Aspects of Topology - Christenson, Charles O Voxman, William L - 1998

| ASPECTS
OF
TOPOLOGY
C.O. Christenson
~ W.L. Voxman
5B Associates
w, Idaho, U.S.A.
Digitized by the Internet Archive
in 2022 with funding from
Kahle/Austin Foundation
https://archive.org/details/aspectsoftopologOO0O0O0chri_j6l7
Second Edition
ASPECTS OF
1IQOPOLOGY
CHARLES O. CHRISTENSON
WILLIAM
L. VOXMAN
Department of Mathematics
College of Letters and Science
University of Idaho
Moscow, Idaho
BCS Associates Moscow, Idaho

1998
First Edition was published by Marcel Dekker, Inc.
Edwin Hewitt, Editor
Second Edition
Bryan A. Smith, Editor
Copyright ©1998 by Charles O. Christenson and
William L. Voxman; published by BCS Associates, Moscow, Idaho, USA.
All rights reserved.

No part of this book may be translated or reproduced in any form without
the written permission of the authors and publisher.
Library of Congess Cataloging-in-Publication Data:

Christenson, Charles O.
Aspects of topology/ Charles O. Christenson, William L. Voxman-2nd ed.
p. cm.
Includes bibliographical references and index.
ISBN 0-914251-07-9 (alk. paper). —ISBN 0-914351-08-7 (soft: alk. paper)

1. Topology. I. Voxman, William L. II. Title.
QA611.C436 1998
514—de2 98-45593
ISBN: 0-914351-07-9 Hard Cover ISBN: 0-914351-08-7 Paper Cover
Manufactured in the United States of America.
10987654321
Printed on acid free recycled paper

The authors dedicate this work to their former mentors,
5 Armentrout, Ai Giever, A. Kruse, and T. Price.
[hada feeling about Mathematics—that I saw it all. Depth
beyond Depth was revealed to me—the Byss and the Abyss.
I saw—as one might see the transit of Venus or even the
Lord Mayor's Show—a quantity passing through infinity
and changing its sign from plus to minus. | saw exactly how
it happened and why the tergiversation was inevitable—but
it was after dinner and | let it go.
Winston S. Churchill
Preface to Second Edition
Over twenty years have passed since the original edition was published. Dur-
ing those years our students and colleagues have from time to time forcefully
brought to our attention various shortcomings. In this reincarnation of the
book we have attempted to minimize confusion by rewriting and/or rephras-
ing many of the proofs and explanatory passages. No doubt, however, in the
process new ambiguities have arisen. These should be viewed as our way of
ensuring that topology instructors will continue to be needed in the future.
Substantive changes have occurred in chapters 8, 13, 15 and 18. The
old chapter 18 has been partially tucked into chapter 8 and pruned. We
have added a new chapter 18 that gives a brief introduction to the theory
of dimension. The last two of sections of chapter 13 have been reorganized
significantly. We have rewritten the section of chapter 8 dealing with CW-
complexes, and the of chapter 15 dealing with the fundamental group of the
complement of a graph to make them more readable.
We have also tried to make the index more complete and usable, and
all of the many figures have been redrawn.
Many of the problems at the end of chapters have been reformulated,
added, and/or shuffled. We view these problems and the students agonizing
over them to be essential to any course based on this book. In the first
edition, we marked problems we thought would give students difficulty by
*. However, over the years it became clear to us that what is difficult and
what is easy is often a highly individual matter, and the *’s were serving
no purpose other than to scare some students from attempting problems.
None of the *’s have been retained. Nevertheless we can say with assurance
that students will find some problems easy, and some nearly impossible. Our
advice to the student: Do not be discouraged when you run into a problem
you can not solve—but keep thinking about it!
Vv
There is more material in the text than any instructor should cover in a
year’s course. For what it is worth, we generally cover chapters 1,2,3,4,6,7,8,
12,13, and depending on the interest and ability of the class (and the whims
of the instructor) an additional two to four chapters.
We are grateful to our students over the years who have directly and
indirectly pointed out ambiguities and other sources of confusion in the first
edition. We particularly thank K. Meerdink, C. Steenberg, and S. Stockett.
We also thank our colleagues John Cobb and Mark Nielsen for their help and
stimulating discussions. Finally we thank Bryan Smith, who not only helped
with the original edition, but has functioned as editor for this edition.
C.Christenson
W.Voxman
Moscow, Idaho, August 1998
vl
CONTENTS
Peeinces 44, Sian gk. Pir sc kg ek pk cae eal
Chapter OF 'Prelmmaries 90". . . "2, attietiisegedt
en’)
af). ae
A. Basic Notation
B. Cartesian Products and Relations
C. Functions
D. The Axiom of Choice and Some of its Variants
E. Cardinal Numbers and Ordinal Numbers
F. Minkowski’s Inequality
Chapter I: The Basic:Constructe wnu ao, Se, ote? sere 8 11
. Topological Spaces and Continuity
. Further Examples of Topological Spaces
. Metric Spaces: A Preview
. Building New Spaces from Old Ones
. A Potpourri of Fundamental Concepts
. Continuity
Bs
coi
Coe
Sey . Homeomorphisms
Problems
Chapter 2: Connectedness and Compactness ...........47
. Connectedness: General Results
. Connectedness: Slightly Deeper Results
. Path Connectedness
. Components
. Locally Connected Spaces
. Simple Chains
. Compactness
= . Two Characterizations of Compactness
[o2)
@)
ke)
[esl
23)
(ep)
an)
I. Compactness: Local, Countable, and Sequential
J. The One Point Compactification
Problems
Chapter 3: Metric Spaces . Pk
A. Compactness in a Metric Setting
B. Complete Metric Spaces
C. Convex and Hausdorff Metrics
Problems
vii
viii Contents
Chapter 4: Separation,Properties 0% 4, s+) Guedesele Boe 97

A. Normal Spaces ;
B. Uryson’s Lemma; Tietze’s Extension Theorem
C. Further Results Concerning Normal Spaces
D. The Separation Axioms
Problems
Chapter 5: 1-Manifolds and some Plane Theorems ...... . 121
A. 1-Manifolds
B. On the Contractibility of S!
C. The Jordan Curve Theorem
D. The Schonflies Theorem
E. The Annulus Theorem
Problems
Chapter 6: The Product Topology and Inverse Systems ..... 151
A. The Product Topology Revisited
B. Inverse Systems: the Preliminaries
C. Compact Metric Spaces are Continuous Images of the Cantor Set
D. The Dyadic Solenoid
Problems
Chapter 7: Functions Spaces, Weak Topologies, and Hilbert Space 175
A. The Point-Open Topology
B. The Compact-Open Topology
C. The Weak Topology: I
D. The Weak Topology: II
E. Hilbert Space
Problems
Chapter 8: Quotient Spacesmiuautt eet ie cee hice. mt S
A. The Quotient Topology
B. Identifications
C. Identification Maps
D. The Strong Topology and k-Spaces
E. CW-Complexes
F. Upper Semicontinuous Decompositions: An Introduction
Problems
Chapter 9:Continua. 24, 2 2 . SRR sl EH Te 227
A. The Arc
B. Peano Continua
C. Chainable Continua
D. Decomposable and Indecomposable Continua
E. The Pseudo-Arc
Problems
Contents ix
Chapter 10: Paracompactness and Metrizability ........ 249

A. Paracompactness
B. Partitions of Unity
C. A Sampling of Metrizability Theory
D. Moore Spaces
E. Completion of Metric Spaces
Problems
Chapter 11; Nets and Filters. ..2 < <5 ou.) 506 MoD Zu
A. On the Failings of Sequences
B. Nets
C. Filters
D. Nets vs. Filters
Problems
Chapter 12: The Algebraization of Topology .......... 283
A. The Fundamental Group (The First Homotopy Group)
B. Elementary Properties of the Fundamental Group
C. Continuous Functions and Homomorphisms
D. Categories and Functors
E. The Seifert-van Kampen Theorem
F. Direct Limits
Problems
Chapter 13: Coveriuig Spaces”...
we ee apt Natla ae er y Poe Ne
A. The Lifting Theorems
B. 1(S!,s) = Z (A Diversion)
C. Regular Covering Spaces
D. Map Liftings
E. Covering Morphism and Translations
F. Universal Covering Spaces
Problems
Chapter 14: Some Elements of Simplicial Theory ........ 341
A. The Polyhedral Category
B. Basic Constructions
C. Simplicial Approximation Theorems
D. Edge Path Groups: How to compute 7(|A’|,vo) for any Complex
Problems
Chapter 15: Further Applications of Homotopy ........ . 375
A. The Extension Problem (Revisited)
B. The Separation Problem
C. Unicoherence
D. 7 (€3,G), for a Polygonal Graph G in €°
E. Wild Sets Do Exist
Problems
X Contents
@hapter .16::2-Manifoldspidarasls bow emeatadein

etal 201 23: 401
A. The Triangulation of 2-Manifolds
B. The Classification of Compact 2-Manifolds
C. A Characterization of €?
Problems
Chapter 17: An Introduction to n-Manifolds
A. Some Preliminaries
B. n-Annuli
C. Cellular Sets
D. The Generalized Schoénflies Theorem
E. Compact n-Manifolds
Problems
Chapter 18: Dimension Theory
A. Little Inductive Dimension
B. Big Inductive and Covering Dimension
C. Some Final Results
Problems
Appendix’: : >... -< « s enonegen 459
A. Free Groups
B. Group Presentations
C. Homomorphisms and Tietze’s Theorem
D. Direct Limits of Groups
Bibliography 471
Index 481
~~ 7 .
- al ro
=
:
PRELIMINARICS
7 a ae Ee amet
' ; a
Bits Vapor «sz trade oe} 16 He mahal) : iaagtaciit the

ie Sti, Coe ww tle
14 ue hagen: of <2 obrag Cher © emit
ee a ee ey? Batak @ =. fh) Penqiices
ibe
aberial: Op mercies yates hehe (4 bee, oe Me be
toe, dies L001, Oeil 08 Dijiverk piel Je {0 i
The. pce ffsem ype @ fe) 4 (A (hate pea Mey
rele, os fianas say ts rein . ie i et of
ie me i (1 > her A al oe wh ate ary Oe
i. Ve wi 2 n
i 7
A HAS & Lr vi .
e 7 26@e ss ’ a
e » e
ih)7 s
“ad -: i wise _
‘
—a s -
> a ? -
Le é > a" Td =
— : ,
- = Pan ihe. af
é th) eenmw O¢
-! an a ae @ 7.
acy orhee aahta
(rage 1 © NNireneient Veg

— 7 : -
‘ ue te dar flee aw
F —,-.
A te il: aC ap
‘ ie 7 Pex
a :
¢ ps Phual ale i : a” =
Pe Adee a
hp-<enile’ vas on 6
4, Tie Oe
ts Yim * :
\ taped ae a Tan) T agus fe -
QD Dreet Lust whim
HNO. weed ;
Chapter 0
PRELIMINARIES
In this chapter we introduce much of the notation used throughout the book.
In addition, we give a sketch of the elements of set theory that is needed in
the ensuing chapters. No attempt is made to give a full presentation of this
material; for more complete expositions of set theory, see the books by Monk
[1969], Halmos [1960], as well as Hrbacek and Jech [1984].
The proofs of most propositions found in this chapter are left to the
reader, presumably as review exercises. Our recommendation for the use of
this chapter is to initially scan it, and then refer back as necessary while
reading the rest of the book.
A. BASIC NOTATION
The following symbols are used frequently:
is a subset of
is a proper subset of
is an element of
is not an element of
the set of integers (also considered as an additive group)
the set of positive integers
+-
NNRWAWN
the set of natural numbers
1 the set of real numbers
the set of rational numbers
the interval {x € R! : 0<a< 1}
n
the set {(21,22,.-., 2a): 41 E R'}
ee
ee the empty set
P(X) the set of subsets of X
1
2 0. Preliminaries
Standard set notation is employed. The words “set,” “family,” ”

and “collec-
tion” are used interchangeably throughout the book.
(0.A.1) Definition. Suppose A = {Ay : a € A} is a family of sets indexed
by A. The union of the sets Ag is defined to be {x : x € Aq for some a}
and is denoted by either U{Aa : a € A} or Ugeg Aa-
The intersection of the sets Aq is defined to be {x : x € Ag for each a}
and is denoted by (\{Aa : a € A} or ()y¢, Aa:
(0.A.2) Definition. If A and B are sets, the complement of A in B is
defined to be {x € B : x ¢ A} and is denoted by B \ A.
(0.A.3) Definition. A set X is a singleton if and only if X has just one
element.
The following proposition gives two essential and easily established re-
sults involving set operations, often referred to as the DeMorgan laws.
(0.A.4) Proposition. Suppose {Aq : a € A} is a family of subsets of a set

X. Then
(i) X\ (U{40 a€ A}) =(\{X\Aa : a € A}, and

(ii) X\ ((fAe ; a€ A}) =|J{X\4a : weA}.
(0.A.5) Proposition. Suppose {Ag : a € A} is a family of sets and Aisa

set. Then
(i) An (tA :ae€ Ay) =(J{AnAa : @€ A}, and

(ii) AU ((\fAq : ae Ay) =(){AUAg : @€ A}.
(0.A.6) Definition. A subset A of R! is an interval if and only if whenever

x,y€ Aandz<z<_y, then z € A. We denote the various kinds of intervals
as follows:
(a, b) =-1@7eR’:a<¢<hi
[a,5] = {reR’ :a= 2 < 6}
[a, b) = (ce Rt a<¢ <5}
(a, ] = {ren -a—2 <b)
(-co,4) = {ZeR’ a <a}
(=o0,a] “= (gen. 2c}
(O60) = ee ee Oy
[b, 00) =e € ee Sy
—co,co) = RI
(0.A.7) Definition. The dyadic rationals are the set {m/2” : m,n € Z}.
Cartesian Products and Relations 3
B. CARTESIAN PRODUCTS AND RELATIONS

(0.B.1) Definition. Suppose X and Y are sets. The Cartesian product of
X and Y, denoted by X x Y, consists of all ordered pairs (x,y) where z € X
andy € Y. That is, X x Y = {(z,y) : 2 € X andy € Y}.
The Cartesian product of n sets X1, X2,..., Xn is defined by
Xt X Agix & Xq = {(21,22,..-;2n) 1.04 © X; for each ae
The following set equality is occasionally useful.
(0.B.2) Proposition. Suppose A Cc X and B CY. Then
(X x Y)\ (Ax B) = ((X \ A) x Y) U(X x (Y \B)).
Cartesian products are used to define relations on a set X.
(0.B.3) Definition. A subset of X x X is called a relation on X. If

(v1, 22) € R we write 1 Ra2.
(0.B.4) Definition. If R is a relation on a set X, then

(i) Ris reflexive on X if and only if Raz for each x € X,
(ii) Ris symmetric if and only if yRa whenever xRy, and
(iit) Ris transitive if and only if zRy and yRz implies that zRz.
A relation R is an equivalence relation on X if and only if it is symmetric,
transitive, and reflexive on X. Suppose that # is an equivalence relation on
X and z € X. Then the equivalence class of x is {y € X : tRy} and is
denoted by [z] or [z]p.
(0.B.5) Proposition. Suppose FR is an equivalence relation on a set X.
Then {[z] : 2 € X} forms a partition of X, i.e., U{[z] : « € X} = X, and
if z,y € X, then either [z] N [y] = @ or [2] = [y].
Relations are used to define various kinds of orderings.
(0.B.6) Definition. Suppose RF is a relation on a set X. Then F is a partial

ordering for X if and only if
(i) R is reflexive and transitive, and
(it) ifeRy and yRa, then x= y.
In this context R is frequently replaced by the symbol <. Le., write x < y
instead of cRy. Furthermore, we denote “< is a partial ordering for the set
X,” by (X,<) and write x < y if and only ifz <y anda fy.
(0.B.7) Definition. A partial ordering < for a set X is a linear ordering if

and only if for each z,y € X either zr <yory<z.
4 0. Preliminaries
(0.B.8) Definition. A linear ordering < for a set X is a well-ordering if

and only if whenever A # 9 and A C X, there is'an a € A such that a < g,
for each x € A.
(0.B.9) Definition. Suppose that < is a partial ordering for a set X and
Ac X. Then u € X is an upper bound for A if and only if a < u for each
a € A; uis a least upper bound for A if and only if u is an upper bound for
A and if v is also an upper bound for A, then u < v. The least upper bound
u is denoted by either lub A or sup A.
(0.B.10) Definition. Suppose that < is a partial ordering for a set X and
Ac X. Then bis a lower bound for A if and only if b < a for each a € A; bis
a greatest lower bound for A if and only if b is a lower bound, and whenever c
is also a lower bound for A, then c < b. The greatest lower bound is denoted
either by glb A or inf A.
(0.B.11) Definition. Suppose that < is a partial ordering for X and that
Ac xX. Then m € A is a mazimal element in A if and only if whenever
a€Aandm <a, then m=a;n € Ais a minimal element in A if and only
if whenever a € A anda<n, thena=n.
C. FUNCTIONS
(0.C.1) Definition. Suppose that X and Y are sets. A function from X
into Y, denoted by f : X > Y, is a subset f of X x Y with the property
that for each x € X, there is a unique y € Y such that (z,y) € f. Usually
f(z) = y is written instead of (x,y) € f. The set X is called the domain of
f and {y€Y : f(x) =y for some z € X} is the image of f; the image of f
will often be denoted by Im f or f(X).
(0.C.2) Definition. Suppose f : X + Y and A C X. The image of A
under f, denoted by f(A), is defined to be {f(z) : 2 € A}.
(0.C.3) Definition. Suppose f : X — Y is a function. Then f is onto Y
(or is a surjection) if and only if f(X) =Y.
(0.C.4) Definition. Suppose that f : X > Y is a function. Then f is one
to one (or an injection), denoted by 1-1, if and only if whenever 21, 272, € X
and 2; # Zo, then f(r1) # f(x2). If f is both 1-1 and onto, f is called a
byection.
(0.C.5) Definition. Suppose f : X — Y is a function and A C X. The

restriction of f to A is the function f|4 : A + Y given by f|,4 = {(z,y) €
fe ge A}.
(0.C.6) Definition. Suppose f : X > Y is a function and B C Y. The

inverse image of B under f is defined to be {x € X : f(x) € B}, and is
denoted by f~1(B).
Functions 5
(0.C.7) Proposition. Suppose that {Aq : a € A} is a collection of subsets

of a set X and {Bg : B € WV} is a collection of subsets of a set Y. If
f :X —Y isa function, then
(i) f ApS, 4 CNaea f(Aa);
(12) fi lJ eeiad = Cry f(Ag),
(ee (Ass Ba) = Neew pe (Bp);
(iv) £7 (Upew Ba) = Upew f*(Ba),

(v) f-'(Y
\ Bg) = X \ f-*(Ba),
(vs )ifofisvonte; thentY \ f(A.) C.f(X \ Ag),
(vii) if f is 1-1, then if(a Na) Se eras) cand
(viii) f(f-(Ba)
MAa) = Bg f(Aa).
Suppose f : X > Y is a bijection. Then for each y € Y, f~?(y) consists
of a single point in X. A function g : Y — X can be defined by setting
g(y) = z, where f(x) = y. In this context, g is usually denoted by f~! and
is called the inverse function of f.
(0.C.8) Definition. Suppose f : X + Y and g: Y — Z are functions. The
composite function gf : X —> Z is defined by gf (x) = g(f(z)).
(0.C.9) Proposition. If f : X > Y andg: Y — Z, then (gf)~'(B) =
f-1(g71(B)) for each B C Z.
(0.C.10) Definition. Let X be a set, the identity function idx : X + X is
defined by idx (x) = 2, for each zx € X. When the context is clear, we write
id instead of idx.
(0.C.11) Proposition. If f : X > Y and g: Y ~ X are functions such
that gf = idx, then g is onto and f is 1-1. If in addition fg= idy, then f
and g are bijections.
(0.C.12) Definition. Let W, X, Y, and Z be sets. Suppose that f : X
Y,g:Y 3Z,andh: X > Z are functions. The diagram
V4
x —_“_—-y
is commutative if and only if gf = h. Suppose that f: X > Y,9:Y > Z,

p:X—3W andq: W — Z are functions. The diagram
xX ¥
Pp g
WwW Z
6 0. Preliminaries
is commutative if and only if gf = gp. Similar definitions hold for other

configurations. ’
Functions are used to define infinite Cartesian products.
(0.C.13) Definition. Suppose {X, : a € A} is a family of sets. The

Cartesian product of the sets Xq is defined by
[] Xe ={f +A UneaXa : f(a) € Xo for each a € A}.

acA
An element f € [],¢, Xq is usually denoted by {ra} or {Za}aea where it

is understood that f(a) = Zq for each a € A.
(0.C.14) Notation. Suppose {X, : a € A} and {Y, : a € A} are
families of sets indexed by the same set A, and suppose for each a € A,
foe Xe S YooPhetfinction: |] fa Pee ane eee Yotaetdetined sby
I] fa({ta}) = {fo(Ta) fae:
D. THE AXIOM OF CHOICE AND SOME OF ITS VARIANTS

(0.D.1) The Axiom of Choice. Suppose that {Xq : a € A} is a family
of nonempty sets. Then there is a function f : A + U je, Xa such that
f(a) € Xq for each a € A.
Note that the “choice” function f selects a point x, from each Xqg. From
(0.D.1) it follows that the Cartesian product (0.C.13) is not empty. There
are a number of propositions that are equivalent to the axiom of choice.
After the following definition we list a few of these which we use later.
(0.D.2) Definition. Suppose that A is a subset of a partially ordered set

(X,<). The set A is a nest in X if and only if for each a,b € A, we have
a <borb<a. That is, A is a linearly ordered subset of the partially ordered
set.
A proof of the next proposition may be found in most books on set

theory.
(0.D.3) Proposition (Zorn’s Lemma). If X is a partially ordered set

such that every nest in X has an upper bound (in X), then X contains a
maximal element.
(0.D.4) Proposition (Kuratowski’s Lemma). Suppose (X, <) is a par-

tially ordered set. Then each nest A in X is contained in a maximal nest M,
i.e., M is a nest, A C M, and if z € X \ M, then M U {z} is not a nest.
Proof. We first suppose that A is a nest in (X,<). Next we let M =
{M c P(X) : M isanest in (X,<) and A C M}. Then apply Zorn’s
Cardinal Numbers and Ordinal Numbers if
lemma to find a maximal element in M. Partially order M by declaring

M, < Mz if and only if My C Mp. Let C = {M,. : a € A} be a nest in M.
We show that D = U{M, : a € A} is an upper bound (in M) for C.
Certainly A C D and M, C D for each a; hence it remains to establish
that D is a nest in (X,<). Suppose that B,C € D, B € My, € C, and
C € Ma, € C. Then since either Mz, C Ma, or Ma, C Ma,, we have that
either B,C’ € Mg, or B,C € Ma,. Since My, and M,, are nests, it follows
that B < C or C < B and, consequently D is a nest in (X,<).
Thus D is an upper bound for C, and therefore, by Zorn’s lemma, M
has a maximal element. 0
(0.D.5) Remark. Kuratowski’s lemma is usually applied in the following

context. Suppose that X is a set and W is a subset of P(X). Partially order
P(X) by inclusion: if A,B € W, then A < B if and only if A C B. A nest
in W is a collection C of sets in W such that if A,B € C, then either AC B
or. DB GA.
We outline a proof of the following proposition.
(0.D.6) Proposition (Zermelo’s Theorem). Every set can be well-
ordered.
Proof (sketch). Suppose that X is a set. Let C be the set of all ordered
pairs of the form (A,<,4) where A C X and <zy is a well-ordering for A.
Partially order C by setting (A,<,) < (B <p), provided that
ACB,
(ii) ifc,y€ AC B then « <p y if and only if z <, y, and
(iii) ifye B\ AandxeA, thenz <py.
Zorn’s lemma may be used to show that (C, <) has a maximum element
(Y, <y), and it is then easy to establish that Y= X. O
(0.D.7) Definition. Suppose that (X,<) is a well-ordered set and z € X.
Then the initial segment determined by x, denoted by Xz, is defined to be
{yEeX :y<az}.
(0.D.8) Proposition (Transfinite Induction). Suppose that X is a well-
ordered set and that F is a nonempty subset of X. Suppose further that
whenever x € X and X, C E, then x € E. Then E£ is all of X.
Proof. If E # X, then X \E #9. Let z be the first element of X \ E.
Since every element of X, is an element of LE, we have X; C E. Thus, by
hypothesis, x € E, which contradicts cx¢ X\E. O
E. CARDINAL NUMBERS AND ORDINAL NUMBERS
It is beyond the scope of this book to provide a systematic treatment of

either cardinal or ordinal number theory. For our purposes, it suffices to
postulate (and accept) the existence of certain sets mentioned below.
8 0. Preliminaries
(0.E.1) Definition. Suppose X and Y are sets. We say X and Y are

equipotent if and only if there is a bijection f: X > Y.
We postulate the existence of a class C of sets with the property that

every set X is equipotent with precisely one of the sets in C. If X is equipotent
with C € C, we write C = card X.
(0.E.2) Definition. A set X is countable if and only if it is equipotent
with Z*, a finite set {1,2,...,n} or is empty. Otherwise X is said to be
uncountable.
(0.E.3) Proposition.
(i) The set of rational numbers Q is countable.
(it) The union of countably many countable sets is countable.
(iit) The finite product of countable sets is countable.
With regard to the ordinal numbers we postulate the existence of an
uncountable well-ordered set (O,<), containing a largest element , and
with the property that ifa € © anda < 2, then {G8 € O: GB < a} is
countable. The element 2 will be referred to as the first uncountable ordinal.
Since (0, <) is well-ordered, there is a first element w with the property
that {a € © : a < w} is countable but not finite. We call w the first infinite
ordinal.
The smallest ordinal in © will be denoted by 0, the second smallest by 1,
etc. Thus the ordinals may be thought of as being strung out in the following
manner:
Oped lyin scarily Cla a he een, Dt Gd > DS ioe Mile
An important property of the ordinals is the following.
(0.E.4) Proposition. If A C O is countable and 0 ¢ A, then supA < 2.

Proof. It follows from the properties of O that for each a € A, the set
{8 €0O: B <a} is countable. Let X = {6 : B <a for some a € A}. Then
by (0.E.3), X is countable (it is the countable union of countable sets). Since
O is well-ordered, there is a smallest element y in © that is not contained in
X. Clearly y is an upper bound for X, and furthermore, 3 < ¥ if and only
if 8 € X. Hence, {5 € O : 6 < 7} is countable, and consequently y can not
be 2. Since sup A < 7, the proof is complete. O
F. MINKOWSKI’S INEQUALITY
The following inequality (Minkowski’s inequality) will frequently be useful:
Minkowski’s Inequality
where a; and b; are arbitrary real numbers and n € Zt.

This equality follows readily from another famous inequality, that of
Cauchy and Schwarz.
(0.F.1) Proposition (Cauchy-Schwarz Inequality). If a,,a2,...,an,

by, b2,..., bn are real numbers, then
nm
Sy abi <
a=1
Proof. It suffices to show that
(4) <(4) (Ex)

If b; = O for each 7, then clearly equality holds. Suppose then that not
all the };’s are 0. If a is any real number, then
n n n
0< S\(ai+ab;)? = Sa? + 2a) ~aibi +0? 5°

i= 1 i=l i=] t=
Let
Then we have
n n 2
Qoins 2idi)
0<y Dial aibi)” 4
— heaps) = aibi)” Eh yeha
iz a ee
a pas
Sia Dia OF seal Ba.
or
(4) <(E4) (Ze)

(0.F.2) Proposition (Minkowski’s Inequality). Suppose that a1, a2,.. On;
bi, bg,..-, bn are real numbers then
10 0. Preliminaries
Proof. From the Cauchy-Schwarz Inequality we have

n n
So (ai + bi)? = So(aj + 2aib; + 07)

71 <=!
n nm n
= Soa? +25 aii + 50oF

1 a=] *=1
lA
The desired inequality is now obtained by taking square roots of both

sides. O
Minkowski’s inequality can be extended to infinite sequences.
(0.F.3) Proposition. Suppose that a;,a2,a3,... and 61, bz, 63,... are se-
quences of real numbers such that
co CO
y a;2 and y b;2

i=1 i=1
are convergent. Then
The following corollary to (0.F.2) and (0.F.3) will be of special interest.

(0.F.4) Proposition. Suppose that 7,2%2,... and yj, y2,... and 2}, 29,...
are sequences of real numbers. Then for each n € Z*,
Proof. Let aj = x2; — yi, b: = yi — 2%, and apply the previous two
propositions. O
Chapter 1
THE BASIC CONSTRUCTS
A. TOPOLOGICAL SPACES AND CONTINUITY
To a significant extent, topology can be viewed as the study of the meta-

morphosis that sets or spaces undergo when they fall under the influence of
continuous functions. The idea of continuity should be familiar from calcu-
lus, but the notion of space, or more accurately topological space, is likely
to be vague at best. The reader is certainly aware of some special sets such
as the real line, R', the plane R?, closed intervals, spheres, balls, doughnuts
(tori), etc. Our first task is to give such sets structures that will enable us
to define the continuity of functions between any two of them.
The reader should remember the € — 6 concept of continuity as used
in calculus, wherein a function f : R! > R! is said to be continuous at a
point c € R' if and only if for each positive number e, there is a positive
number 6 such that |f(x) — f(c)| < € whenever |x — c| < 6. Stated more
geometrically, this becomes: a function f is continuous at a point c if and
only if for each open interval (w, z) containing f(c), there is an open interval
(a,b) containing c such that f((a,b)) C (w, z).
ae te ee f(e)
a c b w —_—_——— 2
f((a,6))
The reader should verify that the two definitions are indeed equivalent. The
advantage of the latter definition is that if we can establish the notion of
“open interval” or, more generally, “open subset” of a given set, then our
definition of continuity can easily be generalized.
1
12 1. The Basic Constructs
Recall that an open set in R is any set that can be written as the
union of a family of open intervals (in particular, this will make the empty
set open, since it is the union over an empty family of open intervals). It is
easy to verify that
i) the union of any y collection of open
p sets is open, and
(ii) the intersection of any finite number of open sets is open.
We can now say that a function f : R! + R! is continuous at a point
c if and only if for each open set U containing f(c), there is an open set V
containing c such that f(V) C U. The reader should check that this defi-
nition of continuity is compatible with the previous ones. The appropriate
designation of open subsets of a set X determines what is called a topological
structure for X. Much of the beauty of topology stems from the fact that
properties (i) and (77) above are all that is needed to determine whether or
not a subfamily U of P(X) (the collection of all subsets of X) is a topolog-
ical structure, and that the idea of continuity is readily defined using this
structure.
(1.A.1) Definition. Suppose X is a nonempty set and U is a collection

of subsets of X. The pair (X,U) is a topological space if and only if the
following conditions are satisfied:
(i) If each Uy € U when a € A, then (U,¢, Ua) € Ui;
(ii) If Uy € U where a € K and cardK < oo, then ((\xe% Ua) € U;
(iti) @€U and X EU.
The family U is called a topology (or a topological structure) for X, and the
elements of U are called the open sets of the topological space (X,U).
Note that if each point x in a set U is contained in an open set V, C U,

then U is open (why?). We will make frequent use of this observation.
The reader should also observe that the family of unions of open intervals
in R! yields a topology for R’.
(1.A.2) Definition. If / and V are topologies for a set X, then U is smaller
than Y (or V is larger than U) if and only if UC V.
(1.A.3) Definition. Suppose (X,U/) and (Y, V) are topological spaces and
f:X > Y. Then f is continuous at a point c € X if and only if whenever
f(c) € V € YV, there is aU € U containing c such that f(U) C V. If f is
continuous at each point c € X, we say f is continuous.
It is often convenient to use the following characterization of continuity.

(1.4.4) Theorem. Suppose (X,U/) and (Y,V) are topological spaces and
f :X —Y. Then f is continuous if and only if for each V € V, f-1(V) €U.
Proof. Suppose f is continuous and V € V. If f-1(V) = @ we are
through. Hence, suppose € f~'(V). Then f(x) € V, and since f is
Topological Spaces and Continuity 13
continuous, there is aset U; € U such that x € Uz and f (U,) C V. Therefore

we have U; C f~'(V). It follows from (i) of (1.A.1) that f-1(V) is open,
singe fo" (Vil= Wile se fo"(V)}:
Conversely, suppose f—!(V) € U for each V € V. Let x € X and suppose
f(z) €V €V. Then we have f-'(V) €U, x € f-!(V), and f(f7!(V)) CV.
Consequently, f is continuous at z, and since xz was arbitrary, we conclude
that f is continuous. O
Frequently it is awkward to describe in full the topology U one wants

for a set X; this situation is rendered more tractable by the notion of a basis.
(1.A.5) Definition. If (X,U/) is a topological space, then a basis of U is a

subcollection B of U with the property that if € X and U is an open set
containing z, then there is a V € B such that x € V C U. In other words,
each U € U can be written as a union of sets from B.
As an example, the open intervals of R! constitute a basis for the topol-

ogy determined by the open sets in R! discussed earlier. This topology for
R! henceforth will be referred to as the usual topology for R! (we will soon
introduce other topological structures on the set of real numbers).
Next, let us define the usual topology for the plane R?. Here it is
particularly advantageous to give the topology in terms of a basis. For each
point z = (x,y) € R? and each real number e€ > 0, let
| Pe {(u,v) ER? : J/(r—u)? + (y —v)? <e}
Then B = {S.(z) : z € R* and e > 0} is a basis for the usual topology U

for R?, where U consists of all possible unions of members of B. Typical
elements from B and U are indicated below. The sets S.(z) are called open
disks in R?.
That B is a basis for a unique topology on R? is a consequence of the

following theorem and exercise.
(1.A.6) Theorem. Suppose B is a collection of subsets of a set X. Then

B is a basis for some topology on X if and only if (i) X = U{B : Be B},
and (ii) whenever Bi,By € B and x € B, Bo, there is a B € B such that
1S BC B, al Bo.
Proof. Suppose B is a basis for a topology U on X. If x € X, then

ze xX EU. Since B is a basis forU, there is a B € B such that re BC X,
and hence X = U{B : B € B}. If By,By € B and x € B, NM Bz, then since
B, and Bz also belong to U, we have that B,; 1 Bz € U. Consequently, there
isa Be Bsuchthatre BC BN Be.
Conversely, let U = {U C X : U is a union of members of B}. Clearly
X,@ €U and arbitrary unions of members of U are in U. If U,V € U and
xz € UNV, there are sets B, Bi,By € B with z € B;} CV, 2 € By CU, and
z€BCB,NB, CcUNV. Hence UNV is a union of members of B. We
leave the trivial induction proof that the finite intersection of members of U/
is in U to the reader. Henceforth, in similar situations we will frequently not
mention that a further argument is necessary. O
(1.A.7) Exercise. Suppose that B is a basis for two topologies U and U on

a set X. Show that U =U. That is, B is a basis for a unique topology.
It is important to note that a given topology may have distinct bases.

For each z = (z,y) € R? and each c€ > OQ, let T(z) = {(u,v) € R?
|c — ul < € and |y — v| < e}. Then
B = {T.(z) : z € R? ande > O} isa
basis for precisely the same topology as that generated by B = {S,(z) : z€
R? and c > 0}. This follows immediately from the next exercise.
(1.A.8) Exercise. Show that two bases B and B generate the same topology
on a set X if and only if whenever x € B € B, there is a B € B such that
z € BC B, and whenever x € B € B, there isa B € Bsuchthatr € BC B.
(1.A.9) Definition. Bases that generate the same topology on a set X are
said to be equivalent.
It follows from the previous exercise that B = {(x,y) : x,y are rational}
and B = {(z,y) : 2,y are irrational} are equivalent bases for the usual
topology on R!.
Continuity may also be expressed in terms of basis elements.
(1.4.10) Theorem. Suppose (X,U/) and (Y,V) are topological spaces and
B is a basis for U and B is a basis for V. A function f : X - Y is continuous
at a point c if and only if whenever f(c) € B € B, there is a B € B such that
c€ Band f(B) CB.
Proof. Suppose f is continuous and f(c) € B © B. Since B is also in V
and f is continuous, there are sets U € U and B € B such that ce BCU
and f(c) € f(B) Ccf(U) CB.
For the converse, suppose f(c) € V € V. Then there is a B € B such
that f(c) € B C V, and by hypothesis, there isa B € B CU such that c € B
and f(B)cCBcV. Oa
Further Examples of Topological Spaces 15
(1.A.11) Exercise. Suppose (X,U/) and (Y,V) are topological spaces and
Bisa basis for Y. Show that a function f : X — Y is continuous if and only
if f—!(B) is open in X for each B € B.
B. FURTHER EXAMPLES OF TOPOLOGICAL SPACES

The reader should verify that each of the following are topological spaces.
1. Discrete topology. Let X be any nonempty set and U = P(X).
2. Indiscrete topology. Let X be any nonempty set and U = {0, X}.
3. Half-open interval topology. Let X be R!. Sets of the form [a, b) give
a basis for a topology U on X (which is distinct from the usual topology on
Ri}
4. Open ray topology. Let X be R! and U be the topology determined
by a basis consisting of sets of the form (a,0o) where a € R'.
\...Usual topology for R”.. For 2 = (2,49;5...,2n) € R” and.e > 0,
etd ZA iis one winstn) CEC wena) )5) alae U4)? =e}. Lien B =
{S.(z) : z € R” ande > 0} is a basis for a topology on R”, the usual
topology. The set S.(z) is called an (open) e-ball about z.
6. Order topology. Let (X,<) be a linearly ordered set. For each
z,y € X with x < y, let (x,y) = {ce X : 2 <cc< y}. Sets of the form
(x,y) together with sets of the form {c € X : c < x} and sets of the form
{c € X : x <c} constitute a basis for a topology on X, the order topology.
Note that the usual topology on R! is an order topology.
7. Finite complement topology. Let X be an infinite set. Define a
topology U on X by taking the collection that consists of the empty set
together with all subsets A of X with the property that X \ A is finite.
8. Countable complement topology. Let X be an uncountably infinite
set. Define a topology U by taking the empty set together with all subsets
A of X with the property that X \ A is countable.
9. Fortissimo topology. Let X be any uncountable set and choose a
point p of X. Then U = {U Cc X : X \Uis countable or p ¢ U} is a
topology for X.
10. Let X be an arbitrary nonempty set and p a particular point in X.
Then U, defined by U = {UC X : U=X orp ¢ VU}, is a topology for X.
(1.B.1) Notation. (a) Henceforth €” will always denote R” with the usual
topology (see also the remark after (1.C.4)).
(b) We will frequently abuse the notation and say that X is a topological
space, when we mean that (X,U) is a topological space for some topology
U.
This section concludes with a brief introduction to a few of the more

convenient properties that a topological space may possess.
(1.B.2) Definition. A topological space is a Hausdorff (or Tz) space if

and only if each pair of distinct points a and b are contained respectively in
disjoint open sets U and V.
(1.B.3) Exercise. Determine which of the above spaces 1-10 are Hausdorff.
(1.B.4) Definition. A topological space (X,U/) is second countable if and
only if U has a basis consisting of a countable number of sets.
For instance, €! is second countable, since the family of open intervals

with rational endpoints constitutes a countable basis for the usual topology
on R!. An uncountable discrete space is obviously not second countable.
(1.B.5) Exercise. Determine which of the above spaces 1-10 are second
countable.
(1.B.6) Definition. If (X,U/) is a topological space and x € X, then an
(open) basis for U at x (often called an (open) neighborhood basis for U at
x) is a subcollection B, C U with the properties that
(i) ifc € U EU, there is a V € B, such that c € V CU, and
(it) = E (\yep, V-
(1.B.7) Definition. A topological space (X,U) is first countable if and only
if there is a countable neighborhood basis at each point x € X.
Note that €” is first countable, since for each z € €”, the set {51 /;(z) :
i € Zt} is a countable neighborhood basis at z. In fact, by Problem 1 of
section B every second countable space is also first countable.
(1.B.8) Exercise. Find a space that is first countable but is not second
countable.
C. METRIC SPACES: A PREVIEW

Topological spaces for which a concept of distance between points is defined
are of particular interest. These spaces are known as metric spaces, and they
play a fundamental role throughout the book.
(1.C.1) Definition. A metric space is a pair (X,d) where X is a set and

d: X x X — (0,00) is a function that, for all z,y,z € X, satisfies the
properties:
(1) d(x,y) = 0 if and only if 2 = y (reflexive property);
(ii) d(x, y) = d(y,x) (symmetric property);
(112) d(x,z) < d(x,y) + d(y, z) (triangle inequality).
Metric Spaces: A Preview 17
The function d is called the distance function or the metric for the metric
space (X,d).
The first condition implies that the distance from a point to itself is
0 and that the distance between distinct points is positive. Condition (ii)
states that if you walk from x to y, you go the same distance as if you walk
from y to x. The third condition is as a generalization of the idea that the
shortest distance between two points is a straight line, (or that the length of
one side of a triangle is less than the sum of the lengths of the other sides).
(1.C.2) Notation. Suppose that (X,d) is a metric space, € X, ande > 0.
We denote the set {y € X : d(z,y) < €} by S4(z). When the metric is clear
from context, d is frequently omitted and we simply write S,(z).
The distance function d can be used to define a topology for X. In the
next theorem we show that if (X,d) is a metric space, then B = {S¢(x) : x €
X, € > 0} is a basis for a topology on X. This topology is called the metric
topology or the topology induced by the metric d. In the future, whenever
we speak of a metric space we will assume that it has the metric topology
(unless we specifically impose a different topology).
(1.C.3) Theorem. If (X,d) is a metric space, then B = {S,(x) : © €
X, € > 0} is a basis for a topology on X.
Proof. By (1.A.6) it suffices to show: ifp € S.(z)N S,(y), then there
is a positive number v such that S,(p) C S.(z)N Sy(y). Let A1 = d(z,p)
and A2 = d(y,p). Then if v is any positive number less than both e — A, and
ps — Ag, it is easy to see that S,(p) is the required basis element containing
p.
We list some examples of metric spaces. The reader should verify that
each example satisfies properties (i), (ii), and (72) of (1.C.1).
(1.C.4) Examples.
1. (The usual metric for R'.) Let X be R’ and define d(z,y) = |x —y].
2. (The usual metric for R?.) Let X be R? and define d(z,y) =
(z1 — yi)? + (2 — y2)? where & = (21,22) and y = (yi, y2). Use the
Minkowski inequality, (0.F.2), to show that d satisfies the triangle inequality.
3. (The tazicab metric). Let X be R? and define d(x,y) = |z1 — yi| +

|v — yo| where x = (21,22) andy=(y1,y2).
4. (The usual metric for R"). Let X be R” and define d(z,y) =
>y-1
(ti — yi)”, where = (21,22,...,2n) and y = (1, Y2,---,Yn)- Use
the Minkowski inequality to show that d satisfies the triangle inequality.
5. (The discrete metric). Let X 4 @ be arbitrary and define d(z,y) = 1
ife 7 y and dz. y)=O0u ts =v
6. Let X = {f : f:Y — R' and f is bounded} and define d(f,g) =
sup,cy |f(y) — g(y)|. In this construction, Y may be any set. (A function
f :Y — R! is bounded if and only if there is a positive number M such that
lf(y)| < M for each y € Y.
In the future, we will assume that €” comes equipped with the usual
metric unless we specify another (equivalent) metric. Note that the metrics
in examples 2 and 3 induce the same topology. That is, if U is the topology
generated by the usual metric for R? and U is the topology arising from the
taxicab metric, then U = UW. We delay the proof of this until after (1.C.11)
which gives a more convenient criterion for proving these topologies coincide
than (1.A.8) does.
(1.C.5) Exercise. Prove or find a counterexample: (a) all metric spaces
are second countable; (b) all metric spaces are first countable.
(1.C.6) Exercise. Show that all metric spaces are Hausdorff.
We just observed that a topology can be derived in a natural way from a
metric. An interesting and quite important question (treated to some extent
in Chapter 10) concerns the reverse implication—given a topological space
(X,U), when does there exist a metric for X that induces U? A topological
space (X,U) is metrizable if and only if it is possible to define a metric d on
X such that the metric topology induced by d coincides with WU. Note that
Example 5 above guarantees the existence of at least one metric for any set,
but clearly the topology associated with this metric is discrete. There is a
great advantage in having a metric function that induces a given topology,
but fairly stringent conditions are necessary to guarantee the existence of
one.
(1.C.7) Exercise. Suppose X is an arbitrary set with more than one point
and with the indiscrete topology U. Is (X,U) metrizable?
(1.C.8) Exercise. Is R! with the open ray topology metrizable?
(1.C.9) Exercise. Explain why Theorem (1.C.11) below is easier to apply

than (1.A.8).
It is possible that two distinct metrics d and d on a set X will induce
the same topology. In this case, d and d are said to be equivalent metrics.
Building New Spaces from Old Ones 19
(1.C.10) Example. Suppose that (X,d) is a metric space. If we define

d: X x X — (0,00) by setting d(x,y) = min{d(zx,y), 1}, it is easy to show
that dis a metric. The equivalence of d and d follows immediately from the
next theorem.
(1.C.11) Theorem. Suppose d and d are metrics for a set X. Then d and
d are equivalent if and only if for each x € X and for each r > 0, there are
numbers s > 0 and § > 0 such that S4(x) c S4(x) and S4(x) C $4(z).
Proof. Suppose d and d are equivalent, z € X andr > 0. Since S4(z)
is an open set containing z, there is a basis element Se(y) in the topology
generated by d such that z € S¢(y) C S42). Ifs =e- d(z,y), then
S4(x) C S4(x). Similarly, an § > 0 can be found so that Six 1G S2(z),
To prove the converse, let U/ be the topology induced by d and U be the
topology induced by d. Suppose U € ¥Y and zg is an arbitrary point of U.
Then there is an r > 0 such that S4(x) C U. By the hypothesis there is an
8 > 0 such that x € S¢(x) Cc S4(x) C U; thus, U is a union of U/-open sets
and consequently U Cc UW. An analogous argument shows that (Vile Fale
As an application of (1.C.11) we prove the claim that the metrics in

parts 2 and 3 of example (1.C.4) are equivalent. Let U be the topology for
R? induced by the usual metric, U the topology induced by the taxicab
metric, = (%,%2) € R’, and r > 0. It is clear that if § = r, and
y= fee E S4(a), (ie., d(2, oo [zi—yi| + |22 — yo| < 8), then d(z,y) =
s/ (@1'= 91)? wee (21 —y1)? + (@2= yo)? = |t1 — yi | + [te —
yo|<8=r. ae S4(x) C S4(x). Now, if s=r/2 and y = (y1,y2) € S$(z)
(ie., d(z,y) = V(t1 — yi)? + (Z2 — y2)? < 5), then d(z,y) = [x1 — yi| +
|z2 — yo| < 2/(a1 — y1)? + (a2 — yo)? < 28 =r. Hence, S$(x) C S#(z). B
(1.C.11) the metrics are equivalent.
(1.C.12) Definition. A metric space (X,d) is bounded if and only if
sup{d(z,y) : z,y € X} is finite.
(1.C.13) Corollary. Every metric space is equivalent to a bounded metric
space.
Proof. The proof is immediate from (1.C.10) and (1.C.11).
D. BUILDING NEW SPACES FROM OLD ONES

The Relative Topology
There is a natural way to turn a subset A of a space (X,U) into a

topological space in its own right. To do this we must determine which
subsets of A are to be designated as open (in A). The next definition does
this in an obvious way.
(1.D.1) Definition. Suppose that (X,U/) is a topological space and that

A is a subset of X. A set V C A is said to be open in A (or A-open or
relatively open) if and only if V = UNA for some U:€ U. The collection of all
A-open sets forms a topology for A, called the relative topology. Henceforth,
whenever a subset A of a space (X,U/) is considered as a topological space,
it is assumed that A has the relative topology. In this context A is often
called a subspace of (X,U)
(1.D.2) Exercise. Show that the relative topology is in fact a topology.
It should be noted that A-open sets need not be open in the original
space X. For instance, if X= €! and A = (0,1), then (0, $) = (—2, $)N[0, 1)
is open in A but not in X. In fact, A itself is an open subset of itself, but
not open in €!. Observe, however, that if A is chosen to be an open subset
of (X,U), then A-open sets also turn out to be open in X.
(1.D.3) Theorem. Suppose U C A C X where A is open in X and U is
open in A. Then U is open in X.
Proof. Since U is open in A, there is an open set V in X such that
VnA=U. However, since A is open in X, we have that AN V is also open
in X, and the theorem follows. O
A second trivial but useful result is the following.
(1.D.4) Theorem. Suppose W and Z are subspaces of a topological space

X, and U C WZ is open in both W and Z. Then U is open in W UZ.
Proof. Since U is open in both W and Z, there are open sets O, and
Oz in X, such that O, NW =U and O02 Z =U. The reader should verify
that (0; NO2)N(W UZ) =U, which equality implies that U is an open set
in the relative topology forW UZ. OQ
(1.D.5) Exercise. Show that if X and Y are topological spaces, f : X > Y

is continuous, and A C X, then f|,4 is continuous.
(1.D.6) Exercise. Show that if (X,d) is a metric space and A C X, then

the relative topology for A is metrizable. In fact d|4,4 is a metric that
induces the relative topology.
The (Finite) Product Topology
The point set R” is the Cartesian product of R! with itself n times. It

is reasonable to wonder whether the usual topology defined for R” might
be related in some way to the usual topology for R!. This relationship is
best seen if we first generalize and consider the notion of a finite Cartesian
product of arbitrary spaces. In a later chapter, infinite Cartesian products
will be treated.
Building New Spaces from Old Ones 21
(1.D.7) Definition. Suppose (X1,U4), (X2,U2),..., (Xn,Un) are topolog-

ical spaces and [];"_, X; is the Cartesian product of the sets MaliXo vest iat
The product topology for Lge Xj; is the topology with a basis consisting of
all sets of the form U = U, x U2 x --- x Un, where U; € U; for i= 1,2,...,n.
(1.D.8) Remark. If (X1,4), (X2,U2),..., (Xn, Un) are topological spaces,

then, unless otherwise stated, i hea X; will be assumed to have the product
topology.
(1.D.9) Exercise. Show that the product topology is in fact a topology,

i.e., show that the designated sets actually form a basis for some topology.
(1.D.10) Theorem. If U/ is the usual topology for R” and V is the product
topology for €! x €! x --- x €}, then U = V.
Proof. To see that U C V, let U € U. Then U = YU, cy S., (x). To
show U € J, it suffices to show that for every x and for every « > 0,
the set S.(x) contains a basic open set of V which contains x. Now let
Guz (xig23y.-4)/2%,)'€ R®. and ie. >.0.,For each,2;, let.Uz; = {y 6. €1 .
|z; —y| < €//n}. Then z € U, x U2 x ---
x Un, and if u = (uw, U2,.--, Un) €
U, x U2 x --- xXUn, we have
du,o) =
Hence U EU.
Similarly, to show that V € U it will suffice to show that for a typical
basis element of the product topology, B = U, x U2 x --- x Un, and for
z = (%1,22,..-,2n) € B, there is an € > 0 such that x € S,(x) C B.
Hence, suppose that z € B. Then for each z;, there is an €; > 0 such that
{ye €1 : jn; —y| < ex} CU;. Let e = min{e, : i =1,2,...,n}. We show
thatsSe@ec Uy x Ua eX Uns Ve Sle 225 104 Fn) S.(z), then
n
\x5 = zi| < » (ei = za)? <<16R<G,
11
Therefore z; € U; foreachi. O
(1.D.11) Definition. For each i = 1,2,...,n define the ith projection map,
Der Meee Aa x Xn) ea) = ae
piesc
> Xi DY Pi(@ig
(1.D.12) Theorem. Suppose (X1,l), (X2,U2),.--, (Xn, Un) are topolog-

ical spaces and [];"_, Xi is the associated product space. Then the projection
maps p; are continuous. Furthermore, the product topology is the smallest
topology for which all of the p; are continuous, i.e., if / denotes the product
topology and Y CU is any topology on the set Piet X; for which each p; is
continuous, then V =U.
Proof. Suppose U; is an open set in X;. Then p; (Ui) =A Mg eX
U; x Xia1 X +++ X Xn is clearly a member of U, and therefore p; is continuous
(1.4.4).
Suppose now VY C U and each p; is continuous with respect to V. To
demonstrate that V = U it suffices to show that each basis set U = U; x
Uy x --- x Uy, in U belongs to V (why?). Since p; is continuous for each 2, it
follows that p~!(U;) is in V. But U = U, x Uz x --- x Un = (V1,
|(Ui),
andhenceUeEV. QO
(1.D.13) Exercise. Suppose for 1 = 1,2,...,n, fj : X; — Yj; is continuous.
Show that the map [J fi : []j_, Xi: > []jL, Yi defined by
(II Ft) (1, 23, 20 : Zn) = (fi(z1), fo(z2), ea » fn(n))
is continuous. Is every (set) function from [];_, X; to []j_, Yi of this form?
The Disjoint Union Topology
We now consider one further example of how a new topology may be

generated from old ones. Suppose that {(Xq,Ua) : a € A} is a collection
of pairwise disjoint topological spaces, i.e., if a,G € A and a # B, then
XaAX~g =O. Let X = Ugeq Xa- Define a topology U for X be declaring a
set U C X to be open if and only if UNM Xq, is open in Xq for each a € A.
(1.D.14) Exercise. Show that the family U/ defined above forms a topology.
The topological space obtained in the manner just described is custom-
arily called the free union of the spaces X,, and the corresponding topology
is frequently referred to as the disjoint union topology. Requiring that the
spaces in question be pairwise disjoint can be avoided, since any collection of
sets may be replaced by a disjoint collection of sets as follows. Suppose that
{X, : a € A} is a family of sets. For each a € A, define X, = Xq x {a}.
Then clearly for a ~ B, XgN Xz = . Now suppose {(Xq,Ua) : a € A} is
a collection of topological spaces. If a € A, let Xq = Xq x {a} and define
a topology Ug for Xq by declaring a subset Ug x {a} C Xq x {a} open if
and only if U, is open in Xqg. Then the free union of the X, is defined to
be the free union of the disjoint spaces (oxhae) described above.
(1.D.15) Exercise. Suppose {A, : a € A} is a collection of disjoint

subsets of a topological space X. Let U denote the relative topology for
U{Aa : a € A} and let V denote the disjoint union topology for the same set,
where each A, is endowed with the relative topology. Discuss the relationship
between U and V. [Hint: Think about A, = {x} for x € X_]
A Potpourri of Fundamental Concepts 23
E. A POTPOURRI OF FUNDAMENTAL CONCEPTS
Although the concepts introduced in this section are relatively simple, the
reader should give them careful consideration. A firm grasp of these ideas is
needed before one can successfully proceed to the more intriguing aspects of
topology.
Closed Sets
(1.E.1) Definition. Suppose (X,U/) is a topological space. A subset A of

X is closed if and only if X \ A is open.
Note that if U is open in X, then X \ U is closed. It follows easily
from (1.A.1) and the De Morgan rules that finite unions as well as arbitrary
intersections of closed sets are closed. Indeed, if C is a family of subsets of
X that is closed under finite unions and arbitrary intersections (and hence
contains X and @), then C is the family of closed sets for the topology U =
{U : U=xX\C for C € C}. Always keep in mind that subsets of a space
may be neither open nor closed! E.g., [0,1) c €!. Furthermore, infinite
unions of closed subsets may fail to be closed: indeed, the set of rational
numbers is a countable union of singletons (each of which is closed in €'),
and yet the rationals are not closed in €!.
(1.E.2) Exercise. Suppose X is a topological space, A C X has the relative
topology and F' c A. Show that F is closed in A if and only if there is a
closed set C' in X such that F= CNA.
(1.E.3) Exercise. Show: If F is closed in G and G is closed in a topological
space X, then F is closed in X.
An arbitrary subset of a topological space may be “closed off” in the

following way.
(1.E.4) Definition. Suppose (X,U/) is a topological space and A C X. The
closure of A (in X) is defined to be (\{F : F is closed in X and A C F},
and is denoted by A* (or when the context is clear, simply by A).
(1.E.5) Theorem. If A and B are subsets of a topological space X, then

(i) ACA,
(ii) A is closed, _
(iii) if A is closed, then A = A,
(iv) A is the smallest closed subset of X containing A, in the sense that
if AC Bc A and B is closed, then B = A,
(v) if AC B, then AC B,
(vi) AUB = AUB, and
(vii) ANB C ANB. ‘
Proof. (i) Trivial. (27) Recall that the intersection of closed sets is closed.
(iii) Since A is contained in each set of the intersection, and A (being closed)
is one of those sets, then A C A. On the other hand, A is always contained
in A. (iv) Since B is closed, we have A C B. (v) Trivial. (vi) Since the
finite union of closed sets is closed, we have by (iv) that AUB C AUB. On
the other hand, A C (AUB), and hence by (v), A C AUB. Similarly, B is
contained in AU B and therefore AUB C AUB. (vii) Since ANBCA
and ANB C B it follows from (v) that ANB CANB. a
(1.E.6) Exercise. Find an example to show that AN B is not necessarily

equal to AN B.
(1.E.7) Theorem. Let X1, Xo,...,Xn be topological spaces and suppose

Aye \vlor v= 1,2.*. ite Len
gg So SO May Mn ee Ae
(Warning: Not every closed set in []j_, Xj is of this form!)

Proof. We prove the theorem for n = 2; an easy induction argument
yields the more general result. First, observe that if F’ and G are closed
subsets of spaces X and Y, respectively, then F x G is closed in X x Y. To
see this, note by (0.B.2) that (X x Y)\(F'xG)= (X\F) xYJU[X x (Y\G)]
which is the union of two basic open sets. Since a x AG is closed, it follows
immediately from part (iv) of the previous theorem that A; x Az C 7 veccay
For the converse we show (X, x X2)\(A,
x Az)C (X1 x eV AyoeAs).
Suppose that (x,y) € (X1 x X2)\Ai x Ao. We note (X1 x X2)\ (Ai x A2) is
an open set, and clearly [(X1 x X2) \(A1 x A2)] (Ai x A2) = @. Hence there
is a basic open set U x V such that (x,y) € UxV C (X1 x X2)\(Ai x Ag). In
particular, (U x V)M(A; x Az) = 9, and therefore UN Ay = @ or VN Ag = 9.
(If there were an a; € UNM A, and ag € VN Ag, then (a1, a2) €UxV and
A; * Ao.2-) It follows that either X, nes 3))At or X2\V D Ag, so either x ¢ Ay
or y ¢ Ag. Consequently (x,y) ¢ A, x Aadtl
(1.E.8) Definition. A topological space (X,U) is T; if and only if each

point in X is closed.
(1.E.9) Exercise. Show that a space X is T, if and only if for each pair
of distinct points z,y € X, there is an open set containing xz that does not
contain y and an open set containing y that does not contain z.
A Potpourri of Fundamental Concepts 25
Accumulation Points
Perhaps no concept in either topology or analysis is of more importance

than the notion of an accumulation point. As we proceed it will become
increasingly apparent that the fabric of topology is snagged on this idea.
(1.E.10) Definition. Suppose X is a topological space and A is a subset
of X. A point z € X is an accumulation point of A if and only if each open
set containing z has nonempty intersection with A \ {zx}.
It should be noted that in the literature, accumulation points are some-
times referred to as limit points. We will reserve the term “limit point” for
a different, but related concept.
(1.E.11) Examples.
1. If X = £' and A = (0,1), then any z such that 0 < x < 1 is an
accumulation point of A.
2. If X = €! and A is the set of rationals, then every point of X is an
accumulation point of A.
3. If X = €! and A = Z, then no point of X is an accumulation point
of A.
(1.E.12) Exercise. Let X = R! and A = (0,1). Find the accumulation
points of A with respect to each of the different topologies we have imposed
on R!.
(1.E.13) Definition. Suppose A is a subset of a space X. The derived set,
A’, of A is defined by A’ = {z : x is an accumulation point of A}.
(1.E.14) Theorem. Suppose A is a subset of a space X. Then
(i) A= AUA’, and
(ii) A is closed if and only if A’ C A.
Proof. (i) If a point x does not lie in AU A’, then there is an open set
containing z that misses A (and hence A’). Consequently,X \ (AU A’) is
open so AUA' is closed. Thus, we have A C AUA’. If x ¢ A, then since A is
closed, x belongs to an open set that does not intersect A. Thus « ¢ AUA’,
and hence A = AUA’. (it) If A is closed, then A = A = AUA’ and therefore
A' C A. Conversely, if A’ C A, then A = AUA!' = A and hence A is closed.
O
It immediately follows from the above that: z € A if and only if every

open set containing x has nonempty intersection with A.
Interior, Exterior, and Frontier
Suppose A is any subset of a topological space X. It will be useful to consider

the following three sets that we construct from A.
(1.E.15) Definition. Suppose A is a subset of a space X. The interior

of A, denoted by intA or A®, is the union of ‘all open sets of X that are
contained in A. The exterior of A, denoted by ext A, is the union of all open
sets in X that are contained in X \ A. The frontier of A, denoted by Fr A,
consists of all points zx € X with the property that each open set containing
zx intersects both A and X \ A.
For example, if A = (0,4] c €!, then intA = (0,4), ext A = (—00,0)U
(4,00), and FrA = {0, 4}.
(1.E.16) Theorem. If A is a subset of a space X, then int A, ext A, and
FrA are pairwise disjoint and X = (int A) U (ext A) U (FY A).
Proof. If x € X and & not in either intA or in ext A, then from the
preceding definition each open set containing x must intersect both A and
X \ A. Therefore z € FrA. That the three sets are pairwise disjoint is also
immediate from the definition. O
Although any of the three sets int A, ext A, and Fr A may be empty, it
should be obvious that int A is the largest open set contained in A and ext A
is the largest open set contained in X \ A.
(1.E.17) Exercise. Show that if A C X, then Fr A = AN(X \ A).

Frontiers of products are related to products of frontiers and closures as
follows.
(1.E.18) Theorem. Suppose A C X and B CY. Then
Fr(A x B) = ((FrA) x B) U(A x (FrB)).
Proof. Suppose (z,y) € Fr(Ax B). By (1.E.17) Fr (A x B) is contained

inAxB =AxB,soré€
Aand ye B. If x ¢ FrA and y ¢ FYB, then
by the previous exercise, ¢ X \A andy¢Y\B. LetU=X\(X\A)
and V = Y \(Y \B). Then (z,y) € U x V and U x V fails to intersect
(X x Y) \(A x B), contradicting (x,y) € Fr(A x B). Consequently, either
z € FrA ory € FYB, and we have Fr(A x B) C ((FrA) x B)U(A x (Fr B)).
Suppose now that (x,y) € (Fr A) x B and let U x V be a basic open set
such that (x,y) € U x V. We want to show that (U x V)M(A x B) #0 and
(Ux V)N((X x Y)\(Ax B)) #9. Since z € FA and y € B, we have that
UnNA#0 and VNB F 9; hence it follows that (Ux V)N(A x B) £9.
Since x € Fr A, we have UN (X \ A) £@. Let z € UN(X \ A). It follows
from (0.B.2) that {2} x Vc (Ux V)N((X x Y) \(A x B)). Consequently,
((FrA) x B) C Fr(A x B).
In a similar manner, one may show A x (FrB) CH(AxB). O
Continuity 27
(1.E.19) Exercise. Generalize (1-E.18) to finite products.

We conclude this section with an important definition.
(1.E.20) Definition. If X is a topological space and z € X, then a subset

N of X is a neighborhood of « if and only if there is an open set U in X such
that reU CN.
The reader should consider each of the preceding definitions to see
whether the phrase “open set containing x” may be replaced by “neigh-
borhood of z” without changing the meaning of the definition.
F. CONTINUITY
In this section and its problem set we examine the concept of continuity in
light of our newly developed topological ideas. It is reasonable to suppose
that continuity can be expressed in terms of each concept that can be used
to determine a topology. We first establish a few useful generalizations of
some familiar results.
(1.F.1) Theorem. Suppose X, Y, and Z are topological spaces, and that

f:X > Y andg:Y — Z are continuous functions. Then gf : X > Z is
continuous.
Proof. Suppose U is open in Z. Then g~!(U) is open in Y, and
fut(g
7(@)) 38 Open in KX. But f-*(gs(U)) is just (gf) (A). 8
The solutions to the next exercise use nothing more than manipulation of e’s
and 0’s.
(1.F.2) Exercise. Show that the following functions are continuous:
(i) f :€1 — €!, where f is defined by f(z) = 27;
(ii) f : E+ \ {0} > &1, where f is defined by f(z) = 1/z;
(iii) f : €1 > E1, where f is defined by f(x) = |z|;
(iv) f :€! > E!, where f is defined by f(x) = az, a€ R’.
(1.F.3) Theorem. Suppose f and g are continuous functions mapping a
topological space X into €1. Leth = f +9, k = f-g, and (if g(x) # 0
for all x) let 7 = f/g. Then h, k, and j are continuous. (Here, of course,
(f + 9)(z) = f(x) + g(z), etc.)
Proof. Suppose z € X. We first establish the continuity of h at z.
Let U be an open interval of radius € and center h(x). Since f and g are
continuous at x, there are open sets V and W containing x such that f(V)
is contained in the interval (f(z) —€/2, f(z) +€/2) and g(W) is contained in
(g(x) — €/2, g(x) + €/2). If Y =V NW, it is easy to check that h(Y) CU.
In order to show that k is continuous at x, we begin by demonstrating the
continuity of f? = f-f. This, however, follows immediately from the exercises
above and the previous theorem, since f? can be written as a composition of f

and the function p defined by p(x) = x?. Since f-g = (1/4)((f+g)?—-(f-g)”),
we can again apply (1.F.2) along with the first part of this theorem to see
that k is continuous.
The function 7 = f/g can be considered as the product of f and 1/g.
The continuity of 1/g is immediate from the observation that 1/g is merely
a composition of g and the function q defined by q(x) = 1/z. Then, since 7
is a product of continuous functions, the proof of the theorem is complete.
Oo
The next theorem shows that continuity may be expressed as elegantly

in terms of closed sets as in terms of open sets.
(1.F.4) Theorem. Suppose (X,U) and (Y,V) are topological spaces and
f:X —-Y. Then the following conditions are equivalent:
(i) f is continuous;
(ii) for each V € V, f~1(V) is open in X;
(iit) for each closed set D C Y, f~!(D) is closed in X.
Proof.
(i) = (it) This is just (1.A.4).
(it) = (tit) Suppose D is closed in Y and let
V = Y \ D. Then V is
open, and hence f~!(V) is open. But f~!(D) = X \ f~1(V), which proves
that f—!(D) is closed in X.
(iit) = (ti) Suppose that V € V. Then Y \ V is closed, and hence
=1(Y \ V) is closed in X. But f-(Y \V) = X \ f-*(V) and therefore
f-(V) isopenin X. O
We next give a criterion for continuity that is expressed in terms of
the closure operator. Similar results for the interior operator, the derivation
operator, and the frontier operator can be found in the problems.
(1.F.5) Theorem. Suppose (X,U) and (Y, V) are topological spaces and
f:X —Y. Then f is continuous if and only if f(A) C f(A) for each subset
A of X.
Proof. Suppose f is continuous, and A is a subset of X. Since A C
f-1(f(A)) C f7!(f(A)) and the latter set is closed by (ii) above, we obtain
Ac f71(f(A)). It follows that f(A) c f(A).
Conversely, suppose B is a closed set in Y. Consider A == f-\(B ).
By hypothesis [Male )) ¢ f(f-(B)) c B = B. Hence, ee
Hh e
f-f(f-1(B)) C f-!(B), so f-!(B) is closed. O
The next result will see continual service during succeeding chapters.
Continuity 29
(1.F.6) Theorem (Map Gluing Theorem). Suppose A and B are closed

subsets of a topological space X. Let Y be an arbitrary space and suppose
f : A — Y and g: B + are continuous functions such that funn =
g|ans- Then the function h : (AU B) > Y defined by
F(z) ife@eA
Maye
g(a) if pe B
is continuous.
Proof. The condition f|4qng = glang is necessary in order for h to

be well-defined. For continuity, we suppose C is a closed subset of Y and
apply (1.F.4). Thus f~1(C) is closed in A, and hence in X since A is closed
in X. Similarly, g~'(C) is closed in both B and X. However, h7!(C) =
f~1(C) Ug71(C), and therefore is closed in AU B and X, which completes
the proof. O
(1.F.7) Theorem. Suppose X, X2, and Y are topological spaces and

f:Y 7+ X, x Xe, where X, x X> has the product topology. Let f; = pif
and fz = pof, where p; and pz are the projection maps. Then f is continuous
if and only if f; and fz are continuous.
Proof. If f is continuous, then so are f; and f2 by (1.F.1). To prove the
converse, suppose U x V is a basis element in the product topology. We show
f-1(U x V) is open in Y. First note that U x V = (U x X2)N(X1 x V) =
pyCU) ps (VY), At follows.that f7(U x V) =f, (WU) op, V)) =
f-'py*(U)Nf-!py'(V) = f7'(U)
fz *(V). The latter is an open set under
the hypothesis that f; and f2 are continuous. O
In metric spaces, continuity may be expressed elegantly in terms of

sequences. We remind the reader that a sequence in a set X is a function
f :D-—X where D is a subset of N. If for each i € D, f(i) = 2;, then the
sequence f is usually denoted by {z;}. Unless otherwise stated, it will be
assumed that D = Z*.
(1.F.8) Definition. A sequence {z;} in a topological space X converges to
a point x € X if and only if for each neighborhood U of z, there is a positive
integer Ny such that x; € U whenever 1 > Ny. In this case, z is called a
limit point of the sequence {x;}.
Observe that if U is any neighborhood of z, then the members of a
sequence {z;} that converges to z, are eventually trapped in U.
(1.F.9) Theorem. Suppose (X,d) and (Y,d) are metric spaces. The func-
tion f : X — Y is continuous at a point x € X if and only if whenever a
sequence {x;} in X converges to z, then the sequence {f(x;)} converges to
f(x). [Note: It is easy to see that this theorem also holds for first countable
spaces X and Y.]
Proof. Suppose that f is continuous at xz, and let V be any open set
containing f(z). Since f is continuous, there is a neighborhood U of x such
that f(U) C V. The convergence of {z;} implies that there is some integer
N such that z; € U whenever i > N. But then for each i > N we have that
f (zi) € V and hence {f(x;)} converges to f(z).
To prove the converse, suppose f is not continuous at « € X. Then
there must be an « > O such that no neighborhood of x is mapped by f
into S.(f(x)). For each positive integer n select a point tp € Sj/n(z) for
which f(rn) ¢ S-(f(x)). Clearly, the sequence {z,} converges to z, but the
corresponding sequence {f(xrp)} fails to converge to f(z). O
The preceding theorem gives some rigor to the concept of continuity
as often taught in elementary calculus: a function is continuous at a point
xz € €! if and only if whenever points in €! approach z, then their images
under f approach f(z). Note that half of the theorem holds for arbitrary
topological spaces. The other half does not hold in general. The concept
of net, defined in Chapter 11, was devised in order to obtain an analogous
theorem that holds for arbitrary topological spaces.
(1.F.10) Definition. Suppose (X,d) is a metric space and A, B are

nonempty subsets of X. The distance between the sets A and B, d(A, B), is
defined to be glb{d(a,b) : a€ A and be B}.
In the two exercises that conclude this section, we see that both d and
a distance function closely related to d are continuous.
(1.F.11) Exercise. Let (X,d) be a metric space. Show that the metric
function d: X x X — [0,00) is continuous.
(1.F.12) Exercise. Suppose (X,d) is a metric space, and A C X. Show
that the function V : X — [0,00), defined by U(x) = d({x}, A), is continu-
ous.
G. HOMEOMORPHISMS
Let f : X — Y be a continuous function from one topological space onto
another. In spite of the continuity of f, the space X may differ considerably
from Y: a flagrant example of this occurs when X is arbitrary and Y consists
of a single point. Not only may Y be a good deal simpler than X, but,
surprisingly, as we shall see later, the reverse situation can happen as well.
In Chapter 9, a continuous function is constructed that maps the unit interval
I = [0,1] onto the unit square J x I. In fact, maps can be found that carry
I onto cubes, as well as onto a host of far more exotic spaces.
Homeomorphisms Sil
In these cases, the continuous function in question will not be 1-1. Nev-
ertheless, even with this further imposition, the spaces X and Y may still
have little similarity. To see this, suppose X is an infinite set with the discrete
topology and Y is any space with the same cardinality as X. Then every
1-1, onto map from X to Y is continuous, even though X has uncountably
many open sets and Y could have as few as two.
Suppose now, however, that f : X — Y is a continuous bijection such
that f—! (which is a function from Y to X since f is a bijection) is also
continuous. The reader should verify that not only is there a 1-1 correspon-
dence between the points of X and Y, but also between the open sets of the
respective topologies. Thus, the topological structure of X has an almost
exact counterpart in Y. Bijective, continuous mappings with continuous in-
verses are called homeomorphisms, and the corresponding spaces are said to
be homeomorphic or topologically equivalent. An embedding of X into Y isa
continuous 1-1 map f : X > Y such that f-! : f(X) > X is continuous. At
least a germinal feel for topology should arise from the following discussion
of homeomorphisms.
(1.G.1) Definition. A function f : X — Y is open (respectively, closed) if
and only if f maps open (respectively, closed) sets onto open (respectively,
closed) sets.
(1.G.2) Exercise. Show that homeomorphisms are both open and closed.
(1.G.3) Exercise. Show that a continuous open (or closed) bijection is a
homeomorphism.
(1.G.4) Exercise. (a) Show that if X; x--- x Xn has the product topology,
then the projection maps, p; are open, but need not be closed. (b) Show
that i: [0,1] + €? defined by i(x) = (x,0) is an embedding.
(1.G.5) Exercise. Show that two metrics d and d for a set X are equivalent
if and only if the identity map id: (X,d) — (X,d) is a homeomorphism.
For a topologist, homeomorphic spaces are essentially identical. Thus
little distinction (if any) is made between the following pair of spaces (con-
sidered as subspaces of €7), since one can be mapped homeomorphically onto
the other.
Similarly, a topologist is resigned to the fact that no distinction is to be

made between the following trio of spaces.
In a like vein, since a doughnut can be transformed into a coffee cup via
a homeomorphism (see figure), a topologist has classically been defined as
an otherwise enlightened individual who fails to note any difference between
these two objects.
oe ee
A word of caution, however. At this stage it might appear that two

spaces X and Y are homeomorphic if, intuitively, X can be molded into
Y without either X or Y being torn apart. While it is true that a home-
omorphism will result whenever one space is stretched, bent, or otherwise
deformed into another, this does not constitute the only manner in which
homeomorphisms may arise. For instance, it should be clear that the two
subspaces of €3 shown in the following figure are homeomorphic even though
no amount of tugging (in €°) will transform one of them into the other.
Thus, in order for two spaces to be homeomorphic, it suffices for the

homeomorphism to jump immediately from one to another without any in-
termediate steps.
Homeomorphisms 33
(1.G.6) Exercise. Show that (—1,1) and €! are homeomorphic via the
map h : €} — (1,1) defined by h(x) = x/(1 + |z)).
The problem of establishing the existence or nonexistence of a home-
omorphism between two topological spaces is by no means a trivial one.
Readers might enjoy discovering this for themselves by attempting to show
that €1 and €? are not homeomorphic. In order to attack this problem
in general, we shall investigate in the ensuing chapters certain topological
invariants, i.e., properties of topological spaces that are preserved by home-
omorphisms. If a space X possesses a certain property that is topologically
invariant, and the space Y does not, then clearly X and Y cannot be home-
omorphic. A couple of obvious invariants (which are not particularly useful)
are the cardinality of the sets in question and the cardinality of their re-
spective topologies. Note that neither of those invariants helps in deciding
whether or not €! and €? are homeomorphic.
Metrizability is a prime example of a topological invariant.
(1.G.7) Theorem. Suppose (X,U/) is homeomorphic to (Y, VY) and (X,U)

is metrizable. Then (Y, V) is metrizable.
Proof. Let d be a metric for X which induces the topology U, and let
h: X -+Y bea homeomorphism. Define d: Y x Y — (0,00) by d(yi, y2) =
d(h-1(y,), h~*(y2)). Since d is clearly a metric, it remains to show that the
topology W generated by d coincides with V. Suppose that y € V € V. Then
h-!(y) € h-1(V) €U, and hence there is an € > 0 such that S4¢(h~1(y)) C
h-1(V). Note that if d(z,y) < ¢, then we have d(h71(z),h7!(y)) < € and
consequently h~'(z) € h7*(V), soz € V. Therefore, S4(y) CV and V CW.
Now consider S4(y). If z € Sd(y), then we have h~'(z) € S¢(h7"(y)),
and therefore h(S¢(h—!(y)) C S4(y). Since S¢(h—(y)) € U, it follows that
h(S¢(h-!(y)) € V, and hence WCY. O
(1.G.8) Exercise. Show that the properties T2, first countability, and
second countability are topological invariants. Use the topological invariance
of T> to show that €!, and R! with the finite complement topology, are not
homeomorphic.
(1.G.9) Definition. A subset D of a space X is dense in X if and only if

every nonempty open set in X intersects D. A space X is separable if and
only if X contains a countable dense subset.
(1.G.10) Exercise. (a) Show that D is a dense subset of a topological

space X if and only if X = D. (b) Show that separability is a topological
invariant. (c) Show that €” is a separable metric space.
(1.G.11) Theorem. If (X,d) is a separable metric space, then X is second
countable.
Proof. Let x1, 22,... be a countable dense subset of X. Then it is not

hard to verify that the collection {$1/,(x;) : n € Z*, i € Z*} is a countable
base for the topology generated by d. O
The following theorem has numerous applications.
(1.G.12) Theorem. Suppose f,g : X — Y are continuous and Y is a T>

space. If f(x) = g(x) for each x in a dense subset D of X, then f = g.
Proof. Suppose for some z € X, we have f(z) # g(z). Since Y is To,
there are disjoint open subsets U and V of Y that contain f(z) and g(z),
respectively. However, this is impossible, since W = f-!(U)Ng~!(V) would
be a nonempty open set that would lie in the complement of the dense set
D. Therefore, f=g. O
We conclude this chapter with the definition of a class of spaces whose

properties will serve as a focus for much of our attention during the remainder
of the text.
(1.G.13) Definition. A separable metric space X is an n-manifold if and

only if each point of X is contained in a neighborhood that is homeomorphic
tO;E
Since €” is homeomorphic to each e-ball in €", the last part of the

definition can be rephrased as: each point of X has a neighborhood that is
homeomorphic to an open Euclidean e-ball in €". (See problems 1 and 2
of section G, or problem 1 and the observation that translations and scalar
multiplications by c > 0 give homeomorphisms.)
In general, n-manifolds are strongly geometric and have considerable
visual appeal; nevertheless, they may become exceedingly complex. Observe
too that a point with little wanderlust living in an exotic n-manifold, might
well consider its domicile to be nothing more exciting than €”.
(1.G.14) Examples of 1-manifolds.

Lie:
2. For each n € Z, let An = {(z,y) € E? : y= n}. Then M =
U{An : n € Z} is a 1-manifold.
3. The 1-sphere, St = {(z,y) € E? : x? + y? = 1}, considered as a
subspace of E?.
4. Any open subset of €?.
(1.G.15) Examples of 2-manifolds.
LO,
2. The 2-sphere, S? = {(z,y,z) € E* : a? +474 27 = 1}.
Homeomorphisms 35
3. A torus (the outer surface of an innertube, or alternatively S! x S oe
4. A Moebius strip (without the edge).
5. Any open subset of €?.

(1.G.16) Examples of 3-Manifolds.
Lae.
2. The 3-sphere, S* = {(2,y,z,w) € E* : 2? +y? +27 +w? = 1}.
3. The inside of a torus (i.e., the air inside an innertube).
4. The inside of a can (i.e., the beer inside a can).
5. Any open subset of €°.
(1.G.17) Exercise. Make a guess as to which of the following subspaces of
E? are manifolds (0 = (0,0), and S,(0) = {y € E? : d(y,0) < r}):
(a) 51 (0);
(b) 51 (0);
(c) S2(0)
wo
(d) S2(0) \ $1(0);

(e) {(t,y) : -l1<2<1, y=O0}U{(z,y) : 2 =0, O<y
< I}.
(1.G.18) Exercise. Let X be the union of the following rays in R?:
AS {oiy<s >0}, 1Beate) Oy andiG iG, 0)= <0}.
Points x in X other that (0,—1) and (0,1) are given neighborhood bases
consisting of (small) open intervals centered at z. Base neighborhoods of
(0,1) are of the form {(z,y) : 6 < 2 < Oandy =O}U{(ey) 7 0< 2 <
a and y = 1}, and base neighborhoods of (0,—1) are of the form {(z,y) :
b<x<Oandy=O0}U{(z,y) : 0<2<aand y= -—1}. Show that X with
the resulting topology is separable, 7), first countable, and locally Euclidean
(ie., show that each point has a neighborhood homeomorphic to €'), but
that X is not T> and, hence, is not a 1-manifold.
PROBLEMS
Section A
1. (a) Suppose {U, : a € A} is a collection of topologies for a set X.

Show that (\{Ua : a € A} is a topology for X.
(b) Is the union of topologies for a set X necessarily a topology?
(c) Suppose {U, : a € A} is a collection of topologies for a set X.
Establish the existence of a unique largest topology that is smaller
than each U, and a unique smallest topology that is larger than
each U,.
2. Find all possible topologies for the set X = {a,b,c}.
3. True or false? If (X,U) is a topological space and C is a family of open
sets in X, then (){C' : C € C} is open.
4. For each subset A of a nonempty set X, let F(A) be the topology on X
whose open sets are 9, X, and all subsets of X containing A. Assume
that X has at least two elements.
(a) Show that A C B if and only if F(B) C F(A).
(b) Suppose that A;,A2,...,A, are subsets of X and the F is
a topology for X such that F(A;) C F for each i. Show that
FAS.
5. Find four equivalent bases for the usual topology on R?.
Problems 37
6. Suppose U; and U2 are topologies for a set X and id: (X,U;) > (X,U2)
is the identity map. Show that id is continuous if and only if Uy Cc U4.
7. Suppose R! has the usual topology. Show that a function f:R'3=R})
is continuous if and only if f~*(—oo,a) and f~1(a, 00) are open for each
aé Rl.
8. Show that a subset A of R’ is open in the usual topology if and only if A
can be written as the countable union of pairwise disjoint open intervals
(intervals of the form (—o0o,a), (6,00), (—00, 00) are allowed).
9. Suppose (X,U/) and (Y, V) are topological spaces. A function f : X 3 Y
is open if and only if f(U) € V whenever U € U. Find an open function
that is not continuous.
10. Find an example of bases B and B for topological spaces (X,U) and
(Y, V), and a continuous function f : X > Y such that there is a basis
element B € B for which f~1(B) ¢ B.
Section B
1. Let (X,U) be a topological space and B Cc U. Show that B is a basis

for U if and only if for each r € X,B, = {BE B:2€ Bhisa
neighborhood basis for / at z.
2. Given: X a first countable space, Y an arbitrary space, and f: X > Y
continuous and onto. Is Y necessarily first countable?
3. Determine which subsets A of €! have the property that both A and
€! \ A are open.
4. Suppose UY and U are topologies for a set X such that (X,U) is Tz and
U CU. Show that (X,U) is To.
5. Suppose (X,U/) is second countable. Show that any basis for (X,U)
contains a countable subcollection that is a basis for UW.
6. Show that (X,U) is a topological space, and determine if it is T», first
countable, or second countable where:
(a) X is an uncountable set and U/ is the countable complement topology;
(b) X is an uncountable set, p is a particular point in X, and UY con-
sists of @ together with those subsets A of X such that either X \ A is
countable or p ¢ A;
(c) X = [1,1], and U is the topology generated by a basis consisting
of sets of the form [—1,), (a, 1], and (a,6), where a < 0 and 0 < 6.
MN Let X = {(2,y) : O0< 2 <1, 0 < y < 1} and define an order on X
by declaring (a,b) < (u,v) if and only if a < u, or a = u and b < v.
This ordering is called the lexicographic ordering. Show that this is a
linear ordering and determine if X (with the order topology) is T2, first
countable, or second countable.
8. For each x € Z and eachn EN, let B® = {y : y=rn+a, r € Z}.
Show that B= {B? : c € Z,n € N} is a basis for a topology on Z.
9. Suppose X is first countable (respectively, second countable), Y is a
topological space, and f : X — Y is continuous, open (i.e., f takes open
sets to open sets), and onto. Show Y is first countable (respectively,
second countable).
10. True or false? Suppose X is second countable. Then every nest of
distinct open sets in X is countable. |
Section C
1. Suppose (X,d) is a metric space and p is a point of X. Show that for

each r > 0, {z € X : d(x,p) > r} is open.
2. Give an e€ —6 definition of continuity for a function between two metric
spaces that is equivalent to (1.A.3).
3. Suppose (X,d) is a metric space and d: X x X > [0, co) is defined
by d(z,y) = d(z,y)/(1 + d(z,y)). Show that d is a metric for X and
that d and d are equivalent. [Hint: If a > b > 0 and m >n > 0, then
a/(1+a) > 6/(1+b) and (b+ m)/(a+n) > b/a.]
4. Suppose X is a set consisting of a finite number of points. Describe all
possible metrizable topological structures for X.
5. The post office metric. Let X = R? and d be the usual metric. Denote
(0,0) by 0. Define d: X < [0, co) by d(p,q) = d(0,p) + d(0,q) for
p,q€ X and p#q, and d(p,p) = 0 for allpe X.
(a) Show that d is a metric.
(b) Show that all points other than 0 are open.
(c) Describe the basis neighborhoods whose center is at 0.
6. If X is aset, a function d: X x X — [0,00) is called a pseudo-metric if
and only if
(a): «d(nme); II= Orfor allie X,
(it) d(x,y) = d(y, 2) for all (2, y) € X x X, and
(it) d(z,z) < d(z,y) + d(y, z)for all x,y,z € X.
Suppose’ X = {7 Pf ie SS and fe ro on [0,1]}. Define
d:X x X + (0,00) by d(f,g) = So \f (2) (x)|dz. Show that d is a
pseudo-metric but not a metric. Is there a een analogous to (1.C.3)
for pseudo-metric spaces?
Problems 39
te Let X = R?, and for points z = (21,22) and y = (y;, y2) define
1/2 if z] = yy and xq F yo or if x1 A y, and 22 = Op

d(z,y)=41 if 2} Ay, and rq F y2
0 otherwise
Show that d is a metric and that the sets shown in the figure possess
different “area” if d is used to measure the length of sides.
ee
Describe the topology induced by this metric.
Suppose V is a vector space over R!. A function ¢ : V — [0,00) is
called a norm for V if and only if for all x,y € V we have
(i) (x) = 0 if and only if c = 0,
(ii) o(a +) < d(x) + 419),
(iti) d(ax) = |al(xz) where a is a real number.
The pair (V,¢) is called a normed vector space and ¢(z) is frequently
denoted by ||z||.
Suppose (V,@) is a normed vector space and d : V x V — (0,00) is
defined by d(z,y) = ||z — y||. Show that d is a metric for V.
French railway metric. Let X = R?. Suppose z and y are points in X.
If x and y are on a line passing through the origin, define their distance
to be the usual Euclidean distance. If x and y do not lie on such a line,
then define their distance as in the post office metric (problem 5). Show
that the French railway metric is a metric and (informally) describe the
induced topology.
Section D
i. Determine the relative topology that is induced by €! on the integers.

2. Suppose B is a basis for a topology on a set X and A C X. Show that
{BN A: B€ B} is a basis for the relative topology on A.
Suppose B is a basis for a topology on X and B is a basis for a topology
on Y. Show that {B x B : B € B, B € B} is a basis for the product
topology on X x Y.
Show that if X and Y are T2 spaces, then so is X x Y.
Suppose (Xj, di), (X2,d2),..., (Xn, dn) are metric spaces, and suppose
Y =X, x X_ X--: x Xn. Define
p: Y x Y — (0, 00) by setting
(a1, Base we erp Viste: oa BS a) i
Show that p is a metric and the topology induced by p coincides with

the product topology on Y.
Consider the following diagram where the p,’s and q;’s are projection
maps, and the f;’s are continuous. Show that a unique continuous func-
tion ¢: X,; x X_ > Y; x Y2 may be defined which makes the following
diagram commutative.
iis
Xy = Xj x X92 is X29
Yi q1 Vi x Y5 q2 Y>
Suppose X,, X2,..., Xn and Z are topological spaces and Y = |]j_, Xi.
Let g: Z > Y be a function. Show that g is continuous if and only if
pig is continuous for each projection map p;.
(a) Is the product of a finite number of first countable spaces first

countable?
(b) Is the product of a finite number of second countable spaces second
countable?
(a) Is the free union of a family of first countable spaces first countable?
(b) Is the free union of a family of second countable spaces second
countable?
10. Show that the free union of metrizable spaces is metrizable.
dil. Suppose X and Y are topological spaces and V C X x Y is open. Show
that V[z] = {y : (z,y) € V} and V[y] = {x : (z,y) € V} are open in
Y and X respectively. Is the converse true: if AC X x Y and A[z] and
Aly] are open for each z € X and y € Y, must A be open in X x Y?
12. Are there any open sets in the product topology of X x Y that are not
of the form U x V where U is open in X and V is open in Y?
Problems 41
Section E
Show that a subset U of a space X is open if and only if ANU = @

implies that ANU = @ for each AC X.
Suppose X is a space and AC BC X. Show that AB = AX NB.
Find a topological space that has neither the discrete nor indiscrete
topology in which subsets are open if and only if they are closed.
Suppose f and g are continuous functions from a space X into a T>
space Y. Show that {x : f(x) = g(zx)} is closed.
show (A(te)°” AmB; and (A°)"= A.
Suppose X is a finite T; space. Show that X must have the discrete
topology.
Show that a set U C X is open if and only if FrU Cc (X \U).
Show that a set A is open and closed if and only if Fr A = 9.
Suppose X is a topological space and A C B C X. Find an example
where int A in B is not the same as int A in X.
10. (a) Show that (AU B)! = A’ UB’. (b) Show A = AUFY(A).
11. Suppose that (X,U/) is a topological space, A C X, and z is an accumu-
lation point of A. Discuss whether or not z must be an accumulation
point of A with respect to topologies UW’ and U", where U' CUCU".
12. Let X = {a,b,c} and let U = {0,X, {a}, {b}, {a, b}} be a topology for
X. Find the derived set for each subset of X.
13. Let X be an infinite set with the finite complement topology. Find the
derived set, interior, exterior, closure, and frontier for the subsets of X.
14. A space X is called a door space if and only if each subset of X is open
or closed.
(a) Find a door space that does not have either the discrete or the
indiscrete topology.
(b) Suppose X is a T2 door space. Show that X has at most one
accumulation point and that nonaccumulation points are open.
15. A space X is called a semi-door space if and only if for each A C X,
there is an open set U such that UC ACU or UC (X\ A) CU.
(a) Find a semi-door space that is not a door space.
(b) Show that a T> semi-door space is a door space. [Hint: Show that
there is at most one point that is not open.]
16. Find a metric space which illustrates that S.() need not be the same
as {ye X : d(z,y) < €}.
Be: (a) Find a topological space that is not 7}.
(b) Find a T; space that is not T».

(c) Determine which of the examples given in Section B are T; spaces.
18. Show that derived sets are closed in 7) spaces.
19. Let U be the finite complement topology on an infinite set X. Show that
if CU and (X, U) is Ty, then U = U. Furthermore, show that if (X, V)
is T; then U Cc V. (Hence, U is the smallest T; topology for X.) If S is
the family of all T; topologies for X, show that U =(){7T : T € S}.
20. Show that every uncountable subset of a second countable space has an
accumulation point.
a1: Show there are closed subsets of [0,1] (usual topology) which are un-
countable and consist only of irrational points.
22. Suppose X is aset andc: P(X) > P(X) is a function with the following
properties:
(1) c(0) = 9;
(it) AC c(A) for each A € P(X);
(iit) e(c(A)) = c(A) for each A € P(X);
(iv) c(AU B) =c(A) Uc(B) for each A and B in P(X).
Let U = {X\c(A) : A€ P(X)}. Show that U is a topology for X with
the property that for each A € P(X), A =c(A). A function with the
properties of c is called a closure operator.
23. Let X be a set and I: P(X) + P(X) have the properties that:
(i) I(X) =;
(it) I(I(A)) =I(A) for each A € P(X);
(iit) I(A) C A for each A € P(X);
(iv) I(AN B) =1(A)NI(B) for each A and B in P(X).
Let U = {I(A) : A € P(X)}. Show that UW is a topology for X, and
that for each A € P(X), J(A) = int A. A function with the properties
of I is called an interior operator.
24. Formulate and prove results similar to those in problems 22 and 23 above
for
(a) a frontier operator, and
(b) a derivation operator.
25. Prove that in a T; space, a point x is an accumulation point of a set A
if and only if every neighborhood of z contains infinitely many points
of A.
26. Suppose A and B are disjoint closed subsets of a metric space. Show that
there are disjoint open sets U and V containing A and B respectively.
7 Suppose A is an open subset of a topological space X. Prove or disprove
that int(A) = A.
Problems 43
28. Let A and B be subsets of a topological space X. Show that Fr (AUB) c

Fr (A) UFr (B). Show that the analogous statement for countable unions
is false.
Section F
Use (1.A.3) to give an alternate proof of (1.F.1).

Suppose X is aset and A C X. The characteristic function f 4 associated
with A is a map from X into {0,1} which assumes the value 0 for points
not in A and is 1 for points of A. If X is a topological space and A C X,
show that f4 is continuous if and only if A is open and closed (assume
that {0,1} has the discrete topology).
Prove or disprove: A function f : X — Y is continuous if and only if
f(A’) C [f(A)]’ for each AC X.
Show that a function f : X — Y is continuous if and only if we have
f~'(int B) Cc int (f~1(B)) for each BCY.
Prove that the following are equivalent:
(a) f : X + Y is continuous;
(b) f-1(B) c f71(B) for each B CY;
(c) f(A’) C f(A) for each A C X;
(d) Fr{f-1(B)] c f-!(FrB) for each BCY.
Suppose f,g : X — €! are continuous. Show that h: X — €' defined
by h(x) = max{ f(x), g9(x)} is continuous.
A map f : X > E! is upper semicontinuous if and only if for each b € €},
{a : f(x) < b} is open.
(a) Suppose {fa}aea is a family of continuous functions each mapping
the space X into (0,1) c €!. Show that h: X — E&} defined by
h(x) = inf{fa(x) : a € A} is upper semicontinuous.
(b) Suppose f : X — &! has the property that for each rational r,
{x : f(x) <r} is open. Show that f is upper semicontinuous.
Let f : X — Y and AC X. Show that it is possible for f|4 to be
continuous even though f is not continuous at any point of A.
Construct a function f : €! x €! > €! such that f is continuous in each
variable separately, but is not continuous on €! x &?.
10. Suppose (X,1/) and (Y, V) are topological spaces. A function f : X + Y
is weakly continuous at a point x € X if and only if whenever f(x) €
V €Y, there is aU €U such that « € U and f(U) C V. The function
f is weakly continuous if and only if it is weakly continuous at each
a4 1. The Basic Constructs
a € X. Show that f is weakly continuous if and only if for each V € Y,

foU(V) Cc int (f-*(V).
EE Suppose (X,U/) and (Y, V) are topological spaces. A function f : X > Y
is feebly continuous if and only if whenever V € V, f~'(FrV) is closed
in X. Show that weak continuity and feeble continuity are unrelated,
but that a function f is continuous if and only if it is both feebly and
weakly continuous.
12. Suppose (X,/) and (Y, V) are topological spaces. A function f : X + Y
is strongly continuous if and only if f(A) C f(A) for each A C X. Show
that f is strongly continuous if and only if f~'(B) is closed for each
BCay..
13. (a) Show that theorem (1.F.6) may fail to hold if A and B are not both
closed in X.
(b) Prove an analogue to (1.F.6) where A and B are both open in X.
(c) State and prove an analogue to (1.F.6) where {O. : a € A} isa
family of open subsets of X.
(d) State and prove an analogue to (1.F.6) where {Cy : a € A} isa
family of pairwise disjoint closed subsets of X.
14. (a) Find an example of a topological space X and a sequence {z;} in
X such that {z;} has more than one limit point in X.
(b) Show that if X is Hausdorff then a sequence can have at most one
limit point.
(c) Discuss the convergence of sequences in an uncountable infinite set
X with the countable complement topology.
15. Find a counterexample to the following proposition: Suppose X and Y
are topological spaces and f : X — Y is a function with the property:
whenever {z,,} is a sequence in X and z is a point in X to which
{x,,} converges, the sequence {f(z,)} converges to f(x). Then f is
continuous.
Section G
Show that €” is homeomorphic with {x € E” : d(z,0) < 1}.

A subset A of €” is convez if and only if the line segment {tz+(1—t)y :
0 < t < 1} is contained in A whenever x,y € A. Show that bounded
convex open subsets of €” are homeomorphic. Is the modifier “bounded”
necessary?
Suppose X and Y are topological spaces and f : X — Y is bijective.
Show that f is a homeomorphism if and only if f(A) = f(A) for each
Ane
Problems 45
4, Fill in the blank: Let id : (X,U) > (X,U'). Then id is open if and only
if U Pee
Suppose X and X are homeomorphic and Y and Y are homeomorphic.
Show that X x Y and X x Y are homeomorphic.
Find a mapping: (a) that is open but not closed; (b) that is closed but
not open; (c) that is continuous but neither open nor closed.
Let A be the set of irrationals in €' with the relative topology. Is A
separable?
Suppose A is a dense subset of X and G is a dense open subset of X.
Show that ANG is dense in X.
Suppose A is dense in X and B is dense in Y. Show that Ax B is dense
in An ey’,
10. A map f : X > Y isa local homeomorphism if and only if for each point
x € X, there is an open set U containing x that is mapped homeomorphi-
cally by f onto an open subset of Y. Show that local homeomorphisms
are continuous and open.
11. Prove the following converse of (1.G.12). Suppose Y has the property
that for any space X, any dense subset D of X, and any two continuous
functions f,g: X — Y which agree on D, then f = g. Then Y is T».
12. A metric space X is totally bounded if and only if for each « > 0, there
dEC pOints 2 4i05,.05ee ouch that HX =A) SAe,)) tednsy 1 2509.57:
Show that a totally bounded metric space is separable.
13. Find an example of a nonseparable metric space that does not have the
discrete topology.
14. Suppose (X,d) is a separable metric space and that A C X. Show that
(A, d|,) is separable. In general, is it true that a subspace of a separable
space is separable?
15. Suppose R! has the half-open interval topology U, and A = (0,1) is
given the relative topology (with respect to U). Show that R! and A
are homeomorphic.
16. True or False: If U is an open subset of X and h: U > X is an
embedding, then h(U) is open in X.
Li. Assume the statement in problem 16 is true whenever X = €”, and
prove that it is true for n-manifolds.
18. A subset A of a space X is semiopen if and only if there is an open set
U in X such that U CAC U.
(a) Show that A is semiopen if and only if A = AND for each dense
subset D of X.
(b) Suppose D C X. Show: D is dense if and only if A= AND for

each semiopen set A in X.
19: (a) A topological space is 0-dimensional if and only if whenever x € V,
and V is open, there is an open set U with empty frontier such that
x € U CV. Show that the rationals and the irrationals (with the
relative topology) are 0-dimensional sets.
(b) The higher dimensions are defined inductively. A space X is said
to have dimension < n, if and only if for each point z € X and
for each open set V containing z there is an open set U such that
x €U CV, and dimension(FrU) < n—1; X has dimension n if
and only if dimension(X) < n, but it is false that dimension(X) is
<n-—1. Show that €! has dimension 1.
(c) Show that dimension is a topological invariant.
20. Are restrictions of open (closed) mappings open (closed)? Is the re-
striction of an open (closed) mapping to an open (closed) subset open
(closed)?
21. Find a set X and distinct topologies UU, and U2 for X such that there is
a homeomorphism h: (X,U) > (X,U2).
22. Suppose (X,U/) and (Y,V) are topological spaces. Show that a map
f : (X,U) - (Y,V) is open if and only if f(int A) C int (f(A)) for each
AC:
23. Find a proper closed subset of €! x €! which projects to a proper open
subset of €! under both projection maps.
24. Suppose X and Y are topological spaces and y € Y. Show that the
subspace X x {y} of X x Y is homeomorphic with X.
20. How many different spaces are represented by the (san serif capitals)
alphabet (as subspaces of the plane)?
Chapter 2
CONNECTEDNESS AND
COMPACTNESS
The two principal topological properties studied in this chapter occupy a

central position in both topology and analysis. The concepts of connected-
ness and compactness may be viewed as generalizations of basic properties of
intervals. Connectedness represents an extension of the idea that an interval
is all in “one piece,” while compactness may be construed as a generalization
of the fact that a closed interval [a, b] is both closed (as a subset of €!) and
bounded.
A. CONNECTEDNESS: GENERAL RESULTS

Let X = (—1,2) U [3,7) U {10} have the relative topology inherited from
€1. Intuitively, X is not in one piece. Note that we can write X as the
union of disjoint (relatively) open sets. For example, let U = (—1,2) and
V = (3,7) U {10}. On the other hand it does not seem possible to split
Y = (2, 20] into a union of disjoint Y-open sets. These observations lead to
the following definition.
(2.A.1) Definition. A pair (U,V) of nonempty open subsets of a space X

is a separation of X if and only if X =UUV and UNV = 9.
(2.A.2) Remark. Suppose X is a topological space, A and B are disjoint
nonempty closed subsets of X¥, and X = AUB. Then A and B effect
a separation of X, since their complements, B and A, are open. If X is
separated by subsets A and B, then A and B are both open and closed.
(2.A.3) Definition. A topological space X is connected if and only if no

separation of X exists. If a separation of X does exist, then X is disconnected.
A subset A of X is connected if and only if A, with the relative topology, is
connected.
47
48 2. Connectedness and Compactness
(2.A.4) Exercise. Show that a space X is connected if and only if X does

not contain a proper nonempty subset that is both open and closed.
(2.A.5) Exercise. Suppose A is a subspace of a space X. Show that A is

disconnected if and only if there are closed (open) subsets C; and C2 of X
such that A C C; UC2, ANC, NC, = 9, and both ANC; and AN C4 are
nonempty.
(2.4.6) Exercise. Show that if (U,V) is a separation of a space X and

C Cc X is connected, then either C CU or CCV.
(2.A.7) Examples.
Connected Spaces. (The justification for examples 1-4 follow from later
results.)
1. Intervals in €}.
2. Disks (spaces homeomorphic with {(z,y) € E? : 2? +y? < 1}).
yar
4. A sine curve.
5 . Any set with the indiscrete topology.
Disconnected Spaces.
1. (0,1) U (3, 4) (with the relative topology from &').
2. (0,1) U (1, 4] (with the relative topology from €').
3. The rational numbers (with the relative topology).
4. Any set (containing two or more points) with the discrete topology.
Although it seems quite reasonable from a visual standpoint, the “cri-
terion” that a space is connected if it consists of solely “one piece” is often
impossible to apply in practice. For instance, while it seems that €? is con-
nected, what about €? \ {(z,y) : x,y € Q}? Is this subspace in “one piece”
in spite of all the holes? In €! no such complications arise, as we see from
the following theorem. But first we recall that an interval is a subset Z of R
with the property: ifz,y,z€ R,z,ye€TZ,andr<z<y, then ze TZ.
(2.A.8) Theorem. A nonempty subset B of €' is connected if and only if

B is a point or an interval.
Proof. We use the contrapositive to show that B connected implies B
is a point or interval. For that purpose, suppose then B is a subset of €!
that is neither an interval nor a point. Then there are points a,b,c € €!
such that a < c < b, where a,b € Bandc ¢ B. Let U = (-o,c)NB
and V = (c,oo) N B. Since U and V are clearly nonempty disjoint B-open
subsets whose union is B, they form a separation of B, and consequently B
is not connected.
Connectedness: General Results 49
To prove the converse we use contradiction. Suppose B is an interval

that is not connected. (We will eventually show that B cannot be an inter-
val.) It follows from (2.A.5) that there are closed subsets C and D of €! such
that CONDON B=0, BC CUD, and both CN B and DNB are nonempty.
Let c€ CNB and de DNB, and assume (relabel if necessary) that c < d.
Let s = sup{x € B: x € Cand < zc < d}. Note that s € C since C is
closed, hence s # d. Now let t = inf{fy € B : y€ Dands < y < d}. We
consider two possibilities: s = ¢ and s < t. If s = t, then s € CN D (why?)
and consequently s is not in B. If s < t, then BN(s,t) = 0. In either case, it
easily follows that B is not an interval. We leave it for the reader to supply
the prove that a singleton is connected. O
(2.A.9) Exercise. Fill in the details of the foregoing proof.
(2.A.10) Theorem. Suppose {A, : a € A} is a collection of connected
subspaces of a topological space X, and for each a,B € A, AgN Ag # 0.
Then U{Aaq : a € A} is connected.
Proof. Let Y = U{Aa : a € A} and suppose Y is not connected. Then
there are disjoint, nonempty, Y-open sets U and V whose union is Y. Since
U # 0, there is an a € A such that Ag NU # 9, and similarly there is a
B € A such that AgNV # @. By (2.4.6), either Ag C U or Ag C V, and
since Ag NU # 9, it follows that A, C U. Similarly, we have that Ag C V.
However, this is impossible, since AgN Ag #%. O
(2.A.11) Exercises.
(a) Why is €' connected?
(b) Suppose that {Aq : a € A} is a collection of connected subspaces
of a space X, and that B C X is also connected. Show that if
BNA, # 9 for each a € A, then BU(U{Aq : a € A}) is connected.
The following theorem, quite useful in spite of its trivial proof, gives
a convenient characterization of connectedness. For the remainder of this
chapter we let S denote the set {0,1} with the discrete topology.
(2.A.12) Theorem. A space X is connected if and only if there is no
continuous function f : X — S which is onto.
Proof. If f : X -> S is onto, then the sets f~'(0) and f~'(1) form
a separation of X. On the other hand, if X is not connected there is a
separation (U,V) of X. Define f : X — S by f(x) = 0 if e € U and
f(x) =1ifaeéV. Then f is clearly a continuous map of X ontoS. O
(2.4.13) Theorem. If A is a connected subspace of a topological space X

and B is a subset of X such that A C B C A, then B is connected.
Proof. We apply (2.4.12). Suppose f : B > S is continuous. We must
show that f fails to be onto. Since f|,4 is continuous and A is connected,
we may assume (without loss of generality) that f(a) = 0 for each a € A.

Suppose that for some b € B, f(b) = 1. Then f~'(1) is a B-open set
that contains b but misses A. This, however, is impossible, since } is an
accumulation point of A. O
Connectedness is even better then a topological invariant; it is actually
preserved by continuous functions.
(2.A.14) Theorem. If f is a continuous function from a connected space

X onto a space Y, then Y is connected.
Proof. If Y is not connected, then by (2.4.12) there is a continuous map
g from Y onto S. Hence gf is a continuous function from X onto S, which
contradicts the fact that X is connected. O
We have three immediate corollaries, the last of which should be familiar
from elementary calculus or high school algebra.
(2.A.15) Corollary. Connectedness is a topological invariant. O
(2.A.16) Corollary. If X is a connected topological space and f : X > &}
is continuous, then f(X) is a point or an interval. O
(2.4.17) Corollary (The Intermediate Value Theorem). Suppose X
is a connected space and f : X —> €! is a continuous function. Suppose
further that a,b € f(X) and a < b. Then if z is any number such that
a<z<b, there is at least one point c € X such that f(c) = z.
Proof. Suppose no such point c exists. Then f(X) is neither a point
nor an interval, which contradicts (2.4.16). O
B. CONNECTEDNESS: SLIGHTLY DEEPER RESULTS

Corollary (2.4.17) proves to be an important tool in establishing some basic
theorems in calculus. However, rather than pursue these results, we give an
illustration of how to employ this corollary in a non-calculus setting.
An interesting problem, various aspects of which have kept many topol-

ogists and analysts employed, concerns the study of fixed point properties.
(2.B.1) Definition. A space X is said to have the fized point property if

and only if for each continuous map f : X — X, there is an z € X such
that f(z) = x. (Of course the point x may well be different for different
functions.)
Theorems involving the fixed point property are not only aesthetically
pleasing in themselves, but also have wide ranging applications to other areas
of mathematics. Our first fixed point result is a special case of the Brouwer
fixed point theorem, which states that the space obtained by taking the
Connectedness: Slightly Deeper Results 51
product of J = [0,1] with itself a finite number of times has the fixed point
property.
(2.B.2) Theorem. The interval J has the fixed point property.

Proof. Suppose f : J + I is continuous. We will show the existence of
a point c € J such that f(c) =c. If f(0) = 0 or f(1) = 1, we are finished.
Hence, assume 0 < f(0) and f(1) < 1. Define a continuous function g from
I into €' by setting g(x) = z — f(x). Note that g(0) = —f(0) < 0 and
g(1) = 1— f(1) > 0. From the Intermediate Value Theorem it follows that
there is a point c € J such that g(c) = 0. However, this implies c— f(c) = 0,
and hence we have f(c)=c. O
Suppose the topological spaces X,, X2,...,Xn, each have a common
topological property P. It is natural to inquire whether the product space
[][;_, X: also enjoys P. We have seen in the problems that this is sometimes
the case, e.g., if P is the topological property T2 or second countability. The
next theorem shows that connectedness is also preserved when constructing
finite products.
(2.B.3) Theorem. Suppose Xj, X2,...,X, are topological spaces. The

product space Y = [];_, X; is connected if and only if each X; is connected.
Proof. We leave the trivial proof that Y connected implies each X;
connected to the reader (Exercise (2.B.5)). For the converse, we prove the
theorem for n = 2, omitting the easy induction proof that yields the more
general result. If Y were not connected, then by (2.A.12) there would be
a continuous onto map f : Y > S. Let a = (a;,a2) and b = (b;,b2) be
points in Y with f(a) = 0 and f(b) = 1. Define functions f; : X1 — S
and fz : X2 — S by setting fi(z1) = f(z1,b2) for each a; € Xj, and
fo(z2) = f(ai,%2) for each tz € X2. It is easy to show f; and f2 are
continuous. Since X, and X92 are connected, we conclude f; and fz must be
constant maps. But this gives 1 = 0, since 1 = f(b, b2) = fi(bi) = fila) =
f (a1,b2) = fo(b2) = fo(a2) = f(ai1,a2) = 0. We trust the reader will see the
contradiction and conclude that Y must be connected. O
(2.B.4) Corollary. The space €" is connected. O
(2.B.5) Exercise. Complete the proof of Theorem (2.B.3).

(2.B.6) Theorem. If B is a subset of €” (n > 1) and B contsists of a
countable number of points, then €” \ B is connected.
Proof. Let c € €” \ B and select a line L, in €" \ B containing c. (Why
does such a line exist?) For each point z € €” \ B, there is a line Ly, in
E” \ B that contains z and intersects L,. Thus €” \ B may be considered as
the union of L, and the L,’s. Hence £,, \ B is connected by Qari) “oO
As mentioned previously, a central problem in topology is that of clas-

sifying spaces up to homeomorphism. This involves finding a method for
deciding whether each arbitrary two topological spaces are homeomorphic.
In this generality the problem can not be solved, as mathematical logicians
have shown (Markov, [1960]). However, if one reduces the problem to that of
trying to determine whether a particular pair of topological spaces is homeo-
morphic, a positive result is sometimes obtainable. As indicated in Chapter
1, a common procedure is to find a topological invariant that one, but not
both, of the spaces possesses. Theorem (2.4.14) shows that connectedness
is a topological (actually a continuous) invariant. The reader is asked to
show in the next exercise that the fixed point property is also a topological
invariant.
(2.B.7) Exercises.
(a) Show that the fixed point property is a topological invariant. That
is, show that if h : X —+ Y is a homeomorphism and X has the fixed point
property, then so does Y.
(b) Show that [0,1] and S' are not homeomorphic.
The topological invariant, connectedness, may be exploited to show that
E! is not homeomorphic with €?, even though both spaces are connected.
(2.B.8) Theorem. The space €! is not homeomorphic to €?.
Proof. Suppose a homeomorphism h : €? + &! exists. Let c € €? and
note that E? \ {c} is connected by (2.B.6). However h(E? \ {c}) = €1 — {h(c)}
is not connected, which contradicts (2.4.14). O
(2.B.9) Exercise. Show that (0,1) does not have the fixed point property,
and hence (0,1) is not homeomorphic with (0, 1].
(2.B.10) Exercise. Use the topological invariant, connectedness, to give
another proof that (0,1) and [0, 1] are not homeomorphic.
We conclude this section with a rather strange, but intuitive, result that
will see frequent use in succeeding chapters.
(2.B.11) Theorem. Suppose X is connected and A is a connected subset

of X. Suppose further that X \A = U UV, where (U,V) is a separation of
X \ A. Then AUU is connected.
Proof. Note first that A is nonempty, for otherwise (U,V) would give
a separation of X. Now suppose C; and C» are disjoint open and closed
subsets of AUU such that AUU = C) UC). Since A is connected, either
Ac C, of A C C2 (without loss of generality, assume that A C C,). Since
C, # 0 it suffices to show Cp = 9. From C,2N A = Q, it follows that
72 C U. Hence C2 would be open and closed in U (and AUU) . But U is
Path Connectedness 53
open and closed in X \ A. Therefore, by (1.D.4), Cy is open and closed in

(X \ A) U(AUU) = X. Since X is connected, it follows that Cz is empty.
Hence there is no separation possible for AUU. 0
C. PATH CONNECTEDNESS
In this section we introduce a different concept of connectedness—one that
will find repeated applications in later chapters. A layman’s notion of con-
nectedness would probably center on the idea that a space X is connected if
one can move from one point in X to another without ever leaving X. This
idea is conceptualized mathematically by the following two definitions.
(2.C.1) Definition. Suppose X is a topological space. A path in X is a

continuous function f : J > X. If f(0) =a and f(1) = 6, then f is said to
be a path from a to b; a is the initial point of f, and b is the end or terminal
point. The set {a, b} is called the set of endpoints.
The reader should note that a path is defined to be a function and not
the image of the function.
(2.C.2) Definition. A space X is path connected if and only if each pair of

points in X are the endpoints of a path in X, i.e., for each x,y € X there is
a continuous map f : J — X such that f(0) = 2 and f(1) =y.
It should be clear to the reader that €” is path connected for every
choice of n € Zt.
(2.C.3) Theorem. Suppose a,b, and ¢ are points in a space X and there
are paths from a to b, and from b to c. Then there is a path from a to c.
Proof. Let f : I + X andg:I—- X be paths such that f(0) = a,
f(1) = 5, g(0) = 6, and g(1) =c. Define hh: [- X by
f (2t) if0<t<$;
A(t) =
gat=1) fe sre.
That h is continuous follows from the map gluing theorem (1.F.6), as well
as (1.F.3) and (1.F.1). Clearly h(0)=aandh(1)=c. O
(2.C.4) Exercises.
(a) Show that if there is a path from z to y, then there is a path from
y to x.
(b) Suppose {Aq : a € A} is a collection of path connected subsets of a

space X such that (}{Aq : a € A} #0. Show that U{Aa : a € A} is path
connected.
(2.C.5) Exercise. Show that every path connected space is connected.
Although path connected spaces are connected, the next example shows
there are connected spaces that are not path connected.
(2.C.6) Example. Let Y = {(z,sin1/z) € €? : 0 < x < 1/7} and suppose

Z={(x,y) X = Y UZ with the
€ €? : t=O and —-1<y <1}. The space
relative topology inherited from €? is frequently referred to as the topologist’s
sine curve.
It follows immediately from (2.4.13) that X is connected. However, X is

not path connected. To see this, we show that no path can stretch from the
point (1/7,0) to the point (0,0). Suppose to the contrary that there is a
path f : I > X with endpoints f(0) = (1/7,0) and f(1) = (0,0). Since f(J)
is connected, every point on the sine curve (for 0 < x < 1/7) must be in the
range of f. Thus we may select a sequence of points in J, 71 < 22 < %3°-°--,
such that the sequence {z;} converges to 1 and f(z;) is 1 if 7is odd and is
—1ifzis even. This however is impossible by (1.F.9) and Problem (1.F.14b).
(2.C.7) Definition. A topological space X is locally path connected if and
only if each point « € X has a neighborhood base of path connected open
sets, i.e., whenever x € U where U is open, then there is a path connected
open subset V of X such that r EV CU.
(2.C.8) Theorem. If X is a connected, locally path connected topological

space, then X is path connected.
Proof. Let x € X, and A= {z€ X : there is a path from z to z}. We
show that A is nonempty, open and closed. Then, since X is connected, this
will imply A = X (2.A.4). It is clear that A is nonempty since x € A. To
show A is open, suppose z € A. Since A is locally path connected, there is
an open set V containing z such that every two points in V can be joined by
a path lying in V. Hence (2.C.3) implies that V C A, so A must be open.
To see that A is closed, we show that X \ A is open. Suppose z € X \ A and
let V be a neighborhood of z that is path connected. If VN A 4 0, then we
may join z to x by a path (2.C.3), which contradicts the fact that z € X \ A.
Hence VN A=9Q, and X\Aisopen. O
Components 55
(2.C.9) Corollary. Connected open subsets of €” are path connected. O

The concept of a path is perhaps less intuitive than it originally seemed.
For example, we have noted earlier that there are continuous functions map-
ping the unit interval onto the unit square. By definition, such functions
are paths even though they are probably not quite what the reader had in
mind when we described paths as a means of traveling from one point to
another in a topological space. A notion that is undoubtedly much closer to
the reader’s preconception of what a path should be is that of an arc.
(2.C.10) Definition. Suppose X is a topological space. An arc in X is an

embedding h: I + X. As with other paths we call h(0) the initial point and
h(1) the terminal point.
We shall eventually prove, (2.G.11), that for Hausdorff spaces an arc
is merely a path that is 1-1. Spaces in which each two distinct points may
be connected by an arc are called arc connected. Local arc connectedness is
defined in a manner analogous to local path connectedness.
(2.C.11) Exercise. Prove or disprove: If a space X is path connected, then

X is also arc connected.
(2.C.12) Exercise. What is the difficulty in replacing the word “path” in

(2.C.3) and (2.C.8) by the word “arc”? Explain how to get around these
difficulties.
(2.C.13) Exercise. Prove that connected n-manifolds are arc connected.
(2.C.14) Exercise. Let Xo = {a € E? : x = (2,0) for x € I}, and define
X; = {a € €? : @ lies on the line between 0 and (1,1/i), i € Z*+} for each
i € Z+. Show that X = Ujen Xi is path (arc) connected but not locally
path (arc) connected.
The reader should have no difficulty in finding examples of locally path
(arc) connected spaces that are not path (arc) connected. One such example
is given in (2.D.2).
D. COMPONENTS
We can view spaces that are not connected as consisting of a (possibly infi-
nite) number of connected pieces. This leads to the following notion.
(2.D.1) Definition. Suppose X is a topological space and A is a subspace
of X. Then A is a component of X if and only if
(i) A is connected, and
(ii) if B is a connected subspace of X containing A, then B = A.
In other words, components are maximal connected subspaces.
Note that, if X is connected, then the only component of X is X itself.

At the other extreme, if X is a discrete space then each point of X is a
component.
(2.D.2) Example. Let X be the subspace of the plane €? consisting of

11 vertical line segments, as shown. Then each vertical line segment is a
component of X.
(2.D.3) Exercise. Find the components of the rational numbers with the
relative topology. Do the same for the irrational numbers.
The basic properties of components are easily established.
(2.D.4) Theorem. Suppose X is a topological space.

