U - w) # 0. It is c l e a r that the formula

K(~L(V)) = pf~E(V) (v • M)

defines a character < of ~L(M). Then the formula

~(v) = pf~E(V) - KCL(V) (v e E x ~L(M))

defines a continuous character ~ of the group E × SL(M). Naturally,

we m a y extend ~ to some X • (E × L) Then XIM = ~IM = 0 and

X(U) = ~(u) = pf~E(U) - ~$L(U) = pfSE(U) - K$L(W)

= pf$E(U) pf~E(W) = pf~E(U - w) # 0. •

(15.7) r ~ . Every metrizable and complete locally convex vec-

tor g r o u p is r e f l e x i v e .

~ . Let F be a m e t r i z a b l e and complete locally convex vector

group. Let {Un}n= 1 be a b a s e at z e r o consisting of symmetric convex

sets. Fix an a r b i t r a r y n = 1,2, ... and d e n o t e M n = span Un . We c a n

find a linear subspace Ln of F such that F = Mn ~ Ln algebraic-

ally. Since Mn is an o p e n subgroup of F, it f o l l o w s that Ln is
discete and F = Mn ~ Ln topologically. Hence the canonical projec-
tions Mn : F + Mn and Ln : F ~ Ln are continuous. Let Pn be the
Minkowski functional of Un in Mn and let En be the quotient s p a c e

Mn/pnl(0) endowed with the canonical norm. Let then Sn : Mn En be

the c a n o n i c a l projection and let Fn denote the group F endowed with

the topology induced b y the homomorphism

M ~L : F ~ E × Ln
¢n = S n ~ n × n n "

co

Denote G = II (En x Ln) and consider the l i n e a r mapping ~ =

n=t
 143

(#n)n=l : F ~ G. It is clear that F m a y be identified, as a topolo-

gical group, w i t h the limit of the inverse s e q u e n c e id : F n + 1 ~ F n.

Therefore # is a t o p o l o g i c a l isomorphism between F and %(F). F r o m

(1.8), (15.2) and (14.11) we see that G is reflexive. In view of
(14.3), (14.4) and (14.6), to c o m p l e t e the proof, we only have to show
that $(F) is d u a l l y e m b e d d e d and d u a l l y c l o s e d in G.

So, take any X E ~(F) Then X~ E F and, c o n s e q u e n t l y , X~ E

Fn for some n because F = U F . Since the t o p o l o g y of Fn is

n=l
induced by #n' we can w r i t e X¢ = <#n for some < ~ (E n x Ln) Let

K ~ (En × Ln) be the n a t u r a l e x t e n s i o n of K and let On : G ~EnX Ln

be the c a n o n i c a l projection. An easy v e r i f i c a t i o n shows that KO n is

an e x t e n s i o n of X. Thus ¢(F) is d u a l l y e m b e d d e d in G.

To prove that %(F) is d u a l l y closed, choose any g ~ G \ ~(F).

Since F is complete, ¢(F) is a closed s u b g r o u p of G. T h e r e f o r e we
can find some U E No(G ) with (g + U) D ~(F) = ~. In other words,
there is a finite subset I c {1,2,...} such that if qI is the ca-
nonical p r o j e c t i o n of G onto GI : = II (En x Ln) , then oi(g) does
nEI
not b e l o n g to the closure of qi(~(F)) in G I. Hence, by (15.6),
there is some X ~ GI with X(qI(~(F))) = {0} and X(oi(g)) # 0. Then
Xq I is a c o n t i n u o u s c h a r a c t e r of G taking #(F), but not g, to
zero. •

~15.8; ~ . The above a r g u m e n t shows that if F is a complete

locally c o n v e x vector g r o u p not n e c e s s a r i l y m e t r i z a b l e , then ~F(F) =
F From (9.4), (8.5) and (14.3) it follows that ~F is open. The
f o l l o w i n g e x a m p l e shows that ~F need not be continuous (see also
(17.6)).

~15.9; ~ . Let E be an i n f i n i t e d i m e n s i o n a l n o r m e d space

e n d o w e d w i t h its w e a k topology. Clearly, there is a convex infinite
dimensional subset K of E* compact in the n o r m topology. Since
w e a k l y c o m p a c t subsets of E are bounded, K is c o m p a c t in Ec but,
b e i n g i n f i n i t e dimensional, it cannot be e q u i c o n t i n u o u s . Thus the eva-
luation map E ~ (Ec) c is not c o n t i n u o u s and, by (2.3), neither is

~E"
 144

~15.10)~f~s. Propositions (15.1) and (15.2) are taken from [57]

(cf. (15.5)). Proposition (15.1) is a n i m m e d i a t e consequence of the r e -
sult of A r e n s [i], Theorem 2, w h i c h asserts that if E is a l o c a l l y
convex space and • is a l o c a l l y convex topology on E* not weaker
than the topology of p o i n t w i s e convergence, then the evaluation map

E + (E T) is o n t o if a n d o n l y if ~ is n o t stronger than the topo-

l o g y of u n i f o r m convergence on convex weakly compact subsets of E.

Arens's paper contains a very detailed study of "reflexive" topologies

on E* and E**.

For Banach spaces, (15.2) appears in [833. Also (15.3) is t a k e n

from [833. Lemma (15.6) and Theorem (15.7) are new. Example (15.9) is
due to G.W. Mackey and was given in [i].

16. Nuclearity of d u a l groups

The a i m of t h i s section is to p r o v e the following fact:

~16.1JTaR~JR~m. Let G be a nuclear group. If G is m e t r i z a b l e ,

then G is a n u c l e a r group, too.
pc

~16.2J ~ _ The assumption of m e t r i z a b i l i t y is essential. Let E

b e an u n c o u n t a b l e product of r e a l lines. Then E is a n u c l e a r locally
convex space, hence a nuclear group. By (14.11), we may identify E-
pc
with an uncountable locally convex direct s u m of r e a l lines. As a lo-
cally convex space, an uncountable direct sum of real lines is not
nuclear. So, by virtue of (8.9), the g r o u p Epc cannot be n u c l e a r .

(16.3) r ~ . Let F be a m e t r i z a b l e nuclear vector group and let

P be a precompact subset of F. Given arbitrary c > 0 and m = i,
2,..., one can find a vector space E, two symmetric and convex sub-
sets X,Y of E with

(i) d k X,Y) ~ c k -m (k = 1 , 2 , . . . ) ,

a subgroup K of E and a homomorphism ~ : K ~ (F), such that

¢ ( K N Y) is p r e c o m p a c t and p c $(K N X).

Proof. By (9.3), one can find a base {Un}n= 1 at zero in F,

consisting of s y m m e t r i c , convex sets such that U1 m U2 D ... and

(2) dk(Un+l,Un) < c2 - 2 m n - I k -m (k,n = 1 , 2 , . . . ) .

 145

I
Let I be a set of indices. By Ro we d e n o t e the corresponding

direct s u m of r e a l lines, i.e. the subspace of the p r o d u c t RI con-

sisting of all functions f : I ~ R such that f(i) = 0 for a l m o s t
all i E I. By Z oI we d e n o t e the subgroup of RI consisting of in-
o
teger-valued functions.

Fix an a r b i t r a r y n = 1,2,... and let M n = span U n. There is a

linear subspace Ln of F such that F = Mn ~ Ln (that is, the vec-

tor space F is the algebraical direct sum of its v e c t o r subspaces M
n
and Ln; since Mn is an o p e n subset of F, it f o l l o w s that the to-
pological group F is the topological direct s u m of its subgroups M
n
and Ln). Since every abelian group is a q u o t i e n t of a free one, we

In
may identify Ln with a quotient group of Z° for some set In; let
I
~n : Zo n + Ln be the canonical projection.

There is a f i n i t e subset An of F with P c Un+ 1 + A n because

P is p r e c o m p a c t . Since F = Mn ~ Ln, we have An c Bn + Cn for some

finite subsets Bn of Mn and Cn of Ln; we m a y assume that Bn

and Cn are symmetric. Next, there is a s y m m e t r i c and finite subset

I
Dn of Zon with #n(Dn) = C n. Thus

(3) p c U n + 1 + B n + @n(Dn).

It is c l e a r t h a t we c a n find a symmetric and finite subset Bn of Mn

such that

(4) dk(COnV Bn, cony Bn) < c2 - 2 m n - I k -m (k = 1 , 2 , . . . ) .

In
Similarly, there is a f i n i t e and symmetric subset Dn of Rn such
that

(5) dk(COnV Dn, cony Dn) < c2 -2mn k -m (k = 1 , 2 , . . . ) .

Let us d e f i n e

X2n_l = Un+ 1 +conv Bn, Y2n-i = Un + c o n y Bn,

X2n = cony Dn, Y2n = conv D n.

 146

Now, take any k = 1,2, . . . . Applying (2.6) (a) and t h e n (2) and (4),
we get

(6) d2k_l(X2n_l,Y2n_l) = d 2 k _ l ( U n + 1 + c o n y Bn, U n +conv Bn)

dk(Un+l, U n +conv Bn) + dk(COnV B n, U n + c o n y B n)

£ dk(Un+l,Un) + dk(COnV Bn, c o n v Bn)

< c2 - 2 m n - I k - m + c2 - 2 m n - I k-m = c2-2mn k -m.

Hence

(7) d 2 k _ l ( X 2 n _ l , Y 2 n _ I) < c 2 - m ( 2 n - l ) ( 2 k _ i) -m

From (6) we also obtain

(8) d 2 k ( X 2 n _ l , Y 2 n _ I) ~ d 2 k _ l ( X 2 n _ l , Y 2 n _ I)

< c2 - 2 m n k - m = c 2 - m ( 2 n - l ) ( 2 k ) -m.

From (7) a n d (8) w e d e r i v e

(9) d k ( X 2 n _ l , Y 2 n _ l ) & c2 - m ( 2 n - l ) k -m (k,n = 1 , 2 , . . . ) .

