LONDON MATHEMATICAL SOCIETY

LECTURE NOTE SERIES

Editor: PROFESSOR G. C. SHEPHARD, University of East Anglia

This series publishes the records of lectures and seminars on advanced

topics in mathematics held at universities throughout the world. For the
most part, these are at postgraduate level either presenting new material
or describing older material in a new way. Exceptionally, topics at the
undergraduate level may be published if the treatment is sufficiently
original.
Prospective authors should contact the editor in the first instance.
Already published in this series
1. General cohomology theory and K-theory, PETER HILTON
2. Numerical ranges of operators on normed spaces and of elements
of normed algebras, F. F. BONSALL and J. DUNCAN
3. Convex polytopes and the upper bound conjecture, P. McMULLEN
and G. C. SHEPHARD
4. Algebraic topology: A student's guide, J. F. ADAMS
5. Commutative algebra, J. T. KNIGHT
6. Finite groups of automorphisms, NORMAN BIGGS
7. Introduction to combinatory logic, J. R. HINDLEY, B. LERCHER
andJ. P. SELDIN
8. Integration and harmonic analysis on compact groups,
R. E. EDWARDS
9. Elliptic functions and elliptic curves, PATRICK DU VAL
10. Numerical ranges H, F. F. BONSALL and J. DUNCAN
11. New developments in topology, G. SEGAL (ed.)
12. Proceedings of the Symposium in complex analysis, Canterbury
1973, J. CLUNIE and W. K. HAYMAN (eds.)
13. Combinatorics, Proceedings of the British combinatorial Con-
ference 1973, T. P. McDONOUGH and V. C. MAVRON (eds.)
14. Analytic theory of abelian varieties, H. P. F. SWINNERTON-
DYER
London Mathematical Society Lecture Note Series 15

Introduction to
Topological Groups
P.J.HIGGINS

Cambridge • At the University Press • 1974

CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo

Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK

Published in the United States of America by Cambridge University Press, New York

www. Cambridge. org

Information on this title: www.cambridge.org/9780521205276
© Cambridge University Press 1974

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 1974

Re-issued in this digitally printed version 2008

A catalogue recordfor this publication is available from the British Library

Library of Congress Catalogue Card Number: 74-82222

ISBN 978-0-521-20527-6 paperback

Contents

Page
Preface v

CHAPTER I Preliminaries
1. Historical notes 1
2. Categories 2
3. Groups 8
4. Topological spaces 9

CHAPTER II Topological groups

1. The category of topological groups 16
2. Subgroups and quotient groups 18
3. Products 25
4. Fundamental systems of neighbourhoods 26
5. Separation axioms 29
6. Open subgroups 34
7. Connectedness 35
8. Compactness 42
9. Profinite groups 51
10. Locally compact groups 55

CHAPTER m Integration on locally compact groups

1. Abstract integrals 61
2. Some results on approximation 65
3. Convex cones 69
4. The existence of Haar integrals 72
5. The uniqueness of Haar integrals 73
6. Integrals on product groups 77
7. Unimodular groups 79

iii
CHAPTER IV Examples and applications
1. Extensions of Haar integrals 83
2. Examples of Haar integrals 84
3. Representations of compact groups 92
4. Topics for further reading
102
References
103
Index
104
Index to numbered results
108

IV
Preface

In 1969, and again in 1972, I gave an introductory course on topo-

logical groups at the University of London. The audience consisted large-
ly of first-year postgraduate students in algebra or number theory and
the course was designed to meet their needs, which I took to be a know-
ledge of basic facts about various general types of topological groups and
enough about the Haar integral to enable them to appreciate its use in such
contexts as representation theory and algebraic number theory. Some of
the students had little or no background in topology and for this reason
topological concepts were developed ab initio, whereas the rudiments of
group theory were assumed.
After the 1969 course there was some demand from new students
for copies of the notes and so it was decided to produce a duplicated set
of notes of the 1972 lectures. I am extremely grateful to Robert Coates,
Michael Rutter and Anthony Solomonides for all their hard work in pre-
paring the mimeographed version on which the present notes are based.
I have made a number of minor changes and amplified some passages when
this seemed advisable for a wider audience. I have also added an informal
section on the representation theory of compact groups which I intended to
include in the course but for which there was no time. Its purpose is to
illustrate one practical application of the Haar integral and to encourage
further reading.

King's College, London P. J. Higgins

1974
I • Preliminaries

1. Historical notes
Abstract topological groups were first defined by Schreier in 1926,
though the idea was implicit in much earlier work on continuous groups of
transformations. The subject has its origins in Klein's programme (1872)
to study geometries through the transformation groups associated with
them, and in LieTs theory of continuous groups arising from the solution
of differential equations. The 'classical groups' of geometry (general
linear groups, unitary groups, symplectic groups, etc.) are in fact Lie
groups, that is, they are analytic manifolds and their group operations
are analytic functions. On the other hand, Killing and Cartan showed
(1890) that all simple Lie groups are classical groups, apart from a finite
number of exceptional groups.
In 1900 Hilbert posed the problem (No. 5 of his famous list) whether
every continuous group of transformations of a finite-dimensional real or
complex space is a Lie group. The twentieth-century habit of axiomatising
everything led to a more abstract formulation of this problem. A topologi-
cal group is a topological space which is a group with continuous group
operations, and the question is: What topological conditions on a topologi-
cal group will ensure that it has an analytic structure which makes it a
Lie group? Since integration was a major tool in the study of Lie groups,
especially their representations, it became important to establish the
existence of appropriate integrals on general classes of topological groups.
This was achieved by Haar in 1933 for locally compact groups with coun-
table open bases. Von Neumann (1934) gave another proof for arbitrary
compact groups, making the representation theory of compact Lie groups
immediately available for all compact groups and so solving Hilbert's
problem in this special case. Haar's method of integration was extended
to all locally compact groups by Weil (1940). However, there are serious
obstacles to extending the representation theory to locally compact groups,
and it was not until 1952 that Hilbert's problem was settled by Gleason,
Montgomery and Zippin. Their answer can be formulated as follows:
a topological group is a Lie group if and only if it is locally Euclidean;
alternatively, it is a Lie group if and only if it is locally compact and
does not have arbitrarily small subgroups, that is, the identity element
has a compact neighbourhood containing no non-trivial subgroups.
Although the theory of topological groups was developed mainly
in order to study groups of Lie type and its impetus came from problems
in analysis, it soon proved to be useful also in purely algebraic contexts.
Certain algebraic constructions lead to groups having natural topological
structures which are somewhat pathological from an analyst's point of
view. Examples are power-series rings, Galois groups of infinite field
extensions, and p-adic groups. The pathology lies in the existence of
arbitrarily small subgroups, but in most important cases the groups are
actually locally compact and integration is therefore possible on them.
The algebraist must be familiar with these facts and this course is de-
signed to make them available to him. It will, I hope, also serve as an
introduction to topological groups and the Haar integral for students of
other branches of mathematics. However, the general flavour of the
development is more algebraic than is usual in such an introduction, and
the use of analytical arguments has been kept to a minimum. The only
pre-requisites for the first three chapters are a few facts from elementary
group theory. Chapter IV is less rigorous and demands more of the
reader.

2. Categories
Definition. A category 6 is a structure comprising the following
data:
(i) a class whose members A, B, C, . . . are called the objects
of 6;
(ii) for each pair of objects A, B, a set @(A, B), called the set
of morphisms from A to_ B (in G); we write f : A -• B to mean that
f €<3(A, B);
(iii) for each triple of objects A, B, C, a law of composition

e(A, B) x e(B, c) -e(A, c);

that is, for f : A -* B and g : B -* C, there is defined a 'composite'
morphism fg : A -+ C. (Note: we have adopted a right-handed notation
for morphisms which is contrary to the current practice of many authors.)

These data are subject to the following axioms:

I. Associative Law. If f : A -* B, g : B - * C and h : C -* D,
then (fg)h = f(gh).
II. Identities. For each object A, there is a morphism
e A € 6(A, A) such that for all f : A -• B, e.f = f, and for all g : C -* A,
ge A = g.

Examples, (i) The category S of sets. Its objects A, B, C, . . .

are sets, and the morphisms f : A -* B are functions (maps) from A to
B. Composition is ordinary composition of functions. The identities are
the maps

eA
A = tA : a *-*a, for a in A.
A '
(Note: in any category in which the morphisms are certain maps between
sets, and composition is the ordinary one of maps, associativity follows
immediately.)

(ii) The category 8 of groups. Its objects are groups, and the
morphisms A -* B are group homomorphisms. Composition and identi-
ties are as in S. (The point is that if f, g are group homomorphisms,
so is fg; and t. is always a group homomorphism.)

(iii) The category T of topological spaces and continuous map-

pings, defined as follows: Given a set X, a topology °U or^ X is a col-
lection of subsets of X, called 'open subsets of X', satisfying:
(a) the intersection of two open sets is open (whence so is any finite
intersection),
(b) an arbitrary union of open sets is open, and
(c) the empty set 0, and X itself, are open subsets of X.
We then call (X, cll) a topological space. Notation: if °\i is fixed, we
shall say simply 'X is a topological space'.
The objects of t are the topological spaces. Their morphisms
are mappings f : A -* B (A, B topological spaces) which are continuous,
i. e. such that for all U open in B, the pre-image Uf (={xeA; xfeU })
is open in A. Composition and identities are as in S. (Check: f, g
continuous => fg continuous.)
The language of categories enables one to define certain familiar
concepts very generally. This is helpful in comparing and contrasting
analogous situations in different mathematical contexts.

Definition. Let 6 be a category. Then f : A -• B in (B is an

isomorphism (in 6) if it is invertible in (£, i. e. if it has an inverse
g : B -• A in G such that gf = e_>, fg = e A . If such a morphism g
B _x A
exists, it is unique, and we write g = f . We also say that A and B
are G-isomorphic and write A ^ B.
e
Examples. The isomorphisms in S are the bijections ( 1 - 1
correspondences).
The isomorphisms in g are the group isomorphisms (bijective
group homomorphisms). If a group homomorphism f is bijective, its
_ "I

set-inverse f is automatically a group homomorphism.

The isomorphisms in JT are the homeomorphisms, i. e. bijective
maps f such that f is continuous and f"1 is continuous. There are con-
tinuous bijections in T which are not isomorphisms, i. e. their set-
inverses are not continuous.
Salient facts about 8. Subsets: If S Q T, the rinclusionT map
j : S -* T in S, (sj = s) is an injection and has the following universal
property:
if f :A-*T in S, and Af c s , then 3! f* :A-*S such that f = f*j.

Pictorially:
+- T
3 ! f* such that the diagram
commutes.

( 3 ! means 'there exists a unique... T)

 Quotient sets: If A, B are sets, a correspondence from A to
B is a subset y of A x B. If also 6 is a correspondence from B to
C, 6 c B x c, we obtain a correspondence yd Q A x C by letting
(a, c) e y6 if and only if 3b e B such that (a, b) e y, (b, c) e 6. We
also define y"1 Q B x A by the rule: (b, a) € y"1 if and only if
(a, b) € y. The identity correspondence L. : A -• A is {(a, a); a e A }.
In this notation a function f : A -• B is a correspondence such that
ff"1 D i A , and f^f c L B .
An equivalence relation on a set A is a correspondence y Q Ax A
such that y 2 i. (y is reflexive), y= y"1 (y is symmetric) and
yy c y (y is transitive). Given an equivalence relation p on A, A is
partitioned into equivalence classes (a) = |b eA; (a, b) e p ) . The
quotient A/p is the set of equivalence classes, and there is a canonical
map q : A -^ A/p, a K (a). This q is a surjection ('an onto map').
For f : A -* B, ff"1 is an equivalence relation on A whose classes
are the 'fibres' of f, namely the sets of form bf , for b e Af

Definition, ff" is the kernel of f, denoted by Ker(f).

The quotient map q : A -* A/p, where p is an equivalence rela-
tion, has the following universal property: if f : A -• B and
p c Ker(f) = ff"1, 3! f* : A/p -• B such that qf* = f.
q
• A /n
3! f* such that the diagram commutes.

Proposition 1 (S). (First isomorphism theorem.) Let f : A -* B

be any map. Let p be its kernel with quotient map q : A -* A/p. Let

A/p — — - - M
I = Af c B, with inclusion j : I -• B. Then (by the universal properties
for quotients and subsets), 3 ! f* : A / p - ^ I such that qf*j = f. Further,
f* is an isomorphism in S.

Products in S: If I, S are any sets and x : I -* S any map, we

often speak of x as a family (indexed set) of elements in S. We use the
notation {x. j . T for the family, where x. = ix, and we call I the index-
ing set of the family. For example a sequence of real numbers (xn)neN
is a function x : N -*R. If {A.} is a family of sets, we define:

n A = {{a } ; V i e I, a 6 A }.
iel x x 1€1
* l

If I is finite, 1 = {1, . . . , n } , then II A. is S-isomorphic with

iel *
A x A x ... x A .
1 2 n
Suppose that a = {a. }. is a member of A = n A.. We call
T
I I ei .T l
a. the i coordinate of a, and TT. : A -• A., a K a. is called the i
coordinate projection. These projections have a universal property: if
for each i e I we are given a map f. : B -* A., then 3 ! f : B -* A such
that fTr. = f. for each i. (This map is given by bf = {bf. }. ..)
We are thus led to the categorical concept of product. Let © be
a category. Suppose that I is a set, and for each i e I we are given an
object A. of 6. Suppose that A is also an object of G, with morphisms
7r. : A -• A. for i e l . We say that A is a product in 6 of the A., with
projections ir.9 if for every family of morphisms i f . } . ,, with
f. : B-*A. in G, 3 ! f : B-*A in 6, such that in. = f. for every i e l .
Such a product need not exist. However, any two such products
must be isomorphic in 6 by the 'standard argument' for universal pro-
perties, which we write out in full here (it will be left as an easy exercise
at all future occurrences).
If A is a product of the objects A., with projections TT., let us
(temporarily) write f., 7 r w ^ f to mean that the morphisms f. : B -* A.
determine (uniquely) the morphism f : B •• A satisfying fn. = f. for all
iel. If A' is another product of the A., with projections 7r! : A' -* A.,
let

77., vlss^a and TT!, 77.*\/\/*^3 .

1' 1 1' 1

Then a : A -* AT s a t i s f i e s an] = 77., and /S : AT -• A s a t i s f i e s 077. = 77!.

It follows that a/377. = 77., for all i e l , and hence

77., 77.-W^O?iS .

But clearly, 77., 77.w*-i and it follows from the uniqueness clause that
1 1 x\
a/3 = «... Similarly /3o? = 1 Thus A ^ AT and s i n c e airl = 77. and
A Pi. fQ 1 1

/S77. = 77!, the isomorphism respects the projections.

We may now say that S has products (i. e. any family of sets has
a product in S).
We shall always assume the Axiom of Choice: 'If the sets A.
f
(i € I) are non-empty, their product is also non-empty. This is equiva-
lent (given the other axioms of standard set theory) to Zorn's Lemma.

Definition. A partially ordered set is a set S together with a

correspondence y Q S x S such that: y 2 1 (y is reflexive), yy Q y
-1
(y is transitive), y n y QL (y is anti-symmetric). We usually write
b
a < b to mean (a, b) e y.
Let S denote a fixed partially ordered set.

Definition. T c s is a chain if Vt, tf e T, either t < tf or

t' < t (i. e. T is totally ordered by the induced relation y n T 2 . )

Definition, x e S is maximal in S i f V y e S , y > x = > y = x.

Definition. T £ S is bounded above in S if 3 s e S such that

Vt € T, t < s.

Definition. If every chain in S is bounded above in S, we say

that S is inductively ordered.
Zorn T s Lemma asserts: 'Every inductively ordered set has a
maximal element.'
N. B . , it is not enough to know that all countable ascending chains
(s < s . . . < s . . . ) are bounded above. E.g. , take S = all countable
subsets of R, ordered by Q. Any countable union of sets in S is also
in S, hence countable ascending chains are bounded above in S (by their
unions). However, no set in S can be maximal because, R being un-
countable, we can always adjoin one more element.

3. Groups
We state some familiar facts about groups in categorical language.
Subgroups. Let G be a group. Let H be a subset of G with a
group structure defined on it; then H is a subgroup of G if and only if
the inclusion map j : H -* G is a morphism of groups. There is at most
one group structure on a given subset H for which this is the case. Sub-
group inclusion j : H -* G has a universal property, as for sets: if
f : L -• G is a morphism in 9 such that Lf c H, then 3! f* : L -• H in
S such that f*j = f.
Quotient groups. For a morphism f : G ~* H of groups it is usual
to define the kernel of f as the normal subgroup {x e G; xf = e } of G,
where e = e^ denotes the identity element of H. This is in conflict with
-i
our earlier definition in S, namely, Ker f = ff , but the two are closely
related, each determining the other. (If we write K = lx eG; xf = e },
then K is the equivalence class of ff"1 containing e^. On the other
hand, ff"" is just the equivalence relation on G whose classes are all
the cosets of K.) There is no danger in using the same name, Ker f, for
both concepts - the context will always make it clear which is intended.
For any normal subgroup N or F, the cosets of N are the equivalence
classes of the equivalence relation p defined by apb <=» ab e N. It
is usual to write G/N for the quotient set G/p in this case. If
q : G -* G/N is the corresponding quotient map, then there is a unique
group structure on G/N such that q is a morphism of groups. We shall
always give G/N this structure. The quotient map q : G -* G/N in g
has a universal property, as for sets: if f : G -• H is a morphism in 8
such that N C K e r f (i.e., Nf = {e}), then a! f* : G/N -* H in g such
that qf* = f.

Proposition 1 (g). (First isomorphism theorem.) For f : G -• H

in g, let K = Ker f, and let I = Gf (a subgroup of H). Let^ q : G -• G/N
be the quotient map and j : I -• H the inclusion map. Then
3 ! f* : G/N-*I (in S) such that qf*j = f. Furthermore, f* is_a g-
isomorphism.
Products in Q. Suppose we are given for each i e I a group A.
with identity e.. Form A = n A. in S. Define e. = {e.}, and for
a = {a.}, b = {b.} eA, define a"1 = {a"1 }, ab = {a.b.}. (We note
that a., b. e A. =*• a? 1 , a.b. e A..) Then with these operations A is a
group with identity e A , and the projections IT. : A -• A. are group-
A 1 1
morphisms.
If now f. : B-* A. are group-morphisms, 3 ! f : B -• A (of sets)
such that in. = f.; it is given by bf = {bftf. } = {bf. }. One verifies that
f is in fact a group-morphism, and hence that A is the product in 8
of the A..
4. Topological spaces
Examples of topological spaces:
(i) Any metric space X under the distance topology, i. e. ,
Y c x is open if and only if Vy e Y, 36 > 0 such that the 6-ball with
centre y is contained in Y. (E. g. R n with the usual metric. )
(ii) Any set X with the discrete topology (all subsets are open;
this is the strongest topology on X, where if S , S are two topological
spaces with underlying set X, we say S is stronger than S , or S
is weaker than S , if the identity: S -» S is continuous, and strictly
^ 1 2
so if in addition the identity: S -• S is not continuous). Warning: some
authors use the terms weaker and stronger in the opposite sense.
(iii) Any set X with the trivial topology (0, X are the only open
sets; this is the weakest topology on X).
If X is a space with the discrete topology, or Y is a space with
the trivial topology, then any mapping f : X -* Y is continuous.
Subspaces. Let X be a topological space and Y £ X a subset.
Consider all sets of the form A n Y, with A open in X. These may be
taken as the open subsets of a topology on Y, the subspace (or induced)
topology on Y. It is the weakest topology such that the inclusion map
j : Y -• X is continuous. TSubspaceT will always mean 'subset with the
induced topology'.
 Example. The topological n-sphere S is defined to be the sub-
n+1 . ...
set {x; I x.1 = 1} of R with the subspace topology, R having
i=l
the usual metric topology.
Let Y be a subspace of X. Then j : Y -• X has the usual uni-
versal property: if f : Z -* X in F and Zf £ Y, then 3! f* : Z -• Y
in T such that f*j = f. (For we may construct f* in S, and continuity
follows from the observation that, if A is open in X, then (A n Y)f* =
(Afl)f*~l = Af"1 is open in Z.)

Exercise. The subspace topology on Y is the only one for which

j has this property (standard argument; Y is then determined up to T-
isomorphism).
Note the contrast with 9 J a subset of a group G need not possess
any group-structure making j a group-morphism, but if one exists, it
is unique; whereas a subset of a space X always has a topology (which
is not unique) making j continuous. Similar remarks apply to quotients
in ¥ which we now define.
Quotient spaces. Let p be an equivalence relation on a topologi-
cal space X, and let q : X -• X/p be the quotient map. The quotient
(or identification) topology on X/p is the strongest one for which q is
continuous, i. e. , A Q X/p is open if and only if Aq~ is open in X.
By Quotient (or identification) spaceT we shall always mean Quotient set
with the quotient topology1, and X/p will always denote this space. In
this situation, the quotient map q has the universal property that if
f : X -* Z is continuous and p c Ker f, then 3 ! f * : X/p ->Z in f such
that qf* = f. The quotient topology is the only topology on X/p such that
q has this property. (The proof of these facts is left as an exercise.)

Example. On R with the usual topology, define p by the rule:

(a, b) e p «=» a - b e Z. Then R/p = R/Z is a group and, with the
quotient topology, is homeomorphic to the subspace S of R 2 , via the
canonical bisection Z + t K (cos 2?rit, sin 27rit). (Exercise: Show that
this is a homeomorphism.)
N. B. The first isomorphism theorem is false in T. For example,
the identity map S -* S , where S is a stronger topology than S over

10
the same set X, is a continuous bijection but not a ct-isomorphism.
Thus some care is required in this last exercise.
We will shortly prove a modified form of the first isomorphism
theorem. (See Proposition 1 (2") below.)

Definition. If X is a topological space, Y Q X is closed if its

complement in X is open. Closed sets satisfy the following properties:
X, 0 are closed;
if A, B are closed, so is A u B;
if A.1 is closed for each i e I, so is n A..
iel 1
The above may be taken as alternative axioms for a topology on X in
terms of closed sets. Then f : X -• Y is continuous if and only if for
all B£Y, B closed in Y=> Bf"1 closed in X.

Definition. Let X, Y be spaces. A map f : X -• Y is said to be

closed (respectively open) if for all A c x, A closed =* Af closed in Y
(respectively A open => Af open in Y).