(i) Each component of X is closed.
) Distinct components of X are disjoint.
(iit) Each point of X belongs to exactly one component of X.
) If X is homeomorphic to a topological space Y, then there is a 1-1
correspondence between the components of X and the components
of Y.
Proof. (1) This is immediate from (2.4.13). (ii) This is immediate from
(2.4.10). (iti) Let c € X. The union of all connected sets that contain zx is
a component of X. (iv) This follows from the fact that images of connected
sets under continuous maps are connected (2.A.14). O
Parts (17) and (277) of the above theorem can also be expressed as saying
that the components form a partition of X.
(2.D.5) Definition. A nontrivial space X is totally disconnected if and only

if each of its components consists of a point.
While it is clear that a space with the discrete topology is totally dis-
connected, there are other examples as well. By (2.D.3) the set of rationals
(or irrationals) with the relative topology is a totally disconnected space.
Locally Connected Spaces 57.
(2.D.6) Exercise. A connected space X (with more than two points) is

said to have an explosion point p (also called a dispersion point) if and only
if X \ {p} is totally disconnected. Find such a space and show that such a
space can have at most one explosion point. [Hint: Use (2:Bsi1)
(2.D.7) Definition. A subspace C of a topological space X is a path com-

ponent of X if and only if C is path connected and is not contained in a
larger path connected subspace.
(2.D.8) Exercise. Show that for open subsets of €”, path components and
components coincide. Is the modifier “open” necessary?
E. LOCALLY CONNECTED SPACES

We begin by considering a subspace of €?. Let An = {(z,1/n) :0<a2< 1}
for each n € Z*, and let Ap = {(z,0) : 0<2< 1}.
Ai
The components of X = Ao U (UP2., An) are precisely the sets Ao, Ai,....
We know from (2.D.4) that Ao must be closed. Since it is clear that Apo is
not open in X we see that components need not be open. (Another example
of this behavior was given in (2.D.3).) For certain types of spaces, how-
ever, components will be both open and closed. These are known as locally
connected spaces, and they include the n-manifolds. The formal definition
of such spaces is couched in somewhat different terms (emphasizing the lo-
cal nature of the concept), but the fact that the two notions are essentially
equivalent is readily established in (2.E.2).
(2.E.1) Definition. A space X is locally connected if and only if for each
point z € X and each open set U containing 2, there is a connected open
set V such that s € V CU.
Note that the definition above is equivalent to saying that there is a
neighborhood base at each point consisting of connected open sets. It is
clear that local connectedness is a topological property.
(2.E.2) Theorem. A topological space X is locally connected if and only

if every component of each open subspace of X is open. (In particular, the
components of X itself are open.)
Proof. Suppose X is locally connected and C is a component of an open
subset U. Let c be a point in C. By (2.E.1), there is a connected X-open
subset V of U that contains c. Then V must lie in C (C is a maximal
connected subset of U), and therefore C’ is open.
Conversely, let x € X and let U be any open set containing z. Then
the component V of U that contains z is open and z € V CU. Hence, X is
locally connected. O
(2.E.3) Theorem. The (finite) product of locally connected spaces is locally
connected.
Proof. Suppose X,,X2,...,Xn are locally connected and let X =
[]j-1 Xi. Suppose that U is open in X and & = (21,42,...,%n) € U. Let
U,; x U2 x --- x U, be a basic open set in X that contains x and is con-
tained in U. Since each X; is locally connected, there are connected open
sets V; such that 2; € V; C U;. Consequently, by (2.B.3), we have that
V =V, x V2 x--- x V, is a connected open set. Clearly, re V CU. O
(2.E.4) Corollary. The space E” is locally connected. O

(2.E.5) Corollary. All n-manifolds are locally connected. O
(2.E.6) Exercise. Show that local connectedness is a topological invariant,
but that it is not necessarily preserved by continuous onto functions.
(2.E.7) Theorem. If X is locally connected and f : X — Y is continuous,

onto, and closed, then Y is locally connected.
Proof. Suppose U is open in Y and C is a component of U. We will
show that C is open. For each z € f~(C), let Cz be the component of
xin f—1(U). By (2.E.2), Cy is open, and since f(z) € C, the connected
set f(Cz) also lies in C. Therefore f-!(C) = U{C, : x € f-1(C)} and
consequently f~'(C) is open. Thus f(X \ f—~!(C)) = Y \ C is closed (f is
closed and onto), and this implies C is open. O
F. SIMPLE CHAINS
We introduced paths as a means of mathematically traveling from one point
to another in a topological space. A geometrically pleasing concept, vaguely
akin to path, is that of a chain.
(2.F.1) Definition. A finite family {A,, A2,..., An} of subsets of a space

X is a simple chain in X in case A; A; # @ if and only if |i -—j| <1. A
simple chain C = {A;, A2,..., An} is said to connect points a and b in X if
and only if a € A; and b € A,. Each set A; is called a link of the chain.
Simple Chains 59
It should be observed that the links of a simple chain need not be con-
nected; thus the accompanying figure illustrates a perfectly acceptable simple
chain. This makes the characterization of connectedness given in problem
F’.1 seem even more spectacular.
CS pres Re
2 : ees A2
A3
Frequently, simple chains are forged from links that are open sets; a
common procedure for their construction consists of starting with an arbi-
trary entanglement of open sets and then extracting a suitable simple chain
from the confusion. That this may be done is a result of the following elegant
and useful theorem.
(2.F.2) Theorem. Suppose X is a connected space and that a and b are

any two points in X. Let U = {U, : a € A} be a family of open sets whose
union is X. Then there is a simple chain with links from U that connects a
and b.
Proof. Let D be the set of x in X for which there is a simple chain (with
links in UY) that runs from a to z. The set D is certainly nonempty, since a
itself is found there. We will show that D is both open and closed. Since X
is connected, this will imply (2.4.4) that D= X.
Suppose x € D and {U;, U2,...,U,} is a simple chain with a € U; and
xz € U,. We clearly have U, C D, and it follows that D is open.
To see that D is also closed, we show D contains all of its accumulation
points (recall (1.E.14)). Suppose z is an accumulation point of D and U €
U contains x. Since z is an accumulation point of D, U must intersect
D. Hence, if z is any point in the intersection, there is a simple chain
{U1,U2,...,Un} of elements in YU connecting a to z. Let r be the first
integer such that Up. U # 0. Then {U;,U2,...,U,,U} is a simple chain
between a and x, and consequently re D. O
(2.F.3) Definition. A polygonal arc in €” is an arc whose image consists

of a finite number of straight line segments.
(2.F.4) Corollary. Any two distinct points contained in a connected open
subspace U of €” may be joined by a polygonal arc in U.
Proof. For each z € U, choose €, small enough so that the open ball
S., (x) is contained in U. Then V = {S., (x) : z € U} is a family of open sets
whose union is U. Consequently, members of V may be found that form a

simple chain {V;, V2,..., Vn} running from a to b. For eachi = 1,2,...,n—1,
let z; € Vi N Viz1. Let L; be a line segment in V, that runs from a to 21.
For i = 2,3,...,n—1, let L; be a line segment in V; from z;_; to 2;. Finally,
let L, be a line segment in V, connecting z,_1 to b. These segments yield a
polygonal arc extending fromatob. O
G. COMPACTNESS
The notion of compactness, to which we now turn, is the generalization of
a second property of some intervals—closed and bounded—promised in the
introduction to the chapter. In calculus, great use is made of the fact that a
continuous function defined on a closed interval is bounded, and attains it’s
maximum and minimum on the interval. On the other hand some functions,
such as f(z) = 1/z, defined on (0,1] fail to be bounded. The property
of being a closed set is well defined in a general topological space, but the
concept of boundedness is tied to the notion of a metric. Hence in order
to generalize, we must couch the definition in terms other than closed and
bounded. The astute reader will notice that definition (2.G.2) draws on the
Heine-Borel Theorem of analysis for inspiration.
(2.G.1) Definition. Suppose X is a set and B is a subset of X. A cover of

B is a collection of subsets of X, C = {Cg : a € A}, whose union contains
B. A subcover of B is a subcollection of C that also is a cover for B. If X is
a topological space and the sets Cy are open, we say that the cover is open.
If A is a finite set, the collection C is called a finite cover of B.
The notion of cover provides topologists with one of their most successful
tools. Covers play a prominent role in the study of such diverse areas as
metrizability, dimension theory, and Cech homology, in addition to giving us
a means for defining compactness.
(2.G.2) Definition. A topological space X is compact if and only if every

open cover of X has a finite subcover.
It is easy to that that if A is a subspace of X, then A is compact if and

only if every cover of A by open sets of X has a finite subcover.
Compactness 61
Any finite space is compact, no matter what its topology and, indeed,
any space with only a finite number of open sets is compact.
(2.G.3) Exercise. Show that every closed subspace of a compact space is

compact.
(2.G.4) Example. Suppose X = R! with the open ray topology, and let
A = [0,1]. Then A is a compact subset of X, but A is not closed (in the
open ray topology).
(2.G.5) Exercise. Suppose A C X, A is compact, and X is Ty. Show that

A is closed.
We obtain a nontrivial example of a compact space, by giving a non-

standard proof of the Heine-Borel Theorem. It is important to recall that
E+, and €”, are assumed to have the usual metric.
(2.G.6) Theorem (Heine-Borel). Intervals that are closed and bounded

in €! are compact.
Proof. Let U = {Uq : a € A} be an open cover of [a,b] by open sets
of €!. Set V = {Vg : for some a € A, Vg is a component of U,}. Clearly
V also covers [a,b], and since €’ is locally connected, each member of V is
open. By (2.F.2) there is a simple chain {V,, V2,..., Vn} of sets from V that
connects a and b. Since V = U;_, V; is a connected subset of €1, it follows
from (2.A.8) that V is an interval. Furthermore, since a,b € V, we have that
[a,b] C V. For each i, 1 <i < n, select a; € A such that Vj C Ug;. Then
{Ug.y2++,Ug, ¥ 18 4 fmnite subcoyer of U. “O
(2.G.7) Exercise. Show that compact subsets of €' are bounded and
closed. Note that €! itself is not compact, since a cover consisting of sets of
the form (—n,n) for n € Zt has no finite subcover.
The next exercise is one of the most important in the entire book. The
result will be generalized at a later time with the aid of the axiom of choice.
However, as it stands, its proof represents a substantial if not formidable
task, which should more than adequately test the reader’s understanding of
both compactness and the product topology.
(2.G.8) Exercise. Show: If X; and X2 are compact spaces, then X; x X2

is compact.
Compactness, like connectedness, is preserved by continuous functions.
(2.G.9) Theorem. If X is a compact topological space and f:X > Y is

continuous and onto, then Y is compact.
Proof. Suppose C is an open cover of Y. Then {f~*(O) : O € C} is an
open cover of X. The compactness of X allows us to select a finite subcover
{ifs (Oda> 1 =A; 2).25,n} fromayial (Omen res CP wiSincer fais "ontoyrthe
collection {O1,O2,...,On} is a finite subcover of Y. O
(2.G.10) Corollary. Compactness is a topological (and a continuous) in-
variant. O
The following exercise is an important result. Among other things, it
says: many 1-1, onto, continuous functions are automatically homeomor-
phisms.
(2.G.11) Exercise. Let f be a continuous function from a compact space

into a Ty space. Show that f is closed. If, in addition, f is 1-1, show that f
is an embedding.
We now establish a complete and useful characterization of compact
subsets in Euclidean spaces—the n-dimensional Heine-Borel Theorem.
(2.G.12) Theorem. Suppose A C €”. Then A is compact if and only if A

is closed and bounded.
Proof. Suppose A is closed and bounded. Let €} (= €') denote the i-th
coordinate space of €” and let p; : E” —> E} be the projection map. It is easily
seen that p;(A) is bounded in €} for each i, and hence, for each i = 1,2,...,n
there is a closed interval [w;,z;] C €} such that p;(A) C [w;,2:]. However,
this implies A C [w1, 21] x [we2,22] x «+: X [Wn, Zn]. Since the product of
closed and bounded intervals is compact by (2.G.6) and (2.G.8), it follows
from (2.G.3) that the closed and bounded set A is compact.
Now suppose A is compact. By (2.G.9), p;(A) is compact for each i,
and thus by (2.G.7), p;(A) must be bounded. Since this is true in each
coordinate, A is bounded also. That A is closed follows from (2.G.5). O
As an immediate corollary we have another theorem that should be
familiar from calculus.
(2.G.13) Corollary. Suppose A is compact and f : A > €! is continuous.

Then f attains its maximum and minimum. O
The next theorem, useful in itself, also has a proof that is typical of
many theorems having compactness as a hypothesis.
(2.G.14) Theorem. Suppose A x B is a compact subset of a product space

X xY and W is an open subset of X x Y which contains Ax B. Then there are
open sets U and V in X and Y, respectively, such that Ax BCUxV CW.
Proof. Fix a € A. For each b € B, let Uy x Vi be a basic open set in
X x Y such that (a,b) € U, x V, C W. Then {V, : b € B} is an open
cover of B, and hence there is a finite subcover {Vp,,Vs,,..., Vo, }of B. Let
U* =(\i_, Us,, and V* = U_, V,. Repeat this procedure for each a € A to
obtain an open cover {U*% : a € A} of A and a corresponding family of open
Two Characterizations of Compactness 63
sets {V* : a € A} each of which contains B. Note that U?xV? C W for each
a € A. Once more, we use compactness (this time of A) to obtain a finite
subcover {U%!,U%,...,U°™} of A. Set U = Wi US and V1) eV" to
complete the proof. O
H. TWO CHARACTERIZATIONS OF COMPACTNESS

The notion of compactness may be characterized in a number of ways. The
next theorem gives what is perhaps the most common and useful of these
characterizations. The reader should realize from the proof, if not before-
hand, that (2.H.2) is just a restatement of the definition of compactness using
DeMorgan’s rules.
(2.H.1) Definition. A collection of sets {C. : a € A} has the finite

intersection property if and only if (\{Cya : a € K} # @ whenever K is a
nonempty finite subset of A.
(2.H.2) Theorem. A topological space X is compact if and only if we have

(\{Ca : a € A} 4 @ whenever C = {Cy : a € A} is a collection of closed
subsets of X with the finite intersection property.
Proof. Suppose X is compact and C = {Cy : a € A} is a collection of
closed subsets of X. We will show: if \{Ca : a € A} = 9, then C does not
satisfy the finite intersection property. For each a € A, let Up = X \ C4.
Then it is immediate from the hypothesis and one of the DeMorgan rules that
User Ua = X, and hence {U, : a € A} is an open cover of X. Since X is
compact, there is a finite subcover, U,,U2,...,Un. But another application
of a DeMorgan rule yields ();_, C; = @. Hence C does not have the finite
intersection property.
Suppose X is not compact. Then there is an open cover {U, : a € A}
of X having no finite subcover. For eacha € A let Cy = X \Uy. It is easy to
see that the collection C = {Cg : a € A} has the finite intersection property,
but the intersection of all members of C is empty. O
The next theorem gives an illustrative application of (2.H.2).

(2.H.3) Theorem. Suppose {A; : i € Z*} is a countable family of compact
subsets of a Ty space X, such that for each 7, Ai+1 C Aj. If there is an open
set U such that ()72, Ai = A C U, then there is an integer N such that for
tN A GU,
Proof. First note the conclusion follows trivially if some A; = §. So we

consider the case where each A; is nonempty. Now suppose C; = A; \ U is
nonempty for each i. Then {C; : i € Z*} is a collection of closed sets (2.G.5)
with the finite intersection property. This follows easily from the observation
that if n <_m, then Am C An and hence Ay \U C An \U. Since for each
i, C; C A; and A, is compact, we have from (2.H.2) that ()2, C; # 0. But

Ne Cr AEA AAGT) ="(QEPADN VeeOme contradiction 9H
The next exercise gives a couple of extensions and generalizations of the
previous theorem.
(2.H.4) Exercise.
(a) Let A be a partially ordered set with the property that if a, 6 € A,
then there is a \ € A suchthat A > a and A > B. Suppose that {A, : a € A}
is a collection of compact subsets of a T> space such that Ag C Ag if and
only if a > B. Show that if {Aa : a € A} C U, where U is open, then
there is a yp € A such that A, C U for each a > p.
(b) Suppose that {A, : a € A} is a family of closed compact sets such
that (\{Aq : a € A} is a subset of an open set U. Show that there is a finite
set {Q1,Q2,...,Qn} C A such that (\_, A; CU.
(c) Suppose in (2.H.3) one also assumes that each A; # 0. Show that
(d) Can one make a similar improvement in parts (a) and (b)?
Our second characterization of compactness, based on nests of closed
subsets, is somewhat subtler than its predecessor. Some equivalent of the
axiom of choice is apparently needed for its proof.
(2.H.5) Theorem. A space X is compact if and only if each nest of

nonempty closed subsets in X has nonempty intersection.
Proof (Catlin [1968]). The proof of the “only if” part of the theorem is
rendered trivial by (2.H.2).
Suppose each nest of closed nonempty subsets of X has nonempty in-
tersection. We show X satisfies the finite intersection property criterion for
compactness.
To this end, let C be a family of closed subsets of X with the fi-
nite intersection property and let K = {K C X : K is closed andC U
{K} has the finite intersection property}. The Kuratowski lemma (0.D.4)
may be employed to extract a maximal nest NV (with respect to inclusion)
ink. Let D=(\{N : N € N}. By hypothesis, D is not empty; thus, to
complete the proof it suffices to show that DC (){C : C €C}.
If D is not a subset of (}{C : C € C}, then there is a C € C with
COD # D (but COND C D). We obtain a contradiction by showing
that CM D € K, which of course negates the maximality of NY. Note that
CoD #9, sine CND=(\{CNN : NEN} and {CNN : NEN} is
a nest of closed sets (CNN # 9, since CJ{N} has the finite intersection
property). To see that CND € K, observe that whenever C1, C2,...,Cn €C,
then C1 NC2N---NC~ANCND =(\{(CinC2n---AC,NCNN) : NEN},
and the latter set is nonempty, since it is an intersection of a nest of closed
Compactness: Local, Countable, and Sequential 65
nonempty subsets of X. Thus, CU{CND} has the finite intersection property,

and consequently CN DEK. O
1. COMPACTNESS: LOCAL, COUNTABLE, AND

SEQUENTIAL
Although there are nearly as many varieties of compactness as there are
topologists, we will concentrate for now on just three such mutations.
(2.1.1) Definition. A topological space X is locally compact if and only if

for each x € X and for each open set U containing «, there is an open set V
such that z € V CV CU and V is compact.
(2.1.2) Examples.
1. For each n € Z*, E” is locally compact.
2. The set R’ with the open ray topology is not locally compact.
3. For each i € Zt, define A; = {(2,1/i) : 0 < x < 1} and let
Ao = {(x,0) : 0< a <1}. Then U7, A; is locally compact.
Note that in the last example if Ao is replaced by {(0,0), (1,0)}, then
Uso, Ai is not locally compact.
Suppose that X = R! with the finite complement topology, Then X is
compact but not locally compact (the closure of any open set is X). In order
to obviate this problem, local compactness is sometimes defined by merely
requiring that every point have a compact neighborhood. In a T» space, the
two concepts coincide, as the reader is asked to establish in the next exercise.
(2.1.3) Exercise. Show that a T) space X is locally compact if and only if
each point in X has a compact neighborhood.
(2.1.4) Exercise. Suppose that X1, X2,...,Xn are locally compact spaces.
Show that [];"_, X; is locally compact.
(2.1.5) Exercise. Show that local compactness is a topological invariant.
Is it a continuous invariant?
(2.1.6) Definition. Suppose that f : Zt — X is a sequence. A sequence
g: Z* + X is a subsequence of f if and only if there is a strictly increasing
function h: Z+ — Z+ such that fh = g. We denote f(n) by zp and h(2) by
n;. The subsequence g is defined by g(t) = rn, and is generally denoted by
{Zn,}-
(2.1.7) Definition. A topological space X is sequentially compact if and
only if every (infinite) sequence in X has a convergent subsequence.
In the next chapter, we show that the notions of compactness and se-
quential compactness coincide in metric spaces. However, in general there
is no precise relationship between these two concepts. Problem 16 from sec-

tions A-B in Chapter 6 gives an example of a compact space that is not
sequentially compact. Furthermore, the set [0,{) with the order topology
is sequentially compact (by (0.E.4), since all sequences in [0,) have a least
upper bound in [0,). However [0,2) is not compact (why’?)).
The notion of countable compactness is somewhat akin (although it is
not obvious a priori) to the idea of sequential compactness.
(2.1.8) Definition. A topological space X is countable compact if and only

if every countable open cover of X has a finite subcover.
Compactness clearly implies countably compactness; however, it follows

from the next theorem and the preceding remarks that [0,9) is countably
compact, but not compact.
(2.1.9) Theorem. If X is sequentially compact, then X is countably com-

pact.
Proof. Suppose X is sequentially compact and U = {U,,Up,...} is a

countable open cover of X. If no finite subcover of U exists, then for each
n, there is a point xz, € X \Uj_, Ui. The sequence {x,} has a subsequence
{Zn;} that converges to some point z* € X. Since U/ is a cover of X, there
is aU; € U such that 2* € U;,. However, for i sufficiently large (and greater
than k), we have that x,, € Ux, which is impossible. O
We now introduce a property of topological spaces that might be viewed

as a weak version of compactness.
(2.1.10) Definition. A topological space X is Lindeléf if and only if every

open cover of X has a countable subcover.
A classic theorem involving Lindelof spaces is the following.
(2.1.11) Theorem (Lindelof). If X is a second countable space, then X

is Lindel6f.
Proof. Let U be an open cover of X and let B be a countable basis
for X. Then for each x € X, there are sets Uz € U and B, € B such that
xz € B, CU;z. Since {B, : x € X} is a countable family and covers X (many
B,’s may be duplicated), there is a corresponding countable collection of the
Umsithatecoversiex.
Note that it follows from (1.G.11) that separable metric spaces are Lin-
delof.
(2.1.12) Exercise. Show that a metric space X is Lindel6f if and only if X
is second countable. [Hint: For each positive integer n, cover X with sets of
the form S,/,(z).]
One Point Compactifications 67
(2.1.13) Exercise. Show that the concepts of Lindel6f, second countability,

and separability coincide in a metric space.
J. ONE POINT COMPACTIFICATIONS
As we proceed, the convenience of compactness will become increasing ap-

parent. Compactness is such a powerful property that one frequently finds
it useful to embed noncompact spaces as dense subsets of compact spaces.
(2.J.1) Definition. A compactification of a space X is a pair (Y, h) where

Y is a compact topological space and h : X — Y is an embedding whose
image is a dense subset of Y. That is, h is an embedding, Y is compact, and
A(X) =.
Compactifications were originally inspired by problems in analysis where
it was sometimes found desirable to add appropriate “boundary points” to
a region. Actually, there are many ways of doing this; for instance, we will
see in (17.E.1) that every compact connected n-manifold may be viewed as
a compactification of {x € E” : |z] < 1}.
In this section we are concerned with what is probably the simplest
and most common means of compactifying a space, the one point (Aleksan-
drov) compactification. The following situation motivates the definition that
eventually follows.
In €? consider the sphere S = {(z,y,z) : 2? +y?4+(z—1/2)? = (1/2)?}
which rests on top of the zy plane. Denote the north pole of S by p, i.e.,
p = (0,0,1). The following homeomorphism “wraps” the plane around the
sphere by embedding €? as S \ {p}.
; Ey x y a? +y?
(c= l+a?+y?’ l+a2?+y?? 14+a?4+y?
Clearly, (S,h) is a compactification of the plane.

It is instructive to consider the relation between open sets in S and the
corresponding open sets in €? resulting from the homeomorphism h. If U
is open in €?, then h(U) is open in S, and conversely, open sets of S \ {p}
have homeomorphic images in €*. Consider, however, an open set V of S$
that contains p; its complement is compact, and hence h~'(S\V) is compact
in €2. Thus V may be described as corresponding to the complement of a
compact set in €? together with the point p.
(x,y)
(2.J.2) Definition. Suppose that X is a topological space, and let oo be

a point not in X. Define a topology for X* = X U {oo} as follows: a set
U c X* is open if and only if either
(i) co gU and U is open in X, or
(ii) co €U and X* \U is a closed and compact subset of X.
The set X* with this topology is called the one point compactification
(Aleksandrov compactification) of X.
(2.J.3) Exercise. Show that the one point compactification, X*, of X has
the following properties:
(i) X* is a topological space;
(ii) X is a subspace of X*;
(iit) X* is compact;
(iv) X is locally compact and T> if and only if X* is To;
(v) X is dense in X* if and only if X is not compact.
We consider a considerably more sophisticated compactification, the
Stone-Cech compactification, in problem 9 from sections A—B of Chapter 7.
PROBLEMS
Sections A and B
1. (Riesz-Lennes-Hausdorff Separation Criteria) Suppose A is a subspace

of a topological space X. Show that A is disconnected if and only if
there are subsets C and D of X satisfying:
(ie AG GD
(ii) ANC#6, AND, and
(iti) (CND)U(CND)=6
Problems 69
De Suppose {U,, : n € Z*} is a countable collection of connected subspaces

of a space X such that U,MUn41 4 @ for each n. Show that WU, :
n € ZT} is connected.
Show that an infinite set with the finite complement topology is con-
nected.
Prove or disprove: If A;, A2,... is a collection of closed connected sub-
sets of the plane such that An41; C Ap for n = 1,2,..., then (lai An
is connected.
Suppose F is a proper subset of a space _X, and suppose A is a connected
subset of X such that AN E # @ and AN (X \ E) # @. Show that
ANKE #@and ANF (X \ E) £0.
Suppose A C &€! has the fixed point property. Show that A is a point
or a closed and bounded interval.
The intermediate value theorem has somewhat of a converse. Suppose
f : [a,b] + €'. Show that f is continuous if and only if
(i) whenever 21,22 € [a,b] and f(z) < c < f(x), there is an
g €)[1,%9,| such that f(2)=c, and
(it) f—*(y) is closed for each y € €!.
Find an example of a function f : €! > &} for which the conclusion of
the intermediate value theorem holds, but f is not continuous.
Show that every connected open subset of €? can be written as a disjoint
union of “open line segments.”
10. Suppose X is a JT; space. Show that every connected subset of X con-
taining more than one point is infinite.
th Suppose A C €! is an interval and f : A > €! is continuous and 1-1.
Show that f is either strictly increasing or strictly decreasing.
12. Suppose X is a space such that every pair of points in X lie in a con-
nected subset of X. Show that X is connected.
tS. Suppose (X,U/;) and (X,U2) are connected. Is (X,U, NU2) connected?
14. Show that €! with the open ray topology is connected.
15. Suppose h : €! + €! is a homeomorphism and has no fixed point. Show
that h” (the composition of h with itself n times) is also fixed point
free. Find a homeomorphism h : €2 > €? that is fixed point free but
for which h? has a fixed point.
16. A metric space is well-chained if and only if for each e > 0 and for each
pair of points x,y in X there is a finite subset {1,72,...,%p,} in X such
that 2; = 2, v, = y, and d(x;,2;41) < € fori = 1,2,...,2—1. Show
that if X is connected, then X is well-chained, but that the converse is
false.
17. Suppose f : €! > &! is a homeomorphism and f” = id. Show that

if f is order preserving, then f = id, and if f is order reversing, then
je Sa) .
18. Find a continuous function f : J — I with precisely two fixed points, 0
and 1.
19. Show that if a continuous function f maps a half-open interval onto
itself, then f has a fixed point.
20. Show that each open subset of €” can be written as a countable union
of disjoint connected open sets.
21. What can you say about the connectedness of the disjoint union of
connected spaces?
Section C
1. Show that path connectivity is a topological invariant (in fact, a contin-
uous invariant).
2. Suppose Xj, X2,...,X, are topological spaces. Show that likes X; is
path connected if and only if each X; is path connected.
3. Prove or disprove: If A is a path connected subset of X, then A is path
connected.
4. Prove or disprove: If A and B are path connected subsets of a space X
and ANB # G, then AU B is path connected.
5. <A topological space X is said to be contractible if and only if there is
a continuous map H : X x I > X and a point z* € X such that
H(a,0) = weand H(#,1) = a* for each € xX. Prove-thatiévery
contractible space is path connected.
Section D
1. Define an equivalence relation on a space X by setting x ~ y if and only
if for each separation (U,V) of X, x and y are both in U or both in V.
(a) Show that ~ is an equivalence relation. The equivalence classes are
called quasicomponents.
(b) Show that quasicomponents are closed.
(c) Show that each component of X is contained in some quasicompo-
nent.
(d) Describe the quasicomponents of the following subset of €?:
A= ie4n]U {(0,0), (1,0)}, where A, = {(2,y) : O0<a2<1,y= 1/n}.

Problems 71
2. Show that a space X has an infinite number of components if and only

if there is a countably infinite family of nonempty disjoint sets in X that
are both open and closed.
Find an example in the plane where components and path components
do not coincide.
Show that every component of a space X is a union of path components.
Suppose a topological space X has the property that for every open set
U and for each point p € U, there is an open set V with pe V CU
such that if y € V, then y may be connected to p by a path lying in U.
Show that for every open set W containing p, there is a path connected
neighborhood Z with pe ZC W.
It is possible that spaces X and Y fail to be homeomorphic even though
there are continuous bijective maps f : X ~ Y andg: Y 7 X.
For example, let X = (U>-.9[3n,3n + 1)) U (Ue2,{3n + 2}), and Y =
(Up 9(32, 3n + 1)) U (UP, {3n + 2}). Find continuous bijective maps
f:X —7Y andg:Y — X, and show that X and Y are not homeomor-
phic.
Let X = {x € R! : wis a dyadic rational}, let Y = {y € R!
y is rational and y ¢ X}, and let Z = {z € R! : z is irrational}. Then
R! = X UY UZ. Define a topology U for R! whereby
(i) X and Y are open,
(ii) if U C X or U CY, then U is open if it is open in the usual
(E} relative) topology for X or Y, and
(iit) a neighborhood basis for a point z € Z consist of sets of the
form {z}U{we XUY : |w—2z| <r} wherer
> 0.
Show that X and Y with the relative topology are totally disconnected
and that Z is discrete. Let A= R! \ X and B= R!\Y. Show that A
and B are totally disconnected but that their union is not.
A function f : X — Y is connected if and only if the image under f of
every connected set is connected. Prove or disprove: If f : € 1 + €1 is
connected, then f is continuous.
Suppose f : X — Y is connected (see problem 8), and f~'(y) is con-
nected for each y € Y. Show that if X and Y are T2, then for each
y € Y, f—'(y) is closed.
Show that if f : €1 — €! is connected (see problem 8), and f~'(y) is
connected for each y € €!, then f is continuous.
in Show that a path component of an n-manifold is an n-manifold.
12. Suppose X is T, with an explosion point p, and f : X — X is con-
tinuous. Show that if f—1(p) is finite (or empty), then f has a fixed
point. [Hint: if {a1,22,...,2n} C X \ {p}, then X \ {21,22,...,2n} is

connected.] |
13. A function f : X — Y is super continuous if and only if f —1(B) is open
for each B C Y. Show that if f is supercontinuous, then f is constant
on each quasi-component of X.
Section E
TS Show that open subsets of locally connected spaces are locally connected.
Is the same true for arbitrary subspaces of locally connected spaces?
Suppose X = AUB, where A and B are locally connected. (a) Must X
be locally connected? (b) If A and B are also closed, must X be locally
connected?
Show that if x and y are points belonging to different components of a
locally connected topological space X, then there is a separation (U, V)
of X such that r€ U andy eV.
Does the conclusion of problem 3 hold if the hypothesis “locally con-
nected” is dropped?
Show that in a locally connected T> space, components and quasicom-
ponents coincide.
Prove or disprove: If a space X is locally path connected, then X is
locally connected. What about the converse?
Suppose X is connected and locally connected and f : X — I is contin-
uous and bijective. Show that X and J are homeomorphic.
Section F
I Show that a space X is connected if and only if for each x,y € X and
for each family U of open sets whose union is X, UY contains a simple
chain from z to y. [Warning: You may not assume that the links are
connected]]
Show that every two points in a connected n-manifold can be joined by
a simple chain whose links are homeomorphic to €”.
Suppose X is a locally connected T; space and p, x and y are points
in X. Show that x and y are in distinct components of X \ {p} if and
only if every simple chain of connected open sets from zx to y has a link
containing p.
A metric space K is chainable if and only if for each € > 0, there is a
simple chain of open sets covering K such that each link has diameter
less than e. Are the following spaces chainable?
Problems 73
(a) I
(b)" lx T
(c) The unit circle
(d) The topologist’s sine curve
Section G and H
L: Show that if X is an infinite set with the finite complement topology U,

then (X,U/) is compact.
Suppose B is a basis for a topological space (X,U). Show that X is
compact if and only if every cover of X by basic open sets has a finite
subcover.
Find examples to show: a) the intersection of two compact subsets of
a topological space may fail to be compact, and b) the closure of a
compact subset may fail to be compact.
Find a space in which points are closed, but where the intersection of
compact subsets is not necessarily compact.
Suppose (X,7) is a compact space with the property: whenever C is
a compact space and f : C — X is a continuous bijection, then f is a
homeomorphism. Show that each compact subset of (X,7) is closed.
{Hint: Let C be a compact subset on X and let YU be the smallest
topology for X that contains JT and {X \ C}.]
Show that [0,1] c €' may be covered by a family of closed intervals in
€! that has no finite subcover.
Suppose C is a compact subset of a T; space X and that z € X \C.
Show that there are disjoint open subsets U and V with x € U and
CE.
Suppose C' and D are disjoint compact subsets of a Tz space X. Show
that there are disjoint open subsets U and V of X such that C' Cc U and
iD hecai
Show that if X is T) and C is a family of compact subsets of X such
that every finite intersection of members of C is connected, then (){C :
C € C} is connected.
10. Show that a first countable space is T2 if and only if every compact
subset is closed.
tie Show that X is compact if and only if every open cover of X has an
irreducible subcover (a subcover that fails to be a cover if any element
is removed).
12: Suppose A C X and let Y be a compact space. Show that if U is a

neighborhood of A x Y in X x Y then there is a neighborhood V of A
such that V x Y CU.
13. Show that if f : J - X is continuous, onto and open, where X is T>
and has at least 2 points, then J and X are homeomorphic. [Hint: Find
the smallest closed interval [0,a] such that f((0,a]) = X.]
14. Suppose X is T2 and Y is compact and T2. Show: f : X — Y is
continuous if and only if {(x, f(x)) : € X} is closed in X x Y.
15. Suppose (X,//) is a compact T> space. Show that
(a) IfU GU’, then (X,U’) is not compact.
(b) IfU" GU, then (X,U") is not To.
16. Let X = Z* and let U be the topology for Z* that consists of 0, X,
and sets of the form {1,2,...,n}. Show that every closed and compact
subset of (X,U/) is empty.
Wes Let X = X; x Xo x--: X Xp and suppose that a subspace A of X is
compact if and only if A is closed. Show that each X; has the same
property.
ts: Suppose X is a compact metric space. Show that X is connected if and
only if X is well-chained (see problem A-B 16).
19. Show that the following space X is an example of a noncompact subspace
of €? that has the fixed point property. For each n € Zt, let A, =
((ayy) €'€? * & = ln, Oy 21). Let B =A (ag) 6 62 0 <<
Ly = 0}. Then C= (U2 A) UB.
20). Suppose X is a compact T> space and f : X — X is a continuous map.
Show that there is a nonempty closed set A C X such that f(A) = A.
ZL. Show that a compact locally connected space has only a finite number
of components.
Section |
Ue Suppose X is a connected, locally connected, locally compact T> space.

Show that if z,y € X, then there is a compact connected set containing
both z and y.
Suppose X is a locally compact space, Y is a T2 space, and f: X ~ Y
is continuous, open, and onto. Show that if C is a compact subset of Y,
then there is a compact subset D of X such that f(D) =C.
Let X = [—1,1]. Define a topology for X by declaring sets of the form
[-1,b), (a,b), and (a,1] to be open, whenever a < 0 < b (include of
course X and #)). Show that X is compact, but that no open set (a, b)
is locally compact.
Problems 75
4. Suppose X is a Ty space and A is a dense locally compact subspace.

Show that A is open.
Suppose X is an infinite set and A C X. Define a topology for X by
declaring @ and all sets that contain A to be open. Is this space locally
compact?
Let C' be a compact subspace of a locally compact metric space (X, d).
Show that there is an € > 0 such that {y € X : d(y,C) < €} is compact.
Show that every countably compact metric space is separable.
Show that if every closed ball in a metric space is compact, then the
space is locally compact and separable.
Let C = {{xn} : {xn} is a convergent sequence in €!}. Define a func-
tion d: C x C — (0,00) by d({tn}, {yn}) = sup{|rtn — yn|}. Show that
d is a metric and that (C,d) is separable but not locally compact.
ta), A space X is pseudocompact if and only if each continuous function
f:X 7 &' is bounded.
(a) Show that pseudocompactness is a continuous invariant.
(b) Show that compact implies pseudocompact, but that the converse
does not hold.
ite Let X = R! and let U be a topology for X that consists of 0, X, and
sets of the form (—n,n) where n € Zt. Show that each € X has a
neighborhood basis consisting of compact sets, but that not all points
have a closed compact neighborhood.
1Z Show that the product of a compact space and a countably compact
space is countably compact.
13. Show that a 7; space is countably compact if and only if every infinite
subset has an accumulation point.
14. Show that the continuous image of a countably compact space is count-
ably compact.
15. Show that a second countable T, space is compact if and only if it is
sequentially compact.
16. Suppose X is a countably compact JT; space and that U;,U2,... are
open subsets of X such that ();°, Ui = {x}. Show that the sets U; need
not form a neighborhood basis at 2.
ive Let (Z+,U) have for a basis {{2n —1,2n} : n = 1,2,...} and let
(Z*,V) have a basis {{1,n} : n = 1,2,...}. Show that (Zt,U) is
locally compact and (Z*+,V) is not. Define f : (Z*+,U) > (Zt,V) by
f(2n —1) =1 and f(2n) =n. Show that f is open but not continuous,
and consequently local compactness is not preserved under open maps.
18. Show that a TJ; space X is countably compact if and only if every infinite
open cover of X has a proper subcover.
Section J
1. Let X be the rationals with the relative topology and let X* denote the
one point compactification of X. Show directly that X* is not T>.
2. To what is the one point compactification of (0, 1] homeomorphic?
3. Show that the one point compactification of Zt U {0} is homeomorphic
toOLOLU Tn = neta}. ;
4. Suppose X is a locally compact T> space and f : X + €' is continuous.
Show that there is a continuous function f : X* — €! such that f(z) =
f(z) for each x € X if and only if for each € > 0, there is a compact
subset K, of X such that |f(x) — f(y)| < € whenever x,y ¢ K..
5. Suppose X is a Ty space and X™* is first countable. Is X necessarily
locally compact?
6. Let X = (0,1) U (1,2) U (2,3) U (9,10) be given the relative topology
(with respect to €!). Describe the one point compactification of X.
Chapter 3
METRIC SPACES
To consolidate and extend some of the ideas introduced in the previous two
chapters, we now focus our attention on metric spaces. We considera variety
of metric spaces beginning with those that are compact.
A. COMPACTNESS IN A METRIC SETTING

Sequences are particularly useful in the theory of metric spaces. We have
already seen an example of this in (1.F.9), where the continuity of a function
was completely determined by the function’s behavior with respect to con-
vergent sequences. Subsequences are the key for characterizing compactness
in a metric setting.
(3.A.1) Definition. A point z in a topological space X is a cluster point

of a sequence {z;} if and only if for each neighborhood U of z and for each
i € Z*, there is an integer k > 1 such that rz, € U. That is, the sequence
meets U infinitely often.
(3.A.2) Exercise. Suppose (X,d) is a metric space and z is a cluster point

of a sequence {z;} in X. Show that {z;} contains a subsequence converging
to 2.
The following example, the Porcupine space, shows that in a more gen-
eral setting it is possible for a sequence to have a cluster point that is not
the limit of any subsequence.
(3.4.3) Example and Exercise. Let ¢;,¢2,... be a sequence of disjoint
closed rays in the plane emanating from o = (0,0). Let X = U7, 4. It is
fet
78 3. Metric Spaces
convenient to adopt the following notation: If a,b € £; and d(o,a) < d(o,b),
where d is the usual metric for €?, we write a = b and define
(a,b) ={(1-—tha+tb: t € (0,1)}, and
[a,b) = {(1—t)a+tb: t € [0,1)}.

Now, for each x € X \ {o}, x € £; for some 7, and we define an open
neighborhood base N; for z by
Nu=A(a,
0) be: O <<G-—.0).
Let €; represent a point on @; of distance e; from o. We define a neighborhood

base at o to be
Noah UG XU = U [o,€;) where card(’) < oo and each €; > 0

i€Z+\K
Let B = U,ex Ne. Show:

(a) B is a base for a topology on X.
(b) There does not exist a countable neighborhood base at o.
Define a sequence {z;} in X as follows: x, is a point on 4, that has
(Euclidean) distance 1/2 from 0. points 22 and x3 are on @; and é2, respec-
tively, at distance 1/4 from o. The points 24, r5, zg, and 27 are on 4), 9,
é3, and £4, respectively, at distance 1/8 from 0, etc.
Show:
(c) The sequence {z;} has o as a cluster point.
(d) There is no subsequence of {x;} that converges to o.
In a metric space the notions of compactness, sequential compactness,
and countable compactness coincide. Before proving this, we introduce one
additional concept that appears frequently in analysis.
(3.A.4) Definition. A metric space (X,d) is totally bounded if and only if

for each € > 0, there is a positive integer NV, and a finite subset {z1,22..., Gn}
of X such that X =)" Se(z;):
(3.A.5) Remark. You should note that total boundedness is not a topolog-
ical property. For instance, the open interval (0,1) with the usual metric is
totally bounded. However, it is homeomorphic with €!, which is not totally
bounded with its usual metric. Nevertheless; if certain conditions are im-
posed on a metric space, then total boundedness does have some topological
interest. This is illustrated by the following theorem, which establishes an
important property of sequentially compact metric spaces.
Compactness in a Metric Setting 79
(3.4.6) Theorem. If (X,d) is a sequentially compact metric space, then

(X, d) is totally bounded.
Proof. Suppose € > 0. Pick an arbitrary point 2; € X. If X = S. (27),
we are done; if X # S.(x1), choose a point rg € X \ S.(x,). Again, if X =
S-(z1) US. (x2) we stop; otherwise, select a point z3 € X \ (S.(#1) US. (22)).
Continuing, we eventually meet an integer N, such that X = ee Del Li odt
that did not happen, we would have a sequence that fails to have a convergent
subsequence since d(z;,z;) >e for alli,j. O
(3.A.7) Corollary. If (X,d) is a sequentially compact metric space, then
(X,d) has a countable basis.
Proof. For each n € Zt, X may be covered by a finite family of neigh-
borhoods Sj /n(Z1),---,Si/n(k, ). The collection of all these neighborhoods
is clearly a countable basis for X. O
(3.A.8) Theorem. Suppose (X,d) is a metric space. Then (X, d) is count-

ably compact if and only if (X,d) is sequentially compact.
Proof. We have already seen that sequential compactness implies count-
able compactness (2.1.9). Suppose then that (X,d) is a countably compact
metric space which is not sequentially compact. Let {x;} C X be a se-
quence having no convergent subsequence. We will construct a countable
open cover of X that has no finite (in fact no proper) subcover. First ob-
serve that (3.A.2) implies that the sequence {z;} does not have a cluster
point. Let A = {z; : 1 € N}. Clearly A is infinite (otherwise {xz;} would
have a cluster point). For each x € X, there is an open neighborhood V,
of z such that V, 9 A is finite. Since points are closed, there is an open
neighborhood U, of x such that U;N A =9@ if x ¢ A, and U,N A = {z} if
zéA. Let Up = W{U, : © € X\ A}. Then {{U.} : ae A}U{Up} isa
countably infinite cover of X which has no proper subcover. O
(3.A.9) Definition. If A is a subset of a metric space (X,d), we define the

diameter of A, denoted by diam A, to be sup{d(z,y) : z,y € A}.
The next theorem and its corollary (3.4.13) are amazingly useful, both
in topology and in analysis.
(3.4.10) Theorem. Suppose U/ is an open cover of a sequentially compact
metric space X. Then there is a A > 0 such that for each x € X, there
is a set U € U with S)(x) C U. (Equivalently, there is a A > 0 such that
whenever diam A < 4, there is a U € U that contains A.) Such a number A
is called a Lebesgue number for the cover U.
Proof. If no such X exists, then for each n, there is an z, € X such that
Si /n(Zn) fails to be in any member of U/. Since X is sequentially compact,
there is a cluster point x of the sequence {z,} (why?). Since UY is a cover
80 3. Metric Spaces
of X, we have z € U for some U € U. Let r = d(x,X \U) > 0 and choose

m large enough so that d(tm,z) < r/2 and 1/m < r/4. Then Sj/m(tm) is
contained in U, which contradicts the way that 2 was chosen. O
The following is our principal theorem for this section.
(3.A.11) Theorem. Suppose (X,d) is a metric space. Then X is compact

if and only if X is sequentially compact.
Proof. If X is compact, then X is countably compact. Hence by (3.A.8)
we have that X is sequentially compact.
Conversely, suppose X is sequentially compact. Let / be an open cover
of X. By (3.A.10), there is a \ > 0 such that if « € X, then S)(z) C U for
some U € U. Furthermore, by (3.A.6), there is a finite subset {11,22,...,2n}
of X such that X = Ui, S,(2i). Now choose for each i = 1,2,...,n, a set
U; € U with S)(2;) C U;. Then {U; : i =1,2,...,n} is a the desired finite
subcover. O
The following corollary has many applications in analysis.
(3.A.12) Corollary (Bolzano-Weierstrass Theorem). Every bounded

sequence {z;} of points in €” (with the usual metric) contains a convergent
subsequence.
Proof. Let A be the set whose members are precisely the elements of
the sequence. Then A is certainly closed and bounded, and hence compact.
The corollary now follows immediately from theorem (3.4.11). O
(3.A.13) Corollary. Suppose (X,d) is a compact metric space and U is an

open cover of X. Then there is a Lebesgue number A forU. O
Theorem (3.4.6) can be applied to prove the following basic result.
(3.4.14) Theorem. Suppose A and B are disjoint subsets of a metric space

(X,d). If A is compact and B is closed, then d(A, B) > 0. Furthermore, if
d(A, B) =r, then there is an a € A such that d(a, B) =r.
Proof. Suppose to the contrary d(A,B) = 0. Then for each positive
integer n, there are points a, € A and by € B such that d(an, bn) < 1/n. By
(2.4.11), there is a subsequence {a,,} of {a,} that converges to some point
a € A. But then the corresponding subsequence {b,, }must also converge to
a contradicting the fact that the closed sets B and A are disjoint. Therefore,
d(A, B) > 0.
To prove the second part of the theorem, we suppose d(A, B) = r. Now
define dg : A — E* by dg(a) = d(a,B). Then dg is continuous (1.F.12)
and since A is compact, we have that dg(A) is compact and hence closed
in €'. Therefore, r € dg(A), and consequently there is an a € A such that
dp (ay dla,B\= 7. 2D
Compactness in a Metric Setting 81
If d(A,B) = r, can we always find an a € A and b € B such that

d(A,b) or d(a,b) = r? The answer is no (see problems A-12 and B-28).
The following example in €? shows that the compactness of A is a necessary
condition in the preceding theorem.
(3.4.15) Definition. A function f : (X,d) > (Y, p) between metric spaces

is uniformly continuous if and only if for each € > 0, there is a 6 > 0 such
that if x,y € (X,d) and d(z,y) < 6, then p(f(z), f(y)) <e.
Note that 6 depends only on ¢€ and not on the individual points z and
y. This is the motivation for using the modifier “uniformly.” Obviously,
uniform continuity implies continuity, but the converse does not hold.
(3.4.16) Exercise. Find an example of a homeomorphism between two

metric spaces that is not uniformly continuous.
(3.4.17) Theorem. Suppose (X,d) and (Y,p) are metric spaces. If X is

compact and f : X — Y is continuous, then f is uniformly continuous.
Proof. Let « > 0. For each x € X there is a 6; > O, such that if
d(xz,y) < dz, then p(f(z), f(y)) < €/2. Cover X with sets of the form
S5,(x), and let A be a Lebesgue number for this cover. Now suppose w
and z are points of X such that d(w,z) < ». Then both w and z belong
to some member of the cover, say S5,(x). It follows that p(f(w), f(z)) <
p(f(w), f(x)) + (f(z), f(z) <€/2+e/2. 0
Strangely enough, from a strictly topological point of view, all contin-
uous functions between metric spaces are in a sense uniformly continuous.
That is, a continuous function f mapping a metric space (X, d) into a metric
space (Y, p) can be converted into a uniformly continuous function by switch-
ing to an appropriate equivalent metric on X. Thus, uniform continuity of
a given function is a metric property rather than a topological one.
(3.4.18) Theorem. (Levine [1960]). Suppose (X,d) and (Y,p) are metric
spaces and f : X + Y is a continuous function. Then there is a metric d*
on X, equivalent to d, that makes f uniformly continuous.
82 3. Metric Spaces
Proof. Define d*(a,b) = d(a,b) + p(f(a), f(b)). To see that d* is a

metric, we need only check that d*(a,c) < d*(a,b) + d*(b,c) for a,b,c € X.
But this is immediate, since d*(a,b) + d*(b,c) = d(a,b) + p(f(a), f(6)) +
d(b,c) + p(F(0), f(c)) > d(a,c) + (f(a), f(c)) = a*(a,¢).
It follows from (1.G.5) and (1.F.9) that d and d* are equivalent provided
that a sequence {z,,} converges to a point z with respect to the d metric if
and only if it converges to x with respect to the d* metric. Suppose that
{zn} converges to x with respect to the metric d. Let « > 0. There is an
integer N,, such that d(tn,x) < €/2 whenever n > Nj. Since f is continuous,
the sequence {f(z,)} converges to f(x). Hence there is an integer N2 such
that p(f(2n), f(z)) < €/2 whenever n > No. Thus if n > max(N;, N2) we
have that rz, € S? (x). Conversely, if {z,} converges to z with respect to
the metric d*, then {z,,} converges to x with respect to the d metric, since
d(xn,2) <d*(rn,2) for each n.
To see f : (X,d*) > (Y,p) is uniformly continuous, let « > 0 be given
and set 6 = e. Then if d*(a,b) < 6, we have d(a,b) + p(f(a), f(b)) < 6 =e,
which of course implies p(f(a), f(b))<e. O
B. COMPLETE METRIC SPACES

Although compactness is convenient when dealing with convergence, less
stringent conditions are frequently sufficient to achieve adequate control over
sequences in metric spaces. Complete metric spaces possess enough struc-
ture to enable one to establish many important theorems that have wide
applications in both topology and analysis. For instance, we can use an
important fixed point result (problem B-6) that holds for complete metric
spaces, for proving theorems about the existence and uniqueness of solutions
to differential equations.
(3.B.1) Definition. Suppose (X,d) is a metric space. A sequence {z;} in

X is a Cauchy sequence if and only if for each € > 0 there is an integer N
such that d(%m,2%n) < € whenever m,n > N. The metric space (X,d) is
complete if and only if each Cauchy sequence {x;} in X converges to a point
Cale An
(3.B.2) Exercises.
(a) Give an example of a metric space that. is not complete.
(b) Show that completeness is not a topological property. [See the next
theorem.]
(c) Show that completeness is preserved by homeomorphisms that are
also isometries. [If (X,d) and (Y,p) are metric spaces, a function
f : X — ¥ is an isometry if and only if d(z,y) = p(f(z), f(y)) for
all z,y € X.]
Complete Metric Spaces 83
It is easy to show that if {x;} is a Cauchy sequence for which some

subsequence converges to a point z, then the entire sequence must also con-
verge to x. Thus it follows from (3.A.11) that compact metric spaces are
complete. However, there are no dearth of complete metric spaces that are
not compact, e.g., €! is complete as the next theorem shows.
(3.B.3) Theorem. The space (€!,d) is complete, where d is the usual

metric (d(z,y) = |x — y)).
Proof. Suppose {r,} is a Cauchy sequence in €!. Consider the set
A= {xy : n € Z*}. Since A is bounded (why?), A is compact (2.G.6). Now
apply (3.4.11) to obtain a convergent subsequence of {z,}. The remarks
preceding the theorem now tell us that {z,,} also converges. O
No doubt, you observed when working Exercise (3.B.2(b)) that €! and
the open interval (—1,1) are homeomorphic and that (—1,1) is not com-
plete. This rather distressing situation (after all, this text purports to study
topological invariants) is remedied with the introduction of the concept of
topological completeness. It is easy to see that the method used to turn the
metric property of completeness into a topological property can be applied
to any metric property.
(3.B.4) Definition. A metric space (X,d) is topologically complete if and

only if there is an equivalent metric p for X such that (X, p) is complete.
Of course, a complete metric space is automatically topologically com-
plete. A natural question to ask is whether all metric spaces are topologically
complete. Although the answer is no (10.E.5), it will be shown in Chapter
10 that any metric space may be embedded isometrically as a dense subset
of a complete metric space.
(3.B.5) Exercise. Show that a closed subset of a complete metric space is

complete.
(3.B.6) Exercise. Show that a complete subspace of a metric space is
closed.
We prove later that any locally compact separable metric space is topo-
logically complete (10.E.3). In fact, it can be shown that every locally com-
pact metric space is topologically complete (Dugundji [1966)).
(3.B.7) Exercise. Find a locally compact metric space that is not complete.
The following theorem gives an important property of complete metric

spaces.
(3.B.8) Theorem. Suppose A; > A2 D A3 D --: is a decreasing sequence
of nonempty closed subsets of a complete metric space (X,d) and the limit
of the diameters of the A; is 0. Then (\72, Ai = {x} for some x € X.
84 3. Metric Spaces
Proof. For each positive integer i select a point 2; € A;. Since the
diameters of the A;’s converge to 0 and the A;’s are nested, it is clear that
{x;} is a Cauchy sequence and hence converges to a point z € X. We
show that ()$2, Ai = {x}. If ¢ ¢ ()72, Ai, let A; be the set with smallest
index that fails to contain z. By (3.A.14), d(z,A;) =r > 0. However, this
is impossible, since it implies d(z,A;) > r for 1 > j, which prevents the
sequence from converging to x. Therefore, we have x € (3 A;. That z is
the only point in the intersection follows readily from the observation that
if y is any other point in X, then d(z,y) = s > 0; however, for sufficiently
large i, diam A;<s. O
(3.B.9) Corollary. Suppose (X,d) is a complete metric space and let
D,,D2,... be a sequence of open dense subsets of X. Then (}%<, D; is
dense in X.
Proof. Suppose U is a nonempty open set in X. It suffices to show
that UN (lex D;) # 9. Since D; is open and dense, D; NU must be
nonempty and open. Hence, there is a positive number €; < 1 and a point
xz, € X such that S.,(z1) C Dj NU. The open set S,,(r1) has nonempty
intersection with D2, and consequently there is a point x2 and a positive
number €2 < 1/2 such that S,,(%2) C D2 N S_,(1). Similarly, S,,(%2) N D3
is open and nonempty, and therefore there is an e3 < 1/3, etc. Repeated
application of this procedure leads to a decreasing sequence of closed sets
with diameters converging to 0. By the previous theorem, there is a point z
common to each S,,(x;). Hence, x is contained in both U and (2, Di. O
Spaces other than complete metric spaces may also have the property
described in the foregoing corollary; such spaces are known as Baire spaces.
(3.B.10) Definition. A topological space X is a Baire space if and only if

the intersection of each countable family of open dense sets in X is dense.
(3.B.11) Exercise. Show that if X is a locally compact T> space, then X

is a Baire space. [Hint: Use an argument similar to that given in the proof
of (3.B.9).]
The following theorem gives an important property of Baire spaces.
(3.B.12) Theorem. Suppose X is a Baire space and C1, C2,... is a count-

able closed cover of X. Then at least one of the C;’s contains a nonempty
open (in X) set.
Proof. This theorem follows easily from the DeMorgan rules. Since
X = Uj, Ci, it must be the case that (){2,(X \ C;) is empty. However,
X \ C; is open for each i. Hence, since X is a Baire space, at least one of the
sets, say X \ C; is not dense, which implies that X \ (X \C;) = C; contains
an open set.
(3.B.13) Definition. A subset A of a space X is nowhere dense in X if

and only if A contains no nonempty open subsets of X, i.e., int A = 0.
(3.B.14) Definition. A subset A of a space X is a set of the first category

in X if and only if A can be written as a countable union of sets nowhere
dense in X. The subset A is of the second category if and only if A is not of
the first category.
This use of the word “category” should not be confused with category
used in the context of functors and categories (see Chapter 12).
(3.B.15) Examples.
1. The rationals are of the first category in €!.
2. The irrationals are of the second category in €}.
If the irrationals could be written as a countable union of nowhere dense
sets, then with the addition of the individual rational points as nowhere dense
subsets, €! itself could be represented as a countable union of nowhere dense
subsets and would be of the first category (in itself). This contradicts the
following classical theorem.
(3.B.16) Theorem (Baire Category Theorem). Any complete metric
space is second category in itself. That is, a complete metric space can not
be written as a countable union of nowhere dense sets.
Proof. Suppose X is a complete metric space and X is of the first
category. Then X can be written as a countable union of nowhere dense
sets A,, Ag,. ... Since each A; is nowhere dense, X ‘ A; is both open and
dense. From the completeness of X, we have by (3.B.9) that ();,(X \ Ai) is
nonempty. But this contradicts (\f2,(X \ Ai) = X \ (Uf, 4i) = 0. Hence
it follows that X must be of the second category. O
(3.B.17) Definition. A point x of a space X is isolated if and only if x has
a neighborhood that contains no other point of X, i.e., {x} is open in X.
(3.B.18) Corollary. Every countable complete metric space has an isolated
point. O
Note that every point of a discrete topological space is isolated. At the
other end of the spectrum are the perfect spaces.
(3.B.19) Definition. A closed subset A of a topological space X is perfect

if and only if A is closed and each point of A is an accumulation point of A.
In other words, a set A is perfect if and only if A = A’
(3.B.20) Exercise. Suppose X and Y are topological spaces. Show that

X xY is perfect if X is perfect.
As an application of the Baire category theorem, we consider an inter-

esting example given by Knaster and Kuratowski [1921]. This space is of
86 3. Metric Spaces
importance in dimension theory, since it is a zero dimensional set that is not

totally disconnected. It is also an ingenious example of a subspace of the
plane having an explosion point.
Before constructing the example, we need first to investigate one of the
most unusual sets in all of mathematics, the Cantor set. To define the Cantor
set we begin by forming the following subsets of the closed interval [0, 1]:
A; = (0, 1/3] U [2/3, 1}; Az = (0, 1/9]U (2/9, 1/3]
U (2/3, 7/9] U[8/9;1};"A3=
(0, 1/27] U [2/27, 1/9]U [2/9, 7/27]U [8/27, 1/3]U [2/3, 19/27] U [20/27, 7/9]U
(8/9, 25/27]U[26/27, 1]; etc. In general, A;+1 is obtained from A; by removing
the middle third from each component of A;. The Cantor set K is defined by
K =)72, Ai. Clearly, endpoints of the intervals in the various A,’s belong
to K. It is equally obvious from the construction that for every point z € K,
there are points of K distinct from x that are arbitrarily closed to z. Thus,
K possesses no isolated points. Since K is closed (actually, K is compact) it
is complete (3.B.5). Hence, we have by (3.B.18) that K is uncountable, and
thus K contains points other than the end points already noted.
We would be derelict if we failed to mention that the Cantor set was
actually discovered independently by H.J.S.Smith in 1875 (Hawkins, [1970]).
Nevertheless, “Smith set” fails to have the lyric sound of a “Cantor set.”
(3.B.21) Exercise. Show that the Cantor set is totally disconnected.
(3.B.22) Exercise. Show that the Cantor set is homeomorphic to the set
of points in €1 of the form 5>°°, k;/3*, where k; is either 0 or 2.
We now construct the Knaster-Kuratowski example to which we previ-
ously alluded.
Consider the Cantor set K C I ( view J as a subset of the x axis in
E*) and set p = (1/2,1/2). For each c € K, let L(c) be the line segment
connecting p and c. Let E be the subset of K that consists of the endpoints of
the deleted intervals in the construction of K’, and let F = K \ E. Define sets
E, F, Xz, and Xp asfollows: E=U{L(c) :.cé€ E}; F=U{L(c): ce #'}s
Xp = {(2,y) € E : yis rational}; Xp = {(z,y) € F : yis irrational}.
Finally, set X = X_U Xp. This set is sometimes irreverently referred to as
Cantor’s leaky teepee.
We now sketch a proof showing that X is connected, even though intu-
itively X seems too sievelike to be in “one piece.”
y P
If X is not connected, then there are open sets U and V in €2 such that
X CUUV and XNUNV = 9. We assume thatp € U. For each t € (0, 1/2),
let Ky = {c € K : L(c) contains a boundary point (x,t) of U}. We claim
that for each ¢, Ky; is nowhere dense in K. Suppose not; then there are
points c1,c2 € K such that 0 4 (c1,c2)NK C Ky. Lete € EN (c1, C2).
Then the boundary point (z-, t) of U has an irrational y coordinate, i.e., t is
irrational. (If t were rational then (z,,t) would be in X and hence in either
U or V, both alternatives being impossible.) Now suppose t € F'N (ci, c2);
we conclude in a similar way that t is rational. Hence, K; must be nowhere
dense in K.
Let Ko = {c € K : L(c) has no boundary points in U}. Since p € U,
we have that for any c € Ko, L(c) is contained in U, and consequently,
Ko # K (otherwise, VM X would be empty). Thus, there are points c; and
co in K such that (c1,c2)N KN Ky = 9. Therefore, one may assume that
Ko = 9, since otherwise we could replace the space X with a “new” X that
sits over a part of the interval (c1,c2) (the “new” X is a subset of X that
is homeomorphic to X). With this minor point out of the way, we observe
that K = (U{K; : tis rational}) U E, (why?), or, in other words, K is a
countable union of nowhere dense sets, which contradicts (3.B.16). Hence,
X must be connected.
It is easy to establish that the removal of point p totally disconnects X.
Each component of X \ {p} must be contained in L(c) for some c (any subset
of X that contains points in two L(c)’s can be separated by a line running
between the L(c)’s, and through the missing point p). But clearly the points
of X lying in each L(c) are totally disconnected, because either points with
rational second coordinates or points with irrational second coordinates are
missing. Thus components consist of individual points.
As another vivid illustration of the Baire Category Theorem in action,
we prove the following rather delicate and unexpected result.
(3.B.23) Theorem. Suppose (X,d) is a connected, locally connected, com-
plete metric space. Then X cannot be written as a countable union of
nonempty pairwise disjoint closed subsets.
Proof. Suppose the contrary, X = 72, Ci where the C;’s are nonempty,
closed and pairwise disjoint. Let D = X \ (Uj2, int Ci) = Uf2, FrCi. Then
D is a closed subset of X, and hence, by (3.B.5) and (3.B.16), D is second
category (in itself).
Claim. If U is an open set in X such that UN FrC; # ) for some 7, then
UnN(D\FrC;) # 0.
To establish the claim, first observe that by hypothesis we may assume
U is connected. Since U intersects FrC;, we have that UM (X \ Cj) is
nonempty, and hence, there is a positive integer 7 #7 such that UNC; # 0.
88 3. Metric Spaces
If U fails to intersect FrC;, then C; NU = (int Cj) NU, which implies that
UC; is both open and closed (in U), contradicting the connectedness of
U. Therefore, UN Fr C; # 0, and the claim is proven.
Since D is second category, not all of the FrC; can be nowhere dense
(in D). By the connectedness of X, Fr C; # 0 whenever C; # 0. Thus, there
is an open set V in X and an integer j such that 9 AVNDC PEC; = bY C,.
Therefore, we have that VN FrC; # @ and VN (D \ FrC;) = 6, which
contradicts the claim. O
C. CONVEX AND HAUSDORFF METRICS

Convexity is a fairly strong condition that is possessed by some metrics. We
do not deal extensively with this concept, although it serves as a backdrop
for a significant amount of work in mathematics.
(3.C.1) Definition. A subset C of €” is convez if and only if each pair of

points a and b in C can be connected by a straight line segment that lies in
C, i.e., if a and b are points of C, then {ta+ (1-—t)b:0<t<1} CC,
where a = (a1, 42,...,@,) and ta = (ta, tag,..., tan).
The standard n-ball, B”, in E” is defined to be the set of points in E€”

whose distance from the origin is less than or equal to 1. It should be obvious
that B” is convex.
The following theorem will prove useful in later chapters.
(3.C.2) Theorem. Suppose A is a compact convex subset of €” with

nonempty interior. Then A is homeomorphic to B”.
Sketch of proof. Let z be an interior point of A, and choose « > 0
small enough so that the set B.(z) = {w € E” : d(w,z) < e€} is contained
in the interior of A. For each z € A \ {z}, let L(x) be the ray starting
at z that passes through z. We assume for the moment that L(r) N FrA
consists of a single point ¢. Define h(x) to be the unique point on L(z)
with the property that d(z,z)/d(z,£) = d(z,h(x))/e. A somewhat tedious
argument (and hence left to the reader) can be employed to show that h is a
homeomorphism between A and B,(z). Since B,(z) is homeomorphic to B”,
it follows that A and B” are homeomorphic. It remains to show L(z)N Fr A
consists of a single point for each z € A \ {z}.
Suppose ¢ and @ are distinct points in. L(x) MNFrA and that the order
of the points on L(x) is z,¢,#. Since Z € Fr A, there is a point w outside of
A and near @ (and not on L(2)), such that a line segment starting at ¢ and
ending inside of B,(z) passes through w. This contradicts the convexity of
A; hence, no such ¢ exists. O
Convex and Hausdorff Metrics 89
We can extend the notion of convexity to arbitrary metric spaces as

follows.
(3.C.3) Definition. Suppose (X, d) is a metric space and x,y € X. A point
m € X is a midpoint of x and y if and only if d(z,m) = d(m,y) = d(x, y).
The metric space X is said to be convez if and only if each pair of points has
at least one midpoint. The metric space X is strongly convez if and only if
each pair has a unique midpoint. The metric space X is without ramifications
if and only if no midpoint of x and y is also a midpoint of @ and y, where
Sia.
Consider the following subset A of the plane.
It should be apparent that the usual metric for €*, restricted to A, is

not convex. Nevertheless, little ingenuity is required to discover a convex
metric for A—simply define the distance between two points p and q to be
the (usual) distance traversed when one moves along A from p to q.
An obvious question: for what spaces do convex metrics exist?
A remarkable result due to Moise [1949] and Bing [1949] ensures the
existence of a convex metric for all compact, connected, locally connected
metric spaces. The reader now has sufficient background to follow the argu-
ments given in those papers.
The convex metric constructed in the Bing and Moise papers need not be
strongly convex. For example, readers should be able to convince themselves
that S! does not admit a strongly convex metric.
Strongly convex metric spaces without ramifications are less abundant.
In fact Rolfson [1970] showed that any compact 3-dimensional strongly con-
vex metric space without ramifications is homeomorphic to B?.
We now introduce an entirely different type of metric, one that is defined
on a set consisting of closed subsets of a given metric space. Earlier, we
saw how one can define the distance between two subsets of a metric space
(X,d) by setting d(A,B) = inf{d(z,y) : c € Aand y € B}. This distance
is quite useful in many contexts, but it has the disadvantage that distinct
but intersecting sets always have distance 0; hence, this particular distance
function does not lead to a metric for families of subsets of X. One way of
creating such a metric is described as follows.
(3.C.4) Definition. Suppose (X,d) is a metric space with finite diameter
and H is a collection of nonempty closed subsets of X. For C,D € H let
90 3. Metric Spaces
dg(D) = sup{d(z,C) : x € D}. Then p: H x H — (0,00) defined by

p(A, B) = max{d4(B),dp(A)} is the Hausdorff metric for H, and (H,p) is
called a hyperspace of X.
(3.C.5) Exercise. Show that p is a metric.

Note that if {x} € H for each x € X, then for points z,y € X, we have
p({x}, {y}) = d(x, y), and hence X is isometrically embedded in (H, p). This
observation justifies the name hyperspace. The problem of determining the
actual nature of (H, p) for a given collection of closed subsets of the metric
space (X,d) is of considerable interest. Let us examine what happens in a
relatively simple case.
Suppose X is the unit circle S! with the usual topology, and let H_ be the
collection of all closed connected subsets of X. We show that the hyperspace
(H,p) is homeomorphic to the unit ball B? = {(z,y) € E? : e*+y? <1}. A
homeomorphism h between the two spaces is given as follows. Define h(S')
to be the origin (0,0). If C is a proper closed connected subset of S?, let c
be its midpoint and let r = (length C)/27. Then h maps C’ onto the point
of B? lying on the radius passing through c, and at a distance of 1 — r from
the origin.
h(C)
There is no problem in verifying that h is a homeomorphism.

Other examples of this sort are given in the problem set. Schori and
West [1972] resolved a long-standing problem involving hyperspaces by show-
ing that if X = J and H is the collection of all closed subsets of X, then
the resulting hyperspace is the Hilbert cube. (The Hilbert cube is the space
obtained by taking the product of J with itself a countable number of times
and will be discussed in Chapter 7.) Amazingly, Schori went on to show in
subsequent papers that the hyperspace associated with any compact, con-
nected, locally connected metric space is also the Hilbert cube. The reader
is referred to Nadler [1978] for a systematic study of hyperspaces.
Problems O1
PROBLEMS
Section A
Show that a metric space is compact if and only if every infinite subset
has an accumulation point (cf. chapter 2, problem I-13).
Suppose X is a first countable space, {x;} is a sequence in X, and z is
a cluster point of {x;}. Show that z is a limit point of a subsequence of
{ai}:
A subset A of a metric space (X,d) is metrically isolated if and only if

d(A, B) > 0 for each closed set B contained in X \ A. Show that if A is
metrically isolated, then A is closed and has compact frontier.
Suppose that (X,d) and (Y,p) are metric spaces. Show that a function
f : X — Y is uniformly continuous if and only if for any two sequences
{Zn} and {yn} in X, n—limco d(Ln, Yn) = 0 implies n—-co
lim e(f (tn), fYn)) =
0.
Find an example to show that if (X,d) and (Y, p) are metric spaces and
f : X - Y is continuous, there need not be an equivalent metric for Y
under which f is uniformly continuous.
Suppose A and B are disjoint compact subsets of a metric space (X, d).
Show that there is an a € A and b € B such that d(A, B) = d(a,b).
Suppose that (X,d) and (Y,p) are metric spaces. Show that a function
f : X > Y is uniformly continuous if and only if whenever A and B are
nonempty subsets and d(A,B) = 0, then p(f(A), f(B)) = 0.
Suppose (X,d) and (Y, p) are metric spaces. Ifh : X — Y is a uniformly
continuous homeomorphism, must h~! be uniformly continuous?
A metrizable space (X,U) is topologically totally bounded if and only if
there is a metric d for X such that the topology induced by d coincides
with U and (X,d) is totally bounded. Show that a space (X,U) is
topologically totally bounded if and only if (X,U) is separable.
10. Determine which of the following are uniformly continuous as functions
from €! > é!.
(a) f(z) =2’,
(b) f(z) = Vial,
(c) f(z) =e".
bt. Is the sum (product) of two uniformly continuous functions (into &+)
uniformly continuous?
92 3. Metric Spaces
12. Find a subspace of €! and sets A and B that satisfy the hypothesis of
(3.4.14) and for which there is no point b € B with d(A,b) = d(A, B).
Section B
1. Suppose (X,d) and (Y,p) are metric spaces and f : X — Y is an

isometry. Prove or disprove:
(a) f is I-I.
(b) f is continuous.
(c) f is open. What if X = Y?
(d) f is onto. What if X =Y?
2. Let (X,d) be a compact metric space and suppose f : X — X is an
isometry. Show that f is onto.
3. Suppose that (X,d) is a metric space with the property that whenever
f :X — X is an isometry, then f is onto. Is X compact?
4. Suppose (X,d) is a metric space with the following two properties:
(i) for each « € X andr > 0, {y : d(z,y) < r} is compact.
(it) for each z,y € X, there is an isometry % : X — X such that
P(r) =y.
Show that every isometry from X into X is onto.
5. Show that in (3.B.8), (72, Ai may be empty if the limit of the diameters
of the A; is not 0.
6. Let r € (0,1) and suppose (X,d) is a complete metric space, S is a
closed subset of X, and f : S + S is a map with the property that
d( f(x), f(y)) < rd(a,y) for all z,y € S. (Such a map is called a con-
tractive map.) Show that f has a unique fixed point.
7. Suppose d is a metric for a space X with the property that every closed
set of finite diameter is compact. Show that (X,d) is a complete metric
space.
8. Suppose X is a metric space with the property that if S is any nonempty
closed subset of X and f : S 4 S is any contractive map (see problem
6), then f has a fixed point. Show that X is complete. [Hint: Let
{Zn} be a nonconvergent Cauchy sequence and for each x € X, let
L(x) = inf{d(z,r,) : tn # x}. Note that L(x) > 0. Choose r such
that 0 <r < 1. Define o : N > N inductively by setting o(0) = 0
and defining o(n) to be an integer > a(n — 1) such that d(a;,2;) <
rL(Xo(n—1) for all integers 7,7 > a(n). Let S = {to(n) : n = 0,1,2,...}
and let f : S — S be defined by f(24(n)) = Lo(n-1);]
9. Show that any subsequence of a Cauchy sequence is Cauchy.
10. Show that a metric space is compact if and only if it is complete in every
equivalent metric.
Problems 93
Eb: Show that a metric space is compact if and only if it is bounded in every
equivalent metric.
12: Show that every subset of a first category set is first category.
13. Prove or disprove: every uncountable space is second category (in itself).
14. Show that if A is a nowhere dense subset of a topological space X, then
X \ A is dense in X. Is the converse true?
15. Suppose X and Y are metric spaces and that A C X and BC Y. Show
that if A x B is nowhere dense in X x Y, then either A is nowhere dense
in X or B is nowhere dense in Y.
16. Suppose (X,d) is a metric space and that A is a countable subset of X.
Show that A is of the first category in X if and only if A has no isolated
point in X.
17: Prove or disprove: every open subset of a second category space is of
the second category.
18. Show that the countable union of first category sets is of the first cate-
gory.
19. Show that every countably infinite complete metric space contains an
infinite number of isolated points.
20. The Cantor set may be used as follows to describe a map f : I > €! that
is open but not continuous. On each component C, of the complement
of the Cantor set in J, let fa : Cag > €! be any strictly increasing
continuous function that has 0 in its image. On the Cantor set itself,
define f to be identically 0. Show that f is open but not continuous.
a1, Show that if U is an open subset of a space X, then FrU is nowhere
dense.
22 Prove the following ‘converse’ of (3.B.8). Suppose (X,d) is a metric
space and the intersection of each decreasing sequence of nonempty
closed balls whose radii converge to zero is nonempty. Then (X,d)
is complete.
2o- Another subset of the plane having an explosion point may be con-
structed as follows. As before, start with the “cone” over the Cantor
set. Let S denote the family of all half-closed segments each connecting
a point of the Cantor set with p = (1/2,1/2) but excluding p, and let
@ : [0,2] + S be 1-1 and onto. Let A be the family of all compact
subsets of €% that intersect an uncountable number of members of S
(but miss p). Let w : [0,2] + A be 1-1 and onto (why is this possible’).
For each a € (0,2), let d(a) = Sq and ~(a) = Ag. Let ao be the
smallest element of [0,2) such that Ap N Sa, # @, and select a point
Ca, in this intersection. Let a; be the smallest element in [0,) (and
94 3. Metric Spaces
# ag) such that Ai N Sq, # @. Select a point ca, in the intersection.

Continuing in this fashion, show that for each 6 € [0,9], exactly one
point cg is selected in Sg. Let X = {p} U {eg : 6 € [0,)}. Show that
X is connected, but X \ {p} is totally disconnected.
24. Show that an uncountable second countable space can be written as the
union of a perfect set and a countable set and that these two sets may
be chosen to be disjoint.
25. Show that a subset A of a complete metric space is countably compact
if and only if A is closed and totally bounded.
26. Show that the Knaster-Kuratowski example has the fixed point property
(the fixed point need not be p).
Ze Show that if A is a subset of a separable metric space, then A has at
most a countable number of isolated points.
28. Find a subspace of the French Railroad metric space that is complete
and subsets A and B satisfying the hypothesis of (3.4.14) in which there
is no point b € B such that d(A,B) = d(A,b).
Section C
Show that the conclusion of (3.C.2) need not hold if A has empty inte-
rior.
Suppose p,q,r € (X,d). Then q is said to be between p and r if and only
if d(p,r) = d(p,q) + d(q,r). A subset A C (X,d) is linear if and only if
there is an isometry (into) ¥: A > &}.
(a) Suppose p,g,r € X. Show that {p,q,r} is linear if and only if one
of the points p, q, or r is between the other two.
(b) Give an example of a metric space consisting of four points that is
not linear but for which every proper subset is linear.
An arc a contained in (X,d) is a segment if and only if Ima is linear.
Suppose (X,d) is a strongly convex metric space without ramifications.
Let pg and pr be segments in X such that pq pr \ {p} 4 9. Show that
either pg C pr or pr C pg. [Here, pq denotes the segment with endpoints
p and q.]
If (X,d) is strongly convex and p,q,r € X, show that q is between p
and r if and only if q € pr. [See problem 3 for notation.]
Suppose Uj, U2,...,Un are open subsets ‘of a topological space X, and
define (Ui,...,Un) ={AC X : ACU, Uj and ANU; # O for each ¢}.
Show that sets of this form determine a basis for a topology for the set
V={ACX : Ais closed and A # 9}. This topology is called the
Vietoris topology.
Problems 95
6. Show that for a compact metric space X, the Vietoris topology (see
problem 5) and the topology generated by the Hausdorff metric coincide
on the set H = {C : C compact in X}.
Suppose f : X — Y is continuous, where X and Y are metric spaces
and Y is also compact and connected. Let Hx = {A Cc X;AF
Q and is connected and compact}, and suppose Hy is defined similarly.
For each A € Hx, let f(A) = {f(a) : a € A}. Show that:
(a) f : Hx > Hy defined by A 4 f(A) is a well defined function
(called the induced map).
(b) f is continuous
(c) f need not be onto even if f is onto.
The function f is said to be weakly confluent if and only if for each
B € Hy there is a component A of f~!(B) such that f(A) = B.
(d) Assume that f is onto. Then f is onto if and only if f is weakly
confluent.
Let X = IJ, and let H be the family of all subsets of J that consist of one
or two points. Show that H with the Hausdorff metric is homeomorphic
to a (solid) triangle.
Let X = I and let H be the family of all subsets of J that consist of
one, two, or three points. Show that H with the Hausdorff metric is
homeomorphic to a tetrahedron.
10. Let X be a metric space and H be the family of all compact subsets
of X with the Hausdorff metric p. Suppose that C),C2,C3 € H and
Ci; C C2. Show that p(C3, C3 U C;) = p(Cs, C2).
Lb; Let X = I and H be the family of all closed connected subsets of X

with the Hausdorff metric p. Show that (H,p) is homeomorphic to a
(solid) triangle.
_ he chee comers te wok
—s wan (AN eelag eae pad yal

i, Sn Hue Cy Kg tien hee cteniedt aca pity
ile Ud potin: ses = aie oY mabece ae Yo er 3)‘o .
a ee oo on ve ee ee, ee
oll = Vttaas Aart Plclekdenegmms
« of medi eT 3
idee st.) siete Wola)
tet” odes ai iaanaiiciegole
7 ond eh 402) ceniie hyp "eahe Adi Sees
cnetWHFilet
eA AeMetemetse
tte lenny Shr AYE sf baat =he - ;
sideniar- ewes el sretharThobansH
af) date WS tady wode oa
eee C lignes? naa “
hx elvan sari %o
ef fsBo. eit Pt otbaw) =Kid,
SH od) Cette? VPS sale ahhh Sh AS NR
id a 01
wethoedec omnosntad
SP tin tors, cg Lee & ell bbe) edad tae ot a Kod
bie Kehyth CHD & i aoe aoe wdetol ober xtatiy A! Yaw)
eS heyy iy oe 0, Sade rode 25 ~
A> Gt 1ary = ual Cllet4 Ginid ti\ot Foams eet eRe one
au? tapes FH, Ay cdl Palmee pore loteegh af ati :
~ 7
_ >
atecMRBRDAPTS&
2 -
0 ont Py Che
a le Get
(A.o2(
ent —Siang
aft 1 Oh Shae tres
~ > @© < Bb Stteecf

_ ag ee |
ie eed sp 6, ae
2 > by 2 @ ter ekg 1;
2 ‘ . Se ORS Ne Ue et
=
‘i i'tegy Ocede Ss
=
Chapter 4
SEPARATION PROPERTIES
In this chapter we examine a variety of spaces that have some, but in general
not all, of the attributes of metric spaces. Normal spaces may lack a distance
function, but they nevertheless have sufficient topological structure to yield
some of topology’s major theorems. We also introduce a number of other
types of spaces that are of interest precisely because they lack some of the
properties inherent to metric spaces.
A. NORMAL SPACES
We have previously seen that in €*, the distance between closed subsets A
and B may be zero even if A and B are disjoint.
UsAU:
- +,
V=BU'.’.’.
Nevertheless, note as illustrated above that in the case shown, we can find
disjoint open sets U and V that serve to keep A and B apart. In fact, in
metric spaces, all that is needed for such a “separation” to exist is that the
closure of each of the sets does not intersect the other. This is the content
of the next theorem.
o7
98 4 Separation Properties
(4.A.1) Theorem. If A and B are subsets of a metric space X such that

ANB =BnA=4Q, then there are disjoint open sets U and V of X such
that ACU and BCYV.
Proof. For each a € A, there is a positive number r, such that the open
ball S,,(a) C X \ B. Cover A with sets of the form S,,/2(a) and B with
similarly defined open sets of the form S,,/2(b). Let U be the union of sets
of the former type and V be the union of sets of the latter form. It is easy
to verify that U and V are the desired open sets. O
The property of metric spaces described in (4.4.1) turns out to be very

useful, and, as is frequently the case with useful concepts, it is made the
basis for a definition.
(4.A.2) Definition. A topological space X is completely normal if and only

if for every pair of subsets, A and B, with AN B = AN B = J, there are
disjoint open sets U and V that contain A and B, respectively.
A slightly weaker, but no less important notion is the following, which

the reader will recognize as a sort of Hausdorff property for closed sets.
(4.A.3) Definition. A topological space X is normal if and only if for each

pair of disjoint closed subsets A and B there is a pair of disjoint open sets
U and V such that ACU and BCYV.
Observe that points need not be closed sets in an arbitrary topological
space, so normality need not imply T>. Note also that if X is a T; space that
is not T>, then X fails to be normal.
Clearly, any completely normal space is normal; however, the converse

does not hold. An example, the Tihonov Plank, will be given in section C.
(4.A.4) Exercise. Show that a space X is normal if and only if for each
closed subset A and for each open set U with A C U, there is an open set V
such that ACV CV CU.
(4.4.5) Exercise. Show that a space X is completely normal if and only
if every subspace of X is normal. [Hint: Consider X \ (AN B).]
Metric spaces are, of course, normal. However, in view of the Tihonov
plank cited above, there are normal spaces that are not metrizable. A number
of further examples may be derived from the following result, whose proof is
an almost classic illustration of the use of compactness.
(4.4.6) Theorem. If X is a compact Hausdorff space, then X is normal.

Proof. Suppose A and B are disjoint closed subsets of X. Then A and B
are compact. For each pair of points a and b with a € A and b € B there are
disjoint open sets U(q,») and V(q,») that contain a and b respectively. Fix a
Normal Spaces 99
point a € A. The family of open sets {V(.4) : b € B} forms an open cover of

B from which a finite subcover AVeasiy V(a,b2)>+-+» Via,b,)}may be extracted.
Let U, be the intersection of the corresponding sets Ua Ca bzyoeae ae
G.ba)
and set Vomatlile 5Via,b;). Repeat this procedure for each a € A to obtain
an open cover {U, : a € A} of A. Let {Ug,,...,Ua,,} be a finite subcover.
For each U,; there corresponds an open set Vo, that contains B. Then
U = Uji, Ua, and V = ()2, Va, are disjoint open sets that contain A and
B, respectively. O
We use the next exercise in the proof of Theorem (4.4.8) below.

(4.A.7) Exercise. Show that if {z;} is a sequence in a compact space X,
then {z;} has a cluster point in X (compare with the remarks following
(2.1.7)).
(4.A.8) Theorem. If A; D5 A: 3D -:: is a countable family of closed,
connected, nonempty subsets of a compact Hausdorff space X, then the
intersection A = (rs A; is compact, connected, and nonempty.
Proof. Since A is closed it is compact. Now suppose A C U UV, where
U and V are disjoint open sets. By (2.H.3), there is an integer N such that if
i> N, then A; C UUV. Since each A; is connected, it follows that A; C U
or A; C V for i > N, which in turn implies that A C U or AC V. Thus A
is connected.
That A # @ is immediate from (2.H.2). O
(4.A.9) Exercise. Generalize the previous theorem as follows. Let A be
an index set that is partially ordered by <. Suppose further that for each
a, 8 € A, there is a \ € A such that a X \ and 8 X X (such a set is said to
be directed). Suppose that {Ag : a € A} is a family of closed, connected,
nonempty subsets of a compact Tz space X and that A, C Ag if and only if
6 <a. Show that (),-, Aa is compact, connected and nonempty.
Suppose A and B are disjoint closed subsets of a space X. Then, if X

is normal, there are disjoint open subsets U and V such that A C U and
B CV. Can the open sets U and V be chosen so that X = U UV? To do
so obviously requires that X be disconnected. A sufficient condition for the
existence of such a dramatic cleavage is given in our next result. It should
not seem too surprising that some powerful tool—Kuratowski’s Lemma—is
used to bring this about.
(4.A.10) Definition. Disjoint subsets A and B of a topological space X
are separated in X if and only if there is a separation (U,V) of X such that
Ac U and BC VY.
The reader should note that a disconnected (as a subspace) pair of
subsets of a connected space X can never be “separated in X.”
(4.A.11) Theorem. Suppose X is a compact T2 topological space and A,

B are disjoint closed subsets of X. Then, A and B can be separated in X if
and only if no connected subset of X intersects both A and B.
Proof. The implication > is clear. For the converse we first consider
the case where A and B are singletons, i.e, A = {x} and B = {y}. For
the purpose of seeking a contradiction, suppose that {x} and {y} cannot be
separated in X. Let K = {Kg : a € A} be the collection of all closed subsets
of X in which {x} and {y} are not separated. Note that X € K. Partially
order K by defining K. x Kg if and only if Ky C Kg. By the Kuratowski
lemma (0.D.4), there is a maximal nest K* = {Kala € L C A} in K. Let
K =(\ger Ka. We first show that K € K (and hence by the maximality of
K*, we will have that K € K*). If K ¢ K, there are disjoint closed subsets
K, and Ky of K, such that + € Ki, y € Ko, and Ki UK2 = K. Since K
itself is closed in the normal space X, so must be Ky and K2. Thus, there
are disjoint open sets U and V in X with K,; C U and K2 C V. By part
a of (2.H.4), some member Kg of K* is contained in U UV. However, this
is impossible, since it implies that {x} and {y} can be separated in Kg by
Un Kg and VN Kg. Consequently, we have K € K.
Since z and y lie in K, it follows from the hypothesis that K cannot be
connected. Therefore, K can be written as the union of disjoint nonempty
closed subsets H,; and Hg, and since K is “minimal” in K*, both x and y
must be contained in either H; or in H2, say H,. On one hand, if {x} and
{y} could be separated in Hy by sets Hj and H3, then Hj and H}UH>2 would
separate {x} and {y} in K, which is impossible by the hypothesis. On the
other hand, if {x} and {y} cannot be separated in H,, then H; belongs to
the maximal chain K*, which is a contradiction since H, is a proper subset
of K =(),¢, Ka. Hence we have reached a contradiction so the special case
of the theorem is proven.
Now suppose A = {2}, and B is a closed set disjoint from A, where A
and B satisfy the hypothesis of the theorem. For each b € B, the special
case just proven is applied to obtain a separation (U,,V,) of X such that
z €U,andb€ Vy. The family {V, : 6 € B} is an open cover of the compact
space B and consequently yields a finite subcover {V;,,V,,,..., Vs, }. Then
U = ()j_, Us, and V = Uj_, Vs, form a separation of X where x € U and
Bev.
By employing the method used in the preceding paragraph, the reader
should have no difficulty in establishing the general case. O
A particularly elegant application of the above theorem is found in the

proof of the following rather innocent appearing, but nevertheless deep result.
(4.4.12) Theorem. If U is an open subset of a compact connected T>

space X and C is a component of U, then CX NFrU # 9. (C% is the closure
OfC ness)
Uryson’s Lemma; Tietze’s Extension Theorem 101
__ Proof. Suppose C* N FrU = @. Since CX C U = UU Fr(U) implies

CX ¢c U, and since C is a component of U, we have C* = C, ie, C is
closed in X. The space X is normal so there is an open set V such that
CCVCVCU. Since V is compact and since no connected subset of V
intersects both C and FrV (recall that C is a component of U), it follows
from (4.4.11) that there is a separation (F,G) of V such that C C F and
FrV CG. Then (F,GU(X \V)) is a separation of the connected space X,
which is impossible. O
B. URYSON’S LEMMA; TIETZE’S EXTENSION THEOREM

We now prove two of the great classical theorems in general topology, Uryson’s
lemma and Tietze’s extension theorem. Suppose X is a space with the prop-
erty that for each pair of disjoint closed subsets A and B, there is a continuous
function fag : X — [0,1] such that fag(A) = 0 and fag(B) = 1. Then X
is normal since f4,({0,1/3)) and f4,((2/3, 1]) are disjoint open subsets of
X that contain A and B, respectively. Uryson’s remarkable lemma gives us
the following converse.
(4.B.1) Theorem (Uryson’s Lemma). A topological space X is normal
if and only if for each pair A and B of disjoint closed subsets of X, there is
a continuous function f : X — [0,1] such that f(a) = 0 for each a € A and
f(b) =1 for each b € B.
Proof. We temporarily delay the proof in order to obtain the following
crucial lemma.
(4.B.2) Lemma. Let D be the set of all dyadic fractions in (0,1), ie., let
D = {p/2” : n,p € Z* and p < 2”}. If A and B are disjoint closed subsets
of a normal space X, then there is a collection of open sets {Ua : d € D}
such that d, < dz implies A C Ug, C Ua, C Ua, CUa, CX \ B.
Proof. Use (4.4.4) to obtain an open set Uj, so that the inclusions
AcE Uyjo°C Oj j2 © X \ B hold. Another two applications of this exercise
yield open sets U;/4 and U3/4 with the properties that
AC Ui /4 CUy/4 CU; /2 CU 4/2 a U3/4 CUs/4 CUB:

The inductive step should now be apparent. Suppose the sets Uy (satisfying
the lemma) have been defined for all membersof D with denominators less
than or equal to 2*. Once again, by (4.4.4), there are open sets U;_1/2k+1,
j =1,2,...,2*, with the property
AC Uj /(ae+1) C U1 (241) SU pa ee ee
The Uq’s are now defined for all d’s of the form p/2**1, and this completes
the induction, and the proof of the lemma. O
Proof of (4.B.1). Let {Ua : d € D} be the collection of open sets

obtained in the preceding lemma. Define a function f : X — [0,1] by
(2) eee if z ¢ Ug for some

d€ D,
t)=
0 otherwise.
Note that if € A, then x € Ug for each d € D and hence f(z) = 0;

furthermore, if z € B, then x ¢ Ug for all d € D and therefore f(x) = 1. To
complete the proof it suffices to establish the continuity of f.
Suppose z € X ande > 0. Assume first that 0 < f(x) < 1. Since the
dyadic rationals are dense in (0,1), there are numbers d;,dz € D such that
f(z) —-e < dy < f(z) < d& < f(z) +e. Let V = Ug, \Ua,. Then V is
an open set and V contains z, since f(x) > d, implies that x ¢ Ua, and
f(x) < dz implies that x € Ug,. We assert that f(V) is contained in the
é-neighborhood about f(x). To see this, let y € V. Then since y € Ug,,
f(y) must be at most equal to dz and hence strictly less than f(r) +e. Since
y € Ua,, f(y) must be greater than or equal to d; and thus strictly greater
than f(z) — €. Consequently, f is continuous at x whenever f(z) lies in the
interval (0,1). A similar proof may be supplied by the reader for the cases
fey = Vand fier a
The following corollary will be useful when we prove Tietze’s extension
theorem.
(4.B.3) Corollary. Suppose A is a closed subset of a normal space X and

f : A — €' is a continuous function such that |f(a)| < c for each a € A.
Then there is a continuous function h : X —> €! such that
(i) |h(x)| < (1/3)e for each x € X, and
(it) |f(a) — h(a)| < (2/3)c for each a € A.
Proof. Consider the following sets: Ay ='{a € A : f(a) > (1/3)c} and
A_={aeA: f(a) < —(1/3)c}. Since Ay and A_ are closed and disjoint,
there is a map h : X > [—(1/3)c, (1/3)c] C €' with h(A;) = (1/3)c and
h(A_) = —(1/3)c (Why?). A routine verification shows that h has all the
requisite properties. O
Uryson’s Lemma; Tietze's Extension Theorem 103
Uryson’s lemma does not guarantee that f-1(0) = A and f-1(1) = B

To force this outcome, an additional condition is imposed on the sets A and
ee
(4.B.4) Definition. A subset A of a space X is a G5 set of X if and only
if A may be expressed as the countable intersection of open subsets of X.
(G, respectively, 6 is from the German word Gebiet=“region,” respectively,
Durchschnitt = “intersection.” )
Note that closed subsets, as well as open subsets, of a metric space
(X,d) are G5 sets. (If A C X is closed, then A = ()*°, S1/,(A) where
Sijn(A) = {ze X : d(x,A) <1/n}.)
(4.B.5) Definition. A normal T; space X is perfectly normal if and only if
each closed subset of X is a Gs; set.
(4.B.6) Exercise. Suppose X is perfectly normal, and A and B are disjoint

pir sets. Then there is a continuous function f : X — [0,1] such that
—1(0)= A and f~'(1) = B. [Hint: Let A = (\?2., Un, where U,, is open
in
fo and Uni C Un. For each n, find Jn: ss
— [0,1] such that f,(A) = 0
and f,(.X.\Un)-= ‘1s Define.fa(a =eise . Define a function fg in a
similar manner and set f = fa/(fa+ My ]
(4.B.7) Definition. A subset A of a topological space X is an F, set of X
if and only if A'can be expressed as the countable union of closed subsets of
X. (F, respectively a, is from the French word fermé= “closed,” respectively,
summe = “union.” )
Note that a subset A of a topological space X is a G5 set if and only if
X \ Ais an F;, set.
A great number of problems in mathematics can be reduced to the
question: If A is a subset of a topological space X and f : A > Y is
a continuous function into a topological space Y, can f be extended to a
continuous map with domain X? Uryson’s lemma is one (rather trivial)
solution to an extension problem. A more interesting solution to a different
extension problem is given in the next theorem. For a discussion of some
of the problems that are equivalent to the extension problem see Hu [1959,
pp.1-34].
(4.B.8) Theorem (Tietze’s Extension Theorem). Suppose X is a nor-
mal space, A is a closed subset of X, and f : A > €' is a continuous function.
Then there is a continuous map F : X > &' such that F(a) = f(a) for each
aé A. Furthermore,
(i) if |f(a)| < c for each a € A, then F may be chosen so that |F(x)| < ¢
for each x € X, and
(ii) if |f(a)| < c for each a € A, then F may be chosen so that |F'(x)| < c
for each z € X. }
Proof. The proof is based on repeated applications of (4.B.3). Three
cases are considered.
Case 1. Suppose |f(a)| < c for each a € A. By (4.B.3), there is a map
go : X — €} such that |f(a) — go(a)| < (2/3)c for each a € A and |go(z)| <
(1/3)c, for each x € X. Application of (4.B.3) to f — go yields a function
gi : X — €, where |gi(z)| < (1/3)(2/3)c for z € X and |f(a) — go(a) —
gi(a)| < (2/3)(2/3)c for a € A. Continue inductively and obtain a sequence
of functions go, 91, 92,... such that for each n € N, |gn(x)| < (1/3)(2/3)"c on
X and |f(a) — go(a) —--: — gn(a)| < (2/3)(2/3)"c on A. Define F: X + E}
by
F(a\= a Oil z).
For each n, let sn(x) = )7j.5 9i(z). Note that {s,} converges uniformly
to F(z) since |S 4 gn(x)| < 07-9 (1/3)(2/3)"c = c. Hence, there is an
integer N such that for n > N, |sp(x) — F(x)| < € for each x € X. We give
a common advanced calculus argument to show that F is continuous. Let
zo € X be given. Choose N so |s,(xz) — F(x)| < €/3 when n > N. Since
Sn(x) is continuous, there is a neighborhood U of zo such that if x € U,
then |sy(z) — sn(o)| < €/3. Hence |F(x) — F(ao)| < |F(x) — sn(z)| +
|s~(2) — Sn(Zo)| + |Sn~ (20) — F'(zo)| < € whenever x € U, which establishes
the continuity of F. That F is an extension of f and |F(x)| < c for each
xz € X may be readily verified by the reader.
Case 2. Suppose |f(a)| < c for a € A. By Case 1 there is a continuous
extension F' of f such that |F(z)| <c. Let B = {x : |F(x)| = c}. Note that
B is closed and disjoint from A. Apply (4.B.1) to obtain a function h: X >
E' such that h(A) = 1 and h(B) = 0. The desired extension G of f is defined
by pointwise multiplying the functions F and h, i.e., G(x) = h(x) - F(z).
Case 3. Suppose f is not bounded. Take any homeomorphism h :
€} - (-1,1). By Case 2, the map hf : A — (—1,1) may be extended to
G: xX - (-1,1); then F = h“!G extends f to allof X. O
Euclidean 1-space €' is not the only range space for which the foregoing
theorem is valid. In fact, quite a number of spaces may be used in place of
€'. Such spaces are called Absolute Retracts.
(4.B.9) Definition. A normal space Y is an Absolute Retract (AR) if and

only if for each normal space X, each closed subspace A of X, and each
continuous map f : A + Y, there is a continuous function F : X + Y such
that F(a) = f(a) for each a € A.
This use of the word “retract” will make more sense shortly, but first
let us prove a basic result concerning AR’s.
(4.B.10) Theorem. If the spaces Yi, Y2,...,Yp_ are AR’s, and if the product
space |];_, Yi is normal, then the product space is an AR.
Proof. Suppose X is a normal space and A is a closed subset of X. Let f
be a continuous map from A into Flzas Y;. For each i, the map p;f : A > Y;
may be extended to a map F; : X + Yj. Define F : X > []/_,Y; by
Fry = (F(a) Pols)... crac): DO
(4.B.11) Corollary. Euclidean space €” isan AR. O
(4.B.12) Exercise. Show that J x J x--- x J is an AR.
(4.B.13) Exercise. Show that the property of being an AR is a topological
invariant.
(4.B.14) Definition. A subset A of a topological space X is a retract of X

if and only if there is a continuous function r : X — A such that r(a) =a
for each a € A (i.e., the identity map on A has a continuous extension r). In
this case, r is called a retraction or a retraction mapping.
(4.B.15) Exercise. (a) Show that a subset A of a topological space X is a

retract of X if and only if the following diagram commutes, where 7 is the
inclusion map, and all maps are continuous.
X
A id A
(b) Show that a subset A of X is a retract of X if and only if, for each
topological space Z, each continuous map f : A + Z can be extended to a
continuous function from X to Z.
(4.B.16) Examples.
1. Any point in a space X is a retract of X.
2. Any closed interval of €1 is a retract of €'.
3. If x € €” and € > 0, then the closed ball S.(x) in €” is a retract of
Gon
(4.B.17) Exercise. Show that if X has the fixed point property and A is
a retract of X, then A has the fixed point property.
(4.B.18) Definition. A normal space X is an absolute retract (ar) if and

only if whenever X is homeomorphic to a closed subset B of a normal space
Y, then B is a retract of Y. That is, you can view X as being a retract of

any normal space that it ‘lives in’ as a closed set.
(4.B.19) Theorem. A space X is an AR if and only if X is an ar.

Proof. Suppose X is an AR. Let B be a closed subset of a normal space
Y, and suppose B and X are homeomorphic. We seek a retraction r from Y
onto B. Let h: B > X be a homeomorphism.
Since X is an AR, h may be extended to a continuous map H: Y > X,

where H|p = h. Then r = h~'H is the desired retraction. The converse is
considerably more difficult, and its proof is deferred until Chapter 8. O
The concept of Absolute Neighborhood Retract, AN R, is closely related

to AR.
(4.B.20) Definition. A normal space X is called an Absolute Neighborhood

Retract (ANR) if and only if for each normal space Y, for each closed A C Y,
and for each continuous map f : A > X, there is a neighborhood U of A
and a continuous extension F': U + X such that F|, = f.
Although AR’s are obviously ANR’s, there are numerous examples of

ANR’s that are not AR’s. Perhaps the most prominent of these is the Eu-
clidean n=sphere'S” = (@i,.0.,@agi) eo ay Sh oe ny
(4.B.21) Example. The n-sphere S” is an ANR. Suppose Y is normal
and A is a closed subset of Y. Let f : A + S” be a continuous function.
Since S” c E"*!, f may be regarded as a map from A into the AR, €"+!.
Consequently, there is a continuous extension of f, F : Y 4 E"+1. Set
U = F-1(€"+!\ {0}) and on U define a function F by F(y) = F(y)/||F(y)I|,
where ||F'(y)|| denotes the usual distance between F(y) and the origin. Then
F is the desired extension of f to the neighborhood U of A. That S” is not
an AR follows from the (n + 1)-dimensional Brouwer fixed point theorem
(14.C.22); see problem B-9.
We will use the following result in later sections.

(4.B.22) Lemma (Brouwer Reduction Theorem). Suppose X is a

second countable space and C = {Cy : a € A} is a nonempty family of
closed subsets of X with the property that whenever C; D Cy D C3 D-::
is a countable decreasing family of members of C, then ();2, Ci € C. Then
there is an irreducible set C* in C, i.e., C* € C but no proper subset of C*
belongs to C.
Proof. Let U;, U2,... be a countable basis for X. Let C; be an arbitrary
member of C and inductively select a sequence of sets from C by choosing
C,\€ C such that C, C Chi (X \U,,). (If no such set in C. exists, then
define C, = Cn-1.)
We assert that C* = (7°, C; is irreducible. If this is not the case, then
there is a proper subset C of C* that belongs to C. Let x € C* \ C. Since
this latter set is C*-open there is a basis member U,, such that x € U,, and
COU, = 0. Now since C € C we have C C ON UR) GCrain(
Xan U;),
and hence C, C Cn_-1M(X\Un). But, C, C (X\U;n) which says CpNUn = 0.
On the other hand # 4 C*NU, = Un N(NC,) C UnNCy. That is impossible
since ynCr#e2. 0
The fact that S” is an ANR plays an important role in the proof of the
following extension result, which we will use in Chapter 15.
(4.B.23) Theorem. Suppose (X,d) is a compact metric space, C is a

closed subset of X, and f is a continuous map from C into S” that cannot
be extended to all of X. Then there is a closed subset C* of X such that f
cannot be extended continuously to C* UC but can be extended continuously
to C UD, where D is any proper closed subset of C™.
Proof. Let D = {D : D is closed and f has no continuous extension to
CUD}. We show that D has an irreducible member. For that purpose
suppose D, > D2 D --: is any countable decreasing family of elements of
D, and let D = (2, D;. If D ¢ D, then there is a continuous extension
F : (CUD) > S” of f. Since S” is an ANR, F can be extended continuously
to an open neighborhood U containing CUD. However by (2.H.3), for some
integer i, we have D; C U, contradicting the fact that f cannot be extended
continuously to C U D;. So D € D and by the Brouwer reduction theorem
(4.B.22) there is an irreducible element C* in D; clearly, C* is the required
subset of X. O
The notion of ANR is also frequently defined in a metric space context.
(4.B.24) Definition. A metric space (X,d) is a metric absolute neighbor-

hood retract, denoted by ANRy, if and only if for each closed subspace B
of a metric space Y and each continuous function f : B — X, there is a
neighborhood U of B and a continuous extension Ff’: U + X of f.
(4.B.25) Theorem. Suppose X is an ANRm, Y is a metric space and

h:X -Y is an embedding such that h(X) is closed in Y. Then there is a
neighborhood U of h(X) and a retraction r: U > h(X).
Proof. Consider h~! : h(X) > X. Since X is an AN Ry, there is a
neighborhood U of h(X) and an extension g: U + X of h'. Then hog is
the desired retraction. O
(4.B.26) Remark. The converse of (4.B.25) holds if X is compact (10.C.3),
but its proof requires material from Chapters 8 and 10.
The following rather technical lemma is needed in the tedious proof of
(4.B.28).
(4.B.27) Lemma. Suppose X; and X2 are closed subspaces of a topological
space X = X,; U X2. Let A be a subset of X and for 7 = 1,2 let U; be an
relatively open subset of X; that satisfies AN X; C U;. Then U; UU? is a
neighborhood (in X) of A.
Proof. There are open sets V; and V2 in X such that U; = V, NX; and
Uz = V2N Xq. Note that Vin (X1 \ X2) CUT Aen (Xo \ X1) C U2, and
VinvVs = VY, NVAN(Xy UX) Gs U, UU 2. Hence, we have A = (AN(Xy \X2))U
(AN(X2\X1))U(ANXINX2) es (Vin(X1\X2))UVan(X2\X1))UM V2) EG

U; UU». Since Vi 11 (X1 \ Xo), Vo A (Xe \ X1), and Vi N V2 are open subsets
of X,sois their union. O
(4.B.28) Theorem [Borsuk, 1932]. Suppose X, and X2 are closed sub-

spaces of a metric space (X,d) such that X = X,;U X>2. Suppose further X,,
X2, and X;M X2 are ANRyy’s. Then for each embedding h of X as a closed
subset of a metric space Y, there is a neighborhood V of h(X) and a retract
R:V > h(X).
Proof. Let Y be a metric space and let h : X — Y be an embedding
such that h(X) is closed. By (4.B.25), h(X; M X2) is contained in a y-
open set U, and there is a retraction r : U + h(X,M X2). Since metric
spaces are completely normal, there are open sets U, and Up in Y such
that h(X1) \ h(X2) C Ui, h(X2) \ h(X1) C U2, and 1, NU, = 0. Let
U, = U;\h(X2) and Us a U2\h(X). Then we have that h(X1)\h(X2) ec Ui,
h(X2) \ h(X1) G U2, U, NU, = 0, and (U; U U2) M (h(X1) M h(X2)) = 0. For
i = 1,2 define sets P; = [U \(U; UU2)] UA(X;) and functions r; : P; > h(X;)
by
fe: ifz €U \ (U; UU2)
Clearly r; is continuous.
| Now let Ey = (YUU; UU2) \ U2 and Ey = (UUU,; UU2) \ U;. Then for
i = 1,2 we have P; = [(U\(U, UU2)]UA(X;) = (E;\(U; UU2)) UA(X;). Hence
Further Results Concerning Normal Spaces 109
P; is a closed subset of E;. By (4.B.24), there are E;-open subsets V; such

that P; C V; C E;, and there are extensions 7; : V; > h(X;) of r; for i = 1, 2.
We have that h(X, UX2) CUUU,; UUz = EB, VU E,, Ey = (Fy U E>) \ U2,
and BE, = (£, UE) \U;, where the E;’s are closed in E, U Eo. Furthermore,
h(X1UX2)NE; = h(X;) C P; Cc Vj. It follows from (4.B.27) that V = VjUVe
is a neighborhood of h(X, U X2) in E, U Ey = U UU, UUs, and since this
latter set is open, V is a neighborhood of h(X, U X2) in Y. Note now that
U\(U,UU2) C PNP, C WAV, C Ey NE2 = ((U\U2)\U1)N((U\U1) UU) =
U \ (U; UU2), and hence Vj NV2 = E, N Ez = U \ (U; UU2). Since the E;’s
are closed in E, U Ep, we have that V; N (Va WivVeys. Vi. U (Vi NV2) Cc
V, U(£, N Ey) = Vi and similarly for V2. Since the V;’s are closed in V, it
follows from the definition of the r;’s that the function R: V > h(X, U X2)
defined by
fTi(z) iffreVy
Ri(z).=
Po (x) ifx€ Vo
is continuous, and a retraction of V onto h(X; UX2). O
(4.B.29) Remark. Once we have (10.C.3), it will follow that (4.B.28) may
be restated so that if X is a compact metric space, then X = X; U XQ is an
ANR»m.
C. FURTHER RESULTS CONCERNING NORMAL SPACES
(4.C.1) Definition. If A= {Aq : a € A} is a cover of a topological space

X, then an open shrinkage of A is an open cover {Vy : a € A} of X such
that V, C Aq for each a € A.
Certain open covers of normal spaces are amenable to being shrunk.

The next theorem is a remarkable result that is quite useful in a number
of distinct contexts. Its proof is dependent on the axiom of choice, which
appears in its well-ordering guise.
(4.C.2) Theorem. Suppose U = {U, : a € A} is an open cover of a

normal space X such that each point in X lies in at most a finite number
of the U, (such a cover is called point finite). Then there is an open cover
Vv ={V. : a € A} that is an open shrinkage of U.
Proof. We assume that A has been well ordered with first element ao,
second element a, etc. We use transfinite induction to construct a family
{V, : a € A} of open sets such that
(i) for each a € A, Vly, and
(it) for each a € A, the family {Vy : \ < a} together with the family
{U5 21d 2a} cover, X.
We then show that the family {V, : a € A} is an open cover of X.
For purposes of motivation we observe that X \U,5.,, Ua is closed and
is contained in U,,. Hence by (4.4.4), there is an open set Va, such that
COON se Ca) Varo Vay C Uay- In order to replace Ua, by a similar
construction, we make use of the fact that {Va,,Ua,,.--} is an open cover
Of
Now assume for each 3 < a, we have constructed an open set Vg such
that Vg C Ug and that {Vg : 8 < a} and {U, : y > a} form a cover of X.
We show that this permits the construction of an open set V, such that
(i) Va C Ug, and
(ii) the families {Vg : @ < a} and {U, : y > a} form a cover of X.
Note that X \ (U{Vg : B < a}UU{U, : y > a}) C U, and X normal
implies that there is an open set V, such that
X (ls = Bo} UM, o> ol NEV AGV ols.
Then the combined collections {Vg : 8B < a} and {U, : 7 < a} cover
X, and this completes the inductive step. To see that the family {Vg :
B € A} covers X, suppose z € X and let U,,,U,,,...,U,,, be the sets
of UY that contain x. Assume that 7, > yj for 1 = 1,2,...,n —1. Then
{Ver si«vieig Vere «3 9| yn Unagas
oe} covers: X 5.whichsimplies thater y@ Maal
Vee, a Cl
A topological property is said to be hereditary if and only if it is retained

by each subspace. For example, any subspace of a metrizable space is still
metrizable, subspaces of Hausdorff spaces are Hausdorff, and subspaces of
first countable spaces are first countable. As we see next, normality is not
hereditary.
Ordinal numbers are useful in the construction of several important

examples in topology. Certain basic features of ordinals are given in Chapter
0, and a more thorough introduction may be found in Monk [1969]. We
denote the first infinite ordinal by w and the first uncountable ordinal by
Q. Let X = [0,w] = {a : ais an ordinal and 0 < a < wh, and let Y =
[(0,Q] = {a : ais an ordinal and 0 < a < 1}. Each of these sets is given
the order topology. The reader may verify (see Exercise (4.C.3) below) that
both X and Y are compact T> spaces. Consequently, X x Y with the product
topology (called the Tihonov Plank is a compact Hausdorff space and hence
The Separation Axioms 111
is normal by (4.A.6).
1 2 3 4 w
We will exhibit a subspace of X x Y that is not normal. (This, incidentally,

implies that X x Y is not completely normal (4.A.5) and therefore not metriz-
able (4.A.1).) For a simpler example, see problem C-5. The desired subspace
is obtained by removing the point (w, 1). Hence let W = (X x Y)\ {(w, )}.
Then A= {(a,2) : 0< a < w} and B= {(w,8) : 0 < B < 9} are dis-
joint closed subsets of W. We show that it is impossible to enclose A and
B in disjoint ‘open subsets. Suppose U is any open set containing A. For
each n € X \ {w}, let B, be the least element of Y such that (n,B) € U
whenever ( > £,. Since the collection S = {G, : n € X \ {w}} is countable,
the least upper bound so of S will be strictly less than 2 (0.E.4). However,
{(n,B) : B > so, n < w} is a subset of U. Now suppose V is an open set
containing B. Choose an ordinal # such that so < 6 < 92. Then of course,
(w, 3) € B, and it should now be clear that the points “close to” (w, 3) must
lie in both U and V; hence W is not normal.
(4.C.3) Exercise. Show that [0,w] and [0,9] are compact. [Hint: Consider
A = {a € [0,w] : [0,a) is compact } and show that if A = sup A, then
A€ A, and A =w.]
D. THE SEPARATION AXIOMS

A number of properties other than normality have been given sufficient at-
tention to warrant mention. We examine a few of these.
(4.D.1) Definition. A topological space X is regular if and only if whenever

z € X and F is aclosed subset of X not containing z, there are disjoint open
sets U and V such that x € U and F CV.
(4.D.2) Exercise. Show that a space X is regular if and only if for each
point « € X and for each open set U containing x, there is an open set V
such that rE V CV CU.
Note the similarity between definition (4.D.1) and the definition for
normality. In fact, it should be clear that normal T; spaces are regular.
Requiring that points be closed is obviously not a particularly strong
imposition on a topological space, but a still weaker condition defines an
even lower form of topological life. A space X is To if and only if whenever
x,y € X, an open set may be found that either contains x but not y or
contains y but not z; there is no guarantee, however, as to whether the open
set will contain specifically x or y.
A hierarchy of conditions (with increasingly restrictive structure) may
be set up as follows:
To space—defined above.
T, space—defined in Chapter 1.
T2 space—defined in Chapter 1.
T3 space—a regular T) space.
T4 space—a normal T) space.
Ts space—a completely normal T} space.
Metric spaces.
The T;’s are frequently referred to as separation axioms; basically, they
reflect the degree to which points or sets may be kept apart by open sets.
Strangely enough, an inordinate amount of mathematical energy has been
expended in defining separation conditions that lie somewhere between the
ones listed above (15, 22, etc.); however, with a few notable exceptions, this
sort of exercise would appear to be about as significant as it is interesting.
(4.D.3) Exercise. Show that metric > T; > T, > T3 > T7 > T; > To.
In the remainder of this chapter we see that none of the above implica-
tions is reversible.
(4.D.4) Examples.
1. Any set (consisting of more than one point) with the indiscrete topol-
ogy fails to be Tp.
2. The set R! with the open ray topology is Ty but not 7}.
3. The set R' with the finite complement topology is T; but not Th.
4. Let X = R'. For each point x € R! and each €!-open interval U
containing z, set U = {y € U : y is rational} U {z}. The family of all such
sets form a basis for a topology that is Tz but not regular.
5. (Niemytzki’s Tangent disk, or bubble topology). Define sets X and L
by X = {(z,y) € R* : y > 0} and L = {(z,y) € R? : y = 0}.To forma
topology for X let B be the set of all U C X such that either
The Separation Axioms 113
(t) UC X \ Land U is a member of the usual topology for €?, or

(ii) U is of the form {x}UD where z € L and D is an open disk tangent
to L at the point z.
Then B is a basis for a topology for X that is T; but not T,. We use the Baire
Category Theorem to show that X is not normal. Let A = {(z,0) € L :
zx is rational} and B = {(z,0) € L : z is irrational}. Certainly A and B are
disjoint closed subsets of X (any subset of L is closed in X). Suppose that
U and V are open subsets of X containing A and B, respectively. We show
that U and V must intersect. For each (z,0) € B, let Dz be a disk of radius
rz that lies entirely in V and is tangent to L at (z,0). Forn = 1,2,..., set
Wr = {(x,0) € B : rz >1/n}. Then L = AU(U?2, W,,). Since L (with the
usual topology for £1) is a complete metric space, it follows from (3.B.16)
that some W ,,contains an open interval. However, since every open interval
in L contains a rational point, U and V must intersect.
(4.D.5) Exercise. Show that the space X in Example 5 is regular, T;, and
separable. Show that L with the relative topology is not separable.
Under certain conditions regular spaces are normal.
(4.D.6) Theorem. A regular, Lindelof space X is normal.

Proof. Suppose A and B are disjoint closed subsets of X. By regularity,
for each a € A, there is an open set U, containing a such that U,NB = and
for each b € B, there is an open set V, containing b, whose closure misses A.
Clearly, we have A c U{U. : a€ A} and BC U{Y : b € B}. Since both
A and B are Lindeldf (closed subspaces of a Lindeléf space are Lindeldf),
countable subcovers {U;, U2,...} and {Vi, V2,...} can be found for A and B
respectively. Now we separate the U’s from the V’s. a
Define W, = U; and Z; = Vi \ Wy. Now let Wz = U2 \ Z; and
Zz = V2 \ (W, UW2). In general, set W, = Un \ (Z1U-+--UZp_1) and
Zn = Vn \ (W,
U---UW,,). It should be apparent that W = U72, Wi and
Z =U, Zi are disjoint and open. O
(4.D.7) Exercise. Let X be any uncountable set and let x* € X. Define a
subset U ofX to be open if and only if (a) X \U is countable, or (b) «* ¢ U.
Show that X with this topology is T;, but that X is not first countable and,
hence, is not metrizable.
We mention one other type of space that has special interest in analysis.
Its definition is reminiscent of Uryson’s lemma.
(4.D.8) Definition. A T, space is a Tihonov space (or T31) if and only
if for each closed subset C' of X and for each point x € X \ C, there is a
continuous function f : X > I such that f(z) = 0 and f(C) =1.
If the requirement that X be T; is removed in (4.D.8), then X is said
to be completely regular.
(4.D.9) Exercise. Show that Ty > T31 => T3.

We will presently describe a T31 space that is not T4. For a T3 space
that fails to be T31, the reader may consult Steen and Seebach [1970] or
attempt to devise a more digestible example.
(4.D.10) Exercise. Suppose X and Y are topological spaces. Show that
X x Y has property A if and only if both X and Y have property A, where
A ranges over the following: (1) To; (2) T1; (3) To; (4) Ts; (5) T31.
(4.D.11) Exercise. Show that the properties listed in (4.D.10) are heredi-
tary.
Observe that in both of the preceding exercises, we came to a halt before

T, was reached. One may well argue that “normal” is a misnomer from the
standpoint that normality is neither hereditary nor does it behave decently
under product formation. To illustrate the latter point, we consider two
elegantly simple, but rewarding examples.
(4.D.12) Example (Half-Open Interval Topology). Let X = R' and let the
topology for X be determined by a basis consisting of all half-open intervals
of the form [a, b) where a < b.
We show that X is 75. It is clear that X is T;. Suppose that A and B
are subsets of X such that ANB = @ and BNA = 9. For each z € A there is
a half-open interval [x, e,) contained entirely in X \ B, and similarly for each
y € B there is a half-open interval [y,e,) C X \ A. It is readily seen that the
sets U = U{[z,ec) : x € A} and V = U{[y,e,) : y € B} are disjoint open
sets containing A and B, respectively.
(4.D.13) Exercise. Show that R! with the half-open interval topology is

first countable, separable, and Lindelof.
(4.D.14) Exercise. Show that R! with the half-open interval topology is

not second countable, and hence by (4.D.13)’and (1.G.11), is not metrizable.
The next example is even more curious.
(4.D.15) Example (Sorgenfrey’s Half-Open Square Topology). With X

defined to be the topological space described in the previous example, let
Problems 115
Y = X x X (with the product topology). A typical basis element for Y is

illustrated below. Since X was seen to be Ts, it follows from (4.D.3), (4.D.9)
and (4.D.10) that Y is Ts.
(4.D.16) Exercise. Let Y be the space defined in the previous example

and let L C Y be the diagonal line {(z,y) : y = —x}. Show that any subset
of L is closed, and use an argument similar to that which was employed in
connection with the bubble topology to demonstrate that Y is not normal.
Consequently, Y is a T.31 Space that is not Ty.
PROBLEMS
Section A
1. (a) Show that if X is a normal space, Y is a topological space and

f :X - Y is continuous, closed and onto, then Y is normal.
(b) Is normality a continuous invariant?
(c) Suppose that Xj, X2,..., Xn are topological spaces. Show that if
[[j_, Xi is normal, then X; is normal for each i.
2. Suppose C;,C2,... are closed subsets of a normal space X such that
jor cachik. = 15.2;220pC pn (UlG, ring ky) = (). Show that there are
disjoint open sets U;,U2,... such that C;, C Un for each n.
3. Show that a topological space X is normal if and only if for each pair of
disjoint closed subsets A and B there are open sets U and V containing
A and B respectively such that UNV = 0.
4. Suppose X is a 0-dimensional second countable space and A and B
are closed disjoint subsets of X. Show that there is a separation of X,
(U,V), such that ACU and BCV.
5. Suppose (X,d) is a metric space. The metric d is said to be normal if and
only if whenever A and B are closed disjoint subsets, then d(A, B) > 0.
Show that if d is normal, then (X, d) is complete.
6. Suppose that (X,d) is a metric space. Show that d is normal (see

problem 5) if and only if every continuous function f : (X,d) — (Y,p)
is uniformly continuous, where (Y, p) is a metric space.
7. Show that if a topological space X is the union of a finite number of
closed, normal subspaces, then X is normal.
Section B
1. Show that subspaces of perfectly normal spaces are perfectly normal.
2. Show that [0,1) is an AR.
3. Prove Uryson’s lemma using Tietze’s extension theorem.
4. Suppose X is a normal space and A and B are disjoint closed subsets
of X. Show that there is a continuous map f : X — [a,b] such that
f(A) =azand fils) =.
5. Let S = {0,1}. Show that a space X is path connected if and only if
each continuous map f : S + X can be extended to (0, 1].
6. Suppose (X,d) is a metric space. Show that (X,d) is perfectly normal,
i.e., fill in the details for the remarks above (4.B.5).
Show that a retract of a T> space is closed.
Show that if a (finite) product X of ANR’s is normal, then X as an
ANR.
9. Assume the (n + 1)-dimensional Brouwer fixed point theorem. Use
(4.B.17) to show that S” is not an AR.
10. Find a closed subset of a normal space that is not a Gs set.
11. Show that a pseudometric space is perfectly normal.
12. Suppose X and Y are topological spaces, A is a retract of X and B is
a retract of Y. Show that A x B is a retract of X x Y.
13. Suppose X is T, and an AR. If A C X is a retract of X then A is an
AR.
14. Show that an open normal subset of an ANR is an ANR.
15. Do the following closed subspaces of the plane €? have the fixed point
property?
16. Does there exist a countably infinite connected normal space that is T;?
17. Use Tietze’s Theorem to give an elegant proof of problem 7 from Section
A.
Problems ilaly/
Section C
Prove a converse of (4.C.2): if every point finite open cover has an open
shrinkage, then X is normal.
Show that closed subsets of a normal space are normal.
Show that a space X is normal if and only if for each finite open cover
{U,,U2,...,Un} ofX there is an open cover of X, {V, V2,..., Vn}, with
the property that V; C U; fori = 1,2,...,n. Avoid using the axiom of
choice in your proof.
Suppose X is a normal space and Aj, A2,..., A, are closed subsets of
X such that ()i_, Ai = 0. Show that there are open sets Vi, V2,...,Vn
such that A; C V; and ()i_, Vi = 0.
Let X, be any uncountable set and X2 be any infinite set. Suppose
that X, and X>2 have discrete topologies and let Xf = X1 U {c0;} and
XZ = Xz U {2} be the corresponding one-point compactifications.
Show that Xf x X} is normal, but (Xf x X3) \ {(co1, o02)} is not.
Section D
Show that a regular second countable space is completely normal.