L e t us n o w c o n s i d e r the v e c t o r space

E = II (M n × Ron).
n=l
I
It c o n s i s t s of s e q u e n c e s (Un, f n ) n~= 1 where un g Mn and fn g R on

for e v e r y n. Let K be the s u b g r o u p of E consisting of all se-

I
quences (u n ,f n ) n~= l such that fn E Z o n and Un+~n(fn) =Ul+~l(f I)

for e v e r y n. Next, let ~ : K ~ F be the h o m o m o r p h i s m given by the

formula ~((Un, f n)n=l) = u I + ~l(fl). Finally, set

X = {(Un,fn)n= 1 E E : u n E X2n_l and fn ~ X2n for all n} '

Y = {(Un, f n ) noo= l ~ E : u n E Y2n-i and fn E Y2n for all n}

 147

we shall prove that the E,X,Y,K and ¢ thus defined have the de-
sired properties.

Inequality (i) follows from (5), (9) and (2.7). To p r o v e the in-
clusion p c ~(K N X), take any g ~ P. By (3), to each n there
correspond some u n E Un+l, bn e Bn and d n E Dn, s u c h that g = un +

b n + ~n(dn). Then the sequence (u n + b n , d n ) n = 1 belongs to K N X,

and

co

~ ( ( u n + b n , d n ) n = I) = g.

It r e m a i n s to p r o v e that ~(K N Y) is p r e c o m p a c t . Let us take a n y

n = 1,2, . . . . The subspace M n = span Un is a b s o r b e d by U n. Since
Bn is finite, we c a n find another finite subset B~ of Mn such
that conv Bn c B~ + U n. Then

(i0) Y2n-i c 2U n + B~.

In
It is c l e a r that the set D~ = Z ° N conv Dn is finite.

Take an a r b i t r a r y g ~ ~(K N Y). we have g = @ ( ( U n , f n ) n = I) for

some sequence ( U n , f n ) n = 1 ~ K N Y. Then un Y2n-i and fn ~ Y2n N

I
n
Zo for e v e r y n. Applying the d e f i n i t i o n of ~ and then (i0), for
every n, we get

I
g Un + ~ n ( f n ) e Y 2 n - i + ~n(Y2n N Zo n) c 2U n + B~ + ~n (D "n ) .

Since g ~ ~(K N Y) was arbitrary, we d e r i v e

~(K n Y) c 2U n + B~ + #n(D~) (n = 1,2 .... ).

This proves that ~(K A Y) is p r e c o m p a c t becasue {2Un}~= 1 is a b a s e

at z e r o in F. -

~16.4J r~mm~. Let K be a subgroup of a v e c t o r space E and let

X,Y be two symmetric, convex subsets of E such that

(i) d k ( X , Y ) ~ ck -m (k = 1,2 .... )

where c > 0 and m = 5,6,... are some fixed numbers. Let # be a

 148

homomorphism of K into a t o p o l o g i c a l group G. Then there exist a

vector space E', a subgroup K" of E', two s y m m e t r i c , convex sub-
sets X',Y" of E" and a h o m o m o r p h i s m ~" : K" ~ G , such that

(2) [~(K n y ) ] O c ¢'(K" N X'),

(3) ¢'(K" N Y') c [~(K N X)] °,

(4) dk(X',Y') ~ =< CTm k -m+5 (k = 1,2,...)

where ¥m is some u n i v e r s a l constant depending on m only.

~ . Set F = span X. From (i) and (2.14) it f o l l o w s that t h e r e

are p r e - H i l b e r t seminorms p,q on F with X c Bp-< Bq c y and

d k ( B p , B q) & cc m k -m+2 (k = 1,2, • . . )

where cm is some u n i v e r s a l constant depending on m only. By (2.15),

there is a n o t h e r pre-Hilbert seminorm r on E such that

(5) d k ( B p , B r) & 1 0 2 C C m k-m+5 (k = 1,2 .... ),

(6) d k ( B r , B q) ~ 10 -2 k -3 (k = 1,2 .... ).

From (6) we get

(7) 1
5 Z kdk(Br,Bq) ~ 5.10 -2 ~ k-2 < 4"
k=l k=l

We m a y i d e n t i f y G- with a subgroup of T G. Let PG : RG ~ TG

be the c a n o n i c a l projection. Set E" = R G and K" = pGl(G-); let ¢"

be the r e s t r i c t i o n of PG to K'.

For e a c h g ~ ¢(K N Bp) , choose some Wg ~ K N B P with %(Wg) =

g. Then the f o r m u l a

if g E ¢(K N Bp)

(¢f)(g) : If0 (wg)

if g ~ %(K N Bp)

defines a linear operator ¢ : F # + R G. Let us d e n o t e

 149

1
B °r = {f • F # : If(u) I < ~ for all u • B r}

1
B p° = {f E F # : If(u)l < ~ for all u e Bp}

L = {4 • R G : ~(g) = 0 for all g g ~ ( K N Bp)}.

Set X" = ¢(B~) + L and Y" = ¢(B~) + L. Since L is a l i n e a r sub-

s p a c e of R G, we h a v e dI(L,L) = 0. Applying (2.6) (a), (2.8)(a) and

(2.16), we get

dk(X',Y') ~ d k ( ¢ ( B ~ ) , ¢ ( B ~) + L) + d I ( L , ¢ ( B ~) + L)

dk(¢(B~)'¢(B~)) + dI(L'L) < dk( B °r' B p)

= ° = dk(Bp'Br)

for k = 1,2, . . . . Hence, by (5), we o b t a i n (4) w i t h Ym = 102Cm"

To p r o v e (3), t a k e a n y ~ e K" N Y" and g • ~(K N X). We may

write ~ = Cf + ( for some f • B° and ( • L. Then
P

I¢'(~)(g)l = IPG(~)(g)I = tP(((g))l & I~(g)l

1
< l(¢f)(g)l + [~(g) l = [f(Wg)[ =< 7"

Thus ¢'(~) E (%(K D X)) ° which proves (3)

It remains to prove (2). So, take any

X • (¢(K N y))O. Let H =
1
K N F and < = X~IH. Then IK(H n Bq) I ~ 4 " By 17) a n d ( 8 . 1 ) , there
is some f • B°r with pflH = <. Then the formula

~(g) = Ix(g) if g ¢ ~(H n Bp)

L f(wg) if g E ~(H N Bp)

defines some function ~ e R G. It is e a s y to v e r i f y that PG(~) = X

and ~ - %f • L. Hence ~ ~ K" N X" and ¢'(~) = X. "

Let F be a n u c l e a r vector group. If F is m e t r i -

^

zable, then F is a n u c l e a r group.

pc
 150

PmsQf. Choose a r b i t r a r y U ~ No(F~pc )' c > 0 and m = 1,2,...,

T h e r e is a p r e c o m p a c t subset P of F with pO c U. According to
(16.3), there exist a vector space E, two s y m m e t r i c and c o n v e x sub-
sets X,Y of E with

dk(X,Y) & ck -m-5 (k = 1,2,...),

a subgroup K of E and a homomorphism ~ : K ~ F, such that

~(K N Y) is p r e c o m p a c t and p c ~(K N X). Next, by (16.4), there
exist a vector space E', a subgroup K" of E', two symmetric
c o n v e x subsets X',Y" of E" and a h o m o m o r p h i s m ~" : K" ~ F, such
that

[~(K N y)]O c ¢(K" N X'), %'(K" N Y') c [#(K N X)] °,

dk(X',Y') ~ CYmk-m (k = 1,2,...)

where Ym is a u n i v e r s a l c o n s t a n t d e p e n d i n g on m only. Now, it re-

mains to observe that

¢'(K" n Y') c [%(K N X)] ° c pO c U,

and that ~'(K" N X') e No(Fpc ) because [~(K N y)]O c ¢'(K" D X') and
~(K D Y) is precompact. •

P,~,Jf~f~16.1). Every compact subset of G is c o n t a i n e d in the

closure of some p r e c o m p a c t subset of G. T h e r e f o r e we may identify G
pc
with (G)c" So, we have to p r o v e that G- = Gc is nuclear provided
that G is complete.

A c c o r d i n g to (9.7), there are a m e t r i z a b l e and c o m p l e t e nuclear

vector g r o u p F, a closed subgroup H of F and a closed s u b g r o u p Q
of H, such that G is t o p o l o g i c a l l y i s o m o r p h i c to H/Q. From (9.4),
(8.3) and (8.6) we infer that H is d u a l l y e m b e d d e d and d u a l l y c l o s e d
in F. From (16.5) and (7.5) it follows that F^/H ° is a nuclear
group. Hence, by (8.5), it is l o c a l l y q u a s i - c o n v e x . Therefore, Dy (15.7)
and (14.8), the m a p p i n g ~H : F ^ / H ° ~ Ha is a t o p o l o g i c a l isomorpnism.
o
Thus H- is a nuclear group. A p p l y i n g (7.5), we see that QH is nuclear~
too. To c o m p l e t e the proof, it remains to o b s e r v e that, due to (14.10),
o
the m a p p i n g ~Q : (H/Q) ~ QH is a t o p o l o g i c a l isomorphism. •

(16.6) ~ . The m a t e r i a l of this s e c t i o n is new.

 151

17. Stron9 r e f l e x i v i t y

A reflexive group G is c a l l e d strongl Y r e f l e x i v e if e v e r y c l o s e d

s u b g r o u p and e v e r y H a u s d o r f f q u o t i e n t group of G and of G" is re-
flexive. S t r o n g r e f l e x i v i t y is an e s s e n t i a l l y stronger property than
reflexivity. For instance, e v e r y infinite d i m e n s i o n a l B a n a c h space is a
reflexive, but not s t r o n g l y r e f l e x i v e group ((5.3) and (15.2)).

(17.1) P ~ o P o s r l ~ . Let H be a closed s u b g r o u p of a s t r o n g l y re-

flexive group G. Then

(a) H is d u a l l y c l o s e d in G;

(b) H is d u a l l y e m b e d d e d in G;

(c) the c a n o n i c a l m a p p i n g s CH : G'/H° ~ H- and ~H : (G/H)- ~ H °

are t o p o l o g i c a l isomorphisms;

(d) H and G/H are s t r o n g l y reflexive.