Exercises, (i) The projection R2 -»R, (x, y) h x , is open but

not closed.
(ii) The injection R-+R 2 , x K (x, 0), is closed but not open.
(iii) An isomorphism in 7* is open and closed.
(iv) If a continuous bijection is open (or closed) it is a T- isomor-
phism.
(v) If a continuous sur jection f : X -• Y is open (or closed) it is
a quotient map (that is, we can find an equivalence p on X and a homeo-
morphism h : X/p -• Y such that the following diagram commutes:
f

X/p

where q is the 'genuine' quotient map X -• X/p).

(vi) If a continuous injection f : Z -• X is open (or closed), it is
an inclusion map (that is, 3T c x and a homeomorphism h : Z -• T,

11
where T has the subspace topology, such that the following diagram
commutes: 17 t _v

where j : T -* X is the inclusion of T as a subspace in X.

Note: Quotient maps and inclusion maps are not themselves necessarily
open (or closed).

Proposition 1 (T). Let f : X -• Y be a morphism in T. Let^

p = Ker f (= ff"1), let q : X -* X/p be the quotient map (where X/p
has the quotient topology), and let Z = Xf (as subspace of Y), with
j : Z -* Y the inclusion map. Then a! f * : X/p -• Z in_ T such that
qf*j = f. If f is open (or closed) then f* is a T-isomorphism.

Proof. The existence and uniqueness of f* in 2" follow from the

universal properties of subspaces and quotient spaces. It is clearly a
bisection.
Suppose now that f is open. Let A be open in X/p, i. e. , let
Aq~ be open in X. Then Aq" f is open in Y, and so Aq" fj~ is
open in Z (because j is continuous). But qf*j = f, so fj~ = qf*, and
Aq~ qf* = Af* is open in Z. Hence f* is open and is a continuous bi-
jection, so it is a homeomorphism.
If f is closed, we show in a similar way that f* is closed, hence
a homeomorphism.
(We use the facts that Sjj" 1 = S, Tq"Xq = T, for all S Q Z,
T C x/p.)
Bases for topologies
Definition. If X is a topological space, a basis for the open sets
of X is a family {B.}. _, where each B. is an open subset of X, such
that any open subset of X can be written as the union of a subfamily, say
u B., for some J C I , (If J is empty, we define the above union to be
iej l
the empty set.)

12
 Examples, (i) In a metric space, all open balls; (ii) in R ,
all open rectangles; (iii) in R , all open discs with rational centres, and
radii of the form 1/n, n e Z.

Note: If X, Y are spaces and {B.} is a basis for the open sets
of X, a map f : Y -* X is continuous if and only if every B.f is open
in Y.
We can also define a basis for the closed sets of X to be a family
{D. } of closed subsets such that any closed set can be written as n D.
1
iej *
for some J Q I. (Here, if J = 0, we define the intersection to be the
whole space X.)
If {B. }. T is a family of subsets of X, there is a topology on X
with the B. as open basis if and only if the following TBasis Conditions1
are satisfied:
Bl u B. = X ;
iel x
B2 V i, j € I, 3 K c i such that B. n B. = u R ,
1 J
keK K
and then the topology has precisely all sets of the form u B., for JCi?
iej l
as open sets.
Let {C } * be an arbitrary family of subsets of X. Consider
all sets of the form

... nca(n), with

i. e., all finite intersections of the sets C , where we interpret the empty
intersection to be X. Then clearly these sets satisfy Bl and B2 above,
and so form a basis for a topology "U on X. We say that the sets C
form a sub-basis for U It is clear that °U is the weakest topology on
C
X such that all the C are open. If X has the above topology U, a
map f : Y -* X (Y a topological space) is continuous if and only if every
C f"1 is open.
Products in T. Suppose that {X.}. - is a family of topological
spaces. Let X = IUC in S, with projections ^ i X ^ X . . Let U be
the topology on X with sub-basis all sets of the form Y.7r7 , with Y.
open in X.; then clearly *U makes all the n. continuous; it is in fact
the weakest topology for which this is so.

13
 Now let f. : Z -*X. be continuous maps. Then a! map
f : Z -*X such that for all i € I, f7r. = f.. For each set Y . ^ 1 of the
sub-basis, we have (Y.TT7 )f~ = Y.f7 , which is open in Z (because f.
is continuous). It follows that f is continuous. This shows that in the
category T, (X, °lt) is a product of the given objects X.. Since products
in T are unique (up to ct- isomorphism) we see that °\i is the only topology
on the product set which has this universal property in T. We call this
topology 'U the product (or Tychonoff) topology on X. A basis for the
open sets of "U is obtained by taking all finite intersections of sets
Y.TT"1, with Y. open in X.. Now Y.TT"1 = II Yl, where Y! = Y. and

Y! = X. for j =£ i. Thus a finite intersection of such sets is precisely a

set of the form T = II T., where each T.l is an open subset of X.l and,
iel l
for all but a finite number of i, T. = X.. The open sets of the product
topology °li are therefore all unions of sets T as above.

Examples. (i) R is the product in T of n copies of R.

(ii) Definition. The n-dimensional torus T n is the product
X
VS x s1 x x s ^ i n r . In particular, T° is a single point; T 1 = S1;
n copies
T is the surface of a doughnut. In general T can be parametrized
by n complex numbers (z , . . . , z ) such that |z.| = 1 and T n is a
subspace of C n . (See Exercise (iv) below.)
Note: If {X.}. T is a family of discrete spaces then II X. is not in
general discrete. E. g., consider a countable number of discrete spaces
X., i e N, each having two elements. All open sets of X = II X. have
1
ieN l
an infinite number of elements, so X is certainly not discrete. (In fact
X is homeomorphic to the Cantor set.) For a more general result see
Exercise (v) below.
Exercises, (i) If X = II X. in 5T, then the projections
iel 1
7T. : X -* X. are open maps and each X. is (isomorphic to) a quotient
space of X under the projection map TT..

14
 Definition. In any category 6 if X = n X. and Y = n Y.
iel 1 iel 1
with projections £. : X - * X . and 77. : Y "• Y. respectively and if
f. : X. -• Y., then the morphisms |.f. : X -* Y. define a unique f : X -• Y
such that £.f. = frj. for all i e l . This f will be denoted by n f. and
1 1 1
iel l
will be called the product of the morphisms f..
(ii) If the f. are as above in some category (B and if in addition
we have morphisms g. : Y. -*Z. in ©, then II (f.g.) =(II f.)( n g.).
1 1 1
iel l l iel l iel l
(iii) If I is finite, and the morphisms f. : X. -* Y. in T are
open, then II f. is open.
iel l
Warning: If h. : Z -• X. are open in T then it is not true in general that
the induced map h : Z -* II X. is open, even when I is finite.
iel 1
(iv) If X. is a subspace of Y. for each i then IIX. is a sub-
space of IIY. (i. e. the product of inclusion maps is an inclusion map).
Warning: If Y. is a quotient space of X. with quotient map q. :X.-*Y.
then the map q = II q. : IIX. -* IIY. is not in general a quotient map.
iel l 1
*
A counterexample, due to Dieudonne, is as follows: Let Q be the
rational numbers considered as a subspace of R and let Y be the quo-
tient space obtained by identifying all the integers in Q. If q : Q -• Y
is the resulting quotient map and t is the identity map on Q then
iXq:QxQ-*QxY is not a quotient map.
(v) If X. (i e I) are discrete spaces then IIX. is discrete if
and only if at most a finite number of the X. have more than one point.

15
11-Topological groups

1. The category of topological groups

Definition. A topological group is a set G with two structures:
(i) G is a group with respect to • , " , e,
(ii) G is a topological space,
such that the two structures are compatible i. e., the multiplication map
JU : G x G -* G and the inversion map v : G -+ G are both continuous.
In this definition the set G x G carries the product topology.

Examples, (i) R regarded as an additive group with the usual

metric topology.
(ii) R*, the set of all non-zero real numbers, regarded as a
group with respect to multiplication, with the usual topology.
(iii) Any abstract group is a topological group with respect to
the discrete topology.
(iv) C*, the set of non-zero complex numbers, as a multiplicative
group, with the usual topology.
(v) S Q C, S = {z : |z| = 1}, is a topological group with
respect to multiplication and the subspace topology. More generally we
have:
(vi) If G is any topological group and H is a subgroup, then H
is a topological group with respect to the subspace topology. (Note:
(a) Z, the group of integers, is discrete as a topological subgroup of R.
(b) Q, the group of rationals, is not discrete as a topological subgroup of
R.)
(vii) The general linear group GL (R) (the set of all non-singular
n x n matrices with coefficients in R) is a group with respect to multi-
plication and has a topology induced by the inclusion in R n . Operations
2
of multiplication and inversion are given by n rational functions of
2 2

2n and n real variables, respectively, and are continuous. Thus

GL (R) is a topological group.
16
 (viii) GL (C) is a topological group in a similar way.
(ix) The algebra H of quaternions is a vector space over R
with basis 1, i, j , k. It is trivially a topological group with respect to
+ , -, 0, isomorphic with R . Multiplication of quaternions is given by
i2 = j 2 = k2 = - 1 , ij = - ji = k, jk = -kj = i, ki = -ik = j , and bilinearity,
the basis element 1 acting as identity element. If h = al + /Si + y] + 6k eH,
we define the conjugate of h to be h = al - /3i - yj - 6k, and the norm
of h to be |h| = (a 2 + /32 + y2 + 6 2 )*. Since hH = | h | 2 l , we see that
every non-zero quaternion h has a multiplicative inverse h" = |h|~ 2 h.
The fact that multiplication is associative can be checked directly on the
basis elements. Alternatively, if, for each quaternion h= a l + j6i + y] + 6k
we form the complex matrix M(h) = (a _jf* y ^.), we find that
M(h h ) = M(h )M(h ), and associativity in H follows from associativity
of matrix multiplication. The outcome of this is that H*, the set of non-
zero quaternions is a multiplicative group. It also has a topology inheri-
ted from H ^ R4 and it is a topological group because multiplication and
inversion are given by formulae involving only rational functions of the
coordinates.
(x) The 3-sphere S is embedded in H* as the subgroup of
all quaternions of unit norm. Hence S is a topological group with res-
pect to quaternion multiplication.
(xi) Let F be a field and let | | be a valuation on F, i. e., a
function from F to the set of non-negative real numbers satisfying:
(a) |x| = 0«=» x = 0,
(b) |xy|= |x||y|,
(c) |x + y| < |x| + |y|,
for all x, y, z e F. If we define d(x, y) = |x - y | then, by (a) and (c),
d is a metric which induces a topology on F. It is easy to see that F
is now a topological group with respect to +, -, 0, and that F* is a
topological group with respect to •, " , 1, the proofs being exactly the
same as for R. A case of particular importance in number theory is the
following. Let F = Q and let p be a fixed prime number. Every non-
zero rational number x can be written in the form x = p r m/n, where
m, n, r € Z and p Im, p^n. The integer r is uniquely determined by x.
If we define |x| = p~ r for x ± 0 and |o| = 0 , then | | is a valua-

17
tion, called the p-adic valuation, and the resulting topology on Q is the
p-adic topology. Thus every prime p provides two topological groups
(Q, +) and (Q*, •) with respect to the p-adic topology.
Note: In the definition of topological group one cannot omit the
assumption that inversion is continuous. For example, if we give Z
the topology with open sets 0, Z and all intervals [n, +<*>), then + is
continuous but - is not.

Definition. A morphism f : G -* H of topological groups is a

continuous group homomorphism.
Clearly topological groups and their morphisms form a category,
which we denote by T£.
Note: An isomorphism in T£ is a continuous group homomor-
phism having an inverse which is also a continuous group homomorphism.
It is not true in general that a group isomorphism which is continuous is
a T£-isomorphism. For example, let RT be the reals with the discrete
topology. Then the identity map IT -*R is a continuous group isomor-
phism, but its inverse, the identity map R -*Rf is not continuous.

Examples, (i) (R, +) <=* (R^°S, •), the isomorphism being given

by x ^ exp x from R to R p o s and x H- log x for its inverse.

(ii) The orthogonal group O2(R) = { ( ^ J *™ J) : 6 eR }

is a subgroup of GL (R). There is a 2"S-isomorphism O (R) -* S1 £ C*
given by (_* J K X + iy.

2. Subgroups and quotient groups

Any subgroup H of a topological group G is a topological group
with respect to the subspace topology, and the inclusion map j : H -» G
is a morphism of TQ. This morphism has the universal property: given
any f : L -* G in 7*8 with Lf Q H, 3 ! f* : L -• H in 2*9 such that
f*j = f.
A basic principle: Let G be a topological group and let g be a fixed
element of G. The constant map x K g and the identity map x H*X are
continuous maps from G to G, so they induce a continuous map
x H* (g, x) from G to G x G. Composing this with the continuous multi-

18
plication G X G ^ G we get a continuous map I : x ^ gx from G to
G, called left multiplication (or left translation) by g. This map has
inverse I _i which is also continuous, so / is a homeomorphism
G -* G (an isomorphism in 7*, but not in T£). Similarly all right trans-
lations r :x Kxg are homeomorphisms G -* G. As a consequence G
must be a homogeneous space, that is, given a, b e G there is a homeo-
morphism G -* G sending a to b. (L _l or r _x, will do.) Thus G
looks topologically the same at all points. We can now use translations
to transfer topological information from one point to another in any topo-
logical group, and this basic method is used in almost every proof in the
subject.
Note: Since not all topological spaces are homogeneous it is not
always possible to find a group structure on a given topological space
which will make it into a topological group.

Notation. If A, B c G, and g e G, where G is a topological

group, then
(i) Ag = Ar = {ag; a e A }; Ag is called the right translate
of A by g;
(ii) gA = Alg;
(iii) AB = u Ab = u aB;
beB aeA
(iv) A" 1 = {a" 1 ; a e A } .

Proposition 0. Let G be a topological group A, B Q G, g c G.

Then
(i) G is a homogeneous space;
(ii) A open implies Ag and gA open;
(iii) A closed implies Ag and gA closed;
(iv) A open implies AB and BA open;
(v) A closed and B finite implies AB and BA closed.

Proof, (i): already proved.

(ii), (iii): I , r , being homeomorphisms, are both open and
closed.
(iv): AB = u Ar, is a union of open sets and hence open. Simi-
b D

19
larly for BA.
(v) AB and BA are each the union of a finite number of closed
sets and hence closed.
Quotient groups and coset spaces. Let G be in T§ and let H be a sub-
group of G. Let G/H = {xH; x e G } , the set of left cosets of H. The
map q : G -• G/H defined by x K x H defines a quotient topology on G/H.
With this topology G/H is called the left coset space of G with respect
to_ H. The quotient map q is an open map; for if S is an open subset
of G, then Sqq" = the union of all left cosets of H which meet S
= SH, which is open by Proposition 0(iv),
and it follows that Sq is open in G/H by the definition of the quotient
topology. Now suppose that H is normal in G. Then G/H has a canoni-
cal group structure. If ju and \x' are the multiplications in G and G/H
and i>, vr the inversions in G and G/H respectively then \x\ vr are
uniquely defined by the commutativity of the following diagrams in S.

G- G

G/H x G/H- — G/H G/H G/H

We claim that G/H with this group structure (and the quotient topology)
is a topological group, i. e. JLIT and v% are continuous.

Proof, i>q is continuous and so by the universal property of q

in T there exists a unique map 6 : G/H -» G/H in T making q0 = yq.
However, vy satisfies the condition q^' = vq and so v' = 0 e T.
Also /-tq is continuous and so by the same argument /z' is con-
tinuous provided that q x q is a quotient map. In general it is not the
case that the product of quotient maps is a quotient map (see Exercise
(iv), p. 15); however, in our case q is open and so q x q is open (see

20
Exercise (iii), p. 15). In addition q x q is continuous and a surjection
and hence a quotient map. Thus MT is continuous and G/H is a topo-
logical group.

Proposition 1 (T§). Let G be a topological group, H a sub-

group, G/H the left coset space, q : G -* G/H the quotient map. Then:
(i) G/R is homogeneous,
(ii) q is open,
(iii) if^ H < G (i^_e. , H is a normal subgroup of G) then G/H
is a topological group, with the usual universal property: if f : G -• L
in r g and Hf = {e}, then 3 ! f* : G/H-^ L such that qf* = f.

G/H

(iv) Let f : G -• L in TQ have kernel K and image I = Gf £ L,

with quotient map q : G -• G/K and inclusion map j : I -* L; then
3! f* : G/K-»I in_ 2"g such that qf*j = f. The map f* is always a
bijection, and if f is open (or closed) then f* is a STg-isomorphism.

Proof, (i) is an exercise; (ii) and (iii) have been proved above;
(iv) follows from the corresponding results for 9 a n d ^*.

Examples, (i) Define f : R -* S1 by t K exp(27rit). Then f is

a morphism from the additive group of reals, with the metric topology, to
the multiplicative group S1 (considered as a subspace of the complex
numbers).
The composite map R -• S1 -*C=RXR has the continuous compo-
nents cos(27rt), sin(27rt), hence f is itself continuous. Now Ker(f) = Z,
and Im(f) = S1. Also f is open. (Proof: the open intervals (a, b) = I
(b > a) form a basis for the topology on R. Write Z(I) = b - a.
If 1(1) > 1, f(I) = S1 is open in S1.

21
 If 1(1) = 1, f(I) = S less a point; but C less a point is open in
the whole space, so f (I) is open in S .
If 1(1) < 1, f(I) is an arc without end-points, and so is open in
S , being the intersection of S with an open disc with the missing end-
points on its edge.)
Hence, by Proposition l(iv), R/Z ai S1 as topological groups.
N. B. f is not closed; e. g. , in + (1/n); n integral > 1} is
closed, but {f(l/n)} is not closed, as its closure contains f(0).
(ii) In a similar way we have a map g : C* -•R p o s in T'g,
given by z *+ |z | which is an open surjection, with kernel S , so

CVS1 ^R p 0 S in r s .

(iii) Again, the map C* -• S , given by z K z / | z | induces a

TQ-isomorphism C*/R pos -• S1. (See also Exercise (ii) p. 26 ).
(iv) Considering M (R) as real (n x n)-space, the determinant
map 6 : A N- det(A) is given by a polynomial, hence is continuous. By
restriction to the subspace of invertible matrices we have a continuous
homomorphism of groups

6 : GLn(R)-*R*,

with kernel denoted SL (R) (the special linear group). For x eR*, the
matrix diag(x, 1, . . . , 1) has determinant x, so 6 is a surjection. We
claim that 6 is open, and thus

GLn(R)/SLn(R) ^ R * in ?$.

(Proof that 6 is open: Let U be an open set in GL (R) and let

A be a matrix in U. For real t * 0, (tA)6 = tn(A6) * 0. Since U is
open in GL (a subspace of M (R)), there is an open interval J round 1
such that tA e U for all t e J. Now as t ranges over J, t takes every
value in some open interval J' round 1, so t (A6) takes every value in
some open interval J" round A6. Thus U6 D J" and U6 is open in R*.
(v) In GLn(R) the set of all'scalar matrices' tl is a closed
normal subgroup, isomorphic as topological group to R*. The quotient
group by this subgroup is (definition) the protective general linear group

22
PGL (R). It is, of course, a topological group with respect to the quotient
topology.

Corollary to Proposition 1. Let G be a topological group. Let

H, L be subgroups with H ^ L . Then the two topologies on L/H as
(a) quotient space of L and (b) subspace of G/H are the same. Hence,
if H <J G, then every subgroup of G/H is isomorphic in T§ to a quotient
group L/H.

Proof. Let q : G -* G/H be the quotient map and let qT be the

induced map L -* Lq. By Proposition l(T) it is enough to show that qT
is an open map. So suppose that A is an open subset of L. Then
A = L n B where B is open in G. Since q is an open map (Proposition
1(5T8)), Bq is open in G/H, whence Bq n Lq is open in Lq. However,
since L is a union of left cosets of H, we have Bq n Lq = (B n L)q = Aq,
so Aq is open in Lq, as required.

Exercise. Prove that if H O G, L <| G and H £ L , then

(G/H)/(L/H)sG/L in r g .
Warning: The analogue in !Tg of the third isomorphism theorem
of group theory would say that if A and H are subgroups of the topo-
logical group G, with H < G, then A/(A n H) ^ (AH)/H in r g . This
is false, as the following example shows. Let G = R, H = Z, A = AZ,
where A is irrational. Then A n H = {0 }, so A/(A n H) = A is iso-
morphic to Z with the discrete topology. However A + H = Z + AZ
is dense in R, i. e. , every open set in R meets it. (Proof: If an addi-
tive subgroup C * 0 of R has a least positive member d then C = dZ.
If it has no least positive member then it contains a strictly decreasing
positive sequence c > c > c > . . . . Given any interval (p, q), C
contains some c with 0 < c < q - p, namely, c = c. - c , for suf-
ficiently large i. Some multiple nc of this element lies in (p, q), so
C is dense. In our case Z + AZ is not of the form dZ since this would
imply 1 = dm, A = dn, for some m, n e Z, whence A would be rational.)
It now follows that (A + H)/H = (Z + AZ)/Z is dense in R/Z ^ S1. It is
isomorphic to Z as a group, but it is not discrete since every non-empty

23
open set in S meets it in an infinite set, so none of its finite subsets
are open.
We briefly describe another notorious example which is perhaps
easier to visualise. It concerns subgroups of the two-dimensional torus
T = S x s which is a topological group with respect to the product
group structure and the product topology (see the next section for products
in 7*8). The function 0 : R -* T2 defined by t K (exp 2ffit, exp 27riAt)
A
2
is a morphism of topological groups. Its image I. is a subgroup of T
whose nature changes radically according to the value of the real constant
A. If A is rational, all is well; putting A = ±m/n where m and n are
coprime positive integers we find that Ker 0 = nZ and I is a topologi-
A A

cal circle which winds round the torus m times one way and n times the
other way. Indeed, the restricted map 0* : R -•I is a continuous, open
surjection, and therefore L ^R/nZ, which is ^S-isomorphic to S . We
A

leave the details as an exercise. If, on the other hand, A is irrational,

then Ker 0 = {0 } and, as a group, L is isomorphic to R. It winds
A A

round the torus infinitely many times, never passing through the same
point more than once. However, the topology on I does not match the
A
topology on R because, although I passes through the identity
A
0 (o) = (1, 1) only once, it passes arbitrarily close to it after winding
round the torus suitably often. In fact, every neighbourhood of 0.(0)
A in
2
T contains points 0.(t) for an unbounded set of values of t. Hence
A
any neighbourhood of 0.(0) in L corresponds to an unbounded subset of
A A

R, and the two topologies are quite different. Again we leave the detailed
proof of these facts as an exercise. (In order to find points 0. (t) close
A

to 0A(O) one looks for pairs of integers (s, t) such that s + At is close
to 0. There are plenty of such pairs because, as we showed above,
Z + AZ is dense in R.)
This example confirms the need for a topological condition in the
first isomorphism theorem. It also gives another example showing that
the third isomorphism theorem of group theory does not extend to topo-
logical groups. We put G = R x R, H = Z x Z and consider the line
A = {(t, At); t eR } in G with irrational slope A. Then A n H = {(0, 0)}
so A/(A
24 n H ) ^ A ^ R in TQ. However, (A + H)/H is, by the corollary
to Proposition 1, a topological subgroup of the quotient group G/H. We
shall see in the next section that G/H ^ T in T£, the isomorphism
being induced by the map 6 : (x, y) K (exp 2?rix, exp 27riy) from G to
T . The subgroup of T corresponding to (A + H)/H is the image of
A under 6, which is precisely the subgroup L described above. It
follows that (A + H)/H and A/(A n H) have different topologies though,
of course, they are canonically isomorphic in g.