Show that.a normal space is completely regular if and only if it is regular.
Suppose X is a space such that each z € X has a neighborhood V with
the property that V is regular. Show that X is regular.
Show that there is no countable connected T31 space containing more
than one point.
Suppose X = Zt. For each a,b € Zt which are relatively prime, define
Bia,b) = {an+b:ne€ (Ae (6)by
(a) Show that the collection of all such Bia) form a basis for a T>
topology U for Zt.
(b) Show that if U = Bia») and V = By.,a), then ac is an accumulation
point of both U and V.
(c) Show that (Z*,/) is connected.
Suppose X is a first countable space. Show that X is T> if and only if
every convergent sequence has a unique limit.
Show that every F, subset of a compact space is Lindelof.
bo (a) Show that locally compact T; spaces are regular.
(b) Show that locally compact Tp spaces are regular. [Hint: Show that
a locally compact Tp space is T>.]
9. Suppose X is a completely regular space, A is a compact subset of X,

and B is a closed subset of X disjoint from A. Show that there is a
continuous function f : X — [0,1] for which f(A) = 0 and f(B) =
10. Show that F, subsets of a T, space are T4.
11. Suppose X is an infinite T> space. Show that X has an infinite discrete
subspace.
12. Suppose X is a T3 space and f is a continuous, open and closed function
from X onto Y. Show that Y is To.
13. A space X is an Uryson space if and only if whenever z,y € X and
x # y, there are open subsets U and V of X such that x EU, ye V,
and UM V = @. Show that T3 spaces are Uryson spaces and find an
Uryson space that is not regular.
14. Show that perfectly normal spaces are completely normal.
15. Is Example (4.D.12) perfectly normal?
16. Show that every T, space (X,U/) can be embedded in a separable T,
space. [Hint: Let Y = X UZ* (disjoint union) and define a topology V
for
Y byV= {0} U{UUA:U EU, ACZ*, and Z* \ A is finite}.]
17. A topological space X is T:12 if and only if for each pair of disjoint
compact subspaces A and B of X, there are open sets U and V such
that ACU, B*e Vand AnV = 0 and DiiU= Gr Showsthat
[i Ty2 okie
18. Suppose X is a set. A uniform structure for X is a nonempty family U

of subsets of X x X (relations in X) such that:
(i) for each U EU, id, CU,
(ii) IFU €U then U1 EY,
(iit) if U € U, then there isa V €U such that VoV CU,
(iv) if U,V EU, then UNV EU, and
(v) fUEUandUCVCX xX, then V EU.
(Inverses and compositions of relations are defined in an analogous manner
to that of functions).
The uniform topology Ty of X determined by the uniform structure U/
is: Ty = {W C X : foreach z € W, there isaU € U with U(x) Cc W},
where U(x) = {z € X : (x,z) € U}.
(a) Show that if X is a set with a uniform structure UW, then Ty is a
topology.
(b) Find uniform structures that yield the usual topology for R', the
discrete topology, and the indiscrete topology.
19. Show that if // is a uniform structure (see preceding problem) for a set X,
then the topological space (X, Tz) is completely regular. [Hint: Suppose
Problems 119
that z € X and that C is a closed subset of X not containing z. Then

there is a sequence {U, : n € Zt} of elements of Y/ such that U, = Uz,
and U, o Un C Un-1. Let Vo = idy. For each dyadic rational r € (0, 1]
expressed in the form ye 2,** where 0 < ny < ng < ->- < ny, let
Vp = Uny 0 Uny_, 0° 70Ug, OU. li r < 8s, then V, C V, (first show
that Un © Vin2-» C Vim41)2-"). Define ¢: X — [0,1] by
sup{r: 2 ¢V,(z)} ify #2,

o(y) =
0 if y=,
and show that ¢ is the desired function.]

20. Suppose X and Y are topological spaces and that f : X + Y. The
function f satisfies property * if and only if for each open cover U of Y,
{int f-!(U) : U € U} is an open cover of X.
(a) Show every continuous function g: X — Y satisfies property *.
(b) Show that if Y is T,, then a function g : X — Y is continuous
whenever g satisfies property *.
(c) Show that property * is not equivalent to continuity.
- f A / a a - f |
} . - - .
, (nN. ae Dat
pag :
> on 7 mney oe ee ae =—
ameeithin
grr :
te rsanianioniny hen theelis raniahadlaed ;
mi
‘5
ye
? ‘wa “a peony
ee oe
7 “bee At
ve
ojieee fie
Paeoer a ct
Weriltionrat
tm 16) weit a wid a-ak Pe: 7
coe Sst | 0. ebaiented
ngtv a eereiynt
ne 7
ss > > eabbpis joeihoviiap> One tho Srnceeesg Site deca Th
wach <0 gar 8 @.T,; Tad cai Cte eniieee Oe i
a shes “© tte 22 open aoe ll en T mb ©
: D2 Fer av © eos PN Oe ae
) 2 eet ere) 7
a
’ (ta ee)
Ss th ae
= = Gia, gtruaee fd
th Vite h.
et if Tow a 4
yi # -
ad Vina D: is
pp! Dae Sore eeeJF

25 smal), (Gach! egy
Chapter 5
1-MANIFOLDS AND
SOME PLANE THEOREMS
We begin this chapter with a typical, though relatively easy classification

theorem, (5.4.3). Next, we show that the plane cannot be retracted onto
S', (5.B.6). The bulk of the chapter is devoted to a sketch of an “elementary”
proof for the Jordan curve theorem, the Schénflies theorem, and the annulus
theorem in the plane. (Elementary does not mean simple, or easy; it just
means that the concepts involved are not very sophisticated.)
A. 1-MANIFOLDS
We defined an n-manifold M in chapter 1 as a separable metric space with
the property that each point of M is contained in an open set homeomor-
phic to €". Our goal in this section is to classify connected 1-manifolds
(both of them). In Chapter 16, we perform the somewhat more complicated
task of classifying compact 2-manifolds. Curiously, it may be shown that a
reasonable classification of 4-manifolds is impossible (Markov [1960]).
(5.A.1) Definition. A triode is a space homeomorphic to
{(t,y)
€ E? : toe peu ee 7 r=, (sey I,
The next exercise gives an obvious characteristic of 1-manifolds.
(5.A.2) Exercise. If M is a 1-manifold, then M does not contain a triode.

(5.4.3) Theorem. If M is a connected 1-manifold, then M is homeomor-
phic to either €1, or S! = {(z,y) € E? : a? + y? = 1}
Proof. We consider two cases.
Case 1. Suppose that M is compact. Then there is a finite open cover
U, of M each of whose members is homeomorphic to the open interval (0, 1).
121
122 5. 1-Manifolds and Some Plane Theorems
We induct on the number of members in the cover. If the cover U contains

just two sets (Why can it not consist of just one?) U, and Ug, then it is
not difficult to see from the above exercise and the compactness of M that
the intersection of U; and U2 consists of precisely two components C; and
C2, which are open (2.E.2) and homeomorphic to (0,1). Let c; and cz be
points in C, and C2 respectively, and map the arc from c; to c2 lying in UV;
onto the upper hemisphere of S! and the arc from c; and c2 lying in U2 onto
the lower hemisphere of S!, where end points are matched up in the obvious
fashion. This yields the desired homeomorphism.
Suppose now that the theorem is true if the cover U described above
consists of n > 2 sets, and let U,,...,Un41 be a cover of X by sets home-
omorphic to (0,1). We may assume that no U, is contained in the union
of the rest of the U;. Pick any pair of the U;’s that intersect, say U; and
U;. The reader should be able to see (think triode) that U; NU; must be
connected and homeomorphic to (0,1). Hence U; JU; may be mapped home-
omorphically onto (0,1). Let U; = U; UU; and replace U; and U; by Us in
the covering set. Since this reduces the number of members of the cover,
induction may be applied to finish the proof.
Case 2. Suppose that M is not compact. Let D = {dj,do,...} bea
countable dense subset of M. Since M is arcwise connected (2.C.12), there
is a homeomorphism from [0,1] onto an arc A; with initial point d; and
terminal point dz (actually, A; is the image of an arc). Let d,, be the first
d; in the sequence dj,d2,... that does not lie in A,;. Since no triodes are
permitted in M, an arc from d,, to A; must meet either d; or dz before
intersecting any point in the “interior” of A,. If the arc intersects d; first,
map [—1,0] homeomorphically onto the arc Az running between d,, and d;
where 0 is matched up with d; and —1 with d,,. On the other hand, if d2
is the first point struck, then map [1,2] homeomorphically onto the arc A»
connecting d,, and dz, where 1 is sent to dz and 2 is mapped to d,,.
Let d,, be the first member of the sequence d,,d2,... that is not con-
tained in A; U Ay. An arc Az is sent out from d,, in search of A; U Ag.
Again, it first meets an end point of A; U Ag, and depending on which one,
the appropriate interval of €' is mapped onto A3 following the pattern used
previously for Ag. In this fashion, a homeomorphism h may be constructed
from €' into M that at least covers the dense subset D (Why is the domain
all of €'?). It remains to show that h is onto. Suppose that h is not onto,
and let x be a point not found in the range of h. Let U be an open set con-
taining z that is homeomorphic with (0,1). Select points d and d in D that
lie on “opposite sides” of x (this is possible, since U is essentially (0,1)). The
On the Contractibility of S? 123
homeomorphism h yields an arc from d to d that misses x. Complete this

arc with another one from d to d that lies in U to form a space, containing
a subspace S, homeomorphic to S! and passing through x. However, this is
impossible, since S cannot be all of M (M is not compact) and points not
in S may be connected to S only by creating a triode in M. Consequently,
h must be onto, completing the classification of connected 1-manifolds. O
B. ON THE CONTRACTIBILITY OF S'
In this section, we establish in an elementary manner the “obvious” fact that

the plane cannot be retracted onto S!, (5.B.6). That such a retract cannot
exist is at least intuitively clear, since its existence would apparently indicate
that a hole could be torn in €? by a continuous function. The proof that this
cannot happen is most elegantly done in an algebraic setting (see Chapter
12); but in the spirit of this chapter we follow a longer but more naive path.
The notion of homotopy is exceedingly important in topology and will
be examined more carefully in Chapter 12.
(5.B.1) Definition. Suppose f and g are continuous functions from a

space X into a space Y. We say f and g are homotopic if and only if there
is a continuous function H : X x I > Y such that H(z,0) = f(x) and
H(z,1) = g(x) for every x € X. For eacht € I, define H;: X — Y by
Ayp(e)r= H (a;
t).
A homotopy may be thought of intuitively as a continuous deformation

from one function to another.
In the following theorem, S! is considered to be a subset of the complex

plane C, where S! = {z € C : |z| = 1}. We follow the presentation of
Brown [1974].
(5.B.2) Theorem. Suppose f : S! + S! is homotopic to a constant map.

Then there is a continuous function ¢ : S! + € such that f(x) = e'#() for
all x € S!.
Proof. We suppose H : S! x I + S! is a homotopy, where H(z,0) =c
and H(x,1) = f(x) for each x € S!. Since S! is compact, H is uniformly
continuous, and there is a 6 > 0 such that |H (x,t) — H(z, u)| < 2 whenever
|t — ul < d and ze S!. Let 0 = to < th < te < ++» < tn-1 < tn =1 bea
partition of J such that |t;41 —t;| < 6. Note that Ho = cis of the form e’¥(),
where wp : S! — €! is a constant map, and that |H;,(z) — Ho(zx)| < 2 for all
z. We show that H;,(z) is of the form e’?(*). Since |H:,(x) — Ho(z)| < 2,
we have H;, (x) 4 —Ho(a) and hence H;, (x)/Ho(x) # —1. Define a function
\:S! > €&} by setting A(x) equal to the number of radians between 1 and
Hi, (x) /Ho(2) if Hz, (2)/Ho(z) is on or above the z axis, and, to the negative
of that number if H:,(2)/Ho(zx) is below the x axis. Then, H:, (x)/Ho(z) =
e\(2), and consequently H;,(x) = Ho(x)e) = e(¥(2)+A)) = etdrl(z)|
where ¢;(z) is defined to be equal to w(x) + A(x). The same procedure may
be used to show that H;,(r) = e*?2(*), H,,(x) = e*3(*), and eventually that
f(x) = W(x) =e), oO
(5.B.3) Definition. A topological space X is contractible if and only if

there is a homotopy H : X x I> X and a point cin X such that for each
xz € X, H(z,0) = x and H(z, 1) =c. In other words, a space is contractible if
and only if the identity map is homotopic to a constant map. The homotopy
HZ is said to be a contraction.
(5.B.4) Example. Any convex subspace A of €” is contractible. Let p € A

and define H: Ax I> A by H(z,t) =tp+(1—t)z.
(5.B.5) Theorem. The unit circle S! is not contractible.
Proof. Suppose S! is contractible. Then the identity map is homotopic

to a constant function and hence, by the previous theorem, there is a function
o: S! -» €! such that x = e'*(*) for each x € S!. Hence ¢ is one to one,
and in particular, ¢(z) # ¢(—x). Define g: S' + {—1,1} by
Then g is continuous. Furthermore, g is onto {—1,1} since g(—x) = —g(z).

This however contradicts the connectedness of S! (2.4.12). 0
The Jordan Curve Theorem 125
(5.B.6) Corollary. There is no retraction from €? onto S!.

Proof. Suppose r : E€? - S! is a retraction. Let p = (0,0) and define a
homotopy H : S'xI + €? by H(z,t) = tp+(1—-t)x. ThenrH :S'xI > S!
is a contraction, which contradicts (5.B.5). O
As another corollary of the preceding theorem we obtain a slightly
stronger version of the 2-dimensional Brouwer fixed point theorem.
(5.B.7) Corollary. Let B? = {(z,y) € E€? : x? + y? < 1} be the unit disk,

and suppose f : B? — €? is a continuous map such that f(S!) C B?. Then
f has a fixed point.
Proof. First define a retraction r : €? \ {(0,0)} > S! by r(x) = 2/||z|

(\|z|| = 2? +23, where x = (21, 22)). If f(x) 4 a for all c € B?, then S!
may be contracted by the homotopy
feeb bao 0<t<}

Hiei =
r((2—2t)x — f((2-2t)x)) Z<t<l
which contradicts (5.B.5). O
(5.B.8) Exercise. Show that the homotopy H in the preceding corollary is

well defined and, in fact, yields a contradiction.
(5.B.9) Exercise. Show that contractibility is a topological invariant.
C. THE JORDAN CURVE THEOREM

A homeomorphic image of S! is called a simple closed curve. The Jordan
curve theorem, one of the most celebrated classical theorems in topology,
asserts that the complement of a simple closed curve S in the plane consists
of two components, each of which has S as its frontier. At first glance, it
seems that this assertion is obvious; nevertheless, as we shall see, the proof
is quite involved. In fact, Jordan himself gave an invalid proof. The first
generally accepted demonstration was presented by Veblen [1905]. If one
uses tools from algebraic topology, a relatively simple and elegant proof is
possible which generalizes to higher dimensions. However, in order to gain
experience in working with manifolds, we have chosen to follow a longer, but
more geometric, path. Most of the proofs have been left as exercises. To
help convince the skeptical reader that there is more afoot than might have
been anticipated, we include a relatively simple closed curve for meditation.
Does it have an outside or an inside?
Initially we consider a special type of simple closed curve, one that con-
sists entirely of vertical and horizontal line segments. These will be referred
to as rectilinear simple closed curves. The first theorem is a simple exercise
in analytic geometry.
(5.C.1) Theorem. The line L, defined by x = 0, separates the plane into

exactly two components and is the frontier of each.
Sketch of Proof. First we show €?\L consists of at most two components:
Given any three points (21,41), (2, y2), and (23,y3) with z; 4 0, at least
two of the points must have either both first coordinates positive or both
negative. Then the line segment joining them misses L, and hence they lie
in the same component.
Secondly, we show that the sets A = {(z,y) € €? : x < 0} and B=
{(x,y) € E* : « > 0} are distinct components: Let a = (—1,0) and b = (1,0)
be points in A and B, respectively. Suppose f : I > E? is an arbitrary
path from a to b. The intermediate value theorem on p; f (where p; is the
projection onto the first coordinate) shows that there is a point c € J such
that f(c) € L. It follows that every path from a to b intersects L. That
is, A and B are distinct path components, and hence, by (2.D.8), they are
distinct components of €? \ L.
Finally, we show that L is the frontier of A and B: Clearly, any open

set containing a point of L intersects both A and B, and, if p € E? \ L, then
the component containing p is an open set in €? that misses L. 0
(5.C.2) Definition. Suppose A and B are homeomorphic subsets of a
topological space X. Then A and B are equivalently embedded in X if and
only if there is a homeomorphism h : X > X such that h(A) = B.
(5.C.3) Theorem. Let L in €? be defined by x = 0. Any line L in €? is
equivalently embedded to line L.
Proof. Suppose the equation of L is x = c. Then the space homeo-
morphism h defined by h(z,y) = (x — c,y) is the desired map. If L has the
equation y = az + b, then h(z,y) = (z,y — (ax + b)) defines the desired
homeomorphism. O
(5.C.4) Corollary. Each line in €? separates €? into two components and

is the frontier of each.
(5.C.5) Definition. A rectangle is a quadrilateral in €? with two vertical

and two horizontal sides (the “inside” of the rectangle is not included).
(5.C.6) Theorem (Jordan Curve Theorem for Rectangles). Suppose

R is a rectangle in €*. Then
(i) R separates €? into exactly two components and is the frontier of
each, and
(ii) E? \ R has exactly one component with compact closure (which is
henceforth called the interior of R.
Proof. Exercise. O
Our next goal is to generalize Theorem 5.C.6 to obtain the Jordan curve
theorem for rectilinear simple closed curves. We begin by “simplifying” a
given simple closed curve.
(5.C.7) Theorem. Suppose R is a rectilinear simple closed curve in e-

with more than four sides. Then there is a rectangle R’ and a rectilinear
simple closed curve R” such that
(7) R'N R" = K is aside of FR’,
(ii) R=(R'\ K)U(R"\ kK), and
(iii) R” has fewer segments than R.
Sketch of Proof. Let T be the shortest (in length) horizontal segment
that has adjacent vertical segments lying below it. Then essentially one of
the following four cases must occur. In each case the side K is indicated
with a dotted line.
It follows from the condition imposed on T that badly labeled configu-

rations like the one below cannot occur. The theorem follows. O
ir
We now indicate a homeomorphism of €? onto itself that. maps a given

rectilinear simple closed curve onto one with fewer sides. Enclose the rect-
angle R’ of the previous theorem in a pentagon as indicated below.
V1 v2
Us5 U3
U4
It should be apparent that a homeomorphism h from €? onto itself can

be defined that satisfies the following:
(i) his the identity outside of vj v2v304U5;
(ii) the segments Usa,
ab, bv3 are moved onto the segment K and the
remainder of R’ and its interior are correspondingly pushed down-
_ ward onto the triangle vsh(c)v3;
(iit) The triangular area v3u4v5 is pushed by h onto the area with ver-
tices v3, U4, Us, and h(c).
At this stage a new R’, R”, and K are found for h(R) (as in the previous
theorem), and the process is repeated to derive a second homeomorphism
whose image has still fewer segments. Eventually, this process must stop
when a rectangle is obtained. Hence, we have the following theorem.
(5.C.8) Theorem (Rectilinear Two-Dimensional Schénflies Theo-
rem). Suppose R is a rectilinear simple closed curve lying in the plane.
Then there is a homeomorphism h from €? onto itself such that h(R) = R*,
where R* is a rectangle (but not yet an arbitrary rectangle). O
As an immediate corollary we have the following special case of the

Jordan curve theorem.
(5.C.9) Corollary (Jordan Curve Theorem for Rectilinear Curves).

Suppose R is a rectilinear simple closed curve in €2. Then €2 \ R has exactly
two components. Precisely one of the components has compact closure and
R is the frontier of each component. O
(5.C.10) Definition. If R is a rectilinear simple closed curve in €?, then

the component of €? \ R that has compact closure is called the interior of
R, denoted by J(R), while the unbounded component of E? \ R is called the
exterior of R, denoted by F(R).
Note that interior (exterior) when used in this context is a distinct

concept from the interior (exterior) as defined in Chapter 1.
In the previous chapter, we devoted considerable attention to the prob-
lem of extending continuous functions. The problem now facing us is how to
extend a homeomorphism to a homeomorphism—a far more formidable task.
For instance, it is clear that the subsets of the plane in the figure below, (each
of which is the union of a rectangle and a point), are homeomorphic, but
the reader should be able to prove (using perhaps a compactness and con-
nectedness argument) that there is no way of extending a homeomorphism
between A and B to all of €?.
Nevertheless, homeomorphisms between rectilinear simple closed curves may

be extended to the entire plane.
(5.C.11) Theorem. Suppose FR and R’ are rectilinear simple closed curves,

and h is a homeomorphism from R onto R’. Then there is a homeomorphism
H : €2 > €? such that H|p =h and H(RUI(R)) = R'UI(R’).
Sketch of Proof. If R and R’ are both rectangles, the homeomorphic
extension of h is indicated by the figure, where the length of ¢’, is determined
by PSS lz) - On'll.
fe
& ly —~*,
In the pie case, let h, and hz be spacea hoesoulsepetae that take

R and R' onto rectangles R; and R2 respectively. Then he|R hhy A R, Maps
R, homeomorphically onto R2. By the special case for rectangles, this map
may be extended to a space ay eee kK. Loen.e= hy khy satisfies
all the requirements of the theorem.
a,
Ses
k
(5.C.12) Definition. A 2-cell is any topological space homeomorphic to

the subspace of €? given by B? = {(z,y) € E?. : a2 +y? < 1}.
(5.C.13) Corollary. If R is a rectilinear simple closed curve, then RUI(R)
isa 2-cell. O
(5.C.14) Definition. A 2-annulus is any space homeomorphic to S! x I.
(5.C.15) Notation. If t> 0, R; will denote {(z,y) € E? : |z| <t, |y| < t}.
(5.C.16) Theorem (Rectilinear Annulus Theorem). If R and R’ are
rectilinear simple closed curves with R' C I(R), then T = (RUI(R))N(R'U
E(R')) is homeomorphic to the annulus S!' x I. In fact, there is a space
homeomorphism that carries T onto FR \ int Ry.
Proof. Construct a small rectangle connecting R and R’ as indicated in
the figure. Note that AFDC and BF EC are rectilinear simple closed curves
whose intersection consists of the line segments F’ and C.
A ait
A= U4U5 U1
B= U1 U4
V1
B C = 1403
U5
D= U2VU6G U3
V4
E= U2U3
fee U1 V2
Define a homeomorphism h, that maps the paths A, F, D, and C onto the

segments A’, F’, D', and C’ in the figure below. Then (5.C.11) may be
applied to extend h; to a homeomorphism (which we still call h,) that maps
AFDC UI(AFDC) onto A'F'D'C' UI(A'F'D'C'). Define a second home-
omorphism hz that agrees with hy on CU F and maps the segments B and
E onto B’ and E", respectively. Again apply (5.C.11) to obtain an extension
of hg (still called hz) that sends BCE F UI(BCEF) homeomorphically onto
BICLESE UI (BI GSE’):
ps
Use the map gluing theorem to combine h; and hz into a homeomor-
phism h mapping (RUI(R))N(R' UE(R’)) onto the annular region between
A'B' and C'D'. The homeomorphism h may be extended to I(DE), (5.C.11).
In order to extend h to E(AB), let g : €? > E* be an extension of h|p given
by (5.C.11). Then we have g : E(AB) - E(A'B'). Now define h : €? + €
by
‘ h(a). ilove AUTH),
huey =
ee if2e RUE(R). Oo
(5.C.17) Exercise. Work out the details of the above sketch.

The proof we give for the Jordan curve theorem depends on being able
to approximate simple closed curves with rectilinear ones. We obtain this
approximation using brick partitions.
(5.C.18) Definition. A brick partition of € is a collection 7 of solid

rectangles (rectangles together with their interiors) that satisfy the following
properties:
(i) If R € T, then there are six line segments Aj, A2,...,A6, each of
which is either horizontal or vertical, and:
(a) FrR = A, U---U Ag;
(b) non-end points of each A; lie in exactly two members
6) aBe
(c) end points of the A,’s lie in exactly three members
Oe
(ii) The union of the rectangles in 7 is all of €?. ;
(iii) If RE T and FrR = A, U---UAg and if R € T and R # R, then
either RO R = 0 or for exactly one of i= 1,2,3,4,5,6, RNR = Aj.
The mesh of the partition is the least upper bound of diameters of the
members of 7. If the least upper bound does not exist, we say that the mesh
of 7 is infinite.
It is clear that for each positive number e, there is a brick partition of

€? with mesh e. The next theorem is quite useful for our purposes. Its proof
is a consequence of the previous work with 1-manifolds.
(5.C.19) Theorem. Suppose A is a compact subset of €? and T is a brick

partition of €?. Let M be the union of the elements of J that intersect A.
Then the frontier of M is the union of a finite number of rectilinear simple
closed curves, no two of which intersect.
Sketch of Proof. Since the number of members of 7 that intersect A
is finite, there can only be a finite number of components of Fr M. Show
that each component is a 1-manifold. It then follows from (5.4.3) that each
component is in fact a rectilinear simple closed curve. O
(5.C.20) Theorem. Suppose S is a connected compact subset of €2 such

that €? \ S is connected. Let U be an open set containing S. Then there is
a rectilinear simple closed curve S$ such that S$ Cc I($) c (S$ UI(S)) CU.
Sketch of Proof. Since S is compact, the distance from S to the com-
plement of U is positive. Choose a brick partition of €? with small enough
mesh so that if a brick intersects S, then the brick lies entirely in U. Let M
be the union of all of the partition elements that intersect S$. Let R be the
component of the frontier that contains S in its interior (why does such a
component exist?). By the previous theorem, R is a rectilinear simple closed
curve. If J(R) is contained in U, we are done. If not, then there are a finite
number of components of Fr M (other than R) that contain points of €? \U
in their interiors. We shall call such components holes. The holes satisfy the
following conditions:
(7) each hole is bounded by a rectilinear simple closed curve that lies
in I(R), and
(ti) each hole contains points of €? \ U.
Pitta ies
In each such hole choose a point of €? \ U, and construct an arc from

this point to ext(U) N E(R) that misses S' (recall that €? \ S is connected).
The distance from S to the union of the arcs and the frontiers of the holes is
positive. Hence, there is a brick partition of fine enough mesh so that each
brick that hits S misses these sets (and is contained in U). We may assume
that the union of the bricks that intersect S is contained in the union of
the bricks of our original partition that struck S. The boundary component
of this new partition that contains S is the desired rectilinear simple closed
curve S. 0
(5.C.21) Theorem. There is no retraction of €? onto a simple closed curve

S.
Proof. The proof mimics that of (5.B.6). Let S be a simple closed curve
in €?. By (5.B.5) and (5.B.9), S is not contractible. Suppose r : €? + S isa
retraction. Let p be any point in €? that does not lie in S and connect each
point of S with p.
Define a homotopy H: S x I > €* by H(z,t) = tp+ (1 —t)z; note

that Ho is the identity and H, is a constant map. Then rH: Sx I-— Sis
a contraction, which is impossible since S is not contractible. O
The following theorem shows, however, that if S is a simple closed curve,

then there is a neighborhood U of S that retracts to S. For every positive
number t, B; will denote the set {(x,y) € E€? : 2? + y? < t}. It should be
clear that By is homeomorphic to I x I for each ¢ (if not, consult (3.C.2)).
(5.C.22) Theorem. Suppose S is a simple closed curve in €*. Then there
is an open set U containing S and a map r: U — S that is the identity on
S,ie., S is a retract of U.
Proof. Let h: S 4 S! be a homeomorphism. Then h maps S into By,
and by (4.B.12) and (4.B.13), h can be extended to a continuous function H
with domain €?. Let V = (int Bz) \ By/2, and let r: V > S’ be a retract
of V onto S!. Then U = H~1(V) is open in €? and h~!rHy is the desired
retraction. O
(5.C.23) Theorem. If S is a simple closed curve in €”, then €? \ S is not

connected.
Proof. Suppose S is a simple closed curve in €* such that €? \ S is
connected. This will lead to the existence of a retraction from €? onto S,
which contradicts (5.C.21).
By the previous theorem, there is an open set U containing S and a
retraction r of U onto S. It follows from (5.C.20) that there is a rectilinear
simple closed curve $, between S and U, such that S c (S UI(S)) CU.
Furthermore, by (5.C.8), there is a space homeomorphism h which maps
SUI(S) onto a rectangle R and its interior. Then ¢ = hr|(gurgy)h* |(RUI(R))
is a retraction of RU J(R) onto the simple closed curve h(S), and since ¢
can be extended easily to E(R), we have contradicted (5.C.21). O
We next work toward determining the number of components of €? \ S,
and showing that S is the frontier of each component.
(5.C.24) Theorem. If A is the image of an arc in €?, then €? \ A is

connected.
Sketch of Proof. Suppose €? \ A is not connected. Let R be a rectangle
that contains A in its interior. Since RU E(R) is connected, it must lie in
some component K of €? \ A. Note that C = (E€? \ A) \ K is nonempty and
its frontier lies in A. Let ro € C. By (4.B.12), (4.B.13), and (4.B.19), there
is a retraction r: €? + A. Let p: E? \ {zo} > R be the obvious projection
of €* from zo onto R. Define a map h: €? > R by
nieve feed iLE Z

ple) ifa €€* \ C.
By the map gluing theorem we see that h is continuous. Moreover, h is a

retraction of €? onto R, which contradicts (5.0.21). O
(5.C.25) Theorem. If S is a simple closed curve in €? and U is a component
of 6*\S, then Fr! =:S;
Proof. Clearly FrU C S, and €? \ FrU is not connected. If FrU # S,
then there is an arc A such that FrU C AC S. Hence, (E? \ A) c €?\ Fru.
Since €? \ A is connected, either (€? \ A) C U or (E€7\ A) Cc E?\\U. However,
E2\ \AA = €%, and consequently either €? Cc U or €? c E? \U. Since neither
of these posnibilibies is viable, it must be the case that FrU = S.__O
The following theorem will be used to ensure the existence of no more

than two components in the complement of a simple closed curve.
(5.C.26) Theorem. Suppose S is a simple closed curve in €? and let « and

y be points in S and €? \ S, respectively. Then given any € > 0, there is an
arc A, such that
(i) A, has end points y and z, where z € S,
(it) d(x,z) < €, and
(tii) AeNS = {z}.
Proof. Exercise. QO
We are now ready to prove the principal result of this section.
(5.C.27) Theorem (Jordan Curve Theorem). Suppose S is a simple

closed curve in €?. Then €? \ S has exactly two components, and S is the
frontier of each. Furthermore, exactly one of these components has compact
closure.
Proof. It only remains to establish that €? \ S has at most two compo-
nents. Suppose to the contrary €* \ S has at least three components. We
will contradict (5.C.23) by constructing a simple closed curve S in €? such
that E2\$ is connected. Let C), C2, and C3 be components of E?\ S. Then
S is the frontier of each component. Let c, and cz be points in C; and C2,

respectively. Select two points s; and s2 in S$ ,and let 6 = d(si, 82). Careful
repeated application of the previous theorem yields disjoint arcs c)U1, CiV2,
Cot1, and cgt2 (see the following figure) such that:
(i) Oily
Ory ily WO Ss.
GipdGa vir 6/8 tors = 2}
(iit) d(s3;t;) < 0/8 fort = 1, 2,
(iv) for t = 1,2, civ; lies in C; and cot; lies in C2, except for the end
points v; and ¢;.
Let S be the simple closed curve formed from the arcs €1V1, Vi1t1, C2ti, Cote, teva,
and c,v2. We show that the removal of S§ fails to disconnect the plane.
Suppose z and 2 are points in €? \ S. If z and 2 both lie in C,, remove

a small arc Aj from S near cp (and lying in Cy). Then z and Z may be
connected to a point w € C3 by arcs that miss $ \ Ao (5.0.24). Note that
these arcs intersect S before they meet S (why?), and hence we can cross S
to enter C’3 at the first point where these arcs strike S, and then continue
to w. Thus z and 2 are connected by an arc that misses S. If z € C; and
Z € Cy, then small arcs are removed from S near c; and cg, and a similar
argument is applied. If at least one of the points z or 2 lies outside C, and
C2, then an analogous but even easier argument may be given. O
(5.C.28) Exercise. Suppose K is a compact subset of a connected open
subset U C €?. Show that there is a rectilinear simple closed curve R such
that K C I(R) and R CU. (Hint: Use brick partitions and (5.C.8).]
D. THE SCHONFLIES THEOREM

Although a simple closed curve S is homeomorphic to the unit circle S!,
does it necessarily follow that there is a homeomorphism h : €2 > €? that
carries S onto S'? In (5.D.9) we show that such a homeomorphism exists;
The Schonflies Theorem 137
however, the proof is by no means trivial. In fact, similar theorems for higher
dimensions are false without additional hypotheses. For instance, in Chapter
15, we construct an example of a 2-sphere that cannot be carried by a space
homeomorphism onto the standard 2-sphere. The Alexander horned sphere
shown below is one of the most famous such examples.
Intuitively, the reason why a homeomorphism from the horned sphere

onto the unit sphere cannot be extended to all of €° is that. it is impossible
to form a “membrane” with the ring R as its boundary which misses the
horned sphere. If a space homeomorphism h : €? + €° mapping the horned
sphere onto S? were to exist, R would be “freed” and such a membrane could
be found for h(R) outside of S?. Then h~! would carry this membrane back
to one that is glued to R but fails to intersect the horned sphere.
(5.D.1) Definition. An n-sphere S (that is, a topological space homeomor-

DMctoS” = 1(21,23,--.,2nd1) C6 2 ae: ee eel } lying in €°
is tame (in €"+") if and only if there is a homeomorphism h : E"+! > E”*1
such that h(.S) = S". Spheres that are not tame are called wild.
The Alexander horned sphere is an example of a wild 2-sphere. In this
terminology, the principal goal of this section is to show that every 1-sphere in
the plane is tame. Bing [1963] ingeniously showed in his side approximation
theorem that a tame sphere in €° can be approximated either from the
inside (the bounded complementary domain) or the outside (the unbounded
complementary domain) by a polyhedral 2-sphere (roughly, one that is the
union of a finite number of triangles).
We will eventually obtain an analogous theorem for the 1-sphere (5.D.8)

en route to a proof of the Schénflies theorem.
The existence of “necks” is a major obstacle in proving that an arbitrary
simple closed curve can be mapped via a space homeomorphism onto S!.
We will want to achieve very delicate approximations of simple closed curves
by rectilinear curves; however, necks can prove to be troublesome, as the
following figure might suggest. Note that points at the end of a neck may
be quite far from the corresponding points of the approximating rectilinear
curve.
A degree of control on necking is gained from the following theorem,

(5.D.2) Theorem. Suppose S is a simple closed curve in €? and ¢€ > 0.
There is a positive number 6 such that if z,y € S and d(z,y) < 6, then at
least. one component of S \ {x,y} has diameter less than «. (Henceforth, we
denote the smaller component of S\ {x,y} by L(z,y), and call 6 the necking
number for €).
Proof. Let x be a point of S and let C, and C, be the components
containing x of S./4(z) MS and S.72(x) NS, respectively.
The Schdnflies Theorem 139
GE **ieeenee ace
Choose 0 < Az < €/2 so that A, is less than the distance between C,
and S\C',. Note that if y € C, and z is a point of S such that d(y, z) < Az,
then z € C, and at most one component of S \ {y, z} has diameter greater
than e. Cover S with the sets {C, : x € X}, and select a finite subcover
{Cz,,Cz5,..-,Cz,}. Let 6; be a Lebesgue number for this cover, and let
6 = min{d;,z,,-.-, Az,}. Then it is easy to verify that 5 is the required
positive number. O
(5.D.3) Notation. If h is a homeomorphism between subsets A and B of

E?, then |h| will denote sup{d(a,h(a)) : a € A}. We let H(A, B) denote
inf{|h| : h is a homeomorphism from A onto B}.
(5.D.4) Theorem. Suppose S is a simple closed curve in €? and € is a
positive number. There is a number 6 > 0 such that
(i) if S is‘a simple closed curve with H(S,S) <6, and
(ii) if 2 and y are points in S with d(z,y) < 6,
then diam L(z, y) is less than e.
Proot. Corresponding to €/3 there is a necking number b 088 <.é,
such that if d(x, y) <6 where x,y € S, then diam L(s, y) 6/3. Leo = 6/3
and suppose S$ is a simple closed curve such that H(S,S) <6. Leth: SS
be any homeomorphism with d(s,h(s)) < 6 for each s € S. Suppose w and
z are points of S and d(w,z) < 6. We show that diam L(w,z) < e. First
note that d(h~!(w),h7!(z)) < d(h-!(w),w) + d(w,z) + d(z,h-1(z)) < 36,
and hence diam L(h~!(w),h71(z)) < €/3. Now h(L(h~*(w), h7*(z))) = A is
clearly an arc from w to z, and we need only establish that its diameter is less
than «. Suppose s,t € A. Then d(s,t) < d(s,h—1(s)) + d(h1(s),h—*(t)) +
d(h—'(t),t) <d+6¢/3+6<e, and thusdiamA<e. O
Proofs of the following important corollaries are left as nontrivial exer-
cises. They should provide some good entertainment for the reader.
(5.D.5) Definition. If A is a closed subset of a connected locally connected

topological space X, then any component of X \ A is called a complementary
domain of A.
(5.D.6) Corollary. Suppose S is a simple closed curve in €7 and € > 0.

There is a 6 > 0 such that if S and S’ are rectilinear simple closed curves
in a complementary domain of S with S Cc I(S’), H(S, S) <6, and with
H(S,S') < 6, then for each z,y in the annulus T determined by S and S",
where d(x,y) < 6, there is a rectilinear arc A in T such that
(i) diamA < e€, and
(ii) A has x and y for end points. O
(5.D.7) Corollary. Suppose S is a simple closed curve in E? and let e > 0
be given. There is a 6 > 0 such that if S and S' satisfy the conditions of the
previous corollary, then the annulus between S and S’ may be written as the
union of a finite number of rectilinear simple closed curves D,, D2,...,Dn
together with their interiors, where
(i) I(Di) NI(D;) = @ for i F J,
(it) diam D; < «,
(iit) D;.D; = unless |i —j|< 1, or |i —37| =n — 1, and
(0) oD at nee (and D,; MN D,) is an arc with one end point on S, one
end point on S’, and the arc minus its end points lies between $
and S$’! O
The next theorem says that any simple closed curve can by approxi-
mated (on either side) as close as you wish by a rectilinear simple closed
curve.
(5.D.8) Theorem. Suppose S is a simple closed curve in E”, € is a positive
number, and U is a complementary domain of E€2\ S. Then there is a
rectilinear simple closed curve S$ contained in U such that
(i) H(S,S) <, and :
(ii) if U = I(S), then S C E(S), and if U = E(S), then S c I(S).
Sketch of Proof. We use the following notation within the proof: If A is
the image of an arc in €? with end points x and y, then we write A = [x,y],
and denote A \ {x,y} by (x,y). The proof is first outlined, and then some
of the details are indicated.
1. Suppose U = E(S). Let € be a positive number. For each z € S,
rectilinear arcs are constructed from a point pz € E(S) to points gz and r,;
that lie on “opposite sides” of z, The resulting simple closed curve formed
by the arcs [pz,dz], [gz, 72] and [rz,pz] will have diameter less than e, and
furthermore, it will intersect S only in the arc [qz, rz].
The Schdnflies Theorem 141
2. Since {(qz,rz) : x € S} is an open cover of the compact set S,

there is an irreducible finite subcover A = {(gz,;rz;) : i= 1,2,...,n}. (A is
irreducible in the sense that if (q2;,r2;) € A, then (q2,,rz;) is not contained
in the U (q2; Tx; 3)

ixj
t=1
Tro
3. For 7 = 1,2,...,n, let T; be the simple closed curve formed by the

union of [pz;,@z;], [dz;,7z;], and [rz;,pz;]. The rectilinear parts of these
simple closed curves are modified so that nonadjacent T;’s do not intersect,
and so the rectilinear portions of adjacent T;’s intersect at only one point.
4. Show that the rectilinear simple close curve running around the
outside of the T;’s is the desired one.
Now we indicate a few of the details.

1. Let z € S and let C, be the component of SM S,/2(x) that contains
x. Select points a, and b, in Cz that lie on opposite sides of z.
Si
Let D, be the component of S,/2 \ (5 \ (az, bz)) that contains (az, bz),
and select points gz and 7, such that g, € (az,z) and fz € (a,b). Let
Az = min{d(gz, x),d(7z,z)} and let 6, be a necking number for Az.
Now pick a point pp € D, MN E(S) and let Q; and R, be rectilinear
arcs in E(S)M D, with initial point p, and terminal points in S5, (gz) and
Ss, (72), respectively. From the respective terminal points, send out straight
line segments toward g, and f,, and let g, andr, be the first points in S that
these segments encounter. Replace these segments by rectilinear ones. Now
devise a method to obtain arcs [pz,qz] and [pz,rz] that have only the point
Pz in common and that lie (with the exception of g; and rz) in Dz N E(S).
2. Relabel the arcs [gz;,72z;] so that as you go around S in one direction,

points are encountered in the following order:
Qazi Tons Ix) Toy) zr3) Tao A243 T 235 Qz5; Vo4) CERO RLS | Urn ee Ae) dx
For each 2, let T; denote the simple closed curve composed of the arcs
[Pei ) dai], (az; ’ Tz;] and [rz Px; )-
3. We say that T; and T; (i # j) are adjacent if and only if |i—j| = 1, or
1=1and j =n, or j = 1 andi = n; otherwise, they are nonadjacent. Now
alter the T;’s to meet the conditions. All repairs are to be done so that the
diameters of T; remain less than € and that each T; still intersects S only in
The Schénflies Theorem 143
the arc [qz;,72;]. To illustrate the basic procedure, we indicate how T, and
T3 may be pulled apart.
First, note that we may assume that 7; and T3 intersect only a finite
number of times, since slight adjustments of say T3 will alleviate the situation
indicated in the next figure.
Now induct on the number of points in the intersection of T; and T3.

Obviously, there is nothing to do if n = 0 (or 1), where n represents the
number of points of intersection. Assume, then, that T; and T3; may be
separated in the desired fashion whenever there are n or fewer points of
intersection, and suppose 7; and 73 intersect in n + 1 (actually it must be
n+ 2) points. We construct another T3 that meets 7, in fewer points.
Begin at gz, € T3 and proceed toward rz, (along the rectilinear portion
of T3). Let s be the first point found in common with T,. At this point, there
must be some segment of T; that crosses T3. Starting at s, move along T) (in
I(T3)) until T; leaves I(T3) at some point §. Now T3 is altered as indicated
by the dotted line in the next figure. We again start at qz,, but this time
we refuse to enter 7\ at s; instead we move toward §, always staying close to
T,, and finally we continue along T3 until r,, is reached. The new curve has
two intersections fewer with 7), and induction takes case of the remaining
ones.
oo00000000000000000000
0
The process is repeated for all pairs of nonadjacent T;’s. A similar

procedure is applied to adjacent ones. If sufficient care is exercised, the
diameters of the 7;’s remain less than e.
4. We now have the following situation:
Let R be the rectilinear curve indicated in the following figure. A home-
omorphism from S onto R is defined as follows.
The arc (gz, Tx,]is mapped onto [tn, t1], the arc [rz,, dz,] is mapped onto
[t1, ta], the arc [gz,,7z3] is mapped onto [t2,t3], etc. By the construction it
is clear that this homeomorphism moves no point more thane. O
(5.D.9) Theorem (The 2-dimensional Schénflies Theorem). Suppose

S is asimple closed curve in €?. Then there is a homeomorphism h : €? > €?
such that h(S) = S?.
Proof. First, we want to define a homeomorphism from SUI(S) onto B?
by approximating S from the inside with an increasing sequence of rectilin-
ear simple closed curves. The annular region between succeeding rectilinear
curves is subdivided and the pieces mapped onto ‘truncated wedge shaped’
pieces of annular regions formed by concentric circles in the interior of S?.
If this is done with appropriate care, a homeomorphism may be found from
S plus its interior onto the unit disk.
Consider, first, the following subsets of S' (the points are specified by in-
dicating their radian measure): A; = {0,7}; Ap = {0,7/2,7,37/2}; A3 =
{0, 7/4, 2/2, 37/4, 7, 5/4, 32/2, 77/4}; etc.
The Schonflies Theorem 145
m/2
‘Ay ‘A2
37/2
Let g be any homeomorphism from S! onto S, and for each i = 1,2,...

let A; = min{d(g(x), 9(y)) : z,y € A;}. Let {e;} be a decreasing sequence of
positive numbers such that €; < min{A;, 1/2*} fori = 1,2,---. Apply (5.D.4),
(5.D.6), (5.D.7) and (5.D.8) to obtain a sequence of rectilinear simple closed
curves {.S;} and a sequence of homeomorphisms {h;} such that
(2) Dek: I( S41) G (Si41 U I(Si41)) e I(S) foreach t= 102 4
(ti) hy: S > S; and for each x € S, d(h;(z),x) < €;/2, and
(iii) if a; € S; and x41 € Sj41 and d(2z;, 2441) < €;, then there is an arc
A; with end points z; and z,4, that (except for the end points) lies
in the open region between S; and S;,; and diam A; < 1/1.
Let {T;} be a sequence of concentric circles in S!, where each T; has
radius 1 — 1/(2i). For each i, define r; : S' > T; to be ene radial projection.
nm /2
31/2
We begin the construction of a homeomorphism from SU J(S) onto the

unit disk: Define h : U72, Si + UZ, Ti by h(si) = rig
th; ‘(s;) for each s; €
S;. Extend h to the interiors of the various annuli as follows. Map r;(0) = al
and r (7) = a} into S; by setting b} = h~1(a}) and b) = h7*(a). Now,
similarly, the points of T2, a? = r2(0), a3 = re(m), a3 = r2(1/2), and aj =
r2(37/2), are mapped to points 67, 63, b3 and by of Sy by h~*. Rectilinear
arcs of diameter less than 1/1 may be found that join bj to bj and b} to b3 (by
(ii) and (iii) above). Call these arcs B} and B3. First extend h by mapping
the arcs B} and Bi} onto the arcs Aj and A} indicated in the previous
figure. Then use (5.C.11) to further extend h to the two 2-cells bounded
by Si, S2, Bi and B} so that they are mapped homeomorphically onto the
corresponding cells in the disk bounded by T;, T2, A} and A}. (Clearly
(5.C.11) applies, even though the T;’s are not rectilinear. The skeptical
reader may See replace the T;’s and S! by eS in this proof.)
The points a = 73(0)jpas.= ran), as = cst /2), 2 = 13(37/2),
a3 = r3(7/4), a8 = r3(3n/4), a? = 13(57/4) and a3 = r3(Tm/4) are now
mapped by h~! onto points b?, 7 = 1,2,...,8 in S3. Arcs of diameter less
than 1/2 are found and mapped onto the obvious candidates in the disk.
The four pieces of the annular region between Sz and S3 are carried by an
extension of h to the appropriate wedges in the interior of S?.
Continue for each positive integer 7, and the end result is a homeomor-
phism h from I(S) onto I(S!). Let h = g~! on S, and we have constructed
a function from SUI(S) onto S'UI(S*) that is clearly 1-1, onto, and is con-
tinuous except possibly at points of S. To remove any doubts concerning the
behavior of h at a point 2 € S, we show that if {x;} is a sequence of points in
I(S) that converges to x, then {h(a;)} converges to h(x). It may be assumed
that for each 7, x; belongs to the annulus bounded by S; and S;4,. If x; is not
in S;, select a point c; € S; (c; isin the same rectilinear region as z;) such that
d(x;,c;) < 1/i. Consider d(h(x),h(xi)) < d(h(x), h(ci)) + d(h(c;), h(x;)).
Clearly, for large i, one need only worry about d(h(x),h(c;)). Let «€> 0
be given. Corresponding to € there is a 6 > 0 such that if s;,s2 € S and
d(s1,82) < 6, then d(g~!(s1),9~+(s2)) < €/2. Furthermore, there is a posi-
tive integer N such that for i > N, d(h;*(y),y) < 6/4 for each y € S; and
d(r;(x),x) < €/2 for each z € S.
Now, h(c;) equals r;g~!hj*(c;). For i > N we have d(h;'(ci),ci) < 6/4
and d(c;,z) < d(ci,zi) + d(aj,z) < 6/4+ 6/4. Hence, d(h7(c),2) <
d(h; *(ci), ci) + d(c,z) < 35/4, and thus d(g-1h;*(c:),g71(a)) < €/2. Fi-
nally, from the triangle nea we have d(rjh~1g;'(ci),g7!(z)) is less
than d(rig~1h; '(ci), 971A;*(ci) +d(g7h;* (ci),g~(a) which is less than
e. The continuity of h follows. (Why is it legitimate to use the sequence {c;}
instead of the original sequence {z;}?)
An analogous procedure may be used to obtain both a sequence {S;} of
rectilinear simple closed curves that converges to S from the exterior of S,
and an analogous homeomorphism. 0
(5.D.10) Corollary. If S is a simple closed curve lying in €?, then there is

a homeomorphism H : (S UI(S)) > B?. G
An important property common to all Euclidean spaces €” is the fol-

lowing: If U is an open subset of €” and h: U > &” is an embedding, then
h(U) is open in E€”. This property is frequently referred to as the invariance
Problems 147
of domain, and it is not difficult to construct examples that show that not
all topological spaces satisfy this condition. Invariance of domain turns out
to be one of the most powerful features of €”, and its proof is appropriately
sophisticated. The Schénflies theorem may be used to show that invariance
of domain holds in €?. That the invariance of domain property also holds in
€” for n > 2 is proven in Chapter 15.
(5.D.11) Exercise. Show that if U is an open subset of €? and ifh: U > E?

is an embedding, then h(U) is open in E€?.
E. THE ANNULUS THEOREM

(5.E.1) Exercise. Suppose R is a rectilinear simple closed curve in €?.
Show that there is a compact set K and a homeomorphism h : €? -> €? such
that h(R) = S' and hlg2\x% = id. [Hint: Note that the homeomorphism
constructed in (5.C.8) is equal to the identity off a compact set. Inscribe
everything in a large circle and project the rectangle obtained onto the circle.]
(5.E.2) Exercise. Show that the homeomorphism obtained in the Schénflies
Theorem can be chosen so that there is a compact set K C &? such that
hle2\K is the identity.
(5.E.3) Exercise. Suppose S and S are simple closed curves in €?. Let U
be an open subset of €? such that U is homeomorphic to €? and SUS CU.
Show that there is a homeomorphism h : €? — £€? such that h(S) = S and
hlea\u
= id.
(5.E.4) Exercise. Suppose S and S are simple closed curves in €2 such
that S Cc I(S). Show that there is a homeomorphism h : €? - E€? such that
h(S) = S! and h(S) is a circle of radius 1/2 centered at the origin.
(5.E.5) Exercise (Annulus Theorem). Suppose S and S are simple
closed curves in €? such that $ C I(S). Show that J(S) MN E(S) is an annulus.
PROBLEMS
Section A
1. A 1-manifold with boundary is a separable metric space M with the

property that each point of M has a neighborhood homeomorphic to the
closed unit interval. Classify all connected 1-manifolds with boundary.
2. Find an example of a space X such that every point has a neighborhood
homeomorphic to (0,1) but such that X is not a T) space.
3. Let 9 be the first uncountable ordinal. Order L = [0,) x [0,1) by

(a,r) < (G,s) if and only if (¢) a < 8 or (ti) a= 8 andr<s. Then L
with the corresponding order topology is called the long line. Essentially,
we have filled in the gaps between the ordinals by the intervals (0,1).
Prove:
(a) Each point of L has a neighborhood homeomorphic to (0, 1).
(b) Lis To.
(c) L is connected.
(d) L is not compact (in fact, not Lindeléf).
(e) L is countably compact.
(f) L is not separable.
Section B
1. Show that J x J has the fixed point property.

Show that there is no retraction of B? onto S!.
3. Show that if f : B? > €? is a continuous function for which z does not
lie on the ray from the origin through f(z) (except when f(z) = (0,0)),
then f has a fixed point.
4° (For n= 1,257.57 tct Ag = {(2,9) € 6 7 a Sif 0 og = Te oe
Ap = {(a,y)
€ E? : (cg =O and 0 < y < 1) or O << 1 and y =0)}.
Let X = U2, Ai. Show that X is contractible to the point (0,1), but
there is no contraction H such that H((0,1),t) = (0,1) for all t.
Show that contractible spaces are path connected.
Show that any space can be embedded in a contractible space.
Suppose U is a bounded open subset of €? and that h: €? > €? isa
homeomorphism such that h|p,.,,= id. Show that h(U) =U.
Sections C and D
1. A polygonal simple closed curve in €? is a simple closed curve in €?

that consists of the finite union of straight line segments (not necessarily
horizontal or vertical). Suppose we have a polygonal simple closed curve
ne
(a) Show that by a change of coordinates, things can be arranged so
that each pair of vertices have distinct z and y coordinates.
(b) Divide the bounded region of the polygonal simple closed curve
into a finite number of squares and triangles by means of horizontal
and vertical lines through each vertex, and prove that the bounded
Problems. 149
region is homeomorphic with a 2-cell. (This argument may be gen-

eralized to polyhedral 2-spheres in €3 (Moise [1952]).)
2. Show that n disjoint simple closed curves in €? haven+1 complementary
domains.
3. Let 7, 11, Y2,---,;%n be the images of disjoint simple closed curves in
E* such that for i = 1,2,...,n, 7% is contained in the bounded comple-
mentary domain D of y. Let D; denote the bounded complementary
domain of 7; and define P = (yUD) \ (Uj_, Di). Show that P can be
written as the union of a finite number of disks with disjoint interiors.
4. A map f : X + Y isa local homeomorphism if and only if for each
xz € X, there is an open set U containing z such that f|y : U > f(U)
is a homeomorphism and f(U) is open in Y. Show that f : €! > S!
defined by f(x) = (cosz,sinz) is a local homeomorphism but not a
homeomorphism.
5. Let B? = {(z,y) € €? : x? +y? < 1} and suppose f : B? > B? isa
continuous function with the properties that
(i) f maps S! (= Fr B?) homeomorphically onto itself, and
(it) flint ge is a local homeomorphism (see problem 4 above).
(a) Show that f is onto (recall (5.C.21)).
(b) Suppose z € int B?, y € S! and that A is an arc between z and y.
Show that f~!(A) consists of arcs (possibly only one) whose only
intersection is at f—*(y).
(c) Suppose f(z) = f(#) = w, where z,£,w € intB”. Let w* be
the end point on S! of the radius G (on any radius G in case
w = (0,0)) that contains w. Select arbitrary points w; and w2 on S!
that are distinct from w* and connect these to w with straight line
segments B and C, respectively. Let = f~'(w*), 1 = f~*(w1),
ta = f~'(we) and find arcs Az, Az, Bz, Bz, Cz, Cz with end
points {z,%}, {%,2}, {v,21}, {2,21}, {v, 22}, {2,72}, respectively
such that f(Az) = f(Az), f(Bc) = f(Bz), and f(Cz) = f(Ca).
Finally, let D, and D2 be the arcs on S' with end points {x,,Z}
and {r2,Z} and whose intersection consists of precisely 7.
(a) Show that ¢ belongs to the bounded complementary domain of
Dy UW By. Cz D>.
(b) Show that £ does not belong to the bounded complementary domain
of either A, UD; UB, or Az UC, U Do.
(c) Deduce from (1) and (2) that f must be 1-1.
(d) Show that f is a homeomorphism.
Show that there are only a countable number of mutually disjoint triodes
in the plane (Pittman [1970]) [Hint: Let D= {D; : i € Z+} be a basis
for €! consisting of a countable number of open disks. If T is a triode,
let Dr, € D such that 0 € Dry and Dri N{X,Y,Z} =9.
xX
a
ZL
Let X7, Yr, Zr be the first points of OX N Fr Dri, OY NFr Dri, OZ N

Fr Dr. Show that Dri \ (OXT UOYr U OZr) = Ur2UU73 VU Ura, where
the Ur,,’s are disjoint open sets. For 7 = 2,3,4 select Dr; € D such that
Dr, C Ur,;. Thus, a quadruple of sets in D is associated with T’.]
Ge Show that no arc separates S?.
8. Show that each simple closed curve S in S? has two complementary
domains. Furthermore, show that the closure of each complementary
domain of S is homeomorphic with B? = {(z,y) € €? : a? +y? < 1}.
A theta curve is any space homeomorphic to the union of the images
of three arcs, each two of which intersect precisely at their end points.
Show that each theta curve in €? separates €? into exactly three com-
ponents.
10. If A and B are homeomorphic compact sets in €? compare H(A, B) and

the Hausdorff distance Dy(A, B).
Chapter 6
THE PRODUCT TOPOLOGY

AND INVERSE SYSTEMS
A. THE PRODUCT TOPOLOGY REVISITED

In previous chapters we repeatedly encountered the product topology asso-
ciated with a finite number of topological spaces. Our present task involves
extending definition (1.D.7). Recall from set theory: if A= {Xq : a € A} is
a family of sets, then the Cartesian product of the Xq’s, [ acA +a; 18 defined
by
[] Xe = {fAUXa : f(a) € Xa for each ae A}.
acA ‘
The axiom of choice says that [],<, Xo # @ if and only if each Xq # 0.

We denote a typical element f € [],¢, Xa by {fa}, with the understanding
that if a € A, then f(a) = rq. The problem is to find a suitable topology for
Taeca Xa. An initial choice for such a topology (considering the definition
given when A is finite) might be that obtained from a basis whose elements
are of the form Ree U,, where U, is an open set in Xq for each a € A.
However, this topology (frequently called the box topology) does not serve as
a suitable generalization of the finite product topology. It has more open
sets than is necessary, or desirable.
What, then, should open sets be like in [],<, Xa? In the finite case it
was shown that the projection maps pa : [],¢, Xa 4 Xa are continuous.
We would certainly like this to hold for the infinite product as well. However,
if continuity of these maps were the only desideratum for determining open
sets, then assignment of the discrete topology to the Cartesian product would
settle the matter. But, the discrete topology has too many open sets to be
of interest. What we want is a topology for the product space that has a
minimum number of open sets, but for which the projection maps are still
continuous. Fortunately, there is no problem in doing this. In order for pa
to be continuous, p,'(U~) must be open for each open subset Ug of Xq.
151
152 6. The Product Topology and Inverse Systems
Let S = {pz1(Ua) : a € A and Ug is open in Xq}. Note that ifa # 6

then pz!(Ua) Npa!(Us) ¢ S, and hence fails to be a topology or even a
basis for a topology. Nevertheless, this deficiency is easily remedied if we
add to S all finite intersections of members of S. The resulting family of sets
meets all the requisites necessary to be the basis for some topology. Observe
that basic open sets are of the form ()j_, pa; (Ua;) (frequently denoted by
Ug, X-::* Ua, x []{Xa : a€ A and a # a4,...,Qn}). Clearly, when A is
a finite set we obtain definition (1.D.7) as a special case.
(6.A.1) Definition. Suppose {Xq : a € A} is a family of topological
spaces. The product topology for [|,,<, Xa is the topology which has a basis
consisting of all sets of the form ()j_, pz) (Ua;), where for each a; Ug, is
open in Xq,.
(6.A.2) Notation. If {Xq : a € A} is a collection of topological spaces

then (unless otherwise stated) [],¢, Xa will denote the Cartesian product
of the X,’s with the product topology.
In the first exercise that follows, the reader is asked to confirm that the
product topology is indeed the smallest topology for which continuity of the
projection functions is maintained. The exercises, theorems and problems of
this section give further justification for this choice for the topology.
(6.A.3) Exercise. Suppose {Xq : a € A} is a collection of topological
spaces and [],¢, Xa has the product topology U. Show: if V is another
topology for [],¢, Xa such that V GU, then not all of the projection maps
are continuous.
(6.4.4) Exercise. Show that the projection maps are open for the product
topology U and the box topology V.
(6.A.5) Exercise. Suppose {Xq : a € A} and {Yq : a € A} are families of

topological spaces and for each a € A, fy : Xa — Yq is continuous. Define
T1fo : MacaXa> Taca Ya by I]fa({2a}) = {fa(va)}. Show that []fa
is continuous. Show that if X. = X for each a, and ¢: X > [J],
e,q Yo is
defined by ¢(z) = {fa(x)}, then ¢ is continuous.
The procedure used to obtain a basis for the product topology from finite
intersections of subsets is utilized frequently enough to merit special consid-
eration.
(6.A.6) Definition. Suppose (X,U) is a topological space. A subbasis for

(X,U) is a collection S of open subsets of X with the property that the
collection of all finite intersections of members of S forms a basis for (X,U).
Note that if S is any family of subsets of X, then the collection of finite
intersections of members of S forms a basis for some topology on X.
The Product Topology Revisited 153
(6.A.7) Exercise. Show that the topology obtained from a subbasis is the
smallest topology that contains the subbasic sets as open sets.
(6.A.8) Exercise. Show that sets of the form (a,0oo) and (—oo,b), where
a,b € R, form a subbasis for €!. Formulate and prove a similar result for
the order topology.
Continuity is easy to express in terms of a subbasis. The following

theorem is the most useful of the various possible formulations.
(6.4.9) Theorem. Suppose X and Y are topological spaces and S is a

subbasis for Y. Then a function f : X — Y is continuous if and only if
f—1(S) is open for each S € S.
Proof. Since members of S are open, it follows from the continuity of
f that f—1(S) is open for each S € S. Conversely, suppose f~!(S) is open
for all S in S and let U be a member of the basis for Y determined by S.
Then U is a finite intersection of members of S, and since intersections are
well behaved under inverse images, (0.C.7), the theorem follows. O
We now determine which of our previous results relating to finite prod-
ucts also hold for infinite products. Perhaps the most difficult. question to
resolve is that of compactness. The result that the arbitrary product of com-
pact spaces is compact is known as Tihonov’s theorem, and this theorem fol-
lows readily from the characterization of compactness given by Alexander’s
Lemma below. Note that while it is obvious that any cover of a compact
space by subbasic open sets must have a finite subcover, it is not at all
obvious that the existence of such subcovers is sufficient to guarantee the
existence of a finite subcover when one is given a cover by open sets that do
not necessarily belong to the subbasis.
(6.A.10) Theorem (Alexander’s Lemma). Suppose S is a subbasis for
the topology of a topological space X. Then X is compact if and only if
every cover of X by members of S has a finite subcover.
Proof. Suppose X is not compact. Then there is an open cover U of X
from which no finite subcover can be extracted. Let H be the collection of
all open covers of X from which no finite subcover can be extracted. Then
the family U is a member of H. Let D be a nest of covers in H (ordered by
inclusion) and observe that V = {V : V € D and D € D} is an upper bound
for D that lies in H. By Zorn’s lemma (0.D.3) (some form of the axiom of
choice is inevitable in this proof), there is a maximal element M in H.
The family M has the following peculiar property. Suppose M € M
and V;,,V2,...,Vn are open sets in X such that Vi 1V20---NVn C M.
Then, for some 7, we have V; € M. If this were not the case, then since
M is maximal in H, for each i there would be sets M;,, Mi,,...,Mi,, such
that Vi U Mj, U---UMj,, = X. Hence, X \ Vi C Uji, Mi,, which gives
X \ (Mea Vi) = U(X \ Vi) C Uli Uji Mi; However, that is absurd,
for we would then have
x=(Yn--avV)Uul{ UUM, |cMul(UU™, |.

1=17=1 i=1 j=1
which implies that M is not in H.

Suppose now x € M € M. Since S is a subbasis, there are sets
S1,52,---,S5n € S such that x € S;NS2N---NS, C M. By the pecu-
liar property of M, we have S; € M for some i. Thus each xz € X is in
some S$; € S which is also in M. Hence SMM is an open cover of X. This
poses a problem: since SMM is a cover of X by subbasic open sets, it has
a finite subcover; since SM M is a subset of M, any finite subcover by sets
in SM M is a finite subcover of M—contradicting M € H. This dilemma
can be resolved only by admitting that X was compact.
The other half of the theorem is completely trivial. O
(6.A.11) Theorem (Tihonov’s Theorem). Suppose {X, : a € A} isa

family of topological spaces. The product [],¢, Xa is compact if and only
if X_. is compact for each a € A.
Proof. Suppose X,g is compact for each a € A. Let S be the subbasis
for [[,c, Xa consisting of all sets of the form {py'(Ua) : a € A and Uy
is open in X,}. By Alexander’s lemma, it suffices to show that each cover
of [].c, Xa by members of S has a finite subcover. Suppose then iSaer’S
covers [[,¢, Xa, and for each a let Cg = {U C Xa : Pai(U) € Si We
show that for some a, Cg covers Xq. If that is not the case, then for each
a € A we can choose a point tg € Xq \(U{U : U € Cy}). Then the point
{za} € T]xcq Xa is not covered by S, an impossibility. Thus, there is
an a € A such that C, is an open cover of Xq. Since Xq is compact, a
finite subcover {U,,,Ua,,..-,Ua, }can be extracted from Cy. The collection
{p,!(Ua,),-++;Pa (Ua, )} constitutes a finite subcover of S.
The proof of the second half of this magnificent theorem is quite easy
and is left to the reader. O
Tihonov’s theorem has been described as the single most important

theorem in general topology. Whether or not such hyperbole is warranted
might be a matter of some debate; nevertheless, its exalted position in both
topology and analysis is beyond dispute. Kelley [1950] showed that Tihonov’s
theorem is equivalent to the axiom of choice.
(6.4.12) Theorem. Suppose {X, : a € A} is a collection of topological

spaces. The product X = [|[,¢, Xa is connected if and only if each Xq is
connected.
The Product Topology Revisited 155
Proof. If X is connected, then since each pq is continuous it follows

that each Xq is connected. For the converse, define points z = {tq} and
y = {ya} in X to be equivalent (x ~ y) if and only if {ae A : tg F Ya} is
finite. The remainder of the proof is split into a number of lemmas, each of
which represents a fairly trivial exercise.
Lemma 1. If r= {z.} € X and D= {yeé X : x ~ y}, then D is dense
in A
Lemma 2. Suppose K C A and X = Tleex Xa- For each a € A\ K, let
dq be an arbitrary point in Xq. Define a map f : X + X by PUGaheeK
{Ya teens Where Yo = dy if ae A\K, and yo. = fq if a € K. Then f is
continuous.
Lemma 3. If K is finite, then f(X) is connected.
Now suppose F' is any nonempty open and closed subset of X. It must
be shown that F is in fact X (2.4.4). Select a point u = {ua} € F, and
let v = {vq} be any point in X equivalent to u in the above sense. We
show that v € F’, and it will then follow from Lemma 1 that F is dense
in X, and since F is closed, F will be all of X. To see that v € F, let
K = {aj,Q2,...,Q@n} have the property that if a € A\ K, then ug = va,
and let X = Xo, Mee te ay so OL CACh O26 AN KO let G == tn
and let f : X — X be the continuous map defined in Lemma 2. By Lemma 3,
f(X) is connected, and since u € f(X)NF, f(X) must actually be a subset
of F (otherwise, f(X) MF and f(X)M(X \ F) would form a separation of
an
f(X)). Thus, we haveve f(X) CF. O
More care must be exercised in dealing with families of locally compact
and locally connected spaces. For example, if we take X, = {0,1} (with the
discrete topology) for each n € Zt, then [],¢z+ Xn is totally disconnected,
and thus not locally connected (components are points, which are not open
in the product topology) even though the individual spaces X,, are locally
connected. The next theorem shows how this situation can be partially
remedied.
(6.A.13) Theorem. Suppose {X, : a € A} is a family of topological

spaces. The product X = [],¢, Xa is locally connected if and only if
(i) each Xq is locally connected, and
(ii) all but a finite number of Xq are connected.
Proof. Suppose X is locally connected, and a € A. Let tg € Ug C Xa,
where U, is open in X,. Choose any point x € X such that pa(r) = Za.
Then p,'(Uq) is an open set in X containing x. Since X is locally connected,
there is a connected open set V such that x € V C p,'(Uq). Note that pa(V)
is an open connected set containing tq and lying in U,. Part (ii) follows
from the fact that pg(V) = Xg for all but a finite number of 8 € A—even
though V is not necessarily a basic open set.
For the converse, let K C A be a finite subset such that a € A\ K
implies Xq is connected. Suppose z € V C hve a Xa Where V is open. We
may assume (why?) V is of the form Vy, x Va, X-:: X Va, X []Xy, where
7 € A\{a1,a2,...,an}. Now, fori = 1,2,...,n, let Uz; be a connected open
set such that tq; € Ua; C Va;. For each B; € K, let Ug; be a connected open
set in X that contains g,. Then U = Ug, x:::xUa, xUg, x---xUz,, XT] X47;
where y # aj, 3; is open and connected. Since x € U C JV, it follows that
Tlaca Xa is locally connected. O
The proof of the following theorem is left to the hyperactive reader.
(6.4.14) Theorem. Suppose {Xq : a € A} is a family of topological

spaces. Then [],¢, Xa is locally compact if and only if
(i) each Xq is locally compact, and
(ii) all but possibly a finite number of X, are compact.
As we will presently see, the arbitrary product of metric spaces need not
be metrizable. However, if the index set is countable, we have the following
important result.
(6.4.15) Theorem. Suppose (Xj, di), (X2,d2),... is a countable collection

of metric spaces. Then the product [],-7+ Xi is metrizable.
Proof. Since equivalent metrics generate the same topology, we are
permitted to replace each d; with d;, where d;(z,y) = min{1,d;(x,y)}. Let
0: Viez+ Xi x Tieg+ Xi > [0, 1) C [0, 00) be defined by
where x = (%1,22,...) and y = (y1, y2,...). The details needed to show that
p is a metric are omitted, and we concentrate on showing that the metric
topology induced by p coincides with the product topology for iez+ XK
Let M denote the topology induced by p and let P be the product topology.
We first show that P_C M. Supposex = (2),22,...) is a point in
Iliez+ Xi and let U = S21(a1) x. --- x S(¢,) x T],,, Xi be a typical
basic P-open set containing x. Let « = min{e,/2,...,€,/2"}. We show
S?(z) C U. Suppose y = (y1, y2,---) € S?(x). Then p(x,y) < € and hence
>a Gi(2i, yi)/2* < €. Therefore, d;(2;, yi) Si, 168 tasked, ae, ond
consequently S?(a) CU.
We now show M C P. Let U = S?(x) be a typical basic M-open set
containing x. Choose N large enough so that 57°, 1/2' < €/2. Then one
can easily check that Sein (a1) X +++ x Sern (Ln) x Ilisnai elles a0)
Inverse Systems: The Preliminaries 157
(6.A.16) Exercise. Show that p is a metric.

(6.4.17) Exercise. Suppose (X1,d1), (Xo, dz)... are complete metric spaces.
Show that ([],;<z+ Xi, p) is a complete metric space.
(6.4.18) Examples.
(a) Uncountable products are useful in building a number of counterexamples
in topology. For instance, the uncountable product of S = {0,1} with itself
I] aera = X, yields a compact (and hence countably compact) space that
is not sequentially compact. A sequence in X that fails to have a convergent
subsequence can be constructed in the following manner.
Represent each element of J by its binary expansion; for each dyadic
rational, choose the binary expansion that ends in a sequence of zeros. For
each n EN, let z, € [] Sq be defined by z,(a@) = n-th digit of the binary
expansion used for a € J. Consider the sequence {z,}, and let {rn,} be
an arbitrary subsequence of {z,}. Let ag € I be any element of J having
no;th digit O and n2;41’th digit 1 for each j, then the subsequence {zp,}
cannot converge, since its projection into the ag-th coordinate space does
not converge.
(b) Observe also that [],-R: Jo is not first (hence second) countable, even
though each factor space is. To see this suppose U;,U2,... is a countable
basis at the point {z,}, where ry = 0 for each a. We may assume that for
each i, U; has the form Uj, x Ui, x --- x Ui,.;) X eer I, (why?). Since
R! is uncountable, there is a 8 € R’ that is not equal to any of the i; for
i€ Zt and 1 <j <n(i). Let Vg be a proper open set of Ig containing 0.
Then p~*(Vg) is an open set in [],¢Ri Ja that contains {z_} but does not
contain any of the U;.
Since [],¢Ri Ja is not first countable, it is not metrizable. Hence, the
product of metrizable spaces need not be metrizable. [Of course, [],cRi Ja
is also not metrizable because it is compact but not sequentially compact.]
(c) The product topology can be used to give an alternate description for
the Cantor set K. Let A = [];2, Xi where X; = {0,2} with the discrete
topology. Define f : A > K by f({2i}) = 92, i /3* (here we consider K
as in (3.B.22)). Since f is 1-1 and onto, it suffices to show f is continuous
(by (2.G.11) and Tihonov’s Theorem (6.A.11)). Let ¢ = )072, 2:/3' be a
point of K and suppose U is an e-neighborhood of c. Choose N large enough
so that >; y2/3' < «. Then the open set {21} x --- x {tw} x [Tiny Xi is
mapped by f into U, which establishes the continuity of f.
B. INVERSE SYSTEMS: THE PRELIMINARIES

Inverse systems have long played an important role in algebraic topology.
Though once regarded as a topological curiosity of rather restricted signifi-
cance, in recent years a variety of significant papers involving such systems
have appeared. We investigate a few basic properties of inverse systems,

and apply these results (in the following sections) to prove, among other
things, the scarcely believable theorem that every complete metric space is
the continuous image of the Cantor set.
(6.B.1) Definition. A relation > directs a nonempty set D if and only if
(i) > is a partial order, and
(ii) for each a and b in D, there is ad € D such that d > a and d > b.
The pair (D, >) is called a directed set. (We will always assume D is infinite.)
Note that property (ii) gives a “sense of direction” to D. When no
confusion is likely to arise, we will shorten the notation and call D a directed
set. In such circumstances it should be tacitly understood that there is a
relation > lurking in the background.
(6.B.2) Definition. Suppose D is a directed set and {X, : a € D} is
a collection of topological spaces indexed by D. Suppose, further, for each
(> a, there is a continuous function fog : Xg + Xq such that
(t) faa is the identity map, and
(it) fopfay = fay whenever 4 > 8 > a.
We call (Xa, fag,D) an inverse system. The maps fag are called bonding
maps, and the spaces Xq the factor spaces of the inverse system.
(6.B.3) Definition. Suppose (Xq, fag, D) is an inverse system. The inverse

limit of the system (Xqa, fag, D) is the subspace Xo of [],¢p Xa consisting
of those points {z, }with the property that za = fag(zg) whenever a, 3 € D
and 3 >a. The elements of the inverse limit are called threads.
(6.B.4) Remark. If D = N (with the usual order), then the system

(Xn, fmn; N) is usually described by (Xn, fn, N), where fn : Xn 3 Xn-1.
In this case, the bonding map fmn is completely determined by the maps f;,
1€.; ieee = Fm+1 a Seet iy:
(6.B.5) Examples.
1. Let D =N and for each n € N let X, = €!. Define each bonding
map f, to be the identity function.
fi fa fs fa
Xo=E} X,=€! Xo=8 X3=€!
2. Let D = N and for each n € N let X,, = Z with the discrete topology.
Define f, by setting f,(p) = p — 1, for each p € Z.
. fi fe g fs M fa
3. Let D = N and for each n € N let X, = N with the discrete

topology. Define f, : Xn 4 Xn-1 by fn(p) =pt+1.
4. Let X be a topological space and let {X, C X : a € D} bea
collection of subsets of X, where D is a set directed so that B > a if and
only if Xg C Xq. Define the bonding maps to be the inclusion maps.
5. Suppose that D = N and let Xo = {0,1}, X1 = {00,01, 10,11},
X2 = {000, 001,010,011, 100, 101,110,111}, etc. Give each X; the discrete
topology. The bonding maps are as indicated in the following figure.
fi fa fs
Xo
(6.B.6) Exercise. Guess the inverse limit in each of the preceding examples.
(Hint: In the last example, try the Cantor set.|
Since Xoo is given the relative topology inherited from |], -p Xa, a basis
for X,. may be defined by taking intersections of members of the basis for
Heep Xa with X... However, an amazingly simple description of this basis is
obtained in the next theorem. If (Xa, fag, D) is an inverse system and X qq is
the inverse limit, then the restriction of the projection pg : Bees Die ee
to Xo is denoted by pa. Theorem 6.B.8 says that the analog of a subbasis
for the product topology is actually a basis for the inverse limit.
(6.B.7) Lemma. If (Xa, fog, D) is an inverse system, then po = fapps
whenever 0 > a.
Proof. If {x} € Xoo, then fasia({tr}) = faa(@e) = 2a = Bal({z,}).

Oo ’
(6.B.8) Theorem. Suppose (Xq, fas,D) is an inverse system and Xo is

the inverse limit. Then B = {p,1(U) : a € D and U open in Xq} is a basis
for -X4;
Proof. It is clearly that we have the elements of B are open in Xq since
p'(U) = pzi(U) N Xe and pz1(U) is a subbasic open set for [],¢p Xa-
Furthermore, it is easily seen that B = {()*_,p
Pa,1(U;) : ke Zt, a; € Dand
U; open in Xq, }forms a basis for Xoo, since gen PRUNE eA wane D
and U; open in Xq; } is a basis for [],¢p Xa. We will show that each element
of B is in B, so that B = B.
Let B € B. Then
k k
Back
= [ Pa, = (i Ui) N Xoo] = Inral) rx,
We re a € D so that a > a; for each i = 1,2,...,k and then define

Wa ‘ee fz (Ui). Clearly V is open in Xg. We bees that
k
Ba (V)==Ba (Afeel vi)= (|2a" faca(Us) = ()BahUi) = B.
t=
Hence BEB. O
(6.B.9) Notation. Suppose (Xq, fag, D) is an inverse system. For each

par.d, 6 in D with a < Bilet Aye dey) © [oem +0 * Sa. = Japa)}~
It is clear that X.. C Ags; usually one would expect Agg to be much larger
than Xo
(6.B.10) Exercise. Show that if X, is a Tz space for each a, then Agg is

closed in [[,
ep Xa and Xq =(){Aag : a,8 € Dia < f}.
Inverse systems are commonly used to create rather exotic topological

spaces. We see examples of this in Section D and in Section E of Chapter 15.
The inverse limit of a system may (and generally does) differ radically from
the factor spaces that make up the system. This may even be the case when
the maps themselves are quite simple. Some results have focused on reducing
the number of distinct bonding maps while maintaining uncomplicated factor
spaces. For instance, the pseudo-arc (an unusual space defined in Chapter
9) can be viewed as the limit of the system (Xp, fn, N), where each X,, is
the interval J and all the f,’s represent the same map (Henderson [1965]).
Of course, the inverse limit of a system may be rather mundane as well;
in fact, the limit space may be empty even though the bonding maps are
embeddings (6.B.5, example 3). However, we see next that the inverse limit
cannot be empty if the factor spaces of the inverse system are compact and
To.
(6.B.11) Theorem. Suppose (Xq, fag, D) is an inverse system, where each
Xq is compact and T2. Then X. is nonempty, compact and T».
Proof. We use the finite intersection property characterization of com-
pactness to find a point in X.. Let A = {Agg : a < B}. Then JA is a collec-
tion of closed subsets of the compact space [],¢p Xa (6.B.10). We show that
A has the finite intersection property. Suppose Aq,g,,---,Aa,, is any finite
subcollection of A. Since D is directed, there is a 1 € D such that » > ;
for? = 1,2,...,n. Pick an arbitrary point 7, € X) and fori = 1,2,...,n
let ta; = fa;(Z,) and zg, = fg,,(). Define a point {ye} € [xen Xa
by Settingsya; = wa and wei rgnior/ += 1,2, ..., nvyt=iey; and let
ting Yq be arbitrary for the remaining coordinates. Then {ya} € ()j_, Aa:a;
since fais: (Ya;) = Faia; (p;) = faa: fa:r(£a) = foir(Z,) = La; = Yai:
Thus, A has the finite intersection property. It follows from (2.H.2) that
Na<p 40g # 0. However, by (6.B.10), we have (1),
<gAag = Xoo. Further-
more, Xo is compact, since it is a closed subspace of the compact space
Haep Xa. That X is Tz comes from problem 13 of Sections Band C. O
(6.B.12) Exercise. Show that if (Xa, fag,D) is an inverse system such
that each fag is 1-1 and onto, then py is 1-1 and onto.
(6.B.13) Exercise. Show that if (Xn, fn, N) is an inverse system such that
fn is onto for each n, then py is onto for each n.
Complications arise with regard to connectedness. The following exam-

ple illustrates that the inverse limit of a system of connected spaces need not
be connected.
(6.B.14) Example. Let A, = {(x,y) € E€? : -n< a <n, y = 0} for
each positive integer n, and let X, = €* \ An. Define an inverse system
(Xn, fn, Zt), where each f, is the inclusion map. It is easy to verify that
the inverse limit is homeomorphic to the union of the sets A = {(z,y) € E? :
y > O} and B= {(z,y) € E? : y < O}.
(6.B.15) Theorem. Suppose (Xq, fag, D) is an inverse system of compact,
connected JT» spaces. Furthermore, assume fq is onto for each a € D. Then
Xoo is connected.
Proof. We show: if X. = AUB, where A and B are closed nonempty
subsets of X,., then ANB #9. Since X. is compact, so are A and B, and
therefore for each a, fa(A) and pa(B) are closed subsets of Xq. However, pa
is onto, and consequently, Xq = pa(A)Upa(B). Let Ya = Pa(A)NPa(B) and
note that Y, cannot be empty since Xq is connected. Define gag = fagly,-
Then gog maps Yq into Yq, since gaa(Y¥a) = fap(Ys) = faa (ba(A)NPa(B)) C
fopPp{A) NM fapPe(B) = pa(A) N pa(B) = Ya:
Each Y, is compact and T> so (6.B.11) applied to the inverse system

(Ya; 9ae,D) gives a point {Ya} € Yoo. Since Pal{Ya}) € Pa(A) A Po(B) for
each a, it follows from the next exercise that {yz}E ANB. O
(6.B.16) Exercise. Suppose (Xa, fag,
D) is an inverse system, {Za} € Xoo,
and A is a closed subset of Xo. Show: if fa({Za}) € fo(A) for each a € D,
then {rq} € A.
C. COMPACT METRIC SPACES ARE

CONTINUOUS IMAGES OF THE CANTOR SET
We saw, (6.B.5) and (6.B.6), that the Cantor set may be considered as
the inverse limit of a countable number of finite discrete spaces. We will
presently see that any totally disconnected compact metric space may be so
represented. In fact, inverse systems can be employed to show all perfect,
totally disconnected compact metric spaces are homeomorphic to the Cantor
set.
(6.C.1) Theorem. If X is a totally disconnected compact Tz space, then

the family of open and closed subsets of X forms a basis.
Proof. Let x € X and O be a (proper) open set containing x. Then {zr}
and X \ O are disjoint closed subsets and there is no connected set C' that
intersects both {x} and X \ O. By (4.A.11), there is a separation (C1, C2)
of X such that {x} C C,; and X \O C C2. Then {x} C Ci C O, and C, is
open and closed. O
(6.C.2) Definition. Suppose X is a metric space and U is a cover of X.

The mesh of U is defined to be the least upper bound of the diameters of
the sets in U.
(6.C.3) Definition. Suppose U and VY are covers of a space X. Then V is

a refinement of U if and only if for each V € VY, there is aU € U such that
V CU. If V refines U, we write V < U.
(6.C.4) Theorem. Suppose X is a totally disconnected compact metric

space. Then there is a sequence of finite covers U,l,,... of X such that
(7) Ug > Uy, >:::,
(it) meshU,, < 1/n, (meshl < 2).
(1i7) all members of the covers are open and closed, and
(iv) for each n, if U,V €U, andU #V, then UNV = 9.
Proof. We first construct Up, By (6.C.1), for each x € X, there is an
open and closed set U, of diameter less than 2. The family of U,’s covers
X, and since X is compact, there is a finite subcover {Uz,,Uz,,.. Eta &
These sets are not necessarily disjoint, but that may be corrected by setting
Compact Metric Spaces are Continuous Images of the Cantor Set 163
OUR (US U;). An inductive procedure for completing the proof

jJ=
should now be clear. O
(6.C.5) Theorem. Suppose X is a totally disconnected compact metric

space. Then there is an inverse system (Xn, fn, N) such that for each n, X,
is a finite discrete space and X. is homeomorphic to X.
Proof. Let Uo, l4,... be the sequence of open and closed covers obtained
in the previous theorem. For each n, let X, = U, and give X,, the discrete
topology. Define the bonding maps f, as follows. If Un € Un, let fn(Un) =
Un—1, where U,_; is the unique member of U,_; such that U, C Un-1.
We now define a homeomorphism h from X onto X... Suppose x € X.
For each n, let U, be the unique member of U,, that contains z. Observe
that Up D U; D U2 D --- and, consequently, (Uo, Ui1,...) € Xoo. Define h by
h(x) = (Uo,Ui,...). The map h is onto: Suppose (Vo,Vi,...) € Xoo. Then
for each 1, we have V; D V;41, and since for increasing i, diam V; approaches
0, it follows from (3.4.11) and (3.B.8) that ();2, Vi consists of a single point
z. Clearly, h(z) = (Vo,MWi,...). Now, for distinct points x and y there are
distinct sequences Up D U; D--- and Vo D Vi D --- that converge to x and
y, respectively, and hence f is 1-1.
It remains to show that h and h7! are continuous. Since X and X.
are compact T> spaces, it suffices to show that h is an open map (why?).
Clearly B= {U : U €U,, n EN} is a basis for X. Note: if Un € Un, then
WAU3) = (ia == Xa ek {UK Xn * 7) Xo, which is obviously
openin Xy,. O
We next describe a ‘function’ from one inverse system to another. This

is a family of maps between the respective factor spaces paying due respect
to the inverse system structure.
(6.C.6) Definition. Suppose A = (Xa, fag,D) and B = (Ya,9ag,D) are

inverse systems with the same directed set D. Suppose further for each
a € D there is a continuous map ha : X, — Yq such that the following
diagram is commutative whenever { > a.
5i ES Gt
We hg
Yu
a Jap
Ye
The family of maps H = {hq : a € D} is called a map between A and B.
(6.C.7) Definition. Suppose H is a map between inverse systems A and

B. Define [] ha : Taep Xa > Hap Ya by [[hal{ta}) = {Ra(ta)}. Then
hoo = |] hax. is called the limit of H.
(6.C.8) Theorem. Suppose H is a map between inverse systems (XenfessD)

and (Yo,9ag;D). Then the limit of H, ho, maps Xo into Yo. and is con-
tinuous.
Proof. Suppose {za} € Xoo, and let {ya} = hao({ta}). If a < B,
then it must be shown that gas(ysg) = Ya. However, this follows easily,
since gag(¥e) = Japha (zg) = hefae(ze) = he(Za) = Ya, and hence {Ya}
is a thread in Y... Clearly, hoo is continuous, since it is the restriction of a
continuous map. O
(6.C.9) Theorem. Suppose H = {ha : a € D} is a map between the

inverse systems (Xq, fag, D) and (Yo, gap, D).
(i) If he is 1-1 for each a € D, then Ngo is 1-1.
(ii) If D=N and each hg is 1-1 and onto, then hq is 1-1 and onto.
Proof. (i) This is immediate from the definition of h.,., and problem A
WP.
(ii) To see he is onto, let {yn} € Yoo and for each n, set tpn = hz! (yn).
Then we have that hyj(Zn) = yn and it follows easily from the commutativity
of the following diagram that {x,,} is a thread. Hence ho({Zn}) = {yn}.
Xo Xi X29 X3+—
ho oil he h3
Yo Yi Y2 ee Oo
We now show that all perfect, compact, totally disconnected metric

spaces are homeomorphic. We begin with the following lemma, which may
be established using (6.C.1) and an easy induction argument.
(6.C.10) Lemma. Suppose X is a perfect, compact, totally disconnected

Tz space and n is a positive integer. Then X can be written as a disjoint
union of n open and closed (nonempty) subsets. O
(6.C.11) Theorem. If X and Y are totally disconnected, compact, perfect

metric spaces, then X and Y are homeomorphic. (Hence, X and Y are
homeomorphic to the Cantor set.)
Proof. By (6.C.4) and the previous lemma, there are covers Uo and Uo
of X and Y, respectively, that satisfy the following conditions:
(i) the members of each cover are open and closed;
(it) the elements of each cover are mutually disjoint;
(iti) the mesh of each cover is less than 2;
(iv) Up and Uo have the same number of elements.
Let ho by any 1-1 map from Up onto Uo. Apply (6.C.4) once again to obtain
refinements U4, and UW; of Uo and Ves respectively, whose meshes are less
Compact Metric Spaces are Continuous Images of the Cantor Set 165
than 1 and which also satisfy conditions (i) and (ii) above. Lemma (6.C.10)
allows us to assume that if Up € Up, then Up and ho(Uo) both contain the
same number of elements from U; and U4, respectively. Hence, there is a
1-1 function h; from U4, onto U/; with the property that if Uy; € U,, Uo € Uo,
and U; C Uo, then hi(U1) C ho(Uo). This procedure can be repeated in the
obvious inductive fashion to obtain two inverse systems and a map between
them, as indicated in the following diagram. Each U,, is assumed to have the
discrete topology and each f; and fi is defined as in (6.C.5).
Uo U; Up Us
ho hy ho hg
4 fe 2 fs 2
It is clear, (6.C.8) and (6.C.9), that hg is a homeomorphism. By
(6.C.5), X is homeomorphic to Uz and Y is homeomorphic to WU. O
We now prove one of the most startling and least intuitive results in all
of topology.
(6.C.12) Theorem. Suppose (X, d) is a compact metric space. Then there

is a continuous function f mapping the Cantor set K onto X.
Proof. Since X is compact and metric, it is easy to construct a sequence
of finite covers Up, U4,,U2,... of X with the following properties:
(i) each U € U, is the closure of an open set;
(ii) if U € Up, then diamU < 1/2";
(iit) U, refines U,-1.
Let Up = {Uo1,U02,...,Uor,}. Members of Uo are “disjointified” by
means of the trick of letting Vo; = Uo; x {i} for 1 <i< ko. A subset Ax {i}
is to be open in Vo; if and only if A is open in Up;. Let Vo = Ue—1 Voi and
assign the disjoint union topology to Vo.
Notation soon becomes a problem, but if the reader keeps alert it will
help. Let Uy = {Ui1,Ui2,...,Uix,}. For each Uj; € U,, choose an integer
i such that Uj; C Uo. Again we “disjointify.” Let Viij = Uij x fo pexeta}
(where i is selected so that Ui; C Upi), and let V; be the (disjoint) union
of these sets with the disjoint union topology. Define f; : Vi — Vo by
fi(u,i,7) = (u,7); then f; is obviously continuous. It should now be obvious
what is afoot, but in the interest of clarity we proceed one step further.
Let Uz = {U21, U22,...,U2k,}. For each U2, € U2, choose an integerJ
such that U2, C U;;. Then for each “selected triple” U2, C U1j C Uoi, define
a set Voijx = Urn x {i} x {j} x {k}. Assign to V2ij, the obvious topology and
let V2 be the disjoint union of all these sets, again with the disjoint union
topology. Define a map fz : V2 > Vi by fe(u,i,j,k) = (u,t,j). Continuing
in this manner, we obtain an inverse system (Vn, fn, N).
A second inverse system (Xn, 9n,N) is constructed by defining X, = X

for each n and setting the bonding maps gy equal to the identity. A map
between the two systems is described as follows. For each n € N, let hy :
V, > Xn be defined by hn(u,i,j,...,p) = u. Thus we have the following
commutative diagram.
yy (ei is
ho hy ho hg
Xo Xi Xo X3+~—
n 92 93
Denote the respective inverse limits by Vi. and Xo, and the limit of
the hn by hoo. Clearly, hoo is onto, Voo is a compact metric space and X is
homeomorphic with X... We now investigate properties of Voo.
We first show V,, is totally disconnected. Suppose x = (Zo, %1,...) and
y = (Yo, ¥1,---) are distinct points of Vo. We are looking for an open and
closed set in V, that contains z but not y. Since x and y are distinct, there
is a first integer n such that rp # yn. Suppose Zo # Yo, where rp = u x {i}
and yo = t x {i}. If u # G, then an integer n may be chosen large enough
so that any two members of U, that contain z and y respectively must be
disjoint. Suppose u € Uny and t € Ung. Then the corresponding sets Vji;...p
and Vpjij...¢ are disjoint and both open and closed. Furthermore, Vij...) is
pulled back by p;' to an open and closed set in V.. which contains x but
not y; thus x and y have been separated in Vo.
If u = t, then 2 ~ ¢. Therefore, we have zo € Vo; and yo € Voz;
furthermore Vo; and Vo; are disjoint, open, and closed. Again, p—'(Vo;) is an
open and closed subset of V. containing x but not y.
If zo = yo, let m be the first integer such that zm 4 Ym. Suppose, for
instance that m = 2. Then since the case u # % can be handled as above,
we have z2 = u x {i} x {j} x {k} and yo = ux {i} x {j} x {k}, where
k Ak. Thus, to € Voij, and also yo € Vaijh> where Voi;, and Voisk are
disjoint, open, and closed. Then p~*(V2;;,) is an open and closed subset of
Vo. containing x but not y. Hence, Vx is totally disconnected.
Although V,. may not be perfect, it follows from (3.B.20) that Vi. x K
is perfect. Therefore V.. x K is homeomorphic to K (6.C.11, 6.C.13), and
hence the following maps exist: a homeomorphism H : K + (V. x K), a
continuous surjection q : (Voo XK’) > Voo defined by q(v, y) = v, a continuous
surjection hg : Voo —- Xoo, and a homeomorphism Y : X, ~ X. Then
f = WVhe¢d maps K-ontock.
(6.C.13) Exercise. Show that V., is metrizable.

The Cantor set may be generalized as follows. A product space [ive Weer
is a Cantor space if and only if for each a € A, Xq = {0,1} with the discrete
The Dyadic Solenoid 167
topology. It can be shown that every compact T> space is the continuous
image of a closed subset of some Cantor space (see Aleksandrov and Uryson
[1929], for example).
D. THE DYADIC SOLENOID

We close this chapter by giving in some detail the construction of a curious
space, the dyadic solenoid. We first provide a geometric description of this
space, and then show how the geometric version can be realized as an in-
verse limit. The latter characterization is both neater and somewhat more
manageable.
(6.D.1) Definition. A solid torus is any space homeomorphic to S! x B?.

The standard solid torus in €? is obtained by rotating the disk in the yz
plane with center (0,2,0) and radius 1 about the z axis.
The solenoid is constructed by successively intertwining solid tori inside

of solid tori. In Euclidean 3-space, one starts with a standard solid torus
T°. In the interior of this torus, another solid torus T! is wrapped around
twice (see the figure that follows). Then in the interior of the second torus
a third one is wrapped around in precisely the same manner in which T!
was wrapped around inside T°. This procedure is repeated for each positive
integer n, and the solenoid is defined to be the intersection of all the tori.
It is assumed that the diameter of the cross sections of the tori approaches
0 with increasing 7. In spite of the decreasing diameters, the intersection of
the tori is nonempty (4.4.8). The successive tori T°,T?,... yield an inverse
system (7, fn, N), where the f, are simply inclusion maps.
(6.D.2) Exercise. Show that the inverse limit Tos ot (1, fas) is home-
omorphic to D = (\2, 7".
We next investigate an inverse system, whose limit corresponds to the
solenoid D = (\%2, T’, that proves to be more useful than the one above.
Suppose that (Cn, 9n,N) is the inverse system where, for each n, Ope
is the unit circle represented in polar coordinates by {(r,@) : r = 1}, and
Gn 2 Cn —> Cn-1 is defined by g,(1,0) = (1,28).
61 (p)
We construct a map between the systems (T”, fn,N) and (Cn, gn,N) as
follows. Let p € T° and let Q be the plane containing the z axis and p.
Define 09(p) to be the angle measured counterclockwise in radians between
the positive « axis and Q?. We assume the range of 4 is [0,27). Now
suppose p € T!. Let Q) be the plane passing through the z axis and p.
Define 6,(p) to be the angle measured counterclockwise in radians formed
between the x axis and Qh subject to the following conditions:
(i) if 6;(p) = 0, then so does 69(p);
(ii) the range of 0, is [0,47), and 6, is continuous on T! \ 6;1(0);
(iii) 61(p) = 1 (H) if and only if Q; = Qj and p and # belong to the
same component of QT" (see the preceding figure).
If p € T?, then let Of be the plane containing the z axis and p, and
define 92(p) as the angle measured counterclockwise in radians between the
x axis and Ope where 62 satisfies the conditions:
(i) if 62(p) = 0, then so does 6; (p);
(ii) the range of 2 is [0, 8m), and 62 is continuous on T? \ 651 (0);
(iit) O2(p) = 42(p) if and only if Q? = Q3 and p and # belong to the
same component of Q2 NT”.
Similar functions are defined for each positive integer n. Note that 6n(p)
either equals 6,_1(p) or equals 6,_1(p) + 2"7. Define a map h, : T" + C,
by hn(p) = (1, 6n(p)/2”). Then hp is clearly continuous, and hence we have
the following diagram.
The Dyadic Solenoid 169
ho h1 h2 hg
In order to show that the diagram is commutative, we consider two

cases,
(a) Suppose 6,(p) = 6n—1(p). Then
Jnhn(p) SUBeae iz(vt

a)
re = (, 22) = Ten
1 ep):
(b) Suppose 6,(p) = On— 1(p) + 27. Then
Gnlin(p) = Gn (1=P) = (1a)

is (1,222) i) rf (1,20) +2n).
Dias Qn-1
However,
hn fa(p)= (1,SEP?)
and since scurrying around the unit circle 27 radians lands you back at the
same point, the commutativity of the diagram is established.
We show h is 1-1 and onto: Suppose t = (p,p...) and t = (p,p,...) are
distinct points in T,. Then clearly ho(p) 4 ho(p), and hence hoo(t) # hoo(t).
Therefore h is 1-1. In order to show hq is onto, let c = (c9,c1,...) € Coo,
where ¢, = (1)%;,) foreach nm. Note that r, =7p,=1/2 of Tr = (tye /2) + 9,
For each, tet o;,-= 2". hen eithera,, = f_2q00l Sp yoy ee e. Let
Hy, = 6751(an), i.e., Hn is the appropriate component of the cross section of
T,, determined by the plane that contains the z axis and forms an angle of
Zn radians with the z axis. Readers can convince themselves that the point
t = (p,p,...), where {p} = ()7—_, Hp, is mapped by hoo onto c.
Recall that a topological space X is homogeneous if and only if, given
any two points z and y in X, there is a homeomorphism h : X — X that
carries z to y. Note how the use of inverse limits yields an easy proof of the
following theorem:
(6.D.3) Theorem. The dyadic solenoid is homogeneous.
Proof. We use the above notation, where the solenoid is considered to
be Css. Let x = (49,21,...) and y = (yo,41,...) be points in C,,, where
Zn = (1,Qn) and yn, = (1, Bn). Consider the mapping system
Co = C1 = C2 = Co
ko ki ko k3
Co sangs C1 more E C3
where k,(1,9) = (1,8 + Bn — Qn). It is a routine matter to check that the
diagram commutes and that each k, is a homeomorphism. Thus, by (6.C.8),
(6.C.9), and (2.G.11), koo is a homeomorphism, and a direct computation
shows that k..(z)=y. O
More results involving inverse limit systems will be obtained in Chapter
9. At this point, the following articles should be well within the grasp of
the rara avis that is interested in some curious developments in inverse limit
theory. Brown [1960], Jolley and Rogers [1970], Nadler [1970], and Kresimar
and MardeSi¢é [1968].
PROBLEMS
Sections A
1. Show that the projection maps of [[,<, Xa are open, but need not be
closed.
2. Show that [],<, Xa has each of the following properties if and only if
each X, does:
(a) To,
(b) Ty,
(c) T,
(d) regularity,
(e) complete regularity
3. Let {Xq : a € A} be a family of topological spaces, and for each a let
B. be a basis for Xq. Describe a basis for [[,<, Xa in terms of the
Bas:
4. Show that [].,<, Xa is first (second) countable if and only if each Xq is
first (second) countable and all but a countable number of the X, have
the indiscrete topology.
5. Let P be a topological property and X a nonempty set. Then a topology
U on X is maximal with respect to P if and only if whenever V is a
topology for X and U G V, then V does not have property P; UW is
minimal with respect to P if and only if whenever V is a topology for
X and VY G U, then V does not have property P. Show that: (a)
Problems il
the product of a family of maximally compact spaces is maximal with

respect to compactness; (b) the product of a family of minimally T>
spaces is minimal with respect to T>.
A space X is super compact if and only if X has a subbasis S with the
property that any cover of X by members of S has a subcover consisting
of just two sets.
(a) Show that I is supercompact; show that a triode is super compact.
(b) Show that any super compact space is compact.
(c) Show that the product of super compact spaces is super compact.
(d) Find a compact space that is not super compact.
Suppose X, and X» are nondiscrete T; first countable spaces. Show
that if po : X; x X2 — X-2 is closed, then X, is sequentially compact
(p2 is the projection map).
Show that sequential compactness is preserved under continuous maps.
Let X = [er Ja, and let Y = [0,2) with the order topology. Show
that Y is countably compact and X x Y is countably compact, but that
X x Y is neither compact nor sequentially compact.
10. Suppose {X, : a € A} is a family of topological spaces, X is a topo-
logical space, and for each a € A, fy : X — Xgq is continuous. Show
that there is a unique continuous function f : X > [[,¢, Xa such that
Pof = fa-
11. Suppose {X, : a € A} is a family of spaces, and Y is an arbitrary
space. Let {qa : Y 4 Xq : a € A} be a family of continuous open
maps from Y onto X,. Suppose for each topological space X and for
each family of continuous maps {fa : X + Xq : @€ A} there is a
unique continuous map i : X — Y such that qaf = fa for each a € A.
Show that Y is homeomorphic with [],¢, Xa-
12, Let {Xq : a € A} and {Y, : a € A} be two families of topological
spaces and let {fa : Xa — Ya : a € A} be a family of continuous func-
tions. Show that there is a unique function f : [],¢, Xa > []aca Ya;
such that the following diagram commutes for each a, and f is contin-
uous. Furthermore, show that if each fq is 1-1 (or onto), so is ifs
Yo fe [Teen Yo
(Hint: Use problem 10.]

13. Suppose {X,. : a € A} is a family of topological spaces, A = Ay U Ag,
and Ay MAg = 0. Show that [T],<, Xo and ([Tycq, Xe) X (Teen, Xe}
are homeomorphic.
ie 6. The Product Topology and Inverse Systems
14, Show that the Tihonov theorem does not Aes if []acq Xa is given the
box topology.
15. Suppose {Xq : a € A} is a family of topological spaces. Show that
Toca Xa i8 separable if and only if each Xq is separable and all but at
most 2%° spaces are a point (2° is the number of functions from Z into
the set {0,1}). [Hint: See Comfort [1969].
16. Show that II I, is compact but not sequentially compact.
ael
Sections B and C
Show that every totally disconnected compact metric space is homeo-

morphic to a subset of the Cantor set.
Show that there is a subspace of the irrationals in [0,1] that is homeo-
morphic to the Cantor set.
Suppose (Xq, fag, D) is an inverse system such that each Xq is compact.
Show that pa(Xoo)= ese fap(Xa).
Let D ={A ¢c RR’; Ais finite}. For each Ae D, let X4 =
A — Z : i is 1-1} have the discrete topology. Order D by A < B if
and only if A Cc B. Observe that X44 is countable, and if A < B, then
fap: Xp — Xza defined by fas(g) = gja is onto. Find the inverse limit
of (Xa, fas, D).
Suppose (Xq, fag, D) is an inverse system of compact T> spaces. Show
that if X. has no more than k components for each a, then X, has no
more than k components.
Show that J is the inverse limit of an inverse system where all of the
Nas ale o>:
Suppose (Xq, ieDyis an inverse system and A C X is compact. For
each a, let Ag = 6y'(A), and A* = [],¢p Aa. Define gag= fap\Ag-
(a) Show that (Aq, gag, D) is an inverse system.
(b) Let A. be the inverse limit of (Ag, gag,D). Show that A, =
Ailes SA.
Let (Ya, fag, D) be an inverse system of compact T2 spaces. Suppose
each fog is monotone (a function f : X + Y is monotone if and only
if f~*(y) is connected for each y € Y). Show that for each a € D,
a! Xoo + Xq iS monotone.
Suppose (Xq, fas, D) is an inverse system and there is an integer k such
that card Xq < k for each a. Show that card Xx < k.
10. Is the hypothesis “pg is onto” necessary in (6.B.15)?
Problems 173
11. Find examples which show that the hypotheses of (6.C.9) are necessary.
12. Consider the set (no topology) X = ][ {0.1}. Show that it is not
Tel
possible to impose metric topology on X so that X is compact.
13. Suppose (X., faZ, D) is an inverse system and X,. # @. If each XQ is
To, respectively T), respectively T>, respectively T3, respectively T3312,
then so is Xq.
Section D
1. Consider the Cantor set as an inverse limit and show that it is homoge-
neous.
*. ma a
OE Med die astond> cartUE hel

Romiavee nA
geren
ig tevengen iTevita |
| so laeeal - ¢
ices:end Seer _ “6 noisee2
» efites © mA % Pe
Aaystiathah37 2ade-qrosle bite Abin emere! ponagtoe pote’) ads okt) |p
‘ —) CO eta) 62699 M4 OG cow Poe hee :
7 : >> Gant Ci '3Ue © ne es
7 F
,
» = ~~ € oa -® co=—ay Mi 7 < - a
= a _ a1 - oe
7 ad - -
: ’ in Ftd Ay Sc7AAG ret ae

aa “
a = : i
———_ 2
2@ os Oe ho ae | a» r i¢
, Gm ie oe.) ete Oe Ss ~* .
7 7 vee Sy 7 <7 wae
= -_ a0bw > sell
= LY 2
ice A ' ‘
-»
a fies :
= a
on -
.
an o
a q
.
-~ . ‘
—_
t=
Chapter 7
FUNCTION SPACES, WEAK

TOPOLOGIES AND HILBERT SPACE
Inverse systems gave some rather amazing examples of subspaces of product

spaces. Less exciting, but probably more ubiquitous are two further offshoots
of the product topology: function spaces and the weak topology. Function
spaces are essential in the study of both topology and analysis, and one
version of the weak topology may be regarded as generalizing certain basic
properties of the product topology. Hilbert space represents an extraordi-
narily rich topological space, one whose properties are still the subject of
substantial topological research.
A. THE POINT-OPEN TOPOLOGY

Product spaces serve as a convenient starting point for the study of function
spaces. The “points” of a function space consist of the members of a family
of functions. Various topologies may be given to such collections. If X and
Y are topological spaces, then Hom(X,Y) will denote the set of all (not
necessarily continuous) functions from X into Y. One way to topologize
Hom(X,Y) is to observe that each of its elements may be regarded as an
“X-tuple,” i.e., Hom(X,Y) = [],¢x Ye, where Y, = Y for each x; thus, we
may (and do) assign the product topology to Hom(X, Y).
In general, a thorough understanding of Hom(X,Y) with the product
topology lies in clearly understanding the subbasic sets. Note that with
each point « € X and each open set U C Y, p;'(U) can be described as
S(z,U)={f:X ~Y : f(t) €U CY,}. All sets of this form give the
usual subbasis for Hom(X,Y). Since such sets depend on the point x and
the open subset U, the product topology, when imposed on Hom(X,Y), is
called the point-cpen topology.
175
176 7. Function Spaces, Weak Topologies, and Hilbert Space
(7.A.1) Exercise. Show that subbasic sets for the product topology on
Hom(X,Y) are of the form S(z,U) ={f:X + Y : f(x) € U}, where U is
an open subset of Y and ze X.
The projection maps retain their importance in the context of function

spaces, but undergo a change in notation and terminology. For each x € X,
let e, : Hom(X,Y) — Y, denote the natural projection defined by e2(f) =
f(z) where f € Hom(X,Y). Thus e, evaluates a given function f at the
point z. In this context, the projection maps e,; are called evaluation maps.
The point-open topology is the smallest topology for which the evaluation
maps are continuous (6.A.3). Of course they are also open (and usually are
not closed). Evaluation maps will play a role in a number of subsequent
proofs.
The point-open topology is frequently given yet another name: the topol-
ogy of pointwise convergence. The next theorem motivates this terminology.
(7.A.2) Theorem. Suppose Hom(X, Y) is given the point-open topology,

{fn} is a sequence of points in Hom(X,Y), and f € Hom(X,Y). Then the
sequence {f,} converges to f if and only if {f,(z)} converges to f(x) for
each x € X.
Proof. Suppose {f,} converges to f and x € X. If U is an open set
in Y containing f(z), then f € e;'(U). The sequence {f,} converging to f
implies there is a positive integer N such that f, € e;!(U) for n > N. This
in turn implies f,(z) € U forn > N. Hence {f,(x)} converges to f(z).
Conversely, suppose {f,(x)} converges to f(z) for each € X. Consider
an arbitrary subbasic open set S(z,U) = {g: X > Y : g(t) EU C Yz}
which contains f. Since {f,(x)} converges to f(x), there is a positive integer
N such that f,(z) € U for n > N. Consequently, ifn > N, then fp €
S(z,U), which completes the proof. (Why is it sufficient to consider only
subbasic open sets in the proof?) O
B. THE COMPACT-OPEN TOPOLOGY

One unsatisfactory aspect of the point-open topology is that it does not in
any way reflect the topology of X. Hence, it is quite possible that a sequence
{fn} of continuous functions in Hom(X,Y) may converge to a function f in
Hom(X,Y) that is not continuous.
(7.B.1) Exercise. Suppose X = Y = [0,1]. Find an example of a sequence
of continuous functions in Hom(X,Y) that converges to a function that is
not continuous.
To avoid this situation, we generalize the construction that yielded the

point-open topology.
The Compact-Open Topology i727
(7.B.2) Definition. Suppose X and Y are topological spaces. For each

compact subset A of X and each open set U of Y, define
S(A,U) = {f € Hom(X,Y) : f(A) CU}.

The compact-open topology for Hom(X,Y) is the topology with subbassis
S = {S(A,U) : A compact in X and U open in Y}.
Although the definition of the compact-open topology may appear to be
somewhat arbitrary, this topology is perhaps the most important and useful
of the function space topologies in analysis. Furthermore, as we shall see,
(7.B.11), it is a natural generalization of another reasonable topology. One
should note that the point-open topology is a subset of the compact-open
topology. Hence, if the evaluation maps e, are defined as before, continuity
of each e, is immediate.
The following is an easy but mildly interesting exercise.
(7.B.3) Exercise. Suppose Hom(X,Y) has the compact-open topology.

a) Show that Hom(X,Y) is To, Ti, or T> if and only if Y is.
b) Show that e, is open. [Hint: What is the relation of the compact-open
topology to the box topology?]
(7.B.4) Notation. Suppose X and Y are topological spaces. We denote
{f:X —-Y : f is continuous} by C(X,Y).
(7.B.5) Exercise. Let X be a T; topological space with a subbasis S. Show
that X is T3 if and only if for each S € S and for each z € S, there is an
open set U such that r EU CU CS.
(7.B.6) Theorem. Suppose Y is 73 and X is any topological space. Then
the compact-open topology on C(X,Y) is T3.
Proof. Clearly C(X,Y) is T;. (Why?) By exercise (7.B.5) it suffices
to show that if A is a compact subset of X, V is an open subset of Y, and
g€W =S(A,V)NC(X,Y), then there is an open set U in C(X,Y) such
that g€ U CU C W. For each z € A, we have g(x)€ V. Since Y is regular,
there is an open set V, in Y such that g(x) € Vz C Vz C V. Then the family
{V, : « € A} is an open cover of the compact set g(A). Let {Vz,,...,Vz,}
be a finite subcover. We will show that U = S(A,U;_, Vz;) AC(X,Y) is
the desired open set containing g and with closure lying in W. Certainly
g € U. To see that U Cc W we show es k ¢ W implies k ¢ U. But
we first note that Vz; C Uj, Ve; = Uji, Ve; C V, which in turn implies
{h € C(X,Y) : h(A) C UR, Ve} CW. k ¢ W,
If & € C(X,Y) and
definition of S(A, vy)there is an a € A such that k(a) ¢ V, so
then by thede
k(a) ¢ Ur,Vo; Hence koe tke C(X,Y)-: k(e)ie Y \ Ujen Vex } =U,
which is an open set in C(X,Y) disjoint from U (if k eUN U, then k(a) €
(U2 AV. 1a (v Vs V2) =). Thus k ¢ U, and consequently UcW.