Proof. (a) Since G is s t r o n g l y reflexive, G/H is reflexive. In

particular, G/H admits s u f f i c i e n t l y m a n y c o n t i n u o u s characters, which
means that H is d u a l l y c l o s e d in G.
(b) Since G is reflexive, it follows from (a) and (14.2) that
~G]H is a t o p o l o g i c a l i s o m o r p h i s m of H onto H °°. Let y : H °° ~ H
be the inverse isomorphism. Choose any × g H and c o n s i d e r the conti-
~H ° H° --
nuous i s o m o r p h i s m : (G-/H °) ~ H °°. Then XY¢ g (G-/H °) . By
Ho
assumption, G-/H ° is reflexive, so that we m a y w r i t e XY~ =e
G /H °(
for some ~ ~ G-/H °. Let ~ : G- ~ GA/H ° be the c a n o n i c a l projection.
Then ~ = ~(K) for a c e r t a i n K ~ G and a d i r e c t v e r i f i c a t i o n shows
that KIH = X.

(c) Since G is s t r o n g l y reflexive, G-/H ° is reflexive, hence

locally q u a s i - c o n v e x . It follows from (a), (b) and (14.8) that ~H is

a topological isomorphism. Similarly, ~ : G~-/H °° ~ (H°) - is a topo-

Ho
logical isomorphism. C o n s i d e r the c a n o n i c a l c o m m u t a t i v e d i a g r a m

(~G ~
G > G

1
G/H 13
;
> G--/H °°
 152

It is clear that 8 is a t o p o l o g i c a l isomorphism. A direct verifica-

tion shows that the c o m p o s i t i o n

Ho H° Ho - ^ .
> (H°) -- > (G -/H°°) - ~ > (G/H)

is equal to (#H) -I.

(d) Let K be a closed s u b g r o u p of H. Then K is a closed

s u b g r o u p of G, t h e r e f o r e it is reflexive. We shall prove that H/K
is reflexive.

C o n s i d e r the c a n o n i c a l d i a g r a m

H > G

H/K > G/K

It is clear that j is a t o p o l o g i c a l embedding. Since, by (a) and(b),

H is d u a l l y c l o s e d and d u a l l y e m b e d d e d in G, it follows e a s i l y that
j(H/K) is d u a l l y closed and d u a l l y e m b e d d e d in G/K. So, in virtue of
(14.6), we have ~H/K(H/K) = (H/K)--

Since G is s t r o n g l y reflexive, G/K is reflexive, hence locally

quasi-convex. Therefore j(H/K) is locally q u a s i - c o n v e x , too (sub-
groups of locally q u a s i - c o n v e x groups are locally quasi-convex). Con-
sequently, H/K is locally q u a s i - c o n v e x and from (14.3) and (14.7) we
infer that ~H/K is open and continuous. Thus H/K is reflexive.

Finally, let Q be a c l o s e d s u b g r o u p of G/H. That Q is re-

flexive follows from the above and the obvious fact that Q may be
i d e n t i f i e d w i t h a H a u s d o r f f q u o t i e n t of a c l o s e d s u b g r o u p of G. The
group (G/H)/Q is r e f l e x i v e because it is a q u o t i e n t of G. •

A nuclear group G is c a l l e d b i n u c l e a r if G- is n u c l e a r and ~G

maps G onto G -~ (we do not assume ~G to be continuous).

~17.2) ~ . Let G be a b i n u c l e a r group. If G is ~ e c h - c o m -

plete, then it is s t r o n g l y reflexive.

i~uaf. That G is r e f l e x i v e follows from (8.5), (14.3) and (14.4)

because C e c h - c o m p l e t e space are k-spaces. Ew,~ry c l o s e d s u b g r o u p of G
 153

and of G is d u a l l y c l o s e d and d u a l l y e m b e d d e d due to (8.6) and (8.3),

respectively. From (7.5) and (8.5) it follows that Hausdorff quotient
groups of G and of G- are locally quasi-convex.
Let H be an a r b i t r a r y c l o s e d s u b g r o u p of G. in v i e w of (14.2),
we only need to prove that the four groups H, G/H, H ° and G-/H ° are
reflexive.

As in the proof of (17.1) (c), we see that the c a n o n i c a l m a p p i n g

8 : G/H ~ G--/H °° is a t o p o l o g i c a l isomorphism. The group G--/H °° is
locally quasi-convex, being topologically isomorphic to the locally
q u a s i - c o n v e x group G/H. So, by (14.8), the m a p p i n g

¢ : G - ' / H OO ~ (H°) -
Ho

is a t o p o l o g i c a l isomorphism. Since G is Cech-complete, (14.10)

plies that cH : (G/H)- + H ° is a t o p o l o g i c a l isomorphism, too. It is
not hard to v e r i f y that the d i a g r a m

¢
G--/HOO H° > (H°) -

GIH aGIH > (G/H)-"

is commutative. Thus ~G/H is a t o p o l o g i c a l i s o m o r p h i s m of G/H onto

(G/H)--, i.e. G/H is reflexive. Then (G/H)- is reflexive, too, and

H° is r e f l e x i v e b e c a u s e cH is a t o p o l o g i c a l isomorphism.

The r e f l e x i v i t y of H follows from (14.6), (14.3) and (14.4). Con-

sequently, H- is reflexive, and G-/H ° is r e f l e x i v e becasue CH :

G-/H O ~ H ^ is a t o p o l o g i c a l i s o m o r p h i s m due to (14.8). •

(17.3) c ~ R O n r ~ . Let G be a m e t r i z a b l e and complete nuclear group

and let (An)n= 1 be a s e q u e n c e of LCA groups. Then the product

G × II A n is a s t r o n g l y r e f l e x i v e group.
n=l

~ . Naturally, P : = G × II A is Cech-complete. We have to

n=l n
show that P is binuclear. That P is nuclear follows from (7.10) and
 154

(7.6). The nuclearity of P is a consequence of (14.11), (16.1),

(1.8) and (7.8):

P = (G × II An) -- G ~ Z A-.
n=l n=l n

In view of (1.8) and (14.11), it s u f f i c e s to p r o v e that G is r e f l e x -

ive.
According to (9.7), we can find a metrizable and complete nuclear
vector group F, a closed subgroup H of F and a closed subgroup Q
of H, such that G -- H/Q. From (15.7) and (16.5) it f o l l o w s that
F is a b i n u c l e a r group. Hence, by (17.2), it is strongly reflexive.
Finally, (17.1) (d) implies that H/Q is r e f l e x i v e . •

We shall now give some examples of g r o u p s which are not strongly

reflexive; they show that (17.3) cannot be much generalized. Denote by

R~ the countable product of r e a l lines. We begin with the following

technical result:

£17.4) ; ~ . Let M be a dense linear subspace of a metrizable

locally_ c o n v e x space E. Then the identity mappings Epc ~ E c ~ Mpc
Mc are all topological isomorphisms.

Pmmmf. It is e n o u g h to s h o w t h a t id : M- + E- is continuous.
c pc
Let X b e an a r b i t r a r y precompact subset of E. We have to find a

compact subset Y of M with y O c X °. Naturally, there is a pre-

compact subset A of M with X c A. Next, there is a n u l l sequence

(Un)n= 1 in M with A c conv {Un}n= 1 (see [80], p. 151). Obviously,

the set

Y = {tu n : -i < t < i; n = 1,2 .... }

is c o m p a c t . Let X e yO. By (2.3), we have X = Pf for some f ~ E*.

1
From (2.2) it f o l l o w s that f(u) = X(U) for u ~ Y. Thus If(u)I ~

for a l l u ~ Y and, consequently, for all u ~ c o n v Y. Since X c

1
c o n v Y, for each u ~ X, we have IX(u) l = IPf(u)l ~ If(u) l ~ 4'
which means that X E X °. •

~II.5) J~mr. Let E be a locally convex space and let E be

T

the dual group (or, w h i c h is t h e same, the dual space) e n d o w e d with t h e

 155

topology of u n i f o r m convergence on compact convex sets. Then id:E

T
Ec need not be continuous. For instance, let E be the subspace of

R~ consisting of finite sequences. It is n o t h a r d to s e e that every

compact convex subset of E is f i n i t e dimensional. Therefore E may
T

identified with E, whereas Ec is simply the d i r e c t sum of real

lines. Notice that the evaluation map is a t o p o l o g i c a l isomorphism of

E onto (ET)T.

q17.~) ~ . O.G. Smolyanov [84] proved that the space D = ~(R)

of t e s t functions on the real line contains a closed linear subspace L
such that N/L is topologically isomorphic to a n o n - c l o s e d dense sub-
space M of R ~. By (17.4), we have MA -- (R~) -, and hence M -~ --

(R~) -~ -- R ~, so t h a t M is n o t a reflexive group. ConseqUently, ~ is

not strongly reflexive. Observe that D is a r e f l e x i v e nuclear space,
and the dual space D" is n u c l e a r , too. Thus ~ is a r e f l e x i v e binu-
clear group ((7.4) and (15.3)). This example shows that the assump-
t i o n of the metrizability of G in (17.3) is e s s e n t i a l .

Let N = L0 be the annihilator of L in 0". We may write

M -- N / L -- N " / L 00 -- (L0) " = N" -- N-.

Since M is n o t a reflexive group, it f o l l o w s that ~N is not c o n t i n -

uous (cf. (15.8)).

The space D(R) may be replaced here by the space N(~) of t e s t

functions on a n a r b i t r a r y open subset ~ of Rn because the latter

one has a quotient topologically isomorphic to N(R) (see [90] or [85]).

Other examples of s p a c e s with q u o t i e n t s isomorphic to non-closed dense

subspaces of Re are given in [91], Theorem 2 and [92], Theorem 7.

~17.7) ~ . Let ~R be the countable direct s u m of real lines.

We shall prove that the group E = ~R × R ~ is n o t strongly reflexive.
Elements of E may be identified with left-hand finite sequences of
real numbers. The family of s e t s of t h e form

{(Xn) E E : IXnl < En for n ~ no},

where n o = 1,2,... and (~n)n~n is a s e q u e n c e of p o s i t i v e numbers,

O
is a b a s e at zero in E. By (14.11), the dual group E^ may be iden-

tified with the space R ~ × ~R of r i g h t - h a n d finite sequences. Thus E

 156

is a r e f l e x i v e binuclear group.
For each n ~ Z, let e E E be the sequence with 1 in the
n
n-th place and 0 elsewhere. Let en, n E Z, be the c o n t i n u o u s lin-

ear functional on E given by e n ( e m) = 6mn. We s h a l l identify the dual

space E* with R ~ x ~R in the usual way.