3. Products
Let {G. }. be a family of topological groups. Their product
G = I1G. (as sets) has a natural group structure (product in 8) an(* a
natural topology (product in T). With this topology on G, a map
f : X -* G is continuous if and only if each f 77. : X -• G. is continuous.
The group operations ju, v are defined on G so that the following
diagrams commute for each i:

G x G-

77. X 77.
1 1

G. x Gf

whence, as 77., ju., v. are continuous for all i, so are \m. and i>77..
Thus M, v are continuous and G is a topological group.
Now given morphisms f. : H -• G. in 7*8, the unique induced map
of sets

f : H -* G with f77. = f. for each i

is a group-morphism (product in 8) and continuous (product in T)f hence

a 7*8-morphism. It follows that G is the product of the G. in TQ.

Examples, (i) Let R be the additive group of real numbers with

the usual topology. Then

25
 R n s* R x . . . x R (n copies)

in 2*9, that is, the product topology coincides with the metric topology
on real n-space. The reason for this is that every open ball in R con-
tains an open n-cube centred on the same point, and vice versa.
(ii) S is an Abelian topological group, whence so is
T = S x . . . x s 1 (n copies). (By definition T° is the trivial topo-
n 1

logical group with one element.) We observe that S =. R/Z in 7*9 or,
what is the same thing, the morphism 0 : R -• S1 with kernel Z defined
by t ^ exp 277it is open. The product morphism 0 n : R n -* T n is there-
fore open, and its kernel is Z n . It follows, by Proposition 1 (2"g) that
T n ^ R n / Z n in r g . More generally, if G=nG. in r$ and H. is a
normal subgroup of G. for each i, then H =IIH. is a normal topological
subgroup of G and G/H .= n(G./H.) in 2"g. One simply needs the facts
that a product of subspaces is a subspace of the product of the containing
spaces, and that any product of open maps is open.

Exercises, (i) If G = A x B in TQ, let A = {(a, e-JeG; aeA}.

0 D
Prove that A is a normal subgroup of G and that A ^ A, G/A ^ B
in rg.
(ii) Prove that C* ^ R p 0 S x s 1 in r g .
4. Fundamental systems of neighbourhoods
Definition. Let X be a topological space and let A Q X. An
element a € A is said to be an interior point of A if there is an open
set N of X with a e N £ A.

Definition. A neighbourhood of x e X is a subset of X containing

x as an interior point. (An open neighbourhood of x is then any open set
containing x.)

Definition. A fundamental system of neighbourhoods of x in X

(f. s. n. of x) is a collection $ of neighbourhoods of x such that every
neighbourhood of x contains a member of SF. If each member of $ is
open, we speak of a fundamental system of open neighbourhoods of x,
(f. s. o. n. of x). (E.g. in R the intervals (-1/n, 1/n) for n c N form
a f. s. o. n. of 0.)

26
 Clearly, if for each x e X we are given a fundamental system of
open neighbourhoods of x, ^(x) say, then u ^(x) is a basis for the
xeX
open sets of X.
Now suppose that G is a topological group and $ is a fundamental
system of open neighbourhoods of the identity element e. Then for xeG,
= {Ux; U e ^ } is a fundamental system of open neighbourhoods of
x (because r is a homeomorphism!).
Any fundamental system ^ of open neighbourhoods of e in G
has the following properties:

FN1: If U J e J , then a W e J such that W C U n V .

FN2: If a e U e J then 3 V e $ such that Va c u.
FN3: H U e J , then a V e J such that V" 1 V £ U.
FN4: If U c ff, x € G, then a V e J such that x" 1 Vx £ U.

Proof. In (1) U n V, in (2) Ua" , are open neighbourhoods of e,

so contain an element of 5\
(3): The map f : G X G - * G given by (a, b) ^ a~1b is continuous.
Thus Uf" is open, contains (e, e), and hence contains a set of the form
A x B, where A, B are open and contain e. Since A n B is an open
neighbourhood of e, a V e ( F so that V Q A n B. For this V we have
V x v c uf" 1 , that is, V ^ V C U.
(4): The map f : G -• G given by a H* X~ ax is continuous, so
1
Uf" is open and contains e, hence contains some V e 3\ For this V
we have U 2 V f = x - 1 Vx.
Notes, (i) Any fundamental system of open neighbourhoods of e,
say ff, also satisfies: VU e IF, 3V, W e ff such that V" 1 £ U , W 2 C u.
(We shall write W for WW when there is no risk of confusion with
W x W.)
(ii) The set IF of all open neighbourhoods of e is a fundamental
system of open neighbourhoods of e, and has the extra properties that if
U, V € $ then U n V e J and U" 1 e(F. These imply that every U € $
contains a symmetric W e ^ namely W = U n U" 1 . (A neighbourhood W
of e is called symmetric if W = W.)

27
 Proposition 2. Let G be an abstract group and let £F be a non-
empty collection of subsets of G, each containing e, such that £F satis-
fies the conditions FN1-FN4. Then there is a unique topology on G such
that G is a topological group and $ is a fundamental system of open
neighbourhoods of e.

Proof. Uniqueness is clear, because if $ is a fundamental

system of open neighbourhoods of e, then <B = {Ug; U e J , g e G ) is
a basis for the topology. So, for given SF, we first check the basis con-
ditions for (B (see p. 13).

Bl: $ is non-empty, so a U e J , e e U . Thus g e Ug, and

U Ug = G.
g€G
B2: Let U, V € SF, a, b e G, and let c e U a n V b .
(If Ua n Vb = 0, we are there.) Let u = ca" 1 e U, v = cb" 1 e V. Then
(by FN2) a u x , V1 e ff such that IJ^u C U, V i v Q V, whence U ^ Q Ua,
V x c c Vb. By FN4 we can choose W € 7, W £ U n V . Then
We C Ua n Vb. Hence the sets f U g - . U e ? , g e G ) do give a basis for
the open sets of a topology on G.
To show that G is a topological group under this topology, it is
enough to show that the map f : G x G -* G, defined by (a, b) K a" X b
is continuous. Since (B is an open basis, we need only show that (Ua)f~
is open, for U e J , a e G.
Let (b, c) e (Ua)f"1, i. e., b" 1 c = ua for some u e U. By FN2,
a V e i F such that Vu c u, and by FN4, 3 W e i F such that b"XWb Q V.
For this W we have b" 1 Wca" 1 c u. By FN3, 3 Z e J such that
X 1 1 1 X
Z" Z c w, and then b" Z" Zca" £ U. Thus (Zb)" (Zc) Q Ua, i. e.
1
Zb x Zc c (Ua)f" , which is therefore open.

Corollary. Any group G can be made into a topological group

by specifying, as fundamental system of open neighbourhoods of e, a set
*F of subgroups satisfying:

FS1: If U, V e 9r, then 3W e ff such that W c u n V.

FS2: If U e IF, x e G, then S V e f F such that x ' V x £ U.

28
 (Proof. FN2, FN3 hold automatically with V = U.)
Thus any collection of subgroups of G containing all conjugates of
its members and all finite intersections of its members (in particular any
chain of normal subgroups of G), defines a topological group structure
on G, in which the open sets are all unions of right (or left) cosets of the
given subgroups.

Examples, (i) Q (the additive group of rationals) is a topological

group under the p-adic topology. If, for t e Z, we define U.= {mp /n;p|n},
then we have a countable chain of (normal) subgroups

giving a fundamental system of open neighbourhoods of 0 for this topology,

(ii) In Q* under the p-adic topology, the subsets V. = 1 + U.,
— L L

for t c Z form a fundamental system of open neighbourhoods of 1.

(iii) For any abstract group G, let SF be the set of all subgroups
of finite index. Then $ determines a topological group structure on G
such that $ is a fundamental system of open neighbourhoods of e.
(iv) In any group, the terms of the lower central series (or the
derived series) similarly determine a topological group structure.
(v) If R is any commutative ring, let R(x) be the power-series
oo

ring, with elements all formal (infinite) sums Z a x , where a e R.

O
n n
Define additive subgroups U = xHR<x>, for all n. Then U 2U 2U 2 . . . ,
and this sequence defines a topology which makes R(x> an (additive)
topological group. (Exercise: R(x) is in fact a topological ring, i. e. ,
multiplication of power series is continuous with the topology thus defined.)
5. Separation axioms
Definition. Let X be a topological space.
(i) X is a T -space if for any pair of distinct points a, b in
I —
———
—
X there exists a neighbourhood of a not containing b.
(ii) X is a T -space (or a Hausdorff space) if for any pair of
distinct points a, b in X there exist neighbourhoods of a and b which
are disjoint.
Clearly X is a T -space if and only if every one-point subset is closed.
29
Note that every discrete space is Hausdorff, and every Hausdorff space
is a T -space.

Proposition 3 (T). If X is a topological space then the following

are equivalent:
(i) X is Hausdorff.
(ii) The diagonal map 6 : X -* X x X given by x K (x, x) is a
closed map.
(iii) For any pair of continuous maps f, g : Y -• X, their differ-
ence kernel Z = {y; yf = yg} is a closed subset of Y.

Proof. (i)=* (ii): Let Y be a closed subset of X and let

(a, b) i Y6. Either (1) a * b or (2) a = b and a ft Y.
In case (1) there exist disjoint open sets A and B containing
a and b respectively and then A x B is a neighbourhood of (a, b) in
X x x which does not meet Y6.
In case (2), since Y is closed, there exists an open set A con-
taining a which is disjoint from Y. In this case A x A is an open
neighbourhood of (a, b) = (a, a) in X x x not meeting Y6. Hence Y6
is closed and 6 is a closed map.
(ii)=* (iii): Let f, g : Y - ^ X be in f. Since X is closed and
6 is a closed map, A = X6 is a closed subset of X x X. Now f and
g induce a continuous map (f, g) : Y -• X x X and the difference kernel
Z of f and g is the pre-image under (f, g) of the closed set A, so
Z is closed.
(iii) => (i): Taking f and g to be the projections X x x - > X ,
their difference kernel A is closed. Hence if a * b are points of X,
there exists an open neighbourhood N of (a, b) in X x X which is dis-
joint from A. We may now choose open neighbourhoods A and B of a
and b respectively in X so that A x B is contained in N and so is
disjoint from A, giving A n B = 0 as required.

Proposition 3 (2*9). Let G be a topological group and $ ji

fundamental system of neighbourhoods of e in G. Then the following
statements are equivalent:

30
 (i) G is Hausdorff.
(ii) 6 : G •* G x G is a closed map.
(iii) For f, g : H -• G in ¥§, their difference kernel is a
closed subgroup of H.
(iv) For any f : H -* G in TQ, Ker f is a closed subgroup of H.
(v) {e } is a closed subgroup of G.
(vi) G is T , i. e., every one-point subset of G is closed.
(vii) n ff = { e } .
(viii) The intersection of all neighbourhoods of e is_ {e }.

Proof. (i) => (ii) follows directly from Proposition 3 (T) whilst
(ii) => (iii) requires only the extra result that the difference kernel of two
group homomorphisms is a subgroup.
(iii) => (iv): Kerf is the difference kernel of f and g, where
g : H -• G is defined by h K e for all h in H.
(iv)=* (v): Take f as the identity map on G.
(v) =^ (vi): G is a homogeneous space.
(vi)=» (vii): Let g^e be an element of G. Since {g} is
closed and e j£ {g} there exists U e $ such that U n {g} = 0. Hence
g i n % and so n $ = {e}.
(vii) =» (viii) is trivial.
(viii)=> (i): Let a * b be elements of G. Since ab" 1 * e there
exists a neighbourhood U of e which does not contain ab" 1 . We may
now find a neighbourhood V of e such that V"1V £ U. Then ab" 1 ^V~V,
so Vab" n V = 0 and Va n Vb = 0. But Va and Vb are neighbourhoods
of a and b respectively and so G is Hausdorff.

Definition. A space X is a T -space iij given any two distinct

points a and b in X, there exists an open set containing exactly one of
them.

Exercise. Every T topological group is Hausdorff.

Another property of spaces akin to the separation properties T ,
T , T is regularity. A space X is regular if every neighbourhood of
1 2
each point x contains a closed neighbourhood of x. This is easily seen
to be equivalent to the following condition: for every closed set A in X

31
and every point b £ A, there exist open sets U, V in X such that
A £c uu b eeVV and U n V = 0. Thus a regular T -space is Hausdorff
since we may take A to be a point.

Exercise. Prove that every regular T -space is Hausdorff.

In view of the fact that topological groups need not be T -spaces
it is somewhat surprising to find:

Proposition 4. (i) Every topological group G is regular,

(ii) Every left coset space G/H is regular.

Proof, (i) By homogeneity it is enough to show that if A is

closed and e i A then there exist open sets U, V with A £ U, e e V
and U n V = 0. Now the complement of A is a neighbourhood of e;
we can therefore find (by FN3) an open neighbourhood V of e such that
V"1V n A = 0. But this implies V n VA = 0, so we may take U = VA
which contains A and is open by Proposition 0(iv).
(ii) The proof is similar. Given a closed set A in G/H and a
point xH not in it, let A = Aq~ , where q is the quotient map G -*• G/H.
Then A is closed in G, and x £ A; we can therefore find an open neigh-
bourhood V of e in G such that V"1VxnA = 0, and hence Vx n VA = 0.
Now A = Aq" is a union of left cosets of H; therefore so is VA, and
we deduce that in fact VxH n VA = 0. It follows that the images in G/H
of the open sets VxH and VA are disjoint open sets containing xH and
A, respectively.

Proposition 5 (T). (i) If X is a Hausdorff space and f : Y -• X

is a continuous injection then Y is Hausdorff. In particular, subspaces
of Hausdorff spaces are Hausdorff.
(ii) If (X. }. T is a family of non-empty spaces, then X = nx.
is Hausdorff if and only if all the X. are Hausdorff.

Proof, (i) If a * b are points of Y then af * bf. Let A and

B be disjoint neighbourhoods of af and bf respectively; then their pre-
images are disjoint neighbourhoods of a and b.
(ii) Suppose that X is Hausdorff and let j e I. For each i € I
define f. : X. -* X. in T as follows: (a) f. is the identity map on X.;

32
(b) for i =£ j pick some fixed element x. e X. and let f. be the corres-
ponding constant map. These f. induce an injection f : X. -»X in T
and so by part (i) X. is Hausdorff.
Conversely, suppose that each X. is Hausdorff and that a and b
are distinct points in X, with coordinates {a.} and {b. } respectively.
Since a * b there exists an i such that a. * b. and hence there exist
disjoint neighbourhoods A and B of a. and b. respectively in X..
Clearly ATT7 and B7i7 are disjoint neighbourhoods of a and b in X.

Proposition 5 (2"g)« Let G, G. be topological groups and let H

be a subgroup of G. Then:
(i) G Hausdorff => H Hausdorff;
(ii) G/H is Hausdorff ^ H is a closed subgroup;
(iii) H Hausdorff and G/H Hausdorff => G Hausdorff;
(iv) IIG4 is Hausdorff <=* every G. is Hausdorff.
1

Proof, (i) and (iv) follow from Proposition 5 (2*).

(ii): By the definition of quotient topology, G/H is a T -space if
and only if H (and therefore all its cosets) are closed in G. If H is a
normal subgroup then G/H is a group and the result follows from Pro-
position 3 (7*9). If H is not normal we need a slight elaboration of the
argument of Proposition 3 (2"g), as follows.
If G/H is Hausdorff then it is T , so H is closed. Conversely,
if H is closed, and aH, bH are distinct cosets, it suffices to find a
neighbourhood W of e in G such that WaH n WbH = 0. Since a"1bH
is closed and does not contain e, we can find neighbourhoods U, V, W
of e in G such that U n a"xbH = 0, V~1V Q U and a' 1 Wa Q V. It is
easy to check that W has the required property.
(iii): If G/H is Hausdorff, H is closed in G by (ii). If also H
is Hausdorff then {e} is closed in H. Hence {e} is closed in G,
and G is Hausdorff.

Examples, (i) R and all its subgroups are Hausdorff.

(ii) Since Z is closed in R, S1 ^ R/Z is Hausdorff. (Alter-
natively, S is Hausdorff because it is a subgroup of the Hausdorff group
C*.)

33
 (iii) T n = S1 x s 1 x . . . x s 1 is a Hausdorff group.
(iv) R/Q is not a Hausdorff group since Q is not closed in R.
(v) If G is any topological group, let E be the closure of the
trivial subgroup, that is, the intersection of all closed sets containing e.
Then E is closed, and we shall show that it is a normal subgroup of G.
The sets E~ , x" E for x e E, and y"XEy for y e G are all closed in
G, being images of E under homeomorphisms G -* G; they also all
contain e. Hence E c E" 1 , E c x"1E for x € E, and E c y"XEy for
y e G. It follows easily that E" 1 c E, xE c E for x e E, and yEy"1CE
for y € G, whence E is a normal subgroup. The group G/E is a
Hausdorff group since E is closed.

Exercises, (i) Show that G/E is the universal Hausdorff group

on G, i. e. , for any morphism f : G -* H in 7*9, where H is Hausdorff,
3 ! f* : G/E -» H in r g such that f = qf*, where q : G -• G/E is the
quotient map.
(ii) Prove also that if G denotes the group G with its topology
replaced by the trivial topology then the map g h* (gE, g) from G to
(G/E) x G° embeds G as a topological subgroup in (G/E) x G°.

6. Open subgroups
Proposition 6. Let G be a topological group. Then:
(i) every open subgroup of G is closed and every closed sub-
group of finite index is open;
(ii) every subgroup of G containing a neighbourhood of e is_
open;
(iii) if H is a subgroup of G then G/H is discrete if and only
if H is open.

Proof, (i) If H is open in G then by Proposition 0 so are all

cosets of H. Also the complement of H in G is a union of cosets and
hence is open. Therefore H is closed. Similarly if H is closed of
finite index, its complement is the union of a finite number of cosets,
each closed. Hence H is open.
(ii) If H contains a neighbourhood of e it contains an open
neighbourhood U or e. Hence H = HU, which is open by Proposition 0.

34
 (iii) G/H is discrete «==* all points are open in G/H
<=* H and all left cosets of H are open
in G
^ ^ H is open in G.

7. Connectedness
Definition. A topological space X is connected if it is non-empty
and cannot be written X = A u B where A and B are non-empty, dis-
joint open subsets of X. Thus a non-empty space X is connected if and
only if the only subsets of X which are both open and closed are 0 and
X.
A disconnection of X is a pair of complementary, non-empty,
open-closed (i. e. open and closed) subspaces.

Corollary to Proposition 6. (i) A connected topological group

has no proper open subgroups.
(ii) A connected topological group is generated as an abstract
group by any neighbourhood of e (or by any non-empty open subset).

Proposition 7 (2"). (i) If X is connected and f : X -* Y is a con-

tinuous function then Xf is connected.
(ii) I1X. is connected if and only if every X. is connected.

Proof, (i) Let Z = Xf and suppose that A is an open-closed

subset of Z then Af~ is open-closed in X. Hence Af"1 is 0 or X
and so A is 0 or Z.
(ii) If X = IIX. is connected then so are the X. by (i). Suppose
now that all the X. are connected and that A is a non-empty open-closed
subset of X. We show that A = X.
Let a = {a.} eA and let j e I. We define 6 : X. -• X to be the
map whose components are the constant maps 0. : X. -• X. (x. *-* a.) for
i * j and the identity map on X.. The image X.0 is connected by (i) and
contains a. Hence An X.6 is non-empty and open-closed in X.6, so
A 2 X i . Therefore, if af differs from a in one component then aT e A.
Clearly, by induction, if af differs from a in a finite number of com-
ponents and a e A, then aT € A. Now A is open in X so there exists
a basic neighbourhood of a inside A. Suppose this is IIU. where

35
a. € U., U. is open in X. and for all but a finite number of i, U. = X..
Given any x € X, by changing a finite number of components of x we
obtain a point in IIU. and hence in A. Hence x e A and it follows that

Proposition 7 (2"S). Let G, G. be topological groups and let H

be a subgroup of G. Then
(i) G connected =» G/H connected.
(ii) H connected and G/H connected => G connected.
(iii) I1G. is connected <£=> every G. is connected.

Proof, (i) and (iii) follow from Proposition 7 (2").

(ii) Suppose that H and G/H are connected and that G has a
disconnection G = A u B, where we may assume that e € A. Since H is
connected so are all its left cosets. Now each coset meets either A or
B and so must be contained in one of them (because A, B are open-
closed). Therefore A and B are each a union of left cosets of H. If
q : G -* G/H is the quotient map, it follows that Aq n Bq = 0. Since q
is an open map, Aq and Bq are open-closed, so we have constructed
a disconnection Aq u Bq of G/H. This is a contradiction, so G is
connected.
Note: A subgroup of a connected group need not be connected.
E. g. , R is connected but Z and Q are not.

Proposition 8 (T). If X is a topological space and x € X then

the union of all connected subspaces of X containing x is connected and
is closed in X.
(This subspace will be called the component of X at x and will
be denoted by Comp-.-x or Comp x.)

Proof. Let C , i el, be all the connected subspaces of X which

contain x, and let C = u C1. Then C * 0 since {x} is connected.
id
Suppose that C has a disconnection A u B where x e A. Since
A n C. * 0 and C. is connected we have A 2 C . for all i, which implies
that A 2 C, a contradiction. Hence C is connected.
Now let C be the closure of C in X, i. e., the intersection of

36
all closed subspaces of X containing C. We shall verify that C is
connected, which will imply that C £ C thus proving that C is closed.
Suppose that C = A u B is a disconnection of C with x € A.
Then A n C * 0 and C is connected, so A 2 C . Now A is closed in
C which is closed in X, so A is closed in X. Since A 2 C , this
implies that A 2 C , so A u B is not a disconnection of C. Hence C
is connected.

Corollary. The components of X form a partition of X.