Hence we haveg EU CU CW. oO
(7.B.7) Definition. Suppose X is a set and (Y,d) is a metric space. Then

a function f : X — Y is bounded if and only if diam f(X) is finite.
(7.B.8) Notation. Suppose X is a topological space and (Y, d) is a metric
space. We denote {f € C(X,Y) : f is bounded} = {f € Hom(X,Y) : f is
continuous and bounded} by B(X,Y).
Another topology, the sup topology, is frequently assigned to sets of
functions. The sup metric is used, e.g., in the Weierstrass Approximation
Theorem of analysis.
(7.B.9) Definition. Suppose X is a topological space and (Y, d) is a metric
space. The sup topology for B(X, Y) is the topology generated by the metric
D(f,9) = sup{d(f(x),g(x)) : « € X}.
(7.B.10) Exercise. Show that D is a metric.
The versatility of the compact-open topology is illustrated in the follow-
ing theorem.
(7.B.11) Theorem. If (Y,d) is a metric space and X is a compact topologi-
cal space, then the sup topology on C(X, Y) coincides with the compact-open
topology.
Proof. Let JT be the compact-open topology for C(X,Y) and let S be
the sup topology. We first show that J C S. (Compactness of X is only
needed in this part to guarantee that C(X,Y) = B(X,Y).) Suppose S is
a subbasic open set of 7, ie., S = S(A,V) MN C(X,Y), where A is compact
in X and V is open in Y. Let g € S. It suffices to find an e€ > 0 such
that S.(g) C S. Since g(A) is compact in Y, there is an € > 0 such that
d(g(A),Y \V) =e. If h € S,(g), then d(g(x), h(x)) < € for each x € X, and
hence h(x) € V for each « € A. Consequently h € S and we have shown
Se(g) CS.
For S C 7, suppose S,(f) is a basic open set in S. We find a finite
number of subbasic open sets S;,...,Sp ofT such that f € ()_, Si C S.(f).
Since X is compact, so is f(X), and therefore there are a finite number of
sets S.6(f(21)),---,Se/e(f(an)) that cover f(X). For each 1,1 <i <n, let
Ci = f~*(Seje(f(xi))) and V; = S./3(f(ai)). Then C; is compact, since it is
a closed subset of the compact space X. Finally, let S$; = S(C;, V;)NC(X,Y).
It is clear that f € ();_, Si, and hence it remains to show (\j_, S; C S.(f).
If g € fj, Si then g(C;) C Vj for each i. Let x € X, and let Cj be a
member of the cover {C,...,Cn} of X that contains z. Then both g(x)
and f(z) are in V;, and consequently, d(g(z), f(x)) < 2€/3. Hence, we have
sup,cx{d(f(x), 9(x))} < and therefore g€ S.(f). O
The Compact-Open Topology 179
The sup topology (or equivalently the compact-open topology on C(X, Y))
when X is compact and Y is metric) is frequently referred to as the topology
of uniform convergence. The following exercise gives the motivation for this
terminology.
(7.B.12) Exercise. Suppose X is a compact space and (Y,d) is a metric
space. Show: if {fn} converges to f in C(X,Y) (with the sup topology),
then {fn} converges to f uniformly, i.e., given « > 0, there is an N € Zt
such that if n > N, then d(fn(x), f(x)) < € for alla € X.
(7.B.13) Definition. Suppose X and Y are topological spaces. The func-
tion e : Hom(X,Y) xX — Y defined by e(f,x) = f(z) is called the evaluation
function associated with Hom(X,Y).
Note that e(f,z) = e:(f), where e, is the evaluation map discussed
previously. Our principal result involving e is the following.
(7.B.14) Theorem. If X and Y are topological spaces and X is locally

compact, then e : C(X,Y) x X — Y is continuous, where C(X,Y) has the
compact-open topology.
Proof. Suppose g € C(X,Y) and x € X. Let U be an open subset
of Y containing e(g,x) = g(x). Since X is locally compact, there is an
open set V with compact closure such that c € V C V C g74(U). Let
A=S(V,U)NC(X,Y). Since A x V is open in C(X,Y) x X and contains
(g,z), to complete the proof it suffices to show that e(A x V) C U. However,
this is trivial: if f € A andv € V, then e(f,v) =f(v)Ee f(V) CU. O
Suppose X, Y, and Z are topological spaces and f: X x Y > Zisa
continuous function of y for each fixed x € X. Define f* : X + C(Y, Z) by
(f*(x))(y) = f(x,y). If C(Y, Z) is given the compact-open topology, then f
and f* enjoy the following close relationship.
(7.B.15) Theorem. Suppose f : X x Y > Z and f* : X + C(Y, Z) are
defined as above. If f is continuous, so is f*. And, if f* is continuous and
Y is locally compact, then f is continuous.
Proof. Suppose f is continuous, z € X, G = S(A,V) N C(Y, Z) where
A is compact in Y and V is open in Z, and f*(x) € G. We seek an open
set U in X such that x € U and f*(U) CG. Note that this is equivalent to
finding an open set U in X such that for each u € U, (f*(u))(A) C V or,
somewhat more aesthetically expressed, f(U x A) CV.
Since z € f*~1(G), we have {x} x A C f7'(V). Hence, by (2.G.14),
there is an open neighborhood U of z such that U x A C f~'(V). Therefore,
FOE .
Now suppose f* is continuous. Let h: X x Y + C(Y,Z) x Y be defined
by h(x,y) = (f*(x),y). Then h is continuous. The continuity of f follows
from the fact that f = eh, where e is the evaluation function. O
(7.B.16) Definition. Suppose F = {f,: X + Xq : a € A} is a collection

of functions from a space X into arbitrary spaces Xq. The family F separates
points from closed sets in X if and only if for each closed subset B of X and
each x € X \ B, there is ana € A such that f(r) ¢ fa(B).
The following result will prove to be central in proofs of a number of
metrization theorems as well as in our brief examination of the Stone-Cech
compactification (see problem A-B 11).
(7.B.17) Theorem. If X is a T; space and F = {fa : X + Xq : a€ A} is
a family of continuous functions that separates points from closed subsets,
then €: X > ||,¢, Xa, defined by é(r) = {f.(x)}, is an embedding.
Proof. By (6.4.5), é is continuous. If « # y, there is an a € A such
that fa(x) # fa(y) (points are closed and F separates points from closed
sets). Hence, é is 1-1. To complete the proof, we show: if U is a member of
a basis for X, then é(U) is open in é(X). First, we construct a basis suitable
for our purposes. Let B = {f71(V) : a € A, V openin X,}. Clearly
B is a collection of open sets. That B is a basis may be seen as follows.
Suppose z € W, where W is an arbitrary open set in X. Obviously, x does
not belong to the closed set X \ W, and hence there is an a € A such that
fa(x) ¢ fa(X \W). Let V = XQ \ fa(X \W). Then it is easily established
that x € f7'(V) C W; therefore B is a basis.
Now consider é(f;1(V)) and observe that fa = paé, where py is the
a-th projection map. Since é(fy1(V)) = pg'(V) Né(X) and pz! (V) Né(X)
is an open set in €(X), €is an embedding. O
An interesting corollary is that any Tihonov space can be embedded in
a product of unit intervals.
(7.B.18) Corollary. IfX is Tihonov and F = {f : X +I : f is continuous},

then €: X > [[;c¢Jy is an embedding (where I+ = I for each f € F, and
the f-th coordinate of é(z) is f(z)).
(7.B.19) Exercise. Suppose X and Y are Tihonov spaces and h: X > Y
is continuous. Let F = {f :X 71: f is continuous} andG={g:Y 31:
g is continuous}. Show there is a continuous map # : [| ;e¢1¢ > J]
such that the following diagram is commutative.
. . . . 9¢G I
x ls Y
é é!
|
Iyer I5—j—[I
ge 1s
[Hint: If {zs} € []pexJy, let the g-th coordinate of #({zs}) be the (gh)-th
coordinate of {zs}.] Show also that #(é(X)) Cc é(Y).
The Weak Topology: II 181
C. THE WEAK TOPOLOGY: |

Let {Xq : a € A} be a family of topological spaces. We were careful when
defining the product topology on J] ack Xa to ensure that the projection
maps pa : [| ack Sa —* Xq would be continuous while having as few open
sets as possible. We generalize this procedure as follows.
Let {Xq. : a € A} be a family of spaces and X be an arbitrary set.
Suppose, further, for each a € A, there is a function fy : X > X,q. Our goal
is to define a topology for X that guarantees the continuity of each fg. Of
course, if X is given the discrete topology, then each f, will be continuous.
However, as was the case with the product topology, we would like to keep
the topology as small as possible. This is achieved, as with the product
topology, by using a subbasis. Let S be the family of all sets of the form
fxz1(Ua), where U, is open in X, and a € A. Then S is a subbasis for
a topology for X, called the weak topology (or initial topology) induced by
the family {f. : a € A}. It should be noted that in this terminology, the
product topology for Y = [],¢, Xa is just the weak topology for Y induced
by the natural projection maps {pq : a € A}.
One of the most useful properties of the weak topology (and hence of
the product topology) is given in the next theorem.
(7.C.1) Theorem. Suppose X is given the weak topology induced by the

family {fg : X 4 Xq : a€A}. Let g: Y > X, where Y is an arbitrary
topological space. Then g is continuous if and only if fag is continuous for
each a € A.
Proof. Obviously, if g is continuous, then so is each fag. Suppose, then,
fag is continuous for each a € A, and let U be an open set in X. It suffices
to show that g~!(U) is open in Y. By (6.A.9), it may be assumed that
U = fz1(Uq) for some a € A, where U, is open in Xq. Since g~'(U) equals
g ‘(fz '(Ua)), which is open by hypothesis, it follows that g is continuous.
O
Note that (7.B.17) is easily generalized (with practically no change in

the proof) to handle the case where X has been given the weak topology
induced by some family of functions.
D. THE WEAK TOPOLOGY: II

In the literature, one frequently encounters another topology called the weak
topology, a topology induced by subsets of the given set rather than by a
family of functions. Suppose some set X is a union of a collection of sets
{A, : a € A}, where each Aq enjoys its own topology. We have already seen
that if the Ag are disjoint, then X may be given the free union topology.
Suppose, however, the Aq are not necessarily disjoint, but that:
(i) the topologies of Ag and Ag coincide on Ag Ag for each a, G € A,

i.e., the relative topology that A.M Ag inherits from Ag is the same
as that which it inherits from Ag, and
(ii) either (a) Ag M Ag is open in both Ag and Ag for all a, 8 € A or
(b) Aa N Ag is closed in both Ag and Ag for all a, 8 € A.
The topology {U Cc X : UN Ag is open in Ag for each a € A} is called the
weak topology for X associated with (or induced by) the spaces {Ag : a € A}.
It should be apparent that the free union topology is simply a special case
of this topology.
This weak topology is used in the construction of what are known as
cell complexes (studied in Chapter 8). Suppose during an idle moment one
started to build a space by starting with some intervals, then began adding
some triangles, then some tetrahedra, then some four-dimensional analog,
etc. Since each piece of the resulting space has its respective Euclidean
topology, the weak topology may be assigned to the union of the pieces.
One virtue of this particular version of the weak topology is given by

the following theorem.
(7.D.1) Theorem. Suppose X has the weak topology induced by a family

of subsets {Ag : a € A} and Y is an arbitrary space. A function f : X > Y
is continuous if and only if f;4, : Aa — Y is continuous for each a € A.
Proof. Since restrictions of continuous maps are continuous, it suffices
to show that if all the restrictions are continuous, then so is f. Suppose U
is an open subset of Y. Since f~!(U) NM Ag equals fia UO), which is open
in Ag for each a, it follows from the definition of the weak topology that
f~1(U) is open in X. Therefore, f is continuous. O
The following theorem is useful in the next chapter.
(7.D.2) Theorem. Suppose X; C X2 C X3 C --- is a sequence of topo-

logical spaces and X = UP”, Xn is given the weak topology induced by the
X,. If each X, is normal and closed in X,j41, then X is normal.
Proof. First we note that X,M Xm is closed in both X, and X,, and
so it is legitimate to speak of the weak topology on X. Now, suppose A and
B are closed disjoint subsets of X. To show that X is normal, we construct
Hilbert Space 183
a continuous function f : X — [0,1] such that f(A) = 0 and f(B) = 1,

(4.B.1). Choose N large enough so that for each n > N, we have X,NA #0
and X,0B #9. Since Xy is normal, Uryson’s lemma yields a continuous
function fy : Xn — [0,1] such that fy(Xn NA) =0 and fy(X, NB) =1.
Define fy : Xn U(Xn41 NA)U(XnN4i N B)-> [0, 1] by
0 ifs Ee XnqiNA
fn(z)=¢1 if 6 win 8
fn(c) if2e Xn
By the map gluing theorem, 7Nn is continuous. We apply Tietze’s theorem to

extend fy to a continuous function fyi: Xn41 — (0, 1], where of course
fn4i(Xn41 A) =0 and fgi(Xn41N B) = 1.
Inductively, for i > N we obtain a function f; : X; — [0,1] such that

fix; = f; whenever N < j < i and furthermore f;(X;M A) = 0 and
fi(XiN B) = 1. Define f : X — [0,1] by f(x) = fr(x) whenever x € Xp.
Then clearly f is well defined, f(A) = 0 and f(B) = 1. It remains to show
that f is continuous. However, since fix, = fn, the continuity of f follows
immediately from the preceding theorem. O
E. HILBERT SPACE
One of the more interesting spaces in topology, both because of its versatility
and because of its unpredictability, is the classical Hilbert space. Hilbert
space serves as an important bridge in spanning the hiatus between analysis
and topology; we, however, will focus our attention on some of its topological
properties. Although it has been shown that Hilbert space is homeomorphic
to the countable product of €! with itself (Anderson [1966]), we begin with
the classical description of Hilbert space.
Let H denote the set of all infinite sequences {2,, 72,...} in R’ that are
square summable (i.e., )>;~; z? < co). We frequently denote such sequences
by (an) or (x1, 22,...). A metric p is defined on H that is completely anal-
ogous to the usual Euclidean metric for €”. If @ = (tn) and y = (yn) are
elements of H, then define
It is immediate that p is reflexive and symmetric. The triangle inequality

is a direct consequence of (0.F.4). (Square summable sequences are used
to ensure that p(x,y) is a real number.) The metric space (H, ) is called
Hilbert space.
To gain an idea of the “size” of (H,p), note that for each n, €” can
be embedded isometrically into H via the map ~, : E€” — H defined by
W((t1,22,---;Ln)) = (@1,.--, Ln, 0,0,...). Further indication of the vastness
of H is exemplified by the fact that any separable metric space (including H
itself) may be embedded in a compact subset of that has empty interior,
(7.E.8).
Before sampling a few of the more exotic features of (H, p), we first show
that it is both separable and complete.
(7.E.1) Theorem. Hilbert space (H, p) is a separable metric space.
Proof. We exploit the separability of €” to find a countable dense subset
of H. Let D, = {(ri,fo,.ec5 7a 0s 0s. 2) © 7. 2° ris rational! foreach
n € Zt, and let D = U7, Dn. It is clear that D is countable. It remains
to show that D is dense. Let € > 0 be given and suppose x = (z,) € H. We
show that DNS,(x) # 9. Choose N large enough so that }0° v4, 1? < €?/4.
For i = 1,2,..., N, let rj be a rational number such that |z; —1;| < «/V4N.
Then, if r = (ri,r2,...,7n,0,0,...), we have
p(z,r) < > (ai — 0)? < €/2+€/2,

i=N+1
and hencer€ DN S,(z). O

(7.E.2) Theorem. Hilbert space (H,p) is a complete metric space.
Proof,” Let =(oi teen = (tie Oe De aa Calicny ge-
quence in (H, p). For each k, the sequence {xi}, is a Cauchy sequence in
E}, since |x, — 27,| < p(x', x7) for each i and j. Consequently, for each k, the
sequence {x}, }S, converges to a point yz € E!, (3.B.3). Let y = (yi, y2,---).
We simultaneously show that y € H and that y is the limit of the sequence
pian)
Note, for fixed p, limm—oo >} (27-2)? = OP, (2? —limmoo 2)? =
‘1 (2? — yi)”. Now, since z!,x?,... is a Cauchy sequence, there is a pos-
itive integer N such that if m,n > N, then )°?_, (a? — 2)? < (€/2)? for
all integers p. Consequently, ifn > N we have limmoo )-h_, (2? — 2™)? =
ee ieee ele)
Now use the Minkowski inequality (0.F.2) to conclude
Hilbert Space 185
and hence JF (yi)? is a monotone bounded sequence in p. Thus, y is

in H. Furthermore, since for n > N, we have /)<2, (a? — yi)? < €/2, it
follows that p(x",y) approaches 0 as n increases. Therefore, the sequence
{x"} converges toy. O
(7.E.3) Theorem. Hilbert space (H,p) is a strongly convex metric space
without ramifications.
Proof. We first show H is strongly convex. Let x = (zp) and y = (yn)
be elements of H. The obvious candidate for a midpoint between x and y
is m = ($(tn + Yn)): confirmation that m is a midpoint is trivial, but the
uniqueness of m must be verified. Suppose b also lies midway between x
and y. Then p(x,b) = p(y,b) = $p(x,y). Hence
4530 (rn = bg = 45 (un = B10 = Ss. (aa = Yn)?
The following well-known identity comes immediately to mind (and we will

use the identity even if it doesn’t):
(u+v—2w)? = 2(u—w)? + 2(v —w)? — (u—v)?.
Consequently, we have
bi Int Yn —- malar By te GS De) ~ 2a

(en— tn) = 0,
n= n=1 (ea n=
which leads us to conclude that b, = (1/2)(%@n + yn); hence, m is the unique

midpoint.
We now show H is without ramifications. Suppose m is a midpoint of
x and y and also of x and z. Then p(x,m) = p(z,m) = p(y,m). From
the preceding paragraph, we have $(In + Yn) = ™n = $(n + 2n) for all n.
Therefore, yn = 2n for alln, and hencey=z. O
From the definition of Hilbert space and from the foregoing theorems,
one might be led to believe that (H, p) has the same properties as Euclidean
space. The next two theorems should dispel any such notions.
(7.E.4) Theorem. Hilbert space (H, p) is not locally compact. In fact, for
each x € H and each € > 0, S,(x) fails to be compact.
Proof. Suppose x = (11,22,...) € H and e > 0. For n = 1,2,... let
Yn, = (21, 22,---)En—-1,En + 6, £n41,--+). Note that for each,7y,°¢ See),
but for i # j, p(yi,
Yj) = /2e. Thus, the sequence {y,,} has no convergent
subsequence. Hence S,(x) is not sequentially compact, so by (3.A.11) can
not be compact. O
Invariance of domain fails badly in (H, p), i.e., open sets in H may be
mapped homeomorphically onto non-open subsets of H. In fact, H itself can
be embedded (even isometrically) into itself as a nowhere dense subset.
(7.E.5) Theorem. Suppose A = {(21,%2,...) € H : 2 = O} and define
W:H A by W((21,22...)) = (0,21, 22,...,). Then w is an isometry and
int A= 0.
Proof. Clearly w is an isometry. To see that A is nowhere dense in H,
let x = (%1,22,...,) € H and let € > 0; we show that S,(x) is not contained
in A and hence A has empty interior. Choose b # 0 so that |x; — b| < €/2
and let y = (b, £2, 73,...). The proof will be complete once we establish: if
6 = min{|b|,€/2}, then Ss(y) C S.(x) and Ss(y)N A= 0.
Suppose z € S;(y). Then we have p(x, z) < p(x, y)+p(y, z) < €/2+€/2,
and consequently, Ss(y) C S.(x). If w = (wi, we,...) € Ss(y)N A, then
w, = 0, and hence p(w,y) > |b| > 6 > p(w, y)—an unusual occurrence at
best. O
Another strange property of Hilbert space is that compact subsets of

H fail to have interior points. (This follows easily from (7.E.4).) Fur-
thermore, if K is a compact subset of H, then H is homeomorphic to
H \ K. In fact, if Ky, Ko,... is a sequence of compact subsets of H, then
H \ Us2, Ki is still homeomorphic to H. We shall not prove these latter
oddities, but the interested reader might begin a serious study of Hilbert
space with the previously mentioned article of Anderson or the book by van
Mill [1989]. There, the relationship between H and J]7~, €} (where €} = &}
for each 7) is investigated and it is shown that H and []<, €} are home-
omorphic. A hint that this might be a nontrivial result is seen from the
following considerations. Let Uy denote the usual topology for (H,p). Let
H = {(01,%2,...) € TJ,€} : Ve,27 < oo}, and let Uy be the rela-
tive topology for H considered as a subset of the product space []7<, €}.
Then Uy # Uy. To see this, let e; = (1,0,0,...), eg = (0,1,0,0,...), etc.
Clearly, there are no limit points of the sequence {e1, e2,...} in (H,Uy). On
the other hand, it follows readily from the nature of basic open sets in the
product topology that the sequence {e;,€2,...} does converge to (0,0,...)
in (H1,U,3).
Hilbert Space 187
(7.E.6) Exercise. Show that i: H — H is continuous but that i-? is not.

One compact subspace of (H,p), the Hilbert cube, Hc is of particular
importance. Define Hc to be {(11,22,...)€ H: 0< an, <1/nforne Ze¥.
Although Ho is compact (7.E.7) and hence lacks interior points in H, it
nevertheless has sufficient capacity to swallow any separable metric space
(including H) (7.E.8).
Unlike H, Hc is easily shown to be a product space.
(7.E.7) Theorem. The Hilbert cube Ho is homeomorphic to []%2, hi,
where for each 7, J; = I.
Proof. Clearly, p((21, %2,...)) = (#1, 22/2, 23/3,...) defines a function
wv: T]2,l > Hc. It is easy to see that is 1-1 and onto. Continuity
may be seen as follows: Suppose x = (21, 22,...), € > 0, and V = S,(y(z)).
Choose an integer N large enough so that )>. y 1/n? < e?/4, and let U =
U, x Up x --» x Un x €1 x €} x --+ be a neighborhood of 2, where U; =
{z; EI : |x; —z;| < €/V4N} foreach i = 1,2,...,N. If z= (2, 20,...) €
U, then (ai /i) — (z/i))? < SN(ei - x)? < 2/4. Furthermore,
Dens ((ti/t) — (zi/1))? < €/4, Thus, we have 4(U) C V, and hence ~ is
continuous. That y~ is a homeomorphism follows from (2.G.11). O
Next we show that any separable metric space can be embedded in He.
(7.E.8) Theorem. If (X,d) is a separable metric space, then there is an

embedding €: X > Hc.
Proof. Since (X,d) is separable, it has a countable basis B. Now let
A = {(U,V) : U,V € Band U C V}. Since X is normal, Uryson’s lemma
may be applied to each pair (U,V) € A to obtain a map fyy : X — I where
fuv(U) = 0 and fyuy(X \V) = 1. Note that F = {fuv : (U,V) € A} isa
countable family of functions that separates points and closed sets (7.B.16).
For each f € F, let I be the unit interval J and define é: X > [[;¢¢Jy by
setting é(z) = {f(x)}sex- Since [];-z Js is homeomorphic to Ho and é is
an embedding (7.B.17), the proof is complete. O
As an encore to the preceding beautiful result, we show that Ho has
the fixed point property. We will utilize the Brouwer fixed point theorem for
I” (to be proven in Chapter 14).
(7.E.9) Exercise. Let (X,d) be a compact metric space and let f: X 4 X
be continuous. Show: if for each € > 0, there is a point x. € X such that
d(f(x-),%e) < €, then f leaves some point of X fixed.
(7.E.10) Theorem. The Hilbert cube, Hc, has the fixed point property.
Proof. Consider Hg as the product space [0, 1] x [0, 1/2] x [0, 1/27] x ---
(why is this possible?). Let f be a continuous map from Ho into Hc,
and let € > 0 be given. Choose N large enough so that >, 1/n? < ,
and let A = [0,1] x [0,1/2] x --- x [0,1/2%]. Define a map fi AoA
by f(21,22,-..,;2n) = (y1,Y2,---,YN), wheres BN ONO...) =
(15 °09,,4045
(41, Y2s-++>YNyYN+1)YN+2,---). Then f has afixed point a = (a), @2,...,an)
by the Brouwer fixed point theorem (14.C.22). Let a = (a1, a2,...,an,0,0,...).
Then p(f(a),a) < €, and consequently, by (7.E.9), Hc has the fixed point
property. O
A more general result is obtained in problem E.1.

We recommend the book by Young [1988], as well as that by Debnath
and Mikusinski [1990] for the readers who are interested in an introduction
of the use of Hilbert space in analysis. Of course we would be remiss if we
didn’t also mention Halmos [1982], although it would be best if the reader
had more background than we have given before tackling that book.
PROBLEMS
Sections A and B
1. Suppose X and Y are topological spaces and cardX is infinite. Show

that Hom(X,Y) (with the point-open topology) is locally connected if
and only if Y is locally connected and connected.
2. Suppose X and Y are topological spaces. Show that if C(X,Y) (with
the compact-open topology) is T3, then so is Y.
3. Show that if X is a locally compact T) space, and X and Y are second
countable, then C(X,Y) (with the compact-open topology) is second
countable.
4. Show that the sup topology on B(X, Y) depends not only on the topolo-
gies of X and Y but also on the metric chosen for Y (construct exam-
ples).
5. Let X and Y be a compact metric spaces. Show that the set of contin-
uous onto maps from X to Y is a closed subset of C(X,Y).
6. Let X and Y be separable metric spaces. Show that C(X,Y) (with the
compact-open topology) is separable.
7. Suppose X is a topological space, Y is a bounded metric space, and K
is the collection of all compact subsets of X. For each f € C(X,Y),
K €K, and c > 0, let S(f, K,e) = {g € C(X,Y) : d(f(z),g(x)) < €
for all x € K}. Then {S(f,K,e) : f € C(X,Y), KEK, e>O0}isa
subbasis for a topology called the topology of uniform convergence on
Problems 189
compacta. Show that the topology of uniform convergence on compacta

coincides with the compact-open topology.
Suppose X and Y are topological spaces. Let j : Y > C(X,Y) be
defined by j(y) = cy where cy : X ~ Y is the constant map from
X onto y. Then j is called the natural injection of Y into C(X, Y).
(Suppose C(X, Y) has the compact-open topology.) Prove:
(a) 7 is an embedding of Y into C(X,Y));
(b) if Y is To, then j(Y) is closed;
(c) j(Y) is a retract of C(X,Y).
Suppose X is a completely regular T> space and (Y,d) is a bounded
metric space containing a nondegenerate path. Show that the compact-
open topology on C(X, Y) coincides with the sup topology on C(X,Y)
if and only if X is compact.
10. Suppose X is a compact space and Y is a metric space. Show that
C(X,Y) with the compact-open topology is an AR if and only if Y is.
it. Let X be a Tihonov space. Show that A= {f:X — €! : f is bounded
and continuous} separates points from closed sets, and therefore, by
(7.B.17), 6: X — []se¢]y is an embedding. The Stone-Cech com-
pactification of X is the closure BX of é(X) in [|,;¢¢J;. Show that if
f : X — Y is continuous and Y is a compact T2 space, then there is a
unique extension of f o (€~*)|g(x) to 8X. (Hint: Use (7.B.19)].
12. A space X is real compact if and only if X can be embedded as a closed
subset of a product of copies of €1.
(a) Show that compact T> spaces are real compact.
(b) Show that an arbitrary product of real compact spaces is real com-
pact.
13. Suppose X is a topological space and Y is a metric space. Let B(X, Y)

be defined as in (7.B.9). Show: Y complete implies B(X, Y) complete.
14. Suppose X is a topological space, Y is a metric space, F C C(X,Y)
and « € X. The family of functions F is equicontinuous at x if and only
if for each € > 0 there is an open set U containing x such that if f € F
and @ € U, then d(f(Z), f(x)) < ¢. If F is equicontinuous at each point
of X then F is said to be equicontinuous.
Prove:
(a) If X is compact and Y is a compact metric space then F C C(X,Y)
is equicontinuous if and only if F is totally bounded.
(b) Let X be a compact space and let F be a bounded subset of
C(X,R”) = B(X,R”). Then, there is a compact subset C' of
R” such that if f € F and z € X then f(x) €C.
(c) (Ascoli’s Theorem). Let X be acompact space. Then C C B(X, R”)

is compact if and only if C is closed, bounded, and equicontinuous.
Section C
1. Let X = 0,00) have the relative topology from €1. Define f : R' > X
by f(x) = x”. Determine the weak topology for R’ induced by f.
2. Suppose Y is a topological space and f : X — Y. Let W be the weak
topology for X induced by f. Show that A C X is closed (with respect
to the weak topology) if and only if A = f~!(B), where B is closed in
YE
3. Generalize in some way the result of problem 2, where the weak topology
for X is induced by the family of functions {f, :X + Yo : a € A}.
4. Suppose (X,U) is a T3 1 space. Show that U coincides with the weak
topology generated by {f : X — €! : f is continuous}.
5. Suppose S is an arbitrary set and f : S — T is an onto function, where T'
is a connected topological space. Show that if S has the weak topology
induced by f, then S is connected.
6. Repeat problem 5, replacing “connected” with “compact” and “count-
ably compact.”
7. Suppose A is a subset of a space X. Show that the relative topology
for A coincides with the weak topology for A generated by the inclusion
functioni: A> X.
8. Suppose (X,U) is a topological space. Let {Ya : a € A} be a family of
spaces and {f,:X — Ya : a € A} a corresponding family of functions
with the property that for each space Z and for each function f : Z > X,
f is continuous if and only if fy f is continuous for each a € A. Show
that U/ is the weak topology for X generated by {fg : X + Yo : a € A}.
Section D
1. Let X have the weak topology with respect to {Ag : a € A}. Show
that C is closed in X if and only if CM Ag is closed in A, for each a.
2. Suppose X is a topological space and {Ag : a € A} is an open cover of
X. Show that the weak topology induced by the cover {A,} coincides
with the given topology on X.
3. Suppose (X,//) is a topological space and A = {A : a € A} is a closed
cover of X. Let W be the weak topology on X induced by A. Show:
a) UCW.
Problems 191
b) Find an example where W ¢ U.

c) Show: if A is locally finite (each x € X is contained in a U-open
set which intersects only a finite number of the Aq), then W CU.
4. Suppose (A;,d;) and (Az, d2) are metric spaces bounded by 1. Suppose
dy) AyNAg = 42}A,;nA, aNd A, N Ag F O is closed in A; and Ag. Define a
function by
d, (x,y) rye Ay
d(z,y)= d2(x,y) zy € Ag
inf {d(z,z) +do(z,y)} 2 € Ay \ Ao and y € A \ Aj.
zE€AiNA2
(a) Prove that d is a metric.

(b) Show that the metric topology coincides with the weak topology
induced by {Aj, A>}.
5. Suppose X = LU,¢, Aa where the Ay’s satisfy the necessary conditions
for X to be given the weak topology induced by the family {Ag : a €
hy
Prove or disprove:
(a) If each Ag is connected, then so is X.
(b) If each A, is compact, then so is X.
(c) If each A, is second countable, then so is X.
(d) If each A, is first countable, then so is X.
(e) If each A, is separable, then so is X.
(f) If each Ay is To, Ti, etc., then so is X.
6. Suppose X is a topological space and A; C Ap: C --- is a sequence of
subspaces of X such that X = U?2., An. Show that the weak topology
on X induced by the A, does not necessarily coincide with the original
topology for X.
Section E
1. Suppose {Aq : a € A} is a collection of compact T2 spaces with the

property that for each finite subcollection {Aq,,---,Aa,}, 1 eee
has the fixed point property. Show that [],¢, Aa has the fixed point
property.
Show that int Ho = 0.

3. Suppose that X {u,v,w,y}, and define a metric for X by setting
d(u,y) = d(v,y) = d(w,y) = 1 and d(u,v) = d(u,w) = d(v,w) = 2.
Show that (X,d) cannot be isometrically embedded in H.
Let X = ][f2, €}, where €} = €' for each 7. Define a metric

Ys
at,y) = > G@asi

ee easene:
i=1 1 Slay yi|)2*
Show that the topology induced by d coincides with the product topol-
ogy.
Show that []?°, €} with the metric given in problem 4 contains all sep-
arable metric spaces X via the function f(x) = (d(z,z1),d(z,22)...),
where x € X and {z;} is a countable dense subset of X.
Find an example of an infinite set of points A in H with the property
' that p(x,y) = 2 for each x,y € A.
Show there is a point x and a closed set C in (H, p) with p(x, C) = € > 0,
but p(xz,c) > € for each cE C.
Prove or disprove: If X, Y, and Z are topological spaces with X x Y
homeomorphic to X x Z, then Y is homeomorphic to Z.
Let X = []7°, €},
a
where €} = €' for each i. Define a metric
co
has Ge as
Le, teil
(2,y) > i! (1 + [zi — ysl)
Show that (X,d) is a complete, separable, metric space which is not
totally bounded.
10. Show that (X,d) in problem 9 is homeomorphic to a subspace of Hilbert
space, and Hilbert space is homeomorphic to a subspace of (X, d).
Chapter 8
QUOTIENT SPACES
A. THE QUOTIENT TOPOLOGY

A partition of a set X is a collection G of mutually disjoint nonempty subsets
of X whose union is X. Partitions arise quite naturally in mathematics and
are quite important in topology. They lead to the creation of many complex
and interesting spaces.
Suppose G is a partition of a topological space X. Define X/G to be
the set whose “points” are the members of the given partition G. Then X/G
is called a quotient or decomposition set. The function P : X + X/G that
maps a point z to the unique set of G that contains z is called the quotient
map. We will frequently denote P(x) by [x]. Note that P(x) = P(y) if and
only if xzand y belong to the same member of the partition. We want to
topologize X/G so that (i) P is continuous, and (i7) X/G has a maximal
number of open sets; contrast this with the rationale behind the definition
of the product topology.
(8.A.1) Definition. Suppose X is a topological space, G is a partition of

X and P: X > X/G is the quotient map. The topology
U={UCX/G : P~'(U) is open in X}
is called the quotient topology or decomposition topology for X/G.
(8.A.2) Exercise. Show:

(i) The quotient topology is a topology.
(ii) If X/G has the quotient topology, then P is continuous.
(iii) If UW is the quotient topology and V is any other topology for which
P is continuous, then V CU.
(iv) If X/G has the quotient topology and A C X/G, then P~'(A) is
closed in X if and only if A is closed in X/G.
193
194 8. Quotient Spaces
We now consider a useful construction that leads to partitions and quo-

tient spaces. Suppose f : X — Y is continuous and onto. Since f is a
function with domain X, it is clear that G; = {f~'(y) : y € Y} is a par-
tition of X. If z € X, note that P(x) = [z] = f~'(f(z)). In this context,
the function ®; : X/Gy > Y defined by ©¥([z]) = f(z) is of special interest
(why is ¢ well defined’).
xX 2 “a
P Dy
X/Gs
(8.4.3) Theorem. Suppose f : X — Y is a continuous and onto map, and

X/G; has the quotient topology. Then ®; is continuous.
Proof. Suppose U is an open subset of Y. Then f~'(U) is open in X.
Sincef“(Mi= (Ox. 0P)jr ()= P-1(@;7(U)) and X/G' has the quotient
topology, we have 6; '(U) openin X/Gy. O
(8.A.4) Remark. There is a striking parallel between quotient spaces and
quotient groups.If G; and G2 are groups and h: G; — G2 is a homomor-
phism from G; onto G2, then G; may be partitioned into sets of the form
{h-1(y) : y € Go}. Let G,/Kerh denote the resulting quotient set. There
is a natural group structure that may be assigned to G;/Kerh whereby the
function ®, : G,;/Kerh — G2 becomes an isomorphism (first isomorphism
theorem). In the topological context, it now becomes reasonable to ask
whether ®f is the topological counterpart of an isomorphism, i.e., a homeo-
morphism. It is clear that @¢ is 1-1 and onto. Although in many situations
®; is a homeomorphism, the following example shows this is not always the
case.
(8.4.5) Example. Let X = [0,27), and suppose f : X > S! is defined

by f(x) = (cosz,sinz). Then f is a continuous 1-1 map from X onto S!,
and since Gy is essentially X, it follows that X/Gy is homeomorphic to
X. However, X is not homeomorphic to S! (S1 is compact and X is not).
Therefore, in this example, ®¢ is not a homeomorphism.
The following theorem gives two conditions that force ®¢ to be a home-

omorphism.
(8.A.6) Theorem. Suppose f : X + Y is a continuous function from

X onto Y and f is either open or closed. Then 9%; : X/Gf > Yisa
homeomorphism.
Identifications 195
Proof. Suppose f is closed, and let A be a closed subset of X (Gra

suffices to show that ®, is closed in Y. This is immediate, however, since
;(A) = f(P~1(A)). The case where f is open is no more difficult. 0
The following corollary is useful in showing that some spaces are home-
omorphic to quotient spaces; it will be exploited in the next section.
(8.A.7) Corollary. Suppose X is compact, Y is T>, and f : X > Y is

continuous and onto. Then ®; isa homeomorphism. 0
B. IDENTIFICATIONS
Partitions of a space X are frequently formed by “identifying” certain points
in X with others. An equivalence relation ~ on X leads quite naturally
to such a partition, where the members of the partition are the equivalence
classes under ~. In this case, we denote the resulting decomposition or
quotient space by X/~. For example, if X = [0,27], we can induce an
equivalence relation ~ on X by declaring 0 ~ 27, and x ~ y if and only if
x = y for z,y € (0,27). The equivalence classes consist precisely of single
points in the open interval (0,27) together with the set {0,27}. What does
the corresponding quotient space look like? Intuitively, the end points of
the interval have been “glued” together, so it would seem that the resulting
space should be a circle; that this is indeed the case may be established by
considering the following commutative diagram, where f(x) = (cos z, sin z).
x A S!
X/G;= X/~
Since [0,27] is compact, (8.4.7) tells us @y is a homeomorphism. Conse-
quently, the quotient space is topologically a circle. The only trick here was
to find a continuous onto map for which Gy is ~.
Many interesting topological spaces may be constructed in a similar
fashion. For example, although the torus is usually defined to be the product
space S! x S!, an alternate description can be obtained using quotient spaces.
Start with the square X in the zy plane (€7) whose vertices are the points
(0,0), (0,2), (2,0) and (2,2). Identify each point in X of the form (0,y) with
the corresponding point (2, y) on the opposite side of the square. Intuitively,
that has the effect of rolling up the square into a tube. Now glue the ends
of the tube together by identifying points of the form (z,0) with (z,2) and
one obtains a quotient space homeomorphic to the torus S Longe}
This may be observed while considering the following diagram, where f is

defined by f(x,y) = ((cos7z, sin mz), (cosy, sin ty)) and X/G¥ is the quo-
tient space just described.
XG i S1 x S!
P Dy
XIGy
More exotic spaces may also be constructed when starting with a square.
For instance, the Klein bottle can be defined as the space obtained by forming
a tube and then identifying points of the form (z,0) with (2—z,2). This has
the effect, intuitively, of twisting the tube before gluing. It can be shown that
the resulting space is not embeddable in €%, although it can be embedded in
&4
(2—z,2) (2,2)
In the problem section, the reader is asked to construct a number of

other curious spaces using the techniques just described. Notice that each
of the spaces that we have created by identifying certain boundary points
of the square is a 2-manifold (why?). In Chapter 16, it is shown that all
connected compact 2-manifolds may be obtained in this fashion. Of course
it is also possible to make identifications of the boundary points of a square
so that the associated quotient space is not a 2-manifold.
Next we use quotient spaces to construct the “cone” over a given space.
Identifications 197
(8.B.1) Definition. Suppose X is a topological space. Let ~ be the equiv-

alence relation on X x J determined by: (z,t) ~ (y, s) if and only if either
(x,t) = (y,s) ort = s = 1. The cone overX is the quotient space (X xI)/~.
SA
ES stor
(8.B.2) Exercise. Suppose X is a subspace of €” C €"+!. Sometimes the
Cone over X is defined to be the subspace of €"* consisting of all points
on the line segments from (0,...,0,1) to (2,0) where x € X. Problems 14
and 15 show that the Cone of X is not necessarily homeomorphic with the
cone of X.
a) If X Cc €” andi: cone(X) — Cone(X) is the obvious map, show
that 7 is continuous.
b) What reasonable conditions can be imposed on X C €” to guarantee
that i is a homeomorphism.
(8.B.3) Definition. Suppose X is a topological space. The suspension of

X is the quotient space obtained by identifying in X x [—1, 1] all points of
the form (z,1) with each other, and all points of the form (x, —1) with each
other.
AN
se
Cones and suspensions may also be derived as special cases of an im-
portant construction that involves “sewing spaces together” with the aid of
a given continuous function.
Recall from Chapter 1 that if X and Y are disjoint spaces, then the free
union of X and Y is the topological space Z = X UY, where aset W C Z
is open if and only if WN X and WNY are open in X and Y, respectively.
Suppose X and Y are disjoint spaces and A is a closed subset of X. Let
f :4—Y bea continuous function. Define an equivalence relation in X UY
by identifying each point z € f~*(y) with y. Denote the resulting quotient
space by X Uy Y. Thus A has been “glued” to its image f(A), and in the
process, X and Y have been sewn together. The space X Uy Y is called the
adjunction of X andY by f. (Of course, if X and Y are not initially disjoint,
then the usual procedure may be utilized to “disjointify” them.)
DGC 9 6
The cone over X may be considered an adjunction space: Let p be any

point not dn A. Sete Wa xox, Al Xoo and dete jena.
The adjunction space W Uy {p} is clearly homeomorphic to the cone over X
described in (8.B.1). Suspensions may be formed in a similar manner.
(8.B.4) Exercise. Suppose X and Y are disjoint spaces, A is a closed
subset of X, and f : AY is continuous. Let P: X UY > X U; Y be the
quotient map. Show: P\x\,4 is an embedding, Py is an embedding, P(X\ A)
is open in X Uy Y, and P(Y) is closed in X Uy Y.
The proof of the following theorem is surprisingly intricate.
(8.B.5) Theorem. Suppose X and Y are normal, A is a closed subset of

X,and f: A-—Y is continuous. Then X Uy Y is normal.
Proof. If D, and D2 are disjoint closed subsets of Z = X Us Y, then by
(8.B.4) Di NP(Y) and D2NP(Y) are disjoint closed sets in P(Y) and P(Y)
is homeomorphic to Y. Since Y is normal, there are open sets U; and U2 in
P(Y) with disjoint closures such that D} N P(Y) C U, and D2NP(Y) C Up.
Furthermore, B, = (P~!(D, UU1))N X and By = (P~!(D2UU2))NX are
disjoint closed subsets of the normal space X. Let V; and V2 be disjoint open
sets in X containing B, and Bg, respectively, and set W; = P(V, \ A) UU,
and W = P(V2 ‘ A) U Us.
Clearly, W; and W, are disjoint, Dj C W;, and Dz C We. Hence, it

remains to show that W,; and W2 are open in X Uy Y. The set W, will be
open if and only if P~'(W,) NX and P~!(W,) NY are open in X and Y,
respectively. Since P~'(W,) = P~!(P(V, \ A) UUi) = (Vi \ A) UP“1(()),
we have P~'(W,)N X = (VY, \ A)Uf—1(U1). Of course, Vi \ A is open in X
and f—!(U,) may be written as Z MA, where Z is open in X and ZC V4.
Therefore, P~'(W1) NX = (Vi \ AVU(ZN A) = (YW \ A)U(ZNYNA)
= (V;\A)U(ZNV,), and hence P~'(W,)NX is open in X. Since P-!(W,)N
Y = Uj, it follows that W, is open in X Us Y. One may show W, is open in
a similar manner. O
The reader’s mounting concern that (4.B.19) would be forever bereft of

proof can now be allayed.
Identification Maps 199
Proof of (4.B.19). It remains to show that absolute retracts are Absolute

Retracts. To this end, suppose X is an absolute retract, Z is a normal
space, A is a closed subset of Z, and f : A + X is a continuous map. We
will find a continuous extension of f to Z. Form the space Z Uy X and let
P:ZUX + ZUs X be the quotient map. By (8.B.5) and (8.B.4), ZU; X
is normal, and X is embedded as a closed subset of Z Us X. Since X is an
absolute retract, there is a retraction r: ZU; X — P(X) that leaves points
in P(X) fixed. Now F: Z + X defined by F = (Pix)7'rP,z is the desired
extension. O
(8.B.6) Exercise. Suppose X and Y are T; spaces and X Uy Y is an

adjunction space. Show X Uy Y is 7}.
C. IDENTIFICATION MAPS
(8.C.1) Definition. Let X and Y be topological spaces and let f : X > Y
be onto. The map f is an identification map if and only if the topology for
Y coincides with U = {U CY : f7'(U) is open in X}.
Clearly every identification map is continuous. For examples: the quo-
tient map P : X — X/G is an identification map, as well as continuous and
open or closed mappings.
A key property of identification maps is given in the next theorem.
(8.C.2) Theorem. If X, Y, and Z are topological spaces, f : X — Y is an

identification map, and g: Y — Z, then g is continuous if and only if gf is
continuous.
Proof. Certainly if g is continuous, then so is gf. Now suppose gf is
continuous. Let U be an open set in Z. Since gf is continuous, we see that
(gf)~!(U) = f7!(g-1(U)) is open. But since f is an identification map,
f7(g-!(U)) being open implies g~!(U) is open. O
(8.C.3) Exercise. Obtain a characterization of identification maps by stat-
ing and proving a converse of (8.C.2).
The following exercise can sometimes be used to show when a continuous
function from X to Z induces a continuous function from X/Gy to Z.
(8.C.4) Exercise. Suppose X, Y, and Z are topological spaces. Suppose
further f : X — Y is an identification map and h: X — Z is continuous.
Show: if hf—! is a function, then hf~? is continuous.
We now characterize identification maps in terms of the function, ®,,
defined in section A.
(8.C.5) Theorein. Suppose f : X — Y is continuous and onto. Then f is

an identification map if and only if ®y : X/Gy — Y is a homeomorphism.
Proof. We show: if f is an identification map then ®; is open. Let U

be an open subset of X/G,;. Since f is an identification map, it suffices to
show f~!(®,(U)) is open in X. However, f—1(@,(U)) = P~*(U) and since
P is continuous, the result follows.
Now suppose yf is a homeomorphism. We need to show that U is open
in Y if and only if f—!(U) is open. Let U be open in Y. Since f is continuous
it is clear that f~!(U) is open in X. Assume now U C Y and f7'(U) is
open in X. We want U to be open. But f~*(U) = P-1(6;*(U)) is open
which implies 6; °(U) is open, since P is the quotient map. Then U must
be open since @;(U) is a homeomorphism. O
The following theorem is somewhat related to (2.E.7).
(8.C.6) Theorem. Suppose X is locally connected and f : X — Y is an
identification map. Then Y is locally connected.
Proof. By (2.E.2), it suffices to show that components of open sets of
Y are themselves open. Let C be a component of an open set U C Y. Since
f is an identification map, we need only show that f—!(C) is open. Suppose
x € f~1(C) and let D, be the component of f~!(U) that contains z. Then
f (Dz) is connected, lies in U, and intersects C. Therefore, f(D,) C C, and
hence D, C f—1(C). Since f~!(U) is open and X is locally connected, it
follows that D, must be open (2.E.2). Thus, f~!(C) is a union of open sets,
and consequently Y is locally connected. O
D. THE STRONG TOPOLOGY AND k-SPACES

The weak (or initial) topology induced by a family of functions was intro-
duced as a generalization of the product topology. In a similar manner, a
topology (called the strong or final topology) may be defined that generalizes
the quotient topology. Suppose A = {Xq : a € A} is a family of topological
spaces, Y is a set, and for each a € A there is a function fy : X¥, > Y. As
the reader can surely guess by now, we will define a topology for Y so that
each f, is continuous and Y has a maximum number of open sets. This is
accomplished by declaring a subset U of Y to be open if and only if f71(U)
is open in Xq for each a. Since inverses are well behaved, the resulting
collection of subsets of Y is easily seen to be a topology for Y that meets
all of our requisites. This topology is called the strong (or final) topology
associated with the family of functions {fy:Xa—+Y : a€ A}. The reader
should note that if f : X — Y is a surjection from a topological space X
onto a set Y, and Y is given the strong topology associated with {f}, then
f is an identification map. In such circumstances it is fairly common to see
the strong topology referred to as the identification, or quotient topology
induced by f.
The Strong Topology and k-Spaces 201
(8.D.1) Exercise. Replace the word “open” by “closed” in the definition

of the strong topology, and show the resulting topology coincides with the
strong topology.
(8.D.2) Exercise. Suppose Y is given the strong topology associated with

a family of functions {fa : Xq —- Y : a € A}, and Z is an arbitrary space.
Show: A function f : Y + Z is continuous if and only if ffx :Xq 7 Z is
continuous for each a.
The strong topology is used to establish a class of spaces known as

k-spaces.
(8.D.3) Definition. Suppose (X,U) is a topological space and suppose

C=1C CX: C is compact}. For each C € C, lei tc: C — X be the
inclusion map. We call X a k-space if and only if the strong topology K for
X induced by {ic : C € C} coincides with U. The topology K is called the
k-topology for X associated with U.
Proofs of the following remarks are trivial.
(8.D.4) Remarks.
1. If (X,U) is a topological space, then U/ is always a subset of the
k-topology for X associated with U.
2. Suppose X is a k-space and Y is an arbitrary space. A function
f : X + Y is continuous if and only if fl¢ : C > Y is continuous
for each compact subset C' of X (8.D.2).
(8.D.5) Theorem. Suppose that (X,U) is a topological space and that
C={C : Cis acompact subset of (X,U)}. Then (X,U) is a k-space if and
only if a subset A of (X,U/) is closed whenever ANC is closed in C' for each
G.aC,
Proof. Suppose (X,U) is a k-space and A C X. Then by hypothesis,
the set A is U/-closed if and only if ig'(A) = ANC is closed for each C € C.
Conversely, suppose a subset A C X is U-closed whenever ANC’ is
closed in C for each C € C. We need to show that U = K, where K is the
k-topology on X. By (8.D.4 1.), UC K. We use (8.D.1) to establish the
reverse inclusion. Let A be K-closed. Then ig'(A) = ANC is closed for
each C’ € C. Hence, by hypothesis, A is U-closed. Thereforek CU. QO
(8.D.6) Exercise. Show that “open” can be substituted for “closed” in the
hypothesis of the preceding theorem.
The next theorem gives us a rich source of k-spaces
(8.D.7) Theorem. (i) Locally compact spaces are k-spaces. (ii) First
countable spaces are k-spaces. (In particular, every metric space is a k-
space.)
Proof. (i) Let (X,U) be a locally compact space. By (8.D.4) it suffices

to show that K C U, where K is the k-topology for (X,U). Hence suppose
U c X is K-open. That is, UNC is open in C for each compact subset C' of
X. Let z € U. By local compactness, there is a U/-open subset V of X that
contains z and has compact closure. Since V is compact, we have UNV is
open in V and hence UN (V NV) =UNV is open in VV =V. Therefore,
UNV is U-open in X (V is U-open in X) and contains z. Hence U may be
written as a union of U/-open sets, and so U € U. Thus X is a k-space.
(it) Let (X,U) be first countable. Again, it suffices to show K C U.
Suppose F Cc X is K-closed, i.e., is C-closed for each compact subset C’ of
X. We will show F is U/-closed and apply (8.D.5). Claim: FY c F. Suppose
a € FY and {rn} is a sequence in F that U/-converges to x (why does such
a sequence exist?). Now C = {r, : n € Zt}U {z} is compact in (X,U) and
thus CM F is closed. Hence z € F NC (why?). Consequently, FY“ CF, or
F isU-closed. O
All of the usual questions concerning products, subspaces, etc., may be

asked about k-spaces. Some of these are taken up in the problem section,
but most are left to the reader.
The reader should be cautioned that terminology involving weak and
strong topologies is not always consistent. Especially lamentable is the fact
that the weak topology induced by a family of spaces is disturbingly close
to the concept of the strong topology. In fact, if {Ag : a € A} is a family
of disjoint topological spaces, and if X = U{Aag : a € A}, then the weak
topology for X induced by the Ag coincides with the strong topology for X
generated by the inclusion maps i, : A + X (which in turn is nothing more
than the free union topology). Thus, in reading the literature in topology
(as in theology) one should be quite careful to check definitions before trying
to apply the results.
E. CW-COMPLEXES
(8.E.1) Definition. Forn >1: The standard closed n-dimensional cell,

also known as the standard closed n-cell or the standard n-ball is
a {( Sin Goat sey Gre 2 wi t:+++22 <1};
The standard open n-dimensional cell, also called an open n-cell or simply
an n-cell, is
IB) = (ata
ety by ree ajit--- +22 <ol}:
CW-Complexes 203
The standard n — 1-sphere is
See eh,
Pos Ene Cas etre 1)
In addition, Bo = 1(B°) = {0} € €1 and S-! = . An (open) n-cell,
respectively, n-sphere in a topological space X is a subspace homeomorphic
to I(B”), respectively, S”.
The CW-complexes offer an excellent illustration of how different con-
structions (in this case, the quotient topology and the weak topology induced
by a family of spaces) may be successfully intertwined to produce spaces
of considerably more interest than their progenitors. The notion of CW-
complex was introduced by Whitehead [1949]. Essentially, CW-complexes
are formed by gluing together closed n-cells. For instance, to a collection of
0-cells (points) one might carefully glue a number of closed 1-cells, and then
to the resulting space a number of closed 2-cells might be adjoined, etc.
(8.E.2) Example.
There are two natural points of view to take when discussing how a
topological space is built out of cells. The first, is to have our building blocks
in hand, and then to construct a space “dimension by dimension.” First, put
the 0-cells in place, next glue in the closed 1-cells, etc. This is perhaps the
most intuitive way to proceed. However, there is some awkwardness in the
formulation since one can not describe the gluing maps for the closed n-cells
until one has set the closed 0- through (n — 1)-cells in place.
The second approach is to start with a topological space, break it up
into open cells, and specify how we want these open cells and their closures
to relate to each other and to the topology of X. We start with this latter
approach.
(8.E.3) Definition. Let X be a topological space. A cell decomposition of
X is a set € of subspaces of X such that: each element c € € is an (open)
cell, and (as a set) X is the free union of these cells. The n-skeleton of X is
the subspace X" = U{c € € : dim(c) < n}. Clearly
Gap OCG Ge? CRG ane.

and (as a set) X =|) X”. For each cell c let ¢ C X be the closure of c in X
and defie the boundary of c by bd(c) = é\c.
(8.E.4) Definition. Let X be a topological space with a cell decomposition
€ and let c € € be an n-cell. A continuous function F : B” + X where
F(S"-1) ¢ X™7! and F|;(g=) is a homeomorphism onto c is called a char-
acteristic map of c, and f = F|gn-1 : S""=! X"1, is called a sewing map
for c.
(8.E.5) Theorem. If X is T2, € is a cell decomposition for X, and
F : B” > X is a characteristic function for a cell c € €, then
Gear (BY) exXeakvec
Gixbd(c) (Say epee
(iii) @ and bd(c) are compact and F : B” — € is an identification map.
Proof. (i) and (ii): Since F is continuous, F(B") C F(I(B")) C ©,
(1.F.5). Furthermore, F'(B”) is compact, hence closed since X is T>. Since
e C F(6"); we have ¢-G.iF(B"); Hence F(B") = ¢ =e bd(c).).Since
RUB") )\-= & clearly F(SP—*) Dbd(c); Furthermore; F(S8a" nc Ai
by the definition of characteristic map. Then we have F(S"~!) C bd(c),
since cU X"~! = O. (iii): By (2.G.11), F is a closed map, and hence an
identification by the remarks after (8.C.1). O
(8.E.6) Definition. Suppose X is a T> space with a cell decomposition €,
and each cell in the decomposition has a characteristic map. We call X a
CW-complez if and only if the following hold:
(C) (closure finite) For each cell c € €, @ intersects only finitely many
cells.
(W) (weak topology) If A C X and AMZ is closed for each cell c, then
A is closed in X.
A CW-complex is called finite if it contains only a finite number of cells,
otherwise it is infinite. A CW-complex is said to be n-dimensional if it
contains an n-cell but no cells of higher dimensions. If it is not n-dimensional
for any n we Say it is co-dimensional.
Condition (W) implies that the given topology on X coincides with the
weak topology induced by the closure of the cells. This avoids pathologies
as in example (a) below. Condition (C) circumvents the infelicity that can
be observed in example (f) below. In all the examples that follow we merely
describe the cell decomposition and leave it for the reader to decide that
characteristic maps exist.
(8.E.7) Examples.
(a) Let X = [0,1] with the usual topology. For each zx € [0,1] let {z} bea
0-cell. This cell decomposition does not yield X as a CW-complex since the
CW-Complexes 205
corresponding weak topology is the discrete topology—not what we started

with. In the future we will call 2 a 0-cell when we mean {z}.
(b) Let X = [0, 1] with the usual topology. Let 0 and 1 be the 0-cells and let
(0,1) be the only 1-cell. X with this cell decomposition is a CW-complex.
(c) Let X = S” with the usual topology. Let the north pole p be the only
O-cell and let S” \ {p} be the only n-cell. This cell decomposition makes X
a CW-complex.
2}
(d) Let X = iS (2-sphere with tangent line at t). Let x; and x2 be
the only 0-cells, the open line segment from 2 to x2 be a 1-cell, and S? \ {t}
be a 2-cell co. Then X with this cell decomposition is a CW-complex. Note:
while t =bd(c2) C X?, it is not a union of cells.
(e) Any cell decomposition of a closed subspace of E” into a finite number
of cells is a CW-complex.
(f) Let X = (a 2-sphere)
There is one 2-cell (S? minus the closed arc from zo to xz,). There are a
countable number of 0-cells, namely, ro, £1, £2, 73,°--. There are a countable
number of 1-cells, (the open arc from 2; to 2:4; for each i > 1. The 2-sphere
X with this cell decomposition has the weak topology with respect to the
closure of its cells, but is not closure finite. Note that X! with the weak
topology is not a subspace of X ({zo} is open in the weak topology but
not the relative topology on X!). Hence X! (with the relative topology) is
not a CW-complex. Closure finiteness prevents this bizarre occurrence, see
(8.E.11) iv.
(8.E.8) Lemma. Suppose X is a CW-complex with cell decomposition €,
€zp C €, and B= U. ce, ©, and suppose further c* CB for all cE Ez. If
c € €, then there is a finite subset €7, C € such that:
(i) cE Cp,
(ii) ife € Cpr, then € C F, where F, = U-ce,, ©, and
(iii) ifc € €g then F, C B.
Proof. We induct on the dimension of the cell. If c is a 0-cell, then
€r, = {c} works. Assume true for all cells of dimension < n and let c be an
n-cell. By condition (C), bd(e) is the union of finitely many cells c; each of
dimension <n. Define €p, to be {c}U U.ebd(c) fFe+ 4
(8.E.9) Theorem. Let X be a CW-complex with cell decomposition €.
Suppose €4 C X and A= Unce, ¢c. The following are equivalent:
(i) A is a CW-complex.
(ii) A is closed in X.
(iii) for each c € €4, C* CA.

Proof. (ii) = (iii): Let c C €4. Then c C A, and hence & C AS A
(iii) > (i): Clearly €4 is a cell decomposition for A; we keep the same
characteristic maps. Condition (C) holds automatically. To prove condition
(W), let C Cc A have the property: @/ C is closed in ¢ for all c € €4. Let
c € €\€,4. We will show that €N C is also closed. If we apply (8.E.8)
to each open cell in bd(c), then we obtain a finite collection of cells whose
union Fod(c) satisfies €¢7 A C Fyq.) C A. Since Pode) is the union of a
finite number of closed cells of A, it follows that Fy dce) NC is closed in X
by the hypothesis on C. Now @/NC is closed since ENC = (EN A)NC =
ena (c) NC). Therefore C is closed in X and hence in A. As a consequence
A satisfies condition (W). In particular if C = A, then A is closed in X, and
so (iii) > (ii). It is trivial to show that (i) > (iii). O
(8.E.10) Definition. Let X be a CW-complex and A C X. We say that
A is a CW-subcomplez or more simply a subcomplex of X if and only if A
satisfies the hypothesis and conditions of the previous theorem.
There are some immediate and useful consequences of (8.E.9).

(8.E.11) Corollary. Suppose X is a CW-complex. Then the following
hold:
(i) If Ay and Az are subcomplexes, so is A; M Ag and A; U Ag.
(ii) X” is a subcomplex.
(iii) X"-1 U (Ugea C2) is a CW-subcomplex where {c? : a € A} is any
collection of n-cells from X.
(iv) The relative topology on X” is the same as the weak topology that
X™” inherits from the closure of its cells.
(v) AC X is closed if and only if AN X” is closed for eachn. O
(8.E.12) Exercise. Suppose X is a CW-complex, and suppose X"~! is the
(n — 1)-skeleton of X, {cR : a € A} is the collection of n-cells of X, and
(Bolignat : S”™-1 - X"-1 : a € A} is the collection of sewing maps. Then
X” = (Yaea BR) Up X"! where F is the obvious map from YacaS"™! 3
AO,
(8.E.13) Corollary. If X is a CW-complex, then X is normal.
Proof. An immediate consequence of (8.E.12) and (4.B.4) is that X”
co
is normal for each n. By (8.E.12)(v), X = U X” with the weak topology,

=1
and then by (7.D.2) we have X is normal. i 0
The next theorem, while of interest in itself, is the remaining crucial
result that will enable us to build CW-complexes starting with standard
balls.
CW-Complexes 207
(8.E.14) Theorem. Let X be a CW-complex.

(a) If X has only a finite number of cells, then X is compact.
(b) If C C X is a compact subset of X, then C is contained in a union
of a finite number of closed cells (hence is contained in a finite subcomplex).
Proof. (a): This is trivial since X can be written as a finite union of
compact sets, the images of the closed cells.
(b) Let F C C consist of one point from each (open) cell that meets C.
If c be any cell of X, then €/ F is finite by condition (C). Therefore F' is
closed, hence compact. But F has the discrete topology since every subset
of F is also closed. Thus F' must be finite. Then C C Wve pCa Where Cz is
the cell such that r€c,NC. O
(8.E.15) Construction of CW-complexes starting with standard

balls. Start with the disjoint union of a nonempty collection of 0-balls.
This is the the 0-skeleton X°. Note that, trivially, X° with the disjoint
union (discrete) topology is a CW-complex. Now let {B1 : a € Ai} bea
collection (possibly empty) of closed 1-cells. For each B} pick a continuous
map fa : So = Fr(B1) > X° (the sewing map). Define f : (UacaS®) 4 X°
to be the obvious map, and let X! = (UacaB) U; X°. The 6-th character-
istic map is Fg = Po ig where ig : B! > UBL is an inclusion map and P
is the quotient map from (|JB1) U X° > ((JB}) U; X°. It is clear that each
characteristic map sends the bounding 0-sphere to at most two points, so
condition (C) is clearly satisfied by X!. Furthermore, X! satisfies condition
(W). By (8.B.5) X! is Ty, and hence by (8.B.4) X! is normal. Thus X? is
T>, and hence is a CW-complex with 0-skeleton X°. One may now continue
in the same fashion. The only additional complication is that starting with
mn = 2 you must observe that the images of the sewing maps are compact
and use (8.E.14) to obtain condition (C).
(8.E.16) Definition. A CW-complex is locally finite if and only if each
xz € X has a neighborhood that lies in a finite subcomplex of X.
(8.E.17) Examples.
1. Any CW-complex that is locally compact is locally finite (see the
next theorem).
2. As an example of a CW-complex that fails to be locally finite, let
I,, = I for each n € Z* and let X be the space obtained by identifying all
{0} € In together. Then X with the obvious cell structure is not locally
finite.
Locally finite CW-complexes are easily characterized.
(8.E.18) Theorem. A CW-complex X is locally finite if and only if X is

locally compact.
Proof. Suppose X is locally compact and z € X. Then X has a compact

neighborhood which by (8.E.14) is contained in a finite subcomplex.
Conversely, since finite subcomplexes are compact (8.E.14), local finite-
ness implies local compactness. O
At this point it should not be surprising that any CW-complex may be
considered as a quotient space. Suppose X is a CW-complex. For n € N, let
Xn = (Uaea, Br) UX"-}. Let X_; = and X be the disjoint union of the
spaces X". Define P: X > X by setting P(x) = P,(z) if x € X", where
P,, is the quotient map Xn > (ie By) U; X"-1, It is easy to see that P
is continuous. We show next that P is an identification.
(8.E.19) Theorem. The map P : X > X is an identification.
Proof. Suppose U Cc X and P~!(U) is open in X. We show U is
open by verifying UN X” is open in X” for each n. Since the maps P,, are
identifications, it follows that UN X” will be open if and only if P-'(UNX")
is open in X,. Since P>}(UN X") = P(U) N ((Uaca,B27 UX"), the
result follows. O
It is now immediate from (8.C.5) that X is homeomorphic to X fiGe:

The next result is used in a later chapter to show that locally finite CW-
complexes are metrizable.
(8.E.20) Theorem. If X is a locally finite CW-complex, then

(i) P is a closed map, and
(ii) P~1(x) is compact for each x € X.
Proof. We prove the first part of the theorem and leave the second half
to the intrepid reader. Suppose K is a closed subset of X. We show that
P(K) Cc P(K). Suppose z € P(K). Since X is locally finite, there is a
neighborhood U of z that is the union of only a finite number of closed cells.
Let T = P(K)NU. Note that x € T. For each n > 0 anda E Ap, let
Ke = K (8%), Gas (Kaq)} dnd Vee= PAU) Ox B2)» It follows from
(8.E.16) that with the exception of a finite number of a € A, all the V,
are empty. Therefore, only a finite number of the G, can intersect U. If $
denotes the union of these Gq’s, then clearly S is compact and furthermore
T = P(kK)NU = SNU. Thus T is compact and hence closed. Since
z€T=TC P(K), the first part of the theorem is established. 0
F. UPPER SEMICONTINUOUS DECOMPOSITIONS:

AN INTRODUCTION
In recent years, there has been considerable interest in quotient spaces result-
ing from what are called upper semicontinuous decompositions. For readers
Upper Semicontinuous Decompositions: an Introduction 209
whose interest is sparked by this section, we highly recommend the very

readable book by Daverman [1986].
(8.F.1) Definition. Let G = {gq : a € A} be a partition of a topological
space X. We call G an upper semicontinuous decomposition (usc) of X if
and only if for each g, € G and for each open set U containing gq there is a
saturated open set V such that gy C V CU. (A subset V of X is saturated
if and only if V = P~!P(V), where P is the quotient map of X to X/G.)
Motivation for this terminology might be based on a consideration of

decompositions arising from the graphs of upper semicontinuous functions.
(8.F.2) Definition. A function f : X > E! is upper semicontinuous if and

only if for each a € €', {x : f(x) < a} is open.
A graph of one such function where X = €! is seen in the following
figure.
(2, f(z))
(2,0)
The collection of segments connecting points (z, 0) with their image (2, f(x)),
together with individual points of €? that do not lie in any of these segments
yields a decomposition that is upper semicontinuous.
Geometrically, the preceding example of an upper semicontinuous de-
composition illustrates what is happening: little sets in the decomposition
may approach big ones, but the converse is not allowed (this will be made
more precise by theorem (8.F.10)). That such decompositions are by no
means rare is be seen from the following exercise.
(8.F.3) Exercise. Show: A decomposition G of a space X is usc if and

only if for each open set U of X, the union of all the sets of G contained in
U is open.
The following characterization of usc decompositions is frequently useful.
(8.F.4) Theorem. A decomposition G of a space X is usc if and only if

for each closed set D of X, the union of members of the decomposition that
intersect D is closed.
Proof. Suppose G is a usc decomposition of X and D is closed in X.
Let F =U{g: 9 € Gand gND #9}. Since X \ D is open, it follows that
U=U{g : 9 € Gandg C X \ D} is open by (8.F.3). Hence F is closed
since F =X \U:
Conversely, suppose whenever D is closed U{g : 9 € G and gND #9}

is closed. Let g € G be a subset of U where U is open in X, and V = U{g :
g€Gandg CU}. ThenV=X\U{g: 9 € Gand gN(X \U) #9}, and
hence V is open and saturated. It follows from the definition that G is usc.
O
(8.F.5) Examples—Exercises.
(a) Let X = €?. For each x € &! let g, = {(z,y) € E? : —co < y < oo},
and G = {g, : z € €'}. Show that G is usc and X/G is homeomorphic to
Ex:
(b) Suppose that X = €! and G is the usc decomposition of €1 whose
nondegenerate elements (i.e. those that are not singletons) consist of closures
of components of the complement in J of the Cantor set. Show that &} is
homeomorphic to €1/G.
(c) Suppose X = S! and the only nondegenerate element of a decom-
position G consists of the three vertices in S! of an equilateral triangle.
Describe X/G.
(d) Suppose X = S! and G is the collection of all triples s;, 82,83 € S!
that form the vertices of an equilateral triangle. Describe X/G.
(e) Suppose X = €° and the only nondegenerate element of G is the
unit circle. Are €° and X/G homeomorphic?
(f) Suppose X = €? and G consists of the origin, and all 2-spheres of
the form {(z,y,z) : 2? +y?+z* =r?} where 0 <r < oo. Describe X/G.
The notions of lim inf and lim sup are quite useful in the study of usc
decompositions.
(8.F.6) Definition. Suppose {A,} is a sequence of subsets of a space X.

We define lim inf{An} to be {x € X : every neighborhood of x intersects
all but a finite number of the A,}, and lim sup{A,} to be {x € X : every
neighborhood of « intersects infinitely many of the A,}. A set L is said to
be the limit of the A, if and only if lim inf{A,} =lim sup{A,} = L.
(8.F.7) Examples.
1. For each n € Z", let A, = {(z,y) € €? : g =I/n and 0 <y <1}.
Then lim inf{A,} =lim sup{A,} = {(z,y) € €? : 2 =0,0<y <1}.
2. For each n € Z*, let Aon = {(z,y) € E? : © =1/2n and 0 < y < 1}.
and Agny1 = {(2,y) € E€? : = 1/(2n+1) and -1 < y < 0}. Then
lim inf{A,} = {(0,0)}, and lim sup{A,} = {(z,y) € €? : « = 0 and
—Leeyly.
3. Let An = {(2,y) € E? : = 1/n for each n € Zt and suppose that
y € [0, 1/3] U [2/3, 1]}. Note that neither lim sup{A,} nor lim inf{A,} is
connected.
(8.F.8) Exercise. Suppose {A,} is a sequence of subsets of a space X.

Show: lim inf{A,} Clim sup{A,}, and both of these sets are closed.
(8.F.9) Theorem. Suppose {A,} is a sequence of connected subsets of a
compact T2 space X, and lim inf{A,} # 9. Then lim sup{A,} is connected.
Proof. If T =lim sup{A,} is not connected, then T may be written as
the union of disjoint closed subsets D and F.. By (8.F.8), D and F are also
closed in X. Since X is normal, there are disjoint open sets U and V such
that D C U and F CV. Observe that the sequence {A,} is eventually in
UUV: for if not, there would be a sequence of points {tn,;}, 2n; € An;, lying
outside of U UV with a cluster point z in X \(U UV). However, it then
follows that x € T, a contradiction.
Since lim inf{A,} 4 0, there is a point common to both lim inf{A,}
and either U or V, say U. Then, U intersects all but finitely many of the
sets {A,}. Note, however, that for sufficiently large n, if A, NU # 0, then
AnNV = 9 (the A,,’s are connected and eventually lie in U UV). Since this
implies VN lim sup{A,,} = @, we have reached another contradiction. O
The next theorem helps make precise what we mean by saying that small
sets may approach large sets in a usc decomposition, but big sets must keep
their distance from a given small set.
(8.F.10) Theorem. Let G be a usc decomposition of a T3 space X into
closed sets. Suppose {g,} is a sequence of elements in G, g € G, and
gM lim inf{g,} 4 0. Then lim sup{g,} is a subset of g.
Proof. Suppose x € gNMlim inf{g,} and y € gNlim sup{gn}, where
94 9(9 € G). Since X is regular, there are disjoint open sets U and V
in X such that z € U and g € V. We may assume V is saturated. Since
x €lim inf{g,}, there is an N > 0 such that for each i > N, we have
9, 1U #9. On the other hand, this cannot happen since an infinite number
of the gn lie entirely in V. Thus lim sup{g,} is contained ing. O
In compact metric spaces, the property described in the previous theo-
rem characterizes usc decompositions.
(8.F.11) Exercise. Suppose G' is a decomposition of a compact metric

space into closed subsets. Show: G is usc if and only if lim sup{gn} C g,
whenever {gn} is a sequence of elements of G with lim inf{g,} OG # 0.
We next show a natural way of generating usc decompositions. If f is

a closed continuous map from X onto Y, then {f~'(y) : y € Y} forms a
usc decomposition of X. This is an easy consequence of the following basic
result.
(8.F.12) Theorem. Suppose G is a decomposition of a space X. Then G

is a usc if and only if the quotient map P : X > X/G is closed.
Proof. Suppose G is a usc decomposition of X and F C X is closed.

Then P-1(P(F)) = {9 : 9€ Gand gNF #9}. The latter set is closed by
(8.F.4). It follows that P(F) is closed since X/G has the quotient topology.
Conversely, suppose P is a closed map and F is a closed subset of X.
We show that U{g : 9 € G and gN F F 9} is closed. Since P is closed, so
is P(F). Therefore, U{g : 9 € G and gN F # 0} = P™*(P(F)) is closed.
Hence G is usc. O
(8.F.13) Corollary. If f : X — Y is closed, onto and continuous, then

G={f-'(y) : y € Y} is a usc decomposition of X.
Proof. Consider the following diagram
7G
By (8.4.6), ®¢ is a homeomorphism. If A is a closed subset of X, then
P(A) = 67° f(A), which is closed in X/G. Hence P is a closed map and G
1S isc, © 4E)
Components of compact metric spaces also yield usc decompositions.
(8.F.14) Theorem. Suppose that X is a compact metric space and let

G = {gC X : gis acomponent of X}. Then G is a usc decomposition of
X.
Proof. Apply (8.F.11): Suppose {g,} is a sequence of elements of G,
g € G and (lim inf{gn}) NG #9. By (8.F.9), L =lim sup{g,} is connected,
and since L intersects g, it must lie in the component g. O
In general, if f : X — Y is continuous, onto, and an identification, then

the map ®; is a homeomorphism. Thus, any information gleaned from X/G
is immediately applicable to Y. Hence it is natural to inquire which prop-
erties of X are inherited by X/G. Of particular interest is the question of
whether X and X/G are themselves homeomorphic. If G is a usc decompo-
sition of X, then by (8.F.12), X/G will retain those properties of X that are
preserved by closed continuous mappings, e.g., connectedness, compactness,
local connectedness, separability, etc. Nevertheless, a number of things may
go awry. For instance, not even the To separation property need be retained.
(8.F.15) Exercise. Suppose X = (0,00) is given the topology whose open
sets are of the form (0,b), where b € R. Let G = {A, B}, where A is the set
of positive rationals and B is the set. of positive irrationals. Show: G is usc,
X is To, and X/G is not Tp.
More topological structure is preserved under usc decompositions if

members of the decomposition are closed. However, some difficulties still
remain.
(8.F.16) Example. Suppose that X = €? and let G be the usc decompo-

sition of X whose only nondegenerate element is the real line, considered as
L = {(z,0) € €? : -co < & < oo}. Then X/G is not even first count-
able. To see this, suppose U; D> U2 3 --- is a countable basis at the point
P(L) € X/G. Then P~'(U,), P~1(U2),... is a decreasing sequence of open
subsets of X containing L. For each n € Z*, choose yn, > 0 such that
(n, Yn) € P~*(Un). Then V = €? \ {(n, yn) : n € Zt} is a saturated open
set containing L, but no U,, is contained in P(V).
It follows from the preceding example that first and second countability,
and in particular metrizability are lost under usc decompositions. However, if
additional conditions are imposed on the nondegenerate elements, somewhat
greater control is gained over the resulting decomposition space.
(8.F.17) Theorem. Suppose G is a usc decomposition of a space X into
compact subsets.
(i) If X is To, then X/G is Tp.
(ii) If X is second countable, then X/G is second countable.
(iii) If X is metrizable, then X/G is metrizable.
Proof. (i): This is easy, since compact subsets of a T2 may be enclosed
in disjoint open sets. ;
(ii): Let B be a countable base for X, and B be the family of all finite
unions of members of B. For each B € B, let B= U{g eG: 9g Cc Bh.
Then {B : B € B} is clearly a countable collection of saturated open sets in
X. We show that {P(B) : B € B} is a basis for X/G. Suppose P(g) € U,
where g € G and U is open in X/G. Since g is compact, there is a finite
number of members of B that cover g and lie in P~'(U). If V is the union
of these sets, then V € B. Hence, the corresponding set V contains g, and
we have P(g) € P(V) CU.
(iii): This will be an immediate consequence of the Stone metrization
theorem (10.C.7). O
Upper semicontinuous decompositions into compact sets are also well
behaved in the following sense.
(8.F.18) Theorem. If G is a usc decomposition of a space X into compact

subsets and K is a compact subset of X/G, then P~'(K) is compact.
Proof. We apply (2.H.2). Suppose C = {cg : a € A} is a family
of closed subsets in P~!(K) with the finite intersection property. Let F
be the family of all finite intersections of members of C. Then F also has
the finite intersection property, as does the collection {P(F) : F € F}.
Since K is compact, there is a point x € (){P(F) : F € F}. Then clearly,

{P-!(x)NCy : a € A} isa family of closed subsets of the compact set P~'(z)
and enjoys the finite intersection property. Hence, the intersection of all of
the members of this collection is nonempty. Consequently, (\{Ca : a € A}
is nonempty, which says that P~!(K) is compact. O
(8.F.19) Exercise. Suppose G is a decomposition of a metric space into

compact subsets. Show: the following are equivalent:
(i) G is usc, and
(ii) If whenever xj,yi€ gi € G (i=1,2,...) and the sequence {z;} con-
verges to a point x € g € G, then there is a subsequence {yn,} of
{yi} that converges to a point y € g.
Connectedness does not fare so well under decompositions into compact

subsets. For instance, if X = [0,1] and G is the usc decomposition of X
whose only nondegenerate element is the set g = {0,1}, then P(g) does not
pull back to a connected subset. This is corrected in the obvious manner.
(8.F.20) Theorem. If G is a usc decomposition of a space into connected

subsets, and D is a connected subset of X/G, then P~!(D) is connected.
Proof. Suppose P~!(D) is not connected. Then P~1(D) may be written
as the disjoint union of closed (in P~!(D)) sets A and B. Since each member
of the decomposition is connected, it follows that both A and B are saturated.
Select two closed subsets A and B of X such that AN P~!(D) = A and
BO P-'(D) =B. Since P is a closed map, P(A) MD and P(B)MD split D
into disjoint closed subsets, contradicting the connectedness of D. O
(8.F.21) Definition. Decompositions whose elements are connected and

compact are called monotone decompositions.
(8.F.22) Corollary. If G is a monotone usc decomposition of a space X

and C’ is a compact connected subspace of X/G, then P~!(C) is compact
and connected. O
The following rather amazing result shatters any notions that the reader
might have that monotone usc decompositions are a panacea for the problem
of inheritability.
(8.F.23) Theorem (Hurewicz [1930]). Suppose K is a compact metric

space. Then there is a monotone usc decomposition of €° such that K is
homeomorphic to a subset of €7/G.
Proof. Let T be a standard tetrahedron in €2 and let f,; and fy be

nonintersecting edges of T.
Let C; and C2 be Cantor sets in £; and £ respectively. By (6.C.12), there are

maps f; :C, ~ K and f2: C2 > K that are continuous and onto. For each
z € K, let A, = f;'(z) and B, = fy 1(a). Elements of the decomposition
are defined as follows. If « € K, let gz be the subset of €? consisting of
the union of all possible line segments running from points of A, to points
of B,. The reader may show with the aid of (8.F.19) that the monotone
decomposition of € with nondegenerate elements {g, : z € K} is usc.
If G is the usc decomposition constructed above, then K may be embed-
ded in €?/G: Let j : C; — E? be the inclusion map and define H : K > €3/G
by H(z) = Pjf;'(z). It is clear that H is continuous and 1-1. Since K is
compact, it follows that H is the desired embedding. O
The implications of the preceding result are startling. For example, it

follows that the Hilbert cube may be embedded in €?/G for some monotone
usc decomposition G of €°. Hence the decomposition space under a mono-
tone usc decomposition may differ drastically from the original space. Ob-
viously, additional conditions are necessary in order that the decomposition
space associated with a space X be homeomorphic to X. R.L.Moore [1925]
showed that if the plane is decomposed upper semicontinuously into cellular
sets, then the resulting decomposition space is again a plane. An identical
result holds for cellular usc decompositions of 2-manifolds. Whyburn [1936]
seems to be the first to pose the obvious question: If G is a cellular decompo-
sition of €3, is €?/G homeomorphic to €3? This issue remained unresolved
for many years until Bing constructed the dogbone space outlined below. Ar-
mentrout [1970] as well as others subsequently constructed decompositions
of €3 for which the decomposition space was not homeomorphic to €°. This
line of research led to the following result of Bing and Armentrout: If G is
a usc decomposition of a 3-manifold into cellular sets. Then M/G is home-
omorphic to M if and only if G is shrinkable. The necessary definitions and
exposition can be found in the aforementioned book by Daverman, as can
subsequent results by Edwards with regard to approximating decomposition

maps in manifolds of dimension > 5 by homeomorphisms.
We conclude this section with a brief sketch of the celebrated dog bone
space, a decomposition of €3 unearthed by Bing [1957a]. The nondegenerate
elements of this decomposition are constructed as follows. Start with a solid
double torus Tp lying in €? (see next figure). Inside Tp intertwine four solid
double tori as indicated in the figure. Let 711,712,713, 714 denote these tori,
andulet Li ee T,;. In each of the T),, intertwine four more tori in exactly
the same fashion as the previous four tori were embedded in To. Let T2 be
the union of these sixteen tori. This procedure is repeated inductively to
obtain a sequence of sets Ty, 7,..., where each T; is the union of 4* tori. It
may be shown that ();~, T; is a collection of arcs (actually, a “Cantor set” of
arcs in the sense that the intersection of the nondegenerate elements and a
vertical plane yields a subspace homeomorphic to the Cantor set). The dog
bone decomposition G of E? consists of these arcs, together with single points
in the complement of ();", T;. Bing showed that €3/G is not homeomorphic
to €°, which seemed remarkable enough at the time, but a few years later he
showed that €2/G x €! is homeomorphic to €*. How does that strike your
intuition?
PROBLEMS
Sections A and B
1. Suppose (X,d) is a pseudometric space. Define an equivalence relation

on X by setting x ~ y if and only if d(z,y) = 0. Show that X/~ is
metrizable.
Problems PAT /
2. Suppose X and Y are topological spaces, f : X + Y is continuous and

onto, and V is a topology for X/Gy that is strictly smaller than the
quotient topology UW. Show: ®y is not continuous with respect to V.
In €?, declare two points to be equivalent if and only if they lie on
the same circle with center at the origin. What previously encountered
space is homeomorphic with €?/~?
Describe how a 2-holed torus (not solid) can be derived as a quotient
space (starting with the square in €).
Describe how a Moebius strip can be viewed as a quotient space.

Suppose ~ is an equivalence relation on a space X, ~ is an equivalence
relation on a space Y, and f : X — Y is a continuous map with the
property that if 21 ~ x2, then f(x) = f(x2). Show that the function
f :X/~ — Y/s defined by f((z]) = [f(z)] is well defined and continu-
ous.
Is the hypothesis “Y is T,” necessary in (8.4.7)?
The projective plane P» is the quotient space obtained from B? by iden-
tifying antipodal points of S'. Show that P: is homeomorphic to the
quotient space obtained from €? \ {0} by identifying points x and y if
and only if they lie on the same line passing through the origin.
Suppose X = Uc, Xo is given the weak topology (induced by the
family {Xq : a € A}). Let Y be the free union of {Xq : a € A}, and
let h : Y ~ X be the obvious map. Show that h is continuous and
Y/G»p is homeomorphic with X.
10. Let B? be the unit disk in €?, S! be its frontier, and f : S' > S! Cc B?
be continuous and onto. Must B? Uy B? be a 2-manifold? If f is a
homeomorphism, is B? Us B? a 2-manifold?
tt; In S?, every simple closed curve separates S*. Is the same true in the
torus, the Klein bottle, the projective plane? (See problem 8.)
12: Show that the space obtained by sewing a disk homeomorphically along
its boundary to the boundary of a Moebius band is homeomorphic to
the projective plane. (See problem 8.)
13. Show that none of the T; properties are necessarily retained when taking
quotients.
14. Show that the following spaces are not homeomorphic: Let A be the
cone over {(n,0) : n € Z*}; let B be the-union of {(n,0) : n € Zt}
and all points in the plane lying on line segments connecting points of
{(n,0) : n € Zt} with (0, 1), (with the relative topology from £7) [Hint:
Use first countability.]
15. Let X = (0,1) x [0, 1] and identify all points of the form (x, 1) with each
other. Show that the resulting quotient space is not homeomorphic to
Y, where Y is the union of the interior of the triangle with vertices
(0,0), (1,0), and (0,1), the open line segment {(z,0) : 0< a < 1}, and
the point (0,1) (with the relative topology from €”).
16. Identify the boundary points of a square so that the resulting quotient
space is not a 2-manifold.
Section C
1. Show: retraction maps are identification maps.

2. Suppose f : X > Y is 1-1. Show that f is an identification map if and
only if f is a homeomorphism.
3. Show: If fa : Xa — Yq is continuous, open, and onto, for each a € A,
then [] fa: [loca Xa > [aca Yo is an identification.
4. Find an identification map that is neither open nor closed.
Section D
1. Find an example of a space that is not a k-space.

2. Suppose X is a k-space, Y is a topological space, and f : X > Y is an
identification. Show that Y is a k-space.
3. Suppose G is a partition of a locally compact space X. If X/G is To,
show that X/G is a k-space.
4. Suppose X is a T, k-space. Let {Cy : a € A} be a collection of compact
subsets of X. Let Y be the disjoint union of the Cy with the disjoint
union topology. Show that Y is locally compact and X is homeomorphic
to a quotient space of Y.
5. Show that the product of uncountably many copies of €! is not a k-space.
(Consider A = {{ra} € []aeq E€* : for some integer n > 0, tq =n for
all but at most n coordinates and xg = 0 otherwise}.)
6. If f : X — Y is onto, continuous, and open, show that Y has the strong
topology.
Problems 219
7. Suppose X is a set and (Y,U) is a topological space. Let f : X 3 Y be

onto and give X the weak topology, W. Let S be the strong topology
for Y (with respect to f and W). Compare U and S.
Are closed subsets of k-spaces necessarily k-spaces?
Suppose X is a space for which every compact subset of X is closed. Let
X™* be the 1-point compactification of X. Show: every compact subset
of X* is closed if and only if X* is a k-space.
10. Suppose (X,/) is a topological space and let K be the k-topology asso-
ciated with (X,U/). Is (X,K) a k-space?
Section E
1 Let X ={(a,9,2) € €* = 2? £4? +2? = 1) Pind

(a) two distinct cell structures for X such that the resulting CW-
complexes have the relative topology, and
(b) two distinct cell structures for X such that the resulting CW-
complexes yield different topologies for X, neither of which is the
relative topology.
2. Show that €” with the usual topology is a CW-complex.
Show that S! with the usual topology is a CW-complex.
Is the following space a CW-complex (with the relative topology from
ea
5. Is there a locally finite CW-complex X such that X"\ X"~! # O for

each n?
Represent the Moebius strip as a CW-complex.
7. Represent the torus and the Klein bottle as CW-complexes.
Show that the projective plane is a CW-complex with one 0-cell, one
1-cell and one 2-cell, Describe the characteristic maps.
9. Show: A CW-complex is connected if and only if X! is connected.
Section F
1. Show that if X is normal, and G is an upper semicontinuous decompo-

sition of X into compact sets, then X/G is normal.
2. Show that if X is separable and G is an upper semicontinuous decom-
position of X, then X/G is separable.
3. Let G be a decomposition of €?, whose only nondegenerate element is
the x axis. Show that X/G is not locally compact.
4. Show that if G is an upper semicontinuous decomposition of a regular
space X into compact sets, then X/G is regular.
5. Suppose G is a decomposition of a space X such that G has only finitely
many nondegenerate elements each of which is closed in X. Show: G is
usc. Is “closed” necessary?
6. Show: A decomposition G of a space X is usc if and only if for each
g € G and each open set U containing g, there is an open set V such
that g C V C U and with the property that if g € Gand gNV £9,
then g C U.
7. If Gis a usc decomposition of a normal space X into closed sets, show
that X/G is T>
8. Show: If G is a usc decomposition of a Tp space such that G has only

one nondegenerate element, then X/G is Ty. Can To be replaced by T;?
9. IfG is ausc decomposition of a locally compact T> space X into compact
sets, show that X/G is regular.
10. IfG is a usc decomposition of a locally compact Tz space X into compact
subsets, show that X/G is locally compact.
11. Let {C; : 1 € Zt} be a family of connected subsets of a T) space X
and LJ;=, Ci be compact. Show: If lim inf{C;} # 0, then lim sup{C;} is
connected.
12. Suppose G is a monotone usc decomposition of €? and N is the union of
the degenerate elements of G. Show that P| embeds €° \ N in €3/G.
Chapter 9
CONTINUA
This chapter is devoted to a historic and important part of point-set topol-

ogy: the study of continua. A continuum is a connected compact topological
space. Although the literature involving continua is extensive, we limit our
presentation to establishing a few of the better known theorems and to ex-
amining a few of the more bizarre examples of continua that arise. For the
reader whose interest is tweaked by this chapter, we recommend the book
by Nadler [1992].
A. THE ARC
Our study of continua begins with an investigation of arcs. Recall that an
arc is defined to be a homeomorphism from the closed interval J into a space
X. However, here, as is often the case in mathematics, the definition is
not always easy to apply. The primary goal of this section is to obtain a
characterization arcs that is sometimes more convenient than the definition.
An arc is obviously a metric continuum, but then so are a host of other
spaces that are not arcs (as usual, the arc will be confused with its image
when it is convenient to do so). Intuitively, it seems that an arc is as “skinny”
as a continuum can be. However, circles, trees, etc., are just as slender. The
question, then, is how do these spaces differ from an arc? Notice that if any
point is removed from a circle, the resulting space is still connected. Further-
more, various points can be extracted from a tree without disconnecting it.
On the other hand, if any point is plucked from an arc (with the exception
of the two endpoints), the resulting space is not connected. Trivial as this
observation may be, it is the keystone of the arc. We shall work toward
221
222 9. Continua
proving that any metric continuum with exactly two “non-cut points” is an
arc. :
a tree
(9.A.1) Definition. Suppose X is a connected topological space and that

xz € X. We say x is a cut point of X if and only if X \ {x} is disconnected.
Points that are not cut points are called non-cut points. A cut point = is
said to separate points a and b in X if and only if there is some separation
(U,V) of X \ {x} witha €U and be V.
A basic property of J is that it is linearly ordered; furthermore, the
usual topology for J coincides with the topology induced by the usual or-
der relation. Hence it should not be too surprising that efforts to obtain a
characterization of the arc will involve the introduction of a linear order on
certain metric continua—a linear order for which the original topology and
the order topology are identical.
(9.A.2) Definition. Let X be a connected topological space and a,b € X.

Suppose S(a, b) is the set {a,b} U{p € X : p separates a and b in X}. The
separation order on S(a,b) is defined as follows. If s;,s2 € S(a,b), we say
$1 < Sq if and only if
(2) S) = G OF > = 6,.0r
(it) s, separates a and sq in X, or
(iti) 51 = 89.
The following lemma and its corollary shows there is more symmetry in
the definition than appears at first glance.
(9.A.3) Lemma. Suppose X is a connected topological space, a,b € X,

$1, $2 € S(a,b) \ {a,b}, and U;,,Vs, is a separation of X \ {s;} with a € U,,
and 6 € V;,. Then, s1 € Us, if and only if sz € Vs, in which case U;, C Us,
BOG Ved Vga
Proof. Suppose s; € Us,. By (2.B.11), Vs, U {s2} is connected. Since
b € (Vs, U {82}) NVs,, by (2.4.6) we have (V;, U {s2}) C Vz,. Hence so € Vs,
and V;, C V;,. The other half is similar. O
(9.A.4) Corollary. Suppose X is a connected topological space, abe x

and 81,82 € S(a,b). Then s; < sg if and only if
(2) {= @.0r 65 ='0- or
The Arc 223
(ii) sg separates s; and b in X, or

(zit) S81 = S92.
Proof. Direct from the definition and the above lemma. 0

One should note that the separation order when applied to I yields the
usual order.
(9.A.5) Theorem. The separation order < is a linear order on S(a, b).
Proof. The proof is in two steps.
1. The separation order < is a partial order.
We prove only the transitivity of <. Antisymmetry is established by
Exercise (9.4.6). Suppose s; < sg and sz < 83. Only the trivial cases are
omitted if we assume 51, S2,s3 € S(a,b) \ {a,b} and s; # so # 83. By the
definition, there are separations U;,, Vs, of X \{s;} with a € Us, and b € V,,.
We want s3 € V;,. But by (9.A.3) V;, C Vs,. Hence s3 € V,, C Vay.
2. The separation order < is a linear order.
Suppose s1,52 € S(a,b). We must show either s; < sg or 82 < 3}.
Again, we ignore the trivial cases by assuming s1, 82 € S(a,b) \ {a,b}, and
Ss; # S2. Let (U,V) be a separation of X \ {si}, with a € U and be V.
Either s2 € V (in which case s; < s2 by the definition and we are finished),
or $2 € U (in which case corollary (9.A.4) gives so <s,). O
(9.A.6) Exercise. Show: if s1 < s2 and sg < $1, then s; = sg.
Since the separation order is a linear order, there are two natural topolo-
gies for S(a,b): the order topology and the relative topology from X. In
general, these topologies need not be the same, but if we start with a T; con-
tinuum having exactly two non-cut points, then the two topologies coincide.
(9.A.7) Theorem. Suppose X is a 7; continuum with exactly two non-cut

points a and b. Then S(a,b) = X, and the order topology induced by the
separation order is precisely the original topology on X.
The proof of the theorem is based on the following lemma.
(9.4.8) Lemma. Suppose K is a T; continuum, p is a cut point of K, and

(U,V) is a separation of X \ {p}. Then each of U and V contains at least
one non-cut point of X.
Proof. Suppose U consists entirely of cut points (of X). For each x € U,
let (Uz,Vz) be a separation of X \ {x}, where p € V;. Partially order
U = {Uz : x € U} by inclusion and let W = {Uz, : a € A} be a maximal
nest in U (0.D.4). Claim: (){U;, : a € A} = 0. Suppose to the contrary
z€({Uz, : «a€ A}. For each rq note
(i) PEU,Ws0V, Ufre) CX \ {2} =U, 0M;
224 9. Continua
(ii) p € Ve, MVz, so the connected set Vz, U {ra} is contained in V;,
and hence, 2q € V;.
Consequently, U, U {z} is contained in Uz, UVz,, and since z € U;z,, it
follows that U, is a proper subset of Uz, for each tq. This contradicts
the maximality of the nest W. Thus (){Uz,, : a € A} = 9, and furthermore
{Uz,,U{ta} : a € A} = 0 (why?). Therefore, it follows that {Vz, : a € A}
is an open cover of the compact set X that has no finite subcover (why’),
which is impossible. Hence, we conclude that U must have a non-cut point.
The argument for V is similar. O
Proof of 9.4.7. It is obvious that X = S(a,b), since if p € X \ {a,b},
then p is a cut point and thus there is a separation (U, V) of X \{p}. Lemma
(9.4.8) (together with the hypothesis of the theorem) guarantees that a and
b are separated by p, and hence p € S(a,)).
Let 7 denote the given topology for X, and let F denote the order
topology. To show that F C 7, it is sufficient to prove that sets of the form
West aca <p} and W = {x : @ > p} are in JT. Suppose p € X \ {a, db}.
Then there is a separation (Up, Vp) of X \ {p} such that a € U, and b € Vp.
We show that W = Up, and a similar argument may be used to show that
W = Vp.
Suppose z € W. If x € Vp, then (by the definition of the separation
order) z > p. This contradicts c € W; hence we have W C Up.
Now suppose z € U,. By (9.A.3) this implies that p € V, which by
definition says z < p. Hence x € W, and U, = W.
We now show 7 C F. Let O be an element of the given topology T
and x € O. It suffices to find an (<)-interval (c,d) such that x € (c,d) C O.
If no such interval exists, then the collection {[p,q]N (X \O) : x € (p,q)}
has the finite intersection property. Since X is compact, the intersection
of all the members of this family is nonempty and is contained in X \ O.
(Note: [p,q] is closed, since F C 7.) However, it is easy to show that
(\[p,q] : « € (p,q)} = {x} C O). This should strike the reader as being
a bit unusual. Therefore, the desired open interval exists, and consequently
T CF. (The pedant should check the cases t =a, andzr=b.) O
Now that the candidate for arc status has been successfully ordered, the
task remains of defining a homeomorphism from K onto J. The next lemma
gives us a way to start such a map.
(9.4.9) Lemma. Suppose A is a countable linearly ordered set such that

(i) A has no smallest or largest element, and
(17) ifa,b € A with a < b, there is ac € A such that a<c< b.
Then, there is an order-preserving bijection h : A + D, where D = (ie/2" %
O<k < 2” and kn G2}.
Peano Continua 225
Proof. Let A = {a,,@2,...} and define h(a,) = 1/2. Let n; be the first
integer such that an, < a; and nz be the first integer such that a; < dn,.
Define h(an,) to be 1/2? and h(an,) to be 3/2?. Et cetera. O
(9.A.10) Theorem. Suppose X is a metric continuum with exactly two
non-cut points a and b. Then X is homeomorphic to J.
Proof. By (2.1.13), there is a countable dense subset A of X\{a, b}. Since
K is connected, it follows that A (ordered by the separation order) satisfies
the two conditions of the hypothesis of the preceding lemma. Consequently,
there is a function h mapping A onto D that preserves order. Let h(a) = 0
and h(b) = 1. Suppose p € X \ (AU {a,b}); we must define h(p). Since p is
a cut point, there is a separation (Up, V,) of X \ {p}, such that a € U, and
bE V,. Note that 0 < sup{h(x) : x € ANU,} < inf{h(z) : c € ANV,} <1,
and an elementary argument yields sup{h(r) : x € ANU,} = inf{h(z) :
xz € ANV,}. Define h(p) to be this common value. It is not difficult to check
that h is the desired homeomorphism. 0
Characterizations are not only aesthetically pleasing in themselves, but

are often useful in proving theorems. For example, we use (9.A.10) in the
proof of (9.B.1).
In the problem section, a characterization of 1-spheres is obtained. The
2-spheres, however, are somewhat more difficult to handle. Perhaps the most
elegant characterization of the 2-sphere along these lines is due to Bing [1946],
who showed that a locally connected metric continuum X is homeomorphic
to S? if and only if X is separated by each simple closed curve contained in
X and is not separated by any pair of points in X. This is known as the
Kline sphere characterization.
Can 3-spheres be so neatly categorized? One of the great unsolved
problems in mathematics deals specifically with this problem. The Poincaré
conjecture asserts (in terminology that will become comprehensible once
Chapter 12 is completed) that every compact simply connected 3-manifold
is homeomorphic to S*. Since the early 1900’s, this conjecture has plagued
many of the world’s outstanding mathematicians, and as yet no one has
been able to prove or disprove it. Different versions of its generalization to
dimensions (n > 5) have been proven by various people in the 1960’s. See
Rushing[1973] for a discussion and three different proofs. Friedman [1982]
gave a proof in the 4 dimensional case. This is an excellent illustration of
one of mathematics more intriguing aspects; the simplicity of the statement
of a proposition often belies the difficulties encountered in its resolution.
B. PEANO CONTINUA
We now turn our attention to a major class of continua, the Peano continua.
A Peano continuum is a locally connected metric continuum. Arcs are Peano
226 9. Continua
continua, but hardly representative of them. The result toward which we are
heading is the venerable (1913) Hahn-Mazurkiewicz theorem, which states
that every Peano continuum is a continuous image of the arc J. Thus there is
a continuous map from J onto the n-dimensional cube, or even more amusing,
there is a continuous function mapping J onto the Hilbert cube. The reader
whose intuition concerning continuous functions has suddenly taken a jolt is
in good company. Peano’s amazing discovery in 1890 that J could be mapped
continuously onto the unit square created havoc in the mathematical world
(but scarcely anywhere else, needless to say). Among the major casualties
brought about by Peano’s result was the then existing concept of dimension.
As a first step toward proving the Hahn-Mazurkiewicz theorem, we will
show that Peano continua are arc connected, an interesting result in itself.
The proof is based on the notion of simple chains. Recall from chapter 2: if
a and b are points in a topological space X, then a simple chain of sets from
a to 6 is a finite collection of subsets {U;,U2,...,Un} of X, where a € Uj,
bE U, and U;NU; # O if and only
if |i — j| < 1.
(9.B.1) Theorem. If X is a Peano continuum, then X is arc connected.
Proof. Let a and b be distinct points in X (if we can not find distinct
points in X, it is arc connected by default). It is necessary to construct an
arc from a to b.
We will construct a sequence of simple chains from a to b, Ci,C2,... each
composed of connected open links such that
(2) HU;,. 1s adlink of C,, then diam U;,;< 1/2".
(it) If Un, is a link of Cy, then ee CUneig 10F Some Unt © tnt
(Hence, U,,; can not intersect Un_1¢ if€<j—loré>j+1)]
(iit) If Un C Un-1,j, then no U,, for k > 7 has a point in common
with Un_-1i,m form < j — 2, and no U,,, for k < i has a point in
common with Un_-1m for m > 7 +2.
Suppose such a family of simple chains has been constructed. For n > 1
define
Ch = U Uni
Now, C, = lee (Why?), and each C,, is a metric continuum containing

a and b. Define
Then A is also a metric continuum (connectivity follows from (2.H.3), and

the other properties are obvious).
To show A is an arc, it suffices to show that every point in x € A\ {a, b}
is a cut point.
Peano Continua 997,
For each n, we have xz € Unis C Un; for some 7. Define
j-2
Dn = (J Uni and En = (J Uns.
A=1 i>j+2
It is clear that D, NE, = 0. Furthermore, by property (iii) above, we have
for each n. Let
oe UL(Dan A) and E = Ua A).

nil ns
Then D and E are open in A and D # 0 # E. It follows from (iii) that

DANE = 9. Now from (i) and the definition of D and E we get DUE = A\{z}.
Hence A is an arc from a to b—assuming that a suitable sequence of simple
chains can be constructed.
Construction of the Simple Chains: For each x € X, let U, be the
component of Sj /2(x) that contains x; then diam U, (and U,) is < 1. Apply
(2.F.2) tou = {U, : x € P} toget asimple chain Cy = {Ui1,U12,..-,U ix, }
from a to b whose links are connected open sets with diameter less than or
equal to 1.
Now for each i = 1, 2,...,«,—1, select a point x; € U;,;NUi,i41. We will
construct for each 7 a simple chain from 2; to 2;41 as follows: Let 2 € Uj j.
Case 1. If x € Uy; \ (U1,i-1 UU 1,141), we consider the set S5(rz) C U1;
where 6 = min(1/4,d(z, P \ U;,;)/2) and let Uz be the component of 55(z)
that contains x. By construction U, is open, connected and has diameter
less than or equal to 1/2, and Uc Ui
Case 2. If € Ui; U;1,;-1, one proceeds in a manner similar to case 1,
but selects 6 = min(1/4, d(z, P \ U1,;)/2,d(a,P \ Ui ,i-1)/2).
Case 3. If € U1; U1,i41, proceed as in case 2, mutatis mutandis.
Now, for each 7 < Kk, — 2, apply (2.F.2) to obtain a simple chain from
x, to £41, and similarly from a to x; and from z,,~; to b.
Now we construct Cy from those k; chains as follows. Use the links in
the simple chain linking a to x; until some link @ intersects a link m of the
simple chain from 7; to 2. Let w be the last link of the second simple chain
to meet . Then discard all of the links of the first chain past @ and all links
of the second before w and continue on the second simple chain until a link
of the third simple chain (linking x2 to z3) is encountered, etc.
It is clear from the construction that properties (2), (#7), and (7iz) are
satisfied, and that one can repeat the construction inductively to obtain the
desired sequence. O
228 9. Continua
Before proving the Hahn-Mazurkiewicz theorem, we present the reader

with the following exercise.
(9.B.2) Exercise. Suppose X is a Peano continuum and « > 0. Show:

there is a 6 > 0 such that if z,y € X and 0 # d(z,y) < 4, then there is an
arc from x to y with diameter less than e.
(9.B.3) Theorem. (Hahn [1924]; Mazurkiewicz [1920]). If X is a Peano
continuum, then X is a continuous image of J. Conversely, if X is T2 and
f : I — X is continuous and onto, then X is a Peano continuum.
Proof. Since, logically speaking, one miracle is equivalent to another, it
should not be surprising to find that the first half of the theorem is derived
from a previous result (6.C.12): Every compact metric space is a continuous
image of the Cantor set, K. We will extend the domain of a map from
the Cantor set onto X to all of J. Suppose f : K + X is the map given
by (6.C.12). Note that I \ K consists of a countable collection of disjoint
open intervals, Z = {1), I2,...}. Denote In by (an, bn), and assume that the
collection Z has been linearly ordered in a manner that respects decreasing
length.
If f(an) = f(bn), then define f(r) = f(an) for each z € (Gn, bn). If
f(an) # f(bn), the extension is slightly more complicated. From (9.B.2),
corresponding to « = 1/2”, there is a 6, such that if d(z,y) < 6,. Thus arcs
may be found from z to y with diameter less than 1/2”. Since f : K > X is
uniformly continuous, corresponding to dn, there is a yp, such that if |r—y| <
Yn, then d(f(z), f(y)) < dn. We may assume without loss of generality that
a> Bare
Corresponding to 7; there are at most a finite number of intervals with
diameter greater than 7. For each such interval J;, let A; be the image of
any arc a; in X with domain [a;,b;] running from f(a;) to f(b;), (9.B.1),
and extend the map f to I; by setting f(x) = a;(x) for x € (a;,b;). For any
interval I, with y2 < diamJ,, < 7, there is an arc A,, of diameter less than
1/2 that connects f(a@m) to f(bm). Map Im onto A in the obvious fashion.
If Jj, is such that y3 < diam I, < yz, there is an arc A,, with diameter less
than 1/2? that joins f(am) and f(bm). Again extend f to map I, onto Am.
An easy inductive argument yields an extension F' of f that maps I onto X.
We leave to the reader the task of showing that F is continuous.
The converse follows easily from (2.G.11) and (2.E.7) except for one
detail: the proof of the metrizability of X, which is deferred until the next
chapter (10.C.8). O
The foregoing theorems may be combined to yield an easy proof of the
following result.
(9.B.4) Theorem. A T> space X is arc connected if and only if it is path
connected.
Peano Continua 229
Proof. Certainly arc connectedness implies path connectedness. Sup-

pose then a #b € X and let f : I > X be a path with endpoints a and b.
Since f(J) is a Peano continuum (9.B.3), there is an arc in f(J) from a to b
(9.B.1), and consequently X is arc connected. O
The Hahn-Mazurkiewicz theorem is typical of many existence type the-
orems in that it is of no particular help in actually defining a map from I
onto a Peano continuum X. We sketch Peano’s construction of a space filling
curve, a map from J onto X =I x I.
Divide J x J into nine squares and partition J into nine subintervals.
The piecewise linear map f; from J onto the subset of J x J indicated in the
figure is easily defined: send the ith segment to the appropriate diagonal of
the ith square, starting where the (i — 1)th segment ended.
Now divide each of the nine squares into nine more squares and partition I
into 81 pieces. Define a map f2 from J into I x J by mapping each segment
in succession onto a diagonal of the appropriate square.
230 9. Continua
A pattern has begun to emerge which the reader should be able to continue.
(9.B.5) Exercise. Show that the sequence of functions f,, fo,... converges
uniformly to a (continuous) function f : ] + X. Note: The pointwise limits
{fn(z)} converge for each z, and {f,} is a Cauchy sequence in B(J,I x I)
with the sup metric.
The next theorem gives a rather elegant characterization of Peano con-

tinua.
(9.B.6) Theorem. (Fraser [1972]). Suppose (X,d) is a compact metric
space. For each x € X and each ec > 0, let B.(x) = {ye X : d(z,y) < €}. If
for each x € X and each e > 0, S4(x) = B,(x), then X is a Peano continuum
(cf. problem B 13).
Proof. It must be shown that X is connected and locally connected;
this will be accomplished if we can establish that B,(x) is connected for all
x € X and all e > 0. Suppose B,(x) is not connected for some x € X and
some € > 0. Then there are disjoint nonempty closed sets A; and A» such
that B.(x) = A; U Az. Suppose z € A. By (3.4.14), there is ay € Ap
and an € < € such that d(z,y) = d(x, Az) = € > 0. Then y € B(x), but
y ¢ S:(x) which contradicts the hypothesis. Thus, B,(z) is connected. O
We mention that Fraser also established a partial converse: if X is a
Peano continuum, then X has a metric d under which B.(x) = S,(x) for all
zeExX ande>0O.
Before leaving the topic of Peano continua, we describe a special mem-
ber of this class: the Sterpinski universal plane curve, C. This continuum
is universal in the sense that if M is any 1-dimensional Peano continuum
embeddable in €*, then M can be embedded in C, even though C is 1-
dimensional (Sierpinski, 1916). The construction of the Sierpinski curve is
based on a Cantor like procedure, which we describe geometrically.
Start with the unit square J x J and from it extract the interior of the
rectangle R indicated in the following figure.
The second step consists of the removal of the interiors of eight more
squares, and the third step 64 such interiors are eliminated, etc. That which
Chainable Continua 231
remains after the removal of all these squares is called Sierpinski’s universal
plane curve.
If instead of removing squares from the unit square, one removes rect-
angular prisms (in a somewhat different fashion, Engelking [1978]) from the
unit cube, then the resulting space is called the universal curve. The uni-
versal curve is a l-dimensional Peano continuum with the property that
any 1-dimensional continuum can be embedded in it. Interestingly, it can
be shown that simple closed curves and the universal curve are the only
1-dimensional Peano continua that are homogeneous, Anderson [1958].
C. CHAINABLE CONTINUA
We have already seen the wide applicability of arguments using simple chains.
Our study now focuses on spaces that may be defined in terms of simple
chains.
(9.C.1) Definition. A simple chain C of open sets in a metric space is
called an e-chain if and only if each link of C has diameter less than e.
(9.C.2) Definition. A metric space X is chainable if and only if for each

€ > O there is an e-chain that covers X. The space X is chainable from a to
b if and only if for each € > 0, there is an e-chain {C),...,C,} covering X
such that a € C, and DE C,.
(9.C.3) Examples.
1. The unit interval I is chainable from 0 to 1;
2. {0,1} with the relative topology is not chainable;
3. (0,1) (with the usual metric) is chainable, but not from a to b for
any a,b € (0,1);
4. €' with the usual metric is not chainable, even though it is homeo-
morphic to (0, 1).
5. The topologist’s sine curve is chainable.
6. The rationals in [0, 1] are chainable from 0 to 1.
In spite of examples 2 and 3, chainability does have some hereditary
traits.
232 9. Continua
(9.C.4) Theorem. Suppose X is a chainable (hence, metric) continuum,

and K Cc X is asubcontinuum. Then K is chainable.
Proof. Let € > 0. Since X is chainable, there is an e-chain {C1, C2...,Cn}
that covers X. Let i be the first integer with C; K # 0 and let 7 be the
largest integer with C;AK #0. We show that {C;NK, CiziNK,...,CjNK}
is an e-chain in K that covers K. This is clearly the case unless there are
links Cy, Cp41 with i < p <j for which C,N K and Cp41 1 K fail to inter-
sect. However, if (Cp 1K) M (Cp41 1K) = O, then Ujemep(Cm K) and
Unti<me<j(Cm MK) form a separation of K, which is impossible.
Chainability is also surprisingly well behaved with respect to some types
of quotients.
(9.C.5) Theorem (Bing [1951]). If G is a monotone usc decomposition of

a chainable continuum X, then X/G is a chainable continuum.
Proof. Compactness and connectedness are trivial since the quotient is
continuous. Metrizability follows from (8.F.17). Let p be the metric for X,
d the metric for X/G and suppose e« > 0. Since the quotient map is uni-
formly continuous, there is a 6 > 0 such that d(P(z), P(y)) < €/5 whenever
p(z,y) < 6. Since X is chainable, there is a d-chain {Aj, Ag,...,An} that
covers X. It is easy to see that there is an increasing sequence of integers
1=1nj,N2,...,n; =n such that some element of G intersects An; and An,,,
but no member of G intersects Ap; and An,;,,)+1- For integers i and k, let
Uzp={P(9)vi go © Grand g°C A; C Ayer G- Gilg} Then x isgopen
and {Uj.5; Uninds Une nas es Une, | fong —0e< 3h 1 <9 Bias an
e-chain that covers X/G. O
Intuitively, it seems that chainable continua are not very thick. In fact,
it can be shown that any chainable continuum is homeomorphic to a plane
continuum, i.e., all chainable continua are embeddable in the plane (Bing
[1951a]). Furthermore, Bing showed (in the same paper) that if K is a chain-
able continuum, then there is an uncountable collection of disjoint continua
lying in the plane, each of which is homeomorphic to K.
A rather surprising result is that chainable continua have the fixed point
property.
(9.C.6) Definition. Suppose X is a chainable continuum. Then a sequence

of simple chains C,,C2,... each of which covers X, is call a defining sequence
of chains for X if and only if
(7) Cy is a (1/2”)-chain with the property that disjoint links C?, Ce
have disjoint closure, and
(it) Cn4i is a proper refinement of Cp, i.e., the closure of each link in
Cnr+1 is contained in some link of Cp.
(9.C.7) Exercise. Show that each chainable continuum has a defining

sequence of chains.
(9.C.8) Theorem. (Hamilton [1951]). If X is a chainable continuum, then

X has the fixed point property.
Proof. Suppose f is a continuous function from X to X, C),Co,... is a

defining sequence of chains for X, and for each k € Zt Cy = {Cf,...,C* }.
By (7.E.9) it suffices to find, for each « > 0, a point p € X such that
d(p, f(p)) < €. Let € > 0. Choose an integer k large enough so that 1/2" < «.
Now define:
Ae Xx, : if f(x) € CF and x € CF, then j <i}
B= {x €X : for some integer i, x € cr and f(z) € CF}
C = {x €X| if f(x) € CF and x € CF, then j > i}
Note that A and C are closed (their complements are open: why?) If B = 9,
then it is easy to see that A and C are disjoint closed sets whose union is
X, which contradicts the connectedness of X. Hence B # 9. Any point in
Bwillserveasp. O
We mention here that Dyer [1956] shows an arbitrary product of chain-

able continua has the fixed point property. An important consequence of
this result is a rather novel approach to proving the Brouwer fixed point
theorem. Most readers should have little difficulty with Dyer’s article, which
illustrates nicely how chainable continua may be manipulated.
Irreducible continua, which we introduce next, are somewhat distant
relatives of chainable continua.
(9.C.9) Definition. Suppose X is a topological space and H, and Ho are

closed subsets of X. Then a continuum K C X is irreducible from H, to H2
if and only if
(i) KNH, 490, KN HA, £9, and
(ii) for each proper subcontinuum L of K, we either have LN H; = 0
or LN Hz = 9.
Note that an arc with endpoints a and 6 is irreducible from a to b. More

generally, we have the following result.
(9.C.10) Exercise. Suppose X is a metric continuum and a,b € X. Show

that if X is chainable from a to b, then X is irreducible from a to b.
The converse of the preceding exercise is false, as may be seen from our
semiuniversal example, the topologist’s sine curve X.
234 9. Continua
The space X is irreducible from a to b but not chainable between these points
(although X is chainable between a and c).
The next theorem shows that irreducible continua are quite common.
(9.C.11) Theorem. Suppose X is a Tz space, H; and Hz are disjoint closed

subsets of X, and K is any continuum intersecting both H, and H2. Then
there is a subcontinuum L of K that is irreducible from H; to Ho.
Proof. Let M = ia : L is a continuum intersecting H, and Hz and
Denk }. Partially order M by inclusion (Ly < Ly if and only if ee Iz)
and let N be a maximal nest in M (0.D.4). Then L=(\{L: LEN} isa
continuum, and for each i € Z+, we have that LNH; =(\{LNH; : LEN}
is nonempty, (2.H.2). It follows that L is the desired irreducible continuum.
O
The concept of unicoherence is introduced next. This topic will reappear

in Chapter 15.
(9.C.12) Definition. A connected space X is unicoherent if and only if

whenever X = H, U Hy where H, and Hp are closed and connected, then
HH, Hg is connected.
It is easy to see that arcs are unicoherent and circles are not. It is not
so obvious what the status of €”, disks, spheres, etc., should be. The next
result throws little light on this problem, although it does yield a number of
examples of unicoherent spaces.
(9.C.13) Theorem. Every chainable continuum is unicoherent.

Proof. Suppose X is a chainable continuum that is not unicoherent.
Then X may be written as the union of closed connected sets C’ and D,
where CM D = AUB, and A and B are nonempty, disjoint, and closed.
Let r = d(A,B). By (9.C.7), there is a simple chain C; covering X of mesh
less than Rae with the property that if C; and C; are nonadjacent links in
C1, then C; NC; = 0. It follows from the proof of (9.C.5) that the links of
C, that intersect C form a subchain Ci that covers C’. Let Cj, be the first
link of C 1 that intersects either A of B. If A was struck first, let Gr be the

last link of C 1 that intersects B. It follows from the connectedness of C that
there is a link Ch, a an that misses AU B and 7; < ky < my.
Let r; be a Lebesgue number for C;. Then there is a simple chain
C2 of mesh less than min{r), 1/27} that refines C,, covers C, and has links
QR Opin Ginx (je the < m2) where
(i) C3, NA+ and C2, NB FO, and

(27) C3, ‘- Ca.
You should be able to convince yourself that C2 exists since C is connected,
and no harm is done if the numbering of the links in reversed.
This procedure is repeated inductively in order to obtain a decreasing
peed of links Ci, , CZ,,--. that do not intersect AUB. Let p = (\72, Ci, =
ge , and set a= d(p, D). If n is any positive integer such that 1/n < s,
then the link Cz, misses D; however, this is impossible, since D is connected
and there are links in C, preceding C7? that intersect AMD, and ones
following Cf, that meet BND. O
The next theorem gives an important property of locally connected uni-
coherent spaces.
(9.C.14) Theorem. Suppose X is a locally connected unicoherent space.

If a closed subset A separates two points p and q in X, then so must some
component of A.
Proof. Three lemmas are used. The proofs are due to Stone [1949].
(9.C.15) Lemma. Suppose D; and D» are disjoint connected subsets of a
locally connected unicoherent space X such that FrD; C Fr D2. Then Fr D,
is connected.
Proof. Let {Ca : a € A} be the set of components of X \ (Di U D2).
Then for each a € A, we have FrC, C Fr(D, UD2) C Fr D UF D2 = FYDy.
Hence CN Dy # @ for each a, and so by (2.A.11) A = DAO User Ca is
connected. Note that X \D, C A Cc X \ Dj; therefore, A=xX \ D, and
consequently, X \ D; is connected. Since X = X \ D, U D, is unicoherent,
we have X \ Di N D, =FrD, is connected. O
(9.C.16) Lemma. If D,; and D2 are connected open subsets of a locally

connected unicoherent space X such that FrD, 9 Fr D2 = 9, then D; N D2
is connected.
Proof. Suppose D; N Dz is not connected, and let C be a component of
D, ND». Let x be a point in (D;N D2) \C. Since C is closed in D; N Do, it
follows that x ¢ C. Let C* be the component of X \ C that contains x. Note
that C* is open (X is locally connected) and that CM C* = 9, and therefore
Cnc* =6.
236 9. Continua
Let C be the component of X \ C* containing C. Then FrC* = FrC

and hence, by (9.0.15), FrC* is connected. On the other hand, we have
rC.«c Free Fr (D, U D2) Cc WD, UFrD,. Since FrD; UFr Dz is the
union of two disjoint closed sets, either FrC* C FrD; or FrC* C Fr Da, say
FrC* c FrD,. However, z € Dy) NC* and C Cc Di N(X \ C*). Therefore,
D,NFrC* # O and it follows that D, N FrD; # 9, which contradicts the
openness of D;. O
(9.C.17) Lemma. If D; and D2 are disjoint closed subsets of a locally

connected unicoherent space X neither of which separates the points p and
q in X, then D, U D2 does not separate p and q.
Proof. The proof is left as an exercise for the reader. O
Proof of (9.C.14). Suppose A is a closed subset of X that separates the
points p and g. Let D be the component of X \ A containing p. Clearly,
q ¢ D, and furthermore, since Fr D C FrA C A, it follows that q ¢ D. Let
C be the component of X \ D that contains g. Then C and D are disjoint
open sets and FrC’ Cc Fr D.
We show that FrC is connected. Suppose FrC’ is not connected. Then
FrC = F, UF), where F, and F 2 are nonempty disjoint closed subsets of
X. Observe that C'U F) U D is connected (2.A.13), contains p and q, and
does not intersect F,. Similarly, CU F; UD is connected, contains p and q,
and does not intersect /2. Thus neither F; nor Fp separates p and q, and so
by (9.0.17), F; U Fh = FrC does not separate p and q, which is obviously
absurd. Hence, Fr C is connected.
To conclude the proof, let K be the component of A that contains Fr C.
Then K separates pandqginX. O
In the problem section, the reader is asked to prove the converse of this
theorem.
D. DECOMPOSABLE AND INDECOMPOSABLE

CONTINUA
This topic leads to some rather picturesque examples of pathological topolog-
ical spaces. A continuum X is decomposable if and only if X can be written
as the union of two proper subcontinua; X is indecomposable if and only if
X is not decomposable. The more familiar continua are certainly decompos-
able, and the reader would probably be hard pressed to find an example of
a continuum that is indecomposable (other than a point).
(9.D.1) Exercise. Suppose X is a T> continuum. Show: X is decomposable
if and only if X contains a proper subcontinuum with nonempty interior (in
X). Thus all nontrivial Peano continua are decomposable.
Decomposable and Indecomposable Continua 237
The first step in constructing an indecomposable continuum is to intro-

duce the idea of composant.
(9.D.2) Definition. Suppose X is a continuum and x € X. Then the
composant of x, Kz, is the union of all proper subcontinua of X that contain
Be
(9.D.3) Examples.
1. Suppose X is an arc with endpoints a and b. Then K, = X \ {b},
Ky, = X \ {a}, and if c is an interior point of X, K, = X.
2. If X = S! or S?, then K, = X for each x € X.
(9.D.4) Exercise. Find the composants of a triode and of the topologist’s
sine curve.
(9.D.5) Exercise. Show: if X is a decomposable T> continuum, then there

is an x € X such that K, = X.
Composants are characterized by the next theorem.