Let 9 be some fixed irrational number. For each m= 1,2,..., set

U m = e - m + em' Um = (i - %)-l(e -m - % e m ) ,

w m = %e_m + em, wm (i - % ) - l ( e m e_m).

co
Let K be the subgroup of E generated b y the set H = {Co} U { U m } m = 1

O {Wm}m= 1 and let K* be the subgroup of E* generated by the set

W * co *
H* = {Co} U {Um}m= 1 U { W m } m = I. A direct calculation shows that

K = {u ~ E : f(u) e Z for all f ~ K*},

therefore K is a (weakly) closed subgroup of E. Observe that the

linear mapping ¢ : E ~ E* defined b y the conditions

(1 - ~)-le_m when m < 0

~em = o when m = 0

(i %)-le_m when m > 0

is a t o p o l o g i c a l isomorphism which carries K over onto K*. Conse-

quently, the m a p p i n g pEte : E + E is a t o p o l o g i c a l isomorphism which

carries K over onto K °. This means that E-/K ° -- E/K.

Let { be the group E endowed with the topology induced b y its

canonical embedding into R e × R ~. It is not h a r d to see that id :

E/K + E/K is a t o p o l o g i c a l isomorphism. Let ~ : E + E/K be the ca-

nonical projection. The set

H" : {tu : t E E0,1]; u e H}

is o b v i o u s l y compact in ~. Consequently, 9(H') is a c o m p a c t subset

of E/K. We shall prove that
 157

o
(i) [¢(H')]E/K = {0}.

Take any × ~ (E/K) with IX(~(H')) I ~ ~. Then ×$ ~ E , and,

by (2.3), there is some f E E* with Pf = X~. From (2.2) we o b t a i n

(2) f(u) = X~(U) for all u e H'.

Now, take any v e H and denote I v = {tv : t e [0,11}. Then ~(I v)

is a subgroup of E/K~ since IX~(Iv) l =< !4' it follows that

IX~(Iv) I = 0 and (2) implies that f(v) = 0. Since v E H was arbi-

trary and H is l i n e a r l y dense in E, we see that f = 0, which means
that X = 0. This proves (i).

Let T H* be the group of all functions f : H* + T endowed with

the topology of p o i n t w i s e convergence. Let ~ : E / K ~ T H* be the ho-

momorphism given b y the formula

¢(~(u))(v*) = pv*(u) (u ~ E, v* g H*)

(it c a n be shown that ¢ is a t o p o l o g i c a l embedding). Take any u =

Z Xne n E E a n d set f = ~(~(u). Then, according to our defini-
nEZ
tions, for e a c h n ~ Z, we h a v e

* -i
f(u n) = p[(l - %) (X_n %Xn)],

f(w n) = p[(l - %)-l(x n X_n)],

which c a n be w r i t t e n as

W
x n = f(Un) + f(w n) + k n + i n,

X_n= f(u n) + % f ( w n) + k n + %1 n,

for some k n , 1 n ~ Z. Since Z Xne n e E, there is an index n > 0

o
such that, for e a c h n > no, one has X_n = 0, which is c l e a r l y pos-
sible if a n d o n l y if

(3) f(u n
) + % f ( W n*) ~ Z + %Z.

The above argument shows that ¢(E/K) consists of all t h o s e func-

 158

tions f : H* ~ T which satisfy (3) for s u f f i c i e n t l y large n. So,

%(E/K) is a n o n - c l o s e d dense s u b g r o u p of the c o m p a c t group T H*. Now,

(i) implies that (E/K) is discrete. Therefore, if E/K were re~ex-

ive, it w o u l d have to be compact and then ~(E/K) would be a compact

subgroup of T H*. This proves that E/K cannot be reflexive. It is

not h a r d to see that (E/K) is compact and eE/K is a topological
i s o m o r p h i s m of E/K onto a n o n - c l o s e d d e n s e s u b g r o u p of (E/K) (cf.
(14.7)).
From (14.8) we see that @K : E-/K° + K- is a topological iso-
morphism. Since E-/K ° -- E/K, and E/K is not reflexive, it follows
that K is not reflexive, either. It can be shown that eK is an open,
but not continuous, i s o m o r p h i s m of K onto K , that K is di-
screte and that each p r e c o m p a c t subset of K is finite. The h o m o m o r -
phism ~K : (E/K)- ~ K ° is not open; c o n s e q u e n t l y , ~ : E ~ E/K is
not c o m p a c t - c o v e r i n g .
The above example shows that in (17.3) the c o u n t a b l e product of
LCA groups cannot be r e p l a c e d by the c o u n t a b l e direct sum.

~17_8~ p~,~R~. By (17.3), the groups ~R and R~ are both strong-

ly reflexive. Therefore (17.7) shows that the p r o d u c t of two s t r o n g l y
r e f l e x i v e groups need not be s t r o n g l y reflexive. It is quite easy to
show that the p r o d u c t of a s t r o n g l y r e f l e x i v e g r o u p and a c o m p a c t or
d i s c r e t e one is s t r o n g l y r e f l e x i v e again; it is not known if the prod-
uct of a s t r o n g l y r e f l e x i v e g r o u p and the real line m u s t be strongly
reflexive.

A s t a n d a r d a r g u m e n t shows that each c l o s e d linear s u b s p a c e and

each H a u s d o r f f q u o t i e n t space of ~(R ~) and (~R) ~ are reflexive (the
m e a n i n g of the symbols is obvious). The author does not know if the
same is true for the spaces (~(R~)) ~ and ~((~R)~).

(~7.9) ~ . A subset of ~R is o p e n if and o n l y if it inter-

sects e a c h finite d i m e n s i o n a l s u b s p a c e in a r e l a t i v e l y o p e n set ([26],
P r o p o s i t i o n 1 or [93], P r o p o s i t i o n 4 on p. 477). So, eR is a k-space
and R~ is a k-space because it is metrizable. Nevertheless, ~R x R ~

is not a k-space; the f u n c t i o n (Xn)ne Z + ~ XnX_n is k-continuou~

but not continuous, n=l
 159

(17.1o) ~ . Let G = R for all n ~ Z, let H = {0} for

n n
n < 0 and Hn = R for n ~ 0. It is then clear that ~R × R ~-
Z (G n : Hn). So, (17.7) shows that countable reduced products of L C A
n~Z
groups need not be strongly reflexive. Nevertheless, if G is a r e f l e x -
ive b i n u c l e a r group (in p a r t i c u l a r , a countable reduced product of LCA

groups) and K is a c l o s e d subgroup of G, then #K : G-/K° ~ K- is

a topological isomorphism, cK : (G/H)- ~ K° is an open isomorphism,

~K is an o p e n isomorphism of K onto K and eG/K is a t o p o l o g i ~

al e m b e d d i n g of G/K into (G/K) The proofs of these assertions
are left to the reader.

(17.11J~r2~,~K. Let ~ be the first uncountable cardinal and, for

each e < ~, let G be the group Z/2Z. Let G be (algebraically)
the g r o u p Z G . As a b a s e at zero in G we take the family
e<R

{HE}E< ~ where

H E = {(g~)~<~ : g~ = 0 for all ~ < E].

Then G is a n o n - d i s c r e t e Hausdorff group and e v e r y H8 is an o p e n

subgroup of G.

The groups G/H E are d i s c r e t e , and the canonical projections

~E : G ~ G/H E induce a homomorphism ~ of G into the limit K of
the inverse system

{~YE : G/Hy + G/HE; E £ ¥ < R}

where ~¥E are the canonical projections. Evidently, ~ is a t o p o -

logical embedding. As a matter of fact, G is c o m p l e t e and, t h e r e f o r e ,
is a t o p o l o g i c a l isomorphism of G onto K (see [72], Example 1.6
and [102], §5). So, G may be treated as a c l o s e d subgroup of the
product II (G/HE). We shall show that G is not reflexive; this

means that the countable product of L C A g r o u p s in (17.3) cannot be re-

placed b y an u n c o u n t a b l e one (see, however, (17.14)).
The dual group G m a y be identified algebraically with the sub-
group Q of the p r o d u c t II G~, consisting of all elements g =

(g)a<a with the p r o p e r t y that there is s o m e E < ~ (depending on g )

 160

such that ge = 0 when ~ & ~. It is not hard to see that every pre-
compact subset of G is finite; therefore G- m a y be i d e n t i f i e d w i t h
Q topologically. For each ~ < ~, let g8 = (gi~)~<~ be the e l e m e n t

of ~<~II Ge with g~8 ~ 0 and g~ = 0 when ~ # ~. It is clear that

the set {0} U {g }8<~ is compact, and its q u a s i - c o n v e x hull in G

is G- itself. Consequently, G ~- is discrete. Naturally, we may i-

dentify G with Z G . This means that eG is an open, but not
e<~
continuous, i s o m o r p h i s m of G onto G -~

The p o s s i b i l i t y of extending the P o n t r y a g i n - v a n K a m p e n d u a l i t y

t h e o r e m to n o n - l o c a l l y c o m p a c t a b e l i a n groups was i n v e s t i g a t e d in sev-
eral papers. A brief survey can be found in [38], in the notes after
§§23-25. Here we shall r e s t r i c t o u r s e l v e s to a r e v i e w of results con-
nected w i t h strong duality.

Let (A n)n=l
~ be a s e q u e n c e of LCA groups Set G = II A n and H =
n=l
Z A . Let P and Q be a r b i t r a r y c l o s e d s u b g r o u p s of G and H,
n=l n

respectively.
Kaplan [49] p r o v e d that G and H are reflexive, and that there

are c a n o n i c a l t o p o l o g i c a l isomorphisms G -- E A and H = II A

n=l n n=l n
Kaplan [50] proved that P is d u a l l y c l o s e d and d u a l l y e m b e d d e d in G.
Varopoulos [93] p r o v e d that Q is d u a l l y c l o s e d and dually embedded
in H. Kaplan [50] p r o v e d that limits of d i r e c t and inverse s e q u e n c e s
of LCA groups are reflexive. Similar results were independently ob-
tained by V i l e n k i n [99].