Proof. Certainly u Comp x = X since x e Comp x.

xeX
Suppose that Comp x n Comp y * 0, say z e Comp x n Comp y.
Since Comp x is connected and contains z we have Comp x Q Comp z.
But this implies that x is contained in Comp z, so Comp z c Comp x.
Hence Comp x = Comp z = Comp y.
N. B. Components are not in general open, so a single component
and its complement do not in general give a disconnection of a space. If
a space has only a finite number of components then these are all open
and closed. In this case every division of the components into two classes
gives a disconnection of the space.

Definition. A space X is said to be totally disconnected if each

component of X has just one point, i. e., there do not exist connected
subspaces containing more than one point. E. g., a discrete space is
totally disconnected. (But note: totally disconnected does not imply dis-
crete! )

Proposition 9 (T). Let X, X. be topological spaces, with

I
X.l * 0 for each i. Then
(i) if X is totally disconnected then every subspace of X is^
totally disconnected;
(ii) ITX. is totally disconnected if and only if each X. is totally
disconnected.
Proof, (i) is obvious.
(ii) Suppose that I1X. is totally disconnected and X. * 0 for

37
each i. Let A. be a connected subspace of X. for each i. Then
IIA. is connected (by Proposition 7 (T)) and is a subspace of IIX. (see
Exercise (iv), p. 15), thus can only have one point. Therefore each A.
has only one point, i. e. , each X. is totally disconnected.
Conversely suppose that X. * 0, and X. is totally disconnected
for each i. Let A c nx. be connected. Then Air. is a connected sub-
space of X. (by Proposition 7 (T))9 so has only one point. Therefore
A has only one point, i. e., nx. is totally disconnected.

Corollary. Any product of discrete spaces is totally disconnected.

Proposition 9 (^g). Let G, G. be topological groups, and H

a subgroup of G. Then
(i) if G is totally disconnected then so is H;
(ii) if G/H and H are totally disconnected then G is totally
disconnected;
(iii) IIG. is totally disconnected if and only if each G. is totally
disconnected.

Proof, (i) and (iii) are contained in Proposition 9 (2").

(ii) Let A be a connected subspace of G. Let q : G -* G/H be
the quotient map. Then Aq is connected, so has just one element, that
is, A is contained in some left coset of H. But each coset of H, being
homeomorphic to H, is totally disconnected. Hence A has just one
element.

Theorem 1. Let G be a topological group and let H be the com-

ponent of G at^e. Then H is a closed, connected, normal subgroup of
G, and G/H is a totally disconnected (Hausdorff) group. The components
of G are just the cosets of H. Every open subgroup of G contains H.

Proof. By Proposition 8, H is connected and closed in G. For

any h e H and any g e G , the sets h -1 H and g -1 Hg are connected,
being homeomorphic to H; also e e h~ H and e € g" Hg. Since H is
the component of G at e, this implies that h~1H £ H and g"XHg Q H
for all h e H and all g e G, which shows that H is a normal subgroup
of G. Now G/H is a topological group, Hausdorff since H is closed.

38
Consider its component at its identity element. This is a normal sub-
group of G/H and so, by the corollary to Proposition 1, it is isomorphic
in 2^9 to a group L/H, where L is a topological subgroup of G con-
taining H. Now L/H and H are both connected, so L is connected,
by Proposition 7 (2"9). It follows that L £ H and so the component of
G/H at its identity element is trivial.
The component of G at an arbitrary point x is just xH, because
I x is a homeomorphism from G to G. Similarly, in G/H, the com-
ponents are just the cosets of the trivial subgroup, i. e. , one-point sets.
Hence G/H is totally disconnected.
Finally, if U is an open subgroup of G, then U is open and
closed (Proposition 6) and contains e. Hence UfiH is a non-empty,
open-closed subset of the connected space H, and so must be the whole
of H, i. e. , U 2 H.
Note: Any totally disconnected group is Hausdorff because the
trivial subgroup is a component and is therefore closed (see Propositions
8 (T) and 3

Examples, (i) (R, +) is a connected group. For suppose that

X is a non-empty open-closed subset of R, with X * R. Let x e X
and let A be the set of all real numbers a > 0 such that X contains
the closed interval [x-a, x+a]. Then 0 eA and A is bounded above
(otherwise X = R). Hence A has a least upper bound m. If m e A
then [x-m, x+m] Q X and since X is open, X contains a neighbourhood
of each of x - m, x + m and so contains an interval [x-m', x+m'] with
m' > m, which is impossible. Hence m £ A and therefore one of x - m ,
x + m is not in X, say x + m £ X. Since X is closed, some neigh-
bourhood of x + m does not meet X, so 3 m ' , 0 < m' < m, such that
T
[x-m , x+m'] $Z X. This m' is an upper bound for A, and we have the
required contradiction since m is the least upper bound.
(ii) (R*, •) has two components, because R p o s , being isomor-
phic in 7*9 to (R, +), is connected, and is an open subgroup. The quo-
tient R * / R P is a discrete group of order two.
(iii) (R , +) is a connected group by Example (i) and Proposition
7 (2"*9)« Alternatively we may argue as follows. A space X is called

39
path-connected if, given a, b e X, there is a continuous function p : I-*X
(where I denotes the closed unit interval [0, 1] in R) such that Op = a,
lp = b. Any path-connected space is connected. For if X = A u B is a
disconnection of X, we may choose a € A, b e B and obtain a path from
a to b as described above. Then I = Ap u Bp is a disconnection
of I. This is impossible because I is connected (Exercise). Now in
R , any two points a and b can be joined by a line-segment, which is
a continuous image of I under the map p : t H* (1 - t)a + tb. Hence R n
is path-connected.
(iv) R n \ {0 } is connected if n > 2. Indeed, it is path-connected
because any two points can be joined to a third point by line segments not
passing through 0.
(v) S is connected if n ^ 1 since there is a continuous sur-
jection x K x / | x | from R n + 1 \ { 0 } to Sn. (Note: S° is a discrete
two-point space.)
(vi) T n is connected for all n > 0 since there is a continuous
surjection from R n to T . Alternatively, it is connected because it is
a product of connected groups S .
(vii) In a free group F, the lower central series F = F DF 3 . . .
is a chain of normal subgroups. (F is generated by all commutators of
weight n.) By the corollary to Proposition 2, F is a topological group
with respect to a topology in which {F } is a fundamental system of open
neighbourhoods of e. Now Magnus has proved that DF = {e } (see
'Combinatorial Group Theory' by Magnus, Karrass and Solitar (Inter-
science), Theorem 5. 7 and Corollary). Since the groups F are open in
F, they all contain the component of F at e (Theorem 1) and it follows
that F is totally disconnected.
(viii) Q, as a subgroup of R is totally disconnected. For let
a, b e Q with a < b and suppose that X is a connected subspace of Q
containing a and b. Then a r € R such that a < r < b and r £ Q.
The sets A = lx eX; x < r } and B = {x e X; x > r } are now easily
seen to form a disconnection of X.
(ix) Q, with the p-adic topology, has as a fundamental system of
open neighbourhoods of 0 the chain of subgroups Z 3 pZ 3 p2Z 3 . . .
3 p Z 3 . , . . The intersection of these subgroups is { 0 }, so Q is

40
totally disconnected as in Example (vii).
(x) The orthogonal group O is the group of all n x n real
matrices M such that MM = I . These matrices represent linear
transformations R -+R which preserve distances (and therefore angles);
the columns of such a matrix are orthogonal vectors of unit length. As
2

a subspace of R n , O inherits a topology which makes it a topological

group; it is a topological subgroup of GL (R). Now the determinant map
from O to R* is continuous, since det M is a polynomial in the n2
coordinates of M, and it is a group homomorphism. Its image is the
group {±1 } and its kernel is the group O of all orthogonal matrices
of determinant 1 (also denoted by SOn in the literature). Since {±1 }
+
is a discrete subgroup of R*, O is an open normal subgroup of O ,
so O is not connected. We shall prove that O is connected and is
n ^ n
therefore the component of O at the identity. Thus O has exactly
two components.
To prove that O is connected, we use induction on n. For n = 1
n
+
the assertion is trivially true. We assume, therefore, that O is con-
+ +
nected and consider the embedding O "* O ,, given by
M ;
Oil

It is clear that O is ^9-isomorphic to its image in O , , , so we may

I n i n • X

consider O as a topological subgroup of O + , . If we can prove that

the left coset space O + 1 / O is connected, then Proposition 7 (TQ)
and the induction hypothesis will imply that O ,, is connected, and the
+ +
proof will be complete. In fact, we shall prove that O + 1 / O is homeo-
morphic to the sphere Sn, which we have already shown to be connected.
For M e O , , , let 0(M) denote the last column of M. This is
a vector of length 1 in R n , so 0 is a function from O ,.. to Sn, and
it is clearly continuous. Any vector of length 1 in R is part of an
orthonormal basis (apply the Gram-Schmidt process to any basis starting
with the given vector). It follows that any column vector of unit length
is the last column of some orthogonal matrix, and hence the map
0 : O + 1 -+ Sn is a surjection. If M and M are orthogonal matrices

41
with the same last column, then M = M M where M is of the form
A . v i i

Since M e O ,, its last row has length 1, so u = 0 and

M € O . The converse is clear, and it follows that the fibres of

/ : O + , , -• S n are precisely the left cosets of O . Thus (j> induces a
<>
i I n

continuous bijection ty : O ,. /O -* S . Our proof will be complete if

we can show that i// is a homeomorphism and, by Proposition 1 (%), it
is therefore enough to prove that 0 is a closed map. We postpone the
proof of this final step until the next section so that we can use a com-
pactness argument.
Exercises, (i) Prove that any product of path-connected spaces
is path-connected.
(ii) Prove that GL (C) is connected.
(iii) Prove that GL (R) has two components.
(iv) Show that a quotient space of a totally disconnected space is
not necessarily totally disconnected.
(v) (Harder) Show that a quotient group of a totally disconnected
group is not necessarily totally disconnected.
(vi) Prove that if a finite topological group is connected then it
must have the trivial topology. (Note: There exist finite connected spaces
with non-trivial topology. An interesting example is the so-called 'small
circle': let X = {a, b, c, d} have open sets 0, X, {a}, {c}, {a, c } ,
{a, b, c} and {a, c, d}. Then X is clearly a topological space and
is in fact a quotient space of S under the map defined by: 1 H* b;
-1 Kd; e 2iflt K a for 0< t < | ; e 2?rit K c for | < t < 1. It follows
that X is path-connected. However, X has no group structure making
it a topological group since it is not a homogeneous space.)

8. Compactness
Definition. A space X is compact if it has the Heine-Borel
property: every cover of X by open sets, X = u X., can be reduced to
iel *
a finite sub-cover, i. e., there exist

i(l), i(2), . . . , i(n) el such that X = X. ( 1 ) uX. ( 2 ) u . . . uX. (n) .

42
Alternative definition (trivially equivalent to the above): X is compact
if every family {C. } of closed subspaces of X has the 'finite inter-
section property', which asserts that if each finite subfamily of {C. }
has non-empty intersection, then the whole family has non-empty inter-
section.
Notes: 1. In older books this property is called TbicompactnessT.
2. Many authors call a space 'compact' only if it is also Hausdorff.
3. X is said to be sequentially compact if (i) it has the Bolzano-
Weierstrass property: every infinite subspace S of X has a point of
accumulation (i. e., there exists b e X such that every neighbourhood of
b contains an infinite subset of S), or, equivalently, (ii) every sequence
of elements of X has a limit point (i. e. , given {x }, x e X, we can
find b e X such that every neighbourhood of b contains x for infinitely
many n). This property was called 'compactness' by earlier authors.
[Exercises, (i) If X is compact then it is sequentially compact, (ii) If
either (a) X is a metric space, or (b) X has a countable basis of open
sets, then X is compact if and only if X is sequentially compact. ]
4. A subspace of R is compact if and only if it is bounded and
closed. This is the content of the Heine-Borel Theorem. In particular
[0, 1] is compact. More generally, a subspace of a complete metric
space is compact if and only if it is closed and 'totally bounded' (i. e.,
for any given e > 0, it can be covered by finitely many open e-balls).
For properties of metric spaces, see 'Functional Analysis' by Kolmogorov
and Fomin (Graylock), Volume 1.
5. A discrete space is compact if and only if it is finite. (Exer-
cise! )

Proposition 10 (7*). (i) Any closed subspace of a compact space

is compact.
(ii) Any compact subspace of a Hausdorff space is closed.
(iii) The image of a compact space under a continuous map is
compact.
(iv) The union of a finite number of compact subspaces of a space
is compact.
(v) If C is compact, H is Hausdorff and f : C -* H is con-
tinuous then f is a closed map.

43
 (vi) If f : C "• H as in (v) is a bijection then it is a homeomor-
phism.
(vii) If f : C -• H as in (v) is a surjection then H has the
quotient topology.

Proof, (i) Let X be compact, and let Y Q X be closed; let

Y = u Y., where each Y. is open in Y; then Y. = Y n X. for some open
X. in X. The X. together with X\Y form an open cover of X; since
X is compact this has a finite subcover, from which we may construct
a finite cover of Y by deleting X\Y (if it appears) and replacing X.
by Y.. This is the required finite subcover of {Y. }.
(ii) Let H be a Hausdorff space, and let C be a compact sub-
space of H. Let x i C be fixed. We have to find a neighbourhood X
of x with X n C = 0. For every c e C we can find an open neighbour-
hood N of c and an open neighbourhood X of x in H such that
c c
Nc n Xc = 0. The sets Nc cover C, so by compactness there exist
finitely many c, d, . . . , g e C such that C £ Nc u Nd u . . . U N . Let
X = X n X J , , , n x ; this is a neighbourhood of x, and X n C = 0.
Hence C is closed.
(*") Exercise (look at the pre-image of an open cover of the image
space).
(iv) This is also a routine exercise.
(v) Let f : C -* H be continuous with C compact and H Haus-
dorff. If X £ C is closed, it is compact by (i), so Xf is compact by
(iii) and so closed by (ii).
(vi) From (v), f is closed, continuous and bijective; it is there-
fore a homeomorphism.
(vii) By Proposition 1 (T), since f is continuous and closed, it
induces a homeomorphism C/(ff"X) = H.
Examples, (i) We know that any bounded closed set in R is
compact. Any subset of R n defined by a set of equations or inequalities
of the form f(x , . . . , x ) = a, or < a, or > a, where f : R -* R is
i n n-4-1
continuous and a e R, is closed. In particular S is closed in R
since it is defined by x2 + x2 + . . . + x = 1. It is also bounded, so it

44
is compact.
(ii) Real protective n-space P n is defined as (R n \{0})/p
where xpy means that 3X^0 with x = Ay. The compact subspace S
of R n \ {0 } maps onto P n under the quotient map. Hence P n is
compact. (P is in fact a quotient space of S , obtained by identifying
diametrically opposite points.)
(iii) T n = R n / Z n is compact because the closed cell given by
0 < x. < 1 in R is closed and bounded, so compact, and it maps onto

(iv) The orthogonal group O is the subspace of R defined

by the equations

2x..x., = 6., for all i, k.

. IJ lk jk *"

It is therefore closed in R n . It is also bounded since 2 x2. = 1 for all

+ * 1J
j . Hence O is a compact group. The subgroup O is open, therefore
closed; so O + is also a compact group for all n ^ 1.
We can now complete the proof that O is connected (see Example
(x) of §7 above). We there constructed a continuous map 0 : 0 +, -• S n ,
and we only need to show that 0 is a closed map. But this now follows
from Proposition 10 (J*), part (v), since O , , is compact and S n is a
Hausdorff space. (A similar inductive argument can be applied to the
connectedness of the unitary and symplectic groups; see Chevalley [2]
for the details.)
Theorem 2 (Tychonoffs Theorem). Any product of compact
spaces is compact.

We shall give a proof using terms familiar to algebraists. At the

same time we shall describe the connection between ideals and filters so
that the analytically minded reader can translate into more familiar terms;
alternatively, he will find other proofs in [6] and [7].

Preliminaries. Let S be any set and let £ be any lattice of

subsets of S (i. e. £ is closed with respect to n and u, and 0 e <£,
S € £). An ideal of £ is a subset 3 c £ such that

45
 (i) if A, B € 3 then A u B e 3;
(ii) if A e 3 and B c A, B € <C then B € <J.
3 is a proper ideal if 3 ± £, i. e., if S ft 3.
The union of any chain of proper ideals is a proper ideal, so by
r
Zorn s lemma every proper ideal is contained in a maximal ideal.
('Maximal ideal* will always mean Tmaximal proper i d e a r . )

Lemma 1. If 3TC is a maximal ideal of £ and if A, Be<C and

A n B eSTC then either A e3TC or B eWL

Proof. We show that if Aj^m then B e l , If A ft TO, put

3 = \X e £ ; X C A u M, for some M €311}.

Then 3 is an ideal of £. Since A € 3 and Jit Q 3, 3 properly con-

tains "311, whence 3 = «&. In particular S e 3, so S = A u M, for some
M e WL. Hence B Q (A n B) u M, and since AnBcSTC and M € SfR,
this implies that B e 3TC.
Dually, if «C is any lattice of subsets, a filter in £ is a set
ff Q £ satisfying:
(i) A,Be^AnB€(F,
(ii) if A € ( F , B e £ , and B 2 A, then BeJ,
(iii) 0 t S.
By Zorn's Lemma, every filter is contained in a maximal one
(maximal filters are called ultrafilters). By duality we clearly have:

Lemma I 1 . If ff is an ultrafilter and A u B e f f , then A € ff

r
or B e 9 .
Now let X be a topological space. We have two lattices of sub-
sets of X, namely
(B(X), the lattice of closed subsets, and
0(X), the lattice of open subsets.
These are dual-isomorphic (i. e. unions and intersections are interchanged)
under the complement map: A »-*x\A. In this correspondence, filters in
(B(X) correspond to proper ideals in $TX), and ultrafilters correspond to
maximal ideals.

46
 Lemma 2. The following are equivalent conditions on a topological
space X:
(i) X is compact
(ii) If 3 is a proper ideal in 0(X), then

u (3) * X.

(iii) If ff is a filter in 6(X), then

n (5F) * 0.

Proof. (i)=» (ii): Let X be compact, and suppose that u(#) = X.

Then by compactness there are finitely many elements A , . . . , A e$
whose union is X. But 3 is closed under finite unions, so X e 3, and
3 is not a proper ideal.
(ii)^ (i): Assume (ii), and let X = uX. be any open cover of X.
Let 3 be the collection of all open subsets A of X such that A is
covered by finitely many of the X.. Then 3 is an ideal in ®(X), and
each X. e 3, so X £ u ( # ) . Hence, by assumption, 3 cannot be proper,
so X e 3- Thus any open cover of X can be reduced to a finite sub-
cover, and X is compact.
(ii) <=^ (iii) by taking complements.

Proof of Tychonoff's Theorem. We phrase the proof in terms of

ideals, but it can equally be written in terms of filters.
Let I be a set, and for each i e I let X. be a compact topo-
logical space. Let X = nx., and suppose that 3 is a proper ideal in
0(X). We show that u (3) * X, and then by Lemma 2 we see that X is
compact.
By ZornTs Lemma, 3 is contained in some maximal ideal 3TI.
It is therefore enough to show that u (3HX) ± X for an arbitrary maximal
ideal M.
For each i e I, define
<
Mi= {A eO(X.); ATT"1 e l ) ,

where TT. is the projection X -* X.. This is clearly a proper ideal of

47
 hence as X. is compact, u(3TC.)*X.. Choose x. eX.\upH.).
We claim that x = (x.) t u (3Tt). For if x € A e 3H, A is open, so there
is a basic open set Y such that x e Y Q A. (Hence Y e3Tl.) Now we
may choose finitely many indices i(l), . . . i(m) e I, and open sets

Z
i(r)CXi(r) <r = l, - m)

such that, if Y ( r ) = ^^rf^r) , we have

Y = Y m n . . . n Y, v.
(1) (m)
Then for each r, x e Y Q Y, y and therefore x., , eZ., y But by
Lemma 1, since Y e3Il, some Y, , e3TC, and then x., , eZ., , e^d., y
which contradicts our choice of the x.. Hence u pit) * X, and X is
compact.
The converse of TychonoffTs Theorem is trivial. If each X. is
non-empty, and IIX. is compact, then each projection is continuous,
hence its image is also compact; but this is the whole of X..

Proposition 10 (7*S). Let G, G. e 7"g, and let H be a subgroup

G
9L - Then
(i) G compact and H c l o s e d ^ H compact.
(ii) G compact^ G/H compact,
(iii) H compact and G/H compact^ G compact.
(iv) IIG. compact <=^> each G. compact.

Proof, (i), (ii), (iv) follow from Proposition 10 (T) and Theorem
2. For the proof of (iii) we need:

Proposition 11. Let G e ^ g , let C be a compact subset of G

and let A be an open set containing C. (We call such a set an open
neighbourhood of C.) Then there is an open neighbourhood V .of e in
G such that VC c A.

Proof. Every point x € C is an interior point of A, so there is

an open neighbourhood W of e such that W x Q A. Choose an open
x x
neighbourhood V of e such that V V QW. The open sets V x
X X X X X

48
cover C, hence so do a finite number; thus we may select V , . . . , V ,
n in
corresponding to x , . . . , x , with C £ u V.x..
n .=1 11
Now take V = n V., an open neighbourhood of e. Then
VC c W i x i u • . . u VV n x n , and each VV. c y.V. c w., so W . x . Q A.
Thus VC £ A as required.
Proof of (iii) in Proposition 10 (2*Q) above. Suppose that H and
G/H are compact. Write q for the projection G -> G/H. Let 3 be an
ideal of 6(G) such that u (3) = G. It is enough to show that G e 3.
For any x c G, xH is compact, and is covered by 3 (as u($)=G);
hence xH Q J, where J is a finite union of sets in 3. But 3 is an ideal,
so J e 3, and J is open.
By Proposition 11 we may choose an open neighbourhood V of e
such that VxH c j . Then VxH is open, hence is a member of 3, and
it is a union of cosets of H.
Let 3' = \A e ©(G/H) : Aq" 1 n V e 3 }. Then 31 is an ideal in
0"(G/H), non-empty since

(VxH)qq"1 = VxH e # => (VxH)q € # \

But (VxH)q contains (exe)q = xq, so for any x, xq€U(3 T ) and

u (<JT) = G/H. Since G/H is compact, G/H e 3\ i. e. , G e 3, and G
is therefore compact.

Proposition 12. In a compact group G, every open subgroup H

has finite index.

Proof. G is a union of disjoint cosets of H, which are open;

hence it is a union of finitely many of them. (Alternatively, G/H is
compact (Proposition 10) and discrete (Proposition 6), hence finite.)

