ELEMENTARY TOPOLOGY

SECOND EDITION
 MICHAEL C. GEMIGNANI
State University of New York at Buffalo

ELEMENTARY TOPOLOGY
SECOND EDITION

w ADDISON-WESLEY PUBLISHING COMPANY

Reading, Massachusetts * Menlo Park, California * London •Don Mills, Ontario
This book is in the
ADDISON-WESLEY SERIES IN MATHEMATICS

Consulting Editor
LYNN H. LOOMIS

Printed in the United States of America. Published simultaneously in Canada. Library

of Congress Catalog Card No. 73-168763.
To my w ife , Carol
PREFACE

This book is intended as a first text in topology. If the reader has com­
pleted at least three semesters of a calculus and analytic geometry sequence,
he should have sufficient background to understand this book. The reader
might gain a deeper appreciation of the contents, however, if he has also
had at least one semester of real analysis, or its equivalent.
If anyone asks, “What is topology?”, the most correct answer is,
“Topology nowadays is a fundamental branch of mathematics and like
most fundamental branches of mathematics does not admit of a simple
concise definition.” Topologists are indeed investigating widely different
problems and are using a multitude of techniques. Topology is today one
of the most rapidly expanding areas of mathematical thought.
Historically, however, topology has its roots in geometry and analysis,
that is, the study of real and complex functions. Geometrically, topology
was the study of properties preserved by a certain group of transforma­
tions, the homeomorphisms. Geometry itself can be considered as the
study of properties preserved by certain types of functions; e.g., Euclidean
metric geometry is the study of properties preserved by rigid (that is,
distance-preserving) transformations (known sometimes as congruences).
(Of course, as with topology, it is somewhat unfair to try to define geom­
etry as the study of one particular thing.)
Certain of the notions of topology are also abstractions of concepts
which are classical in the study of real or complex functions. Open sets,
continuity, metric spaces, etc., were a basic part of analysis before being
generalized in topology.
Topology, then, has its roots in at least two areas of mathematics;
but now topology has reached the point where a mathematician engaged
in topological research is not only justified in calling himself a topologist,
but he must specify whether he is a point set topologist, differential
topologist, algebraic topologist, or some other topological specialist.
Since topology is now a study in its own right, we would be justified
in merely introducing topological concepts without giving some idea of
where or how these concepts arose. Since this is an introductory text,
vii
viii Preface

however, we have tried to motivate the concepts introduced so that the

reader can see where these concepts came from originally. Where a con­
cept is primarily geometric, we have tried to treat it geometrically; where
analysis is the inspiration for a concept, an analytic approach is used. It
is hoped that the reader of this book will not only learn the fundamentals
of topology, but will appreciate how abstract topological notions de­
veloped from classical mathematics.
Topology, the child of geometry and analysis, now serves as a powerful
tool not only in these areas, but in almost all areas of mathematical study.
But besides being an instrument for use elsewhere, topology has a beauty
and a content of its own. Topology is valuable in its own right in so far
as any well-developed mathematical study is valuable, or, in fact, any
aesthetically pleasing creation of the human mind is valuable.
In this second edition we have attempted to correct all errors, typo­
graphical and otherwise, found in the first edition. Numerous exercises
have been added as well as a section dealing with paracompactness and
complete regularity. The Appendix on infinite products has been extended
to include the general Tychonoff Theorem; a proof of the Tychonoff
theorem which does not depend on the theory of convergence has also
been added in Chapter 7.

Northampton, Mass. M. G.
September 1971
CONTENTS

Chapter 1 Preliminaries
1.1 Sets and fu n c t io n s .................................................... 1
1.2 Partial and total orderings; equivalence relations . . . . 4
1.3 C a r d in a lit y ............................................................ 9
1.4 G r o u p s ...................................................................... 13

Chapter 2 Metric Spaces

2.1 The notion of a metric s p a c e ............................................ Id
2.2 N e ig h b o r h o o d s ............................................................. 19
2.3 Open s e t s ................................................................. 21
2.4 C losed s e t s ................................................................. 24
2.5 Convergence of s e q u e n c e s ............................................ 26
2.6 C o n t i n u it y ................................................................. 29
2.7 “D ista n ce”between two s e t s ............................................34

Chapter 3 Topologies
3.1 The notion of a t o p o l o g y ................................................ 40
3.2 Bases and s u b b a s e s .................................................... 43
3.3 Open neighborhood s y s t e m s ............................................ 49
3.4 Finer and coarser t o p o l o g i e s ............................................ 52
3.5 D erived s e t s ............................................................. 55
3.6 M ore about topologically derived s e t s ............................... 59

Chapter 4 Derived Topological Spaces.Continuity

4.1 S u b s p a c e s ................................................................. 64
4.2 The topologically derived sets in s u b s p a c e s ....................... 67
4.3 C o n t i n u it y ................................................................. 70
4.4 H o m e o m o r p h is m s .........................................................75
4.5 Identification s p a c e s .................................................... 79
4.6 Product s p a c e s ............................................................. 34

ix
X Contents

Chapter 5 The Separation Axioms

5.1 To- and T i- s p a c e s .................................................... 91
5.2 T V s p a c e s ................................................................. 95
5.3 T 3- and regular spaces .................................................97
5.4 7V and normal s p a c e s ............................................... 102
5.5 N orm ality and the extension of f u n c t i o n s ......................... 106

Chapter 6 Convergence
6.1 The need for a generalized notion of convergence . . 113
6.2 N e t s ......................................................................... 116
6.3 Subsequences and s u b n e t s .......................................... 119
6.4 Convergence of n e t s ................................................... 122
6.5 Limit p o i n t s ............................................................ 127
6.6 Continuity and c o n v e r g e n c e .......................................... 130
6.7 F i l t e r s .................................................................... 133
6.8 Ultranets and u ltr a filte r s ............................................... 136

Chapter 7 Covering Properties

7.1 Open covers and refin em en ts...........................................142
7.2 C ountability p r o p e r t i e s ............................................... 144
7.3 C o m p a c t n e s s ............................................................ 152
7.4 The derived spaces and compactness. The separation
axioms and c o m p a c t n e s s ............................................... 156

Chapter 8 More About Compactness

8.1 Com pactness in Rn ................................................... 163
8.2 Local c o m p a c t n e s s ................................................... 168
8.3 C o m p a c t ific a t io n s ........................................................174
8.4 Sequential and countable c o m p a c t n e s s ..............................179

Chapter 9 Connectedness
9.1 The notion of c o n n e c t e d n e s s ...........................................183
9.2 Further tests for c o n n e c t e d n e s s ...................................... 187
9.3 Connectedness and the derived s p a c e s ..............................192
9.4 Components. Local c o n n e c t e d n e s s .................................. 196
9.5 Connectedness and com pact T V s p a c e s ............................. 202

Chapter 10 Metrizability. Complete Metric Spaces

10.1 M etrizable s p a c e s ........................................................208
10.2 Cauchy s e q u e n c e s ........................................................213
10.3 C om plete metric s p a c e s ...............................................217
10.4 Baire category t h e o r e m ............................................... 223
10.5 Paracompactness. C om plete R e g u l a r i t y ......................... 228
 Contents xi

Chapter 11 Introduction to Homotopy Theory

11.1 H om o topic f u n c t i o n s ...................................................233
11.2 L o o p s .................................................................... 240
11.3 The fundamental g r o u p .............................................. 246
11.4 The fundamental group and continuous functions . . . 253

Appendix on Infinite Products................................... 261

Index of S y m b ols.............................................. 265
I n d e x ......................................................... 267
1
PRELIMINARIES

1.1 SETS AND FUNCTIONS

It is assumed that any reader of this book has already had some experience
with sets; hence most of what is said in this section will be for the sake of
review rather than for the purpose of presenting new material.
We will not deal with sets axiomatically. A set will be taken to be
any well-defined collection of objects; the objects in a set are called ele­
ments, or points, of the set. If x is an element of the set S, we write x G S.
We denote the phrase is not an element of by g.
Sets may be denoted either by explicitly listing their elements inside
of braces (for example, {1, 2, 3} is the set having 1, 2, and 3 as elements)
or by giving the rule by which a typical object of the set is determined
(for example, {x |x is a red schoolhouse} is the set of all red schoolhouses
or, alternatively, the set of all x such that x is a red schoolhouse).
A set S is said to be a subset of a set T if each element of S is an element
of T. We usually denote S is a subset of T by S C T. Two sets S and T
are equal if they contain exactly the same elements; that is, S = T if
S C T and T C S. The phrase is not a subset of is denoted by <Z.
The empty set, that is, the set which contains no elements whatsoever,
is denoted by 0.
If S and T are any two sets, then the complement of S in T is the set
of all elements of T which are not elements of S ; we denote the complement
of S in T by T — S : Similarly, the complement of T in S, denoted by
S — T, is the set of all elements of S which are not elements of T.
The two most basic set operations are union and intersection. If {aS*},
i G I, is any family of sets indexed by some set I, then the union of this
family of sets is {x \x G Si for at least one i G /}. (We will rigorously
define the notion of an index set later in this section; for now the reader
can consider I to be merely a set of labels distinguishing the various mem­
bers of the family of sets.) The union of , i G I, may be denoted by
U / Si, or U {Si |i G 1}. The intersection of this family of sets is {x \x G Si
for every i G I, that is, x is an element of every member of the family of
sets}. The intersection of {Si}, i G I, may be denoted by f li Si, or
1
2 Preliminaries 1.1

0{Si |i E /}. Where only a few sets are involved, say {Si, S 2, S 3), the
intersection and union of these sets may be denoted by S i D S 2 PI S 3 and
S i U S 2 U S 3, respectively.
It is assumed that the reader is moderately familiar with these set
operations, at least so far as any finite family of sets is concerned. We
now prove the DeMorgan formulas for an arbitrary family of sets.
Proposition 1. Suppose {Si}, i e I, is a family of subsets of some set T.
Then
a) U/ (T- Si)= T - D j Sf,
b) (1/ (T- St) = T- U j -Si.
Proof, a) To prove that any two sets are equal, we must show that they
contain the same elements. Suppose x E U/ (T — Sf); then x E T — Si
for at least one i E I. Therefore x £ Si for at least one i E /. Then x is
not in Si for every i e I ; hence x ? FI/ Si. Consequently, x E T — fl j Si.
We have thus proved that every element of Ui (T — Si) is an element of
T — fl/ Si) that is,
U (T — Si) C T - PI Si.
i i

Suppose x E T — fl/ Si. Then there is some i G I for which x & Si

(or else x would be an element of fl/ Si). Therefore x E T — Si for some
i E I. Then x E U/ (T — Si). Consequently, T — fl/ £; C U j {T — Si);
hence
T — fl S i = U (T — Si).
I I
The proof of (b) is left as an exercise.
If S and T are any two sets, then the Cartesian product of S and T is
defined to be the set of all ordered pairs (s, t) such that s E S and t E T.
The Cartesian product of S and T is denoted by S X T.
If S and T are any sets, then a subset R of S X T is said to be a rela­
tion between S and T. A subset of S X S is said to be a relation on S. If R
is a relation between S and T, that is, if R C S X T, then if (s, t) E R,
we may also write sRt, or say that s and t are R-related. Some special
types of relations will be discussed in the next section. Although strictly
speaking a relation is a set, at times ajphrase or symbol defining the rela­
tion will be used in place of the actual set. For example, although is equal
to defines a relation on the collection of subsets of some set, we usually
write simply S = T if S and T are equal subsets, rather than explicitly
refer to any relation.
A function f from a set S into a set T is a relation between S and T
such that each element of S is /^related to one and only one element of T.
1.1 Sets and Functions 3

If (s, t) E /, then we may write t = /(s). Functions are usually defined

by giving a rule which enables us to find f(s) whenever s is given. Again,
rarely is explicit mention made of the fact that a function is a set. Func­
tions are also called maps or mappings.
If / is a function from S into T, then S is called the domain of /, T the
range of /, and {t E T |t = /(s) for some s e S} the image of /. The image
of / may be denoted by /(£).
If / is a function from S into T and W C S, then the restriction of f
to IF, denoted by / | IF, is a function from IF into T defined by / | W(w) =
f(w) for each w E IF.
If / is a function from S into T, we may write /: S —+ T. If f(S) = T ,
then / is said to be onto. If f(s) = /(s') implies s = s' for any s, s' E S,
then / is said to be one-one; that is, / is one-one if each element of T is the
image of at most one element of S.
Suppose /: S —* T. If t E T, then

r \ t ) = { s e S \ m = t}.
If U C T, then
f - \ U ) = {s E <S |/(s) E U}.

By f ~ 1 we mean {(*, s) | (s, t) E/}. Note that /-1 is a relation between

T and S , called the inverse relation of /, and that it is a function from T to
S if and only if / is one-one and onto.
Suppose /: S —► T and g: T —> IF. Then g of is defined by

{(s, w) I s E S, w E IF, such that there is some t E T with t = f(s) and

w = g{t); that is, w = #(/(«)) for some s E $}.

g of is a function from S into IF and is called the composition of g with /.

There are two special types of functions, sequences and indices, which
the reader should already have encountered at least informally. A sequence
u in a set S is any function from the set N of positive integers into S.
If u is a sequence in S , then u(n) is usually denoted by un; the sequence
itself may be denoted by uf {un} , n E N, or {ui, u2, w3, . . .}.
Sometimes the elements of one set are used to label the elements of
another set, this often being a convenient way to express a collection of
objects, or sets. For example, the elements of {1,. 2, 3} are used to label
the elements of {^i, t2, t3}. A one-one and onto function / from some set
/ onto a set S for the purpose of labeling the elements of S is called a
system of indices for S , and I is called the set of indices, or the index set.
The set S is said to be indexed by /, and we may represent this relationship
by writing S as {$*•}, i E /.
1.2 Orderings; Equivalence Relations 5

to is denoted by < , then < defines a relation on R having the properties

that
pi) x < x, for any x G R ;
P2) x < y and y < x implies x — y, for any x) y e R ) and
P3) x < y and y < z implies x < z for any x, y, and z in R.
Any relation on any set S which shares properties PI through P3 is
called a partial ordering on S. If S has a partial ordering defined on it,
then S is said to be a partially ordered set. We may denote a set 8 with
partial ordering < by S, < .
Example 1. Let P(S) be the family of subsets of a set S. Then C defines a
partial ordering on P(S). We verify that C satisfies PI through P3.
PI) If W is any subset of S, then W C W.
P2) If W and T are any two subsets of S such that W C T and T C W,
then W = T.
P3) If W, T} and Z are any sets such that W C T and T C Z, then
each element of W is an element of T. But since T C Z, each element
of T is also an element of Z; therefore each element of W is an
element of Z, that is, W C Z.
Note that in Example 1, it is not true that given any two subsets W
and T of S, either W C T or T C W. It is true, however, that given any
two real numbers s and t, either s < t or t < s. In P(S), C, any two
elements are not necessarily comparable, whereas in R, < , any two ele­
ments are comparable. If S is any set with a partial ordering < , then the
partial ordering of S is said to be a total ordering if given any elements s
and t of S, either s < t or t < s. The partial ordering < on the set of
real numbers is a totaPordering, but C does not define a total ordering on
P(S) in Example 1.
Since the less than or equal to relation on the set of real numbers is
the prototype of a partial ordering, we will generally denote a partial
ordering by < , unless there is a special symbol called for.
Suppose S is a set partially ordered by < , and W C S. Then W can
also be considered to be partially ordered by < through the device of
letting w < w' for any two elements of W if and only if w < wf considering
w and w' as elements of S. We say the ordering < on S induces an order­
ing on W.
Let S, < be any partially ordered set, and suppose W C S. An ele­
ment w of aS is said to be an upper bound for W if w < u for each w E W.
An element v of S is said to be a lower bound for W if v < w for each w E W.
It is not necessarily true that every nonempty subset of a partially
ordered set S, < has an upper or a lower bound.
6 Preliminaries 1.2

Example 2. Let S = {x |0 < x < 1} be partially ordered by < . Then

if W = S, W has no upper bound, nor any lower bound. Suppose that u
is an upper bound for W; then 0 < u < 1. Therefore
0 < u < (u + l )/2 < 1.
Hence (u + l )/2 is an element of W which is greater than u\ thus u could
not be an upper bound for W. Similarly, W has no lower bound.
The partially ordered set R, < of real numbers has neither an upper
nor a lower bound since, given any real number, we can find both a larger
real number and a smaller real number.
Suppose that IT is a subset of a partially ordered set S, < . Then an
element V of S is said to be a least upper bound for W if U is an upper
bound for W and U < u if u is any upper bound of W. An element L
of S is said to be the greatest lower bound of W if L is a lower bound for W
and if v is any lower bound for W, then v < L. The least upper bound and
greatest lower bound for W may be denoted by lub W and gib W, re­
spectively.
It is not always true that any nonempty subset of S, < which has
an upper bound has a least upper bound.
Example 3. Let Q be the set of rational numbers partially ordered by <.
Let W be the set of rational numbers less than V 2. Then 3 is an upper
bound of W; but since y/2 is an irrational number, it can be shown that
W has no least upper bound in Q. Note that W does have a least upper
bound in the full set of real numbers, namely, y/2.
Every nonempty subset of the set of real numbers which has an upper
bound (lower bound) has a least upper bound (greatest lower bound).
Example 4. Let P($), C be the partially ordered set described in
Example 1. Suppose {t/;}, i G I, is any collection of subsets of S. Then
this collection has a least upper bound U/ U{ and a greatest lower
bound D j U{. Note that gib P(S) = <t> and lub P(S) = S.
Let S, < be any partially ordered set, and let W C S. An element
M of W is said to be maximal in W if M ^ w for each w e W — {M} . An
element m of W is said to be minimal in W if w ^ m for each
w G W — {m} . An element M of S is said to be maximal (minimal) if
M is maximal (minimal) in S.
Example 5. Let R be the set of real numbers and W = {1, 2, 4}. Then 4
is maximal in W and 1 is minimal in W. R contains no maximal or minimal
element.
Suppose P(R) is the collection of subsets of R partially ordered by C.
Let W = {{1}, {2}, {4}}. Then each element of W is both maximal and
minimal in W. R is a maximal element and 0 is a minimal element of P(R).
1.2 Orderings; Equivalence Relations 7

If a subset IF of a partially ordered set S, < is totally ordered by the

ordering on W induced by < , then W is said to be a chain in S. That is,
W C S is a chain in S if given any two elements w and w' of W, either
w < w', or w' < w.
Example 6. Let S = {1, 2, 3, 4, 5} and P(S) be the family of subsets of S
partially ordered by C. Then

{{1}, {1,2}, {1,2, 3}, {1,2, 3,4},S}

is an example of a chain in P(S).

One of the fundamental axioms in the theory of sets (and hence in
mathematics) is the axiom of choice. As its name implies, the axiom of
choice is a true axiom, assumed and not proved, although there are dif­
ferent ways in which it can be formulated. The axiom of choice properly
so-called is stated as follows.
The axiom o f choice. Suppose {Si}, i E I, is a family of nonempty
sets. Then there is a function / from I into U / Si such that f(i) E Si
for each i e I.
The axiom of choice essentially says that given any collection of non-
empty sets, it is possible to form a set by choosing one element from each
set in the collection. It all sounds simple enough, but it is hardly simple;
it has stemmed from and led to some of the deepest thinking in the founda­
tions of mathematics. The purpose of this book is not to delve into this
problem, however.
The axiom of choice has several apparently different but actually
equivalent formulations. The particular formulation we will be interested
in later in this book is known as Zorn’ s lemma.
s lemma. Suppose S, < is a partially ordered set with the prop­
Z orn ’
erty that every chain in S has an upper bound. Then S contains a
maximal element.
A partial ordering is an example of a special kind of relation that can
be defined on a set. Another particularly important type of relation is an
equivalence relation. The prototype for an equivalence relation is =, just
as < is the prototype for a partial ordering. Since ambiguity is likely to
result if = is used to denote an arbitrary equivalence relation, E will be
used instead. A relation E on a set S is said to be an equivalence relation
on S if E satisfies the following properties:
El) sEs for any s e S.
El) If s and s' are any elements of S such that sEs', then s'Es.
E3) If s, s', and s" are any elements of S such that sEs' and s'Es",
then sEs".
8 Preliminaries 1.2

Compare E l through E3 with the properties of = . Note that the

only difference between a partial ordering on S and an equivalence relation
on S is that property P 2 has been replaced by property E 2.
Example 7. Let T be the set of all plane triangles. Then is similar to
defines an equivalence relation on T. An equivalence relation on T is
also defined by is congruent to; still another equivalence relation on T is
defined by has the same area as.
The most important property of an equivalence relation is given in
the following proposition.
Proposition 2. Let S be any set. A partition (P of S is any collection of
nonempty subsets of S such that each element of S is contained in one
and only one member of (P. Suppose E is an equivalence relation
on S. For each s E S y set s = {t E S \sEt}. Then the collection of
s for all s E S is a partition of S, called the partition induced by E.
Moreover, given any partition (P of S, there is an equivalence relation
E on S such that (P is the partition induced by E.
Proof. Suppose that E is an equivalence relation on a set S. We must
show that {S},sG S, is a partition of S. Since sEs for each s e S by El,
then 5 E s', hence each element of S is contained in at least one member of
{s}, s E S. We now must show that each element of S is contained in
only one member. Suppose that s E s and s E t . Choose any s' E s. Then
sEs '; also tEs since s E t . By E3, tEs and sEs' implies tEs'\ hence s' E t.
Therefore s C t. A similar argument, however, shows that t C s, and
hence s = t. Thus s is the only member of {s}, s E S, which contains s
for each s E S. Therefore { s} ,sE S, is a partition of S.
Suppose that (P is a partition of S. Define a relation E on S by letting
sEs' if and only if s and s' are contained in the same member of (P for any
s and s' in S. It is left as an exercise to prove that E is an equivalence
relation on S. By definition of E, (P is clearly the partition induced by E.
If E is an equivalence relation on a set S, then if sFJs', s and s' are
said to be E-equivalent, or simply, equivalent. The set of elements of S
which are equivalent to an element s of S is said to be the E-equivalence
class of s, or simply the equivalence class of s. It is the collection of In­
equivalence classes which forms the partition of S induced by E.
Example 8. Let / be a function from a set S into a set T. Define sEs' if
/(s ) = /(s ') for any s and s' in S. Then E is an equivalence relation on S.
Denote the set of ^-equivalence classes by S/E; if s E S, denote the
equivalence class of s by s. We may associate with/a function/: S/E —> T,
defined by /(s ) = /(s) for any s E S/E. Since s' = s if and only if /(s) =
/(s '), / is well defined. Note that whereas / may not have been one-one,
/ is one-one.
1.3 Cardinality 9

EXERCISES

1. Prove that E in the second part of the proof of Proposition 2 is an equivalence

relation on S.
2. The following refer to Example 8.
a) Verify that E is an equivalence relation on S.
b) Prove that f is well defined, that is, single valued, and one-one.
3. Suppose that N is the set of positive integers. Let n |m denote n divides m,
that is, m — nk for some positive integer k. Prove that | defines a partial
ordering on N. Does N contain a maximal element (with respect to this
partial ordering)? a minimal element?
4. Let N, | be the partially ordered set described in Exercise 3.
a) Prove that any two-element subset of N has a greatest lower bound and
a least upper bound.
b) Which of the following subsets of N are chains in V? Find a maximal
and a minimal element, an upper and a lower bound, and a least upper
bound for each subset:
i) {1,2, 4, 6, 8}, ii) {1,2, 3, 4, 5},
iii) {3, 6, 9, 12, 15, 18}, iv) {4, 8, 16, 32, 64, 128}.
5. A subset W of the set Z of integers is said to be closed under addition if given
any elements w and w' of W, w + w' E V/. Prove that there is a maximal
subset of Z which is closed under addition and does not contain 9. Do this
using Zorn’ s lemma. 1

1.3 CARDINALITY
Two sets S and T are said to have the same number of elements, or to have
the same cardinality, if there is a one-one function / from S onto T. That
is, S and T have the same cardinality if the elements of S can be put into
one-one correspondence with the elements of T.
A set S is said to be finite if S has the same cardinality as <t>, or if there
is a positive integer n such that S has the same cardinality as {1, 2,. . . , n}.
Otherwise, S is said to be infinite. Furthermore, a set S is said to be count­
able if S has the same cardinality as a subset of N , the set of positive
integers. Otherwise, S is said to be uncountable. Thus any finite set is
certainly countable.
Proposition 3
a) Any subset of a finite set S is finite.
b) Any subset of any countable set $ is countable.
Proof
a) Since S is finite, either S = 0 or there is a positive integer n such
that S has the same cardinality as {1, 2, . . . , n}. If S = </>, then the
only subset of S is <t>, which is finite. Suppose that S ^ 4>. Then there
10 Preliminaries 1.3

is a one-one function / from S onto {1, 2, ... ,n} for an appropriate n.

Suppose W C S. If W = <f>, then W is finite. If IF 9^ <t>, let i\, i2, .. . ,im
be the elements of {1, 2, . . . , n} in the image of W. Then defining
g: W —> {1, 2, . . . , ra} by g{w) = j, where f{w) = ij, for each w e W,
we see that W is finite.
The proof of (b) is left as an exercise.

Proposition 4. Let {A n} , n e N ybe a countable collection of countable

sets. Then \JN An is also countable (N represents the set of positive
integers).
Proof. We may enumerate the elements of each of the A n in an array
as shown.

The element anm is the rath element of An. If we run out of elements in
any set, i.e., if any of these sets are finite, we just put down x*8 in the spot
where an element should go.
We now must find a one-one function/from Uv An onto some subset
of the set N of positive integers. Set/(au) = l,/(ai2) = 2, and/(a2i) = 3.
In general, follow the path indicated in the diagram and correspond the
fcth element reached with k. Eventually every element of Ujv will be
reached; hence Un An can be put in one-one correspondence with a subset
of N , and is therefore countable.

Corollary 1. If A and B are countable sets, then A X B is countable.

Proof. Let A = {ai, a2, a3, . . .} and B = {&i, b2, b3, . . .}. Set

An — {(an, b) |b G B} for each n G N.

Then each An has the same cardinality as B , and hence is countable.

Therefore, by Proposition 4, Uat An is a countable set. But then A X B =
Uat An, as the union of a countable number of countable sets, is countable.
1.3 Cardinality 11

Corollary 2. The set Q of rational numbers is countable.

Proof. If q is any positive rational number, then we may consider q as the
quotient of two positive integers ra/n, where the fraction m/n is in lowest
terms. Associate q with the ordered pair (ra, n). We then see that the
positive rational numbers may be associated with a subset of N X N,
where N is the set of positive integers. But N X N is countable by Corol­
lary 1. Therefore, by Proposition 3, the set of positive rational numbers
is countable. The set of negative rational numbers, however, has the same
cardinality as the set of positive rational numbers (corresponding q with
—g, where q is any positive rational number), and hence the set of negative
rational numbers is' countable. But Q is the union of {0}, the set of positive
rational numbers, and the set of negative rational numbers, all three of
which are countable sets; therefore, by Proposition 4, Q is countable.
Corollary 3. The set Z of integers is countable.
Proof. Z is a subset of Q, the set of rationals, and hence is countable by
Proposition 3.
Although we now have a goodly number of sets we know to be count­
able, we have not yet shown that any set is uncountable. The following
example shows that the set of real numbers is uncountable.
Example 9. The set S of unending decimals between 0 and 1 which contain
only 0 or 1 as digits is uncountable. For proof, suppose that S is countable.
Then we can find a one-one correspondence between S and the set N of
positive integers; hence we can make a table like the following, in which
the first column gives a positive integer and the second column the element
of S associated with it by a suitable function /.
n
1 0.011010- ••
2 0.1110111 •••
3 0.101101 •••
4 0.0000000111

We now form an element .X1X& 3X4 * * •of £ as follows: If the first

digit of /(l) is 0, let x\ = 1, and if the first digit of /(l) is 1, let x\ = 0.
Similarly, if the second digit of /(2) is 0, let x 2 = 1, and if the second digit
of /(2) is 1, let x 2 = 0. In general, if the nth digit of/(n) is 0, let xn, the
nth digit of our new element of S, be I, and if the nth digit of /(n) is 1,
let xn = 0. Then .xix2xs •••could not be /(n) for any positive integer n,
since .XxX2x 3 •••differs from each/(n) at least in the nth digit because of
the way it has been constructed. Hence / could not be onto, and S is there­
fore uncountable.
12 Preliminaries 1.3

But S is a subset of R, the set of real numbers. If R were countable,

then, by Proposition 3, S would also be countable; therefore R is un­
countable.
We have only gone as far in our discussion of cardinality as it was
felt necessary to go in order that the reader understand the contents of
this book. The discussion has been somewhat informal and much has
been left unsaid. For a more complete discussion of cardinality and
cardinal numbers, the following texts are recommended.
1. J. L. K e l l e y , General Topology, Van Nostrand, New York, 1955. The
appendix to Kelley gives a concise axiomatic treatment of set theory
and ordinal and cardinal numbers. It may be a bit too concise for the
reader.
2. G. B i r k h o f f and S. M a c L a n e , A Survey of Modern Algebra, Macmillan,
New York, 1953. Chapter X II gives a nice introduction to cardinal
numbers and their arithmetic.
3. E, K a m k f , Theory oj Sets} Dover, New York, 1950. This book is one
of the classics in set theory and is a must in the library of any serious
mathematician.

EXERCISES

1. Let ft be the class of all sets. (Technically, the collection of all sets is not a
set, so in such cases we use some word like class.) Prove that has the same
cardinality as defines an equivalence relation on ft. An equivalence class is
called a cardinal number.
2. Prove (b) of Proposition 3.
3. Prove that no finite set S has the same cardinality as one of its proper sub­
sets W. (A subset W of S is said to be proper if IF ^ S.) Does this remain
true if S is infinite? Prove that any two infinite subsets of V, the set of
positive integers, have the same cardinality.
4. The cardinality of a set S is said to be strictly greater than the cardinality of
a set T if there is a subset W of S which has the same cardinality as T, but
no subset of T which has the same cardinality as S.
a) Prove that the cardinality of any uncountable set is strictly greater than
the cardinality of any countable set.
b) Let S be any set and let P(S) denote the collection of subsets of S. Prove
that the cardinality of P(S) is strictly greater than the cardinality of S.
c) Show that given any set whatsoever, there is a set of strictly greater
cardinality.
5. Prove: The set R of real numbers has the same cardinality as a subset of the
set P(N) of all subsets of the positive integers. Prove that P(N) has the
same cardinality as a subset of R. It can then be shown that R and P(N)
have the same cardinality.
14 Preliminaries 1.4

Example 9. Let R be the set of real numbers. Then R, + is a group with

identity 0. R is not a group with respect to multiplication, since 0 has no
inverse with respect to multiplication; but R — {0} is a group with
multiplication as the operation and 1 as the identity.
Example 10. There are groups which contain only a finite number of
elements. The table below gives the “
multiplication table ”for a group
of only four elements:

# Si s2 S3 s4

Si Si s2 S3 s4

S2 S2 Si S4 S3

S3 S3 s4 Si s2

S4 s4 S3 s2 Si

It would actually take a great deal of computation to verify directly that

this is indeed the operation table for a group; therefore, if the reader
does not immediately recognize this group, he will more or less have to
accept its being a group on faith. Note that the identity of this group is
Si and that each element of the group is its own inverse.
Suppose that S , # and T, $ are groups. There may be many functions
from S into T, but perhaps only a few of these are related in any way to
the group structures of S and T. When studying groups, however, we
wish to consider functions which somehow respect the operations of the
groups; such functions are called homomorphisms. More formally, a
function f : S —> T is called a homomorphism if

f ( s i # s 2) = f ( s 1) $ f ( s 2)

for any elements s x and s 2 of S. If / is a one-one and onto function as well

as being a homomorphism, then / is said to be an isomorphism, and the
groups S, H and T, $ are said to be isomorphic. Isomorphic groups have
essentially the same group properties.
If S, H and T, $ are any groups, then we can define an operation & on
S X T as follows: If (s, t) and (s', t') are any elements of S X T, define

(s, t) & (s', t') = (s # s', t $ t').

We call the group thus formed the direct sum of S , # and T, $ (see Exer­
cise 5). We denote the direct sum of S, §and T, $ by S © T.
1.4 Groups 15

Again, we have only set forth as much about groups as will be required
to understand the text. For a more complete treatment of the theory
of groups, the following books are suggested.

1. i r k h o f f and S. M a c L a n e , A Survey of Modern Algebra, Macmillan,

G. B
New York, 1953. Chapter VI is a good introduction to groups. Chap­
ters I and II can also be used as a reference on the structure of the
real numbers.
2. W. L e d e r m a n n , Introduction to the Theory of Finite Groups, Oliver and
Boyd, London, 1961. This is another excellent book that should be in
anyone’ s mathematics library.

EXERCISES

1. Let S be any set. Prove that the set of one-one functions from S onto £ is
a group with composition as the group operation. Suppose that / and g are
any one-one functions from S onto S. Is it necessarily true that/°0 = g°p.
2. Suppose / to be a homomorphism from the group S, # into the group T, $.
Prove that f(S), $ is a group contained in the group T, $. (If S, # is any group
and lb is a subset of S such that W, # is also a group, then W, # is said to be a
subgroup of S, #.) Let k' be the identity of T, $ with respect to $. Prove that
f~ l (k') is a subgroup of S, #.
3. Let Z, + be the additive group of integers. Prove that any subgroup of Z , +
consists of all the multiples of some fixed integer n; that is, if IP is a subgroup
of Z, +, then there is an integer n such that W = {nz | z G Z}.
4. Suppose £, # is a group and {T*}, i E /, is a family of subgroups of S. Prove
that H i Ti is a subgroup of S.
5. Prove that if S, # and T, $ are groups, then S @ T, the direct sum of S and T,
is also a group. Prove that S, # and T, $ are each isomorphic to some subgroup
S © T. Find a homomorphism / from S @ T onto S, jf.
6. Suppose a set S with operation # has an identity k with respect to #. Prove
that k is the only identity in S with respect to #. Prove that if # is associative
and s G S has an inverse t, then t is the only inverse of s in S.
 2
METRIC SPACES

2.1 THE NOTION OF A METRIC SPACE

Most of the important notions in point set topology are generalizations
of concepts which were first studied in the context of metric spaces. By
way of motivation, and also because metric spaces are still extremely
important in their own right, we would do well to consider the fundamental
properties of metric spaces.
A metric space is a set in which we have a measure of the closeness
or proximity of two elements of the set, that is, we have a distance defined
on the set. A metric is nothing more than the ordinary notion of distance.
More precisely, we make the following definition.
Definition 1. Let X be any set. A function D from X X X into R ,
the set of real numbers, is said to be a metric on X if
i) D(x, y) > 0, for all x, y e X ;
ii) D{x, y) = D(y, x), for all x, y e X;
iii) D(x, y) = 0 if and only if x = y; and
iv) D(x, y) + D(y, z) > D{x, z), for all x,y,z e X.
A set X with metric D is said to be a metric space, and may be denoted
by X, D.
Note that the metric D has properties that we intuitively associate
with distance; in fact, as has already been remarked, a metric is merely
a formalized expression of distance.
Example 1. Let R be the set of real numbers. One possible metric for R
is the absolute value metric; that is, define D(x, y) = \x — y|.
Example 2. The metric in Example 1 should already have been familiar
to the reader (although he may not have called it a metric). Another
metric usually encountered in more elementary courses is the “ Pytha­
gorean”metric on the coordinate plane R 2. If (xi, y{) and (x2, y%) are
any two points of R 2, then we can define a metric D on R 2 by setting

((xi, 2/1), (x2, 2/2)) = V (xiz 2) 2 + (2/1 — 2/2)2-

16
2.1 The Notion of a Metric Space 17

Example 3. The metric defined on ft2 in Example 2 is not the only metric
which can be defined for R 2.Again letting (xu and (x2, be any
two points of R2, we can define metrics and on R 2 asfollow

y),
i (x2, y 2)) = — + I2/1 ~
if (xi, iV) = (x2>2/2);
D 2((xx,y x), (x2,y 2))
if 0c 1 , y1 ) 5^ 2/2);
D 3((x 1, y 1
), (x2, y 2)) = max (|xi — x 2\, \yx —y 2\).

Example 4. If X si any metric space with metric D, and F is a

of X, then F can also be considered to be a metric space using the same
metric as X. More precisely, Y with metric D | F(i.e. D defined for pairs
of elements of F) is a metric space. F, | F is said to be a subspace of
the metric space X, D.
Example 5. This example is given to illustrate that a metric can be defined
on a set which is neither Rn for some n nor a subset of Rn. Let X be the
set of all functions from the closed interval [0, 1] into itself. If / and g are
any such functions, define

D(f, g) = least upper bound (|/(x) — g(x)\ | G [0, 1]}.

Since any subset of the real numbers which has an upper bound has a
least upper bound and

0 < |/Or) — g(x)| < 1 for all f , g e X and

then D(f, g) is defined for all /, ge

metric for X by showing that D satisfies each of the properties required
for a metric in Definition 1:
i) D{f, g) > 0 for all /, gG X .S

{|/0*0 — g(x) | |z e [0, 1]}

is greater than or equal to 0, the least upper bound of this set, D(f, g),
is greater than or equal to 0.
ii) D(f, g) = D(g, f) for all /, ge
that ]f(x) — g(x)\ = |</(x) — f(x)\.
iii) D(f,g) = 0 if and only if / = g.If / = g, the
x G [0, 1], and hence {|/(x) — g(x)\ e [0, 1]} = {0}. Therefore it
follows that D(f, g) = 0. On the other hand, if D(f, g) = 0, then

lub {|/(*) - )||x

g(x G [0, 1]}
18 Metric Spaces 2.1

S in c e \f(x) — g(x)\ is a lw a y s g r e a t e r t h a n o r e q u a l t o 0, it f o llo w s th a t

(1/0*0 — 00*01 | x G [0, 1]} = {0};

h e n c e |f(x) — g(x) | = 0 f o r a ll x e [0, 1]. T h e r e f o r e f(x) = g(x) f o r

a ll x G [0, 1]; th a t is, f = g.

iv) D (f, g) + D(g, h ) > Z>(/, h) f o r a ll/ , g, h e X. T h is in e q u a lity f o llo w s

fr o m

1/0*0 — < Ifix ) — g(x) | + |g(x) - h(x) | f o r a ll x G [0, 1].

T h e d e t a ils a re le f t a s a n ex ercise .

EXERCISES

1. Prove that D\, £>2, and Ds as defined in Example 3 are really metrics for R2.
2. Let (xi, y 1) and (X2,2/2) he any points of R2. Which of the following do not
define metrics for R2? Explain your answer in each case.
a) D((xi, yi),(x2,2/2)) = min(|*i —— |).
b) D((xi,yi), (x2,y2)) = {xx — x2)2 + (yi —
c) D((xi,yi),(x2,y2)) = D
\{xi,yi),(x2,y2)) — Dz({xi,y\), (x2,y2)),
where D\ and D3 are as defined in Example 3.
. d) D((x 1, yi),(x2, yi)) = |xi| + \x2\+ + \y2\.
3. Suppose X, D is a metric space. We may define a metric D' for X X X as
follows: If (x, y) and (a/, y') are any elements of X X X, set

)x(',2/0) = D(x,x')+D(y,y').
,y
\{x
D
Prove that D' is really a metric for X X X . Define a metric for X n, that is,
X X •••X (n times) X X.
4. Suppose X, D is a metric space. If x and y are any elements of X, which of
the following define metrics on X?
a) Di(x, y) = kD(x, y), where k is any positive real number.
b) Z>2(x, y) = kD(x, y), where k is any real number.
c) Ds(x, y) = Dn(x, y), where n is any positive integer.
d) D^x, y) = Dr(x, y), where 0 < r < 1.
5. Supply the details for the proof of Example 5(iv).
6. For any two distinct points P 1 and P2 of R2, set

M(Ph Pi) = {F e B2 I P) = d(P2,

where d is a metric on R2. Describe geometrically M(P 1, P2) for each of the
metrics on R2 introduced in Example 3 as well as for the Pythagorean metric.
2.2 Neighborhoods 19

Figure 2.2

2.2 NEIGHBORHOODS
Let X, D be a metric space. If x is any point of X , then we may want
to consider all the points of X within a certain distance of x, that is, the
set of points of X which are within some degree of nearness to x.

Definition 2. If X } D is a metric space, x e X, and p is any positive

real number, then the D-p-neighborhood of x is defined to be the set
of all points y of X such that D(x,y) < p; that is, the D-p-neigh­
borhood of x is defined to be

{ y ^ X I D(x,y) < p}.

Where there is no danger of ambiguity, the D-p-neighborhood of x will

be called the p-neighborhood of x and will be denoted by N(x, p).
Example 6. Let R 2 be the coordinate plane. Figures 2.1 through 2.4
illustrate the 1-neighborhoods of (0, 0) with respect to the metrics Z),

(0 , 0 )

Figure 2.3 Figure 2.4

20 Metric Spaces 2.2

y=y'

D\y D 2, and Z)3 of Examples 2 and 3. The reader should be sure to verify
these figures.
Note that the Dx-l-neighborhood of (0, 0) is a subset of the D-l-
neighborhood of (0, 0). Since

•D((*i, 2/i), (*2,2/2)) < # i ( ( * i , 2/ i), (*2,2/2))

for any two points (xx, 2/x) and (x2, 2/2) of (Exercise 2), the Di-p-
neighborhood of any point (x', y') of R 2 is a subset of the D-p-neighbor-
hood for any positive real number p. However, a simple calculation
shows that the D-p/v^-neighborhood of (x', y') is a subset of the
Z>x-p neighborhood of (x', y') (Fig. 2.5). Note, however, that if p < 1,
there is no positive number q such that either the D-g-neighborhood of
(x'f y') or the Z>x-g-neighborhood of (x', y') is a subset of the D 2-p-neigh-
borhood of (x', y') which consists of (x', y') alone.
Example 7. Let X, D be the metric space described in Example 5. Suppose
f El X and p > 0. We may draw a p-collar about the graph of / as shown
in Fig. 2.6. Then the D-p-neighborhood of / will consist of all functions
from [0, 1] into [0, 1] whose graphs lie within the p-collar of /.

EXERCISES

1. Confirm Figs. 2.1 through 2.4.

2. Prove that D((xi, y\), (x2)y2)) < Di((x\, t/i), (x2)y2)) for any two points
(xi,y 1) and (x2,2/2) in R2 as claimed in Example 6. Also in Example 6,
carry out the computation which shows that the D-p/y^-neighborhood
of (x, y) is a subset of the Dx-p-neighborhood of (x, y) for any (x, y) E R2.
22 Metric Spaces 2.3

Figure 2.8

p — D(x, w). Then if z E N(w, q), we have D(w, z) < p — D(x, w).
Therefore
£>(#, z) < D(x, w) + D(w, z) < D(x, to) + p — D(x, w) = p.
We thus have that zE N ( z,p ) ; hence N(w, q) C N(x, p). Therefore
N(z, p) is an open set.
Proposition 2. Let X, D be a metric space. Then
a) X and <j) are both (D-) open sets,
b) the intersection of any two open sets is again an open set, and
c) the union of any family of open sets is again an open set.

Proof
a) If x e X and p > 0, then N(x, p) C X. Therefore X is an open
set. Since contains no points whatsoever, it is true that for each
x El <j> (there is no such x) and any p > 0, N(x, p) C <t>; hence </> is
also an open set.
b) Suppose U and V are open subsets of X and x E U n V. Since
U is open, there is a positive number p i such that N(x, p i) C U.
Since V is open, there is a positive number p 2 such that
P2) c V. Set p = min (p1} p 2). Then N(x, p) C U fl V;
therefore U n F is open.
c) Let {Ui}, i E /, be any family of open subsets of X and x E U/ E 7».
Then x E Ui for some i. Since Ui is open, there is a positive
number p such that N(x, p) C Ui. But then N(x,p) C U j Ui;
hence Ur Ui is open.
Proposition 3. Let X be a set with metric D. A subset U of X is open
if and only if U is the union of a family of p-neighborhoods.
2.3 Open Sets 23

Proof. Assume U to be the union of a family of ^-neighborhoods. Since

each p-neighborhood is open by Proposition 1, U is the union of a family
of open sets. Therefore U is open by Proposition 2(c).
Suppose that U is an open subset of X. Then for each x G U, we can
find a t least one N(z, p) such that N(x, p) C U. Since 2V(x, p) C U for
e a ch x & U,\Ju N(x, p) C U. On the other hand, each x G U is an element
o f at least N(x, p ) ; hence U C U u N(x,p). Therefore U = U u N(x,p).

If X is any set, then X and </> have been shown to be D -open sets for
any metric D which can be defined on X. If and D 2 are any two
metrics for X, it is not necessarily true that each Di-open set is D 2-open,
or that each D 2~open set is Di-open.

Example 9. Let D and D 2 be the metrics defined on R 2 as in Examples

2 and 3. For any (x, y) G R 2, the D 2-l-neighborhood of (x, y) is precisely
{(x, y)}, since (x, y) is the only point which is less than D 2-distance 1
from itself. Therefore {(x, y)} is D 2-open, since it is a Z)2-neighborhood.
But {(x, y)} is not D-open, since for any p > 0, the D-p-neighborhood of
(x, y) contains infinitely many points besides (x, y). (See Exercise 4 also.)

EXERCISES
1. L et X, D b e a m etric space. S u p p o se th at x an d y are tw o d istin ct p oin ts of X.
P rov e th at th ere are op e n sets U and V in X su ch th at x G U, y G V and
U n V = <t). [Hint: L et U = N(x, £Z>(x, y)).]
2. D eterm in e w hich o f the fo llow in g su b se ts o f the plan e R 2 w ith the P y th a ­
gorean m etric are open.
a) {(x, )x| <0}
y
b) {(x, y )|x 4- y >5}
c) {(x, )|
y x 2 + y 2< 1, or (x, y) = (1, 0)}
d) {(x,y )| x > 2 an d y <3}
3. Let p b e an y p o sitiv e num ber. P rov e that an y D -p-n eigh borh ood o f R 2 is
/>i-, Z>2-, and Z>3-open.
4. P rov e that a n y su b set o f R 2 w hich is D -open is D i- o p e n and, conversely, that
an y su b se t o f th e plan e w h ich is D i- o p e n is D-open.
5. M ak e a p p rop ria te sk etch es for the p ro o fs o f (b) and (c) in P ro p o sitio n 2.
6. Let D i and D 2 b e p ossib le m etrics for a set X. D\ an d D 2 are said t o be
equivalent if e v e ry D i- o p e n set is D 2-open and ev e ry D 2-open set is D i-open .
P rov e th at D 1 an d D 2 are eq u iv a len t if and on ly if, giv en an y x G X and
an y p > 0, there are p o sitiv e n u m bers p i an d P 2 su ch that the D i- p ^ n e ig h ­
b o rh o o d o f x is a su b se t o f the D 2-p-n eigh borh ood o f x, and the D 2-P2-
n e igh b o rh ood o f x is a su b se t of th e D i-p -n eigh borh ood of x.
7. L et L b e an y stra igh t lin e in R 2. P ro v e th a t R 2 — L is op en w ith resp ect
to all m etrics in tro d u ce d on R 2 th u s far in th e text. T r y to find a m etric on
R 2 fo r w h ich R 2 — L is n ot n ecessa rily open.
24 Metric Spaces 2.4

2.4 CLOSED SETS

Definition 4. Let X, D be a metric space. A subset F of X is said to
be closed if F is the complement in X of an open set; that is, F =
X — [/, where U is open.

min (|*|, |l-x|) Figure 2.9

Example 10. The closed interval [0, 1] is a closed subset of the real line
R with the absolute value metric. For suppose x e R — [0, 1].' Set
p = min (|1 — x\y |x|). Then N(x, p) C R — [0, 1]. Therefore R — [0, 1]
is open, and hence [0, 1] is closed (Fig. 2.9).
Example 11. If X is a set with metric D, if x e X, and if p is any positive
number, then the closed p-neighborhood of x, denoted by C1N (x, p), is
defined to be the set of all y G X such that D(x, y) < p, that is,

C1N (x,p) = {y eX | D(x, y) < p}.

It is left as an exercise to show that C1N (x, p) is a closed subset of X.

Note in Example 10 that [0, 1] = C1N (J, J); therefore the fact that
[0, 1] is closed follows from the more general considerations of this example.
The following proposition gives the basic properties of closed sets in
a metric space.

Proposition 4. Let X, D be a metric space. Then

a) X and 0 are closed subsets of X,
b) the union of any two closed sets is closed, and
c) the intersection of any family of closed sets is again a closed set.

Proof
a) X = X — 0. Since 0 is an open set, X is the complement of an
open set and hence is closed. Now, 0 = X — X, hence 0 is also
the complement of an open set, and is therefore closed. (Note that
X and 0 are both open and closed. It is quite possible for a set to
be both open and closed; the complement of such a set would also
have the property of being both open and closed.)
b) Let F and F' be any closed subsets of X. Then F = X — U and
F' = X — U', where U and XJf are open subsets of X. Then

F UF' = (X - U) U(X - IT) = X- (U n U').

But U n Ur is open by Proposition 2(c); therefore X — (U fl U') =
F U F' is closed.
2.4 Closed Sets 25

c) Let {Fi}, i E /, be any family of closed subsets of X. Then

Fi = X — Ui, where Ui is an open subset of X for each i E /.
It follows that

ni F i i= n (x - j7,o= x - ui c/*
Since Uj is open, (1/ F* is closed.

Proposition 5. If X is a set w ith m etric D a n d x E X, th en {x} is a

c lo se d su b se t o f X.

Proof. Since {x} = X — (X — {z}), if we show that X — {x} is open,

we will have shown that {x} is closed. Suppose y E X — {x}. Set p =
D(x, y). Then N(y,p) C X — {x} ; hence X — {x} is open.

EXERCISES

1. Show that the union of an arbitrary family of closed subsets of a metric

space need not be closed. Find an example to show that the intersection of
any family of open sets need not be open. [Hint: Use Proposition 5.]
2. Let X, D be a metric space and F, D | F be a metric subspace of X (see
Example 4). Prove each of the following.
a) A subset W of F is open in F (that is, is D | F-open) if and only if W =
Y f| U} where U is an open subset of X.
b) A subset C of F is closed in F if and only if C = Y fi F, where F is a
closed subset of X.
c) If F is an open subset of X, then a subset of F is open in F if and only if
it is open (in X).
d) If F is a closed subset of X , then a subset of F is closed in F if and only
if it is closed (in X).
e) A subset of F may be open or closed in F without being open or closed in X.
3. Prove that a subset F of a metric space X, D is closed if and only if X — F
is open.
4. Decide which of the following subsets of R2 with the Pythagorean metric
are closed.
a) {(x, y)|x = 0, y <5}
b) {(x, y) |x = 2 or x = 3, j/ is an integer}
c) {(x, y)jx2+ y2< 1, or (x, y)= (1, 0)}
d) {(x, y)j y = x2}
5. Suppose that {Fi}, i E I, is a family of closed subsets of a metric space X, D
with the property that given any x E X, there is p > 0 such that N (x, p)
intersects finitely many of the Fi. Prove that (J j Fi is closed. Try to find
and prove an analogous statement for a family of open subsets of X.
26 Metric Spaces 2.5

6. Prove that a straight line in R2 is closed with respect to all of the metrics
on R2 introduced thus far. Prove a circle in R2 (circle in the usual geometric
sense) is closed with respect to all of these metrics. List some other standard
geometric objects which are always closed.

2.5 CONVERGENCE OF SEQUENCES

The reader should already have been introduced in previous courses to
the notions of convergence of sequences, limits, and continuity, at least
as far as the real numbers with the absolute value metric is concerned.
We now extend these ideas to general metric spaces.
D efinition 5. Let X, D be a metric space and S = {sn} , n G N, be a
sequence in X. (The capital N will be used almost exclusively in this
text to denote the set of positive integers.) Then S is said to converge
to a point y of X if given any positive number p , there is a positive
integer M such that if n > M, then sn G N(y, p) (Fig. 2.10). If S
converges to y, then we may write sn —► y ; y is said to be the limit of S.
Definition 5 could be restated as follows: sn —> y if all but a finite
number of the sn are in N(y,p) for any positive number p. Or again,
sn —* y if all but finitely many of the sn are closer to y than any given
distance.
Exam ple 12. Consider the sequence defined by sn = (1, 1/n) in the
coordinate plane (Fig. 2.11). If any of the metrics Z), D i, or D 3 are used,
then this sequence converges to (1, 0). For in these cases,

D(sn, (1,0)) = Diitn, (1,0)) = D 3(sn, (1,0)) = 1/n

for any n e N. Given any positive number p , if we let M be any integer

greater than 1/p, if n > M, then D(sn, (1, 0)) < p. On.the other hand,
this sequence does not converge to (1, 0) with respect to the metric Z>2.

Figure 2.10 Figure 2.11

For if p < 1, then the D 2-p-neighborhood of (1,0) contains only (1,0),
and hence excludes all of the points of the sequence.
Example 13. Let X , D be the metric space of functions described in
Example 5. Let $ be the sequence in X defined by sn (x) = xn. If we were to
plot the graphs of sn for successively greater n (Fig. 2.12), it would appear
that the sequence S converges to the function f G X defined by

if x 9^ 1,
m =
if x = 1.

Such is not the case, however. For if we draw a p-collar about the graph
of / for p = ^ (Fig. 2.13), we see that no sn has its graph wholly within
the collar; therefore S cannot converge to /. Note, however, that S con­
verges “ pointwise”to /; that is, for each x G [0, 1], sn(x) —>/(x), where
(sn(x)}, n E N, is considered as a sequence in the plane with the
Pythagorean metric.

Proposition 6. If S = {sn} , n G N, is a sequence in a metric space

X, D such that sn —> y and sn —> y', then y = y'. That is, a sequence
in a metric space can converge to at most one limit.
Proof. We will suppose y ^ y', and prove a contradiction. Set

V = hD (V> V0
(Fig. 2.14). Then
N(y, p) n N(y', p) = <t>.

For if w G N(y, p) n N(y', p), then w e have

D(y, y') < D(y, w) + D(w, y') < p + p = D(y, y'),

a contradiction. But since sn —> y and sn y', both N(y, p) and N(y , p)
28 Metric Spaces 2.5

each contain all but finitely many of the sn. Therefore some sn must be
in both N(y,p) and N(y',p), contradicting the fact that N(y,p) and
N(y', p) have no points in common. Therefore y = y'.
Open sets were defined in terms of p-neighborhoods. Since con­
vergence is also defined using p-neighborhoods, we might suspect that
convergence can also be characterized solely in terms of open sets (rather
than p-neighborhoods). The following proposition gives such a char­
acterization.

Proposition 7. A sequence S = {sn}, n e N, in a metric space X, D

converges to y if and only if any open set which contains y contains
all but finitely many of the sn.

Proof. Suppose S converges to y and U is any open set which contains y.

Since U is open, there is p > 0 such that N (p, p) C U. But since sn —> y,
all but finitely many of the sn are elements of N (y, p); therefore all but
finitely many of the sn are elements of U.
Conversely, suppose that given any open set U which contains y, all
but finitely many of the sn are elements of U. Let p be any positive num­
ber. Then N(y, p) is an open set which contains y\ hence all but finitely
many of the sn are elements of N(y, p). Therefore sn —> y.