Varopoulos [93] showed that ~Q : (H/Q) A ~ QO and ~H : H - / Q ° ~ Q -

are t o p o l o g i c a l isomorphisms. Noble [72] p r o v e d that P is reflexive.
He i n t r o d u c e d a special class of t o p o l o g i c a l groups, the s o - c a l l e d ~-
-groups (see (14.5)) and p r o v e d that if K is a k-group, then ~K is
continuous. It can be shown that P,Q,G/P and H/Q are all k-groups.
B r o w n et al. [26] showed that G is s t r o n g l y r e f l e x i v e if A1 is com-
pact and An is i s o m o r p h i c to R or to Z for n ~ 2 (or, dually,
if A1 is d i s c r e t e and An is i s o m o r p h i c to R or to T for n ~ 2).
Venkataraman [953 p r o v e d that P is reflexive, and that ~p : G-/P °
P is a t o p o l o g i c a l isomorphism.
 161

(17.12) ; ~ R g . As a s p e c i a l case of (17.3) we o b t a i n the follow-

ing p r o p o s i t i o n :

(*) countable products of LCA groups are s t r o n g l y reflexive.

It turns out that (*) is an almost immediate consequence of the results

of [50] and [93] (see [12]). It was p o i n t e d out to the author by E.
Martin-Peinador that, as a m a t t e r of fact, (*) is a d i r e c t consequence
of the results of [50] only, which allows one to o b t a i n a relatively
elementary proof of (*). Another elementary proof of (*) can be ob-
t a i n e d by f o l l o w i n g the w a y of [26].

(17.13) ~ v l [ . Let A be an open subgroup of a Hausdorff abelian

grou p G. Then A is d u a l l y closed and d u a l l y embedded in G ([72],
Lemma 3.3). Venkataraman [94],Corollary 6.3, proved that if G is re-
flexive, then A is reflexive, too (for k-groups, this was proved
by N o b l e [72], Corollary 3.4). It can be shown that G is r e f l e x i v e
(resp. strongly reflexive) if and only if A is reflexive (resp. strong-
ly reflexive); the proof will be g i v e n in [29].

A group locally isomorphic to a r e f l e x i v e group need not be re-

flexive itself (see (15.3) and (5.1) or (5.3)). The author does not
know w h e t h e r a group locally isomorphic to a strongly reflexive one
must be s t r o n g l y reflexive.

(17.14) B ] ~ . Negrepontis [71] p r o v e d that direct and inverse

limits of c o m p a c t l y generated LCA groups are r e f l e x i v e (cf. (17.11)).
There is, however, an error in the proof of P r o p o s i t i o n 4.4 on p. 250.
The author does not know w h e t h e r closed subgroups of uncountable d~ect
sums of real lines are d u a l l y closed or d u a l l y embedded. Also, it re-
main s an open q u e s t i o n whether closed subgroups or H a u s d o r f f quotients
of u n c o u n t a b l e products of R's or of Z's are reflexive. Brown et
al. [263 p r o v e d that each c l o s e d subgroup of the c o u n t a b l e product of
R's can be w r i t t e n in the form

{(x n) : xn = 0 for n e A and xn ~ Z for n ~ B}

for some d i s j o i n t subsets A,B of {1,2,...}. Most likely, this re-

sult c a n n o t be e x t e n d e d to u n c o u n t a b l e products.

~17.15J !o'1"~. The m a t e r i a l of this section is new, with the follow-

ing exceptions. Strong reflexivity was introduced in [26]. Parts (a)
and (b) of (17.1) were communicated to the author by M.J. Chasco and
 162

E. M a r t l n - P e i n a d o r . Parts (c) and (d) of (17.1) are, in fact, taken

from [26J. The special case of (17.3) when G is a n u c l e a r Frechet
space and all g r o u p s An are trivial appears in [81. Example (17.11)
is t a k e n from [72]; as a m a t t e r of fact, this is the e x a m p l e given by
Leptin in [61].

18. Groups w i t h b o u n d e d n e s s

One of the m a i n obstacles in e x t e n d i n g Pontryagin duality to non-

-locally compact groups is that the q u o t i e n t homomorphism G + G/H need
not be c o m p a c t - c o v e r i n g (cf. (17.6) and (17.7)). To avoid this diffi-
culty, N.Ya. Vilenkin [98J introduced the s o - c a l l e d groups w i t h bound-
edness. In this s e c t i o n we give an o u t l i n e of Vilenkin's theory and
show h o w groups with boundedness a l l o w one to o b t a i n an analogue of
(17.2) for b i n u c l e a r groups which are not Cech-complete. The m a i n re-
sults are (18.4) and (18.7).

Let G be an a b e l i a n group. By a b o u n d e d n e s s on G we m e a n a
family of subsets of G, called bounded sets, satisfying the follow-
ing conditions:

i) if X is bounded, so is -X;

2) subsets of b o u n d e d sets are bounded;

3) if X,Y are bounded, so are X U Y and X + Y;

4) finite sets are bounded.

For instance, the family of all finite (resp. precompact or r e l a t i v e l y

compact) subsets of a t o p o l o g i c a l group forms a boundedness. Another
example is the f a m i l y of all subsets of a topological vector space
which are b o u n d e d in the u s u a l sense i.e. a b s o r b e d by neighbourhoods of
zero. Equipped with this b o u n d e d n e s s , a topological vector space E w i ~
be d e n o t e d by E .
o
Let G be a g r o u p w i t h b o u n d e d n e s s and H its subgroup. The
boundedness on G induces the b o u n d e d n e s s on H and on G/H. As
bounded subsets of H we take the sets H n X where X is a b o u n d e d
subset of G. Bounded subsets of G/H are d e f i n e d as the c a n o n i c a l
images of b o u n d e d subsets of G.
If L is a c l o s e d linear subspace of a t o p o l o g i c a l vector space
E, then it m a y h a p p e n that (E/L)o # E /L; for instance, in the no-
tation of (17.6), there are b o u n d e d subsets of 0 /L which are not
canonical images of b o u n d e d subsets of D.
 163

Let G be a t o p o l o g i c a l group w i t h boundedness. We say that G

is a Q - g r o u p if it is locally q u a s i - c o n v e x and if q u a s i - c o n v e x hulls of
b o u n d e d sets are bounded. If E is a locally c o n v e x space, then e v e r y
closed, c o n v e x and s y m m e t r i c subset of E is q u a s i - c o n v e x (see the
proof of (2.4)); t h e r e f o r e E is a Q-group. It is obvious that any
s u b g r o u p of a Q - g r o u p is a Q-group. However, a H a u s d o r f f q u o t i e n t of a
Q - g r o u p need not be locally q u a s i - c o n v e x (see (5.3)).
Let ~ : G + H be a h o m o m o r p h i s m of groups w i t h boundedness. We
say that ~ is b o u n d e d if the images of b o u n d e d sets are bounded. We
say that ~ is b o u n d i n q if to each b o u n d e d X c H there corresponds
some b o u n d e d y c G with X n ~(G) c ~(y). Finally, ~ is called
b i - b o u n d e d if it is b o t h b o u n d e d and bounding. Bi-bounded topological
i s o m o r p h i s m s will be c a l l e d b - i s o m o r p h i s m s .
Let G be a t o p o l o g i c a l group w i t h boundedness. The dual group,
d e n o t e d by G d, is d e f i n e d in the f o l l o w i n g way. As a set, Gd con-
sists of all c o n t i n u o u s c h a r a c t e r s of G. As a base at zero in Gd we
take the polars of b o u n d e d subsets of G. The b o u n d e d n e s s on Gd con-
sists of e q u i c o n t i n u o u s sets. It follows i m m e d i a t e l y from our defini-
tions that Gd is a Q-group. N o t i c e that if % : G + H is a b-iso-
morphism, then the dual m a p p i n g ~d : H d ~ G d, defined in the usual
way, is a b - i s o m o r p h i s m , too.
The e v a l u a t i o n map G + G dd will be d e n o t e d by 8G. We say that
G is an i n v o l u t i v e g r o u p if ~G is a b - i s o m o r p h i s m of G onto G dd.

(~8.1) l ~ J i a ~ s ~ . Let G be a s e p a r a t e d Q-group. Then ~G is a

b - i s o m o r p h i s m of G onto 8G(G).

This is a direct c o n s e q u e n c e of our definitions.

(18.2)~osi, I~. A locally c o n v e x space E is s e m i - r e f l e x i v e if

and only if E° is an i n v o l u t i v e group.

Proof. By (2.3), we have the t o p o l o g i c a l isomorphism PE : Eb ~

Ea) d which, again by (2.3), allows us to identify algebraically
E ) with (Eo)dd. Thus, E is semi-reflexiveif and ortly if im ~E
O
Eo) dd. Now, it suffices to apply (18.1) b e c a u s e Eo is a Q-group. •

~18.3) PI~FOSI~X~. A d u a l l y closed, d u a l l y e m b e d d e d s u b g r o u p of an

i n v o l u t i v e g r o u p is involutive.
 164

Pz~f. Let H be a d u a l l y closed, dually embedded subgroup of an

involutive group G. In v i e w of (18.1), it is e n o u g h to show that
im 8H = Hdd" Take an a r b i t r a r y continuous character ~ of H d. The

canonical homomorphism ~ : Gd ~ Hd is c o n t i n u o u s , therefore ~ is

a continuous character of G d. It is o b v i o u s that ~ ~ H O°. Since

H is d u a l l y closed in G, we h a v e ~G(H) = H °° (cf. (14.2)). So,

there is some h ~ H with ~G(h) = ~¢ and it is not h a r d to verify

that ~H(h) = ~ (cf. the p r o o f of (14.6)). •

Let G = Z (G i : H i ) be a r e d u c e d p r o d u c t of g r o u p s w i t h b o u n d -
ieI
edness. We i n t r o d u c e b o u n d e d n e s s into G in the f o l l o w i n g way. For
each i ~ I, let ~i : G + Gi be the c a n o n i c a l projection. A subset
X of G is b o u n d e d if ~i(X) is a b o u n d e d subset of Gi for all
i ~ I and if ~i(X) c Hi for all but finitely many i.