Proposition 13. In a Hausdorff space H, any two disjoint com-

pact sets A and B have disjoint open neighbourhoods A', BT in H
(i. e. , there exist disjoint open sets A', B' with AT 2 A, BT 2 B).

Proof. Fix a € A. For each b e B there are disjoint open

neighbourhoods XL of a, V, of b. The V, cover B, hence so do a

49
finite number of them, say V, , . . . , V , corresponding to b , . . . , b
and U1 , . . . , Un . Let V = u V., U= n U.. Then U, V are disjoint
i=l * i=l *
open sets with a € U, B Q V, by construction.
Now, for each a e A, let S , T be disjoint open sets such that
a a
a e Sa . B C Ta. (We have just proved that such sets exist.) The Sa
cover A, hence so do a finite number, say S1 , . . . , Sm , corresponding
m m
to a , . . . , a and T , . . . , T . Then AT = u S., and BT = n T.
i m i m .=1 ] .=1 j
are disjoint open sets with A C A ' , B Q B\ as required.

Proposition 14. In a compact Hausdorff space H, the component

at any point x is the intersection of all open-closed sets containing x.

Proof. Let C. (i € I) be all the open-closed sets containing x,

and let C be their intersection. Let C*, C* be the complements in H
of C , C, respectively. Each C. contains comp(x), as the latter is
connected; hence C 2 comp(x). It is therefore sufficient to show that
C is connected (because this implies C £ comp(x)).
Suppose that C = X u Y, where X, Y are disjoint sets, open-
closed relative to C, with x e X. It is enough to show that X = C. Now
X, Y are closed in C, which is closed in H, so X, Y are closed in
the compact space H, and are therefore compact. Hence by Proposition
13 there are disjoint open subsets XT, YT of H, with Xf 2 X, YT 2 Y.
Now X' u Yf D C, and C* = uC*, so H is covered by
f
{X , Y', C* }. Hence by compactness H can be written as a finite union,
say

H = C* U C* U . .. u C * U X ' U YT
l 2 n ^v
= D * U X ' U Y1, K
'

where D = C n C2 n . . . n C n is an open-closed neighbourhood of x

and therefore contains C. It follows (using (1)) that C ^ D ^ X ' u Y ' . If
we write A = D n X\ B = D n YT, we deduce that D = A u B, with A, B
disjoint and open-closed in D, which is open-closed in H, whence A, B
are open-closed in H. But x e A , so C £ A , by definition of C, and
we now have C = C nA = C n D n X ' = C nX' = X, as required.

50
9. Profinite groups
Inverse limits. Let © be any category, and (I, <) a partially
ordered set. Suppose that for each i e I we have an object A. in G,
and that for each pair i < j we have a morphism f.. : A. -• A. in G
ji j i
satisfying the coherence conditions:
(i) each f.. : A. -* A. is the identity,
(ii)
v if i < Jj < k,' then f.kj.f..ji = f.ki. : A.k -• A..
I

We then call {A., f.. } an inverse system in G.

Let A c G . If S = {A., f.. } is an inverse system in <B, and if
for each i, a. : A -+ A. is a morphism in 6 such that whenever i < j
we have a.f.. = a., we shall call {A, a. } a cone over S. If also
J Ji i i
{B, r. } is a cone over S, a morphism of cones
0 : {B, r.} - {A, <r.}

is a morphism 6 : B -• A in C such that for every i, 0a. = T..

If {A, a. } is universal amongst cones over S, i. e. if for every
cone {B, T. } over S there is a unique morphism of cones
6 : {B, T. } -• {A, a.}, we shall call {A, a. } the inverse (or projec-
tive) limit of S. If such an A exists, by the standard argument it is
unique up to isomorphism (in fact {A, a. } is unique up to isomorphism
of cones), and we shall write

A = lim A. in this case.

Example. If I is trivially ordered (i < j if and only if i = j),

then A is the product in 6 of the A., and the morphisms a. : A-*A.
are the projections.
In the category 7*8, inverse limits always exist.

Proof. (This also works in S, T, g.) Let {A^ f^} bean

inverse system in T§. Let P = IIA. in 7*8> with projections n. :P-*A..
For each j < k, the difference kernel of TL f . and ir. : P -» A.
is a subgroup of P. Let A denote the intersection of all these subgroups,
and let a. :A-*A. be the maps induced by the ir.. Then clearly A is
a topological group, the a. are continuous group-homomorphisms, and
by construction, if j < k, then a. f. . = a.. In fact
J 9 j > k k] 3

51
 A = { a = {a } € P ; if j < k , a,f,. = a . } .

Now, given morphisms r. : B - * A . in 2*8 such that, whenever

j < k, T f . = T., let T : B -* P be the induced map, b K br = {br. }.
K K
J j i
Then clearly Br c A, and the universal property holds. (The uniqueness
of rT : B -* A subject to r'a. = r. follows from the uniqueness of
r : B -• P subject to TIT = T... ) Hence A = ljm A. in T§.
Proposition 15. Let G = limG. in 2*9. Then
1
(i) if all the G. are Hausdorff, so is G;
(ii) if all the G. are compact Hausdorff, so is G;
(iii) if all the G. are totally disconnected, so is G.
Proof, (i), (iii): If the G. are Hausdorff (resp. totally dis-
connected), then so is I1G. by Proposition 5 (resp. Proposition 9),
whence so is the subgroup G Q I1G..
(ii) If all the G. are compact Hausdorff, then so is I1G., by
Proposition 5 and Theorem 2. We claim that G is a closed subgroup of
P = IIG^ hence compact. For G. is Hausdorff, so the difference kernel
of two continuous maps from any space into G. is closed by Proposition
3. Hence by the construction of inverse limits, G is the intersection of
a family of closed subgroups of P, and so is closed.

Definition. G e f g is a profinite group, if it is of the form

lim G. in T§ for some inverse system of finite discrete groups G..

Theorem 3. Let_ G e TQ. Then G is profinite if and only if it

is compact and totally disconnected.

Proof. One way round is immediate, using Proposition 15, since

finite discrete groups are compact Hausdorff and totally disconnected.
Conversely, let G be compact and totally disconnected. Since all
components, i. e., all points of G, are closed, G is T , and so in fact
Hausdorff (Proposition 3 (2^8)). By Proposition 14, {e } = comp(e) is
the intersection of all open-closed neighbourhoods of e. We claim that
every open-closed neighbourhood of e contains an open (and so closed)
normal subgroup. To prove this, let A be any open-closed set with

52
e € A. Then (as G is compact Hausdorff), A is compact, and open.
Hence by Proposition 11 there is an open set V containing e, such that
VA c A.
Let W = V n V' 1 , so that W is open, e e W, and W = W"1.
Then WA = W"1A Q A, and so by induction, WnA c A for all integers
n. Since e e A, it follows that Wn c A, for all integers n. If H is
the subgroup generated by W, then H = u W Q A, and H is open by
neZ
Proposition 6(ii). Since H is open in the compact group G, it has finite
index (Proposition 12) and so has only finitely many distinct conjugates
x~ Hx in G. The intersection K of these is thus an open normal sub-
group of G, and K Q H Q A, as claimed.
Let N.(i el) be all the distinct open, normal subgroups of G.
The argument above shows that n N. = {e}. Since G is compact, and
id l
N. is open, the index of N. in G is finite for all i (Proposition 12).
Now order I by letting i < j if and only if N. 2 N., and put A. = G/N.,
a finite discrete group. If i < j then there exists a canonical morphism
f.. : A. -•A. given by xN. KxN.. The {A.) and {f.. } form an inverse
system of topological groups. Let A = lim A. in 2*8 with canonical
maps a. :A-*A.. Each A. is finite, so A is a profinite group; it is
constructed as the subgroup of I1A. consisting of those elements
a = {a. } = {x.N. }. such that x.N. Q x.N., whenever i < j . Now G
has quotient maps T. : G -* A. for all i, and T.f.. = r. for i < j . There-
i i 3 Ji i
fore there exists a unique r : G -• A in ^S such that TO. = r. for all
i €l, namely x -* {xN. }. r We show:
(i) r is an injection. For if x e G and xr = e, then xr. = e
for all i, whence x e ON. = {e }.
(ii) r is a surjection. Let a = {x.N.}. - be any element of A.
Each N. is open and hence closed, so each x.N. is closed. The inter-
section of any finite number of the N.'s, say N. n N. n . . . n N. , is
again an open normal subgroup N, for some k. Hence x,N, £x. N.
for a = 1, . . . , r, and so the intersection of these cosets is non-empty.
By the compactness of G there exists x e n x.N.. But then xN. = x.N.
.eI l I i l l

53
for all i € I and so xr = a.
We have now proved that r is a continuous bisection G -• A. However G
is compact and A is Hausdorff. By Proposition 10, r is an isomor-
phism in 2"9 and so G is a profinite group.

Theorem 4. Let G be any compact group and let G be the

Q

connected component of G jit e. Then

(i) G is a connected compact group;
(ii) G/GQ is a profinite group;
(iii) G is the intersection of all open, normal subgroups of G.
Proof. Put together Theorems 1 and 3.

Example, (i) Consider Z with the p-adic topology. The sub-

groups pnZ form a fundamental system of open neighbourhoods of 0.
Now n pnZ = {0 } and so Z is totally disconnected. However Z is
not compact because, e.g., the cosets pZ+12p 2 Z+p+12p 3 Z+p 2 +p+12...
have empty intersection (when p # 2). In other words the congruences

(1 - p)x s 1 (mod pn)

have no common solution x for all n, even though every finite number of
them have a common solution. (Exercise: show that Z is not compact
with respect to the 2-adic topology.)
However, if we form the p-adic integers Z = lim (Z/pnZ) we
obtain a compact group containing Z as a subgroup (exercise). In Z
the above congruences do have a common solution, and this was the
original purpose of HenselTs construction of the p-adic integers.

Exercise. Every element of Z is uniquely expressible in the

form

b + b P + b p2 +
o i 2 ••• (infinite sum)

where b. € Z, and 0 < b. < p. (Infinite sums are defined as limits of

finite sums in the usual way, relative to the p-adic topology.)

54
 Example, (ii) Infinite Galois groups. Let K be any field, L
any Galois extension of K (i. e. , algebraic, normal and separable), not
necessarily of finite degree.
Let G = Gal (L : K) be the group of automorphisms of L fixing
all elements of K. Clearly L is generated as a field by all finite Galois
extensions L. of K with L. Q L. For each L. the subgroup
N. = Gal (L : L.) of G is normal and of finite index in G. Furthermore
G/N. = Gal (L. : K) in S.
Define a topology on G by taking {N. } as a fundamental system
of neighbourhoods of e. (This may be done since N. n N. = N, , where
1 K
J
L, is the subfield of L generated by L. and L.. See §4 above.)
It is not difficult to show that there exists a one-to-one correspon-
dence between the closed subgroups of G, with respect to this topology,
and the fields between K and L; this correspondence has the properties
one expects from the Galois theory of finite extensions.
Exercises, (i) G = Gal (L : K) is profinite; G = lim G/N..
Crucial points in the proof are:
(a) If a. e Gal (L. : K) for each i, and a., a. agree on L. n L.
for all i and j , then 3 a e G inducing all a..
(b) The topology on G is the same as the inverse limit topology,
(ii) If G is in TQ, and if A and B are compact subsets of
G, then AB is compact.
(iii) If A is a closed subset of the topological group G, and B
is a compact subset of G, then AB is closed.
(iv) If H is a compact subgroup of the topological group G,
then the quotient map G -* G/H is a closed map.

10. Locally compact groups

Definition. A topological space X is locally compact if every
point of X has a compact neighbourhood.
(Note: Some authors require that every point has a closed, com-
pact neighbourhood. The two definitions coincide if X is Hausdorff, or
if X is regular. In particular, they coincide for topological groups, by
Proposition 4.)

55
 Examples, (i) R is locally compact,
(ii) Any compact space is locally compact,
(iii) Any discrete space is locally compact,
(iv) Q C R is not locally compact.
(v) Infinite-dimensional Hilbert space is not locally compact,
(vi) An infinite product of copies of R is not locally compact
(as is implied by the next proposition).

Proposition 16 (ST). (i) Any closed subspace of a locally compact

space is locally compact.
(ii) If X is locally compact, and f : X -* Y is a continuous,
open surjection, then Y is locally compact.
(iii) If X. is a non-empty topological space for every i € I, then
II X. is locally compact if and only if each X. is locally compact and
iel l x

all but a finite number are compact.

Proof, (i) Let X be locally compact, let Y be a closed sub-

space of X, and let y € Y. There exists a compact neighbourhood N of
y in X. Now N n Y is a neighbourhood of y in Y, and it is compact
since it is closed in N.
(ii) Let y € Y; then since f is surjective, y = xf for some
x € X. There exists a compact neighbourhood N of x in X. Since f
is continuous and open, Nf is a compact neighbourhood of y in Y.
(iii) Suppose the X. are locally compact and all but a finite
number compact. Let x e IIX. = X, where x = {x.}, x. e X.. There
exists for each i a compact neighbourhood N. of x. in X.. For all
but a finite number of i we may take N. = X.. Hence IIN. is a com-
pact neighbourhood of x in X. Conversely, suppose that X. * 0 for all
i, and that X = IIX. is locally compact. The projections TT. : X -* X.
are continuous, open surjections and so each X. is locally compact by
(ii). Now there exists an x e X and a compact neighbourhood N of x
in X. For all but a finite number of i we have NTT. = X., so almost all
the X. are compact.
Note: The image of a locally compact space under a continuous
map is not, in general, locally compact. (For any X, let X be X

56
with the discrete topology. Then X is locally compact and the identity
map X -* X is continuous.)

Proposition 16 (fg). Let G, G. be topological groups and let

1
H be a subgroup of G. Then:
(i) G is locally compact if and only if there exists a compact
neighbourhood of e;
(ii) G locally compact and H closed in G => H locally com-
pact;
(iii) G locally compact => G/H locally compact;
(iv) H locally compact and G/H locally compact =*• G locally
compact;
(v) IIG. is locally compact if and only if all the G. are locally
compact and all but a finite number are compact.
Proof, (i): G is homogeneous.
(ii) and (v) follow directly from Proposition 16 (T), whilst (iii)
needs only the additional information that the quotient map G -* G/H is a
continuous, open, surjection.
(iv): It is enough to find closed neighbourhoods U and V of e
in G with the following properties:
(a) V"1V £ u (therefore V ^ U ) ;
(b) xH n U is compact for all x € U (therefore xH n V is compact);
(c) if q : G "• G/H is the quotient map, then C = Vq is a compact
subspace of G/H.
Given such a U and V we argue as in Proposition 10 (TQ) (iii)
to show that V is compact. The argument is, however, a little more
difficult. Let # be an ideal of GXV) and suppose that u(#) = V. Let
5T = {A € <3(C) : Aq"1 n V e Q }. We shall deduce that u(3T) = C
Let x e V; then xH n V is compact and is covered by elements
of 3. Hence there exists J e $ such that J 2 xH n V. Now
xH n U Q J u (G\V) = X, which is an open set of G. By Proposition 11,
there exists an open neighbourhood W of e in G such that
W(xH n U) Q X. We may in fact choose W to be a subset of V. Now
WxH n V £ J; for suppose wxh = v where w e W, h e H and v e V.
Then xh = w"1v e V"1V £ U. Hence v = wxh e W(xH n U) c x. There-

57
fore v € X n V = J. It follows that WxH n V e 3.
Now Wx is open in G (because W is open), so (Wx)q is open
in G/H. Therefore (Wx)q n C e #T since Wxqq"1 n V = WxH n V e Q.
In particular xq e u(# f ) for any x e V, so u(# f ) = C. But C is com-
pact so C e $\ Therefore V e 3 and by Lemma 2 (p. 47) V is com-
pact.
It remains to find closed neighbourhoods U and V of e satis-
fying conditions (a), (b), (c). The subgroup H of G is locally compact;
therefore there exists a neighbourhood U of e in G such that H n U
is compact. Since G is regular (by Proposition 4) there exists a closed
neighbourhood U of e in G with U £ U . Since U OH is closed
in U n H it follows that U n H is compact. Again by regularity there
exists a closed neighbourhood U of e with U~ U Q U . Now if x e U
then xH n U = x(H n x" U). Since x" U is closed and contained in U ,
it follows that H n x~ U is closed in H n U and is therefore compact.
Hence xH n U is compact.
Now choose a neighbourhood V of e in G such that V V £U.
Since q is an open map, V q is a neighbourhood of H in G/H. Using
the regularity of G/H (Proposition 4(ii)), there exists a closed, compact,
neighbourhood C of H in G/H with C £ V q.
Now put V = V n Cq" 1 . It is easy to see that U and V satisfy
conditions (a), (b), (c).

Proposition 17. Let X be a locally compact Hausdorff space.

Let C be a compact subspace of X. Then every neighbourhood of C
in X contains a compact neighbourhood of C; i. e . , given N2U2C
with U open, we can find compact Cf and open UT such that

Proof. We first suppose that X is compact. Let N be a neigh-

bourhood of C; we may assume that N is open. Then X\N is compact.
By Proposition 13 there exist open sets V 2 C , W 2 X\N with V n U = 0.
Put CT = X\W; this is the required compact neighbourhood of C inside
N.
Now consider the general case. Each point c e C has a compact
neighbourhood N, whose interior we denote by M . Since C is compact,
c
58
it is covered by a finite number of the M , say C £ M UM U...UM .

If we now p u t X = N UN U...UN , then by Proposition 10, X

is compact and it is a neighbourhood of C. Given a neighbourhood N of

C, let N = N n X and apply the special case above: there exists C \
a compact neighbourhood of C in X , with CT £ N . Since X is a
T
neighbourhood of C in X and C is a neighbourhood of C in X it
T
follows that C is a neighbourhood of C in X.

Corollary. Any open subspace of a locally compact Hausdorff

space is locally compact.

Exercise. A subspace Y of X is said to be locally closed if

every point y e Y has a neighbourhood N in X such that N n Y is
closed in N. Prove the following:
(i) Closed sets and open sets are locally closed.
(ii) In any space X, a locally closed subspace is a closed sub-
space of an open subspace, and conversely.
(iii) In a locally compact Hausdorff space X, the subspace Y
is locally compact if and only if it is locally closed. (See Nachbin [7],
p. 10, in case of difficulty.)

Examples. The following are locally compact groups.

1. (R, +).
2. Any discrete group.
3. Any compact group.
4. GL (R) (since it is an open subspace of R n ).
5. SL (R) (since it is a closed subspace of GL (R)).
6. PGL (R) (by Proposition 16 since it is a quotient group of
GLn(R)).
The corresponding complex groups are also locally compact.

(Exercise. SL (R) is not compact if n > 2.)

7. (R p o s , • ) , (C*, •).
8. The Taffine group': this is the group of transformations
R n -»R n generated by all linear isomorphisms and all translations. The

59
translations form a normal subgroup isomorphic with R n , whose quotient
group is isomorphic with GL (R); both these groups are locally compact,
so the whole affine group is locally compact (by Proposition 16).
9. The additive group Q of p-adic numbers is locally compact.
This group can be defined as lim (Q/pnU), where U is the group of
rational numbers with denominator prime to p. It contains
Z = lim (Z/pnZ) as a subgroup, and in fact Z is an open, compact
neighbourhood of 0 in Q . (Exercise.)

60
Ill- Integration on locally compact
groups

1. Abstract integrals
We seek to construct an 'integral' on a locally compact group.
This may be done by a variety of methods; we wish to avoid measure-
theoretic methods and so we take as our model the construction of the
Riemann integral on (R, +). There we have the set (ft of all Riemann-
integrable functions (vanishing outside some interval), and a map
J : (ft -* R defined by

which satisfies:
(i) Linearity: J(Xf + /xg) = Xjf +
(ii) Positivity: f > 0 ^ Jf > 0.
(iii) Translation-invariance: if for some fixed a e R and all
x e R, g(x) = f (x + a) then

Jg= ff.

These three properties actually characterize the Riemann integral up to a

scalar multiple. (Exercise.) The same is true if we consider only con-
tinuous functions on R, and this suggests that we should try to construct,
for more general groups, an integral with similar properties for real-
valued continuous functions.
From now on we shall only consider locally compact spaces and
groups, and we shall always assume that they are Hausdorff.
For a space X, 3C(X) denotes the set of all continuous functions
X -•R with compact support, where

support(f) = closure of {x e X;f(x) * 0 }.

(The reader will notice that we have reverted to a left-handed notation

61
for functions X -*R. This notation is more familiar in such a context
and will cause no conflict since we shall not need to compose such func-
tions. ) Equivalently, 3C(X) is the set of continuous functions X ->R
which vanish outside some compact set.
JC(X) is a real vector space. It is also partially ordered by <,
where f < g means that f(x) < g(x) for all x. In fact it is a lattice:

f, g € JC(X)=^ sup(f, g), inf(f, g) e JC(X) (exercise),

(but it has no largest or smallest member). In particular, since

|f| =sup(f, -f),

f e JC(X)=^ |f| € JC(X).

Denote by 3C+(X) the set of all f e 3C(X) satisfying f > 0, i. e. , f > 0

and f * 0 (in other words, f(x) > 0 everywhere, f(x) > 0 somewhere!).
Then 3C+(X) is a convex cone in JC(X) (see p. 70 for the definition of
convex cone).
It is by no means clear that 3C(X) is not the trivial vector space,
so we hasten to establish

Proposition 18 (Urysohn's lemma). Let X be a non-empty

locally compact Hausdorff space. Let C Q U Q X, with C compact and
U open. Then there exists a continuous function f : X -»R such that
(i) f(x) = 1 for all x eC,
(ii) f(x) = 0 for all x i U,
(iii) 0 < f(x) < 1 for all x e X,
(iv) support(f) is compact and contained in U.

Proof. By Proposition 17 there exists a compact neighbourhood

C(0) of C inside U. Write C(l) = C. We observe that, by Proposition
17, if Cn is a compact neighbourhood of the compact set CT, then there
exists a compact neighbourhood Cw of CT lying inside C n , such that
CTt is a compact neighbourhood of Cm. We shall then say that Cm
separates Cr and C n .
Choose C(l/2) separating C(0) and C(l); then choose C(l/4)
separating C(0) and C(l/2), and C(3/4) separating C(l/2) and C(l).

62
Continuing in this fashion we get C(0) for every 9 of the form 0 = i / 2 n
with 0 < i < 2 n . Then for any real a, with 0 < a < 1, define

C(a) = n c(0) (where 0 is of the form i/2 n ).