EXERCISES

1. Discuss the convergence of each of the following sequences in the spaces

indicated.
a) sn = 1+ 1/n, in the space of real numbers with the absolute value metric
b) sn = (2, 2), in the plane R2 with the metric D2 (Example 3)
c) sn = (2, n), in the plane R2 with metric D3
30 Metric Spaces 2.6

Figure 2.15

Exam ple 14. Let R be the space of real numbers with the absolute value
metric. Then the function defined by f(x) = 2x + 3 from R to R is
continuous. A simple calculation shows that f( N (a, p/2)) C N (/(a), p),
for any a E R.
It is usually quite awkward to prove the continuity of a function
directly from Definition 6. We therefore need propositions which will
help us determine whether or not a function is continuous, but which are,
in general, easier to apply than Definition 6.
P roposition 8. Let / be a function from the metric space X , D into the
metric space F, D f. Then / is continuous if and only if, given any
open set U of F,
rlm= {xe
is an open subset of X.
Proof. First suppose that / is continuous and that U is an open subset
of F. Let x G f ~ l(U); then f(x) G U. Since U is open, there is a positive
number p such that N(f(x), p) C U. Since / is continuous, there is a
positive number q such that

f(N{z, q))C N(/{x), p) C

Therefore N(x, q) C f ~ l(U). Since, for x we have found

such that N(x, q) C f ~ l (U), then/ '{(7) is open.
Suppose, on the other hand, that is an open subset of A' when­
ever U is an open subset of Y. We have previously shown that if /(a) e Y
and if pis any positive number, then N(f(a), p) is an open subset of Y;
therefore f~ 1(N(J(a), p)) is an open subset of X. There is therefore a
positive number q such that

N(a, q) c r l (N(J(a),p)).

We have then that for this q, f(N(a, q)) C p ) ; hence / is con-

tinuous.
2.6 Continuity 31

The following proposition is quite similar to Proposition 8, but is

often easier to apply.
P roposition 9. Let / be a function from the metric space X , D into
the metric space F, D'. Then / is continuous if and only if given any
D'-p-neighborhood U in Y ,f~ l (U) is an open subset of X.
Proof. Suppose / is continuous. If U is any D'-p-neighborhood in F,
then U is an open subset of F. Therefore/~1(f/) is an open subset of X
by Proposition 8.
Suppose that f~ 1(U) is an open subset of X whenever U is a D'-p-
neighborhood in F. Let V be any open subset of F. By Proposition 8 we
will have shown that / is continuous if we show that f ~ 1(V) is an open
subset of X. Now V is the union of D '-p-neighborhoods (Proposition 3),
say V = U/ Ui, where each Ui is a D'-p-neighborhood. Then

r
\
v
)
=(u
1 t/,) = u r\v$.
But each is by hypothesis an open subset of X ; hence f ~ x(V) is
the union of a family of open subsets of X and consequently is open.
Therefore / is continuous.
Exam ple 15. Using Proposition 9, we will show that the function / from
R 2 with the Pythagorean metric to R, the set of real numbers with the
absolute value metric, defined by /(x, y) = x, is continuous. If a e R
and p is any positive real number, then N (a, p) is the open interval
(a — p, a + p). Then /-1 (N(ayp)) is easily seen to be

{(x, p)|a — p < x < a + p}

(Fig. 2.16), an open subset of R 2. Therefore, by Proposition 9, / is con­

tinuous.

Exam ple 16. The identity function i [defined by i(x, y) = (x, y)] from
R 2with metric D onto R 2 with metric D x is continuous, since any D -open
subset of R 2 is Dx-open, and conversely (Section 2.3, Exercise 4). That
32 Metric Spaces 2.6

is, if U is any open subset of R 2, Di, then = i(U) = U is also

an open subset of R 2, D ; hence i is continuous. Since i~ l = i, and since
each D -open set is also Z^-open, we see that i~~l is continuous as a func­
tion from R 2} D 1 onto R 2, D. It is quite possible, however, for a one-one
function from a metric space, X, D onto a metric space F, D' to be con­
tinuous without f~ l being continuous, as will be demonstrated in Ex­
ample 17.
The following proposition relates continuity with the convergence of
sequences.
Proposition 10. Let / be a function from the metric space X , D into the
metric space F, D f. Then / is continuous if and only if given any
sequence S = {sn}, n e N, in X such that sn —> y>

K S) = {/(«n)}i n G AT, converges to f(y) in F.

Proof. Suppose that / is continuous, but that there is a sequence S = {sn},

n G iV, in I such that sn —> y, but /(£) does not converge to /(?/). Since
f(S) does not converge to f(y), there must be a positive number p such
that N(f(y), p) excludes infinitely many of the f(sn). But since / is con­
tinuous, there is a positive number q such that f(N{y}q)) C N(f(y), p ).
By assumption, sn —* y; hence N(y, q) contains all but finitely many of
the sn. This implies that N (f(y)}p) contains all but finitely many of the
/(sn), a contradiction.
Conversely, suppose that given any sequence S = {$n} , n e N, in X
such that sn —»y, then f(S) converges to f(y)] assume that / is not con­
tinuous. Then, since / is not continuous, there is a point/(a) in F and a
positive number p for which there is no positive number q such that
f(N(a, q)) c N
Consider the family {Un}, n g N , of neighborhoods of a, where Un =
N(a, 1/n). For each n we can select sn G Un such that/(sn) S iV(/(a), p ) ;
this selection is possible because / is not continuous. Then sn —> a (Exer­
cise 1); but {/(sn)}, n e N, is a sequence in F which does not converge to
/(a), since X(/(a),p) by construction of {sn}, u g N , contains no /(sn)
whatsoever. This contradicts our initial hypothesis that / preserves the
limits of sequences; hence / could not be discontinuous. Therefore / is
continuous.
A continuous function is thus seen to be one which in some sense
preserves the convergence of sequences. (See Exercise 7 also.)
Example 17. The identity function i from R 2 with metric D onto R 2 with
metric D 2 (Example 3) is not continuous. For the sequence defined by
sn = (1, 1/n) converges to (1,0) with respect to metric D (Example 12),
2.6 Continuity 33

but the sequence (z(sn)} = {sn}, n G N does not converge to 1, 0) =

(1,0) with respect to D 2. Note, however, that
f " 1: R 2y D 2- + R 2,D
is continuous. This follows from the fact that any sequence in R 2 which
converges with respect to D 2 is essentially a constant sequence (Section
2.5, Exercise 4), and any constant sequence converges with respect to
any metric whatsoever on R 2. We thus see that even the identity function
from a set X with one metric onto the same set with a different metric
may fail to be continuous.

EXERCISES

1. In the converse part of Proposition 10, prove sn —»a.

2. Discuss the continuity of the function in each of the following:
a) the function defined by f(x) = 5x -f 7 from the space R, D onto itself,
where R is the set of real numbers and D is the absolute value metric;
b) the function defined by f(x, y) = x -f- y from R2, D\ onto R, D [with
R, D as in (a)];
c) the function defined by f(g) = g(0) from the space X, D of Example 5
onto the closed interval [0, 1] considered as a subspace of the space of real
numbers with the absolute value metric;
d) the identity function i from R2, D\ onto R2, Ds.
3. Suppose / to be a function from a metric space X , D into the metric space
7, D' such that D(x, x') > kD'(f(x), f(x')), where A; is a constant positive
real number. Prove that / is continuous.
4. Suppose that / is a continuous function from X, D into 7, D', and g a con­
tinuous function from 7, D' into Z, D". Prove that g °/ is a continuous
function from X, D into Z, D".
5. Assume/ to be a function from X, D onto a subspace W of 7, Z)'. Prove that
/ is continuous as a function from X, D into 7, Df if and only if / is con­
tinuous as a function from X, D onto Wy D' | W.
6. Suppose W is a pubset of 7, D. Prove that the function i : W~ ^ Y defined
by i(w) = w for each w G W is continuous as a function from Wy D | W
into 7, D.
7. Prove that a function / from a space X, D into a space 7, D' is continuous if
and only if given any convergent sequence S in X, f(S) is a convergent se­
quence in 7. [Hint: It must be shown that if S converges to x in X, then
/(£) converges to/(x).l
8. The concept of equivalent metrics was introduced in Section 2.3, Exercise 6.
Prove that metrics D and D' on a set X are equivalent if and only if the
identity map from both X, D onto X, D' and from X, D' onto X, D is
continuous.
34 2.7

9. a) Suppose f : R 2 ->R 2 takes any circle in R2 onto a circle. Need / be

continuous if R2 has the usual Pythagorean metric?
b) Suppose f : R 2—*R 2 takes collinear points into collinear points. Need
/ be continuous?
10. Prove that the set /-1(N(a, p)) C R2 of Example 15 is open with respect
to the usual Pythagorean metric.

2.7 ‘
‘DISTA N CE” BETWEEN TW O SETS

D efinition 7. Let X, D be any metric space. Suppose that i £ l and

A C X . Then define

D(x, A) = greatest lower bound {D(z, a) \a E A}.

If A and B are subsets of X, define

D(A, B) = greatest lower bound {D(a, b) |a E A, b E B}.

The following equalities follow immediately from the definition:

D(x ,A) = D({x}, A)

and
D(A, B) = gib {D(a, B) |a e A} = gib {D(A, b) |b E B}.

Note, however, that D is not a metric for the set of all subsets of X . It
is quite possible, for example, to have sets W and Y such that D(W, Y) = 0,
but W n Y = <£ (a contradiction to Definition liii) as we see from the
following.
Exam ple 18. Let I?2 be the plane with the Pythagorean metric D. Set

Y = {(x, y)|x 2 + y2< 1} and = {(

Figure 2.17 Figure 2.18

2.7 D istance" Between T w o Sets 36

(Fig. 2.17). Then D(Y, W) = 0, since there are points of Y arbitrarily

close to (1,0), but (1, 0) is not a point of F. If we let Z = {(—1, 0)}, then

D(Z, W) = 2 > D(W, Y) + D(Y,Z) = 0 + 0 = 0.

Therefore the “ metric" D on the set of all subsets of R 2 does not even
satisfy the triangle inequality.
Note too that the distance between a set and any of its nonempty
subsets is always 0.
Even though the “ metric" for the subsets of a metric space is not
really a metric according to Definition 1, it is still of great use in helping
us describe the properties of metric spaces.

Proposition 11. Let X, D be a metric space. A subset F of X is closed

if and only if given any point x in X — F, D(x, F) ^ 0.
Proof. If F is a closed subset of X, then X — F is open. Therefore, given
any x E l — F, there is a positive number p such that N(x, p) C X — F.
But then D(x, F) > p; hence D(x, F) 5^ 0.
Suppose, on the other hand, that given any point x in X — F r
D(x, F) 0. Then setting p = D(x, F), we have N(z, p) C X — F.
That is, for each x e X — F, we have a positive number p such that

N(:r, p) C X — Fj

which is to say that X — F is open. Therefore F = X — (X — F) is

closed.

P roposition 12. Let X, D be a metric space. Suppose F is a closed

subset of X and i e X - F . Then there are open sets U and V of X
such that x E U, F C V , and U n V = <f>.

Proof. Since F is closed and x G X — F, D(x, F) 5^ 0 (Fig. 2.18). For

each y e F, set
Uy = N (y,

Then V = Up Uy is an open set which contains F. Also U =

N(x, %D(x, F)) is an open set which contains x. In order to complete the
proof that U and V satisfy the terms of Proposition 12, we must show that
U n V = <j>. Suppose U n V 7* <t>, and select w from U fl V. Then
D(wyy) < %D(x, F) for some y e and D(w, x) < %D(xyF). It then
follows that

D(x} y) < D(w, x) + D(w, y) < \D(xyF) + iD(x, F) = D(x, F).

36 Metric Spaces 2.7

But D(x, F) = gib {D(xf z) \z £ F}, and y e F ; hence D(a;, ?/) > Z)(x, F),
a contradiction. Since the assumption that U fi V 5* </> led to a con­
tradiction, it must be that U n V = </>.
We now prove an even stronger result.

Proposition 13. Let X, D be a metric space. Then if F and F' are two
closed subsets of X such that F n F r = </>, there are open sets U and
V of X such that
FcU, F' C V , and U nV = 0.
(Note that this proposition contains Proposition 12 as a special case,
since each one-element subset of X is a closed subset by Proposition 5.)

Figure 2.19

Proof. For each y £ F, set py = \D(y, F'), and for each y' £ F ', set
TV = %D(y',F). Set U = U F N(y,py) and V = U F'N(y',p'y') (Fig.
2.19). Since both U and V are the union of a family of open sets, both are
open. Then, since F C U and F' C V, it remains to show that U fi V = <£.
Suppose we can find w £ U D V. It follows that D(y, w) < p y for some
y £ F y and that D(y', w) < py* for some y' £ F'. We may suppose
Ptt ^ Vv’- Then
D
(y
,y') < D(y, w) + D(y\ w
)< py +

But D(y, yr) > D(y, F'), hence a contradiction. It follows then that
U n V = <t>.
Propositions 12 and 13 represent what are called separation properties
because they measure our ability to “ separate”or distinguish disjoint
closed subsets. Oddly enough, as the following example demonstrates,
Proposition 13 does not imply that the “ distance”between two disjoint
nonempty closed subsets of a metric space is always greater than 0.
 2.7 "Distance" Between Two Sets 37

Example 19. Let R 2be the coordinate y

plane with the Pythagorean metric
D y and let
{(.x, y)| y =
and
F' = { ( * , y )| = 0 }

(Fig. 2.20). That is, F is the graph of

the rectangular hyperbola y = l/x,
while F f is the x-axis. Both these
sets are closed and F n F' = </>.
Since y = l/x has the x-axis for an
asymptote, D(F, F') = 0. Never­
theless, F and F' can still be sep­
arated in the sense of Proposition 13.
We saw in Proposition 11 that a subset F of a metric space X, D is
closed if and only if each point of X which is 0 distance from F is an element
of F. This inspires the following definition.
Definition 8. Let X, D be a metric space and A C X. We define the
closure of A } denoted by Cl A, by
Cl A = (x e I | D(Xy A) = 0}.
Example 20. Let R be the space of real numbers with the absolute
value metric. Set A = {1/n \n = 1, 2, 3, . . .}. Since 1/n —> 0, then
d(0, A) = 0 (Exercise 5). Since Z)(l/n, 1/n) = 0 for each n, then
A C Cl A. On the other hand, if y is any number other than 0 or an
element of A f then it is readily verified that D(p, A) > 0. Therefore
Cl A = A U {0}.
Proposition 14. Cl A as given in Definition 8 is a closed subset of X.
Proof. Suppose that Cl A is not closed. Then X — Cl A is not open;
therefore there is an element x G X — Cl A such that for any positive
number p, N(x, p) fi Cl A ^ 0. Select w E N(x, p) n Cl A. Then since
N(x} p) is open, there is a positive number q such that N(wyq) C N(x, p).
But since w G Cl A, then D(w, A) = 0; therefore there is at least one
element
aGA n N(Wy q) C N(x, p).
But this means that, for any positive number p, there is an element a G A
such that D(Xy a) < p. It follows then that gib {D(x, a) |a e A} =
D(xy A) = 0; thus x E Cl A, a contradiction, since x G X — Cl A.
Cl A must therefore be a closed subset of X.
2.7 "D istan ce" Between Tw o Sets 39

c) Fr{x |x is a rational number} C R, as in (a)

d) Fr{(z, y) j x = 3} C R2, with metric D2 (Example 3)
7. Prove that a subset A of a metric space X, D is open if and only if FrA fl A =
<j>. Prove that A is closed if and only if FrA C A.
3
TOPOLOGIES

3.1 THE NOTION OF A TOPO LO GY

The fundamental properties of open subsets of a metric space are outlined

in Proposition 2 of Chapter 2. Mathematicians have found from experience
that families of subsets having these same properties arise in contexts
other than those of metric spaces; hence it is reasonable to study these
properties in their own right, abstracted from the limitations that metric
spaces impose. In particular, the properties of open sets in metric spaces
inspire the following definition.
Definition 1. Let X be any set. A collection t of subsets of X is said

to be a topology on X if the following axioms are satisfied:

i) X and <£are members of r.
ii) The in te r s e c t io n o f a n y t w o m e m b e r s o f r is a m e m b e r o f t .

iii) The union of any family of members of r is again in r.

The members of r are then said to be T-open subsets of X, or merely
open subsets of X if no confusion may result.
Example 1. If X, D is a metric space, then the D-open subsets of X form
a topology on X. This topology is called the metric topology induced on
X by D. It was, of course, this topology that we studied in Chapter 2.
Example 2. Let X be any set. Then the family of all subsets of X forms
a topology on X. This topology consisting of all of the subsets of X is
called the discrete topology on X. The discrete topology contains the maxi­
mum possible number of open sets since, relative to the discrete topology,
every subset of X is open.
Example 3. If X is any set, then the collection {X, <£} of subsets of X also
forms a topology on X. This topology is called the trivial (by some, the
indiscrete) topology on X. It contains the fewest possible open sets com­
patible with having a topology on X.
The discrete and trivial topologies represent opposite extremes.
Topologies which are of genuine interest usually lie somewhere in between;
for example, the topology induced on the set of real numbers by the
40
3.1 The Notion o f a Topology 41

absolute value metric is neither the trivial nor the discrete topology. We
now give an example of a topology which is neither discrete nor trivial,
but also is not related to any metric.

Example 4. L etX = {a, b}. Define r = {X, </>, {a}}. It is easily verified
that t is a topology on X. Suppose D to be any metric on X , and set
p — D(a, b). Then AT(6, p) = {&}, and it follows that {b} is a D-open set.
But {b} is not a r-open set; hence r could not be the topology induced
on X by D.

Definition 2. A set X with topology r is called a topological space.

Just as X, D was used to denote a set X with metric D, so X, r will be
used to denote a set X with topology r.
As in metric spaces, so in a topological space X, r we say that a
subset F of X is r-closed (or merely closed) if F = X — U, where U
is a r-open set. (Compare this to Definition 4 of Chapter 2.)
Proposition 1. Let X, r be a topological space. Then the closed subsets
of X have the following properties.
a) X and <f> are closed subsets of X.
b) The union of any two closed subsets of X is again a closed subset
of X.
c) The intersection of any family of closed subsets of X is again a
closed subset of X.
The proof is the same as the proof of Proposition 4, Chapter 2.
The next proposition shows that rather than defining a topology on
a set by specifying the open subsets, we may equally well determine the
topology by specifying the closed subsets.
Proposition 2. Let X be any set, and suppose that J is a family of
subsets of X such that
i') X and <t> are in SJ;
ii') the union of any two members of J is a member of J ;
iii') the intersection of any family of members of J is a member of $L
If we now define a subset U of X to be open if and only if U = X — F,
where F is some element of $F, then the set r of open sets thus formed
is a topology on X with as the set of (r-) closed subsets of X.
Proof. We first show that r is a topology on X by verifying that r satisfies
Definition 1.
i) X and </> are in r. Since X and are in $7, and since X = X — </>
and <£ = X — X, then X and <f> are in r.
42 T o p o lo g ie s 3.1

ii) The intersection of any two members of r is a member of r. Sup­

pose U and V are members of r. Then
U = X - Fx and V = X - F 2,

where F x and F 2 are in £F. Therefore

u n v = (X - Fi) n ( x - f 2) = x - (Fi u f 2).

But, by (ii'), Fi U F 2 G £F; hence (7 n F G r.

iii) The union of any family of members of r is a member of t . Suppose
{Ui}, i G /, is a family of members of r. It follows that for each
i G I, Ui = X — F*, where F»G SF. Then

U Ui = U (X - Fi) = X - PI F t.
i i i

But, by (iii'), f l j F* G hence U j C/»G r.

Therefore r satisfies the definition of a topology on X.
It remains to be shown that ^ is the set of closed sets for the topology
r. Suppose F is r-closed. Then F = X — [/, where Z7 G r. But £/ =
X — F', where F' G SF; therefore

F = X — (X — F') = F' G iF.

On the other hand, if F G 9r, then X - F g t . Then, since F = X —

(X — F), F is r-closed. The members of SF are therefore precisely the
r-closed subsets of X.

Example 5. We define a family $F of subsets of 7Z2, the coordinate plane,

as follows: Let F G £Fif and only if F = F 2, F — 0, or F is a set consisting
of finitely many points together with the union of finitely many straight
lines. By hypothesis, X and 0 are in It follows from the fact that two
straight lines can only intersect in either a straight line (if they coincide),
a point, or the empty set, that the intersection of any family of members
of SF is again a member of IF. Since the union of finitely many lines and
points with finitely many more lines and points still consists of finitely
many lines and points, the union of any two members of 3 is also a mem­
ber of JF. Therefore, $ satisfies (i') through (iii') of Proposition 2, and
hence determines a topology on R 2. The topology which $ determines is
in fact the smallest topology in which lines and points are closed sets.
It is not, however, the topology induced on R 2 by the Pythagorean metric
D; for a subset of R 2 can exclude at most finitely many lines and still be
open in the topology determined by but {(x, y) \x2 + y2 < 1} ex­
cludes infinitely many lines and still is D-open.
3.2 Bases and Subbases 43

EXERCISES

1. Prove that the intersection of finitely many open sets is open and that the
union of finitely many closed sets is closed.
2. Suppose X, t a topological space, and A C l . The interior of A, denoted by
A°, is defined by A° = U {U E r | U C A}. Prove the following:
a) (A°)° = A°.
b) A°C A.
c) (AnB)° = A°nB°.
d) A subset U of X is open if and only if U = U°.
3. Let R2 be the plane with the Pythagorean metric. Let 31 be the set of all
p-neighborhoods in R2. Which properties of a topology for R2 does 31 fail
to satisfy? Let t be the set of all unions of elements of 31. Prove that r is a
topology for R2. What topology is this?
4. Find all possible topologies for the set {1, 2, 3}.
5. Suppose X a set with more than one element. Prove that there is no metric
on X which induces the trivial topology on X. Find a metric for X which
induces the discrete topology on X.
6. Let N be the set of positive integers. Define a subset F of N to be closed if
F contains a finite number of positive integers, or F = N. Show that the
closed subsets of N thus defined satisfy the conditions of Proposition 2, and
hence can be used to define a topology on N. Prove that this topology is
not induced by any metric. [Hint: Show that the topology does not satisfy
Proposition 12 of Chapter 2.]
7. Prove that the topology defined on R2 in Example 5 is really the smallest
topology in which lines and points are closed sets. Would it be possible to
have a topology on R2 in which every line was a closed set, but every one-
point subset was not? in which every one-point subset was closed, but
every line was not? Try to find a topology satisfying the latter condition.
8. a) Define explicitly, that is, characterize completely the members of, the
topology on R2 which has the fewest members and relative to which each
one point subset of R2 is closed,
b) Characterize the smallest topology on R 2 relative to which each straight
line of R2 is closed. Is this the same topology found in (a)? Does it
contain the topology found in (a)?

3.2 BASES AND SUBBASES

Quite often it is impractical to explicitly specify all the open sets in order
to define a topology on some set. Note that when we defined an open sub­
set of a metric space, it was done in terms of p-neighborhoods and not by
listing each open set separately. Proposition 2 has also shown us that we
could equally well define a topology by giving the closed sets instead of
44 T o p o lo g ie s 3.2

the open ones. We now investigate other methods of determining a

topology on a given set. In the first of these methods, a collection of sets
is furnished which “generates”the topology (in much the same way that a
basis of a vector space generates the vector space).

Definition 3. Suppose that X, r is a topological space. A subset (B of

r (i.e., a collection of open sets) is said to be a basis for the topology
r if each member of r is the union of members of (B.

There is no analog of linear independence in Definition 3. Any topology

has at least one basis, namely itself. Generally it is of no consequence
whether or not a basis is in any sense minimal.

Example 6. Suppose R 2 to be the plane with the Pythagorean metric D.

Then the p-neighborhoods of R 2 form a basis for the topology induced by
D (see Section 3.1, Exercise 3). In fact, if X, D is any metric space, then
the p-neighborhoods of X form a basis for the topology induced by D.

Note that if X, D is any metric space x e X and p > 0, then there is

a rational number q, 0 < q < p, with N(x, q) C N(x, p). This fact can be
used to prove that

{N(x, q) |x G X, q a positive rational number}

is also a basis for the topology induced by D (Exercise 6). It can also be
proved that
{N(x, 1/n) |x G X and n a positive integer}

is a basis for the metric topology (Exercise 6). Thus we see that bases
for a topology can be quite diverse.

Proposition 3. Suppose that X, r is a topological space and that (B is

a basis for r. Then the intersection of any two members of (B is the
union of members of <B, and X itself is the union of members of (B.

Proof. Since both X and the intersection of any two members of (B are
members of r, such sets must be the union of members of (B.

The case often occurs when rather than being given a topology for a
set X, we are merely given a collection of subsets of X. For example, in
our study of metric spaces, it was the p-neighborhoods which arose most
naturally; the open sets were defined after the p-neighborhoods had been
introduced. We might, then, reasonably ask, When is a collection of
subsets of X the basis for a topology on X ? The following proposition
answers this question.
3.2 Bases and Subbases 45

Proposition 4. Let X be any set. Assume (B to be a family of subsets

of X such that
i') X is the union of members of (B;
ii') the intersection of any two members of (B is the union of members
of (B.
Define r = {U C X \ U is the union of members of (B}. Then r is a
topology on X and (B is a basis for r. (The topology r for which (B is a
basis is, in fact, unique; see Exercise 3.)

Proof. We must verify that r satisfies Definition 1.

i) X is the union of members of (B, by (i'), and </> is the union of the
empty subfamily of (B. Therefore X and $ are members of r.
ii) Suppose that U and V are inr. Then U = Ui B{ and V = U/ B j,
where I and J are appropriate index sets and where Bi and Bj are
members of (B for each i e I and j e J. Then

UnV= U (Bi n Bj).

i, J

Since each Bi n Bj is the union of members of (B by (ii'), U fl V

is the union of members of (B, and hence is in r. The intersection
of any two members of r is again a member of r.
iii) If {[/*}, f c E i f , is any family of members of r, then Uk is the
union of members of (B for each k G K. Therefore Uk Uk is the
union of members of (B, and hence is in r. That is, the union of
any family of members of r is again a member of r. Therefore r
satisfies the definition of a topology on X. Since each member of
t is by definition the union of members of (B, (B is a basis for r.

Example 7. Let R be the set of real numbers. Clearly R is the union of

open intervals. Since the intersection of any two open intervals in R is
either empty or again an open interval, condition (ii') of Proposition 4 is
satisfied by the collection of open intervals in R. The family of open
intervals in R thus forms the basis for a topology on R. This topology is
the same as the topology induced on R by the absolute value metric
(Exercise 1).
Suppose X to be any set, and S any collection of subsets of X. Com­
bining Propositions 3 and 4, we see that S is the basis for a topology on X
if and only if S satisfies conditions (i') and (ii') of Proposition 4; but not
every family of subsets of X satisfies these conditions. We may ask,
therefore, in what topologies on X the given sets are open. There is, how­
ever, generally no unique topology on X for which the given sets are open.
46 T op ologies 3.2

For example, the p-neighborhoods in R 2 with the Pythagorean metric D

are open in the topology induced by D, but they are also open with respect
to the discrete topology. We may therefore rephrase our question by
asking, What is the “ smallest" topology r on X , that is, the topology
having the “ fewest" open sets, for which S C t? This question is answered
in the following proposition.

Proposition 5. Let X be any set and suppose S to be a collection of

subsets of X. Set

<B = {B |B is the intersection of finitely many sets in S or B = X}.

Then (B is the basis for a topology r on X defined by

r = {U | U is the union of members of ©}.

Moreover, S C t, and if r' is any topology on X such that S C t',

then t C t '; that is, r is the smallest topology on X for which S is a
collection of open sets.
Proof. We first show that (B satisfies (i') and (ii') of Proposition 4.
i') X is itself the union of members of (B, since X G (B by hypothesis.
ii') If Bi and B 2 are both in (B, then both B x and B 2 are the inter­
section of finitely many members of S. Therefore B x C\B2 is it­
self the intersection of finitely many members of S, and is hence
in (B. The intersection of any two members of (B is thus again a
member of (B. By Proposition 4, then, (Bis the basis for a topology,
the topology r as defined above, on X. Clearly S C r .
If S is to be a collection of open subsets in any topology on X, then
all finite intersections of members of S must also be open sets (Section 3.1,
Exercise 1), and hence any union of a family of these intersections must
also be open. Since r is the smallest topology on X in which these con­
ditions are fulfilled, it is therefore the smallest topology on X for which S
is a family of open sets.
Proposition 5 inspires the following definition.
Definition 4. Let X, r be a topological space. A subset S of r is said
to be a subbasis for r if the set

(B = {B |B is the intersection of finitely many members of S}

is a basis for t .
Proposition 5 tells us that any collection of subsets of X whose union!
is X is the subbasis for a unique topology on X.
3.2 Bases and Subbases 47

{x|x<6} {x\a<x}

C {x|«<*<6} ^ Figure 31

Example 8. We saw in Example 7 that the family of open intervals in the

set R of real numbers is the basis for a topology on R. Each open interval
(Fig. 3.1) is, however, the intersection of two half-lines (rays) without
endpoints, and each such half-line is open in the topology determined by
the open intervals. Thus the set of all half-lines without endpoints forms
a subbasis for the open-interval topology.

Example 9. Let R 2 be the coordinate plane with metric Z>3 as in Chapter

2, Examples 3 and 6. As in any metric space, the D 3-p-neighborhoods in
R 2 form a basis for the topology induced on R 2 by Z>3. Note that each
Z)3-p neighborhood of R 2 is the intersection of finitely many open half­
planes (Fig. 3.2), and that each of these half-planes is open in the topology
induced by Z>3. The collection of open half-planes is therefore a subbasis
for this topology.

Figure 3.2

Thus far we have four means of specifying a topology on a set X:

(1) by explicitly giving the open sets, that is, the members of the topology;
(2) by explicitly giving the closed sets; (3) by giving a basis for the topol­
ogy; °r (4) by giving a subbasis for the topology.

EXERCISES

1. Prove that the topology on R, the set of real numbers, for which the collec­
tion of open intervals is a basis is the same as the topology induced on R by
the absolute value metric.
2. Let R2 be the coordinate plane and let D, D\, Z>2, and Z>3 be the metrics
described in Chapter 2, Examples 2 and 3.
48 T opologies 3.2

a) Prove that the collection of open half-planes is a subbasis for the topol­
ogies induced by D and D\.
b) Prove that the collection of closed half-planes (that is, a half-plane and
its bounding line) is a subbasis for the topology induced by Z>2.
c) Suppose that X, t is a topological space and S a subbasis for r. Prove
that r is the only possible topology on X for which S is a subbasis; that is,
if rf is a topology on X for which S is also a subbasis, then r = r'.
d) Using (c), prove that the topologies induced by D, D\ and Ds on R2 are
equal.
3. Let X be any set. Suppose a collection (B of subsets of X is the basis for
topologies r and r' on X. Prove t = r. Thus any collection of subsets of X
which satisfies (i') and (ii') of Proposition 4 is a basis for one and only one
topology on X.
4. Prove that each of the following are bases for topologies on the prescribed
sets.
a) the set of intervals of the form [a, b) in the set of real numbers
b) X = {/1/ is a function from [0, 1] into [0, 1]} and the collection of sub­
sets of X of the form

B s ={ f e x If(x) = 0 for X e S},

where is some subset of [0, 1]
c) X = {p |p is a polynomial with real coefficients} and the collection of
subsets of X of the form
Bn = { p E X \ degree of p = n},
where n is a nonnegative integer
5. Prove that
{N(x, q) |q is a rational number, x 6 l }
and
{N(x, 1/n) |x E X and n a positive integer}

are both bases for the topology induced on X by D as is claimed in Example 6.

6. Let N be the set of positive integers. Find explicitly all the open sets in the
smallest topologies on N for which each of the following is a collection of
open sets.
a) N and 0 b) N, {1,2}, {3,4,5} c) N, {1,2}, {3,4,5}, {1,4,7}

7. Let X f D be a metric space. For each xt y E X , define Hi(x ,y) to be

{w E X | D(x, w) > D(y, w)} and tf2(z, y) = {w E X | D(xf w) < D(y, w)}.
a) Prove that Hi(x, y) and H 2(x, y) are open with respect to D.
b) Describe these sets relative to two distinct points of R2with the Pythag­
orean metric.
c) Prove or disprove: {H\{x> y) | (x, y) E X X X, x j* y} is a subbasis for
the topology induced on X by D.
3.3 Open Neighborhood Systems 49

3.3 OPEN NEIGHBORHOOD SYSTEMS

Although we already have four ways to specify a topology on a set, we
have not as yet formally introduced one of the most widely used manners
of determining a topology. Actually, however, we have already en­
countered this method, since it is nothing more than the generalization
of the p-neighborhoods of a point in a metric space.
Definition 5. Suppose X, r to be a topological space, and suppose
that for each point x G X we have a collection 91* of open sets having
the following properties:
i) 91* ^ 0.
ii) x G N for ea ch N G 91*.
iii) If N i and N 2 are in 91*, then there is X 3 G 91* such that iV3 C
Ni n n 2.
iv) Given N G 91* and any y G N, there is N' G 9l y such that N' C N.
v) A subset U of X is open if and only if for each x G U, there is
N G 91* such that N C U.
Then the collection of families of members of r (one for each x G X)
is called an open neighborhood system for r.

Example 10. Suppose X, D is a metric space. Set 91* = {N(x, p) |p > 0},
for each x G X. We will verify that the collection of 91* forms an open
neighborhood system for the topology induced on X by D.
i) Since N(x, 1) G 91* for each x g I , 91* 9^ 0 for each x G X.
ii) D(x, x) = 0 for any x G X implies that x G N{x} p) for any
p > 0. Therefore x G N for any N G 91*.
iii) Suppose N\ snd N 2 are in 91*. Then N\ = N(x, pi) and N 2 =
N(x} p 2), where p i and p2 are positive numbers. We may suppose
Pi > P2- Then
N x n N 2 = N(x, p 2) e 91*.

iv) This is essentially Proposition 1, Chapter 2.

v) This is the definition of open set in the topology induced by D.

The following proposition relates open neighborhood systems and

bases for a topology.

Proposition 6. Suppose that X, t is a topological space. Then if 91

is any open neighborhood system for r, the collection of subsets of X
contained in 91 forms a basis for r. On the other hand, if (B is any
basis for r, then, setting 91* = {B G (B |x G B} for each x G X, we
obtain an open neighborhood system for r.
3.3 Open Neighborhood System s 51

N 2 CV. Therefore
x g N 1h N 2 c U n v .

By Definition 5(iii), there is N3 E 9lx such that

x e N 3c N 1n N 2c U n V ;

hence U n V is also an open set. The intersection of any two

open sets is again an open set.
iii) Suppose {Ui}, i E /, is a family of open sets, and x E U j Ui.
Then x E Ui for some i; hence there is N E 9lx such that x G N C
[/f c U / The union of any family of open sets is thus again
an open set. Therefore r is a topology on X.
The reader should compare the proof of Proposition 7 with the proof
of Proposition 2, Chapter 2. Why might one expect to see many simi­
larities?

Example 11. Let R be the set of real numbers. For each x E R, let 3lx
be the set of all half-open intervals having x as a left-hand endpoint; that is,

9lx = {\x, a) |a E R, x < a}.

The reader should verify at once (Exercise 3) that the collection of 3lx
satisfies (i) through (iv) in Definition 5. In accordance with Proposition 7,
then the collection of 9lx determines a topology on X. Exercise 2 shows
that this topology is unique.

Example 12. Let R 2 be the coordinate plane. For each x E R 2, let 3lx
be the set of interiors of all triangles which contain x in their interior.
Then the collection of 9lx forms an open neighborhood system for a topol­
ogy on X.

Example 13. The following topology has applications in algebraic geom­

etry. Let S be a ring. For each s E S, define

3IS — {s + A | A is a nonzero ideal of S},

that is, is the set of cosets of s. It can be verified that the collection
of 91* satisfies (i) through (iv) of Definition 5 and hence forms an open
neighborhood system for a topology on S (Exercise 5).
The reader should note that even though the definition of an open
neighborhood system seems more cumbersome than other methods of
specifying a topology, in actual practice it is often the easiest and most
natural way.
3.4 Finer and Coarser Topologies 53

Definition 6. Let X be any set, and suppose that r and r' are two
topologies on X. Then r is said to be finer than r' if r' C t, that is,
any r'-open set is also a r-open set. r is said to be strictly finer than
r' if r is finer than r', but t ^ r'. If r is finer than r', then we may
say that r' is coarser than r.
Proposition 8. Let X be any set. Then two topologies r and r' on X
are equal if and only if r is finer than r' and r' is finer than r.
Proof, t finer than r' means r' C r. r' finer than r means r C r'. There­
fore t — t '.

Example 14. Let X be any set. Then the discrete topology on X is finer
than any topology on X and is strictly finer than any other topology on X.
The trivial topology on X is coarser than any topology on X.

Example 15. Let r and r' be any two topologies on some set X. Let
§
>= t n r' and S' = r U r'.

Then S and S' are subbases for unique topologies ri and t 2, respectively,
on X. (By Proposition 5, any collection of subsets of X is a subbasis for
a unique topology on X. The reader should be certain that he under­
stands that r n t' is the family of all subsets of X which are both r-open
and r'-open, e.g. X and 0, and not intersections of r-open and r'-open sets.)
T\ is coarser than both r and r', since any ri-open set is both r-open and
r'-open. r 2, on the other hand, is finer than both r and r'. The reader
should also see Exercise 5.
Proposition 9. Let r and r' be two topologies on some set X. Suppose
that 91 and 91' are open neighborhood systems for r and r', respectively.
Then r C r' if and only if for each x £ X and N £ 91*, there is X' £ 9l'
such that N ' C N.
Proof. Assume U to be any r-open set and x £ U. Then there is N £ 91*
such that x £ N C U. Now if there is N' £ 9l' such that x £ X' C X,
then x £ X' C X; hence £7 is also r'-open. Therefore r C r'. On the other
hand, if r C r', then since X £ r, N e t ' ; hence there is X' £ 9l' such
that X' C X. (Note that the reason for virtually every step in this proof
is Definition 5v.)
Corollary 1. Let r and r' be two topologies on the set X. Suppose 91
and 91' are open neighborhood systems for r and r', respectively.
Then r = r' if and only if for each x £ X and X £ 91*, there is X' £ 9l'
such that X' C X, and for each X' £ 9l', there is X £ 91* such that
X cX '.
54 Topologies 3.4

Proof. Applying Proposition 9, we see that this corollary merely states

that r = r' if and only if r C r' and r' C r.
Corollary 2. Suppose X to be a set and D and D r possible metrics on
X. Then the topology induced by D is the same as the topology in­
duced by D f if and only if for each x e X and positive number p,
there are positive numbers p x and p 2such that the D-px-neighborhood
of £ is a subset of the D'-p-neighborhood of x, and the D'-p2-neigh-
borhood of x is a subset of the D-p-neighborhood of x.
Proof. The p-neighborhoods of points in a metric space form an open
neighborhood system for the metric topology (see Example 10). Corol­
lary 2 then is merely a restatement of Corollary 1 applied to metric spaces.
Example 16. Suppose that R 2, the coordinate plane, is given either the
Pythagorean metric D, or the metric D i of Chapter 2, Example 3. It was
shown in Chapter 2, Example 6 that the hypotheses of Corollary 2 apply ;
hence the topologies induced by D and D x on R 2 are the same.
Example 17. Let r be the topology on R 2 which is defined by the
open neighborhood system described in Example 12. It is easily verified
(Fig. 3.3) that r is the same topology as that induced on R 2by the Pythag­
orean metric (Exercise 7).

Figure 3.3

The reader might be tempted to conjecture that two topologies r and

t' on some set X are equal if and only if given bases (Band (B' for r and
t'j respectively, for each B e (B there is B' e (B' with B' C B, and for
each B' e there is B e (Bsuch that B C B'. This conjecture is, how­
ever, false, as is-shown by Example 11 and Section 3.3, Exercise 3. Each
interval of the form [xf a) contains some interval of the form (p, q) [though
of course (p, q) could not contain x]9 and each interval of the form (p, q)
contains some interval of the form [x, a). The set of all half-open intervals
of the form [x} a) forms a basis for a topology r', and the set of open
intervals also forms a basis for a topology r. If the conjecture were correct,
these two topologies should be equal, which they are not.
3.5 Derived Sets 55

EXERCISES

1. In Example 15, prove as asserted that t\ is coarser than both r and r', and
that T2 is finer than both r and r'.
2. Prove that the topology induced on R)the set of real numbers, by the absolute
value metric is the same as the topology for which the set of all open intervals
is a basis.
3. Prove that the topologies induced on R2 by the metrics D\ and Z>3,
Chapter 2, Example 3, are equal.
4. Define a metric D' on the set R of real numbers by D'(x} y) = 3|x — y\.
How does the topology induced on R by D' compare with the topology in­
duced on R by the absolute value metric? Answer this same question with
D' replaced by D", where D" is defined by D"(x, y) = \x — y\2.
5. In Example 15, prove that t \ is the finest topology which is coarser than
both r and r', and that T2is the coarsest topology which is finer than r and r'.
6. Find all possible topologies on {x, y, z}. Order these topologies as to fineness
and coarseness. Construct a diagram which illustrates the relationships be­
tween the topologies.
7. In Example 17, prove that r is the same as the topology induced on R2 by
the Pythagorean metric.
8. Suppose (B and (B' are bases for topologies r and r', respectively, on a set X.
Suppose that each member of (B' contains a member of (B. Are t and r'
necessarily comparable? If so, in what way?

3.5 DERIVED SETS

Let X, r be a topological space. Then associated with any subset A of X,
there are a number of sets which are topologically related to or “
derived”
from A. We have already encountered some of these sets in the discussion
of metric spaces.
Definition 7. If A c X, where X, r is a topological space, then we
define
a) the closure of A, denoted by Cl A, to be the intersection of all
closed sets which contain A (cf. Chapter 2, Definition 8 and
Proposition 15);
b) the interior of A, denoted by A°, to be the union of all open sets
which are contained in A (Section 3.1, Exercise 2);
c) the frontier of A, denoted by Fr A, to be
{x |each open set which contains x contains points of
both A and X — A}
(Section 2.7, Exercise 6);
56 Topologies 3.5

d) the exterior of A, denoted by Ext A, to be X — Cl A ; and

e) the derived set of A (sometimes called the weak derived set), denoted
by A', to be

{x |if x e U, U an open set, then A n (U — {x}) ^ <j>;

that is, if x G U, then U — {x} contains some point of A}.

Example 18. Let R 2 be the plane with the topology induced by the
Pythagorean metric D. Let

A = {(x, y) |x2 + y2 < 1}

(Fig. 3.4). Then

C IA = {(x, y ) \ x 2 + y 2 < 1}, A° = A,

Fr A = {(x, y) |x2 + y2 = 1},
Ext A — {(x, y) |x2 + y2 > 1}, and A' — Cl A.

The reader is not expected to see all of these equalities until more in­
formation has been obtained about these topologically derived sets, but
he should verify as many as possible. He should also examine this example
for possible relations that might hold between the sets topologically
associated with A. These reflections also hold for the following example.

Example 19. Let R be the set of real numbers with the topology induced
by the absolute value metric, and let A be the set of rational numbers.
Then Cl A = R, A° = </>, Fr A — R, Ext A = <f> (note that both A°
and Ext A can simultaneously be empty), and A' = R. If R is given the
trivial topology, then these sets topologically associated with A are
exactly the same as in the metric topology. We can conclude then that
what the topologically derived sets happen to be for any one subset of R
3.5 Derived Sets 57

does not give much information about the topology. If, however, we
know, say, Cl A for every A C X, then the topology is completely deter­
mined, as we shall see from Proposition 11.

Proposition 10. Suppose X, r is a topological space and A and B are

any subsets of X. Then
i) A C Cl A; ii) C1(C1 A) = CI A;
iii) C1(A U B) = Cl A U Cl B ; iv) Cl0 = 0;
v) A is closed if and only if A — Cl A.

Proof. Cl A is the intersection of a family of sets each of which contains

A; therefore A C Cl A, and (i) is proved. Since Cl A is the intersection
of a family of closed sets, Cl A is closed. If A is already closed, then A is
one of the closed sets which contains A ; hence Cl A C A. Since A C Cl A
by (i), A — Cl A. We have therefore proved (v), (ii), and (iv).
It still remains to prove (iii). Since A C C1(A U B) and B C C1(A U B),
we have

Cl A C C1(C1(A uB)) = Cl (A u B) and Cl B c Cl (A u B).

Therefore Cl A U Cl B C C1(A U B). On the other hand, since Cl A U
Cl B is the union of two closed sets, it is closed. Thus Cl A U C1J5 is a
closed set which contains A u B] consequently, C1(A U B) C Cl A U Cl B.
Therefore Cl (A u B) = Cl A u Cl B.

Proposition 11. Let X be any set, and suppose that Cl is a function

from the set of subsets of X into the set of subsets of X such that Cl
satisfies (i) through (iv) of Proposition 10. Then if we define a subset
of X to be closed in accordance with (v), the collection $ of closed
subsets thus obtained satisfies (i') through (iii') of Proposition 2 and
hence determines a topology r on X. Moreover, Cl A is the closure
of A with respect to r for each subset A of X.
Proof. We must verify (i') through (iii') of Proposition 2.
i') Since X C Cl X C X, X - Cl X; therefore X is closed. Then
Cl <t> = <t> by (iv).
ii') Let F and Ff be any two closed sets. Then F = Cl A and F' =
Cl B , where A and B are subsets of X. Hence

F U F' = Cl A u Cl B = Cl (A U B),

and it follows that F U F' is also a closed subset.

iii') Let {Fi}, i E /, be any family of closed subsets of X. Then
Fi = Cl Fi for each i. Now f)i Fi C Fi for each i; it follows
58 Topologies 3.5

therefore that

C1 (/ ic C1 F
F
[For fl/ Fi C Fi implies (flj F,) U Fi — Fi, and thus

Cl ((f) F-) u F.) = Cl (I) F t) U Cl Fi

= Cl Fi = Cl (0 Fi) u Fi = Fi.]
Therefore
ci (n f^ c n f„
By (i), however, fl/ Ft-C C l(flj F,); hence

D i Fi = 01(0, Ft).

We have shown then that fl, Ft-is closed. The family 57of closed
subsets of X therefore satisfies (i') through (in') of Proposition 2
and hence defines a topology on X.
Now if A is any subset of X, then since C1(C1 A) = Cl A, Cl A is
closed with respect to r. But A C Cl A by (i); hence the r-closure of A
is a subset of Cl A. If F is any set such that Cl F — F and A C F, then
Cl A C Cl F = F. Therefore the intersection of all such F, the r-closure
of A, contains Cl A; that is, Cl A C r-closure of A. Hence Cl A is the
same as the r-closure of A.

EXERCISES

1. Suppose I , r a topological space. If A and B are any two subsets of X,

show that it is not true in general that C1(A D B) = Cl A fl Cl B.
2. Let X, t be any topological space. Compute Cl X, X°, Fr X, Ext X, X',
and the corresponding sets for 0.
3. Suppose X is any set and °is a function from the set of subsets of X to the
set of subsets of X with the following properties:
i) A°C A, ii) (A°)° = A°,
iii) ( i fl 5)° = .4°f| B°, andiv) X° = X, where A and B are any
subsets of X.
Define a subset U of X to be open if and only if U° = U. Prove that the
set of open sets thus defined gives a topology on X.
4. Suppose that X is any set and that r and r' are topologies on X with r finer
than r'. If A C X, we will denote the closure of A with respect to r by Cl A
and the closure of A with respect to r' by Cl' A. Analogous notation will
be used with regard to the other sets topologically derived from A. Prove
3.6 Topologically Derived Sets 59

that
a) Cl ACC1' A; b) A°'C A°.
c) Find relations between the corresponding other sets topologically derived
from A.
5. Try to find a method for specifying a topology on a set X by specifying Fr A
for each A C l Do likewise for Ext.
6. Suppose that X is a set with the discrete topology and that A C X. Find
sets topologically associated with A.
7. Is it possible for two distinct subsets of a topological space to have exactly
the same topologically derived sets? Support your assertion.
8. Suppose t and r' are topologies on a set X. Determine if each of the following
conditions implies either r C r' or r' C r. In the following, A stands for any
subset of X; we use ' to indicate that a derived set is being taken relative
to r'.
a) Fr A C Fr' A b) Cl A C Cl' A c) Ext A C Ext' A

3.6 MORE ABOUT TOPOLOGICALLY DERIVED SETS

In this section we continue the discussion begun in Section 3.5. Through­
out this section X, r will be assumed to be a topological space.
Proposition 12. If A c X, then
a) Cl A = A u A'; b) ClA= A° u Fr A;
c) Fr A = Fr(X — A); d) Cl A- Fr A = A°.

Proof
a) Assume that x e Cl A, but that x £ A. We prove first that for
each open set U which contains x, U D A ^ 0. If A n U = 0,
then X — U is a closed set which contains A ; hence Cl A C X — U.
But since x e U, x could not be in Cl A, a contradiction. We
have then that for each open set U which contains x, A n
{U — {x}) <t>. But then x G A'. Therefore
Cl A C A U A'.
Now suppose that y e A U A'. If y e A, then y e Cl A, since
A C Cl A. Suppose further that y E A', and that F is a closed
subset of X which contains A, but not y. Then X — F is open;
hence X — F is an open subset which contains y such that
((X — F) — {y}) n A = 0. Therefore y could not be in A', a
contradiction. Thus y is contained in any closed set which con­
tains A, and hence y e Cl A. This gives A u A ' c C I A ; it
follows that
Cl A = A U A'.
60 Topologies 3.6

b) Suppose x e Cl A, but x & Fr A. Since x S Fr A, there is some

open set U which contains x such that either V C A, or [/ C X — A.
If U C X — A, then X — U is a closed set which contains A ;
therefore Cl A C X — U, contradicting the assumption that
x E Cl A. It must be then that U C A, and hence x e A°. There­
fore
Cl A C A0 U Fr A.
Assume that y E A° u Fr A, but that y Cl A. Since A° C A C
Cl A, y must be in Fr A. Since y 2 Cl A, there is a closed set F
which contains A, but not y. Then X — F is an open set which
contains y, but does not intersect A; hence y could not be in
Fr A, a contradiction. Therefore y must be in Cl A. It follows
that A° u Fr A C Cl A ; hence
C IA - A° U Fr A.

c) Suppose x E Fr A. Then any open set U which contains x meets

both A and X — A. Then U meets X — A and X — (X — A) = A,
and hence x e Fr(X — A). T hm - jrc Fr(X—^ A). Similarly, if
x E Fr(X — A), x E Fr A.
d) Proposition 12(d) will follow from (b) if we show that
A° n Fr A = <£.
Suppose A° D Fr A ^ <t>, and choose x in this intersection. Then
since x E Fr A, every open set which contains x meets X — A.
Since x E A°, however, there is an open set U which contains x
such that U C A. Clearly these two possibilities mutually exclude
one another; thus it is impossible to have x in both A° and Fr A.

The following terminology is introduced as an aid in making certain

statements about topological spaces.

Definition 8. Let Z, r be a topological space. If x E X, then any open

set which contains x is said to be a neighborhood of x. (Some texts
define a neighborhood of x to be any set which contains x in its interior,
and refer to what we have defined to be a neighborhood as an open
neighborhood. Such variations in terminology should be expected, how­
ever, in topology, since topology is still a rather young branch of
mathematics and much terminology still has not become universally
accepted.)
The following proposition relates neighborhoods and the topologically
derived sets.
Proposition 13. Suppose that X, r is any topological space and A C X.
3.6 Topologically Derived Sets 61

Then
a) x 6 Cl A if and only if every neighborhood of x meets A ;
b) x E A° if and only if some neighborhood of x is contained in A;
c) x E Ext A if and only if x has some neighborhood disjoint from A ;
d) x e Fr A if and only if every neighborhood of x meets both A
and X — A ;
e) x G A' if and only if every neighborhood of x meets A in some
point other than x.
Proof
a) By Proposition 12(a), Cl A = A U A'. If x E Cl A and x G A,
then every neighborhood of x meets A (in x). If x E A', then every
neighborhood of x meets A also. On the other hand, if every
neighborhood of x meets A, then either x E A, or x G A'; hence

x e A U A' = Cl A.

b) If x E A°, then there is a neighborhood (open set) U of x such that

U C A by definition of A °. Conversely, if there is a neighborhood
V of x with U C A, then x e A °.
c) If x e Ext A, then since Ext A = X — Cl A is open, Ext A is a
neighborhood of x disjoint from A. On the other hand, if there is a
neighborhood of x disjoint from A, then x G X — Cl A = Ext A.
d) and e) are merely the definitions of Fr A and A' stated in terms
of neighborhoods.
Example 20. Let R be the set of real numbers with the topology induced
by the absolute value metric D. Let A = (0, 1]. Given any a e A,
a ^ 1, it is possible to find a neighborhood of a which lies entirely in
A [since A — {1} = (0, 1) is open]. On the other hand, no neighborhood
of 1 lies entirely in A; hence A
°= (0, 1). There are only two points, 0
and 1, with the property that every neighborhood of each of these points
meets both A and R — A. Therefore Fr A = {0, 1}. Since Cl A =
A°U Fr A, Cl A = [0, 1]. If U is a neighborhood of any point x in [0, 1],
then U intersects A in some point other than x; hence A' = [0, 1]. Note
that although A °n Fr A = <£,it is not true in general that A n A7= <t>.
We also have (see Fig. 3.5).
Ext A = R — Cl A = {x e R |x > 1, or x < 0}
A = {x|0<£ <1}
Ext A 0 4°= (0,1) 1 ExtA

C\A = [0, 1]

Figure 3.5
62 Topologies 3.6

Example 21. Let N be the set of positive integers, and define a subset U
of N to be open if U contains all but finitely many positive integers. The
set r of open subsets of N forms a topology for N (Section 3.1, Exercise 6).
Let A be the set of even positive integers. Then Cl A = N, since any
open set which contains any integer must contain at least one (in fact an
infinite number) of even integers. For if the open set excluded all even
integers, it would exclude infinitely many positive integers and hence
would not be open. It is true that A° = 0, since any subset of A excludes
infinitely many positive integers and hence could not be open. Also,
Fr A = Cl A - A° = N - 0 = N.

Note that Fr A can be larger than A. Then Ext A = N — Cl A = 0.

Finally, A' = N, since any neighborhood of any integer contains both
even and odd integers.
The notion of denseness is important in topology. Although we will
not develop the concept in this chapter, this is an appropriate place to
define it.
Definition 9. Let A, r be a topological space. A subset A of A is said
to be somewhere dense if
(Cl A)° * 0,

that is, if the closure of A contains some open set. A is said to be

nowhere dense if A is not somewhere dense. A is said to be dense if
Cl A = A.

If A is any subset of a topological space such that A° ^ 0, then A

is somewhere dense, since A° C A C Cl A.

Example 22. Let R be the set of real numbers with the topology induced
by the absolute value metric. The set A = [0, 1) is somewhere dense,
since A° = (0, 1) ^ 0. The set of integers Z is nowhere dense, since Z
is closed [R — Z is the union of open sets of the form (n — 1, n), n an
integer, and hence is open] and no neighborhood of any integer contains
only integers; that is,

Cl Z = Z and Z° = 0.

Since any neighborhood of any number contains a rational number, the

closure of Q, the set of rationals, is all of R (Proposition 13a). Therefore Q
is dense in R.
The following proposition gives a simple criterion for determining if
any given set is dense.
4
DERIVED TOPOLOGICAL SPACES.
CONTINUITY

4.1 SUBSPACES
We have already noted that if X, D is a metric space and Y C X, then
F, D | Y is also a metric space (Section 2.1, Example 4). Recall that a
subset W of Y is D | F-open if and only if W = Y n U, where U is a
D-open subset of X (Section 2.4, Exercise 2). However, F is not merely
a subset of X, but is a subspace of X, and the topology which D | F in­
duces on F can be defined by means of the topology which D induces on X.
Suppose F is a subset of a topological space X } r. It is reasonable then to
inquire whether there is any topology on F which is “ induced”by r.
Using what we learned about metric spaces, the following definition seems
in order.

Definition 1. Let I , r be a topological space and F C X. A subset

W of F is said to be open in Y if
p\ h
W = Y n U, w h ere P et.