(18.4) R r n [ s . Let G =
Z (G i : H i ) be a reduced product of
iEI
topological groups with boundedness. Suppose that Hi is dually c l o s e d
in Gi for a l m o s t all i. Then Gd is c a n o n i c a l l y b-isomorphic to

Z ( G ~ : H~).~ Consequently, if all g r o u p s Gi are involutive, then G

i~I
is i n v o l u t i v e , too.

Proof. Take any X e Gd. Since X is continuous, one has

I×(o)1 ¼ for some U E No(G ) . According to the d e f i n i t i o n of the

topology on G, we m a y assume that

(i) U = {(gi ) ~ G : gi ~ Ui for all i ~ I}

where Ui E No(Gi) for all i and H I• c U.l for almost all i. Let

Xi = XIGi Gd1 for e a c h i E I. Then Xi E H oi for almost all i,

which means that

(Xi)i~ I e G" : = Z (G~ : Hg).I

ieI

consider the m a p p i n g ~ : G d ~ G" given by #(X) = (Xi)is I. It is

clear that ~ is an a l g e b r a i c a l isomorphism of Gd onto G" We

shall prove that ¢ is a b - i s o m o r p h i s m .
 165

1° ¢ is c o n t i n u o u s . Take any W e No(G'). According to the

definition of the topology on G', we m a y asume that

W = {(Xi) ~ G" : Xi E W i for all i ~ I}

where W i ~ No(G ~) for all i and H9 c W i for almost all i. For

1
each i ~ I, there exists a bounded subset Xi of Gi with Xgl c W..l

If Xi ~ Hi , then H~1 ~ X o
i because Hi is d u a l l y closed in G i. T h i s
means that Xi c Hi for almost all i. Then the set

(2) X = {(gi ) E G : gi~ Xi for all i g I}

is b o u n d e d in G. We h a v e X ° e No(Gd ) and it is clear that

¢(X °) c W.

2° ~ i s open. Take any V E No(Gd ) . There is a b o u n d e d subset

X of G with X O c V. We m a y assume that X has form (2) where
Xi is a b o u n d e d subset of Gi for all i and Xi c Hi for almost
all i. We m a y w r i t e

XO = {(Xi ) ~ G d : Xi E Xgl for all i E I}.

If X i c Hi, then H oi c X9i" So ' we h a v e H~i c X91 for a l m o s t all i,

which means that ~(X °) e No(G').

3° # is b o u n d e d . Let X be a b o u n d e d subset of G d. There is

some U ~ No(G) with X c U °. we may assume that U has form (i).

Then

UO = {(Xi ) ~ G d : Xi ~ U O for all i e I}.

o o it
If H i c Ui, then U i c Hi; since this holds for almost all i,
follows that %(U °) is a b o u n d e d subset of G'.

4° ~ is b o u n d i n g . Let Y be a b o u n d e d subset of G'. We may

assume that

y = {(Xi) E G" : Xi E Yi for all i ~ I}

where Y- is a b o u n d e d s u b s e t of G.~ for all i and y~ c H O for

1 1 1 1
all i outside a certain finite subset J of I. To each i e I
 166

there c o r r e s p o n d s some U i e No(Gi) with Yi c Ugl. Let A be the

subset of G c o n s i s t i n g of all s e q u e n c e s (gi) such that gi E U i for

i ~ J and gi ~ Ui + Hi for i ~ J. Then A E No(G ) and A° is the

b o u n d e d subset of Gd c o n s i s t i n g of all s e q u e n c e s (Xi) such that

Xi E U~I for i ~ J and Xi E (U i + Hi )° for i ~ J. Since Hi is

a s u b g r o u p of Gi, we have (U i + Hi )° = U°1 N Hg.I This implies that

Yi c (U i + Hi)° for i ~ J. Hence y c ¢(A°). •

An i n v o l u t i v e group G is called strongly involutive if every

closed s u b g r o u p and e v e r y H a u s d o r f f q u o t i e n t group of G and of Gd
is involutive.

(18.5) [ ~ . Let G be a topolo§lical group with boundedness. For

each c l o s e d s u b g r o u p H of G, the c a n o n i c a l m a p p i n g cH : (G/H)d ~ H o
is a b - i s o m o r p h i s m .

The v e r i f i c a t i o n of this simple fact is left to the reader.

(18.6) PR~ssrrlom. Let H be a c l c s e d s u b g r o u p of a s t r o n g l y in-

volutive group G. Then

(a) H is d u a l l y c l o s e d and d u a l l y e m b e d d e d in G;

(b) the c a n o n i c a l m a p p i n g s CH : Gd/H° ~ Hd and cH : (G/H)d HO

are b - i s o m o r p h i s m s ;

(c) the groups H and G/H are s t r o n g l y involutive.

The proof is similar to that of (17.1). We leave it to the read-

er.

Let G be an abelian t o p o l o g i c a l group. By [G] we shall denote

the g r o u p G e q u i p p e d w i t h the b o u n d e d n e s s c o n s i s t i n g of precompact
sets.

(18.7)~r~J~f. Let G be a r e f l e x i v e group. Suppose that both the

groups G and G are n u c l e a r and complete. Then the group [G] is
s t r o n g l y involutive.

The a s s u m p t i o n s of (18.7) are satisfied, in particular, for G =

~(D) or G = eR x R e (cf. (17.6) and (17.7)).
 167

Proof. It follows d i r e c t l y from our a s s u m p t i o n s that the i d e n t i t y

mappings i : [G] d ~ [G-] and j : [G-] d + [G--J are b-isomorphisms.
Then the sequence

[G] eG > [G __ ] •
j-i ~ l.d
> [G ]d > [G] dd

shows that [G] is involutive. By (8.6) and (8.3), closed s u b g r o u p s of

[G] and of [G] d are d u a l l y c l o s e d and d u a l l y embedded. Hence, by
(18.3), each closed s u b g r o u p of [G] and of [G] d is involutive. It
remains to show that H a u s d o r f f q u o t i e n t groups of [G] and of [G] d are
involutive.
So, let H be a c l o s e d s u b g r o u p of [G]. The canonical mapping
¥ : [G]/H + [G]dd/H°° is a b - i s o m o r p h i s m (see (14.2)). Next, (14.8)
implies that the c a n o n i c a l m a p p i n g ¢ : [G]dd/H °° ~ (H°) d is a to-
Ho
p o l o g i c a l isomorphism. It is clear that ¢ is bounded; that it is
Ho
follows from (8.2). Finally, (18.5) says that cH : ([GS/H) d + H ° is
a b-isomorphism. Now, the s e q u e n c e

[G]/H Y > [G]dd/HO ° CH o > (H°) d (~H)d > ([G]/H) dd

shows that [G]/H is involutive. The proof that H a u s d o r f f quotients of

[G] d are i n v o l u t i v e is similar. •

(15.S) ~ . If E is a n o n - r e f l e x i v e B a n a c h space, then E is

a r e f l e x i v e group, but E is not i n v o l u t i v e (see (15.2) and (18.2)).
T h e r e f o r e it is i n t e r e s t i n g that a m e t r i z a b l e l o c a l l y c o n v e x space E
is a s t r o n g l y r e f l e x i v e group if and only if E is s t r o n g l y involu-
tive, namely, if and o n l y if E is a n u c l e a r F r e c h e t space. This fol-
lows from (17.3 , (18.7) and (6.1).

(18.9) HOI,Y . The m a t e r i a l of this section is taken from [98].

Theorem (18.7) is new.
 BIBLIOGRAPRY

i. R. Arens, Duality in linear spaces, Duke Math. J. 14 (1947),787-793.

2. A. A r k h a n g e l s k i i , Open and c l o s e - t o - o p e n mappings. Relations

among spaces, T r u d y Moskov. Mat. Obsch. 15 (1966), 181-223.

3. K. Ball, Some remarks on the g e o m e t r y of convex sets, GAFA

Seminar "86-87, S p r i n g e r Lecture Notes in Math. I]17 (1988), 2 2 ~
- 231.

4. M. B a n a s z c z y k and W. Banaszczyk, Characterization of nuclear

spaces by m e a n s of a d d i t i v e subgroups, Math. Z. 186 (1984), 125-
-133.

5. W. B a n a s z c z y k , On the e x i s t e n c e of unitary representations of

c o m m u t a t i v e nuclear Lie groups, Studia Math. 76 (1983), 175-181.

6. --, On the e x i s t e n c e of exotic Banach-Lie groups, Ann. Math. 264

(1983), 485-493.

7. --, Closed subgroups of nuclear spaces are w e a k l y closed, Stu-

dia Math. 8 0 (1984), 119-128.

8. --, P o n t r y a g i n d u a l i t y for subgroups and q u o t i e n t s of nuclear

spaces, Math. Ann. 273 (1986), 653-664.

9. --, On the e x i s t e n c e of c o m m u t a t i v e B a n a c h - L i e groups w h i c h do

not admit continuous unitary representation, Coll. Math. 52
(1987), 113-118.

i0. --, The S t e i n i t z t h e o r e m on r e a r r a n g e m e n t s of series for nu-

clear spaces, J. R e i n e Angew. Math. 403 (1990), 187-200.

Ii. --, Polar lattices from the p o i n t of v i e w of nuclear spaces,

Rev. Mat. Univ. Complut. M a d r i d 2 (special issue) (1989), 35-46.

12. --, C o u n t a b l e p r o d u c t s of LCA groups: their c l o s e d subgroups,

q u o t i e n t s and d u a l i t y p r o p e r t i e s , Colloq. Math. 59 (1990), 53-57.

13. --, A Beck-Fiala-type theorem for e u c l i d e a n norms, to appear in

E u r o p e a n J. Combin. ii (1990).

14. --, Rearrangements of series in n o n - n u c l e a r spacess, in pre-

paration.