0<a
Since each C(0) is closed, C(ot) itself is closed, and so compact. Define
C(a) = X if a < 0, C(a) = 0 if a > 1. Then for any a < j8, C(a) is
a closed neighbourhood of C(/3), compact if a > 0. Now define
f (x) = sup | a ; x e C ( f f ) l . Immediately we see that f (x) = 0 for all x
outside C(0), so that support(f) is compact and contained in U. Further-
more, f(x) = 1 for all x € C(l), and 0 < f(x) ^ 1 everywhere.
To complete the proof, we need to show that f is continuous. For
real j8 and y we observe that on the one hand

f(x) > p ^ x e n C(a) = D(j3),

and D()3) is a closed set; while on the other hand

f(x) > y$=$ x e C(a) for some a > y

^=^ x e interior of C(a f ) for some a1 > y
*=* x e u Int C(af') = E(y),
a*>y
and E(y) is an open set. Thus

y < f(x) < ) 3 ^ x e E(y) n (X\D()S)),

which is open; so the pre-image of a basic open set in R is open in X,

and f is continuous.

Corollary. If X is a non-empty locally compact Hausdorff

space, then

JC(X) * {0} and JC+(X) * 0.

Definition. A positive integral on X is a non-zero linear func-

tional \x : 3C(X) -•R such that jtx(f) > 0 whenever f > 0. We remark in
passing that both 3C(X) and R are partially ordered vector spaces, that
is, vector spaces over R with a partial order < satisfying

63
 (i) f < g ^ f + h < g + h,
(ii) f < g=* Af < Ag for any real A > 0.
The obvious morphisms to consider between such structures are the order-
preserving linear functions. Thus a positive integral on X may be des-
cribed as a non-trivial morphism of partially ordered vector spaces
from 3C(X) to R. Integrals do not, however, preserve the lattice opera-
tions.

Example. Let X = R and let a be any element of 3C+(R). Then

the functional JLI : 3C(R) -*R defined by

is a positive integral on R.

Definition. If G is a locally compact Hausdorff group, then a

right Haar integral on G is a positive integral \i 'invariant under right
translations'. This means that if f, g e 3C(G) and if f(x) = g(xa) for
all x e G, where a is a fixed element of G, then ju(f) = jLt(g).
Translation of functions introduces a new element of structure on
3C(G). For any function f : G -•R, and any a e G, we denote by fa the
a 1 a
function from G to R defined by f (x) = f(xa" ). We call f the right
a
translate of f by a. If f has support S C G then f has support Sa,
and it follows that if f € 3C(G) then also fa e JC(G). Thus the group G
acts on the vector space 3C(G) making it a (right) G-module over R,
which means that
(i) JC(G) is a vector space over R;
a
(ii) for each a e G, the map f *•• f is a linear map from 3C(G)
to itself;
(iii) fe = f for all f e 0C(G);
(iv) (f a ) b = fab for all f e JC(G) and all a, b e G.
For G-modules M, N, the natural definition of a morphism M-* N is a
linear map 0 : M -» N such that 0(fa) = (0(f))a for all f e M and all
a € G. If we let G act trivially on R (i. e., r a = r for r eR, a e G),
then R is also a G-module and we may describe a Haar integral on G
as an order-preserving morphism of G-modules from JC(G) to R.

64
2. Some results on approximation
Our aim is to prove that there exists an essentially unique right
Haar integral on any locally compact group G. To do this we shall try
to approximate functions f e 3C(G) by sums of translates of scalar mul-
tiples of a function with small support. (Cf. approximation by step func-
tions in the case of the Riemann integral.) The basis for such approxi-
mation is:

Proposition 19. Let f, F e 3C (G). Then there exist real num-

bers a , a , . . . , a > 0 and x , x . . . . . x e G such that
l' 2' ' n l' 2' ' n
x.1
l a . F ^ f.

For any such a. and x., and any right Haar integral jx on G,

^ (2 «.)
Proof. There exists t e G such that F(t) * 0. Let j8 = | F(t)> 0.
Then there exists an open neighbourhood V of e in G such that
F(x) > 0 for all x e Vt (by the continuity of F). Let C = support(f);
then C is compact, so is covered by a finite number of sets Vc (ceC),
say C £ Vc u Vc2 u . . . u Vc .
If x e V c . then x c ^ t e V t , so that F(xc~Xt) > 0, i.e.,
x
i -l
F (x) > 0, where x. = t c But f is bounded (the set of its values in
R is the image of a compact set!), say f(x) < M for all x e G. Hence
n x
lv/r -
I M p 1(x) > M > f (x)
X
~ M *
for all x € C (since x c some Vc^, and 2 -g-F l(x) > f(x) trivially for
x ft C. This proves the first part of the proposition.
x
If now f < X « . F i we have
x.
/i(f) < /i(2 a.F l)
X.

for any right Haar integral [x.

65
 Corollary. If F e JC+(G) then JU(F) > 0 for all right Haar
integrals JU.

Proof. Since M is non-trivial, there exists f e 3C(G) such that

0. Since either ju(f) ) 0 or ju(-f) > 0, we have

|| /x(-f)) > 0.

By the Proposition, there exist a. > 0 such that

0 < M(|f|)^(Z

So M(F) > 0.
For f, F € 3Q+ = 3C+(G), let (f : F) be defined as
(f : F) = inf {(SOL); a. > 0, and for some x. e G, Z ^ F * > f }. By
Proposition 19, (f : F) exists and is non-negative; also if \x is any
right Haar integral on G, then \x{i) < (f : F)M(F).

Proposition 20. Let f, f F, g e 0C+(G). Then:

(i) (f : F) > 0,
(ii) for each x € G, (fX: F) = (f : F),
(iii) (fi + f2 : F) < (fx : F) + (f2 : F),
(iv) if a > 0, (cd : F) = a(i : F),
(v) (f : F) < (f : g)(g : F).
x.
1=
Proof, (i): f < J ) a . F => sup(f) < (Za.)sup(F)
=> Z « . > sup(f)/sup(F).
Hence (f : F) > sup(f)/sup(F) > 0.
x. „ (x.x)
(ii): f < £ a . F l<^ f x < Z « . F l .
1
x. 1 y.
(iii): If f < A = Z a.F \ and f < B = Z j8.F J , then
i X j J
f + f < A + B, and (f + f2 : F) < Z a + Z j8.. Thus, taking lower
bounds, (f + f
: F) < (f : F) + (f : F).
x. x.
(iv): f < Z « . F x if and only if af < 2 ( « a . ) F \

66
 x. y.
J
(v): Suppose that f < Z a.g , g < Z^F . Then
3
x. y.x * (y^x.)
g * < 2 £.F J \ and f < Z a.jS.F J . Thus
J
i U *J
(f : F) < Z a.jS. = (Z a)(l 0 ), and hence (f : F) < (f : g)(g : F).
J
i,j i j
We now normalize (f : F) by fixing a standard function g € 3C+(G)
and defining

/xF(f) = (f : F)/(g : F).

Proposition 20 implies immediately:

Proposition 20\ Let f, f., F e 3C+(G). Then:

(i) MF(f) > 0,
(ii) for x € G, MF(fX) = /xF(f),
(iii) MF(fx + f 2 ) ^ / x F ( f l ) + /xF(f2),
(iv) if a > 0, /xF(af) = a/iF(f),
(v) ( g ^ ) " 1 < MF(f) < (f : g).

Proposition 21. (Uniform continuity). For any topological group

G, and f e 3C(G), f is uniformly continuous on G, that is, for arbitrary
e > 0 there is a symmetric open neighbourhood V rf e in G such that

|f(x)-f(y)| < e

whenever x, y € G and either xy~ eV or x y e V.

Proof. The function f has compact support C. For every

xeC there is a neighbourhood N(x) of e in G such that
f(x) - f(y) | < e/2 whenever y e xN(x). Choose, for each x e C, a
symmetric open neighbourhood M(x) of e such that M(x) Q N(x).
By compactness, there exist x , x , . . . , x e C such that
x i M(x l ) u X2M(x2) U . . . u x n M(x n ) 3 C. Put U = M(x 1 )nM(x 2 )n... nM(xn),
and suppose that x y e U. If both x and y are outside C then
|f(x) - f(y) | = 0 < e. Otherwise we may suppose that x e C, in which case
x e x.M(x.) for some i. But then y e xU ^ x.M(x.)U £ x.N(x.) and so

|f(x.) - f(y)| < e/2.

67
Since also x ex.M(x.) Q x.N(x.), we have

|f(x.) - f(x)| < e/2,

whence
|f(x) - f(y)| < e.

A similar argument shows that there is a symmetric neighbourhood U'

of e such that

|f(x)- f(y)| < e

whenever xy~ e IT. The neighbourhood V = U n UT then has the re-

quired property.

Proposition 22. Let G be a locally compact Hausdorff topologi-

cal group. Let f , f e X ,(G), and let e > 0. Then there is an open
neighbourhood V of e in G such that, whenever F e 3C+(G) has
support c V , nF(fi + f2) - MF(fx) + MF(f2) - e.

Proof. Let C be the union of the supports of f and f . Then

C is compact, and by Urysohn's Lemma we may select q e 3C+(G) such
that q(x) = 1 for x e C ,
Let a > 0, and define p = f + f + aq, so that for x e C,
p(x) > a > 0. Hence by continuity, p(x) > 0 on some open neighbourhood
of C.
r f.(x)/p(x) if x €C
For i = 1, 2, define h.(x)
1
= ] It is easy
' 0 if x £C.
to see that h. is continuous, and that support(h.) is closed in the com-
pact space C, hence is compact. So h. e JC (G), and for all x,

By Proposition 21, we can find a symmetric open neighbourhood V

of e (depending on a) such that |h.(x) - h.(y) | < a/2 (i = 1, 2) when-
ever xy" e V or x" y e V.
Suppose that F e JC+(G), with support(F) Q V, and that
x
i
p ^ 2 0.F J (with /3. > 0, x. e G).

68
 x
i I I
Then F j(t) = 0 unless t fVx., and then |h.(x.) - h.(t) | < a/2;
so
x.
f.^ph.^.h.F^
x.
j a/2)F J .
j
It follows that (f. : F) < Z jS.(h.(x.) + a/2), (i = 1, 2). Thus
(fx : F) + (f2 : F) < (1 + a)2 A., because h^x.) + h2(x.) < 1. Taking
limits, (f : F) + (f : F) < (1 + a)(p : F), and hence

/i
F(f2) ~
f2)

where R = jzF(fx + f2) + (1 + F

< (fx + f2 : g) + (l + a)(q : g).

But q, g, f , f are fixed, so we may choose a small enough to ensure
that aR < e. The corresponding V is the neighbourhood of e required.

Corollary. Given f , . . . , f e 3C+(G), and e > 0, there is a

symmetric open neighbourhood V of e such that, whenever F e 3C (G)
has support Q V,

Proof. This is a straightforward induction on n.

3. Convex cones
The facts about the 'approximate Haar integrals' M^ which we have
r
now proved can be used in many ways to establish the existence of a Haar
integral (see, e. g. , Nachbin [7] or Hewitt and Ross [3]). The geometrical
approach which we adopt here has the advantage of being more conceptual,
though no shorter, than the standard methods. We first introduce some
elementary notions from convexity theory and prove a general result on
the existence of supporting hyperplanes of convex cones.

69
 Definition. If V is a real vector space, a non-empty subset
E of V is a convex cone if
(i) A> 0=^ AE C E,
and (ii) E + E Q E.
(It follows from (i) and (ii) that if A, \x > 0, and A + JU = 1, then for all
u, v e E, also Au + JLXV e E; i. e., E is convex in the usual sense.)

Example. 3C+ = 3C+(G) is a convex cone in the real vector space

JC = 3C(G).
The set E as above is an open convex cone if in addition
(iii) for any c e E, and any v e V , 3 6 > 0 such that
c + Av e E whenever | A | < 6.
Note that if an open convex cone E contains 0, then E = V.
Because E is convex, (iii) is equivalent to
(iii)T every line (coset of a 1-dimensional subspace) in V meets
E in an open interval (possibly empty or infinite).
Any non-zero linear functional \i on V has kernel a hyperplane
(maximal proper subspace) H. Conversely, given a hyperplane H in V,
there is a non-zero linear functional /x on V with kernel H, and such
a JU is unique up to scalar multiple. (In fact /z : V -* V/H = R is such a
functional, and any MT with kernel H factors uniquely through ju.)
Now take V = 3C. If M is a positive integral, then it maps 3C+
to the set of non-negative real numbers. If /x is a right Haar integral,
then it maps 3C to the set of positive real numbers, and for f e 3C,
x € G, we have ju(f) = ju(fx). Hence /i(f - fx) = 0 and f - ^ e H = Ker(/i).
Let £ be the R-space (in JC) spanned by all functions f - f (for
f € 3C, x e G). Then if \x is a right Haar integral with kernel H ,
(i) H 2 £,
(ii) H n 3C+ = 0.
Conversely, suppose that H is a hyperplane in 3C satisfying (i)
and (ii) above, and let JLI be a linear functional with kernel H . Then
either [i or -JU is a right Haar integral. For if f e JC and x e G,
then f - fx € H, so /x(f) = M ^ ) , that is, M is right-translation-invariant.
Also, given f f e 3C+, and 0 < A < 1, we have Afx + (1 - X)f2 e JC+.
Thus, by condition (ii), Aju(f ) + (!*- A)/x(f ) * 0 for all 0 < A < 1. It

70
follows that /Lt(f ) and n(f ) have the same sign, and therefore either
\i or -\x maps 3C to the positive reals, hence is a Haar integral.

Proposition 23. Let V be a real vector space, E a convex cone

in V, and W a subspace of V which does not meet E. Then:
(i) D = E + W is a convex cone which does not meet W.
(ii) If E is open, so is D.
(iii) If E is open, there is a hyperplane H 2 W such that
E n H = 0.
(iv) If E is open, and its complement V\E is convex, there
is a unique H as in (iii).

Proof, (i) X > 0 =» XD Q XE + XW Q D, and

D + D £ E + E + W + W£D.
(ii) Let d = e + w € D, where e e E, w € W. Let v e V, and
choose 6 > 0 such that e + Xv e E whenever |x| < 6. Then
d + Xv e E + W = D for such X.
(iii) The set S of subspaces U 2 W such that U n E = 0 is
non-empty and inductively ordered by inclusion (note that the union of a
chain of subspaces is also a subspace). Hence, by Zorn r s lemma, there
is a maximal element H e S . We have to show that H is a hyperplane.
Maximality of H says that if v £ H, then H + Rv meets E,
hence Rv meets D = H + E. By (i) and (ii) D is an open convex cone
not meeting H; hence 0 £ D, and so either v e D or - v e D ,
Thus V = D u H u -D, and this union is disjoint. (For
D n H = 0=> (-D) n H = 0; and if u e D n (-D), then u e D and -u e D,
so 0 = u + (-u) e D, a contradiction.)
To show that H is a hyperplane, we need to prove that
v £ H =* H + Rv = V. We therefore suppose that v £ H and w £ H + Rv.
Then, because w £ H we have w e D or w e -D, and similarly for v.
We may assume that v c D , w e -D. Since D and -D are open convex
cones, the line joining v and w meets these cones in two disjoint, non-
empty open intervals. The union of these intervals cannot exhaust the
line, so some point of the segment between v and w, say A.V + /XW(X+/J=1),
must lie in H. But v £ H, so M * 0 and hence w e H + Rv, a contradic-
tion.

71
 (iv) E is an open convex cone, and E* = V\E is a convex cone,
whence so are (-E*), and V\(E u -E) = E* n (-E*) = H, say. Clearly
u e H ^ -u e H, and H is a subspace. Also V = E u H u (-E), a dis-
joint union, and so proceeding as in (iii), we see that H is a hyperplane.
Further, W n E = 0, so W n (-E) = 0, and W Q H. Similarly if HT is
any subspace containing W and not meeting E, then Hf £ H. In par-
ticular if HT is also a hyperplane, then H' = H, so H is unique.

4. The existence of Haar integrals

Theorem 5. Let G be a locally compact Hausdorff topological

group. Then there is a right Haar integral \i on_ G. (Similarly there is
a left Haar integral v, but note that JU + v in general.)

Proof. Let JC = JC(G), JC+ = JC+(G), and let £ be the R-sub-

space of JC spanned by all functions f - fX (f e JC, x e G).
Every f e JC can be written as f - f , with f , f e JC , (e. g.
fi = l ( | f | + f), f2 = | ( | f | - f)). Also observe that if f € JC+ and x € G,
then fx = h e JC+, and fX - f = h - h x ~ \ Thus every f e £ has the
xi
form f = I (f. - f. x ), for some f. e JC + , and x. e G.

L e m m a A. JC n <C = 0.

Proof. This is essentially Proposition 22, Corollary. Suppose

that f, f. € JC + , x. € G, and

i
X
i
Then f + 2 f • = Z f., and we are looking for a contradiction.
l
i i l
Let e > 0. By the corollary to Proposition 22 we can find an open
set V such that, whenever F e JC has support contained in V,
X X
i i
Hr (f + 2f.
. 1 )^M Trr(f) + ZM
. r F (f.
I )-e.

Such an F exists. For as G is locally compact, we can find an open set

U with compact closure U c v. By Urysohn's Lemma, we can find

72
F € 3C+ such that support(F) Q V.
With such an F, we have

= MF(f + Zf i 1 )
1

-M F (f) + Z M F ( f i V e

> (g : f)"1 + l^(f.) - e, by Proposition 20T.

i F l

Hence e > (g : f)~ for all e > 0, a contradiction.

Proof. This is essentially Proposition 19. We know that (B is

a convex cone, and 6 n £ = 0, by Proposition 23. Let f € 6, say
f = p + q9 p c 3C and q € <£. Let k € JC. By Proposition 19 we can
find a. > 0, x. e G, such that |k| < £ a.p ^ Thus for A > 0,

x.
f ± Xk => p + q - Xl a.p 1

x.
where qT = q + A 2 a?, (p - p ) € «C.
Hence f ± Ak e 6 whenever 1 - \J, a. > 0, i. e., whenever
0 < A < (£ a?.)"1, which shows that 6 is open.
It now follows by Proposition 23 that there is a hyperplane H in
JC such that H 2 £ and H n e = 0 (so in particular H nJC+ = 0). By
the remarks preceding Proposition 23, H determines a right Haar inte-
gral on G.
5. The uniqueness of Haar integrals
We shall use Theorem 5 to prove the uniqueness of the right Haar
integral. It is possible, at the expense of some extra detailed analysis,
to prove both existence and uniqueness at the same time, but the effort
seems hardly worth while. The only advantage is that it does away with
the use of the choice axiom in the proof of existence; but since Urysohn's

73
Lemma is essential and relies on the choice axiom this is not a genuine
logical advantage.

Proposition 24 (Dieudonne). Let f e 3C (G), where G is a

locally compact Hausdorff group, and let C = support(f); let W be any
neighbourhood of e in G. Then there exist x , x , . . . , x e C and
functions f , f , . . . , f e 3C+(G) such that
(i) f = 2f.
(ii) support(f.) c Wx. for each i.

Proof. By Proposition 17, there exists a compact neighbourhood

N of e in G contained in the interior of W. Since C is compact there
exist x , . . . , x e C such that Nx , . . . , Nx cover C. Choose
h € 3C+(G) such that h.(x) = 1 for x € Nx. and support(h.) £ Wx..
(UrysohnTs Lemma guarantees the existence of the h..) Put h = 2 h.;
then h(x) > 1 for all x e C and we may define

f(x)h.(x)/h(x) if x €C

0 if x i C.

Then f. is continuous, support(f.) Q support(h.) Q Wx. and Zf. = f.

Proposition 25 (Uniform approximation theorem). Let f e JC+(G)

and e > 0. Then there exists a neighbourhood V of e in G such that
for every symmetric F cJC+(G) ( i . e . , F(x~l) = F(x)) with support(F)£V
there exist real numbers a , . . . , a > 0 and x , . . . , x e G such
that
x.

|f(x) - laF \x)\ <e

for all x e G.

Proof. Choose a neighbourhood V of e so that

| f (x) - f (y) | < ^ whenever y eVx

(by Proposition 21), and let F e 3C,(G) be symmetric with support(F)£V.

74
Then we have support(Fx) £ Vx, so

|f(x)-f(y)|F x (y)<|F X (y)

for all x, y e G. Viewing the two sides of this inequality as functions of

y, we may rewrite it as

|f(x)FX - fF X | < | F x for all x e G. (1)

Now let 6 > 0 and choose another neighbourhood W of e such that

| F(y) - F(z) | < 6 for all y e Wz.

Then

| Fx(y) - FX(z) | < 6 for all y e Wz and all x € G.

By Proposition 24 (applied to this W) we can find x , . . . , x e G,

f , . . . , fn e 3C+(G) such that f = I f and support(f.) c Wx.. For each
i it follows that

^OrtlF^y) - FX(x.)| < 6f.(y) for all x, y e G.

x.
Now we observe that, since F is symmetric, F \x) = F (x.), and sum-
ming over i we get
|f(y)F X (y)-Zf.(y)F Xi (x)| < Sf(y)

for all x, y e G, from which follows

x.

|fFX - 2 f.F \x) | < 6f for all x e G. (2)

Combining (1) and (2) we get

|f(x)FX - I f.F \x) | < | F X + 6f for all x e G. (3)

Now, by Theorem 5, there exists a right Haar integral \i on G. We
apply \x to (3) to obtain

|f(x)ju(F) - ZM(ft)F l(x)\ < |

75
 < I M(F) + 6(f : F)jti(F)

for all x € G. Now divide by ju(F) and put a. — /u(f.)/JLI(F). This gives
x.
|f(x) - I a.F \x) | < | + 6(f : F) for all x e G.

But 6 is independent of F, so we can choose 6 small enough for

6(f : F) < •«. The resulting a. and x. give the required approximation.

Corollary. Let f e 3C(G) and let 6 be the cone 3C+(G) + £

(as in the proof of Theorem 5). Then there exists h e 3C+(G) such that
for every e > 0 either f + eh e (S or f - eh e-G.

Proof. Let C = support(f) and let D be a compact neighbour-

hood of C (Proposition 17). Let h £ 3C+(G) with h(x) > 2 for xcD
T
(such an h exists by Urysohn s lemma). Write f in the form f - f ,
f , f € JC (G), and apply the approximation theorem to f and f ,
choosing the respective neighbourhoods V , V of e small enough
for V^C c D, V2C CD (this is possible by Proposition 11). Let
Vc v n V2 be symmetric and choose F e 3C+(G) with support(F )£V
(by Urysohn). Put F(x) = F (x) + F (x" 1 ). Then F e JC+(G), F is
symmetric, and support(F) £ v. Both f and f can be approximated
as in the Proposition, using this F. Hence there exist x , . . . , x e C
and a , a , . . . , a e R (not now necessarily > 0) such that
x.
|f(x) - I a.F l(x) | < 2e for all x e G ,
X
i
Since f and F vanish outside D it follows that
x.
l
\i-laF \ < eh.
x. x.
X
Put a = 2 «.; then k = aF - 2 ot.F = Z «.(F - F l) e <£, and

f - eh < a F - k < f + eh.