Proposition 1. If X, r is a topological space and F C X, then the set

of all subsets of F which are open in F forms a topology for F.
Proof. We shall show that the set of all subsets of F which are open in F
satisfies the definition of a topology on F (Chapter 3, Definition 1).
i) X and </>are members of r. Therefore F n X = F and F n <£ = <t>
are open in F.
ii) Suppose U and V are open in F. Then U = Y n U' and V =
Y n Vf, where Uf and V' are open subsets of X. Then

u n V = (F n U') n (F n V') = Y n (Ur n V').

But U' D V' is an open subset of X; hence U fi V is an open

subset of F.
iii) Suppose {[/»}, i E 7, is a family of open subsets of F. Then
Ui = F n where is an open subset of X for each i e 7.
64
4.1 Subspaces 65

Then
u Ui = u (y n ui) = y n (u •

Since U j is the union of open sets, it is open, and therefore

U j U{ is open in F. The set of subsets of Y which are open in Y
therefore forms a topology on Y.
Proposition 1 enables us to make the following definition.
Definition 2. The topology on Y described in Definition 1 and Prop­
osition 1 is called the subspace topology on Y. Y with this topology is
said to be a subspace of X. If X, r is a topological space and Y C X,
then Y will be assumed to have the subspace topology when con­
sidered as a topological space, unless explicitly stated otherwise.
Example 1. If X, D is a metric space and Y C X, then the topology in­
duced on F by D | Y is the same as the subspace topology on Y induced
by the metric topology on X. It was in fact this example which inspired
us to define the subspace topology as we did. (See the remarks opening
this chapter.)
Example 2. Let N be the set of positive integers with the topology defined
by declaring a set to be open if it contains all but at most finitely many
elements of N. Let Y — {1, 2, 3, 4, 5}. We now show that the subspace
topology on Y is the discrete topology. Suppose n G F. Then

U(n) = {n} U {N - Y)

is an open subset of N, since it excludes only four positive integers. There­

fore U(n) n Y — {n} is open in Y. Every one-point subset of Y is there­
fore open in F, and hence every subset of Y (being the union of one-point
subsets) is open in F. Consequently Y has the discrete topology. We
thus see that it is quite possible for a subspace to have the discrete topology
even when the space itself does not have the discrete topology.
Example 3. If a set X has the discrete topology, then every subspace of
X has the discrete topology. If X has the trivial topology, then every
subspace of X has the trivial topology. The proof of these assertions is
left as an exercise.
Proposition 2. If F is a subspace of X and W is a subspace of F, then
W is a subspace of X. That is, if F is given the subspace topology from
X, and then a subset W of F is given the subspace topology considered
as a subset of the topological space F, then W would be given the same
topology as though W were considered as a subset of X and were
given the subspace topology.
66 Derived Topological Spaces 4.1

Proof. Let ry be the topology on W considered as a subspace of Y, and

let Tx be the topology on W considered as a subspace of X. We must
show that r x = ry. Suppose U e Tx• Then U — W n U', where Uf
is open in X. But then

u = w n U' = (W n Y) n U' = w n (F n £/'),

since W C Y; hence U G t y - On the other hand, if [ / E r y , then U =
W (1 U', where U' is open in Y. But since U' is open in Y, IV = Y CI 17",
where £7" is open in X. It follows that

u = w n U ' = W n ( Y n U") = (Wn Y) n U" = TFn c/",

and hence ?7 G r j. Therefore r x = Ty.

EXERCISES

1. a) Prove that every subspace of a topological space with the discrete topology
has the discrete topology,
b) Prove that every subspace of a space with the trivial topology has the
trivial topology.
2. Suppose that X, r is a topological space and that F is a closed subset of X.
Prove that a subset W of F is closed in F if and only if TP is a closed subset of
X. Make and prove the corresponding statement about open subsets of X.
3. Assume X, r a topological space and (B a basis for r. If Y C X , define
(By = {BD Y\ £e(B}.
Prove that (By is a basis for the subspace topology on Y.
4. Suppose R is the space of real numbers with the topology induced by the
absolute value metric. Prove that the following subsets of R do not have
the discrete subspace topology.
a) the set of rational numbers
b) {x |x = 0, or x = 1/n, where n is a positive integer}
c) {x |x = qw, where q is a rational number}
d) any subset of R which is somewhere dense
5. Let X, D be any metric space. Prove that if Y is a finite subset of X , then
the subspace Y has the discrete topology. Is this true if we replace finite by
countable?
6. Prove or disprove: A subset A of a topological space X, r is nowhere dense
if and only if the subspace A has the discrete topology.
7. Suppose X, t is a topological space with the property that every two-point
subspace of X has the trivial topology. Prove that X has the trivial topology.
Show that the corresponding statement about the discrete topology is not
true.
4.2 Derived Sets in Subspaces 67

8. Find the coarsest topology on R2 for which each finite subspace of R2 has the
discrete topology. Find the finest topology on R2 for which each finite sub­
space of R2 has the trivial topology.

4.2 THE TOPOLOGICALLY DERIVED SETS IN SUBSPACES

Suppose that X, r is a topological space and that F is a subspace of X.
If A is a subset of F, then A is also a subset of X. We may wish to know
the sets topologically associated with A either with respect to the topology
on X or with respect to the subspace topology on F. As we shall see, the
corresponding sets are not always equal, nor should we expect them to be,
since it is not at all true that a set open in F is necessarily open in X.
Nevertheless, the subspace topology on F is defined in terms of the topol­
ogy on X ; there should therefore be some relationships between the
corresponding sets. It is the purpose of this section to investigate these
relationships.
Example 4. Assume that R is the set of real numbers with the topology
induced by the absolute value metric. Let

F = (0, I) and A = { x | 0 < x < l and x is rational}.

Then the closure of A in F is (0, 1), while the closure of A in R is [0, 1].
Fr F in F is </>, since no subset of F, open or otherwise, contains any
elements of F — F = 0. But Fr F in R is (0, I}.

Example 5. Let the plane R 2 have the topology described in Example 5

of Chapter 3. Let

F = {(x, y) |x2 + y2 < 1} and A = {(x, y) |x2 + y2 < 1}.

There is no closed subset of 1R2, except R 2, which contains A ; that is, no

union of finitely many lines and points contains all of A. Therefore the
closure of A in R 2 is all of R 2. Suppose F is a subset of F which is closed
in F and which contains A. Applying Proposition 3 below, then F =
F n F'j where F' is a closed subset of R 2. But F' contains F; hence
A C F'. Since R 2 is the only closed subset of R 2 which contains A , it
follows that F' = R 2. Thus F = Y n R 2 = F . We have then that the
closure of A in F is F.
Proposition 3. If X, r is a topological space and F is a subspace of A,
then a subset F of F is closed in F if and only if F = Y n F'f where
F' is a closed subset of X.
Proof. Suppose F is closed in F. Then Y — F is open in F ; therefore

Y — F = F n f7,
68 Derived Topological Spaces 4.2

where U is an open subset of X. But then Y — (Y — F) = F — Y CI

(X — U). Since U is open, I - [/is closed; hence F is of the desired
form.
Suppose further that F — F n Fr, where F' is a closed subset of X.
Then
Y - F = Y n ( X - F').

But X — F' is open; hence Y — F — Y n (X — F') is open in Y. There­

fore F is closed in Y.
Note that this proposition tells us that the subspace topology on Y
could equally well have been defined by defining the subsets of Y which
are closed in Y in the same way that the subsets of Y which are open in
F are defined (substituting closed for open, of course) in Definition 1.
Proposition 4. Let X, r be a topological space, and let F be a sub­
space of X. If A C F, then

C IA in F = F n C l A (inX).

Proof. Cl A is closed in X and A C Cl A ; hence F n Cl A is a closed

(in F) subset of F (Proposition 3) which contains A. Therefore Cl A in
F C F n Cl A. Now Cl A in F is a closed (in F) subset of F, and thus,
again by Proposition 3, there is a closed subset F of X such that Cl A in
F = F n F. But A C F; hence Cl A C F. It follows that

F n C l A c F n ^ - C l A i n F.

We therefore have Cl A in F = F n Cl A.
The reader might be tempted to conjecture that A° in F = F n A°.
This is not true, as is shown by the next example.

Example 6. Let R 2 be the coordinate plane with the topology induced by

the Pythagorean metric, F = {(x, y) |y = 0}, that is, F is the x-axis,
and A = Y. Then A° in F = Y, while

F n A ° = F n 0 = 0,

since F contains no open subset of R 2. We might also note that Fr A in

F = 0, while Fr A = A ; hence it is also false that Fr A in F = F n Fr A.
Proposition 5. Suppose that X, r is a topological space and that for
each x e X , we have a collection 3lx of subsets of X such that the 31*
form an open neighborhood system for X. Let F C X. Then setting

31' = {F n N I N E 9ly}
4.2 Derived Sets in Subspaces 69

for each y E Y, we obtain an open neighborhood system for the sub­

space topology of Y.
Proof. We must show that the 91' satisfy Definition 5 of Chapter 3.
i) Since there is at least one N E 3ly for each y E Y C X, there is
N fl Y E 31'; hence 9l' ^ <t>.
ii) Definition 5(ii) follows at once from the fact that y E N for each
N E 9l„ .
iii) Suppose N[ and N 2 are in 91'. Then N[ = Y n N± and N 2 =
Y n N 2 for some N i and N 2 in 91^. There is, however, N$ EWy
such that N s C N i C \ N 2. Therefore
N's = Y n Ns C N[ H N'2 and N’
z Effly.

The proofs of (iv) and (v) are left as exercises.

Since N E 91^ is an open subset of X, N' — Y C\ N is an open (in Y)
subset of F. Therefore we do have an open neighborhood system for the
subspace topology on F.

EXERCISES
«
i

1. Prove (iv) and (v) in PropositiQn 5.

2. Assume U to be an open subset of a topological space X, r and A C U. Is it
true that Fr A in U = Fr A n U? Does this equality hold if U is a closed
subset of X rather than an open subset?
3. Why is it true that Cl A in F = Cl A fl F, but that A° in F ^ A°n F?
(See Proposition 4 and Example 6.) "Try to reproduce the proof of Proposi­
tion 4 for A°instead of Cl A and see where the proof fails.
4. Let R2 be the coordinate plane with the usual Pythagorean metric. Let
F = {(x, y) |x2+ y2 < 1, or a; = 0 or 1 and y = 0 or 1}.
For each of the following subsets of F compare the sets topologically derived
from these sets in F with those topologically derived in X. That is, compute
Cl A in F and compare it with Cl A in X, etc.
a) (Or, y) |x = 0 and y = 1/n, n a positive integer, or y = 0}
b) (Or, y) |x2+ y2 < 1} c) {(z, y) |either x or y is irrational}
5. Suppose that F is a subspace of A, r and that A C F. Prove
a) Fr A in F C Fr A D F; b) A°C A°in F.
6. Compute Ext A, A', A°, and Fr A in F, for A and F in Example 5.
7. Find a necessary and sufficient condition for each subset A of a subspace W
of a space X, r to have the same frontier relative to W as A has relative to X.
Find such a condition on W in order to have A' in W equal to A' (in X) for
each subset A of W.
70 Derived Topological Spaces 4.3

4.3 CONTINUITY

We now come to one of the central notions in all of topology: continuity.

We have already encountered continuity in connection with metric spaces.
Then a continuous function was a “ nearness-preserving”function. Neigh­
borhoods of a point in a general topological space are in a sense measures
of nearness, just as the term “ neighborhood”implies. As we have seen,
however, many topological spaces are not metric spaces, nor can they be
made into metric spaces by defining an appropriate metric. We therefore
need a definition of continuity which will reduce to the metric definition
of continuity when we are dealing with a metric space, generalize ap­
propriately the idea of a “ nearness-preserving”function, and not depend
on metrics (or anything else which is not common to all topological spaces)
for its definition. In Chapter 2 we found at least one criterion for the
continuity of a function from one metric space to another which does not
include any mention of the metrics in its statement (Proposition 8, Chap­
ter 2). We will therefore use this proposition for a generalized definition of
continuity.

Definition 3. Let A, r and F, r' be topological spaces. Then a function

/ from X to F is said to be continuous if given any open subset U of F,
then f ~ l (U) is an open subset of X.

Exam ple 7. If X is any space with the discrete topology and F is any
topological space, then any function / from X into F is continuous. For
if U is any subset (open or not) of F, then f ~ 1(U) is an open subset of X,
since every subset of X is open.

Example 8. If X is any space with the trivial topology and / is any func­
tion from X onto a space F, then / is continuous if and only if F has the
trivial topology. For if F has the trivial topology, then F and <£ are the
only open subsets of F; hence is open in X (being either </> for
U = </>, or X for U = Y) for any open subset U of F. On the other hand,
if F does not have the trivial topology, then there is an open subset U of
F which is neither F nor <j>. Then/_1(C7) is neither X nor </>, and hence is
not an open subset of X. Therefore f could not be continuous.
The following proposition is the generalized version of Definition 6 of
Chapter 2.

Proposition 6. A function / from a topological space A, r to a space

F, r' is continuous if and only if given any f(x) e F and any neigh­
borhood V off(x), there is a neighborhood U of x such that

f(U) C V.
4.3 Continuity 71

Proof. Suppose / is continuous. Then if V is any neighborhood of f(x),

V is an open subset of F. Therefore /-1(F) is an open subset of X which
contains x; that is, /-1(F) is a neighborhood of x. Setting U = /-1(F),
we have the desired result.
Suppose that given any f(x) E Y and any neighborhood V of f(x),
there is a neighborhood U of x such that f(U) C V. Let W be any open
subset of F; we must show that/-1 (W) is an open subset of X. Suppose
2 e / -1(TF). Then /(z) E W, that is, W is a neighborhood of f(z). Then
there is a neighborhood U of z such that f(U) C W. But then U C f ~ l (W).
We therefore have that for each z E / -1(TF), z e U C f ~ l (W), where U is
an open subset of X. Hence f ~ 1(W) is the union of open subsets of X,
and thus is an open subset of X. Therefore / is continuous.
Propositions 7 and 8 give further criteria for the continuity of a
function.
Proposition 7. Suppose that X , r and F, r' are topological spaces, and
that / is a function from X to F. Let (B be any basis for r'. Then / is
continuous if and only if for each B E (B, f ~ 1(B) is an open subset of X.
(Compare this with Proposition 9, Chapter 2.)
Proof. Assume / continuous. Then since each B e (B is an open subset of
F, f ~ l (B) is an open subset of X. Suppose instead that f~ 1(B) is an open
subset of X for each B E (B. Let V be any open subset of F. Then V —
U i Bi, where each J i s a member of (B and I is a suitable index set. It
follows that

r \ v ) = r 1( u s y = u r 1^ ) .

But f ~ l {Bi) is open in X for each i E I] hence /-1(F) is the union of a

family of open sets and is therefore open. Consequently, / is continuous.
Corollary. Suppose that / is a function from X, r into F, r' and that
{91^}, y E F, is an open neighborhood system for r'. Then / is con­
tinuous if and only if given any N E d l y for any y E F, /-1(A0 is an
open subset of X.
Proof. The collection of all N contained in some 91y forms a basis for r'
by Proposition 6 of Chapter 3. The corollary then follows at once from
Proposition 7.

Example 9. Let R 2be the coordinate plane with the usual metric topology.
A "rotation”of R 2 is best described using polar coordinates. If A 0 is an
angle measured in radians, define

7 ^ 0(r, A) = (r, A + A 0)
72 Derived Topological Spaces 4.3

for any point (r, A) (expressed in polar coordinates) of R 2. The reader

may recall from analytic geometry that R a 0 is a rotation through angle
A0. Then R A\ = R(- Aoh the rotation through angle — A 0. Any rotation
preserves congruences; in particular, if U is the interior of some triangle
or square, then #!*(£/) is also the interior of a triangle or square. Since
the family of interiors of triangles, or the family of interiors of squares,
forms a basis for the standard topology on R 2, any rotation is continuous.
The inverse of any rotation, also being a rotation, is continuous.
Proposition 8. Let X, r and F, r' be topological spaces. Then a func­
tion / from X to F is continuous if and only if given any closed subset
F of F, f ~ x(F) is a closed subset of X.
The proof is left as an exercise.
Proposition 9. Suppose that X is any set and that r and r' are topol­
ogies for X. Then r is finer than r' if and only if the identity function
i from X to X defined by

i(x) = x for all x g I

is continuous from the topological space X, r to the topological space

X, r'.
Proof. Assume i continuous. If U G r', then = i(U) = U is in r.
Therefore r' C r , that is, r is finer than r ' . Suppose r is finer than r ' .
Then if U e r',

i~\U ) = U e r

(since any r'-open set is r-open). Therefore i is continuous.

The following proposition shows that the composition of continuous
functions is continuous.
Proposition 10. If / is a continuous function from the space X, r to
the space F, r' and if g is a continuous function from F, r' to Z, r",
then g of is a continuous function from X , r to Z , r".
Proof. Suppose U is an open subset of Z. Then g~ 1(U) is an open subset
of F, since g is continuous. But then since / is continuous,

r ' ig - ' iU ) ) = (0 O/ )-!([/)

is an open subset of X. Therefore g°f is continuous.

Propositions 11 and 12 pertain to continuous functions as they are
related to subspaces. Proposition 11 deals with a function which is known
to be continuous on certain subspaces of a space X.
4.3 Continuity 73

Proposition 11. If / is a function from X, r to F, r', X = A U B , and

/ 1A and / 1Z? are both continuous (where A and B are considered as
subspaces of X), then if A and B are both open or both closed, / is
continuous.

Proof. We will prove Proposition 11 for the case when A and B are both
closed. The case when A and B are both open is left as an exercise. We
use Proposition 8. Let F be any closed subset of F. We must show that
f~ l (F) is a closed subset of X. (/ | A)~1(Z?T) is closed in A and (/ |B)~l (F)
is closed in B, since / 1A and / |B are both assumed to be continuous.
But since A and B are closed, (/1A)~l (F) and ( / 1B)~l (F) are closed
subsets of X (see Section 4.1, Exercise 2). Since A U B = X,

r\F )= ( f,)
f \ A r 1(F
)u

which is a closed subset of X since it is the union of two closed subsets

of X. Therefore / is continuous.

Example 10. Let R be the set of real numbers with the topology induced
by the absolute value metric. Define /: R —> R by

if x > 0,
if x < 0.

Set A = {x |x > 0} and B = {x |x < 0}. Then A U B = R, A and B

are closed, and / 1A and / 1B are easily seen to be continuous. There­
fore, by Proposition 11, / is continuous.
Note that A and B must either both be closed, or both be open. One
cannot be closed and the other open. For if we continue to let R be the
space of real numbers with the absolute value topology and set

A = {x |!* > 0},

B = {x j\x < 0},

and define g: R R by
, ^ (3, if x G A
= L if X E: B i

then A is closed, B is open; but g is not continuous.

The next proposition answers the following questions:

a) Suppose that / is a continuous function from X, r onto a subspace
Y of Z, r". I s / then continuous as a function from X to Z?
b) Suppose that / is a continuous function from X, r to F, r' and that
IF is a subspace of X. Is / | IF continuous?
74 Derived Topological Spaces 4.3

Proposition 12. Suppose / is a continuous function from X, r to F, r'.

a) If Y is a subspace of Z, r", then / is a continuous function from
X to Z.
b) If TP is a subspace of X } then / 1W is a continuous function from
TP to F.
Proof
a) Suppose U is an open subset of Z. Then F n U is open in F;
hence /-1(F n £/) is an open subset of X. But/(x) E F for every
x G X, and thus
/“‘
(tf) = n
'(Y
f~

is an open subset of X. Therefore / is continuous as a function

from X to Z.
The proof of (b) is left as an exercise.
Example 11. Let TP be a subspace of an X , r. The identity function
restricted to TP, i |TP, is sometimes called the inclusion mapping of TP
into X. If Tw denotes the subspace topology on TP, then

i | TP: TP, rw TP, Tw

is continuous; hence, applying Proposition 12(a), i | TP: TP, tw X, t is

continuous. Looking at it another way, i: X, r —> X, r is continuous, and
therefore, by Proposition 12(b), i | TP is also continuous.
Note that Proposition 12(a) implies that we never would lose any
generality by assuming that a continuous function was onto.

EXERCISES

1. Prove Proposition 8.
2. Prove Proposition 12(b).
3. Suppose that / is a function from a space X, t to a set Y. Define a subset U
of Y to be open if/-1(£7) is an open subset of X. Prove that the set of open
subsets of Y then forms a topology r'. Further, prove that / is continuous
from X, t to F, r'. Show that r' is the finest topology for which / is con­
tinuous.
4. Let / be a function from a set X to a topological space F, r'. Define a subset
U of X to be open if U = f~ l (V) for some open subset V of F. Prove that
the family of open subsets of X thus obtained forms a topology r on X.
Prove that /: X, r —> Y, t is continuous. Prove that r is the coarsest
topology for which / is continuous.
5. In Example 11, show that the subspace topology is the coarsest topology
for which i I TF is continuous.
76 Derived Topological Spaces 4.4

Example 13. Let R 2 again be the plane with the Pythagorean metric
topology. Then any triangle in R 2 is homeomorphic to any circle. We
can prove this by positioning the triangle and circle so that the center of
the circle lies inside the triangle. We can then "project”the circle onto
the triangle as shown in Fig. 4.2. An argument similar to that of
Example 12 shows that this projection is a homeomorphism.
Proposition 13. Let / be a one-one function from a space X , r onto a
space F, r'. Then the following statements are equivalent:
a) / is a homeomorphism.
b) A subset U of Y is open if and only if f ~ l {U) is open in X.
c) A subset F of F is closed if and only if f ~ l (F) is closed in X.
d) If (B is a basis for r, then/((B) = {f(B) |B e CB} is a basis for r'.
Proof. Statement (a) implies statement (b): If/ is a homeomorphism, then
both / and f ~ l are continuous. Therefore if U is any open subset of F,
/ 1(C/) is open in X. Suppose U is a subset of F such that/-1 {U) is open
in X. Since / is onto, /(/ 1(C/)) = U. But / = (Z-1)”1, and since f ~ l
is continuous and/-1 (U) is open in X,

( r ' r ' ir ' m ) = u

is open in F. Therefore U is open in F if and only if f ~ l (U) is open in X.

Statement (b) implies statement (c): Suppose F is a closed subset of F.
Then F — F is an open subset of F ; hence/-1(F — F) is open in X. But
/-1(F — F) = X — f ~ x(F), and thus /-1(F) is a closed subset of X.
Suppose that F is a subset of F such that/-1(F) is a closed subset of X.
Then
f - \ Y - F) = X - f- \ F )

is an open subset of X ; hence F — F is an open subset of F ; hence F is

closed. Therefore (b) implies (c).
Statement (c) implies statement (a): By Proposition 8, (c) states that
/ and/-1 are continuous; hence / is a homeomorphism.
4.4 Homeomorphisms 77

Statement (a) implies statement (d): Suppose U is any open subset

of F. Since / is continuous, is an open subset of X. Therefore
= U/ Bi, where each Bi e (B and I is a suitable index set. Now

f(r\u))= u= /

But each Bi is open in X, and it has been shown that (a), (b), and (c) are
equivalent (a implies b implies c implies a); hence, by (b), f(B{) is an open
subset of F. Then U is the union of members of /((B), and each member of
/((B) is an open subset of F. Therefore/((B) is a basis for r'.
Statement (d) implies statement (a): Suppose U is an open subset
of F. Then U — U if(Bi), where Bi e (B and I is again a suitable index
set. It follows that

r \ u ) = r 1 (u/(Bi)) = u r 1^ ) ) = u bu

which is an open subset of X , Therefore / is continuous. On the other

hand, if V is any open subset of X, then V — \Jj B/ where each Bj e (B.
Thus
m = / ( u B_,) = u /(By)

is the union of open subsets of F and is therefore open. But / = (/-1)—1;

hence (/“1)”1(F) is open in F if V is open in X. Therefore/-1 is con­
tinuous, and / is a homeomorphism.

We see from Proposition 13 that homeomorphic spaces are essentially

equivalent from a topological point of view. If two spaces are homeo­
morphic, there is not only a one-one function from one space to the other,
but also a natural one-one correspondence between their open sets (Prop­
osition 13b). Put another way, if X, r and F, r' are homeomorphic spaces,
then by suitably relabeling the points of F, we obtain X , and r' becomes r.
We must keep in mind, however, that homeomorphic spaces can appear
quite different from other points of view than the topological. We have
seen, for example, that a circle and a triangle are homeomorphic. From a
geometric point of view, a circle and a triangle are quite different, even
though from a topological point of view they are indistinguishable. Recall
that Euclidean geometry is primarily concerned with the properties of
objects which are preserved under rigid motions, that is, in Euclidean
geometry, we are interested in studying properties common to all objects
which are congruent. Almost all geometric studies can be classified ac­
cording to the type of properties they study; in particular, these types of
properties are those which are preserved by certain kinds of functions.
Topology, considered as a branch of geometry, studies properties preserved
by a very special type of function, the homeomorphism.
78 Derived Topological Spaces 4.4

Proposition 14. Let T be the class of all topological spaces. Then the
relation R defined on T by “ is homeomorphic t o ”is an equivalence
relation on T.
Proof. If X , r is any topological space, then X, r is homeomorphic to X, r
by the identity function. Therefore X, r is ^-equivalent to X, r.

Suppose X , r is homeomorphic to F, r' by some homeomorphism /.

Then f ~ l is a homeomorphism from F, r' to X, r. Hence if X, r is 12-
equivalent to F, r', then F, r' is ^-equivalent to X, r.
Now suppose that / is a homeomorphism from X, r to F, r', and that
g is a homeomorphism from F, r' to Z , r". Since both / and g are one-one
and onto, £ 7°/: X —»Z is one-one and onto. Since /, g, f ~ 1)and g~l are all
continuous, g°f and (g°f)~1 = are also continuous (Proposition
10). Therefore
9*f:X,r - + & T *

is a homeomorphism. Hence the relation R is transitive, and, consequently,

R is an equivalence relation on T.
From a topological point of view then, any two homeomorphic spaces
are equivalent, or interchangeable. The question of determining whether
or not two given spaces are homeomorphic is often extremely difficult.
As a matter of fact, it is usually very difficult to determine whether there
is even a continuous function from one space onto another. In most in­
stances, this problem has not been solved.
Proposition 15. Suppose that X is any set and that r and r' are two
topologies for X. Then r = r' if and only if the identity function i
on X is a homeomorphism from X, r to X, r'.
The proof is left as an exercise.

EXERCISES

1. Prove Proposition 15.

2. Let R be the set of real numbers with the absolute value topology.
a) Prove that any open interval (a, b) is homeomorphic to the interval (0, 1).
[Hint: Usef(x) = (x — a)/(b — a).]
b) Prove that the ray (a, °o) is homeomorphic to (1, co).
c) Prove that (a, oo) is homeomorphic to (— —a).
d) Prove that R is homeomorphic to (—tt/2, 7t/2). [Hint: Usef(x) = tan-1 x.]
e) Prove that (1, oo) is homeomorphic to (0, 1). [Hint: Use g{x) = 1/x.]
We thus conclude that any two open intervals of the real line are homeo­
morphic.
f) Prove that any two closed intervals of the real line are homeomorphic.
4.5 Identification Spaces 79

3. Homeomorphic spaces have essentially the same topological properties, that

is, properties related exclusively to their topologies. Although the reader
has yet encountered very few topological properties, he should be able to
make an intelligent conjecture about whether the spaces in each of the follow­
ing pairs are homeomorphic to one another. If the spaces are homeomorphic,
try to describe a homeomorphism. If they are not homeomorphic, try to
find a topological property which one of the spaces has, but which the other
space does not have.
a) the open interval (0, 1) and the closed interval [0, 1] considered as sub­
spaces of the real numbers with the absolute value topology
b) (0, 1) and [0, 1] considered as subspaces of the real numbers with the
discrete topology
c) a circle C considered as a subspace of the plane R2 with the usual topology
and the interval (0, 1) from (a)
d) {x |x is a rational number} and {n |n is an integer}, both considered as
subspaces of the real numbers with the absolute value topology
4. Let X be the space of functions described in Example 5, Chapter 2, with the
topology induced by the metric D.
a) Prove that X cannot be homeomorphic to the space R of real numbers
with the absolute value topology. [Hint: Prove that X has greater car­
dinality than [0, 1], which has the same cardinality as R. Do this by
assuming that there is a one-one correspondence between the elements of
[0, 1] and the elements of X and then constructing a function which does
not correspond to any element of [0, 1]. More particularly, if x<^ /, set
g{x) t* /(x), for each x E [0, 1].]
b) Find an embedding of R as a subspace of X.
5. Let N be the set of positive integers. Define a subset F of N to be closed if
F contains a finite number of positive integers, or F = N (cf. Exercise 6 of
Section 3.1). Prove that any infinite subspace of N is homeomorphic to AT.
Is it possible to find a nontrivial topology on the set R of real numbers such
that every uncountable subspace of R is homeomorphic to R*!

4.5 IDENTIFICATION SPACES

If X is any set and R is an equivalence relation on X y then R determines

a partition of X into /^-equivalence classes. We will denote the set of
equivalence classes by X/R. If X also has a topology r, we might inquire
if r can be used in a natural way to give a topology on X/R.
We note that there is a natural function/from X to X /R defined by
f(x) = x, where x is any element of X and x is the /^-equivalence class
of x. If X is a topological space, it is reasonable to want a topology on
X/R which would at least make/continuous. Of course, if X /R is given
the trivial topology, then / is continuous. But the trivial topology is
pretty much what its name implies, trivial. Furthermore, the trivial
topology on X /R is not necessarily related to r, and we are looking for a
80 Derived Topological Spaces 4.5

topology which is derived from r. We know that the function / will be

continuous if and only if given any open set U of X/72, is open
in X. We will use this fact to define a topology on X / R ; that is, we will
say that a subset U of X/R will be open if f ~ l (U) is open in X .
Definition 5. Suppose X, r is a topological space and R is an equiv­
alence relation on X. Let X /R denote the set of 72-equivalence
classes. Define the function / from X to X/R by f{x) = x, where x
is any element of X and x is the 72-equivalence class of x. Then / is
called the identification mapping from X to X/R. Define a subset U
of X /R to be open if f~~L(U) is open in X. The topology thus obtained
on X /R (Proposition 16) is called the identification topology on X/R.
(Some topologists refer to this topology as the quotient topology, and
of X /R as a quotient space.)

Proposition 16. The collection of open sets of X /R actually forms a

topology for X/72, that is, the identification topology is really a
topology.

Proof. We verify that the collection of open sets in X/R satisfies Defini­
tion 1, Chapter 3.
i) /-1(</>) = 0 and f ~ 1(X/R) = X. Since X and 0 are both open
subsets of X , 0 and X /R are open subsets of X/R.
ii) Let U and V be open subsets of X/R. Then f~ 1(U) and/-1(F)
are open subsets of X. Now f ~ l {U) C\f~l (V) is also an open
subset of X. But

r \ U ) n r \ V ) = r 1(UnV).

Therefore U r I F is also an open subset of X/R.

iii) Suppose {C7t-}, i G 7, is a family of open subsets of X/R. Then
is open in X for each i G 7, and thus Ui f ~ x{Ui) is an
open subset of X. But since

u r\ui)= r1(u uty

U / Ui is an open subset of X/R. Therefore the collection of open
subsets of X/R forms a topology on X/R.
Proposition 17. Let X, r be a topological space, let 72be an equivalence
relation on X, and suppose that X /R has the identification topology
t '. Then the identification map

/: X, t > X/72, r'

is continuous. Furthermore, r' is the finest topology on X/72 for
which / is continuous.
4.5 Identification Spaces 81

Proof. By definition of the identification topology, f ~ l (U) is an open

subset of X whenever U is an open subset of X/R. The identification
topology has been specifically defined so as to make / continuous. Suppose
r " is a topology on X /R which is strictly finer than r'. Then there is
U G t " such that U & r'. Then is not an open subset of X (if it
were, U would also be in r'); hence / is not a continuous function from X , r
onto X/R, r".
The identification topology is so called because it may be viewed in
the following manner: Let X, r be a space, and let R be an equivalence
relation on X. Then we obtain X/R by identifying all ^-equivalent
elements of X with one another, that is, we make an equivalence class a
point of a new set. We put a topology on X /R by defining a subset U
of X/R to be open if all elements of X contained in all the members of
U form an open subset of X.

Example 14. Let the closed interval [0, 1] have

the usual (absolute value) topology. An equiva­ 0 1
lence relation on any set can be specified either
by giving the equivalence classes, that is, a par­ {0,1}
tition of the set, or by defining the relation. For
[0,1], we will let 0 be equivalent to 1, and every
other element of [0, 1] be equivalent only to it­
self. Then the equivalence classes are {0,1} and
{x}, for x 9^ 0, 1. By identifying 0 and 1, we
obtain a circle (Fig. 4.3). We have joined the
endpoints of [0, 1] by making the endpoints a
single point of a new topological space, which
is a simple closed curve. We therefore have a
continuous function from [0, 1] onto a circle.

Example 15. Suppose X is any set and F, r' is a topological space. Let
g be a function from X to F. Suppose we want to find a topology on X
which will make g continuous. Of course, the discrete topology will do,
but as with the trivial topology, this possibility is not of much interest.
Since g will be continuous if and only if given any open set U of F, g~l (U)
is open in X, we will define a subset V of X to be open if there is an open
subset U of F such that V = g~l (U). Then the family of open subsets
of X forms a topology r o n I ; moreover, this topology is the coarsest
topology which makes g continuous (Section 4.3, Exercise 4).
Define an equivalence relation R on X by letting x R x' if g(x) = g(x')
for any x and x' in X. If x is the ^-equivalence class of x, then there is
a natural function from X /R into F defined by 7}(x) = g(x). The function
g is well-defined, for if x = x', then
g(x) = g (x)= g(x') = g
82 Derived Topological Spaces 4.5

Let / be the identification mapping from X to X/R. Then if g is onto and

X/R is given the identification topology r", g is a homeomorphism from
X /R , t " onto F, r'. This is proved as follows: Since g is onto, g is onto.
Suppose gixi) = g(x2). Then g(xi) = g(x2). Therefore xi = x2, and
hence g is one-one.
It remains to show that g and g~ l are continuous. Suppose U is any
open subset of Y. Then

r\g-\U)) =

which by definition is an open subset of X. Since Z-1 (g~l (U)) is an open

subset of X , and since X/R has the identification topology, g ~ l (U) is
open in X/R. Therefore g is continuous. Assume that V is any open sub­
set of X/R. Then f ~ l (V) is an open .subset of X. Hence

f ~ W ) = g~\U),

where U is some open subset of F, and then

g(V) = g{g~\U = U

is an open subset of Y. Since g = (^“1)“1, (g ~ l)~l (V) is open in F;

hence g ~ l is continuous. Therefore g is a homeomorphism.

If the reader is familiar with some group theory, he may find it in­
structive to recall the relationship between quotient groups and homo-
morphisms. If G and G' are groups and h is a homomorphism from G
onto (?', then G' is isomorphic to the quotient group G/K, where K is the
kernel of h. The quotient group G/K is nothing but the set of equivalence
classes of the relation R, defined by g R g' if h(g) = h(g'). There is a
function h from the set of R -equivalence classes G/K onto G\ defined by
h(g) = h(g) where g is the equivalence class of g G G. If h is onto, then h
is one-one and onto. An operation is then defined on G/K by means of
the operation on G such that h and h~l are homomorphisms; hence h is
an isomorphism.
This procedure is quite important in mathematics. That is, starting
with a function g onto a structured set F from an unstructured set X, we
might wish to find a structure on X so that g becomes a structure-preserving
function with the structure on X derived from the structure on F in a
natural way. It might be that X already has a structure and that g is
structure preserving, hence making it unnecessary to define another
structure on X. This is the case for example, if X and F are groups and g
is a homomorphism. In any event, taking the equivalance classes deter­
mined by g [that is, x is equivalent to x' if g(x) = g(x')], we have a func­
tion g from the set of equivalence classes onto F which is one-one. We
also have the identification mapping from X onto the set of equivalence
4.5 Identification Spaces 83

classes. We then find a structure on the set of equivalence classes so that

both the identification mapping, g, and g~ l are structure preserving.
Note the close parallel in this respect between quotient groups and homo-
morphisms in group theory, and identification spaces and continuous
functions in topology (also see Exercise 2).

EXERCISES

1. Verify that the identification space obtained from [0, 1] in Example 14 is

really homeomorphic to a circle. An actual homeomorphism might be given
by “ wrapping”[0, 1] around a circle in R2 of radius 1/27r.
2. A fundamental theorem of group homomorphisms states: There is a homo­
morphism from the group G onto the group G' if and only if there is a normal
subgroup K of G such that G' is isomorphic to the quotient group G/K.
Provide an example to show that the following analogous statement about
topological spaces is not true: There is a continuous function g from the
space X, t onto the space F, r if and only if there is an equivalence relation
R on X such that the identification space X/R is homeomorphic to F, r'.
Explain why this statement fails to be true. Is it true if the phrase “if and”
is omitted? Is it true if the phrase “
and only if”is omitted?

Fig. 4.4 {{a; |x EABUCD}} U Fig. 4.5 {P = {w |w is diagonally opposite

{{a; |x = x} |x & AB U CD}} P,otw = P} , if P is on the circumference} U
{ {P |P = P } ,if P is not on the circumference}

Fig. 4.6 {{(a;, y) \x = y + m, Fig. 4.7 {{z |z is on the perimeter of C}} U

where m is an integer}} {{z |z = z} |z is not on the perimeter of C}}
3. Each of Figs. 4.4 through 4.7 is to be considered as a subspace of the plane
R2 with the usual Pythagorean metric topology. Under each figure is given
a partition of the set which the figure represents. Draw a picture of the
identification space determined by each partition.
84 Derived Topological Spaces 4.6

4. Let g be a function from a space X, r onto a set 7. Define a subset U of Y to

be open if g~l (U) is an open subset o l x ? It was shown in Section 4.3, Exer­
cise 3 that the open subsets of Y then form a topology r' on Y and that
g : X, t —+ Y, t ' is continuous. Prove that Y, r' is homeomorphic to the
identification space X/R, where R is the equivalence relation associated with
the function g.
5. Find a quotient space of R2 homeomorphic to each of the following.
a) a rectangle with its interior
b) a sphere
c) a straight line

4.6 PRODUCT SPACES

The reader has undoubtedly encountered the concept of the Cartesian

product of finitely many sets in previous studies. If Si, S 2f . . . , Sn are
sets, then the Cartesian product of these sets X"=i Si is defined by
n
X S{ = {($1, $2, •••j ® n) | E: S i, i — 1, . . . , n}.
i= 1
That is, the Cartesian product, or simply the product, of the Si is the set
of ordered n-tuples of elements of the Si. The coordinate plane is nothing
but the Cartesian product R X R, where R is the set of real numbers.
The ith place in an ordered n-tuple is usually called the it\i coordinate.
If, however, the reader has already adjusted to the n-tuple definition
of the product of n sets, then he may find it somewhat hard to begin the
study of product topological spaces by having to learn a new and more
general definition of the product of a family of sets—one which allows us
to take the product of infinitely many sets as well as finitely many. In
the body of this text, we will extend the definition of the product so that
we can deal with the product of countably many sets. The Appendix
gives the definition and some properties of more general products. Wher­
ever possible, proofs about product spaces will be given in a form which
easily adapts to the more general definition of a product. It should be
kept in mind, however, that not all statements about finite or countable
products are true when applied to the product of an arbitrary family of
sets or topological spaces.

Definition 6. Let {Si}, i E I, be a family of sets, where I is a countable

index set (either finite or infinite). We may then choose I either to be
{1, 2, . . . , n}, where n is an appropriate positive integer, or to be the
set of positive integers. By X/ Si we will mean the set of ordered
tuples of the form ($i, s2) . . . , Si, . . .), where Si e Si for each i E I-
If
$= ($i, . . . , Si, . . .) E X Si)
86 Derived Topological Spaces 4.6

is continuous. Moreover, r is the coarsest topology on X/ Si for which

each pi is continuous.
Proof. The product topology has been specifically defined so as to make
each projection continuous. If any topology on Xj Si were strictly coarser
than the product topology, then some member of S, the subbasis for r,
would not be open; hence at least one of the projections could not be
continuous.

Example 17. It is not true that the product space of countably many
discrete spaces necessarily has the discrete topology, although the product
topology will be discrete if only finitely many spaces are involved. Let I
be the set of positive integers. For each i e /, let Si — {1,2} with the
discrete topology. Set U — X/ where = {1} for each i e I.
Then U is the product of open subsets of the Si (specifically, U =
{(1, 1, . . . , 1, . . .)}), but U is not open. The proof that U is not open is
left as an exercise. It is, however, a rather easy corollary of Proposition 19.

Example 18. Let R be the space of real numbers with the absolute value
topology. We will show that the product topology on the plane R 2 =
R X R is the same as the topology on R 2 induced by the Pythagorean
metric D. The topology induced by D is the same as the topology induced
by the metric D s of Chapter 2, Example 3 (Section 3.2, Exercise 2). A
typical subbasis element for the product topology on R 2 is shown in
Fig. 4.8. This means that a typical basis element for the product topology
is given by Fig. 4.9 (see Proposition 19). Each element of the basis for the
product topology is Z)3-open, and each Z)3-p-neighborhood of any point
of R 2 is exactly a basis element of the product topology. We can easily
verify that a basis for the product topology is also a basis for the topology
induced by Z)3; hence the topologies are the same.
 4.6 Product Spaces 87

Note that not every open set of the product topology is of the form
U X F, where U and F are open subsets of R. For example,

{(*, y)Iz 2 + y2 < 1

}

is open, but is not a product set.

Proposition 19. Let Xj Si, r be the product space of the countable

family of spaces {Si, r*}, i e 7. Set

(B = {X/ Vi |F* is open in Si, and Vi = Si

for all but at most finitely many i] •
Then (B is a basis for r.
Proof. Let S be the subbasis for r described in Definition 7. Then a
basis for r is obtained by taking all finite intersections of members of S
(Chapter 3, Definition 4). Suppose f7x, . . . , Un are elements of S, with
Uk — X/ IF*, where IF* = S* for all i, except possibly i = 4, /c = 1, . . . ,n.
Then
E7i n C/2 n •••n = X F*,
i
where F* = /S*, except possibly for i lf i 2) . . . , 4- This means that each
element of the basis derived from S is also a member of (B; but each mem­
ber of (B is the intersection of finitely many members of S. Therefore (B is
a basis for r.
Corollary. If 7 is finite, then a basis for r consists of all sets of the
form Xj F*, where F* is open in S*.
Proposition 20. Suppose X/ >S*, r is the product space of the nonempty
spaces {Si, r*}, i e 7. Then £*, r* is homeomorphic to a subspace of
X/ aS*, r for each i e 7.
Proof. We lose no generality in proving this proposition for $i, T\, since
the same proof could be used for any i e 7. Let 1/2, . . . , yi, ... be fixed
points of $2, . . . , Si, . . . , respectively. Define the function from S 1
into Xj aS* by
= (*, 2/2, , Vi, •••
)

for each # E aSx. Let F be the subspace of X/ aS*, defined by

Y = {(x, y2} . . . , 2/<, . . ,) | x G aSx}.

Then g x takes aSx onto F ; moreover, gx is one-one. It remains to show that

qi and qj”1 are continuous.
Assume U an open subset of F. Then [/ = F n C/', where C/' is an
open subset of Xj aS*. If p x is the projection into the first component, then
 Derived Topological Spaces 4.6

Pi(U') is an open subset of S i (Exercise 2). But it is readily seen that

qTl (U) = Pi(U'). Therefore qTl (U) is an open subset of Si, and thus qi
is continuous. On the other hand, suppose V is an open subset of Si.
Then
91(F) = (gr1)-1^ ) = Y n X W i ,

where W{ = Si, i > 2, and W\ = V. It follows that (gr1)""1(F) is an

open subset of Y ; hence qT1 is continuous. Therefore qi is a homeo-
morphism.
Example 19. Let R be the set of real numbers with the absolute value
topology, and let R 2 be the plane with the product topology. Figure 4.10
illustrates one possible embedding of R as a subspace of R 2. One generally
thinks of the x-axis as being the real line, whereas, strictly speaking, it
is a space which is homeomorphic to the real line. Note that even when
restricting oneself to the procedure of Proposition 20, there are uncountably
many subspaces of R 2 homeomorphic to R.

y
1
1
1
1
I *->(*»0)
------------ |------ » ----- *
I
I
I
! Figure 4.10

Proposition 21. Suppose / is a function from a space X, r into the

product space Xj Si, r'. Define /;: X, r —> Si, Ti by /;(x) = Pi°f(x)
for each x g I , where pi is the projection into the ith component.
Then / is continuous if and only if /* is continuous for each i e
Proof. If / is continuous, then /*•= pi°f is the composition of two con­
tinuous functions and therefore is continuous.
Assume now that/; is continuous for each i e /. We will use Proposi­
tions 7 and 19. We first note that

f(x) = (fl(x),f2
(x), . . . Ji(x),

for each x e X. Suppose X/ V i is any member of the basis (B for r' de­
scribed in Proposition 19, where Vi = Si for each i E I, except i\, . . . , im.
Now f ~ l (X/ Vf) is the set of all points x of X such that f(x) e X/ Vi.
But this is easily seen to be f lif^ iV i) . For every i, except i\, . . . , im,

ST1{Vi) = X (because Vi = Si).

Since/; is continuous for each i E 7,/~1(Fij.) is open in X for/ = 1, . . . ,m.

4.6 Product Spaces 89

Therefore
r 1 ( X F s) = / ^ ( F * ) n ••• n frm\ v im),

which is open in X since it is the intersection of finitely many open sets.

Hence, by Proposition 7, / is continuous.
Proposition 21 is extremely important in the study of product spaces.
Note that its proof would not have gone through if we had defined a sub­
basis of the product topology to consist of sets of the form X/ Vi, where
Vi is open in Si, since we could not have been sure then that f l j f r W i)
was an open subset of X. (Note, however, with this topology that each
projection is still continuous.) Proposition 21 is, in fact, another good
reason why the product topology was defined as it was.
Example 20. Let R be the space of real numbers with the absolute value
topology, and let R 2 be the plane with product topology from R. Define
f : R - > R 2 by
fix) = (sin x, 3x + 1)

for each x e R. Then f\(x) = sin x and /2(x) = 3x + 1 for each x e R.

Since/i and/2are both continuous functions from R into R,/is continuous.

EXERCISES
1. Suppose that {Si}, i E I, is any countable family of sets, and that Si = </>,
for some i. Prove Xr Si = </>.
2. A function / from a space X, t to a space Y, r is said to be open if /(F ) is
open in Y whenever V is an open subset of X. Prove that the projection pi
from the product space Xr Si, r into Si, Ti is open for each i G I.
3. Prove that the product space of a countable family of spaces, each with the
trivial topology, has the trivial topology.
4. Each of the sets involved in the following is to be considered as a subspace
of the space R of real numbers with the absolute value topology. Sketch each
of the following spaces.
a) [0, 1] X [0, 1] b) (0, 1} X [0, 1] c) (0, 1} X R
d) {x |x > 0} X {x |x is an integer greater than 1}
5. Let C = {(x, y) \x2+ y2 = 1} C R2 with the Pythagorean topology.
Describe
a) C X C ; b) C X [0,1]; c) C X (0,1).
6. Prove that the set U in Example 17 is not an element of the product topology.
7. Let X, t be any space, and let X X X have the product topology. The
diagonal A of X X X is defined by A = {(x, x) \x G X}. Prove that X is
homeomorphic to A.
8. Let {Si, Ti}, i G I, be a countable family of spaces, and let P be a permuta­
tion of I. Prove that the product spaces Xi Si and Xi Spa) are homeomorphic.
90 Derived Topological Spaces 4.6

That is, the order in which the components are used in the product does not
affect the topological character of the product space.
9. Consider the space N described in Exercise 6 of Section 3.1 and Exercise 5
of Section 4.4. Prove or disprove: The product space N X N is homeomorphic
to N.
 5
THE SEPARATION AXIOMS

5.1 J0- AND Ji-SPACES

Propositions 12 and 13 of Chapter 2, and Section 2.3, Exercise 1 furnish
us with examples of “ separation”properties for metric spaces. A “separa­
tion”property really does imply separation in the following sense: Given
any two nonintersecting subsets A and B of a topological space X, where
A and B are subsets of a certain type, there are other nonintersecting
subsets U and V of X, generally open sets, such that A C U and B C V
(Fig. 5.1). In other words, being separated in a topological space is a bit
stronger than merely being disjoint. There are various degrees of separa­
tion. As we saw in Chapter 2, it is possible to separate disjoint closed
subsets of a metric space in a rather strong way. But not all topological
spaces are metric spaces; hence not all topological spaces have strong
separation properties. Although most important topological spaces are at
least T 2 (see below), many are not.

Figure 5.1

Many topologists will not even consider a topological space which is

not T 2, and some won’ t touch anything which is not at least normal. But
since it is hoped that the reader has not yet developed personal prejudices,
at least in the area of topology, and since, too, the reader has not yet
begun to specialize to the point where he feels justified in throwing out
^ whatever does not fall in his sphere of interest, we shall even study some
of the lesser separation axioms. The first separation axiom follows.
Definition 1. A topological space X, r is said to be T0 if given any
two distinct points x and y of X, there is a neighborhood of at least
one which does not contain the other.
91
5.1 To- and T\-Spaces 93

space each one-point subset is closed, whereas this is not necessarily true
in a pseudometric space. In fact, it is definitely false with a pseudometric
space unless the pseudometric is a metric. For if x and y are distinct points
such that D(x, y) = 0, then every open set which contains x contains y,
and every open set which contains y contains x ; hence x and y cannot be
separated. Therefore {x, y) is a subset of Cl{x} and Cl{y}, and it follows
that Cl{x} 9^ {x} and C l{y} ^ {y}. A pseudometric space, then, is
generally not even T 0.
A slightly stronger separation property than T0 is given by the
following.
Definition 2. A space X , r is said to be Ti if for any two distinct
points x and y of X, there is a neighborhood of x which does not con­
tain y and a neighborhood of y which does not contain x.
Proposition 2. A space X, r is T x if and only if for each x E X,

Cl{x} - {x}.

Proof. Suppose X is TV and suppose there is x e X such that z E Cl{x},

z 9^ x. Then every neighborhood of z must contain x. Hence there is no
neighborhood of z which excludes x, a contradiction since X is assumed to
be TV Therefore Cl{x} = {x} for all x E X.
Now assume that Cl{x} = {x} for each x E X and that y and z are
distinct points of X. If every neighborhood of ^contains then

* e Cl{y} = {y};

hence y = z, a contradiction. Therefore there is a neighborhood of y

which excludes z. Similarly, there is a neighborhood of z which does not
contain y; hence X is TV

Corollary. A space X, r is T\ if and only if every one-point subset of

X is closed.

The proof is left as an exercise.

Example 3. Every metric space is TV since every one-point subset of a

metric space is closed (Chapter 2, Proposition 5).

Example 4. Let N be the set of positive integers with the topology defined
by making every finite subset of N closed (Section 3.1, Exercise 6). Then
every one-element subset of N is closed; hence N is a TVspace. Suppose
x and y are two distinct positive integers. Then any neighborhood of
either x or y contains all but finitely many positive integers. Therefore,
while it is possible to find a neighborhood of x which excludes y and a
94 The Separation Axioms 5.1

neighborhood of y which excludes x, it is not possible to find a neighborhood

U of x and a neighborhood V of y such that U fl V = </>.
It should be clear from the definitions that any TVspace is also a
TVspace. In Example 1 the reader can find an example of a space which
is To but is not T x.

EXERCISES

1. Prove the corollaries to Propositions 1 and 2.

2. Suppose X is any finite set. Prove that the only topology on X which makes
X into a TVspace is the discrete topology. If X is a set of n elements, what
is the fewest number of members a topology can have which makes X into
a To-space?
3. Let X be any set and D be a pseudometric on X. Define a relation R on X
by x R y if D(x, y) = 0, for any x,y G X.
a) Prove that R is an equivalence relation on X.
b) Let X have the topology induced by D (defining open sets as if D were
a metric). For each x E X, let x denote the ^-equivalence class of x.
Define D(x, y) = D(x, y), for any x, y E X.
i) Prove that D is a metric for X/R.
ii) Prove that the topology induced on X/R, considered merely as the
set of equivalence classes, is the same as the identification topology
on X/R.
iii) Find a natural one-one correspondence between the open sets of X, D
and the open sets of X/R, D.
c) Suppose X, t is any topological space. Find an equivalence relation R on
X such that the identification space X/R is To and there is a natural
one-one correspondence between the open sets of X and the open sets
of X/R.
4. Let X = (1, 2, 3}. Find all topologies on X which are either To or T\.
5. Suppose a space X with topology r is To or T\. Prove that if r' is any topol­
ogy on X which is finer than r, then the space X, r' is also To or T\.
6. a) Prove that every subspace of a TVspace is T\.
b) Prove that every subspace of a To-space is To-
c) Prove that the product space of a countable family of nonempty To-
spaces is To if and only if each component space is To. Prove the cor­
responding statement for Ti-spaces.
7. Prove that if a space X is homeomorphic to a space Y and X is To(T\), then
Y is also.
8. Prove that a space X, r is To if and only if distinct one-point subsets of X
have distinct closures.
5.2 72-Spaces 95

9. Prove or disprove: A space X is To if and only if every proper subspace of

X is To. Does this statement become true if X contains more than two
points? Prove or disprove the corresponding statement with T\ substituted
for To and with the added assumption that X contains at least three points.
10. Prove or disprove: A space X is T\ if and only if {x}' = <t> for each x G X.

5.2 TVSPACES
A still stronger separation property than being either T0 or T\ is the
following.
Definition 3. A space X, r is said to be T 2 if given any two distinct
points x and y of X, there are open sets U and V such that x E U,
y e Vj and U n V = <t>. A TVspace is often called a Hausdorff space.
Example 5. Every metric space is a T2-space (Section 2.3, Exercise 1).
Example 6. Let X be any set which is totally ordered by a relation < .
Let S be the family of all subsets of X of the form {x\x < a} or {x \a < x},
for all a G X. Then S is a subbasis for a topology on X called the order
topology induced by < . The set X with the order topology is always 772.
For suppose a and b are distinct points of X. Since X is totally ordered,
we may assume a < b. If there is c E X such that a < c < 6, then

{x |x < c} and {y |c < y}

are disjoint neighborhoods of a and b, respectively. If there is no c G X

such that a < c < b, then
{x |x < b} and {y |a < y}

are disjoint neighborhoods of a and 6, respectively.

Note that for the set R of real numbers, the order topology (for which
the family of open intervals forms a basis) and the absolute value topology
are the same.
Proposition 3
a) Any subspace of a 7T2-space is 772.
b) Let Y = X/ Xi, t be the product space of the countable family of
nonempty spaces (X»-, r t}, i e /. Then Y is T2 if and only if each
X i is T2.
Proof
a) Suppose IF is a subspace of the 772-space X, and let x and y be
distinct points of W. Then there are open sets U and V in X such
96 The Separation Axioms 5.2

that x e U, y G V, and U n V = <t>. But x e U nW , y e V nW ,

and
(U n W) n (V n W) = (U n V) n w = * n if = 4 .
Therefore U C\W and F fi W are disjoint neighborhoods in W of x
and y, respectively. Hence IF is T2.
b) Assume that each X\, r t-is T 2, and let x and 2/ be distinct points of
F. We will use £*•and y* to denote the ith coordinate of x and y,
respectively. Since x 9* y9x{ 9^ yi for at least one i G /, say for i'.
Therefore there are open sets Ui> and Vv in Xv such that

Xi> e U i r , y {r G V {’, an d Uv fl V v = 0.