(9.D.6) Theorem. If X is a continuum and z € X, then
K, = X \ {y : X is irreducible from z to y}.
Proof. Suppose y € K,. Then there is a proper subcontinuum C’' con-

taining xz and y, and hence X is not irreducible from z to y.
The remainder of the proof is somewhat easier, and is omitted. O
(9.D.7) Theorem If X is a JT) continuum and z € X, then K, = X, ie.,

composants are dense subsets.
Proof. Suppose U is an arbitrary open set in X. We will show that
UnK, #9. Since X is regular, there is a nonempty open set V such that
V CU. IfzeV, then zc € UN K; and we are through. If x ¢ V, let C be
the component of X \V containing z. Then C is a proper subcontinuum of
X that contains z, and hence C C K,. However, by (4.4.12), we have that
CNV £0, and therefore K, intersects U. 0
In (9.D.3), we saw that an arc has only three distinct composants. This
represents a special case of the following theorem.
(9.D.8) Theorem. Suppose that X is a decomposable T; continuum and
a,b € X. If X is irreducible from a to b, then X has precisely three distinct
composants.
Proof. Suppose X = AUB, where A and B are proper subcontinua of
X with a € X \ B and be X \ A. It is trivial to verify that X has at least
three composants, namely K,, K, and K,, where z is any point in AN B.
(Note that K, = X since both AC K, and BC 1
238 Xo). Continua
Suppose now y € A. We show: either Ky = K, or Ky = Ka. We

first consider the case where b € Ky. If b € Ky then there is a proper
subcontinuum D of X such that both y and b lie in D. Hence AND # 9,
and consequently AU D is a continuum containing a and b. Since X is
irreducible from a to b, it follows that AUD = X. Furthermore, we have
Ac K, and BC Ky, and so Ky = X = K;.
For the other case, suppose b ¢ K,. To see that Ky = Ka, let r € Ka
and let E be a proper subcontinuum of X containing z and a. Note that
b¢ E. Since a € AN EL, we have AU E£ is a proper subcontinuum of X in
which are found z and y- Therefore, x € K, and hence Ky C Ky.
Finally, if w € Ky, there is a proper subcontinuum F' of X such that
w,y € F and b ¢ F. Since y € ANF, it follows that AU F is contained in
K,, and thusweéK,. O
As a corollary to the next theorem, we have the contrasting result

that indecomposable continua have an uncountable number of distinct com-
posants.
(9.D.9) Theorem. If X is a T) continuum with a countable basis, then each

composant K, of X is the union of a countable number of proper subcontinua
Of X..
Proof. Suppose z € X and U;,U2,... is a countable basis for X \ {x}.
For each positive integer 7, let C; be the component of X \Ui containing x (if
such exists), and observe that C; C Kz, since C; is a proper subcontinuum.
We claim that K, = U72, Ci. To see this, suppose y is an arbitrary point
in K, and D is a proper subcontinuum containing both z and y. Let U; be
a member of the basis with the property that U; C X \ D. Then since D is
connected, D must lie in C;, which completes the proof. O
(9.D.10) Corollary. If X is an indecomposable T) continuum with a count-

able basis, then X has uncountably many distinct composants.
Proof. It will be shown in Chapter 10 (10.C.1) that a T) continuum
with a countable basis is metrizable. Since a continuum is compact, we
may consider X to be a complete metric space. Hence the Baire category
theorem applies. Suppose X could be written as the union of a countable
number of composants K,,,K;z,,.... Since each composant is the union of a
countable number of proper subcontinua, so is X. By (3.B.16), at least one
of these proper subcontinua has nonempty interior, and hence, by (9.D.1),
X is decomposable. Thus X has uncountably many composants. O
The next exercise says that the situation is even worse.
(9.D.11) Exercise. Show: If X is an indecomposable T) continuum, then
distinct composants are disjoint.
A major clue to the construction of an indecomposable continuum is

revealed by the following theorem.
Decomposable and Indecomposable Continua 239
(9.D.12) Theorem. A T> continuum X with a countable basis is inde-

composable if and only if there are points a, b, and c in X such that X is
irreducible between each pair of these points.
Proof. Suppose X is indecomposable and let K,, Ks, and K, be distinct
(and therefore disjoint) composants of X. If D is a proper subcontinuum of
X containing a and b, then D C KM Ko, which contradicts the disjointness
of K, and Ky. Thus, X is irreducible from a to b. The same argument
applies to b and c, and a and c.
Now suppose points a, b, and c exist such that X is irreducible between
each pair of them. Consider the corresponding composants K,, K, and K,. If
X is decomposable, then it follows from the proof of (9.D.8) that K,, Ky, and
kK, are distinct. It also follows from this proof that each of these composants
must in fact be X. Thus X is indecomposable. O
The problem of finding an indecomposable continuum has now been

reduced to starting with three points and constructing a continuum that is
irreducible between each pair of them. How might one create an irreducible
continuum of this nature? The answer is easy to obtain: it follows from
(9.C.9) that if a continuum is chainable between points a and 6, then it is
also irreducible from a to b. The obvious way to build a chainable continuum
is to employ simple chains, and this is what we do next.
The entire construction is carried out in the plane. Select any three
noncollinear points a, b and c in €”, and construct a simple chain C; con-
sisting of open disks that starts at a, passes by b and ends at c. Inside C;
construct a simple chain C2 of open disks that starts at b, passes through
c, and ends at a. Then inside C2 construct a third chain C3 that begins at
c, runs through a, and terminates at b (see the figure that follows). The
whole process starts over again with a simple chain C4 that lies inside C3 and
follows the pattern a-b-c. In general, for each positive integer n, construct a
simple chain C3,4, that runs the a-b-c route, C3n42 that passes from 6 to c
to a, and C3, that goes from c to a to b. Clearly, for each n € Zt, we may
assume that meshC,, < 1/n.
For each n € Zt, let Dn = U{C : C € Cy}. Then X =(\?_, Dn is an
indecomposable continuum.
a
eeOO
e Oxc es
WAT
t
AY CBO) rier
ee
Bs,
a te esow
unt
Pt
oe Ast
Loman
IAI
CRTC ATU Bee WY Va.NUR DANGER DXi>)
a nn as *
240 9. Continua
(9.D.13) Exercise. Show that the space just constructed is an indecom-

posable continuum.
Unbeknownst to the reader, we have previously exhibited another in-

decomposable continuum, the dyadic solenoid. We now prove, using inverse
systems, that the solenoid is indecomposable.
Consider the dyadic solenoid, S, as an inverse limit of the inverse system
(X;, fi, N) where C; is the unit circle in the complex plane (£7) and a typical
bonding map fj; : C; > C;-1 is defined by f;(z) = z?. Let p; : S 4 C; be
the customary i-th projection map restricted to the inverse limit S. The
following ingenious proof of the indecomposability of S is due to Nadler
[1973].
(9.D.14) Theorem. The dyadic solenoid S is an indecomposable contin-
uum.
Proof. Let A and B be subcontinua of S such that AU B= S. We
show that either A or B fails to be a proper subset of S, or equivalently,
either A C Bor B C A. The heart of the matter lies in the following
assertion: For each positive integer i, either p;(A) C p;(B) or p;(B) C p;(A).
Suppose the assertion is false for some integer k, and let a € p,(A) \ pe(B)
and b € px(B) \ px(A). Since each bonding map f; is onto, it follows from
(6.B.13) that p; is also onto for all i. Thus, we have a € px41(A) Upp 41 (B).
If Ja Se Pre+i(B), then a = fr4i (Ja) (E fe+1Pr+1(B) — Dr (B), which is
impossible. Therefore, \/a € px41(A)\fe41(B). Using similar arguments, we
may conclude that —/a € px(A)\pe4i(B) and Vb, —Vb € fri1(B)\pe41(A).
It is immediate from the connectedness of A and the fact that /a and
—/a are antipodal, that p,41(A) includes at least one of the semicircles of
Cr41 with endpoints fa, and —/a. But either Vb or —Vb must also fall
in this semicircle, contradicting the fact that both Vb and —¥V4 lie in the
complement of p,+1(A). Consequently, the assertion is established.
There are now two possibilities: p;(A) C p;(B) for infinitely many i, or
pi(B) C p;(A) for infinitely many 7. It follows from the first exercise below
that in the former case, p;(A) C p;(B) for all positive integers i, and, of
course, in the latter case p;(B) C p;(A) for each i. The proof is completed
by observing that A is the inverse limit of (p;(A), fi+1\9,,,(4),N) and B is
the inverse limit of (p;(B), fi+11p,,,(4), N), and therefore, by (9.D.16) below,
AC Bor B CA> sb
(9.D.15) Exercise. Show that if p;(A) C #;(B) for infinitely many i, then
pi(A) C p;(B) for all i.
(9.D.16) Exercise. Finish the proof of the previous theorem by showing
that AC Bor BCA.
The Pseudo-Arc 241
E. THE PSEUDO-ARC
In this section, an important addition is made to our small band of indecom-

posable continua. The pseudo-arc (actually there are many, but they are all
homeomorphic) has probably been the most scrutinized of all the indecom-
posable continua, and its properties are for the most part as spectacular as
they are difficult to prove. The idea behind the construction of a pseudo-arc
is somewhat similar to the trick employed in creating our first example of
an indecomposable continuum: simple chains are used to define the space.
In the previous example, simple chains were carefully woven between three
points in the plane; in the case of the pseudo-arc, chains are stretched out
between two points. The stretching, however, is of a very contorted nature
(similar to this explanation). It will soon become clear that “pseudo-arc” is
an appropriate label for this space.
Any space that results from the following construction is called a pseudo-
arc.
Let a and b be any two points in the plane, and let C,,Co,... be a
sequence of simple chains of connected open sets satisfying the following
conditions.
(i) the point a belongs to the first link of each chain, and 6 to the last;
(ii) the closure of each link of C,41 is contained in some link of C,,;
(iit) meshC,, < 1/n for each positive integer n;
(@v) if C2** and C®t are links of Cyi1 (& < m), and Cpt" cc?
and C2+! c C™, where |p — q| > 2, then there are links C?+* and
Crt! where k < s < t < m, such that CPt’ is contained in a
link adjacent to Cf and Cf't? is contained in a link adjacent to C?
(note carefully that s < t).
aS
CORAL AORN
= KOO BOOOOOOR OOO
(2COOP MOS OOM OOIOOP SOO).
=) b
CBSE
RRRRRIED (a)
HOTTIN
& (=) 5
LPN OTTO TN
COODIOROOOOOOMOOOIOOOR
OOS
& =SS
iS)
—
COO SOOTHING)
Cr C3 C3 C4 C3
If An = U{C : C is a link of Cy}, then A = ()?~, An is the pseudo-arc

associated with the chains C,,C2,....
We first show that A is indecomposable. In fact, we prove something
even more astonishing: every subcontinuum of A is indecomposable.
242 . 9. Continua
(9.E.1) Theorem. The pseudo-arc A is hereditarily indecomposable.

Proof. Suppose there is a subcontinuum B of A that is decomposable.
Then B may be written as the union of two proper subcontinua H; and Hp.
Use is made of the zigzagging property of the chains (iv) to show that H; is
not connected.
It is obvious that there are points x and y in B and an integer n such
that d(z,H,) > 2/n and d(y, Hz) > 2/n. This of course implies that x € H
and y € H,. Consider the parts of the chains C, and C,41 that run between
z and y. Denote these by {Cf,CZ,1,...,C"} and {Cp*?, CR mo Suara
Assume z € C7. Since the diameter of each link of C, is less than 1/n, it
follows that CR, ,Hi = 0, and hence CZ, ,NH2 # 9. Similarly C?_,NH2 = 9
and C?_, NH; #0. The twisting of the chains now leads to a contradiction.
As a result of C,41’s kinky relationship with C,, there are links Cat.
Cr andiCl:” (crcid< e)rwitnee mame OC. and Ct ac Ga:
Thus, Ca misses H,, but the other two links do not, which implies that
Hy, is not connected and establishes our contradiction. O
Moise [1948] proved an even stronger result: Every subcontinuum of A

is homeomorphic to A. Naturally, arcs also possess this property. However,
a very striking and dramatic difference between arcs and pseudo-arcs arises
from a result of Bing [1948]: The pseudo-arc is homogeneous. Moreover, Bing
[1951b] showed that any homogeneous hereditarily indecomposable contin-
uum is a pseudo-arc.
This latter result is used to show that there are many pseudo-arcs. In
fact, if X is either E” (n > 1) or Hilbert space, then most of the continua
lying in X are pseudo-arcs.
(9.E.2) Theorem. Suppose X = €" (n > 1) or Hilbert space. (Assume

X has a bounded metric.) Let C(X) be the family of all continua in X and
A(X) be the subfamily of pseudo-arcs. If C(X) is given the Hausdorff metric,
then A(X) is a dense Gs subset of C(X).
Proof. We show first that A(X) is dense in C(X). Suppose C € C(X)
and €« > 0. It suffices to find a pseudo-arc A such that p(A,C) < «. There
is a broken line ab in X such that p(ab,C) < €/2 (why?). Let D be an
(€/2)-chain from a to b that covers ab and whose links are open balls in X.
Inside U){D : D € D}, construct a pseudo-arc A which contains the points
a and b. Then p(A,C) < e.
A broken line
Next we show that A(X) is a Gs subset of C(X). For each i € Zt,

let G; be the collection of all members of C(X) that cannot be covered by
The Pseudo-Arc 243
a (1/i)-chain. Then G; is a closed subset of C(X) since if any sequence

G1,G2,... in G; converges to a continuum G, G must lie in G;: otherwise
a (1/1)-chain covering G would also contain some member of the sequence
Gi,Go2,.... It is clear that C(X) \ (U32, Gi) consists of all the chainable
continua in C(X).
Let K, be the subfamily of C(X) consisting of sets K with the property
that K contains a subcontinuum K which is the union of two continua K,
and K», where
(1) there is an x € Ky such that d(x, K2) > 1/i, and
(ii) there is a y € K2 such that d(y, K) > 1/i.
Note that K; is closed in C(X), and if H € C(X) \ (U32, Ki), then H is
hereditarily indecomposable.
Finally, observe that if P represents the collection of points in X, then
P is a closed subset of C(X).
Let W =C(X)\(PUUZ, G UU, Ki). Then W is a dense Gs subset
o C(x): o
We conclude this chapter with a few additional curiosities involving con-
tinua. Although it was a nontrivial matter to construct just one hereditarily
indecomposable continuum, the one produced is hardly unique. Bing [1951b]
has shown that there are as many topologically different hereditarily inde-
composable continua as there are real numbers.
Two results, reminiscent of the fact that compact metric spaces are
continuous images of the Cantor set, have been established by Mazurkiewicz
[1920] and Bellamy [1971]. Mazurkiewicz proved that every chainable con-
tinuum is a continuous image of the pseudo-arc, and Bellamy demonstrated
that every metric continuum is a continuous image of some indecomposable
metric continuum, but that there is no continuum of which every indecom-
posable continuum is a continuous image.
Schori [1965] used inverse systems to construct a universal chainable
continuum, universal in the sense that if X is any chainable continuum,
there is a subcontinuum of U homeomorphic to X. Anderson and Choquet
[1959] have exploited inverse systems to concoct a plane continuum with the
property that no two of its subcontinua are homeomorphic. Needless to say,
the literature is replete with such aberrant examples and peculiar theorems.
The curious reader might wish to consult the following articles for further
enlightenment: Bing [1948, 1951a,b], Fugate [1965], Burgess [1959, 1961],
and Jones [1951], as well as the books by Nadler [1978] and [1992].
244 9. Continua
PROBLEMS
‘
Section A
1. Describe the possibilities for S(a,b) when X is a circle, a tree, or a circle

with a tail.
2. Find an example of a space X containing points a and b for which the
relative topology on S(a,b) does not coincide with the order topology.
3. Suppose X is a metric continuum such that for each two distinct points
z,y € X, X \ {x,y} is not connected.
(a) Show: no point separates X.
(b) Show: if (U,V) is a separation of X \ {x,y}, then U U {x,y} and
V U {x,y} are connected and at least one of these sets is an arc.
(c) Show: both U U {x,y} and V U {z,y} are arcs.
(d) Show: X is homeomorphic to S?.
4. Prove or disprove the following assertion. If X is a Tz continuum con-
taining a point z such that X \ {x} contains two components but there
is no point y such that X \ {y} has more than two components, then X
is an arc.
5. Consider X = [1,3)U {5} c €1. Let <x be the restriction of the usual
order to X. Is X with the order topology induced by <x, the same as
X considered as a subspace of €1?
6. The long interval T is defined to be the long line (see problem A-3
of Chapter 5) together with {2,0}. Extend the order in the obvious
manner. Show that Z is T>, T3, connected, compact, has exactly 2
non-cut points, and is not an arc.
Section B
1. A metric space X is uniformly locally connected if and only if for each

€ > 0 there is a 6 > O such that if d(z,y) < 6, then z and y lie in
a connected set of diameter less than «. Show that a compact locally
connected metric space is uniformly locally connected.
2. Show: a metric space is locally connected if it is uniformly locally con-
nected.
jw We say a metric continuum X has property S if and only if for each e > 0,
X is the union of a finite collection of connected sets, each of diameter
less then €. Show that a metric continuum X is a Peano continuum if
and only if X has property S.
Problems 245
4. A space X is connected im Kleinen at a point p € X if and only if for

each open set U containing p, there is an open set V such that p € V C U
and any pair of points in V lies in a connected subset of U, Show: if X
is connected im Kleinen at each point p, then X is locally connected.
Show that the following space is connected im Kleinen at z, but there
is no basis of connected open sets at 0.
0 75 1/4 T/3 1/2 1
Suppose X is a nondegenerate Peano continuum and Y is an arbitrary

Peano continuum. Establish the existence of a continuous onto map
Pk eX.
Show: connected open subsets of Peano continua are arc connected.
Let S be the unit square. Show that if f : S > J, then there are disjoint
arcs A and B in S each with at most one boundary point of S such that
JOA B) =F.
Let A be the set of all vertical line segments lying between two disks.
Show there is an arc in €? whose image intersects each line segment,
but that there is no arc in €? whose image intersects each line segment
exactly once.
10. Let A;, Ag,... be a sequence of subsets of a space X. Recall:

(i) limsup A, = {x € X : each open set containing x intersects in-
finitely many A,}, and
(ii) liminf A, = {x € X | each open set containing z intersects all but
finitely many A, }.
246 9. Continua
Find lim sup and lim inf of the following sequence of subsets of the plane
AioAs Ag
tt. Suppose A;, A2,... is a sequence of subsets of aspace X. Show liminf Ay

and lim sup A, are closed subsets of X.
12. Suppose A, A2,... is a sequence of subsets of a space X. In the case
where liminf A, = limsup A, we write lim A, for their common value.
Show: if X is compact, and T2, and each A, is connected, then lim A,
exists and is connected.
13. Why is the topologist’s sine curve not a counterexample for Theorem
(9.B.6).
Section C
Show: a chainable continuum cannot contain a 1-sphere.

Show the only nondegenerate metric locally connected chainable contin-
uum is the arc. (Hint: Use problem B-7 and the characterization of the
arc given in Section A.]
Suppose X is a chainable continuum and C is a simple chain of open
sets covering X. Show there is a 6 > 0 such that any 6-chain D properly
refines C (i.e., if D € D, there is aC’ € C such that DC C).
Show that the topologist’s sine curve has the fixed point property.
A continuum X is a triode if and only if there is a proper subcontin-
uum H Cc X such that X \ H is the union of three nonempty pairwise
separated sets. Show that a chainable continuum contains no triodes.
Define a space to be open-unicoherent if the word “closed” in the def-
inition of unicoherent is replaced by “open.” Show that if a locally
connected, connected space is unicoherent, then it is open-unicoherent.
Suppose X is locally connected and connected with the property that
if a closed subset A separates points p and'q in S, then so does some
component of A. Use the following steps to show that X is unicoherent.
(a) Suppose that A and B are closed and connected, AU B = X and
ANB = HUK, where H and K are disjoint, closed, and nonempty.
Problems 247
Show there is a component D of X \ A such that FrD is not con-

nected.
(b) Show that X \ D is connected.
(c) Show that there is a component C of X \ D such that FrC is not
connected.
(d) Show that FrC C FD.
(e) Let p € C, q € D, and let F be a component of Fr C that separates
pand q. Let z € (FrC)\F and show that CU{z}UD is connected,
contains p and q, and misses F to conclude the proof.
8. Suppose X is a unicoherent T continuum and Y is a T> space. Show:

if f : X — Y is a continuous onto map such that f~!(C) is connected
for each connected subspace C' of Y, then Y is unicoherent.
Sections D and E
1. Suppose X is a metric continuum. Show that there is an uncountable

set A C X such that X is irreducible between any two points in A.
2. Suppose X is a T> continuum and Y is a subcontinuum. Show: if

L CY is a subcontinuum of Y such that Y \ L is not connected, then
Y is indecomposable.
3. Suppose X is a J> continuum such that every subcontinuum Y of X has

the property that no subcontinuum of Y separates Y. Show that X is
hereditarily indecomposable.
4. Suppose X is a hereditarily decomposable, hereditarily unicoherent con-

tinuum that contains no triodes. Let F be any collection of subcontinua
of X. Show that (\{F : F € F} is a subcontinuum.
5. Suppose X is an indecomposable continuum and let p € X. Show that
{x € X : X is irreducible between p and 2} is dense in X.
6. Show that the union of a countable number of proper subcontinua of an
indecomposable continuum X can not separate X.
7. Show that each composant of a metric continuum is an F; set.
8. Let K be the Cantor set. Let So be the union of all semicircles in the
upper half-plane with both endpoints in K and such that the endpoints
are symmetric with respect to the line = 1/2. For i = 1,2,..., let
S; be the union of all semicircles in the lower half-plane whose ends are
on K and are symmetric with respect to the line z = 5/(3'- 2). Let
248 9. Continua
ee OBlaar S;. Show that X is chainable and indecomposable.
So
S3
S2
Sy
Show that the plane contains uncountably may pairwise disjoint non-
degenerate continua, none of which contains an arc. [Hint: Use the
pseudo-arc.]
10. Construct the first two stages of a pseudo-arc starting with six links in
the initial chain.
11. Show that every topological space X that is a nested intersection, i.e.,
co
Pm in X;, where X; D X;4; can be viewed as an inverse limit of the

as |
spaces X;. Show that the inverse limit of nonempty spaces may be
empty if the bonding maps are not onto.
Chapter 10
PARACOMPACTNESS AND
AND METRIZABILITY
Most results involving paracompactness and metrizability are derived from

judicious refinements of open covers. Paracompactness is defined in terms
of such coverings, and metrization theory is almost wholly dependent on the
successful exploitation of appropriate covers.
The notion of paracompactness is a relatively recent one as topological
ideas go, and was first introduced by Dieudonné [1944]. In the hierarchy of
topological spaces, paracompactness lies between normality and metrizabil-
ity. It retains many of the virtues of compactness, but is somewhat more
general. In addition to its usefulness in the theory of metrizability, paracom-
pactness served as a catalyst in the liberation of dimension theory from the
confines of separable metric spaces.
We give an admittedly somewhat cavalier treatment of metrization: only
a few of the more celebrated results are stated, and even fewer proofs are un-
dertaken. Presumably, however, enough material is presented to either whet
or extinguish the reader’s appetite. The reader is invited to consult Nagami
[1970] for a terse but nevertheless more complete and elegant presentation.
In sections D and E, Moore spaces and the completion of metric spaces
are dealt with briefly.
A. PARACOMPACTNESS
(10.A.1) Definition. A family C of subsets of a space X is locally finite if
and only if each xz € X lies in a neighborhood that intersects at most a finite
number of members of C.
As an easy but important application of this definition, we claim that
the union of a locally finite family of closed sets is closed.
249
250 10. Paracompactness and Metrizability
(10.A.2) Exercise.
(a) Suppose {Aq : a € A} is a locally finite collection of closed subsets
of a space X. Show that U,¢, Aa is closed.
(b) Suppose {Aq : a € A} is a locally finite collection of subsets of a
space X. Show that {Aq : a € A} is locally finite.
The simplest type of cover refinement is defined next.
(10.A.3) Definition. If (/ is an open cover of a space X, then a cover V of

X refines U if and only if for each V € Y there is aU € U such that V CU.
The cover V is an open (closed) refinement of U if and only if V is an open
(closed) cover.
(10.A.4) Example. Suppose U/ is an open cover of a compact metric space
X and 6 is a Lebesgue number for U/. Then any cover Y of X with mesh less
than 6 refines U/.
(10.A.5) Definition. A T2 space is paracompact if and only if each open

cover of X has a locally finite open refinement.
If an open locally finite refinement of an open cover exists, then it can

be chosen in a particularly elegant fashion.
(10.A.6) Theorem. Suppose U/ = {U, : a € A} is an open cover of a

space X and W = {Wg : B €T} is an open locally finite refinement of UW.
Then there is a locally finite open refinement V = {V, : a € A} of U such
that Va, C U, for each a. (Such a refinement is called a precise locally finite
refinement of U.)
Proof. For each 8 € IT, let f(@) be a member of A with the property
We C Usig). Then for each a € A, let Vy = U{We : f(8) = a}. There is
of course no guarantee that V, is nonempty, but this is immaterial, since in
any case V = {Vy : a € A} forms an open cover of X that refines UW. In fact,
we have Vy C U, for each a € A. Since W is locally finite, soisV. O
Certainly, compact T2 spaces are paracompact. The compactness con-
dition may be relaxed somewhat when the space is T3.
(10.A.7) Theorem. If X is a Lindel6f, T; space, then X is paracompact.

Proof. Suppose U = {U,g : a € A} is an open cover of X. For each
xz € X, select an a; € A such that z € Ug,. Since X is regular there are
open sets V, and W, such that zs € Vz; C Vz C WG W G&U,../ Then
Vv ={V, : 2 € X} is an open cover of X, and since X is Lindeléf, we may
extract a countable subcover {Vz,,Vzr.,..-} from V. Let T; = W;,, and for
each integer n > 1, set Th = Wz, N(X \V2,)N---N(X\Vz,-1). We show
T ={T, : n € Z*} is a locally finite open refinement of U/ that covers X.
Paracompactness 251
That 7 is a cover of X may be seen as follows. Suppose « € X. If

« ¢ T; = W,z,, let n € Z* be the first integer such that « € W,,. Then
since r ¢ W,, U--- UW,,_, and since Vie U- mWIg 2, CW We sues,
we have x € T,,. Hence, 7 is an open cover of X.
Clearly 7 refines U/; thus it remains to show T is locally finite. Suppose
zéX. Then z € V,,, for some n, and since T,4; Vz, = @ for each j € Zt,
it follows that 7 is locally finite. O
A far more significant result is that metric spaces are paracompact.

This result was established by Stone [1948]. Stone’s demonstration of this
theorem is exceptionally intricate and has since been greatly simplified by
Rudin [1969], whose proof we give here.
(10.A.8) Theorem. If (X,d) is a metric space, then X is paracompact.

Proof. Suppose U = {U, : a € A} is an open cover of (X,d) and A is
well-ordered (0.D.6). For each n € Z*, define Van (by induction on n) to be
the union of all sets of the form Sian (x) such that
(i) a is the first member of A with x € Ug,
(ii) c ¢ Vg; if j<n, and
(iii) S$)on (2) C Ua.
We show that the family V = {Van : n € Zt} is a locally finite open
refinement of U that covers X.
Clearly, V refines U. To see that V covers X, observe that if z € X,
there is a first member a of A such that 2 € U, and an n so large that (iii)
holds. Then by (ii), either € Vg; for some j < n or z € Van.
We complete the proof by showing Y is locally finite. Suppose x € X
and let a be the first element of A for which z € Von for some n and choose
j large enough so that S}/2; (x) C Van. Clearly, V will be locally finite if we
can establish:
(a) ifi >n+j, then S¥/.,.4;(z) intersects no Vg; and
(b) ifi <<n+ J, then Styoni (x) intersects Vg; for at most one /.
We establish (a): Since i > n, by (ii), each ball of radius 1/2* used in the
definition of Vg; has center y outside of Van. Since So 95 (x) C Van, we have
d(x,y) > 1/2’. However, since i > 7 +1 andn+j > j +1, it follows that
Stands (z) a)SH: (Y) 2
We establish (b): Let p € Vg; and q € Vj; where 6 < y. We show
d(p,q) > 1/2"*9-1!. There are points y and z such that p € S¥)9:(y) C Vei
and q € So oi(2)°G. Via Byatiil) S393 (y) C Up, and by (ii) z ¢ Ug. Hence
d(y,z) > 3/2* and consequently d(p,q) > 1/2'>1/2"t7-". Oo
Although metrizability ensures paracompactness, the next lower species

in the pecking order of topological spaces, Ts, is not quite strong enough (see
Problem A-10). The converse of Stone’s theorem is false (any nonmetriz-
able compact T> space will do—for instance, an uncountable product of unit
intervals). However, paracompactness is powerful enough to guarantee nor-
mality.
(10.A.9) Theorem. If X is paracompact, then X is normal.
Proof. We show X is regular, and the reader may extend the techniques
slightly to obtain the normality of X. Suppose z € X, and F is a closed
subset of X not containing z. For each p € F, let U, be an open set
containing p whose closure misses x (X is T2). Then {U, : p € F} together
with X \ F forms an open cover U of X. Since X is paracompact, there is
a locally finite open refinement V = {Vy : a € A} of U. Note that ¢ ¢ Vg
for each a € A for which NF #0. LetW=U{VEeV: VearFk £9}.
Then F Cc W, W is open, and x ¢ W. Furthermore, since VY is locally finite,
a ZW by (10.A.2). O
(10.A.10) Definition. If U/ is a cover of X and A C X, then the star of A
with respect to U denoted by St(A,U), is defined to be the union of all sets
in U that intersect A.
(10.A.11) Definition. Suppose U is a cover of a space X. A cover V of X

is called a star refinement of U if and only if for each V € Y there isaU €U
such that St(V,V) C U,ie., {St(V,V) : V € V} refines U.
A slightly different but equally useful concept is that of barycentric
refinement.
(10.A.12) Definition. Suppose U/ is a cover of a space X. A cover V of X

is a barycentric refinement of U if and only if {St(z,V) : x € X} refines U.
(10.A.13) Exercise. Show: a barycentric refinement of a barycentric re-
finement is a star refinement.
Our principal result is the following.
(10.A.14) Theorem. If X is paracompact, then each open cover U of X

has an open star refinement.
Proof. By the previous exercise, it suffices to show that U possesses a

barycentric refinement. Since X is paracompact, there is an open locally
finite refinement W = {W, : a € A} of U. By (10.4.9), X is normal, so
(4.C.2) there is an open cover V = {V, : a € A} such that V, C Wy
for each a € A. Clearly, V is also locally finite. For each z € X, define
W, = (\{Wa : z € Va}. It is not difficult to see that this is a “finite”
intersection, and hence W, is open.
Paracompactness 253
For each & € X, let F, = U{Va : x ¢ Va}, and note once again
that by (10.A.2) F, is closed. Finally, set Z, equal to W, \ F,. Then
Z={Z, : c € X} isa barycentric refinement of YU. O
We note in passing that the converse of this theorem also holds, but do
not belabor (or prove) the point.
The following inequality might well be more than a frivolous conjecture:
card(known mathematicians) < card(known varieties of compactness).
(10.A.15) Definition. A locally compact T> space is o-compact if and only

if X can be expressed as a countable union of compact subsets.
Certainly, €” is o-compact for all n, and equally obvious is the fact that
an uncountable discrete metric space is not o-compact. Two basic properties
of o-compact spaces are given next.
(10.A.16) Theorem. Suppose the space X is o-compact. Then there is

a sequence U,,U2,... of open subsets of X each with compact closure such
that X = U7, U, and, for each n, ere
Proof. Since X is o-compact, we have X = hee Cy, where each C,

is compact. For each z € C;, let V, be an open neighborhood of x with
compact closure (X is locally compact). Then V = {V, : x € Ci} covers
C, and a finite subcover may be extracted from VY. Let U, be the union of
the members of this finite subcover. Note that Ua is compact. The same
procedure is repeated to obtain an open set U2 with compact closure that
contains the compact set U, UC. The proof may now be concluded with
an easy inductive argument. O
(10.A.17) Theorem If X is a o-compact space, then X is Lindelof.
Proof. Suppose X is o-compact and U is an open cover of X. Since X is

a-compact, we have X = U7, Cn, where each C;,, is compact. For each n, U
is a cover of C,, and hence there is a finite number of sets Un, ,Un.,--- Unica)
in U that cover C,. Then {Un, : n € Zt, 1 < 1% < k(n)} is a countable
subcover of U. O
The notions of paracompactness and o-compactness are closely related.

Since T3, Lindeléf spaces are paracompact, it follows from (10.A.17) and
the fact that locally compact T2 spaces are T3, that o-compact spaces are
paracompact. There is even a partial converse that will prove useful in
Chapter 17.
(10.A.18) Theorem. If X is a locally compact paracompact space, then

X may be expressed as a free union of o-compact spaces.
Proof. For each x € X, let U,; be an open neighborhood of xz with
compact closure. Let U = {U, : « € X}. By (10.A.6), there is a precise
open locally finite refinement V = {V, : z € X} of U. Partition X into
disjoint subsets by means of the following equivalence relation, ~. Two
points a,b € X are equivalent if and only if a and b can be connected by a
simple chain with links in V.
Let C = {Cy : a € A} denote the set of equivalence classes derived
from ~. Each C, is open (it is the union of V’s from V) and hence is locally
compact. To see that Ca is o-compact, let W; be any member of VY lying in
Cy and set W2 = {U{V €V: VOW, #90}. Then W2 C Cy; furthermore,
W, is a finite union (W ; is compact and V is locally finite), and consequently,
W. is compact. Inductively define W, = U{V € V : VON W,y-i F Of.
Clearly, W , is compact and Cg =7_,W,-
(oil “OU
B. PARTITIONS OF UNITY
As a diversion before taking up the subject of metrizability, we consider

briefly the idea of a partition of unity.
(10.B.1) Definition. A partition of unity of a space X is a family of

continuous functions, {fag : a € A}, each of which maps X into [0,1] such
that for each x € X:
(i) {a : fa(x) # 0} is finite, and
(ii) \oaea fa(x) = 1 (whence the terminology).
A partition of unity {fa : a € A} is subordinate to an open cover
U = {Ug : 8B ET} of X if and only if for each a € A, there is a 8 € T such
that fa\xX\Ug =):
(10.B.2) Theorem. If X is paracompact, then each open cover of X has a

partition of unity subordinate to it.
Proof. Let U = {Ug : B € T} be an open cover of X. By (10.A.9),
X is normal. Therefore, we may apply (4.C.2) twice to obtain open covers
V = {Vg : B € T} and W = {Wg : 6B € T} such that for each @ € T,
Ws CVe C Vg C Ug. Tietze’s extension theorem can be used to generate
functions gg : X — [0,1], where 98lW = 1 and gg|x\v, = 0. Since X is
paracompact, we may assume (with a little care) that V and W are locally
finite. The gg’s are normalized by defining a function fg : X — (0, 1] for each
B, where fg(x) = ga(z)/ (ees 9a(2)). (Why does }' 4c gg make sense?)
Then the family {fg : 6 € T} is the desired partition of unity. O
A Sampling of Metrizability Theory 255
(10.B.3) Definition. A function g : X — €! is lower semicontinuous if

and only if for each a € €', {x : g(x) > a} is open.
The following result is a particularly useful consequence of (10.B.2).
(10.B.4) Theorem. Suppose X is paracompact. If f : X — €! and

g : X — €! are upper and lower semicontinuous real valued functions, re-
spectively, such that f(x) < g(x) for each x € X, then there is a continuous
function h : X - €! such that f(x) < h(x) < g(x) for each x € X.
Proof. We first construct an open cover of X and then find a partition
of unity subordinate to it. For each rational r we define an element U, of
the cover by U, =4ge. fig < riya 3, 9(2) > r}i. Since f 16 upper
semicontinuous and g is lower semicontinuous, it follows that U, is open,
and U = {U, : r is rational} covers X. By the previous theorem, there is a
partition of unity {f, : r is rational} subordinate to U. Then the function
h defined by h(x) = >°_ rf,(z) is the desired mapping. O
C. A SAMPLING OF METRIZABILITY THEORY
Recall that a topological space (X,U/) is metrizable if there is a metric for

X such that the induced metric topology coincides with U. Although metric
spaces lead naturally to topological spaces, not all topological spaces are
metrizable:, for instance, nonnormal spaces are not metrizable (metric spaces
are normal, and normality is a topological invariant). This leads one to the
problem of finding conditions sufficient to ensure the metrizability of a space.
The purpose of this section is to provide some fairly representative solutions
to these problems.
Perhaps the major classical metrization result and still one of the most
useful is the following theorem due to Uryson.
(10.C.1) Theorem (Uryson’s Metrization Theorem). If a space X is

second countable and 73, then X is separable and metrizable.
Proof. Let B be a countable basis for X and define A to be the countable

set {(U,V) : U,V € BandU C V}. Since X is normal ((2.1.11) and (4.D.6)),
Tietze’s extension theorem can be exploited to obtain for each (U,V) € A
a continuous function fyy : X — [0.1] that maps U to 0 and (S\ V) to 1.
The family F = {fuv : (U,V) € A} separates points from closed sets. By
(7.B.17), X is homeomorphic to a subset of [](y,yye4Z. But by (6.4.15)
this latter space is metrizable. Hence X is homeomorphic to a subspace of
a metrizable space, and so is metrizable. O
(10.C.2) Corollary. Suppose X is a locally compact separable metric

space. Then the one point compactification X* of X is metrizable.
Proof. Since X is a separable metric space, it is Lindel6f (2.1.13), and
therefore, since X is also locally compact, X may be written as a countable
union of open subsets with compact closure; thus, X is o-compact. By
(10.A:16), X = U7, Unwhere Un C Un41 and each U, is open and has
compact closure. Observe that {X \U, : n € Z*+} forms a countable basis
at oo (X* = X U{oo}), since any compact subset of X is contained in a finite
number of the U,. Consequently, X* is second countable and 73; therefore
X* is metrizable. O
(10.C.3) Exercise. Suppose X is a compact metric space such that when-
ever h : X — Y is an embedding of X into a metric space Y, there is a
neighborhood U of h(X) in Y and a retraction f : U + h(X). Show X is
an AN Rm.
Uryson’s metrization theorem is adequate for working with separable

metric spaces, but for many purposes more general metrization results are
needed.
(10.C.4) Definition. Suppose X is a topological space. A sequence of open
covers U,,U2,... of X is called a development for X if and only if for each
xe X, {St(z,Un) : n =1,2,...} is a neighborhood basis at z. A T; space
X is developable if and only if X has a development.
Historically, the first documented metrization theorem appears to be

the following.
(10.C.5) Theorem (Aleksandrov and Uryson [1923]). A To space X is
metrizable if and only if there is a has a development U,,U,... for X such
that if U,V €U, and UNV #9, then UUV C W for some W € U,-1.
Rather than prove this theorem, we show how others have taken advan-
tage of it to establish their own metrization results. For instance, Frink [1937]
utilized the Aleksandrov-Uryson theorem to derive the following surprisingly
useful theorem.
(10.C.6) Theorem. A T, space X is metrizable if and only if for each
x € X, there is a decreasing sequence of open sets Uf D UZ D --- such that
(i) {U? : 1 € Z*} forms a neighborhood basis at x, and
(ii) for each z € X and each rae ny,oe is an es N(2,i) > 1 such
that for each y € X, if U?n(a,2) Cre. i) #9, then Ce
U. poe:
Proof (Frink [1937]). For each x € X select a sequence of positive
integers as follows. Let af = 1, af = n(z,a7Z), at = n(x, a3), etc. Note that
at < aj <.---. For each i, let Vv = Ujs, and set V; = {V? : « € X}.
Clearly, {VZ : i= 1,2,...} forms a neighborhood basis at z.
A Sampling of Metrizability Theory 257
We show that Vj, V2,... is a development for X and satisfies the hy-
pothesis of the Aleksandrov-Uryson Theorem. Suppose x € X and i € Zt.
To establish that the V;’s form a development, an integer k € Z+ must be
found such that St(z,V,) C V;7. Let k = aZ,, and suppose that z € V,’.
Since ay > k = az,,, we have V,’ = ny C Vys_,- Therefore, x € Uae,, ©
Uge = Vi? (a7; = n(x, a7)). Thus, Vi, V2,... is a eau @ X.
Now suppose Vi, V,4., # 0, or equivalently, OFC aaz) ae a¥) au:
Then if n(z,a?) < n(y,a¥), we have U?niga) Oe a?) 4 0, which by hy-
pothesis implies V,., = U%,z GUSaay Uzaon Ve: An analogous ar-
gument results in a similar Poreieon ifn(y,a/) < n(z,a?). Hence, by the
Aleksandrov-Uryson theorem, X is metrizable. The converse is straightfor-
ward and left as a problem (see Problem C-4). O
The following question is of considerable interest. Suppose X is metriz-
able and f : X > Y is continuous and onto. Is Y metrizable? Stone used
Frink’s result to obtain a particularly elegant solution to this problem.
(10.C.7) Theorem. Suppose (X,d) is a metric space and f : X > Y is

continuous, closed, and onto. If Fr f~1(y) is compact for each y € Y, then
Y is metrizable.
Proof (Stone [1956]). For each y € Y and n € Z*, we define the sets
= $4,(Frf(y), UY = NY Uint f(y), VY = ULF-M@) + f@)
UY}, and WY = f(V,Y). Note that WY = Y \ f(X \U¥), and since f is closed,
both WY and V,¥ are open.
We first show: for each y € Y, the set {WY : n € Z+} satisfies the
conditions of (10.C.6). Obviously, the WY’s form a decreasing sequence of
open sets, and verification that they comprise a neighborhood basis at the
point y can be safely entrusted to the reader. Establishment of condition (ii)
of Frink’s theorem is considerably trickier. Suppose y € Y andi € Z*. If
int f(y) # @, select an arbitrary point x, € int f~'(y) (if int f~(y) = 0,
don’t worry). Now choose n to be sufficiently large so that
(it > 2s,
(ii) if Frf-!(y) 40, then d(Fr f~!(y),X \ Vi) > 2/n, and
(iii) if int f(y) £9, then S1m(ay) C f(y).
We assert that n = n(y,i). Suppose WYN WZ 4 9. Two steps are used to
show W2 c WY.
Step 1. We show f—1(z) C V4: Let p € Windas and select w €
foi(p) CVZNVY CUZNUY. Ifw € int f—*(z), then f—'(z) NV,Y #90, and
heneesf A(z )yre ViGiVe..
If w ¢ int f—!(z), then w € NZ. In this case w € X \ int f~'(y), for if
not, we have (1) f~!(y) c V,’, (2) ty € V,? C (int f~'(z) U NZ), and since
ty ¢ f—‘(z), and it follows that (3) zy € Nz. This, however, contradicts
condition (iii) above. Consequently, w € NZ MN}, and there are points

u € Fr f7!(y) and v € Frf~1(z) such that d(w,u) < 1/n and d(w,v) < 1/n;
hence, d(u,v) < 2/n. Therefore, by (ii) above, we have v € Vx, and since
f(z) OVY #0, from which f~'(z) C Vi follows. O
Step 2. We show UZ Cc U!: By step 1, we know that f~*(z) C UY, =
N¥, Uint f-1(y), and hence, f—!(z) c N¥,. Let u € UZ. Choose v € f—*(z)
such that d(u,v) < 1/n < 1/2i. Since v € N¥,, there is an s € Fr f(y)
such that d(s,v) < 1/2i. Hence, d(s,u) < 1/i and u € N? C U?. Therefore,
UCU,
To complete the proof of the theorem, we merely observe that V,7 Cc V,’
is immediate from step 2, and consequently we have W7 CWY. O
An important corollary is the following (which was used in the proof of

(9.B.3)).
(10.C.8) Corollary. Suppose X is a compact metric space and Y is Tp. If

there is a continuous, onto function f : X — Y, then Y is metrizable.
(10.C.9) Corollary. Locally finite CW-complexes are metrizable (cf. with

(8.E.20)). O
(10.C.10) Definition. A space X is fully normal if and only if each open

cover of X has an open barycentric refinement.
(10.C.11) Remark. It follows from the proof of (10.A.14) that any para-
compact space is fully normal. Stone [1959] showed that T2 fully normal
spaces are paracompact.
We use the Aleksandrov-Uryson theorem to prove: fully normal devel-

opable spaces are metrizable.
(10.C.12) Theorem. A space X is metrizable if and only if X is fully

normal and developable.
Proof. The following notation is used: if 4 and V are covers of X, then

UNV={UNV :U EU and V € VY}. Suppose D,, Do,... is a development
for X. Let H, be a barycentric refinement of D; and set Uz = Hi MD>2. Note
that U2 refines D2 and barycentrically refines D,.
Let H2 be a barycentric refinement of U2 and set U3 = H2MD3. Contin-
uing in this fashion, we obtain a sequence of open covers U2,U3,..., where
each Uj; refines D,,D2,..., Dj; and for each i, U4; is a barycentric refinement
of Ui. It is not difficult to see that {U; : i = 2,3,...} satisfies the require-
ments of the Alexandrov-Uryson theorem. Again, the easy converse is left
to the reader (Problem C-4). O
Moore Spaces 259
Bing gave an especially convenient metrization theorem in terms of a

condition that is somewhat stronger than normality.
(10.C.13) Definition. A family of subsets {H, : a € A} of a space X
is discrete if and only if {Hq : a € A} is locally finite and the H,’s are
mutually disjoint.
(10.C.14) Definition. A topological space X is collectionwise normal if

and only if for each discrete family of subsets {Ha : a € A} there are
mutually disjoint open subsets {Ga : a € A} such that Hy C Gy for each
aeéA.
A proof of the next theorem may be based on showing that developable

collectionwise normal spaces are paracompact.
(10.C.15) Theorem (Bing [1947]). A space X is metrizable if and only if

X is collectionwise normal and developable.
(10.C.16) Definition. A cover U of a space is o-locally finite if and only

if / can be expressed as the union of a countable collection of families, each
of which is locally finite.
Uryson’s metrization theorem is an easy corollary of the following result,
which is one of the most powerful of all the metrization theorems. A proof
of this result may be found in Kelley [1955].
(10.C.17) Theorem (Nagata [1950], and Smirnov [1953]). A space X is
metrizable if and only if X is T3 and has a o-locally finite basis.
D. MOORE SPACES
There is continuing interest in investigating an old concept: Moore spaces. In
fact one can argue that Moore spaces have helped give rise to a whole branch
of topology—set theoretic topology. Moore spaces are a generalization of
metric spaces.
(10.D.1) Definition. A T3 space with a development is a Moore space.

Clearly, metric spaces are Moore spaces: however, there are examples
of Moore spaces that are not metrizable. One such example is given below.
That Moore spaces are “almost” metric spaces can be inferred from the
following theorems, which hold in both metric spaces and in Moore spaces.
(i) Suppose M is a Moore space and A C M. Then A is compact if
and only if each infinite sequence in A has an accumulation point
in M.
(ii) If a Moore space is Lindel6f, then it is second countable.
(iii) If M7 is a Moore space and every uncountable subset of M has an
accumulation point, then M is second countable.
Thus far, we have the following relationships:
Paracompact + developable ae developable + fully normal
— metrizable
(10.C.12)
Hence, in particular, a paracompact Moore space is metrizable. Can the

requisite of paracompactness be weakened? This leads to the celebrated
Moore Space Conjecture: Every normal Moore space is metrizable.
Jones [1937] showed that every separable normal Moore space is metrizable,
however the general conjecture has deep set theory ramifications. It is nev-
ertheless in some sense settled. Quoting from van Mill and Reed [1990], “We
now know that the normal Moore space conjecture is in the hands of the
gods and large cardinals (see Tall [1984] and Fleissner [1984]).” We next
give an example of a Moore space that is not metrizable.
The space to be constructed is connected and locally Euclidean (if it
were metrizable, it would be a 2-manifold). In €? start with a Cantor set on
the x axis of ee and let Aj; Ai, Aj; Aiii, Aj12, Aj21,; Aj22;3 ... denote the
“middle thirds.”
We lay out a network of roads from the point (3, 5 ), all of which are
directed toward the Cantor set in such a way that every point of the Cantor
set is a limit of precisely one route and each route eventually “arrives” at
precisely one point of the Cantor set.
Moore Spaces 261
From each point of the Cantor set, extend a ray in the negative y di-
rection parallel to the y axis, as indicated in the preceding figure. Let M be
the union of these rays and the road system, and let N = M x {-1, 1].
To obtain a space that is locally €?, something must be done about the
branch points. Consider a cylinder about a typical branch point p. This
“neighborhood” of p is replaced by a hexagon. The sides @), £2, £3 are glued
to £), £4 and £5 respectively, and thus the “fat Y’s” around a branch point are
replaced by a twisted two-dimensional hexagon. This procedure is repeated
for each branch point. Let X be the space derived from N in this manner.
What is the topology for X? For points outside K x [-—1, 1], the usual relative
plane topology is used, i.e., disks form a basis at each point. Suppose w €
K x [-1,1], and say w = (z,0). A typical neighborhood for w will have the
following form. Let € > 0 be given. Starting at a point a distance € from the
Cantor set, we select the route in X that leads to w: Near a point (which is
not a branch point), we have the following situation (a)
(a)
As we enter a hexagon, we simple proceed as shown in (b), where the dark-

ened lines indicate which branch was taken en route to w and the shaded
area denotes the neighborhood in question. This gives us one half of the
neighborhood of w.
The “other half” of the neighborhood is easily described as simply con-
sisting of an open disk that lies in the product of [—1,1] and the ray ema-
nating from w.
It is easy to see that the two halves form a whole disk (without boundary).
(10.D.2) Exercise. Show that X with the topology described above is con-
nected (even arc connected), regular, T2, and each point has a neighborhood
homeomorphic to €?.
Completion of Metric Spaces 263
(10.D.3) Exercise. Show that X is developable.

(10.D.4) Exercise. Show that X is not metric.
E. COMPLETION OF METRIC SPACES

We deal briefly with the idea of completing a metric space. A quick review of
Chapter 3 might be in order here. In that chapter it was noted that although
completeness fails to be a topological invariant, topological completeness by
definition is preserved under homeomorphisms. There do exist, however,
metric spaces that are topologically complete. Nevertheless, all metric spaces
are “almost” complete in the sense that they can be embedded as dense
subsets of complete metric spaces. This result is a generalization (both in
content and in proof) of one of the standard constructions of the real numbers
from the rationals. Recall that a map f : (X,d) — (Y,d) is an isometry if
and only if d(z1,22) = d(f(z1), f(z2)) for each 71,22 € X.
(10.E.1) Theorem. Every metric space can be isometrically embedded as

a dense subset of a complete metric space.
Proof. The details of the proof, although not difficult, tend to be tedious.
We suppose that the reader has had previous experience with similar details
in the Cauchy sequence approach to the construction of the real line, and
hence we merely outline the major steps in the proof.
Suppose (X,d) is a metric space.
Step 1. Define C to be the class of all Cauchy sequences in X.
Step 2. Define d: C x C > (0,00) by d({sn}, {tn}) = limnsco d(Sn, tn).

Step 3. Note that d has all the properties of a metric except that
d({sn},{tn}) = 0 does not necessarily imply that {sn} = {tn} (dis a pseudo-
metric).
Step 4. Rectify the problem in step 3 by defining the equivalence rela-

tion: {sn} ~ {tn} if and only if d({tn},{tn}) = 0. Let C be the correspond-
ing set of these equivalence classes, and redefine d in the obvious manner on
the equivalence classes. Then d is a metric.
Step 5. Define 6 : X — C, where 6(z) is the equivalence class represent-

ing the constant sequence rz, = 7 for all n.
Step 6. Note that 0(X) is dense in ae since if [{s,,}] € GC. then for large
k we have that 6(s;) is “close to” {sp}.
Step 7. Show C is complete. Note: since 6(X) is dense in C, it suffices

to show that Cauchy sequences in 0(X) convergeinC. O
Although a subspace of a complete metric space is not necessarily com-

plete, is it topologically complete? The reader should realize that this need
not be the case. The following theorem, however, gives a surprisingly general
condition for which the answer is yes.
(10.E.2) Theorem. Suppose (X,d) is a complete metric space and U is an

open subset of X. Then U is topologically complete.
Proof. We establish the theorem by showing U is homeomorphic to

a closed subset of a complete metric space, and therefore is topologically
complete. Define a function f : U > &! by f(u) = 1/(d(u,X \ UV), and let
W = {(u, f(u) € X x E1 : ue U}. Then it is not difficult to show that W
and U are homeomorphic and W is closed in X x €'. O
(10.E.3) Corollary. Locally compact separable metric spaces are topolog-

ical complete.
Proof. Suppose X is a locally compact separable metric space. Then by

(10.C.2), its one point compactification, X*, is a compact metric space and
hence complete. However, X is an open subset of X*, and the result follows.
O
(10.E.4) Theorem. If A is a G5 subset of a complete metric space X, then

A is topologically complete.
Proof. The trick is exactly the same as that used in the previous theo-
rem: the set A is embedded as a closed subset of a complete metric space.
Since A is a Gs subset of X, A may be written as an intersection of open
subsets of X, say A = ee U,. By the previous theorem, each U,, is
topologically complete, and hence by (6.A.17) the product ie U, is a
complete metric space. Embed A in []*°_, Un by »: A > [°° Un, where
w(a) = (a,a,...).
Again, it is not unreasonable to ask the reader to check
that A and (A) are homeomorphic and 7(A) is closed in []°°., Un. O
(10.E.5) Exercise. Show that topologically complete spaces are Baire

spaces and hence the rationals are not topologically complete.
(19.E.6) Exercise. Show that the irrationals are topologically complete.
If one can complete metric spaces, can one “complete” Moore spaces?
The question is meaningless unless a reasonable definition can be given for
“complete Moore space.” Theorem (3.B.8) is used to motivate the following
definitions.
Problems 265
(10.E.7) Definition. Suppose that € = {C,,C2,---} is a nested develop-

ment for a Moore space M (i.e.,C; D C2 D ---). We say M is complete relative
to € if and only if for any decreasing sequence of close sets F, D Fy D --+ with
the property that for each n, F, D G, for some Gp € Cn, the intersection
Nh=i Fn 4 0.
(10.E.8) Definition. A T, space M is a complete Moore space if and only
if M has a nested development € such that M is complete relative to €.
In closing, we mention (without proof) a few theorems concerning com-

plete metric spaces that also hold for complete Moore spaces.
(i) The countable intersection of dense open sets in a complete Moore
space is dense in M.
(ii) If M is a locally connected complete Baire space, then connected
open subsets are arcwise connected.
(iii) If M is a complete Moore space for which there do not exist un-
countably many disjoint open sets, then M is separable.
There are examples of Moore spaces that cannot be embedded in com-

plete Moore spaces, and this has prompted a great deal of investigation
of additional properties that a Moore space must possess in order to have
a completion. A very readable account of this problem may be found in
Armentrout [1967]. A number of open problems about Moore spaces are
mentioned in the book edited by van Mill and Reed [1990].
PROBLEMS
Section A
1. Show that a closed subset of a paracompact space is paracompact.

2. Show that every regular second countable 7; space is paracompact.
3. Show: if each open subset of a paracompact space X is paracompact,
then every subset of X is paracompact.
4. Show that [0,2] is paracompact, but contains a nonparacompact subset.
[Hint: Try [0,).]
5. Show that the product of a compact T) space and a paracompact space
is paracompact.
Show: if X x Y is paracompact, then X and Y are paracompact.
7. Suppose X is a T2 space such that each open cover of X has a locally
finite closed refinement. Show that X is paracompact.
8. A space X is metacompact if and only if each open cover of X has a

point-finite open refinement (i.e., if / is an open cover of X, there is an
open refinement V of U/ such that each x € X belongs to at most a finite
number of members of V.) Show that a space is compact if and only if it
is both countably compact and metacompact. [Hint: Use Zorn’s lemma
to show that every point-finite cover has an irreducible subcover.|
9. Show: a paracompact space X is compact if and only if X is countably
compact.
10. Show that the space [0,{2) is T; but not paracompact.
Section B
1. Show that a T> space X is normal if for each locally finite cover U/, there
is a partition of unity subordinate to U.
2. Suppose X is paracompact andU = {U, : a € A} isa locally finite open
cover of X. Show: if w : A — (0,00) is a function, then there is a map
f : X — (0,00) such that for each « € X, f(x) < sup{(a) : c € Ug}.
3. If {U,,U2,...,Un} is finite open cover of a normal space X, show there
are continuous functions fj, fo,..., fn such that:
(alesis X= [0,1 doreach tHal,.2;
5 am
(bi 4a: fy se Ohew@ for esen a= Lite gad
(©) Shy AG) = 1.
(Hint: Use (4.C.2) and (4.B.1).]
4. Use Problem 3 to show that every compact n-manifold M can be embed-
ded in €™ for some m € Z*. [Hint: Find an open cover {U;,U2,...,U:}
of M such that for each 7, there is a homeomorphism g; : U; > €”. Find
continuous functions fj, fo,..., f¢ as in Problem 3, and fori = 1,2,...,t
define
nore fi(z) - gi(x)(scalar product) if 2 € U;
| ? ifce M\ {y: fi(y) #0}.

Finally define H : M > E*xE€™ by H(z) = (fi(z),..., fe(z), ¢1(z),..., 64 (2)
and show that H is the desired embedding.]
Section C
1. For each n € Z* let I, = I. Let X be the space obtained from 4],

n € Zt} by identifying all left-hand endpoints to a common pomt 2*,
The starfish metric d for X is defined as follows. If t € In, y € re
Problems : 267
and n # m, then d(z,y) = |x — 2*| + |y — 2*|; if n=m, then d(z,y) =

|x — y|. Does the topology induced by the starfish metric coincide with
the porcupine topology?
2. Show that Uryson’s metrization theorem is a corollary of Bing’s metriza-
tion theorem.
3. Suppose a space X has a development. Show that X has a nested
development (i.e., a development C,,C2,... with the property that C; >
Ci+1 for each 7).
4. Prove the converses of Theorem (10.C.6) and (10.C.12).
Show that every metric space is developable. Find a T; space that is
not developable.
6. Suppose X is an uncountable set with the discrete metric. Is X* metriz-
able?
7. Use the Nagata-Smirnov metrization theorem to show: if X is normal
space having a locally finite open cover U/ with the property that each
U €U is metrizable, then X is metrizable.
8. Show: A locally metrizable, T> space X is metrizable if and only if X is
paracompact. (A space X is locally metrizable if and only if each point
x has an open neighborhood that is metrizable.)
9. Suppose X is a metric space, Y is T, and f : X — Y is closed, con-
tinuous, and onto. Show: Y is metrizable if and only if the set of
accumulation points of X is compact.
10. Suppose X is a metric space, Y is Tz, and f : X — Y is continuous and
onto. Show that Y is metrizable if and only if X is compact.
11. Find a a-locally finite basis for €?.
12. Prove or disprove: If X is a topological space and A and B are metrizable
subspaces of X such that X = AU B, then X is metrizable.
13. Show that metric spaces are collectionwise normal.
Section D
1. Show that the “bubble” space described in Chapter 4 is a Moore space

that is not metrizable.
2. Let K be the Cantor set and order the intervals of J \ K by length (and
from left to right for those of the same length). Denote the intervals
by Ai; Aj, A132; Aji, A112} Aj221, A122} ... Let ay be the point in the
plane one unit below the midpoint of A;. Draw segments emanating
from a; that go half way toward the centers of Ai; and Ajo. Let ay
and a12 be endpoints of these segments. Continue as indicated in the

figure. Let M be the union of these segments
Aii1 Ai A112 Ay Ai21 Ai2 A122
a1
and K. If x € K, then from a, there is only one arc in M that arrives at z.

Denote this arc by A,. Topologize M by defining those sets to be open which
consist of open arcs (w, z] lying in A, together with the obvious half-open
arcs leaving each branch points (but not including the next branch point).
\v
(a) Show that M with this topology is a Moore space.

(b) Show that M is not metrizable (note that M is separable but not
second countable).
3. Show that Moore spaces have the following property: There is a count-
able family € of open covers of X such that if H and K are disjoint
closed subsets of X, one of which is finite, then there is a cover C € C
such that no element of C intersects both H and K.
4. Prove (i) in the text following (10.D.1).
5. Prove (ii) in the text following (10.D.1).
6. Prove (iii) in the text following (10.D.1) (Jones [1937]).
Problems 269
Section E
1. Define a completion of a metric space X to be any complete space in

which X is embedded isometrically as a dense subset. Show that all
completions of a given metric space are isometric.
2. Show: the completion of a metric space X is separable if and only if X
is separable.
3. Suppose (X,d) is a compact metric space, A is a dense subset of X,
and f : A—Y is continuous, and Y is a complete metric space. Show
that f may be extended continuously to X if and only if f is uniformly
continuous.
4. Give an example of a topologically complete metric space that is not
locally compact.
5. Suppose (X, d) is a metric space and A is a subset of X such that (A, dj 4)
is complete. Show that A is closed.
6. Suppose Y is a completion of (X,d). Show that Y is second countable
if and only if (X,d) is.
7. Suppose X is a dense subset of a complete metric space (Y,d). Let
d = d\x xx. Show that there is an isometry wp : (C,d) — (Y,d), where
w(6(x)) = 2 for each x € X. (Notation is as in (10.E.1.))
8. Suppose X and Y are complete metric spaces, D and E are dense subsets
of X and Y respectively, and f : D > E is an isometry. Show that f
can be extended uniquely to an isometry F: X ~ Y.
9. Show: the completion of a metric space (X,d) is a compactification of
X if and only if (X,d) is totally bounded. (Hint: Show that a metric
space X having a totally bounded dense subset is totally bounded.]
uepncennn pais FeO, es
nl pronctmey. "2, : ct i
pole ae
#¥ - 246
bagA) Se ont OP 2) aevaque a ee

isola gt ands welttscapipoa
idazsce bam oft malt wud@><%,7,) Yo artform ead: ‘a
hy in Ay ee ® ” @ TH 2 af
Wine
ab Ty were cchho ly See Pacme lo rade wna |fi iy ,poe

ma |, bt) ¢ \ = eo ae ra badd woo! ¥ak
taW wolhaion "S'S diane ev “a
Vvreerhinh LAr ee i had h aegqqv7e. 2
Witvetes XY ha Kt
PUT! bohoets el ima
lt mone @
Chapter 11
NETS AND FILTERS
A. ON THE FAILINGS OF SEQUENCES

Sequences in a nonmetric setting may behave in an erratic fashion. To illus-
trate their eccentricities, we list four propositions involving sequences that
are valid for metric spaceS but which are partially false in a more general
context. Examples are then given to demonstrate what can go awry.
(11.A.1) Proposition. If X is a metric space and x € X is a cluster point

of a sequence {z,,} in X, then there is a subsequence of {z,,} that converges
to°¢e:(3.A.2)
(11.A.2) Proposition. Suppose X and Y are metric spaces. A function
f : X > Y is continuous at a point z if and only if the sequence {f(x,)}
converges to f(z) whenever {z,,} converges to x. (1.F.9)
(11.4.3) Proposition. A subset U of a metric space X is open if and only
if whenever a sequence {z,,} converges to x € U, there is an integer N such
that x, € U when n > N. (Easy exercise)
(11.A.4) Proposition. A subset C' of a metric space X is compact if and
only if each sequence in C’ has a cluster point in C. (3.A.11)
(11.A.5) Example. Let X = (Zt x Zt) U {(0,0)}. For each i € Zt, let
C; = {(i,n) : n € Zt}. Define open sets as follows:
(i) {(m,n)} is open for each (m,n) € Zt x Zt;
(ii) a set U containing (0,0) is open if and only if U has the property
that there is an integer Ny such that (X \U)NC; is finite (or
empty) for each i > Nu.
Let f : Zt > X \ {(0,0)} be any onto map, and set 2, = f(n) for each
n € Zt. It is clear that (0,0) is a cluster point of the sequence {z,}. We
show: no subsequence of {x,,} converges to (0,0).
271
272 11. Nets and Filters
Suppose to the contrary {rn,} is a subsequence of {z,} converging to

(0,0). Note that if infinitely many of the zn, lie in Cy for some k € Zt,
then the sequence {z,n,;} fails eventually to be in the open neighborhood
Use, CiU {(0, 0)} of (0,0). Thus, each Cy contains at most a finite number of
members of the sequence {rn,;}. Let U = {(0,0)}U(X \Uf{{2n,} : 1€ Z*}),
and observe that U is an open neighborhood of (0,0) which does not contain
a single member of the “converging” subsequence {Zp,}.
(11.A.6) Exercise. Show that the space X of the above example is T> and
Lindel6of.
(11.A.7) Example. Suppose X = R! has the following topology: a set

U Cc X is open if and only if X \U is countable or U = 9. Note that the only
convergent sequences are the ones that are eventually constant. Let Y = €!
and let f : X - Y be the identity function. If s € X and a sequence {z,}
converges to z, then the sequence {f((r,)} will converge to f(x), but f is
not continuous at any point z.
(11.4.8) Example. Let X = R’ have the topology described above and

U = {x}, which is a nonempty nonopen subset in X. Note that whenever
a sequence {z,,} converges to x € U, there is an integer N such that for
n> N, ap, € U. Hence, (11.A.3) need not hold in a nonmetric setting.
(11.A.9) Example. Let X = [0,9) with the order topology. Then every
infinite sequence in X has a cluster point, but X is not compact.
We are too compassionate to expose all of the failings of sequences,

but the foregoing examples should be enough to convince the reader that
sequences are somewhat less than adequate. Nets and filters were invented
to obtain theorems that should, but do not, hold for sequences.
B. NETS
The notion of sequence may be generalized as follows.
(11.B.1) Definition. Suppose D is a directed set (6.B.1) and X is an

arbitrary set. A net in X is a pair (NV, D), where N is a function mapping
D into X.
We often refer to the net (IV, D) by simply writing N : D > X, or just

N when the directed set D is clear from context. If a € D, then N(qa) will
be denoted by zy; when there is little chance for confusion, the function N
is denoted by {rq} or {Za}aep.-
The following example points out that sequences are special cases of
nets.
Nets 273
(11.B.2) Example. Let D = Z* with the usual ordering. Then any

function N : D -+ X is both a sequence and a net.
(11.B.3) Example. Let X = [0,00) and let D = P(X) be ordered by
inclusion. If one defines N : P(X) — X by N(A) = inf{z : x € A}, then N
is a net but is not a sequence.
As was the case with sequences, the most important property associated
with nets is convergence.
(11.B.4) Definition. If N : D > X is a net, then N is eventually in a set

A C X if and only if there is an a € D such that for each B > a, N(@) € A.
A net N converges to a point x € X if and only if N is eventually in each
open set containing 2.
Note that this definition when specialized to sequences, agrees with

(1.F.8). Does the net in (11.B.3) converge?
Before attempting to investigate the interplay between nets, continuous
functions, and topological spaces, we introduce the concept of subnet.
(11.B.5) Definition. Suppose N : D > X is anet. A subnet of N is a net

(N, &), where there is a function P : E > D such that
(i) NP =N, and
(ii) for each d € D, there is an e € E such that P(z) > d for all z >e.
Note that while P need not be strictly increasing, the tag end of P eventually
surpasses any given element of D. Note further that we are not assuming
is a subset of D!
(11.B.6) Exercise. Show that subsequences are subnets.
(11.B.7) Definition. A net N : D > X is frequently in a set A C X if and
only if for each a € D, there is a 8 € D, B > a, such that N(B) € A. A
point x € X is a cluster point of N if and only if N is frequently in every
neighborhood of z.
The reader should note the similarity between cluster points of nets and
cluster points of sequences.
We now proceed to show how nets may be used to generalize the first
and last propositions for metric stated for metric spaces in Section A. The
other two are saved for the problem set.
(11.B.8) Theorem. Suppose VN: D > X isanet in X. A point r € X isa
cluster point of a net N if and only if there is a subnet of N that converges
OMG:
Proof. Suppose z is a cluster point of N = {xq}. To find a convergent
subnet, we first define E = {(a,U) : a € D, U is open and z,zrq € U}.
The fact that x is a cluster point ensures that E is nonempty. We order E
as follows: (a},U) < (a2,V) if and only if ay < ag and V C U. Define

P:E-D by P(a,U) =a. It is clear that (NP, E) is a subnet of {rq}.
To show that this subnet converges to 2, we suppose that W is an open
neighborhood of x and then choose a* € D such that ra- € W. So we have
(a*,W) € E. Note: if (a,U) > (a*,W), then U C W. Since zq € U, it
follows that NP(a,U) = N(a) =z. €U CW.
Conversely, suppose that N : D > X is a net containing a subnet
NP: E- X that converges to a point x € X. Suppose U is an arbitrary
neighborhood of x and a € D. We want an & € D, @> a, with N(a) € U.
There is an e; € E such that P(e;) > a, and an eg € E such that eg > e;
and NP(e2) € U. Let @= P(ez2). Then N(@) EU. O
Resolution of the problems encountered in Examples (11.A.7) and (11.4.8)
above are quite routine and left to the reader. We now consider the relation-
ship between nets and compactness.
(11.B.9) Theorem. A space X is compact if and only if each net in X has

a convergent subnet.
Proof. By (11.B.8), it suffices to show that XY is compact if and only if
each net in Y has a cluster point. Suppose every net in X has a cluster point.
We use the finite intersection property characterization of compactness to
prove that X is compact. Let A be a collection of closed subsets of X with
the finite intersection property. Expand A to a family 8 that includes all
finite intersections of members of A. Direct B by containment (B > B if
and only if BC B). For each B € B, select a point zg € B (the axiom of
choice permits this) and define N :B > X by N(B) = zz.
Clearly, N is a net in X and hence must have a cluster point xz. We
show that z € (\{F : F € A}. Let B be any element of B and suppose that
B€BandB> B. Then N(B) € B CB. Therefore, the net N is eventually
in B for each B, and since B is closed, we conclude x € B. It follows that
ref\{B: Be B}cfi{F: Fe A}. Thus X is compact.
Conversely, suppose X is compact and N : D > X is a net. We use
the finite intersection property for compact sets to find a cluster point for
N. For each a € D, let Fx = {N(8) : B > a}. It is easy to see that
F ={F, : a € D} isacollection of sets with the finite intersection property;
furthermore, F = {Fa : a € D} is a family of closed sets that also has
the finite intersection property, and hence has nonempty intersection. Let
y €f\{Fa : a € D}. We claim y is a cluster point for N: If y is nota
cluster point for , there is a neighborhood U of y and an a € D such that
if 86 > a, then N(3) ¢ U. This, however, is impossible, since y € UN Fg.
0
Filters 275
(11.B.10) Exercise. Let (D,>) be a directed set and {Ay : a € D} bea

family of nonempty, compact, connected subsets of a Tz space X such that
Aq C Ag if and only if a > 8. Show that flee p Aa is nonempty, compact,
and connected.
C. FILTERS
The concept of filter, while seemingly far removed from the idea of se-
quence, nevertheless provides an important and useful theory of convergence.
We have seen that neighborhoods of a point are critical in determining con-
vergence at the point. Filters may be considered as an abstraction of the
following two properties of these neighborhoods:
(i) the intersection of two neighborhoods is again a neighborhood, and
(ii) if G is a neighborhood of z, then any set containing G is also a
neighborhood.
(11.C.1) Definition. A filter F on a set X is a collection of nonempty

subsets of X such that
(i) if PF,Fy € F, then Fy N Fy € Ff, and
(ii) ifF €F, and F CG, then
Ge Ff.
If F is a filter, then B C F is called a filter base for F if and only if each
member of F contains a member of B.
(11.C.2) Exercise. Suppose B is a collection of subsets of a set X. Show:

If B has the finite intersection property, then B is a base for a unique filter
on X.
(11.C.3) Examples.
1. A neighborhood base at a point x € X is a filter, called the neigh-
borhood filter at zx.
2. Let N: D— X beanetin X. Then F={ACX : N is eventually
in A} is a filter.
3. The family B = {(a,0o) : a € R'} is a base for a filter.
4. If X is an infinite set, then the family of all sets with finite comple-
ments is a filter in X.
(11.C.4) Definition. Suppose F is a filter in a set X. We say F is even-
tually in a subset A C X if and only if A € F. A filter F in a topological
space X converges to a point x € X if and only if F is eventually in each
neighborhood of z.
Thus one might picture such a filter as funneling down towards z.
Suppose F is a filter in a set X and f : X — Y is a function from X
into an arbitrary set Y. Although f(F) ={f(F) : F € F} is not necessarily
a filter (why?), it is at least a base for a filter {G CY : f(F) CG for some
F € F}. Abusing notation slightly, we also denote this filter by f(F).
(11.C.5) Exercise. Let X and Y be topological spaces and f: X > Y.

Show: f is continuous at a point Zo if and only iffor each filter F converging
to Zo, the filter f(F) converges to f (zo).
Filters like nets may cluster at a point.
(11.C.6) Definition. A filter F clusters at a point x if and only if x belongs

to the closure of each member of F.
Subnets, when translated into filter terminology, become “finer” filters
(11.D.2). If F and G are filters on X, F is said to be finer than G if G C F.
(11.C.7) Theorem. A filter F clusters at a point z if and only if there is
a filter finer than F that converges to z.
Proof. Suppose F¥ clusters at x. Let B be the family of all sets of the
form FU, where F € F and U is a neighborhood of z. If G is the filter
with base B, then G is finer than F and converges to x (why does U € G?).
Suppose now F C G and G converges to x. Since the neighborhood base
at x is contained in G, every member of the base must intersect each F € F
(since 0 ¢ G). Hence, F clusters at z. O
(11.C.8) Definition. A filter / in a set X is an ultrafilter if and only if
whenever V is a filter such that U C V, then U = V, i.e., there does not exist
a filter in X that is finer than U.
(11.C.9) Theorem. Every filter F in a set X is contained in some ultrafil-

ter.
Proof. Zorn’s lemma or some variation of it is unavoidable. Let F be a
filter in X and let 6 = {G : G isa filter finer than F}. Order G by inclusion
(Ga < Gg if and only if Ga C Gg), and let {Ga : a € A} bea nest in ©. It
is easily checked that U{Ga : a € A} is a filter finer than each filter in the
nest, and hence each nest has an upper bound. Therefore, 6 has a maximal
element, the desired ultrafilter. O
The next theorem gives us one of the more peculiar properties of ultra-
filters.
(11.C.10) Theorem. A filter / in a set X is an ultrafilter if and only if

for each A C X, either A or its complement in X belongs to U.
Proof. Suppose U is an ultrafilter, and let / = UU {A}. IE ANU £0
for each U € U, than U is a base for a filter UW. However, since U/ is finer than
U, U must be equal to U/, and hence A € U. On the other hand, if there is a
U €U such that ANU =96, then U C X \ A EU. The other implication is
left to the reader as an exercise. O
(11.C.11) Exercise. Finish the proof of (11.C.10).

Nets and Filters PT
(11.C.12) Exercise. If Y/ is an ultrafilter on X and f : X 3 Y is onto,

show that f(U/) is an ultrafilter.
(11.C.13) Exercise. If each filter in X has a cluster point, show that each
ultrafilter in X converges.
Compactness may be characterized in terms of filters.

(11.C.14) Theorem. A space X is compact if and only if each filter in X
has a cluster point.
Proof. Since X is compact, it has the finite intersection property. Sup-
pose F is a filter on X. Then {F : F € F} has the finite intersection prop-
erty, and hence the intersection of all these sets will be nonempty. Clearly,
any point in this intersection is a cluster point of F.
Assume that each filter clusters in X, but there is an open cover U with
no finite subcover. We exhibit an ultrafilter that does not converge, thus
contradicting (11.C.13). Let B be the collection of all complements of finite
unions of members of U, ie., a typical B € B has the form X \ Uj_, Ui,
where U; € U. Then B is a base for a filter G. By (11.C.9), G is contained
in an ultrafilter H, which by (11.C.13) converges to a point x. There is
aU € & such that xz € U, and the definition of convergence implies that
U € H. On the other hand, it is immediate from the construction of G that
X\UEGCH. Thus, we have U € H and X \U € H, which is impossible,
since UN(X\U)=0¢H. O
Problems involving compactness are often easily solved with the aid of
filters. For instance, the next two exercises yield a very elegant proof of the
Tihonov theorem.
(11.C.15) Exercise. Suppose X,|a € A} is a collection of topological
spaces and {zq} is a point of [[,¢, Xa. Show: A filter F converges to
{ro} in [],¢, Xa if and only if for each a € A the filter pa(F) converges to
Pa({Za}) in Xq, where pq is the a-th projection.
(11.C.16) Exercise (Tihonov Theorem). Show: If {Xq : a € A} isa
family of compact spaces then J],¢, Xa is compact. [Hint: Use ultrafilters.]
D. NETS VS, FILTERS
Nets were first introduced by E.H.Moore and H.L.Smith [1922], and refine-
ments of the theory were added by Kelley [1950a]. The French, exhibiting
their customary enthusiasm for American inspirations, have demonstrated a
marked predilection for filters. The concept of the filter found its origin in
the work of the noted French mathematician Cartan [1937]. The purpose of
this section is to underscore the futility of any filter vs. net contest by show-
ing that nets arise from filters in a most natural way, and, no less naturally,
filters may be derived from nets. Our personal feeling is that nets are “more
intuitive,” but filters are easier to work with, often yielding elegant proofs.
Suppose N = {zra}eep is a net in a set X. For each a € D, let
Tx = {tg : B > a}. Then it is easy to show that {T, : a € D} has the
finite intersection property and therefore is a base for a unique filter, which
we denote by Fy.
(11.D.1) Theorem. A net N is eventually in a set A C X if and only
if Fn is eventually in A. Consequently, if X is a topological space, and N
converges to € X, so must Fy.
Proof. If N is eventually in A, there is an a € D such that for 6 > a,
zg € A. Thus, T, C A, and hence A € Fy.
If Fy is eventually in A, then A € Fy, which implies there is an a such
that T, C A. Hence, for all G > a, we have rg € Ty C A, ie., {xq} is
eventually in A. O
We previously alluded to a possible link between subnets and filter re-

finements. This relationship is clarified in the next theorem.
(11.D.2) Theorem. Suppose NV : D > X is a net in X with associated

filter Fy. Suppose further P : E — D, where E is a directed set and
NP:E-—- X isasubnet of N. Then the filter Fyp refines Fy.
Proof. Suppose Tg € Fn where a € D. There is a @ € E such that
P(y) > a for each y > 8. Let Sg = {zp(,) : y > B}. Then Sg € Fnp and
Sg C Ty. Therefore, T, € Fup. O
We now see how filters give rise to nets. Suppose F = {Fy : a € D}

is a filter in a set X. Order D so that a, < ag if and only if Fy, C Fa,.
Clearly, D is a directed set. For each Fy € F, select a point rg € Fy. Then
Nr: D-— X defined by Nr(a) = Zq is clearly a net.
The proof of the next theorem is trivial, and is omitted.
(11.D.3) Theorem. A filter F is eventually in a set A C X if and only if

the net N-¢ is eventually in A; consequently, if X is a topological space and
F converges to z € X, so must N-.
One aesthetic mishap does occur however:
(11.D.4) Exercise. Find a refinement G of the filter Fy associated with a

net NV that is not associated with any subnet of N.
Problems 279
PROBLEMS
Sections A and B
Suppose X and Y are topological spaces. Show: a function f : X > Y

is continuous if and only if whenever a net {x.} converges to a point
x*, {f(xa)} converges to f(z*).
Show that (11.A.3) is valid when sequences are replaced by nets.
A net {rq} in X is a universal net if and only if for each A C X, {xq} is
eventually either in A or in X \ A. Show: If a universal net is frequently
in a set, then it is eventually in the set.
Show: Every universal net converges to each of its cluster points.
Show: A space X is compact if and only if every universal net in X

converges to an element of X.
Show that every subnet of a universal net is a universal net.
Let {Xq : a € A} bea family of topological spaces and X = [],¢, Xa:

Show that a net {zg : 6 € V} in X converges to a point x* € X if and
only if {pa(zg)}sew converges to pa(x*) for each projection map pa.
Suppose (NV, D) is a net in a space X, and A is the set of cluster points

of (N, D). Show that A is closed.
Find an example of a space X and a universal net (X,D) in X.
10. (Tihonov Theorem via nets.) Let X = [[,¢, Xa with each Xq
compact. Suppose D is a directed set and N: D> X is anet in X. If
CCA, let pr : []geqn Xa > [ger Xa be the projection map. A point
y € [leer Xa is called a partial cluster point of (N,D) if and only if y
is a cluster point of pp N.
Note: If f = A then y is a cluster point of (NV, D).
Let P be the set of all partial cluster points in (V, D) partially ordered
by inclusion. Then y; C yz if and only if Dom(y:) C Dom(yz2) and yz is an
extension of y.
(a) Show that P satisfies the hypothesis of Zorn’s Lemma.
(b) Apply Zorn’s Lemma to P and show that the resulting maximal
member of P is a cluster point of (N, D).
c) Conclude that X is compact.
Section C
1. If F is a filter, the adherence of F, Adh(F).is defined to be (ert ae

Show: If a filter F in X converges to x, then x € Adh(F).
2. Show that a T) space is compact if and only if Adh(F) # 9, for each
filter F in X.
3. Show that the collection of subsets of R!, each of which contains an
interval of the form (0, co), is a filter.
4. Show that the family of all subsets of R! whose complements are finite
is a filter.
5. Suppose F is a filter in a set X. A subset S C F is a subbase for F if
each set in F contains the intersection of a finite number of sets in S.
If G is a collection of subsets of X with the finite intersection property,
show that G is a subbase for some filter F in X.
6. Suppose F is an ultrafilter in a set X. Show:
(a) (\rer F consists of at most one point.
(b) IfM\pereF = {zr}, then F={ACX : cE A}.
7. Show: A space is T if and only if each filter in X converges to at most
one point. Find an example of a filter with no cluster points.
8. Suppose (X,d) is a pseudometric space and F is a filter in X. Show
that F converges to some point z € X if and only if, for each € > 0,
there is an F € F such that diamF < e.
9. Suppose Ff is a filter in a space X. Denote the set of points of conver-
gence of F by Lim. Show: If F is an ultrafilter in a space X, then
Adh(F) = Lim(F).
10. Show: A neighborhood filter of a point z is an ultrafilter if and only if
x is an isolated point.
11. Prove or disprove: Suppose X is a topological space and A Cc X. A
point z € X is an accumulation point of A if and only if there is a
sequence {z,} in AN (X \ {x}) that converges to x.
12. Show: A filter F in a set X is an ultrafilter in X if and only if whenever
ANF
#0 for each F € Ff, then
A € F.
13. Suppose F is a filter in a set X and whenever AU B € fF, then either
Aé F or Be F. Show that F is an ultrafilter.
14. Find an example of a filter that is not an ultrafilter.
15. Show: The intersection of a family of filters in a set X is a filter in X.
16. Let (X,U) be a door space. Prove or disprove: U/ must be an ultrafilter.
Problems 281
Lie Show: If (X,7) is a completely regular space, then there is a uniform

structure U, for X such that T = Ty (see problem 18 from section D
of chapter 4). [Hint: Think about the filter U/ that has the subbase
S = {Us : f : X — [0,1], « > 0} where f is a continuous function and
Use = {(x,y) : |f(x) — Fly) < €}.]
Section D
1. Show that a net associated with an ultrafilter is a universal net.

2. Show that the filter generated by a universal net is an ultrafilter.
3. Show that every net has a universal subnet.
4. Show: A point z is a cluster point of a net {z,} if and only if ¢ is a
cluster point of the filter associated with z.
Suppose N = {za}aea is a net in X, F is the associated filter base, and
G = {Gg}gew is arefinement of F. Show that the following construction
leads to a subnet of N: Order W by declaring 6; < (2 if and only if
Gg, C Gg,. Rename the points ry as yg’s, requiring only that r_ cannot
be named yg unless it belongs to Gg. Then {yg}gew is a subnet of N.
: [
ire
~ ata
|
J ct anbeap regener
| = ; —
aa oe enel near
= . =, D4 Fed coe
D ik ,
7 P ag ihe otet ee
-_ > Fe 6. «4 @ierce 7, ae
; 7
‘% vara 1! Coy, Carty A

ee)
iv = = an £¢ a
DPA ot WSN 0%
a - way st _
ie vo tani eonly¢ wren
y 04 “ —
ot
.
cheoms dy) CH raion |

Ee
= at oe _ 7
: —— a _
: Mid YS Op tae,
:
: ou andl @ais ag Ve
7
Chapter 12
THE ALGEBRAIZATION OF
TOPOLOGY
In this chapter we approach the interstices between the sublime realm of

topology and the nether world of algebra. Hitherto, we have tackled topologi-
cal problems by imposing one or more conditions on the topological structure.
In algebraic topology, one tries to answer topological questions by translat-
ing them into algebraic ones (which may or may not be easier to resolve). A
way will be found to assign groups to topological spaces and to assign group
homomorphisms to continuous functions in such a manner that composition
is preserved and that identity homomorphisms are associated with identity
maps. Actually, there are a number of ways of doing this, but we restrict
our attention to what is perhaps the most geometrically intuitive method.
Algebraic topology is an active and difficult area of mathematics; our
presentation here only hints at the profundity of the subject matter. Once
adequate topological and algebraic machinery is set up, an extraordinary
variety of deep and interesting results may be obtained.
A. THE FUNDAMENTAL GROUP

(THE FIRST HOMOTOPY GROUP)
We have stressed throughout the book that a central problem in topology is
the determination of whether two given spaces are homeomorphic. Invari-
ant properties such as connectedness, compactness, metrizability, etc., are a
definite aid in attacking this problem, but there is a need for far more pow-
erful tools than these. For instance, how might one distinguish topologically
between the interior of a square (an open disk) and the interior of a square
with some points removed in the middle?
283
284 12. The Algebraization of Topology
Although both spaces are connected, locally connected, locally compact,

metrizable, etc., they do not appear to be homeomorphic. At this junc-
ture, the reader might like to take a stab at proving the spaces are indeed
topologically distinct. The resolution of this problem becomes trivial when
appropriate tools are available.
(12.A.1) Definition. Suppose X is a topological space and Zp is a point
in X. The family C of loops in X based at zo is defined to be the set of all
paths in X that start and end at zo, ie, CL={f:I 4X : f is continuous
and f(0) = to = f(1)}. If xo € X, then the constant loop cz, : I + X is
defined by cz,(t) = Zo for all ¢ € I.
It is possible to use CL in order to extract essential information about
the space X. For example, suppose X is the interior of a square with some
points removed and Y is the whole interior of the square. Let zp and yo be
points of X and Y, respectively. Consider the loops @; and ¢2 of X as shown
in the following figure. Note that @; can be “shrunk” (or pulled back) to xo
without any trouble; however, £2 becomes hung up on the hole if any such
shrinking is attempted. In Y, any loop based at yo can easily be shrunk to
yo. Thus, it appears that a distinction between these two spaces might be
based on the behavior of certain loops.
What does all of this have to do with groups? Suppose for each space X
the loops based at a point xp could somehow be considered as elements of a
group. Then in the preceding examples, the group element associated with
the loop £2 would seem to have no counterpart in the group derived from the
loops in Y based at yo. Consequently, the group associated with X would
differ from that associated with Y. If, in addition, it could be established
that homeomorphic spaces yield isomorphic groups, then one could conclude
that X and Y are not homeomorphic. The primary goal of this section is to
find a reasonable way of assigning groups to topological spaces.
ve
We want to consider two loops 2 and é3 of L to be equivalent if 23 can be

“pushed” or “deformed continuously” into £2 without moving the point zo.
Such “deformations” are given a mathematical formulation in the following
definitions.
The Fundamental Group (The First Homotopy Group) 285
(12.A.2) Definition. A topological pair is an ordered pair (X, A), where

X is a topological space and A is a subspace of X. If (X, A) and (Y,B) are
topological pairs and f : X + Y is a function such that f(A) C B, then we
write f : (X,A) > (Y,B), and call f a mapping of pairs.
(12.A.3) Definition. Suppose (X, A) and (Y, B) are topological pairs and
that f and g are continuous functions from (X, A) into (Y,B). We say f
is homotopic to g relative to A if and only if there is a continuous function
F: (X xI,Ax I) = (Y,B) such that for each + € X, F(z;0)'= f(z),
F(az,1) = g(x), and for each a € A F(a,t) = f(a) = g(a) for allt € I. In
this case we write f ~ g(mod A) or f ~ g. The map F is called a relative
homotopy between f and g. If A = @, then we say that f is homotopic to g
(denoted f ~ g).
If f : X + Y is homotopic to a constant map, then f is said to be null
homotopic.
(12.A.4) Notation. Suppose F : (X x I,A x I) > (Y,B) is a relative

homotopy between f and g. Then for each t € I, define F; : (X, A) - (Y, B)
by Fi (2) = F(z,t). Note that Fo = f and Fi =g.
Fy (A)=f(A)=9(A)
F\()=9(z)
Fo(#)=f(z)
(12.A.5) Notation. If (X,A) and (Y,B) are topological pairs, we denote

{f : (X,A) > (Y,B) : f is continuous} by C((X, A), (Y, B)).
(12.A.6) Theorem. The relation of relative homotopy (mod A) is an equiv-
alence relation in C((X, A), (Y, B)).
Proof. Suppose f,g,h € C((X, A), (Y, B)).
Reflexivity. Define F : (X x I, Ax I) > (Y,B) by F(z,t) = f(z) for all
x and t.
Symmetry. Suppose f ~ g, with F the homotopy between f and g.
Define G: (X x I,A x I) > (Y,B) by G(z,t) = F(z,1—t). Then Go = g,
G, = f, and G;(a) = f(a) for allt € J anda € A; therefore g ~ f.
Transitivity. Suppose f ~ g and g ~ h where the homotopies are given

by F and G, respectively. Define H : (X x I,A x I) - (Y, B) by
He.) F(a, 2t) for0<t<

z, =
( G(a,2t-—1) for$<t<1
Then H is the desired homotopy between f and h. The continuity of H

follows from the map gluing theorem. O
(12.A.7) Exercise. Suppose (X,A), (Y,B), and (Z,C) are topological

pairs, f ~ g where f,g € C((X,A),(Y,B)), and h € C((Y,B),(Z,C)).
Show that hf ~ hg.
We now return to loops by restricting our attention to the following

special situation. Suppose X = I, A = {0,1}, Y is an arbitrary space, and
B = {yo}, where {yo} C Y. The preceding theorem yields an equivalence
relation on the family CL of loops based at {yo}. (Note that if f tg,
then geometrically one may think of the homotopy as moving the loop f
continuously onto g without disturbing the base point.) If f € L, then [f]
will denote the corresponding equivalence class and G the family of these
equivalence classes. We define a binary operation on G, and then show that
this operation converts G into a group.
Suppose [f],[g],€ G. A way must be found of “multiplying” [f] and
[g] to obtain an element [f] - [g] also in G. This is easily done. The loop f
starts at yo, runs around a circuit, as the parameter t € I goes from 0 to 1,
returning to yo at 1; the same, of course, holds for g. A loop that represents
[f] - [g] will look like f on the first half of J (even though it must race with
the parameter twice as fast as f did to arrive back at yo when t = 5) and
then becomes an accelerated version of g on the second half of J. This new
loop is denoted by f * g. More precisely, we have the following definition.
(12.A.8) Definition. If f, g € CL, then f * g is the loop defined by
2 for0
(f *g9)(t) = e : : t
g(2t—1) for 2
Note that f(2t) = g(2t—1) =yo att= ’ so f *g is continuous by the

map gluing theorem. The next theorem shows that this definition induces
a well defined operation in G, namely [f] - [g] = [f * g]. We will use the
customary abbreviation and denote [f] - [g] by [f][g].
(12.A.9) Theorem. Suppose f,g,f,g € L, je ats and gag: Then

[fllo]= [f]la)-
Proof. It suffices to show f * g ean f * g. By hypothesis, there are
homotopies F': J x I+ Y andG:Iwith
xI4Y Fy=f, F, =f, Go=g,
and G, = g. Furthermore, F(0,t) = F(1,t) = G(0,t) = G(1,t) = yo. Define
Hel xl ¥ hy
It is easy to verify that H is the desired relative homotopy between f *g and

Fie ts eS
Frequently, it is necessary to “multiply” paths that are not loops. This

can be done, provided the paths match up properly, i.e., the first path ends
where the second starts.
(12.A.10) Definition. Suppose f : J] + X and g: I > X are continuous

functions satisfying f(1) = g(0). The product f *g: I — X is defined by
_ f f(2t) if0<t<}3
enO= er font <1
g(1)
fits) faig(*y)
Ee, Gig Saye aoteg)
1)=g9(0
(12.A.11) Exercise. State and prove the analog of (12.A.9) for matching
paths.
(12.A.12) Exercise. Suppose f,g : J + X are continuous functions such

that f(1) = g(0). Leth: X — Y be continuous. Show that ee hf hg.
Furthermore, if ee f and g fee9 then [h(f * g)]= [h(f * 9)] (paths
D1, )2,21 > Y are defined to be equivalent if and only if pi.~ + P2):
Theorem (12.A.9) assures a well-defined operation for G. The rest of
this section will be devoted to finding inverses, identity elements, etc., in
order to show that (G,-) is indeed a group.
The homotopies to follow are rather tedious to untangle analytically.

However, contemplation of the accompanying figures should enable the reader
to penetrate the freshman algebra that lies behind the rather arcane formu-
lae appearing in the next few theorems. In any case, it is advisable to check
the given homotopies for their relevance and accuracy.
It is easy to guess which equivalence class acts as an identity element.
(12.A.13) Theorem. Let e = [c,,] be the equivalence class represented
by the constant map c,, : I + Y and let f € L. Then (c,, * f), ~, f and
{0,1}
(f # cy), f;hence e- [f] = f =[f]-e.
Proof. Define a map H:I x I> Y by
Yo if0<t<—2s+1
H(s,t) = oe
Ti ) iif —2 Paid. ie
Sitsol.
As usual, the map gluing theorem rescues continuity.
: f
(1/2,0) :
Note that Ho = c,, +f, Hi = f, and H(0,t) = H(1,t) = yo for all ¢ € I.
A similar homotopy demonstrates that ioe Pens GG
(12.A4.14) Theorem. Suppose f,g,h € £L. Then [f]((9][h]) = ([F][g]) [A],
and thus (G,-) is associative.
Proof. It suffices to show that f * (g * ee (f *g)*h. Define
H(s,t) = 4 g(4s
—-2 +1) for
4s — t --
(FA) cs ae
Me
t
(1/4,1) (1/2,1)
s
(1/2,0) (3/4,0)
Note that Hp = f *(g*h) and H, =(fx*g)*h. ODO
(12.A.15) Exercise. State and prove analogs of the previous two theorems
when f, g, and h are paths that match properly.
Finally, inverses for elements of (G,-) must be found. This is done with
the aid of the well-known principle that if you walk down a path in one
direction and subsequently retrace your steps, then you might as well not
have left in the first place.
(12.A.16) Definition. Suppose f : J — Y is a path in Y. The reverse of

f, f", is defined by f(t) = f(1—t) for each t € I.
(12.A.17) Theorem. If f € £, then [f][f"] = [f"][f] = [ey] = e. That is,

Linear ied:
Proof. It suffices to show (f * I”) es Gyo ee * f). A glance at the
next figure suggests the following homotopy:
t
Yo TOU S08 Sa,
t 1
1a f (2s — t) fot SiS
(s,t) = 1 a
{' (28+%t—1), for =< s< ——
2 2
2-t
Yo for igo
t Cyo (1/2,1) C¥o
f (2s) f’ (2s—1) ;
Note that Ho = (f * f") and H; = c,,. A similar figure and homotopy can
be given to establish that f” * FO) Cv0- Oo
The previous theorems may be combined to yield the principal result of

this section.
(12.A.18) Theorem. Suppose Y is a topological space and yo € Y. Let

L be the collection of loops based at yo, and let G be the set of equivalence
classes obtained from L by declaring two loops equivalent if and only if they
are homotopic mod{0,1}. Then G with the binary operation defined by
(f]- [9] =(f *g], where f,g€£,isagroup. O
The group G is called the fundamental group of Y based at yo (or al-

ternately the first homotopy group of Y based at yo); we follow the usual
custom of denoting G by 71(Y, yo).
The elegance of forthcoming proofs that involve the fundamental group
is not at all presaged by the somewhat messy details used to show that (G, -)
is a group.
B. ELEMENTARY PROPERTIES OF THE

FUNDAMENTAL GROUP
(12.B.1) Definition. Suppose X is a space and zo € X. Then the pair

(X, xo) is called a pointed space. The point Zo is called either the base point
or the preferred point of (X, 20).
As we have just seen, it is relatively easy to ‘associate a group with each

pointed topological space. However, it is quite another problem to actually
recognize one of these groups; in fact, except for a number of special cases,
this has not been successfully done.
Elementary Properties of the Fundamental Group 291
Before any attempt is made to calculate a fundamental group, we show

the role of the base point is minimal in path connected spaces.
(12.B.2) Theorem. Suppose X is path connected and Xo,2; € X. Then

m™(X,2o) and m(X, 21) are isomorphic.
Proof. Suppose xp and 2; are in X, and k: I > X is a path from zo to
£,. For each loop based at xo, we want to associate a loop based at x1, and
visa versa. Let f : J + X be an arbitrary loop at zo, and observe that there
is a natural choice for a corresponding loop based at x,, namely k” « f *k
(walk backwards along k to get from x, to xo, then proceed along f, and
finally return to 2 via k).
Define functions ky : m(X,20) > m(X,21) by ky([f]) = [k” * f * kl]

and ki : m(X,21) > ™(X, 20) by ky ([g]) = [k * 9 *k"]. The reader should
verify that these are well-defined homomorphisms and are inverse to each
other (use (0.C.11)). O
When X is path connected, we frequently denote the fundamental group
of (X, xo) by ™(X).
The reader should be aware that a different choice of the path k between
Zo and 2 in the above theorem may give a different isomorphism. However,
if k and k are homotopic mod{0, 1}, then the isomorphisms they determine
are the same. (Problem A-B 14).
The next theorem shows that the assignment of fundamental groups to
topological spaces is well behaved with respect to finite products.
(12.B.3) Theorem. Suppose (X,z9) and (Y,yo) are pointed topological

spaces. Then 7(X x Y, (Zo, yo)) is isomorphic to 7 (X, 20) x ™(Y, yo), the
direct product of groups.
Proof (outline). For each loop in X x Y with base point (xo, yo), corre-
sponding loops must be located in X and Y. This is accomplished via the
projection maps px and py, which send X x Y onto X and Y, respectively.
If f : I + X xY is a loop in X x Y, then pxf and pyf are loops in X
and Y. Define a function w : ™(X x Y,(z0,yo)) ™(X,20) x ™(Y, yo)
by #([f]) = ([pxf], [py f]). It is not difficult to verify that this map is well-
defined.
In order to show w is a homomorphism, we observe that
W((f]lg]) = YF * 9]) = (lex(F* 9)]; [py (F * 9)])

= ([pxf* pxg], [pyf* pyg))
= ([px f][px 9], py fpr9}) = ¥([F)) d([9))
The third equality follows from (12.A.12).
To see that ~ is one to one, suppose w([f]) = ({g]). Then ([px f], [pyf]) =
([px9];[pyg]), which implies that [pxf]= [pxg] and [pyf]= [pyg]. Let H
and K be the corresponding homotopies. Define F: I x I ~ X x Y by
F(s,t) = (H(s,t), K(s,t)). It is easy to verify that F is a suitable homotopy
between f and g, and hence [f] = [g].
Finally we show w is onto: if ([f1], [fo]) € 71(X,20) x m™(Y, yo), then
the equivalence class determined by the following loop in X x Y
O0<t<i
k(t) = rete.
(xo, fo(2t-—1)) $<t<l
is mapped onto ((f1],[fo]). 0
The fundamental groups of some rather simple topological spaces are
calculated in the next section. Before succumbing to such formalities, how-
ever, we attempt to discover in an intuitive fashion the fundamental group of
S!. (Two complete but distinct proofs of the fact that 71(S1,
sq) is infinite
cyclic are given in Chapter 13 and 14.)
A fairly obvious candidate for a generator of 1(S1, sq) is the equivalence
class [a] determined by the loop shown in the next figure, where a wraps
once around S! in the direction indicated.
If a loop # is wrapped n times around S! in the direction of a, it is

reasonable to expect that the corresponding equivalence class in 7(S!, so)
is [a]". The inverse of [(] can be derived from the loop that runs around
the circle n times in the opposite direction. Of course, not all loops are so
simple, since paths may reverse directions repeatedly. Imagine the ‘edge’
of S' to be covered with Velcro. Represent the path by a rubber strip that
wraps around to and fro according to the path 6. Then magically replace the
Velcro with a frictionless substance and the strip becomes the image of an a”
for some n. Consequently, it seems reasonable that 1(S!, so) is isomorphic
to the group of integers under addition.
(12.B.4) Exercise. Assume that ™(S!,s9) ~ Z. Find the fundamental
group of the torus.
Continuous Functions and Homomorphisms 293
C. CONTINUOUS FUNCTIONS AND HOMOMORPHISMS
Continuous functions induce homomorphisms in a particularly simple way.

Suppose f : (X,20) > (Y,yo) is continuous. If a is a loop in X based
at Zo, then fa is a loop in Y based at yo. Furthermore, it is easily seen:
if a aos then fa igen fc& (12.A.7). Thus, f induces a well-defined map
fa 2™1(X, 20) 9 m1 (Y, ie) defined by f.([a])= [fa]. We call f, the homo-
morphism induced by f. The ease of establishing the following assertions is
only surpassed by their importance.
(12.C.1) Exercise. If f : (X,20) > (Y,yo), g : (X,20) — (Y,yo) and

h: (Y,yo) > (Z, 20) are continuous functions, show:
" ppg see: ROP Omer p heat
b)Falla) = (F(la))-* fo fa] € (X20)
(hf). =
a hs tej
g, then jf. = Gg.
(e lie “m1 (X, Zo) — 1 (X, 20) is the identity homomorphism, where
id :(X,20)> (X, Zo) is the identity map.
We now give a key concept of this section.
(12.C.2) Definition. Two topological spaces X and Y are said to be homo-

topically equivalent if and only if there are continuous maps f : X — Y and
g:Y +X such that fg ~ idy and gf ~ idx. Topological pairs (X, A) and
(Y, B) are homotopically equivalent if and only if there are continuous maps
f : (X,A) — (Y,B) and g: (Y,B) — (X,A) such that fg ~ idy(mod B)
and gf ~ idx(mod A). We call f (or g) a homotopy equivalence and say that
f is a homotopy inverse of g (or g is a homotopy inverse of f).
(12.C.3) Examples.
1. Homeomorphic spaces are clearly homotopically equivalent.
2. Suppose D is the unit disk minus its center and E = S'. Let
f:E—- D be the inclusion map, and g: D > E be the obvious projection
map. Then it is easy to see that gf = idg and fg ~ idp; hence, D and E
are homotopically equivalent, and f and g are homotopy equivalencies.
We are now able to prove a result that implies that homeomorphic spaces
have isomorphic fundamental groups, or equivalently, if the fundamental
groups of two spaces differ, the spaces cannot be homeomorphic.
(12.C.4) Theorem. If pointed spaces (X, xo) and (Y, yo) are homotopically
equivalent, then 7(X,z0) and 71(Y, yo) are isomorphic (cf. Problem C 17).
Proof. Suppose f : (X,20) > (Y,yo) and g : (Y,yo) > (X,2o) are
homotopy inverses. Then from (12.C.1) parts (c), (d) and (e) we obtain
PCP = (fg)« = idx s(Y,yo) and Oxf = (of)« = idx, (X,x0)- Consequently,
each of f, and g, is an isomorphism. O ;

(12.C.5) Corollary. If h : (X,20) — (Y,yo) is a homeomorphism, then
hy : 71 (X,2%0) 4 7 (Y, yo) is an isomorphism. O
A large number of spaces have trivial fundamental groups; included
among them are the contractible spaces.
(12.C.6) Definition. A space X is contractible if and only if there is a
point zo in X such that the identity map id : X — X is homotopic to the
constant map c,, : X — X. In this case, X is said to be contractible to zo.
[Note: we are not assuming the homotopy is mod{Zo}.]
(12.C.7) Example. Every convex subset of €” is contractible. Suppose
that A C €”" is convex. Let ro € A. Then F': Ax IJ —- A, defined by
F(a,t) = ta + (1 — t)zo is the desired homotopy. In fact a weaker condition
star convex at Xo is sufficient, i.e., for each point a € A the line segment
between a and Zo lies in A.
In (12.C.7), the contraction homotopy does not move Zo. This is not
always the case as may be seen in the next example.
(12.C.8) Example. For eachn € Zt, let An = {(1/n,y) € E27 : 0<y< 1}

and set 45 = 4(0,9) Ge: OS yl}. Then X = KYO} OU, An)
contractible to p = (0,1) € A, although p must move during the contraction.
The next theorem lists some basic properties of contractible spaces.
(12.C.9) Theorem.
(i) Contractible spaces are path connected.
(ii) If X is contractible, Y is a topological space, and f,g: Y ~ X are
continuous, then f and g are homotopic.
(tit) A space X is contractible if and only if X is homotopically equiva-
lent to a point.
Proof. (i) Suppose that X is contractible to zo, and that 2; € X. Let
H : X x I + X be a homotopy such that Hp = idx and H, = c,, where
Ce is defined by cz,(z) = 2. Then f : I + X defined by f(t) = H(21,t)
is a path from x; to zo. It now follows easily from (2.C.3) that X is path
connected.
(ii) Let H : X x I -+ X be the homotopy given by part (i). Define
w:YxI-+X by wly,t) = A(f(y),t). Then ~ is a homotopy between f
and c,,. All maps from Y into X are homotopic to the constant function
Cz,, and hence to each other.
(177) Suppose that X is contractible to x9 € X. Let g: {ao} + X be
the inclusion map and let f : X — {zo} be the constant function. Then
gf ~ idx and fg =id;,,}. Thus, X and {x9} are homotopically equivalent.
Continuous Functions and Homomorphisms 295
Conversely, suppose X and {p} are homotopically equivalent, where

p € X. Then there are continuous maps f : X > {p} and g : {p} + X
such that fg ~ id,,} and gf ~ idx. Then the constant map ¢s(p) =o fsis
homotopic to idx, and hence X is contractible. 0
(12.C.10) Definition. A path connected space X is simply connected if

and only if 7(X,2zo) = {e}.
It is not completely obvious that all contractible spaces are simply con-
nected (especially in view of Example (12.C.8) above). Suppose X is con-
tractible to x9 € X. Then certainly by the previous theorem, every loop in
X based at ro is homotopic to cz, : J + X; however, there is no guarantee
that the base point will not be moved during the homotopy. The simple
connectedness of contractible spaces follows from the next theorem (it also
follows from problem 17 in section C).
(12.C.11) Theorem. If X is a path connected space with the property
that each continuous function f : S' + X can be extended to the disk B?,
then X is simply connected. Conversely, if X is simply connected, then each
continuous function f : S' + X can be extended to the disk B?.
Proof. First note (or proceed to (12.C.13)) that if three sides of J x I are
identified to a point then the resulting quotient space, B, is homeomorphic to
B? (and the fourth side becomes S!). Let h : B + B? be a homeomorphism.
Let xo € X and let g: I > X bea loop based at xo. Define G: I x {0,1}U
{0,1} x I -y X by G(s,0) = g(s) and G(0,t) = G(1,t) = G(s,1) = ao.
Then f = GP-1h-1i : S! + X where i: S! — B? is the inclusion and
P:IxI — B is the quotient map. By hypothesis f has an extension
F : B? - X. Let G be an extension of G to I x I defined by G = F~!hP.
It is easy to verify that e gives a relative homotopy between g and c,,.
Conversely, suppose X is simply connected. Let f : S' + X bea
continuous map. Define f : I + X by f(t) = f((cos2zt, sin 2rt)). Since x
is simply connected, there is a homotopy H : I x I + X such that Ho = f,
Hy = cz, where ro = f(0) = f(1), and H(0,t) = H(1,t) = ao for all t € I.
Define F : B? ~ X by F = HP~'h-!. The reader may check that F' is
well-defined, continuous and an extension of f. O
(12.C.12) Corollary. If X is contractible to xo, then ™(X, 20) = {e}.

Proof. Let H : X x I > X be the contraction homotopy, where Ho =
idx and H, = c;,. Suppose f : S' — X is continuous. It suffices to find
an extension of f to B?. Define G: S' x I > X by G(s,t) = H(f(s),t).
Let B be the space obtained by identifying the subset S! x {1} of S! x I
to a point, and note that B is homeomorphic with B?. Let h : B’ > B be
a homeomorphism which is the “identity” on S'. Let P: S' x I > B be
the quotient map. Extend f to F : B? + X by defining F(x) = GP~th(z).
Observe that G is well-defined, since H collapses X x {1} to zp. Thus, by

(12.C.11), X is simply connected. O
(12.C.13) Exercise. Show that if three sides of J x I are identified to a
point, the resulting quotient space is homeomorphic to oe
We now give an algebraic proof of the no retraction theorem (5.C.21).
(12.C.14) Theorem. Suppose S is a simple closed curve lying in a con-
tractible subset X of €2. Then there is no retraction of X onto S.
Proof. Let s € S and suppose that r : (X,s) — (S,s) is a retraction. Let
i: (S,s) + (X,s) be the inclusion map. Then ri = ids, and consequently
Tix = idz,(s,s). However, since X is contractible, 7 (X5:5):=e}s Thusi ry
maps the trivial group onto the infinite cyclic group 7(S,s) {see (13.B.3)],
an obvious contradiction. O
(12.C.15) Corollary. There is no retraction of a disk onto its boundary.
(12.C.16) Corollary (Brouwer Fixed Point Theorem). Each contin-
uous map f : B? — B? of B? into itself has a fixed point.
Proof. Suppose f has no fixed point; hence, f(z) # z for each x € B?.
A retraction r of B? onto its boundary is constructed as indicated in the
figure. Each point z in B? is projected onto the point in S! obtained by
intersecting S! and the ray from f(x) through x (why is r continuous?).
B?
This of course contradicts (12.C.15). O

Suppose f : (X,zo) — (Y,yo) is a continuous function. Although the
homomorphism f, : 71(X,z0) > 7(Y, yo) is induced by f, it does not fol-
low that f, necessarily retains all of f’s characteristics. For instance, even
though the inclusion map f : S! - E? is one to one, f, is the trivial map.
Furthermore, for any onto map f : I > S!, the corresponding homomor-
phism f, fails to be surjective. Nevertheless, we will see that f and f, are
more closely related in some situations.
(12.C.17) Theorem. Suppose A is a retract of a space X and that a € A.

Let r : (X,a) — (A,a) be aretraction andi: (A,a) — (X,a) be the inclusion
map. Then r, is onto and 7, is one to one.
Proof. Note that ri = ida. Consequently, by (12.C.1), we have that
Txt» = td,,(A,a), and hence i, must by one to one andr, onto. O
Categories and Functors /
297
(12.C.18) Definition. A subspace A of a space X is a deformation retract

of X if and only if there is a retraction r: X — A and a relative homotopy
H:X xI-—X such that Ho = id, H, = ir, and H(a,t) = a for each a € A
and t € I (where i: A -> X is the inclusion map).
Note: (i) A is a deformation retract of X if and only if (A, A) is homo-
topically equivalent to (X, A). (ii) a point zo in a space X is a deformation
retract of X if and only if X is contractible to zo (mod 29).
The following theorem gives an important property of deformation re-
tracts.
(12.C.19) Theorem. If A is a deformation retract of X and zo € A, then

™(X,2o) is isomorphic to 7 (A, zo).
Proof. Let H be the deformation retraction homotopy given in (12.C.18),
and set r = H,. Then by (12.C.17), r, is a homomorphism onto. Since r is
homotopic to the identity (mod A), we have i,r, = id,,(x,2)) and hence ry
is also one toone. O
Deformation retracts are frequently used when computing fundamen-

tal groups. For instance, it is easy to describe the necessary deformation
retraction in the following examples.
(12.C.20) Examples.
1. The fundamental group of an annulus is isomorphic to that of a circle.
2. The fundamental group of a Moebius strip is isomorphic to that of a
circle.
(12.C.21) Exercise. Show that a disk D (open or closed) is not homeo-

morphic to D minus its center.
D. CATEGORIES AND FUNCTORS
Homotopy theory represents an example, par excellence, of how one can

pass from one class of mathematical “structures” (topological spaces and
continuous functions) to another (groups and homomorphisms). As we shall
see presently, this is merely a special case of a far broader theory involving
categories (certain objects together with certain maps) and functors (ways
of moving from one category to another).
The reader in his previous studies has most likely encountered a variety
of mathematical systems: vector spaces and linear transformations, groups
and homomorphisms, topological spaces and continuous functions, etc. A
category (to be defined below) may be viewed as a very general system with
just enough properties to encompass all of the above systems (and many
more) under its aegis.
(12.D.1) Definition. A category (0, &) consists of a collection O of sets

called the objects of the category together with a collection ¥ of sets whose
elements are called morphisms (or mappings) with the property that for
each A,B € QD, there is a set Homy(A,B) € §, satisfying the following
properties:
(i) for each triple A,B,C € 0, there is a composite function o which
assigns to every pair of morphisms f € Hom»(A, B) and g € Homy(B,C)
a morphism go f € Hom»(A,C);
(ii) if f € Homg(A,B), g € Homp(B,C), h € Homg(C,D), then
(hog)o f =ho(gof) € Homy(A,
D);
(iii) if A € 9, there is a morphism 14 € Homg(A, A) such that if
f € Homg(A,B) and g € Homg(B,A), then fol, = f and1l4og = g.
The morphism 1, is called the identity function on A.
(12.D.2) Examples.
1. Let & be the class of all topological spaces. For eack A,B € &,
set Home(A,B) = {f : A> B : f is continuous}. If F is the class of all
Homg(A, B) for A, B € &, then (&, ¥) is a category. This is the category of
topological spaces and continuous maps.
2. Let Zp be the class of all pointed topological spaces. For each
(A,a),(B,b) € Tp, let Home, ((A, a), (B,b)) = {f : (A,a) > (B,b) : f is
continuous}. This is the category of pointed topological spaces and pointed
continuous maps.
3. Let ER be the class of all topological pairs, and for each two such
pairs, define Hom, as in the previous two examples. This is the category
of pairs of topological spaces and continuous mappings of pairs.
4. Let @ be the class of all groups, and for each G,H € ©, define
Hom@(G, H) to be the set of all homomorphisms from G into H.
5. Let G be the class of all topological spaces, and for each A, B € 6,
let Home(A, B) be the set of homeomorphisms from A onto B. Compare
with example 1, and note that in example 5 the map population is so severely
restricted that there results a large number of empty Hom’s.
6. Let II be the class of all topological spaces, and Homy(A, B) be the
set of equivalence classes under homotopy. Then II is called the homotopy
category.
We next describe mappings between categories; such mappings will be

called functors. The mappings must preserve the structure, i.e., due respects
are paid to the identity functions, and composition.
(12.D.3) Definition. Let (€,¥) and (D,) be categories. A covariant
functor F : (€, ¥) + (®,H) consists of a map F from the class € into the
class ® and a map (also denoted by F’) from the class U4 p-¢ Homg(A, B)
to the class Ucpes Homg(C, D) such that
(i) if f € Home(A,
B), then F(f) € Homa(F(A), F(B));
(4) F(a) = Lea);
(vii) F( fog) = F(f) o F(g).
The Seifert-Van Kampen Theorem 299
We have a contravariant functor F, if conditions (i), (ii), and (iii) are

replaced by
(i) if f € Home(A,B),then F(f) € Homa (F(B), F(A):

(it') F(1a) = lea);
(itt') F(f og) = F(g) 0 F(f).
(12.D.4) Example. Let (&,%) be the category of topological spaces and
continuous functions, and let (9, 9t) denote the category of sets and set
functions. Define a covariant functor F : (&,¥) 4 (MR, Mt) as follows. For
each topological space (X,U), let F(X,U) = X, and for each continuous
function f : (X,U) — (Y,V), define F(f) to be the set function f : X > Y.
We call F the forgetful functor.
(12.D.5) Exercise. Express the relationship between pointed topological

spaces and their fundamental groups in terms of categories and functors.
Categories and functors form the essential vocabulary for the study of
homology theory (which is treated in courses in algebraic topology). We will
find this terminology useful in Section F as well as in Chapter 14.
In Section B, we saw that the homotopy functor sends finite products
of topological spaces to direct products of groups. In Section F, we shall
see that direct limits are also fairly well behaved. It will also be shown that
wedge products of topological spaces (disjoint unions of pointed spaces with
the base points identified) are sent to free products of groups.
One of the advantages of studying categories and functors is that one
can recognize the similarity of structures in the different categories, e.g.,
products of topological spaces vs. direct products of groups, wedge products
of topological spaces vs. free products of groups, etc. Once the corresponding
structures are identified, then it becomes natural to ask which structures are
preserved by a given functor. Thus, category theory may be employed to
help bring into focus a broad range of natural questions concerning functors
between pairs of categories.
E. THE SEIFERT-VAN KAMPEN THEOREM
Before reading this section, the reader should be (or become) familiar with
Sections A and B of the Appendix. Difficulties encountered in calculating
fundamental groups soon converge on the insurmountable. However, some
relief is afforded by the Seifert-van Kampen Theorem, two versions of which
are given below. These results were established independently by Seifert
[1931] and van Kampen [1933] and were generalized by Olum [1958]. The
main emphasis of this section is to make the statement of the theorem plau-
sible, and to show how the theorem may be applied. The proof itself is
omitted.
We start with a nonrigorous calculation of the fundamental group of the

figure eight shown in the next figure.
a B
Loops based at xo are evidently reducible to the form ---*a™ « B” xa? x

8% x--+, where one alternately travels a finite number of times around a@ and
than a finite number of times around B (a” = axax---*a). Consequently,
it is reasonable to conclude that 71(X, 20) is a free group on two generators
(or, equivalently, the free product of two infinite cyclic groups), and such is
indeed the case.
A somewhat more sophisticated point of view is needed for attacking
problems involving the fundamental groups of more complicated spaces such
as the Klein bottle, the projective plane, complements of knots, etc. A
favorite tactic used in obtaining the fundamental group of a space X is to
split X into appropriate pieces whose fundamental groups are known, and
then to exploit this knowledge to find the fundamental group of the original
space X. Suppose X = X, U X2, where X, and X2 are open subsets of X.
Suppose further that X, X1, X2 and Xo = Xi M X2 # @ are path connected
and Zo € Xo. Let Go = (Xo, Zo), Gio 7™1(X1, Zo), Gas m1 (Xe, 20), and
G = 7™(X,2o). The homomorphisms given in the following commutative
diagram are those induced by the inclusion maps between the appropriate
spaces.
To achieve a first approximation of 7(X,z09), one might imitate the ar-

gument used in predicting the fundamental group of the figure eight and
conclude that G is the free product of G,, and Gz. Complications arise, how-
ever, since the loop @ in Xo (in the preceding figure) is essentially counted
twice in the free product, once as g; = 9;([a]) and again as g2 = 02([a}).
This problem is overcome by introducing relations in G; * G2 of the form
9; ({a])2([a])~? for each [a] € 7 (Xo, 20).
How are loops in X accounted for that do not lie completely in either
X, of X2: for instance, the loop a in the next figure?
The basic procedure followed is to break a into paths a;, each of which
lies entirely in Xo, X, or Xo.
Since Xo is path connected, for each 7 a path @; may be constructed in Xo

from Zo to p;. Note that a ~ (a1 *B7)*(31 *a2* 33) *(B2*a3*
G5) *-- -*(Bg*a7).
Thus, G may be viewed as being generated by loops, each of which lies
entirely in either X, or Xo.
With all this in mind, it should seem plausible that G is the “amalga-
mated” product of G; and Go, i.e., the free product of G; and G2 modulo
the relations mentioned previously. More precisely, suppose that {A; Ra}
and {B; Rg} are presentations of G; and G2, respectively. Then we have
G = {A,B;Ra, Re, 91([a])02([o])~* where [a] € Go}. The following is the
classic formulation of the Seifert-van Kampen theorem.
(12.E.1) Theorem (Seifert-van Kampen). Suppose X = X; U X2 isa

path connected space, where X; and X2 are gpen, path connected subsets
such that Xo = X;M X2 is nonempty and path connected. Let ro € Xo.
Further, let 6; : 7 (Xo0,20) 7 ™(X1, 20) and 62 : ™(Xo0,2%0) 7 ™1(X2, 20)
be the homomorphisms induced by the respective inclusion maps. Suppose
{A;Ra} and {B; Rpg} are presentations of 7(X1,29) and 7(X2, 20), re-
spectively. Then {A, B: Ra, Rp, 61([a])@2([a])~! where [a] € 71 (Xo, 20)}
is a presentation for 7(X, Xo).
Note that in the special case where Xo is simply connected, G is simply

the free product of the groups G; and G2. Suppose X is the figure eight
described earlier. To see that the fundamental group of X is in fact a free
group on two generators, first write X as the union of the path connected
open sets
Xy X2
and then observe that 71(X1,20) = Z, ™(X2,20) = Z, and by (12.C.19)

we have 7(X1 M X2,%0) = {e}. Consequently, m1(X1,2%0) and 71(X2, 20)
each have presentations of the form {a; }. Thus, by the Seifert-van Kampen
theorem, 7(X) is a free group on two generators.
The fundamental group of S” (n > 1) is also easily calculated. Note that
S” can be written as the union of two overlapping (open) “hemispheres,” each
of which has trivial fundamental groups (the hemispheres are contractible),
and hence 71(S”) = {e}.
There is a more modern, but less transparent statement of the Seifert-

van Kampen Theorem that has the virtue of frequently being easier to apply
than its classical counterpart. A proof is not given here, but the interested
reader may find a detailed demonstration in either Massey [1967] or Crowell
and Fox [1963].
(12.E.2) Theorem. Suppose X = X, UX» and 0 # Xp = X1N Xo where

Xo, X; and X2 are path connected open subsets of the path connected space
X. Let x € Xo. As before, there is a commutative diagram
where G = 7(X, 20), the G;’s are the respective fundamental groups of the
X;’s, and the indicated homomorphisms are those induced by the inclusion
mappings.
Then
(i) wi (G1) Uwe(G2) generates G, and
(ii) If H is an arbitrary group and y; : G; ~ H, i = 0, 1, 2 are homo-
morphisms satisfying W = ~16, = W262, then there is a unique homomor-
phism A: G > H such that w = Aw;, 1 = 0, 1, 2.
There are a number of corollaries that follow rather easily from the main
result.
(12.E.3) Corollary. If X; and X>2 are simply connected, then so is X.
Proof. This is immediate from (7). O
(12.E.4) Corollary. If Xo is simply connected and G; and Gy» are free

groups with free bases A = {2,22,...} and B = {y1,y2...}, respectively,
then G is a free group with basis {w1(r1),wi(Z2),...,we(y1), w2(y2),.--}-
Proof. Let H be a free group with basis {c1,c2,...,d,d2,...}. Define

yy, :A—> A by vi(x;) = G, 1 = 1,2,..., and #2: B A by yo(y;) = d;,
j = 1,2,.... Extend y homomorphically to G; and 2 to G2.
Let Wo : Go H be the only homomorphism possible (the trivial one)
and note that wo = 716, = W262. By (12.E.2), there is a homomorphism
\:G— H such that y; = Aw;, i = 0,1, 2. We show that 4 is an isomorphism
by exhibiting an inverse. It follows from the “freeness” of H that there is
a homomorphism p : H — G such that p(c;) = wi(ci), 1 = 1,2,... and
p(dj) = we(yj), 9 = 1,2,.... Then pr(wi(zi)) = pdi(ti) = p(ci) = 1 (xi)
and Ap(c;) = A(wi(zi)) = Yi(zi) = cj. Since the same sort of thing occurs
with w2 and the y; and d;, we have Ap and p) are identity maps. O
(12.E.5) Corollary. Suppose X2 is simply connected. Then
(i) wy is onto, and

(ii) if {a1,22,...} generates Go, then kerw is the consequence of R =
{01 (x1), 01(x2),..-}.-
Therefore, 7; (X) is the quotient group 7 (X1)/(R), where (R) is the normal
subgroup generated by R.
Proof.
(i) This is trivial by part (2) of (12.E.2).
(ii) Since w1(01(x;)) = wo(xi) = w2(2(zi)) = e, we have that R is
contained in kerw,. Suppose z € kerw,;. Let H = Gi/(R), ¥1 : Gi 7 A
be the natural projection, and y2 : G2 > H be the trivial homomorphism.
Observe Wo = 16), and so Wo = W141 = W262 (all of these maps are trivial).
Consequently, by (12.E.2), there is a homomorphism A : G + H such that
w, = Aw. This implies y,(z) = A(wi (z)) =e. Thusze(R). O
We apply (12.E.5) to calculate the fundamental group of the projective

plane P,. The projective plane is the space obtained by identifying diametri-
cally opposite points on the edge of the disk, as indicated in the next figure.
Q@
Q@
As usual, the first step consists in finding suitable open subsets. Let c denote
the center of the disk and let xo be any other point in the interior of the disk.
Let X; = P2 \ {c} and let X2 be the interior of the disk (considered as a
subset of P2). Note that after identification has taken place, a is a circle, and
furthermore there is a deformation retract from X, onto a. Therefore, G; =
™(X 1,20) = Z, and G; is generated by the equivalence class represented by
y = B’ xaxB. The group 7 (X9, xo) is trivial and 7(X1N X2, 20) is infinite
cyclic with generator [5]. Note that @,([5]) = [7]?. Therefore, by (12.E.5),
™(P2,%0) = ™1(X1,20)/K, where K is the normal subgroup generated by
[y]?. Thus 7 (P2, 20) is cyclic of order 2.
Direct Limits 305
F. DIRECT LIMITS
In this section, direct systems and direct limits of topological spaces are
defined and characterized by a “universal mapping property.” The homotopy
functor will transform a direct system of pointed topological spaces into a
direct system of groups (see Section D of the Appendix). Amazingly, (with
some reasonable conditions on the topological spaces) the fundamental group
of the direct limit of the topological spaces is isomorphic to the direct limit of
the fundamental groups. This result will be indispensable when we consider
the examples in section E of chapter 15.
(12.F.1) Notation. Let {X, : a € A} be a family of topological spaces.