Now either a > 0, in which case

f + eh > -k € £,

76
so that

f + eh e £ + JC+(G) = 6,

or a < 0, in which case

f - eh < -k e £

and

f - eh € - e .

Theorem 6. If JU, JUT are right Haar integrals on a locally com-

pact Hausdorff group G then there exists X > 0 such that JUT = XJJL.
(Remark: The equivalent statement holds, of course, for left
Haar integrals. But the right and left Haar integrals of a group need not
be the same.)

Proof. By Proposition 23(iv) it is enough to show that the com-

plement of 6 is convex, since we would then obtain a unique hyperplane
H containing JB and not meeting 6; since ker /x and ker /x1 are such
hyperplanes, they must coincide; therefore M = XptT and X will obviously
be positive.
In other words, it is enough to show that if f , f e JC(G) and
f + f e <£ then f € 6 or f £ ©. Suppose that f + f £ © and let
f - f = f. By Corollary to Proposition 25 there exists h € JC+(G) such
that for all e > 0 either f + eh € 6 or f - eh € - e , i. e., either

ii - f2 + eh € 6 or f2 - fx + eh e 6. (*)

But 6 is open; so there exists e > 0 such that f + f - eh € 6;

adding this to (*) we get either 2f e 6 or 2f e 6, i. e., f € (3 or
f ee.

6. Integrals on product groups

We now exhibit the relationship between the right Haar integrals
ix on G, v on H and the right Haar integrals on G x H.

77
 Notation. We shall write the more explicit form

J G f (x)dM(x)

for what we have so far denoted by

Proposition 26 (Fubini's theorem). Let G, H be locally compact

Hausdorff groups with right Haar integrals /x, v respectively. Let
f € 3C(G x H). Then the integrals

P(f) = !
and

a(f)= JHG

exist, are equal, and are right Haar integrals on G x H.

Proof. There exists a compact set C £ H such that f(x, y) = 0

for y ft C (project support(f) onto H). So, for each x e G, the function
0 : y Kf(x, y) is in 3C(H), and ;//(x) = v(<t> ) exists. To show that
x x
p(f) exists, we must show that xj/ e JC(G). Clearly i// has compact
support (again by projection!). Given e > 0 there exists a neighbourhood
V of e in G such that

|f(x', y) - f(x, y)| < e for all xT eVx, y e H,

since f is uniformly continuous on G x H. Choose k e JC (H) such that

k(y) = 1 for y € C. Then

|f(x\ y) - f(x, y) | < ek(y) for all y e H , x ' e V x .

Apply v to get

|*//(xT) - i//(x) I ^ ei/(k) for all xT e Vx,

which is just the statement of continuity for \js.

So i(/€ 3C(G) and p(f) = M(^) exists; this holds for any
f € JC(G x H). Now p is clearly linear, positive and right-invariant;
therefore it is a right Haar integral on G x H.

78
 Similarly a is a right Haar integral on G x H. So for some

X> 0

or(f) = Ap(f) for all f e JC(G x H).

To show that X = 1, choose p € 3C+(G), q € JC+(H) and put

f(x, y) = p(x)q(y) for all x e G, y e H.

Then, for this f,

p(f) = cr(f) = /x(p)i/(q) * 0.

It follows that X = 1, and hence that p = v.

7. Unimodular groups
Let G be a locally compact Hausdorff group, and let /x be a
right Haar integral on G. Let s c G and consider the left translate
s
f of f by s (i. e. Sf(x) = f(s"1x)). In general n(si) * [x{i) (example
later!). However, we can define a new linear functional SJU on 3C(G)
by Sjix(f) =/Lt(sf). This functional s\i is, in fact, a positive integral
(check!) and is right invariant:
s
H(fl) = /iftf*)) = M((Sf)t) = M(Sf) = SM(f).

Thus JU is a right Haar integral and so there exists a positive real num-
ber A(s) such that \x = A(S)M; A(s) is independent of the particular
right Haar integral JLX chosen. (Exercise.)
The function A : G -^R^ 08 so defined satisfies

A(st) = A(s)A(t) = A(ts)

and is continuous (Exercise). It is, therefore, a morphism in TQ. A is

called the right modular function of G. A group with A(s) = 1 for all s
is called unimodular.
If G is unimodular then Sjti = /x for all s, so jti is left invariant;
thus the right and left Haar integrals coincide; conversely, if the left and
right Haar integrals on G are the same, then G is unimodular. Obviously

79
any Abelian group is unimodular. A little unexpectedly we also have:

Proposition 27. All compact Hausdorff groups are unimodular.

First proof. Let f(x) = 1 for all x e G ; then f e 0C(G) and is

invariant under all translations. Hence

A(s)/i(f) = M(Sf) = M(f) * 0, so A(s) = 1 for all s € G.

Second proof. A : G -• R is a morphism in 7*8 ; its image

must be a compact subgroup of R . But the only compact subgroup
of R p o s is {1} (Exercise).
Note: For a compact group G we can always normalize the Haar
integral so that

The relation between right and left Haar integrals on a group G

can be made explicit via the modular function. For f e 3C(G) let i be
defined by f(x) = f(x~l); then f € 3C(G). Define v by v(f) = /x(f),
where ju is a right Haar integral. Immediately, v is a positive integral.
It is also left-invariant:
/\ -l
S
f(x) = ^(x" 1 ) = f f s ' V 1 ) = f(xs) = fS (x),
whence
S S S

Therefore v is a left Haar integral.

An immediate consequence is that \i is invariant under inversion
(i. e. , M(f) = M(f) for all f) if and only if the group is unimodular.
We get the left modular function
/
as1>sfollows: by an argument simi-
lar to the one above, we see that fs" = s" f, whence

vif) = M(fS) = M(S f) = Ais^W) = Ats" 1 )^).

Thus the left modular function is A; A(s) = A(s" ) = A(s)~ .

80
 Proposition 28. For a right Haar integral /z with right modular
function A,

JGf(x)dM(x)=

and this formula defines a left Haar integral, JGf(x)d^(x).

Proof. The left hand side is n(f) = v(f) and we have seen that
v is a left Haar integral. Let f*(x) = f(x)/A(x); then the right hand side
is fi(f*) = P(f), say. Now p is clearly a positive integral. It is also
left invariant:

Ms x)
whence

p( s f) = M((sf)*) = M(^gJ.
s
(f*)) = - J L j M(
s
(f*)) = M(f*) = p(f).

So, by Theorem 6, there exists X > 0 such that

p(f) = Xv(t) for all f e 3C(G).

To show that X = 1 we must choose an appropriate function and compute

the two sides. Let e > 0 and choose a neighbourhood V of e in G
such that

e
11 - -T7-T-
A(s) I < whenever s e V

(this can be done since A is continuous). There exists a symmetric

function f € 3C+(G) with support(f) Q V and ju(f) = 1. For this f,

We can now apply /x to obtain

and since

i/(f) = /x(f) = M(f) = 1 and p(f) = Xi/(f) = X,

we have

81
 |l - AI < e for any e > 0,

i. e. X = 1. This proves the proposition.

82
IV - Examples and applications

In this final chapter we aim to give the reader some idea of what
the Haar integral looks like in various special cases and how it is used in
practice. The material is informally presented so as to convey informa-
tion more rapidly than would be possible in a self-contained, rigorous
account. It will be an advantage for the reader to have some knowledge
of such topics as Lebesgue integration, measure theory, normed linear
spaces and the representation theory of finite groups. However, this is
not essential for a general understanding of the chapter, and it is hoped
that the items chosen will act as a stimulus to further reading.

1. Extensions of Haar integrals

Let G be a locally compact, Hausdorff group and let M be a
right Haar integral on G. Then 3C = 3C(G) is a normed space with
respect to ||f || =/x(|f |); however, 3C is not complete. There is a
standard procedure, due to Daniell,for completing 3C which we now des-
cribe briefly.
Let U be the set of all functions f : G -*R u { + 00} which are
pointwise limits of increasing sequences {f }, f € JC. For f e U,
define /x(f) = lim/x(f n ) (possibly +°°). The value of /u(f) is independent
of the sequence chosen. Let -U = {-f; f e U }, the set of pointwise
limits of decreasing sequences of functions in 3C. For f € -U, define
ju(f) = -/x(-f). This definition is consistent with the previous one.
A function f:G-*Ru{±«>} is said to be summable if
(i) there exist functions g € -U and h e U such that

g < f < h,

and (ii) sup jiz(g) = inf ju(h) < °°, where the inf and sup are taken
g h
over all g and h in (i).

83
 If f is summable we define ju(f) = sup /x(g) = inf M(h). Denote by
g
11 \
£ = £ (G) the set of all summable functions. If f e £ x then also
|f| ecC1. For any f eJC1, define ||f || =/x(|f |). Then JC1 is a semi-
normed space (there may exist non-zero functions whose norm is 0),
and is complete (Riesz-Fischer Theorem).
A null-function in <£ is a function f such that || f || = 0. If we
now define L = £ / {null-functions }, then L is a Banach space.
Similarly, we define <£p (for p > 1) to be the space of all functions f
such that |f|p is summable and define ||f|| = {/i(|f | P ) } 1 / P . Then
lP = £p/{null-functions } is again a Banach space, and 3C is a dense
subset of L P , for each p. Further, L 2 = L (G) is a Hilbert space,
with (f, g) = Jc f(x)g(x)d/x(x). This theory extends easily to complex-
valued functions, giving complex Banach spaces.
We can now define a right Haar measure on G, that is, a measure
invariant under right translations. We say that A Q G is measurable if
XA, the characteristic function of A, is summable. The measure of A
is then defined to be L^ X A W ^ ( X ) . It is possible to recover the Haar
integral from the Haar measure by the standard procedure for the Lebesgue
integral. For further details see Nachbin [7] and Widom [12],
2. Examples of Haar integrals
1. G = (Rn, +). This group is unimodular (being Abelian) and
the extended Haar integral is J f(x)dx, the n-dimensional Lebesgue
integral. R
2. G = S = R/Z. Again this is a unimodular group. A func-
tion f : S -*R is essentially the same thing as a periodic function
f * : R -•R with period 1. The Haar integral JU is given by

= !\ f*(x)dx.
1 \n

Alternatively, we may take S = { e ;0^^<2TT} and the Haar integral

is then ju(f) = •=- J f(e )d#, where the factor ^— produces the norma-
lized integral (S1 being compact).
3. Induced integrals, (i) Let M be a right Haar integral on
the locally compact, Hausdorff group G and let H be an open subgroup

84
of G. Any f e JC(H) extends to Te JC(G) by the definition
f (x) if x e H

0 if x ft H.
(Exercise: check this!)
Defining MT(f) = MCO we get a right-invariant integral and hence
a right Haar integral on H.
N. B. This method does not work for all subgroups or even for
all closed subgroups; for example the Lebesgue integral on R induces
on Z by this method the trivial integral, but the Haar integral is never
trivial.
(ii) Suppose that G is a locally compact, Hausdorff group and
N is a closed, normal subgroup of G. Then Q = G/N is again locally
compact and Hausdorff. Suppose that /x and AX are right Haar inte-
grals on N and Q respectively. We can construct /x a right Haar
integral on G, as follows.
Let f € 3C(G), and for x e G define

f*(x) = JNf(yx)d/xN(y).

Because /LLT is right-invariant, f*(x) is constant on each coset of N.

Hence f* induces a function F:Q-»R defined by f(Nx) = f*(x). It
follows that I € JC(Q) (check!).
Defining Mr(f) = Mr(F), it is obvious that Mr is a positive integral
and easy to verify that jtip is right-invariant. Thus it is a right Haar
integral. (A particular case is the Haar integral on a product A x B. )
4. Finite discrete groups. For a finite discrete group G,
X
3C(G) = L (G) is the set of all functions f : G -*R. The normalized (left
and right) Haar integral is given by

f(x),

where n is the order of G.

Now L (G) is isomorphic to the group ring RG (which is a real
vector space with basis G), so the multiplication in RG, induced by the
multiplication in G, gives a multiplication on L (G) as follows. If

85
f, g € L (G) then the corresponding elements of RG are 2 f (x). x and
2 g(x). x. Define f * g e L (G) to be the function corresponding to their
product:

2 2 f(x)g(y).xy= I { 2 f(x)g(x"1z)}z
xeGyeG zeG xeG
= 2 i 2 f(zy-1)g(y)}z,
zeG ycG
i. e . , f * g = h where h(z) = 2 ftogfx^z) = 2 tizy'^gly).
xeG yeG
By analogy, for any locally compact Hausdorff group G, with

right Haar integral ju, we can define a multiplication * on L (G) by

(f * ]U g)(z)= SGi(zy-l)g(yW(y).
(Since this definition is for f, g in <C one needs to show that it behaves
correctly on taking quotients with respect to the null functions.) Simi-
larly, for the left Haar integral v, define:

(f * i ; g)(z)= iGf(x)g(x~lz)dp(x).

Exercises, (i) (f *M g)(z) = jQ f(y"1)g(yz)djLt(y).

(ii) *^ is associative on L 1 , and ||f *^ g|| x < ||f 1^ • ||K||x.
This shows that L 1 is a Banach algebra.
(iii) L is commutative if and only if G is Abelian.
(iv) L has an identity element if and only if G is discrete.
5. Arbitrary discrete groups. For any discrete group G,
JC(G) is the set of all functions with finite support, and the Haar integral
is ju(f) = 2 f(x). This is invariant under inversion, so G is uni-
X€G
modular.
LX(G) is the set of all functions f with countable support S, such
that
2 |f(x)| < + « .
xcS
For f €LX(G), we may take /z(f) = 2 f(x) and ||f|| = 2 |f(x)|. For
xeS

86
example, if G = (Z, +), then LX(Z) is the set of all doubly infinite se-
+ 00 + °°
quences a = {a } _ such that 2 la | converges, and ju(a) = £ a .
n n e z n n
-00 +00 _oo
X
If a, b€L (Z) then a * b = c where cn = Z ambn-m (the
m=-°°
Cauchy product of Laurent series).
6. Let G=(R P O S , •). K f eJC(G) then p(f) = J~ f(x)dx
(Lebesgue integral) is a positive integral but it is not right invariant. In
fact if y €R p o s , then

p(fy) = / ^ ( x y ' ^ d x
= J~yf(z)dz
= yp(f).

A general procedure for finding the Haar integral in such cases is given by

Proposition 29. Let G be a locally compact, Hausdorff group,

and let p be a positive integral on G. Suppose that p(f ) = g(y)p(f),
for all f € JC(G) and y e G, where g is a function from G to^ R p o s .
Then the equation ju(f) = J - S dp(x) defines a right Haar integral JU
L* gpg
on G.
Proof. Since fyz = (fy)z, and since p(f) * 0 for some f e JC(G)
we see immediately that g(yz) = g(y)g(z).
Hence ^(fy) = jQ l ^ j J dp(x)

g(xy"1)g(y)

Thus |i i s a right Haar integral (cf. the relationship between left and right
Haar integrals and modular functions).

87
 Corollary. The Haar integral on R p o s is given by

Alternatively we can obtain the Haar integral on R p o s from the isomor-

phism

G = ( R p o s , •) = (*, +) = H

given by log : G -• H and exp : H -* G. If we transfer f e 3C(G) to H

by this isomorphism we get f* e 3C(H) where f*(x) = f(exp(x)). If v
is a Haar integral on H we get a Haar integral ju on G by putting

\+2 f(exp(x))dx = J"IM dy.

7. Let G = (C*, •). We know that C* = R p o s x s 1 , so taking

polar coordinates (r, 0) we have

But dxdy = rdrd0, so

M(f) = JJ %-^-dxdy = J 2 M 2 dxdy,

r R |z|
where z = x + iy.
8. The general linear groups. Let G = GL (R). Then G is
2
an open subset of R n because it is defined by the inequality det(x..) ± 0
2 ^
We write T=(t..) for a variable matrix, M for R n and JM<iT for
the n2-dimensional Lebesgue integral.
Given f e 3C(G), define p(f) = JGf(T)dT = JMF(T)dT where I
is the extension of f which has value 0 outside G. Clearly p is a
positive integral on G, and we shall apply Proposition 29 to convert it
into a Haar integral. We therefore let U € G and consider
Tfl
P(f ) = JGf(TU)dT. For fixed U, the map S *-*SU is a linear map
from M to M and we need to find its Jacobian. The matrix of this
linear map, with respect to the standard basis of M, has rows enumera-
ted by pairs of integers (i, j) where i, j run from 1 to n. Similarly

88
for the columns. The entry in the (i, j)-th row and (k, l)-th column of
the matrix is

a«su)..)
=
)
Arranging the basis of M in lexicographical order we therefore obtain
the matrix

\0 0

where UT is the transpose of U. Thus the Jacobian of the transformation

is (det U)n and we have
-i
p(fU ) = |detur n p(f),

or, equivalently, p(f^) = |detu! n p(f). It follows, by Proposition 29, that

the functional JLX defined by
f(T)
M(f) = dT
~ |detT|"
is a right Haar integral on G. By symmetry, jti is also a left Haar inte-
gral and therefore the general linear group is unimodular (though it is
neither Abelian nor compact).

Exercise. Obtain a formula for the Haar integral on GL (C).

9. The affine groups. Let G = GA (R), the affine group on
R . Then G has a normal subgroup N = R n which consists of all the
n

translations of R n . The quotient group G/N is isomorphic with

GL (R) and G is a split extension of N by the general linear group.
This means that G is isomorphic with the group of all pairs (t, T),
where t € N, T e GL (R) and multiplication in G is given by the for-
mula

(t, T). (V, T')=(t 4-t'T, TT).

89
Here we identify N with R , using additive notation, and tTT denotes
the transform of the vector tT by the matrix T. For the computation
we shall identify G with this group of pairs.
We now apply Example 3(ii) to find the Haar integrals on G. If
f € JC(G), we first compute f*(t\ TT) = /^((t, I). (t\ TT))d(t, I), where
I is the identity of GL (R). From the definition of the multiplication
in G, we deduce that

f*(t', TT) = J n f(t + t \ Tf)dt

R
= J n f(t, T')dt.
R
This is a function of TT alone (i. e. f* is constant on the cosets of N)
and, according to Example 3(ii), we obtain a right Haar integral JU by
integrating it with respect to the right Haar measure on GL (R). This
gives

= J J f(t? T dtdT
GL
GLn(R) R n ,| d e t ( T )!n
|n >

by Example 8, the integrals being n-dimensional and n -dimensional

Lebesgue integrals.
For the left Haar integral on G = GA (R) we must first compute

f V , T ) = JNf((tT, T).(t, I))d(t, I)

= J n f(tT + tTT, TT)dt.
R
T
Substituting s = tT , we obtain

ft 41 TI)
< n ' n f<t? + s' T ' ) d s
IX

The left Haar integral v is therefore obtained by integrating this function

of TT over GL , which gives i/(f) = J J f t? T )
^ ,, dtdT.
n 2 n+1
R n R n |det(T)|
So GA (R) is an example of a group which is not unimodular. Its
right modular function is A(t, T) = | det(T) |.

90
 10. The orthogonal groups. The computation of Haar integrals
on orthogonal groups is harder. O is not an open subset of R n , so
the n2-dimensional Lebesgue integral is useless; it always gives zero.
What is needed is a parametrisation of O by local Euclidean coordinates.
n
+
In fact O , as a closed subgroup of GL , acquires a natural structure
as a Lie group and one gets local parameters by taking Euclidean coordin-
ates in its tangent spaces. There is then a standard method of constructing
the Haar integral (see e. g. Chevalley [2], Chapter 5).
Alternatively, one can extend the theory of Haar integrals to
locally compact homogeneous spaces and then use the embeddings
O -* O -* O -* . . . to obtain the Haar integral on O inductively,
using the analogue of Example 3(ii). Here we need to know the invariant
integrals on the coset spaces O /O -, - Sn~ . For example, on S2
there is a familiar surface integral, invariant under the action of O ,
and this together with the Haar integral on O = S1 gives an easy
2
+
formula for the Haar integral on O .
11. Profinite groups. Any profinite group G is a compact
Hausdorff group, so has a unique normalized Haar integral. Any con-
tinuous function f : G -*R is uniformly continuous, so for e > 0 we
can find a neighbourhood V of e such that |f(x)-f(y)|< e whenever
xy~ € V. Now V contains an open normal subgroup H of finite index
n, say. On each coset of H, f varies by at most e, so if we choose
arbitrary representatives x. (i = 1, 2, . . . , n) for the cosets of H, the
real number
n
l
l f(x }
5 i
n
i=l *
is determined to within e. The integral /u(f) is then the limit of this
sum as H -* e in the sense that, given e > 0, there is a neighbourhood
V of e such that

for all open normal subgroups H ^ V ,

The corresponding Haar measure on G is determined by the fact
that the measure of G is 1 and for an open normal subgroup H of index

91
n the measure of each coset of H is 1/n. For example, if G = Z ,
the group of p-adic integers, the subgroups H = p G ( n = 0 , 1, . . . )
form a fundamental system of neighbourhoods of 0, and if H x, H y are
any two cosets of these subgroups, either they are disjoint or one is con-
tained in the other. It follows that every open set S is the union of a
(countable) collection of disjoint cosets of the H and the measure of S
is the sum of the known measures of these cosets. In fact, if r denotes
the number of cosets of H contained in S, then the measure of S is
lim (r /p n ).
n-*°°
3. Representations of compact groups
A (complex) representation of a topological group G is a mor-
phism of topological groups p : G -• GL (C) for some n. The integer n
is called the degree of the representation. If we consider GL (C) as
Yi

acting on an n-dimensional complex vector space V in the usual way (on

the right), then a representation p of degree n gives an action of G on
V by the rule v. g = vp(g) for v e V, g € G. This makes V a right
G-module over C, that is, v. (gh) = (v. g). h, v. e = v and, for each g,
v H* v. g is C-linear. Also the action is continuous in the sense that the
map or : V x G -* V defined by (v, g) H* v. g is continuous, where V
has its canonical topology as a C-space (V = C n , and the topology is
independent of the basis). We call V a CG-module for short.
Conversely, given an n-dimensional CG-module V with continuous
action V x G -* V, one can obtain a representation p : G -* GL (C) by
choosing a basis v1 , 2v , . . . , vn for V and letting v..l g = Z. p..(g)v.
IJ j
for
i = 1, 2, . . . , n. The matrix p(g) = (p..(g)) is then a continuous function
of g and satisfies p(gh) = p(g)p(h). Two representations which come
from isomorphic modules in this way (in particular from different bases
of the same module) are called equivalent. Note: a morphism 6 : V -*V
of CG-modules is a C-linear map satisfying (x. g)0 = (x0. g) for all
x e V , g e G; it is automatically continuous.
For discrete groups G, we may omit all mention of continuity
without altering the meaning of the above definitions, and we are then
reduced to the purely algebraic concepts of classical representation theory.
In this theory the group algebra A = CG, whose elements are finite formal

92
sums Z a.g. with a. cC, g. e G, plays a fundamental role. This A
j i 1 1 l
is an associative algebra with respect to the multiplication induced by
multiplication in G of the basis elements. Any G-module M over C
becomes an A-module if we define

for m € M, a. e C and g. € G. Conversely, any A-module is obviously

a G-module over C since G is embedded in A. Thus the terms 'G-
module over C and 'CG-module' are interchangeable.
Now A itself has a right A-module structure induced by its mul-
tiplication, and in the special case when G is finite A is actually a
finite-dimensional A-module. The corresponding representation is called
the regular representation of G and it is faithful, that is, it is an em-
bedding of G in GLn(C) (where n = | G | ) .
We now state the basic theorems of the complex representation
theory of a finite group G. (See Serre [10] for direct proofs of these.)