Set U = X/ Hi, where Hi — Xi, i if, and Hi> = ; and set

V = X/ Gi, where Gi = i ^ F, and = F*v. Then C7 and
V are neighborhoods of 3 and y, respectively. Since any point of
U differs from any point of V at least in the i'th coordinate,
U fl V = <t>. Therefore Y is T2.
By Proposition 20, Chapter 4, each Xi, Ti is homeomorphic to a sub­
space of Y (regardless of whether F is T 2). If F is T2, then every subspace
of F is T 2, by (a). Therefore if F is T2, each X i, T i is homeomorphic to a
T 2-space, and hence is T2 (Exercise 1).
Example 7. The reader might conjecture from Proposition 3 that if X, r
is a T 2-space and if R is an equivalence relation on X, then the identifica­
tion space X/R is also T 2. This is not true. For example, let R 2 be the
plane with the Pythagorean topology. Let E be the equivalence relation
defined by the partition

{{(*, y)I y o<}, {(x, I 0}

Here {(x, y) \y < 0) is open, whereas {{x, y) |y > 0} is not open. There­
fore R 2/E is homeomorphic to the set X = {0, 1} with the topology
{X , <f>, {0}}, which is not T2. Since the identification mapping from R 2
onto R 2/E is continuous, we have also shown that the property of being
T 2 is not preserved by continuous functions (although, of course, it is
preserved by homeomorphisms).

EXERCISES

1. Suppose that the space X, r is homeomorphic to the space F, r' and that X is
T2. Prove that F is T2.
2. Prove that a space X, r is T2 if and only if, given any two distinct points
x and y of X, there is a neighborhood U of x such that y & Cl U.
5.3 7*3- and Regular Spaces 97

3. Prove that a space X , r is T2 if and only if the diagonal A = {(x, x) |a; £ J}

is a closed subset of the product space X X X .
4. Suppose that X, r is a TVspace and that A C X. Prove that z E Cl A if and
only if x E A, or each neighborhood of x contains infinitely many points
of A.
5. Assume that / is a function from a set X onto a TVspace F, r'. Assume
further that X is given the topology r, defined by taking a subset U of X to
be open if U = /-1(F), where F is an open subset of F. Is X with this
topology necessarily T2?
6. Suppose / is a function from a TVspace X, r onto a space F, r' such that / is
one-one and f~ l is continuous. Prove that F is a TVspace.
7. Let X be a set partially ordered by <. Define S to be the collection of subsets
of X of the form {x |x < a} or {y |a < y}, for all a G X. Is S necessarily
the subbasis for a topology on X? If it is a subbasis for a topology r on X,
is X, r a TVspace?
8. Suppose X is a TVspace. Is it necessarily true that given any subbasis S for
* the topology on X and any two distinct points x and y of X, there are disjoint
members U and F of S with x E U and y E F? Answer this question with
subbasis replaced by basis.

5.3 r 3- AND REGULAR SPACES

Definition 4. A space X, r is said to be !T3 if given any closed subset F
of X and any point x of X which is not in F, there are open sets U and
V such that x G U, F C F, and U D F = <f> (Fig. 5.2). A space X, r
is said to be regular if X is both T% and T x. (The author is quite aware
of the lack of uniformity in the literature about what constitutes a
7Y or a regular space. In some places, T 3 and regular are synonymous.
In others, T x is assumed as part of Ts, and in still others, T 3 does not
imply TV The author has therefore felt justified in making the
definition to suit himself.)

X-U

We first state and prove a very important criterion for being TV

98 The Separation Axioms 5.3

Proposition 4. A space X , r is T 3 if and only if given any x e X and

any neighborhood U of x, there is a neighborhood F of x such that
Cl F C U (Fig. 5.3).
Proof. Suppose that X is T3 and that t/ is a neighborhood of x. Then
X — U is a closed subset of X which does not contain x. Therefore there
are open sets W and F such that X — U C W, x E V, and W n V — <t>.
Since X — U CW , X — W C U. Moreover, since W n V = <t>, we have
F C X — W C U. But then X — W is a closed set which contains F ;
hence
V c C I V c X — W cU .

Suppose instead that given any x G X and any neighborhood U of x,

there is a neighborhood F of x such that Cl F C U. Let x G X, and let F
be any closed subset of X which does not contain x. Then X — F is a
neighborhood of x; hence there is a neighborhood F of x such that
Cl F C X — F. Then X — Cl F is an open set which contains F, and F
is an open set which contains x. Since F C Cl V,

(X - Cl F) n F = 0.

Therefore X — Cl F and F are suitable open sets for “

separating”F and
x; hence X is T3.

Example 8. Any metric space X, D is regular. Although this has been

proved previously, we may prove it again using Proposition 4. For if
x G X, and if U is any neighborhood of x, then U contains a D-p-neigh­
borhood of x for some positive number p. Choose a number q such that
0 < q < p. Then the D-g-neighborhod of x is a subset of the D-p-neigh-
borhood of x, and

Cl N(x, q) C {y | D(y, x) < q} C N(x, p) = {w | D(w, x) < p} C U.

Therefore X , D is 7Y We have already seen that any metric space is T x;

hence any metric space is regular.

Example 9. If X is any set of more than one point, then if X is given the
trivial topology, X is T3 in a vacuous sort of way. For the only closed
nonempty subset of X is X itself, and it follows that there is no point of
X in X — X. It is to avoid such cases as this that one usually requires
a space to be regular rather than merely T 3. We also see from this example
that T3 does not imply T2. However, if every one-point subset of A is a
closed subset of X, then X is T2 if X is T3.

We now give an example of a space which is T 2 but not T3, or regular.

This demonstrates that regularity is a stronger separation property than
merely being T 2.
5.3 73 - and Regular Spaces 99

Example 10. Let R be the set of real numbers. We will define a topology
on R by giving an open neighborhood system. If x is any real number
other than 0, let 3lx be the family of all open intervals which contain x.
If x — 0, we will let 9l0 be the family of all sets of the form

(—p, P) — {1/n | is a positive integer},

where 0 < p. The collection of 9lx for all x e R gives an open neighbor­
hood system for a topology r on R (Chapter 3, Proposition 7). It is easily
verified that R, r is T2 (Exercise 1).
We now show that X is not T 3. Take x = 0 and F = {1/n \n is a
positive integer}. F is a closed subset of R, r, since there is no point y of
R such that each neighborhood of y contains a point of F , except those
points in F itself. (Note that in the usual topology for R , each neighbor­
hood of 0 would contain points of F; thus 0 would be in Cl F. We have,
however, purposely excluded the points of F from the neighborhoods of 0.)
Suppose V is a neighborhood of 0 of the form

(—p, P) ~~ (1/tt |n is a positive integer}

(any neighborhood of 0 contains such a neighborhood because of the

manner in which the 3lx define the topology r). Then (—p, p) contains
infinitely many of the 1/n. Hence any open set U which contains F would
have to overlap V (Fig. 5.4); thus we could not find an open set U which
contains F such that XJ D V = <£. Therefore R, r is not T 3.

V U

Figure 5.4

We now investigate how the property of being regular or T3 behaves

with respect to the derived topological spaces.

Proposition 5
a) Every subspace of a regular space is regular.
b) Suppose Y = Xj X i is the product space of the (countable) family
of nonempty spaces {Xi, r»}, i e I. Then Y is regular if and only
if each Xi, n is a regular space.
Proof
a) Suppose W is a subspace of a regular space X, r. Let F be closed
in W and x G W — F. Since F is closed in W, F = W fl F', where
F' is a closed subset of X. Then x e I - F'. *Since X is T3,
100 The Separation Axioms 5.3

there are open sets U and V in X such that

xeU, F' C V, and U n V = </>.

Then W n U and W n V are disjoint subsets of W which are open

in W such that x E W n U and F C W n V. W is therefore T%.
By Section 5.1, Exercise 6, W is also T i, since X is T x. Therefore
W is regular.
b) Suppose each X*, n is a regular space. By Exercise 6 of Section 5.1,
F is T\. Suppose that x e Y and that U is a neighborhood of x.
We lose no generality in assuming that U is a basis element for the
product topology, since any neighborhood of x contains a neigh­
borhood of x which is a basis element. Then U = Xj Ui, where
each Ui is open in X{. Each Ui is therefore a neighborhood in X i
of x^ the.ith coordinate of x and Ui — X i except for i l9 ..., im.
Since each X i is T3, there is an open neighborhood Vi. of X{., j =
1, . . . , m, such that

For each i e 7, set = X{, i ij and let the V%. be as above,

j — 1,..., m. Then

By Proposition 4, then, F is 7Y Hence F is regular.

If F is regular, then since each X i, Ti is homeomorphic to a sub­
space of F, and each subspace of F is regular, each X i is regular.
As with TVspaces, it is not true that if X, r is regular and R is an
equivalence relation on X, then the identification space X/R is regular
(Exercise 2). However, the following is true.
Proposition 6. If X, r is a regular space and F is a closed subset of X,
then if R is the equivalence relation defined by the partition
{{2 I X G ^ } } U {y| y|
the identification space X/R is 772- (Note that this identification has
the effect of shrinking F to a point. See Figs. 5.5 and 5.6.)

Proof. Suppose x and y are distinct points of X /R , where x and y denote

the equivalence classes of x and y, respectively. If x and y are not in F ,
then since X is regular and hence also T2, there are open sets U and V
in X — F such that
x E U} y E V, and U n V = 4>.
5.3 7*3- and Regular Spaces 101

Figure 5.5 Figure 5.6

The sets U and F may be chosen in X — F, since X — F is a TYsubspace

of X, and since a subset of X — F, an open set, is open in X — F if and
only if it is open in X. Then U — {u\u e U} and V = {v |v e F} are
open disjoint subsets of X/R such that x G U and y e F.
If either x G F, or y e F, then the other point could not be in F. For
if x and y are both in F, then x = y = F, contradicting x ^ y. Suppose
x = F. Then there are open subsets U and V of X such that

x = F C U, t/ e F , and J7 n F = </>.

It follows that U = { u \ u ^ U } and F = {y |v E F} are disjoint open

subsets of X/R such that x e C/ and y E. V. Therefore X//? is 7Y

Figure 5.7

Example 11. In Example 14 of Chapter 4, a circle is obtained by identify­

ing 0 and 1 in [0, 1]. Since {0, 1} is a closed subset of [0, 1], we know that
the circle is at least a T 2-space. Since the circle is a subspace of a metric
space R 2, and any metric space is regular, we have the stronger result that
a circle is a regular space. The torus C X C (Fig. 5.7), where C is a circle,
is regular since it is the product of regular spaces. The torus is also seen
to be regular because it is a subspace of R 3.

EXERCISES

1. In regard to Example 10,

a) verify that the family of 3lx forms an open neighborhood system for a
topology on R ;
b) show that R with this topology is T2.
5.4 74 - and Normal Spaces 103

Example 12. Any space X, r of more than one point with the trivial
topology is T4 (there are no nonempty disjoint closed subsets of X), but
is not normal, since no one-point subset of X is closed. Any metric space
is normal (Propositions 5 and 13 of Chapter 2).
Note that any space with the discrete topology has all of the separation
properties introduced so far.
We now prove another criterion for a space to be T4.
Proposition 7. A space X, r is T 4 if and only if given any closed subset
F of X and any open subset U of X with F C U, there is an open set
V such that
F C V C Cl V C U.

(Note the similarity between this proposition and Proposition 4 re­

garding TV spaces. Note also the difference, namely, F is a set rather
than a point. This difference helps us explain the rather bad “ hered­
itary”properties of normal spaces.)

Proof. Suppose X is T 4, F is a closed subset of X, and U is an open set

which contains F. Then X — U is a closed set and (X — U) n F = <£.
Hence there are open sets W and V such that

X- U C W, FCV, and W nV= <t>.

Then F C V c C I V c U (since X - U C W and W n V = 0). There­

fore V has the desired properties.
Suppose now that given any closed set F and any open set U which
contains F, there is an open set V such that F C V C Cl V C U. Let F
and F' be any two disjoint closed subsets of X. Then X — F is an open
set which contains Ff. By hypothesis, then there is an open set V such that

Ff C V C Cl V C X- F.

Hence X — Cl V is an open set which contains F and is disjoint from V.

Therefore X is 7Y
It would be very nice to have an analog of Propositions 3 and 5 for
normal spaces. Unfortunately, not only is it false that the product of
normal spaces is normal; it is even false that every subspace of a normal
space is normal. Examples of normal spaces for which some subspace is
not normal are somewhat sophisticated for this text. The following ex­
ample, however, gives a regular space which is not normal, but which is
the product of normal spaces.
Example 13. Let R be the set of real numbers with the topology r as
described in Example 11 of Chapter 3. Let R 2 be given the product topol-
104 The Separation Axioms 5.4

Figure 5.9

ogy. Then a typical basic neighborhood of (x, y) E R 2 is as shown in

Fig. 5.9. We know that R, r is normal (Exercise 1). Each basic neigh­
borhood U of (x, y) is not only open, but also closed. For if (x', yr) is any
point of R 2 — U, then it is readily seen that there is a basic neighborhood
of (x', y') contained entirely in R 2 — U; hence R 2 — U is open, and thus
U is closed. Therefore C IU = U. Hence, by Proposition 4, R 2 is jT 3 .
R 2 is T i since each one-point subset of R 2 is closed. Alternately, R, r is
regular; thus R 2 with the product topology is regular by Proposition 5.
Let Y = {(x, y) |y + x = 0}. Then the subspace topology of Y is
discrete, since if (x, y) e Y, there is a basic neighborhood U of (x, y) such
that
U n Y = {(x, y)}

(Fig. 5.10). Since Y is a closed subset of X, each subset of Y is a closed

subset of X (since a subset of Y is closed in Y if and only if it is closed in X ,
but every subset of F is closed in F). Let

F = {(x, y) |x + y = 0 and x is rational}

and
Fr = {(x, y) |x + y = 0 and x is irrational}.

Since F and F' are subsets of F, they are closed. Also, F n F' = 0. If
R 2 is T4, there must then be open sets U and V such that

F CU , F' C V, and U n V = 0.
Although a rigorous argument of the impossibility of such sets will not
be given here, it can be made fairly clear why there cannot be such sets.
For if (x, y) E F, then there would be a basic neighborhood of (x, y) con­
tained in U. There are, however, points of F' “ arbitrarily close”to (x, y).
Hence some basic neighborhood contained in V of a point in F' would be
bound to overlap with the basic neighborhood of (x, y) in U. Therefore
U n V could not be </>.
5.4 Ta- and Normal Spaces 105

Although we do not have that every subset of a normal space is

normal, we do have the weaker statement which follows.

Proposition 8. If F is a closed subset of a normal space X , r then the

subspace F is normal.

Proof. Since X is T x, Y is T x because every subspace of a TVspace is T x.

Since Y is closed, a subset F of Y is closed in Y if and only if F is closed
in X. Therefore if F and F' are disjoint closed subsets of F, they are also
disjoint closed subsets of X. There are thus open sets U and V such that

FCU, F' C Vj and U n V = <t>.

But then
F C F n U, Ff c F n V,
and Y n U and F n V are disjoint subsets of F which are open in F.
Therefore F is T4; hence F is normal.

Proposition 9. If the product space Xj X i of the family of nonempty

spaces {Xi, n }, i G I, is normal, then Xi, Ti is normal for each i e I.
Proof. If Xj X i is normal, then X/ X i is T x. But then each Xi, Ti is
homeomorphic to a closed subspace of X/ X i (Exercise 6). Such a subspace
then is normal by Proposition 8. Hence Xi, Ti is normal (Exercise 5).

EXERCISES

1. Prove that the set R of real numbers with the topology described in
Example 11 of Chapter 3 is normal.
2. Prove that a space X , r is T4 if and only if given any two disjoint closed sub­
sets F and Fr of X, there are open sets U and V such that

FCU, F'CV, and C\ U fl Cl V = </>.

3. Suppose X, t is a normal space and F is a closed subset of X. Let X/F be
the identification space formed by identifying all the points of F with one
another; more figuratively, X/F is the identification space obtained by
squashing F to a point (as in Proposition 6). Prove that X/F is normal.
4. Prove that every subspace of a metric space is normal. Thus if we were to
find a normal space which had a subspace which was not normal, we would
know such a space was not a metric space.
5. Prove that the property of being normal is preserved by homeomorphisms
but not by continuous functions.
6. We saw in Proposition 20, Chapter 4, that if Xi Xi is the product space of
the family of nonempty spaces {.Xi, r*}, i E I, then each Xi, Ti is homeo-
106 The Separation Axioms 5.5

morphic to a subspace of Xr X*. Prove that if X/ X* is T\, each X*, rz*is homeo-
morphic to a closed subspace of Xi Xi.
7. Are any of the spaces given in Exercise 6, Section 5.3, normal besides that
given in (c) ?
8. Review the proof of Proposition 5. Discuss why an analogous proof will not
hold for normal spaces. That is, try to find out what goes wrong in attempting
to apply to normality the techniques which gave us the “ hereditary”prop­
erties of the other separation axioms.
9. Suppose X , r and 7, r' are normal and / is a continuous function from a
subspace A of X into 7. Let Z = X U 7 have the topology r" defined by
using r U r' as a basis. Prove that the identification space formed from Z
using/, that is, by setting x equivalent to/(x), is normal.

5.5 NORMALITY AND THE EXTENSION OF FUNCTIONS

One of the central and most difficult questions in all of topology is that of
function extensions. Specifically, suppose that 7 is a subspace of a space
X, t and that / is a continuous function from 7 into some space Z , r'.
Does there exist a function F from X into Z such that F is continuous and
F(y) = f(y) for each y e 7? That is, is there a continuous function

F : X , t - > Z , r'

such that F | 7 = /? The answer is sometimes yes and sometimes no.

For most instances, the answer is not known.
Example 14. If 7 is a subspace of X, r and i is the identity function on 7,
then i is a continuous function from 7 into X. Of course i can be extended
to the identity function I for all of X. In this case, I is an extension of i
since 1 17 = i. This is a rather trivial and therefore uninteresting type
of extension.
A function may have several extensions. For if we let X = {1, 2}
with the discrete topology and 7 = {1}, then I' defined by 7'(1) = 1,
/'(2) = 1, is an extension of i which is different from I.
Example 15. Let X be the closed interval [0, 1] with the usual absolute
value topology, and let 7 = Z = {0, 1} with the subspace topology
(which is the discrete topology in this case). Let i be the identity function
on 7. Then i is continuous as a function from 7 to Z. Although we can­
not prove it at this time (we will be able to do so later in the book), i can­
not be extended to a continuous function from [0, 1] into Z. We may see
this informally as follows: If there were a continuous function F from X
onto Z (as there would have to be if i could be extended), then, since {0}
and {1} are both open subsets of X, F ~ l {{0}) and F _1({1}) would both
5.5 The Extension of Functions 107

be open subsets of X; moreover,

X = F - \ { 0}) U ^ ( { l } ) and F ~ l ({0}) fl / ^ ( { l} ) = <j>.

Thus X = [0, 1] would be expressible as the union of two disjoint, non­

empty, open subsets. The reader should try to express [0, 1] as the union
of two such subsets in order to see the intuitive difficulties of such a de­
composition.
Topological spaces which are T4 are, however, bound up essentially
with some very important extension properties. In fact, T4-spaces can
be characterized by certain of their extension properties. This is proved
in the following proposition, one of the most important propositions in
topology.
Proposition 10 {Urysolin's lemma). A topological space X, r is T4 if
and only if given any disjoint nonempty closed subsets A and B of X,
there is a continuous function / from X into Z = [0, 1] (with the
absolute value topology) such that

/(a) = 0 for any a E A and f(b) = 1 for any b G B.

Before proving this proposition, we note that it is indeed a proposition

dealing with function extensions. Explicitly, if X, r is a 7Vspace and Y
is a subspace of X which can be expressed as the union of two disjoint
nonempty closed subsets of X, then the function g: Y —> [0, 1] such that

g(a) = 0 for all a E A and g(b) = 1 for all b e B

can be extended to a continuous function /: X —> [0, 1]. Although this

may appear as a rather modest result about function extensions because
of the restrictions that have been placed upon Y and g) very general and
important results flow from Urysohn’ s lemma. We shall mention a few
of these after the proof.
Proof (Proposition 10). Suppose X has the property described and A and
B are any two disjoint nonempty subsets of X. Then there is a continuous
function / from X into Z = [0, 1] such that

/(a) = 0 for all a e A and /(b) = 1 for all b G B.

Now U' = {x |0 < x < 1/2} and Vr = {x | 1/2 < x < 1} are disjoint
open subsets of Z; therefore, since / is continuous, U = f~ l (U') and
V = f ~ l (V') are disjoint open subsets of X. But A C U and B C V, and
hence X is T4.
Suppose now that X is T4. Recall that a space X, r is T4 if and only
if given any closed subset F of X and any open set U which contains F,
108 The Separation Axioms 5.5

there is an open set V which contains F such that

F CV CCIV CU

(Proposition 7). Suppose A and B are disjoint nonempty closed subsets

of X (Fig. 5.11). Consider the set of rational numbers q such that 0 < q < 1
and q is of the form q — n/2k, where
n and k are positive integers. For
example, 1/2, 3/22 = 3/4, and 5/23
= 5/8 are such rational numbers.
With each such rational q we will
associate an open subset U(q) of X
such that
1) A C
2) B n U(q) = <t>;
3) if q < q', then Cl U(q) C U(q').
Since X is T± and A and B are disjoint closed subsets of X , there are
disjoint open sets U and V such that A C U and B CV . We let U — U (0)
and X — B = U( 1). Using Proposition 7, we can find an open set, which
we let be UG), such that

Cl 17(0) C 17(4) C C l 17(4) C 17(1).

Similarly, we can find UG) and £/(f) such that

C117(0) c UG) c C1 UG) C UG)

and
Cl UG) C UG) C C l 17(f) C 17(1).

We will continue finding the U(q) by induction on k, the exponent of 2

in q = n/2k. Note that we have already defined U(q) for k = 1 and
k = 2.
Assume we have defined U(q) for k. We now define U(q) for k + 1
(and thus for n = 1, 3, . . . , 2k+l — 1). Note that the definition of U(q)
needs to be given only for odd n; for if n were even, the numerator and
denominator of q could be divided by 2. Because the U(q) have already
been constructed for q = n/2k, n odd, we have

a u ( ^ ) cV ($ r)

[sincenis,odd,€ i U((n - l)/(2fc+1)) = C l U(((n - l)/2)/2*), whichhas al­

ready been defined]. We therefore can find an open set, which we let be
5.5 The Extension of Functions 109

U(n/2k+l) such that

Cl C7((n - l)/2k+l) C U(n/2k+1) C Cl U(n/2k+l) C U((n + l)/2k+l)

(Fig. 5.12). We thus have an inductive definition of U(q) for each q as

described. By construction, the collection of U(q) have properties (1)
through (3) given above.

Figure 5.12

We now define a function / from X into Z = [0, 1] such that/(a) = 0

for all a E A and f(b) = 1 for all b E B. If x E X } define f(x) = 1, if
x E B. If x is not in B , then x e U(l). For each x not in B, define

f(x) — greatest lower bound {q \q = n/2k and x E U(q)}

(this set of real numbers has a lower bound, 0, and hence has a greatest
lower bound). Certainly 0 < f(x) < 1 . If x E A, then x E !J(0); there­
fore /(0) = 0. We now prove that / is continuous.
Suppose f(x0) = yo•First, assume that y0 is neither 0 nor 1. Then,
given any positive number p, there are rationals q and q' of the form n/2k
such that
Vo e (g, qf) C (2/o “ P, Vo + V)

(that is, the set of “

binary”rationals is dense in [0, 1]. Alternately, any
real number can be approximated to an arbitrary degree of accuracy by
a rational of the form n/2k). Then (Fig. 5.13) V = U(q') — Cl U(q) is a
neighborhood of x0, and

f(V ) C (2/0 — P, V o + P)-

If y0 is either 0 or 1, then the corresponding neighborhoods of 0 and 1,

respectively, are [0, q') and (q, 1]; but the argument is the same. What we
have shown is that, given any neighborhood H of 2/0 (any neighborhood
5.5 The Extension of Functions 111

Proof. Let pi be the projection into the tth coordinate from Rn into R
[defined by pi(xu . . . , £*, . . . , xn) = Xi\. Then, setting fi = p i °/, /*•is a
continuous function from A into R. Each fi therefore has a continuous
extension Fi to all of X. Define

Fix) = Fn{x))

for each x e X. Then F is an extension of /; moreover, F is continuous,

by Proposition 21 of Chapter 4.

Example 16. Let R be the space of real numbers with the absolute value
topology and R 2 be the plane with the product topology. Both of these
spaces are normal. Let Z C R be the set of integers, and let Z 2 C R 2 be
the set of points of R 2 of the form (m, n), where m and n are integers. The
subspace topology on both Z and Z 2 is the discrete topology, and hence
any function/from Z into Z 2 is continuous. Both Z and Z 2 are of the same
cardinality; thus we have a one-one function / from Z onto Z 2, and / is
continuous. By the corollary to Proposition 11, then / has a continuous
extension F from R into R 2. The reader might find from a little experi­
mentation that this is a case where the proof that F exists is much simpler
than trying to construct a specific F.
Example 17. A continuous function from the interval [0, 1] into any
space X, t is called a path in X. The space [0, 1] X [0, 1] (with the product
topology) is a normal space, and

A-= {(z, y)Iy= 1} U

is a closed subset of [0, 1] X [0, 1]. If / is any continuous function from A

into R 2, then / has a continuous extension F (Fig. 5.14). Technically,
this means that any two paths in R 2 are homotopic (Chapter 11).

EXERCISES

1. Using only Proposition 10, prove the following: Suppose A and B are disjoint
closed subsets of a normal space X, r and / is a continuous function from A\J B
112 The Separation Axioms 5.5

into Rn such that / | A and / |B are each constant functions. Then / has a
continuous extension F to all of X.
2. Any space which can be substituted for R in Proposition 11 is called an
absolute retract. Which of the following are definitely absolute retracts?
Which could not possibly be absolute retracts?
a) Rn, with the product topology from the space R of real numbers with the
absolute value topology
b) I n, with the product topology, where I = [0, 1]
c) any finite set with the discrete topology
d) (0, 1) with the usual topology
e) R2 — {(0, 0)} with the Pythagorean topology
3. Which of the following statements about absolute retracts are true?
a) If X, t is an absolute retract and x and y are any points of X, then there
is a continuous function/ from [0, 1] into X such that {x, y} C/([0, 1]).
b) The product space of a countable family of nonempty spaces is an absolute
retract if and only if each component space is an absolute retract.
4. Prove that the set of binary rationals as described in the proof of Proposi­
tion 10 is dense in the space of real numbers.
5. A space X is said to be completely normal (sometimes called T5) if every
subspace of X is normal. Prove that X is completely normal if and only if
X is Ti and given any two subsets A and B of X such that Cl A fl B =
A n Cl B = (j>, there exist disjoint open sets U and V such that A C U
and B C V.
6
CONVERGENCE

6.1 THE NEED FOR A GENERALIZED NOTION OF CONVERGENCE

The reader will recall that we have already discussed convergence of
sequences in metric spaces in Chapter 2. He may therefore suspect that
extending the theory of convergence to general topological spaces will
merely consist of rewording the definition of a convergent sequence in
terms of a general space. For example, we might say: A sequence {sn},
n G N, in a space X , r converges to a limit y in X if every neighborhood
of y contains all but finitely many of the sn.
Actually, however, not only will we find it necessary to generalize the
notion of convergence of a sequence, but we will also have to generalize
the very notion of a sequence as well. The purpose of this section is to
illustrate this point. We begin by proving a proposition concerning se­
quences in metric spaces.
Proposition 1. Let A be a subset of a metric space X, D. Then
y G Cl A if and only if there is a sequence {$„ }, n G AT, such that

sn * y and sn G A
for each n G N.

Proof. Suppose y G Cl A. Then if y G A, let {sn},n G N, be the sequence

defined by sn = y for all n G N. Then

sn —> y and $n G A

for all n G N. Suppose y G Cl A — A. By Proposition 12, Chapter 3,

Cl A = A U A'y therefore y G A'. Then each neighborhood of y con­
tains at least one element of A. Let Un be the D-l/n-neighborhood of y
for each positive integer n. For each n, select sn G Un n A. The sequence
{sn}, n G N, thus obtained converges to y, and each sn is in A.
On the other hand, suppose y is such that there is a sequence {sn},
n G N, such that sn —► y} and sn G A for each n G N. Since sn —> y, each
neighborhood of y contains all but finitely many of the sn. Therefore each
neighborhood of y contains some point of A. By Proposition 13, Chapter 3,
then y G Cl A.
113
 114 Convergence 6.1

If the notion of a sequence were sufficient for the study of general

topological spaces, we would expect that this proposition relating closures
and sequences generalizes to arbitrary topological spaces. That is, if
A C X, where X, t is a topological space, then y e Cl A if and only if, etc.
There are a number of reasons why we would like this proposition to
generalize. The fact is, though, that it does not generalize using sequences.
This is shown by the following example.
Example 1. Let X be the set of all functions from the set R of real numbers
into R. We make no assumption about the continuity of these functions.
We will define a topology on X by specifying an open neighborhood system.
Suppose / is any element of X. Let F be any finite subset of R and p be
any positive real number. Define
U(/, F, p) = {g e X | |g{x) — f{x) | < p for all x G F}.

Let 91/ be the set of all [/(/, F, p) for all finite subsets F of R and all
positive numbers p. Note that for a given/, [/(/, F, p) depends on both F
and p ; hence U(/, F, p) is not a p-neighborhood in the metric sense.
We will now show that this definition of 91/ for each / e X gives us an
open neighborhood system for a topology on X. In accordance with
Proposition 7 of Chapter 3, we must show that (i) through (iv) of Defi­
nition 5, Chapter 3 are satisfied.
Statements (i) and (ii) are clearly satisfied.
iii) Suppose U(f, Fi,px) and [/(/, F 2, p%) are any two members of
91/. Then
F 1U >min(pi,
F2

is an element of 91/ which is contained in U(f,Fu P l ) n

U(f,F2,P 2)- For suppose

g eU(f, F, uf2,min(pi,
we may assume Pl < p 2. it follows that

Ifl'fr) ~ / ( * ) | < Pi < P2

for each x iand each x e F 2. Therefore

eF

9e V(f, p i) n
Fu

iv) Suppose U(f, F, p) G 3ly and e U(f, F, p). Let

F = {*1,
and
Qi=P~ II/O*) - g(Xl)\, i= I,..., n.
6.1 A Generalized Notion of C onvergence 115

Set p' = min^!, . . . , qn). Then

U(g, F, p*)C U(f,F, p)

(Exercise 1). Therefore the 31/ do form an open neighborhood

system for a topology on X.
Let A be the set of all elements / in X such that/(a;) = 0 or 1 for any
x G R, and f(x) = 0 for at most countably many x e R. Suppose g is the
function defined by g{x) = 0 for all x G R. We will show that g e Cl A.
For let U(g, F, p) be any basic neighborhood of g. Let h be the function
defined by h(x) = 0 for x e F and h(x) = 1 if x 2 F. Then

h e A n U(g, F, p).

Therefore any neighborhood of g meets A; hence g e Cl A.

Suppose there is a sequence {/n}, n g N , such that f n E A for each
n E N, and f n —> g. Let Bn — {x |f n(x) = 0}. Each Bn is a countable
subset of R ; hence Un Bn is also a countable subset of R. Since R is un­
countable, we can find z e R — Un Bn. Let F = {z} and p = 1/2.
Consider U(g,F,p). Then no matter what n is, z & Bn, and hence
fn(z) = 1. Therefore
Ifniz) - g(z)| = 1.

There is no positive integer n, then, for which f n G U(g, F }p). But

U(g, Fj p) is a neighborhood of g ; thus if f n g, U(g, F, p) would have
to contain all but finitely many of the/n. It is impossible then that/n —> g.
What we have here, then, is a topological space for which Proposition 1
of this chapter does not hold. What are we to do? There are two alter­
natives, either (1) we can restrict ourselves merely to sequences and say
that Proposition 1 has no generalization, or (2) we can try to generalize
the notion of a sequence in such a way that Proposition 1 can be generalized
to any topological space. It is this latter alternative that we choose.

EXERCISES

1. In Example 1, prove (iv) in the proof that the 91/ form an open neighborhood
system.
2. A topological space X , r is said to be first countable if there is an open neigh­
borhood system for r such that 91* is a countable collection for each x G X.
a) Prove that every metric space is first countable.
b) Prove that Proposition 1 holds for any topological space which is first
countable. Thus the space in Example 1 is not first countable.
c) Prove that any subspace of a first countable space is first countable.
116 Convergence 6.2

d) Prove that the product space of a countable family of nonempty spaces is

first countable if and only if each component space is first countable.
e) Which of the spaces mentioned in Section 5.3, Exercise 6 are first
countable?
3. Do you think that the following might be an appropriate generalization of
sequences? Let I be any set. Then an I-sequence in a space X, r will be a
function s from I into X. We will denote the /-sequence by {s*}, i E /, where
Si denotes s(i). We will say that {$;}, i E /, converges to y if any neighbor­
hood of y contains all but finitely many of the S{. For example, suppose
I = (0, 1) and s is the identity function on I considered as a function from /
into the space of real numbers. Does {si}, i E I, converge to 1 according to
our definition of convergence of a I-sequence? Does it seem as though it
should if this is a suitable generalization of the notion of a sequence?
4. Let X be any uncountable set. For each x E X, define

9l x = {N C X |x E N and N excludes at most countably many points of X}.

Prove that the collection of 9lx forms an open neighborhood system for a
topology on X. Does Proposition 1 apply to X with this topology?
5. Find two distinct topologies on the set R of real numbers such that the only
sequences which converge relative to each of the topologies are those sequences
which are constant from some term on, and these sequences converge only
to their constant value. (Use Exercise 4 and the notion of convergence in­
troduced in the first paragraph of this chapter.) Since the two topologies
have the same convergent sequences converging to the same limits, sequences
alone are inadequate to characterize either topology.

6.2 NETS

Note that the positive integers form a partially ordered set (in this case,
totally ordered) such that if n and n' are any integers, there is an integer m
with n < m and n' < m. Since all partial orderings share the properties
of aless than or equal to,”denoted by < , we will use < to denote any
partial ordering. The set of positive integers do have other properties
which are not shared by every partially ordered set; for example, the
positive integers are totally ordered, well-ordered, and countable. But in
any generalization, experience and the problem to be solved indicates
which properties must be generalized and which are incidental to the
question at hand. The experience and labor of many mathematicians over
many years leads us to the following definition.

Definition 1. Let I be any partially ordered set. (Recall that < is

used to designate any partial ordering.) I is said to be a directed set
(more accurately, an upward directed set) if given any i and any j in /,
118 Convergence 6.2

Y < W(W C Y). If X is finite, then P(X) is finite as well, and hence it
is quite possible to have a finite directed set.
If x E X, set
P(X,x) = { W e P ( X ) \ x e W } .

Then P(X, x) is also a directed set (directed by < just as P{X) is}. One
possible function s from P(X, x) into X would be a selection function
where, if W E P(X, x), then s(TF) E W. Such a selection function would
then give a net in X, W E P(X, x).
If X, t is a topological space, then we might set

T(X, x) = {U | U is a neighborhood of x}.

The set T(X, x) is also a directed set [in fact, a directed subset of P(X, x)].
Using a selection function s: T(X,x) —> X, where s(U) E U for each
U E T(X, x)y we get a net {$£/}, U E T(X, x)9in X. We might suspect that
this net converges to x, since it has the property of being in every neigh­
borhood of x residually (Exercise 2).
Example 4. Let {$n}, n E N, be the sequence in the set R of real numbers
defined by sn = (—l)n. If R is given any topology whatsoever, then {sn},
n E Nj has the property of being in every neighborhood of 1 cofinally.
This sequence is also in every neighborhood of —1 cofinally, but it is
residually in every neighborhood of both 1 and —1 if and only if every
neighborhood of 1 is a neighborhood of —1 and every neighborhood of
—1 is a neighborhood of 1.

EXERCISES

1. Which of the following sets with the orderings as given are directed sets?
a) the set of positive integers partially ordered by “ divides”
b) the interval [0, 1] ordered by <
c) the interval (0, 1) ordered by <
d) the set {1, 2, 3, 4}, where the order is defined by the relation

R = {(1,1), (2,2), (3,3), (4,4), (2,3)}.

2. Prove the assertion in Example 3 that {$£/} has the property of being in any
neighborhood of x residually. Using this example, make a tentative definition
of what we mean by saying that a net converges to an element y of a space X, t.
3. Suppose I and V are sets which are directed by < and <', respectively.
Suppose (i, i') and (j, f) are elements of I X V. Define

«,o < ti, f)

if i ^ j and i' <'/. Prove that I X I' with the relation < as defined is a
6.3 Subsequences and Subnets 119

directed set. Y ou must prove that / X I' is both partially ordered and
directed.
4. Let {sn}, n E N, be a sequence in the set of integers with the property that
if m < n, sm < sn. Which of the follow ing properties m ust such a sequence
have residually? cofinally? neither cofinally nor residually?
a) The property of being odd; b) The property of being positive;
c) The property that sn < n; d) The property that sn is prime.
Suppose {sn}, n E N, is the sequence defined b y sn = 4n + 1. Which of the
properties (a) through (d) does this sequence have residually? cofinally?
5. Prove that a net {$»}, i E I , has a property P cofinally if and only if for any
i E I, there is an element k of I such that i < k and sk has property P.
6. The follow ing define functions from [0, 1] into R 2 and hence define nets in
R 2. If R 2 has its usual topology, indicate any points to which you feel the
nets should converge. Explain inform ally the reasons for your answers in
each case.
a) fix) = Or, 2) b) f(x) = ( l/(x + 1), x 2) c) fix) = (cos x, sin x)

6.3 SUBSEQUENCES AND SUBNETS

Fundamental in any study of either sequences or nets is the concept of a
subsequence, or its generalization, a subnet.

Definition 2. Suppose {$*•}, i E I, is a net in a set X. Let J be a di­

rected set and k a function from J to I such that
i) if i < j', then k(j) < kif);
ii) if i, i' E I, then there is j E J such that i < k(j) and i' < k(j).
That is, k is order-preserving, and considered as a net in I, k is cofinal
in I. Then the composition s°k from J into X is said to be a subnet
of the net {$*}, i E I. The subnet s°k is usually written as {s^y},j G /.

Note that each Skj is also an Si for some i (specifically, for i = kf),
and that the skj have the property of being cofinal (but not necessarily
residual) in the set of S{.

Example 5. Let {sn}, n E N, be any sequence. Let 2N be the set of

positive even integers, and let k be the identity mapping from 2N into N.
We first verify that k has properties (i) and (ii) of Definition 2.
i) If m and n are positive even integers and m < n, then /c(m) = m
and kin) = n.
ii) Given any integer n, there is a positive even integer greater than n ;
therefore there is m E 2N such that

n < k(m) = m.
120 Convergence 6.3

The function s°/c is therefore a subnet o f {$n}, n G N. In the case

where a subnet is a sequence, it is customary to call such a subnet
a subsequence; thus s°fc is a subsequence of {sn}, n E N . Ex­
plicitly, m G 2AT, is the same as {s2n}, n e N . A subse­
quence of a sequence is a sequence in its own right, and a subnet
of a net is itself a net.

Example 6. Let X, r be any topological space, and let x E X. Suppose

P(X, x) is as defined in Example 3. Let s be any selection function from
P(X, x) into X ; that is, s(TT) G W for each W e P(X, x). Then {sw} ,
W G PCX’ , x), is a net in X. Let T(X, x) be (as in Example 3 also) the set
of neighborhoods of x. Then

T(X, x) C P(X, x).

Let k be the identity mapping from T(X, x) into P(X, x). Then s°k does
not define a subnet of {sw}, IE G P(X, x), unless {x} is itself a neighbor­
hood of x. For if {x} is not open, then {x} & T(X, x); hence there is no
U G T(X, x) such that {x} < k(JJ) as is required by (ii) of Definition 2.
Example 7. Let {sn}, n G N } be any sequence. Let k be a function from
N into N defined by

k{n) = n for n < 10, k{n) = 10 for all n > 10.

Then s°k satisfies (i) of Definition 2, but not (ii). On the other hand, if
we define k':N —»N by

k'{n) — n if n is even, fc'(n) = 2 if n is odd,

then s°fc' satisfies (ii), but not (i), of Definition 2.

Example 8. Let {st}, i G /, be a net where I is the set of real numbers

greater than or equal to 1. Let k be the function from the directed set
I X I (directed as in Section 6.2, Exercise 3) defined by k(i, i') = ii'.
We verify that k satisfies (i) and (ii) of Definition 2.
i) If (ily i2) < {h , i±), then i\ < i 3 and i2 < i±. Since all of the
numbers concerned are greater than or equal to 1,

k(ii, ^2) = W 2 < His, u) = hU-

ii) If i and i' are any real numbers greater than or equal to 1, i' < i,
then i < i 2 and i' < 22; hence i < A:(z, 2) and i' < A;(i, i). There­
fore «o/c is a subnet of {$;}, i G /. Note that here the set which
“indexes”the subnet is actually “ richer”than the original index set.
We now prove some of the fundamental properties of subnets.
6.3 Subsequences and Subnets 121

Proposition 2. Suppose a net {$,•}, i G /, has a property P cofinally.

Then there is a subnet {skj}, j G J, of {sy}, i E /, which has the prop­
erty P residually.

Proof. L e t«/ be the set of all j G / such that sy has property P, and let A;
be the identity map from J into I. If J has the order induced from /,
then J is at least a partially ordered set. Since the property P is cofinal,
given any j, f G J C /, there is j " G I such that j < j' < and
sy" has property P. Therefore

/' G J j < j", and / <

hence J is a directed set. Since fc:J —»I is the identity mapping, k is

certainly order-preserving. Since P is cofinal, k also satisfies (ii) of Defini­
tion 2. For if i and i' are any elements of /, there is j G / such that i < j,
i' ^ and sy has P, and hence

j i < k(j), and i’< k(j).

Therefore 5°k is a subnet of {sj, i G /. By definition of each has

the property P; thus , j G /, has the property P residually.

Proposition 3. Suppose a net { s j , i G /, has a property P residually.

Then every subnet of {st-}, i G /, also has the property P residually.

Proof. Since {sy}, i G /, has P residually, there is z0 G I such that if

i 0 < i, then $y has P. Suppose {s^y}, j G /, is a subnet of {sy}, i G I.
Applying (i) and (ii) of Definition 2, we can find^o £ J such that if^’
o ^ j,
then i 0 < k(j). Hence if j 0 < j , Skj has the property P. The net {s^y},
j G /, therefore has the property P residually.

Corollary. A net {sy}, i G /, has a property P residually if and only if

every subnet of {$y}, i G /, has the property P residually.

Proof. Each net is a subnet of itself (Exercise 1); hence if each subnet of
{sy}, i G /, has P residually, then {sy}, i G /, does also. The converse is
Proposition 3.

EXERCISES

1. Prove that every net is a subnet of itself. [Hint: Use J = 1 and let k be the
identity mapping.]
2. Prove the converse of Proposition 2. That is, if a subnet {skj}, j G J, of the
net {sy}, t G I, has a property residually, then {«y}, i G / , has the property
cofinally.
122 Convergence 6.4

3. Let I and J be directed sets and {« *■

} and {tj} be nets indexed by I and J,
respectively, in some set X. Let I X J be directed as in Section 6.2, Exercise 3.
Define a function s X t from I X J into X X X by (s X t)(i, j ) = ($;, tj).
a) Prove that s X t defines a net in X X X.
b) Prove that if {$» • }, i G I, and {tj}, j G /, both have a property P resid-
ually or cofinally, then {(s*, tj)}, (i, j) G I X J, has the property P in the
same way that both nets have it.
c) Suppose M and M' are directed sets and k and k! are functions from M
and M' into I and J , respectively, such that s°k and t°kr are subnets of
{$*•}, i G I and {tj}, j G J. Define the obvious function

k X k ' : M X M' I X J.
Prove that (s X t)°(k X kr) is a subnet of {(«*, tj)}, (i, j) G I X J.
4. Find an example of a net which has a property P cofinally, but such that no
subsequence of the net has the property residually. This in turn will be ac­
complished if we find a net no subnet of which is a subsequence. To find such
a net, consider Example 1 of this chapter. Let g G X be the function which
is identically 0, and let T(X, g) be the set of all neighborhoods of g. Let s be
a selection function from T(X, g) into X. Prove that the net {$f} j V G T(X, g)
has the property of being in every neighborhood of g residually, but that no
subsequence of this net has the property; in fact, prove that there are no
subsequences of {sy}, V G T(X, g), at all.
5. Find a net in the set N of positive integers which has the property cofinally
of being equal to every positive integer. Describe explicitly the subnet of
this net which is residually equal to 3.

6.4 CONVERGENCE OF NETS

Thus far we have primarily studied the idea of a net in an arbitrary set
without regard to any topological structure that might be on the set. But
just as, in Chapter 2, we were interested in the convergence of sequences
in metric spaces, so now we are interested in finding a notion of con­
vergence for nets in general topological spaces which generalizes the notion
of convergence of sequences. We note that the criterion for convergence
of a sequence in a metric space can be restated: A sequence {sn}, n G N,
converges to y if and only if given any neighborhood U of y, then {sn},
n G N, is residually in U. We therefore make the following definition.
Definition 3. Let X , r be any topological space, and suppose {$*•}, i G /,
is a net in X. Then { s j , i G I, is said to converge to a point y of X if
{si}, i G I, is residually in every neighborhood of y. If {«*}, i G I,
converges to y, we write Sj —> y. In other words, s* —> y if given any
neighborhood U of y, there is i 0 G I such that if i 0 < i, S; G U. The
point y is called the limit of {st-}, i G I.
6.4 Convergence of Nets 123

Example 9. If X is any topological space, x G X, and T{X, x) is the

directed set of neighborhoods of x, then for any selection function
s: T(X, x) —»X, the net {$£/}, U G T(X, x), converges to z. For let F be
any neighborhood of x. Then since Su G U for each U G T(X, x) and
V < U means U C F, if F < 17, sc/ G F.
Example 10. Again consider Example 1. We have seen that no sequence
of elements of A converges to g. We now show that there is a net {$»},
i G /, such that Si —> g and Si G A for each i G I. Let T(X, g) be as
defined previously. We wish to prove the existence of a selection function
s from T(X, g) into X such that
s(W) G IF nA for each W G T(X, p).
This will be accomplished if we show that for any finite subset F oi R,
the set of real numbers, and for any positive number p,
U(g, F, p) fi A 9* <f>
(for the family of U(g, F, p) is 3lg, and hence any neighborhood of g con­
tains a neighborhood of this form). This was already done, however, in
proving that #G Cl A. Therefore the net {sw}, IF G T(X, g), where
sw G IF n A, converges to g. We thus see that even though no sequence
of elements of A converges to g G Cl A, there is a net of elements of A
which converges to g. We seem therefore to be well on our way to gen­
eralizing Proposition 1.
The reader may feel that at least sequences are sufficient for doing
whatever has to be done pertaining to the ordinary space R of real numbers
(that is, R with the absolute value metric), and that nets are only of use
in dealing with “ screwball ”topological spaces such as that given in Ex­
ample 1. This is not at all the case, but to help convince the reader that
nets are of great value even in real analysis, we give the following example. *

a = xQ b=xn

E xx T2 x3 xn_x -I
Figure 6.1

Example 11. Let [a, b] be a closed interval in R, the space of real numbers
with the absolute value metric. A partition P of [a, 6] is a finite collection
of points
x0 = a < Xi < x2 < •••< xn- i < xn = b

* Actually, Riemann integrals can be adequately handled entirely in terms of

sequences, but the use of nets is more elegant. Many important notions, how­
ever, depending on the concept of convergence cannot be handled adequately
without appeal to something more general than sequences.
124 Convergence 6.4

(Fig. 6.1). The mesh of P , denoted by m(P), is defined to be

max(|x; — Xi+1|, i = 0, . . . , n — 1,
where n is the number of points in the partition);

thus the maximum mesh of any partition of [a, h] would be b — a. Suppose

P i and P 2 are two partitions of [a, b]. Then P x is said to be finer than P 2l
if P 2c P i ; if P 2 C P i , then m(Px) < m(P2), since P i has at least as
many points as P 2. Set P x < P 2 if P i is finer than P 2. If we let (P denote
the family of all partitions of [a, b], the reader can show that < makes (P
into a directed set.
Let / be any function from [a, 6] into R . For any partition P, say
P consists of
x0 = a < xi < •••< zn_ i < xn = b,
define
n—1
s(/, P) = Y j f(— X i)
i=0
and
«(/, P) = 2 f(xi+-
i=0
Then aS(/, —) and s(/, —) define nets in R, that is,

W ,P )}, PG(P, and («(/,P)}, Pe <P.

If the former net converges, its limit is called the upper Riemann integral
of / over [a, 6]; if the latter net converges, its limit is called the lower
Riemann integral of / over [a, b]. If both nets converge to a common limit,
this limit is called the Riemann integral of / over [a, b], commonly denoted
by $f(x) dx.
Admittedly, this example has been somewhat sketchy. The interested
reader, however, can find this topic developed at length in most books in
real analysis. It should indicate, though, that nets do furnish a powerful
and effective tool in defining and studying a concept known to the reader
from elementary calculus.
We now prove some of the more fundamental properties of the con­
vergence of nets.
Proposition 4. Suppose {$*}, i e /, is a net in X such that {$*}, i G /,
is residually constant; that is, there is y e X and i 0 G I such that if
i 0 < i, Si = y. Then Si —> y.
Proof. Since any neighborhood of y contains y, {si}, i G /, is residually in
any neighborhood of y\ hence {$*•}, i G /, converges to y.
6.4 Convergence of Nets 125

Proposition 5. If sy —> y, then every subnet of {$y}, i G /, also con­

verges to y.
Proof. Since $y —> y, {sy}, i E 7, has the property of being in every neigh­
borhood of y residually. Then, by Proposition 3, every subnet of {sy},
i G 7, also has the property of being in every neighborhood of y residually.
Therefore every subnet of {sy}, i G 7, converges to y.
Proposition 6. If every subnet of a net {sy}, i G 7, has a subsubnet
which converges to y, then sy —> 2/. This is to say that if {sy}, i G 7,
does not converge to ?/, then there is a subnet of {sy}, i G 7, no subnet
of which converges to y.
Proof. Since {$y}, i G /, does not converge to y , there is a neighborhood U
of y such that there does not exist any i 0 E 7 such that i 0 < i implies
sy G U. Let J = {j E I \ sj & U } , and let k be the identity mapping from
J into 7. Then s°k is a subnet of {sj, i G I (Exercise 1). But each
is not an element of U. Therefore there cannot be a subnet of ,j G J,
which converges to y.
We now prove the long-awaited generalization of Proposition 1.
Proposition 7. If A is any subset of a topological space X, r, then

zeCIA

if and only if there is a net {sy}, i G /, such that

Si —> x and Si G A for each i G I.

Proof. Suppose first there is a net {sy}, i G /, such that

Sy —»z and sy G A for each i G /.

Then each neighborhood of x contains at least one point of A. There­

fore x G Cl A.
Suppose x G Cl A. Let T(X, x) be the directed set of neighborhoods
of x, and let s be a selection function from T(X, x) into X such that

s(W) e W n A for each W E T(X, x).

We know that such a selection function exists because x E Cl A ; hence

every neighborhood of x contains some point of A. Then the net {W},
W E T(X, x), converges to x (Example 9).
Proposition 7 strengthens our opinion that we have not only general­
ized sequences properly, but have also generalized the notion of con­
vergence properly.
126 Convergence 6.4

It was shown that any sequence which converges in a metric space

converges to a unique limit (Proposition 6, Chapter 3). We might then
wonder, In what types of spaces do convergent nets have unique limits?
The next proposition answers this question.
Proposition 8. A space A, r is T2 if and only if given any convergent
net {s2}, i G /, in A , the limit of {$*•}, i G /, is unique.
Proof. Suppose A, r is T 2, but that there is some net {«*•}, i G /, in A such
that {$*}, i G /, converges to distinct points £ and y. Since A is T 2, there
are neighborhoods 1/ and F of x and y, respectively, such that U n V = </>.
Since Si —»a; and s* —»y, { s j , i G /, is residually in both [/ and F. There­
fore there are and Zq such that i 0 < i implies S{ G U and ^ < i implies
Si G F. Since I is a directed set, there is7 G / such that z0 < j and ig ^ i-
Therefore sy G 1/ D F, a contradiction.
Suppose A, r is not T2. Then there are distinct points x and y of A
such that every neighborhood of x meets every neighborhood of y. Let
T(A, x) and T(A, y) be the directed sets of neighborhoods of x and y.
Then T(A, x) X T(A, y) is a directed set [directed by defining (Uy V) <
(U'y V') if U' C U and F' C F as in Section 6.2, Exercise 3]. Since U n
F 5^ </> for each (£7, F) G T(A, a:) X T(A, y), there is a selection function

s: T(A, «) X r(A, y) —> A

such that s([/, F) G C/ fl F. Then

{« (^F)},(^ F) e T(X, xy),

is a net in A which converges to both x and y (Fig. 6.2).

Figure 6.2

Example 12. Lest it seem peculiar to the reader that a net should be able
to converge to several points, let him remember that if A is a set with the
trivial topology, then any sequence {sn}, n G N y in A converges to every
point of A. For if x G A, then the only neighborhood of x is A, and every
sequence in A is residually in A. Admittedly, however, the nicest spaces
are those where convergent nets have unique limits. This is why many
topologists restrict their attention only to spaces which are at least T 2.
6.5 Limit Points 127

EXERCISES

1. In Proposition 6, prove that s°k is a subnet of {s;}, i G I.

2. Suppose X, D is a metric space and {s»}, i G /, is a net in X.
a) Suppose Si —> x. Prove that a subsequence of {st-}, i G J, converges to z.
b) Prove that if every subsequence of {$»} converges to x, then s; —»x.
c) Prove (a) and (b) when it is merely assumed that X, r is a first countable
space (Section 6.1, Exercise 2).
3. Let X be a set. Suppose a “ rule”of convergence is given which satisfies
i) Si —> x implies every subnet of {$» •}, i G 7, converges to x, and
ii) if a net {st-}, i G 7, is residually constantly equal to y, then st-—► y.
Define a subset A of X to be closed if and only if for each net {$»}, i G 7,
such that Si G A for each i G 7 and Si —► y, y G A.
a) Show that the set of closed subsets of X defines a topology r on X.
b) Show that each net which converges according to the original rule of
convergence also converges with respect to r.
c) Show that some net which did not converge with respect to the original
rule might still converge with respect to r.
4. Prove that the only nets which converge in a space with the discrete topology
are nets which are residually constant.
5. Let N be the set of positive integers. Discuss the convergence of nets in N
when N is given each of the following topologies.
a) r = {N, <t>,{0}}
b) t = {U C N \ U contains all but finitely many elements of N}
c) the subspace topology from R with the absolute value topology
6. Let N, the set of positive integers, have the topology in Problem 5(b). Show
that every net in N has a convergent subnet.
7. Suppose X is any set and r and t are two possible topologies for X. Prove
that r' C t if and only if every net in X which converges with respect to r
also converges with respect to r'.
8. Find a criterion in terms of nets for a space to be Ti. Do likewise for To-

6.5 LIMIT POINTS

Definition 4. Let {si}, i G /, be any net in a space X , r. A point y of X
is said to be a limit point of {Si}, i G /, (not to be confused with limit)
if {si} , i G /, is cofinally in every neighborhood of y. That is, y is a
limit point of {Si}, i G /, if given any neighborhood U of y and any
elements i and i r of /, there is i ff G I such that i < i", V < z", and
S{••G U.