We denote the disjoint union of the X, with the disjoint union topology by
t rehie-
(12.F.2) Definition. A direct system of topological spaces (Xq, fag, D)

consists of a directed set D, a family of topological spaces {Xa}aep, anda
family of continuous functions {fag : a,3 € D and a < (} such that
(i) fag: Xa + Xp, whenever a < ,
(ii) for each a € D, fog = id, and
(iii) fa <8 <7, then fay = for fap.
bet = [ee If tr, € X_q and zg € Xz, then declare xz, and zg to be
equivalent (x, ~ xg) if and only if there is ay € D such thata<y,B<¥
and foy(Za) = fa+(za)- _
Denote X/~ by X@ or limgepXq. Then X is called the direct limit
of the direct system (Xa, fag,D). If ta € Xa, we denote P(z,) by [ra],
where P is the quotient map from X to X/ ~.
(12.F.3) Exercise. Show that ~ is an equivalence relation on X.
It should be noted that unlike inverse limits, the direct limit cannot be
empty, provided that any one of the X,’s is nonempty.
(12.F.4) Notation. If {Xn}nen is a family of topological spaces and {f; :

X;, > Xisih}ien is a family of continuous functions, then the direct system
(Xi, fam; N) will often be denoted by (Xj, fi, N) where fnm is defined to be
fm-1°** fntifn-
(12.F.5) Examples.
1. Let X,, = €' for each n € N and define f, : €! — E' by f(r) =0 for
all € €!. Then X@ is a point with the usual topology.
2. Suppose for each n € N, Xn C Xn41 and fy, : Xn 4 Xn4i is the
inclusion map. Then X% is homeomorphic to Y = i aes Xn, where Y is
given the strong topology.
3. Suppose for each n, X, = €” and define f; : €* — Et! by

f(a1,.--,2i) = (a1,.--.,2i,0) (see problem 1 of,section F).
(12.F.6) Theorem. Suppose X° is the direct limit of the direct system
(Xa, fap,D). For each a € D, define fa : Xa + X° by fa = Pia, where
la : Xa > UeepXa is the inclusion map and P : UaepXa 7 X~ is the
quotient map.
5 eee U6
aé€D
a
xX@&
Then
(i) fa is continuous,
(ii) for each a, 8 € D witha < £ the following diagram is commutative,
xX”
f
fi
Xo faa Xp
(itt) XC Usep falXa):

Proof.
(i) This is trivial.
(17) Suppose that rq € Xq and let rg = fag(tqa). Hence rq ~ zg since
fop(tp) = fop(ta). Then fafap(ta) = Pigfap(ta) = P(xg) =
P(fa) = falta).
(i). "Thisisiclear, since: Pi, (ra). falta). Gl
(12.F.7) Theorem. Let (Xa, fag,D) be a direct system of topological

spaces. Suppose Y is a topological space and {ga : Xz, 7 Y:a€ D}
is a family of continuous functions such that Y = U,ep ga(Xa) and the
following diagram is commutative whenever a, 3 € D anda < £.
Direct Limits 307
Then there is a unique continuous onto function h : X° — Y so that the

tetrahedron is commutative for all a < .
ow
co
pen Ga Y
fap ha.
Xe
Proof. It is clear how h must be defined: Suppose zx € (Uae pXa) i)~

= X~, and z, is an element in Nee such that [rq] = «x. If the following
diagram is to be commutative,
xo E Y
DEE
there is no choice but to set h(x) = ga(xq). Hence, h (if it is a function)

is unique. Furthermore, it is easy to show that if h is a map, then the
tetrahedron is‘commutative.
To see that h is well-defined, suppose tg € Xq and xg € Xz have the
property that [z.] = [zg]. Then there is a y such that a < y, 6B < 7, and
fay(@a) = faq(aa) = 27. We show that ga(za) = ga(zs) = gy(2y). By the
commutativity of the following diagram, we have ga(®a) = 9y(fay(La)) =
g(x), and similarly gg(zg) = 9+(z-).
14
Xo
a fay Xy
That h is onto may be seen as follows: Suppose y € Y. By hypothesis,

there is an a € D and an Zq € Xq such that y = ga(Zq). Then A([ra]) = y.
It remains to show that h is continuous. Suppose U is open in Y. Since
X@ has the quotient topology, h~!(U) will be open in X® if P~*h71(U)
is open in Ulacp
Xa. However, P~'h71(U) = Uaenga'(U) which is clearly
open in GREE oe Oo
The next theorem gives a characterization of direct limits that is trivial

to establish in spite of the rather awesome hypothesis.
(12.F.8) Theorem (Universal Mapping Characterization). Suppose
(Xa, fag, D) is a direct system of topological spaces, Y is a topological space,
{ga : Xy + Y : a € D} is a family of continuous functions such that the
following diagrams are commutative, and Y = Uaepga(Xa)-
VW
bas
Xa ;Jar XB
Suppose further for each topological space Z, for each family of continuous
functions {hg : X_q + Z : a € D} with the property that the following
diagrams are commutative
es
Xo fap Xp
and Z = Ugep ha(Xa), there is a unique continuous surjection dz : Y > Z

for which the following diagrams are commutative.
Then Y is homeomorphic to X%.

Proof. Consider the following diagram:
DEES
h Y oxo h
Direct Limits 309
By the uniqueness of the map in (12.F.7), we have ¢xoh = idyo. From

the uniqueness in the hypothesis, it follows that hdx. = idy. Thus h is a
homeomorphism. O
The reader will want to reread the entire section in order to observe that
nothing goes awry if one assumes that all topological spaces are endowed
with a base point, and that all continuous functions preserve base points. To .
nudge the less than diligent reader, we present the following exercise.
(12.F.9) Exercise. Restate and prove the analog of (12.F.7) for the cate-
gory of pointed topological spaces.
The reader should now check the Appendix and observe that the analog
of each of the preceding theorems also holds in the category of groups and
homomorphisms.
Next we examine what is produced when the homotopy functor operates
on a direct system.
(12.F.10) Theorem. If ((Xa,2a), fag, D) is a direct system of pointed

topological spaces, then (71(Xa, Za); (fag)+, D) is a direct system of groups.
Proof. By (12.C.1) (or, since 7 is a covariant functor), ifa < B <4,
then (fay)x = Usyfaa)e: = (fax)x(foa)x: Furthermore, if a < 0, then
(fap)x : 71 (Xa,La) + ™(Xg,2g) isahomomorphism. O
Now for the jewel in the crown.
(12.F.11) Theorem. Let ((Xa,20), fag,D) be a direct system of topo-

logical spaces. Suppose fa(Xq,£o) is open in (X™, 2) for each a, and the
fag’s are inclusion maps. Then there is an isomorphism h : lim(™ (Xa, Zo) >
m™(X@©, 209). Furthermore, the following diagram is commutative.
h
fim (11 (Xq, 20) ——> ™(X%,Zo)
Ge AY
ne
(ga)» (fp)«
7 (Xa, 20) —G>> (Xp, Zo)
Proof. By the previous theorem, (7 (Xa, £0), (fag)x;D) is a direct sys-

tem. Furthermore, for each a < 3, the following diagram is commutative.
us OG ro)
(fa) (fa).
71 (Xoy20) re ee 7 (Xg, Xo)

Thus, if we can show ™(X, 20) C Ugen(fa)*(™1 (Xa; %0)), then there will
be an epimorphism h : iim (m7 (Xq,20)) 4 ™m (XS, ro) by Theorem (Ap.D.2).
Let [€] € m(X™,20). Since {fa(Xa) : @ € D} forms an open cover
of €(I), there are a finite number of f.(Xq)’s that cover (J). Let 6 be
an index larger than all of the a@’s in the finite cover. Then we have that
€(I) C fg(Xg). Since the fag’s are inclusion maps, fg is one to one. Hence,
we may define a path m = fz 1¢. Then (fg)«[m] equals [é].
We now show h is one to one. Let x € iim(m (Xq,2o)) and A(x) = e;
we show that x = e. Since fim(m7 (Xa, 20)) C Usep(Ga)x
(71 (Xa, Zo)), there
isanaé€ Danday € ™(Xqa,2Zo) such that (ga).(y) = 2.
The diagram
iim (m4 (Xa, 20) > ™ (X®, 20)
(Ga)« (fa)
(Xe, Zo)
is commutative, and therefore (fa).(y) = e. Now let wu: J + Xq be a loop
based at xo such that [y~] = y. Since (fa).(y) = e, there is a homotopy
H:IxTI - (X®,%9) such that Ho = fi. and, = ¢,. Let Bybe
sufficiently large so that Im H C fg(Xg) (and a < 8). As before, fg is one
to one, and hence fz 'H is a homotopy between fag’ and Cz). Therefore,
(fas)«(w) =e. By the commutativity of the following diagram,
fim (77 (Xa, 20)

(ga) (gs).
™1(Xa, Zo) Gear ™ (Xg, 20)
T= (Ga)x(y) = (Ga)x([¥]) = (98)*(fos)«[¥]= (9a)elero]= (ga)x(e)= e,

which completes the proof. O
PROBLEMS
Sections A and B
1. Show: If f,g: X - Y are homotopic and A C_X, then f\4 and g|, are
homotopic.
2. Find an example of a space X containing points xo and 2; such that
7™(X,29) and 71(X,21) are not isomorphic.
Problems 311
3. Show that a continuous function f : X — Y is null homotopic if and

only if f can be extended to the cone on X.
Suppose zy € X and C is the path component of X containing x9. Show
that 7(X,z9) and 7(C,z9) are isomorphic.
Suppose X; and X2 are disjoint topological spaces and X = X, U Xq is

given the disjoint union topology. Show that if z; € X, and zz € Xo,
then 11(X, 21) = 7(X1,21) and ™(X,z2) = 7 (X2,22).
Suppose {Y, : a € A} is a family of topological spaces and let X be an

arbitrary topological space. Suppose further f,g : X — J] achra are
continuous functions and let pg : [],-, Yo — Yg be the projection map.
Show: f is homotopic to g if and only if p,f is homotopic to pag for
all a. (For an elegant approach to the “if” part, use problem 10 from
section A of chapter 6.|
Suppose X and Y are topological spaces and f,g : X — Y are contin-
uous. Show that Y can be embedded in a space Z by a function h so
that hf is homotopic to hg.
Let X and Y be topological spaces and suppose C(X, Y) has the compact-
open topology.
(a) Show that if X is locally compact, then f ~ g if and only if f and

g belong to the same path component of C(X,Y),
(b) If X is arbitrary, does (a) hold?
A path connected space X is 1-simple if and only if every two paths p and
qin X with p(0) = q(0) and p(1) = q(1) induce the same homomorphism
from ™(X,p(0)) + m(X,p(1)), ie. o({a]) = P’][e]le] = [a"Jlellal.
Show that X is 1-simple if and only if 7;(X) is abelian.
10. Find the fundamental group of an open tin can (both ends open).
11. Find the fundamental group of the Moebius strip.
12. Prove or disprove: if f,g: X + Y and f ~ c and g ~ cz where c; and

Cy are constant maps, then f ~ g.
13. Investigate the relationship between the infinite product of topological

spaces and the corresponding product of the fundamental groups.
14. Suppose in Theorem 12.B.2 that k and m are paths from zo to z; and
that k Oy OD,
Show the corresponding isomorphisms ky and my are the
same.
Section C
1. Give an example of an embedding f : (X,20) — (Y,yo) such that f, is

neither 1-1 nor onto.
2. Give an example of a continuous onto map f : (X,zo) > (Y,yo) such
that f, is neither 1-1 nor onto.
Show that retracts of a contractible space are contractible.
Suppose A is a deformation retract of a space X and B is a deformation
retract of a space Y. Show: A x B is a deformation retract of X x Y.
5. Suppose X and Y are topological spaces and f : X — Y is continuous.
The mapping cylinder of f, My is ((X x I)UY)/ ~, where ~ is the
equivalence relation generated by (z,1) ~ f(x) for each x € X, and
COX S \UY is the disjoint union of X x J and Y with the disjoint union
topology (i.e., M; is the adjunction space of X x J and Y via the obvious
function f : X x {1} > Y). Show:
(a) Y and My are homotopically equivalent, and
(b) Y is a deformation retract of My.
6. Prove that the Hilbert cube is contractible. Generalize.
What is the fundamental group of a starfish?
Show: If X is contractible, Y is path connected, and f,g: X > Y are
continuous functions, then f and g are homotopic.
9. Suppose F is a homotopy between maps f,g : X — Y, where X and
Y are path connected spaces. Let rp € X, yo = f(xo) and Jo = g(x).
Show that the following diagram is commutative, where k(t) = F'(zo, t)
and that f, is an isomorphism if and only if g, is.
m(X, 25) —* > mY; 90)

gs ky
™(Y,
Go)
(See the proof of (12.B.2) for the definition of ky.)
10. Show: If A is a deformation retract of X and B is a deformation retract
of A, then B is a deformation retract of X.
11. Show that if A, B, and AN B are nonempty, open, path connected
subsets of AUB, and if A and B are simply connected, then so is AUB.
12. Use problem 11 to show 7™(S?) = {e}.
13. Let n and s be the north and south poles of an n-sphere. Prove {n, s}
is not a retract of S”.
Problems 313
14. Suppose 2p € S'. Show that S! x {zo} is not a deformation retract of

SiS
15. Two continuous functions f,g: I + Y C €? are adjacent in Y if and
only if there is a partition 0 = 29 < 2, < 22 <-:: <2, =10fI such
that f([zi-1, 2i]) U 9([zi-1,xi]) C Dj; for each i, where D; is a disk in
Y. If f and g are adjacent in Y, are f and g homotopic?
16. Do there exist continuous functions.f and g such all of that the following
hold?
(i) fs (S* x {0})

U({zo} x 1) U(S? x {1}) 3S? x J,
(it) g: (S' x I) > (S! x {0}) U ({z0} x I) U(S! x {1}), and
(35) OF == 1a.
Pi Suppose X and Y are topological spaces and f : X — Y is a homotopy
equivalence with homotopy inverse g: Y > X. Let zp € X. Show that
m™(X,Zo) is isomorphic with 7(Y, f(zo)). [Warning: While gf ~ id,
the homotopy need not keep Zo fixed.]
18. Suppose T is a solid triangle in €? with boundary S. Show that there
is no continuous function f : T — S which maps each edge of T into
itself.
19. Prove that a path connected space X is simply connected if and only if
Ca for each pair of paths a and @ in X such that a(0) = 6(0) and
a(1) = B(1).
20. Justify the claims made in the notes after (12.C.18).
Section D
1. Find (somewhere in mathematics) an example of a contravariant functor.
2. (The Higher Homotopy Groups) Suppose X is a topological space. Let
OI” = {(a1,...,2n) € I” : for somei, 2; = 0 or 1}, and let Qn(X, 20) =
C((I", 01”), (X, z0)). If T58 € Qn(X, xo), define
i (2riya55. ve ay) Ty
Che Oar. e.' iia) =
g(22, mam Ml fei oie an)
NIFTA
© A. Ty IA
IA Nile
a
Partition 2,(X,z0) into equivalence classes via homotopy relative to

OI". Denote the set of equivalence classes by 7,(X, 2). Finally, if [f]
and [g] € ™(X, 20), define [f] - [9] = [f * 9].
(a) Show that - is well defined and that 7,(X, zo) with this operation
is a group.
(b) Prove the analog of (12.C.1).
(c) Show that for each n > 1, the construction above gives rise to
a functor from Zp to the category of abelian groups and homo-
morphisms. (It should be no surprise that this is called the n-th
homotopy functor.)
(d) Show that ifn > 1, then t,(X xY, (xo, yo)) & t(X, Lo) Xtn(Y, yo).
(e) Show that if (X,2o) and (Y, yo) are homotopically equivalent, then
T(X, Zo) and 7,(Y, yo) are isomorphic.
Remark: In general, it is exceedingly difficult to compute the higher homo-
topy groups. In fact, their computation even for n-spheres is still not fully
done. Some of the known results leave one gasping, e.g., 715(S°) = Z x Z120
and 716(S®) = Z x Z2 x Zo X Ze (Zn is the group of integers mod n).
3. Suppose (€, #) is a category, f € Home(A, B), and g € Home(B, A).
Then f is a right inverse of g if and only if go f = 7zd,, in which case g
is a called a left inverse of f.
(a) Show that in the category of sets (hence also in the category of
topological spaces and continuous functions), if f has a left inverse,
then f is 1-1, and if f has a right inverse, then f is onto.
A map f is called a €-equivalence if and only if f has both a left and a right
inverse.
(b) Show that if f has a left inverse g and a right inverse h, then g = h.
(c) Show that a homotopy equivalence is a €-equivalence, where € is
the homotopy category.
4. Define a category as follows: Let 9 = Zt be the class of objects. If
m,n € O, define Homs(m,n) = {(m,a,n) : a is a common divisor of
m and n}. Define (n,b,q) o(m,a,n) = (m,d,q), where d is the greatest
common divisor of a and b. Show that (0,8) is a category, where
§ = {Homp(m,n) : n,m Ee OD}.
Section E
ce
Calculate the fundamental group of the five-petaled rose:
2. Calculate the fundamental group of CY) .

Calculate the fundamental group of a 2-sphere with three holes in it.
4. Calculate the fundamental group of the torus (using the Seifert-van
Kampen Theorem).
Calculate the fundamental group of the Klein bottle.
6. Calculate the fundamental group of €° \ S!.
Problems 315
7. Calculate the fundamental group of a torus with a hole in it (a punctured

inner tube).
8. For each n > 2, show how to construct a space having a cyclic funda-
mental group of order n.
Section F
1. Show that X° in part 3 of (12.F.5) is not Hilbert space.

2. Show: S! cannot be written as a direct limit of subspaces, unless one of
theX, sasS*.
3. Compute ™(X), where xX =CY
YX X):::.
a” Form eZ; let -X,, = [—1,1),; and:for, m.> n; let dian 2 Xp Xn
be defined by ‘@a,(¢,) =,/2" ™ Show that lim(X,,,¢am,;Z") is
homeomorphic to €}.
5. Show that any CW-complex can be considered as a direct limit in a
nontrivial way.
6. Show that any T>) k-space can be considered as a direct limit in a non-
trivial way.
rametb owwara sea
| plas
|
that
ie m4! os it site Aint
eo sha ead4% | haath tro wad

7 Ras=moe fh ay) onl pbc beinote v
— , =i) os of¥, oui ee mney aE.
Sree
= hy ghae tn mt
eo rn et
Cima. "he OC Fone ru ee we gt x
el Aisi, ee nr path
i:
R ash
ott
RaaAt. Sees Carrs, dds
Sof A eae
Pa wl perce
a wt NegalPiet > at bf
oA 5 Sl
Sen
awe nalatoarate
aOR ra made=
ies
ha ox Sis es Saint 16. srk las ed

: 1@ a -astio «Indie © a opea bien Cle
Chapter 13
COVERING SPACES
The beautiful and surprisingly close interplay between algebra and topology
is well illustrated in the theory of covering spaces. This theory is very rem-
iniscent of Galois theory and has numerous applications in both topology
and geometry. In particular, some of the results may be used to obtain an
elegant computation of the fundamental group of S'. Throughout this chap-
ter the reader is especially advised to construct numerous figures illustrating
the definitions as well as the hypotheses and proofs of theorems.
A. THE LIFTING THEOREMS

(13.A.1) Definition. Suppose X and X are connected, locally path con-
nected topological spaces and p: X — X is a continuous map. The pair
(X, p) is a covering space of X if and only if for each x € X, there is a
connected open set U containing z such that
(i) p-*(U) = Uaea, Sa, where the S,’s form a nonempty collection of
pairwise disjoint open sets in x , and
(ii) pls, : Sa + U is a homeomorphism (onto) for each a € Az
The space X is called the base space, and the open set U is called
a canonical neighborhood; p is referred to either as the projection or the
covering map.
(13.A.2) Exercise. a) Show that covering maps are open. b) A map

f : X - Y is a local homeomorphism if and only if for each x € X, there is
an open set V, containing x such that f(V;) is open in Y and the restriction
flv, : Vz — f(Vz) is a homeomorphism. Show that covering maps are local
homeomorphisms.
317
318 13. Covering Spaces
The decision to define covering spaces only in the context of connected

locally path connected spaces is made with some reluctance. Less restrictive
definitions are often given (Spanier {1966]) and a number of the basic theo-
rems are still valid in a more general setting. However, spaces that satisfy
these connectedness conditions yield an elegant theory without requiring a
constant shuffling of hypotheses. This together with blatant expediency is
our justification for confining the presentation to such spaces.
(13.4.3) Example. Suppose X = S! and X = €!. Define p: €! > S! by

p(t) = (cos 2rt, sin 27t). In this case, p wraps €' around S’ an infinite num-
ber of times. In fact, for each n, the interval [n,n + 1] goes once around S!.
Any connected open proper subset of S! serves as a canonical neighborhood.
(13.A.4) Example. The projective plane P); may be defined as the quotient
space obtained by identifying antipodal points of S?. Then (S?,P) is a
covering space of P, where P : S? + Pa is the quotient map.
(13.A.5) Example. Any connected locally path connected space is a cov-

ering space of itself with the identity map serving as the projection.
(13.A.6) Exercise. Suppose (X, p) is a covering space of X and (Y,q) is a

covering space of Y. Find a covering space of X x Y.
(13.A.7) Exercise. Find three distinct covering spaces of the torus (one of
which is the plane).
The following theorem is one of the most useful results in covering space
theory. It states that a path a lying in the base space X may be “lifted in
a unique fashion” to a path in the covering space. Here, “lifted” means that
the projection of the lifted path coincides with a, and “unique” means that
if pa = pa and a(0) = a(0), then a = a.
(13.4.8) Theorem (Unique Path Lifting Theorem). If (X,p) is a
covering space of X, a: I + X is a path in X with initial point zo, and
#9 € p—'(ao), then there is a unique path @: I > X with initial point Zo
and pa@ =a.
Proof. The idea of the proof is quite simple. We subdivide J finely
enough so that each piece of the subdivision is mapped by a into a canonical
neighborhood, and then use the local homeomorphism p to lift the pieces.
Some care is taken in order to ensure that the appropriate endpoint of the
(7 + 1)-st piece matches up properly with a suitable endpoint of the (i)-th
piece.
Cover X with a family U = {U, : x € X} of canonical neighborhoods.
Then a~!(U/) is an open cover of I. Since I is compact, there is a Lebesgue
number y for this cover. Let 0 = to < ti,:-: < tn = 1 be a partition of
I of mesh less than y. The following assertion is proven by induction: For
The Lifting Theorems 319
each 7, 0 <7 < n, there is a unique continuous map 4; : [0, t;] X such
that @;(0) = Zo and pd; = aljo.,;. Obviously, for i = 0, the map Qp exists.
Assume then we have constructed such a map for some i, where i < n. By
the choice of the partition, we know that a((t;,t:41]) is contained in some
U €U. Let S; be the component of p~'(U) that contains 4;(t;). Since
Pj = ps, is an embedding, it follows that p;' : U + S; is a continuous
function. Consequently, the map 4441 ((0, tisi1] > x given by
Gi(t) if0<t<t;
dizi (t) =
a peagy ties tat
is well defined. Continuity of @;4; follows from the map gluing theorem. It
should be clear that G;+4; has all of the requisite properties. The function
Qn is the desired lifting. We leave the uniqueness for the reader. O
Not only may paths be lifted but so may entire homotopies.
(13.A.9) Theorem (Homotopy Lifting Theorem). Suppose (X, p) is

a covering space of X, Y is locally connected, and H: YxI 4 X isa
homotopy. Suppose, further, there is a continuous map f : Y > X such
that pf = Ho. Then there is a unique homotopy H:Y xI-—-X such that
pH =H and Ho =f.
Proof. The proof is based on the observation that for each y € Y,
HA |,,;.1 is essentially a path in X, and hence, by the previous theorem may
be lifted uniquely to a path in x starting at f(y). Suppose y € Y. Let
arte es X be the unique path with the property o,(0)= f(y) and, for each
t € I, po,(t) = H(y,t). Define H: Y x I> X by H(y,t) = 0,(t). Clearly
pH = H; the problem is to show H is continuous.
Suppose y, € Y. We construct a “tube” in Y x J, containing {y,} x I
and on which H is continuous. Let U = {Ua : a € A} be a cover of X by
canonical open sets. For each t € J, let J; be an open interval about ¢ and
let N; be a connected open neighborhood of y, such that H(N; x It) C Ue
for some U, € U. Let {i,,...,
1, } be a finite subcover of the open cover
{I, : t € I}, and let N;,,...,.Nz, represent the corresponding neighborhoods
of yx. Choose a connected open set N containing y, such that N C ()j_, M;.
Let y be a Lebesgue number for {J;,,...,J¢,} and choose a partition
OS ae ai A ee) SO thal ag, oe re fy TO each 2.5, Let
A; = [ai,ai41] and note that for each i, H(N x Aj) is contained in some
canonical neighborhood U,.
Consider f(N x {0}). Since pf = Ho and N is connected, it follows that

there is an a € A such that f(N) is contained in a component of p~'(Ua,)
and H(N x A,) C Ug,. Denote the restriction of p to this component by pa,-
Then po, is an embedding, and hence the map F': N x A; > X, defined by
F(y,t) = pa} Hy, t), is well-defined and continuous.
Select a2 € a € A such that
(i) H(N x Az) C Ug,, and
(ii) F(N x {a;}) is contained in a component of p~'(Ua.).
If pa, denotes the restriction of p to this component, then F’ may be
extended to N x Az by setting F(y,t) = pg, H(y,t) for y € N andt € Ap.
Note that F is well defined at t = a; (why?). This process is repeated for
each A;; the resulting map F: Nx I 7 oe clearly has the property that
F(y,t) = oy(t) for each y € N and t € I. Thus, we have F = hha and
since F' is continuous, so must be As 8
(13.A.10) Exercise. Show that H is unique.

(13.A.11) Exercise. Suppose (X,p) is a covering space of X. Show that
py is 1-1, where p, : ™(X,£) — 11(X,z) is the homomophism induced by
the projection map p.
The next exercise states an amazing result: no matter which point you
take in the base space, you always have the same number of points lying
above it. This result makes possible many of the beautiful theorems that
follow.
(13.A.12) Exercise. Suppose (X,p) is a covering space of X. Show that

for each z,y € X, cardp-1(z) = cardp~‘(y), and that p7!(x) with the
relative topology is discrete; card p~!(z) is called the number of sheets of the
covering space é 5):
B. 7,(S's)=Z (A DIVERSION)
The previous two theorems are employed to confirm that 7,(S!,s) = Z. In
the earlier discussion, we claimed that the integer assigned to a member of
m(S',s) would somehow indicate the number of times a path was wrapped
around S!. Let s = (1,0). Consider a typical loop a based at s. By (13.A.3),
(E1,p) is a covering space of S!, where p(t) = (cos 2rt, sin 27t). By (13.A.8),
o may be lifted uniquely to a path G starting at 0 € €' and for 0 € J,
po(0) = o(0) = s. Furthermore, since pé(1) = o(1) = s, it follows that é(1)
must be an integer. The reader should recognize that this integer reflects
the net number of circuits that the loop o makes around S! as o runs from
0 to 1. A loop o that starts around S! in one direction, but from time to
m7 (S',s)= Z (A Diversion) 321
time reverses itself, is characterized by zigzags traced out by & in €!. This
leads to the following definition.
(13.B.1) Definition. Let ¢ : ™(S',s) + Z be defined by ¢({o]) = 4(1),

where ¢ is the unique lifting of o that starts at 0. The function ¢ is called
the degree map.
(13.B.2) Lemma. The degree map ¢ is well defined.

Proof. Suppose o and 7 are homotopic loops (mod{0,1}) based at s,
with liftings ¢ and 7, respectively. We show that ¢(1) = 7(1).
Let H : I x I +S! be the homotopy satisfying:
(i) Ho = 9;
(ii) Ay =7;
(iii) H(0,t) = H(1,t) =s for eachte I.
By (13.A.9), there is a unique lifted homotopy H : I x I > €! with the
property that pH = H and Ho = 6. By (iii), H(0,t) and H(1,t) € p-*(s)
for each t. However, p~+(s) consists of the integers, which is a discrete space.
Therefore, since H(1,0) = &(1), it follows that H(1,t) = &(1) for all ¢ (since
H({1} x I) is connected). Similarly, H(0,t) = (0) = (0) = 0. Then, since
pHo = Ho = 0 and pH; = H; = 7, it follows from (13.4.9) and (13.A.8)
that Ho = 6 and H, =f. Finally, we have é(1) = H(1,0) = H(1,1) = (1),
so @ is well defined. O
(13.B.3) Theorem. The degree map ¢: 7(S!,s) > Z is an isomorphism.

Proof. First, we show ¢ is a homomorphism. Suppose [go] and [7] are
elements of 7,(S',s) and let m = G(1) and n = 7(1). Define a function
fh Titer € byt (s))et(e)ean, <Dhen. 7(0)= meandet (1) sete:
Furthermore, it is clear that pr(s) = p(7(s) +m) = 7(s), and only slightly
less obvious that p(¢*7) = ox7 (this can be verified directly). Consequently
we have ¢([o][r]) = ¢([o *7]) = (@«7)(1) = 71) = m+n = ¢([o]) + O([7)).
Next we show ¢ is 1-1. Suppose ¢([o]) = 0, ie., ¢(1) = 0. Then G is
actually a loop in €! based at 0. Define a homotopy H : I x I + €* by
H(s,t) = G(s) -(1-—t). It may be routinely verified that pH is a homotopy
(mod {0, 1}) between o and the constant function at s. Hence [a] = e and ¢
is one to one.
Finally, we show that ¢ is onto. Let n be an integer, and ¢(t) = nt for
each t € I. Set o = p@. Clearly, ¢([7]) =a(1) =n. O
C. REGULAR COVERING SPACES
In (13.A.11), we saw that if (X,p) is a covering space of X and zp € X, then

the induced homomorphism p, embeds 7 (X,#) as a subgroup of 7(X, 20),
where £ € p-!(zo). In this section, and in the corresponding problems, we
investigate the nature of p,(7(X,#)) and the topological information that
it provides.
(13.C.1) Theorem. If (X,p) is a covering space of X and x € X, then

C = {p,(m(X,#)) : & € p7'(x0)} forms a complete conjugate class of
subgroups of 71(X, 20) (i.e., if £,£ € p~1 (xo), then there is a g € ™(X, Zo)
such that p,(1(X,@)) = g~!p,(m1(X,#))g, and furthermore, if C € C and
9g € ™(X, 20), then g Cg EC).
Proof. We first show C is a conjugate class of subgroups. Suppose
ps(t1(X,#)) and p,y(m(X,#)) € C. Let tr: I + X be a path such that
7(0) = & and r(1) = &. Then pz is a loop in X based at zo; furthermore,
[pr]~ = [pr”]. We will show that p,(m(X,#)) = g71p,(m1(X,#))g, where
g = [pt]. To see this, observe: if [a] € 7 (X,#), then p,[o] = g~!p,([G})9,
where [8] = [r «a«7"] € ™(X,). This follows, since
9 *p+([6])9 = [pt] [p8] [pr] = [p7]~* [pr][pal[pt]~*

[pT] = px([a)).
Thus p,({a]) € g~!p,
(m1 (X, ))g. Consequently
px(™(X,2)) C 97 pe (mi (X, €))g.

If 7 is replaced by 7”, then the same argument may be used to show that
Px(1(X,#)) C gpx(m(X,#))g~. Hence g~!p,(m(X,
€))g C px(m(X, 4).
Now we show the conjugate class is complete. Suppose p,(71(X,#)) € C
and g = [6] € m(X,2o). The path @ has a unique lifting that starts at @
and ends at some point ¢ in p-'(zo) (of course, may be #). We show
9 'ps(™1(X,£))g = px(t(X,#)) and consequently g~!p,(m(X,#))g € C.
Suppose [a] € m(X,2). Then g~!p,([a])g = px([r” +a *T]) € py(i(X, Z)).
Conversely, if [y] € ™(X,2), let [a] = [r+ y*7"]. It is easy to verify that
Px([7]) =97*px([a])g. 0
(13.C.2) Definition. A covering space (X,p) of X is regular if and only if
there is a point x9 € X such that for some ¢ € p~!(z9), py(™(X,2)) is a
normal subgroup of 71(X, Zo).
The next theorem and exercise show that xo and Zo are not special.
Map Liftings 323
(13.C.3) Theorem. Suppose that (X,p) is a regular covering space of X

with ro € X the special point of definition (13.C.2). If ¢,@ € p~!(ao), then
Px(™1(X, £)) = py (mi (X, 2).
Proof. Without loss of generality, we may assume ¢ is a point of p~! (a9)
for which p,(7(X,#)) is normal. By the previous theorem, there exists a
g € m™(X,2o) such that g~'p,(m(X,4))g = px(™1(X,#)). Since the sub-
group p,(71(X,#)) is normal, the result follows. 0
The previous two theorems, together with (12.B.2), may be used to
obtain the following result.
(13.C.4) Exercise. If (X,p) is a regular covering space of X, then for any
Zo € X and any & € p-'(20), ps(™1(X, £)) is normal.
The following exercise and theorem are frequently useful.
(13.C.5) Exercise. Suppose H is a normal subgroup of a group G and

a,b€G. Then ab € H if and only if ba € H, in which case aH = b71H.
(13.C.6) Theorem. Suppose that (X,p) is a regular covering space of X,
ro € X, £0,£,£ € p-4(z0), a is a path from Zp to #, and @ is a path from
&o to &. If [pa]p,(7(X,@0)) = [p6]p.(m1(X,@o)), then # = @.
Proof. Since [pB]~!{pa] € p,(™(X,Zo)), it follows from the previous
exercise that [pa][p]~! = [pa][pB"] = [(pa)
*(pB")] € px(m(X, £0). Hence,
(pa) x (p3") lifts to a loop. However, by the unique path lifting theorem, this
is impossible unless 7 =Z. O
D. MAP LIFTINGS
Suppose (x ,P) is a covering space of X. We have seen that paths in X can

be lifted uniquely to paths in X. Can we generalize this property? That is,
given f : (Y,yo) > (X, Zo) can we find a lifting f :(Y,y0) > (X,#) such that
pf = f? Remarkably enough, the answer to this query can be formulated
in purely algebraic terms. This is expressed in the following extraordinary
theorem.
(13.D.1) Theorem. Suppose (X,p) is a covering space of X, ro € X,
Ey) € p~'(xo), and Y is a connected and locally path connected space. Let
f :(Y,yo) + (X, x0) be a continuous function. There is a unique continuous
function f : (Y,yo) > (X,@o) such that pf = f if and only if the inclusion
fx (™ i. Yo)) Cc Dx (771 xe £o)) holds.
Proof. Half of the proof is trivial, for if such an ‘4exists, then f = Ppfiand
hence f, = px fx. This latter equations yields f,(77(Y, yo)) C px (11 (X, £o)).
Now the remarkable part: the creation of a function f mapping (Y, yo)
into (X, £0). Suppose y € Y, and let a be any path from yo to y. Then fa
is a path from zo to f(y). By (13.A.8), fa may be lifted uniquely to a path

T, in X with initial point Zp. Define f(y) = Ta (1). We first establish that f
is well defined, i.e., f is not dependent on the path a.
Ta (1)=f(y)
Ta
Suppose / is another path from yo to y. We want T,.(1) = 7¢(1), where

Tg is the lifting of the path f@ with initial point £9. Observe that ax" isa
loop in Y based at yo and consequently, f(a * 8") =a is a loop in X based
at Zo.
By (13.A.8) there is a path 7g,gr that lifts f(a* 8"). We first show that
Taxgr is a loop. Since f(a x 8") is a loop, by hypothesis there is a loop h
based at £9 such that [ph] = [f(a x B")]. Let H: I x I > X be a homotopy
such that Ho = ph, Hy = flax 0), and HCl} x1) = HAO} x1) = 34.
Observe that ph = Ho and hence by the Homotopy Lifting Theorem, H can
be lifted to a homotopy H:IxI-—X such that Ho =hand H, = Tor
(pH, = f(axG")). Furthermore, H(1,t) € p~! (ao), {1} x J is connected, and
H(1,0) = @. Then H(1,1) = @o since p ‘(zo) is discrete. Consequently
Tax” 18 a loop based at Zo.
Map Liftings 325
Next observe, for 0 < t < 1 we have
PToxpr ((2 — t)/2) = f(a * B")((2 — t)/2) = (fax fB")((2 — t)/2)

= fA" (2((2 — t)/2) -1) = fe"(1-2t)
= fB(t) = pra(t).
Thus by the uniqueness of path lifting, we have that if 0 < t < 1, then
Tg(t) = Taxer((2 — t)/2). Similarly, we can show 7.(t) = Toxgr(t/2) for
0<t<1. Hence, 79(1) = Taxgr(1/2) = Ta(1), so f is well-defined.
We now establish the continuity of 2 Suppose y € Y and that U
is a neighborhood of f(y). Let V be a canonical open neighborhood of
pf(y) = f(y), and let V be the component of p~!(V) that contains f(y).
If W =UNV. then ply is an embedding and p(W) is open in X. Claim:
if N is a path connected neighborhood of y such that f(N) C p(W), then
f(N) CWCU.
Suppose z € N. Let a be a path from yo to y. The path fa may
be lifted to a path 7, with initial point £0; of course, f(y) = Ta(1) by the
definition of f. Let @ be a path from y to z contained in N. Then ax (
is a path from yo to z and f(a x* G) may be uniquely lifted to a path Taxg
with initial point Zo. Since (p|w)~1(fB(t)) C W for each t € J, the lifting of
f(ax B) is given by
oe ees 0 t
cohol ShGt <aAVAny oo
The terminal point of Tag is in W, and since f(z) = Taxa(1), the result
follows. QO
E. COVERING MORPHISMS AND TRANSLATIONS
A given space may have a number of distinct coverings (13.A.7). In this sec-
tion, we investigate the possibility of whether there is a connection between
these covering spaces.
(13.E.1) Definition. Suppose (X1,p1) and (Xo,po) are covering spaces of
X. A covering morphism is a continuous function ¢: MenmeXeithat yields
the following commutative diagram.
io Seek
XxX
(13.E.2) Exercise. Show that each covering morphism ¢ is onto. [Hint:

Use (13.4.8) and the fact that all the spaces involved are path connected.]
(13.E.3) Theorem. If (X1,p1) and (Xo, p2) are covering spaces of X and
@: X, — X2 is a covering morphism, then (Xj, ¢) is a covering space of X2.
Proof. Canonical neighborhoods in X» are constructed as follows. Sup-
pose z € X5. Let U; be an open subset of X that contains p2(z) and is
canonical with respect to p;. Let Uz be an open subset of X containing
p2(z) that is canonical with respect to po. Finally, let V be the path compo-
nent of U; N U2 that contains p2(z). We show that the path component W
of p;'(V) that contains z is a canonical open set with respect to ¢.
Let {Sq : a € A} be the set of path components of 6~!(W). We must
show that ¢|;, : Sa — W is a homeomorphism onto. However, this follows
immediately, since py '(V) = ¢~!p5‘(V) and V is canonical with respect to
both p; and pg.
(13.E.4) Definition. Let (X,p) be a covering space of X andh: X 3 Xa

covering morphism. If h is a homeomorphism we call it a covering translation
(or a Deckbewegung).
Covering Morphisms and Translations 327
It is clear that the set of all covering translations form a group, where
the group operation is composition. This group which we denote by D(X ,P);
is called the Covering translation group, or Deckbewegungs group.
Given a regular covering space (Xx, Pp), there are a variety of groups that
may be associated with it. Two such groups include 7 (X, to) /ps(™1(X, £))
(which superficially does not seem to say very much geometrically) and
Dix ,p). This latter group appears to be somewhat removed from any homo-
topy considerations. The next theorem asserts that an amazing coincidence
occurs: these seemingly unrelated groups are isomorphic. This is one of the
most aesthetically pleasing results in mathematics.
(13.E.5) Theorem. If (X,p) is a regular covering space of X, then D(X,p)

is isomorphic to 7(X, 20) /ps(™1(X,£)), where @ € p~!(z9).
Proof. Suppose h € D(X,p). Then we have h(é) = #; € p~!(ao). Let
a be a path from ¢ to £, and define ¢: D(X,p) > ™1(X, to) /ps(1(X, £))
by $(h) = [pa]p,(71(X,#)). That ¢ is well defined is left to the reader as an
easy exercise (recall (13.C.5)).
We now show that ¢ is a homomorphism. Let g;h € D(X,p), #1 = h(&),
£2 = g(£), and £3 = gh(@) = g(#1). Suppose a is a path from ¢ to £; and 8
is a path from @ to £2. Observe that ga is a path from £2 to #3, and hence
3x* ga is a path from Z to 3.
Hence,
$(gh) = [p(B * 9a)|p.(m (X, #)) = [pB * pgalp,(m1(X, €))
= [pB][polp.(m(X,#)) (since pg = p)
= [pB
lps(m1 (X, 4) [palps(m (X, €))
= $(9)(h).
We now show ¢ is one to one: Suppose g,h € D(X,p), o(g) = ¢(h),

h(@) = £1, and g(@) = £2. Let a and £ be paths from ¢ to £; and £2, respec-
tively. Since [pa]lp,(m(X,4#)) = [p6]p.(™m (X,#)), it follows from (13.C.6)
that @, = @. We finish the proof that ¢ is one to one by showing the
following: whenever g,h € D(X ,p) and there is a point £ € X such that
g(@) = h(#), then g =/h.
Let p(£) = zo and note ~ and g(@) = h(£) = 2, are both elements
of p~!(x9). Since (X,p) is regular, we may apply (13.C.4) and (13.C.3) to
obtain p,(71(X,£)) = px(m1(X,#1)). Hence, by (13.D.1) there is a unique
lifting f of p such that f(¢) = #,. Since both g and A are liftings of p with
this property, we have g = h.
Finally we show ¢ is onto: Consider {a]p,(71(X, £)) where [a] € 1 (X, 20).
Let 7 be the lifting of a that starts at ¢, and set ; = 7(1). We wish to
find an h € D(X,p) such that h(@) = @,. Actually, we show: whenever
&, € p~!(zo), there is an h € D(X,p) such that h(@) = #1.
Since (X,p) is regular, it follows that p,y(m(X,£)) = px(m (X, #1)).

By (13.D.1) there are covering morphisms h and k such that the following
diagram is commutative.
(X, Zo)
By uniquenessof map liftings, we have kh = idy and similarly hk = idx.

Hence h € D(X, p).
(13.E.6) Corollary. If (X ,p) is a simply connected covering space of X,

then D(X,p) is isomorphic to 7(X) and the order of 7(X) is the sheet
number of the covering space.
(13.E.7) Corollary of Proof. If (X,p) is a regular covering space with
sheet number y, then D(X,p) and ™(X,20)/py(m1(X,#)) are isomorphic
with a transitive subgroup of the symmetric group S¥.
Universal Covering Spaces 329
F. UNIVERSAL COVERING SPACES
We now turn our attention to “super covering spaces”—covering spaces of a

space X that are covers for all other covering spaces of X.
(13.F.1) Definition. A covering space (X,p) of X is a universal covering
space if and only if for each covering space (X,p) of X, there is a covering
morphism ¢ such that the following diagram is commutative.
bs Si
Note that by (13.E.3), (X,¢) is a covering space of X.

To define a universal covering space scarcely constitutes a guarantee
that any actually exist. The following theorem gives a simple criterion for
establishing that a given covering space is universal. This theorem, together
with (13.F.5), will ensure the existence of universal covering spaces under
fairly general circumstances.
(13.F.2) Theorem. If (X,p) is a covering space ofX and m(X,@o) = {e},

then (X,p) is a universal covering space for X.
Proof. Suppose (Xp) is any covering space of X. Consider the following
diagram
Zh
SRS TD
Since ™(X,@o) = {e}, we have p,(m(X,#0)) C Bx(m(X,#0)) and
hence, by (13.D.1), there is a lifting ¢ of p such that the diagram is commu-
tative. Thus, ¢ is a covering morphism. O
It follows from the previous theorem that any simply connected space
serves as its own universal covering space (where the projection map is the
identity). However, we will presently obtain more substantial results.
(13.F.3) Definition. A path connected space X is semilocally simply con-

nected if and only if X has a basis B consisting of path connected sets with
the property that if z € B € B and a is a loop in B based at z, then a
may be shrunk to z in X, i.e., there is a homotopy F’: I x I > X such that
Fo = a, F, =c,, and F(0,t) = F(1,t) =2 for all ¢ € J.
(13.F.4) Example. If M is an n-manifold, then M is semilocally simply

connected.
The next result again illustrates how intricately interwoven are the al-
gebraic and topological aspects of covering space theory.
(13.F.5) Theorem. Suppose X is a semilocally simply connected (and

hence, a locally path connected and connected) space and zp € X. Then, for
each subgroup G of 71(X,2o), there is a covering space (Xq,p) of X such
that pen wy (Xa, £) + ™(X, 20) is a monomorphism with Im(p,) = G.
Before establishing this theorem, we observe that it implies the existence
of a universal covering space for X. Simply let G = {e}, and apply the
preceding theorem.
Proof of (13.F.5). For clarity we subdivide the proof into the following
parts.
(A) Definition of the set Xq and the function p.

(B) Construction of a family B of subsets of Xq.
(C) Properties of B.
(D) Demonstration that (Xg, p) is a covering space (with basis B).
(E) Demonstration that p,(7(Xq@,
#)) = G.
(A) Let P be the collection of all paths in X with initial point zo. Define
an equivalence relation in P by declaring paths a and £ to be equivalent
(a ~ ) if and only if (i) a(1) = A(1) and (ii) [a * B"] € G. Denote the
set of equivalence classes {(a) : a € P} by Xq and observe that the map
p:Xq— X defined by p((a)) = a(1) is in fact well defined.
(B) For each @ = (a) € Xq and each open set U in X containing p(#), define
Uz to be {(a*8) € Xg : B: I 3 U, B(O) = a(1)}. The set Uz is well
defined since ifa ~ d, then (a*) = (@x 8) (this follows from the fact that
[(a*
B) (ax B)” == [(ax
B) x(8"*a")| = [ax d"]€ G. Since X is semilocally
simply connected, X has a basis S consisting of path connected sets with the
property that if 2 is a loop in U € S that is based at x, then boy Ce where
the homotopy may wander outside of U. Let B = {Us : U € S, @ € Xq,
and p(£) € U}.
(C) First we show: for U ¢€ B fixed, if 2 € Uz,, then Us, = Uz,. Let
£, = (a) and £2 = (axB) where 8B: I > U and (0) = a(1). Suppose that
(axa) € Uz,. Then (axa) = ((a* B) x (B" *o)) € Uz,. Hence we have
Uz, C Ug,. On the other hand, any member ((a * 8) xy) € Us, may be
written (a * (Gx y)), which of course places it in Uz,. Note that it follows
trivially that if £;,£2 € p-'(U), then either Uz, = Uz, or Us invUpned.
Next. we show that if U is open in X, then p~'(U) = U{Uz : p(#) € U}.
It is immediate from the definition of Ug that U{Uzg : p(¢) € U} c p71 (U).
On the other hand, if 2 = (a) € p-1(U) then p(Z) = a(1) € U and hence
2 = (a) = (ax B) € Uz € {Uz : p(£) € U} where @ is the constant path at
a(1).
Finally, we see that B is a basis for a topology on Xg. Suppose that
2 € Us, Ve., where Z = (y), #1 = (v1), and £2 = (y2). Then 2 = (y1 * #1)
and Z = (72 * G2) where 6,(7) C U and (o(I) C V. Consequently, we have
Yo 1% and y > yx Se. Let W € S, where y(1)€ W andW CUNY.
We show that 2 € Wz C Uz, NVz,. If § € Wz, we have that 7 = (yx) where
BI) CW CUNY. Therefore, § = (71 *(81*8)) = (72*(B2*B)) € Us, AVE.
Thus BG is a basis. We endow Xq with the topology induced by B.
(D) Members of the family {U : U € S} serve as canonical neighborhoods.
The continuity of p follows from the properties of B, which may be easily
verified by the reader. We prove the following assertion: if U € S, = (a),
and Uz; C bP then Plu, : Uz; > U is a homeomorphism.
It is easy to show that Plu; is onto. Suppose y € U, and let 6 be a path
in U from p(£) = a(1) to y. Then p((ax B)) = y.
We show that ply; is one to one: Suppose p((a*8) = p((ax*T)); we show
that (a*3) = (ax7). Since p((a*)) = p((axr)), we have B(1) = (1), and
consequently 8x7" is a loop at x in U. We now establish that [(a*8)x (a*
T)"] € G. This is done by showing that (a « 8) x (ax*T)” = (ax8) x(r" xa")
is homotopically equivalent to the constant path c,, : J — X, which implies
that [(a
* B) x (a*T)"]}=e€G.
Since 8x7” is a loop in U, there is a homotopy H : I x I + X such
that:
(i) H(s;0)= (6 #7" )(s) for each s © I.
(ii) Ai(sy l= eyior each sé I.
(ip) (0; t= (1,t) =aidor each 1.6 I,
By reparameterizing, stacking homotopies and using the map gluing theorem,
we get the desired homotopy (as indicated in the figure below) between
ax Bxt" xa” and c,,. Thus p is an injection.
Therefore the path component containing (cz,) is a covering space of X,

which we again call Xq.
(E) Finally, we show that p,(m™m (Xq,£0)) = G, where #9 = (cz). Before
proceeding further, we take advantage of our previous work to glean more
information about certain path liftings. Suppose o is a path in X with initial
point x9. Then of course o may be lifted uniquely to a path in Xg that starts
at (Cz,). Under ordinary circumstances, nothing more could be said about
this path; however, if ¢ : I 3 Xq is defined by setting ¢(t) = (c+), where
o:(s) = o(ts), then it is clear that ¢ must be the unique lifting of o.
Suppose now that [a] € ™ (XG,£o). Then + = pa is a loop based at
Zo and can be lifted as indicated to a map 7. However, a also represents a
lifting of pa,,so a@ = 7, Therefore, (c,,) = a(1) = 71) = (1) = (7) = (pa).
Since cz, and pa are equivalent, it follows that [(pa) x cZ,] € G. However,
ps({a]) = [(pa) * cZ,], and thus p,(mi(Xc, #0) CG.
If [7] € G, then 7 and cz, belong to the same equivalence class (i.e.,
[7 xct,,] € G), and hence (r) = Zo. Therefore, 7 must be a loop based at Zo,
and since p,([*]) = [r], we conclude that G C p,(m(Xa,#0)). O
(13.F.6) Exercise. Write out the details of the homotopy needed in part
(D).
It follows from (13.F.5) and (13.F.2) that universal covering spaces can
be found and identified as universal for a fairly wide range of spaces.
We now turn our attention to three things. The covering space D4 (and
of course the covering map), the base space X, and the group of covering
translations. In some sense, any two of these determines the third.
(1) We saw (13.E.5) that if (X,p) is a regular covering space over X, then
D(X,p)= ™(X)/pam (X).
(2) In the previous theorem (13.F.5) we saw that given a suitably nice space
X and a given normal subgroup G of m(X) (which determines D rather
indirectly), then we can construct X (and p) so that p,(m(X)) =G p]
and D(X,p) ~ ™(X)/pam1(X).
(3) We next show that given a suitably nice space X and an appropriate
family H of homeomorphisms on X, then we can construct a base space
X and a covering map p, so that D(X,p) = H.
(13.F.7) Definition. Suppose X isa topological space. Suppose further
that H C4 XxX =X sie homeomorphism} and H is a group under the
operation of composition. The group H is said to be properly discontinuous
if and only if for each « € X there is an open neighborhood U of x such that
{h(U) : h € H} is a pairwise disjoint family of open sets.
(13.F.8) Definition. Let H be a properly discontinuous group of homeo-

morphisms on X and xz € X. The orbit of z, denoted by Oy, is defined by
Of={h(z) comet}.
(13.F.9) Exercise. Suppose H is a properly discontinuous group of home-
omorphisms on X.
i) Show: for each x € X, card(O;) = card(H)
) Show: for each x € X, O, is a discrete subspace of X.
iii) Show: the only h € H with a fixed point is the identity function.
) Show: {Oz ea ¢} forms a partition, which we will denote by
G H; OLX.
v) Show: for each x € X, H acts asa transitive group of permutations

on O,. (See problem 11 of sections E and F.)
vi) If X is connected and locally path connected, show that X /Gy with
the quotient topology is connected and locally path connected.
vii) If (X,p) is a covering space of X, show that D(X, p) is a properly
discontinuous group.
(13.F.10) Theorem. If X is a connected and locally path connected space,
and H is a properly discontinuous group of homeomorphisms on X, then
(X, p) is a covering space of X/Gy, where p is the quotient map, and X/Gy
has the quotient topology. Furthermore, D(X ,P) = H and (X , Pp) is a regular
covering space.
Proof. Let @ = O; € X/Gy. Pick @ € p~'(#) and let U be an open
neighborhood given by the proper discontinuity of H. Let V C U bea
path connected open neighborhood of ¢. We claim that p(V) is a canonical
neighborhood for ¢. Clearly, p(V) is
connected and & € p(V). It is open since
Pp'(p(V)) = Une MV) is open. Furthermore, p restricted to a component
of p-'(p(V)) is one to one and an open map; hence the restriction of p
to this component is a homeomorphism. The connectivity and local path
connectedness of X/Gy are “ inherited” from X (13.F.9). The remainder of
the assertions are left as an easy, but not trivial exercise. (See problem 4 of
sections Eand F.) O
We conclude this section with examples and exercises that tie together
some of the material in this chapter.
(13.F.11) Example. For each n € Z* there is (up to isomorphism) exactly
one n-sheeted covering space of S!. Since 7,(S1) » {a; }, we know there is
a subgroup of index n in 7(S') for each n. Hence for each n there is an
n-sheeted covering space. In fact
pn : (S',
89) + (S",
80) defined by (cosa, sin a) — (cos na, sinna)
works. Why is this the only such covering space (up to isomorphism)?
Suppose another, (KX,P) exists. Since X is an n-sheeted covering space,
Pom (x ,£o) is a subgroup of order n in {a; Ne but there is only one such
subgroup, and so P,7(X) = pem(S,). Theorem (13.D.1) gives us covering
morphisms ¢ : (X, #9) — (S!,
so) and w : (S!,
so) 3 (X,@o). The reader can
supply the recurring argument which shows ¢w and ~w¢ are identity maps.
(13.F.12) Exercise. Let X = €!. For each n € Z define h, : €1 > €}

by hn(z) = ax+n. Let H = {h, : h € H}. Show that H is a properly
discontinuous group of homeomorphisms, and X/Gy ~ S). [Furthermore,
since 7(€!) = 0, this gives €! as the universal covering space of S?.]
(13.F.13) Example. Find a 3-sheeted covering space of (Yeo ) We
know that there are subgroups of {a,b;} of index three, so (13.F.5) tells
us such covering spaces exist. We will not try to recognize what space the
construction in that theorem gives, but will play a detective game instead.
Over the point zo there are three points such that small neighborhoods of
each of these points look like We . Now all we need to do is connected each
pee =
of the three /4 and the three /_ in a one to one fashion, and do the same for
the r;’s and r_’s. However, we must impose one additional condition: the
resulting snarl must be connected. Below we show one such realization. We
take p to be the “projection” map (even though you might have to carry out
the construction in €* to avoid the illusory crossings).
as =
od i
rn
Find Dix ,p). Problem 3 from section C may be applied to see that the
example pictured is a regular covering space. ,
(13.F.14) Exercise. Let X = €?. For n,m € Z define hnm : €2 — E? by

hanl@,y) =(2 +n,2 +m). Leto = {hyn n,m, e.2). how: ff isa
properly discontinuous group of homeomorphism and X/Gy ~ the torus.
Problems 335
(13.F.15) Exercise. Show that the Klein bottle also appears as the base
space of €?, where H is the group generated by h ft (defined by hy:(xr,y) =
(x +1, —y)) and the hi,s, where hy is defined by hn(z,y) = (x,y +n). Note:
This gives €? as the universal covering space of the Klein bottle.
In the preceding two exercises the elements of H are actually Euclidean

isometries. It is not possible to generate every 2-manifold as the base space of
E* where D(E?,p) consists of Euclidean isometries (even after excluding the
obvious exception of S?). Indeed, the torus is the only compact connected
orientable (see chapter 16) 2-manifold that can be obtained in that way.
However if one starts with the Poincaré upper 1/2-plane model of hy-
berbolic geometry and selects appropriate subgroups of the Mobius trans-
formations (which preserve hyperbolic distance) then one can obtain the
remaining compact orientable 2-manifolds. The generalization of these no-
tions to three and higher dimensions is a powerful tool used in the study of
manifold topology. See Radcliff [1994] and Thurston [1997].
PROBLEMS
Sections A and B
1. Prove or disprove the converse of (13.A.2b), where f is onto.

2. Suppose (X,p) is a covering space of X and Y is connected. Show: if
f,9: (%yo) > (X, 4) are continuous and pf = pg, then f(y) = g(y)
for each y EY.
3. Suppose (X, p) is a covering space of X and f,g: I — X are continuous
functions with the same initial point. Show that if pf ~ pg (mod {0, 1}),
then f ~ g (mod {0,1}).
4. Suppose X and Y are connected and locally path connected. Show: if
X is compact, Y is Ty, and f : X + Y is a local homeomorphism, then
(X, f) is a covering space of Y.
5 Dhow ait : p ;
ay.
Xy X94
v6
is commutative and es 1, Pi) and (Oe2, P2) are covering spaces of X, then
(X1,p) is a covering space of X2.
If X is a compact space, Y is T;, and f : X > Y is a local homeomor-

phism, show that for any point y € Y, card f —l(y) < oo. Furthermore,
if Y is connected and T2, show that f is onto.
Suppose X is a connected and locally path connected space, U and V
are path connected open subsets of X such that X = UUV,UNV
consists of two disjoint open sets A and B, a € A, and b € B. Show:
if a and b can be joined by paths in U and in V, then 7(X,a) is not
trivial. [Hint: Assign Y = U(U x {2n}) U(V x {2n+1}) : n € Z}
the relative topology inherited from X x Z. Construct a space v by
identifying in Y:
(x, 2n) and (x,2n —1) for each z € A
(y, 2n) and (y,2n + 1) for each y € B
Now, define p : X > X by p((z,n)) = x. Then (X,p) is a covering
space of X. If a is a path from a to b in U and @ is a path from b toa
in U, there is a lifting of ax G@ that is not a loop. Conclude that ax @ is
not null homotopic.] Munkres [1975] has used this and another similar
result to obtain a very clever and readable proof of the Jordan Curve
Theorem.
Suppose (X, p) is a covering space of X. Show ®, : X1Go +> Xisa

homeomorphism [i.e., the base space is essentially a quotient space].
Section C
Find a two sheeted covering space of the figure eight.

Show that the covering space found in problem 1 is regular (or else find
another one that is).
Show: (X,p) is a regular covering space of X if and only if whenever
£ is a loop in X based at xo that lifts to a loop at some %o € p~!(z9),
then 2 lifts to a loop at each € € p~'(zo). [This justifies the terminology
‘regular’.|
Suppose (X, p) is a covering space of X and xp € X. Let a and B be
loops based at x9. Show: if a and @ can be lifted to paths based at
Zo with a common endpoint, then a and 3 determine the same coset of
1 (X,20)/ps(™7(X,
£o)).
Suppose (X,p) is a regular covering space of X and ap € X. Show:
(X, p) is an n-sheeted cover of X if and only if the index of p,(™(X, Zo))
in 7(X,2o) is n. [Hint: Use the preceding two problems.|
Problems 337
Section D
1. Show: if (X , p) is a covering space of X and Y is contractible to yo EY,

then any continuous map f : (Y,yo) — (X,2o) can be lifted to a map
f :(¥Y, yo) > (X, £0).
2. A continuous map p: Y - B is called a fibration if p has the homotopy
lifting property described in Section B (therefore, any covering map is
a fibration). Show: if p: Y > B is a fibration, and o : I + B is a path
such that o(0) € p(Y), then o can be lifted tog: I-34 Y.
3. Show: if a fibration p: Y > B has the unique path lifting property,
then for each b € B and for each path a in p~‘(b), a is constant.
Prove the converse of problem 3.
Suppose for each a € A, fa : Xqg — B is a fibration. Show that
To Vipex Ae > Mee, Pasdetned by flrialacs — A(faltack isa
fibration.
6. Show: if f : X > Y is a fibration, then f(Y) is a union of path compo-
nents of B.
7. Suppose (X ; p) is a covering space of X. A section of p is a continuous
map s: X — X such that ps = idx. Show that p has a section if and
only if p,(11(X,£0)) = m1 (X, x0).
8. Suppose (6.4,p) is a covering space of X. Show that p has a section if
and only if there is ag: X — X such that pg ~ id.
Sections E and F
1. if X and X are universal covering spaces of a space X, show X and X

are homeomorphic. [Hint: Use (13.D.1).]
2. Suppose X is simply connected. Show: if (X,p) is a covering space of
X, then p is a homeomorphism.
3. Suppose (X,p) is a covering space of X and m(X) = {e}. Show that
every point z € X has a neighborhood U with the property that i, :
m(U, x) + ™(X, 2) is the trivial map (2, is the homomorphism induced
by the inclusion map).
4. Suppose (X,p) is a covering space of X and #),£2 € p~'(xo). Show
that there exists an h € D(X,p) such that h(£,) = 22 if and only if
Ds(™ (x, £1)) =a Px(™ (X, 2)).
Problems 5-10 yield the Borsuk-Ulam theorem (problem 8) and some of its
consequences.
Suppose q: S! + S! identifies antipodal points. Show that the quotient

space derived from q is S!. Let p; : S! + S* be the quotient map. Show
that if pp : S? + P» is the map described in (13.A.4) and f : S? > S’ is
antipodal (i.e., f(—xz) = —f(z) for all xz € S*), then there is a continuous
map g: P2 > S! such that the following diagram is commutative.
P? Si
Show gx : 7 (P2) > 7™(S?) is trivial.

Suppose a: I > S! C S? is a path and a(0) = —a(1). Show:
(a) (p2)x([a]) # e and (p1).f.([a]) # e, and
(b) (p2)x([a]) € 7 (Pa, a0) and (p1)+fs([a]) € 71(S*, bo), where pra(0) =
ap and p; fa(a) = bo.
Show that the function f hypothesized in problem 5 cannot exist (i.e.,
there does not exist an antipodal function from S? onto S*).
Show: if f : S? + B? is continuous and f(—x) = —f (zx) for each x € S?,
then for some x € S”, f(z) =o.
10. Show that no subspace of €? is homeomorphic to S?.
11, Suppose (6.4,p) is an n-sheeted regular covering space of X and Zp € X.
Let X,, denote the permutation group on n objects and h € D(X,p).
Note that h(p~(zo)) = p~1(zo), and hence this yields a permutation
of n objects. Define ¢: D(X,p) > Xp, where ¢(h) is the permutation
indicated above. Show that ¢ is a 1-1 homomorphism.
12. Find an upper bound for the number of four-sheeted regular covering
spaces of the figure eight. Find as many of them as possible.
13. Suppose X is a semilocally simply connected space such that 7 (X, 20)
is isomorphic to Z. How many distinct covering spaces does X have,
and how does one know they are distinct?
14. Find an example of a semilocally simply connected compact space that
is not locally simply connected.
15. Use (13.F.3) to show that if X is a semilocally simply connected compact
space, then 7;(X) is finitely generated.
16. Show: if (X,p) is a covering space of X and 7 (X) = {e}, then X is
semilocally simply connected.
Problems 339
6S Find a properly discontinuous group of homeomorphisms on S? which

yields S? as a covering space of P?.
18. Find (up to isomorphism) all 4-sheeted covering spaces of the torus.
Prove you have all of them, and that there does not exist a covering
homeomorphism between any two of them.
19" Find a properly discontinuous group of homeomorphisms on €, each of
which is a Euclidean isometry, that yields the dunce cap as a base space.
The dunce cap is the space obtained by identifying the three sides of a
triangle as indicated.
(Note: 1) The dunce cap is not a 2-manifold. 2) You can actually

construct one with a piece of paper.]
AAS
7 _ eh
ry -ene a
ah catedined Aatiienieeeie nea

eae toh. 7:reew Ry ere
i;‘y vi of
foe“3 vo
; ot pitt oneal
6A a
7 Piso itaeast
“8 : evs De pon reas 7

; bug an mitt< een pallet
Uisitoe®ie Felina , ic at es peeroo Sele-

bd ”
Pal ey = 08 feteglie estheey
tom shark
: >. & la ae* Smet cen PG) ae
a ih mo insbared SRG
2708 & 6? i
y'
Guy tytacy i and at Xi
= | oes mA.) 8 EN oi, =

© VID o perl “4
- #1 li oe 0 rr or aa
a : - “<6,
- wh aAe 1A 7
wt 11 :
O femct 0s >Re =)(Sah
; é Cet “tout
3 - on =
: >a8 aa ite
ta : a (eng ox
.
: 2a if}a i‘|- teeta
‘
Chapter 14
SOME ELEMENTS OF
SIMPLICIAL THEORY
When the Jordan curve theorem was established in Chapter 5, we first proved
the result for simple closed curves composed of straight line segments, and
then generalized to arbitrary simple closed curves. This illustrates one of the
basic techniques in topology: theorems are proven for spaces with some kind
of linear structure, and then efforts are made to extend these results to more
general spaces. Frequently, however, this technique leads to more questions
than answers. In this chapter, we will work almost exclusively with spaces
that have or may be given a local linear structure. The advantages inherent
in dealing with such spaces will become apparent as the theory is developed.
A. THE POLYHEDRAL CATEGORY

(14.A.1) Definition. A set of k +1 points {vp,v1,...,ue} in E” are inde-
pendent if and only if the vectors v; — vo, v2 — Vo,--.,Uk — Vo are linearly
independent (in the algebraic sense).
(14.A.2) Exercise. Show that points {vo,v1,..., vx} in €” are independent
if and only if whenever )>*_, ajv; = 0 and S>*_, a; = 0, then a; = 0 for
each 7.
(14.A.3) Example. The points (1,4), (3,0), (4,4) are independent in €?,
but the points (1,4), (2,8), (3,12) are not.
(14.A.4) Definition. Suppose {vo, v1,..., Ux} is a set of independent points
in €”. The set
k k
g= {Sonn ee a yaa
341
342 14. Some Elements of Simplicial Theory
is called the Euclidean k-simplex with vertices vo,v1,...,Uz- In this context,

o is said to be spanned by {vo,v1,... Rie Such simplexes will often be
denoted by o = (vo, --: Uz). The dimension of a, dima, is k.
If {vo,v1,---, Uk} is a set of independent points in €”, the set
k k
{Soameet 20, Sarma}
i=0 i=0
is called a closed k-simplez.

An r-face of o = (vov1 -+- vx) is the Euclidean r-simplex spanned by any
subcollection of {vo,v1,...,vx¢} consisting of r+ 1 vertices. If 7 is a face of
a, we write tT < ao. Note that every simplex is a face of itself. Quite often,
we drop the term Euclidean in referring to (Euclidean) k-simplexes. It is
convenient to define the (—1)-simplex as the empty set and to decree that it
is a face of all simplexes. The closure of (vovi --- uz), denoted (vov1 «++ UR),
is defined to be {7 : T < (uo, U1. Ux)}-
In the problem set for this section, another kind of simplex, the abstract
simplex, will be introduced.
Analytic geometry is all that is needed to verify that:

(i) (vov1) is an open line segment with endpoints vp and v1;
(ii) (vov1v2) is the interior of a triangle with vertices vo, v1, and v2;
(iii) (vov1v2v3) is the interior of a tetrahedron;
(iv) Uf{o : o € (vou,---vg)} is the closed k-simplex whose vertices are
{0,2 5-22 sVkrs
(v) if (vov1 --- Ug) = (wow1--- we), then each v; is equal to some w,,
and conversely;
(vi) diam (vov1 ---vz~) = max{d(vj;,v;) : 0< 1,7 < k}.
Simplices may be used to form a variety of both commonplace and exotic
spaces. The precise manner in which this is done is specified in the following
definition.
(14.A.5) Definition. A simplicial complex K (often referred to simply as a

complex KC) is a finite collection of simplexes lying in some Euclidean space
€” that satisfies the following properties:
(i) ifo € K and7t <o, then 7 € K, and
(ii) ifo,7 € K, andoNr FQ, then o =T.
If a complex K contains an n-simplex, and for all k > n does not contain a
k-simplex, then we say that the dimension of K is n, and write dim K = n.
The Polyhedral Category 343
(14.A.6) Example. In the figures below we illustrate a 2-complex and a

3-complex.
A 2-complex A 3-complex
Although our primary concern is with complexes defined as above, the

reader should be aware that there is a more general definition (to be given
shortly) involving “infinite complexes.” First, however, let us see what is
permitted under the present definition and what is not. The figure on the left
represents a complex, while the other two figures fail to represent complexes.
(14.A.7) Definition. A subcomplez of a complex K is any subset L of K

that is a complex. If L is a subcomplex of K, then the pair (K, L) is called
a simplicial pair.
(14.A.8) Definition. If K is a complex and L is a subset of K. then the

complex generated by L (denoted by L9) is {o € K :there is ar € L and
aay
(14.A.9) Exercise. Suppose L is a subset of a complex K. Show that L9
is a subcomplex of K.
(14.A.10) Definition. Suppose K is an n-dimensional complex. The m-

skeleton of K is the set K,, = {0 € K : dima < m}.
(14.A.11) Exercise. Show that the m-skeleton of a simplicial complex Kv
is a subcomplex.
It should be emphasized that a simplicial complex is merely a collection

of simplexes, and, as such is not a topological space. Nevertheless, a given
complex K in €” determines a subset of €”, namely the union of all simplexes
contained in K. This subset is called the carrier of K and will be denoted by
|K|. Carriers of complexes are frequently referred to as Euclidean polyhedra
or geometric complexes. We will also on occasion consider || when K is not
a simplicial complex, but just a set of simplexes in €". The topology given
to a carrier is the relative topology with respect to €”. Ifo is a simplex, the
boundary of o, denoted by G, is the carrier of the proper faces of o.
(14.A.12) Exercise. Suppose K is a simplicial complex, and A C |i].

Show that A is closed if and only if AM |o9| is closed in |o9| for each o € K.
Hence we may view |K| as a CW-complex.
The next concept is one of the most basic in topology.
(14.A.13) Definition. A topological space X is triangulable if and only
if X is homeomorphic to the carrier of some polyhedron. In this case, the
homeomorphic image of a simplex is called a curvilinear simplez.
(14.A.14) Example.
[ODS
In Chapter 16, a more general notion of simplicial complex and triangu-
lation is employed. The following definitions show in what sense noncompact
spaces may be triangulated.
(14.A.15) Definition. An infinite simplicial complex K is a collection of

simplexes in some Euclidean space €” (or in Hilbert space) that satisfies the
following properties:
(i) ifo € K andr <o, then7r € K;
(ii) ifo,7 € K andonr £9, then o = 7;
(iii) if v is a vertex of K, then v is a face of only finitely many simplexes
of K (locally finite property).
If K is an infinite simplicial complex, then the carrier of K, |K|, is
defined to be U{o : o € K}. The carrier |K| is assigned the relative
topology.
Condition (iii) ensures that the relative topology and the topology ob-
tained by gluing simplexes will coincide. Boundaries, etc., are defined as
before.
(14.A.16) Definition. A topological space X is (infinitely) triangulable if

and only if there is a homeomorphism from |K’| to X, where K is an (infinite)
complex.
Throughout the remainder of this chapter, we work with finite com-

plexes; however, the reader may wish to check whether the definitions and
theorems that appear can be extended to infinite complexes.
The Polyhedral Category 345
(14.A.17) Definition. Let K and L be complexes. A map f : |K| — |L|

(also denoted f : K - L) is simplicial if and only if
(i) whenever (vov1 --- vg) is a simplex in K, then f(vo), f(v1),---, f(vk)
are the vertices (not necessarily distinct) of a simplex in L, and
(ii) f is linear on each simplex, i.e., if c = yee ivi € (UOU1 +++ UR),
then f(x) = ty Af (vi).
Suppose that (AK, LZ) and (M,N) are simplicial pairs, We call a map
f : (K,L) > (M,N) simplicial if and only if f : K > M is simplicial and
f\lz : L— N is simplicial.
(14.A.18) Example. In the figure below let
K = {v1,
v2, U3, V4, (V1 V2), (v1 U3), (v2v3), (vi v4), (v204), (v1 V203) },
f= {v1, v2, U3; (v1 v2), (v1 v3) },
MS {w1, w2, W3; (wi We), (w1w3), (w2ws)},
iNe= {w1, w2, (w 1 w2) }.
Furthermore, let f be determined by

f(vu) =w1
f (v2) = f(v3) = we
f(v4) = ws.
Then f : (K,L) > (M,N) is simplicial.
U3 W2
(14.A.19) Exercise. Show: a) simplicial maps are continuous, and b) if K

and L are complexes, then every map of the vertices of K to the vertices of
L determines at most one simplicial map.
We are now in a position to define the (finite) polyhedral category (PC).

The objects in the category are Euclidean polyhedra and the morphisms are
obtained as follows. If X and Y are polyhedra, then a map f belongs to
Hompc(X,Y) if and only if there are simplicial complexes K and M such
that |K| = X, |M| = Y, and f : |K| > |M| is simplicial. Maps in the
polyhedral category are called piecewise linear.
The difficulty in establishing that the polyhedral category is a category

lies in showing that the composition of two morphisms is a morphism. The
next section will be devoted to building the machinery needed to sketch a
solution to this nontrivial problem.
It should be noted that virtually every question that may be asked in
the topological category may also be asked in the polyhedral category, and
although polyhedra are fairly manageable spaces, this is offset somewhat by
the fact that the allowable mappings are more restricted. Hence a theorem
is not automatically true in the polyhedral category just because it is true
in the category of topological spaces and continuous maps (and conversely).
Perhaps the most fundamental illustration of this problem is the
Hauptvermutung: If X and Y are homeomorphic Euclidean polyhedra,

then there is a piecewise linear homeomorphism between X and Y.
This conjecture and variations of it was settled in the late 1960’s. The 2-
dimensional case was proven by Papakyriapoulos [1943]. The 3-manifold
case was proven by Bing [1954] and Moise [1954]. Milnor [1961] showed fail-
ure for 7-manifolds. The general situation is exposed in the papers of Cas-
son [1967], Sullivan [1967], Lashof, Rothenberg [1968], Armstrong, Cooke,
Rourke [1968], and Kirby, Seibenmann [1969]. Copies of some of those pa-
pers (and others) as well as an extensive introduction can be found in Ranicki
[1996].
This section concludes with a theorem whose proof follows immediately
from (4.B.28).
(14.A.20) Theorem. Any finite simplicial complex is an ANRy.
B. BASIC CONSTRUCTIONS
In this section, we develop a substantial amount of the structure used in the

study of simplicial theory and then indicate how one can show the polyhedral
category is in fact a category.
The reader should be warned that the whole area of simplicial topol-
ogy is permeated with what euphemistically might be termed quasi-proofs.
Often, it is quite easy to visualize pictorially or geometrically what is tran-
spiring, but very difficult, or awkward, to write out the details. We occasion-
ally succumb to the rationalization that “little is gained, and often a great
deal is obscured by detailed proofs” which has earned a certain degree of
respectability in simplicial theory.
Basic Constructions 347
(14.B.1) Definition. Suppose K is a simplicial complex and o € K. The

star of o in K, star(o, K), is defined to be {6 € K : a < G}. The link of
o in K, link(o, K), is defined to be (star(c,K))9 \ M, where M is the set
of simplexes in K that share at least one vertex with o, and (star(o, K))9 is
the subcomplex of K generated by star(o, K). When the complex K is clear
from context, we will often write star(c), link(c) and (star(c))9.
(14.B.2) Examples.
star(o,K)
link(o,K)
(14.B.3) Exercise. Show: if K is a simplicial complex and vp is a vertex

of K, then |star(vo,
K)| is open in |K| and vo is the only vertex of K in
|star(vp, K)|. Furthermore, show that {|star(v,K)| : v € Ko} forms an
irreducible open cover of |K|.
(14.B.4) Definition. Suppose o and 7 are simplexes in €”. We say o and
T are joinable if and only if the vertices of o together with the vertices of r
form an independent set of points. If o and 7 are joinable, their join, o -T,
is the simplex whose vertices are those of o and rT.
(14.B.5) Remark. If 7 is the simplex of dimension —1, then 0-7 = 0 = T-o.
(14.B.6) Example.
eQ9
(14.B.7) Definition. Suppose K and L are complexes in €”. We say K

and L are joinable if and only if
(i) for each o € K and each r € L, o and 7 are joinable, and
(ii) K-L={o-r;0€K, 7 € L} is a complex. We call K - L the join
of K and L.
(14.B.8) Theorem. Suppose K is a complex and o € K. Then

(i) link(o, K) = {7 € K : o and are joinable and 0-7 € K}, and
(ii) (star(o, K))9 = (0)9 - link(o, K).
Proof.
(i) Recall link(o, K) = (star(o, K))9\M, where M is the set of simplexes
in K that share at least one vertex with o. Let A= {7 € K : o and7 are
joinable and a0-7 € K}. If 7 € A, then 7 is a proper face of the simplex a -T.
Since o - 7 lies in star(a, K), it follows that 7 € (star(o, K))9. Furthermore
t ¢ M (if 7 shares a vertex with o, then 7 and a are not joinable). Thus we
have A C link(o, K).
Now suppose 7 € link(a, K), i.e., 7 << s € K, where o < s and o andr
do not share any vertices. Then, since o and 7 are both faces of s, it follows
that o and 7 are joinable, and furthermore, 0-7 < s; therefore,a-7 € K.
(ii) Suppose 7 € (star(o, K))%. We consider two cases.
Case 1. Suppose 7 € star(o,K), i., o < 7. Let y be the simplex
spanned by the vertices of rT that are not vertices of o. Then by part (i),
y € link(o, K) and o-y =7; hence ao -y € 09 - link(o, K).
Case 2. Suppose 7 € (star(o,K))9 \ star(o,K). There is a simplex
y € star(o, K) such that tr< y and o ¢ Tr. Let & be the simplex spanned
by the vertices o shares with 7, and let 7 be the simplex spanned by the
vertices of 7 are not shared with o. Then we have 6-7 = 7, G € o9, and
7 € link(o, K). Therefore, 7 € G - link(o, K) C (c)9 - link(o, K).
Finally, suppose 6 € 09 and 7 € link(o, K) = {r € K : o and 7 are
joinable and 0-7 € K}. Then G and 7 are joinable, 6-7 < o-7f, and
o-T € star(a, K) by definition. Consequently, ¢-7 € (star(o,K))9. O
(14.B.9) Definition. Suppose K and L are simplicial complexes such that

|K| = |L|. Then K is a subdivision of L if and only if ¢ € K implies that
there is ago € L such that 6 Ca.
\ VY
Note that K is a subdivision of L if and only if |K| = |L| and each
simplex in L is the union of simplexes in K.
Given any complex with positive dimension, there are an uncountable

number of possible subdivisions. We will describe two particular types of
subdivision that are frequently used.
(14.B.10) Definition. Suppose K is a simplicial complex and oa is the

simplex (vov1 --- ux) € K, the barycenter of o is the point b, = a Dee Uj.
(14.B.11) Definition. Suppose a is a simplex in a complex K. We sub-
divide a9 as follows. For each face s < a, let b, denote its barycenter.
The complex consisting of all simplexes of the form (b;,,bs,,..-,0s,), where
81 < 82 < ++: < 8s, and each s; < a, is called the barycentric subdivision of
a9. The barycentric subdivision of a complex K is the complex K“) resulting
from barycentrically subdividing each simplex in K
(14.B.12) Example.
GO)
(14.B.13) Definition. IfK is a complex, the second barycentric subdivision

of K, denoted by K), is the complex obtained by barycentrically subdivid-
ing K), In general, the n-th barycentric subdivision of K is (K("-))®) (see
14.B.14).
More generally, if one performs the construction employed in building a
barycentric subdivision, but uses, in place of each b,, an arbitrary point in
s, then a first derived complex, K', is obtained (see 14.B.15).
(14.B.14) Example.
(14.B.15) Example.
(14.B.16) Exercise. Suppose K is a simplicial complex. Let K’' be a

derived complex of K and K“) be the barycentric subdivision of K. Show
that there is a simplicial map ¢: |K'| > |K“)| that is a homeomorphism.
We outline one way of showing the composition of two piecewise linear
maps is piecewise linear. Suppose K, L, M, and N are complexes such that
|L|= |M|. Let f : |K| — |L| and g: |M| —+ |N| be simplicial maps. We
want to find simplicial complexes K and N such that |K| = |K|, |N| = |.N|
and gf :|K| > |N| is simplicial.
We proceed as follows. First show, (14.B.30), that if |Z]= |], then
there are subdivisions L and M of L and M, respectively, such that L=M.
Next, we indicate how one can show: if f : \K| > |L| is simplicial and L’ is a
subdivision of L, then there is a subdivision K’ of K such that f : |K’| > |L’|
is simplicial (14.B.31). Finally, it is proven: if g : |M| — |N| is simplicial
and |M’'| is a subdivision of M, then there is a subdivision N’ of N so that
g:|M'| > |N'| is simplicial (14.B.34).
Note: if o and 7 are simplexes in €”, then although 0 MT may not bea
simplex, it is a convex set (possibly empty). We now consider some special
convex sets.
(14.B.17) Definition. Suppose A C €”. Then the convex hull of A is the

intersection of all convex subsets in €” that contain A.
(14.B.18) Definition. A linear cell is the convex hull C of a finite set of

points S in €”. The minimal subset V of S with convex hull C is called the
set of vertices of C. The dimension of C is the smallest number m such that
V contains m+ 1 independent points. In this case, we say that C is a linear
m-cell.
(14.B.19) Example.
Note that C has five vertices and is of dimension 2.
(14.B.20) Exercise. Show that a linear n-cell C’ is homeomorphic with

|o9| where o is an n-simplex.
(14.B.21) Definition. Suppose C is a linear n-cell with vertex set V. A
subset F of C is an (n — 1)-face of C if and only if
(i) F is a maximal linear (n — 1)-cell with vertex set a subset of V, and
(ii) C \ F is connected.
An (n — 2)-face of C is an (n — 2)-face of an (n — 1)-face of C, and in
general, an (n — k)-face of C' is an (n — k)-face of an (n — k + 1)-face of C.
We assume @) is a (—1)-face of any linear n-cell C, and that C is the only
n-face of C.
The boundary of C, denoted by C, is the union of the (n — 1)-faces of C.
If C is a linear n-cell, n > 1, then the interior of C, denoted by C, is C \ C
If C is a O-cell, then C=C.
(14.B.23) Definition. A linear cell complex K is a finite set of linear cells

such that
(i) if C € K and D is a face of C,, then D € K, and
(ii) ifC, DEK, then CODE K.
(iii) {C : c € K} is a partition of |K|.
Since in this chapter we deal only with linear cells and linear cell complexes,
the modifier “linear” is dropped with the hope that no confusion will arise
later in the book when more general types of cells are considered.
(14.B.24) Example.
A linear 2-cell with four 1-cell faces,

four 0-cell faces, and one (-1)-cell face.
Not a cell complex A cell Complex
(14.B.25) Definition. The carrier |K| of a cell complex K in €” is defined

by |K| = Ucex C. It is assigned the relative topology in €”.
A corollary of our first theorem is that the carrier of a linear cell complex
is a Euclidean polyhedron.
(14.B.26) Definition. Suppose K and L are cell complexes. We say L is

a subdivision of K if and only if
(i) |A| = |Z], and
(ii) for each cell C € K, there are cells D,,...,Dm € L such that
IC] = Uses |Dil-
(14.B.27) Theorem. Suppose K is a cell complex. Then K’ has a subdivi-
sion L that is a collection of closed simplexes whose interiors form a simplicial
complex with the same set of vertices as K.
Proof. Initially let L = {C € K : C is a vertex or C is a linear 1-cell}.
Label the vertices of K by v9,v1,...,Um- For each cell C; of K, let vc, be
the vertex in C; with smallest index.
For each 2-cell C, we have |C| = |vc - Di|, where D, is the collection
of 0- and 1-faces of C that do not contain vc. Add vc: D, to L. For each
3-cell C € K, we have |C| = |vc- D2|, where Dz is the collection of members
of £ that lie in C and do not contain vc. Add vc: Dz to L.
We continue the construction until all the cells of K have been subdi-
vided into closed simplexes. O
(14.B.28) Example.
(14.B.29) Corollary. If K is a cell complex, then |K| is a Euclidean

polyhedron. O
(14.B.30) Corollary. Suppose K and L are simplicial complexes such

that |K| = |L|. Then there is a simplicial complex M that is a common
subdivision of K and L.
Proof. Observe that the intersection of two closed simplexes is a linear
cell. If N = {|o9|N|79| : o € K, 7 € L}, then N is a cell complex.
Furthermore, N is a cell subdivision of the céll complexes that corresponds
in a natural way to the simplicial complexes K and L. We apply (14.B.27) to
N and obtain a collection of closed simplexes whose interiors form a simplicial
complex M that will be a subdivision of both K and L. O
(14.B.31) Theorem. Suppose K and L are simplicial complexes and that

f : K — Lis simplicial. Let M be a simplicial subdivision of L. Then there
is a simplicial subdivision N of K such that f : N > M is simplicial.
Sketch of Proof. Since f is simplicial, it follows that if o € K, then
f(o) € L. Let K = {o9|N f-(\r9|) : 0 € Ky re M}. If K is a cell
complex, then it is clearly a subdivision of the cell complex associated with
K, and an application of (14.B.27) yields a simplicial subdivision of K on
which f is simplicial.
That K is a cell complex follows from the three following assertions,
which are more or less visually clear, but technically gruesome:
(i) |o9| NM f-*(\79]) is a linear cell;
(ii) faces of |o9| NM f—+(|r9|) are obtained from |s9| NM f71(|u9|), where
s<o-and pi < 73
(iii) (lo9| 9 F~*([791)) 9 (WH 1 F-*(l991)) = (la NwI2) N F-* (I7.9.419) is
a face of both |a9|N f—}(|79|) and |u9|N f-'(\79|). Oo
(14.B.32) Exercise. Suppose X, Y, and Z are polyhedra. Show: if f €
Hompc(X,Y) and g € Hompa(Y, Z) andg is onto, then gf € Hompc(X, Z).
If g is not onto, there is still one complication in showing that the

composition of piecewise linear maps is piecewise linear. This difficulty is
overcome by showing that whenever K is a subcomplex of L and K isa
subdivision of K, there is a subdivision of L that contains K asa subcomplex.
(14.B.33) Theorem. Suppose K is a subcomplex of a simplicial complex L

and K is a subdivision of K. Then there is a subdivision of L that contains
K as a subcomplex.
Proof. Let F,={o€L:o0€ K}, h={oeEL: 09 CL\K}, andin
addition let F; = {0 € L : o ¢ K and there is at < o such that 7 € K}.
We construct the desired subdivision of DL as follows.
(i) For each o € Fj, use the subdivision induced by K.
(ii) Leave each o € F» alone.
(iii) Order the simplexes in F3 so that if 7<j, then dimo; < dimo;.
For each simplex o;, pick a point vu; € a;. Now we subdivide o, by taking
v, ‘By, where By = {711,712,--+, Tin;} is the set of simplexes obtained in (i)
and (ii) which satisfy |71;| C |o1|. Next subdivide o2 by taking v2 - Bo, where
Bz = {721,..-,T2n.}, and each |79;| is a subset of |o3| and 72; is a simplex
obtained from either (i) or (ii) or is a simplex obtained from the previous
subdivision of 0;. Subdivide a2 by taking v2 - Bg. This process is repeated
inductively to obtain L. O
(14.B.34) Corollary. If K and L are simplicial complexes, f : K — L is

simplicial, and K is a subdivision of K, then there is a subdivision L of L
and a subdivision K of K such that f : K + Lis simplicial.
Proof. Note that f(K) yields a subdivision of a subcomplex of L. 0
(14.B.35) Theorem. Suppose that X, Y and Z are Euclidean polyhedra.
If f € Hompo (X,Y) and g € Hompc(Y, Z), then gf € Hompo(X, Z).
Proof. Let K, L, M, and N be simplicial complexes such that |K| = X
(Lk=(M| = Y, (N= "2,"and such that, {: K — Land g; M —_N are
simplicial. By (14.B.30), L and M have a common subdivision P. Observe
there is a subcomplex Q of N such that |Q|= g(|M|). By (14.B.32) there
is a subdivision K of K and a subdivision Q of Q such that 9.fee Ke-3-0
is simplicial. Finally apply (14.B.33) to obtain a subdivision N of N that
contains Q as a subcomplex, and gf : K > N is still simplicial. 0
C. SIMPLICIAL APPROXIMATION THEOREMS
The simplicial approximation theorem will show: if one restricts attention to

homotopy relations, continuous functions may be replaced by simplicial func-
tions. With the aid of this theorem, polyhedral structures may be exploited
to yield a number of interesting results. First, however, further definitions
and lemmas are in order.
(14.C.1) Definition. The mesh of a complex is max{diamo : a € K}.

If K is a complex, it would appear, at least geometrically, that the
mesh of K") converges to 0 as n increases. This is a trivial corollary of the
following theorem.
(14.C.2) Theorem. If K is an n-dimensional complex with mesh y, then

mesh K(!) < —8-¥,
Proof. It is clear that the mesh of K“) is the maximum over the lengths
of the 1-simplexes of K). Furthermore it is easy to see this maximum is
acheived for some 1-simplex of K“) one of whose vertices in is K. However,
if o = (vpv1 --- UR) € K, then
k
1
d(bz,0) = ||be — vo|| =
(0) ~% i=0
1 k
1
rei (de ‘ »)]
1=0
=
k+1
diam o.
In this ||v — ul] represents the Euclidean norm, i.e., the distance from v — u
to the origin. O
Simplicial Approximation Theorems 355
(14.C.3) Theorem. Vertices vo, v1,. ) Uk from a complex K are the ver-
tices of a simplex in K if and only ifnes star(v;) # @ (or equivalently,
Nizo Istar(vi)| 4 0).
Proof. Suppose vo,v1,...,Ug are the Rertices of a simplex o. Then
clearly o € star(v;) for each i,me hence o € ae_o Star(v;) and furthermore,
lo] C Mg [star(s
Now suppose z € |star(v;)| for i = 0,1,...,k. Let o be the unique
simplex of kK that contains z. Since o C |star(v;)| for each i, we have that
o € star(v;), and hence for each i, vu; € o. Since any subset of the vertices of
a simplex spans a face of the simplex, it follows that v9, v1,..., vg determines
asimplex of K. O
(14.C.4) Definition. Suppose f : |K| — |L| is an arbitrary map. A

simplicial map ¢: |K| — |L] is a simplicial approximation of f if and only if
for each vertex v in K, f(|star(v)|) C |star(d(v))].
(14.C.5) Remark. Suppose ¢: |K| — |L| is a simplicial approximation of
f : |K| — |L| and mesh{{star(v)} : v € L} =e. Then d(f(zx), d(z)) < € for
each x € |K|.
Before establishing the existence of simplicial approximations, we ex-
hibit a sequence of lemmas that give some connections between simplicial
approximations and homotopy.
(14.C.6) Lemma. Suppose ¢: |K| > |L| is a simplicial approximation to

f : |K|— |L| (f need not be continuous). Then for each x € |K| there is a
7 € L such that f(x) and ¢(z) € |r9|.
Proof. If x € |K|, then x € o” = (vo,U1 ---Un) for some o” € K. For
each @ = 0,1,7).,n, f(x) © flee f((star(v;)|) Cc” |star(¢(v;))|: Thus,
f(z) € Nig |star(¢(vi))| # 0. It follows from (14.C.3) that {¢(v;)}#_, spans
a simplex v € L. Although f(x) need not be an element of v, we do get
that f(z) is an element of some simplex t € L. Furthermore, for each
i, TM |star(d(v;))| 4 0. Hence, it follows that + € ()j_,star(¢(u)), ie.,
each $(v;) is a vertex of r. Now if x = )>;.,a;v;, then ¢(x) must equal
Dino UP(Vi) € [79]. O
(14.C.7) Lemma. Suppose ¢ : |K| — |L| is a simplicial approximation to
f : |K| > |L| and if for some vertex v of K, f(v) is a vertex of L. Then
f(v) = g(r).
Proof. Recall that |star(f(v))| contains only one vertex (14.B.3).
(14.C.8) Lemma. Suppose ¢ : |K| — |L| is a simplicial approximation to
f :|K|— |L| and f]|, is simplicial for some o € K. Then 4|, = f|c-
Proof. By the preceding lemma, ¢(v;) = f(vi) for each vertex of o.
Suppose Z = )>,, @iv;, where each v; is a vertex of a. Since f is simplicial
on awe have fle) = 507 aif vi) = dp HO)
(14.C.9) Lemma. If ¢: |K| — |L| is a simplicial approximation to a

continuous function f : |K| > |L| and if f||,4; is simplicial where M is a
subcomplex of K, then f ~ ¢ (mod|M)]).
Proof. Let F : |K| x I — |L| be defined by F(z,t) = to(z)+ (1—-t)f (a).
This makes sense, since if x € |K|, then ¢(x) and f(z) lie in the same closed
simplex of L. Furthermore, by (14.C.8), ¢(z) = f(x) where f is simplicial.
Hence, z € |M| implies F(z,t) = t@(z) + (1-t)¢(z) = ¢(a) = f(x). O
This last lemma suggests why it is worthwhile to seek simplicial approx-
imations. Maps that can be simplicially approximated may be represented
in the homotopy category by simplicial maps. Some consequences of this will
be given in the corollaries of the simplicial approximation theorem.
(14.C.10) Example. The following continuous function f cannot be sim-

plicially approximated.
v2 W3
K| - > Ec
V1 Wi w2
Here, the function f maps K onto the perimeter of the triangle w;w2w 3 and
sends both v; and v2 to w;. The problem is that K does not have enough
vertices for f to be simplicially approximated.
(14.C.11) Theorem (Simplicial Approximation Theorem I). Suppose

that f : |K| — |L| is continuous. Then there is an integer n such that
f :|K| = |L| may be simplicially approximated.
Proof. It follows from (14.B.3) that for each vertex v € L, |star(v)| is
open in L. Thus, U = {f~(star(v)|) : v a vertex of L} is an open covering
of the compact space |K|. Let 6 be a Lebesgue number for UW and choose an
integer n large enough so that mesh K(") < 6/2 (14.C.2). For each vertex
w € K ), there is a vertex v € L, such that f(|star(w)|) C |star(v)|.
We define a simplicial approximation ¢ to f : |K(")| > |L| as follows.
If w is a vertex in K'”), let ¢(w) be any vertex in L with the property
that f(|star(w)|) C |star(¢(w))|. The map ¢ has the following property: If
Wo, W1,-..,W, are the vertices of a simplex in K ”), then the collection of
vertices V = {6(wo), d(w1i),.-., (wr) } are the vertices of a simplex in L (of
course, some of the vertices in V may be listed more than once). To see this,
note that ();_, |star(w;)| #0, and that f(|star(w;)|) C |star(d(w;))| for each
i. Thus, ();_, star(¢(w;)) # 0, and hence, by (14.C.3), the vertices in V are
vertices of some simplex in L.
The map ¢ may now be extended linearly to the interiors of simplexes in

kK (first to the 1-simplexes, then the 2-simplexes, etc.), which completes
the proof. O
The previous theorem and (14.C.9) combine to yield the following result.
(14.C.12) Theorem (Simplicial Approximation Theorem II). If K
and L are complexes and f : |K| — |L| is continuous, then there are an
integer n and a simplicial approximation ¢ to f : |K‘")| — |L| such that
@ ~ f (mod A), where A= {x € |K|: f(z) =¢(z)}. O
The next corollary follows easily from (14.C.12), (14.C.2) and (14.C.4).
(14.C.13) Corollary. Suppose K and L are simplicial complexes, f : |K| >
|L| is continuous, and e > 0. Then there are integers m and n and a simplicial
map ¢:|K‘™| — |L(™| that is homotopic to f (mod.A) and moves points a
distance less than « (where A= {x € |K| : f(z) =¢(x)}). O
(14.C.14) Corollary. Suppose K and L are complexes. Then every ho-
motopy equivalence class of Hom(|K’|,|Z|) can be represented by a simplicial
map. In particular, if f : |K| — |L| is a homeomorphism, then there is a
simplicial approximation ¢ of f such that the isomorphism f, is equal to ¢,
(of course ¢ need not be a simplicial homeomorphism). 0
(14.C.15) Corollary. The fundamental group of S” is trivial for n > 2
(cf. the discussion between (12.E.1) and (12.E.2)).
Proof. Suppose S” is represented by the boundary of an n + 1 simplex,
say o. Let vo be a vertex, and suppose that f : J + || be a loop based
at vo. If |o| \ f(2) 4 9, then it is an exercise in analytic geometry to show
that f ~ cy, mod {0,1}). If f() covers all of |a|, we proceed as follows. By
(14.C.11), I may be subdivided so there is a simplicial map 7¥ : [| — |o|
that is homotopic to f. Furthermore, since f(0) = f(1) = vo, we have by
(14.C.7) that 7(0) = 7(1) = vo. Since (|J™]|) is in the 1-skeleton of |o|, it
follows that |o| \ y(\J°™|) # 0. Another exercise in analytic geometry will
yield y ~ cy, (mod {0,1}). O
(14.C.16) Corollary. For n > 2, every map f : S" + S' is null homotopic.
Proof. Since 7(S") = {e}, f can be lifted to the universal covering
space €! of S! (13.D.1). O
(14.C.17) Exercise. Show that every continuous map f » SF + S" (k <n)
is null homotopic.
(14.C.18) Exercise. Suppose f : S* -» X is null homotopic. Show that f

may be extended to a continuous function f : B‘+! + X.
The advantages inherent in triangulable spaces are well illustrated in
the proof of the following result.
(14.C.19) Theorem. If A is a closed subset of S” and f : A > S” is

continuous, then f has a continuous extension f > 8" > S”.
Proof. Assume S" = o"*1. By (4.B.21), there is an open neighborhood

U of Aand an extension f : U > S” of f. Let 6 = d(A, X\U). Subdivide S”
so the mesh of the triangulation is less than 6/2, and call the corresponding
complex K. Clearly, any simplex intersecting A is contained in U. Let |L|
be the union of all simplexes that are in the closure of a simplex intersecting
A. Then Flic : |L| + |S™| is an extension of f. Extend Flic to |L| U |Ko|
by arbitrarily mapping the points of the 0-skeleton Ko to vertices of K \ L.
Assume f has been extended to f : |L|U|Km| 7 S” and let om4i € K \ L.
Then f ¢ Gmaa 3°S™ and 4 /since by (1400.17) flees is null homotopic,
it follows that f can be extended over |(¢m41)9| (14.C.18). Thus, there
is an extension of f to |L|U|Km4i|. Inductively, we extend f to a map
f : |L|U|Km4i] 3 S”, which completes the proof, since |L| U |Kn| = S”.
O
A slightly stronger version of the simplicial approximation theorem II

is frequently useful. Suppose f : |K| — |Z| where K and L are complexes.
If f is already simplicial on some subcomplex M of K, it would be desir-
able to obtain a simplicial approximation ¢ such that ¢|)m) = f\jaj- The
problem, however, is that when K is subdivided as in the proof of the sim-
plicial approximation theorem I, so is M and f is no longer simplicial on
the corresponding subdivision of M. This problem may be avoided by care-
fully subdividing K (almost barycentrically) in such a way that (i) M is not
disturbed, and (ii) outside of M, the subdivision is fine enough so that the
construction in the proof of the approximation theorem still works. We indi-
cated in the previous figure how this may be done. Successive subdivisions
are denoted by (K,M), (K,M)®),....
The following simplicial approximation theorem is due to Zeeman; we

do not prove it, but instead refer the interested reader to Maunder[1970] for
a Clear exposition.
(14.C.20) Theorem (Zeeman [1964]). Suppose (K, M) is a simplicial pair

and L is a complex. Let f : |K| — |L| be a continuous function with the
property that fj,y| is simplicial. Then there is an integer n, and a simplicial
map g : |(K,M)‘”| — |L| such that g|j7) = fly) and g ~ f (mod|M)).
Although g is not actually a simplicial approximation of f, it is at least
“close,” i.e., it is homotopic to f and agrees with f on M.
Zeeman’s theorem may be used to establish the no retraction theorem
for all n, which in turn yields the general Brouwer fixed point theorem.
(14.C.21) Theorem. There is no retraction from B” onto its boundary

Ss a,
Proof. We need a simplicial pair in order to apply Zeeman’s theorem.
Consider B” to be the closure of an n-simplex, on, and S"~! to be the
boundary, on, of co. Suppose there is a continuous map ¢: |a9| > |d,| that
is the identity on |¢,|._ By Zeeman’s result, there is an integer m and a
simplicial map g : |(02,0n)'™)| > |on| that also leaves Gy fixed.
Let b, be the barycenter of an (n — 1)-simplex s lying in d,. We will
show that g~!(b,) is a polygonal arc with one end at b, and the other
end at a distinct point of |d,,|; this will contradict the fact that g acts as
the identity on'|é,|. Let o be any n-simplex in (0%,¢,)'™), and suppose
g-1(bz) N |o9| # 9. Then of course, o must be mapped onto s. Since it
can be shown that exactly two (n — 1)-faces of o map onto s, it follows
that |o9|Mg~1(bs) is the line segment in o joining the barycenters of the
two faces. Thus, if a is an n-simplex, then g~'(bs) M |o9| is either a line
segment or empty, and consequently, there is a sequence of line segments
that cannot cross themselves. Since only a finite number of simplexes are
involved, eventually the string must come to an end at a point in |o,| from
whence it did not start (each (n — 1)-simplex in S"~' is the face of only one
n-simplex in S"). O
(14.C.22) Corollary (Brouwer Fixed Point Theorem). Every contin-
uous function f : B” > B” has a fixed point.
Proof. The proof is the same, mutatis mutandis, as that used in (12.C.16).
O
n
(14.C.23) Corollary. The n-cube /” = I] I has the fixed point property.

Zi
Oo
We need the next result in chapter 18.
(14.C.24) Corollary. For the n-cube I” define T; = {x : x; = 1} and

B, = {a : 2; = 0}. Suppose for each i = 1,...,n, S; is a closed set and
I”\S; has a separation (U;, Vi) with T; C U; and B; C V;. Then Nj,Si FO.
Proof. By hypothesis, for each 2, (U;, V;) is a separation of I” \ 5; with

T; C U; and B; C V;. Since U; and V; are open in J” \ S; then they are also
open in J” For each @ = (21, 29,.:.52n) € 1% let h(@) = (hi (@), «35 en(Z))
be defined by
ev | FE, o4), ane, Ve,
hi(z) a {—d(x, S;) if x € U;.
Define f : I" > I” by f(x) = (zx + i(a),...,2n + ha(z). It-is an easy

exercise for the reader to show that f(a) € J” and f is continuous. The
fixed point isin NM7_,S;. O
D. EDGE PATH GROUPS: HOW TO COMPUTE 7,(|K|,v0)

FOR ANY COMPLEX
We demonstrate a particularly striking use of the simplicial approximation
theorem II by renewing our attack on the calculation of fundamental groups.
The spaces to be considered will all be triangulable, which is not an unduly
restrictive condition since many common and/or interesting spaces may be
triangulated.
First, a new group, the edge path group, is defined. Then we show the
edge path group is isomorphic to the fundamental group. By the very nature
of its definition, the edge path group is (in an illusory way) more computable
than the fundamental group.
Although the details are somewhat involved, the basic ideas are quite
simple. The fundamental group of a space X was based on the notion of
loops, i.e., maps of J into X sending the endpoints to a base point x9. Loops
were defined to be equivalent if they were homotopic relative to {0,1}. Sup-
pose now X is triangulable and vo is a vertex of the triangulation. From
what we have already seen, a loop in X at vg may be simplicially approxi-
mated by a map ¢ that is homotopic to f relative to {0,1}, provided that
the triangulation of X is sufficiently fine. Thus, rather vaguely speaking, it
seems reasonable to consider “linear” maps when trying to associate a group
with the given space. This leads in a fairly natural way to the following
definition, which gives the “simplicial” counterparts to paths and loops.
Edge Path Groups: How to Compute 7(|K|,
vo) for Any Complex 361
(14.D.1) Definition. Suppose K is a simplicial complex and v, 6 are

vertices of K. An edge path in K from v to @ is a sequence of vertices
Wo,W1,---,Wn, Where for each 7, w; and wj41 span a simplex in K (with
dimension either 0 or 1), wo = v, and wy, = d. If v = 8, then the path is
called an edge loop.
Multiplication of edge paths is effected in the obvious manner. For edge

paths a@ = woW1 ::: Wn and 2 = v9t1 --- Um with wn = v9, af is defined to be
the edge path wow) ---Wnv1U2°::Um. The definition of the reverse of a path
is equally straightforward. If a = vgv, --- Up is a path, then a” is defined to
be UnUn—1°**Vo-
An equivalence relation is introduced on the set of edge paths as follows.
Two edge paths with identical end points are equivalent if and only if one may
be obtained from the other by a finite sequence of the following operations.
I. If a = upv, - + U;Vi41Vi42°°*Un and v; = U441, then a@ is replaced by
VoU1 *** UViVi42°°*Un-
Il. If @ = vp,
++ UiVj41°*+Un and w = v;, then a is replaced by
UVOU1 + °° UjWUj41 °°" Un-
Il. If @ = upd, +++ Uj-1UiVi41 +++ Un and {vj-1, Vi, Viti} spans a simplex
in K, then a is replaced by vov, --- Uj—1Uj41 °°" Un-
II-!. If a = vp, - + Uj—-1Vi41 +++ Un and {v;-1v;i41} spans a simplex in
K, then a is replaced by vovq - ++ Uj—1UjVi41 °°*Un-
It is easy to verify that these four operations lead to an equivalence
relation on the edge paths of K.
(14.D.2) Examples.
vo U1 =V4 U3 vo V1 U3
V6 Ug V10 U6 U8 Vv10
The edge path vov1v2U3U4U5V6V7V9V10 is equivalent to the edge path

UV0V1 U5 V6 V7U8 V10; since we have U0 U1 V2U3 U4 U5 U6 U7U9V10 ™ V0U1 U3 U4 U5 U6 U7U9V10 ™
V9 U1 V4U5 VEU7ZV9U10 © VOV1U5UEU7V9V10 V6 U7U8U9V10

~ VOU1U5 ™ VOV1 U5 U6 U7 U8 V10-
(14.D.3) Exercise. Suppose @ and f are equivalent edge paths with end
points v and 0, and & and @ are equivalent edges with end points 0 and uw.
Show that ad ~ Bf.
(14.D.4) Exercise. Show: if a ~ 6, then a” ~ B".

(14.D.5) Exercise. Show: if a is an edge path with end points v and w,
then va~a~aw, aa’ ~v anda’a~w.
Even though the preceding exercises are trivial, they are indispensable
in allowing one to define a group structure based on edge loops.
(14.D.6) Definition. Suppose vp is a vertex of a simplicial complex K. For

each edge loop a based at vo, let [a] denote the corresponding equivalence
class. Define the product of equivalence classes [a][(] to be [af]. If [a] is
an equivalence class, define [a]~! to be [a]. These operations yield a group
with identity element [vo]. This group is the edge path group, and is denoted
by e(K, vo).
The reader may establish easily that e(K,vo) is a group. We eventu-
ally show e(K, vo) is isomorphic to 7(|K]|,vo), and hence is a topological
invariant. First, however, we see how these groups are calculated. Groups
may be determined by listing their generators and a sufficient number of
relations between these generators (see Appendix). However, the student
should realize that finding generators and relations does not always satisfy
one’s curiosity, or increase ones intuition, about the nature of the group.
The following concept is useful when seeking generators and relations
for the edge path group.
(14.D.7) Definition. Suppose K is a simplicial complex. A 1-dimensional

subcomplex L is called a tree if and only if |L| is contractible.
The existence and some properties of trees may be deduced from the
following set of exercises.
(14.D.8) Exercise. Show: Every simplicial complex K contains a maximal
tree (i.e., a tree not properly contained in any other tree).
(14.D.9) Exercise. Show: a tree L in a connected simplicial complex K is

maximal if and only if L contains all the vertices of K.
(14.D.10) Exercise. Suppose L; and Lz are subcomplexes of a simplicial

complex K, L;UL2, = K, M = L,N Lz is connected, and T is a maximal tree
in M. Show: there is a maximal tree T in K such that TN L; is a maximal
tree in L; fori=1,2andTNM =T.
(14.D.11) Definition. Suppose K is a simplicial complex. Order the set

of vertices of K, vp < vy < vg < +++ < un. A 1-simplex (v;v;) is ordered if
and only if vj < vj. The ordered 1-simplex (v;v;) is denoted by s;; or S5y
A 2-simplex (v;,v;v,) is ordered if and only if v; < vj < vp.
Suppose K’ is a simplicial complex, |K| is path connected and L is

a subcomplex of K with a contractible carrier and containing all of the
Edge Path Groups: How to Compute 71(|K’|,
v0) for Any Complex 363
vertices of K. A group G is formed that has as generators the symbols s;;,

one for each ordered 1-simplex s;; in K \ L, and having relations of the form
8ij8jk8j,, one for each ordered 2-simplex (vjvjuz) in K \ L. (If (ujv;) € L,
interpret s;; as e.)
(14.D.12) Theorem. The group G is isomorphic to e(K, vo).
Proof. Define homomorphisms ¢ : G — e(K,vp) and w : e(K,vo) 3 G
which are inverse to each other. For each vertex v; of K, pick an edge path
e; that goes from vp to v;. If s;; is an ordered 1-simplex in K \ L, define
OSes) = [exvivjer]. If (u;vj;v~) is an ordered 2-simplex of K \ L, it is easily
verified that $(s:;)0(s;x)(¢(siz))~1 = e. Consequently, ¢ may be extended
to a homomorphism (Ap.C.13) from G into e(K, vo).
To define the map 7%, we use the following notation. Suppose vu; and v;
are vertices in K that span a simplex in K. Let
sij___ if (u;v;) is an ordered 1-simplex in K \ L,

=i
tij = 4 83; if (vjv;) is an ordered 1-simplex in K \ L,
e otherwise.
Suppose @ = U0U;, Viz *** Ui, Vo is an edge loop in K based at vp. Define
w([a]) = toi, ti:i. -+- ti,0- The remainder of the proof is straightforward, but
the reader should verify: (i) 7 is a well defined homomorphism, and (ii) dy
and ~w@ are the identity isomorphisms of e(K,vo) and G, respectively. O
(14.D.13) Notation. With the notation as above, we henceforth denote
the generators $(s;;) of e(K, vo) by gi;.
To illustrate the relative ease of calculating edge path groups with the
aid of the preceding theorem, we start with the unit circle S!. Since S' is
topologically the same as a triangle, a contractible subcomplex L containing
all vertices may be obtained as indicated in the next figure.
V1
gol gi2 L g
G~{so2;}~{902;}
vO 902 U2
Thus, G is a group with one generator, go2, and no relations, i.e., the infinite
cyclic group Z. Before proceeding further, let us establish that the foregoing
result is no coincidence, i.e., for any simplicial complex K, we have e(K, vo)
is isomorphic to 7 (|K], vo).
(14.D.14) Theorem. If K is a simplicial complex with connected carrier,

then e(K, vo) is isomorphic with 7(|K'|,vo). -
Proof. We associate a map f, : I > |K| with each edge loop, a =
UOUi; Vip *** Vi, Vo based at vp. Let 0 = ao < ay < +++ < Ag-1 < Ak < Ak41 = 1
be a partition of J. Then f, is defined to be the simplicial mapping of
each simplex (aj;a;41) onto (vi; vi,,,). Define ¢ : e(K, vo) > ™(|K|, v0) by
setting ¢((a]) = [fa]. The reader may verify that ¢ is well defined and is a
homomorphism.
The simplicial approximation theorem is used to show that ¢ is onto.
Suppose [f] € m(|K|, vo). Then f : J + |K| and f(0) = vo = f(1). By the
simplicial approximation theorem II and (14.C.7), there is a triangulation of
I whose vertices form a partition of J, 0 = ag < ay < -++ < ay < Qg41 = 1,
and a simplicial map f (relative to this triangulation) such that $273 1K.
and fn,f.Ifa = f(ao)f(a1)--- f(ae)f(ax41), then clearly, ((a}) = [f].
To show that ¢ is 1-1 requires somewhat more effort. Suppose ¢([a]) = e,

where @ = UpUj, Vin *** Vi, Vo, that is, fy : I + |K| is homotopic via H to
Cy, (mod {0, 1}).
Triangulate J x I as indicated in (a) of the next figure, and note that
H restricted to the boundary B of I x J is simplicial. Hence, by Zeeman’s
version of the simplicial approximation theorem, there exists a subdivision
L = (I x1, B)'™ (which leaves B alone) (See (b)) as well as a simplicial map
G:|L| + |K| such that G|g = H|g and G is homotopic to H (mod B).
In L, however, the edge path aga, ---ax41bcao is equivalent to apo. (Fig-
ures (b), (c), and (d) give a few of the steps that may be taken when
L = (I x I, B)“) to show that. these two paths are indeed equivalent. For
n > 2, similar, albeit many more, equivalence operations may be performed
to yield the edge path equivalence.)
c b c b
ao ay a2 a3 a4 a5 a 5
c b
a0
z 3 5
To see that [a] is equal to the identity element in e(K,vo), we observe

that
A = U0VUji, *** Vi, VO ~ VOVi, ** * Viz, VOVOVOVO
= G(a0)G(a1) --- G(ax41)G(b)G(c)G(ao)

eet G(ao) = Vo.
The last equivalence is due to the fact that aga, ---a,%4bcag is equivalent to
do and G is simplicial. Therefore, dis 1-1. O
Observe that it follows from the proof of the foregoing theorem that the
fundamental group 7 (||, vo) depends only on the carrier of the 2-skeleton
| 2| of K.
Despite the apparent ease of calculating fundamental groups via the edge
path group, the reader should realize that all is not well. Many triangulable
spaces require an inordinate number of 2-simplexes, (the torus, for example
requires at least twenty-one 1l-simplexes, fourteen 2-simplexes, and seven
0-simplexes). Thus the number of generators and relations may become
astronomical, even for relatively simple spaces. Computation of the edge
path group is often greatly facilitated by the following theorem, which is due
to Seifert and van Kampen. It is the polyhedral analog of the Seifert-van
Kampen theorem stated in Chapter 12.
Suppose G is a group with presentation {g1,...,9n : T1,--.;Tm} and
H is a group with presentation {h1,...,hp : $1...,8;}. Recall from group
theory that, {91,;.-.49a,/1---;Np 2T1,--¢, Tims S1>--, $7} 18 aypresentation

for the free product of G and H (Ap.B.10).
(14.D.15) Theorem (Seifert-van Kampen Theorem). Suppose K is
a simplicial complex, L; and Lz are subcomplexes of K whose union is K,
|L,|, |L2| and |N| are connected, where N = 11M Lz, i; : |Li| + |K| and
ig : |L2| 4 |K| are inclusion maps, and vp is a vertex of N. Then e(K, vo)
is the free product of e(Lj, vo) and e(L2, v9) with additional relations of the
form (i1).(g)((i2)*(g))~*, one for each element g in a finite set of generators
for e(N, vo).
Proof. By (14.D.10), there are trees Ty, Tr, and Ty,, such that
(i) Tw is maximal in N,
(ii) bu IV =i = 1 ON,
(iii) Ty, UT,, = Tx is a maximal tree in K’, and
(iv) Ty, and Ty, are maximal in L; and Lz, respectively.
Order the vertices of K; this induces an order on the vertices of each sub-
complex of Kk. As before, e(K,vo) is the group generated by symbols of
the form g;; subject to the relations 9;;9;% aA where each g;; corresponds
to an ordered 1-simplex in K \ Tx. However, any 1-simplex in K \ Tx is
certainly a 1-simplex in either L; \Tzr, or Lz \Tz,, and vice versa. Hence if
a typical ordered 1-simplex in L; \Tz, gives a generator denoted by 953 and
a 1-simplex in Lz \ Ty, gives a generator denoted by 955 then e(K, vo) is the
group generated by all of the gi; and g?, with the same relations as before
(but relabeled to agree with the new names of the generators). There is a
complication, however. It is possible that gj; = gj, (in which case gi, and g?,
correspond to the same ordered simplex in N \ Ty). This problem is elimi-
nated by adding relations gi,(g?,)~' whenever gi, = g?,. Thus, e(K, vo) may
be described as the free product of e(L1, v0) and e(L2,v9) with additional
relations (i1)+(9i;)((i2)*(gi;))~* = e for each generator g;; € e(N,vo). O
(14.D.16) Corollary. Suppose K = L; U Lz, where |K|, |Li| and |L| are
connected. If |N| = |Z1 M Le| is simply connected, then e(K, vo) is the free
product of e(Z1,vo) and e(L2,v9). O
(14.D.17) Corollary. If K = L; UL», |K|, |Li|, |L2|, and |N| = |L1 N Lo|
are connected, and |Lg| is simply connected, then e(K, vo) may be obtained
from e(L1, v9) by adding the relation (71),(gi;) =e for each g;; € N \ Ty.
Proof. The free product of e(L1,v9) and e(L2,v9) is isomorphic to
e(L1,v0). By use of (14.D.15), we obtain additional relations of the form
(11) (giz) ((t2)x(giz))~* = e for each gi; € e(N, v0), where g;; corresponds to
an ordered 1-simplex in N. Since (i2).(gi;) = e, we have ((i2).(gi3))7! =e,
and hence also relations of the form (71)4(gi;) =e. O
This latter corollary is especially useful in the calculation of fundamen-

tal groups of 2-manifords. Many cases (including all compact connected
2-manifolds) may be constructed by attaching 2-cells to appropriate com-
plexes. In fact, it will follow from results in Chapter 16 that any compact
connected 2-manifold can be formed by gluing a 2-cell in a suitable man-
ner to the iterated wedge product of S! with itself. (The wedge product of
pointed spaces (X,20) and (Y, yo) denoted by X V Y is the space obtained
from the disjoint union X UY by identifying the points zo and yo.)
Consider once more the torus as a quotient space obtained by identifying
sides of the unit square
Note that under this identification, the sides of the square form the wedge
product of S! with itself, and a 2-cell (the original square) has been attached
to the product. Alternatively, this may be described as indicated in the next
figure. A 2-cell D is attached to S' V S! by gluing ab onto vstv, be onto
vs't'v, cd onto vtsv, and da onto vt's'v.
The end result is the torus. . .