Theorem A. G has a faithful representation, so is isomorphic

to a group of complex matrices.

Theorem B. Every representation of G is equivalent to a

unitary one (that is, to a representation by unitary matrices).

Theorem C. Every representation of G is completely reducible.

The definition of 'completely reducible' is as follows.
1. An A-module V is simple or irreducible if it is not {0 }
and has no submodules other than {0 } and V.
2. An A-module is semi-simple or completely reducible if it
is a (finite) direct sum of simple modules.
The terms 'irreducible' and 'completely reducible' are also applied
to the corresponding representations. In terms of matrices, a represen-
tation is completely reducible if it is equivalent to a representation of the
form

93
 P2(g)
p(g) =

\ PB(g)

where each p, is an irreducible representation.

Theorem D. All irreducible representations of G come from

simple submodules of the group algebra A = CG.

Theorem E. Let p run through one unitary representation from

each equivalence class of irreducible representations. Then the functions
p.. : G -*C are orthogonal: if p, a are two of these unitary representa-
tions, then

xeG

Theorem F. The functions p.. in Theorem E form a basis for

the space of all functions G-*C. Equivalently, the elements 2 p..(x)x
xeG 1J
form a basis for the group algebra A.

It is natural to ask whether these important results can be proved

for suitable topological groups and their (continuous) representations. We
shall give an outline of the proofs of appropriate generalisations of all of
them for an arbitrary compact Hausdorff group G.
The first thing to notice is that the group algebra A = CG of an in-
finite discrete group is an infinite-dimensional module and has, in fact, no
finite-dimensional submodules except {0 } (exercise!); it is therefore
useless for representation theory. However this group algebra can be
regarded as the space of all functions f : G -*C with finite support; the
element of CG corresponding to such an f is 2 f(x)x. Since 'discrete
xeG
and compact' means 'finite 1 , and all functions on G are continuous, this
space of functions is just 3C(G) (except that we are now looking at complex-
valued functions). With this interpretation, the multiplication in A is
convolution of functions (see §2, Example 4 above) and the action of G

94
on A is right translation of functions. We may therefore enquire whether
Theorems A-F can be proved for a special class of infinite topological
groups using as Tgroup algebra' the convolution algebra A = 3C(G) con-
sisting of all complex-valued continuous functions on G with compact
support. Provided that G is locally compact and Hausdorff, we have a
right Haar integral jti : 3C(G) -*C (integrate real and imaginary parts
separately) and the recipe for generalisation is to replace Z 0(x) in
the finite case by J~0(x)d/x(x).
For a compact group G this programme works very well and
gives results which include Theorems A-F as special cases. The algebra
A is now, of course, the algebra of all continuous functions G -*C. It
is a right G-module under right translation:

fs(t) = f(ts"1) (f €A; s, t €G)

and we must study its finite-dimensional submodules. We use the norma-

lized form ju of the right Haar integral and we recall that multiplication
in A is given by

(f * g)(x) = JGf(xy"1)g(y)diu(y).

(This formula, for complex-valued functions, gives a product with all the
usual formal properties.)
We first observe that convolution and translation are compatible:

(f *g) s (x) = (f *g)(xs"1)

= JGf(xs"1(ys'1)""1)g(ys"1)dM(y) (since ju is right

invariant)
= JGf(xy"1)gs(y)dM(y)
= (f*gS)(x),
that is, (f * g) s = f * g s ,
It follows that the principal right ideal f * A of A, for fixed f,
is a G-submodule of A (though still not finite-dimensional). So let us
fix f € A and write M for the submodule f * A = {f * a; a e A }. We

95
say that g e M is an eigenfunction for f if f * g = eg for some complex
number c. The eigenf unctions are the solutions of the integral equation

cg(x)= JGf(xy"1)g(y)dM(y).

For any such eigenfunction g, we have

f * g s = (f * g ) s = (Cg)s = cg s ,

so g s is also an eigenfunction for the same value of c. Hence the space

E = i g e M ; f * g = eg} is actually a submodule of M.
We shall assume the standard theory of integral equations, which
is applicable without change to continuous functions on compact groups.
(For details, see Chevalley [2], Chapter VI, or Smithies [11].) The basic
result is that if k =£ 0 is a fixed symmetric, continuous function from
G x G to C, then the solutions g of the integral equation

cg(x) = J G k(x, y)g(y)dM(y),

for fixed c, form a finite-dimensional C-space E ; furthermore, the

c
spaces E for c * 0 span topologically the space of all transforms
c
JG k(x, y)h(y)dM(y),

in the sense that any such transform is the sum of a uniformly convergent
series Zg., where each g. lies in some E /.^ with c(i) * 0.
Applying this result with k(x, y) = f(xy"x), (f e A), we see that
if f € A is symmetric (i. e. f(t -1 ) = f(t)) and f * 0 then the G-module
M = f * A is spanned topologically (in the above sense) by the finite-
dimensional eigenspaces E , for c * 0, and each of these E is a
submodule. There seems to be no shortage of finite-dimensional repre-
sentations of G; but we must, of course, make sure that they are not
all trivial. We shall show, indeed, that for every t * e in G there is
an f c A and a corresponding eigenspace E such that t acts non-
trivially on E . For, if we are given t * e, we can find a neighbourhood
U of e such that t" £ U and then, by Urysohn's lemma we can find
a symmetric function f e A = 3C(G) such that f > 0, f(e) = 1 and f(x) = 0

96
for x i U. For this f we have

(f * f)(e) = JG f(y"

On the other hand

(f *f)t(e) = (f *f)(t"X)
= JGf(t"1y"1)f(y)dM(y)

since if f(t"1y"1)f(y) ^ 0 then t" 1 y" 1 eU and y e U, whence t""1 eU 2 ,

which is false. Thus (f * f) * f * f and so t acts non-trivially on
M = f * A.
We remark that the action a : M x G -• M given by (h, t) K h
is continuous when M is given the topology of uniform convergence
(exercise!). Since the submodules E span linearly a dense subspace
of M in this topology it follows that t must act non-trivially on some
E . The topology on M induces on this finite-dimensional submodule
Ec the canonical topology. Since G acts continuously on this module

we get at last a continuous representation

Pt:G-GLn(C), (n = dimE c )
such that t fL Ker p.. We write K, for Ker p, and M, for the corres-
ponding submodule of M (E above). Since GL (C) is Hausdorff, Kt
is closed in G; since t ^ K,, the intersection of all the K, is {e };
since G is compact, every neighbourhood V of e in G contains a
finite intersection K = K.^v n K , ^ n . . . n Kw y The direct sum of
the corresponding modules
M
t(l) * Mt(2) ® • • • ® M t(r)
is a finite-dimensional G-module with continuous action, giving rise to
a continuous representation
repr p : G -• GL (C) whose kernel is precisely
K. This proves:

97
 Theorem A'. K G is a compact Hausdorff group, then every
neighbourhood of e contains a closed normal subgroup K such that
G/K is isomorphic (in T^) with a group of complex matrices.
(Note: the continuous injection G/K -• GL (C) is an isomorphism
with the image group because G/1C is compact and GL is Hausdorff.)
An immediate corollary is

Theorem An. Every compact Hausdorff group without small sub-

groups has a faithful representation as a group of complex matrices.
(A group Thas small subgroups' if every neighbourhood of e
contains a non-trivial subgroup; otherwise it is 'without small subgroups'.)
Clearly each of these theorems reduces to Theorem A if G is
finite and discrete. With a little more work one can prove:

Theorem Am. Every compact Hausdorff group is the inverse

limit of a system of compact matrix groups.
Note: Since closed subgroups of GL (C) inherit an analytic
structure from it, every compact matrix group is a Lie group, and
Theorem A"' shows that compact Hausdorff groups can be 'approximated
by Lie groups'. The corresponding statement for locally compact groups is
true under some restrictions but is very much harder to prove. (See
Kaplansky [4] and Montgomery and Zippin [6],)
The other results of representation theory mentioned earlier
(Theorems B-F) are proved by methods which follow the classical proofs
more closely. For example if p : G -* GL (C) is any representation of
a compact Hausdorff group G, let M be the corresponding module, and
let ( , ) be any positive definite Hermitian form on M. Then the form
(, ) c defined by

< u, v > G = JG <u. g, v. g) d/i(g)

is also positive definite Hermitian and has the extra virtue of being in-
variant under the action of G, i. e. (u, v)~ = (u. g, v. g> r , by the left
invariance of the Haar integral on G. (N. B., G is unimodular.) We may
now choose an orthonormal basis of M with respect to (, )~, and the

98
resulting representation, equivalent to p, will be unitary. This proves:

Theorem B\ Every representation of a compact Hausdorff group

is equivalent to a unitary one.

Theorem C\ Every representation of a compact Hausdorff group

is completely reducible.

Proof. This follows easily from Theorem BT, or rather from its
proof. We take a module M for the representation and construct an
invariant positive definite Hermitian form on it. If M is not irreducible
it has a proper submodule M . Let M be the orthogonal complement
of M . By the invariance of the form M is also a submodule of M,
and M = M © M . The result now follows by induction on dim M.

Theorem E \ K G is a compact Hausdorff group, and if p

runs through a complete set of inequivalent irreducible, unitary repre-
sentations of G, then the functions p.. : G -• C are orthogonal:

where \x is the normalized Haar integral on G.

Proof. This is essentially Schur's lemma. (Schur's lemma

states that any non-zero morphism 6 : M -• N between irreducible
modules is an isomorphism. This is almost obvious since the image
and kernel of 6 are submodules of N and M respectively. Since the
image of 6 is not 0 it must be N, and since the kernel of 9 is not M
it must be 0.) Let p, a be representations of degrees r, s arising
from modules R, S with respect to fixed bases. Let a : R -*• S be
any linear map, and let A be the corresponding r x s matrix. By
'averaging over GT we obtain a module homomorphism a : R -* S, where
for r €R,

By Schur's lemma, if p and a are irreducible, a is 0 or an isomor-

phism. Suppose first that p and a are inequivalent irreducible repre-

99
sentations. Then a = 0 for every choice of a. In matrix notation
this says (assuming that the representations are unitary) that

J G p(A) ^K) d/x(x) = 0

for all matrices A, where integration is performed component-wise.

From this follows

J p (x) cr (x) dju(x) = 0 for all i, j , k, I.

KJ lj K6

For the rest of the proof we may assume that p = o is an irreducible

unitary representation of degree r coming from a fixed basis of the
module R = S. If a? is any linear map R -* R then a is a module
endomorphism of R. Let A be an eigenvalue of a in C. Then
a - At is a singular module-endomorphism and so must be 0, by Schur's
lemma. Thus a = At for suitable A € C. In matrix notation this says
that for every r x r matrix A

J G p(x)Ap(x) dju(x) = AI,

for some A depending on A. Taking traces of both sides we find that

trace A = Ar, so A = (deg p)" 1 (trace A). The equations

=a
ik6iZ^degP^"1

now follow by taking for A the single-entry matrices E.,.

Theorem F \ (Peter-Weyl) Every continuous function from a

compact Hausdorff group G to C can be uniformly approximated by
linear combinations of the functions p.. in Theorem E \

Proof. First observe that if p and o are equivalent represen-

tations, then each function a. 7 is a linear combination of the functions
-l
p.. because, for some matrix T, a(x) = Tp(x)T for all x e G. So
the statement of the theorem is unaltered if we admit all the functions
p.. for all irreducible representations p. Indeed, since every repre-
sentation (of finite degree) is equivalent to a diagonal sum of irreducible
ones, it makes no difference if we admit the functions p.. for all repre-

100
sentations p. We therefore let R c JC(G) be the closure (in the topology
of uniform convergence) of the linear space generated by all the coordi-
nate functions p.. of all the representations p of G. It is enough to
show that R = 3C(G).
Now, for any non-zero symmetric function f e 3C(G), the G-
module M,. = f * 3C(G) contains finite-dimensional submodules E which
between them span a dense subspace (see the proof of Theorem Af). Let
E be such a subspace, giving rise to a representation p with respect
to a basis g^ g , . . . , g^ of eigenfunctions. Then g x = Z p..(x)g.,
and so g.(x~l) = £p..(x)g.(e). This shows that g. e R, where
g.(x) = g.(x ). In the obvious notation, this implies that E Q R and
hence that Mf £ R for every symmetric f.
Finally, if h is any function in 3C(G), we can approximate it by
elements of suitable modules Mf as follows. Let e > 0 and choose a
symmetric neighbourhood U of e in G such that

|h(x) - h(y)| < e for xy"1 € U.

This is possible by Proposition 21. By Urysohn's lemma we can find a

symmetric function f > 0 which vanishes outside U, and by normaliza-
tion we may assume that J f(x)d/x(x) = 1. It is an easy exercise to show
that |(f * h)(x) - h(x) | < e for all x e G, and so h is a uniform limit
/s.

of functions of the type f * h e Mf. Since Mf £ R, this implies that

h e R for every h, and therefore R = 3C(G).
Theorem D\ Every irreducible representation of the compact
Hausdorff group G comes from a submodule of 3C(G).

Proof. This follows easily from the proof of the Peter-Weyl

theorem and the orthogonality relations. For let p be an irreducible
representation and suppose that it is not obtainable from a (finite-
dimensional) submodule of 3C(G). Then its coordinates p.. are ortho-
gonal to all coordinates of all irreducible representations which do come
from submodules of 3C(G). But these latter span a dense subspace of
3C(G) and it follows that the p.. are orthogonal to all functions in JC(G),

101
in particular to themselves. So p.. = 0, for all i, j , which is impossible
for a representation.

4. Topics for further reading

(a) Invariant integrals on homogeneous spaces have been men-

tioned in connection with integration on orthogonal groups. This topic is
treated in Nachbin [7], Chapter in.
(b) The representation theory of compact groups is carried
much further in Adams [1] and is neatly done in coordinate-free fashion.
I strongly recommend this book as an introduction to Lie groups.
(c) The character theory of locally compact Abelian groups can
be found in Pontryagin [8] in a rather old-fashioned presentation (but none
the less worth reading). The duality theory there presented is a corner-
stone of abstract harmonic analysis. More accessible accounts can be
found in Loomis [5] and Rudin [9],
(d) The structure theory of locally compact Abelian groups is
closely related to (c), but can be carried through without characters or
integration. This is done in Hewitt and Ross [3], Chapter 2. (This 2-
volume work contains a super-abundance of information on topological
groups, but because of the detail offered, it is sometimes difficult to
read.) See also Montgomery and Zippin [6], p. 101.
(e) The solution of HilbertTs 5th problem. A full account of
this has been written by two of the principals. Their book, Montgomery
and Zippin [6], is very readable and contains a good treatment of much
standard material. A slicker, but somewhat sketchy account is given in
the second half of Kaplansky [4], This little book is a good place to get
the flavour of the subject, and the exposition carries one painlessly
through the most formidable difficulties.

102
References

[I] J. F. Adams. Lectures on Lie groups. W. A. Benjamin, Inc. ,

New York-Amsterdam (1969).
[2] C. Chevalley. Theory of Lie groups I. Princeton University
Press (1946).
[3] E. Hewitt and K. A. Ross. Abstract harmonic analysis. 2 vols;
Springer-Verlag, Berlin-Gottingen-Heidelberg (1963).
[4] I. Kaplansky. Lie algebras and locally compact groups. Chicago
University Press (1971).
[5] L. H. Loomis. An introduction to abstract harmonic analysis.
Van Nostrand Co. , Princeton N. J. (1953).
[6] D. Montgomery and L. Zippin. Topological transformation groups.
Interscience Publishers, Inc. , New York (1955).
[7] L. Nachbin. The Haar integral. Van Nostrand Co., Princeton
N.J. (1965).
[8] L. Pontryagin. Topological groups. Princeton University Press
(1946).
[9] W. Rudin. Fourier analysis on groups. Interscience Publishers,
Inc. , New York (1962).

[10] J . - P . Serre. Representations lineaires des groupes finis.

Hermann, Paris (1967).
[II] F. Smithies. Integral equations. Cambridge University Press
(1962).
[12] H. Widom. Lectures on measure and integration. Van Nostrand
Mathematical Studies, No. 20 (1969).

103
Index to page heading

Affine group 59, 89 Degree of representation 92

Axiom of choice 7 Dense subset 23
T
Dieudonne's proposition 74
Basis conditionsT 13
Difference kernel 30
Basis for open sets 12
Disconnected, totally 36
Bounded above 7
Disconnection of a space 35
Category 2 Eigenfunction 96
Categorical concept of product 6 Equivalence relation 5
Chain 7 Equivalent representations 92
Closed map 11
set 11 (f:F) 66
Closure 34, 36 Faithful representation 93
Compact space 42 Fibre of a set-function 5
, sequentially 43 Filter in a lattice 46
Completely reducible module Finite intersection property 43
(representation) 93 Fubini's theorem 78
Complex representation 92 Fundamental system of neighbour-
Component 36 hoods 26
Cone over an inverse system 51 Fundamental system of open neigh-
Connected space 35 bourhoods of x 26
Continuous function 3 'Fundamental system of open neigh-
bourhood conditions' 27
Continuity, uniform 67 'Fundamental system of open neigh-
Convex cone (open) 70 bourhood subgroup conditions' 28
Convolution 86
Coordinate, i-th 6 G-module 64, 92
projection, i-th 6 Group, affine 59, 89
Correspondence between sets 5 , general linear 16
Coset space G/H, left 20 , orthogonal (O ) 18
(On) 41, 91

104
Group, profinite 52, 91 lim 51
, projective linear 22
Limit, inverse 51
, topological 16
, projective 51
, unimodular 79
Locally closed subspace 59
compact space 55
Haar, right, integral 64
, measure 84
Maximal element of a set 7
Hausdorff space 30
Measure, right Haar 84
Homeomorphism 4
Modular function, right 79
Homogeneous space 19
Morphism in a category 2
of cones 51
Ideal of a lattice 45
of topological groups 18
, proper 10
Identification topology 10
Neighbourhood 26
Inclusion map 11
Normalised Haar integral 80
Induced integral 84
Null-function 84
Induced topology 9
Inductively ordered set 7 Open map 11
Integral, positive 63 Orthogonal group (O ) 18
, right Haar 64 (On) 41
Interior point 26
Inverse limit of an inverse p-adic integers 54
system 51 numbers 60
system in 6 51 topology 18
Irreducible module 93 valuation 18
Isomorphism in a category 4 Partially ordered set 7
Path-connected 40
3C(X) 61 Peter-Weyl Theorem 100
3C+(X) 62 Positive integral 63
Kernel of a set function 5 Product, categorical 6
of a group morphism 8
of morphisms 15
of topological groups 25
Left modular function 80
topology 13, 14
multiplication in a topological
group 19 Profinite group 52, 91
quotient space G/H 20 Projection, i-th coordinate 6
translation 19 Projective general linear group 22

105
Projective limit of an inverse Summable function 83
system 51
Support of a function 61
n-space, real 45
Symmetric neighbourhood of e 27
Proper ideal of a lattice 46
T -space 31
Quotient 5 T -space 29
group of topological T -space 29
group 20
map 5, 11 Topology, discrete 9
topology 10 , identification 10
, induced 9
Real projective n-space 45 , metric 9
Reducible, completely 93 , p-adic 18
Regular representation 93 , product 14
space 31 , quotient 10
Representation, equivalent 92 , stronger 9
, faithful 93 , subspace 9
, regular 93 on a set 3
of a topological , trivial 9
group 92 , Tychonoff 14
Right Haar integral 64 , weaker 9
measure 84 Topological group 16
modular function 79 Torus, T n 14
translate of A 19 Totally disconnected 36
Semi-simple module 93 ordered set 7
Separating 2 compact sets by a Translate, right 19
compact set 62 Translation, left, right 19
Sequentially compact 43 Tychonoff theorem 45
Simple module 93 topology 14
Small subgroups 98
Space, left quotient, G/H 20 Ultrafilter 46
Sphere, Sn 10 Uniform approximation theorem 74
standard argument' for universal continuity 67
properties 6 Unimodular group 79
Stronger topology 9 Unitary representation 93
Sub-basis for a topology 13 UrysomVs lemma 62
Subspace topology 9

106
Valuation on a field 17
, p-adic 18

Weaker topology 9

Zorn's Lemma 7

107
Index to numbered results

Lemma 1 46 Proposition 11 48
V 46 12 49
2 47 13 49
A 72 14 50
B 73 15 52
Proposition 0 19 16(50 56
1(3) 5 16(2*9) 57
1(9) 8 17 58
ion 12 Corollary 59
1(2*9) 21 18 62
Corollary 23 Corollary 63
2 28 19 65
Corollary 28 Corollary 66
3(50 30 20 66
f
3(^9) 30 20 67
4 32 21 67
5(r) 32 22 68
5(srg) 33 Corollary 69
6 34 23 71
Corollary 35 24 74
7(2*) 35 25 74
7(7*9) 36 Corollary 76
8(ST) 36 26 78
Corollary 37 27 80
9(50 37 28 81
Corollary 38 29 87
9(2*9) 38 Corollary 88
10(2*) 43 Theorem 1 38
10(2*8) 48 2 45

108
Theorem 3 52
4 54
5 72
6 77
A, B, C 93
D, E, F 94
A', A n, A m 98
Bf, C ', E' 99
DT 101
FT 100

109