Note that if Si —»y, then y is a limit point of {$*}, i G I. On the other

hand, a net need not converge to a limit point, as the following example
demonstrates.
128 Convergence 6.5

Example 13. Let R be the space of real numbers with the absolute value
topology. Let {sn}, n E N, be the sequence in R defined by sn = (—l)n.
If n is odd, then sn = —1, and if n is even, sn = 1. Then {sn}, n E N,
has both —1 and 1 as limit points. For if U is any neighborhood of 1 and
m and m! are any two elements of N, then there is an even integer m "
greater than both m and m', and $m** E U. Thus 1 is a limit point of {sn} ,
n E N; similarly, —1 is also a limit point. We note that even though the
space involved is T2y a net, here a sequence, may have a number of dif­
ferent limit points.
However, {sn}, n E N, does not converge to either 1 or —1. For sup­
pose sn —> 1. Since R is T2, we may find neighborhoods U and V of 1 and
—1, respectively, such that U fl V = </>. Then {sn},n E N yis residually in
Uy a contradiction to the fact that {sn}, n E N, is cofinally constantly —1
and —l & U . Similarly, sn -h — 1.
We note, however, that if we let N f represent the set of positive even
integers, N " represent the set of positive odd integers, and k' and k" be
the identity mappings from N' into N and N " into N, respectively, then
s°k' and $<>k" are subsequences of {$w}, n E N, which converge to 1 and
—1, respectively. We might conjecture then that even though a net need
not converge to one of its limit points, some subnet of that net might. We
prove this conjecture in the next proposition.

Proposition 9. Let X yr be a topological space and {$*•}, i E /, be a net

in X. Then y is a limit point of {st}, i E /, if and only if {$*}, i E /,
has a subnet which converges to y.

Proof. Suppose {$;}, i E /, has a subnet which converges to y. Then there

is a directed set J and a function k from J into I as in Definition 2 such
that s ok is a subnet of {$*•}, i E I, and —> y. Suppose U is any neigh­
borhood of y. Then {sfcy}, j E J, is residually in U, that is, there is jo
such that j o < j implies Skj E U. Suppose i and i' are any two elements
of I. Then since I is directed, there is i" E I such that i < i”and i' < i".
But k(j0) is an element of /, and {skj} , j e J , is cofinal in {$*}, i E I. We
can therefore find j' E J such that

k(jo) < H f) and i f < k(f).

Since k is order-preserving, j 0 < j'; hence skj> E U. Since < is transitive,

i ^ k(f) and i' < k(j'). In sum then, given i and i' in /, we have found
k(j') E I such that

i < k(j'), i' < k(j'), and skj, E U.

Therefore {st}, i E /, is cofinally in U, and hence y is a limit point of

{$;}, < e /.
 130 Convergence 6.6

2. Prove the corollary to Proposition 9.

3. Let {$;}, i E I be a net in a space X, r and let A be the set of limit points of
{si}, i E I. Prove that {st-1i E 1} U A is a closed subset of X.
4. Find all the limit points of each of the following sequences.
a) sn = 1/n in the set of real numbers with the order topology
b) sn = (-—l)n in the set of real numbers with the trivial topology
c) sn — (—l)n in the set of real numbers with the discrete topology
d) sn = n in the set of real numbers with the absolute value topology
5. Let X, r be a space with the property that any net in X which has a limit
point converges to that limit point. Discuss the various possibilities for the
topology on X.
6. Let R be the set of real numbers with the absolute value topology. A sequence
{sn}, n E N, is said to be bounded if there are real numbers m and M such
that m < sn < M for all n E N. Prove that any bounded sequence inR
has a limit point. Prove that every convergent sequence in R is bounded,
but that not every bounded sequence is convergent.
7. Prove or disprove: There exists a sequence of real numbers which has every
real number as a limit point. Prove or disprove: There exists a net of real
numbers which has R as its set of limit points.

6.6 CONTINUITY AND CONVERGENCE

The follow ing proposition relates the convergence of nets and the con­
tin u ity of functions. I t is a generalization of P roposition 10 of Chapter 2,
thus strengthening the assertion th a t nets are a good generalization of
sequences.

Proposition 10. L e t / be a fu n c tio n fro m a space X, r to a space Y, r '.

Then / is continuous if and o n ly if fo r every net {s2} , i E I, in X such
th a t Si —> x, the net {f( s z) } , i e / , in Y converges to f(x).

Proof. Suppose / is continuous, b u t also suppose there is a net {si}, i E I,

in X such th a t sz —> x , b u t {f(sz)}, i E I , does n o t converge to f(x). T hen
there is a subnet of { f(s z) }, i e I , no subnet of w h ich converges to f{x)
(P roposition 6). L e t

{/(%)}, JG•
/-

be such a subnet. T hen there is a neighborhood V o f fo r w hich {/(sfcj)},

j G J, is re sid u a lly n o t in . Bu t since / is continuou
V
borhood o f x. Since Si * x, {s*}, i e I, is residu ally i n
—
j E J,is a subnet o f {st}, i E /, w hich is n o t re sid u a lly in / - 1 (F ). T here­
fore {Si} , i E I,has a subnet w h ic h does n o t converge to x, a c o n tra d ic
to P ro p o sitio n 5.
6.6 Continuity and Convergence 131

Suppose for each net {s*}, i G I, in X such that s* —> x, /(s») —>/(x),
but / is not continuous. Then there is a neighborhood F of /(x) such that
for no neighborhood U of x do we have f(U) C V. Let T(X, x) be the
directed set of all neighborhoods of x. Let $ be a selection function from
T(X, x) into X such that & V for all U G T(X, x). Then sv -> x,
but f(su) -/->/(x), a contradiction.

We now use Proposition 10 to prove several results about nets in

derived topological spaces.

Proposition 11. Let X, r be any space and R be an equivalence rela­

tion on X. For each x G X, let x denote the equivalence class of x.
Then if Si —> x in X, S{ —> x in the identification space X/R.

Proof. The function defined by x —> x is continuous. We then apply

Proposition 10.

Proposition 12. Suppose {s*}, i G /, is a net in the product space

X j X j. We will denote the^th coordinate of Si by sj; thus {s^}, i G I,
will be a net in Xj. Then

Si~+y = (yi,y2, •••,2/y, . •.)

(remember we have restricted our attention in this text to the product

of countably many spaces) if and only if

4 -> y,.
Proof. Suppose Si —> y. Then the projection mapping from X j X j into
Xj, Tj is continuous. Therefore, by Proposition 10,

Pj(si) = s3
i -> p3(y) =

Suppose si —> yj for each j G J. Let U be a typical basic neighborhood

for y in the product topology. Then U = X j W j, where each Wj is an
open subset of X j ; in particular, each Wj is a neighborhood of yj. Also
Wj = X j for each j G J , except finitely many, say j\, . . . , jn. For each
j G J , except j u . . . , jn, sj G W j. For j 1} ... , jn we can find i l7 . . . , in
in I such that if iq < i, s\q G W jq, g = 1, ... ,n. Since / is directed, we
can find, using induction if necessary, i 0 G I such that if i 0 < i, s{q G Wj ,
q = 1, . . . , n. Thus if i 0 < i, s{ G Wj for each j G /, and hence if < iy

Si G X Wj.

Therefore , i G /, is residually in X j Wj. Since Xj Wj was a typical

basic neighborhood and every neighborhood of y contains such a basic
132 Convergence 6.6

neighborhood, {$*•}, i G /, is residually in every neighborhood of y ; there­

fore Si —> y.

It is not true that if F is a subspace of X, {$*•}, i G 7, converges in X,

and ^ G F for each i G 7, then {$*}, i G 7, considered as a net in Y also
converges. This is not even true for sequences, as we see from the follow­
ing example.

Example 14. Let R be the set of real numbers with the absolute value
topology. Then the sequence defined by sn — 1/n converges to 0 in R ,
but does not converge at all in the subspace (0, 1), since (0, 1) does not
contain the limit 0. The following is true, however.

Proposition 13
a) If S{ —> 2/ in X and Y is a subspace of X such that S{ G Y for each i
and y G F, then Si —> y in F also.
b) If Si y in X and F is a closed subspace of X, then if each Si G F,
then y G F as well, and s* —> y in F.
Proof. The proof of (a) is left as an exercise. Statement (b) follows im­
mediately from (a) and Proposition 7.

EXERCISES
1. Prove Proposition 13.
2. Let {Xj, Tj}, j G /, be a countable family of nonempty spaces, and consider
the product Xj Xj of the sets {Xj}, j G J. How much of Proposition 12 is
true if X/ Xj is given a topology coarser than the product topology? How
much of Proposition 12 is true if X/ Xj is given a topology finer than the
product topology?
3. Prove that a function / from a space X, r onto a space F, r' is a homeo-
morphism if and only if a net {s*}, i G 7, converges to a; G X if and only if
i G /, converges to f(x) in F.
4. Using the results of this chapter, prove that if / is a continuous function from
X, r into F, r', then/(Cl .4) C Cl/(H) for any A C X.
5. Also using methods from this chapter, prove Proposition 21, Chapter 4.
6. Suppose / is a continuous function from X, r into F, r'. Let {st}, i G 7, be a
net in X, and suppose H is the set of limit points of this net. Prove that/(H)
is a set of limit points for {/(«*)}, i G /. Is it necessarily a complete set of
limit points for {/($;)}, i G /, or might there be others as well? [Hint: Con­
sider the sequence defined by sn = 1/n in (0, 1), and the identity function
from (0, 1) into the space of real numbers.]
7. Let X/R be an identification space of a space X. Discuss the conditions R
must satisfy for the following to hold: For any net {s;}, i G /, in X which
converges to a point x) {Si}, i G /, the identification net in X/R converges
only to x.
6.7 Filters 133

6.7 FILTERS
There is an alternative approach to the concept of convergence in a general
topological space through the notion of a filter. While the study of filters
is of great importance in point set topology, we will accomplish much of
what filters might be useful for by using nets. Nevertheless, we will
introduce the notion of a filter now and study some of its basic properties
for two reasons: (1) the notion of a filter is sufficiently important that
anyone studying even introductory point set topology should at least know
what a filter is; and (2) the reader should come to realize that, even in
mathematics, there may be many means to the same end. A proposition
in mathematics may have many proofs, and different machinery can be
developed to accomplish the same task. We will in this section try to
stress the relations between nets and filters and the analogies in their use.
We would expect that since nets and filters are both designed for the study
of convergence, there will have to be many theorems about filters com­
pletely analogous to theorems stated in terms of nets.
Definition 5. Let X be any set. A collection d of nonempty subsets of
X is said to be a filter on X if
i) (t 7^ (f))
ii) if A and B are in d, then A n B is also in d ;
iii) if A e d and A C B, then B e d.
If X, t is a topological space and d is a filter on X , then d is said to
converge to x, denoted by d —> x, if every neighborhood of x is a member
of d. d is said to have a; as a limit point if every neighborhood of x
meets every member of d. That is, d —»x if given any neighborhood
U of x, U E d. # is a limit point of d if given any neighborhood V of
x, and any A E d, then U n A ^ <t>.
Example 15. If X is any set and 7 is any nonempty subset of X, then the
family d of all subsets of X which contain Y is a filter on X. We verify that
d satisfies Definition 5. Since Y </>, each set which contains Y is non­
empty.
i) Since Y c Y, Y e d; hence d ^ </>.
ii) If A and B are in d, then Y C A and 7 C B ; thus 7 C A n B,
and therefore A D B e d.
iii) If A G d ythen 7 C A. Therefore if A C B, then 7 C A C B, and
thus B E d. Hence a is a filter on X.
If X, t is a space and x e X, then T(X, x ), the family of all neighbor­
hoods of x, is not a filter on X, since given any neighborhood U of x, it is
not necessarily true that any subset of X which contains U is also a neigh­
borhood of x. If we let T*(X, x) be the family of all subsets A of X such
that A contains a neighborhood of x, then T*(X, x) is a filter on x. More-
134 Convergence 6.7

over, since T(X, x) C T*(X, x),

T*(X, x) — x.

(Compare this to Examples 3, 6, and 9.)

Example 16. Let X be any set and let 2) be a nonempty collection of non­
empty subsets of X with the property that if B and B' are in 3D, then there
exists B " G 3D such that B " C B n B r. Let

a= {A |B C A, B G 3D}.

Then G, is a filter on X (Exercise 1). Ct is said to be the filter generated by

3D and 3D is said to be a basis for the filter Ct.

Example 17. Suppose { s j , i G /, is a net in a set X. Let 3D be the family

of all subsets of the form B j = {si |j < i) , for all j G I. Then the family
3D is a nonempty collection of nonempty subsets of X having the property
that if Bj and By are members of 3D, then there is B j" G 3D such that

B j" C B j fi B j/

(Exercise 2). 3D thus forms the basis for a (unique) filter G as in Example
16; specifically,

Ct = {A |B j C A for some j G /}.

Ct is said to be the filter generated by the net {s*}, i G /.

Proposition 14. Let {s*}, i G J, be a net in a space X, r and let Ct be the

filter generated by {s*}, i G /. Then
a) G —»x if and only if s* —► x;
b) Ct has £ as a limit point if and only if a; is a limit point of {st} , i G I .
Proof
a) Suppose s* —»x. Then given any neighborhood U of x, there is
j ElI such that Bj (using the notation of Example 17) is a subset
of U. Since Bj C U} U G Ct. Therefore every neighborhood of x is
in Ct, and hence Ct —> x.
Suppose Ct —> x. Then given any neighborhood U of x, there is j E I
such that Bj C U. Hence if j < i, Sj G U. Thus {$;}, i G /, is
residually in every neighborhood of x, and therefore x.
The proof of (b) is left as an exercise.
In Example 17 we associated a filter with any net. In order for this
association to be at all meaningful, Proposition 14 was a necessity. For if
Proposition 14 were not true, then a net might converge without the
corresponding filter converging, or we might have net and filter converging
6.7 Filters 135

to different points. But if nets and filters are merely to be different ap­
proaches to the same concept, this would be intolerable. We now show that
starting with a filter (2, we can associate a net with <2 which has the same
convergence properties as (2.
Let (2 be a filter on a set X. We will construct a net based on (2. (2 is
a collection of nonempty subsets of X. If A and B are in (2, let
A < B if B e A.
It is easy to verify that this makes (2 into a directed set. Since each A G (2
is a nonempty set, we can find a selection function s from the directed
set <2 into X such that s(A) G A. Then {s^}, A G C2, is a net in X called a
net based on (2.
Proposition 15. (2 —> y if and only if every net {sa} , A g C2, based on
(2 also converges to y.
Proof. Suppose a —> y and {sA}, A e (2, is a net based on (2. Let U be
any neighborhood of y; since d —> y, U G (2. Then if 1/ < 4, A C U, for
any i e f l . Hence if U < A, sA G A C U; therefore {s^}, A G d, is
residually in U. Thus sA —> 2/.
Suppose d -p> y. Then there is a neighborhood U of y which is not a
member of (2. If A G <2, select sA G A — U; such a selection is always
possible, for if A — U = 0, then A C U, which would make U a member
of (2. Then {s^}, A G d, is a net based on (2 which does not converge to y.

Proposition 16. Suppose (2 is a filter on a space X , r. Then x is a limit

point of (2if and only if there is a filter a' such that (2 C (2' and a' —> x.

Proof. Suppose x is a limit point of d. Let

33' = {A n U |A G d and U is a neighborhood of x}.

Then 33' is the basis for a filter <2' on X (Exercise 3). Since A D U C U
for each A e d and any neighborhood U of x, U G d '. Therefore every
neighborhood of x is a member of (2', and thus <2' —> x. It remains to be
shown that d C <2'. This follows at once from the fact that X is a neigh­
borhood of x; hence if A e (2, A n X = A is a member of 33', and hence
of d'.
Suppose on the other hand that there is a filter d' such that d C d'
and a' —> x. Let U be any neighborhood of x and A be an element of d.
Then U and A are both members of (2'; hence

A n U G d'

by (ii) of Definition 5. Since A n U G d', then A n U ^ 0. Therefore,

given any neighborhood U of x, and any member A of a, A n U 9^ 0. Hence
x is a limit point of (2.
136 Convergence 6.8

Comparing Proposition 16 with Proposition 9, we find that the filter

analog of a subnet is a finer filter, where a filter d' is finer than a filter d if
a c d'.

EXERCISES

1. Prove that d in Example 16 is a filter.

2. Prove in Example 17 that there is B y £ 3Dsuch that 5^/ C Bj fi By.
3. In Proposition 16, prove 3D' is a filter basis.
4. Prove (b) in Proposition 14.
5. Suppose / is a function from a space X, r into a space 7, r' and d is a filter
on X.
a) Set /(d) = {/(A) | A £ d}. Prove that /(d) is the basis for a filter d'
on 7.
b) Prove that / is continuous if and only if given any filter d on X such that
d —► x, d' [the filter for which /(d) is a basis] —>f(x).
6. Suppose A is a subset of X, r. Prove x £ Cl A if and only if there is a filter d
on X such that d —»x and A £ d.
7. Prove that a space X, r is T2 if and only if any convergent filter on X con­
verges to a unique limit.
8. Which of the following are filters? Which are filter bases? For those which
are neither filters nor filter bases, indicate which properties are lacking.
a) the family of subsets of a set X which contain 7 C X
b) the set of all closed half-planes in R2 which contain (0, 0)
c) the set of all open half-planes of R2
d) the union of two filters on a set X
9. State and prove the filter analogs of Propositions 4 and 5 of this chapter.
10. Find a criterion in terms of filters for a space to be T1.

6.8 ULTRANETS AND ULTRAFILTERS

There is another concept involving nets which will prove useful in the
discussion of compact spaces; this concept is that of an ultranet. Actually,
the filter analog of an ultranet is more natural, since it is a bit difficult to
motivate the notion of an ultranet, except to say that it works. We there­
fore introduce the concept of an ultrafilter first and then pass to the net
analog.
Definition 6. An ultrafilter on a set X is a maximal filter on X. That
is, a filter d on a set X is an ultrafilter on X if given any filter d' finer
than d (i.e., d C d'), d = d'.
6.8 Ultranets and Ultrafilters 137

Example 18. Let X be any set and x e X. Then the family of all subsets
of X which contain x forms an ultrafilter d on X. For if d' is any filter
finer than 01 and A' e d ', then either x e A' or x & A'. Now if x e A',
then A' e Ct. If x & A\ then x E X — A hence I - i ' E ft C ^ But
then
A' n (X — A') = 0 E a',

a contradiction to the fact that each member of (£' is nonempty. Therefore

x is an element of each member of d '; hence d' C d, and consequently

a = a'.

Proposition 17. A necessary and sufficient condition that a filter d on

a set X be an ultrafilter is that given any subset A of X, either

A ed or X — A G ft.

Proof. Suppose d is a filter on X with the property that either A or X — A

is a member of d for any A C X. Suppose d' is a filter on X which is finer
than d. To prove that d = d f, it will suffice to prove that each element
of d ' is also an element of d. Let A' e d'. If A' e d, we are done. If
A' & d, then
X — A' E d C d'.

But then A' and X — A' are both members of d'\ hence

A' n (X — A ') - 0

is a member of d', a contradiction.

Suppose now that d is an ultrafilter on X, but there is a subset A of X
such that neither A nor X — A is a member of d. We will find a filter d'
on X which is strictly finer than d. If A n B </> for each B e d, then
we can take
3D = {A n B |B G d}

as a filter basis for a filter d' which is strictly finer than d (since A e d
but A & d).
Suppose, however, that A n B x = <t> for some B x e d. We will show
that
(X - A) n B ^ </>

for every B e d. Suppose (X — A) n B 2 = </> for some B 2 E (L Then

B , n B 2 = (CBi n b 2) n a ) u ((£i n b 2) n A))

c (B, n u (£2 n(X- = 0,
138 Convergence 6.8

a contradiction since B i n B 2 is a member of (2 and hence must be non­

empty. Therefore
© - {(X - A) n B I B E a}

is a basis for a filter <2' on X which is strictly finer than <2 (since (2' con­
tains X — A).

Proposition 18. If X is any set, then every filter <2 on X is contained

in an ultrafilter.

Proof. Consider the family § 1 of all filters &' on X such that (2 C (2'.
21 can be partially ordered by “is finer than. ” Suppose 0 = {<2*}, k e K,
is a chain in 2f. We now show that
s
3D = {B |B e a l for some k E K }

is a basis for a filter (2" on X. First, 3D is certainly a nonempty collection

of nonempty sets, since each d k is a nonempty collection of nonempty sets.
Suppose B and B' are members of 3D. Then andB ' e <2*, for some
fc and A/ in X . But 0 is a chain, and hence either Qk C (2^, or d'k, C (2&;
assume the latter. Then B and B' are both in (2*, and thus B n B' e <2&.
Hence
5 fl 5' G 3D.

Therefore 3D is a basis for a filter (2" on X. Moreover, (2" is clearly finer

than any member of 0. Hence (2" is an upper bound in 0 for 21.
Each chain in 21 thus has an upper bound. Applying Zorn’ s lemma,
21 therefore contains a maximal element (2. Then (2 is an ultrafilter which
contains (2.

Proposition 19. Suppose / is any function from a set X onto a set Y

and (2 is an ultrafilter on X. Let /((2) be the filter basis as described in
Section 6.7, Exercise 5, and let (2' be the filter on Y that it determines.
Then (2' is an ultrafilter on F.

Proof. Suppose A c Y. In order to show that (2' is an ultrafilter, we must

show that either A or F — A is a member of (2' (Proposition 17). Since
a is an ultrafilter on X , either/-1 (A) o r / -1(F — A) — X — f~ l (A) is
a member of (2. If /-1(A) e (2, then

f i r 1(A)) = A e/(<2) c a'.

If X — /-1(A), then F — A e (2'. Therefore (2' is an ultrafilter on F.

Proposition 20. If (2 is an ultrafilter on a space X , r and y is a limit

point of <2, then (2 —»y.
140 Convergence 6.8

Proof
b) Suppose A C F. Then {s*}, i g is residually in either/-1 (A) or
X — f ~ x(A). Therefore {/(«»)}, i g /, is residually in either A or
F — A, and is hence an ultranet in F.
c) Let U be any neighborhood of y. Since {si}, i G /, is an ultranet,
it is residually in either U or X ~ U. Since 2/ is a limit point of
{si}, i G /, the net could not be residually in X — [/. Therefore
f e } , i G /, is residually in [/; hence s* —»y.
a) Let {$;}, i G /, be any net in a set X and let <2be the filter generated
by {«;}, i E /. By Proposition 18, a C <3', where (3' is an ultra­
filter. We first show that {st}, i g /, is cofinally in A for each
A G (3'. Let A G (3'. If {$*}, i g /, is not cofinally in A, then
{«»}, i E /, is residually in X — A. But then

X — A G a C (3'.

Therefore A and X — A are both elements of (3', an impossibility;

hence {$*•}, i G /, is cofinally in A. Let

J = {(t, A) | i G /, A G C3', and st-G A}.

Since (3' and / are both directed sets ((3' is directed by letting
A < A' if A' C A), J is directed. Define k: J —> I by k(i, A) = i.
Then s°k is easily verified to be a subnet of {s*}, i G /. By defini­
tion of s o k, s o fc is residually in each A G C3'. But (3' is an ultrafilter
and hence contains any subset of X or its complement. Thus s°k
is residually in any subset of X or its complement, and is therefore
an ultranet in X.

EXERCISES

1. In P roposition 22, verify in the proof of (a) that s°k is a subnet of {s;}, i G I.
2. Prove (b) of Proposition 21. Show that the converse of (b) is false.
3. a) Let {s*}, i G /, be a net in X, and suppose (3 is the filter generated b y
{st-}, i g I. Is {si}, i G I, then a net based on (3? Is it the only net based
on (3?
b) Suppose (3 is a filter on a set X and {sA} , A G <3, is a net based on a.
Is (3 necessarily the filter which {s^}, A G (3, generates?
4. Which of the follow ing sequences in R, the set of real numbers, are ultranets?
N ote that the property of being an ultranet is independent of the top ology
on R. r
a) sn = l/n b) sn = n c) sn = l/^i2 d) sn = (— l) n
6.8 Ultranets and Ultrafilters 141

5. a) Prove that a fu n ctio n / fro m a space X, r to a space Y, t' is continuous if

and only if given any ultranet { s j, i E I, such that —»y, /(st) —>f(y).
b) Prove that /: X, r —> Y, t' is continuous if and only if given any ultrafilter
Q in X such that Cfc —»y, the filter generated b y /(&) converges t o /(?/).
6. Prove or disprove: A space X, r is 7^ if and only if every convergent ultranet
in X converges to a unique limit.
7. Let N be the set of positive integers and $7 be the set of subsets of N con­
taining all but finitely many elements of N. Prove that $ is a filter. Prove
that any ultrafilter containing ^ is nontrivial, and hence there exists a non­
trivial ultrafilter on N.
7
COVERING PROPERTIES

7.1 OPEN COVERS AND REFINEMENTS

Some of the most important aspects of certain types of topological spaces
can be expressed as covering properties. The nice definitions given in this
chapter were not always used in the study of topology. As is usually the
case with a new discipline, those who pioneered in topology thought cer­
tain properties were important for a space to have. The best means of
expressing those properties, best from the point of view of most elegant
and most workable, were only developed from years of experience. The
student should be sophisticated enough to realize that areas of mathe­
matical study are not born full-grown but, as with human infants, require
a period of growth of many years before reaching maturity.
Compactness, the most important covering property, was once defined
as follows: A space X , r is compact if for every infinite subset A C X, there
is at least one y E X such that given any two neighborhoods U and U' of
y, U n A and U' D A have the same cardinality. Even the novice in
topology will realize that this is a rather cumbersome definition. As more
became known about the property that this definition was intended to
convey, equivalent expressions of it became known. Compactness is now
defined as a covering property.* Certain other concepts valuable in the
study of topological spaces can also be best expressed as covering properties.
A cover of a space X, r is exactly what its name implies, a collection of
subsets of X which cover X , that is, whose union is X. Usually, however,
we wish the members of the cover to be sets of a particular form, generally,
open sets. We therefore state the following.

Definition 1. Let X, r be a topological space. An open cover of X is a

collection {Ui}, i E /, of open subsets of X such that

U Ui = x.
i
Let {Ui}, i El I be an open cover of the space X, r. A collection

* The old definition of compactness, however, is actually not equivalent to the

definition of compactness as a covering property.
142
7.1 Open Covers and Refinements 143

{Vj}, j E J, is said to be an open subcover of {Ui}, i E /, if

{ V j \ j e J} c {tf<|*E J}

(that is, each Vj is a Ui) and {Vj}, j E J , is itself an open cover of X.

The collection {Vj}, j E J , is said to be a refinement of {Ui}, i E I, if
{F y },jE « /, is an open cover, and for each Vj, there is Ui such that
Vj C Uj.
Note that an open subcover is a refinement, but a refinement is not
necessarily an open subcover.
Example 1. Let R be the set of real numbers with the topology induced
by the absolute value metric. Then

{N(x, 4) |x E R},

that is, the set of all 4-neighborhoods in R, is an open cover of R. The set

{N(n, 4) |n is an integer}

is an open subcover of {N(x, 4) |x E R}. The set

{N(x, 1) |x E R)

is a refinement of {N(x, 4) |x E R}, since every 1-neighborhood is con­

tained in some 4-neighborhood. In fact, every 1-neighborhood in R is
contained in a 4-neighborhood of an integer; hence {N(x, 1) |x E #} is a
refinement of {N(n, 4) |n is an integer}, even though the cardinality of
{N(x, 1) |x e R} is greater than that of {N(n, 4) |n an integer}.
Example 2. Let X be a set with the discrete topology. Then {{x} |x E X}
is an open cover of X. Moreover, this open cover has no proper subcover,
nor any proper refinement. If X has the trivial topology, then the only
open covers of X are (X, 0} and {X}. (See Exercise 1.)
Example 3. Let N be the set of positive integers with the topology deter­
mined by calling a subset U of N open if U contains all but at most finitely
many elements of X. Let {Ui}, i E I, be any open cover of N. Pick any
U{. Then Ui contains all but at most finitely many of the positive integers;
say Ui excludes nlf . . . , np. Since {Ui}, i E /, is an open cover, every
element of N is in at least one of the Ui, and hence there are at most p
other members of {Ui}, i E I, say Uiv . . . , Uip such that

N = Ui U Ui, U •••U Uip.

Thus {Uif Uiv . . . , Uip} is a finite open subcover of {Ui}, i E I . We

therefore see that every open cover of N (with the prescribed topology)
has a finite open subcover.
144 Covering Properties 7.2

EXERCISES

1. Prove the assertions made in Example 2.

2. Let R2be the coordinate plane with the topology induced by the Pythagorean
metric. Which of the following are open subcovers of
{N((x, y), 1)| (x, e
Which are refinements of
{JV((z, y),3) I (x, y)(ER2}?
In the event a collection is not a subcover, or not a refinement, explain which
properties are lacking.
a) {N((m, n), i) |m and n are integers}
b) {N((0, 0), p) |p a positive real number}
c) {N((x, y), l) I X and y are rational}
d) {N({x, y), |) |x and y are rational}
e) the family of all sets of the form {(x, y) | \x — a \ \ y — b\ < 1}, where
(a, b) is any point of R2
f) the family of all subsets of R2
3. Prove that R2 with the Pythagorean topology has a countable cover con­
sisting of p-neighborhoods. Prove that the set of real numbers with the order
topology has a countable cover consisting of intervals of the form (—p, p),
where p > 0.
4. Suppose the open interval (0, 1) is given the absolute value topology. Form
{Un}> n - 1, 2, 3, ... , where Un = (l/(n + 1), l). Prove that {£/„ }, n G N,
is an open cover of (0, 1). Show that no finite number of the Un cover (0, 1),
even though any finite number of the Un may be omitted and what remains
still give an open cover of (0, 1).
5. Suppose r is a topology on the set N of positive integers with the property
that any open cover of N has an open subcover which contains at most two
elements. Describe all possibilities for r.

7.2 COUNTABILITY PROPERTIES

One would rightly suspect that covering properties take the following
general form: If {Ui}, i E 7, is any open cover of a space X , r, then there
is an open subcover (or refinement) of {Ui}, i E I, satisfying some special
condition. One of the most natural conditions the subcover might satisfy
is a cardinality condition. Such a cardinality condition is given in the
following definition.

Definition 2. A space X, r is said to be a Lindelof space if every open

cover of X has a countable open subcover.
7.2 Countability Properties 145

Example 4. The space presented in Example 3 is certainly a Lindelof

space, since every open cover of N not only has a countable subcover, but
even has a finite subcover.
In order to discuss Lindelof spaces more completely, more terminology
is needed.

Definition 3. Let X, r be a topological space. X is said to be first count­

able if there is an open neighborhood system for r such that dlx is
countable for each x G l (See Section 6.1, Exercise 2.) X is said to
be second countable if there is a basis for r which consists of countably
many sets. X is said to be separable if X contains a countable dense
subset. (For the definition of a dense subset, see Section 3.6).

Example 5. Let R be the set of real numbers with the absolute value
topology r. For each x e R, let

3lx = {X(x, 1/n) |n a positive integer}.

It is easily verified that the collection of 0IX forms an open neighborhood

system for 7. However, each 3l x is countable; hence R is first countable.
(Actually, from Section 3.3, Exercise 6, we have the more general result
that any metric space is first countable.) The set of rational numbers forms
a countable dense subset of R, and hence R is also separable. Proposition 5
will tell us that R is second countable as well, and therefore is Lindelof
(Proposition 3).

Example 6 . Let X be any uncountable set with the discrete topology. For
each x e X, set 9lx = {{z}}. Then the collection of 9lx forms an open
neighborhood system for the discrete topology on X. Thus X is first
countable. Since Cl A = A for every A C X (since every subset of X is
closed), the only dense subset of X is X itself. But X is uncountable, and
hence there is no countable dense subset of X; X is therefore not separable.
X is neither second countable, nor Lindelof (Exercise 6).
Proposition 1. Any second countable space is first countable.
Proof. Let X, r be any second countable space, and let (B be a countable
basis for r. Then the collection of sets of the form

91* = {B G (B |x E B} for all xe X

forms an open neighborhood system for r (Chapter 3, Proposition 6).

Since (B is countable, each 3lx is also countable. Therefore X is first
countable.
Proposition 2. Any second countable space is separable.
146 Covering Properties 7.2

Proof. Let X, r be a second countable space, and let (Bbe a countable basis
for r. For each B E (B, select x# E Then (x# |5 E ©} is a countable
subset of X. The proof that it is also dense is left as an exercise.
Proposition 3. Any second countable space is Lindelof.
Proof. Let X, r be a second countable space with (B as a countable basis.
Suppose {Ui}, i E I, is any open cover of X. We select a subcover of {Ui},
i E /, as follows: Number the elements of (B sequentially, that is, B\,
B 2, . . . , Bn, . . . Select Bk from (B if there is a member Ui of the open
cover such that Bk C Ui. For each Bk selected, choose one Ui for which
Bk C U{ and call it Ukv Since the collection of Bk selected must be count­
able, the collection of Uk{ is also countable. It remains to be shown that

{Uk{ |Bk was selected}

is actually a subcover of {Ui}, i E I. Since {Ui}, i E I , is an open cover

of X and each Ui is the union of elements of (B, the collection of selected
Bk actually forms a refinement of {Ui}, i E I; therefore {Uk{ \Bk was
selected} is an open subcover of {Ui}, i E /.
For general topological spaces, no other implications hold between
Lindelof, first and second countable, and separable, other than those
given in Propositions 1, 2, and 3.
The following proposition describes how these properties behave with
respect to subspaces and product spaces.

Proposition 4
a) Any subspace of a first countable space is first countable.
b) Every subspace of a second countable space is second countable,
and hence is also separable.
c) Every closed subspace of a Lindelof space is Lindelof; however, it
is not true that every subspace of a Lindelof space is necessarily
Lindelof.
d) The product space of a countable family of nonempty spaces is
second countable if and only if each component space is second
countable. (This is an example of a proposition which does not
generalize to the product of an arbitrary family of spaces.)
e) The product of a countable family of nonempty Lindelof spaces
is not necessarily Lindelof, but if a product space is Lindelof and
each component space is T\, then each component space is also
Lindelof.
f) Any open subspace of a separable space is separable.
g) The product of a countable family of nonempty spaces is separable
if and only if each component space is separable.
148 Covering Properties 7.2

The next proposition shows that in metric spaces, the properties of

being Lindelof, separable, and second countable are all equivalent. We
have already seen that any metric space is first countable. Example 6 to­
gether with Exercise 6 furnishes an example of a metric space which is not
second countable.

Proposition 5. If X, D is a metric space, then the following statements

are equivalent:
a) X is Lindelof. b) X is separable. c) X is second countable.
Proof. Since it has already been shown in Propositions 2 and 3 that any
second countable space is both separable and Lindelof, it will suffice to
show that if X is either separable or Lindelof, then X is second countable.
Statement (b) implies statement (c). Suppose X is separable and let
{xn |n E N} be a countable dense subset of X. Let B(n, m) = N(xn, 1/m),
where m and n are in N. We shall show that

(B — {B(n, m) |n,m£ N}

is a basis for the metric topology on X. Let U be any open subset of X

and let x be any point of U. Since U is open, there is a positive number p
such that N(x, p) C U. Choose any integer m > 2/p. Since N(x, l/2m)
is open and {xn \n E N} is dense, there is some

xn E N(x, 1/2m)

(Proposition 14, Chapter 3). Then x E N(xn, 1/ra). Since m > 2/p,
1/m < p/2; thus N(xn, l/m) C N(xf p). Therefore N{xn, 1/m) C U as
well. But then U is the union of members of (B [for x was an arbitrary
element of U and N(xn, 1/m) E (B]. Since U was an arbitrary open set, (B
is a basis for the metric topology. Moreover (B is countable, and hence X
is second countable.

Statement (a) implies statement (b). Suppose X is Lindelof. Choose

some p > 0, and let E be a maximal subset of X having the property that
D(a, b) > p for all a , b E E . Such a maximal subset can be shown to
7.2 Countability Properties 149

exist by Zorn’
s lemma. For each a e l? , consider

N(a, p/2) and V = X — \j {Cl N(a, p/4) |a <E E}

(Fig. 7.1). V is open (Exercise 2). Then {V} U {N(a, p/2) |a E E} is a

covering of X by open sets. Since X is Lindelof, there is a countable sub­
covering. But if N(a, p/2) were omitted from the original cover for any
a e E, the remaining sets would fail to cover X since none of them would
contain a. Therefore {N(a, p/2) |a e E} must itself be countable; hence
E is countable.
Carry out the construction described above for p = l/n, n =
1, 2, 3, . . . , and get {En}, n E N, where E n is the set corresponding to
p = l/ n ; that is, E n is a maximal set having the property that D(a} b) >
l/n for any a , b E E n. Let S = Ujv E n. Since S is the union of countably
many countable sets, S is countable. We now show that S is dense in X.
Suppose x e X and q > 0; we will show that there is z e S such that
z e N(x, q). Take n > 1/q. Then there is z e E n such that z E N(x, q).
For if not, then x has the property that D{xyw) > l/n for each w E E n,
but x & E nj and thus E n would not be maximal. If U is any nonempty
open subset of X, choose x e U and q > 0 such that N(x, q) C U. Then
N(x, q), and hence U, contains an element of S. Therefore S is dense
(Proposition 14, Chapter 3).
Corollary. The space R of real numbers with the absolute value
topology is second countable (Example 5) as is the product space R n
for any n (Proposition 4d).
We close this section with a proposition that will be needed in the
proof of a key result in a later chapter.
Proposition 6. A T3 Lindelof space is T±.
Proof. Let X, r be a Ts Lindelof space and let A and B be disjoint closed
subsets of X. If x G A, then X — B is a neighborhood of x. Since X is
T3j there is a neighborhood Ux of x such that Cl Ux C X — B (Chapter 5,
Proposition 4). Similarly, if x E B, there is a neighborhood Ux of x such
that Cl Ux C X — A. If x is not an element of either A or B, then X —
(A U B) is a neighborhood of x; hence we may find a neighborhood Ux of
x such that Cl Ux C X — (A U B) (and thus Cl Ux D (A U B) = </>).
The family of Ux for each x e X is an open cover for X. Since X is Lindelof,
this cover has a countable subcover {UXn |n = 1, 2, 3, . . .}.
Let Ui, U2, . . . be the UXn (relabeled for convenience) which meet A,
and let Fi, V2, . . . be the UXn which meet B. Then for each positive
integer n, Cl Un fl B = </> and Cl Vn n A = 0; moreover A C Uat Un and
B C Uat Vn. Define Wi - Ux and set Yx = Vx - Cl Wx. Let W2 =
U2 - Cl Yx and Y2 = V2 - (Cl Wx U Cl W2). Suppose Wn and Yn
150 Covering Properties 7.2

have been defined. Then set

Wn+1 = Un+1 - (Cl y, u Cl f 2 U •••U Cl Fn)
and
Yn+l = Vn+1 - (Cl Wt U Cl W 2U•••U

Wn is always an open set since

Wn= n (X - (Cl Fj u •••U Cl F„

Un _x))
= Un D (X —C1(FX u •••U F„ _x));
hence Wn is the intersection of two open sets, and is therefore open. Similar
reasoning shows that Yn is open for each n.
Set H = Uj\r Wn and K — Uv Yn. Since H and K are the union of
open sets, they are open. Suppose o E i . Then a E Un for some n, and

Wn = Un ~ ( C l Yx U •••U C l F n _x).

But for any k, Cl Yk C Cl Vk and Cl Vk n A = <t>. Therefore a & Cl Yk

for any k. We have then that a E Wn. Therefore A C Uat 1F„ = H.
Similarly, B C K. In order to show that X is T4, we now have merely to
prove that H n K = 0.
Suppose x E H n K. Then x E Wn n Ym for some m and n. Suppose
m > n. Then

x E Ym = Vm — (ClWx U •••U C l Wn U •••U C lW m);

hence x could not be in Cl Wn, a contradiction. On the other hand, if

m < n, then

X E Wn = Un - (Cl Yx U •••U Cl Ym U •••U Cl Fn_!).

Thus x 2 Cl Fm, again a contradiction. Therefore H and K are disjoint

open subsets of X, which contain A and B , respectively, and hence
X is TV

Example 7. Let R be the set of real numbers with the topology r as

described in Example 11 of Chapter 3. The set Q of rational numbers is a
dense subset of R since any basis element of r [i.e., an interval of the form
[a, 6)] contains a rational number. Therefore R is separable. R is also
first countable [for each x E R, set = {[x}x + 1/n) |n E N}\. R can­
not be second countable, however, For if R were second countable, then
the product space R 2 would also be second countable and hence Lindelof.
But product space R 2 was shown in Example 13 of Chapter 5 to be T3,
but not 7Y If R 2 were Lindelof and T3, then by Proposition 6 it would
have to be 7Y
 7.2 Countability Properties 151

EXERCISES

1. Prove that the set {xb |B E (B) in Proposition 2 is dense in X.

2. The following refer to the proof of Proposition 5.
a) Prove that there is a maximal subset E as claimed.
b) Prove that the set
V= X U
-
is open. There are a number of possible approaches to this problem. One
approach, for example, is to show that given any w E V, N(w, 1) inter­
sects Cl N(a, p/4) for at most finitely many a E E. This means that there
is p' > 0 such that N(w, p') does not intersect any of the Cl N(a, p/4).
3. Prove (g) of Proposition 4.
4. By Proposition 4(g), R2 as described in Example 7 is separable. Let

)| x -f y= 0} C

Prove that A is a nonseparable subspace of R2. Is A closed? Is .4 Lindelof?

Give another proof that R2 is not Lindelof without appealing to Proposi­
tion 6.
5. A point x of a space X, t is said to be a condensation point of a subset A of X
if each neighborhood of x meets A in uncountably many points. Let A~
denote the set of condensation points of A. Suppose X is a Lindelof space.
Prove that if A is uncountable, then A~ ^ <j>. [Hint: Try to construct a
countable open cover of X each of whose members contains countably many
of the elements of A, and hence arrive at the contradiction that A is countable.]
6. Let X be an uncountable set with the discrete topology. Prove that the
collection of as described in Example 6 forms an open neighborhood system
for the discrete topology. Find a metric on X which induces the discrete
topology. Prove that X is not Lindelof, and hence that X is neither separable
nor second countable (Proposition 5).
7. Let X be the set of continuous functions from the space R of real numbers
with the absolute value topology into itself. For each / E X and p > 0,
define
N(f, p)= {geXI\f(x) — I < pfor all G

The family of N(f, p) for all / E X and all p > 0 forms the basis for a topology
t on X. Try to determine if X, r is second countable. Let

Y — {/ E X |/ has derivatives of all orders at each x E X}.

Is Y a second countable subspace of X?

8. Prove directly, that is, without using Proposition 6, that the space R of real
numbers with the usual absolute value metric topology is second countable.
[Hint: Prove that {N(x, q) |q > 0, q and x rational} gives a countable basis.]
152 Covering Properties 7.3

7.3 COMPACTNESS

The most important of all covering properties is compactness. As was

pointed out earlier in this chapter, compactness was not originally viewed
as a covering property, but it is through the use of coverings that com­
pactness can be stated in its most workable form. Compactness is, like
Lindelof, a cardinality condition.

Definition 4. A space X , r is said to be compact if given any open cover

{Ui}, i G 7, of X, there is a finite subcover of {Ui}, i e 7.
Suppose X , r is any space and A C X. An open cover of A is a col­
lection {Ui}, i G 7, of open subsets of X whose union includes A.
Equivalently, {Ui}, i G 7, is an open cover of A if {Ui n A}, d e 7,
is an open cover of the subspace A. A is said to be compact if every
open cover of A has a finite subcover. Equivalently, A is compact if
the subspace A is compact.
Note that in order for a space to be Lindelof, any open cover had to
have a countable subcover. In order for a space to be compact, any open
cover has to have a finite subcover. Certainly then, any compact space is
also Lindelof.

Example 8. The open interval (0, 1) with the absolute value topology is
Lindelof since it is a subspace of a second countable space R. The interval
(0, 1) is not compact, as we see from Section 7.1, Exercise 4. Another
example of a Lindelof space which is not compact is any countably infinite
set with the discrete topology.
An example of a compact space is the space presented in Example 3.

Proposition 7. The subspace [0, 1] of the space R of real numbers with

the absolute value topology is compact.
Proof. Let {Ui}, i G 7, be an open cover of [0, 1], where each Ui is open
in R. Let

T = {x G [0, 1] |finitely many of the Ui cover [0, x)}.

Then T ^ </>and 1 is an upper bound for T. Therefore T has a least upper

bound, say u. If w = 1, we are done (since finitely many of the Ui cover
[0, 1), and hence at most one more of the Ui will be needed to get a finite
cover of [0, 1]). Suppose then 0 < u < 1. Then either u G T or u g T.
Case 1. u G T. Then finitely many of the Uif say Uiv ..., Uin cover
[0, u). There is, however, UV such that u G U^; therefore

Wi', Uil§..., Uin}

7.3 Compactness 153

Ui'

Figure 7.2

is an open cover of [0, wj. It is then clear (Fig. 7.2) that u could not be an
upper bound for T.
Case 2. u & T. Then there is Uj such that u G Uj and finitely many of
the Ui do not cover [0, u) — Uj. Therefore u is not the least upper bound
for T.
Both cases have led to contradictions; hence it could not be that
0 < u < 1. Therefore u — 1, and hence [0, 1] is compact.
We now derive some important criteria for compactness.
Proposition 8. Let X , r be any topological space. Then X is compact if
and only if given any family {F{}, i E I, of closed subsets of X such
that the intersection of any finite number of the F{ is nonempty,
fl / Fi 7* </>.
Proof. Suppose X is compact and let {Fi}, i E /, be any family of closed
subsets of X such that fl / Fi = <t>. Set Ui = X — Fi for each i G I. Then

X — C\Fi = X — 4 > = X = U ( X — Fi) = \JUi.

i i i

Each Ui is the complement of a closed set and hence is open. Therefore

{Ui}, i G I, is an open cover of X. But X is compact; hence there are
finitely many of the Ui, say Ui„ . . . , J7*n, which cover X. Then

Fi i n • • • n Fin = </>.

We have proved that if X is a compact space, then given any family {FJ,
1 E /, of closed subsets of X whose intersection is empty, the intersection
of some finite family of Fi is empty.
Suppose X has the property that if the intersection of any family
{FJ, i G /, of closed subsets of X is empty, the intersection of finitely
many of the F* is empty. Suppose {Ui}, i G /, is any open cover of X.
Then X = U/ Ui. Therefore setting Ft = X — Ui} {Fi}, i G I, is a
family of closed subsets of X whose intersection is empty. Hence we can
find finitely many of the Fi, say F^, ... , Fin, such that

Fi, n • • • n Fin =
Then {Uiv ... , Uin} is a finite subcover of {Ui}, i e I. Therefore X is
compact.
Proposition 9. A space X, r is compact if and only if every net in X has
a limit point.
154 Covering Properties 7.3

Proof. Suppose X is compact and let {xf}, i e 7, be any net in X. Define

Bj = |j < z). Then {Cl Bj}, j e J , has the property that the inter­
section of any finite family of the Cl is nonempty. Since X is compact,
by Proposition 8, f l j Cl Bj ^ <£. Choose y in this intersection. We now
show that y is a limit point of {xf}, i e 7. Since 2/ G Cl for any j G 7,
any neighborhood 17 of y therefore contains at least one point of Bj. Sup­
pose U is a neighborhood of y and j and f are elements of 7. Since 7 is
directed, there is j " G 7 such that j < j " and f < j". But y G Cl B j>>,
and hence there is j such that

X - G f / n Bjr/.

T h en j < j, j' < j, and xj G U. Therefore {xf}, i e 7, is cofinally in U;

hence y is a limit point of {xt) , i G 7.
Suppose, on the other hand, that X has the property that every net in
X has a limit point. Let {TV}, i G 7, be any family of closed subsets of X
such that the intersection of finitely many of the T\ is always nonempty.
Let J be the set of finite intersections of the TV Then J is partially ordered
by < , where A < B if B C A ; moreover, J is then a directed set. Since
each member of J is nonempty, we can define a selection function 5 from
J into X such that s(A) e A for each A G J. Therefore {s a }, A G *7, is
a net in X, and hence has a limit point y. Consider any of the TV If A E J
and Fi < A, then A C TV Thus for each of the Ft-, the net {s^}, i G J,
is residually in TV Since 2/ is a limit point of {sa} , i G J, some subnet of
{sa}, A g <7, converges to y (Proposition 9, Chapter 6). But since {s^},
A G J, is residually in T\, such a subnet would be residually in 7\- for each i
(Proposition 3, Chapter 6). Then by Proposition 13, Chapter 6, y G 7\ for
each i; hence y e fl j TV Therefore fl/ T\ ^ </>. By Proposition 8, then
X is compact.

Corollary. A space X, r is compact if and only if every ultranet in X

converges.

Proof. If X is compact and {si}, i G I, is an ultranet in X, then {$*•},

i G 7, has a limit point. But an ultranet converges to any of its limit
points (Proposition 22, Chapter 6). Conversely, if every ultranet in X
converges and {s*}, i G 7, is any net in X, then some ultranet is a subnet
of {si}, i G 7 (Proposition 22, Chapter 6). Therefore {$*•}, i G 7, has a
subnet which converges to some point y\ hence y is a limit point of {$*},
i G 7 (Proposition 9, Chapter 6). Then X is compact by Proposition 9.

Example 9. We give another proof now that [0, 1] with the absolute value
topology is compact. Since [0, 1] is second countable or metric, we will
have shown [0, 1] is compact if we show that every sequence in [0, 1] has
a limit point. Suppose {sn}, n E N, is a sequence in [0, 1].
7.3 Compactness 155

Case 1. {sn}, n e N, is monotonically increasing, that is,

^ S 2 — ’* * — —

Then {sn |n G iV} has a least upper bound u, 0 < u < 1. If C/ is any
neighborhood of u, it is readily shown that {sn}, n G N, is residually in U,
and hence sn —»u. Therefore u is a limit point of {sn} ,n e iV .

Case 2. {sn}, n e AT, is monotonically decreasing, that is,

$1 > «2 > ‘ > «n > * ’*

Then |n E N} has a greatest lower bound v, 0 < v < 1; moreover

sn —> v. Therefore v is a limit point of {sn}, n E N.

Case 3. If {sn}, n e N, is either monotonically increasing or decreasing

from some point on, that is, for all but finitely many elements, then the
exceptional elements can be discarded without penalty, and Case 1 or
2 applied.

Case 4. {sn}, n e AT, is neither monotonically increasing nor mono­

tonically decreasing from some point on. Then {sn}, w G N, is “ cofinally”
increasing (the quotation marks here indicate that we are applying a
property informally to the sequence as a whole, rather than to individual
members); hence there is a monotonically increasing subsequence of {sn},
n G N. By Case 1, this subsequence converges to a point u of [0, 1]. But
then u is a limit point of {sn}, n e N.
Every sequence in [0, 1] has a limit point, and therefore [0, 1] is
compact.

EXERCISES

1. Prove that the set J in Proposition 9 is a directed set.

2. Prove that the sequences in Example 9 converge as claimed. Formalize the
argument in Case 4.
3. In Example 9, it is asserted that because [0, 1] is second countable, we need
only consider sequences. Prove: A second countable space X, r is compact if
and only if every sequence in X has a limit point.
4. Prove that a space X, r is compact if and only if every filter on X has a limit
point. Prove that X is compact if and only if every ultrafilter on X converges.
5. Let X, t be a space and let (Bbe a basis for r. Prove that X is Compact if and
only if every cover of X by members of (B has a finite subcover.
6. Decide which of the following spaces are compact.
a) the plane R2 with the topology which has for a subbasis

§
> = {U \ U = R2 — L, where L is any straight line}
156 Covering Properties 7.4

b) the plane R2, where an open set is any set of the form R2 — C, where C
contains at most countably many points of R2, and <£is open
c) the subspace of rational numbers in the usual space of real numbers
d) the space in Example 1, Chapter 6
7. In Section 6.5, Exercise 6, the notion of a bounded sequence in R, the usual
space of real numbers, was introduced. Let {sn}, n E N, be a bounded
sequence in R, and let .4 be the set of limit points of {$n}, n E N. Prove that
A U {sn |n E N} is compact.
8. Prove that the union of finitely many compact subsets of any space is com­
pact. Is the intersection of two compact subsets necessarily compact?
9. Prove or disprove: Let X be an infinite space with the property that the
only compact subspaces of X are finite subspaces. Then X has the discrete
topology.

7.4 THE DERIVED SPACES AND COMPACTNESS.

THE SEPARATION AXIOMS AND COMPACTNESS

It is not necessarily true that any subspace of a compact space is compact.

For example, (0, 1) is not compact (Example 8), whereas [0, 1] is compact
(Proposition 7). We do though have some information about which sub­
spaces of a compact space are compact.

Proposition 10. Any closed subset of a compact space is compact.

Proof. Let A be a closed subset of a compact space X, r, and suppose {Ui},

i E /, is any open cover of A. Then since A is closed, X — A is open;
hence {X — A} U {Ui |i e /} is an open cover of X. Since X is compact,
X — A together with finitely many of the [/*, say t/tl, . . . , Uin form a
cover of X. Therefore [U{v . . . , Uin] is a finite subcover of A, and hence
A is compact.
A partial converse to Proposition 10 is given by

Proposition 11. Any compact subset of a T2 space is closed.

Figure 7.3

Proof. Let A be a compact subset of a TVspace X, r (Fig. 7.3) and suppose

x E X — A. We must find a neighborhood of x which does not meet A
158 Covering Properties 7.4

Proof. Let {[/*•}, i E /, be any open cover of Y. Since / is continuous,

{f~~1(Ui)}y i E I, is an open cover of X. Since X is compact, we can find
finitely many XJi} say Uiv . . . , Uin, such that {f~l {Uh), . . .
is an open cover of X. But then {Uiv . . . , Uin} is a finite open subcover
of {[/*•}, i E I. Therefore Y is compact.
Since any homeomorphism is continuous, we have the following.
Corollary. If X, r is compact, then any space homeomorphic to X is
compact.
Example 11. Proposition 11 enables us to find many more compact spaces.
For example, if X, t is a compact space, R is an equivalence relation on
X, and X /R is the identification space, then X/R is compact, since the
identification mapping from X onto X /R is continuous. Since the circle
is an identification space derived from [0, 1] (Chapter 4, Example 14), the
circle is compact. The next proposition will give us even more compact
spaces.
Proposition 13 (Tychonoff theorem). Let Xj X i be the product space
of the countable family of nonempty spaces {Xi, t/ } , i E I . Then
Xj X i is compact if and only if each component space is compact.
Proof. Suppose X/ X{ is compact. Since the projection map

p n X X i^ X i
i
is continuous and onto for each i E I, X i is compact for each i E I (Prop­
osition 12).
Suppose each X i is compact. Let {sy}, j E J, be any ultranet in
X/ X i with the iih coordinate of Sj being denoted by Sj(i). Then

{Pi(Sj )} = fe'OO} > J E J,

is an ultranet in X i by Proposition 22, Chapter 6. Therefore {$>00} con­

verges in X i by the corollary to Proposition 9 of this chapter. But then
{s;}, i E I, converges in X/ X i by Proposition 12 of Chapter 6. Therefore
X/ X i is compact by the corollary to Proposition 9.
Example 12. We have already seen that the closed interval [0, 1] and the
circle C are compact. Using Proposition 13, we can now say that ([0, l])n
is compact for any n. Then cylinder C X [0, 1], the torus C X C, and the
cube ([0, l])3 are all examples of compact spaces (Figs. 7.4, 7.5, and 7.6).