In a similar manner, the projective plane is obtained by attaching a
2-cell to S! as shown in the next figure.
acb—stus
bda—stus
t U
(14.D.18) Exercise. Show that the two-holed torus may be obtained by

an appropriate attachment of a 2-cell to the wedge product of four circles.
More generally, if K is any complex and a = vot ---UnUo is an edge

loop in K having no identical adjacent vertices, then a 2-cell may be attached
to K along a in the following manner. Let P be a regular polygon in ee
with n+ 1 sides. Triangulate P as indicated below and label the vertices
b, Vp, V1,---,Un- Let f be a simplicial map that sends the boundary of P onto
the carrier of the loop a by mapping vertices uv; onto vu; for each 7. Then
|K| = |K|Uy |P| is the carrier of a simplicial complex ie
!
uv 8
Ake
WA ABA
/
U7 /
Vo U5
ANT AASK
7
Let s be the 2-simplex with vertices vj, v, and b, and |s9| be the closure
of s. Let b’ be the barycenter of the 2-simplex (bujv}). Radial projection from
b yields a deformation retraction of |K|\|s9| onto |K| that leaves points of ||
pointwise fixed. Of course, the projection is originally done in |P| and then
transferred over to |K | by f; furthermore, the projection is not piecewise
linear. By (12.C.19), m(|K| \ |s9|,v0) is isomorphic to ™(|K|,vo). Since
m™(|s9|) = {e}, (14.D.17) may be applied, where Lz = s9 and L; = K \s, to
see that e(K ,Uo) may be obtained from e(K \ 8, Uo) by adding the relations
arising from generators in L; M Lz. Note that any ordered 1-simplexe in
L, M Lg serves as a generator for e(L1 N Lz, vo).
In essence, what has occurred is that if a cell is glued to a loop in K,
then the group element determined by this loop is obliterated (reduced to
the identity). O
(14.D.19) Example. Consider the torus T as a square with sides identified.

The complement of the cell C in the torus indicated by the dashed line is
Problems 369
deformable to S' V S!. Hence, since e(C) = {e} and e(S! V.Sa) = gig> ss},
we have e(T’) = {91, 92; 91929‘99;'}-
92
There are two rather startling results that may with some imagination
be extracted from our work thus far. The enterprising reader might wish
to consider the following two assertions, the first of which is not difficult to
establish.
(i) For each finitely presented group G, there is a two-dimensional
polyhedron whose fundamental group is isomorphic to G.
(ii) For each finitely presented group G, there is a 4-manifold that has
G as its fundamental group.
The latter result coupled with the work of Rabin [1958] leads one to
conclude that there is no countable algorithm that classifies 4-manifolds.
PROBLEMS
Section A
1. Show: if K is a finite collection of simplexes lying in €”, then K is a

simplicial complex if and only if
(i) ifo € K andr <a, then 7 € K, and
(ii) if o,7 € K, then 09 179 is the closure of a face of both o and r.
2. Suppose K is an infinite simplicial complex lying in €” (or in Hilbert
space). Show that the relative topology on |K’| coincides with the topol-
ogy
U = {U Cc |K| : Ufa is open in o
for each
o € K}.
3. Show that the result of problem 2 fails to hold if the local finiteness
condition is removed from (14.A.15).
4. Let V be a set of elements (called abstract vertices) and let K be a
collection of finite subsets of V with the property that any subset of a
set in K is also in K. Then K is called an abstract complex. Any set of

n+ 1 vertices a9,a1,...,@n in K is called an abstract n-simpler. Show
how an infinite simplicial complex may determine an abstract complex.
Two complexes K and L (simplicial or abstract) are isomorphic if and
only if there is a simplicial bijective map ¢ : K — L such that 67!
is simplicial. Show: any finite abstract complex with n + 1 vertices is
isomorphic to a subcomplex of the closure of an n-dimensional simplex.
Suppose (X,U/) is a topological space and A = J]yey Ey, where Ef = E'
for each U € U. For each V € U, let ty € [yey Ey be the point in
A whose V-th coordinate is 1, and whose remaining coordinates are 0.
Let V be a finite open cover of (X,U/). A complex Ky associated with V
is defined as follows. The vertices of K consist of the points zy, where
Ve Y. Vertices ry,,...,x2y, determine a simplex in Ky if and only if
1 Vi # @. Show that Ky is an abstract complex and is isomorphic
to a simplicial complex lying in Euclidean space €™. The complex Ky
is called the nerve of the cover V.
Find an open cover V of S! such that the carrier of the simplicial complex
associated with the nerve of V (see problem 6) is homeomorphic to S?.
Suppose K is a simplicial complex. Let U = {star(v) : v is a vertex
of kK}. Let N be the simplicial complex associated with the nerve of U
(see problem 6). Show that there is a homeomorphism from |K| onto
|N| such that h maps each simplex of K onto a simplex of N.
Show: every open set in €” can be triangulated by a countably infinite
complex.
10. Show that a triangulation of a compact manifold must be finite.
Ot; Show that every countable, locally finite, n-dimensional simplicial com-
plex can be embedded in €2"*1,
Section B
Show: if K is a simplicial complex, then |K| x J is triangulable.

Show that the product of two simplexes is a linear cell.
Suppose K is a simplicial complex in €™” and L is a simplicial complex
in €”. Define K x L to be {t x a|r € K, o € L}. Show that K x L
is a cell complex and that with a suitable subdivision, K x L can be
made into a simplicial complex (without the introduction of any new
vertices).
Show that the intersection and union of two polyhedra is a Euclidean
polyhedron.
Problems 371
5. Let s be an n-simplex, and suppose o and 7 are opposite faces of s,

where o is a p-simplex and 7 is a q-simplex and p+ q = n-—1. Show
that every point z of s is of the form x = (1 — a)y + az, where y Eo,
z€7T,anda€ J. Find a homotopy that retracts |s9| \ |79| onto |o9|.
Suppose (K, L) is a simplicial pair. Show that the simplexes of K()
having no vertices in common with L@) form a subcomplex M of K(),
Show that |L| is a deformation retract of |K)\ M| and that |K@)\ M|
is a neighborhood of |Z].
Use problems A-4 and B-6 to show that every finite simplicial complex
isan ANRy.
Suppose X and Y are Euclidean polyhedra and that f : X > Y isa

continuous function. Show that f is simplicial if and only if the graph
of f, {(z, f(z)) € X x Y : « € X}, is a polyhedron.
Suppose X and Y are polyhedra lying in €”. For each xz € X andy € Y,

let £;y be the line segment in €” with endpoints z and y. Show that X
and Y are joinable if and only if (i) 42, N X =z and 4,4,NY = y for
each z € X and y € Y, and (ii) ifz,¢ € X andy, y € Y and é,, 4 Cay,
then é;, gg is empty or is an endpoint.
10. An n-simplex o of a simplicial complex K has a free face 7 if and only

if 7 is an n — 1 face of o but is a face of no other n-simplex of K. An
elementary collapse is the process of replacing K by K \ {a Ur}. If L
is a subcomplex of K, we say K collapses to L, denoted K \, L, if and
only if L can be obtained from K by a sequence of elementary collapses.
Show that if K collapses to L, then L is a deformation retract of K.
Lis The house with two rooms (see next figure) consists of the top labeled T;
the partition, P; the bottom, B; the walls W;, W2,W3, and W4; the two
curtains, K,, K2; and the two tunnels, C;,C2. One enters the bottom
room by crawling down from the roof through tunnel C;. Similarly, one
enters the top room by crawling up from under the floor B through
tunnel C2. Show that the house with two rooms (which is triangulable)
can be obtained by collapsing a 3-cell; hence, the house with two rooms
is a retract of a 3-cell, and thus is contractible.
Show that the house with two rooms is not collapsible to a point, but if
curtain Ky is thickened into a 3-cell as shown in the second figure, then
the resulting complex is collapsible.
Show that the product of the house with two rooms and J is collapsible.
12. The dunce cap D is constructed by identifying the sides of a closed

2-simplex as indicated in the figure.
It can be shown that D is contractible (Zeeman [1963]). Show that D

is triangulable, but that there is no triangulation of D for which D is
collapsible to a point.
Section C
1. Suppose K and L are simplicial complexes and f,, fo: K — L are sim-
plicial maps. The maps f; and f2 are contiguous if and only if for each
(vov1 --- Uz) € K, there is a simplex r € L such that fi (vo),..., fi(ve)
and fo(vo),..-, fo(v~) are vertices of T.
(a) Suppose K and L are simplicial complexes and f,, fo : K > L are
simplicial approximations to a continuous map ¢ : |K| > |L|. Show
that f; and f2 are contiguous.
Problems S713
(b) Show: if K and L are simplicial complexes, and fi, fo: K 3 L are
contiguous simplicial maps, then f; and f2 are homotopic.
2. Show: if K and L are finite simplicial complexes, then there are only a
countable number of homotopy classes of mappings from |K| into |L].
3. Suppose A is a closed subset of S” and f : A> S"~! is continuous. Let
{Cj : a € A} be the family of components of S” \ A. For each a € A,
remove a point cq from Cg. Show that there is a continuous extension
off, £25” \ (icy + ae Ai SS".
4. Suppose K, L, and M are simplicial complexes, f : |K| > |L| and
g : |L| + |M| are continuous maps, and ¢f, ¢, are simplicial approxi-
mations of f and g, respectively. Show that ¢,¢y is a simplicial approx-
imation of gf.
5. Suppose K and L are simplicial complexes and f : |K| > |L| is con-
tinuous. Show: there is a simplicial approximation to f if and only if
the carrier of each simplex o € K is contained in f~!(star(v)) for some
vertex uv € L.
6. Suppose A is a compact subset of E” and h: A > E” is an embedding

such that h|p, , is the identity. Show that h(A) = A.
7. Suppose U is a bounded open subset of €”. Show that FrU is not a
retract of U.
8. Show that id: S" > S” is not null homotopic.

9. Suppose A is a compact subset of €” and B is a subset of A homeomor-
phic to B™. Show: if f : A &” is continuous and maps Fr A into B,
then f has a fixed point.
Section D
1. Suppose T is a tree, ao is the number of vertices of T, and a; the number

of 1-simplexes. Show that ap — a; = 1.
2. Use the edge path group to calculate the fundamental group of the
projective plane P2.
3. Show that the suspension of a polyhedron |K| is a polyhedron.
4. Suppose K is a simplicial complex and |K| is a path connected polyhe-

dron. Use the edge path group to calculate the fundamental group of
the suspension of |].
5. The two holed torus T, may be considered as the quotient space obtained
by identifying edges of the octagon as indicated in the figure.
Find the fundamental group of T».

Chapter 15
FURTHER APPLICATIONS
OF HOMOTOPY
Thus far, we have considered two major areas of topology where the concept
of homotopy was indispensable: the creation of the fundamental group, and
the development of covering spaces. These do not represent the only contexts
in which the notion of homotopy is relevant. This chapter is devoted to
exploring a few additional problems where homotopy theory is crucial.
A. THE EXTENSION PROBLEM (REVISITED)

In chapter 4 a number of problems that are equivalent to finding extensions
of functions were discussed. We have also seen that the proofs of many
theorems hinge on whether or not a given map may be extended. Homotopy
theory is often invaluable in resolving such problems.
The following theorem, due to Borsuk, is a rather sophisticated result
along these lines.
(15.4.1) Theorem (Borsuk’s Top Hat Theorem). Suppose A is a

closed subset of a binormal space X (i.e., X x J is normal) and Y is an ANR.
Let f and g be homotopic maps from A into Y. If there is an extension
F of f to all of X, then there is also an extension G of g to all of X, and
furthermore, G may be chosen to be homotopic to F.
Proof. Let H : Ax I - Y be the homotopy between f and g with
Ho = f and H, = g. Extend H to H:T = (Ax I) U(X x {0}) > Y by
375
376 15. Further Applications of Homotopy
defining H(z,t) = H(z,t) if (x,t) € Ax J, and H (2,0) = F(a), otherwise.

This function is continuous by the map gluing theorem.
Since T is a closed subset of X x J, and Y is an ANR, H may be

extended to an open set U containing T. Denote this extension by H.
By (2.G.14), there is an open set V of X such that AC V andVxI CU.
Since A and X \ V are disjoint closed subsets of a normal space X, we can
apply Uryson’s lemma to define a function ® : X — I that maps all of A to
land X \V to 0. Define G: X x I + Y by setting G(z,t) = H(z,t- (a).
Then Gs =P and Gi is the desired extension of g. O
(15.A.2) Definition. A map f : X > Y is inessential (null homotopic) if

and only if f is homotopic to a constant map. Otherwise, f is essential.
With this terminology we have the following corollary.
(15.A.3) Corollary. Suppose A is a closed subset of a binormal space X

and f : A Y is a continuous function from A into an ANR Y. If f is
inessential, then f may be extended to X.
Proof. Suppose that f is inessential. Then it is homotopic to a constant
map c: AY. Clearly, c can be extended, and consequently, so can f (in
fact, to an inessential map). O
(15.A.4) Exercise. Suppose f : (X,20) — (S',s) is inessential. Show that

f ~cs (mod zo), and hence f, : 7(X,20) + ™(S',8) is trivial.
(15.A.5) Exercise. If A is a closed subset of €" and f : A > S™ is
extendible to €”, show that f is inessential.
The requirement that Y be an ANR in the top hat theorem may be re-
laxed if one is willing to place additional conditions on A and X, Essentially,
we define away the problem.
(15.A.6) Definition. Suppose (X, A) is a topological pair. We say A has

the absolute homotopy extension property (AH EP) if and only if for each
continuous map F': X — Z where Z is arbitrary, and for each homotopy
H:AxI- Z with Hy= F\,, there is a homotopy H : X x I > Z such
that H extends H and Ho =ip
The Extension Problem (Revisited) S77.
The reader may feel that the foregoing definition is a typical mathemat-
ical ruse to get around the problem, i.e., avoid it by restricting attention only
to those spaces that. behave themselves. While this may indeed be the case,
there are nevertheless a number of topological pairs that yield the AHEP,
e.g., polyhedral pairs.
(15.A.7) Theorem. Finite simplicial pairs have the AH EP.

Proof. Suppose (K, L) is a finite simplicial pair and X is an arbitrary
space. Let H : |L| x I X and F':|K|-— X be continuous maps such that
Ho = F\jz). For each simplex o in K \ L, there is a projection p, that maps
|o9| x I onto (|o| x I) U(o x {0}) as indicated in the following figure.
projection point
< |o9|xI
ax{0}
Note that p, is actually a retraction. If proper care is exercised in

matching up the projections (start with the highest dimensional simplices
first), then the map gluing theorem yields a retract p:|K| x I > (|L| x I)U
(\K| x {0}).
If H : ((|L| x J) U(|K| x {0})) > X is defined by
i H(a2,t) ifze|L| andt ef,

Eien:
ee Ny tater ar doe,
then Hp: |K|x I > X is the required homotopy. 0
In general, when does the AH EP infect a space? One important crite-
rion is given in the next theorem.
(15.A.8) Theorem. Suppose (X, A) is a topological pair. The subspace A

has the AHEP with respect to X if and only if (X x {0}) U(A x J) isa
retract of X x I.
Proof. Suppose A has the AHEP. Define F to be the natural embed-
ding of X into (X x {0}) U(A x J), ie., F(x) = (x,0), and let H be the
similar embedding of A x J. Then F'|,4 = Ho, and by the AHEP, H may
be extended to a homotopy H : X x I > (X x {0})U(Ax J) with Ho = F.
Clearly, H is the retraction that was sought.
Now suppose (X x {0}) U(A x J) is a retract of X x I. Let F: X 4 Y

be a continuous function and suppose that H : A x I > Y is a homotopy
with Hp = F|,4. The problem is to extend H to X x J. The map H may
first be extended to (X x {0})U(A x I) by setting H(z,0) = F(x), and since
this set is a retraction of X x J, H may be further extended to all of X x I
(4.3.15), 0
A second criterion for the AH EP is due to Young [1964].
(15.A.9) Theorem. Suppose X is a normal space and C is a closed subset
of X. Then C has the AHEP if C is a Gs subset, and there is an open set
U containing C and a map h: U x I > X such that
(i) h(u,0) = u and h(u, 1) € C for each u € U, and
(ii) h(c,t) = c for each c € C, t € I, ie., C is a deformation retract of
U in X (modC).
Proof. By (15.A.8), it suffices to show that T = (X x {0}) U(C x J) is
a retract of X x I. Since C is a Gs set, it follows from (4.B.6) that there is
amap ©: X 4 / such that ¥—-1(1) = C, and U(X \U) =0, where U is the
open set given in the hypothesis. The retraction r : X x I + T is defined as
follows:
(i) ae,0) (2,0) tr) and ee ar
(ii) r(x,t) = h(a, 2U(a)t) if 0 < U(x) < § andte J;
(ili) r(a,t) = A(z, t/2(1 — U(x))) if }< U(x) < Land t < 2(1— U(z));
(iv) r(z,t) = (h(x, 1),t — 2(1 — U(ax))) if § < U(r) < land1>t>
2(1— U(z));
(v). net) = (Ga) at wo elrand Ped:
The reader will enjoy verifying that r is the promised retraction (the
continuity of r as Y(x) approaches 1 is the only troublesome detail). O
B. THE SEPARATION PROBLEM
The Jordan curve theorem states in part that a simple closed curve lying in
E? separates the plane. Obvious as this result seems from a visual stand-
point, it was no trivial matter to verify. Hence, one might suspect that in
general it could prove difficult to determine whether or not a given subset
of €” separates the whole space. The goal in this section is to establish one
important criterion for such a separation to take place. Homotopy theory
once again proves to be an unwitting accomplice.
We make use of certain maps first described by Borsuk, which commonly
bear his name. These maps are constructed as follows. For each p € &",
define Y, : (E” \ {p}) + S"~} by ¥,(x) = (x — p) \ ||z — p||, where z and p
are considered as vectors, and ||x — p|| represents the usual norm (distance
The Separation Problem 379
of x —p to the origin). If A C €” \ {p}, then W,), is called a Borsuk map for

A based at p.
Borsuk maps take the starring role in the proof of the following theorem,
due to Borsuk.
(15.B.1) Theorem. Suppose A is a compact subset €". Then A separates
€” if and only if there is a map f : A > S"~! that is essential.
Proof. Suppose A separates €”. Let C be a bounded component of
E” \ A (why does one exist?), and pick any point p € C. The Borsuk map
for A based at p will prove to be essential. If W, is inessential, then by
(15.A.3), Y, can be extended to all of €”. In particular, UV, can be extended
to a ball B containing A with center p and radius r > 0. But this allows
us to construct a retraction of B onto its boundary, an impossibility. Let ®
denote the extension of VY, to B, and note that, in particular, @ is defined
at p. Then f : B > FrB defined by
ee
ile: peer TT for r€ B\C,
p+r-®(x) otherwise.
is a retraction, which contradictions (14.C.21).

The converse is more difficult and is based on the following lemma.
(15.B.2) Lemma. Suppose K is a compact subset of €”, C’ is a closed

subset of K, and f :C > S"~! cannot be extended to K. Then, there is an
open subset V of €” such that V C (K \C) and FrV CC.
Proof. Since f cannot be extended to K, there is, by (4.B.23), a closed
subset C* of K such that
(i) f can not be extended to CUC*, and
(ii) f can be extended to the union of C with any proper closed subset
Chir:
We first show C* \ C is open in €”. Suppose C* \ C' is not open and that
z € (C*\ C)NFr(C* \ C). Let U be the interior of a small n-ball that
contains z. Since C* \ U is properly contained in C™, there is an extension
of f, f, that maps CU(C* \U) into S"~'. In particular, f maps C*N (FrU)
into S"-!, and since C*M (Fr U) is a closed subset of the (n — 1)-sphere Fr U,
we may apply (14.C.19) to obtain an extension of flosn(Fru) to all of FrU.
Thus, we now have a map f :((C*\U)U((FrU) UC) > S”“? that extends
f. Let p € UNextC”, and from p project points of C* NU into FrU by a
map P. Finally, extend f to F : (C UC* + S"* by setting
ea ifs EeUN(CUC*),
F(z)=4 .
Peerit eee (CAN Oia rt,
Since this extension violates the properties of C*, C* \ C’ must be open in

E”. If now follows easily that Fr (C* \ C) C C, since C® is closed and C*\C
is open. + QO '
We now complete the proof of (15.B.1). Suppose f : A > S"~! is

essential. Let B be a large ball in €” that contains A. It follows from
(15.4.5) that it is impossible to extend f to B, and hence by the foregoing
lemma, there is an open set U in B \ A whose frontier lies completely in A.
Therefore, U is both open and closed in €” \ A. Hence A separates E”. O
(15.B.3) Corollary. Suppose A is a compact subset of €” and A is another

subspace of €” homeomorphic to A. Then A separates €” if and only if A
separates €”. O
Note that this corollary yields a portion of the Jordan curve theorem,
although it should be observed that neither (15.B.1) nor the corollary gives
the number of components that a separation may cause, nor if this number
is a topological invariant. Results of this nature may be found in Dugundji
[1966].
The next result contains an important consequence of (15.B.1). This
result identifies one of the most powerful characteristics of E”: invariance of
domain. The theorem states that embeddings map open subsets of €” into
open subsets of €”. This property also holds for n-manifolds (see problem
B.10).
(15.B.4) Theorem (Invariance of Domain). If U is an open subset of
E€” and h: U > €” is an embedding, then h(U) is open in E”.
Proof. Suppose z = h(u) € A(U). Since U is open, there is an € > 0
such that if V = S.(u), then V C U. Note, since V does not separate €”,
it follows from the previous theorem that h(V) also fails to separate €”.
Furthermore, h(V) \ h(Fr V) is connected, since h(V \ FrV) = h(V)\A(FrV)
and V \ Fr V is clearly connected. Hence, €” \ h(Fr V) may be written as the
disjoint union of the connected sets €” \ h(V) and h(V) \ A(FrV).
Since these two sets are connected, they must be components of the
open set €” \ h(FrV). However, components of open subsets of a locally
connected space are open; hence, in particular, W = h(V) \ FrV) is open,
and since z € W C A(U), it follows that h(U) is an open subset of €*. O
C. UNICOHERENCE
For a third illustration displaying the versatility of homotopy theory, we

reconsider briefly the concept of unicoherence. Recall that a connected space
X is unicoherent if and only if whenever X = C; UC where C, and C2 are
closed and connected, then C, M C2 is connected. The unit circle S! is an
Unicoherence 381
example of a space that is not unicoherent, since overlapping semicircles do

not have connected intersection. Loosely speaking, unicoherence reflects a
lack of “holes” in the space. The following theorem is considerably more
precise.
(15.C.1) Theorem. Suppose X is a connected, locally path connected

metric space with the property that every continuous map f : X > S! is
inessential. Then X is unicoherent.
We use a pair of lemmas to prove the theorem. The first lemma follows
readily from covering space theory, and the proof of the second is left as an
easy exercise.
(15.C.2) Lemma. If X is a connected, locally path connected space and

f : X — S' is inessential, then there is a map g : X — &! such that
f(x) = (cos g(x), sin g(x)) for each x € X.
Proof. Consider the following diagram,
Here, p : €1 > S! is defined by p(x) = (cosz,sinz). Then (€',p) is a

covering space for S!. Since f,(71(X,2z0)) = {e} C px(m(E',0)), it follows
from (13.D.1) that f may be lifted to a map g such that pg= f. O
(15.C.3) Lemma. Suppose f and g are maps from X into E1. Define f
and @ by f(x) = (cos f(z),sin f(x)) and g(x) = (cosg(z),sing(x)). If we
have f(x) = g(x) for each x € X, then f(x) = g(x) + 2nm for some integer
nandforeachre€ X. O
Proof of (15.C.1). Suppose X is not unicoherent. Then there are closed
connected subsets A and B such that X = AUB and ANB is not con-
nected. Write 4 B as the disjoint union of nonempty closed subsets C;
and C2. For each z € X, let gg = d(x, Ci)/[d(x, C1) + d(x, Cz)]. Define func-
tions f4 : A > S! and fg: B > S! by letting fa(x) = (cos7qe,sin7qz)
and fp(x) = (cos7@z,—sintgz) = (cos(—7qgz),sin(—7gz)). Observe that
fal|ans = falans, and consequently the map gluing theorem ensures that
the function
ba fa(z) ifa@gA
os esi ifzeB
is both well defined and continuous. By hypothesis f is inessential; by
(15.C.2), there is a map g: X — €! such that f(x) = (cosg(z),sin g(z)).
Lemma (15.C.3) gives integers m and n such that g(x) = 7q,+2m7 ifee A
and g(x) = —7q, + 2na if x € B.
However, if x € Cy C ANB, then g(x) = 2mm = 2nz, hence m = n.
On the other hand, if r € C2 C ANB, then g(x) = 7 + 2ma = —7 + 2nz,
and we have shown that 7 = —7. Thus, X must be unicoherent. O
(15.C.4) Corollary. The spaces €”, I” and S"*? are unicoherent for each
mGeNe ge
D. 7,(€°\G), FOR A POLYGONAL GRAPH G IN &
Our goal is to convince the reader of the existence of an algorithm that may .
be used to compute the fundamental group of the complement of a knot, or
more generally of the complement of a finite graph lying in €°. A careful
blend of mathematics and sleight of hand is employed. We hope that the
reader will be convinced that the hand waves can be done mathematically.
It is assumed that only the more naive, or masochistic, reader will attempt
to do more than determine that the constructions can be carried out.
(15.D.1) Exercise. Show that (E? \ {p}) = {e}.
(15.D.2) Definition. A graph in €° is the carrier of a one dimensional
simplicial complex in €?.
(15.D.3) Remark. If G is a graph and vp is a vertex of G that is the face of

only one 1-simplex (vpv;), then there is a homeomorphism of €° \ (G \ (vpv1))
m1 (E? \ G), for a Polygonal Graph G in €3 383
onto €? \G. Actually, there is an isotopy, i.e., a homotopy H : (E3 \ G) x
I + €°\G such that Ho = id, hy is onto E? \ (G \ (vov1)), and H; is a
homeomorphism for all t.
_ Thus, the fundamental group is unchanged if G is replaced by a graph
G in which no vertex is the face of just one 1-simplex (simply apply the
foregoing remark a finite number of times). Furthermore, the Seifert-van
Kampen theorem may be used in conjunction with (15.D.1) to enable us to
eliminate components of G that consist of single vertices.
Hence, the fundamental group of the complement of the graph in the
preceding figure is isomorphic to the fundamental group of the complement
of the following graph.
ser
In the remainder of this section, we assume that all graphs have been
simplified in this way. We make a further observation—subdivision of G
(considered as a complex) does not change 7;(E? \ G).
(15.D.4) Definition. A branch point v of a graph G is a vertex of G that
is the face of three or more l-simplexes of G. The graph just pictured has
three branch points.
(15.D.5) Definition. Let p : €? + E? C E* be the usual projection map
onto the zy plane, i.e., p(z,y,z)= (z,y, 0), and let p = plc.
The map p projects G normally onto the zy plane (or p(G) is a normal
projection) if and only if
(i) p is 1-1 except at a finite number of points of G;
(ii) for each g € G, card(p~'p(g)) < 2;
(iii) if v is a vertex of G, then card(p~'p(v)) = 1.
Observe that (iii) rules out the following configurations.
pa
| _
(15.D.6) Definition. Suppose p(G) is a normal projection and g,g € G,

g # G, and p(g) = p(g). Whichever of g and g has the larger z coordinate is
called an overcrossing, and the other point is called an undercrossing.
It will be (and has been) convenient to indicate the projection of an
overcrossing by a solid line and the image of an undercrossing by a break in
the line.
The line segment contains —> > The line segment contains
an overcrossing an undercrossing
There is no reason to believe that the projection of a given graph into

the zy plane will be normal. The projection might, among other things,
collapse a whole 1-simplex to a point. In the next theorem we see that if a
projection is not normal, then it may be “normalized” by a slight rotation
of the given coordinate system.
(15.D.7) Theorem. Suppose G is a graph in €%. Then either p(G) is a

normal projection, or a small rotation of the coordinate system of €° will
yield a projection p that projects G normally onto the new zy plane.
Sketch of proof. We will stand at the origin and keep track of the
directions in which projections are to be proscribed.
1. Consider the vertices of G three at a time. For each triple that spans
a plane, translate the plane to the origin. The z axis will be selected so that
it misses each of these planes (except at the origin, or course). Any z axis
not contained in one of these planes yields a projection that satisfies parts
(i) and (iii) of the definition of normal projection (15.D.5). However, there
still may be points g € G such that 2 < card(p~'p(g)) = n < oo. We work
on this problem in the following paragraph.
2. For each triple of 1-simplices of G, which when extended form a triple
of skew lines, find the unique direction with the property that projection in
this direction yields a triple point. Draw a line through the origin in each
direction that is determined in this manner. Now there are a finite number
of “unwanted” planes and lines passing through the origin. If the original
a axis is in any one of the forbidden planes or lines, an arbitrarily small
rotation will move the z axis into an allowable position (which of course also
changes the x and y axes, but that is of no concern). Projection relative to
the new coordinate system is normal. Note that topologically we have used
the obvious fact that the union of a finite number of lines and points forms
a nowhere dense subset of the projective plane. O
™7(E? \ G), for a Polygonal Graph G in €3 385
(15.D.8) Notation. Let O be the set of overcrossings, U the set of un-

dercrossings and B the set of branch points. Let R = UUB. For the
remainder of this section we will assume that O and U have n elements, B
has 6 elements and R has n + b = m elements.
For convenience in applying the Seifert-van Kampen theorem we repo-
sition G. Without loss of generality we may assume that G initially lies
in the upper half of Euclidean 3-space (otherwise, shift the z-coordinates
via a translation). Now let the graph drop. As each undercrossing touches
the zy-plane let it sink slightly through the plane, then add (if necessary) a
new vertex on each side of the undercrossing where the graph touches the
zy-plane and let the ins of the graph settle onto the ry-Plane.
Now, at each branch point add (if nena a nearby vertex on each
arc coming to the branch point and then push the branch point through the
xy-plane. After this stage the graph appears as:
Next, at each undercrossing reposition the graph so that at the ¢-th

undercrossing (i = 1,-:-,n) we have a small neighborhood T; that looks
like: eins Sf eran csp 250
Further, if b; is a branch point with k arcs we rearrange the graph so

that in a smal! neighborhood T; (j =n+1,...,m) we have a configuration:
Pick as base point p = (0,0,1). Let A be the slab between z = —e/2 and
z = —3€/2 with p;'(the part of G that lies in z = —e) removed. Add an arc
that starts in A, misses G and goes to p. Clearly 7(A) ~ {ri,r2,.-.,7%m;3 }
where r; is represented by a loop from p to the plane z = —e then to and
counterclockwise around the i-th hole and back to p.
Now consider B, a “thickened” open top box that is large enough so

that it surrounds the part of G lying in z = —e.
(15.D.9) Exercise. Use the Seifert-van Kampen theorem to show for any
eS Sy
7™(AUB) & ph emee Laie cena e
m1 (E* \ G), for a Polygonal Graph G in €3 387
where the circumflex means the rj-th generator has been removed. [Any
loop encircling r; can be slid over AU B and expressed in terms of the other
generators. |
Next, purely for purposes of accounting, we make some more arbitrary
choices. We remove the branch points from G and obtain a number of arcs
and circles. Choose an arbirary direction in each of these and direct the arcs
of G lying in z > —e accordingly.
We now choose the “right hand rule” for determining the exponent of a
loop from the base point that passes once under the i-th arc of G\ R. Stand
on the arc facing the direction of the arrow. If the loop passes under you from
right to left then it gets an exponent of +1, otherwise, -1. [Alternatively,
point the thumb of your right hand in the direction of the i-th arc and curl
your fingers under the thumb. If the fingers are going in the same direction
as the loop when it passes under the i-th arc, then it gets an exponent of
+1, otherwise, -1.]
Li
ih e
For each arc a; (i = 1,...,k) in G\ R pick a loop z; from the base point
that passes under a; once and has an exponent of +1. [We only picture a
small part of x; below.]
In the future we will not draw the z;’s; a label of x; next to an arc will imply
that the loop running under the adjacent arc is in a positive direction. The
circles in the drawing represent a birds-eye view of the r,’s.
(15.D.10) Exercise. a) Show that an open 3-ball with the z-axis removed
has a 1-sphere as a deformation retract.
b) Let H represent the set of points in €? with z > —e. Use part a)
and several applications of the Seifert-van Kampen theorem to show that
7™(H \ G) = Aas ek als
c) Use part b) and several applications of the Seifert-van Kampen the-

orem to show that
m™((H UAUB)\G) = an oyOESTOR a se
MS Wiens es Wri tmat Siesta = ee
where any one relation may be omitted. Furthermore w; = oe See eae

where x; corresponds to the overcrossing arc a; and 2; and x; correspond
to the arcs that meet at the corresponding undercrossing. The e’s are +1;
vj = xj} ---a5* where the xj,’s correspond to the arcs meeting at the j-th
branch point and the e,’s are +1.
At each element of R construct a small (open) rectilinear cell C; that

extends from z = —e/2 to z = —2e and remove from z = —e/2 to z = —e
a slit lying above the part of G lying in z = —e. We picture one of these
corresponding to an undercrossing (of course attach a feeler extending to p).
m(E? \ G), for a Polygonal Graph G in €3 389
(15.D.11) Exercise. a) Show that after attaching each C; we get

m((HUAU
BU (UC;))\ GY Sai. 5k} WI).
Win VIPs 1Ud}
where any on of the relations can be omitted.
b) Show that 71(E€? \ G) is the same.
(15.D.12) Algorithm. With the notation developed above
mi(E2\G) ~ {21,...,
2h} W1.-+;Wm,V1-.., U0}
(15.D.13) Example. For the graph pictured at the beginning of this sec-
tion:
mi (E? \ G) = {21,--. 510; 23 po23T7, XeT\o
Ly -T3, T3T]
Ly Te,
©2025 '21, ©1242] Tp", Ly 2522, T9Z10r4}
(15.D.14) Exercise. Show that if G is a graph (or any compact set in €3

whose complement is path connected), then 71 (€? \ G) ~ ™(S? \ G).
(15.D.15) Definition. A knot is a graph homeomorphic to S’. The group
of a knot K is ™(E€? \ K).
(15.D.16) Examples.
1. The use of the algorithm for calculating fundamental groups of knots
and complements of graphs is illustrated in the following three examples.
m(E°\K)={21; J=Z
2. Trefoil knot.
r3= —1 -1,-1
m(E3\K)2{01,02,03,04j) 2, 030125 B30, © 21}
r5
3 ; ah wee i Po
m1 (E \G)={21,02,03,24,25 ; 232224 »25 ©4%3,%5T1L, Lo I;
(15.D.17) Remark. As we established in (12.C.5), one way of deciding

that two spaces are not homeomorphic is to show that their fundamental
groups are not isomorphic. For instance, one can show that €? minus the
boundary of a 2-simplex is not homeomorphic to €* minus the trefoil knot,
if it can be shown that the group in Example (15.D.16-2) is not isomorphic
to Z. To show that these two groups are not isomorphic, it suffices to find
a homomorphic image of one group that cannot be a homomorphic image
of the other. Let A = 7(€? \ G), where G is the trefoil knot. Observe
that every homomorphic image of Z is abelian (in fact cyclic), and hence,
if a nonabelian (or a noncyclic) homomorphic image of A can be found, or
equivalently if a nonabelian (or a noncyclic) quotient group of A can be
found, then A cannot be isomorphic to Z.
(15.D.18) Reidemeister’s Trick. Suppose G is the trefoil knot and A =

m(E* \.G). Consider A/(x?/ 02,02). Since x, = 2,*, A/(z?,2%,22) has a
presentation
: St oti eo)
HEI ce een Lo, %3, L1%3X%172, T1VM3IQV3, L2L3L22) }
—_—_——”
TQ=T17%321
Since x2 = 214321, we may remove £2 from the list of generators, and

substitution of 212732, for x2 in the relations yields the presentation
ay
{Pis035 27, Tz, £10301 110321, V1 U3L1XL3%1
73, ©123X1 L311 132121}
= ae: 2 a
= (21, 2a; hq, 3, (1123) i:
It is easily seen that each coset can be represented by exactly one of

the following words, 21, 2123, (%123)a1, (%123)*, (x123)?21, e. (Note, for
example, that 732) = (2123)! = (x, 23)”.) That the group is not cyclic can
Wild Sets do Exist 391
now be checked by brute force. The quotient group A/(x?, 23,23) that we
have obtained is the dihedral group of order 6, (which also happens to be
the symmetric group S3), and is neither cyclic nor abelian. Hence, A is not
isomorphic to Z.
E. WILD SETS DO EXIST.
(15.E.1) Definition. An n-cell B is any space homeomorphic to
rm (@iy aes sen) eee ee ai tae +--+ +22 =<)
An n-sphere S is any space homeomorphic to
‘Ou ts i Rea ey ae ee e€r7ri : ait apt +274, = 1}
(15.E.2) Definition. An n-cell C in €” is tame if and only if there is a

homeomorphism h : €” -> €” such that h(C) = B”. If C is not tame,
then C is wild. An n-sphere S in €"+! is tame if and only if there is a
homeomorphism h : €"t! — €"+1 such that h(S) = S” and otherwise is
wild.
(15.E.3) Definition. A compact subset X of S” (E”) is tame if and only

if there is a homeomorphism h : S” + S” (h: E” + E”) such that h(X) is
a Euclidean polyhedron. If X is not tame, then X is wild.
(15.E.4) Remark. It is easy to show that an n-cell or an n-sphere that is

tame in the sense of (15.E.2) is also tame in the sense of (15.E.3). For n-cells
and for 2-spheres, the converse can also be shown (Rushing [1973]).
(15.E.5) Exercise. Suppose A is a polygonal arc in €?. Show that 7 (E3\A)
is trivial. Give a reasonable interpretation of a “polygonal arc” in S* and
prove an analogous result.
(15.E.6) Definition. Suppose X is a locally connected topological space

and A is aclosed subset of X. A complementary domain of A in X is defined
to be a component of X \ A.
In this section, we give examples of
(i) a wild arc in €3 (S3),
(ii) a 2-sphere in €3, the closure of one of whose complementary do-
mains in not a cell,
(iii) a 2-sphere in €? for which the closure of neither complementary
domain is a cell, and
(iv) a wild cell in €% (S*).
The latter three examples will be modifications of the first one.
In all of these examples, one proves wildness by showing that the funda-
mental group of a complementary domain is nontrivial. There are, however,
examples of wild arcs and spheres whose complementary domains have triv-
ial fundamental groups (see Fox and Artin [1948]). The proofs of wildness in
these latter cases involve far more subtle properties of 3-cells than we shall
consider.
(15.E.7) Example (A Wild Arc) (Fox and Artin [1948]). The construc-
tion of this arc involves a certain amount of tedious description. The reader
is cautioned to read the next paragraphs with the figures below in sight.
Let C’ be the right circular cylinder in the figure. The basic idea is to
embed C in each of the regions D; in such a manner that the end point of the
arcs on the left-hand boundary of each D; match up with the appropriate
points on the right hand side. On the base B_, mark three collinear points
t_, m_, and b_. On the other base B,, mark collinear points t,, m4, and
b,. Construct in C’ three disjoint polygonal arcs a, a2, and a3 whose images
are A;, Ag, and A3 with the properties that A; joins t_ to m4, Ae joins by
to t,, and A3 joins m_ to b_. These arcs are arranged as indicated in the
figure (it is possible to describe the arcs analytically, but by now the reader
should be convinced of the sterility of such exercises). Let A = A; UA2UA3.
Suppose S is the solid, double spear point obtained by rotating about
the x-axis the area bounded by z = 0, z = (x/2) +1, and z = (—2/2) +
1. Subdivide S into a countable number of pieces by the family of planes
z = +(2-—1/2™~'), where m = 0,1,.... Let Dn be the segment of S
with x coordinates satisfying (2 — 2?-") <‘x < (2 —2!-") for n > 0 or
(—2+2-") <2 < (-2+2'™) ifm < 0. Let p= (=2;0,0)handq ='(2;,0)0):
For each n, let f, be a homeomorphism from C onto D,, with the fol-
lowing properties:
(i) The simple closed curves bounding the bases of C are mapped to the
simple closed curves bounding the bases of D,. Furthermore, the
indicated orientation of the simple closed curves is to be preserved.
(ii) The bases B_ and B, are mapped respectively to the left and the
right faces of Dn.
(iii) The points fn(t+), fn(m+), and f,(b4) lie in descending order
on a vertical line through the x axis and coincide with the points
fn+i(t_), fav (m_), and fn4i(b_) respectively.
(iv) The set f,(A) projects into the xz plane as
NS where NX
indicates that the point on the solid line has greater y coordinate
than the corresponding point on the broken line.
(15.E.8) Exercise. Convince yourself that the homeomorphism f,, exists
and X = {p}U (Uz fn(A)) U {q} is the image of the arc a in the following
sketch.
In order to facilitate the use of the algorithm that gives the description
of the fundamental group of the complement of a, we redraw that figure and
label subarcs that extend from undercrossing to undercrossing as follows:
nor
We show that a is a wild arc in €?. Suppose a is tame. Then there

is a homeomorphism h : €? + €° that carries a(J) = X onto a polygonal
arc L. The map h also carries €* \ X homeomorphically onto €? \ L, and
consequently the fundamental group of €% \ X is isomorphic to 7(E° \ L),
which is the trivial group by (15.E.5). Thus, to show that a is wild, it suffices
to prove that 7,(€° \ X) is nontrivial.
To compute 71 (€? \ X), we first note that €? \ X can be expressed as a
direct: mit, Let wy XU) (Ove Di) and Y, = €°°.X,, for each 7'EN;
Define the bonding maps fn : Yn > Yn+i by letting f, be the inclusion map
for each n. It is then clear that iim(Yn, pa S
With the aid of the previous section, we have:
3 .
TY (e Wie) = TOs Os Cs (a < U < n— 1);

-1 —1 —1
Qi-1 = C1 CiCpyy, bi = Ci414iCjyyy, Cit = didi410;
pure pes, =e, Un ap tenet ean) on — 1)
Next, we observe that fn :™(E?\Xn) 3 m1 (E2\Xn41) is an injection. This

may be seen from the nature of the presentation and from the next figure,
where it is clear that the loops A,_; and A, are homotopic. The same, of
course, occurs for similar loops A_, and A_(,_1) on the left-hand side of
the spear point. These loops correspond to the relations ase rome ster| a5 (2
b- -d;,G,, =e, and bo ience = 6; boon, dearest =e. Furthermore, for
—n <i<n-—1, fn» maps a; to a;, b; to b;, and c; to c;, and of course
preserves the relations.
Since f,+ is the inclusion map, it now follows from the geometry above,
(12.F.11) and (Ap.D.9) that 7,(€? \ X) is isomorphic to
aaaare Z ; =
lima (O°
\XA; (a; ber, @e Z); ap cncer,
b=
a =i
Cit 14iCi44, Ci-1 = bibi41;
Ay ey
; b; ajc; =e
;
(i Se Z)}.
We show that 7 (E° \ X)is nontrivial by finding a nontrivial homomor-

phic image of it. First, however, note that the generators b; and a; may be
removed since (x) a; = CiCi-16; and (*) bj = ci414;C;,ize This, of course,

also leads to the disappearance of these two relations. The remaining two
relations (xxx) ci41 = bibi4i1b; 1 and (4k) by ajc; = e may be rewritten as
follows: ens
= kkk = ae
1. cit = bibigaby? "=" (aies)(aig.1¢:41)(cj az")
(x)
= ican)
(oz c7*)
= €i6) 167,46:0; 6; |
DApSiagoby
ox aaab: 8)
= aybrag pee ety (Koved
(ennaegs) aed Gee
= qee c) ic;Seal
Cry
Note that in both cases 1 and 2 we have that cjc;_1Ci41C; = Ci41CiCj_-1.-
Thus, it follows from (Ap.C.9) and (Ap.C.13) that
m(E* \ X) = {e; (i € Z); Cie Tr Cicse Cr cc tee. 2)
Now define (for no immediately obvious reason) h : 7 (E32 \ X) > Ss (the

symmetric group on five symbols) by
‘ (12345) ifiis odd,

; (14235) ifiis even.
Since h(c;¢;-1¢441 QA alone) = e, h is a well defined homomorphism by

Exercise (Ap.C.3). Furthermore, the image of h is nontrivial which concludes
the proof. It should be observed that if X is considered as a subset of S°,
then by (15.D.14), 7,(S? \ X) is isomorphic to 7 (E? \ X).
(15.E.9) Example (A wild 2-sphere in S*, the closure of one of
whose complementary domains is not a cell).
Consider the arc X in the preceding example as a subset of S*. In the

previous construction replace the arcs Aj, Az, A3 by slender hollow tubes 7},
T>, T3 similarly situated in C. If this is done carefully, the homeomorphisms
constructed in (15.E.8) will properly match the ends of the tubes when C
is mapped onto the D,’s. Then Y = {p}U (Ufi(T;)) U {gq} is the desired
2-sphere.
(15.E.10) Exercise. Convince yourself that the closure of one of the com-
plementary domains of Y is a 3-cell.
(15.E.11) Remark. It follows from (15.E.4) that both complementary
domains of a tame 2-sphere S in €3 have trivial fundamental groups.
Let D be the complementary domain in the previous exercise whose
closure is a cell. Since it is clear that S* \ D is a deformation retract of
S? \ X, we have that 1(S? \ D) = m™(S? \ X) # {e}; consequently, S3 \ D
is not a cell, and by Remark (15.E.11), Y is a wild 2-sphere.
(15.E.12) Example (A 2-sphere in S* for which the closure of nei-
ther complementary domain is a cell).
In this example, push a bubble on the 2-sphere into the complementary
domain that is a cell. Now cut off the bubble and graft in its place another
copy of the wild sphere (minus a disk) to obtain the following:
Here the closure of neither complementary domain is a 3-cell.

(15.E.13) Example (A Wild Cell). The cell given in (15.E.10) is wild
(why’?).
Problems 397
PROBLEMS
Section A
Show that (S',S? \ {(1,0)}) fails to have the AHEP.

A space X is binormal if and only if X x I is normal. If A is a closed
subset of a binormal space X such that (A x I) U(X x {0}) U(X x {1})
is an ANR, show that (X, A) has the AHEP.
A pair (X, A) has the homotopy extension property (H EP) with respect
to a space Y if and only if for each continuous map F': X > Y and for
each homotopy H : Ax I + Y with Hp = Fy, there is a homotopy
H:XxI-+Y such that H extends H and Hy =F. Suppose A is a
closed subset of X and B is a subset of Y. Suppose further that (X, A)
has the HEP with respect to B and (X x I, (X x {0,1}) U(A x J)) has
the HEP with respect to Y. Show: if f : (X,A) — (Y, B) is homotopic
tog: X — B, then f ~ g.
Find an example of a space Y such that the pair given in problem 1 has
the HEP with respect to Y.
Let A be a closed subset of a metric space X. Show: if f,g: A> S”
are homotopic and if f can be extended to F : X — S”, then g can also
be extended to X.
SupposeX is an ANR and A is a closed subset of X. Show that
(X x {0}) U(A x J) is a retract of X x I if and only if A is an ANR.
Suppose X is an ANR and A is a closed subset of X. Show that (X, A)
has the AH EP if and only if Ais an ANR.
Suppose X is a topological space. Show: (Cone X, X) has the AHEP.
Section B
Suppose A is a compact subset of €”. Show that €” \ A has precisely

one unbounded component.
Show that €” and €” are not homeomorphic, and extend this result to
manifolds.
Show that S” is not homeomorphic to a proper subset of itself.
Suppose A Cc €" and h: A > E” is an embedding. Show: if x € A,
then h(x) € int h(A).
f
[aba Suppose K C €” is compact. Show: K does not separate €” if and only
if C(A,S”) ( with the compact open topology) is path connected.
6. Suppose K Cc €” is compact and D is a retract of K. Show that €” \ D

has no more components than €” \ K.
7. Suppose S is an n-sphere lying in €"*!. Show that €"+! \ S has finitely
many components.
8. Suppose K is a compact subset of €” and x € €” \ K. Show that the
component of €” \ K that contains x is unbounded if and only if ~p|x
is null homotopic.
9. Suppose K is acompact subset of €” and D C K is a deformation retract
of kK. Show there is a 1-1 correspondence between the components of
€” \ K and the components of €” \ D.
10. Let M and N be n-manifolds, U an open set in M, andh: U > N an
embedding. Show that h(U) is open.
11. Suppose U is an open subset of €” and f : U > €” is continuous and
1-1. Show that f is an embedding and f(U) is open.
12. Find an example of a function f : €! > €? that is continuous and 1-1,
but for which the image of some open subset of €! is not open in f(€').
13. If m < n, show there is no continuous 1-1 function mapping an open
subset of €” into E™.
Section C
1. Show: if the union of two continua C; and C2 is S" (n > 1), then CyNC2
is connected.
Show that any contractible, locally path connected space is unicoherent.
Is the space described in (12.C.8) unicoherent?
Suppose X is a compact metric space. Give Z = C(X, S') the compact-
open topology. Show: if Z is path connected, then X is unicoherent.
5. Show: if U; and U2 are connected open subsets of S* and FrU; N Fr U2
is connected, then U; M U2 is connected.
Section D
1. Find a presentation for the group of the following knot, and prove the
group is trivial.
LES
Problems 399
2. Find a presentation for the group of the following knot, and show the
group is nontrivial.
EIS
knot group is nontrivial.
4. Show that the groups obtained in problems 2 and 3 are not isomorphic.
group is nontrivial.
Ss
6. Show that the fundamental group of the complement of the following
graph is {a,b; }.
7. Find a presentation fof the group of the following know and show that
it is differest from the group of the trefoil.
8.
OOOO
Show: if K is a knot, G is the group of K, and [G, G] is the commutator
subgroup of G, then G'/[G, G] is infinite cyclic.
Chapter 16
2-MANIFOLDS
In this chapter, we focus attention on some of the basic properties of 2-

manifolds. Our investigation, while of interest in itself, also serves as an
introduction to some of the techniques used and problems encountered in
working with higher dimensional manifolds. Nevertheless, the reader should
be warned that mastery of the 2-manifold does not presage any great com-
petence for handling its higher dimensional counterparts. Just as we will
see that 2-manifolds are far richer than the 1-manifolds, the gap between
2-manifolds and 3-manifolds is even greater, and a 69-manifold is something
to behold.
We will soon be in a position to classify all compact 2-manifolds. How-
ever, we first establish that all 2-manifolds can be triangulated. This is an
old result of Rado [1925]. A similar result holds for 3-manifolds, but its
proof is more difficult—the intersections of 3-cells being more awkward to
analyze than the intersections of 2-cells. The status of n-manifolds, n > 4,
is somewhat complicated—although most can be triangulated.
A. THE TRIANGULATION OF 2—MANIFOLDS

The initial goal of this section is to establish a simple criterion that enables
one to determine whether a connected open subset of €* is homeomorphic to
E?. Recall that a 2-cell is any space homeomorphic to B? and (in the plane)
every simple closed curve bounds a 2-cell.
401
402 16. 2-Manifolds
(16.A.1) Theorem. If M is an open connected nonempty subset of Ee

with the property that every simple closed curve in M bounds a 2-cell in M,
then M is homeomorphic to €?.
Proof. Since M is o-compact (connected manifolds are o-compact by
(10.A.18)) there is an increasing sequence of compact sets, K1, K2,... that
covers M (10.A.16). By (5.C.28), there is a simple closed curve 7 in
M that encloses K, in its interior, I(7). The same exercise may be ap-
plied again to obtain a simple closed curve 72 such that Ko U1 C I(q2),
and inductively, for each integer n, there is a simple closed curve yp in
M with the property that Kn UYn-1 C I(yn). Since the region between
two nested simple closed curves is a 2-annulus (5.E.5), a homeomorphism
between M and €? can be constructed as follows: for each integer n, let
B? denote the disk of radius n lying in €? with center at the origin. Let
H, : I(y,) > B? be a homeomorphism (5.D.10). Again proceed inductively.
Suppose Hy : (Yn) — B? has been defined. Then there is o homeomor-
phism Hnat : I(Yn4i1)\I(n) B2Ge\ Be such that a, H-eee. is a
homeomorphism from B? onto itself. The map Hy H-ntl p.

can a ex-
tended to a homeomorphism g : B?,, \B2 — B2,, \ B? en Define

An+1 : I(Yn41) ov Baa, by setting
Hy (a) if x € I(Yn)
Ay+1(£) = {
9H nsi(z) if 2 € Tent \I(%n)
Then h: M -+ €? defined by h(x) = H,(x) whenever x € I(yn) is the
desired homeomorphism.
(16.A.2) Exercise. Show that h is a homeomorphism.

(16.A.3) Corollary. Suppose U is a nonempty open simply connected
subset of €?. Then U is homeomorphic with €?.
Proof. Suppose ¥ is a simple closed curve in U. We show that y bounds a
disk in U. Let f : S' + y be ahomeomorphism. Since U is simply connected,
by (12.C.11) f may be extended to a continuous map f : B? = U. We show
that I(y) C U. Let p be an arbitrary point in I(y), there is a retraction r
The Triangulation of 2-Manifolds 403
that maps €? \ {p} onto y (why?). Suppose p ¢ U. Define h : B? + S! by

h(x) = f~!rf(x). Then h retracts B? onto its boundary, which contradicts
(12.C.15). Thus, every point in I(y) lies in U, and hence, each simple closed
curve in U bounds adiskinU. 0
The next goal is to show that all 2-manifolds are triangulable. Since
compactness will not be assumed here, the more general definition of tri-
angulation (14.4.16) is applicable. The conditions imposed on the disks in
the next lemma are sometimes abbreviated by saying the disks are in “gen-
eral position.” The triangulation of 2-manifolds is an early example of the
exploitation of general position. A few additional lemmas are convenient.
(16.A.4) Lemma. Suppose M is a 2-manifold. Then there is a sequence of

disks D,, D2,... whose interiors cover M such that
(i) if i A 7, then FrD;N Fr D; consists of at most a finite number of
points, and
(i) tee 7 Ak AT, then’ fr D; A FrD; NF Dz; =v:
Proof. Since M is a separable metric space, we can cover it with a
countable collection of open disks F,, E2,.... For each i, choose a nested
sequence of closed disks in E; whose union is £;. Sequentially order the
collection of all these disks and denote them by Df, Dj,....
The desired collection is defined inductively. Let D; = Df. Suppose
D,,...,Dn have been chosen so that

(i) if i #7, then FrD; N Fr D; consists of a finite number of points,
(ii) ifi #47 #k Zi, then FrD,;N FrD;N FrD, = 9G, and
(iii) D¥ C D, for each 1.
DefineF,, to be U_, Fr D;. Now note that D7,, is a subset of Ey for at least
one integer k. Let h: Ex — int B? be a homeomorphism and now define K
404 16. 2-Manifolds
to be h(D*,,) U {h((FrDi N FrDj) N Ex) : 1 A j, and i,j < n}. Since K

is compact, we have d(K, FrB?) > 0 and hence there are circles S; and S2
in int B? such that K C I(S) C I(S1) C I(S3). Choose points p and q in
(I(S2) \ I(S2) \ h(F,z M Ex) that do not lie on the same diameter, and draw
radial lines through p and q. These radial lines cut (I(S2) \ [(S1)) into two
cells Cy and C2, which intersect along the radial lines.
Claim. There are arcs a; and a2 from p to q that lie in Cy and C2,
respectively, and which intersect h(F;, MN E,) in at most a finite number of
points. Note that if such arcs can be found, then the union of these arcs
forms a simple closed curve and Dny; = h~!(I(a1 Uaz)) will satisfy the
inductive requirement.
Proof of claim. We will say that an arc in M is admissible if and only if
its intersection with F’, is finite. A point in M will be called a crossing point
if and only if it is on the boundary of more than one D;. Observe that if
P1,P2,--+,Pq is a collection of points in M and if for each i = 1,2,...,q—1,
there is an admissible arc with end points p; and p;,,, then there is an
admissible arc with end points p; and p,. Furthermore, it is easy to see that
every noncrossing point has a neighborhood in which any two points lie on
an admissible arc. (Use (5.D.9), for example.)
In C, construct two small admissible arcs 3; and G2 with end points
{p, zi} and {q, z2}, respectively, such that (31 \ {p}) U (Ge \ {q}) C int C).
Since each point x € int C) is a noncrossing point, there is an arc connected
neighborhood N, of x lying in int C; such that any two points of Nz may
be joined by an admissible arc. By (2.F.2), there is a simple chain of the
N,’s between z; and zg. It should be obvious how to obtain the admissible
arc a; from p to q. Of course, an identical procedure may be used to join
p and q in C2 with an admissible arc ag. As indicated previously, this not
only completes the proof of the claim but also of the lemma. O
(16.A.5) Definition. A punctured disk is any space homeomorphic to B?

minus the interior of the union of a finite number of closed disjoint cells lying
in int B*. (A real disk is also assumed to be a punctured disk.)
(16.A.6) Exercise. Suppose P is a punctured disk. Show that P may be

written as the union of a finite number of 2-cells with disjoint interiors.
The Triangulation of 2-Manifolds 405
(16.A.7) Lemma. Suppose M is a 2-manifold and D = {Dj, D»,.. bis a

countable collection of punctured disks such that
(i) the interiors of each pair of these disks are disjoint, and
(ii) D is a locally finite cover of M (i.e., each point of M is contained
in at most a finite number of punctured disks in D).
Then M may be triangulated.
Proof. By (16.A.6), each punctured disk may be broken up into a finite
number of disks with disjoint interiors. Let C),C2,... be the family of all
disks obtained in this fashion.
A point in M is called a verter if it lies on the frontier of three or more of

the C;. More vertices are introduced as follows. A vertex v; is selected in the
interior of each disk C;, and if necessary, vertices are added to Fr C; to ensure
that each Fr C; has at least three vertices.
It is immediate from the construction and from (ii) of the hypothesis

that C; intersects at most a finite number of other disks, and consequently
each disk contains only a finite number of vertices. In each C;, connect v;
with each vertex on the frontier of C; by arcs that are disjoint except at vj.
406 16. 2-Manifolds
A curvilinear 2-simplex will be the region bounded by a 1-sphere composed

of these arcs with one vertex at v; and which does not contain any arcs in
the interior of its closure. Thus M has been triangulated. O
(16.A.8) Theorem. Any 2-manifold M can be triangulated.
Proof. It suffices to find a cover of M by punctured disks that satisfies
the hypotheses of (16.A.7). Use (16.4.4) to do this. Suppose D,, D2,... is a
sequence of disks satisfying properties (i) and (ii) of (16.A.4). Let D; = D,.
Note that Fr D, “decomposes” Dz into a finite number of punctured disks
whose interiors are disjoint. Let Di card De denote those punctured disks
not in D,. We illustrate two of the possibilities in the next figure.
iD D2
X 2
Similarly, D3 is broken up by Fr D, UF Dy into a finite number of punctured

disks.
Inductively, at the n-th stage, D,, is decomposed by Uir Fr D; into a finite

collection of punctured disks. These are labeled D1, D?,... , Dm, Then
the family of punctured disks D = {Di Lire Lt < 7 =< m,} covers
M. Furthermore, any two elements of D have disjoint interiors. For local
finiteness, note that if z € M, then x € int D, for some n, and int D, can
intersect only those BA ’sfor which l<k<n. O
(16.A.9) Definition. A separable metric space M is a 2-manifold with

boundary if and only if each point z € M has a neighborhood homeomorphic
to B®. The interior of M, denoted by Int M, is {x € M : « has a neighbor-
hood homeomorphic to €?}, and the boundary of M, denoted by Bd M “AS
M \ Int M.
The Classification of Compact 2-Manifolds 407
(16.A.10) Exercise. Show that any compact 2-manifold with boundary

can be triangulated.
It would be of some value for the reader to determine which parts of
the foregoing construction generalize to 3-manifolds, and which parts are
hopeless.
B. THE CLASSIFICATION OF COMPACT 2-MANIFOLDS

By the end of this section we will have an elegant characterization of com-
pact connected 2-manifolds, as well as an algorithm that permits “easy”
computation of their fundamental groups.
Suppose M is a compact connected 2-manifold, and T is any (finite)
triangulation of M. Let S be a collection of disjoint 2-simplices lying in €?
that are in 1-1 correspondence with the 2-simplices of M. For each 2-simplex
a? € S, let hz2 : 07 + M be an embedding that maps o? onto the 2-simplex
of M to which it corresponds in such a manner that vertices are sent to
vertices.
We define a ay € of equivalence relations. Assign the relative topol-
ogy fou8 a=Ute" eT ee S}. List the faces of elements OLS 0h Ons: +» Fan:
If oj and o} are faces eeop and oF then declare x € o} pa lene toy - Oo} if
and only if hg2(x)= h,2(y). This induces an equivalence relation that we de-
note by Rij, which te be trivial. Let € = {Rij : 1 <i,j < 3n and Rj; not
trivial}. Let R be the equivalence relation generated by U{Ri; : Pigg CMe.
It is easy to see that $/ R is homeomorphic to M.
We select a subset € of € with the property that if R is the equivalence
relation generated by U{Rij : Rij € é'}, then S/R is a polygon. We define
€ inductively.
Suppose a? € S. If i, is a face of o?, then there is a unique 2-simplex
a3, one of whose faces oj, is identified with oj, (why?).
oe
2
Q
If the total number of 2-simplices in S is ered Ba 2, then the quadri-
lateral Q feanee from the identification of oj, with oj, (by Ri,;,) will poe
at least one face o}in 3available for appropriate identification with a face of OF,
of a third simplex a? in S. Otherwise M would not be connected! Let Rj,;,
be the relation responsible for this identification.
1
Tiss
_—
1
Tio
2
os
408 16. 2-Manifolds
This procedure is continued (n — 1 times in all) until the simplices of S are

exhausted. Let é ={ Rijs kh = 1)2).. 7) 1} denotettheret of relations
used. If é is the equivalence relation generated by U{Ri,j, : Riz. € € };
then S$/€ is homeomorphic to a regular n-sided polygon lying in the plane.
Note that n must be even (why?).
U7 U6 U5
In order to obtain a 2-manifold homeomorphic to M, we need only identify

the sides of the polygon, making use of the equivalence relation generated
by U{ Rij : Riz € €\€}.
Thus, we have obtained the rather spectacular result that every compact
connected 2-manifold may be obtained by identifying appropriate pairs of
sides of a regular polygon. For higher dimensional triangulable compact
connected manifolds, a similar result holds with slight adjustments in the
proof. What is truly amazing is that there is an analogous theorem due
to Brown for compact connected n-manifolds which does not make use of a
triangulation (17.E.1).
Suppose P is a regular (2n)-polygon obtained as above from a 2-manifold.
We obtain a ‘word’ w that denotes the remaining identifications as follows:
Imagine that P is oriented clockwise, and consider two sides of P that are
equivalent. We give the same label (usually, a or b or etc.) to these sides.
Orient one of these sides arbitrarily and the other side so that the orienta-
tions will agree when the sides are identified. If the orientation of a side a
agrees with the orientation induced by P we place the letter a in the word w.
If the orientation of a is opposite to the induced one, we insert a~! into w.
The word w is obtained by starting anywhere on the polygon and walking
once around in a clockwise direction. Hence for the following picture, a word
representing the identifications would be ab~!cba~!c, as well as cba~!cab7!
(or any cyclic permutation).
By use of this convention we can interpret a collection of letters, e.g.,

aba~‘cbd~*cd~1, as specifying an identification of the sides of a polygon. We
will write M = aba'c--- to mean that the polygon with sides identified via
aba~'c--- is homeomorphic with M. Observe that each letter must appear
exactly twice, and any compact connected 2-manifold will have at least one
representation as a sequence of letters.
The key to the classification of compact 2-manifolds is that each se-
quence of letters can be put into a “normal form” and that different patterns
of letters in normal form yield distinct 2-manifolds.
(16.B.1) Remark. The sequence of letters aa~! will represent a 2-sphere,
2-sphere Projective plane
and the sequence aa will represent the projective plane. All other compact
connected 2-manifolds will have at least two distinct letters in their repre-
sentation.
(16.B.2) Notation. In the following, a, b, c, ... will denote sides of poly-

gons and Greek letters a, 3, y, ... will denote blocks of Roman letters, for
example, a = ab~!c. If a is such a block, then a~! will denote the block
obtained from a by reversing the letters and changing the exponents, e.g., if
a = abb~!cac~, then a~! = ca~!c71bb~!a7!.
A consequence of the following sequence of Lemmas (whose proofs are
sketched, literally), is that compact connected 2-manifolds may be repre-
sented by normal forms.
The normal forms are obtained through a series of cutting and pasting
operations. For instance, the following figure
a
may be “cut in two”

410 16. 2-Manifolds
and then pasted together again along b to yield the next figure.
a
(16.B.3) Lemma. If M = aa“! (where 6 # 0), then M = B.

Proof. A map w is constructed between the two polygons that is simpli-
cial on some subdivision of aa~! and on ZB. The description of such a map
is illustrated in the next figure. Clearly, ~ induces a homeomorphism of M
onto itself.
(16.B.4) Lemma. Suppose a compact connected 2-manifold M is repre-

sented by a polygon with more than two sides. Furthermore, suppose P is
not of the form aaa~1. Then there is a polygon P also representing M for
which all of the vertices (on the boundary of P) are identified.
Proof. Suppose the vertices of P lie in distinct equivalence classes
Vo,Vi,.--,Vg. We obtain a polygon P representing M for which there is
one more vertex in class Vo and one less in one of the remaining classes. Let
vo € Vo be a vertex adjacent to a vertex v; € V;, where i # 0. Let a = v,v
be the 1-simplex on the other side of v;, and let @ be the side of P that
is equivalent with a. Then vpv; # a’, since by hypothesis it is assumed
that all such configurations have been removed. Furthermore, vgv; # a, for
otherwise v9 would be equivalent to v;, which contradicts the way v; was
selected.
Depending on the orientation of a and a one of the following two situa-
tions must occur.
Note that in either case, after cutting and reglueing, there is now an
additional vertex vo in Vo, and the two v;’s have become one.
Of course, under the maps indicated above, a configuration acc! may
have been introduced; however, by (16.B.3), such configurations can be re-
moved without increasing the number of elements in any of the equivalence
classes. Repetition of the foregoing procedure will eventually place all of the
vertices in the equivalence class containing vp. O
From this point on, we assume that if P is a polygon representing a
2-manifold M, than all the vertices of P belong to the same equivalence
class.
(16.B.5) Lemma. If M = ac@c, then M = aaaf-!.
Proof.
(16.B.6) Lemma. If M = aafbya~16b-!, then M = yry~12~dyBa.

Proof.
a a
a ; af oe a? af es
ib
fy
(16.B.7) Exercise. If a and ( have no letter in common and all vertices

of P are identified, show that configurations such as aaa! cannot occur:
the “single” vertex would not have a 2-cell neighborhood.
412 16. 2-Manifolds
(16.B.8) Lemma. If M = ccabab~'a~*, then M = eef fgga1B-}.

Proof.
Integrating the preceding five lemmas with the exercise, we have the
principle result.
(16.B.9) Theorem. Every representation of a compact connected 2-manifold
has a normal form of
(i) aa~? or aa, or
(ii) aabbccdd..., or
(iii)wabas?
b=*cdem* das. «. Oo
Thus, if a compact connected 2-manifold is not a 2-sphere or a projec-
tive plane, it may be described either as a collection of projective planes
glued together, or as an n-holed torus. The forms just described are called
normal forms. It remains to show that distinct normal forms are associated
with distinct (i.e., nonhomeomorphic) manifolds. We accomplish this by
demonstrating that manifolds with distinct normal forms have nonisomor-
phic fundamental groups. é
(16.B.10) Exercise. Use the polyhedral Seifert-van Kampen theorem to
establish that the fundamental group of a 2-manifold, represented by a 2n
polygon with identifications via the word a is {x1,22,23,...,2n : a}, where

the z;’s are the n symbols in the word a (unless a = 2; 2)~).
If M and N are 2-manifolds with the same normal form, then it is easy to
construct a homeomorphism between them. If M and N have distinct normal
forms, we will show that they are not homeomorphic, thus completing the
classification of compact connected 2-manifolds. What we actually prove is
that {71,...,%, : a} and {y1,...,Ym : 8} are not isomorphic whenever a
and @ are different in normal form.
We need the following theorem from group theory.
(16.B.11) Theorem. If G and H are isomorphic groups, then G/|G;G]

and H/[H; 4H] are isomorphic. ([S;.S] denotes the commutator subgroup of
S.) Oo
It follows that if G/[G; G] and H/|H; H] are not isomorphic, then G and
H fail to be isomorphic.
(16.B.12) Theorem. Compact connected 2-manifolds having distinct nor-

mal forms are not homeomorphic.
Proof. If a = £121 2%2%2%3%3°+*XLnXn, then
Lid Wy 5Ai9)
id igh eX}
= (iy Byte
an ea ete ya ye
by the Tietze equivalence IJ (Ap.C.13). When we consider A/[A; A], the first
relation a is equivalent to x”, and the second relation can be used to eliminate
the generator z, (Tietze equivalence J (Ap.C.8)). Hence, abelianization
yields Z@Z@---@ZO Zo, where there are n — 1 factors of Z. Note that if
n = 1, we obtain Zo.
If
hE @1 220, a2 'a3r403 24° + pein th
then {x1,Z2,...,%nj;a@} when abelianized is just Z6 Z@®--- @ Z (since the

relation becomes trivial), where there are n factors of Z.
Recall that 7,(S?) = {e}. Therefore S? is not homeomorphic to any
other compact 2-manifold. Hence, if a # 6 then {z22,...,%, : a} and
{z1,£2,.--,2Zm : 2} cannot be isomorphic, and the corresponding manifolds
are not homeomorphic. O
(16.B.13) Remark. One should note that it is now possible to determine

whether any two compact 2-manifolds M and N are homeomorphic. Since
M and N are compact, they have only a finite number of components, and
each such component is a compact connected 2-manifold. The manifolds M
and N will be homeomorphic if and only if there is a 1 to 1 correspondence
414 16. 2-Manifolds
between the components of M and the components of N in such a way that

corresponding components are homeomorphic.
(16.B.14) Remark. Success in classifying compact 2-manifolds has led to

attempts to carry the work over to compact 3-manifolds. It is relatively easy
to show that every compact connected 3-manifold is a 3-cell with triangles
on the boundary identified. Unfortunately, no one has devised a neat scheme
for deciding what a “normal form” would be. A special case that is of some
interest yields the lens space (Brody [1960]; Reidemeister [1935]).
A somewhat different but related approach to 3-manifolds is to remove
a maximal open 3-cell C from a 3-manifold M. This leaves the identified
boundary of C, which is called a spine of the manifold. Unfortunately once
again, since a manifold has many possible spines, the problem becomes one of
finding a workable “normal form” (Zeeman [1963]). There are several other
approaches to the classification problem. For example: see Stillwell [1980]
for a description of Heegard splittings; see Haken [1962], Waldhausen [1976],
or Hemion [1992] for the classification of Haken manifolds.
C. A CHARACTERIZATION OF &£
In this section we employ the triangulability of 2-manifolds and our knowl-

edge of compact 2-manifolds to obtain a characterization of €?._ We show
that any simply connected noncompact 2-manifold is homeomorphic to €?.
The following two results are used.
(16.C.1) Exercise. Suppose M is a compact 2-manifold with boundary.

Show that each boundary component of M is a simple closed curve, and fur-
thermore, each boundary component can be collared, i.e., if C is a boundary
component of M, then there is a simple closed curve D C IntM and an
embedding h : S' x I -> M such that h maps S! x {0} homeomorphically
onto C, h maps S' x {1} homeomorphically onto D, and h(S! x I) is a
neighborhood of C' in M.
(16.C.2) Exercise. Suppose M is a compact 2-manifold with boundary

and T is a triangulation of M. Let T“) and T) denote the first and second
barycentric subdivisions of T, and let B = {b € M : bis a barycenter of a
0, 1, or 2-dimensional simplex in T}. Let S be the family of closed stars of
the points b € B with respect to T’?). Show that if S’ is any subfamily of S,
then U{S : S € S’} is a 2-manifold with boundary.
(16.C.3) Theorem. If M is a simply connected noncompact 2-manifold,
then M is homeomorphic to €?.
The proof is broken up into three lemmas.
A Characterization of €2 415
(16.C.4) Lemma. Suppose M is a triangulated simply connected non-

compact 2-manifold with boundary, and N C M is a compact connected
2-manifold with boundary. Then N can be embedded in €?.
Proof. Let C = {C,,C2,...,C,} be the boundary components of N. By
(16.C.1), the C;’s are simple closed curves. To each C; we attach a disk as
follows. Let A = {1,2,...,n} be given the discrete topology, and for each
i € A, let B; = B?, S; = FrB;, and f; : S; + C; be a homeomorphism.
Define f : S' x A + N by f(z,i) = fi(x). Then the adjunction space
N = (B? x A)U; N may be viewed as the manifold N with disks attached to
each boundary component. Note that N is a compact connected 2-manifold
without boundary. If we can show that N is simply connected, it will follow
from (16.B.9, 10, and 12) that N is homeomorphic to S?. Then N can be
embedded in €?, since it is properly contained in N.
To show N is simply connected, we first triangulate N. Use (16.A.9)
to triangulate N, and then take the cone from the center of each disk that
was attached to a boundary component of N. We now use (12.C.11) to show
that N is simply connected. Let g : S! + N be a continuous map. We
wish to extend g to B?. Consider S! as a triangulated 1-complex and apply
(14.C.12) to obtain a map g: S! + N that is simplicial on a subdivision of
N and is homotopic to g. We first extend g to B? and then show how this
extension may be used to extend g to B?.
For each boundary component C; on N, let K; be a collar for C; in
N (16.C.1) and let D; denote the disk attached to C;. Since g is simplicial
and each boundary component is collared, it is not difficult to see that g
is homotopic to a continuous function that maps S! into Int N. Thus, we
may assume that g also has this property. However, Int N is a subset of the
simply connected space M and hence g may be extended to a continuous
function G : B? — M.
Let U be the component of B? \ U%_, G~!(C;) that contains S!. Let
{X; : j € L} be the set of components of the frontier of U (in B*). Note
that for each j, there is an i; € A such that X; C G7'(C;,). Furthermore,
for each j € L, there is an open set U; and by (5.C.28) a simple closed curve
S; such that X; Cc 1(§;) C 1($;) US; C U; C G-4(K;,). We may assume
that the 5;’s are mutually disjoint. _ p
For each j7 € L, we have that Gl¢, pigueh De UW: GisUKiys Since
D;, UCi; U Ki; is a disk, Gg, may be extended to a continuous function
Gi ; (S; UI(S;)) aA Di; U CG; U Ki,. Now define G : joke 6 N by
G(z) if c € BP\ jez 1(S;)

ae)=|
G(x) if2€S;UI(S;U1(S;))
416 16. 2-Manifolds
Then G extends g. It follows from (4.B.21) and (15.A.1) that g may also be
extended to B? (also see problem 13 of section B, and C); hence, N is simply
connected. Therefore, by (16.B.9) and (16.B.10) N is homeomorphic to S?,
and since N is a proper subset of N, N can be embedded in €?.. O
(16.C.5) Lemma. Suppose N; and N2 are compact connected 2-manifolds

with boundary, N; C No, and f;: Nj > S? (i =1, 2) are embeddings. Then
there is an embedding fz : N2 — S? such that foln, = fi.
Proof. Let Jj, J2,...,Jn be the boundary components of Ni. By
(16.C.1), each J; is a simple closed curve. For each i = 1,2,...,n, let D; be
the disk in S? bounded by f2(J;) with the property that DiN fo(M1) = fo(Ji)
(why does such a disk exist?). Note that S? = f2(Ni) U(UD;). Then
fils Visco : fo(Ji) 2 fi(Ji) is a homeomorphism. Extend fife |fo(a:) to
a homeomorphism g; : D; > E;, where E; is the disk in S? with boundary
fil Ji) such that £;9 fi(™1) =a fe
Define fo : No 3 S? by
| gifola) if x € [fy1(Di fo(N2))].

Then f2 is the desired embedding. O
We now use these lemmas to supply a proof for Theorem (16.C.3).
(16.C.6) Lemma. If K is a compact connected subset of a 2-manifold M,

then there is a compact connected 2-manifold with boundary N such that
kK CintN GM.
Proof. This follows immediately from (16.C.2) and the fact that K is
compact and connected. O
Proof of (16.C.3). Let T be a triangulation of M and list the closed

2-simplices in T, s1,$2,... (why are there only a countable number?). Let
N, = 8, and let f; : Nj ~ €? be an embedding. If Ni N so # O, let
Ky = N, Use. If Ni and sg have empty intersection, let a; be an arc in
M with one endpoint in N, and the other in so. Then Ky = Ni, Uay U 82
is a connected compact subset of M and by (16.C.6), there is a connected
compact 2-manifold with boundary N2 in M that contains Ko in its interior.
If s3 M No = @, connect these two sets with an arc a2, as before, to form
a compact connected subset K3 = s3 Uag U Nog of M (if s53N No ¥# O,
let K3 = s3U No). Again, (16.C.16) may be.applied to obtain a compact
connected 2-manifold with boundary N3 that contains K3 in its interior.
Thus, there is a sequence of connected compact 2-manifolds with boundary
N,,No,... such that for each 1, Nj C Int Niy1 and U2, Ni; = M. By
(16.C.4), each N; may be embedded in €? via an embedding f;. Lemma
Problems 417
(16.C.5) may be inductively employed to “fix up” the f;’s so as to yield an

embedding F : M — €?. Clearly, F(M) is embedded as an open simply
connected subset of €?, and hence by (16.A.1), F(M) is homeomorphic to
ous
PROBLEMS
Section A
1. ‘Triangulate the torus. (Hint: Triangulate the unit square in a suitable

manner and identify edges.]
2. Triangulate the Klein bottle.
3. Triangulate the projective plane.
4. Triangulate the Mobius strip.
5. Show: if T is a triangulation of a compact 2-manifold, then T has only
finitely many simplices.
6. Suppose T’ is a triangulation of a compact 2-manifold. Let
v = number of vertices in T
e = number of 1-simplices in T
¢ = number of 2-simplices in T’
Define €(M) = v—e+t. The number €(M) is called the Euler char-
acteristic of M. It can be shown that €(M) is a topological invariant
(it does not depend on the triangulation used). Assume that to be the
case and find the Euler characteristic of the torus, the Klein bottle, and
the projective plane.
7. Suppose M is a compact connected 2-manifold and T' is a triangulation
of M. With the notation of the preceding problem, show that
(a) ‘St=e;
ibe = ee — €(M));
(c) vu> 4(7 + \/49—24€(M
Sections B and C
Suppose M and N are compact 2-manifolds. In each space, remove a

small open disk and let h be a homeomorphism from the boundary of
the disk removed in M onto the boundary of the one removed in N.
Then MU, N is called the connected sum of M and N. It can be shown
418 16. 2-Manifolds
that the connected sum does not depend on the disks removed or on the
homeomorphism used. Find the connected sum of two 2-spheres.
Suppose M and N are connected compact 2-manifolds. Let S be the
connected sum of M and N. Show that S is a compact 2-manifold.
Show that the connected sum of a torus and a projective plane is home-
omorphic to the connected sum of three projective planes.
Show that the connected sum of two projective planes is homeomorphic
to a Klein bottle.
Show that the connected sum of a Klein bottle and a projective plane is
homeomorphic to the connected sum of a torus and a projective plane.
Show that the connected sum of the Mobius strip and a torus is home-
omorphic to the connected sum of a Mobius strip and a Klein bottle.
Find the Euler characteristic of the connected sum of a projective plane
and n tori.
Show: The Euler characteristic of the connected sum of two 2-manifolds
is equal to the sum of their Euler characteristics.
What is the universal covering space of S??
Show that the universal covering space of a compact 2-manifold is either
Se One.
Show: if M is a compact 2-manifold and M is neither S? nor the pro-
jective plane, then the universal covering space of M is €?.
12. Let S be the boundary of a 2-simplex. Suppose f is a continuous map
from S into a triangulated 2-manifold with boundary and let «€ > 0 be
given. Show that there is a simplicial map g: S > M that is homotopic
to f such that
(i) g is a local homeomorphism, and
(ii) the homotopy between f and g does not move points more than e.
13. Suppose X is a topological space and f, ie : S' + X are homotopic.
Show: if f has a continuous extension g : B? + X, then f may also be
extended to B?. (Use no cannons!)
Chapter 17
AN INTRODUCTION TO
nN—-MANIFOLDS
Investigations of n-manifolds constitute an important body of research in

topology today. Although the results that we include are not trivial (indeed,
they include some rather profound theorems) we nevertheless barely scratch
the surface of manifold theory.
A. SOME PRELIMINARIES
In Chapter 1, an n-manifold M was defined as a separable metric space with
the property that every point in M has a neighborhood homeomorphic to
E”. This definition eliminates such common spaces as J, half-open inter-
vals, disks, etc. However, these spaces do satisfy the following more general
concept.
(17.A.1) Definition. An n-manifold with boundary is a separable met-

ric space M with the property that each point in M has a neighborhood
homeomorphic with B”. The interior of M (Int M) consists of those points
having neighborhoods homeomorphic to B" \ S"~' (or, equivalently, €”).
The boundary of M (Bd M) is defined to be M \ Int M.
(17.A.2) Remark. With the foregoing terminology, and n-manifold is sim-
ply an n-manifold with boundary, whose boundary is empty.
(17.A.3) Example. Let
OLN zee 22 hy =1,=l<7 <1}
Then C is a 2-manifold with boundary where

iIntC= tape retss a4 7 = 1) “1 <2 <1},
BAC BuGhn2 ee? va? ey? = 1, |z)= 1}.
419
420 17. An Introduction to n-Manifolds
Observe that in the preceding example, Int C and BdC (in the manifold
context) do not agree with the interior and frontier of C when C is considered
as a subspace of €°.
The following theorem states that for connected n-manifolds, the sepa-
rability requirement in (17.A.1) is redundant.
(17.A.4) Theorem. If X is a connected metric space such that each point
has a neighborhood homeomorphic to B”, then X is separable.
Proof. Since X is a metric space, X is paracompact (10.A.8), and
hence, by (10.A.18) may be written as the disjoint union of o-compact spaces.
However, X is connected; consequently, X is itself a o-compact metric space.
By (10.A.17), X is Lindeléf, and therefore separable by (2.1.13). O
As was the case with connected manifolds, connected manifolds with
boundary are arc connected.
(17.A.5) Theorem. If M is a connected n-manifold with boundary, then

M is arc connected.
Proof. Suppose s € M and C = {rz} U{y € M : there is an arc
f :I—4 M with f(0) = x and f(1) = y}. It suffices to show that C is
both open and closed. To see that C is open, suppose z € C and B is a cell
neighborhood of z. Clearly we have B C C, and hence C is open. To see
that C is closed, suppose z € M\C. If B is a cell neighborhood of z, it
must be the case that BNC = @ (for otherwise, an arc from zx to z could be
constructed that passes through a point in BNC).
Since M is connected, it follows that C=M. O
The simplest n-manifolds with boundary are €", S”, and B”. Homeo-
morphic copies of these latter two spaces are of special importance.
(17.A.6) Definition. A space C is an n-cell if and only if C is homeo-

morphic to B”, and C is an open n-cell if and only if C is homeomorphic to
Int B”. A space S is an n-sphere if and only if S is homeomorphic to S”.
An apparently innocuous, but nevertheless very useful result is given by

the next theorem.
(17.A.7) Theorem. If h : B” > €” is an embedding, then h(S"~") is the

frontier of h(B”).
Proof. First, note that h(S"~') = h(B”) \ h(Int B"). Let x € h(S"-!)
and let U be an open set (in €") containing x. If UN (€”" \ h(B”)) = @,
then U C A(B"), and consequently, h~'!(U) is open in B" C €”. However,
by the invariance of domain (15.B.4), this implies that h~!(U) C Int B”, a
contradiction, since z € h(S"~'). Thus h(S"—!) C Frh(B").
Some Preliminaries 421
Now suppose that « € Frh(B"). Again by (15.B.4), h(Int B”) is open

in €”, and hence x ¢ h(Int B"). Therefore, x € h(S"~!), and consequently,
Fri?) Gants" 2) 79
If C is an n-cell, then by definition C is homeomorphic to B”. How-
ever, it is easy to see that an infinite number of homeomorphisms can be
constructed between C and B". Although C may have a very peculiar topo-
logical frontier (or even none at all), we would nevertheless like to be able to
speak of the “boundary of C’,” which, logically, should be the image of S"~}
under any of the homeomorphisms. However, it is conceivable that not all
such images coincide. That this fear is unfounded is the content of the next
theorem.
(17.A.8) Theorem. Suppose C is an n cell. There is a unique (n—1)-sphere

X in C such that h(S"~!) = X for each homeomorphism h : B” - C.
Proof. Suppose h : B” - C is an arbitrary homeomorphism, and let
X = h(S"~'). Suppose f : B? > C is another homeomorphism. We show
that f(S"-!) = X. The map f—'h : B” > B" C €” is a homeomorphism
(and an embedding into €”), and hence we have f~1h(S"—!) = Fr B™ = S"“}
by the previous theorem. Thus, f(S""1)=X. O
(17.A.9) Notation. Henceforth, the X is the previous theorem will be
denoted by BdC.
We have already shown that any compact convex subset of E€” is an n-cell
(3.C.2). The following exercise yields another way that cells may arise.
(17.A.10) Exercise. Suppose C is an n-cell, where C c €” C E€"*!, and
that p € €"*! \ E". Show that the cone over C from p is an (n + 1)-cell.
(17.A.11) Exercise. Show that the suspension of an (n — 1)-sphere is an
n-sphere.
The following extension property of n-cells is frequently useful.
(17.A.12) Theorem. If C is an n-cell and h : BdC — BdC is a homeo-

morphism, then h may be extended to a homeomorphism from C' onto C.
Proof. Let g : C 4 B" be a homeomorphism. Then f = ghg~'|sn-1 is
a homeomorphism from S”~! onto itself. Extend k to B” by defining
ee {InlfGep fz #0
0 a 2 4),
Then h = g”* fg is the desired homeomorphism that extends h. O
(17.A.13) Exercise. Suppose h : BdC + BdC is a homeomorphism,

where C and © are n-cells. Show that h can be extended to a homeomor-
phism from C onto C.
(17.A.14) Exercise. Suppose F' C Int B” is a closed subset of B”. Show

there is a homeomorphism h : €” — Int B” such that h|p = id.
We now show that n-cells are homogeneous (almost), i.e., if z,y € IntC,
then there is a homeomorphism of C onto itself that carries x to y. In fact,
we give a slightly stronger result.
(17.A.15) Theorem. Suppose z,y € IntC’, where C is an n-cell. There

is a homeomorphism h : C - C such that h(x) = y and Alpac = id. In
particular open cells are homogeneous.
Proof. The reader should note that it is sufficient to prove the theorem
for C = B” and y = 0. Let P, : B" \ {x} + S"~! be the obvious geometric
projection map. Define h: B” — B” by
Pelz) ifz#e
nas
0 it a0
Pe)
It is easy to verify that h is the desired homeomorphism. O
In addition to open cells, all connected n-manifolds are also homoge-

neous. This result is contained in the proof of the next theorem.
(17.A.16) Theorem. Suppose M is a connected n-manifold with boundary

and x,y € Int M. Then there is an n-cell C C M such that z,y € Int C.
Proof. Cover Int M with a collection C of open n cells whose closures
are n-cells lying in Int M. By (2.F.2) there is a simple chain C,,C2,...,Cy
in C such that)2 € C; and y € C,, «Leta. 7), and fori = 1, 2,..5. f.— 1.
pick a point. tje7 € C, Ci). Let y= 2,21: For’ = 1;2;....nr, thereas
a homeomorphism h; : C; — C; such that AlpaG, = td and h(ai) = ai41.
Since hj|, ac, 1S the identity, hj; may be extended to a homeomorphism
h;: M > M. Then h= ne vee hohy isa homeomorphism from M onto
itself such that h(x) = y, and h(C}) is an n-cell containing z andy. O
(17.A.17) Exercise. Extend the foregoing result to finite sets of points.
n-Annuli 423
B. n-ANNULI
(17.B.1) Definition. A space homeomorphic to S"~! x I is said to be an

n-annulus, a space homeomorphic to S”~! x (0,1) is an open n-annulus; a
space homeomorphic to S"~* x (0, 1] is a half-open n-annulus.
(17.B.2) Notation. We denote {x € E” : ||x|| <r} by B”.
Clearly, B” \ BY is an n annulus, B” \ BY. is a half-open n annulus,

and Int (B” \ By.) is an open n-annulus.
(17.B.3) Notation. Suppose C and C are n-cells and C C IntC. We let

[C, C] denote C\ C.
It is reasonable to conjecture that [CC] is an n-annulus. In fact, if C
and C are 2-cells, we have seen (5.E.5) that [C, C] is a 2-annulus. However,
for n > 2, counterexamples to this conjecture may be constructed. Never-
theless, if certain conditions are placed on the n-cell és, then more positive
results are obtainable. Recall that an n-cell lying in €” is tame if and only
if there is a homeomorphism h : €" + €” that carries C onto B”, and that
C is wild otherwise: a similar definition applies to (n — 1)-spheres lying in
E” (15.E.2). The annulus problem may be slightly rephrased to yield one of
the most famous conjectures emeriti of topology.
Annulus Conjecture (Theorem): Suppose C’ and C are tame n-cells

lying in €” such that CC IntC. Then [C, C] is an n-annulus.
In 1969, it was proven that the annulus conjecture is valid ifn 4 4

(Kirby, Siebenmann, and Wall [1969]). An exposition of the case n = 4 is in
Freedman and Quinn [1990].
The following is a considerably more modest result.
(17.B.4) Theorem. If C and D are tame n-cells in E€” such that CC Int D.
Then Int [C, D] is an open n-annulus. [Denote Int [C, D] by (C, D).]
Proof. An open n-annulus is homeomorphic with €" \ B”; hence it
suffices to show (C, D) and €”" \ B” are homeomorphic. Since D is tame,
there is a homeomorphism h : €” + €” which maps D onto B". Since h(C)
is a closed subset of the interior of B”, there is by (17.4.14) a homeomorphism
f :€" > Int B" that is the identity on h(C). Since h(C) = f~*h(C) isa tame
n-cell, there is a homeomorphism g : €” + €” such that gf~'h(C) = B”.
The desired homeomorphism is gf~*h|ic,p). 0
Although we are no position to prove the annulus theorem, we do es-
tablish a somewhat weakened version.
(17.B.5) Theorem. Suppose for each tame n-cell C in €” there is a home-

omorphism h : €” + €” and a compact set A C €” such that h(C) = B”
and hlen\4 = id. Let C; and Cy be tame n-cellsin€" such that C, C Int Cp.
Then [C, C2] is an n-annulus.
Proof. Let g : €" — E" be aspace homeomorphism that carries C2 onto
B”. For convenience, we will assume that 0 € Int g(Ci). Apply (17.A.14)
to obtain a homeomorphism f : IntB” — €&", which is the identity on
g(C,). Since fg(Ci) = g(Ci) is a tame n-cell, by hypothesis there is a
homeomorphism h : €” + €” and a compact set A such that
(i) hfg(Ci) = B”, and
(it) hlen\, = id.
We may assume that S”~! C Int A. (One can actually prove that S"~4
must be in the interior of A; see problem 1 of sections B-C.)
Choose € small enough so that Bf, C Int A, and let B be an n-ball with
center 0 contained in the interior of fg(C1) = g(Ci). There is a space home-
omorphism; ky 1/6". €3,snchthat k(B°)| =.Bak([BS Bee))o= (O,Bi
el
and k is the identity on €” \ Int BT,,. Let q = kh. Note that there is a
compact set F’, outside of which q is the identity. Finally, set ¢ = f—'qf :
Int B" — Int B”. Then ¢ is onto, and since q|gn\p= 1d, it follows that there
is a positive number r < 1 such that ¢(z) = z for x € B" \ B". Thus ¢ may
be extended (homeomorphically) to 6: E€" + E”. It is easily checked that
é9(Ci) = B and ¢g(C2) = B”, which proves that [C, C2] is an n-annulus.
O
C. CELLULAR SETS
Cellular subsets of a manifold are defined as follows. (Warning: in the liter-

ature the term cellular set is sometimes used to denote a different concept.)
(17.C.1) Definition. A subset K of an n-manifold M is cellular if and only
if there is a sequence of n-cells Cy, C2,... in M such that C; C Int C;_, for
each i, and ()2,C; = K.
Cellular sets are compact and connected (4.4.8); however, it is clear

from the following examples that they need not be cells.
Ci
Cellular Sets 425
On the other hand, it can be shown that not all cells are cellular. For
example, the wild cell constructed previously (15.E.13) is not a cellular set,
as we prove presently. On the third hand, tame cells are cellular.
(17.C.2) Exercise. Show: if C is a tame n-cell lying in €", then C is

cellular.
If K is a cellular subset of €”, there is a decreasing sequence of cells

whose intersection is K. Can it be assumed that these cells are tame? The
answer is yes; this is an immediate consequence of the following theorem.
(17.C.3) Theorem. If C, and C2 are cells in €” and C, C Int Co, then

there is a tame cell C such that Cy C IntC CC C Int Cy.
Proof. Let D be a tame n-cell that lies in Int C), and let h : Cp > B”
be a homeomorphism. By the invariance of domain, h(D) has nonempty
interior. We may assume that h(D) contains 0. Choose € > 0 such that B?
is contained in Int h(D), and select r, 0 < r < 1, such that h(C1) c BP.
C2
eo
Let g : B” — B” be a homeomorphism that leaves Bd B” fixed and

carries B"™ onto B”. Define a space homeomorphism f : €” — €” by setting
h-gh(z) ifc EC,

f(z) = x
é "
if 7 € €" \ Co.
Then C = f~1(D) is tame, and C' Cc Int C2. To see that C; C Int C, we just
observe that C; C h-(IntB™) = h-1g-1(Int B%) C h-*g*(Int h(D)) =
Int h-1g—4h(D) = Int f-1(D) =IntC. 0
(17.C.4) Exercise. Show [D, C] is an n-annulus, where C and D are defined
as in the foregoing proof.
(17.C.5) Corollary. If K is a cellular set in €”, then there is a sequence of

tame n-cells Ci, C2,... such that Ci41 C Int C;, K = (2, Ci, and [Ci41, Ci]
is an n-annulus.
Next we introduce a concept somewhat more general than cellularity
which can be defined in spaces other than manifolds.
(17.C.6) Definition. A compact, connected subset K of a topological space
X is pointlike if and only if X \ K is homeomorphic to X \ {p}, where p is
some point in X.
In practice, one usually restricts the notion of being pointlike to spaces
where X \ {p} is homeomorphic to X \ {q} for any two points p and q of X.
(17.C.7) Exercise. Show: if M is a connected n-manifold and p,q € M,
then M \ {p} is homeomorphic to M \ {gq}.
In €", subsets are cellular if and only if they are pointlike (problem
C-9). That this is not true for n-manifolds follows from the next exercise
that exhibits a pointlike but noncellular subset of a 2-manifold.
(17.C.8) Exercise. (Christenson and Osborne [1968]). Let M be €? minus

the integers on the positive x axis and minus 1-spheres of radius 1/4 centered
at the negative integers on the z axis. Then M is a connected 3-manifold.
Show that the 1-sphere of radius 1/4 and center at 0 is pointlike but not
cellular.
It is immediate from the next theorem that cellular sets are always
pointlike. But first we need a definition.
(17.C.9) Definition. Suppose X and Y are topological spaces, y € Y, and

f :X — Y. We call f~!(y) an inverse set if and only if f—!(y) contains
more than one point.
(17.C.10) Theorem. If C is an n-cell and K C Int C is cellular, then there

is a continuous surjection f : C + C’' such that
(t) flbac = 1d,
(17) flc\K is a homeomorphism, and
(ii7) the only inverse set under f is K.
Proof. Let C,,C2,... be a sequence of n-cells lying in IntC such that
Ci41 C Int C; and K = ()72, Ci. We define a sequence of homeomorphisms
from C’' onto C' as follows. Let f; : C ~ C be a homeomorphism such that
(i) filpac = id, and
(17) diam f; (C1) <4 Lie
(Why does such a homeomorphism exist?)
If fm has been defined, let fm4i:C — C be a homeomorphism with
the property that fm+ilC\Om = Fmlo\Cn and diamfm4i(Cm4i) < ane
Then f(z) = limm—+oo fm(z) is the desired map. O
Cellular Sets 427
(17.C.11) Corollary. A cellular subset of an n-manifold is pointlike. 0
It follows from (17.C.11) that the cell C constructed in (15.B.13) is

not cellular. If C were cellular, then it would be pointlike and hence the
fundamental group of its complement would be trivial.
It should be noted that in spite of exercise (17.C.8), Christenson and
Osborne [1968] established that in a compact n-manifold with boundary, any
pointlike subset lying in an open n-cell is cellular.
Theorem (17.C.10) may be refined slightly. The easy proof of the next
theorem is left to the reader.
(17.C.12) Theorem. If K is a cellular subset of €", p € K, and W is an

open set in €” containing K’, then a homeomorphism h: €" \ K > E™ \ {p}
exists such that
(i) h(W \ K) = W \ {p}, and
(it) hea ieee”VV o ol
We use the next sequence of theorems to obtain a generalization of the

2-dimensional Schonflies theorem (see section E of chapter 5). Recall: if S
is an (n — 1)-sphere in S”, then a complementary domain is a component of
S”\ S. Throughout the remainder of the chapter, we assume the following
theorem. Its proof is most elegantly established using homology theory, see
Stécker and Zieschang [1988].
(17.C.13) Theorem (Jordan and Brouwer). If S is an (n—1)-sphere in

S”", then S” \ S has exactly two components, and S is the frontier of each.
(17.C.14) Exercise. Suppose h: S"~? x I > S” is an embedding. Show

that S” \ h(S"~1 x I) has exactly two complementary domains.
Note that (17.C.13) does not assert that the closure of each comple-
mentary domain is an n-cell. In fact, we have seen (15.E.12) that ifn = 3,
then there is a 2-sphere in €° for which the closure of neither complementary
domain is a 3-cell. However, for n = 2 the following result holds.
(17.C.15) Exercise. Show that a simple closed curve in S* separates So

into two complementary domains and that the closure of each is a 2-cell.
We now develop criteria under which an (n — 1)-sphere in S” bounds

an n-cell. The results appearing in the next several pages were originally
established by Brown [1960b].
(17.C.16) Theorem. If S is an (n—1)-sphere in S”, Disa complementary

domain of S, C is an n-cell, and f : D > C is a continuous surjection whose
only inverse set is a cellular set in D, then D is an n-cell.
Proof. Let K C D be the unique inverse set under f and let Co be an
n-cell containing K in its interior.
Since K is cellular, by (17.C.10) there is a continuous map g : D > D

with unique inverse set K, and IID\ co Slide Then! fg7* DesCis clearly
a bijection. Furthermore, if A is a closed subset of D, then A is compact
and fg—!(A) is a compact subset of C (and thus closed). Consequently,
(fg~!)7! : C > Dis continuous and hence, by (2.G.11), fg7! is a homeo-
morphism. Therefore Disann-cell. O
In the next theorem, we see that single inverse sets are cellular.
(17.C.17) Theorem. Suppose C is an n-cell and f : C + S” is a continuous
map with a unique inverse set K. If K C Int C, then K is cellular.
We establish two lemmas before proving this.

(17.C.18) Lemma. Suppose C is an n-cell and f : C + S” is a continuous
map with a finite number of inverse sets Ky,..., Km, each of which lies
in the interior of C. Then f(C) is the union of f(BdC) and one of its
complementary domains.
Proof. Since f|pac is an embedding, it follows that f(BdC) has two
complementary domains D and D and is the frontier of each (17.C.13). Since
f(C) is connected, f(C) lies either in D or D, say D. It remains to show
that DC f(C).
We first show f(IntC \ Uj", Ki) is open in S”. If z is an element of
Int C\Uj, Ki, then there is an n-cell B such that c € B C IntC\UZ, Ki,
and f|g : B + S” in an embedding. Therefore, flint5 is an embedding,
and by invariance of domain, f(Int B) is open in S” (why can (15.B.4) be
applied here?). Consequently, Frf(C) C f(BdC)U(U;2, f(Ki)), so Frf(C)
contains at most m points that do not lie in f(BdC).
Cellular Sets 429
Now suppose xz € D\f(C). Since Fr D = f(BdC) Cc f(C), it follows that

z € D\f(C). Select a point z € f(IntC\Ui, Ki), and let a1,...,@m,Qm41
be mutually disjoint (except at the endpoints) arcs from z to z that lie in
D. These arcs must cross Fr f(C) at m+ 1 distinct points, none of which
are in Fr D = f(BdC). This contradicts the last statement in the preceding
paragraph. O
(17.C.19) Lemma. Suppose C is an n-cell and f : C > S” is a continuous

function with only one inverse set K. If K C U C IntC, where U is open,
then f(U) is open in S”.
Proof. Since K is the unique inverse set and K C U, it is immediate
that f(C) \ f(U) = f(C \U). Since C \ U is compact, it follows that
f(C \ U) is closed in S”. Note that S” \ f(U) = f(C \U) UD, where
D is the complementary domain of f(BdC) disjoint from f(C). Thus, the
complement of f(U) is closed and consequently f(U) is open. O
Proof of (17.C.17). By (17.C.18), we have f(C) = f(BdC)UD for some
complementary domain D of f(BdC).
Suppose U is an arbitrary open set in IntC containing K. We fit an
n-cell inside of U that contains K in its interior. By (17.C.19), f(U) is open.
Let V be a “small” neighborhood of f(K) in f(U). Since D # S", there is
a homeomorphism A : S" — S” that carries D into f(U) and is fixed on V.
Define g:C 4 U CC by
it £ if@w@ek
ane rete ee
Then g is an embedding, and hence g(C) is an n-cell contained in U and
containing K in its interior. Since U was arbitrary, K must be cellular. O
(17.C.20) Exercise. Show that the homeomorphism h mentioned above
exists (by returning to €” if necessary) and find a decreasing sequence of
cells whose intersection is K.
Each member of a pair of inverse sets is also cellular.
(17.C.21) Theorem. If f : S" — S” is a continuous surjection and has

precisely two inverse sets Ky and Ko, then both are cellular in S”.
Proof. Let S be an (n — 1)-sphere in S” \ (Ky U Ke) such that the
closures of the complementary domains of S are n-cells. If by some odd
happenstance, S separates K, and Ko, then (17.C.17) may be applied to
conclude the proof. Suppose then, there is a complementary domain D of
S that contains K; U Ky. Let ky = f(K1), ko = f(K2), and E = f(D).
Let U be an open subset in E that contains k; but not kz. Then, as in the
previous theorem, there is a homeomorphism h : S" > 5” that maps £ into
U and is fixed on an appropriately small neighborhood V of k;, Now define

GaSe by
al Oagh ty
Sa ti Gate
£ ifxe Ka
(@) hae
Nis ae
Note that Ky is the only inverse set of g, and hence by (17.C.17) we have
Ko is cellular. A similar argument shows K;, is cellular. O
D. THE GENERALIZED SCHONFLIES THEOREM
In Chapter 5 we showed that a 1-sphere lying in S? (or E*) always encloses

a 2-cell; on the other hand, we gave (15.E.12) an example of a 2-sphere that
fails to bound a 3-cell. What went wrong when the dimension was increased?
The basic problem is that for n > 2, an n-sphere S may become severely
“entangled” with itself. Such entanglements are eliminated if a “protective
collar” can be built around the sphere. The generalized Schonflies theorem
states that a sphere S does in fact bound an n-cell (or n-cells) when such
protective strip(s) are available. We begin with a definition and a mildly
technical lemma.
(17.D.1) Definition. An (n—1)-sphere S in an n-manifold with boundary,

M, is bicollared if and only if there is an embedding h: S"~! x I + M such
that h(S"~' x {1/2}) = S; S is collared if and only if there is an embedding
h: 8"! x I + M such that h(S"! x {0}) = S. An n-cell C C M is
collared if and only if there is an embedding h : S"~! x I — M such that
WS"-! S10) = Ba Cand WS” Se (Ort At Ve.
(17.D.2) Lemma. Suppose h : S"~' x I + S” is an embedding. Let A

and B be the two complementary domains of h(S"~! x I). Then there is a
continuous surjection f : S" + S” such that
(i) A and B are the only inverse sets of f and
(ii) f(h(S"1 x {1/2})) is the equator of S”.
Proof. Recall that S” may be considered as the suspension of S”~!
(17.A.11). Suppose g: S"~! x I + S” is the quotient map which satisfies
q(S”~* x {1}) =a, g(S"~! x {0}) = b, and q(S"-! x {1/2}) is the equator.
Define f :S” > S” by
a ifze A,
f(z) = 456 if z € B,
qh “(a Wait re hi(Sh an),
Compact n-Manifolds 431
(Relabel A and B if necessary. )
SUA ST Oo
(17.D.3) Theorem (Generalized Sch6nflies Theorem). (Brown [1960b]).

If h : S"-1 x'I — S” is an embedding, then the closure of either comple-
mentary domain of h(S"~! x {1/2}) is an n-cell.
Proof. Let A and B be the complementary domains of h(S"~! x I).
By the previous lemma, there is a continuous map f : S” — S” whose
unique inverse sets are A and B. It follows from (17.C.21) that A and B are
cellular. Let D be a complementary domain of S"~! x {1/2}. Again from
the previous lemma, we have the restriction f|; mapping D onto one of the
two hemispheres of S”. Apply (17.C.16) to conclude that Disann-cell. O
(17.D.4) Corollary. Suppose that S is an (n — 1)-sphere in €” and that

hz S™* x0,1) = €% is an embedding. If'h(S"-*x {1/2}) = S, then S
bounds a tame n-cellin €". O
E. COMPACT n-MANIFOLDS
In Chapter 16 we discovered that any compact 2-manifold may be obtained

from a 2-cell by appropriate identifications on the boundary. Surprisingly,
this result can be generalized to compact n-manifolds.
(17.E.1) Theorem (Brown, [1962b]). If M is a compact n-manifold, then

there is a continuous surjection f : B” — M such that
(i) f\lmtee is a homeomorphism, .
(ii) f(Bd B”) 9 f (Int B”) = 0, and
(iii) dim f(BdB") < n—1 (ie. no point of f(BdB”) has an n-cell
neighborhood contained entirely in f(BdB”)).
Thus, just as was the case for 2-manifolds, M may be considered as the
quotient space resulting from identifications on the boundary of the standard
n-cell B”. The proof of the theorem will follow readily once we have estab-
lished Theorem (17.E.3) below. This latter result is of some independent
interest, since it is an example of what is often referred to as an “engulfing”
theorem. In this type of theorem, certain subsets of a space are swallowed up
(or engulfed) by other sets (usually via homeomorphisms). Engulfing theo-
rems are amazingly versatile, and are used in a wide variety ot situations.
The example to be given is one of the more primitive engulfing results, but
it does serve to illustrate the concept involved. In the theorem, note how D
sweeps out to engulf both D and X before it becomes satiated.
(17.E.2) Exercise. Suppose EF is an open n-cell. Show that

(a) any two points in EF may be connected by a tame arc, and
(b) if @ is a tame arc in FE and e > 0, then there is an n-cell B C S,(a)
such that a C Int B.
(17.E.3) Theorem (Primitive Engulfing Theorem). Suppose D is a

collared n-cell in an n-manifold M and that ¢ and 6 are positive numbers
with 6 < 1. Let € = {&,E»,...,E,%} be a finite cover of M by open
n-cells each with diameter less than €, and suppose E;N D # @ for each
i. Let X = {21,%2,...,%m} be a finite collection of points in M, and let
h: B™ + D be a homeomorphism. Then there is a collared n-cell D and a
homeomorphism h : B" + D such that
GDI GD,
(it) Alpe, =h, and
(iii) d(h(x), h(x)) < € for all x € B”.
Proof. Since D is collared, there is a homeomorphism h : B} — M such
that h|gn = h. We may assume that 6 is small enough so that if d(x, Z) < 6,
then d(h(x),h(é)) < . Furthermore, we may assume that X contains only
points in M \ D. For each i, let E,; denote a‘member of € that contains 2;
and let y; be any point in Ex, N(D \ h( 1-6/2))- It follows from (17.E.2)
that 2; and g; may be joined by a tame arc a; lying in E,,. Finally, we
can assume that all of these arcs are mutually disjoint (why?). Let a; be a
Compact n-Manifolds 433
subarc of a; that lies outside of D and joins 2; to some point yi; contained
in A(B?, 572).
Let y = min{mind(ap,a,),mind(a@;,D;)} where 1 < p,q<m,p#4q,

and 1 < 7 < m. By the previous exercise, each @; may be “blown up” to
an n-cell D; that lies in a (y/3)-neighborhood of &;, and contains x; and y;
in its interior. Then the D,’s are disjoint, have diameter less than €, and do
not intersect D.
For each 7, we let hj : M — M be a homeomorphism that satisfies
hilm\p, = id and h,(y;) = 2;. Let g be a homeomorphism of 8B} onto itself
that is the identity on BY_;/. and maps B} \ Int BY_5/2 onto B? \ Int BY 5/25
then h = Amhm—1-:+hi hg is the desired homeomorphism. O
Proof of (17.E.1). For each positive k, let Uy, be a finite open cover of
M by open sets of diameter less than 1/2*. Let X;, be a finite collection of
points that contain at least one point from each U € Uy. By (17.A.17), there
is an n-cell D, in M that contains X,; furthermore, we may assume that
D, is collared (why?). Let hy : B” — D; be any homeomorphism. By the
previous theorem, corresponding to e = 1/2* and 6 = 1/2, there is an n-cell
Dz, and a homeomorphism hz : B” > D2 such that
(i) AgUD, C Ds,
(ii) De = halos. gs and
(iit) d(hy (x), ho(a)) < (1/2)?.

The inductive step should now be clear. If not, clarify it, and show that
f :B” > M defined by f(x) = limn_-+co Mn(z) is the desired homeomorphism.
(Note that f is the limit of a sequence of uniformly continuous functions, and
hence is continuous.) O
We conclude this chapter with a characterization of €”. We need the
following two lemmas.
(17.E.4) (Brown [1961]). Suppose S is a bicollared (n — 1)-sphere lying

in the interior of an n-cell D and U is an open set in the complementary
domain J(S) of S that has compact closure in intD. Leth: D + D be
a homeomorphism such that h|y = id, and suppose S C I(h(S)). Then
[S, h(.S)] is an n-annulus.
Proof. Since S is collared, there is an embedding f : S"~! x I + I(h(S))
such that f(S"—! x {0}) =S and f(S"~! x (0,1)) lies in D \ I(S). Clearly,
there is a homeomorphism g : D > D such that g|y,(s) = id, g(S) C U, and
gf(S"-! x {1/2}) = S. Then g“!hgf(S"~! x [0,1/2]) = [S, h(S)] and hence
[S,A(S)] is an n-annulus. O
(17.E.5) Lemma. Suppose D is an n-cell, S is a bicollared (n — 1)-sphere
in Int D and K is a compact subset of Int D. Then there is an (n — 1)-sphere
S c IntD such that K US C I(S) and [S, S] is an n-annulus.
Proof. We assume that D = B”. Let p be a point in J(S) and let U be
an open neighborhood of p such that U C I(S). There is a homeomorphism
h: D > D such that h\y = id and h(K US) C I(S). Then by (17.E.4),
S =h~1(S) is the desired sphere. O
(17.E.6) Theorem. If M is a connected n-manifold with the property that
every compact subset of M lies in an open n-cell, then M is homeomorphic
tore
Proof. It follows from (10.4.18) and (10.A.16) that there is a countable
family, {K1, K2,...}, of compact subsets of M such that M = U2, Ki and
K; C Int (Ki4i). By hypothesis, there is an open n-cell C,, and hence a
bicollared closed n-cell D, such that K, C Int D,; C iD, C Ci, where Ds
is the union of D,; and its collar. Similarly, there is an open n-cell Cg and
by (17.E.5) a collared n-cell D2 such that C; UK, C Int D2 C De C2
and [Bd D,, Bd D2] is an n-annulus (D2 is the union of Dz and its collar).
Proceeding inductively, we obtain a sequence of n-cells D;, D2,... such that
M = U2, Di and [Bd D;, Bd D;4,] is an n-annulus.
A homeomorphism f : M — €” is constructed inductively as follows.
Let f; : D; — B” be an arbitrary homeomorphism and for 7 = 2,3,4,...,
let f; : D; — BP? be a homeomorphism such that fi|p,_, = fi-1. Then
f : M > €” defined by f(x) = fi(x) if z€ D; is the desired homeomorphism.
O
(17.E.7) Corollary. If E = U;2, Ei, where for each i, E; is open in E, E;

is an open n-cell, and &; C Fj41, then F is an open n-cell. O
Problems 435
PROBLEMS
Section A
1. Give another proof of (17.4.5) using simple chains.

2. Show that invariance of domain holds for n-manifolds.
3. Suppose M and N are manifolds with boundary, andh: M > N isa

homeomorphism. Show that h(Bd M) = BdN.
4. Find a locally Euclidean space that is not T».
5. Suppose M is a compact n-manifold and U C M is homeomorphic to

E”. Identify points in M\U to a single point and show that the resulting
quotient space is homeomorphic to S”.
6. Suppose X is a second countable T2 space with the property that every

point has a neighborhood homeomorphic to €”. Is X an n-manifold?
7. Suppose X is paracompact and every point x € X has a neighborhood
homeomorphic to €”. Show that X is an n-manifold.
8. Suppose M is an n-manifold with boundary. Let z € M and suppose x
has a neighborhood W with the property that there is a homeomorphism
h: W — B® such that h(x) € BdB". Show that x does not have a
neighborhood homeomorphic to Int B”.
9. Show: every compact n-manifold can be embedded in a finite product
of spheres.
10. Suppose M is an n-manifold with boundary and that Bd M # 0. Show

that Bd M is an (n — 1)-manifold.
11. Show: if M and N are manifolds with boundary, then M x N is a
manifold with boundary and Bd(M x N) = (Mx Bd N)U(BdM x N).
12. Show that an n-manifold N cannot be homeomorphic to an m-manifold
M ifmFn.
13. Suppose X x Y is a manifold and (x,y) € Int (X x Y). Suppose further
that X x Z is a manifold. Show there is a point z € Z such that
(ao2)¢ Int CX, x Z).
14. Let X be a topological space and suppose X x €” is a manifold, and M
is an n-manifold where m > n. Show that X x M is a manifold.
15. Suppose X and Y are manifold factors (i.e., there are spaces W; and
W. such that X x W, and Y x W2 are manifolds). Show that X x Y is
a manifold factor and Bd(X x Y) = (X x BdY) U(BdX x Y).
16. (a) Exhibit a homeomorphism of the surface of a torus that inter-

changes circles a and b.
(b) Let S denote the solid torus. Can the homeomorphism of part (a)
be extended to a continuous function f : S > S?
L ts Assume the following result of Hanner[1951]: If X is a metric space, and
U,,U2,... is an open cover of X, where each U; is an ANRy, then X
is an ANRy. Show that every n-manifold is an ANRyy.
18. (a) Suppose M is a compact connected n-manifold, and N C M isa
compact n-manifold. Show that N = M.
(b) A topological space Y dominates a topological space X if and only if
there are continuous maps f and g such that the following diagram
is commutative
AC
Y id is
Suppose M and N are compact connected manifolds, M dominates N,
and N dominates M. Show that N and M are homeomorphic.
19. Find an example showing the conclusion of 18b above can fail if the
hypothesis “manifold” is dropped.
Sections B and C
In (17.B.5) show: not only may we assume that S"~! C Int A, but this
actually must be the case.
Suppose C’ C €” is connected and €” \ C is homeomorphic to €” \ {p}.
Show that C is compact.
Let Ci, C2, and C3 be n-cells with C, C IntC2 C C2 C Int C3. Show:
if [C},C2] and [C2,C3] are n-annuli, then there is a homeomorphism
h:C3.—> BF such that for i = 1,2,3, h(C;). = BP.
Suppose D C &” is an n-cell lying in Int B” such that [Bd D, Bd B”) is
a half-open annulus. Show that D is tame.
Problems 437
5. Suppose K,, Ko,...,Km are pairwise disjoint pointlike subsets of €”,

Show that €” \ Uj", K; is homeomorphic to €” minus m points.
Show that the topologist’s sine curve is cellular.
Let C be an n-cell in €” and let K C IntC be a continuum. Suppose
there is a continuous function f : C > €” and a point p € E” such that
(7) flc\K is a homeomorphism,
(it) f(K) =p, and
(it) pe f(C\ K).
Show that p € Int f(K).
Suppose K is a cellular subset of €3 lying in a plane Q. Show that K is
cellular in Q.
Show: a subset A of €” is cellular if and only if A is pointlike.
. Suppose Kj, K2,..., Km are cellular subsets of an n-manifold M. Let G
be a decomposition of M into points and the sets K;. Show that M/G
is homeomorphic to M.
. Give an example of a 2-manifold M and a pointlike subset p of M that
is not cellular.
12, Modify the example given in (17.C.8) so that all of the 1-spheres are
consecutively linked and display a pointlike subset of the manifold that
is not homotopic to a point in the manifold.
Sections D and E
Call an n-cell in S” tame it there is a homeomorphism h : S” > S”

such that h(C) is the upper hemisphere of S”. Suppose D is a tame
n-cell in E”. Consider S” to be the one-point compactification of €”.
Show that S” \ Int D is a tame n-cell in S”
Suppose D is an n-cell and S is an (n — 1)-sphere contained in Int D.
Suppose further there is an embedding f : S"~' x I > D such that
f(S"-! x {1/2}) = S. Let U be an open subset of J(S) and suppose
h: D > S is a homeomorphism such that h|y = id and S C I(h(S)).
Show that [S,h(S)] is an n-annulus.
Suppose M is a compact manifold and (M \U) C V, where U and V
are open n-cells. Show that M is an n-sphere.
Suppose K is a compact subset of €”. Show that K is cellular if and
only if €"/G is homeomorphic to €", where G is the decomposition of
E” obtained by identifying K to a point.
Show: an n-cell is collared if and only if its boundary is bicollared.
A subset B of a space X is locally collared if B may be covered by a

collection of open (in B) subsets each of which is collared in X. Show
that the boundary of every manifold is locally collared. (The reader may
wish to consult Brown(19621] for a proof of the fact that if a manifold
M is contained in a manifold N and M is locally collared in N, then M
is collared in N.)
A space is invertible if and only if for each x € X and for each neigh-
borhood U of a, there is a homeomorphism h : X — X such that
h(X \U) CU. Prove the following result of Hocking and Doyle, using
the steps indicated below: If M is an invertible n-manifold, then M is
an n-sphere.
(a) Show that M is compact,
(b) Let C be a bicollared n-cell in M and let D= M\C. Define a
homeomorphism h : M > M with the property that h(D) Cc Int C.
(c) Let g : C + S” be an embedding, consider gh(BdC), and then
apply (17.D.3).
Suppose M is an invertible noncompact n-manifold. Show that M and
E€” are homeomorphic.
Suppose {C;} is a sequence of subsets of a space X such that X =
Uso, Ci and for each i, C; C Ci41. Let {hi : X 4 Y : i € Zt} be
a sequence of homeomorphisms such that hiii|c; = hilo; for all 7 and
Use, hi(X) = Y. Show that h = lim;_,.. hj need not be a homeomor-
phism.
10. Show that an arc a contained in €” C €"+! is tame in €"T!.
Chapter 18
DIMENSION THEORY
Let 4 be a class of topological spaces. Intuitively a dimension theory for

& should be a function from 4% to N U {co} that has at least the following
properties:
(i) Dimension is a topological invariant
(ii) The dimension of Euclidean n-space is n. In particular the dimen-
sion of a point is 0.
(ii7) If the dimension of X is n and the dimension of Y is m, then the
dimension of X x Y is <n+m. (Product theorem)
(iv) The theory is applicable to a reasonably broad class of spaces.
(v) If the dimension of X is n, the dimension of Y is m, and both
are closed subspaces of a space Z, then the dimension of X UY is
< max(n,m). (Sum theorem)
(vi) If A C X then the dimension of A is less than or equal to the
dimension of X. (Monotonicity)
We will consider briefly the three most common methods of obtaining
a dimension function: little inductive dimension, big inductive dimension
and covering dimension. These all give the same result for separable metric
spaces, but are not the same in general.
In all three theories, the domain of the dimension function initially will
be the class of all metric spaces, M. However, for little inductive dimension
the important theorems require the restriction to separable metric spaces. It
is both traditional and convenient to change our long-standing understanding
that all spaces are nonempty. For this chapter we consider the empty set to
be a metric space with dimension -1. This will form the starting point for
all the (inductively defined) dimension functions that we consider.
439
440 18. Dimension Theory
This chapter gives just a cursory introduction to the theory of dimension.