15. W. B a n a s z c z y k and J. Grabowski, Connected subgroups of nuclear

spaces, Studia Math. 78 (1984), 161-163.
 169

16. I. Barany, Rearrangements of series in infinite dimensional

spaces, Mat. Zametki 46 (1989), no. 6, 10-17 (Russian), translated
as Math. Notes (preprint 1983).
17. I. B a r a n y and V.S. Grinberg, On some c o m b i n a t o r i a l q u e s t i o n s in
finite d i m e n s i o n a l spaces, Linear A l g e b r a Appl. 41 (1981), 1-9.
18. A.O. Barut and R. R~czka, T h e o r y of group representations and ap-
plications, PWN - P o l i s h S c i e n t i f i c Publishers, W a r s z a w a 1977.
19. J. Beck, B a l a n c i n g families of integer sequences, Combinatorica
i (1981), 209-216.
20. J. B e c k and T. Fiala, I n t e g e r - m a k i n g theorems, Discrete Appl.
Math. 3 (1981), 1-8.
21. J. Beck and J. Spencer, Integral a p p r o x i m a t i o n sequences, Math.
P r o g r a m m i n g 30 (1984), 88-98.
22. W. Blaschke, Uber affine Geometry XI: losing der "Vierpunkt-
problems" von S y i v e s t e r aus der T h e o r i e der g e o m e t r i s c h e n Wahrs-
deinlichkeiten. L e i p z i g e r B e r i c h t e 69 (1917), 436-453.
23. N. Bourbaki, E l e m e n t s de m a t h e m a t i q u e . Premiere partie. Les
structures f o n d a m e n t a l e s de l'analyse. Livre III. T o p o l o g i e gene-
rale. Chap. III. Groupes t o p o l o g i q u e s (Theorie elementaire). Chap.
4. N o m b r e s reels. 3 ed., Hermann, Paris 1960.
24. J. B o u r g a i n and V.D. Milman, New vol~ne ratio p r o p e r t i e s for con-
vex s y m m e t r i c bodies in R n, Invent. Math. 88 (1987), 319-340.
25. J. Braconnier, Sur les groupes t o p o l o g i q u e s l o c a l e m e n t compacts,
J. Math. Pures Appl. 27 (1948), 1-85.
26. R. Brown, P.J. Higgins and S.A. Morris, Countable products and
sums of lines and circles: their c l o s e d subgroups, q u o t i e n t s and
d u a l i t y properties, Math. Proc. C a m b r i d g e Philos. Soc. 7 8 < 1 9 7 5 ) ,
19-32.
27. H . - P. Butzmann, P o n t r j a g i n - D u a l i t a t fur topologische Vektor-
raume, Arch. Math. (Basel) 28 (1977), 632-637.
28. J.W.S. Cassels, An i n t r o d u c t i o n to the geometry of numbers,
Springer-Verlag, B e r l i n 1959.
29. M.J. Chasco and E. M a r t l n - P e i n a d o r , Open subgroups and Pontrya-
gin duality, to appear in Math. Z.

30. I.M. G e l f a n d and N.Ya. Vilenkin, G e n e r a l i z e d functions, vol. 4,

A p p l i c a t i o n s of h a r m o n i c analysis, A c a d e m i c Press, New Y o r k -
- L o n d o n 1964.
31. M. Gromov and V.D. Milman, Brunn t h e o r e m and a c o n c e n t r a t i o n of
volume of c o n v e x bodies, GAFA Seminar Notes, Tel Aviv University,
Israel 1983-1984, Exp. V, 12 pp.
32. V.S. G r i n b e r g and S.V. Sevastyanov, V a l u e of the Steinitz constant,
Funktsional. Anal. i Prilozhen. 14 (1980), no. 2, 56-57 (Russian),
t r a n s l a t e d as F u n c t i o n a l Anal. Appl. 14 (1980), 125-126.
33. P.M. Gruber and C.G. Lekkerkerker, G e o m e t r y of numbers, North -
- Holland, A m s t e r d a m 1987.
34. I. Halperin, Sums of series, permitting rearrangements, C.R.
Math. Rep. Acad. Sci. Canada 8 (1986), 87-102.
 170

35. G.H. Hardy, J.E. L i t t l e w o o d and G. P61ya, Inequalities, Cam-

b r i d g e U n i v e r s i t y Press, C a m b r i d g e 1952.
36. J. Hastad, Dual vectors and lower bounds for the nearest lattice
point problem, C o m b i n a t o r i c a 8 (1988), 75-81.
37. W. H e r e r and J.P.R. Christensen, On the e x i s t e n c e of pathological
s u b m e a s u r e s and the c o n s t r u c t i o n of exotic t o p o l o g i c a l groups,
Math. Ann. 213 (1975), 203-210.
38. E. H e w i t t and K.A. Ross, A b s t r a c t h a r m o n i c analysis, vol. I,
S p r i n g e r - V e r l a g , B e r l i n 1963.
39. E. H i l l e and R.S. Phillips, F u n c t i o n a l a n a l y s i s and semi-groups,
Amer. Math. Soc., P r o v i d e n c e 1957.
40. C.J. Himmelberg, M e a s u r e relations, Fund. Math. 87 (1975), 53-
72.
41. R.C. Hooper, T o p o l o g i c a l groups and integer - valued norms, J.
Funct. Anal. 2 (1968), 243-257.
42. F. John, Polar c o r r e s p o n d e n c e with respect to c o n v e x region,
Duke Math. J. 3, No. 2, U n i v e r s i t y of K e n t u c k y (1937), 355-369.
43. V.M. Kadets and M.I. Kadets, R e a r r a n g e m e n t s of series in Banaeh
spaces, Tartu Gos. Univ., Tartu 1988 (Russian).
44. M.I. Kadets and K. W o ~ n i a k o w s k i , On series whose permutations
have only two sums, Bull. Polish Acad. Sci. Math. 37 (1989), 15-
21.
45. V.M. Kadets, On a p r o b l e m of S. B a n a c h (problem 106 from "The Scot-
tish Book"), Funktsional. Anal. i Prilozhen. 20 (1986), 7 4 - 75
(Russian), t r a n s l a t e d as F u n c t i o n a l Anal. Appl. 20 (1986), 317-
-319.
46. , Series p e r m u t a t i o n in infinite d i m e n s i o n a l spaces (main re-
sults and open problems), C.R. Math. Rep. Acad. Sci. Canada 11
(1989), 151-164.
47. R.V. K a d i s o n and J.R. Ringrose, Fundamentals of the t h e o r y of
o p e r a t o r algebras, vol. I E l e m e n t a r y theory, vol. II A d v a n c e d
theory, A c a d e m i c Press, New York 1983, 1986.
48. N. Kalton, S u b s e r i e s c o n v e r g e n c e in t o p o l o g i c a l groups and v e c t o r
spaces, Israel J. Math. i0 (1971), 402-412.
49. S. Kaplan, E x t e n s i o n s of the P o n t r j a g i n d u a l i t y I: infinite prod-
ucts, Duke Math. J. 15 (1948), 649-658.
50. , E x t e n s i o n s of the P o n t r j a g i n d u a l i t y II: direct and inverse
limits, ibid. 17 (1950), 419-435.
51. Y. K a t z n e l s o n and O.C. McGehee, Conditionally convergent series
in R ~, M i c h i g a n Math. J. 21 (1974), 97-106.
52. J.L. Kelley, General topology, Van Nostrand, New Y o r k 1955.
53. A.A. Kirillov, E l e m e n t s of the t h e o r y of r e p r e s e n t a t i o n s , Sprin-
ger-Verlag, B e r l i n 1976.
54. J. Kisy~ski, On the g e n e r a t i o n of tight measures, Studia Math.
30 (1968), 141-151.
55. Y. Komura, Some e x a m p l e s on linear t o p o l o g i c a l spaces, Math. Ann.
 171

153 (1984), 150-162.

56. S. Kwapie~, On the form of a linear o p e r a t o r in the space of all
m e a s u r a b l e functions, Bull. Acad. Polon. Sci. Set. Sci. Math.
Astronom. Phys. 21 (1973), 951-954.
57. S.H. Kye, P o n t r y a g i n d u a l i t y in real linear t o p o l o g i c a l spaces,
Chinese J. Math. 12 (1984), no. 2, 129-136.
58. J.C. Lagarias, H.W. Lenstra, Jr. and C.P. Schnorr, K o r k i n - Z o l o t a -
rev bases and s u c c e s s i v e m i n i m a of a lattice and its r e c i p r o c a l
lattice, p r e p r i n t 1989.
59. D.G. Larman and C.A. Rogers, The e x i s t e n c e of a c e n t r a l l y sym-
m e t r i c convex b o d y w i t h central sections that are u n e x p e c t e d l y
small, M a t h e m a t i k a 22 (1975), 164-175.
60. H. Leptin, B e m e r k u n g zu e i n e m Satz von S. Kaplan, Arch. Math. 6
(1955), 139-144.
61. , Zur D u a l i t a t s t h e o r i e p r o j e k t i v e r Limites a b e l s c h e r Gruppen,
Abh. Math. Sem. Univ. H a m b u r g 19 (1955), 264-268.
62. A. M~drecki, Minlos t h e o r e m in p-adic locally convex spaces,
preprint.
63. K. Maurin, General e i g e n f u n c t i o n e x p a n s i o n s and u n i t a r y repre-
s e n t a t i o n s of t o p o l o g i c a l groups, PWN - P o l i s h Scientific Pub-
lishers, W a r s z a w a 1968.
64. S. Mazur, On the c o n v e r g e n c e of lacunary polynomials, Studia Math.
89 (1988), 75-78.
65. V.D. M i l m a n and A. Pajor, Isotropic p o s i t i o n and inertia el-
lipsoids and zonoids of the unit ball of a n o r m e d n-dimensional
space, p r e p r i n t IHES, M a y 1989.
66. J. M i l n o r and D. Husemoller, S y m m e t r i c b i l i n e a r froms, Springer-
-Verlag, B e r l i n 1973.
67. R.A. Minlos, Generalized stochastic processes and their exten-
sion to the measure, T r u d y Moskov. Mat. Obshch. 8 (1959), 497-
-518 (Russian).
68. A.F. Monna, Analyse non-archimedienne, Springer-Verlag, Berlin
1970.
69. S.A. Morris, P o n t r y a g i n d u a l i t y and the s t r u c t u r e of locally com-
pact a b e l i a n groups, C a m b r i d g e U n i v e r s i t y Press, C a m b r i d g e 1977.
70. D.H. Muschtari, Some general q u e s t i o n s of the t h e o r y of p r o b a b i -
lity m e a s u r e s in linear spaces, Teor. V e r o y a t n o s t . i Primenen. 18
(1973), 66-77 (Russian).
71. J.W. N e g r e p o n t i s , D u a l i t y in analysis from the point of view of
triples, J. A l g e b r a 19 (1971), 228-253.
72. N. Noble, k-groups and duality, Trans. Amer. Math. Soc. 151
(1970), 551-561.
73. W. Orlicz, D i v e r g e n z von a l l g e m e i n e n O r t h o g o n a l r e i h e n , Studia
Math. 4 (1933), 27-32.
74. D.V. Pecherskil, Series p e r m u t a t i o n s in B a n a c h spaces and rear-
r a n g e m e n t s of signs, Mat. Sb. (N.S.) 135 (177) (1988), 24 - 35
 172

(Russian), t r a n s l a t e d as Math. USSR-Sb.