Often one of the hardest steps in proving that some function is a

homeomorphism is showing that its inverse is continuous. The next
proposition affords us some relief in certain special (though important)
instances.
7.4 Derived Spaces, Separation Axioms, Compactness 159

{1} x c

{0} X C

Figure 7.4

Proposition 14. Let / be a continuous one-one function from a compact

space X, r onto a TVspace F, r '. Then / is a homeomorphism.
Proof. We must show that /“* is continuous. We use Proposition 8,
Chapter 4. Suppose F is any closed subset of X. Since F is closed, F is
compact (Proposition 10); hence f(F) is compact (Proposition 12). Then
f(F) is a compact subset of a TVspace and is therefore closed (Proposi­
tion 11). But
m = (rlr\F).
We have therefore shown that if F is any closed subset of X , (Z-1)”1^ )
is a closed subset of F. Therefore f~ l is continuous; hence / is a homeo­
morphism.
Because of the great importance of the Tychonoff theorem (Proposition
13), we now present another proof which does not depend on the material
of Chapter 6. We first prove another criterion for compactness.
Proposition 15. A space X, r is compact if and only if there is a sub-
basis S such that whenever 6 is a cover of X consisting of elements of
S, then (3 contains a finite subcover of X.
Proof. If X is compact, then r itself serves as a subbasis for r having the
required property.
Suppose now that X has a subbasis S having the property stated; we
now prove that X is compact. Let CL be any collection of open sets which
does not contain a finite subcover of X. We will show that CL cannot be a
cover of X, and, hence, indirectly show that any open cover of X contains
a finite subcover.
Let 3C be the collection of all (B C r such that CL C (B but no finite sub­
set of (B covers X. The set 3C is nonempty since CL G 3C; moreover, C is a
partial ordering of 3C.
Assume X is any chain in 3C, C. Then the union of the members of X
is easily shown to be a member of 3C and is an upper bound for X. There-
 160 Covering Properties 7.4

fore, by Zorn's Lemma, X contains a maximal element 91. Since ft C 91,

if we show that 91 is not a cover of X , then ft itself will not be a cover of X.
Suppose then that 91 is a cover of X. Then each x e X is in some
member of 91; assume x G M G 91. Now S is a subbasis for r and M is a
member of r, hence we can find finitely many members Si, . . . , S n of S
such that
s E S i n •••fl S n C M.

Suppose that no Si is a member of 91, i = 1,. . ., n. Then G r — 91

for i = 1, . . . , n. Because 91 is a maximal element of X, 91 U {Si} must
contain a finite subcover of X (or it would be a member of X which prop­
erly contains 91). Consequently, for i — 1, . . . , n, we can find a finite
subset 9li of 91 such that S», together with the elements of 91;, forms a
finite open cover of X. But since S x D •••fl S n C M, it follows that

{M} U ( 0 91^

forms a finite cover of X . This, however, is a contradiction, since 91

contains no finite subcover of X . This contradiction stems from the
assumption that no Si is a member of 91; therefore a
for some i = 1, . .., n.
The argument above shows that given any e l for which we have
some xEME91, there is some member E S for which E E 91. It
follows then that D 91 covers the same portion of X that 9
§
if 91 is a cover of X , then S n 91 is also a cover of X . But S f l 91 is a sub­
set of S, and thus contains a finite subcover of X . Therefore S n 91, and
hence 91, cannot be a cover of X since 91 contains no finite subcover of X .
We have shown then that any collection of open subsets of X which
does not contain a finite subcover of X fails to cover X . Therefore X is
compact.

Proposition 16 (Tychonoff product theorem, proof of which does not

use nets or filters). If {X t-, r t}, E is a nonempty family of non­
em pty com pact spaces, then the product space X/X,-, r is also compact.

Proof. The set

S= { p T \ U )| E T,-, E

forms a subbasis for the product topology r (recall that pi is the projection
into the ith component). Let (X be any collection of members of S which
does not contain a finite subcover of X/X,-, and for each i E I, set

(Xi = (C7 | U e r ^ p T 1EG}.

 7.4 Derived Spaces, Separation Axioms, Compactness 161

No fin ite subset of di can cover X *; for, otherwise,

{pr1 m,u e

w ould be a subcollection of Ot w hich covers X , and from w hich we could

obtain a fin ite subcover. Since no fin ite subset of d i covers X *, b u t X i is
compact, it follow s th a t di fa ils to cover X i fo r each i G I. Therefore fo r
i G I we can find
Xi e X i— U
L e t x be th a t p o in t of X/X,- w ith Xi as found in the previous sentence as
its zth coordinate. Then x is a p o in t of X w hich is not in the union of
members of d. Consequently, d does not form a cover of X . I t follow s
then th a t any collection of members of the subbasis S which covers X /X ;
contains a fin ite subcover of X /X *; hence by P roposition 15, X /X * is
compact.

EXERCISES
1. Decide which of the following spaces are compact. If practicable, sketch a
picture of the space. The set R of real numbers, the plane R2y or any sub­
space of these spaces will be assumed to have the usual metric topology.
Products will have the product topology.
a) (0, 1) X [0, 1]
b) C X R, where C = {(z, y) |x2+ y2 = 1} C R2
c) {(x,y,z)\x2+ y2+ z2 = 1} C R3
d) {(z, y) |x2+ y2 < 1} C R2
e) N X C, where N is the set of positive integers
f) (1, 2, 3, 4, 5} X C, with C as in (6)
2. Prove that any subset of N in Example 3 is compact.
3. There is a continuous function from [0, 1] onto [0, 1] X [0, 1]. Prove that
this function cannot be one-one.
4. It was shown that the product of normal spaces need not be normal. Prove
that the product of compact normal spaces is normal.
-5. Let / be the function from [0, 1] onto [0, 1] (with the absolute value topology)
defined by f(x) = sin (1/z) if x ^ 0, f(x) = 0 if x = 0. Prove that / is not
continuous. [Hint: Suppose / is a continuous function from a compact space
X onto a compact space Y. Define Gf = {(x, y) |y = f(x)}. Prove that if
/ is continuous, then Gf is a closed subset of the product space X X F. The
easiest way to effect this proof is through the use of Proposition 10, Chap­
ter 6. Then Gf is a closed subset of a compact 7Yspace if X and Y are both
compact and T2 (as in the case in this problem). Therefore what can be
said about Gy?]
162 Covering Properties 7.4

6. Prove that every compact metric space is separable.

7. Suppose X , D is any metric space. A subset 7 of X is said to be bounded if
Y C N(x, p) for some x G X and p > 0. Prove that any compact subset of
a metric space is both closed and bounded.
8. Suppose X, t is a first countable space such that X is T±and every compact
subset of X is closed. Prove that X is TV [Hint: Show that every convergent
sequence in X has a unique limit.]
9. Let Xj D be a compact metric space and let {sn}, n G N, be a sequence in X
such that given any p > 0, there is m G N such that if m < n and m < n',
then D(sn, sn>) < p. Prove that {sn}, nGN, converges in X. Prove that
this is not necessarily true if the assumption that X is compact is removed.
10. Prove or disprove: Suppose X and Y are both compact TVspaces and / is
a function from X into 7. Then / is continuous if and only if / considered
as a subspace of X X 7 is compact.
11. Suppose X, r is a space having the property that whenever a subset A of
X is compact, then X — A is also compact. Which of the following properties
must X also have,
a) T2 b) Ti c) Every subset of X is compact.
8
MORE ABOUT COMPACTNESS

8.1 COMPACTNESS IN Rn
The product space Rn of the space R of real numbers with the absolute
value topology with itself n times, better known as Euclidean n-space, is
perhaps the most important topological space of all (or, more accurately,
family of spaces, since there is a space for each positive integer n). Com­
pact subsets of Rn therefore hold a special place among compact sets and
warrant a special section to study them.
We have already seen that [0, 1] C R is compact. Thus any subspace
of Rn homeomorphic to [0, 1] is compact. More generally, any continuous
image of [0, 1] in Rn is compact. Using the various propositions already
proved, we can find many compact subsets of Rn. However, Rn has many
properties not shared by all topological spaces. We would therefore expect
there to be certain criteria for compactness which are more peculiar to Rn.
Proposition 2 gives such a criterion. Preparatory to Proposition 2, we
first prove the following.
Proposition 1. Suppose x = (xi, . . . , xn) and y = . . . , yn) are
any two points of Rn. Define D(x, y) = max(|x* — yi\, i = 1, . . . , n)
(icf. Examples 3 and 6 of Chapter 2). Then D is a metric on Rn. More­
over, the topology induced on Rn by D is the same as the product
topology on Rn.
Proof. The proof that D is actually a metric is straightforward and is left
as an exercise. Let r be the product topology on Rn and r' be the topology
induced by D. In order to prove r — r', we will use Corollary 1, Proposi­
tion 9, Chapter 3. Suppose x = (xi, . . . , xn) e Rn. Set
n
9U = i X N(xi, Pi) I where pi > 0,
L=i \

N(x{, p^ = (xi — pi, Xi + pi) C R, i = 1, . .., nj,

and
9^ = {N'ix,p )| p >0,
where N'(x, p) is the D-p-neighborhood of x in R71}.
163
164 More about Compactness 8.1

Then taking the collection of all 91* and the collection of 3l' for all e
we get open neighborhood systems for r and respectively.
Suppose '£31*. Then
N

N ' = X N(xi,p)
i=1
and hence is a member of 31*. (For a picture of a typical N* in R 2, the
reader should see Fig. 4, Chapter 2.) Suppose N £ 31*. Then

N = X N(xi, pi),
i= 1

where p* > 0, i = 1, . . ., n. Set p — min(p1, . . . , pn)• Then N'(x, p) e

91' and A'(x, p) C iV. Therefore by Corollary 1, Proposition 9, Chapter 3,
r = r'.
Proposition 2. A subset A of is said to be bounded if there is a
positive number p such that A C N'(0, p\ where 0 is the origin in Rn
and N'(0, p) is the D-p-neighborhood of 0 described in Proposition 1.
A subset A of Rn is compact if and only if A is closed and bounded.
Proof. Proposition 1 has shown that Rn with the product topology is a
metric space (with metric D as in Proposition 1). In Section 7.4, Exer­
cise 7, it was shown that any compact subset of any metric space is closed
and bounded.

Figure 8.1

Suppose A is a closed, bounded subset of Rn (Fig. 8.1). Then, since

A is bounded, A C N'(0, p) for some positive number p; hence

A C Cl N'(Of p).
Now

ACCl N'(D, )C X Cl
p 0
i—1 i—1
But [—p, p] is compact since it is homeomorphic to [0, 1]; hence

X CliV(0,p)
i= i
8.1 Compactness in /?n 165

is compact since it is the product of a family of compact spaces (Proposi­

tion 13, Chapter 7). Therefore A is a closed subset of a compact T2-space
(any metric space is T 2), and hence A is compact (Corollary 1, Proposi­
tion 11, Chapter 7).
Corollary. The closure of any bounded subset of Rn is compact.
Proof. Suppose A is bounded. Then

A c N ' ( 0 , p ) c C lN ' ( U , p + l).

Therefore Cl A C N'(0, p + 1). Cl A is closed, and thus Cl A is closed

and bounded, and is therefore compact.
The next proposition is true in any compact metric space, but has its
application primarily in the study of real functions.

Proposition 3. Let X, D be any compact metric space and suppose

{Ui}, i E I, is an open cover of X. Then there is a positive number p
such that N(x, p) C Ui for some i, for any x E X. That is, there is
p > 0 such that the p-neighborhood of any point in X is a subset of
at least one of the U{. Such a number p is called a Lebesgue number of
the cover, and is dependent on the cover for its value.
Proof. Each element x of X is contained in at least one U\, since {Ui},
i E I, is a cover of X. Since each Ui is also open, for each x E X, we may
select px > 0 such that N{x,px) C Ui for at least one of the Ui which
contain x. Since a selection has been made for each x, {N(x, px/2)}, x E X,
is itself an open cover of X (in fact, it is a refinement of the original cover).
Since X is compact, we can find a finite number of the elements of X, say
X\, . . ., xn, such that
{Nix!, pXl/2), . . ., pxJ2)}

is an open cover of X. Let

p = min(pXl/2, . . . , pxJ2).

We now show that p is a Lebesgue number for {Ui}, i E I. If x E X, then

x E N(xj, pXj/2) for some 1 < j < n. If z E N{x, p), then

D(z, Xj) < D(z, x) + D(x, xf) < p + Vxj/2 < pXj-
Therefore
N(x, p) c N(xj, pXj) C Ui
for some i.
We recall that the definition of continuity of a function from one
metric space to another can be expressed: A function /: X, D —> Y, D' is
8.1 Compactness in R n 167

Proof. Choose any p > 0. Then {N(y, p/2)}, y G 7, is an open cover of F.

Since / is continuous, {f~1(N(y, p/2))}, y E F, is an open cover of X.
Let q be the Lebesgue number of this cover in accordance with Proposi­
tion 3. It is left as an exercise to prove that this q has the desired property
that
f(N(x, q))C p)
for each x e X.

Corollary. Any function from a closed, bounded subset of Rn into any

metric space is uniformly continuous. In particular, any function from
a closed interval fa, b] C R is uniformly continuous.

EXERCISES
*
1. The following refer to the proof of Proposition 1.
a) Prove that D is a metric.
b) Prove that the collection of all 3lx and the collection of all 9lx, are open
neighborhood systems for r and r', respectively.
2. Complete the proof of Proposition 4.
3. A function / from a space X, r into the space of real numbers is said to be
bounded above if fix) < M for some number M and each x G X. What would
we mean if we said that / was bounded belowf Prove that if X is compact
and / is continuous, then / is bounded above and below. Prove that if M =
least upper bound {fix) |x £ X} and m = greatest lower bound {fix) |x £ X},
/ is continuous, then there are w and y in X such that f(w) = M and
fiy) = rn.
4. Which of the functions defined below from R into R are uniformly continuous?
a) fix) = x + 2, for all x £ R
b) fix) = 4x + 7, for all x £ R
* = I* sin (1/x), x ^ 0
•; } \0, if x = 0
5. Let X, D; Y, D and Z, D" be metric spaces. Decide which of the following
statements are true. If a statement is true, prove it; if false, find a counter­
example.
a) If the function / from X to Y and the function g from Y to Z are both
uniformly continuous, then fog: X —> Z is also uniformly continuous.
b) Suppose X and F are both the set of real numbers and D and Df are the
absolute value metric. Then if / and g are uniformly continuous functions
from X to F, then / + g defined by (/ + g)(x) = fix) + g(x) is also uni­
formly continuous.
c) If / is a homeomorphism from X onto F and / is uniformly continuous,
then f~ l is also uniformly continuous.
168 More about Compactness 8.2

6. A metric space X , D is said to be totally bounded if given any p > 0, the

open cover {N(x, p)}f x E X, has a finite subcover. Prove that any bounded
subset of Rm (with the metric described earlier) is totally bounded. Prove
that a compact subset of X, D is closed and totally bounded. Show that a
subset of X, D may be closed and totally bounded yet not be compact.

8.2 LOCAL COMPACTNESS

There are times when a topological space possesses some property “ locally”
which it does not have taken as a whole. For example, a second countable
space has a countable basis for its topology. A space X, r may not be
second countable, but could still have the property that there is an open
neighborhood system for r such that for any x E X, is countable; we
called such a space first countable. In a sense, a first countable space is
a space which is locally second countable. Similarly, a space may not be
compact, but still have the property that each point is contained in each
member of an “ appropriate”family of compact sets.

Example 3. Let R 2 be the coordinate plane with the Pythagorean metric

topology. Then R 2 is not compact, since it is not bounded (Proposition 2).
If x E R 2 and U is any neighborhood of x, then there is p > 0 such that
N(Xy p) C U. Then

N(x}p/2) C Cl N(x, p/2) C N(x, p) C U

(Fig. 8.3). But Cl N(xyp/2) is a closed, bounded subset of R 2y and hence

is compact. We have therefore proved that if x G R 2 and U is any neigh­
borhood of Xy then there is a compact set A [here Cl N(x, p/2)] such that
x E A° [here N(x, p/2)] C A C U. A similar property could be proved for
R 71, n finite.
This example inspires the following definition of local compactness.
8.2 Local Compactness 169

Definition 2. A sp a ce X , r is sa id to b e locally compact if g iv e n a n y

x G X an d a n y n e ig h b o rh o o d U o f x, th ere is a c o m p a c t se t A su ch th a t

x G A° c A C U.

Thus Rn is locally compact. The criterion for local compactness is

much simpler for T2-spaces, as we see from the next proposition.
Proposition 5. Let X, t be a TVspace. Then X is locally compact if
and only if given any x G X, there is a compact set A such that x G A°.
(In other words, the existence of one compact subset A of X such that
x G A° assures us that given any neighborhood U of x, there is a
compact set A' such that x G A'0 c i ' C U.)
Proof. Suppose X is locally compact and x G X. Since X is a neighborhood
of x, there is a compact set A such that

x E A° C A C X.

Suppose instead that given any x G X, there is at least one compact set A
with x G A°. Let U be any neighborhood of x. Then A° fl U is a neigh­
borhood of x and is a subset of U; hence we lose no generality in assuming
that U is already a subset of A°. Now A is compact and T 2, and hence the
subspace A is Ts (Corollary 2, Proposition 11, Chapter 7); moreover,
U D A = U (since U C A° C A) is a nonempty subset of A which is
open in A. Therefore there is V, open in both A and in X , such that

x G V C Cl V (in A) C U C A°

(Proposition 4, Chapter 5). Since A is a compact subset of a TVspace,

A is closed; thus
Cl V (in X) = Cl V (in A).

Then as a closed subset of a compact TVspace, Cl V is compact. Therefore

* G V = V°C Cl V C U,

and Cl V is compact; hence X is locally compact.

Corollary. Any compact TVspace X , r is locally compact.

Proof. X is a compact neighborhood of any x G X.
Example 4. The space X given in Example 10, Chapter 7 is only T\, but
is still locally compact. For if x G X, then any neighborhood U of X is
compact. Therefore x G U = U°C U and 47 is compact; hence X is
locally compact.
170 More about Compactness 8.2

£
Figure 8.4

Example 5. Let Q be the subspace of rational numbers in the space R of

real numbers with the absolute value topology. Then Q is not locally
compact. Let x E Q and suppose A is a compact subset of Q such that
x E A° (Fig. 8.4). Then A contains infinitely many elements of Q. There
is (a, b) CR such that x E (a, b) D Q C A°. Choose an irrational number
t E (a, b). We will now construct an open cover of A which has no finite
subcover. For each q E A, set

TUn\ _ [ {w ^R \q <
~ \ { z & R \ z <g} , if g <
Then {U(q) n A), q E A, is an open cover of A which has no finite sub­
cover. The proof of this fact is left as an exercise.
We see then that no element of Q can be contained in the interior of
any compact subset of Q. Therefore Q is an example of a metric space in
which not every closed bounded subset is compact. For example, [0, 1] D Q
is a closed, bounded subset of Q, but could not be compact; for if it were
compact, then ^ would be contained in the interior, (0, 1) n Q, of a com­
pact subset of Q.
We saw in Chapter 7 that any compact TVspace was TV Since local
compactness is a weaker property than compactness, we should expect
weaker results from local compactness than from compactness, as is the
case with the following.
Proposition 6. Any locally compact TVspace X , r is TV
Proof. We apply Proposition 4 of Chapter 5. If x E X and U is any
neighborhood of x, then there is a compact set A such that x E A°C A CU.
Since A is compact, A is closed, and hence Cl (A°) C A. Setting V = i°,
we have x E V C Cl V C U) where V is a neighborhood of x; therefore
X is TV
We see from Example 5 that a subspace of a locally compact space
need not be locally compact. We do, however, have the following prop­
osition regarding subspaces of locally compact spaces.
Proposition 7. If a space X, r is T 2 and locally compact, then so is
every open or closed subspace.
Proof. Suppose U is an open subspace of X and x E U. Then any neigh­
borhood V of x in U is also a neighborhood of x in X. Therefore there is a
compact set A such that x e A ° C A c F c C / . Hence U is locally com-
8.2 Local Compactness 171

pact. Note that this part of the proof did not depend on the fact that X
was T2; thus we have shown that any open subspace of any locally compact
space is locally compact.
Suppose F is a closed subspace of X and x E F. Let A be any compact
set such that x E A° C A. Since X is T2, A is closed. Then F n A is a
closed subset of the compact set A and hence is compact. But F fi A C F;
hence we also have x G (A n F)° in F C A n F C F. Since F is T 2, F is
locally compact by Proposition 5.

We now prove an even stronger result.

Proposition 8. A subspace F of a locally compact TVspace X, r is
locally compact if and only if it is the intersection of an open set and
a closed set.
Proof. Suppose Y is a locally compact subspace of X (Fig. 8.5). We will
prove that Y is open in Cl F; hence Y — U D Cl F, where U is an open
subset of X. Suppose y E Y; we must find a neighborhood of y (in Cl Y)
which is a subset of F. Since F is locally compact, there is a set U' open
in F such that y E U9 and Cl U' in F is compact. Then U' = Y n V,
where V is open in X. Furthermore, Cl U* in F = F n C1(F n V) is
compact, and hence is closed. Now

FnFcFnCl(FnF);
hence C1(F n V) C F. But

C 1 F n F c C 1 ( F n V).

For if z E Cl F n V and W is any neighbor­

hood of «, V fl W is a neighborhood of z. Since
z E Cl F, every neighborhood of z meets F,
and thus
Figure 8.5
(VnW ) nY = w n (Y nV ) * <t>.
But then every neighborhood of z meets F n V as well; hence
3E c i ( F n V).
Therefore Cl F n V C F. Thus Cl F n F is a neighborhood of 2/ in Cl F
such that CIFnFcF.
It is left as an exercise to show that the intersection of a closed subset
and an open subset of X is locally compact.
We now investigate the behavior of locally compact spaces with regard
to continuous functions. The following example shows that local compact­
ness, unlike compactness, is not preserved by continuous functions.
172 More about Compactness 8.2

A B
€ x>0

Example 6. Let A = {—1} and B = {x |0 < x} (Fig. 8.6). Let X =

A U B be given the absolute value topology. Then X is the intersection
of a closed subset of R , the usual space of real numbers, with an open
subset of R [for example, X = ({—1} U {x |0 < x}) Pi (R — {0})]; hence
X is locally compact. Define a function / from X into R 2 (with the Py­
thagorean topology) by

fM = 1(0, o), if e
^ ' |(x, sin l/x), if x E B.

Then / 1A and / 1B are both continuous, and A and B are both closed
subsets of X; thus/ is continuous (Proposition 11, Chapter 4). Let F be
the image of / considered as a subspace of R 2. The function / is a con­
tinuous, one-one function from X onto Y. But Y is not locally compact.
This can be seen from the fact that (0, 0) is not contained in the interior
of any compact subset of Y. This in turn follows from the fact that each
neighborhood U in F of (0, 0) contains a sequence which does not have a
limit point in Cl U.
A function /: X, r —> Y, r' is said to be open if whenever U is an open
subset of X, f(U) is open in Y (cf. Section 4.6, Exercise 2). With the added
assumption of openness, a continuous function will preserve local com­
pactness.

Proposition 9. If / is a continuous, open function from a space X, r onto

a space F, r', then if X is locally compact, F is also.
Proof. Suppose y e F and U is a neighborhood of y. We must find a
compact subset A of F such that y G A° C A C U. Let y = f(x) for some
x E X. By the continuity of /, there is a neighborhood V of x such that
/(F) C U. Since X is locally compact, there is a compact set B such that
x e B° C B CV . Then

fix) = yef(B°) C f ( B ) c U .

But/(J5°) is open, since / is open; and/(B) is compact, since B is compact

8.2 Local Compactness 173

and / is continuous. Therefore

y c °)cf(B) c u,
f(B

and hence Y is locally compact.

We now use Proposition 9 to study the relation between local compact­
ness and product spaces.

Proposition 10. Suppose Xj X i is the product space of the countable

family of nonempty spaces {Xi, t*}, z*G 7. Then Xj X i is locally
compact if and only if each component space is locally compact and
all of the component spaces except at most finitely many are compact.
Proof. Suppose X/ X i is locally compact. Then the projection

Pi \X X i —> X i
i
is continuous, onto, and open (Section 4.6, Exercise 2) for each i e 7.
Therefore by Proposition 9, each X i is locally compact. We must also
show that all but at most finitely many of the X i are compact. Let A be
any compact subset of X/ X i such that some point y of Xj X i is in ^4°.
Then there is a basic neighborhood X/ Vi of y such that Vi = X i for all
but at most finitely many i and

X Vi CA° CA .
i
We therefore see that P i ( A ) = X i for all but at most finitely many i .
Since pi is continuous and A is compact, X i is compact for all but at most
finitely many z.
Suppose each Xi is locally compact, and all but finitely many of the
X i are compact. Let y G X/ X i} and let yi be the zth coordinate of y. If
U is any neighborhood of y, then U contains a basic neighborhood of y
of the form X/ Vly where Vi is open in X i for each i G 7 and F t = X i for
all z G 7, except for at most finitely many, say z‘ i, . . . , zn. Since each X i
is locally compact, for each z G 7 there is a compact subset Ai of X i such
that yi G A° C Ai C V{. There are at most finitely many more i G 7, other
than z*i, . . . , zn, say in+i, . . . , im, such that X in+1, . . . , X im are not com­
pact. For any i not in {zh, . . . , zn, zn+i, . . . , zm}, we may let Ai = X i.
Then
ic (X A t-)° c X
°
A
eX
y c X F t-.

But Xj Ai is the product of compact sets and is therefore compact (Prop­

osition 13, Chapter 7). Hence Xj X i is locally compact.
174 M ore about Com pactness 8.3

EXERCISES

1. In Example 6 show that each neighborhood U of (0, 0) in Y contains a

sequence which does not converge to any point of Cl U(in Y). Why does this
prove that (0, 0) is not contained in the interior (in F) of any compact sub­
set of Y?
2. Prove that any subspace which is the intersection of a closed subset and an
open subset of a locally compact TVspace is locally compact, thus completing
the proof of Proposition 8. [Hint: Use Proposition 7.]
3. Provide the details for Example 5. In particular, show that {U(q) fl A},
q E A, is an open cover of A which has no finite subcover.
4. Which of the following subspaces of the plane R2 with the Pythagorean
topology are locally compact?
a) R2 - {(0, 0)}
b) {(x, y) |x and y are both rational}
c) {{x, y) |x and y are both integers}
d) R2 — Uw {C |C is a circle of radius 1/n with center (0,0)}, where N is
the set of positive integers
e) R2 — {(x, y) |x2+ y2 < 1, or £ = 0 or 1, and y — 0 or 1}
5. If X, t is a space and R is an equivalence relation on X, is the identification
mapping from X onto X/R, the identification space, necessarily open? If X
is locally compact, must X/R be locally compact? Give examples to prove
your points.
6. Is the union of finitely many locally compact subspaces of any space always
a locally compact subspace? Is the intersection of two locally compact sub­
spaces locally compact?
7. Suppose X, t is a locally compact TVspace which is second countable. Prove
that X is the union of countably many compact subsets A\, A2, . . . , An, . . •
such that An C A°+1. Example: R2 = Uw Dn, where Dn = Cl iV((0, 0), ft).
8. Find an example of a compact space which is not locally compact.

8.3 COMPACTIFICATIONS
Compact spaces are perhaps the most important of all topological spaces.
It is therefore of interest to know if and how any given space can be
embedded as a subspace of a compact space. If any space X, r can be em­
bedded as a subspace IF of a compact space F, r', then X can be embedded
as a dense subspace of some compact space. For IF is a dense subspace
of Cl IF; this follows from Propositions 13 and 14 of Chapter 3. But Cl IF
is a closed subset of a compact space and hence is compact. We there­
fore restrict our attention to considering whether or not a given space
can be embedded as a dense subspace of a compact space. Accordingly,
we make the following definition.
176 More about Compactness 8.3

Proposition 11. Let X, r be any T2- space. Then the Alexandroff com-
pactification F of X is a topological space and is a compactification
of X in the sense of Definition 3.
Proof. If X is already compact, the proposition is trivial. Suppose X is
not compact. If y E F, set

9lj, = {U | U is open in Y and y E U},

that is, 91y is the family of all neighborhoods of y. We will show that the
collection of 91y forms an open neighborhood system for r', a topology on F
in which the open sets are those described in Definition 4. Since, if y E X,
the neighborhoods of y in F are the same as the neighborhoods of y in X,
we need only consider the case when y = P, the ideal point. We now
verify Definition 5 of Chapter 3 for 9lp.
i) Since any one-point subset {x} of X is compact, X — {x} is a
neighborhood of P; therefore 9lp ^ <t>.
ii) By assumption, P e U for each U E 9lp.
iii) Suppose U and U' are neighborhoods of P. Then Y — U = K
and F — U' = K f, where K and K f are compact subsets of X.
Then
(F - U) u (F — U’
)= F - (U n U') = K U K’
.

But K U K' is compact since it is the union of two compact

sets (Section 7.3, Exercise 8). Therefore U f) U' is a neighbor­
hood of P.
iv) Suppose U is any neighborhood of P and z G U. If z = P, then
z e U G 91* and U CU. If z e X , then U — {P} is an open
subset of X; for X — U is compact, and X is T2. Therefore X — U
is closed; hence

X - (X - U) = V - {P}

is open in X, and hence also in F. Therefore

z El U — {P} C U and C7 - {P} G 91*.

The verification of (v) is left as a simple exercise. Therefore the

collection of 91* forms an open neighborhood system for a topology r' on F
which is precisely the family of open sets defined for F .
Since every neighborhood of any point in F meets X, X is dense in F .
It remains to be shown that F is compact. Let {Ui}, i G /, be any open
cover of F. Then P E Ui for some i, say V. Since Ui' is a neighborhood
of P, F — Ui> is a compact subset of X and {Ui}, i £ an °Pen cover
of F — U^. Then finitely many of the Ui, say Uiv ... , Uin, cover
8.3 Com pactifications 177

Y — Ui/; hence
{Ur, Uiiy. . ., Uin}
is a finite subcover of {Ui}, i G /. Therefore Y is compact.
Example 8. The one-point compactification of (0, 1) as in Example 7 is
the circle. Note that (0, 1) also has a two-point compactification [0, 1].
The one-point compactification of the space R of real numbers (with the
usual topology) is again a circle, since R is homeomorphic to (0, 1). It is
given as an exercise to prove that homeomorphic spaces have homeo­
morphic one-point compactifications. Usually the ideal point for the space
of real numbers is taken to be oo.
Example 9. If X is an infinite set with the discrete topology and P is an
ideal point for the one-point compactification Y of X, then the neighbor­
hoods of P will be all subsets of F which contain all but finitely many
points of X ysince the finite subsets of X are the only compact subsets of X.
Note that X is always an open subset of its one-point compactification,
since X is open in X. This implies that the subset containing only the
ideal point of any one-point compactification is closed. We thus see that
if X is T2j then its one-point compactification is at least T i. We now
investigate conditions under which a one-point compactification is T2.
Proposition 12. The Alexandroff compactification of any space X , r is
T 2 if and only if X is T2 and locally compact.
Proof. Suppose X is compact. Then the Alexandroff compactification
of X is X itself. Now X is T2 if X is T 2 and locally compact; on the
other hand, if X is T2, then X is T 2 and is also locally compact by the
corollary to Proposition 5. Assume that X is not compact, and let F be
the Alexandroff compactification of X. If F is T2) then X is T 2y since X
is a subspace of F. Now F is T 2 and compact and is therefore locally
compact. But X is an open subspace of F ; hence X is locally compact
(Proposition 7).
On the other hand, suppose X is T2 and locally compact. Let x and y
be distinct points of F. If x and y are both in X, then since X is T 2) there
are neighborhoods U and V of x and y, respectively, such that U n V = <t>.
Suppose x = P. Then y e X, and hence there is a compact subset A of X
such that y G A° C A. We have then that F — A is a neighborhood of
x = P and that A° is a neighborhood of y with

A° n (F - A) = 0.
Therefore F is T2.
Corollary 1. If X, r is a locally compact T2-space, then the one-point
compactification of X is normal.
178 More about Compactness 8.3

Proof. Any compact T2-space is both T x and P4 (Corollary 3, Prop­

osition 11, Chapter 7).

C orollary 2. Any locally compact TVspace X , r is regular.

Proof. The one-point compactification of X is normal, and hence is also
regular. Since X is a subspace of a regular space, X is regular.
Note that we had already proved Corollary 2 previously (Proposi­
tion 6), but that the use of comp actific at ions gives a simple, elegant proof
for the result. Note too that since we have found examples of 772-spaces
which are not locally compact (e.g., as in Example 5), we therefore have
the one-point comp actific ations of such spaces as examples of compactifica-
tions of T2-spaces which are not T2.

EXERCISES

1. Verify that the circle is the one-point compactification of (0, 1) (Example 8).
2. Prove that if a TV space A, r is homeomorphic to a space X r', then the
one-point compactifications of these spaces are homeomorphic. Show that two
spaces might have homeomorphic one-point compactifications even though
the spaces are not homeomorphic to one another.
3. Describe the one-point compactifications of each of the following subspaces
of R2 with the usual Pythagorean topology. Where practicable, sketch the
compactification.
a) {(2, y)| xG(0, 1],y= 0} b) {(1 1= 1, 2,3,
c) {(x, y)j x2+ y2< 1} d) {(2,+ < 1} U {(0, 1)}
e) {(2, y)j —1 < 2 < 1}
4. Which of the following are compactifications of {(2, -f- < 1} with
the Pythagorean topology? Each of the following spaces is to be considered
as a subspace of Euclidean n-space Rn for an appropriate n.
a) {(2, y, z)\x2+ y 2+ z2 = 1} b) {(2, | + 1, 0 < 2 < 1}
c) {(2, y) x2+ y2 < }1 d) {(2, y,+ + < 1}
e) {(2, y) H < 1, 12/| < 3}
6. Suppose / is a continuous function from a TVspace r into a T2-space Y,
Can / necessarily be extended to a continuous function from the one-point com­
pactification of X into F? Suppose/ can be extended to a continuous func­
tion F from Z, the one-point compactification of X, into Y. Is/(Z) necessarily
a compactification of/(A)? Is/(Z) necessarily the one-point compactification
of f(X) if it is a compactification?
7. Suppose/is a continuous function from A, r onto F, r', and let A' and F' be
the one-point compactifications of A and F, respectively. Define F: A' —> F'
by F(x) = f{x) if x G A, and/(P) = P', where P and P' are the ideal points
of A and F, respectively. Is F necessarily continuous?
8.4 Sequential and Countable Compactness 179

8. a) Describe the two-point compactification of the open interval (0, 1) with its
usual topology. How can this two-point compactification be defined
rigorously, that is, constructed from (0, 1).
b) Sketch the one-, two-, three-, and four-point compactifications of (0, 1) U
(2, 3) U (4, 5).

8.4 SEQUENTIAL AND COUNTABLE COMPACTNESS

There are certain properties which a space can have which are related to,
but do not have the full force of, compactness. We have already seen one
such property, local compactness. The purpose of this section is to intro­
duce two other such properties, sequential compactness and countable
compactness.

Definition 5. A space X, r is said to be countably compact if any count­

able open cover of X has a finite subcover. X is said to be sequentially
compact if every sequence in X has a convergent subsequence.
Any space which is compact is clearly countably compact. Also, since
any open cover of a Lindelof space has a countable subcover, a countably
compact Lindelof space is compact.
Proposition 13 will give another criterion for countable compactness.

Proposition 13. Let X , r be any space. Suppose B C X ; then a point

x e l i s said to be an accumulation point of B if every neighborhood
of x contains infinitely many points of B. Then X, r is countably com­
pact if and only if every countably infinite subset of X has at least one
accumulation point.
Proof. Suppose that B is a countably infinite subset of X, but that B does
not have an accumulation point, and assume X countably compact. Select
Xit x2, . . . , xn, . . . , a sequence of distinct points of B. Set

An = {xn) *^n+1? •••

/ and (Jn X An.

Given any point y e X, y is not an accumulation point of B ; hence there

is a neighborhood Uy of y such that Uy n B contains at most finitely many
elements. Therefore xn e Uy for at most finitely many n. If n is suf­
ficiently large, then Uy PI An = </>, and therefore Uy C Cn. Set Vn = C°.
Then {Fn}, n e N, is a countable open cover of X. But since X is countably
compact, there is a finite subcover, say { V . . . , Vm}, of {Vn}, n e N.
Now xm_|_i is not an element of C x U •••U Cm; hence xm+i cannot be an
element of V\ U •••U Vm, a contradiction since V\ U •••U Vm = X.
Consequently, if X is countably compact, B must have an accumulation
point.
180 More about Compactness 8.4

Suppose now that every countably infinite subset of X has an ac­

cumulation point, but suppose that we can find a countable open cover
{Un}, n G N, for which there is no finite subcover. Then for any n G N,
X — U?=i Uj ^ <t>. Pick xi G X — Ui; suppose x G Uni. Then pick

X2 G X (XJ1 U ••* U Uni).

Suppose Xk has been chosen and xk G Unjc•Choose

xk+ i G X - (£/i U •••U Unk).

By the manner in which they were chosen, these points must all be distinct;
thus the set
B = {xk |k = 1, 2, . . .}

is a countably infinite subset of X. Then B has an accumulation point y,

and y G Un for some n. But if k' is large enough, n < nk>; hence xk 2 Un
for k > k'. Hence Un is a neighborhood of y which meets B in only finitely
many elements, contradicting the fact that y is an accumulation point
of B. There must therefore be a finite subcover of {Un}f n G X, and hence
X is countably compact.
Proposition 14. If a space X, r is sequentially compact, it is countably
compact.
Proof. Suppose a space X, r is sequentially compact and B is any countable
infinite subset of X. Then we can find a sequence {sn}, n G X, where
sn G B for each n, and no two sn are equal. Then {sn}, n G N, has a con­
vergent subsequence; consequently, {sn}, n G X, has a limit point y
(Proposition 9, Chapter 6). But then y is accumulation point of B. Hence
X is countably compact by Proposition 13.
The reader may have already noted that the famous Bolzano-Weier-
strass theorem from real analysis is really a statement that any compact
subset of R or R 2 (that is, a closed, bounded subset) is sequentially compact.
The reader should also note that the distinction between countably
compact and sequentially compact is hairline thin. For in any countably
compact space, any sequence either takes some value infinitely often, or
else the set of points in the sequence is an infinite set and hence has an
accumulation point. This does not imply, however, that a sequence has a
subsequence which converges, but only that it has a subnet which con­
verges, and the distinction in this case is fine indeed (although we are
justified in saying that any first countable space is sequentially compact
if and only if it is countably compact). In “ nice”topological spaces,
sequential and countable compactness are equivalent; examples showing
that countable compactness does not imply sequential compactness are
rather esoteric.
8.4 Sequential and Countable Compactness 181

One of the nicest types of topological space is the metric space. We

have already seen that in a metric space the properties of being second
countable, Lindeldf, or separable are all equivalent. We now will show
that in a metric space the properties of being countably compact, se­
quentially compact, or compact are equivalent. We do this in two prop­
ositions.
P roposition 15. Any countably compact metric space X, D is separable.

Proof. For any positive number p, there is a maximal subset E p of X

such that for any a, b E E py D(a, b) > p. The proof will closely follow the
lines of the proof that (a) implies (b) in Proposition 5, Chapter 7. If E p
were infinite for any p > 0, then it would have an accumulation point y.
But then N(yyp/2) would contain infinitely many points of E p; hence any
two of these points would be closer together than py a contradiction.
Therefore E p is finite for each p. However, given any x E X,
N(x, p) n E p ^ <t>,
or we would have a contradiction to the maximality of Ep.
Take E\/n for each positive integer n. Then Ujv E j/n is a countable
dense subset of X, and hence X is separable.

C orollary 1. Any countably compact metric space is compact.

Proof. By Proposition 5, Chapter 7, any separable metric space is Lindelof.

But any countably compact Lindelof space is compact.
Since any sequentially compact metric space is countably compact
(Proposition 14), we have the following.
C orollary 2. Any sequentially compact metric space is compact.

P roposition 16. If X, D is any metric space, then the following state­

ments are equivalent:
a) X is compact.
b) X is countably compact.
c) X is sequentially compact.
Proof. It remains to be shown that if X is countably compact, then X is
sequentially compact. Suppose that X is countably compact and {s„ },
n e N , is a sequence in X. If sn = y for infinitely many n y then a subse­
quence of {sw}, n E N y converges to y (Exercise 1). Suppose then that no
value is assumed by the sn more than a finite number of times; in fact, we
lose no generality in assuming that the sn are all distinct. Since {sn \n E X}
is an infinite set, it has an accumulation point y. Set Un = N(y, 1/n) for
each n E N. Then Un fl {sn} is infinite for each n E N. Choose

i*
182 More about Compactness 8.4

Choose sn2 E {sn} n U2)ni < n2. In general, choose

snk E {$n} n Ufa njc—i ^ k

w, •

Then {sni, . . . , snjfc, . . .} is a subsequence of {sn}, n E N ywhich converges

to 2/; hence X is sequentially compact.
There are a number of other properties which are in some way related
to compactness, for example, metacompactness, pseudocompactness, and the
very important property of paracompactness. We shall examine this latter
concept briefly in Section 10.5.

EXERCISES

1. Prove that if a sequence assumes some value infinitely many times, then a
subsequence of the sequence converges to that value.
2. Prove that in first countable spaces, countable compactness and sequential
compactness are equivalent.
3. Suppose / is a continuous function from a space X , r onto a space T, r'. Prove
that if X is countably compact, then Y is also. Prove that if X is sequentially
compact, then Y is also.
4. Prove that a closed subset of a countably compact or sequentially compact
space is also countably or sequentially compact.
5. Suppose / is a continuous function from a countably compact space X, r into
the space R of real numbers with the absolute value topology. Prove that
there are numbers m and M such that m < f(x) < M for all x E X, that is,
prove that / is bounded.
6. Let X be any set and suppose r and r' are two possible topologies on X.
Suppose X y t is compact and t ' is coarser than r. Is X, t compact? Suppose
X, r is countably compact or sequentially compact and t' is coarser than r.
Is X, t necessarily countably compact or sequentially compact?
7. Find an example of a separable metric space which is not countably compact.
Is it possible to have a nonseparable metric space which is countably
compact?
8. Need the set of accumulation points of any set be closed? Need a set together
w ith its accumulation points be closed? Prove that any set together w ith its
accumulation points is closed provided the space is T2.
184 Connectedness 9.1

Proof. Suppose [0, 1] is disconnected. Then [0, 1] = U U F, where U and

V are disjoint, nonempty, open subsets of [0, 1]. Suppose u £ U and v E V.
We may assume u < v (relabeling U and F, if necessary). Let S be the
set of numbers s such that s < u or [«, s] C U. Then S has 1 as an upper
bound, and hence S has a least upper bound a with 0 < a < 1. Since
[0, 1] = U U V, either a G U or a e F. Suppose a G U. Since U is open,
there is p > 0 such that (a — p, a + p) C U. Then

[a — p/2, a + p/2] C U

(Fig. 9.1); hence a + p/2 G S. This contradicts the assumption that a is

an upper bound for S. Suppose a e V. Then there is p > 0 such that
(a — p, a + p) C V, and thus a — p/2 is an upper bound for S , con­
tradicting the assumption that a is the least upper bound. Therefore
[0, 1] is not disconnected; hence [0, 1] is connected.

Figure 9.1

Note that since connectedness has been defined in a negative way,

that is, a space is connected if it is not disconnected, most proofs that a
space is connected are by contradiction: a space is assumed to be dis­
connected and a contradiction is proved.
Connectedness, like compactness, is a property which is preserved by
continuous functions.
P roposition 2. Suppose / is a continuous function from a space X , r
onto a space F, r'. If X is connected, then so is F.
Proof. Assume F is not connected. Then F = U U F, where U and F
are nonempty, disjoint, open subsets of F. Then/-1^ ) and/-1^ ) are
disjoint, nonempty (since / is onto) subsets of X whose union is X. But
since / is continuous, f ~ l(U) and f~~l(V) are also open subsets of X; there­
fore X is disconnected, a contradiction. Hence F must be connected.
C orollary 1. Any closed interval in R , any closed line segment in R 2,
and, in general, any image of [0, 1] under a continuous function is
connected.

C orollary 2. If X, r is any connected space and R is an equivalence re­

lation on X, then the identification space X/R is connected.
Proof. The identification mapping from X onto X/R is continuous.
9.1 The Notion of Connectedness 185

Exam ple 2. Since a circle in R 2 (with the Pythagorean topology) can be

thought of as an identification space formed from [0, 1], the circle is also
connected.

We now give some more criteria for connectedness.

P roposition 3. Let X,t be any topological space. Then the following

statements are equivalent.

a) X is connected.
b) X cannot be expressed as the union of two disjoint, nonempty,
closed subsets.
c) The only subsets of X which are open and closed are X and <f>.
d) If A is any subset of X other than X or <£, then Fr A ^ <£.
e) Let Y = {0, 1} have the discrete topology. Then there is no con­
tinuous function from X onto Y.

Proof. Statement (a) implies statement (b). Suppose X = A u B, where

A and B are disjoint, nonempty, closed subsets of X. Then X — A = B
and X — B = A are both the complements of closed sets, and hence are
open. Thus X = A U B is also the expression of X as the union of two
disjoint, nonempty, open subsets of X. Hence X is not connected.
Statement (b) implies statement (c). Suppose 4 is a subset of X which
is both open and closed, but that A is neither X nor 4>. Then X — A is
also open and closed, and nonempty. Thus

X = (X - A) U A

is the expression of X as the union of two disjoint, nonempty, closed sub­

sets, contradicting (b).
Statement (c) implies statement (d). If A is a subset of X other than
X or 0, and Fr A = <£, then since Cl A — 4 ° u F r i, w e have Cl A = A°.
On the other hand, A° C A and A C Cl A, and hence ^4 = ^4L°= Cl ^4;
thus A is both open and closed in X. Therefore if (c) holds, there can be
no subset A of X, other than X or <f>, such that Fr A = <j>.
Statement (d) implies statement (a). Suppose X = U U V, where U
and V are disjoint, open, nonempty subsets of X. Then U and V are also
closed. Therefore
U = U° = C l V.

But Fr U = Cl U — U° (Proposition 12, Chapter 3); hence

Fr U = U - U =
a contradiction of (d).
186 Connectedness 9.1

Statement (a) implies statement (e). Suppose there is a continuous

function from X onto Y = {0, 1}. Then since X is connected, Y must be
also, which is not the case.
Statement (e) implies statement (a). Suppose X is disconnected.
Then X = U U F, where U and V are nonempty, disjoint, open subsets
of X . Define g: X —»Y by g(x) = 0, if x G 17, and by gf(z) = 1, if x G V.
Then ^- 1({0}) = U and 0- 1({l}) = F; hence g is continuous, a con­
tradiction of (e).

We now see why the extension F in Example 15 of Chapter 5 cannot

be continuous.

Example 3. There are a number of ways that we can show that the space
N in Example 3 of Chapter 7 is connected. For example, let A be any
subset of N. Suppose A is infinite, but not all of N. Since a subset of N is
closed if and only if it is finite, the only closed set which contains A is N]
hence Cl A = N. Then

Fr A .= N — A t6 <t>

since A 5* N. Suppose A is finite but nonempty. Then A is closed and

thus Cl A = A. But since A excludes infinitely many elements of N,
A° = <t>; hence Fr A = A 5* <t>. By (d) of Proposition 3, N is connected.

Proposition 4. Suppose X, r is a space such that X = U U F, where U

and F are disjoint, open, nonempty subsets of X. Let A be any con­
nected subspace of X. Then either A C V or A C F.

Proof. If A n U <t> and A n F ^ 0, then A n U and A n F are non­

empty, disjoint subsets of A which are open in A. But

i = ( i n f / ) u ( i n F);
thus A is not connected. Therefore either A (1 U = <f> and hence A C F,
or A n F = <t> and hence A C U.

Example 4. Let R be the space of real numbers with the absolute value
topology. Then the removal of any point y from R disconnects R into two
“rays, ”open half-lines

H + = {x G R |y < x} and H~ = {x E R |x < y).

The removal of y also disconnects any interval which contains y as any­

thing but an endpoint. For if, say, [a, b] contains y in its interior and
[a, b] — {y} is connected, then [a, b] — {y} must lie entirely in either
H + or H~, an impossibility.
9.2 Further Tests for Connectedness 187

EXERCISES

1. Find another proof that the space N in Example 3 is connected.

2. Prove that each of the following subspaces of the space of real numbers with
the absolute value topology is disconnected.
a) any finite subset
b) (0, 1) U (6, 7) U (9, 18)
c) {x |x is irrational}
d) {x |x = l/n, where n is a positive integer or x = 0}
3. Modify the proof of Proposition 1 to show that (0, 1), and hence R, is
connected.
4. A space X, r is said to be totally disconnected if X is not connected and the
only connected subspaces of X are <f> and subspaces which consist of only
one point. Prove that each of the following spaces are totally disconnected.
a) the subspace of rational numbers in the usual space of real numbers
b) any discrete space of more than one point
c) the set of real numbers with the topology described in Chapter 3, Ex­
ample 11
5. Suppose A and B are connected subspaces of a connected space X } t . Show
by producing an example that the following need not be connected.
a) A D B b) A U B c) Fr A d) A°
6. Let X be a space with the property that given any x G X and any neighbor­
hood U of xythere is a neighborhood V of x such that Cl V is a proper subset
of U. Prove that X is connected. Would X necessarily be connected if Cl V
is replaced by V in the first sentence?

9.2 FURTHER TESTS FOR CONNECTEDNESS

In this section we continue to investigate criteria for determining if a
space is connected.
If the reader did Exercise 5 of Section 9.1 and his example for (b) was
correct, it must have been that A n B = </>, as we see from the next
proposition.
Proposition 5. If X, r is a space and X = U/ Ai, where {At}, i e /,
is a collection of connected subspaces of X, then if f l j Ai 9* <t>, X itself
is connected.
Proof. Suppose X = U U V, where U and V are disjoint open subsets of
X. Then for each i, either Ai C U or Ai C V (Proposition 4). If some
A iC Uy then since f l j Ai ^ 0, some element from each Ai must be in
U, and hence every Ai is in U. Therefore we would have U / 4 t = I c P
and V = <j>. Similarly, if some A iC V, then X C V and U = </>. We have
188 Connectedness 9.2

therefore shown that X could not be expressed as the union of two dis­
joint, nonempty, open subsets; hence X is connected.
Example 5. We have seen that the space R of real numbers with the
absolute value topology is connected (Section 9.1, Exercise 3). Any
straight line in Euclidean n-space is homeomorphic to the real line R.
This implies that Rn is connected for any n, since Rn is the union of all
straight lines in R n which pass through the origin. The family of such lines
therefore fulfills the hypotheses of Proposition 5.
Proposition 6. Let X, r be a space such that any two elements x and
y of X are contained in some connected subspace of X. Then X is
connected.

Proof, Let x be a fixed element of X. For any t/ e X , let C(x, y) be a

connected subspace of X which contains x and y. Then {C(x, y)}, y G X,
is a family of connected subspaces of X whose union is X and whose inter­
section is nonempty (since the intersection at least contains x). Prop­
osition 5 tells us that X is connected.

Figure 9.2

Example 6 . Any closed line segment in Euclidean n-space Rn is homeo­

morphic to the closed interval [0, 1], and is hence a connected subspace
of Rn. Using this fact, we can show that Rn — {P}, where P is any point
of R n and 2 < n, is connected. For suppose Q and Q' are any two points
of R n — {P}, Choose
Q " e R n - {P}

such that P & QQ" U Q"Q' (Fig. 9.2). Then QQ" U Q"Q' is connected by
Proposition 5, that is, it is the union of the connected subspaces

QQ", Q"Qf,
and
QQ" n Q"Qf = {£"} * 0.

Therefore Q and Qr are in the same connected subspace of R n — {P}.

By Proposition 6, then Rn — {P} is connected.
9.2 Further Tests for Connectedness 189

We have already seen that since [0, 1] is connected, any homeomorphic

image of [0, 1], for example, a closed line segment in Rn, is connected.
More generally, of course, any continuous image of [0, 1] is connected.
The continuous images of [0, 1] form an important class of spaces known
as paths. More formally, we make the following definition.
Definition 2. Let X , r be any space. A subspace Y of X is said to be a
path in X if there is a continuous function from [0, 1] (with the absolute
value topology) onto F. X is said to be path connected if, given any
two points x and y in X, there is a path in X containing x and y.
Suppose X = Rmy Euclidean m-space. A subset W of Rm is said to
be polygonally connected if given any two points x and y in W, there
are points
x0 = x, xlf . . . , xn—i , xn = y

such that Ui=1 Xi-iXi C Wywhere xi~[xi is the closed segment joining
x,*_i and Xi (Fig. 9.3).

Figure 93 Figure 9.4

Any subset of Rn which is path connected is not necessarily polygonally

connected. For example, the circle

{(x, y)\x2 + y2 = 1} C R 2

is path connected (it is itself a path), but it is not polygonally connected.

On the other hand, any subspace of R n which is polygonally connected is
path connected (Exercise 1). Paths can actually be rather exotic, and may
not look anything like [0, 1]. For example, it can be shown that ([0, l])n
is a path for any finite n.
The next example gives a connected subspace of R 2 which is not path
connected.
190 Connectedness

Figure 9.5 Figure 9.6

Example 7. Let

Y = {(x, y ) \ y = sin (1/x), x > 0} u {(0, 0)} c

(Fig. 9.4). We shall see from Proposition 11 that Y is connected. If P is
any point in Y other than (0, 0), then there is no path in Y which contains
(0, 0) and P. For if there were such a path, it would be possible to show
that the function / from the space of nonnegative real numbers in R for
which Y is the graph is continuous. But / is not continuous (Section 7.4,
Exercise 5). Also see Exercise 2 below.

Proposition 7. If X, r is a path-connected space, then X is connected.

Proof. Suppose X is path connected and x e X. For each i / G l , let
P(x, y) be a path which contains x and y. Then

X = U{P(x, y ) | y e l } and n y) \ y e X }

Therefore X is connected by Proposition 6.

Recall that a subset W of Euclidean n-space R n is convex if, given any
paints x and y of W, the closed segment xy is a subset of W. We note
that the basic neighborhoods in R n, considered as the n-fold product of R,
are convex subsets of R n (Fig. 9.5). The "open balls”of the form

{(X!, . . ., xn) I X? H----- f X* <

where p > 0, are also convex subsets of R n. Of course any convex subset
of R n is polygonally connected, and hence path connected, and therefore
connected (in a very "strong”way).
We now prove a theorem that has wide use in analysis.

Proposttfo* fc Let V be a connected open subset of R n. Then V is

polygonally connected.
Proof. Choose u e U. Let A = {a S U |a can be polygonally connected
to u in 17} (Fig. 9.6) and B — U — A. Then A is open. For since U is
open, given any a E A, there is a basic product neighborhood V of a such
192 Connectedness 9.3

5. Let R2 be the plane with the Pythagorean metric. Prove that R2 — C, where
C is any countable set, is polygonally connected. In particular, prove that
R2 — {(x, y) |x and y are rational}
is polygonally connected. [Hint: Through any point in R2 — < 7 , there is a
line which does not intersect C.]
6. Which of the following subspaces of R2 are connected? Indicate clearly how
you arrived at your conclusion.
a) {(x, y )|y = (1 /n)x,n = 1 ,2 ,3 ,...}
b) {( x ,)y |eitherx or y, but not both, isirrational}
c) {{x, y)xj ^ 1 }
d) {(x, y )|x ^ 1 } U {(0, 1)}
7. Suppose X, r is a space such that X = Ai U ■••U where each At is con­
nected and Ai—i fl Ai </>, i = 2, . . ., n. Is X necessarily connected?
8. Suppose A is a compact subspace of Euclidean n-space Rn, n > 2. Prove
that Rn — A need not be connected.
9. Prove directly (that is, do not refer to Corollary 2 to Proposition 11 of the
next section) that if X is connected, then the one-point compactification of
X is also connected.