For the reader who would like to learn more about this topic we suggest the
books by Engelking [1977] and Nagata [1970], as well as the classic book by
Hurewicz and Wallman [1941]. Most of what is in this chapter can be found
in these books.
Dimension theory leads rather naturally to fractal geometry. Readable
introductions for that topic are Edgar [1990] and Falconer [1990].
A. LITTLE INDUCTIVE DIMENSION

The intuitive idea behind little inductive dimension is quite simple. If you
are a point in €!, it takes a 0-sphere to pen you in; if you are a point in €?,
it takes a l-sphere to enclose you; etc.
(18.A.1) Definition. Let M be the class of all metric spaces. Define a

function, little inductive dimension, ind: M + NU {1,00} by
(i) ind(X) = —1 if and only if X = 9.
(it) ind(X) < n where n € N if and only if for each x € X and each
neighborhood U of z, there is an open set V which satisfies, x €
V CU and ind(FYV) <n-1.
(iii) ind(X) = n if and only if ind(X) < n and ind(X) <n—-1.
(iv) ind(X) = oo if and only if ind(X) £ n for all n.
Little inductive dimension was introduced by Urysohn [1922] and Menger
[1923]. Clearly the definition can be made for arbitrary topological spaces.
Furthermore, some (but not all) of the following theorems hold in this gen-
erality. However, separable metric is necessary for (18.4.7) and most sub-
sequent theorems. Hence the domain of ind is usually considered to be
separable metric spaces.
(18.A.2) Exercise. (a) Prove: If X and Y are homeomorphic metric spaces,

then ind(X) = ind(Y). (b) Suppose X is a metric space. Show: ind(X) = 0
if and only if X has a basis of sets that are both open and closed. (c) Suppose
X is a metric space. Show that ind(X) <n if and only if X has a basis B
where ind(Fr B) < n — 1 for each B € B.
In light of part (b) above, it is tempting to think that being 0-dimensional
is equivalent to being totally disconnected. However, problem A-4 should
disabuse you of any such notion.
(18.A.3) Corollary. ind(€!) = 1.

Proof. Clearly ind(€') < 1 (consider the standard basis). However, we
get ind(E') # 0 since @ and €! are the only sets that are both open and
closed—and they do not form a basis for the Euclidean topology. 0
(18.A.4) Corollary. ind(J)=1. oO
Small Induction Dimension 441
(18.A.5) Theorem (Monotonicity). If X is a metric space and A C X,

then ind(A) < ind(X).
Proof. The theorem is clearly true if ind(X) = —1. Assume it is true
for all spaces of dimension < n — 1. Suppose ind(X) =n, x € A, and U is
a relatively open set containing x. Then there is an open set U of X such
that c € U = ANU. Since ind(X) <n, there is an open set W such that
zé€W CU and ind(AW) <n-1. Let W=WOA. Then W is relatively
open and x € W CU. Since Fr AW CF W, by the induction hypothesis we
have ind(FraW)<n-—-1. oO
The following example shows that taking the union of subspaces can
raise dimension.
(18.A.6) Example. Observe that €! = (irrationals) U (rationals), and also

ind(irrationals) = 0 = ind(rationals), but ind €! = 1.
If the subspaces are both closed, the above behavior can not occur. This
is a consequence of our next goal, the sum theorems (18.A.13) and (18.A.15).
But first we need a few results and definitions.
(18.A.7) Theorem. Let X be a nonempty separable metric space. Then

ind(X) = 0 if and only if for each pair of closed disjoint subsets C,, C2 of X
there is a separation (U;,U2) of X such that C; C Uj.
Proof. Let C; and C2 be disjoint closed subsets of X. By (18.A.2b), for

each x € X, there is an open and closed set O, such that x € O, and either
C1 NO, or C2NO;z is empty.
Clearly {O;}cex is an open cover for X. By (1.G.11) and problem
(1.B.5) there is a countable subcover {O;};¢z+. Define another countable
open cover {Wi}icz+ by Wi = Oi \ Uj; 2,0; C Oi. (Why do the W;’s still
from a cover? Show that if i# 7 then W; NW; = 9.)
It is easy to see that (U;, U2) is the separation with the desired properties
where U; = U{W; : Ci NW; #0} and U2 = {Wi : C1NW; = O}.
Conversely, suppose z € X and U is an open set containing z. Let
C, = {xz} and C, = X\U. Then UV, from the hypothesis satisfies x € U; CU
and Fr O0y-= 0, Oo
(18.A.8) Definition. Let X be a topological space and C1, C2 a pair of

disjoint closed subsets of X. A set P C X is a partition of X between Ci
and Cy» if and only if there are open sets U; and U2 C X such that (Uj, U2)
is a separation of X \ P and C; C Uj.
(18.4.9) Lemma. Let X be a metric space, Y C X a subspace and C1, C2,

disjoint closed subsets of X. Suppose O,, O2 are open in X with C; C O;
and O,NO> = 9. If P is a partition of Y between Y NO, and YN Oz, then
there is a partition P of X between C; and C2 where PNY = P.
Proof. Suppose U1, Ue are open subsets of Y for which Y NO, Cc Ui,
and (U,, U2) is a separation of Aes Clearly C1 NU, =0=CyNU}. Then,
since O; NUyz= yRoro U2 CUMUs= 0, and since Oj, is open, we have
O, NU» =90. Hence, C; NU2= 0. Similarly, C2 i =. .
The Ue s are disjoint and open in U,UU >. Hence O00 > = \= ogni bic
and therefore (Cy U 0) N (C2 U U2) = 0 = (C1 U U,) N (C2 U U5). Since
metric spaces are completely normal (4.4.1), there are open sets U;, U2 of
X such that C; U U; Cc U;, and (U,,U2) is a separation of U; UU2. Let
P= X \(U, UU2). Then P is a partition of X between C; and C2 and
YPriP=Y MOUs) GY VU} =P. O
(18.A.10) Theorem. Suppose X is a separable metric space, Y C X, and

indY = 0. If Cy and C, are disjoint closed subsets of X, then there is a
partition P Cc X \ Y between C; and C2.
Proof. From the normality of X we can find open sets O, and Og, in
X for which C; C O; and O;N O2 = %. By (18.A.7), @ is a partition in Y
between Y NO, and Y MN Og. Now theorem (18.A.9) gives us the partition
Pel aes
(18.A.11) Theorem. Let X be a separable metric space and Y C X. Then
ind(Y) = 0 if and only if for each z € Y and for each X-open set U containing
x, there is an X-open set V such thatr €V CU and YNRV =9.
Proof. Easy exercise. O
(18.A.12) Corollary. Suppose X is a separable metric space and Y C X.

Then ind Y = Oif and only if X has a countable base B such that YNFr B = @
foreach BEB. O
(18.A.13) Theorem (0-dimensional Sum Theorem). If X = U72, X;

is a separable metric space where each X; is closed in X and ind(X;) = 0,
then ind(X) = 0.
Proof. Let C and D be disjoint closed subsets of X. By (18.A.7), it
suffices to find a separation (U,V) of X for which CCU and DCV.
Use normality _(twice) to get X-open sets Up and Vo such that CC Up,
Dc Vo, and Uativn =f,
Suppose we have defined sets U; and V; for i =0,...,7 — 1 such that
(*) U; NV;=@ and X;, CCU;UV; (define Xo = 0),

and fori¢= lste.<jgieel
(s%) — U;1 GU, and V,_) CV;,.

We want to define sets U; and V; with similar properties. Note that
U; oPoex, and V; -1 1 Xj; are closed and disjoint. By (18.4A.7) there is a
separation of Xj byXj-open (closed) sets (U,V) such that U;MEX GUT
and Ve. waP.e1e V. Since X; is X-closed, so are U and V.
Now,
(Uj-1 UU) N(V;-1 UV)

= (Uj-1 NV5-1) U(Uj-1 NV) U(UNV;_-1) U(ONV) = 0.
Apply (18. A.7) to obtain open sets U;,V; C X such that UU; »UU G U; and
Vj-1UV CV; andU; NV; = 0.
co CO
We will let the reader verify that U = U U; and V = U V; is the

i=0 i=0
desired separation of X. O
Note that Example (18.A.6) shows us the following lemma can not be
improved to ind(X) <n—-1.
(18.A.14) Lemma. Suppose X = Y U Z is a separable metric space with
ind(Y) <n-—1 and ind(Z) < 0, then ind(X) <n.
Proof. Let € X and let O be an open set containing z. Since {zr}
and (X \ O) are disjoint closed sets, (18.A.10) yields open sets O1,O2 € X
with z € O,, X \O C Og and [X \ (O; UO2)|
NZ = @. Hence x € O, CO.
Finally, ind(FrO,) <n—1 since FrO; Cc[X \(O, UO2)) CX\ZCY. O
(18.A.15) Theorem (The Sum Theorem). If X = U X; is a separable
metric space where, for each i, X; is closed in X ae) ind(X;) <n, then
ind(X) <n.
Proof. If n= —1 the theorem it trivial, and for n = 0 it is just (18.A.13).
Assume the theorem holds for all n < j, where 7 > 1, and suppose
a Eee X; where each X; is closed in X and satisfies ind(X;) < j. For
each i, by (18.A.2c), (1.G.11), and problem (1.B.5), we may find a countable
base B; for X; such that ind(Frx,B) < j —1 for all B € B;.
Ley = U Frx,B. By assumption, ind(Y) < 7 — 1. Hence for each
BEUB;
i, (18.A.12) yields ind(Z;) < 0 where Z; C X;\ Y C Xj. Now, by (18.A.13),
ind Z < 0 where Z = UZ;. (Why is each Z; closed in Z?). The theorem now
follows from (18.A.14). O
(18.A.16) Corollary of the Proof. If X is a separable metric space and

ind(X) <n, then X = Y UZ where ind(Y) <n—landind(Z) <0. O
We next show that if the subspaces are not closed then the dimension
of the sum can increase, but not by more than one. [Take another look at
example (18.A.6).]
(18.A.17) Theorem. Suppose X;, X2 are subspaces of a separable metric
space X. Then
ind(X, U Xp) < 1+ ind(X) + ind(X.).

Proof. We induct on the dimensions of X; and X2. The theorem is
clearly true if ind(X,) or ind(X2) = —1, or if ind for either is co. Now
assume ind(X,) = m and ind(X2) = n and assume the theorem holds when
(x) ind(X1) < m, ind(X2) <n-—1, and
(**) ind(X,) Si = I ind(X2) < Nn.
Let x € X, U X2, and without loss of generality suppose z € X,. Let

O be open in X with z € O. By (18.A.2c) and (18.A.5) there is an open
set U such that c € U C O and ind((Fru) N X1) < m—1. However,
(FrU) N X2 C Xo, so ind((FrU) N X2) < n. By the induction hypothesis
ind(FrU)N(X, UX2) < m+n, and ind(X, UX2) <m+n+1 follows. O
(18.A.18) Theorem (Decomposition Theorem). Suppose X is a sep-
arable metric space. Then ind(X) < n (where n is finite) if and only if
X = (Jt!
el X; where for each i ind(X;) < 0.
Proof. This is an easy corollary of the preceding two results. O
The product theorem is a straightforward consequence of the sum theo-

rem. Problem A-8 shows that the inequality is essential in general. Problem
A-7 shows that in some circumstances the inequality may be replaced by an
equality.
(18.A.19) Theorem (The Product Thoerem). If X = X, x X2 where

X, and X2 are separable metric spaces, and X, #40 or X2 #9. Then
ind(X) < ind(X,) + ind(X2).
Proof. The result is transparent if either X, or X2 is empty, or if ind

for either is oo.
Suppose ind(X,) = m and ind(X2) = n and assume for the induction
that the theorem holds if
(x) ind(X,) <_m, ind(X2) <n-—1, and

(x) ind(Xi) <m-—1, ind(X2) <n.

Let x = (z,y) € X. Let U x V be an open neighborhood of a, where x € U
and y € V. Without loss of generality assume that ind(Fr U) < m — 1, and
ind(FrV) <n—1. By (1.E.18), Fr(U x V) = (rU x V) U(U x FV). Each
part on the right hand side is closed and by (*) and (+**) has little inductive
dimension < m+n-—1. Then ind(Fr(U x V)) < m+n -—1 follows from
(SACS oe «i
We now give a sharpened version of (18.A.10).
(18.A.20) Theorem. Suppose X is a separable metric space, C,, C2 are

closed and disjoint, Y is a subspace of X, and ind(Y) < n. Then there is a
partition P between C; and C2 with ind(PNY) <n-1.
Proof. If n = 0, then either ind(Y) = —1 and the theorem follows from

the normality of X, or ind(Y) = 0 and the theorem is just (18.A.10).
Hence, suppose n > 0. By (18.A.16), X =WUZ with ind(W) <n-1
and ind(Z) < 0. By (18.A.10), there is a partition P between C; and C2
not meeting Z. Therefore Y M P C W, and as a consequence we have
ind(PNY) <ind(W)<n-1. O
The next lemma is useful in showing that ind(€”) = n
(18.A.21) Lemma. Let X be a separable metric space with ind(X) < n—1.
Suppose for eachi = 1,...,n that C;, D; form a pair of disjoint closed subsets
of X. Then there are n closed sets P,,...,P, such that P; is a partition
between C; and D; and () P, = @.

4=1
Proof. From (18.A.20), there is a partition P; between C and D; with

ind(P,) < (n-—1)-—1. Another application gives a partition P, between
Cy and Dz with ind(N?_,
P;) < (n — 2) — 1. Continuing we eventually get a
partition P, between C, and D, and ind(Nj_,
Pi) < (n-—n) —1. That is,
the intersection is empty. O
We end this section by showing that Euclidian n-space has little induc-
tive dimension n. It is an immediate consequence of (18.A.3) and (18.A.19)
that ind(E”) < n.
(18.A.22) Theorem. For Euclidean n-space €”, ind(Eé”) =n.

Proof. By the remarks above it suffices to show ind(€”) > n. Further-
more, since J" C &", it suffices to show ind(I”) > n.
Suppose ind(J”) < n—1. Then by (18.A.21) there are n-closed subsets
P; each a partition between opposite faces of I” with N7_,P; = 0. But that
contradicts (14.C.24). O
B. BIG INDUCTION AND COVERING DIMENSION
The primary goal of this section is to show that big inductive dimension and
covering dimension agree on the class of metric spaces.
The definition of big inductive dimension is obviously inspired by the
definition of little inductive dimension—just replace “point” by “closed set.”
(18.B.1) Definition. Let M be the class of all metric spaces. Define a
function, big inductive dimension, Ind: M > N U {—1, oo} by
(i) Ind(X) = —1 if and only if X = 0.
(it) Ind(X) < n where n € N if and only if for each closed set C and
each open set U containing C, there is an open set V such that
CcCUCV and Ind(Fr(U)) <n-1.
(iit) Ind(X) = n if and only if Ind(X) < n and Ind(X) <n-1.
(iv) Ind(X) = oo if and only if Ind(X) ¢ n for alln EN
The definition of covering dimension stems from the observation that
for every open cover of €! there has to be some point that is in (at least)
2 elements of the cover; for every open cover of E? there is some point that
is in (at least) 3 elements of the cover; etc. Before formulating this it is
convenient to have the next definition.
(18.B.2) Definition. Let X be a topological space, and U Cc P(X). The

order of U, ord(U), is the largest integer n such that U contains n + 1 ele-
ments with nonempty intersection. If there is no such integer n, we say that
ord(U) = o.
(18.B.3) Definition. Let M be the class of all metric spaces. Define a
function, called covering dimension, cov: M > NU {—1, oo} by
(i) cov(X) = —1 if and only if X = 90.
(it) cov(X) <n where n € N if and only if each open cover of X has
an open refinement with order < n.
(iit) cov(X) = n if and only if cov(X) <n and cov(X) <n—1.
(iv) cov(X) = oo if and only if cov(X) ¢ n for alln EN.
(18.B.4) Exercise. Show that Ind and cov are topological invariants.
(18.B.5) Exercise. Prove: If X is a metric space, then the following are
equivalent:
Big Inductive and Covering Dimension 447
(i) cov(X) <n.

(17) Each finite open cover of X has an open refinement of order < n.
(ii7) Each finite open cover U; = {U; : i = 1...,k} of X has an open
shrinkage of order < n.
(18.B.6) Definition. Let X be a topological space and U = {Uy : a € A}

a family of subsets of X. A swelling of U is any family V = {V, : a € A} of
subsets of X that satisfies:
(a) Ua C Va for each a € A, and
(b) for each finite set of indices, a1,a2,...,Qm,
() Uy, #0 if and only if a Var Ws

al | 41
(18.B.7) Lemma. If C = {C; : i = 1,...,k} is a finite family of closed

subsets of a metric space X, then C has a swelling U whose elements are
open sets. Furthermore, if V = {V; : 1 = 1,...k} is a family of open sets
with C; C V; for each 7, then we can pick U so that U; C V; for each 7.
Proof. Let D, be the union of all finite intersections N'_,C;,, where
1 = 1,2,..., that satisfy C; N (N_,Cj,) = 0. Then D, is a closed set and
D,QC, = 9. By normality there is an open set U; for which Cy C U; and
Ui ND, = 90. Clearly, {U1,C2,C3,..., Cy} is a swelling of C.
Similarly, let D2 be the union of all finite intersections of members of
the family {U4, C2,...,C,} that are disjoint from Cz. Then Dz is closed and
disjoint from C2, etc. The furthermore is left as an appetizer for the reader.
O
The above lemma yields a refinement of exercise (18.B.5).
(18.B.8) Theorem. Let X be a metric space. The following are equivalent:

(i) cov(X) <n.
(ii) Each finite open cover of X has an open shrinkage of order < n the
closure of whose elements is a closed cover of order < n.
(iii) Each finite open cover of X has a finite closed refinement of order
<i) O
(18.B.9) Theorem. If X is a metric space and Y is a closed subspace of

X, then cov(Y) < cov(X).
Proof. If cov(X) = oo, the theorem clearly holds. Hence we assume
cov(X) =n < oo. Let U = {U; : i=1...,k} be a finite open cover of Y.
For each i choose V; in X so that V; is open and U; = YN Vj. Since Y is
closed, V= {X \Y}U{V; :i1=1,...,k} is a finite open cover of X. Then
V has an open refinement W of order < n, since cov(X) < n. It is clear that
O={W;NY :i=1,...,k} is a finite open cover of Y that refines and

has order < n. Hence, cov(Y) <n =cov(X). O
(18.B.10) Theorem. Let X be a metric space. The following are equiva-
lent:
(i) If Y is a subspace of X, then cov(Y) < cov(X).
(ii) If Y is an open subspace of X, then cov(Y) < cov(X).
Proof. Omitting only trivialities, we assume (i7) and cov(X) = n < co
where n # —1. Let Y C X and suppose U = {Uj,...,Um} is a finite open
cover of Y. For each i, let V; be an open set of X for which U; = YNV;. Now
consider V = ();"., V;. Since V is open, by hypothesis we have cov(V) < n.
Hence {Vi,...,Vm} has a finite open refinement R = {Ri,..., Rx} of order
<n. Clearly {Ri NY,..., Rg MY} refines U and has order < n. Therefore
coviy) <n. 0
These results do not give us monotonicity: however, that will be estab-
lished in the next section (18.C.10). We now obtain a weak version of the
sum theorem.
(18.B.11) Lemma. Suppose X = )7<, X; is a metric space where for each
i, cov(Y) < n whenever Y is X-closed and Y C X;. Further assume UE, X;
is closed for each 1. Then cov(X) <n.
Proof. Suppose U = {Uj,...,Um} is a finite open cover of X. We
will inductively define a sequence U/,l4,,... of open covers of X where
Up =U 1 is Ui |Sauisnes:
(*) Ui; e OA ifi > 1 and Uo; (C U; for; == sean vai
(«*) for each i, {C; NUin, are Os NUim} has order <n,
where C; = Wie X; when i > 1 and Co = @. For j = 1,2,...,m we define

Uo,; = U; for j = 1,...,m. Then conditions (*) and (**) are satisfied
by U = {Uo1,-.-,Uo,m}. Suppose such covers have been defined for all
t < £ where £ > 1. Let C be the set consisting of all subsets of {1,...,m}
containing exactly n + 2 elements, and define
AS f)Ung:
SEC JES
We have AN Cy_; = 0 from (**) with i = 21. Now, Y = ANC; is an

X-closed set (Why?) and is a subset of Cy \ Ce_1 C X¢; hence cov(Y) < n.
By (18.B.5)iii, we obtain an open cover {Vo,...,Vm} of order < n with
V; (S YnUGpy: for all 7. If W; = (Ue-1,; Gi ad a e Ur-155, then the family
W = {W,,...,Wm} is an open cover of X and, furthermore, the order

of {Ce Wy,...,Ce Wm} is <n. By (4.C.2) there is an open shrinkage
U = {Uzj,...,Uem} of W. The reader may show that Ui satisfies conditions
(*) and (xx) when i = £.
Now for each x € X there is some j(x) < m such that z is in infinitely
many Ujj(2). Hence by (*), x € UR AP Use (*) and (**) to verify that
for each 7, N&,Ui,; C U;, and {N&,Uia,...,%2,Uim} is a cover ofX with
order <n. Thus, using (18.B.8), we get cov(X) <n. O
(18.B.12) Lemma. If X = U;2, Xi is a metric space, and for each i,

cov(X;) <n, and also Uj=1 X; is X-closed, then cov(X) <n.
Proof. This follows immediately from (18.B.9) and (18.B.12). O
(18.B.13) Corollary (The Sum Theorem). If X = U3, Xi is a metric

space and for each i, X; is closed in X and cov(X;) <n, then cov(X) <n.
O
(18.B.14) Lemma. Let X be a metric space. Suppose for each a pair of

disjoint closed subsets C, D of X, there is a partition P of X between C and
D with cov(P) < n—1. Then cov(X) <n.
Proof. Let U = {U,,...,Um} be a finite open cover of X. Let O be
an open shrinkage of U. By hypothesis, for each 7 there is a partition P;
of X between: O; and X \ U; with cov(P;) < n—1. For each i let W; be
an open set satisfying O; C W; C U; and Fr(W;,) C P;. If P = Ge Pi
then (18.B.13) implies cov(P) < n—1. Use (18.B.8)ii to obtain an open
shrinkage of {PN U;,...,P MU} the closure of whose elements is a closed
cover of order < n. Now use (18.B.7) to get a swelling of this latter cover
and a family {Vi,...,Vm} of open sets of X such that V; C U; for each i,
Pave Vs, and ord({Vi, .1;Vm}) <n-1.
Define Z,=Wi\(VUUj., Wy). Then 2 = {V1,...,V
mn, Zi,000 Sm}
is a closed refinement of U/. The reader may use ord({V1,...,Vm}) <n-1
and Z;NZ; C Wj n[Wi\(VUW;)] C (X\V)NFr (W;) = 0 when (j < i < m)
to show that ord(Z) <n. O
(18.B.15) Corollary. For each metric space X, cov(X) <Ind(X). O
In order to obtain the reverse inclusion, we need a little more machinery.

The next lemma helps in manipulating covers. The theorems that follow
essentially generalize (18.B.5) by replacing the condition “finite” by “locally
finite,” and then dropping the condition altogether.
(18.B.16) Lemma. Suppose X is a metric space. Suppose further that

W = {W.}aea is an open cover of X, and C a locally finite closed cover of
X such that for each C € C, cov(C) <n and C ‘only meets finitely many of
the U,. Then W has an open shrinkage V = {Va}aea of order < n.
Proof. Let U be an open shrinkage of W. Clearly U and C satisfy the

same conditions as W and C. Enlarge C (if necessary) by adjoining Co = 9.
Well-order C with Cp as initial element, to obtain C = {Cg}g<r4i where I is
an ordinal number. We use transfinite induction to define a family of covers
{Ug}a<r+1 where Ug = {Ug,a}aca satisfies:
(@) Uge CU,et BS 7 > OGnd'Up eC Uastor each ae a;
(ii) ord({Cg NUg.ahaca) <n
(itt) Uy,a \UB,0 C Uy<n<g Cn for y< 8 anda e A.
Let Uo,a = Us for all a and define Uo = {Uo,a}aea-. Clearly Uo satisfies
the three conditions above.
Assume covers Ug have been defined for all 89 > 1 that satisfy conditions
(i)—(tit). For each a € A let
Uso ,0 =3 () UB,0
B<Bo
and define Uz, = {Uj, ataea- We claim that U,, is an open cover of X.
If Go is an ordinal of the form y + 1 for some ordinal y this is easy since
U3, = Uy, and hence Uz, = U, is an open cover by assumption. Thus we
assume (3 is a limit ordinal. That is G9 has no immediate predecessor.
Let x € X. Since C is locally finite, there is an open set O containing x
and an ordinal number y < ( such that OM C, = 0 whenever y < T < po.
Since UW, is a cover of X, there is an a € A such that x € U,.. By (iit),
t € Ug.q for y < B < fo; hence x € UZ,,. Consequently U4,, is a cover of
X. We now show that each Uz,, is open. Let r € Uj,,,O be an open set
containing x, and y < fp an ordinal such that UNC, = 0 for y < T < Bo.
Now pick a set U,,q that contains x. By (117) and the definition of Uj,. we
have r € ONU4,q C Ug,.. Hence WU, is an open cover of X.
Now consider {Cg, 1 Ug, ataca- This is an open cover of Cg,. By
the hypothesis on C, only a finite number of elements are nonempty. Hence
the cover has an open shrinkage {Va}aca of order < n. Now define Ug, by
Up, = (UB, .0 \Cao) UVa. The reader may verify that Ug, has the requisite
properties, and that U/p is the desired cover V. O
(18.B.17) Theorem. Let X be a metric space. The following are equiva-

lent:
(i) cov(X) <n.
(ii) Each locally finite open cover U = {Uahaca of X has an open
shrinkage of order < n.
(itt) Each locally finite open cover of X has an open refinement of order
S17:
Proof. (ii) => (tit) is trivial. (iii) => (2) is also trivial after the
observation that a finite open cover is, in particular, a locally finite open
cover, so (18.B.5) applies. Hence all that is necessary is to show (i) => (it).
Let X be a metric space with cov(X) <n, and let U = {Ua}aca bea
locally finite open cover of X. Define T= {F CA: F #0 and F is finite}
and let Cr =(\ yer UanQger(X\Ua). We have cov(C'r) < n from (18.B.9).
It is clear that C = {Cr}rer is a closed cover of X and each Cr meets only
finitely many of the U,’s. The reader should now show that C is locally finite
and apply (18.B.16) to obtain the desired cover V. OD
(18.B.18) Exercise. Prove: If X is a metric space, the following are equiv-
alent.
(i) cov(X) <n.
(it) Each open cover U = {Ua}aea of X has an open shrinkage of order
<n
(iit) Each open cover of X has an open refinement of order < n.
The next, rather tedious lemma is needed to obtain one more charac-
terization of cov. We suggest a peek at (10.4.10) might be appropriate for
any readers not having a photographic memory.
(18.B.19) Lemma. Suppose X is a metric space, and {U;}ien is a sequence

of open covers of X such that for each i, ord(U;) < n and Uj41 < Uj. If
{St(U,U;) : U € Uj, i € N} is a base for X, then cov(X) <n.
Proof. First we build an inverse system on {U;}ien by defining
Ses 4
to be any map that satisfies U C f;(U). Then set
ie = | ae fy dg — U; for 1 <i
and define f;,; to be idy,. Clearly, for each 1 < j and for each U € Uj,
Ure FU). J $ ‘
Suppose O = {Q,,...,O¢} is a finite open cover of X. Define an open
cover O = {O1,Oz,...} by
Ove fu EU; : St(U,Ui) C O; for somes =Apiauyl}s

The O;’s are clearly open and they form a cover since, by hypothesis, the
stars form a base. Now for each k = 1,2,... define subfamilies Vk and W,
of Uk by ,
Ve = {U €Uz : UNOg #0} and WM, = {V EVE = VI(L)

Os) = OF.
j<k
Furthermore, for each V € VY; let i(V) be the largest integer < k for which
ficv),e(V) € Vicv). This exists since f1,4(V) M (Uj; Xj) = 9 and fan(V)N
On=V ThOps 0.
For each W € W; Cc W let
W = U (U{V WOg, = V E Ve; fix(V) = Wand (iV) = i}).
We now work to show that i(V) and W are well defined on W = UW;. By the
definition of O;, and since WN O; = 0, there is aU € Ui; so that WNU £9,
and an i(W) < £@ for which W C St(U,U;) C Oyw). Since U C fi3(U)
whenever 7 < j, we have W Cc W; hence We Oww): Now when 7 # j we
have Wi W; = 0. Thus if W € UW; both W and i(W) are well defined
and W is a refinement of 0.
We next show W = {W : W € W} is actually cover of X. Let x € X.
Pick k so that z € Ox \Uj<, Oj, and U € Us so that x € U. Now, we have
ere UN, Og -C (fuoyn)) € W. Hence W is a cover of X.
Finally, we show ord(W) <n. Suppose c € Wi}. NW2N---MW» where
Wi € Wm; and W; #4 W; when i ¥ j By the definition of W,, we have
mi <k, (i =1,2,...,m), where k is the integer for which «€ Ox \(Uj <4 Oj):
By the definition of W, there are sets V; € Vy; such that fm, %;(Vi) = Wi,
i(Vi) =m, and z € V;N Ox,. Since x € Ox;, we have k < k;. The sets U; all
contain z, where U;= fx.x;(Vi) € Ve for i =1,...,m. Since ord(Uy,) < n, it
suffices to show that U; # U; when i # j. ewes that each U; is in Vy and
that 1(U;) = i(Vi) =m;. Hence U; 4 U; when m; # m;. If mj; = mj; we have
U; # Uj since fm,,4(Ui) = fi:,ei(Vi)=Wi # Wy = fj,h; (Vj) = fmy,e (U5).
O
(18.B.20) Theorem. Let (X,d) be a metric space. The following are

equivalent:
(i) cov(X) <n.
(17) For each metric p on X there is a sequence {U;}icn of locally finite
covers ofX satisfying: (a) ord(U;) < n for each i; (b) diam(U) < 1/i
for each i and each U € Uj; (c) for each i and each U € Ujn1, U CV
for some V € U,.
(217) There is a metric p for X and a sequence {U;}ien of open covers of

X that satisfy: (a) ord(U;) <n; (b) for each i and for each U € U,,
diam(U) < 1/1; (c) For each i, U4, is a refinement of Uj.
Proof. ((i) => (ii)): Let p be a metric for X that is equivalent to d, and
assume that cov(X) <n. Let i = 1. Consider U; = {S?/,(2) sea GaX-}).
Now we obtain U4, by applying (10.A.7) and definition (10.4.5). Assume
m > 1 and that appropriate covers U/; have been defined for i < m. For each
x € X pick a neighborhood U, of diameter < 1/m so that U, CU where
U € Un-1. Now apply (10.4.7) and definition (10.4.5) to obtain Up.
We leave (ii) => (aii) and (iit) = (i) for the reader. [Hint: the first
implication is trivial and the second depends on the previous lemma.] 0
We close this section with a theorem proved independently by Katétov

[1952] and Morita [1954].
(18.B.21) Theorem. For any metric space (X, p), cov(X) = Ind(X).
Proof. By (18.B.15) we have cov(X) < Ind(X). So it suffices to show
that Ind(X) < cov(X). Without loss of generality we may assume that
cov(X) 4 oo or —1. Assume the inequality holds for each metric space Y
with cov(Y) < n —1 and consider a metric space X with cov(X) =n > 0.
Let C',D be a pair of disjoint closed subset of X. It suffices to find open sets
O and O which together with P = X \ (O UO) satisfy:
CCO %WCcO0> 7 ONOSD and cov(P)< 7-2.
Hence P is a partition between C and D and cov(P) = Ind(P) <n-1.

By (4.B.1) there is a continuous function f : X — I with f(C) = 0
and f(D) = 1. It is easy to see that r(x,y) = p(z,y) +|f(x) — f(y)| defines
a metric on X that is equivalent to p. By (18.B.20) there is a sequence
of locally finite open covers of X, {U;}ien such that for each 2 we have
ord(U;) <n, diam(U) < 1/i whenever U € Uj, and for each U € U;j+, there
isa V €U; for which U CV.
Let Oo = C and Oo = D. If i> 1 define inductively
O; = X \ H; where H; = U{U €U; : UN O;-1 # O}, and
0; =X\G; where G; = U{U EU; : UML =O}:
Claim: If U € U; and UN Oni # ), then UNO;_1 = 9. This holds

when i = 1 since, by definition of 7, any set that intersects both C and D has
diameter > 1. Assume 1 < i and suppose U € Ui; and MORK # (). Then
for each V € U;_, which contains U we have VN O,_1 # (, and therefore
V Cc Hj-1. Hence UNO;-1 = 0.
Since the U;’s are locally finite, it follows from the claim and the def-
initions of G;, H;, that Gin O;-1 = 0 = H;NO,-1 for all i. Hence
Oa te H, = int (O;) and ©;1, 0eCEM \ G; = int O;. Furthermore
O;NO; = since G; U H; = X. Consequently,
Oe Jo. and O = We
i=1 i=1
are open, disjoint and contain C and D, respectively. For each 7 define
Lao \ (O; U O;) = G; U H;. Then
The sets O, O, and P are as desired, except it is not clear that cov(P) < n—1.
To prove this last detail; for each 7 we define an open cover of P by
Y; ={U NP : U €uand Cito # 0}. We show that V; is a cover of
PCH,. Ifx € P, then x € P C P; C G; and there is at least one U € Ui;
such that UN O;-1 = 0. If U € Uj41 and UN O; £0, then if V € U; where
U CV, we have Vn O; # (, and hence V is not in G;. Thus an Ons # 0),
which implies VN P € V;. Therefore V;+ is a refinement of V;. By (18.B.20)
cov(P) < n-—1, since diam(V;) < 1/i for allV EV; O
As a consequence of this last theorem: if X is a metric space, a theorem

holds for cov if and only if it also holds for Ind, e.g., with no additional work
we have the following theorem.
(18.B.22) Corollary (The Sum Theorem for Ind). If X = U2, X; is

a metric space and for each 7, X; is closed in X, and Ind(X;) < n. Then
Indy) <n. ol
C. SOME FINAL RESULTS
The first goal of this section is to show that the three notions of dimension
we have considered agree on the class of separable metric spaces. Then we
prove a some additional basic theorems for cov = Ind which hold in general
metric spaces.
(18.C.1) Exercise. Let X be a metric space. Show: Ind(X) < n if and

only if for each pair of disjoint closed sets C, D there is a partition P of X
between C' and D with Ind(P) <n-—-1.
Some final results 455
(18.C.2) Theorem. Suppose that X is a separable metric space. Then

ind(X) = Ind(X) = cov(X).
Proof. It is immediate from the definitions that ind(X) < Ind(X). By
(18.B.21) it suffices to show either Ind(X) < ind(X) or cov(X) < ind(X).
We show Ind(X) < ind(X). The result is trivial if ind = oo. Hence we
assume ind(X) < oo. Furthermore, if ind(X) = 0 the theorem is true as a
result of (18.4.7). Hence, suppose ind(X) = n > 1 and that the theorem
holds for all separable metric spaces of little inductive dimension < n. Let
C and D be a pair of disjoint closed sets in X. By (18.A.20), there is a
partition P of X between C and D with ind(P) <n-—1. By the induction
assumption Ind(P) = ind(P) <n—1. Hence Ind(X) <n by (19.C.1). O
(18.C.3) Corollary. For €", cov(E") =Ind(E") =n. O
Because of (18.C.2) and (18.B.21), we are led to introduce the following
notational convention.
(18.C.4) Notation. If X is a metric space we denote cov(X) = Ind(X)

by dim(X). Furthermore, if X is a separable metric space we also denote
ind(X) by dim(X).
Our next goal is monotonicity for cov and Ind. As is usually the case, we
need a series of preliminary results. We obtain monotonicity for Ind. Then
by (18.B.21) it also holds for cov.
(18.C.5) Exercise. (a) Prove the easy part of the Alexander and Uryson
theorem (10.C.5). Namely, if X is a metric space show the existence of a
development Ve U2,... such that for each n > 1, whenever U,V € U, with
UNV £9 there is aW €U,_; such that U UV C W. (b) Show that the
U,’s above may be chosen so that mesh(U,,) < 1/2". (c) Show that there
are open covers V,, n = 1,2,... such that V = eh Y; is a basis for X and
mesh(V,) < 1/2". (d) Show that there are locally finite open covers By,
n =1,2,... such that B = U2, B; is a basis for X and mesh(B,) < 1/2”.
(18.C.6) Theorem. Suppose X is a metric space and Ind(X) < n. Then
there are locally finite open covers B,, Bz,... of X with Jim. mesh(B;) = 0
such that B = U7, B; is a basis and Ind(Fr(B)) <n — 1 for each B € B.
Proof. By (18.C.5)d there are locally finite open covers U4; such that
U = Us, Ui; is a basis for X and mesh(U;) < 1/2*. For each i, (4.C.2) gives
us an open shrinkage W; of U;. Since Ind(X) < n, for each i and for each
U, € U; there is a By such that
W « C By C Ug and Ind(Fr (By)) <n—1.

For each i, let B; be the collection of all such B,. Then B = ales B; has the
desired properties. O
(18.C.7) Theorem. Suppose X is a metric space and 3), Bo,... is a se-

quence of locally finite open covers of X such that B = Us, Bi is a basis for
X, lim mesh(B;) = 0 and Fr(B) = 0 for each B € B. Then Ind(X) < 0.
2-700)
Proof. Let C and D be disjoint closed sets of X. For each 7 let
Ui = X\({B : Be Ui,Bj,BNC
= 9}.
Clearly each U; is closed. Also each U; is open since each B € B is open and
closed and U5_1B; is locally finite. Hence U;,U2,... is a sequence of open
and closed sets such that
Ui. > Ua De > Grand (Gee:

i=1
Similarly obtain a sequence W,, W2,... with
Wi.D) Wo Die and flee;

(fail
But O = U2, (Ui \ Wi) is open and closed with C C O C X \ D. Hence

Ind(X 102.1 3
(18.C.8) Theorem. If X = X, U X2 is a metric space with dim(X,) <n
and dim(X2) < 0, then dim(X) <n +1.
Proof. We prove the statement for Ind. Let C, D be disjoint closed sets
of X. Since X is normal there are open sets U; and V; such that C Cc Uj,
DCYV,; and U, NV, = 9. Since Ind(X2) < 0, there is an open and closed set
O, of X2 satisfying Ui xe OP ex \ (Vi M X2). Furthermore, since X
is completely normal, there are open sets U2 and V2 satisfying: CUO, C Us,
DU(X2\01) C V2, and U2NV2 = 9. It follows from the above, and from O;
being open and closed in X2, that Fr(U2) C X,. Hence, by (18.B.21) and
(18.B.9), Ind(Fr (U2)) < Ind(X1) < n. Therefore, Ind(X) < n+ 1, since Up
is an open set such that CCU2C X\D. O
(18.C.9) Theorem. Suppose X is a metric space. Then, Ind(X) < n if
and only if there is a sequence of locally finite open covers B,,B2,... with
jim mesh(B;) = 0 such that B= U7=, Bi is a basis and Ind(Fr(B)) <n-1
for each B € B.
Proof. (=>) This is just (18.C.6).
(<=) When n = 0 this is (18.C.7). Hence assume n > 0 and that the
theorem is true for all m <n. Let
C =|){Fr(B) : Be B} and D= X\C.

Some final results 457
We let the reader show that for each i, {Fr(Ba) : Ba € B;} is a locally
finite closed collection. Since Ind(Fr(B)) < n—1 for each B € B, we
apply (18.B.21) and (18.B.13) to obtain Ind(C) < n—1. Define BN D by
BOD={{BND : B eB}. Since D and BND satisfy the hypothesis for
(18.C.7), we have Ind(D) < 0. Now an application of (18.C.8) gives us that
Ind(X) =Ind(CUD)<(n-1)+1l=n. O
(18.C.10) Corollary (Monotonicity). Suppose X is a metric space and
Y CX. Then dim(Y) < dim(X).
Proof. It is an exercise to establish this for Ind. O
(18.C.11) Theorem. Let X be a metric space. Then, dim(X) < n if and
only if X = Uh! X; where dim(X;) < 0 for each i.
Proof. (=) This is a corollary of the proof of (18.C.9).
(<=) This is a corollary of (18.C.8). O
(18.C.12) Corollary. Suppose X; and X2 are subspaces of the metric
space X, then
dim(X, U X2) < dim(X;) SF dim(X2) +1.
We conclude this section with the product theorem.
(18.C.13) Theorem (Product Theorem). Let X and Y be metric spaces

with at least one non-empty. then
dim(X x Y) < dim(X) + dim(Y).
Proof. Only trivial cases are omitted by assuming dim(X) = n and
dim(Y) = m where n and m are finite. Induct on n +m. The theorem is
clearly true if n +m = —1. Hence assume true whenever n +m < d > 0.
By (18.C.9) there are sequences B,, B;,... and M,,Mg,... of locally finite
open covers of X, respectively Y, such that B = U2, Bi and M =U, M;
are bases for X, respectively Y; furthermore Ind(Fr(B)) < n — 1 for each
B € B and Ind(Fr(M)) < m-—1 for each M € M. Let
BY = 1B RMB eB;, and M € M;}.
For each i and j B; x M; is a locally finite open cover of X x Y, and

Usj—1 Bi x M; is a basis for X x Y. By (1.E.18), we have for each B € B
and MewmM
Ind(Fr (B x M)) = Ind(Fr x(B) x M) U(B x Fry(M)).
Then Ind(Fr x(B) x M) = Ind(B x Fry(M)) =n +m — 1 follows from the

induction hypothesis. Hence applications of (18.B.21) and (18.B.13) give us
Indkr(Bx M)jsntm—i. oO
PROBLEMS
Sections A
1. Show that ind(Cantor set) = 0.

2. Let 0 # X c €! and suppose X does not contain a nondegenerate
interval. Show that ind(X) = 0.
Prove: If ind(X) = 0 and @#Y C X, then ind(Y) =0.
Let Hr be the set of points in Hilbert space all of whose coordinates
are rational. Show indHr # 0 but for each x # y € He there is a
separation (U,V) with z € U and y € V. [This example is by Erdés.]
Prove the following variant of the sum theorem: Suppose X = Se gis

t=1
a separable metric space and let N be an integer. If X; is closed in X
and ind X; < M for each i, then ind X < max;(ind X;).
6. Write €? as the union of three subspaces whose little inductive dimen-
sions are 0.
If X> satifies ind(X2) = 0, prove that ind(X, x X2) = ind(X,)+ind(X9).
8. Let Hr be the set of points in Hilbert space all of whose coordinates
are rational. Show 7) indHr = 1; and ti) He is homeomorphic to
Hr x Hr.
9. Show that ind(M) =n where M is an n-manifold.

10. Find ind for Cantor’s leaky teepee.
Section B and C
Prove the sum theorem for Ind without making use of (18.B.21)
Prove the product theorem for cov without making use of (18.B.21)
Prove (18.B.10) for Ind without making use of (18.B.21)
ie
NS
eeeLet X be a compact metric space show: dim(X) < n implies that for
each metric p on X and for each € > 0 there is a finite open cover U
such that mesh(/) < € and ord(U) < n.
5. Suppose X is a metric space and dim(X) < n > 1. Show that for each
k =0,1,2,...,n —1 there is a closed subspace Y with dim(Y) = k.
APPENDIX
This appendix includes material that is frequently omitted from elementary

courses in group theory. We will is assume the reader is familiar with the
basic concepts of group, homomorphism, isomorphism, normal subgroup,
quotient group, etc. Much of the material discussed here may be found in
Fox [1953], Fox, Artin [1948] and Robinson [1980].
A. FREE GROUPS
(Ap.A.1) Definition. Suppose A is a set. A syllable on A is an element of
Ax Z. We denote (a,n) by a”. A word on A is a finite sequence of syllables.
The word with no syllables is denoted by 1.
Let W(A) denote the set of all words on A. Define a binary operation
on W(A) , i.e., a function -, that maps W(A) x W(A) into W(A) by
(at da) ae, be ee a an? eaten

a 2 i bie.
The operation - converts W(A) into a semigroup (the word semigroup on A),
i.e., the binary operation - is associative, and W(A) has an identity element,
i
(Ap.A.2) Definition. Let W(A) be the word semigroup on the set A. Sup-
pose: a € A, wi, we € W(A), w = wia°we, u = wi 2, t = wya"a™ we, and
s=w a"t™ we. We say u is obtained from w by an elementary contraction
of type I and w is obtained from u by an elementary expansion of type I.
Also, the element s is obtained from t by an elementary contraction of type
II, and t is obtained from s by an elementary expansion of type II.
460 Appendix
Two words w and d of W(A) are equivalent, denoted w ~ %, if and only

if w can be obtained from @ by a finite sequence of elementary expansions
and contractions.
The proofs of the following two lemmas are trivial and are omitted.
(Ap.A.3) Lemma. The relation ~ is an equivalence relation. O
(Ap.A.4) Lemma. The equivalence relation ~ agrees with the product on
W(A), i-e., if wi ~ we and w3 ~ wa, then wi -w3 ~~ wWe-ws. O
Note that a° ~ 1, for each a € A.

(Ap.A.5) Notation. If A is a set, F(A) will denote {[w] : w € W(A)},
where [w] is the equivalence class of w determined by ~.
The proof of the next theorem is straightforward.
(Ap.A.6) Theorem. The set F[A] with the binary operation - defined by
[w]-[u] =[w-u]isagroup. O
(Ap.A.7) Examples.
1. If A=9, then F[A] = {[1]}.
2. If A= {a}, then FA] is infinite cyclic.
(Ap.A.8) Definition. Suppose G is a group and E is a subset of G. The
subgroup H generated by E is defined to be (\{S : E C S and S isa
subgroup of G}. If the subgroup generated by E is G, then E is called a set
of generators for G.
Note that every group has at least one set of generators (namely itself).
(Ap.A.9) Exercise. If A is a set, show that {[a] : a € A} is a generating
set for F[A].
(Ap.A.10) Notation. Henceforth, we follow tradition and denote the
equivalence class [a] by a.
(Ap.A.11) Definition. A generating set E of a group G is a free basis for
G if and only if for each group H and for each set. cae g@: E - H, there
is a homomorphism ¢ : G + H such that ¢|2 = ¢. It is clear that such an
extension is unique.
(Ap.A.12) Definition. A group G is called a free group if and only if G

has a free basis.
(Ap.A.13) Remark. The reader should note that a free group with a
free basis containing more than one element is not abelian. Hence, free
groups should not be confused with free abelian groups (which are free in
the category of abelian groups and homomorphisms, but not in the category
of groups and homomorphisms).
Group Presentations 461
The next theorem is crucial.
(Ap.A.14) Theorem. A group G is free if and only if G is isomorphic to

F[A] for some set A.
Proof. We first show that if A is a set, then F'[A] is free with free basis
A. Suppose H is a group and ¢: A > H. Define the map ¢: W(A) ~ H
by g(arb™-.-) = (p(a))” (d(b))™ --- whenever a"b™--- € W(A). Note
$(wia°w2) = $(wi)($(a))°b(we) = $(w1)b(we) = (w;w2), and similarly
H(wrara™we) = 4(w1)(4(a))"(G(a))™4(we) = 4(wi)((a))"*"4(we) =
o(wia"t™we). Hence, the map ¢: F[A] > H defined by ¢({a"][b™]---) =
(¢(a))"(o(b))™ --+ is well defined and a homomorphism. Furthermore, it is
clear that bla = @:
Suppose ~ : G - FIA] is an isomorphism. Then clearly ~1(A) is a
free basis for G.
Conversely, if G is free with free basis E, then G is isomorphic with
FE]. To see this, let 9 : E — F[E] be the inclusion map. Then there
is a homomorphism ¢ : G + FE] that extends 9. Since @ is one to one,
we have 6-1 : E > G, and again since E is a free basis for F[E], there is
a homomorphism w : F[E] > G that extends @~!. Thus wd and ¢y are
extensions of 6-16 = 1|, and 6971 = l|z, respectively, and consequently,
by the uniqueness of the extensions (Ap.A.11), we have both w¢ = idg and
wy = idpg). Hence ¢ and w are isomorphisms. O
(Ap.A.15) Corollary. Every group is the homomorphic image of a free

group. .
Proof. Let G be a group. Then clearly F[G] can be mapped homomor-
phically ontoG. O
B. GROUP PRESENTATIONS
(Ap.B.1) Definition. A group presentation {A ; R} consists of an arbitrary

set A and a set R of elements of F[A]. The group determined by the presen-
tation {A;R} is the quotient group F[A]/(R), where (R) = (\{N|R CN
and N is a normal subgroup of F[A]}. The set R is called the set of relations
for the presentation; and an element g € (R) is called a consequence of R;
the set A is called the set of generators.
In the text, we frequently abuse the terminology by referring to the

“sroup {X ; R}”; of course, by this we mean F[X]/(R).
(Ap.B.2) Definition. A presentation of a group G consists of a group
presentation {A;R} and an isomorphism 7 from the group determined by
the presentation {A; R} onto G.
462 Appendix
Often, we refer to {A; R} as a presentation of a group G without specif-

ically giving the isomorphism ~.
(Ap.B.3) Exercise. Suppose G is a group, {A;R} is a group presenta-
tion, and ¢ : F[A] > G is a homomorphism from FA] onto G such that
ker ¢= (R). Show that 7 determines a presentation of G (consider the fol-
lowing diagram, where p is the quotient map and ? is defined in the obvious
manner).
F(A]
{A; R} 7 G
(Ap.B.4) Exercise. Show: every group has a presentation. {Hint: Use
(Ap.A.15).]
(Ap.B.5) Examples.
1. {x; } and {z,y;2} are presentations for the infinite cyclic group.
2. {x,y;cyxz~‘y~'} is a presentation of a free abelian group on two
generators (note, this is not a free group).
3. {x;x7} is a presentation for Zz (the group of integers mod 2).
4. {x,y;cyz~1y—}, x”, y®} and {xz; 2°} are presentations for Ze.
5. {z,y;2z7, y”, (cy)"} is a presentation for the dihedral group of order
2n.
(Ap.B.6) Remark. Occasionally, we find it convenient to misuse the no-

tation and write a relation in the old fashioned form x7y? = 1, 2° = 1, etc.,
instead of x73, x°, etc.
(Ap.B.7) Definition. A presentation {X ;R} is finitely generated if and
only if X is finite; {X ; R} is finitely related if and only if R is finite. {X ; R} is
finite if and only if both X and R are finite. A group G is finitely generated if
and only if there is a presentation of G that is finitely generated; G is finitely
related if and only if there is a presentation of G that is finitely related; G is
finitely presented if and only if there is a presentation of G that is finite.
(Ap.B.8) Remark. Finiteness does not work the way that it should. There
are examples of finitely presented groups which have subgroups that are
not finitely presented. The enterprising reader might seek an example of a
nonfinitely generated subgroup of {z, y; }.
(Ap.B.9) Definition. Given two disjoint groups G and H, the free product
of G and H, denoted by G x H, is the group generated by all the elements
of G and all the elements of H, subject to the relations g:g293 1 for all
91,92,93 € G with gig2 = g3, and hyhgh;* for all hi,ho,h3 € H with
hy he => hz.
Homomorphisms and Tietze's Theorem 463
(Ap.B.10) Exercise. Show that if {21,...,2m j1r1,... Tn} is a presenta-

tion of a group G and {y,...,Yp;$1,...,$q} is a presentation of a group H,
then G x H is isomorphic to a group with presentation
Sie CaplisaU
ns ls a aePars nesses
C. HOMOMORPHISMS AND TIETZE’S THEOREM
A major problem that arises is to decide when two presentations determine

isomorphic groups. It has been shown (Rabin {1958]) that there is no general
solution to this problem. Partial solutions do exist, however, and in this
section we discuss a few such results.
(Ap.C.1) Definition. Suppose {X ;R} and {Y;S$} are presentations. A

homomorphism f : F[X] + F[Y] is a presentation mapping between the
presentations {X ;R} and {Y;S} (denoted f : {X;R} — {Y;S}) if and
only if f(r) € (S) whenever r € (R).
(Ap.C.2) Theorem. Every presentation map f : {X;R} > {Y;5S} de-
termines a unique homomorphism f, : F[X]/(R) > F[Y]/(S) such that the
following diagram is commutative. (px and py are the quotient maps.)
F(X] —_1—- FIY]

Px
F[X]/(R)—— FIY/(S)
Proof. Note that f, = p,fpx' is a well defined map and is the desired
homomorphism. O
(Ap.C.3) Exercise. Let {X;R} be a group presentation and G be a

group. Show that a homomorphism from f : F[X]/(R) — G is determined
by mapping generators of F[X] to elements of G in such a manner that if
eee ie --.g5” € Rand f(a;') = 9;', then g}*957 -- Ope cal
(AP.C.4) Definition. Presentation maps f,,fo : {X;R} > {Y;S} are

homotopic (denoted by f; ~ fz) if and only if for all c € X we have
fi(z) fo(z—*) € (S).
(Ap.C.5) Exercise. If f,f: {X;R} + {Y;S} and g,g:{Y;S} > {Z;T}
are presentation maps, show:
(i) f ~ f if and only if f, = fh }
(ii), if f= f and g 2g, thengf- of;
464 Appendix
(iii) for each homomorphism ¢ : F[X]/(R) > F[Y]/(S), there is a pre-

sentation map f : {X;R} > {Y ;S} such that f, = ¢, and further-
more, any two such presentation maps are homotopic;
(iv) (ofx = 94Fe3
(v) id: {X ;R} — {X;R} is a presentation map, and id, = idp;x1/,r)-
(Ap.C.6) Definition. Presentations {X ;R} and {Y ;S} are of the same
type if and only if there are presentation maps f : {X;R} — {Y;S} and
g:{Y;S}— {X;R} such that gf ~ id and fg ~ id. In this case, f (or g)
is called a presentation or homotopy equivalence.
(Ap.C.7) Theorem. Presentations {X ;R} and {Y ;S} are of the same

type if and only if F[X]/(R) is isomorphic to F[Y]/(S).
Proof. If ¢ : F[X]/(R) > F[Y]/(S) is an isomorphism, then so is ¢7!.
Use part (iii) of (Ap.C.5) to obtain presentation maps f and g. Since ¢-'¢
and ¢¢~! are identities, we have gf ~ id and fg ~ id.
Conversely, suppose f and g are presentation maps such that gf ~ id
and fg ~ id. Then by parts (i), (iv), and (v) of (Ap.C.5), we have 9,f, = id,
and f,g, = id,. Hence, f, and g, areisomorphic. O
(Ap.C.8) Definition. Suppose {X ; R} is a presentation. Let s € (R) and

suppose {Y ;.S} is a presentation with Y = X and S = RU{s}. The identity
map I : F[X] > FT[Y] is a presentation mapping IJ : {X;R} > {Y;S}.
Similarly I’ : F[Y] + F[X] is a presentation map I' : {Y;S} > {X; RB}.
The mappings J and I’ constitute Tietze equivalencies of type I and I’,
respectively.
(Ap.C.9) Theorem. The mappings I and I’ are presentation equivalencies,

and hence F'X]/(R) is isomorphic to F'[Y]/(S).
Proof. Consider the following diagram.
ery
P Pp Pp
FUXI/(R) FIV ]/(RU {8})5r~ F[X]/(R)

Here, J and I’ are identity maps. It is easy to see that they are both pre-
sentation maps. Furthermore, I'I is the identity presentation map from
{X ;R} onto {X;R}. Therefore, by (Ap.C.5) parts (iv) and (v), we have
F1, = td|rpqjcry. Similarly, L.0) = tdlppcruceqys 0
(Ap.C.10) Remark. The import of (Ap.C.9) is that given a group presen-

tation, one can add a relation that is a consequence of the others, or delete
Homomorphisms and Tietze’s Theorem 465
a relation that is a consequence of the others, without significantly changing

the group of the presentation.
(Ap.C.11) Example. The group {z; 2%} is isomorphic
to {x; 23, cg
(Ap.C.12) Definition. Let {X ;R} be a group presentation. Suppose y
is an element of some set and y ¢ X. Let € € F[X], and let {Y;S} be
the presentation, where Y = X U {y} and S = RU {yé-1}. The func-
tion IJ : F[X] > F[Y] defined by JI(z) = z gives us a presentation map
T:{X;R} — {Y ;S} and II’: F[Y] > F[X], defined by
TGA i ify
2 al 2g GX
yields a presentation map II’: {Y ;S} + {X;R}. The mappings JI and II'
constitute Tietze equivalencies of types II and II', respectively.
(Ap.C.13) Theorem. The mappings JJ and II' are presentation equiva-

lencies, and hence F[X]/(R) and F[Y]/(S) are isomorphic.
Proof. Note that I'IT(x) = id(x). Also we have that
te jt a (SMG
M2) = {
eet =v
and since £y~! = (yé—!)~1 € (S), it follows that II’ ~id. O

(Ap.C.14) Remark. The import of (Ap.C.13) is that a new generator may
be added to a group presentation, provided that a new relation is added that
expresses the new generator as a word in the old ones. Similarly, one can
delete a generator if it is expressible as a word in the other generators.
(Ap.C.15) Example. We have
fopy snyeziy*; (27, y°}

240, yap ye, c= 1, y= Tf
~ {a,y, 23ty=ye, 2 =1,y° = 1,2 =a}
o~{c,y,7;27=9t,c=0 y=, 2 =2y}
Now, since z = yt = y2z7! =y'(y-!a271) =y7'z"", it follows that
{@, y, 2.2=yt,c=e,y=y, 2=2y}
Min. te =Ve=2. yay, 2=2y}

ls ese zgiza=at,z?=2', z=az77}
Beeler prea’ = 1.2? = 2}
eit gf zbrenuee fe = (1 oes fie”)
466 Appendix
Although we do not use the following theorem, it is of fundamental

interest in presentation theory. A straightforword proof can be found in
Magnus, Karass, and Solitar [1966].
(Ap.C.16) Theorem (Tietze Theorem). If {X ; R} and {Y ; S} are finite
presentations for isomorphic groups, then there is a finite sequence of Tietze
Equivalencies J, I', IT, II' that change one presentation into the other.
D. DIRECT LIMITS OF GROUPS
(Ap.D.1) Definition. A direct system of groups, (Ga, fag,D) consists of a

directed set D, a family of groups {Ga}aep and a family of homomorphisms
{fag : a,3 € D anda < PB} such that
(i) fap: Ga Ge,
(ii) ifa < 6B <7, then fo, = feyfeg, and
(iii) foe = td for eachia:
Let X.cpGq be the direct product of the groups Gag, ie. XacpGa =
{{z}vep € [Taep Ga : Ta € Gq and tq = le, for all but a finite number
of a}, where the group operation is defined coordinatewise.
(Ap.D.2) Exercise. Show that for each group H and for each family
of homomorphisms {hg : Ga > H : a € D} such that ha(ga)he(ge) =
he(ga)he(ga) for all a,8 € D and for all gg € Gag and gg € Gg, there
is a unique homomorphism h : X.¢pGa — H such that for each a € D,
hig = ha, where iq is the natural injection from G, into the direct product.
(Ap.D.3) Definition. Let K be the smallest normal subgroup of x.¢pGa
generated by {ia(g)~‘igfas(g) : a,8 € Dia < B, g € Ga}. The direct
limit of the system (Ga, fag,D) is (XaenGa) /K and is denoted by lim Gg.
(Ap.D.4) Theorem. Suppose (Ga, fag, D) is a direct system of groups and
lim G', is the corresponding direct limit. For each a, define a homomorphism
fa : Gy + limG, by fa = pia, where p is the quotient homomorphism
(we denote p(g) by [g] or [g]x) from x.epGq to iimGa = (NeenGe l/h:
Then whenever a < £ the following diagram is commutative, and limG, is
generated by U..,, Im fa:
iim Ge
an
Ga—G5—> Ga
Proof. Suppose rg € Gg. Then ig(ta)~1t¢ fag(za) € K, by the defini-

tion of K. Thus [ia(ta)] = [ig foe(%a)], and hence fo(re) = fa(fas(ta)).
Direct Limits of Groups 467
It is clear that fim Go is generated by U,., Im fa, since X.enGo is

generated by U_.,aeéD Imig. O
(Ap.D.5) Examples.
1. Let D=N. For eachn €N, let G, = Z and set f,;(z) = z for all
moi. Then iim G, is isomorphic to Z.
2. Let D=N. For each n €N, let Gp = {m2" : m € Z} with +
as the group operation, and set f,;(z) = 2*-"z for n < i. Then iim G,, is
isomorphic to Z.
3. Let D=N. For each n €N, let G, = Z and set fyi(z) = 2'-"z for
n<i. Find limG,.
(Ap.D.6) Exercise. Suppose that A, B, and C are groups, and f : A > B,
g : AC are homomorphisms such that f is onto. There is a homomor-
phism h: B > C such that hf = g if and only if ker f C kerg.
(Ap.D.7) Theorem. Let (Ga, fag,D) be a direct system of groups and

iimG, be the corresponding direct limit. Suppose H is a group and that
{ha : Ga + H : a € D} is a family of homomorphisms such that for each
a < £, the following diagram is commutative. Suppose further that H is
generated by U,., Imha.
: eae
Then there is a unique homomorphism h from lim G, onto H such that for
each a < #3, the following tetrahedral diagram is commutative.
im
fim Ge
Re
Ga ——> H
fap we
Ge
Proof. Since iimG is generated by U,., Im fa, it suffices to define h

for the generating set {fa(g) : a€ D,g € Ga}. Let fa(g) be a generator of
468 Appendix
iim G,. Since we want the following diagram to be commutative, it follows

that h must be defined by h(fa(g)) = he(g).
Ve
fim Ge
We show: if k € K, then h(k) = 1,, and hence h may be extended ho-

momorphically to all of fim Go. Suppose k is a generator of K, ie., k =
ia(9a)
‘ig fae(ga)- Then h(k) = 1y if and only if h(fo(ga)) = h( fe fap (9a))-
By the definition of h, h(fa(Ja)) = ha(ga), and h( fa faa(ga)) = ha(faa(9a)):
We have ha(fas(ga)) = he(ga) from the commutativity of the following di-
agram. Hence h may be extended to limGy,.
Jal
oes
Ga fap Ge
The remaining properties of h are easily established.
(Ap.D.8) Theorem. Suppose (Ga, fag, D) is a direct system of groups,

lim Gy is the corresponding direct limit, H is a group, we have a family of
homomorphisms {ha : Ga + H : a € D} such that H is generated by
Ue, Im ha, and for all a < 6 the diagram below is commutative.
aes
Ga fap Ge
Suppose further whenever G is a group and {ho Gay Gora € Dh} isa
family of homomorphisms such that the following diagram is commutative
for each a < £,
7X
G
Go 23 Ge
Direct Limits of Groups 469
then there is a unique homomorphism h : H > G such that the following

diagram is commutative.
Then, H is isomorphic with jim Go.

Proof. The proof is almost identical to that of (12.F.8). O
(Ap.D.9) Theorem. Suppose G, = {21,22,...,%m, 371,12)-> 2 speak
Mn < Mnii1, and Pn < Pnii for each n € N. Furthermore, suppose

thatetntin Gy eGpnsdetined by fryi.n(2,) = ci for 1 <. 4s,
is a monomorphism for each n. Let G = {%1,2%2,...571,T2,-.-}. Then
==
limG, =G.
Proof. The theorem follows easily from (Ap.D.8). 0
S
—> Sire ; .
7 + ey be erouled
bo
8» heed: Pie ke VP
DFID nde aM et © =I
G NOT Tah Sly coy wy if antl aR Fit Yo alae

- ga 22 A aay ft dis
= Ht at uc = of
mae
‘ant 2
paved (CE ae
oA bos PB
gh)
ohaue whined?! Vo o) dupe
wre) oP ahet = Tey ee beehaly saul eds ggechtelt

i co cre es hpreniontom nat
S\ = av
" D {Stl qd) motviteee aeniit ofetenty aT *eor, _
<5 ; /.
Bibliography.
ALEKSANDROV, P. [1927]: Simpliziale Approximationen in der allgemeinen
Topologie, Math. Ann. Vol. 96, pp.489-511
ALEKSANDROV, P. [1939]: Bikompackte Erweiterungen topologishe Raume,
Mat. Sb. Vol. 5, pp.403-423
ALEKSANDROV, P. and Uryson, P. [1923]: Une condition nécessaire et suff-
isante pour qu’ne classe (L) soit une classe (D), C. R. Acad Sci. Paris
Vol. 177, pp.1274-1277
ALEKSANDROV, P. and Uryson, P. [1929]: Memoire sur les espaces topologi-
ques compactes, Verh. Ak. Wet. Amst. Vol. 14, pp.1-96
ALEXANDER, J. W. [1924] : On the subdivision of 3-space by a polyhedron,
Proc. Nat. Acad. Sci. Vol. 10, pp.6-8
ALEXANDER, J. W. [1924]: An example of a simply connected surface bound-
ing a region with is not simply connected, Proc. Nat. Acad. Sci. Vol. 10,
pp.8-10
ANDERSON, R. D. [1958]: One-dimensional continuous curves and a homo-
geneity theorem, Ann. of Math. Vol. 68, pp.1-16
ANDERSON, R.D. [1966]: Hilbert space is homeomorphic to the countable
infinite product of lines, Bull. Amer. Math. Soc. Vol. 72, pp.515-519
ANDERSON, R. D. and BING, R. H. [1968]: A complete elementary proof that
Hilbert space is homeomorphic to the countable infinite product of lines,
Bull. Amer. Math. Soc. Vol. 74, pp.771—792
ANDERSON, R. D. and CHOQUET, G. [1959]: A plane continuum no two
of whose nondegenerate subcontinua are homeomorphic: an application of
inverse limits, Proc. Am. Math. Soc. Vol. 10, pp.347-353
ANTOINE, L. [1921]: Sur ’homeomorphie de deux figures et de leurs voisinages,
J. Math. pures et appl. Vol. 8, pp.221-325
ARENS, R. and DuGUNDJI, J. [1951]: Topologies for function spaces, Pac. J.
Math. Vol. 1, pp.5-31
ARMENTROUT, S. [1966]: Monotone decompositions of E*, Topology Semi-
nar, Wisconsin, 1965. Annals of Mathematical Studies, No. 60, Princeton
University Press, Princeton. pp.1—-27
ARMENTROUT, S. [1967]: Topology Conference, Arizona State Univ., Tempe,
Arizona, pp.22-35
ARMENTROUT, S. [1968]: Concerning cellular decompositions of 3-manifolds
that yield 3-manifolds, Trans. Am. Math. Soc. Vol. 133, pp.307-332
ARMENTROUT, S. [1970]: A decomposition of FE? into straight arcs and single-
tons, Diss. Math. Rozp. Mat. Vol. 68, p.46
ARMENTROUT, S. and PRICE, T. [1969]: Decompositions into compact sets
with UV-properties, Trans. Am. Math. Soc. Vol. 141, pp.433-442
ARMSTRONG,M.A., RouRKE, C.P., and CooKE, G.E. [1968]: The Princeton
notes on the Hauptvermutung, Princeton Press, Princeton
BALL, B.J. [1984]: Arcwise connectedness and the persistence of errors, Am.
Math. Monthly. Vol. 91, pp.431-433
471
472
BELLAMY, D. [1971]: Mappings of indecomposable continua, Proc. Am. Math.

Soc. Vol. 30, pp.179-180
BinG, R.H. [1946]: The Kline sphere characterization problem, Bull. Am.
Math. Soc. Vol. 52, pp.644-653
BING, R.H. [1947]: Extending a metric, Duke Math. J. Vol. 14, pp.511-519
BING, R.H. [1948]: A homogeneous indecomposable plane continuum, Duke
Math. J. Vol. 15, pp.729-742
BinG, R.H. [1949]: A convex metric for a locally connected continuum, Bull.
Am. Math. Soc. Vol. 55, pp.812-819
BinG, R.H. {1951a]: Snake-like continua, Duke Math. J. Vol. 18, pp.653-663
BING, R.H. [1951b]: Concerning hereditarily indecomposable continua, Pac.
J. Math. Vol. 1, pp.43-51
Binc, R.H. [1954]: Locally tame sets are tame, Ann. of Math. Vol. 59,
pp.145-158
BinG, R.H. [1957a]: A decomposition of EZ? into points and tame arcs such
that the decomposition space is topologically different from E*, Ann. of
Math. Vol. 65, pp.484-500
BING, R.H. [1957b]: Upper semi-continuous decompositions of E*, Ann. of
Math. Vol. 65, pp.363-374
BING, R.H. [1959a]: The Cartesian product of a certain non-manifold and a
line is E*, Ann. of Math. Vol. 70, pp.399-412
BING, R.H. [1959b]: An alternative proof that 3-manifolds can be triangulated,
Ann. of Math. Vol. 69, pp.37-65
BING, R.H. [1962]: Pointlike decompositions of E?, Fund. Math. Vol. 50,
pp.431-453
BInG, R.H. [1963]: Approximating surfaces from the side, Ann. of Math.
Vol. 77, pp.145-192
BING, R.H. [1965]: A translation of the normal Moore space conjecture, Proc.
BIRKHOFF, G. [1937]: Moore-Smith convergence in general topology, Ann. of
Math. Vol. 38, pp.39-56
BorsuK, K. [1931]: Sur les retracts, Fund. Math. Vol. 17, pp.152-170
BoRSUK, K. [1931]: Quelques théorémes sur les ensembles unicohérents, Fund.
Math. Vol. 17, pp.171-209
BorsukK, K. [1932]: Uber eine Klasse von lokal zusammenhangenden Raumen,
Fund. Math. Vol. 19, pp.220—242
Borsuk, K. [1933a]: Drei Satze tiber die n-dimensionale Euklicische Sphare,
Fund. Math. Vol. 20, pp.177-190
Borsuk, K. [1933b]: Uber die Abbildungen der metrischen kompactam Raume
auf die Kreislinie, Fund. Math. Vol. 20, pp.224-231
BRAHANA, H.R. [1922]: Systems of circuits on two-dimensional manifolds,
Bropy, E.J. [1960]: The topological classification of the lens spaces, Ann. of
Math. Vol. 71, pp.163-184
Brouwer, L.E.J. [1910]: Beweis des Jordanschen Kurvensatz, Math. Ann.
Vol. 69, pp.169-175
473
Brouwer, L.E.J. [1912]: Beweis des Jordanschen Siitz fiir den n-dimension-
alen Raum, , Math. Ann. Vol. 69, pp.314-319
Brouwer, L.E.J. [1912]: Beweis des Jordanschen Satz fiir den n-dimension-
alen Gebietes, , Math. Ann. Vol. 71, pp.55-56
Brown, M. [1960a]: Some applications of an approximation theorem for in-
verse limits, Proc. Am. Math. Soc. Vol. 11, pp.478-483
Brown, M. [1960b]: A proof of the generalized Schoenflies theorem. Bull.
Brown, M. [1961]: The monotone union of open n-cells is an open n-cell,
Proc. Am. Math. Soc. Vol. 12, pp.812-814
Brown, M. [1962a]: Locally flat embeddings of topological manifolds, Topol-
ogy of 3-manifolds and Related Topics, pp.83-91. Prentice-Hall, Englewood
Cliffs, N.J.
Brown, M. [1962b]: A mapping theorem for untriangulated manifolds, Topol-
ogy of 3-mantfolds and Related Topics, pp.92-94. Prentice-Hall, Englewood
Cliffs, N.J.
Brown, R.F. [1974]: Elementary consequences of the noncontractibility of the
circle, Am. Math. Monthly Vol. 81, pp.247-252
BURGESS, C.E. [1959]: Chainable continua and indecomposibility, Pac. J.
Math. Vol. 9, pp.653-659
BurRGEss, C.E. [1961]: Homogeneous continua which are almost chainable,
Can. J. Math. Vol. 18, pp.519-528
BurRGEss, C.E. and CANNON, J.W. [1971]: Embeddings of surfaces in E?,
Rocky Mt. J. Math. Vol. 1, pp.259--344
Cantor, G. [1883]: Uber unendliche lineare Punktmannigfaltigkeiten, Math.
Ann. Vol. 21, pp.545-591
CAPEL, C.E. [1954]: Inverse limit spaces, Duke Math. J. Vol. 21, pp.233-245
CARTAN, H. [1937a]: Théorie des Filtres, C. R. Acad. Sci. Paris Vol. 205,
pp.595—-598
CaRTAN, H. [1937b]: Filtres et Ultrafiltres, C. R. Acad. Sci. Paris Vol. 205,
777-779
CASLER, B.G. [1965]: An imbedding theorem for connected 3-manifolds with
boundary, Proc Am. Math. Soc. Vol. 16, pp.559-566
Casson, A.J. [1967]: Generalizations and applications of block bundles, Fel-
lowship dissertation Trinity College Library, Cambridge
CaTLIN, D.E. [1968]: A short proof of the nest characterization of compactness,
Am. Math. Monthly Vol. 75, p.751
CHERNOFF, P.R. [1992]: A simple proof of Tychonoff’s theorem via nets,
Amer. Math. Monthly Vol. 99, pp.932-934
CHITTENDEN, E.W. [1927]: On the metrization problem and related problems
in the theory of abstract sets, Bull. Am. Math. Soc. Vol. 33, pp;13-34
CHRISTENSON, C.O. and OsBoRNE, R.P. [1968]: Pointlike subsets of a man-
ifold, Pac. J. Math. Vol. 24, pp;431-435
Coss, J.I. and VoxMAN, W.L. [1972]: Some fixed point results for UV de-
compositions of compact metric spaces, Proc. Am. Math. Soc. Vol. 38,
pp.156-160
474
Coss, J.I. and VOXMAN, W.L. [1980]: Dispersion points and fixed points,
Am. Math. Monthly, Vol. 87, pp.278-281
CouEN, D.E. [1954]: Spaces with weak topology. Quart. J. Math. Ozford,
Vol. 5, pp.77-80
ComFort, W.W. [1969]: A short proof of Marczewski’s separability theorem.
Am. Math. Monthly, Vol. 76, pp.1041—1042
CROWELL, R.H. and Fox, R.H. [1963]: Introduction to Knot Theory, Ginn
and Company, Boston
DAVERMAN, R.J. [1986]: Decompositions of Manifolds, Academic Press, Or-
lando
DEBNATH, L. and MIKUSINSKI, P. [1990]: Introduction to Hilbert spaces with
Applications, Academic Press, San Diego
DIEUDONNE J. [1944]: Uné généralization des espaces compacts, J. Math. pures
et appl. Vol. 23, pp.65—76
Do p, A. [1972]: Lectures on Algebraic Topology, Springer Verlag, Berlin
Do Lp, A. [1963]: Partitions of unity in the theory of fibrations, Ann. of Math.
Vol. 78, pp.223-255
DuGUNDJI, J. [1966]: Topology, Allyn and Bacon, Boston
Dyer, E. [1956]: A fixed point theorem, Proc. Am. Math. Soc. Vol. 7,
pp.662-672
Dyer, E. [1956]: Certain transformations which lower dimension, Ann. of
Math. Vol. 63, pp.15-19
EATON, W.T. and PIXLEy, C. [1975]: S! cross and UV© decompositions
of S* yileds S! x S1, Geometric Topology 1974, Lecture Notes Vol. 438,
Springer-Verlag, Berlin pp.166—194
EpGAR. G.A. [1990]: Measure, Topology and Fractal Geometry, Springer-
Verlag, New York
ENGELKING, R. [1977]: Dimension Theory, North Holland, Amsterdam
FALCONER, K. [1990]: Fractal Geometry: Mathematical Foundations and Ap-
plications,John Wiley & Sons, Chichester
FLEISSNER, W.G. [1984]: The normal Moore space conjecture and large cardi-
nals, Handbook of Set Theoretic Topology, (ed. Kunen and Vaughan), North
Holland, Amsterdam pp.733-760
Fox, R.H. [1943]: On homotopy type and deformation retracts, Ann. of Math.
Vol. 44, pp.40-50
Fox, R.H. [1945]: On topologies for function spaces, Bull. Am. Math. Soc.
Vol. 51, pp.429-432
Fox, R.H. [1953]: Free differential calculus I, Derivation in the free group ring,
Fox, R.H. and Artin, E. [1948]: Some wild cells and spheres in three-
dimensional space, Ann. of Math. Vol. 49, pp.979-990
FRASER, R.B. Jr. [1972]: A new characterization of Peano continua, Prace
Mat. Vol. 16, pp.247-248
FRINK, A.H. [1937]: Distance functions and the metrization problem, Bull.
Am. Math. Soc. Vol. 48, pp.133 -142
475
FuGATE, J.B. [1965]: Chainable Continua, Topology Seminar, Univ. of Wis-

consin, Princeton Univ. Press, Princeton. pp.129-135
FurCH, R. [1930]: Polyedrale Gebilde verschiedenes Metrik, Math. Zeit.
Vol. 32, pp.512-544
GREEVER, J. [1967]: Theory and Examples of Point-set Topology, Brooks/Cole,
Belmont Calif.
HAHN, H. [1914]: Mengertheoretische Charakterisierung der stetigen Kurve,
Akad. Wiss. Wien. Math.-Nat. Klasse Vol. 123, pp.1-57
HAKEN, W. [1962]: Uber das Homoomorphie Problem des 3-mannigfaltigkeiten,
Math. Zeit. Vol. 80, pp.89-120
HALMOS, P.R. [1960]: Naive Set Theory, Van Nostrand, New York
HALMOS, P.R. [1982]: A Hilbert space problem book, Graduate Texts in Math.
#19, Springer-Verlag, Berlin
HAMILTON, O.H. [1951]: A fixed point theorem for psuedo-arcs and certain
other metric continua, Proc. Am. Math. Soc. Vol. 2, pp.173-174
HANNER, O. [1951]: Some theorems on absolute neighborhood retracts, Ark.
Mat. Vol. 1, pp.389-408
HausporrF, F. [1930]: Erweiterung einer Homoémorphie, Fund. Math. Vol. 16,
pp.353-360
HAWKINS, T. [1970]: Lebesque’s Theory of Integration, Univ. of Wisc. Press,
Dal
HEATH, R.W. [1964]: Screenability, pointwise paracompactness and metriza-
tion of Moore-spaces, Can. J. Math. Vol. 16, pp.763-770
HEMION, G. [1992]: The Classification of Knots and 3-dimensional Spaces,
Oxford Press, Oxford
HENDERSON, G.W. [1964]: The pseudo-arc as an inverse limit with one binding
map, Duke Math. J. Vol. 31, pp.421—425
HEWITT, E. [1943]: A problem in set theoretic topology, Duke Math. J.
Vol. 10, pp.309-333
HEwitTT, E. [1960]: The role of compactness in analysis, Am. Math. Monthly
Vol. 67, pp.499-615
HitTon, P.J. and WYLIE, S. [1960]: Homology Thoery, Cambridge Univ.
Press, London
Hirsu, M.W. and ZEEMAN, E.C. [1966]: Engulfing, Bull. Am. Math. Soc.
Vol. 72, pp.113-115
HRBACEK, K. and JEcH, T. [1984]: Introduction to Set Theory, Marcel Dekker,
Inc., New York
Hu, S. [1959]: Homotopy Theory, Academic Press, New York
Hupson, J.F.P. [1969]: Piecewise Linear Topology, Benjamin, New York
HuREwicz, W. [1930]: Uber oberhalb-stetige Zerlegungen von Punktmengen
in Kontinua, Find. Math. Vol. 15, pp.57—60
HurREwICczZ, W. [1935;1936]: Beitrage zur Topologie des Deformationen I[-IX,
Proc. Ak. Wet. Amst. Vol. 38, (1935) pp.112-119, pp.521-528; (1936)
pp.117-126, pp.215—-224
HurEwicz, W. and WALLMAN, H. [1941]: Dimension Theory, Princeton Univ.
Press, Princeton
476
JOLLEY, R.F. and RocsErs, J.T. JR. [1970]: Inverse limit spaces defined by
only finitely many distinct bonding maps, Fund. Math. Vol. 68, pp.117-120
JONES, F.B. [1937]: Concerning normal and completely normal spaces, Bull.
JONES, F.B. [1951]: Certain homogeneous unicoherent indecomposible con-
tinua, Proc. Am. Math. Soc. Vol. 2, pp.855-859
KAMPEN, E.R. VAN [1933]: On the connection between the fundamental group
of some related spaces, Am. J. Math. Vol. 55, pp.261-267
KATETOV, M. [1952]: On the dimension of non-separable spaces I., Czech.
Math. J. Vol. 2, pp.333-368
KELLEY, L.J. [1950a]: Convergence in Topology, Duke Math. J. Vol. 17,
pp.277-283
KELLEY, L.J. [1950b]: The Tychonoff product theorem implies the axiom of
choice, Fund. Math. Vol. 37, pp.75-76
KELLEY, F.J. [1955]: General Topology, Van Nostrand, New York
KIRBY, R.C. and SIEBENMANN, L.C. [1969]: On the triangulation of manifolds
and the Hauptvermutung, Bull. Am. Math. Soc. Vol. 75, pp.742—749
KirBy, R.C., SEIBENMANN, L.C., and Wall, C.T.C. [1969]: The annulus
conjecture and triangulation, Not. Am. Math. Soc. Vol. 16, p.432
KNASTER, B. and KURATOWSKI, C. [1921]: Sur les ensembles connexes, Fund.
Math. Vol. 2, pp.206—255
KNIGHT, C.J. [1964]: Box topologies, Quart. J. Math. Ozford Vol. 15, pp.41-
54
KRESIMAR, D. and MARDESI¢, S. [1968]: A necessary and sufficient condition
for the n-dimensionality of inverse limits, Proceedings of the International
Symposium of Topology and its Applications, Belgrade, pp.124—129
LASHOF, R.K. and ROTHENBERG, M.G. [1968]: Hauptvermutung for man-
ifolds, Proc. 1967 Conference on Top. of Man., Prindle Weber Schmidt,
pp.81-105
LEVINE, N. [1960]: Remarks on uniform continuity in metric spaces, Am.
Math. Monthly Vol. 67, pp.562-563
MAGNus, W., KARRASS, A. and SOLITAR, D. [1966]: Combinatorial Group
Theory, Wiley, New York
Markov, A.A. [1960]: Unsolvability of the homeomorphism problem, Proc.
Int. Cong. Math., 1958, Cambridge Univ. Press, Cambridge
MassgEy, W.S. [1967]: Algebraic Topology, An Introduction, Harcourt, Brace,
and World, New York
MAUNDER, C.R.F. [1970]: Algebraic Topology, Van Nostrand Reinhold, Lon-
don
MAZURKIEWICZ, S. [1920]: Sur les lignes de Jordan, Fund. Math. Vol. 1,
pp.166—209
McAULEY, L.F. [1962]: Upper semicontinuous decompositions of E? into E3
and generalizations to metric spaces, Topology of 3-manifolds and Related
Topics, Prentice-Hall, Engrlewood Cliffs N.J., pp.21-36
MENGER, K. [1923]: Uber die Dimensionalitat von Punktmenger I, Monat.
Fiir Math. und Phys. Vol. 33, pp.148-160
477
MILL, J. VAN and Reed, G.M. [1990]: Open Problems in Topology, North
Holland, Amsterdam
Minor, J. [1961]: Two complexes which are homeomorphic but combinato-
rially distinct, Ann. of Math. Vol. 74, pp.575-590
MolsE, E.E. [1948]: An indecomposable plane continuum which is homeo-
morphic to each of its nondegenerate subcontinua, Trans. Am. Math. Soc.
Vol. 63, pp.581-594
Molsg, E.E. [1949]: Grille decomposition and convexification theorems for
compact metric locally connected continua, Bull. Am. Math. Soc. Vol. 55,
pp.1111-1121
MoIsE, E.E. [1952a]: Affine structures in 3-manifolds II. Positional properites
of 2-spheres, Ann. of Math. Vol. 55, pp.172-176
MoIsE, E.E. [1952b]: Affine structures in 3-manifolds V. The triangulation
theorem and Hauptvermutung, Ann. of Math. Vol. 56, pp.96-114
Moise, E.E. [1954]: Affine structures in 3-manifolds VIII; invariance of knot
types, local tame embeddings, Ann. of Math. Vol. 59, pp.159-170
Monk, J.D. [1969]: Introduction to Set Theory, McGraw-Hill, New York
Moore, R.L. [1925]: Concerning upper semi-continuous collections of con-
tinua, Trans. Am. Math. Soc. Vol. 27, pp.416—428
Moore, E.H. and SMITH, H.L. [1922]: A general theory of limits, Am. J.
Math. Vol. 44, pp.102-121
Morita, K. [1954]: Normal families and dimension theory for metric spaces,
Math. Ann. Vol. 128, pp.350-312
MUNKRES, J.R. [1975]: A First Course in Topology, Prentice-Hall, Englewoods
Cliffs N.J.
NADLER, S.B. JR. [1970]: A note on inverse limits of finite spaces, Can. Math.
Bull. Vol. 13, pp.69-70
NADLER, S.B. JR. [1973]: The indecomposibility of the dyadic solenoid, Am.
Math. Monthly Vol. 80, pp.677-679
NADLER, S.B. JR. [1978]: Hyperspaces of Sets, Marcel Dekker, New York
NADLER, S.B. JR. [1992]: Continuum theory, an introduction, Marcel Dekker,
New York
NAGAMI, K. [1970]: Dimension Theory, Academic Press, New York
NaGaTA, J. [1950]: On a necessary and sufficient condition of metrizability, J.
Inst. Poly. Osaka City Univ. Vol. 1, pp.93-100
NEWMANN, M.H.A. [1964]: Elements of the topology of plane sets of points,
Cambridge Univ. Press, London
NEWMANN, M.H.A. [1966]: The engulfing theorem for topological manifolds,
O.uM, P. [1958]: Nonabelian cohomology and van Kampen’s theorem, Ann.
of Math. Vol. 68, pp.658-668
PAPAKYRIAKOPOULOS, C.D. [1943]: A new proof of the invariance of the ho-
mology groups of a complex, Bull. Soc. Math. Gréce Vol. 22, pp.1-54
PITTMAN, C.R. [1979]: An elementary proof of the triod theorem, Proc. Am.
Math. Soc. Vol. 25, p.919
478
PRICE, T.M. [1966]: A necessary condition that a cellular upper semicontinous

decomposition of E” yield E”, Trans. Am. Math. Soc. Vol. 122, pp.427—435
PRICE, T. [1969]: Mimeographed notes, Univ. of Iowa, Iowa City
RABIN, M.O. [1958]: Recursive unsolvability of group theoretic problems, Ann.
of Math. Vol. 67, pp.172-194
RADCLIFFE, J.G. [1994]: Foundations of Hyperbolic Manifolds, Graduate Texts
in Mathematics #149, Springer-Verlag, Berlin
Ravo, T. [1925]: Uber den Begriff des Riemannshen Flache, Acta Lit. Sci.
Szegad Vol. 2, pp.101-121
RANICKE, A.A. [1996]: The Hauptvermutung Book, Kluwer, Dordrecht
REIDEMEISTER, K. [1928]: Uber Knotengruppen, Abh. Math. Sem. Univ.
Hamburg, Vol. 6, pp.56—-64
REIDEMEISTER, K. [1935]: Homotopieringe und Linsenraume, Abh. Math.
Sem. Univ. Hamburg, Vol. 11, pp.102-109
REIDEMEISTER, K. [1983]: Knot Theory, BCS Associates, Moscow, Idaho
(Translation of 1932 Springer-Verlag edition)
ROBINSON, D.J.S. [1980]: A course in the theory of Groups, Graduate Texts
in Math. #80, Springer-Verlag, Berlin
ROoLFsoNn, D. [1970]: Characterizing the 3-cell by its metric, Fund. Math.
Vol. 68, pp.215-223
ROLFson, D. [1976]: Knots and Links, Publish or Perish, Inc., Berkeley
Rubin, M.E. [1969]: A new proof that metric spaces are paracompact, Proc.
RUSHING, T.B. [1973]: Topological Embeddings, Academic Press, New York
SCHORI, R. [1965]: A universal snake-like continuum, Proc. Am. Math. Soc.
Vol. 16, pp.1313-1316
ScHorI, R. and WEST, J.E. [1972]: 2/ is homeomorphic to the Hilbert cube,
Bull. Am. Math. Soc. Vol. 78, pp.402—406
SCHUBERT, H. [1964]: Topologie, B.G.Teubner, Stuttgart
SEIFERT, H. [1931]: Konstruction dreidimensionaler geschlossener Raume, Ber.
Sachs. Ak. Wiss. Vol. 83, pp.26-66
SEIFERT, H. and THRELFALL, W. [1934]: Lehrbuch der Topologie, B.G.Teubner,
Leipzig [Chelsea, New York (1945)]
SIEBENMANN, L. [1972]: Approximating cellular maps by homeomorphisms,
Topology Vol. 11, pp.271-294
SIERPINSKI, W. [1916]: Sur une courbe cantorienne qui contient une image
biunevoque et continue de toute courbe donnée, C. R. Sci. Paris Vol. 162,
p.629
SINGER, I.M. and THORPE, J.A. [1967]: Lecture Notes on Elementary Topol-
ogy and Geometry, Scott Foresman, Glenview, Illinois
SMIRNOV, YU.M. [1953]: On metrization of topological spaces, Amer. Math.
Soc. Transl. Ser. 1, #91
SPANIER, E.H. [1966]: Algebraic topology, McGraw-Hill, New York
STEEN, A.L. and Seebach, J.A. Jr. [1978]: Counterezamples in Topology, 2nd
Edition, Springer-Verlag, New York
479
STILLWELL, J. [1980]: Classical Topology and Combinatorial Group Theory,

Springer-Verlag, Berlin
STOCKER, R. and ZIESCHANG, H.. [1988]: Algebraische Topologie, B.G.Teub-
ner, Stuttgart
STONE, A.H. [1948]: Paracompactness and product spaces, Bull. Am. Math.
Soc. Vol. 54, pp.977-982
STONE, A.H. [1949]: Incidence relations in unicoherent spaces, Trans. Am.
STONE, A.H. [1956]: Metrizability of decomposition spaces, Proc. Am. Math.
Soc. Vol. 7, pp.690-700
STONE, A.H. [1959]: Metrizability of unions of spaces, Proc. Am. Math. Soc.
Vol. 10, pp.361-366
SULLIVAN, D.P. [1967]: On the Hauptvermutung for Manifolds, Bull. Am.
THURSTON, W.P. [1997]: Three-dimensional Geometry and Topology vol. 1,
Princeton Univ. Press, Princeton
TALL, F.D. [1984]: Normality versus collectionwise normality, Handbook of Set
Theoretic Topology, (ed. Kunen and Vaughn) Chapter 15, North Holland,
Amsterdam, pp.685—732
TieTze, H. [1915]: Uber Funktionen die auf einer abgeschlossenen Menge stetig
sind, J. R. Ang. Math. Vol. 145, 9-14
Tinonov, A. [1935]: Uber einen Functionenraum, Math. Ann. Vol. 111,
pp.762—766
URYSOHN, P. [1922] : Les multiplicités Cantoriennes,C.R.Acad. Paris Vol. 175,
pp.440-442
URYSOHN, P. [1925] : Zum Metrization Problem, Math. Ann. Vol. 94, pp.309-
315
VEBLEN, O. [1905]: Theory of plane curves in non-metrical analysis situs,
Trans. Am. Math. Soc. Vol. 6, pp.83-98
VoxMAN, W.L. [1970]: On the shrinkability of decompositions of 3-manifolds,
VoxMAN, W.L. [1972]: Decompositions of 3-manifolds and pseudoisotopies,
WALDHAUSEN, F. [1976]: Recent results on sufficiently large 3-manifolds, Proc.
of Sympos. in Pure Math. 32(2) pp. 21-38
WHITEHEAD, J.H.C. [1939]: Simplicial spaces, nuclei, and m-groups, London
Math. Soc. Proc. Vol. 45, pp.243-327
WHITEHEAD, J.H.C. [1949]: Combinatorial homotopy I, Bull. Am. Math.
Soc. Vol. 55, pp.213-245
Wuysurn, G.T. [1936]: On the structure of continua, Bull. Am. Math. Soc.
Vol. 42, 49-73
WILDER, R.L. [1929]: Topology of Manifolds, Am. Math. Soc. Collog. Publ.
#32
YounG, G.S. [1964]: A condition for the absolute homotopy extension prop-
erty, Am. Math. Momthly Vol. 71, pp.896-897
480
YounG, N. [1988]: An Introduction to Hilbert Space, Cambridge Univ. Press,

Cambridge
ZEEMAN, E.C. [1963]: Seminar on Combinatorial Topology, Inst. des Hautes
Etudes Sci., Paris
ZEEMAN, E.C. [1964]: Relative simplicial approximation, Proc. Camb. Phil.
Soc. Vol. 60, 39-43
ZEMKE, C. [1974]: Dimension, Decompositions, and Pseudo-Isotopies, Thesis
Univ. of Idaho, Moscow.
INDEX OF TERMS
Absolute homotopy extension property Ascoli’s theorem; 190
(AHEP); 377, 397 Axiom of Choice; 6
Absolute Neighborhood Retract (ANR);
106, 375, 397
W708
Absolute Retract (AR); 104, 189, 199
B”; 88, 202
absolute retract (ar); 105, 199
Abstract complex; 370
BUX.) )ei76
Abstract simplex; 370 Baire category theorem; 85

Baire space; 84
Abstract vertex; 369
Accumulation point; 25 Ball; 15, 202
Adherence of a filter; 280 Barycenter; 349
Adjunction; 198 Barycentric subdivision; 349
AHEP;; 377, 397 Base point; 290
Aleksandrov and Uryson metrization Base space; 317
theorem; 256 Basis; 13, 14
Alexander horned sphere; 137 ato 16

Alexander’s lemma; 153 equivalent; 14,
Annulus: neighborhood; 16
2-; 130 Bd; 406, 419, 421
n-; 423 Big inductive dimension; 446
half-open; 423 Bing’s metrization theorem; 259
open; 423 Bolzano-Weierstrass theorem; 80
Annulus conjecture; 423 Bonding maps; 158
Annulus theorem; 147, 423 Borsuk map; 379
ANR; (see Absolute neighborhood re- Borsuk’s top hat theorem; 375
tract) Borsuk-Ulam theorem; 337-338
ANRy;; (see Metric absolute neighbor- Bound:
hood retract), greatest lower; 4,
AR, ar; (see (A)absolute retract) least upper; 4
Arc. 56 lower; 4
polygonal-; 59 upper; 4
wild; 392 Boundary; 406, 407, 419
Arc-connected space; (see Connected Branch point; 383
space) Brick partition; 132
482 Index
Brouwer fixed point theorem; 50, 125, Filter; 275

187, 296, 359 Net; 272
Brouwer reduction theorem; 107 Sequence; 77
Collapses to; 371
Collectionwise normal; (see Topologi-
C((X, A), (Y, B)); 285
cal space)
COGS Ve Ly7, Commutative diagram; 5
Canonical neighborhood; 317
Compactification; 67
Cantor set; 86, 158, 165, 172, 458
Alexandrov; 68
Cantor space; 166 one point; 68
Cantor’s leaky teepee; 86, 458 Stone-Cech; 68, 189
Carrier; 343, 351 Compact space; 60-65, 80, 81, 98, 154,
Cartesian product; 3, 6 157, 161, 1775 1787 186. 190-197.
Category; 298 212, 221-248, 274, 277, 401, 432
first; 85 countably; 66, 79, 190
second; 85 locally; 65, 156, 186, 201, 207
Cauchy-Schwarz inequality; 9 meta; 266
Cauchy sequence; 82 pseudo; 75
Cell: real; 189
2-; 130, 401 sequentially; 65, 79, 80, 157, 186
n-; 202, 391, 421 a-; 253, 402
boundary of; 204 super; 171
closed; 202, 203 Complement; 2
closure; 204, Complementary domain; 139, 391, 428
collared; 430 Complete conjugate class; 322
open; 203, 430 Complete metric space; (see Metric space)
standard; 202 Completely normal space; (see Topo-
tame; 391, 423 logical space)
wild; 391, 396, 423 Completely regular space; (see Topo-
Cell decomposition; 203 logical space)
Cellular set; 424 Completion of a metric space; 263, 264,
Characteristic function; (See Function) 269
Characteristic map; 204 Complex:
Closed set; 23, CW-; (see CW-complex)
Closure; 23 generated by; 343
of a simplex; 342 linear cell; (see Linear cell complex)
Closure finite; 204 simplicial; (see Simplicial complex)
Closure operator; 32 Component; 55
Cluster point of a: path; 57
Index 483
[Component] Convex set; 44, 88

quasi; 70 Countable set; 8
Composant; 237 Countably compact; (see Compact space)
Cone (cone); 93, 197, 421 cov(X); 446
Connect points by a simple chain; 58 covering dimension; 446
Connected function; (see Function) Cover; 60
Connected space; 47, 53, 99, 154, 161, finite; 60
190, 211, 221-248, 317, 402, 420 open; 60
1-simple; 311 point finite; 109
arc; 55, 226, 420 o-locally finite; 259
im Kleinem; 245 Covering map; 317
locally; 57, 155, 200 Covering morphism; 326
locally arc; 55, 228 Covering space; 317
locally path; 54, 317 number of sheets of; 320
path; 53 regular; 322
semilocally simply; 329 universal; 329
simply; 295 Covering translation; 326
uniformly locally; 244 Covering translation group; 327
Connected sum; 417 Curvilinear simplex; 344
Consequence; 461 Cut point; 222
Constant loop; 284 CW-complex; 204, 315, 344
Contiguous; 372 finite; 204
Continuous; (see Function) infinite; 204
Continuum; 221 locally finite; 207
chainable; 231 n-dimensional; 204
decomposable; 236 oo-dimensional; 204
hereditarily indecomposable; 242 subcomplex; 206
indecomposable; 236
irreducible; 233
Peano; 225 Deckbewegung; (see Covering transla-
universal chainable; 243 tion)
Contractible; (see Topological space) Deckbewegungs group; 327
Contraction; 124 Decomposition:
Contractive map; 92 monotone; 214
Convergent: upper semicontinuous; 209
filter; 275 Decomposition set; 193
net; 273 Decomposition theorem; 444
sequence; 29 Defining sequence of chains; 232
Convex hull; 350 Degree map; 321
484 Index
DeMorgan’s laws; 2 Edge path; 361

Dense; 33 equivalent; 361
Derived set; 25 multiplication of; 361
Development; 256 reverse of; 361
Diameter; 79 Edge path group; 362
dim(X); 455 Elementary collapse; 371
Dimension: Elementary contraction; 459
0-; 46, 115 Elemenatry expansion; 459
n-; 46 Embedding; 31
of a linear cell; 350 e-chain; 231
of a simplex; 342 Endpoint (of path); 53
of a simplicial complex; 342 Equicontinuous family; 189
Directed set; 158 Equipotent; 8
Direct limit: Equivalence; 314
of direct systems of groups; 309, 466 Equivalence class; 4
of direct systems of topological spaces; Equivalence relation; (see Relation)
305, 3941 Equivalent basis; 14
Direct product of groups; 466 Equivalent metric; (see Metric)
Direct system: Equivalent words; (see Words)
topological spaces; 305 Equivalently embedded sets; 127
groups; 309, 466 E(R); 129
Disconnected; (see Topological space) Euclidean k-simplex; 342
Discrete family; 259 Euclidean n-sphere; (see S”)
Disk; 48 Euclidean polyhedron; 343
open; 13 Euler characteristic; 417
Dispersion point; 57, 71 Evaluation map; 176
Distance function; (see metric) Evaluation function; 179
Distance between sets; 30 Eventually in, for a:
Dog bone space; 216 filter; 275
Domain of a function; 4 net; 273
Dominates; 436 Explosion point; (see Dispersion point)
Dunce cap; 339, 372 ext; 26
Dyadic rationals; 2, 71, 101 Exterior; 26
Dyadic solenoid; 167, 240
F(A]; 460
E1; 48, 50, 52, 61, 69-71, 121 F, subset; 103, 247
E15 251,55,.62,/65 face; 342, 351
Edge loop; 361 Factor space; 158
Index 485
Feebly continuous; (see Function) feebly continuous; 44

Fibration; 337 idx; i)
Filter; 275 identity; 5, 45, 298
finer; 276 inessential; 376
neighborhood; 275 injective; 4
ultra; 276 inverse; 5
Filter base; 275 lower semicontinuous; 255
Filter subbase; 280 monotone; 172
Finite intersection property; 63 null homotopic; 376
First category; (see Category) one to one; 4
First countable; (see topological space) onto; 4
First derived complex; 349 open; 31, 37
First homotopy group; 290 restriction; 4
First infinite ordinal; 8 strongly continuous; 44
First uncountable ordinal; 8 super continuous; 72
Fixed point property; 50, 187 surjective; 4
Fox-Artin arc; 392 uniformly continuous; 81
Fr; 26 upper semicontinuous; 43, 209, 255
Free basis; 460 weakly confluent; 95
Free face; 371 weakly continuous; 43
Free group; 460 Functor:
Free product; 306, 462 contravariant; 299
Free union of sets; 22 covariant; 298
Frequently in (net); 273 forgetful; 299
Frink’s metrization theorem; 256 Fundamental group; 290
Frontier; 26
Fully normal; (see topological space)
G5 subset; 103
Function; 4
General position; 403
bijective; 4
Generalized Schonflies theorem; 431
bounded; 18, 178
Generators; 460, 461
characteristic; 43
Geometric complex; 343
closed; 31
glb; 4
composite; 5
Graph; 382
connected; 71
Greatest lower bound; (see Bound)
contiguous; 372
Group presentation; 461
continuous; 11, 12, 27
distance; 17
essential; 376 H; 184
evaluation; 178 Ho; 187
486 Index
H(A, B); 139 inf; 4

Hahn-Mazurkiewicz theorem; 228 Initial point of:
Hauptvermutung; 346 an arc; 50
Hausdorff metric; (see Metric) a path; 53
Hausdorff space; (see Topological space, Initial segment; 7
T2) int; 26
Heine-Borel theorem; 61, 62 Int; 406, 419
HEP; 397 Interior; 26, 406, 419
Hereditary property; 110 Interior operator; 42
Higher homotopy groups; 313 Intermediate value theorem; 50, 69
Hilbert cube; 90, 187 Intersection; 2
Hilbert space; 184, 458 Interval; 2, 48
Hom(X,Y); 175 Invariance of domain; 146, 186, 380,
Homeomorphic spaces; 31 435
Homeomorphism; 31 Inverse:
Homomorphism, induced; 293 left; 314
Homotopic functions; 123, 285 right; 314
relative; 285 Inverse image; 4
Homotopic presentation mappings; 463 Inverse limit; 158, 240, 248
Homotopic, relative to a subset; 285 Inverse set; 426

Inverse system; 158, 243
Homotopy equivalence; 293
Invertible; 438
Homotopy extension property; 397
Isolated point; 85
Homotopy inverse; 293
Isomorphic simplicial complexes; 370
Homotopy lifting theorem; 319
Isometry; 82, 263
House with two rooms; 372
Isotopy; 383
Hyperspace; 90
foot Join; 347

1(B"); 202 eee 347 ‘
I(R); 129 ordan-Brouwer theorem; 427
idx; (see Function) Jordan curve theorem; 125, 135,
Identification map; (see Map)

Im; 4 Klein bottle; 196, 217, 335, 418
Image of a function; 4 Kline sphere characterization; 225
ind(X); 440 Knastet-Kuratowski example; 85
Ind(X); 446 knot; 389
Independent points; 341 Knot group; 389
Induced map; 95 k-space; (see Topology)
Index 487
Kuratowski’s lemma; 6 Long interval; (see Topology)

Long line; (see Topology)
Loops; 284
L(az,y); (see Necking number)
product of; 286
Lebesgue number; 79
Lower bound; (see Bound)
Least upper bound;(see Bound)
Lower semi-continuous; (see Function)
fim; 305
lub; 4
Lim inf; 210, 246
Limit of a map between inverse sys-
tems; 163 Manifold:
Limit point of a set; (see Accumulation 1-; 34, 121
point) 2-; 34,
Limit point of a sequence; 29 3-; 35,
Limit of a sequence of sets; 210 n-; 34, 419
Lim sup; 210, 246 Manifold with boundary:
Lindelof space; 66, 113, 250 1-; 147
Lindelof theorem; 66 2-; 406
Linear cell; 350 n-; 419
boundary of; 351 Map:
face of; 351 between inverse systems; 163
interior of; 351 contiguous; 372
Linear cell complex; 351 gluing theorem; 29
subdivision of; 352 identification; 199, 218
Linear ordering; (see Ordering, linear) lifting theorem; 323
Linear subset; 94 retraction; 105
Link of a simplex; 347 sewing; 204
Link of a simple chain; 58 simplicial; 345
Little inductive dimension; 440 Mapping cylinder; 312
Local homeomorphism; 45, 149, 317 Mapping of pairs; 285
Locally compact space; (see Compact Mappings; 298
space) Maximal element; 4
Locally connected space; (see Connected Mesh of; 354
space) brick partition; 132
Locally finite: 344 cover; 162
CW-complex; (see CS-complex) Metacompact space; (see Compact space)
family; 249 Metric; 16
Locally metrizable; (see Topological space) discrete; 18
Locally path connected; (see Connected equivalent; 18
space) French railway; 39, 94
488 Index
[Metric] ING I
Hausdorff; 90 Nagata;Smirnov metrization theorem;
normal; 115 259, 267
post office; 38 Necking number; 138
pseudo-; 38 Neighborhood; 27
starfish; 266 Neighborhood basis; (see Basis)
sup; 178 Nerve; 370
taxicab; 18 Nest; 6, 64
topology; 17 . Net; 30, 272
usual; 17 universal; 279
Metric Absolute Neighborhood Retract No retraction theorem; 125, 296, 359
(ANRm); 107, 108, 256, 346, 371 Non-cut point; 222
Metrically isolated; 91 non-degenerate element; 210
Metric space; 16, 225 Norm; 39
bounded; 19 Normal form; 412
chainable; 72, 231 Normal metric; (see Metric)
chainable from a to 6; 231 Normal projection; 393
complete; 82, 157, 184, 189 Normal space; (see Topological space)
convex; 89 Normed vector space; 39
perfect; 164 Nowhere dense; 85
separable; 420 (see Topological space) Null homotopic; 285, 376
strongly convex; 89, 185

topologically complete; 83, 264 w; 8
topologically totally bounded; 91 Q; 8
totally bounded; 45 Objects; 298
uniformly locally connected; 244 Open disk; 13
well-chained; 69 Open set; 12
without ramifications; 89, 185 in A; 20
Metrizable space; 18, 255-263 relatively; 20
Midpoint; 89 Open shrinkage; 109
Minimal element; 4 Open unicoherent; 246
Minkowski’s inequality; 9 Orbit; 333
Moebius strip; 35, 418 ord(U); 446
Monotonicity; 439, 441, 457 Order of U; 446
Moore space; 259 Ordering:
complete; 265 lexicographic; 37
Moore space conjecture; 260 linear; 3
Morphism; 298 partial; 3
Index 489
[Ordering] finitely related; 462

well-; 4 Presentation mapping; 463
Ordinal: Primitive engulfing theorem; 432
first infinite; 8 Product space; (see Topology)
first uncountable; 8 Product theorem (for dimension); 439,
Overcrossing; 384 444, 457, 458
Projection map; 21, 31, 151, 317
Projective plane; 217, 304, 318, 418
P(X); 1
Properly discontinuous; 332
Paracompact; (see Topological space)
Property S; 244
Partial ordering;(see Ordering)
Pseudo-arc; 241
Partition; 3, 56, 193
Pseudo compact space; (see Compact
Partition of unity; 254
space)
Partition of X between C and D; 441 Pseudo-metric; (see Metric)
Path; 53 Punctured disk; 404
equivalent; 286
match up properly; 287
product of; 286 Q; 1
reverse; 289 Quasi component; 70, 72
Path connected space; (see Connected Quotient map; 193
space) Quotient set; 193
Peano’s space filling curve; 229
Perfect set; 85 (R); 461
Perfectly normal; (see Topological space) R!: 1
Piecewise linear map; 345 Re}
Poincare conjecture; 225 Ri; 130
Point finite; (see Cover) Real compact; (see Compact space)
Pointed space; 290 Rectangle; 127
Pointlike set; 426 interior; 127
Polygonal arc; (see Arc) Rectilinear annulus theorem; 130
Polyhedral category; 345 Rectilinear simple closed curve; 126
Polyhedron; 343, 345 exterior; 129
Porcupine space; (see Topological space) E(R) of a; 129
Preferred point; 290 interior; 129
Presentation; 461 I(R) of a; 129
Presentation equivalence; 464 Refinement:
Presentation of a group; 461 barycentric; 252
finitely generated; 462 of a cover; 162, 250
finitely presented; 462 precise locally finite; 250
490 Index
[Refinement] Separates points a and b; 222

star; 252 Separation; 47
Regular space; (see Topological space) Separation Axioms; (see Topological space,
Reflexive property; 3, 16 To-Ts)
Reidemeister’s trick; 390 Separation order; 222
Relation; 3 Sequence; 29
equivalence; 3, 195 Sequentially compact; (see Compact space)
reflexive; 3 Sheet number; 320
symmetric; 3 Sierpinski universal plane curve; 231
transitive; 3 a-compact; (see Compact space)
Relations; 461 Simple chain; 58, 226
Relatively open; (see Open) Simple closed curve; 125
Retract; 105, 189 polygonal; 148
deformation; 297 rectilinear; 126
Retraction; 105 Simplex:
Riesz-Lennes-Hausdorff separation cri- boundary of; 344
teria; 68 closed; 342
closure; 342
dimension of; 342
S'; 34, 121 Euclidean; 342
S”; 106, 137, 203 ordered; 362
Saturated set; 209 spanned by; 342
Sch6nflies theorem: 136 Simplicial approximation; 355
generalized; 431 Simplicial approximation theorem; 356,
rectilinear 2-dimensional; 128 357, 359
2-dimensional; 144 Simplicial complex; 342
Second category; (see Category) dimension; 342
Second countable; (see Topological space) infinite; 344
Section; 337 Simplicial map; 345
Segment; 94 Simplicial pair; 343, 345, 377
Seifert-van Kampen theorem; 302, 303, Simply connected; (see Connected space)
366 Singleton; 2
Semilocally simply connected (see Con- Skeleton; 203, 343
nected space) Smith set; 86
Semiopen; 45 Solid torus; 167
Separable; (see Topological space) Sphere: :
Separated subsets in X; 99 1-; 34, 121
Separates points from closed subsets; 2-; 34
180 3-; 35
Index 491
[Sphere] an arc; 55
m= 11061137, 2038391, 421 a path; 53
bicollared; 430 Theta curve; 150
collared; 430 Thread; 158
tame; 137, 423 Tietze equivalence; 464
wild; 137, 391, 423 Tietze’s extension theorem; 103
Spine; 414 Tietze theorem; 464
Standard n-ball; (see B”) Tihonov plank; 98, 110
Standard n-sphere; (see S”) Tihonov space; (see Topological space)
Star: Tihonov’s theorem; 154, 172, 277, 279
convex; 294 Topological invariant; 33
of a simplex; 347 Topological pair; 285
with respect to a cover; 252, 451 Topological space; 12
Stone’s metrization theorem; 213, 257 arc connected; (see Connected space)
Strongly continuous; (see Function) Baire; 84
Subbasis; 152 binormal; 375, 397
Subcomplex: collectionwise normal; 259
of a CW-complex; 206 compact; (see Compact space)
of a simplicial complex; 343 completely normal; 98
Subcover; 60 completely regular; 114
irreducible; 73 connected; (see Connected space)
Subdivision; 348 connected im Kleinen; 245

contractible; 70, 124, 294
barycentric; 349
developable; 256
Subnet; 273
disconnected; 47
Subsequence; 65, 77
door; 41
Subspace; 20
first countable; 16, 151, 170, 201
Sum Theorem (for dimension); 439, 442,
fully normal; 258
443, 449, 454, 458
Hausdorff; 16
sup; 4
Hilbert; 184
Super compact; (see Compact space)
homogeneous; 169
Super continuous; (see Function)
invertible; 438
Suspension; 197, 421
Lindel6f; (see Lindeléf space)
Swelling of U; 447
locally arc connected; (see Connected
Syllable; 459
space)
Symmetric property; 3, 16
locally metrizable; 267
locally path connected; (see Connected
Tame sets; 391 space)
Terminal point of: metrizable; (see Metrizable space)
492 Index
[Topological space] finite complement; 15, 41, 42, 65

Moore; (see Moore space) finite product; 21
n-dimensional; (see Dimension) fortissimo; 15
normal; 98, 113, 182, 198, 206, 252 free union; 22
open unicoherent; 246 fully normal; 258
paracompact; 250 half-open interval; 15, 114
path connected; (see Connected space) half open square; (a.k.a Sorgenfrey’s
perfectly normal; 103 half open square topology)
regular; 111, 113 identification; 200
second countable; 16, 157, 170, 255 indiscrete; 15, 48
semi-door; 41 initial; 181
separable; 33, 184, 420 k-; 201, 315
super compact; (see Compact) larger; 12
Loe ld2 VAR long interval; 244
fT SO ed Wa ee 2 long line; 148
Ty2; 118 maximal; 170
To; 1602 AI 7 metric; 17
T33{h12517 Te 207,250; 255 minimal; 170
T31 ; 114 Niemytzki’s Tangent disk; 112
Dg AID, 2uy of pointwise convergence; 176
Te 11D, 00% of uniform convergence; 179
Tihonoy; 114, 180, 189 of uniform convergence on compacta;
topologically complete; see Metric space) 188
totally disconnected; 56, 162 open ray; 15
unicoherent; 234, 380 order; 15
Uryson; 118 point-open; 175
0-dimensional; 46, 115 porcupine; 77
Topological structure; 12 product; 21, 151, 152, 291
Topologically equivalent; 31 quotient; 193, 200
Topologist’s sine curve; 54 relative; 20, 23
Topology; 12 smaller; 12
box; 151 Sorgenfrey’s half open square; 114
bubble; 112 strong; 201
compact-open; 177 sup; 178
countable complement; 15, 37 tangent disk; 112
decomposition; 193 uniform; 118
discrete; 15, 48 usual; 13, 15
disjoint union; 22 Vietoris; 94
final; 200 weak; 181, 182, 204
Index 493
Torus; 196, 334, 418 Uryson’s metrization theorem; 255

2-holed; 217 Uryson space; (see Topological space)
n-holed; 412 usc decomposition; 209
Totally bounded; 45
Totally disconnected; (see Topological
space) Vertex:
Transfinite induction; 7 abstract; 369
Tree: 222. 362 of a simplex; 342
Triangle inequality; 17 of a linear cell; 350
Trianguable space; 344
pemaaeely 344 weakly continuous; (see Function)
Triode; 121, 150, 246 Wedge product; 367
Well-chained; (see Metric space)

Ultrafilter; (see Filter) Well-ordered; (see Ordered, well-)
Uncountable set; 9 Wild sets; 391
Undercrossing; 384 Wild sphere; (see Sphere)

Unicoherent; (see Topological space). Word; 459
Uniform convergence; 179 equivalent; 460
Uniform structure; 118 Word semigroup; 459
Uniformly continuous function; (see Func-
Hon) X©: 305

Union; 2 Ns 158
Unique path lifting theorem; 318
Universal curve; 231
Universal mapping characterization; 308 Z; 1
Upper bound; (see Bound) faa |
Y
Upper semi-continuous; 209 Zermelo’s theorem; 7
Uryson’s lemma; 101 Zorn’s lemma; 6
ye
vv the
ae ae cae 7
v
ISBN 0-914351-07-9

Aspects of Topology - Christenson, Charles O Voxman, William L - 1998

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Aspects of Topology - Christenson, Charles O Voxman, William L - 1998

Uploaded by

Copyright:

Available Formats

| ASPECTS

BCS Associates Moscow, Idaho

All rights reserved.

Library of Congess Cataloging-in-Publication Data:

Aspects of topology/ Charles O. Christenson, William L. Voxman-2nd ed.

Includes bibliographical references and index.

ISBN 0-914251-07-9 (alk. paper). —ISBN 0-914351-08-7 (soft: alk. paper)

ISBN: 0-914351-07-9 Hard Cover ISBN: 0-914351-08-7 Paper Cover

Manufactured in the United States of America.

Printed on acid free recycled paper

Chapter 4: Separation,Properties 0% 4, s+) Guedesele Boe 97

Chapter 10: Paracompactness and Metrizability ........ 249

@hapter .16::2-Manifoldspidarasls bow emeatadein

Bits Vapor «sz trade oe} 16 He mahal) : iaagtaciit the

(rage 1 © NNireneient Veg

Standard set notation is employed. The words “set,” “family,” ”

(0.A.4) Proposition. Suppose {Aq : a € A} is a family of subsets of a set

(i) X\ (U{40 a€ A}) =(\{X\Aa : a € A}, and

(0.A.5) Proposition. Suppose {Ag : a € A} is a family of sets and Aisa

(i) An (tA :ae€ Ay) =(J{AnAa : @€ A}, and

(0.A.6) Definition. A subset A of R! is an interval if and only if whenever

B. CARTESIAN PRODUCTS AND RELATIONS

Xt X Agix & Xq = {(21,22,..-;2n) 1.04 © X; for each ae

The following set equality is occasionally useful.

(0.B.2) Proposition. Suppose A Cc X and B CY. Then

(X x Y)\ (Ax B) = ((X \ A) x Y) U(X x (Y \B)).

Cartesian products are used to define relations on a set X.

(0.B.3) Definition. A subset of X x X is called a relation on X. If

(0.B.4) Definition. If R is a relation on a set X, then

(0.B.6) Definition. Suppose RF is a relation on a set X. Then F is a partial

(0.B.7) Definition. A partial ordering < for a set X is a linear ordering if

(0.B.8) Definition. A linear ordering < for a set X is a well-ordering if

(0.C.5) Definition. Suppose f : X — Y is a function and A C X. The

(0.C.6) Definition. Suppose f : X > Y is a function and B C Y. The

(0.C.7) Proposition. Suppose that {Aq : a € A} is a collection of subsets

(ee (Ass Ba) = Neew pe (Bp);

(iv) £7 (Upew Ba) = Upew f*(Ba),

is commutative if and only if gf = h. Suppose that f: X > Y,9:Y > Z,

is commutative if and only if gf = gp. Similar definitions hold for other

(0.C.13) Definition. Suppose {X, : a € A} is a family of sets. The

[] Xe ={f +A UneaXa : f(a) € Xo for each a € A}.

An element f € [],¢, Xq is usually denoted by {ra} or {Za}aea where it

D. THE AXIOM OF CHOICE AND SOME OF ITS VARIANTS

(0.D.2) Definition. Suppose that A is a subset of a partially ordered set

A proof of the next proposition may be found in most books on set

(0.D.3) Proposition (Zorn’s Lemma). If X is a partially ordered set

(0.D.4) Proposition (Kuratowski’s Lemma). Suppose (X, <) is a par-

lemma to find a maximal element in M. Partially order M by declaring

(0.D.5) Remark. Kuratowski’s lemma is usually applied in the following

E. CARDINAL NUMBERS AND ORDINAL NUMBERS

It is beyond the scope of this book to provide a systematic treatment of

(0.E.1) Definition. Suppose X and Y are sets. We say X and Y are

We postulate the existence of a class C of sets with the property that

An important property of the ordinals is the following.

(0.E.4) Proposition. If A C O is countable and 0 ¢ A, then supA < 2.

where a; and b; are arbitrary real numbers and n € Zt.

(0.F.1) Proposition (Cauchy-Schwarz Inequality). If a,,a2,...,an,

Proof. It suffices to show that

(4) <(4) (Ex)

0< S\(ai+ab;)? = Sa? + 2a) ~aibi +0? 5°

(4) <(E4) (Ze)

Proof. From the Cauchy-Schwarz Inequality we have

So (ai + bi)? = So(aj + 2aib; + 07)