75 A. Pietsch, N u c l e a r locally convex spaces, Springer-Verlag, Ber-
lin 1972.
76 , O p e r a t o r ideals, VEB D e u t s c h e r V e r l a g der Wissenschaften,
B e r l i n 1978.
77 G. Polya and G. Szego, P r o b l e m s and theorems in analysis, vol. II,
S p r i n g e r - V e r l a g , Berlin 1976.
78 D.A. Raikov, On B - c o m p l e t e t o p o l o g i c a l vector spaces, Studia
Math. 31 (1968), 295-306 (Russian).
79 S. Rolewicz, M e t r i c linear spaces, second e n l a r g e d edition, PWN-
P o l i s h S c i e n t i f i c Publishers, W a r s z a w a 1984.
80 H.H. Schaefer, T o p o l o g i c a l vector spaces, Springer-Verlag, Ber-
lin 1971.
81 The S c o t t i s h Book, ed. by R. Daniel Mauldin, Birkhauser, Boston
1981.
82 S.J. Sidney, W e a k l y dense subgroups of B a n a c h spaces, Indiana
Univ. Math. J. 26 (1977), 981-986.
83 M.F. Smith, The P o n t r j a g i n d u a l i t y t h e o r e m in linear spaces, Ann.
of Math. 56 (1952), 248-253.
84 O.G. Smolyanov, The spaces D is not h e r e d i t a r i l y complete, Izv.
Akad. N a u k SSSR Ser. Mat. 35 (1971), 628-696 (Russian), trans-
lated as Math. USSR-Izv. 5 (1971), 696-710.
85. A n a l y s i s on t o p o l o g i c a l vector spaces and its applications,
M o s c o w 1979 (Russian).
86. J. Spencer, Six s t a n d a r d d e v i a t i o n s suffice, Trans. Amer. Math.
Soc. 289 (1985), 679-705.
87. E. Steinitz, Bedingt k o n v e r g e n t e R e i h e n und konvexe Systeme, J.
Reine Angew. Math. 143 (1913), 128-175.
88. S. Troyanski, On c o n d i t i o n a l l y c o n v e r g e n t series in c e r t a i n F-
-spaces, Teor. Funtsil, Funktsional. Anal. i Prilozhen. 5 (1967),
102-107 (Russian).
89. S.M. Ulam, A c o l l e c t i o n of m a t h e m a t i c a l problems, Interscience
Publishers, New Y o r k 1960.
90. M. Valdivia, The spaces of d i s t r i b u t i o n s ~'(~) is not Br-COmplete,
Math. Ann. 211 (1974), 145-149.
91. , The space D(~) is not B -complete, Ann. Inst. Fourier
(Grenoble) 27 (1977), 29-43. r
92. -364. (F)-spaces and strict (LF)-spaces, Math. Z. 195 (1987), 345-

93. N.Th. Varopoulos, Studies in h a r m o n i c analysis, Math. Proc. Cam-

b r i d g e Philos. Soc. 60 (1964), 465-516.
94. R. V e n k a t a r a m a n , E x t e n s i o n s of P o n t r y a g i n duality, Math. Z. 143
(1975), 105-112.
95. A c h a r a c t e r i z a t i o n of P o n t r y a g i n duality, Math. Z. 149 (1976),
109-119.
 173

96. N.Ya. Vilenkin, Direct decompositions of topological groups I,

Mat. Sb. 19 (61) (1946), 85-154 (Russian).
97. --, Theory of topological groups. II. Direct products. Direct
sums of groups of rank i. Locally bicompact abelian groups. Fi-
bered and weakly separable groups. Uspekhi Mat. Nauk 5 no.4 (38)
(1950), 19-74 (Russian).
98. --, Theory of characters of topological groups with a bounded-
ness given, Izv. Akad. Nauk SSSR Set. Mat. 15 (1951), 439 - 4 6 2
(Russian).
99. , Direct and inverse spectra of topological groups and their
character theory, ibid., 503-532 (Russian).
i00. A. Wald, Bedingt konvergente Reihen von Vektoren im Re, Er-
gebn. math. Kolloqu. 5 (1933), 13-14.
i01. Reihen in topologischen Gruppen, Ergebn. math. Kolloqu. 5
(1933), 14-16.
102. , L'integration dans les groupes topologiques et ses appli-
cations, Actualit~s Sci. et Indust. 869, 1145, Hermann & Cie.,
Paris 1940, 1951.
103. --, Basic number theory, Springer-Verlag, Berlin 1967.
104. Y. Yang, On a generalization of Minlos theorem, Fudan X u e b a o 2 0
(1981), no. i, 31-37 (Chinese).
105. A. Zygmund, Trigonometric series, 2nd ed., vol. I, Cambridge
University Press, Cambridge 1959.
 IMDEXOF

Section i: Z, R, C, S, T
P
<x,g>
I×(A) I
G , G c, Gp, Gpc, G
¢-
o Ao Ao Ao
A°' AG' p' c' pc
Am
gp A
N (G)
0

G -- H

II Gi, Z G.
iEI igI i

g/U
Ui, Z * U.
i~I ieI 1

(G i : H i)
iEI

g
GL(E)

Hi' ~ ~i
ieI iEI

Section 2: E# E* E~, E* E*
' ' c' pc
OE
span A
conv A
XWY
dk(X,Y)
 175

(utv)
d(u,A)
BE, B(E)
dk(¢ : E + F), dk(¢)
Bp, B(p)
E
P
~p
A
Pq

section 3: B
n
vol n

n
d(L)
L*
p(L,U)
Xk(L,U)
D
~i ~ "'" ~ ~n

Section 4: L~(X,p), L~(0,1)

Lc(X,P), L~(0,1)
L~(X,~), L~(0,1)

L~(X,~), L~(0,1)

e 2~i@

Section 5: c
o
1p
e
n

{f}
Lo

Section 6: lul
lwl
Vk(¢ : E ~ F), Vk(¢)
hk(¢ : E ~ F), hk(¢)
 176

Section 7: Co(N,G)

Section 8: R~

6(E,F)

Section 9: Wo(G)

Section i0: S(Zgi;G), S( Z g i )

C(Zgi;G), C( Z q i )

A(Zgi;G), A( Z g i )
F( Z u i )
F0( ~ u i )
~C~), ~(~), o(~)

Section ii: B(X)

P~

Section 14: aG
CH
~H'
Section 17: ~R, R~

Section 18: Gd
E

[G]
#d

~G
 SUBJECT I_~DEX

absolutely summable system 107 image of a measure iii

admissible topology iii invariant subspace 14
annihilator 2 inverse system (sequence) of topological
asterisk topology 9 groups 7
irreducible representation 14
balancing operator 108
basis (of a lattice) 26 Q-continuous mapping 134
bi-bounded homomorphism 163 b-group 134
Binet ellipsoid of inertia 66 Kolmogorov diameters 20
binuclear groups 152 Korkin-Zolotarev bases 44
b-isomorphism 163
Bochner theorem ii0 lattice 26
Borel measure iii LCA group (locally compact abelian) 5
bounded group 49 (see also 162) Legendre ellipsoid of inertia 66
bounded homomorphism 163 Levy-Steinitz theorem 94, 105
bounded representation 50 linear manifold 94
bounded set 162 locally convex vector group 86
boundedness 162 locally isomorphic groups 79
bounding homomorphism 163 locally quasi-convex group 2
Ls(X,~)- representation 45
C-group 134
character 1 Ls(O,I ) - representation 45
character group 1
compact convergence 2 Minlos theorem Ii0
compact-covering mapping 136 monothetic group 56
compact-open topology 2
continuous convergence 141 nuclear group 72
condition unitary representation 14 nuclear space 25
covering radius 42 nuclear vector group 87
cyclic representation 14
cyclic vector 14 Orlicz-Pettis theorem 107

determinant of a lattice 26 p.d. function 15

direct sum of topological groups 8 pointwise convergence 2
direct system (sequence) of topological pointwise linearly independent sequence 58
groups ii polar lattice 26
dual group 1 polar set 2
dual homomorphism 13 Pontryagin-van Kampen theorem 5
dually closed subgroup 5 positive-definite function 15
dually embedded subgroup 5 pre-Hilbert seminorm 23

equivalent representations 15 Q-group 163

exotic group 49 quasi-convex hull 2
quasi-convex set 2
faithful representation 14
Fourier transform of a measure 112 Radon measure iii
rectangular topology 9
Hilbert sum of Hilbert spaces 15 reduced product of topological groups 13
Hilbert sum of unitary representations reflexive group 132
15 regular measure 112
representation 14
 178

set of sums of a convergent series 93 topological isomorphism 5

SNAG theorem 126 topological embedding 5
space of a representation 14
Steinitz operator 105 unconditional convergence 107
Steinitz theorem 105 uniformly continuous representation 14
strongly balancing operator 109 unitary representation 14
strongly continuous representation 14 unitary space 23
strongly exotic group 49 unitarily equivalent representations 15
strongly involutive group 166
strongly reflexive group 151 weak topology on a topological group 107
subseries convergent series 107
zero component 26