9.3 CONNECTEDNESS AND THE DERIVED SPACES

We have already seen that if X, r is a connected space, then any identifi­

cation space derived from X is also connected. In this section we investi­
gate the behavior of connectedness as regards subspaces and product
spaces.
It is, of course, false that any subspace of a connected space is con­
nected. The following proposition gives a criterion for determining whether
or not a subspace of a given space is connected.
Proposition 10. If A is a subspace of the space X, r, then A is con­
nected if and only if A cannot be expressed as S U T, where S and T
are nonempty subsets of X and

S n C lT = C lS n T = < t> .

(Note that no demand is made that S and T be open or closed in A.)

Pr <or Tf A is not connected, then A = S U T, where S and T are dis­
joint, nonempty subsets of A which are open and closed in A. Suppose

x e S n Cl T.

Then since S C A, x G A n Cl T = Cl T in A (Chapter 4, Proposition

 9.3 Connectedness and the Derived Spaces 193

Figure 9.7

4) = T. Therefore x G S r \ T = < l > , a, contradiction. Then S n Cl T —<£;

similarly, Cl S n T — <t>.
Suppose that A — S U T, where Cl S fi T = S n Cl T — </>, and S
and T are nonempty. Then

C IS in A - A n C l S = ( S u T) n C I S
- (S n CIS) u (T n CIS) = s u0 - s.

Therefore S is closed in A; similarly, T is closed in A. Hence A is dis­

connected.
Corollary. Two subsets S and T of a space X, r are said to be mutually
separated if
Cl S D T = S n Cl T = <t>.
Suppose S and T are mutually separated subsets of X and A is a
connected subspace of S U T. Then either A C S, or A C T.
The proof of this corollary is left as an exercise.
Example 8. Let R 2 be the plane with the Pythagorean topology,
x, y)| X 2+ 1}
and
T = { ( x ,y ) \ (x - 2 ) 2 + y2 < 1}.

(Fig. 9.7). Then Cl S n T = Cl T n S = <£. Therefore S U T is a dis­

connected subspace of R 2. Note, however, that

C l S n C l T = {(1,0)} ^ 0.

Proposition 11. Suppose A is a connected subspace of X, r and

AcFcCl A.
Then Y is also a connected subspace of X.
Proof. If Y is disconnected, then F = S U T, where S and T are mutually
separated (Proposition 10). Since A is connected, either A C S or A C T,
by the corollary to Proposition 10. Suppose AcS. Then Cl A c Cl S;
194 Connectedness 9.3

hence F C Cl A C Cl S. But then since Y = S U T, T C Cl S. Since

T n Cl S = <£, we have arrived at a contradiction. Therefore Y is con­
nected.
Corollary 1. If a space X , r contains a connected dense subspace, then
X is connected.
Proof. Suppose A is a connected dense subspace of X. Then Cl A — X
is connected by Proposition 11.

Corollary 2. If X , r is connected, then any compactification Y of X is

connected.
Proof. If F is a compactification of X , then X is a dense connected sub­
space of F; therefore Y is connected, by Corollary 1.
Example 9. We see that the space Y in Example 7 is connected as follows:
Set A = Y — {(0, 0)}, and define a function h from {x |0 < x} C R
into R 2 by
h(x) = (x, sin (1/x)).

Then h is easily seen to be continuous; in fact, h is a homeomorphism onto

its image A. Therefore, since {x \0 < x} is connected, A is connected.
Now (0, 0) is in Cl A since every neighborhood of (0, 0) contains infinitely
many points of A (cf. Fig. 9.4). Then A C Y C Cl A ; hence, by Proposi­
tion 11, F is connected.
y

Example 10. Consider the graph F of the equation r = 1 — 1/A, A > 1,

in polar coordinates (considered as a subspace of R 2 with the usual topol­
ogy) (Fig. 9.8). It may be verified that

Cl F = F U {(r, A) |r = 1}

(again in polar coordinates.) Thus F together with any set of points on the
unit circle forms a connected subspace of R 2.
9.3 Connectednesss and the Derived Spaces 195

We now investigate connectedness and product spaces.

P roposition 12. The product space X/ X* of the countable family of
nonempty spaces
{Xi, T{}, i (E I,

is connected if and only if each X* is connected.

Proof. Suppose each X t is connected, but X/ X i = U U V, where U and
V are disjoint, open, nonempty subsets of X/ X Choose u E U and
v E V (Fig. 9.9). Since U is open, there is a basic neighborhood X/ Wi of
u such that Xj Wi C U, Wi is open in X z, and Wi = X;, except for
ti, . . •, in- Define c? = Vi (the zth coordinate of v) if i j* i lf . . . , zn, but
Ci = Ui if i = z‘ i, . .., zn. Then

c°= (c?, .. ., e% . . .) E X Wi C C/.

Define c1 by letting c} = Vi if i ^ z‘ i, . . . , zn; c/ = if i = ii) and

c} = Ui if i — i 2, .. ., zn. Generally, define cw by setting

Ci Viy i 9^ •••y iny and Ci iZj, i — im-i-i, •••>

Then cn = y.
Let

Am = {x |xim is arbitrary, x,m+i = uim+1, ... ,xin =

and otherwise, Xi = Vi}.

Thencw-1 E Am_ 1 n Am, 1 < m < n\v = cn E. An. A x, . . . , An form a

“chain”from U to V, for, since cm E Am D Am+u

Am G A m.^.\ 5^ (f).

We now show that each Ai is connected.

Define a function gm from X»m onto as follows: g(xim) is the point
of Am with zmth coordinate Xim\ zw+ith coordinate w*m+1, . . . , znth co­
ordinate Uin; and for all other i the zth coordinate Not only is gm
continuous, but also it is a homeomorphism (since it is the inverse of the
projection from Am onto X*m. Since each X t*m is connected, each Am is
also connected. But then Am is connected (Section 9.2, Exercise 7).
But Ut = i Am meets both of the disjoint, nonempty, open sets U and F,
a contradiction of Proposition 4. Therefore Y must be connected.
If X/ X i is connected, then since the projection mapping pi from
Xj X i onto X i is continuous, X i is also connected.
196 Connectedness 9.4

EXERCISES

1. Prove the corollary to Proposition 10.

2. a) Prove that a space X, r is connected if and only if given any continuous
function / from X into R, the space of real numbers with the absolute value
topology, then if a and b are in /(X) and c is a real number such that
a < c < b, there is y G X such that f(y) = c.
b) Use (a) to prove that any connected normal space which contains at least
two points contains at least c points, where c is the cardinality of [0, 1].
c) Prove that any connected separable metric space X, D has either at most
one point, or exactly c points. [Hint: By (b), if X has 2 points, it has at
least c points. Take a countable dense subset S = {sn |n E N}. Each
point of X is the limit of some sequence of elements of S, and each sequence
of elements of S has at most one limit. How many sequences of elements
of S are there?]
3. Determine whether each of the spaces below is connected or disconnected.
a) {(x, y) |y = 1/x} U {(x, y) |y = 0} C R2 with the usual topology
b) {Or, y) j x2+ y2 < 1} U (x, y) |x = 1} C R2 with the usual topology
c) the metric space described in Example 5 of Chapter 2
d) the plane R2 with the topology described in Example 5 of Chapter 3
4. A subset A of a space X, r is said to disconnect X if X — A is disconnected.
We say that A is a minimal disconnecting subset of X if X — A is discon­
nected, but X — B is not disconnected, where B is any proper subset of A.
a) Describe some minimal disconnecting subsets for Euclidean n-space, Rn.
b) If x E X and {x} is a minimal disconnecting subset of X, then x is said to
be a cut point of X.
i) Prove that if / is a homeomorphism from a space X, r onto a space F, r',
then if x is a cut point of X, /(x) is a cut point of F.
ii) Prove that a circle, closed interval, open interval, and half-open in­
terval are not homeomorphic in pairs. [Hint: Examine the cut points
of each.]
5. A connected subset A of a space X is said to be irreducibly connected about
B C A if A is the smallest connected set which contains B. For example,
[0, 1] is irreducibly connected about {0, 1} in the usual space R of real num­
bers. Find sets about which the following subspaces of R2 are irreducibly
connected. If no such set exists, write none.
a) {(x, y)|0 < x < 1, y = 0} b) {(x, 0< x 1, 0 < 1}
c) {(x, y)|x2+ y2 = 1} d) {(x,|x2+ 1}

9.4 COMPONENTS. LOCAL CONNECTEDNESS

A disconnected space may, of course, have connected subspaces. The
structures of the maximal connected subspaces of any space are indis­
pensable in any description of the entire structure of the space. Un-
9.4 Components. Local Connectedness 197

fortunately, even knowing completely the structure of every maximal

connected subset of a space does not determine the structure of the space
as a whole, as we see from the following.
Exam ple 11. Let Q be the space of rational numbers with the absolute
value topology. Then Q is totally disconnected (Section 9.1, Exercise 4);
hence the maximal connected subsets of Q are the one-point subsets. If Q
were to have the discrete topology, then it would still be true that the
maximal connected subsets of Q are the one-point subsets. Note however
that relative to the metric topology, each one-point subspace of Q is closed
in Q, but not open; but that with respect to the discrete topology, each
one-point subset of Q is both open and closed. But clearly we cannot tell
which topology Q has just by knowing that each maximal connected sub­
space of Q is a subspace of exactly one point.

Definition 3. Let X, r be any space. A maximal connected subspace of

X is said to be a component of X.

Thus the components of Q as in Example II are the one-point sub­

spaces of Q ; this is true with respect to either the discrete or the absolute-
value topology on Q.
Exam ple 12. In Example 8, S and T are the components of the subspace
S U T of R 2. In Example 4, H~ and H^~ are the components of R — {x} .
It has not yet been shown that each element of a space X, r is actually
contained in a component of X; that is, although it is clear that each
x El X is contained in at least one connected subspace of X, i.e., {x}, we
must show that there is a maximal connected subspace of X which con­
tains x. This is easily done, however. For let {Ay}, i E I, be the family of
all connected subspaces which contain x. Then U/ Ay is a connected sub­
space which contains x (Proposition 5), and Ur Ay is clearly a maximal
connected subspace of X which contains x.
We saw in Example 11 that a component may or may not be open.
The following shows, though, that a component is always closed.
Proposition 13. A component A of a space X, r is closed.
Proof. By Proposition 11, Cl A is also connected. But A C Cl A and A
is a maximal connected subspace of X; hence A = Cl A. Therefore A is
closed.
Proposition 14. Let {A*}, i E /, be the set of components of a space
X, r. (We see from Example 11 that {A*}, i E I , need not be a finite
set.) Then X = U/ Ay, and if i ^ j, then Ay n Ay = <£.
Proof. Since each x E X is in some component of X (at least one connected
subspace of X contains x, hence a maximal connected subspace of X con­
tains #), Ur Ay = X.
198 Connectedness 9.4

If i 7* j, but Ai D Aj then Ai u Aj is a connected subspace of X

which contains both Ai and A j, a contradiction to the maximality of
both Ai and Aj.
Just as we could in a sense localize the notion of compactness, we can
also localize connectedness. As we shall see, spaces which are locally
connected have particularly nice components.

Definition 4. A space X, r is said to be locally connected if there is an

open neighborhood system for r such that for each i e I , 91* consists
of connected subspaces. Equivalently, X is locally connected if given
any x G X and any neighborhood U of xy there is a connected neigh­
borhood V of x such that V C U (Exercise 1).

Exam ple 13. If X is any space with the discrete topology, then for each
x e I , if we set 91* = {{x}}, we obtain an open neighborhood system for
the discrete topology. Each subspace {x} is, of course, connected. Thus,
even though X is totally disconnected, X is still locally connected.

Exam ple 14. Euclidean n-space R n has a basis for its topology which
consists of convex, and therefore connected, subspaces (c/. the remarks
preceding Proposition 8). But if a space has a basis which consists of
connected subspaces, it is certainly locally connected (Proposition 6,
Chapter 3).
Euclidean n-space is thus both connected and locally connected.
We now give an example of a space which is connected, but not locally
connected.

Exam ple 15. The space

Y= {(x, y)|y= sin (1/*), 0 < x} U {(0

is connected as we saw in Example 9. But it is not locally connected. As

a matter of fact, if 0 < p < 1, then N ((0, 0), p) has an infinite number of
components (Fig. 9.10). The space Y, however, is the image of a locally
connected space under a continuous function; hence we see that local
connectedness is not generally preserved by a continuous function. Spe­
cifically, let

A = {(x, y) | y = sin (1/x), 0 < x}

and
B = { ( - 1, 0)}.
200 Connectedness 9.4

x G C°T a contradiction, since C° ft Fr C = <t>. Therefore x g F°, and

hence x G Fr Y; thus Fr C C Fr Y.
Statement (c) implies statement (b). Suppose C is a component of U7
an open subspace of X. Then
C n F r C c f / H F r U;

hence C n F rC = 4> (since U = U° and U° n Fr U = </>). Therefore

C C C°. But CQ C Cy and hence C = C°. Therefore C is open.
Statement (b) implies statement (a). Suppose U is any neighborhood
of x E l . Let V be the component of U which contains x. Then V is
open; hence V is a connected neighborhood of x which is a subset of U.
Therefore X is locally connected.
Note that in a locally connected space X Tr Tthe components of X are
both open and closed. We immediately have the following corollary.

Corollary. A compact, locally connected space X, t has at most finitely

many components.

Proof. Let {A*}, i G 7, be the family of components of X. Then {A*},

i G 7, is an open cover of X, and hence finitely many of the A„ say
Aiir . . ., A in cover X. But if i ^ j 7 Ai n A j = <£; therefore no A j can
be omitted from {A*}, i G 7, such that the remaining components still
form a cover of X. The components A{ir . .., A,b then must be all of the
components of X.
We have seen that local connectedness is not preserved by continuous
functions. Like local compactness, local connectedness is preserved by
continuous open mappings.

Proposition 16. If / is a continuous open function from a locally con­

nected space X, r onto a space F, r', then F is locally connected.
Proof. Suppose y G F and U is any neighborhood of y. Since / is onto,
there is x G X such that f(x) = y. Then f~ l {U) is a neighborhood of x.
Since X is locally connected, there is a connected neighborhood V of x.
Since / is both continuous and open, f(V) is a connected neighborhood of y
with/(F) C U. Therefore F is locally connected.
Proposition 17 illustrates another property of local connectedness
which is similar to the corresponding property of local compactness
(Chapter 8, Proposition 10).

Proposition 17. Let {X*, T;}, i G 7, be a countable family of nonempty

topological spaces. Then the product space X/ X t*is locally connected
if and only if each X; is locally connected and all but finitely many of
the X i are connected.
9.4 Components. Local Connectedness 201

Proof. Suppose X/ X* is locally connected. Since the projection mapping

Pi from X/ X i onto J t is open and continuous, each X* is locally connected.
Let U be any connected neighborhood of any point y of X/ X*. Then U
contains a basic neighborhood of y of the form X/ F», where V i is open
in X a, and V i — X i for all but at most finitely many i. Then P i ( U ) — X i
for all but at most finitely many i. Since P i is continuous and connectedness
is preserved by continuous functions, X t is connected for all but at most
finitely many i.
Suppose that each X* is connected and all but finitely many of the X e
are connected. Suppose y E X/ X t and U is a neighborhood of y. Then
there is a basic neighborhood X/ V i of y which is contained in U, where V i
is open in X aand Vi = X* for all but at most finitely many t, say i l f ..., in.
For each 4, k = 1,___ , n, there is a connected neighborhood of yik (the
4th coordinate of y) in X*4, call it Wik, such that Wik C Vik. There are
at most finitely many more t, say 4 +i, - * , 4a, such that

^ * + 1 9 * * * »

are locally connected, but not connected. For each of these 4,

k = n + 1, . . . , k = m,

there is a connected neighborhood tFe-fc of ^ with JF^ C Vik. For each

i {4? * * - 5 4+1? * - - ? J

set lFt = X*. Set W = X/ TF*. Then IF is a neighborhood of ?/ and

IF C X/ Vi C F. But W is the product of connected sets and is therefore
connected (Proposition 12). Therefore X j X 4is locally connected.

EXERCISES

1. Prove that the two definitions of local connectedness given in Definition 4

are equivalent.
2. Prove that every open subspace of a locally connected space is locally
connected.
3. Prove that the components of the space Q of rational numbers with the
absolute value topology are the one-point subspaces. Prove that Q is not
locally connected.
4. Let X, t be any space. Define an equivalence relation E on X by letting xEy
if x and y are contained in some (the same) connected subset of X. Show that
the F-equivalence classes are the components of X.
5. Decide which of the following spaces are locally connected. Also describe
the components of each space. Both R and R2 are assumed to have the usual
metric topologies.
a) R2 — C, where C is a compact subset of R2
202 Connectedness 9.5

b) {(*, y)| y= x/n,n= 1,2,3,...} C

c) {(x, y)\y = x/n, n = 1, 2,. .. ; or ?/ = 0} C ft2
d) {(z, y) \y = x/n, n = 1,2,a
, nd ^ 0} C
e) {x| x= 1 /n,n = 1, 2, 3,. . .} U {0} C
f) the space N in Example 3 of Chapter 7
6. A subset A of a space X, r is said to be a path component of X if A is a maximal
path-connected subset of X (see Definition 2).
a) What is the relation between the path components of a space and the
components of the space?
b) What are the path components of the spaces described in Examples 7
and 10?
c) Prove that for open subspaces of Rn, a path component is also a component.
d) What would be meant by saying that a space is locally path connected?
Prove that each path component of a locally path-connected space is open.
7. Prove that if a space has only finitely many components, each component is
open. Thus it is not sufficient for a space to have only open components in
order for the space to be locally connected.
8. If X, r is locally connected, is the one-point compactification of X necessarily
locally connected? Explore conditions under which the one-point compactifi­
cation of a locally connected space will be locally connected.

9.5 CONNECTEDNESS AND COMPACT J2- SPACES

Two of the most important properties in topology are compactness and
being T2; hence a compact TV space is doubly important. In this section
we study the properties of compact TVspaces with regard to connectedness
and local connectedness. We first introduce some pertinent definitions.

Definition 5. A space X, r can be split between two of its points x and

y if there are disjoint open subsets U and V of X such that x e U,
y e F, and X = U U F.
A compact connected T 2-space X, r is called a continuum. The
continuum X is said to be irreducible about A C X if X is a minimal
continuum which contains A.
A point x of a space X, r (not necessarily a continuum) is said to
be a cut point of X if X is connected but X — {#} is not connected.
Otherwise, x is said to be a noncut point. See Section 9.3, Exercise 4.
Example 17. Evidently if a space X, r can be split between two points x and
y then x and y are in different components of X. It is not true, however,
that if x and y are in different components of X that X can necessarily be
split between them. Let

Y= {Cn}
U {(a;,y
)| y = 1, or = —1} c R2
9.5 Connectedness and Compact ^-Spaces 203

Figure 9.11

(with the usual topology), where Cn = {(x, y) |x2 + y2 = (l — (1/n))2},

n = 1, 2, 3, . . . Then the components of F are each Cn and the two
straight lines indicated in Fig. 9.11. It should be intuitively evident that
although (0, 1) and (0, —1) are in different components of F, nevertheless
F cannot be split between these two points.

Exam ple 18. Any closed, bounded, and connected subset of R n for any n
is a continuum. Any path in a T 2-space is a continuum, since any sub­
space of a TVspace is T 2 and compactness and connectedness are both
preserved by continuous functions.
Every point of the closed interval [0, 1] except 0 or 1 is a cut point of
[0, 1]; thus {0, 1} is the set of noncut points of [0, 1]. It is easy to show
in fact that [0, 1] is a continuum which is irreducible about {0, 1}. Sim­
ilarly, if P and Q are any two points of R n, then the closed segment PQ
is a continuum irreducible about {P, Q}.

P roposition 18. Let X, t be a compact P 2-space and {A*}, i e /, be a

family of closed subsets of X such that {A*}, i e /, is directed by <

where A* < Ay if Aj C At* (c/. Definition 1 and Example 3 of Chap­
ter 6). Suppose that each A* has the property that it cannot be split
between x and y. Then D / At cannot be split between x and y.

Figure 9.12
204 Connectedness 9.5

Proof. Let B = f l j Ai. Suppose B — U U V, where U and V are dis­

joint, open (in B) subsets of B, and x G U and y G V. Now U and V are
also closed in B. But B is closed in X (Exercise 1); hence U and V are
closed in X. Since X is compact and T 2, X is T4, and thus there are open
sets G and H such that U C G, V C H, and G Ct H = </> (Fig. 9.12).
For each i G I, Ai <£ G U H, or otherwise we would have

x G A i d Gr y G Ai n H r and (Ai d G) d (Ai C\H) = <t>)

that is, we could split Ai between x and y. For each Air we can therefore
find Xi G Ai — (G U H). Since {Ai}, i e I , is by assumption a directed
set, {xi}, i G /, is a net. Since X is compact, this net has a limit point z
(Proposition 9, Chapter 7). Since any neighborhood N of z contains
{xi}Ti G /, cofinally, any neighborhood X of z meets {Ai}, i G /, cofinally.
But then X meets every For given Ai, there is A j C Ai such that
N f) A j $ (remember that { A ^ ,i G /, is directed); hence X ft Ai ^ <j>.
Therefore
z G Cl - ^1,

for each i, and thus z G B — fl/ ^4,. We therefore have that z G C l f ^ ;

hence G U # is a neighborhood of z. But G u H does not contain any of
the Xi, contradicting the choice of z as a limit point of {xff Ti G I. It must
be then that B cannot be split between x and y.

We have already seen that in a general topological space, the inter­

section of a family of connected subsets need not be connected. This is
not true even in a compact T 2-space, but the following is true.
P roposition 19. Suppose {^1,-}, i G I, is a family of closed connected
subsets of a compact TVspace and {Ai}, i G /, directed by < as in
Proposition 18. Then (1/ AiTs connected.
Proof. If x and y are in (1/ Ai, then each At- cannot be split between x
and y; therefore, by Proposition 18, fl j Ai cannot be split between x and y.
Proposition 19 then follows at once from the following proposition.
P roposition 20. If X , r is a compact TVspace and x and y are in X,
then X cannot be split between x and y if and only if x and y are in
the same component of X.
Proof. Let Cx be the component of x in X and Qx = {z |X cannot be split
between x and z}. We already know that Cz C Qx- Suppose y G Qx. We
must show that y G Cx. Consider the family {Ai}, i G I , of closed subsets
of X such that Ai cannot be split between x and y. This family in non­
empty, since Qx is a member. Then if {Ai}}, j g J , is a chain in [Ai],
i G I, by Proposition 18, (1/ Aif cannot be split between x and y. Apply-
9.5 Connectedness and Compact ^-Spaces 205

ing Zorn's lemma to the directed (and hence partially ordered) set {A*},
i E I, we can find a minimal closed subset B which cannot be split be­
tween x and y. We will now show that B is connected.
Suppose B is not connected. Then B = U U F, where U and F are
open and closed in B and nonempty, but U n V = </>. If x E U and y E V ,
then B splits between x and y , a contradiction. On the other hand, if x
and y are in U (or F) and U could be split between x and y, then B could
also be split between x and y; if U could not be split between x and y,
then B would not be minimal. Therefore B must be connected. But then
B C Cx; hence y E Cx. Then Cx = Qx.

P roposition 21. If X, t is a continuum and A C X, then there is a sub­

continuum Y of X which is irreducible about A. (As the name in­
dicates, Y is a subcordinuum of X if the subspace Y is itself a
continuum.)
The proof is left as an exercise.
Exam ple 19. If P and Q are any two distinct points in R n, then

{P,Q} c Cl N(P,p)

for a suitable p > 0. Now Cl N(P, p) is a continuum. There is therefore

a minimal (or irreducible) subcontinuum of Cl N(P,p) about {P, Q}.
The closed segment PQ is an example in this case of such an irreducible
subcontinuum.

P roposition 22. A compact TVspace X, r is locally connected if and

only if every open cover of X has a refinement consisting of a finite
number of connected sets.
Proof. Suppose X is locally connected and {[/»}, i E /, is an open cover
of X. Let {Vj},j E J , be the family of components of the £7*. Since X is
locally connected, {Ty}, j E J , is an open cover of X (Proposition 15);
moreover, {F/}, j E J, is a refinement of {Ui}, i E /. Since X is compact,
there is a finite subcover { V .. . , V jn] of {Fy}, j E J , and this finite
subcover is a refinement of {£/»•}, i E /, by connected sets.
Conversely, suppose that every open cover of X has a finite refinement
by connected sets. Suppose U is a neighborhood of x E X. In order to
show that X is locally connected, we must find a connected neighborhood
F of x with F C U. Since X is T2 and compact, X is T3; hence there is a
neighborhood W of x such that W C Cl IF C U. The set {U, X — Cl W}
is an open cover of X. There is therefore a refinement {Hi, . .. , H n} of
{U, X — Cl W) by open, connected subsets of X. Either

Hi C U or HiCX - C l W
206 Connectedness 9.5

for i = 1, . . . , n by the definition of a refinement. Let x e Hi. Then

x e Hi C U. Therefore Hi is a connected neighborhood of x which is
contained in U. Hence X is locally connected.
A path is one of the most important types of continua but, as has
already been pointed out, paths can be rather peculiar. The purpose of
the next proposition is to help establish certain properties of paths. We
will then give without proof a proposition which completely characterizes
paths.

P roposition 23. If X, t is a compact TVspace and if / is a continuous

function from X onto a TVspace F, then if X is locally connected,
F is also.
The proof of this proposition is outlined in Exercise 4 below. As an
immediate consequence of Proposition 23, we have the following.

C orollary 1. Any path in a T 2-space is compact, connected, and locally

connected.
Exam ple 20. The space F' in Example 16 is connected and compact, but
is not locally connected, and hence could not be a path.
It can also be proved without much trouble (see Exercise 4) that any
path is also second countable; hence, as we shall prove in the next chapter,
any path in a T2-space is a compact, connected, locally connected metric
space. As a matter of fact, although we will not prove it in this text, the
following is also true.

P roposition 24 (the Hahn-Mazurkiewicz theorem). A metric space is

compact, connected, and locally connected if and only if it is a path.
This is a somewhat surprising result, since it implies that even spheres,
cubes, etc. and their counterparts in R n for any n, are continuous images
of [0, 1]. As a rule, the continuous functions from [0, 1] onto such spaces
are not one-one; for if they were, they would be homeomorphisms (Prop­
osition 14, Chapter 7), which, of course, is not generally true.

EXERCISES

1. Prove that the set B in the proof of Proposition 18 is closed. [Hint: The
proof is extremely simple.]
2. Let X } t be any space. Define a relation E on X by xEy if X cannot be split
between x and y. Prove that E is an equivalence relation on X. An ^-equiv­
alence class is said to be a quasi-component. Let x G l ; denote the com­
ponent of x by Cx and the quasi-component of x by Qx. Proposition 20 states
that in a compact T2-space Cx = Qx. Always Cx C Qx•Find an example of
a space in which Cx 5* Qx.
9.5 Connectedness and Compact ^-Spaces 207

3. In Example 17, prove that F cannot be split between (0, 1) and (0, —1).
4. a) Prove that any continuous function from a compact TVspace onto a
TVspace is closed, that is, f(F) is closed if F is closed.
b) Prove that local connectedness is preserved by continuous closed functions.
c) Use (b) to prove Proposition 23.
d) Use (a) to show that any path F, r' in a TVspace is second countable.
[Hint: Let {Un}, n E N, be a countable basis for the usual metric topology
on [0, 1], which is known to be second countable. Then [0, 1] — Un is
closed for each n. Prove that {F —/([0, 1] — Un)}, n E N, is a basis
for r'.]
5. Prove Proposition 21.
6. Prove that any continuum is irreducible about its set of noncut points.
7. Find an example of a collection {*!,}, i E /, of closed, connected subsets of
a space X, r such that {.4,}, i E /, is directed by < as in Proposition 18,
but fl, .1* is not connected.
8. Find an example of a continuum in R2 (with the usual topology) which is
irreducible about each of the following subsets of R2.
a) {(0,0), (1, 1), (0, 1)}
b) {(£, y) |x and y are rational and x2 + y2 < 1}
c) {( X, y)jx= 0; 1 < y 2
<}
Even before they were found, how could we be certain that such continua
existed?
9. Suppose that X is a compact TVspace which is irreducibly connected about
two points a and b. Prove that if A and B are connected subsets of X each
of which contains a, then A C B or B C A.
 10
METRIZABILITY. COMPLETE METRIC SPACE

10.1 METRIZABLE SPACES

This chapter is concerned with two topics pertaining to metric spaces.
The first question to be considered is, When is a space a metric space?
This of course is a very poor statement of the real issue, since we have
already defined what we mean by a metric space, that is, a set with a
metric D. We have seen, however, that metric spaces have a certain
topology induced by their metric. Suppose that instead of starting with a
set with a metric, we begin with a topological space X y r. We might then
ask, Is there a metric D which can be defined on X such that the topology
induced by D is r? This is in fact a very profound question, and satis­
factory answers to it were not provided until comparatively recently. We
shall only partially answer the question in this text.
Definition 1. A space X, r is said to be metrizable if a metric D can be
defined on X such that the topology induced by D is r. Otherwise, X
is said to be nonmetrizable.
The question now is, When is a space metrizable?
Example 1. As we have seen, a topology can be defined on the space R of
real numbers using the open intervals as a basis. This topology is the
same topology as is induced by the absolute value metric; hence the open
interval topology on R is metrizable.
Example 2. If X , D is any metric space, then the product space X X X can
be defined without direct reference to metric. The product topology on
X X X turns out to be the same topology, however, as is induced by the
metric D', defined by

'((xu z 2),(Vi, y2
)) = D(xu

Therefore if X is metrizable, then X X X is also metrizable. (See Prop­

osition 1, Chapter 8.)
Example 3. If N is the set of positive integers with the topology defined
by calling a set U open if U — N — F, where F is finite, then N is not
metrizable. For any metrizable space must be normal, but N is not even T2•
208
10.1 Metrizable Spaces 209

A theorem which tells us when a space is metrizable is called a metri-

zation theorem. One of the most important metrization theorems is
Urysohn’ s metrization theorem.
Proposition 1. Let X, t be a TVspace. Then the following statements
are equivalent:
a) X is regular and second countable.
b) X is separable and metrizable.
c) X is homeomorphic to a subspace of the product space Xv [0, 1],
that is, the product of [0, 1] with itself countably infinitely many
times, where [0, 1] has the absolute value topology. The product
space Xjv [0, 1] is known as the Hilbert cube.
As a first step toward proving Proposition 1, we prove the following.
Proposition 2. Let {Xn, Z)n}, n e X, be a countable family of second
countable metric spaces. Then the product space Xn X n is second
countable and metrizable.
Proof. Since each X n is second countable, X\ X n is second countable by
Proposition 4, Chapter 7. We define a metric on Xn X n as follows: Let x
and y be any points of Xn X n; denote the nth coordinates of x and y by
xn and yn, respectively. Define

Dfn(xn, yn) = min(Dn(xn, ym), 1).

It is easily proved that the metric D fn thus defined on X n is equivalent to

the metric D n (cf. Section 2.3, Exercise 6). Define

rv / . V" D n(Xni yn)

H(x, y) = z ^n ----------

By comparison with the series we see that D(x, y) is defined for

each x and y in X.v X n. It can be verified in a straightforward fashion that
D is in fact a metric for Xn X n. What must now be shown is that the topol­
ogy induced on Xn X n by D is the same as the product topology. We will
use Corollary 1, Proposition 9, Chapter 3.
Let y eXat X n, and let U be a basic neighborhood of y in the product
topology. Then U = Xn Wn, where Wn is open in X n and Wn = X n for
all but at most finitely many n, say n u . . . , nt. Since Wn is open in X n,
we can find positive numbers pi , ••- , Pt such that

N(yni, Pi) C Wni, i = 1,... , t.

Choose p = min(pl5 . . . , Pt )• Then N (y, p), zn g Wn for each n ;

hence z G U, and therefore N(y, p) C U.
210 Metrizability. Complete Metric Space 10.1

On the other hand, suppose N(y, p) is a D-p-neighborhood of y.

Choose q G N such that

x (i/2)n < p /2
n=q

and choose positive numbers p i, . . . , Pq-i such that

P i + * * * + Vq—i < p/2-

Let F = Xat # n, w here#n = ^ ( p n, pn),n = 1, . . . , g — 1, and/fn = Z n

otherwise. Then F is a basic neighborhood of y in the product topology,
and V C N(y,p). Therefore the product and metric topologies on Xn X n
are equivalent; hence Xn X n is metrizable.

Corollary. The Hilbert cube XN[0, 1] is second countable and

metrizable.
It should be clear to the reader than any subspace of a metrizable
space is metrizable (using the same metric which makes the original space
into a metric space).
We now proceed to the proof of Proposition 1.
Proof (Proposition 1)
Statement (c) implies statement (b). This is true since any subspace
of a second countable metric space is both second countable (and hence
separable) and itself a metric space. Any metric space is regular and, in
metric spaces, separability and second countability are equivalent notions
(Proposition 5, Chapter 7); hence (b) implies (a). The difficult part of
this proof then is to show that (a) implies (c).
Statement (a) implies statement (c). Since X is second countable, X is
Lindelof (Chapter 7, Proposition 3). Since A is a regular Lindelof space, X
is normal (Proposition 6, Chapter 7). Let

« = {Bn |n = 1, 2, 3, . . .}

be a countable basis for r. There are a countable number of ordered pairs

of the form (B m, Bn) such that Cl B m C Bn (since X is regular, there is at
least one such pair). Since there are countably many such pairs, we will
enumerate them

{(Uk, Fife) I A; = 1, 2, 3 , , where Uk = Bmk and Fit = B„ k}.

Then Cl UkC Vk. Therefore Cl Uk and X — Vk are disjoint closed

sets of X. Applying Urysohn’ s lemma (Proposition 10, Chapter 5) to
Cl Uk and X - Vk, we can find a continuous function f k from X into
[0, 1] such that f k(x) = 0 forxe Cl Uk and = 1 if
212 Metrizability. Complete Metric Space 10.1

a contradiction to z' E N(z, p). Therefore x' e G; hence F(G) is an open

subset of Z, which completes the proof.
Proposition 1 is one of the most important metrization theorems in
topology, although there are many others. It does not, however, com­
pletely characterize metrizable spaces, since there are nonseparable metric
spaces (Example 6, Chapter 7); hence there are metrizable spaces which
cannot be homeomorphic to a subspace of the Hilbert cube.
Corollary. Any separable metric space X, D has a metrizable com-
pactification.
Proof. Since X is a separable metric space, X is homeomorphic to a sub­
space Z of the Hilbert cube. The Hilbert cube is, however, compact
(since it is the product of compact spaces). Therefore Cl Z is a closed
subset of a compact metric space, and is therefore itself a compact metric
space. But Cl Z is a compactification of X ; hence the desired result.

Example 4. Note that Euclidean n-space Rn is a separable metric space

for each n, and that Rn is hence embeddable as a subspace of the Hilbert
cube. Since R is homeomorphic to the open interval (0, 1) by some homeo-
morphism h, a specific homeomorphism / of Rn in the Hilbert cube might
be defined as follows: Let x = (xi r . .. , x n) be any point of Rn. Set

fk(x) = h(xk), k = 1,. . . ,n,

and
jfjfc(x) = 0 for k > n.
Define
fix) - (fi(x),f2(x), ■■■,h{x), - - .)•

It is easily verified that / is a homeomorphism from Rn onto a subspace Z

of the Hilbert cube. It can also be confirmed that Cl Z is homeomorphic
to (|Q, l])n.
We can see from Example 7 of Chapter 8 that there can be more than
one metrizable compactification of R , and that hence there may be a
number of distinct embeddings of R (or Rn) as a subspace of the Hilbert
cube.

Example 5. We stated after Proposition 22 of the last chapter that we

would prove that any path in a T2-spaee is a metric space. We do so now.
Any path in a 7T2-space is a compact subspace of a T2-space and hence is
regular (in fact, it is normal). We also saw in the last chapter that any
path was second countable. By Proposition 1, then, any path is homeo­
morphic to a subspace of the Hilbert cube, which is a metric space; hence
any path is a metric space.
 10.2 Cauchy Sequences 213

EXERCISES

1. Which of the following spaces are not homeomorphic to a subspace of the

Hilbert cube? If a given space is homeomorphic to a subspace of the Hilbert
cube, produce a homeomorphism.
a) the open interval (8, 11) with the absolute value topology
b) N, the space of positive integers with the discrete topology
c) {0, 1} with the trivial topology
d) {(x, y) |x2+ y2 = 4} C R2 with the Pythagorean topology
2. The following refer to Proposition 2.
a) Prove that D'n is a metric on X n which is equivalent to the metric Dn.
b) Prove that D is a metric for Xv X n.
3. Find necessary and sufficient conditions on a space X, r which make the one-
point compactification of X a separable metric space.
4. Prove or disprove each of the following statements.
a) If X and Y are each homeomorphic to a subspace of the Hilbert cube, then
the product space I X F is also.
b) If Z is a subspace of X , r and X is separable and metrizable, then Z is reg­
ular and second countable.
c) A locally compact TVspace is metrizable if and only if it is second
countable.
d) The continuous image of a separable metric space is a separable metric
space.
e) The product space of a countable family of compact separable spaces is
homeomorphic to a subspace of the Hilbert cube.
5. Show that there is no finite n such that ([0, l])n could replace the Hilbert
cube in the statement of Proposition 1.
6. Define a sequence {sn}, n G N, in the Hilbert cube by letting sj = 1, and
Sn = 0 if k ^ n, that is, by letting the nth coordinate of sn be 1 and every
other coordinate of sn be 0. Since the Hilbert cube is compact, this sequence
has a limit point. Find a limit point of this sequence. Does the sequence
converge?
7. Suppose a space X has a dense metrizable subspace Y. Is X necessarily
metrizable? For example, suppose D is a metric on Y. For each x, y E X,
let an —> x and bn -»y, an, bn G Y. Define

D(x, y) = lim D(an, bn).

10.2 CAUCHY SEQUENCES

Preparatory to a discussion of complete metric spaces, we will investigate
a notion which the reader may have encountered as early as freshman
calculus, that of a Cauchy sequence (although, as was the case with metric,
the terminology might have been different).
214 Metrizability. Complete Metric Space 10.2

Suppose X , D is any metric space and {sn} , n & N, is any sequence in

X which converges to some point y of X. Then the following proposition
is true.
Proposition 3. Given any number p > 0, there is a positive integer
M such that if k and m are any two integers greater than M, then
D(sk, sm) < p.

Proof. Since sn —> y, we may find a positive integer M such that if n > M,
D($n, y) < p/2. Then if k and m are both integers greater than M , we have

D(sk, sm) < D(skj y) + D(y, sm) < p/2 + p/2 = p.

This result inspires the following definition.

Definition 1. Let X, D be a metric space. Then a sequence {sn}, n e N,
in X is said to be a Cauchy sequence if given any positive number p ,
there is a positive integer M such that if m and k are integers greater
than M, then D(sk, sm) < P-
Proposition 3 states that if a sequence in a metric space converges,
then that sequence is a Cauchy sequence. It is not true, however, that
every Cauchy sequence in any metric space converges.
Example 6. Let {sn}, n e N, be the sequence in the space R of real
numbers (with the absolute value metric) defined by sn = l/n. Then
sn —* 0; hence {sn}, n e N, is a Cauchy sequence in R, or any subspace
of R which contains it. But {sn}, n e N, is therefore a Cauchy sequence
in R — {0} ; however, it does not converge in R — {0}.
When the reader first studied the structure of the space R of real
numbers, he may have taken as a basic axiom any one of the following:
A. Any nonempty subset IT of 72 which has an upper bound has a
least upper bound.
B. Any nonempty subset W of 72 which has a lower bound has a
greatest lower bound.
C. Every Cauchy sequence in 72 converges.

We will now show that all three of these statements are equivalent relative
to 72. In Exercise 2, the reader is asked to show the equivalence of A and
B. Propositions 4 and 5 now prove that A and C are equivalent.

Proposition 4. Assume property A of the space 72 of real numbers.

Then a sequence {sn}, n e N, in 72 converges if and only if it is a
Cauchy sequence.

Proof. If {sn}, n G AT, converges, it is a Cauchy sequence by Proposition 3.

10.2 Cauchy Sequences 215

Suppose that {sn}, n G N, is a Cauchy sequence. Set p = 1. Then

there is a positive integer M such that if k and m are integers greater than
M, 15* — sm| < 1. Let

T = max(|«i|, |s2|, .. . , |sj/+i|).

Then for any positive integer ,s|n| <

n + 1; hen

-(T+ 1
) < sn < 1.

E— ■ I0 . . . . . . .T.+. 1 ] an—-2—an_i
■—= anE-
bn
H -&n-i ] •
b
-(r+D n- 2

Figure 10.1 Figure 10.2

Divide [—(T + 1), T -f 1]into two intervals [—(T + 1), 0] and [0, T + 1].
One of these intervals must contain infinitely many of the sn. Set the
left-hand endpoint of that interval equal to and the right-hand end­
point of that interval equal to b 1 (Fig. 10.1). Divide [ax, 61] into

[ai, (ax + bi)/2] and [(ai + &i)/2, 6J.

One of these intervals contains infinitely many of the sn. Set the left-hand
endpoint of that interval equal to a2 and the right-hand endpoint of that
interval equal to b2. Continuing in like manner, suppose we have found
that the interval [an_!, 6n_ 1] contains infinitely many sn. Divide

[<zn_i, 6n_i]
into
[an- 1, (an- 1 + 6n_i)/2] and [(an_ x + 6n-i)/2, 6n-i]

(Fig. 10.2). One of these intervals contains infinitely many of the sn. Set
the left-hand endpoint of that interval equal to an and the right-hand
endpoint equal to bn. By construction the following statements hold:
a) an- 1 < anyn = 1, 2, 3,. . . ;
h) bn ^ bn—1, 71 1, 2, 3,. . . ,
c) bn - a n = a r ( ( T + 1) - (—( T + 1)) = (i)"“1( r + 1),
n = 1, 2, 3, . . .

Let A = (an |w E iV} and B = {bn |n G A^}. Since A is nonempty

and has an upper bound T + 1, A has a least upper bound, say a. Since
J3 is nonempty and has a lower bound, B has a greatest lower bound, say b.
Then an < a < b < bn for each n G N. If a ^ b, then b — a > 0. But

b - a < b n- a n= ( ir - ^ r + l)
216 Metrizability. Complete Metric Space 10.2

for each n. If then b — a > 0, there is a positive integer M' such that
if n > M',

(i)n~1(T + 1) < b — a,

which is impossible; therefore a — b.

We now show that sn —»a. Let p > 0. Then there is an integer M x
such that n > M i implies bn — an < p/2. Consequently

[«n, bn] C N(a, p/2) = (a — p/2, a + p/2).

Therefore N(a, p/2) contains infinitely many of the sn. Since {sn}, n G N,
is a Cauchy sequence, there is a positive integer M such that if k and m
are greater than M y — sm| < p/2. Since N(a, p/2) contains infinitely
many sn, there is at least one integer m! > M such that sm> G N(a, p/2).
Therefore if n > M y then

\sn — a\ < |sn — *m'| + |«

m/ — a\ < p/2 + p/2 = p.

Hence if n > M, then sn G N (a, p). Therefore sn —> a.

Proposition 5. If property C is assumed for the space R of real num­
bers, then every nonempty subset W of R which has an upper bound
has a least upper bound; that is, property C implies property A.

Proof. Let IT be a nonempty subset of R such that W has an upper

bound T. We can find a sequence {sw}, n G N, such that (1) sn < $n+i for
each n; (2) sn G W for each n; and (3) given any w G W, there is a positive
integer M such that n > M implies w < sn. The construction of such a
sequence is left as an exercise. We now prove that this sequence must be
a Cauchy sequence.
Suppose {sn}, n G N, is not a Cauchy sequence. Then for some p > 0
there is no integer M' such that if m and k are integers greater than ilf',
then |Sk — sm| < P•By construction, if k < m, then Sk < sm] hence

|Sk = Sm Sk*

Therefore there is an integer mi such that smi — Si > p. There is an

integer m2 > mi such that sm2 — smi > p; therefore sm2 — si > 2p.
There is an integer ra3 > m2 such that sm3 — sm2 > p, and hence sm3 —
Si > 3p. Continuing in like fashion, we see that the set of sn, and hence
W, could not have an upper bound, contradicting the assumption that W
has an upper bound. Therefore {sn}, n G N, is a Cauchy sequence.
Since {sw}, n G N, is a Cauchy sequence, it converges to some limit y.
It is left as an exercise to prove that y is the least upper bound of W.
218 Metrizability. Complete Metric Space 10.3

Definition 2. A metric D on a set X is said to be complete if every

Cauchy sequence in X, D converges to a point of X. If D is a com­
plete metric on X , then X , Z) is said to be a complete metric space.
The absolute value metric D on the space R of real numbers is a
complete metric on R\ thus R, D is a complete metric space. Note that
D is not a complete metric for the set Q of rational numbers.
We said that a metric space is complete if its metric is complete. Two
metric spaces may be homeomorphic as topological spaces, but one might
be a complete metric space and the other not. The following example
illustrates this point.

Example 7. Let N be the set of positive integers. Let D be the usual

absolute value metric on N , i.e., D(m, n) = \m — n\ for all m, n E N.
Then N, D is a complete metric space, since only Cauchy sequences in N
are those sequences which are constant from some point on (see Section
10.2, Exercise 6). The topology induced on N by D is the discrete topology.
Now define a metric D f on N by
D'(m, n) = \l/m — l/n\.

It is easily verified that D' is a metric on N which also induces the discrete
topology. Therefore, considered as topological spaces, N, D and N, D' are
homeomorphic (the identity mapping being an explicit homeomorphism).
However, D r is not a complete metric space, since the sequence {sn}, n E N,
in N defined by sn = n for each n E N is a Cauchy sequence, but does
not converge.
The following terminology proves useful in the discussion of complete
metric spaces.
Definition 3. Let A be a subset of a metric space X, D. The diameter
of A is defined to be the least upper bound of

{D(x, y) \

We denote the diameter of A by d(A).

Example 8. If A = {(x, y) |x2 + y2 = 1} C R 2 with the Pythagorean

metric, then d(A) = 2 (Fig. 10.3). If B = {(x} y) |\x\ < 2, \y\ < 1} C R 2,
then d(B) = V20 (Fig. 10.4).

Proposition 6. If A is a subset of the metric space X, D then

d(A) = d(Cl A).

Proof. Let p be any positive number. We will show that

d(Cl A) < d(A) + p.

10.3 Complete Metric Spaces 219

Figure 10.4

Suppose x and y are in Cl A. Then both N(x, p/2) and N(y, p/2) meet A.
Choose

x' G N(x, p/2) nA

and
y' E X(2/, p/2) D A.

Then
£>0, 2/) < D(x, x') + D(x', y) < D(x, x') + D(x', i/') + D(y, y')
< p/2 + d(A) + p/2 = d{A) + p.

Therefore d(Cl A) < d(A). But since A C Cl A, d(A) < d(Cl A ); hence
d(A) = d(Cl A).
We now use Proposition 6 to prove an important criterion for com­
pleteness.
Proposition 7. A metric space X, D is complete if and only if given a
countable family {An}, n e N, of closed, nonempty subsets of X such
that
A i D A 2 D * ••D An D ••• and d(An) —> 0,

r u An 5* <p.
Proof. Suppose X, D is a complete metric space. For each n e f f , choose
an G A n. Suppose p > 0. Then there is an integer M such that n > M
implies d(An) < p/2. If k and m are both integers greater than M, then
ak and am are both elements of A m + i 5hence, since d(AM+i) < p/2,

D(ak, am) < p-

We therefore see that {an}, n G A, is a Cauchy sequence. Since X, D is

assumed to be a complete metric space, the sequence {dn}, n E N, con­
verges to some point y. Then for any n, {an, an+i, . . .} is also a sequence
220 Metrizability. Complete Metric Space 10.3

which converges to y. But An is closed and

^n + lj •••
} C Any

for each n ; therefore y E An (since y E Cl An and Cl An = An). Since n

was arbitrary, y E fW An; therefore Hat An ^ 0.
Conversely, suppose that given any decreasing sequence A i 3 A 2D ••*
of closed, nonempty subsets of X such that

d(A») —> 0, fl ^
N

Let {$n} , n E X, be a Cauchy sequence in X. Set

= {s/b |k > n} and An = Cl Bn

for all n E N. Then {An}, n E N, fulfills the necessary conditions, and

hence flv An ^ 0. Choose y E C\n An. We now show that sn —> y.
Letp > 0. Then there is an integer M such that if n > M,d(Bn) < p.
By Proposition 6, d(An) < p as well. Then D(sn, y) < p for all n > Af;
that is, if n > M, then sn E N(y, p). Therefore sn —> y.
As we have seen, not every metric space is complete. The next prop­
osition enables us to say that any compact metric space is complete.
Proposition 8. If a metric space X, D is compact, then D is complete.
Proof. Suppose we have a decreasing sequence A i D A 2 D •••of closed,
nonempty subsets of X. Then, by Proposition 8, Chapter 7, f|Ar An ^ 0.
Therefore, by Proposition 7, D is complete.
Corollary. Any separable metric space X , D is homeomorphic to a
dense subspace of a complete metric space.
Proof. By the corollary to Proposition 1, X has a metrizable compactifi-
cation Z, D'. But Z, D' is a complete metric space by Proposition 8.
It is not true that any subspace of a complete metric space is neces­
sarily complete (e.g., the rational numbers form an incomplete subspace
of the real numbers). We do, however, have the following.
Proposition 9. Any closed subspace Y of a complete metric space X, D
is a complete metric space.
Proof. Suppose {sw}, nEi V, is a Cauchy sequence in Y. Then {sn},
n E N, is also a Cauchy sequence in X ; hence sn —> y, for some y E X.
But then y E Cl Y = Y, and thus {sn}, n E X, converges in Y.
Suppose {Xn, D n}, n E X, is a countable family of nonempty metric
spaces. Then a metric D can be defined for Xv X n as in Proposition 2.
10.3 Complete Metric Spaces 221

The following proposition gives a necessary and sufficient condition for

Xn A n, Dj to be complete.
Proposition 10. Xn X n, D, is a complete metric space if and only if each
component space X ny Dn is a complete metric space.
Proof. First suppose that X n, Dn is a complete metric space for each n,
and that {sn}, n G N, is a Cauchy sequence in X,v X n. If we denote the
kth coordinate of sn by sn(k), then it is easily verified that {sn(&)}, n G N ,
is a Cauchy sequence in X k. Therefore {sn(k)}, n e N, converges in X k.
The convergence of {sn}, n & N , then follows from Proposition 12,
Chapter 6.
Suppose now that one of the X n, D n, say X 1, D lf is not complete.
Then there is a Cauchy sequence (sn(l)}, n E N, in A x which does not
converge. Select a point an from each A n, n > 2. Then the sequence

{^n}) ti Ei A , in X A n,
N

defined by setting the first coordinate of sn equal to sn(l) and the nth
coordinate of sn equal to ant n ^ 1, (that is, {^n}, n G N, is a constant
sequence in all but the first coordinate) is a Cauchy sequence which could
not converge in X^ X n, since it does not converge in the first coordinate.
Corollary. The Hilbert cube is a complete metric space.
Proof 1. Since [0, 1]with the absolute value metric is compact, it is a com­
plete metric space; hence the Hilbert cube Xn [0, 1] is a complete metric
space by Proposition 10.
Proof 2. The Hilbert space is a compact metric space, and hence is com­
plete by Proposition 8.
We have already seen that any separable metric space is homeomorphic
to a dense subspace of a complete metric space. Actually, the following
stronger result is true.
Proposition 11. Let A, D be any metric space. Then X may be em­
bedded as a dense subspace of a complete metric space Y, D' by an
embedding which preserves distances [that is, if h is the embedding of
X, then D(x, y) — D'(h(x), h(y)) for any x and y in X],
We saw earlier that any topological space was homeomorphic to a
dense subset of a compact space. Here we have a somewhat analogous
theorem for metric spaces, compact spaces being the most important type
of topological space, and complete metric spaces being the most important
type of metric space. A space F, D' as described in Proposition 11 is called
a completion of A, D.
222 Metrizability. Complete Metric Space 10.3

The proof of Proposition 11 in its entirety is long and cumbersome,

and little is to be gained by going through all the sordid details. There­
fore only an outline of the proof is presented here.
Outline of proof (Proposition 11). Let Y' be the set of all Cauchy se­
quences in X. Two Cauchy sequences {sw}, n G N, and {tn} , n e N y will
be considered to be equivalent if D(sn, tn) —> 0. It must be verified that
we have thus defined a genuine equivalence relation on F'. Denote the
equivalence class of a sequence {$n}, n G N, by {sn}"; denote the set of
equivalence classes by F. We define a metric D' on F by setting

£'({««}", { Q " ) = lim D

It must be shown that the required limit always exists and is independent
of the representatives of the equivalence classes. Moreover, it must be
shown that D' is actually a metric. If x G X, then x can be identified
with the equivalence class of the constant sequence {sn = x}, n e N. The
mapping thus defined in distance preserving is hence a homeomorphism.
By a method of diagonalization, it can be shown that each Cauchy se­
quence in F converges, and that each element of F is the limit of a Cauchy
sequence each member of which is the class of a constant sequence. For
more details, the reader might see Theorem 2-72 in Hocking and Young,
Topology (Addison-Wesley, 1961).

EXERCISES

1. In Example 7, confirm that D' is a metric on N and that D' induces the dis­
crete topology on N. Describe a completion of A, D; of N, D'.
2. Find a necessary and sufficient condition for a completion of a space A, D to
be compact. [Hint: Let R be the space of real numbers and let D be the
absolute value metric. Set D'(x, y) = min(D(x, y), l) for all x, y E R. Then
D' is a metric on R which is equivalent to D. Is D' a complete metric on R?
Describe the completion of R, D'.]
3. Prove Proposition 8 by means of Proposition 9, Chapter 7 and the definition
of a Cauchy sequence.
4. In the proof of Proposition 10, verify that each {sn(k)}, n E N, is really a
Cauchy sequence in A*.
5. Let / be a continuous function from a metric space X, D into itself with the
property that
D(f(x),f(y)) <
where 0 < k < 1, for any x, y G A". That is, / “ contracts”distances. Set
f l = /, f 2 = f°f, and, in general, f n = f°fn~l. Choose y G A. Consider the
sequence {sn}, n E N, defined by sn = f n(y).
a) Prove that {s„ }, n E N, is a Cauchy sequence.
10.4 Baire Category Theorem 223

b) Prove that if D is complete, then / has a unique fixed point, that is, that
there is one and only one z E X such that f(z) = z. [Hint: Let z be the
unique limit of {sn}, n E N. The sequence {/(sn)}, n E N, must have the
same limit as {$n}, n E N, but f(z) is also the limit of {f(sn)}9 n E N.
Suppose z and z' are both fixed points of /, and show that this contradicts
the assumption th a t/ is a contracting function.]
c) Show by example that if D is not complete, a contracting function may
not have any fixed points. If D is not complete, might a contracting
function have more than one fixed point?
6. A distance-preserving function is called an isometry. Prove that it is not
possible to isometrically embed a complete metric space X, D as a dense proper
subspace of another complete metric space F, D'. Prove, however, that it
might be possible to embed A as a dense proper subspace of a complete metric
space F, D' if the embedding is not required to be an isometry.
7. A subset A of a metric space X, D is said to be totally bounded if given any
positive number p, the open cover {N(x, p)}, x E A, of A has a finite sub­
cover. Prove that a subspace of a complete metric space is compact if and
only if it is closed and totally bounded. (Cf. Exercise 6 of Section 8.1.)

10.4 BAIRE CATEGORY THEOREM

W e now com e to a theorem of great importance in mathematics, particu­

larly in the construction of existence proofs in analysis. An existence
proof is a proof which shows that something exists, or can be found at
least in theory, even if we cannot actually com e up with a specific example
of what exists. For example, we may wish to know that such and such an
equation has a solution even if we cannot find the solution, or that such
and such a function exists even if we cannot at the moment construct an
example of the function. T o know that something either can or cannot
be done either encourages us to try to do it, or saves us the time and effort
of trying. Unfortunately, there will always be intrepid unbelievers who
will insist upon trying to trisect angles with ruler and compass, square
circles, and solve quintic equations by radicals.
We now state and prove the famous Baire Category Theorem.

Proposition 12. Let X, D be a nonempty complete metric space. Then

the following hold:
a) If X is expressed as the union of countably many subsets A 1}
A 2, . . . , A n, . . . , then at least one of the An is somewhere dense.
That is, for one of the An, Cl An contains an open subset of X.
b) If Ui, U2, •••are countably many dense open subsets of X, then
C\n Un is dense in X , that is, C 1(Da^ Un) = X .
224 Metrizability. Complete Metric Space 10.4

a) If (a) is false, then there is a countable family {An}, n G N, of sub­

sets of X such that X = U# An, but (Cl An)°= <£for each n G N.
For each n then, Cl An ^ X. Select bx G l — Cl Ax. Since
X — Cl is open, there is a positive number pi < 1 such that

N(bu P l ) C X - Cl A x.
Set #x = X(6x,Pi/2) (Fig. 10.5). Then Cl£x C N(bi, pi); hence
Cl Bi n Cl A 1 = </>.

Now Bi is a nonempty open subset of X, and therefore Bi Cl A 2.

Choose b2 G B x — Cl A 2. Since 2?x — Cl A 2 is open, there is
p 2 > 0 such that N(b2, p 2) <Z B x — C\ A 2. We lose no generality
in further requiring that p 2 < Set B 2 = N(b2, p 2/2). Then

B 2 CZ ^1 and Cl B 2 0 Cl A 2 ~~ </>.

Proceeding in like fashion, we can obtain a decreasing sequence of

open /^-neighborhoods B x Z) B 2 D ••o 3 * * * such that
Cl Bn n Cl An = 0 and < 1/n. Then

Cl B i d Cl B 2 D •O Cl Bn D ••• and d(Bn) -> 0.

Therefore by Proposition 7, n#C12?n ^ </>. Pick # E fl# Bn.

Then x G An for some n, since LI# An = X. But then

X G Cl An n Cl £n,

which is impossible, since Cl An and Cl Bn are disjoint. Therefore

(a) is proved.
10.4 Baire Category Theorem 225

b) Suppose {U n }, n e N , is a countable family of dense open subsets

of X. In order to prove that Hat Un is dense, it is sufficient to prove
that each neighborhood of any point of X meets Hat Un. Choose
any x G X and any p > 0; we will show that

N (x, p) n (n U n) y* 0.

(This suffices to prove statement (b), since the collection of p-neigh-

borhoods is a basis for the topology induced by D.) Set

T = Cl N (x, p/2);

then T cN(x,p). We now show that T n (PIat Un) </>. Since T

is closed, the subspace T is itself a complete metric space (Prop­
osition 9). Set An = T — Un. Since

An = T - Un = T n (X - U n),

the intersection of two closed subsets of X, A n is closed in both

X and T.
Suppose An is somewhere dense. Then there is t E T and q > 0
such that

N(t, q) n T C Cl An fi T = An.

Therefore N(t, q) D (T — An) = </>. Now t e T = Cl N(x; p/2)

(Fig. 10.6); hence N(t, q) meets N(x,p/2) in some point z. We
may choose q' > 0 such that

N(z, qr) C N(t, q) D N(x, p/2).

226 Metrizability. Complete Metric Space 10.4

But since Un is dense, AT(z, q') intersects Un, say, in z'. Then

z' G T n N(t, q) C An.

But An = T — Un, and hence z' E T — Un\ that is, 2:' g C7W, a
contradiction. Therefore An must be nowhere dense in T.
By (a) then, T 9* LU An (remember that T is a complete
metric space); thus there is s E T — Un An. Therefore, since
An = T - Un, s e T n (C\N Un). Then T n (Hat Un) 9* </>, and
hence
N(x, p) n ^0 Un^ 7^ <t>•

This completes the proof of (b).

A topological space X , r which is the union of countably many subsets
each of which is nowhere dense in X is said to be of the first category.
Otherwise, X is said to be of the second category. Proposition 12 thus
states that every complete metric space is of the second category.
Example 9. Assign a positive integer n to each real number. Set

An = {x G R |such that the positive integer n has been assigned to x}.

Then R = UAn- Since R with the absolute value metric is a complete

metric space, at least one of the An must be somewhere dense in R.
Actually, we can show that some An is dense in any closed interval.
Note that since the subspace Q of rational numbers is countable, we
could assign a different positive integer to each rational number, and thus
Q could be expressed as the union of countably many subsets of Q each of
which is nowhere dense in Q.
Example 10. The plane R 2 with the Pythagorean metric is a complete
metric space. Any straight line L in R 2 is a closed subset of R 2; moreover,
R 2 — L is an open dense subset of R 2. If L x and L2 are any two lines in
R 2, then
(R2 - L x) n (R2 - L2) = R2 - (Li u L2).
By Proposition 11, however, (R 2 — L x) n (R 2 — L2) is a dense subset of
R 2. In general, we may remove countably many straight lines from the
plane and still have what remains a dense subset of R 2 (though it may
not be open).
Proposition 12 is used in existence proofs in the following ways. A
suitable complete metric space is first constructed. Suppose we wish to
prove that something exists which does not have a certain property P.
We express the set of elements of X which have P as the union of countably
10.4 Baire Category Theorem 227

many nowhere dense subsets of X. Since this union could not be all of X
by Proposition 12(a), there must be some element of X which does not
have P. This approach is used to show that there is a continuous function
from the space of real numbers to the space of real numbers which is no­
where differentiable. If we wish to show that some element of X has a
given property Q, we find countably many conditions such that if an ele­
ment of X satisfies all of the conditions, then that element has Q. If the
countable family of conditions is such that the set of elements of X which
satisfy any one of the given conditions is an open dense subset of X, then
there is an element which satisfies them all, and hence has Q, by Prop­
osition 12(b). For examples and details of some existence theorems from
analysis which use Baire’ s theorem, the reader is referred to Chapter 13,
Section 4.2 of Dugundji, Topology (Allyn & Bacon, 1964).

EXERCISES

1. Use Proposition 12 to show that a complete metric space which is connected

must contain uncountably many points if it contains more than one.
2. Suppose at each point of R2 (with the Pythagorean metric) we draw a circle
with integral radius. Is it necessarily true that the set of circles with radius n
is somewhere dense in R2 for at least one positive integer n?
3. Prove that the union of countably many nowhere dense, closed subsets of a
complete metric space can still be dense. Why is this not a contradiction to
Proposition 11(b)? [Hint: Consider the rationals in the space of reals.]
4. Prove that if X, D is a complete metric space, then the removal of countably
many closed, nondense subsets of X still leaves a dense subset of X. Does
this remain true if the subsets removed are not required to be closed?
5. The results obtained in Exercise 5 of Section 10.3 are also used in some im­
portant existence proofs. Indicate how an existence proof which uses these
results might be constructed.
6. Which of the following subspaces of the usual space R of real numbers are
of the second category?
a) The irrational numbers
b) {x |0 < x < 1}
c) {x |x = l/n, n a positive integer, or x = 0}
d) (0, 1) U (3, 4)
7. Suppose Y and Z are subspaces of some space X , r and Y and Z are both of
the second category. Decide whether each of the following must be of the
second category.
a) X fl Y b) X U Y c) the product space X X Y d) X — Y
8. Prove that every locally compact metric space X, D can be given a metric
D' such that D' is equivalent to D and X, D' is complete. Hence each locally
compact metric space is of the second category.
228 Metrizability. Complete Metric Space 10.5

10.5 PARACOMPACTNESS. COMPLETE REGULARITY

Paracompactness and complete regularity are topological notions of relatively
recent origin, but because they express properties of substantial significance
in important areas of modern mathematics, they are widely used in ad­
vanced mathematical literature today. It is beyond the scope of this text
to discuss these concepts in depth or give much insight into why they are
important, but since the reader is likely to encounter them in more ad­
vanced work in analysis or topology, we are including their definitions
and some of their elementary properties in this section. We first treat
paracompactness.

Definition 4. An open covering {I/*}, i G 7, of a topological space X ,

is said to be locally finite if each point of X has a neighborhood which
meets only finitely many of the U{.
The space X , r is paracompact if X is T 2 and if each open cover of X
has a locally finite open refinement. (For the definitions of open cover
and refinement, see Definition 1 of Chapter 7.)
Since a finite cover is necessarily locally finite, it follows that any
open cover of a compact space has a locally finite open refinement (any
finite subcover of the given open cover will do). Since T2 is also necessary
for paracompactness, we have:

Proposition 13. Any compact TVspace is paracompact.

It is also true that any metrizable space is paracompact though we
will not offer any proof of this fact. In fact, paracompactness and metriza­
bility are very closely related as we will see shortly.
It is not necessarily true that the product of even two paracompact
spaces is paracompact, nor is it true that every subspace of a paracompact
space need be paracompact. We do, however, have the following.

Proposition 14. Every closed subspace of a paracompact space is

paracompact.
Proof. Suppose that A is a closed subspace of the paracompact space
X and {Ui}, i G I, is an open cover of A. Then since each Ui is open
in A, we have Ui = A n Vi, where Vi is an open subset of X, for each
i G I. Also
{IT |i G 1} U {X - A}

forms an open cover of X, and, hence has a locally finite open refinement
{Wk}, k G K. It follows now that {A n Wk}, k G K, is a locally finite
open refinement of {Ui}, i E I. Therefore A is paracompact.
230 Metrizability. Complete Metric Space 10.5

(where Vx. is the set corresponding to [/*.). Then Gy is an open set which
contains y but does not meet V. Let

Then V is an open set which contains F' but does not meet U. Therefore
X is normal.
Local finiteness and paracompactness are both strongly related to
metrizability. We will not review the various metrizability theorems, but
will content ourselves with presenting without proof a theorem due to the
Russian mathematician Smirnov, who is renowned for his work on
metrizability. Two eminent American topologists, John Hocking and
Gail Young, cite this theorem as “ the most natural metrization theorem”
they have seen. We first introduce a definition.
Definition 5. A space X is locally metrizable if each point x G X has a
neighborhood which is metrizable (as a subspace).
Proposition 17. A locally metrizable T 2-space is metrizable if and only
if it is paracompact.
The topological concept of completely regular has particular impor­
tance in that branch of mathematics known as functional analysis. As
with paracompactness, we will content ourselves with presenting the
definition and certain basic facts pertinent to this concept.
Definition 6. A space X is said to be completely regular (sometimes
T 3i) if given any i E l and any closed subset F of X ) x & F, there is
a continuous function f : X —*[0, 1] such that f(x) = 0 and f(y) = 1
for all y e F.
The space X is said to be Tychonoff if X is T\ and completely regular.
Tychonoff spaces lie between regular and normal spaces in that normal
implies Tychonoff, and Tychonoff implies regular. Regular does not
necessarily imply completely regular, nor does Tychonoff necessarily imply
normal. We will soon see, though, that Tychonoff and normal spaces are
rather closely related.
Proposition 17. A completely regular space X is TV

Proof. Suppose F is a closed subset of X and x e X — F. Then there is

a continuous function f : X —> [0, 1] such that f{x) = 0 and f(y) = 1 for
all y e F. Now [0, J) and (J, 1] are disjoint open subsets of [0, 1] which
contain 0 and 1, respectively; hence by the continuity of /, /- 1([0, |))
and /- 1((i> 1]) are disjoint open subsets of X which contain x and F,
respectively. Therefore X is T3.
10.5 Paracompactness. Complete Regularity 231

We have already seen that every locally compact T 2-space is regular

(Proposition 6 of Chapter 8). We now prove the following stronger result.
Proposition 18. Every locally compact T 2-space X is Tychonoff.

Proof. Since X is T 2, X is T\. We now show that X is completely regular.

Let Y be the one-point compactification of X. Then Y is a compact
TVspace (Proposition 12 of Chapter 8) and hence is normal. Suppose F
is a closed subset of X and x e X — F. Then {x} and F are both disjoint
closed subsets of the normal space Y; hence there is a continuous function
/: Y -> [0, 1] such that f(x) = 0 and f(y) = 1 for all y e F(by Urysohn’ s
lemma). By taking / |X: X —> [0, 1] we obtain a function which proves
the complete regularity of X.
We leave the proof of the following proposition to the reader.
Proposition 19. Any normal space is a Tychonoff space.

Since a compact T2-space is normal, we also have

Corollary. Any compact T2-space is Tychonoff.

In fact, compact T2~spaces can be shown to characterize Tychonoff

spaces in the following sense.
Proposition 20. A space X is Tychonoff if and only if it is homeomorphic
to a subspace of a compact T2-space.
We make no attempt to prove Proposition 20. We note, however,
that if we found a Tychonoff space which is not normal, then Proposition
20 tells us that we have found a nonnormal subspace of a normal space.
Proposition 21.

a) Every subspace of a completely regular space is completely regular;

hence every subspace of a Tychonoff space is Tychonoff.
b) If {X n}, n G N, is a countable family of completely regular spaces,
then the product space XnX n is also completely regular; hence the
product of a countable family of Tychonoff spaces is Tychonoff.

Proof. We prove (b) and leave (a) as an exercise. Let F b e a closed sub­
set of XnX n and x e XnX n — F. Since XnX n — F is open and contains
x, there is a basic neighborhood XnWn which contains x and fails to meet
F. All but finitely many Wn are equal to X n. Suppose WniJ . . . , Wnt
are those Wn 5* X n. For i = 1 X n. — WUi is a closed subset of
X n. which does not contain xn., then n^th coordinate of x. For each
i = 1, ...,£, we therefore have a function

Qi •
Xn{ * [0, 1]
232 Metrizability. Complete Metric Space 10.5

such that gi(xn.) = 0 and giiy) = 1 for each y G X n. — Wn.. Define

/: XnX n —i► [0, 1] by setting

f{w) == max{^ o pn.(w) 11 = 1, . . . ,

where pn. is the projection into the /lith component. We leave it to the
reader to demonstrate that / is continuous, f(x) = 0 and f(y) — 1 for all
y G F (in fact, for all y G XnX n — XnWn). This establishes that XnX n is
completely regular.
The importance of completely regular spaces rests partly on the fol­
lowing property.
Definition 7. Let X be a topological space and let C(X, R) denote the
set of continuous functions from X into R , the usual space of real
numbers. We say that C(X, R) separates points if given any distinct
real numbers r and s and two distinct points x and y of X, there is an
element/ of C(X, R) such that f(x) = r and f(y) = s.
Proposition 22. If X is Tychonoff, then C(X, R) as described in Defini­
tion 7 above separates points.
Complete regularity is also related to metrizability, but we will not
develop this aspect of the concept.

EXERCISES

1. Prove Proposition 19.

2. Prove Proposition 22.
3. Prove (a) of Proposition 21.
4. Prove that both paracompactness and complete regularity are topological
properties, that is, they are preserved by homeomorphisms.
5. Prove that if X is completely regular and F and F' are disjoint subsets of X
such that F is closed and Ff is compact, then there is a continuous function
/: X —► [0, 1] such that f(F) = 0 and/(F') = 1.
6. Prove that any locally compact TVspace that is the union of a countable
number of compact sets is paracompact.
7. As a corollary of Exercise 6 show that any locally compact TVspace is para­
compact if it is second countable.
8. Prove that the product of a compact TVspace and a paracompact space is
paracompact.
11
INTRODUCTION TO HOMOTOPY THEORY

11.1 H O M O T O P IC FUNCTIONS

Good mathematical terminology is generally intuitively appealing. For

example, consider this statement: The unit disk

Y = {(x, y) |x2 + y2 < 1} c R 2

(with the usual topology) can be contracted to (0, 0); that is, F is a con­
tractible space (Fig. 11.1). Most likely, the reader has not encountered the
notion of a contractible space before, but having studied the topology
of F, he might feel that such a statement fits in with his notions about F.

The word contractible implies the idea of shrinkability, that somehow

we can reduce F to something smaller; in particular, contractible to (0, 0)
implies that F can in some reasonable way be shrunk down to (0, 0). One
of the most obvious ways to contract the disk F is to slide its points along
radii toward the center (0,0). One such “ contracting function,”call it
j 1/2, of the disk into itself could be described as follows: For each (x, y) G F,
lety 1/2(2, y) be the point of F on the segment (0, 0)(x, y) midway between
(x, y) and (0, 0) (Fig. 11.2). It is easily seen thaty1/2 is continuous. We
may further note that if 0 < r < 1, we may define j r: Y —> Y by letting
233
234 Introduction to Homotopy Theory 11.1

jr(xf y) be the point on (0, 0)(x, y) which is l/rth of the distance from (0, 0)
to (x, y). Then j 1 is the identity mapping on F, and j 0 maps all of Y
into (0, 0).
For each r E [0, 1], we have a continuous function j r from Y into Y.
We can therefore define a function^’ from Y X [0, 1] into Y by

i ( 0 , V), r) y)
fo r each (xt y) E F and r E [0, 1]. Thus

ix=il(^X {l}) and = | (F X {0}).

If we look at the images of j r in Y for each r e [0, 1], we note that as r

proceeds from 1 to 0, we gradually shrink Y to (0, 0). Looking at it from
the point of view of mappings, we are transforming the identity function
on Y into a constant function in a natural sort of way. An informal three-
dimensional representation of what is happening is given in Fig. 11.3.

Figure 11.3 Figure 11.4

Throughout this chapter we will assume that the space R of real

numbers, the plane R 2, and Euclidean n-space in general, have their
standard topologies unless it is specified otherwise. Any subsets of Rn
will be assumed to have the subspace topology from Rn.
Look again at the function j: Y X [0, 1] —> F. It is intuitively clear
that since j r is “
close to ”j r> provided r is close to r', j is continuous. Of
course, the continuity of j can also be demonstrated quite rigorously.
To sum it all up then, for each r e [0, 1], we have a function j r from
F into F such that j\ is the identity function on F and j 0 is a constant
function. As r varies smoothly from 1 to 0, j r varies smoothly from j\ to
j 0. This enables us to define a continuous function j: Y X [0, 1] —> F such
that j | (F X {r}) = j r. We have in a sense smoothly transformed the
identity function on F into a constant function. Thus we now have more
justification for calling F a contractible space.
 11.1 Homotopic Functions 235

This notion of transforming one function continuously into another

is one of the most important ideas in topology. We express it formally in
the following definition.
Definition 1. Let/ and g be continuous functions from a space X, t into
a space Y, r'. Then / is said to be homotopic to g if there is a con­
tinuous function H from X X [0, 1] into Y such that

H | (X X {!}) = / and H | (X X {0}) = g.

The function H is said to be a homotopy between f and g (Fig. 11.4).

Intuitively again, / and g are homotopic if / can be continuously
transformed into g. We see that the identity function on the unit disk is
homotopic to a constant function, that is, the function which maps the
entire unit disk into (0, 0).
Example 1. Reexamine Example 17 of Chapter 5. Note that any two
continuous functions from [0, 1] into R 2 are homotopic. More.generally,
we can say that any two continuous functions from [0, 1] into any absolute
retract (Section 5.5, Exercise 2) are homotopic.
The fact that two functions / and g are homotopic is generally far
easier to see intuitively than to express analytically; that is, it is often
evident that / and g are homotopic even when the actual writing out of
an explicit homotopy between / and g would require considerable labor.
Example 2. We have already seen that the identity function j\ = i on
the unit disk Y is homotopic to the function j 0 which maps all of Y into
(0, 0). We could also show that i is homotopic to the function k: Y —»Y
defined by
k(x, y) = (0,

for any (x, y) G Y. One argument to show that i is homotopic to k could

use functions similar to th ejr previously used to show that i was homo topic
to j 0. Still another method to produce a homotopy between i and k would
be to break [0, 1] up into [0, J] U H, 1]. Let j be the homotopy between
j i = i andy0 as defined earlier. Define

2 r) , ifO < r <

H((x, y),
if i < r < 1.

Geometrically, H first contracts Y to (0, 0) and then slides it along the

segment (0, 0)(0, i) from (0, 0) to (0, $) (Fig. 11.5). Since

H | (Y X [0, *]) and | (F X [|, 1])

d since H| (YX{J}) is we
236 Introduction to Homotopy Theory 11.1

tinuous by Proposition 11 Chapter 4. Since

H | (F X {1}) = i and H | (F X {0}) - k,
H is a suitable homotopy between i and k.

Exam ple 3. An arc is a homeomorphism from [0, 1] into any space; thus
any arc is also a path. Since any two paths in R 2 are homotopic, it is cer­
tainly true that any two arcs in R 2 are homo topic. In particular, if a\
and a2 are distinct arcs in R 2 such that ai(0) — a2(0) and ai(l) = a2(l),
then ai and a2 are homotopic. Suppose that P is some point in the area
bounded by the images of and a2 (Fig. 11.6), and that ai and a2 are
considered as arcs in R 2 — {P}. Then and a2 are not homotopic in
R2 — {P}, since the missing point would prevent us from transforming a x
continuously into a2. Intuitively, in order to transform ai into a2 we
would have to have a break at some stage of the transformation to get
past the barrier posed by the missing point.
In like manner, removing a point
from the interior of the unit disk F
prevents the identity map on Y from a m 1]) a i ( l ) = a 2 ( l )
being homo topic to a constant map.
We therefore see that any two given
functions from one space into an­
other are not necessarily homotopic.
The next proposition shows that the a2(0)=ai(0) _____
notion of homotopy enables us to
classify functions from one space into ^2(10, 1])
another according to the functions to
which they are homo topic. Figure 11.6

Proposition 1. Let X, r and F, r' be topological spaces, and let Yx de-

note the set of all continuous functions from X into F. (The reason
for the notation Yx will be made clearer in the appendix.) Let — denote
“is homotopic to, ”that is, f ~ g will mean that / is homotopic to g.
Then ~ is an equivalence relation on Yx .
11.1 Homotopic Functions 237

Proof. If / G Yx , then / ~ /. The explicit homotopy H: X X [0, 1] —> Y

between / and / is given by fiT(x, r) — f(x) for all x G X and r e [0, 1].
In order to verify that H is a suitable homotopy, we must show that

H | (X X {1}) = H | (X X {0}) = /

and that H is continuous. Now

H |( X X l) = ff(® , « =/(*)

for any x £ and hence H | (X X "(lj*) — /, similarly, ./if | (.X X {0} ) = / .

Suppose U is any open subset of Y. Then

H - ' ( U ) = r 1(U
)[0, 1],

But since / is continuous, f ~ x{U) is an open subset of X; thus/—1(f7) X

[0, 1] is an open subset of X X[0, 1]. Therefore H is continuo
If / ~ g,then g~ f.Since f ~ g, there is a

H : X X [0, 1] -*■
such that
#|(XX{1})=/ and f f | ( I X { 0 } ) = j.

Define H
':X
x[0, 1] —>•F by
H'(x, r)= H(x, 1— r) for any G and r

Then H'(X X {0}) = / and H '|( I X {1}) = g. All

turn H upside down, or reverse its direction, if you prefer. Then H' is a
suitable homotopy between g and /.
If / ~ gandg~ k,then / ~ k.Since f ~ g, there is a homoto
H\. X X [0, 1] —* Ysuch that

H ,| (X X {1}) = / and (X X {0}) =

Since g~ k,there is a homotopy H 2:X X [0, 1] —>

H I2 (X X {!}) = g and f f 2| ( X x {0}) = k.

Define H : X X[0, 1] —» by
_ \H2(x,2r), 0 < r < h,
H{x,r)
H i(x, 2 — 1), h < r < 1.
Note that, for all e X,

H(x, 1) = H i (x,2 - 1) = 1) = /(
H{x, 0) = H 2{x, 0) = k(x);
H(x, \) = 0
, ) = H 2(x , 1 ) =
i(x
H
238 Introduction to Homotopy Theory 11.1

f(X)=H(XX{ 1})

g(X)=H(Xx{i})

■k(X)=H(Xx{ 1})

Figure 11.7

Therefore H is well-defined and will be a homotopy between / and k if it

is continuous. The proof of the continuity of H is left as an exercise.
Essentially what we have done is to “ paste" H x and H 2 together to get a
new homotopy between / and k (Fig. 11.7).
We have therefore shown that ~ is an equivalence relation on Yx .
Definition 2. If / e Yx , then the family of continuous functions from
X into Y which are homotopic to / is called the homotopy class of /.
Because of Proposition 1, we can say the homotopy classes of Yx
form a partition of Yx .

XX[0, 1] Y

Figure 11.8

We close this section by noting that the question of whether two

functions in Yx are homotopic is really a question of whether or not a
certain function can be extended. In particular, we may define

h : X x {0} U X X {1} -> Y

by h(x, 0) = f(x) and h(x, 1) = g(x) for all x G X (Fig. 11.8). Then /
and g are homotopic if and only if h can be extended to a continuous
function H from X X [0, 1] —> Y such that

H I (X X {0}) U ( X X {!}) = h.
240 Introduction to Homotopy Theory 11.2

5. Explain intuitively why the identity function on the unit circle X =

{(x, y) |x2+ y2 = 1} is not homotopic to /: X —> X where f(z) = (1, 0)
for all z E X. Take a rubber band and pin the rubber band at one point to
a table. Then try to push the rubber band back along itself to the point at
which it is pinned.

11.2 L O O P S

It should be well known to the reader that one branch of mathematics

can often be used to give results in another branch. Algebra and geometry
are wedded in algebraic geometry, and calculus is a tool in differential
geometry. There is no branch of mathematics today which is really self-
contained. Even logic has started to draw heavily in recent years on
topological methods to obtain some of its most significant discoveries. It
must also have occurred to the reader that topology is highly geometric;
at the same time, topology often has its inspiration and applications in
real and complex analysis. What we are going to do now is to begin to
develop a method by which algebra can be used to express topological
properties. The method we will study is only one of several applications
of algebra to topology, and, in fact, we will study only a small part of the
method at that. Algebraic topology is both one of the oldest and one of
the newest areas of topological studies; the fundamental group dates back
to the early days of topology (c. 1900, which really was not so long ago),
while much of homology theory only dates back a few years, or less.
If X, r and F, t ' are arbitrary spaces, there is not much algebraic
structure that might be given to either Yx , the family of continuous
functions from X into Y, or the homotopy equivalence classes of Yx .
We therefore would like to find a suitable space X, r so that either Yx or
the homotopy classes could be given an algebraic structure which could
help us to study F. The following definition proves useful.
Definition 3. Let F, r' be any space and y0 E F. Then a continuous
function a from [0, 1] into F such that

a(0) = a( 1) = y0

is said to be a loop in F with base point y0 (Fig. 11.9). Two loops a0

and a i in F with base point y0 are said to be homotopic relative to y0
if there is a homotopy H between a0 and a\ such that for each r E [0, 1],

H(0,r) = 1/(1, r) = y0;

that is, for each r E [0, 1], H | [0, 1] X {r} is a loop in F with base
point yo (Fig. 11.10). We will denote the set of all loops in F with
base point y0 by L(F, y0).
11.2 Loops 241

Figure 11.10

The following notation will occasionally prove useful: Let X, r and

F, t ' be spaces, and let A and B be subspaces of X and F, respectively.
Then/: X, A —> F, B will denote that the function/ from X to F has the
property that /(A) C B. Thus /: [0, 1], {0, 1} —> F, y0 would denote a
loop in F based on y0 if / were continuous.
Now L(F, y0) is a subset of the family F [0,11 of all continuous functions
from [0, 1] into F. The relation on L(F, y0) defined by “ is homo topic
relative y0 to,”which we will again denote by is seen to be an equiv­
alence relation on L(F, y0) by the same argument as was used in Prop­
osition 1. We will denote the set of homotopy (relative to y0) classes of
L(F, y0) by 7Ti(F, y0). If a is any loop in F with base point y0, we will
denote the equivalence class of a by |a|. We shall now determine an
algebraic structure on 7Ti(F, y0); in particular, we shall make 7Ti(F, y0)
into a group.
We define an operation # on 7Ti(F, y0) as follows: Suppose |ai| and
\a2\ are elements of 7Ti(F, y0), that is, |ai| and \a2\ are the homotopy
relative y0 equivalence classes of the loops ai and a2. Define \ai\ # \a2\ to
be the equivalence class of the loop ai # a2 defined by

a i(2r), 0 < r < i,

(ai # a2)(r) =
a2(2r — 1), i < r < 1.

We first verify that # a2 is a bona fide element of L(F, y0). Now

ai # a2 is at least a function from [0, 1] into F. Note that # a2 is formed
by “ going around”a\ once and then going around a2 once. Also «i # a2
is continuous because a\ # a2 is continuous on both [0, J] and [^, 1], and
is well-defined at r = % since (ax §a2)(J) = ai(l) = a2(0) = y0. [One
of the principal reasons that we restricted ourselves to L(F, t/0) was so
that we could “ add”functions in this way and have them well-defined.
If a\ and a2 both did not begin and end at y0, we would have no assurance
that a i # a2 was well-defined at r = J.] Since

(ai # a2)(0) = oi(0) = (ai # a2)(l) = a2(l) = t/0,

242 Introduction to Homotopy Theory 11.2

we see that ai # a2 is really an element of L(F, y0). We are therefore

justified in taking its homotopy equivalence class, which is an element of
7Ti(F, y0). Defining
K l # \a2\ = \di # o 2|,

we have a beginning on an operation on 7ri(F, 2/0).

We are not sure yet, though, that # is really an operation. In defining
\di\ # \a2\} we made use of particular representatives of the equivalence
classes |ai| and \a2\. In order to have a valid operation on x ^ F , y0),
however, |ai # a2| must be independent of the representatives we pick.
That is, the “ sum”of two equivalence classes must depend only on the
equivalence classes we are adding and not on which loops we pick from each
class to compute the sum. Proposition 2 shows that # is a well-defined
operation on X i ( F , y0).

Proposition 2. If ai, a2, a3, and a4 are any elements of L(F, y0) and
d\ ~ U3, then
ai it d2 ~ U3 $ a2.
Similarly, if a2 ~ a4, then

ai it a2 ~ a i if a4-
Therefore
K l # K| = K # o2| = |a3 # 0 2 I = K| # |o2|
and
|«ii # |a2| = K l # l«4|-
Proof. Since x ~ a3, there is a homotopy (relative to y0)
a

H: [0, 1] X [0, 1] 7

between ax and a3. Define H': [0, 1] X [0, 1] —> by

H'(r,
_ lH(2r,)s,
if 0 < r < i,
\a2{2r — 1), if < r < 1. a2
Then
= K ( 2r), if 0 < r < 5,
H'(r,0)
K ( 2r - 1), if i 1
, H(2r, s) «20)
= K # a2)(r) for any r £ [0, 1].

Similarly, H'(r, 1) = (a3 # a2)(r) for any r £ [0, 1].

Also, H'

H’
(0, s) =H
(,0s) = y0 = , s) = o 2(l). Figure 11.11
11.2 Loops 243

Direct computation further shows that H' is well-defined for r = We

have yet to show that H' is continuous.
The continuity of W can be demonstrated in a formal argument, and
the reader is urged to provide such an argument in this case. In general,
however, an appeal to a “ picture”of the homotopy is much easier, and
usually just as convincing. For example, Fig. 11.11 gives a picture of the
homotopy H'. Note how we continuously deform ax into d3 while keeping
a2 fixed. The homotopy is continuous on [0, X [0, 1] and on [J, 1] X
[0, 1], and hence is continuous.
It is left as an exercise to prove that a x jf a2 ~ a x # a4.
Corollary. If a x, a2, a3, and d4 are in L(F, y0), and if a x ~ d3 and
d2 ^ d4, then
ai # a2 ~ as # a4.

Therefore if |ax| = |a3| and \a2\ = \a4\ in 7Ti(F, y0), then

l«l| # M = l«3l if |o4|.

Proof, ai # a2 ~ a3 # a2 ~ d3 # a4-

Although it is not obvious at first glance, the operation # defined on

7Ti(F, y0) is not necessarily commutative. In other words, there is no
particular reason for ax # d2 to always be homotopic to a2 # ax.
We now have a set 7Ti(F, y0) with an operation #, and we claim that
tti(F, 2/0), # is a group. In order to prove this assertion, we must show
that # is an associative operation, that there is an identity in 7Ti(F, y0)
with respect to #, and that each element of 7rx(F, y0) has an inverse with
respect to #.
We first prove the associativity of #.

Proposition 3. If |ax|, \a2\, and \a3\are any three elements of 7Ti(F, y0),
then

(Kl # M) # M = l«i| # (|a2|# |a3|).

Proof. It suffices to show that (dj # a2) # a3 ~ a x # (a2 # a3). By
definition,

( a i # a 2)(2r), if 0 < r < 4,

((ai # a2) # a3)(r) =
a3(2r — 1), if i < r < 1.
Therefore
|ai(4r), if 0 <
((ai # 02) # chOW = W 4 r — 1), if i < r < 4,
^a3(2r — 1), if < r < 1.
244 Introduction to Homotopy Theory 11.2

Similarly,

|
ai(2r), if 0 < r <
a2(4r - 2), if 4 < r < f,
a3(4r - 3), if f < r < 1.
Define
[oi(4r/(l + s)), if 0 < r < i( l + s),
ff(r, 5) = |a2(4r — 1 — s), if i( l + s) < r < J(2 + 5),
(a3(l - 4(1 - r)/(2 - «)), if i(2 + s) < r < 1.

Direct computation shows that H is a suitable homotopy between

(<*1 # a2) # 03 and ax # (a2 # a3). Actually, a convincing argument for the
suitability of can be made from its picture alone (Fig. 11.12).

The reader should have begun to realize by now that pictures are a
very useful way of arriving at homotopies. Sometimes a picture alone is
sufficient to convince one that a homotopy exists. Almost always, at any
rate, a picture helps to obtain an analytic expression of the homotopy.
Perhaps the reader has been brought up to believe that arguing from
pictures is a cardinal sin in mathematics; certainly there is no doubt that
pictures can be misleading if improperly used. Nevertheless, diagrams
intelligently employed can be indispensable tools in a mathematical argu­
ment. This is particularly true with regard to homotopy arguments for
two reasons. First, the notion of a homotopy is highly geometric (as is
much of topology); hence we should expect pictures to be apropos. Second,
actually producing a homotopy analytically may be so cumbersome, while,
at the same time, producing a picture may be so easy and convincing,
that no reasonable person would demand the explicit analytic expression.
Let the reader then accustom himself to the use of pictures in homotopy
theory, while, at the same time, being sure that he understands the theory
and meaning behind any picture he uses, and would be able to produce an
analytic expression if necessary.
11.2 Loops 245

EXERCISES

1. In Proposition 2, prove a\ # a2 ~ ai #
2. Suppose the space F, r' is contractible to the point yo; let A: denote the func­
tion which maps F onto y o. Let f be any continuous function from a space
X , t into Y. Prove that / is homotopic to k', where k' is the function which
maps all of X onto yo. Prove that 7ti(F, yo) contains only one element.
3. Let Y, t be any space, yo E F, and k the function which maps all of [0, 1]
onto yo. Prove that |fc| is an identity for 7ti(F, yo) with respect to #. That is,
prove that if |ai| E 7n(F, yo), then

1*1 # |®i| = ki| # |*| = |oi|.

[Hint: Figures 11.13 and 11.14 give pictures of the desired homotopies.
Express analytically what these pictures say graphically.]

Figure 11.13 Figure 11.14

4. Give an argument to show that if F = {Or, y) \x2+ y2 = 1} C R2 and P is

any point of F, then xi(F, P) contains infinitely many elements. [Hint: Let
k be the loop which maps all of [0, 1] onto P, and let a\ be the loop which
“ wraps”[0, 1] once around F (the identification of 0 and 1, if yon prefer).
Show that ai is not homotopic to k. Can a\ # a\ be homotopic to ai? For
any positive integer n, define na\ = a\ jf •••# (n times) # a\. Try to show
that for any positive integers n and m, na\ is homotopic to ma\ if and only
if m = n.]
5. Find an explicit homotopy between the loop a on the disk {Or, y) \x2-{-
y2 < 1} with base point (0, 1) defined by

'(4x, V l — (4z )2), 0 < x < 5,

(2 - 4x, - V I - (2 - 4x)2), ? < x< &
a Or) = ‘(—4x + 2, - V I - (2 - 4a:)2), i < x < I
(4x — 4, V 1 — (4* — 4)2), § < X < l,

with the loop which maps [0, 1] entirely into (0, 1). There are in fact in-
finitely many such homotopies.
246 Introduction to Homotopy Theory 11.3

11.3 THE FUNDAMENTAL GROUP

Thus far we have seen that Ti(Y, y0), jf is a semigroup (that is, a set with
an associative operation) which also has an identity which we will denote
by \k\ (Section 11.2, Exercise 3). It remains to be shown that each element
of 7ri(F, y0) has an inverse with respect to jf. If a G L(F, y0), we define
a~l by letting a~1(r) = a( 1 — r) for each r G [0, 1]. Then

a“x(0) = a(l) = y0 and a-1(l) = a(0) = y0.

Since a is continuous, a~l is also; hence a~l is an element of L(Y, y0).

Geometrically, a-1 is a going around in the opposite direction (Fig. 11.15).

Proposition 4. If a e L(Y, y0), then

a jf a~l ~ a~l # a k.

Therefore the inverse of \a\ in 7rx(F, y0) is \a ^

Proof. An explicit homotopy between k and a jf a~l is given by

a(2r(l — s)), if 0 < r <

H(r, s) =
a(2(l — r)(l — s)), if \ < r < 1.

The reader should confirm that this is indeed a suitable homotopy. The
geometric idea behind it is that we are starting at y0, then going out along
a to a certain point and finally coming back along a~x, each time shortening
the distance we go until have pulled a # a~l entirely back into y0. The
reader should also produce a homotopy to show a~l # a ~ k.
We therefore conclude that tti(Y, y0), jf is a group with identity |fc| in
which \a\ G tti(Y, y0) has as its inverse \a~1\. We are therefore justified
in making the following definition.
Definition 4. Let Y, t' be any space, and let y0 G Y and iri(F, y0), jf
be as described above. Then

*i(Y, Vo), #
is called the fundamental group based on y0 of the space F.
11.3 The Fundamental Group 247

Of course it is all well and good to know that 7T1(F, y0), # is a group.
This is a good beginning, but hardly any more than that. For to be of
much use in the study of topological spaces, 7Ti(F, y0) must have certain
properties. First, 7Ti(F, y0) should be computable, at least for the majority
of spaces which might be of interest. Second, 7r^F , y0) should tell us
something about the space F, r'; if it gives no information about the struc­
ture of F as a topological space, it clearly has no value in the study of
topology. Third, it is rather repugnant that 7Ti(F, y0) should depend on
the point of F on which it is based; in other words, 7Ti(F, y0) should
depend on F and 7 ' rather than on F, r', and y 0. Otherwise, we might get
different fundamental groups for the same space without any clear idea
of how to choose among them, and we would also be given the repugnant
implication that some point of F was better, or at least in some way sig­
nificantly different than, other points of F. Fourth, if there is a continuous
function/from F onto a space Z ) 7 ", we would hope that there is naturally
associated with / a homomorphism from the fundamental group of F into
the fundamental group of Z. This would enable us to attack the difficult
problem of whether there is a continuous function from one space onto
another; most of all, however, we would expect that if groups are to be
associated with spaces, then homomorphisms of groups will be associated
with continuous functions from one space to the other.
All of the above considerations are very basic and very important. It
will be the goal of the remainder of this chapter to at least partly settle
each one of them. We first consider the question of the computation of
fundamental groups.
There are a number of theorems and techniques for computing funda­
mental groups of spaces. Many of these are beyond the scope of this book.
For most of the simpler spaces, however, it is not very difficult to com­
pute the fundamental group.
Example 4. Let F, t ' be a space which is contractible to one of its points
y0. Then from Section 11.2, Exercise 2, we see that 7Ti(F, y0), # is a group
of precisely one element.
Example 5. Let
Y= {(x, y)\x2 + y2 = R 2
and let y0 be any point of F. Let a be the loop which goes once around the
circle in the counterclockwise direction. Then a cannot be homotopic to k
(the function which takes all of [0, 1] onto y0), since a cannot be “
pulled
back”into y0 without breaking (see Section 11.2, Exercise 4). For each
a G L(F, y0) and each integer n, define

1
a # •••# (n times) # a, if n is positive,
k, if n — 0,

a~l # ••
•# (—n) # a~ly if n is negative.
11.3 The Fundamental Group 249

Example 7. Let

Y = {(x, y, z) \x2 + y2 + z2 = 1} c R 3 and y0 E F.

If P is any point of Y, we can show that F — {P} is homeomorphic to R 2

and hence is a contractible space. For let P' be the point of Y antipodal
to P, and let p be the plane in P 3 tangent to F at P' (Fig. 11.18). For
each w E F — {P}, the line Z(P, w) determined by P and w intersects p
in a unique point m. Define h: Y — {P} —> p by

= Z(P, w) n y
for each w G F — {P}. It is not hard to show that h is a homeomorphism.
Now F is not contractible to any of its points (the intuitive argument
runs, “You can’ t peel an orange without breaking its skin”), but if a is any
loop in F and P is any point in F — a([0, 1]), then a is essentially a loop
in P 2; hence a is homotopic to A;, where \k\ is the identity of 7Ti(F, y0)
(Fig. 11.19). Thus any loop in F based on y0
is homotopic to /c, and therefore 7Ti(F, y0)
consists only of |fc|.
F is hence an example of a noncontractible
space which has a trivial fundamental group.
Note that if P and Q are any two points of F,
then F — {P, Q} is homeomorphic to the space
F' in Example 6 and thus has a fundamental
group isomorphic to the additive group of
integers.
The following proposition aids in the
computation of many fundamental groups.

Proposition 5. Let X, r and F, r' be spaces

respectively. Then

TTi(X x Y, Vo))

is isomorphic to the direct sum of Ti(X, x 0) and 7Ti(F, y0),

7 T i ( Z , X q) 2 /0 )•

That is, the fundamental group of the product space of two spaces is
the direct sum of the fundamental groups of the component spaces.
Proof. Suppose a E L(X X F, (x0, yo)). Let px and py be the projections
of X X F into X and F, respectively. Then

PX 0O' G L(X, x0)

250 Introduction to Homotopy Theory 11.3

and
Py °a E L(Y, 2/o)-
Define
/: 7Ti (X X F, (x0, y 0)) - > tti(X, z 0) © ?ri(F, 2/o)
by
/(M) = (|px»o|, |PK°a|).

We will prove that / is the desired isomorphism.

We first show that / is well-defined, that is, that f(\a\) is independent
of the representative of \a\ that is used. Suppose a ~ a'. Then there is a
homotopy
H: [0, 1] X [0, 1], {0, 1} X [0, 1] -> I X Y, (x0, y0)

between a and a'. It can, however, then be verified by straightforward

computation that px°H and p y °H are suitable homotopies between
px°a and px°a', and pY°a and py°a', respectively. (Compare this to
Section 11.1, Exercise 4.) Therefore / is well-defined.
We now show that / is onto: Suppose

(|al|> Ia2|) €= T T 1 Xq) © 7Ti (Y, y0).

Define a e L{X X Y, (x0, y0)) by

nM = f(ai(2 r),y0), if 0 < r <

|(xo, 02(2 — 1), if < < 1.

Then p x °a ~ d\ and py °a ~ a2 (Exercise 1). Moreover, a is easily seen

to be continuous (a is continuous on both [0, and [%, 1] and is well-
defined at r = J); also
a(0) = (ai(0), 2/0) = (x0, y 0) = a(l),

and hence a is an element of L(X X Y, (x0, y0)). But /(|a|) then is

(|ai|, \a2\); therefore/ is onto.
Now we show that / is one-one: Suppose f(\a\) — /(|a'|). Then
(\px*a\, \v y °o\) = (\px0a'\, IP y °cl'\).

Therefore there is a homotopy H i between px° a and px °cl' and a homo­

topy H 2 between p Y°a and py° a'. Define a homotopy
H: [0, 1] X [0, 1] - * X x Y
by
H (r,s)= (H 1(r,s)

Direct computation shows that H is a homotopy between a and a'; hence

\a\ = \a'\. Therefore / is one-one.
11.3 The Fundamental Group 251

It remains to be shown that / is a homomorphism. Suppose \a\ and

|a'| are elements of 7rx(X X F, (x0, ?/0)). Then

/(|a| # |o'|) = /(|a # a'|) = (|p* •(a # a')|, »(a # o')|)

— (|px°« # Px°a'|, |pr°a # py °a'|) (this latter equality
follows at once from
the manner in which
the addition of loops
has been defined)
= (\Px°a\# |px° a'|,
= (\Px a
°|, |PF°a|) # (|px°a'|, |pK°a'|)
[where # is here “ad­
dition”in 7Ti (X} x0)
© tt^ F , y0)]
= /(W)#/(H).
Therefore / is a homomorphism, and, consequently, an isomorphism.

Figure 11.20

Example 8. We have seen that the fundamental group of the circle Y in

Example 5 (relative to any base point) is isomorphic to the additive group
of integers Z, -f. The torus Y X Y then has a fundamental group which is
isomorphic to the direct sum of the additive group of integers with itself,
that is, Z ® Z, +. Note that Z 0 Z , + has two generators, (0, 1) and
(1,0). These correspond to the homotopy classes of the loops a and b
(actually the images of loops) in F X F shown in Fig. 11.20. Note that
since the fundamental group of F does not depend on which base point
is used, neither does the fundamental group of F X F.

EXERCISES

1. The following refer to the proof of Proposition 5.

a) Prove that px°a ~ and py°a ~ a2.
b) Prove each of the equalities in the chain of equalities used to show that
/ is a homomorphism.
2. A space X , r is said to be simply connected if its fundamental group (with re­
spect to some base point) is trivial. The circle is an example of a space which
is connected but not simply connected. Which of the following spaces are
252 Introduction to Homotopy Theory 11.3

simply connected? In all cases, compute the fundamental groups using the
given base point.
a) (0, 1) C R, using any base point
b) (0, 1) X F, with Y as in Example 5 and using any base point
c) [0, 1] X F', with Y' as in Example 7 and using any base point
d) {(£, y) |x2+ y2 < 1} U {(0, 1)} C R2} with (0, 1) as base point
e) Yn, where n is any positive integer, F is as in Example 5, and any base
point is used
3. Suppose X, t and F, r' are contractible to xo and yo, respectively (Section 11.1,
Exercise 3). Prove that the product space X X F is contractible and simply
connected.
4. Prove that the function H defined in Proposition 4 is a genuine homotopy
between a # a~x and k. Find a homotopy between a~l # a and k.

5. In the examples given in this section, the fundamental group has not de­
pended on the base point which was used to compute it. Actually the following
important proposition is true:
If F, t is arc-connected, and if yo and y\ are any two points of F, then
7ti(F, yo) is isomorphic to m (F, y\).

[Recall that an arc in F is a homeomorphism from [0, 1] into F, and that F

is said to be arc-connected if given any two distinct points x and y in Y, there
is an arc h in F such that h(0) — x and h{1) = y.] Prove this proposition.
[Hint: An isomorphism can be defined as follows: Since F is arc-connected,
there is a homeomorphism j from [0, 1] in F such that j(0) = y i and
j(l) = yo. Define j " 1: [0, 1] -> F by
j~Hr) = j( 1 — r)
for each r E [0, 1]. Defining j jf j~ l in the
“natural”way, it can be shown that j # j~ l is
homo to pic to k, where |A;| is the identity of
7Ti(F, yo). Suppose |a| E 7ti(F, yo). Define
/:tti(F, y0) ->tti(F, y{) by
/(H) = 1/ # aftj-1|
(Fig. 11.21). We know that/(|a|) is a well-defined element of 7ti(F, y\). The
details showing that / is an isomorphism are straightforward, but should be
supplied carefully by the reader to check his understanding of homotopies and
the fundamental group.]
Note that if a space is not arc-connected, then the fundamental group
might depend on the base point. For example, let

F = {(x, y) |x = 3} U {(x, y) \x2+ y2 = 1} C R2.

If a base point yo in {(x, y) |x = 3} is chosen, then all of the loops based on
yo are also in {(x,y) \x = 3} since otherwise we would have a continuous
image of a connected space [0, 1] which was not connected. Therefore if
254 Introduction to Homotopy Theory 11.4

two continuous functions, it is continuous. Furthermore, /(a) is a function

from [0, 1] into Y such that

7(a)(0) = f(a( 0)) = f(x0) = 2/o = 7(a)(1);

hence/(a) is indeed an element of L(Y, y0). For any \a\ G 7Ti(Jf, x0), set

/ * ( N ) = 17(a) I-
We first show that /* is well-defined, that is, that /*(|a|) does not
depend on the representative of \a\ used to compute it, but only on the
equivalence class. Suppose a ~ a', that is, |a| = |a'|. Then there is a
homotopy
H:[ 0, 1] X [0, 1], {0, 1} X [0, 1] —> X, x0
such that
H | [0, 1] X {0} - a and H | [0, 1] X {1} - a'.

Define H f = / o H. Since H' is the composition of continuous functions, it

is continuous. Direct computation shows that H' is a homotopy between
/ o a' and /°a, and hence

l/°a| = /*(|a|) = |/° = /*(|a'|).

Therefore/* is well-defined.
We now prove that/* is a homomorphism. Suppose |a| and \a'\ are any
elements of x 0). Now a §a' is defined by

a(2r), if 0 < r <

(a # aO(r)
a'(2r - 1), if i < r < 1.

Therefore fo (a # a') is the loop defined by

f/(a(2r)), if 0 < r < 2,

t/(a'(2r - 1)), if i < r < 1.
But this is precisely the definition of / °a # / °a'. Then

7(a # a') = /(a) #/(a').

Therefore /*(| a§a'|) = /*(|a|) #/*(|a'|). Since |a # a'| = |a| #

have
/*(|a| # |a'|) = /*(|a|) #/*(|a'|).

Hence/* is a homomorphism.
The following example shows that/* may not be onto even when/ is.

Example 9. We have already seen that there is a continuous function /

from the interval [0, 1] onto a circle Y. But [0, 1] is a contractible space
11.4 The Fundamental Group and Continuous Functions 255

and thus has a trivial homotopy group, whereas the homotopy group of F
is isomorphic to the additive group of integers. Therefore/* in this instance
is merely a function which takes the sole element (the identity) of
it i([0, 1]) onto the identity |A;| of 7Tx(F); hence/* is clearly not onto.
The next two propositions show that the homomorphisms induced by
continuous functions behave fairly respectably in relation to the functions
that induce them.
Proposition 7. Suppose / is a continuous function from X, r into F, r'
and g is a continuous function from F, r' into Z , r". Also suppose
that f (x0) = yQ and g(y0) = z0. Then

(g°f)*: TTi(Xyz 0) -> 7Ti(Z, Zq) is the same as g* <>/*.

That is, the composition of continuous functions gives a corresponding

composition of the homomorphisms that these functions induce.
Proof. Suppose a G L(X, x0). Then

(g°f)(a) = (0°/)°a= 9°(f°a) = = £(?(«)) =

Therefore
(sr° /)*(la l) = (fl'*« /*)(|o|).
Proposition 8. Suppose / and <7are continuous functions from X , r into
F, r' and/(x0) = ^(x0) = t/0. Then if / is homotopic to gy

f* = 9*-
Proof. If / is homo topic to gylet H be a homotopy between / and g. Define

(H*a)(r, s) = H(a(r), s)

for each (r, s) e [0, 1] X [0, 1]. Then H*a is easily verified to be a homotopy
between / °a and g <>a for any a e L(X, a:0). Then J(a) ~ Tj(a) for any
a G L(XyXq) ; hence

/*(|a|) = gr*(|a|) for any \a\ G 7Ti(A, x0).

If fundamental groups are to have much meaning topologically, then

homeomorphic spaces should have isomorphic fundamental groups. As
we have already seen, however, it is quite possible for two spaces which
are not homeomorphic to have isomorphic fundamental groups. We may
therefore suspect that there is a condition even weaker than homeo-
morphism which assures that two spaces have isomorphic fundamental
groups. Experience has shown that this suspicion is indeed quite correct.
The following definition proves to be appropriate.
256 Introduction to Homotopy Theory 11.4

Definition 5. Let X, r and Y, r' be (arc-connected) spaces. Then X

and Y are said to have the same homotopy type, or to be homotopically
equivalent, if there are continuous functions

f:X ->Y and g : Y —> X

such that f°g is homotopic to the identity function iy on Y and g o f is

homotopic to the identity function i x on X .
Being of the same homotopy type is a weaker condition than being
homeomorphic, since X and Y would be homeomorphic if and only if
there were continuous functions f : X —> Y and g : Y —> X such that
f o g = i Y and g°f = ix (hence / = g~l). Of course, if two spaces are
homeomorphic, they are also of the same homotopy type.

Figure 11.23

Example 10. Let X, r be any contractible space and let Y, t ' be a space
consisting of a single point P. Then X has the same homotppy type as Y.
Let x0 be a point of X to which X can be contracted and let k be the
function on X which takes all of X into z 0 (Fig. 11.23). Then ix ~ k.
Let f : X —> Y be defined by f(x) = P for all x e X, and g: Y —> X be
defined by g(P) = x0. Then

9°f = ix-

Therefore X and Y have the same homotopy type.

Note how apparently different two spaces of the same homotopy type
can be.

Example 11. Suppose W is a subspace of a space X, t . Then a continuous

function /: X —► W is said to be a retraction of X onto W if / | W = iw -
We call W a deformation retract of X if ix is homo topic to a retraction of
X onto W. For example, the letter 0 is a deformation retract of the letter
Q; here the retraction / could be described by saying that / takes any
point in the tail of the Q into the point where the tail crosses the 0 part
of the Q, and leaves all other points fixed.
11.4 The Fundamental Group and Continuous Functions 257

If W is a deformation retract of X y then W and X have the same

homotopy type. For convenience, set ix \W = iw• Let / be the con­
tinuous function from X into W which is homotopic to ix . Then
f°iw = f \ W ~ iw and iw °f = f ~ ix-

Therefore W and X have the same homotopy type.

The following proposition is pure set theory and will be stated without
proof. It is of sufficiently wide application that the reader should already
be familiar with it; if such is not the case, he should supply a proof.
Proposition 9. Let / be a function from a set S into a set T. Then / is
one-one and onto if and only if there is a function g from T into S
such th a t/°0 is the identity mapping on T and g°f is the identity
mapping on S, that is, / is one-one and onto if and only if it has a
two-sided inverse.
We use this immediately to prove the following.
Proposition 10. Suppose two spaces X, r and Y, r f have the same
homotopy type. Then tti(X) is isomorphic to tti(Y). (Recall that we
are assuming X and Y arc-connected; hence their fundamental groups
are independent of the base point.)

Proof. Since X and Y are of the same homotopy type, there are con­
tinuous functions /: X —> Y and g :Y —> X such that f°g ~ iy and
g°f ~ i x •Applying Propositions 7 and 8, we have

/* °g* = iy* and g*°f* = ix*-

But ix* and iy* are the identity functions on TTi(X) and Wi(Y), respec­
tively. It follows then from Proposition 9 that/* is one-one and onto, and
hence is an isomorphism.

It is not true that if two spaces have isomorphic fundamental groups

they are then of the same homotopy type (e.g., see Example 7). Never­
theless, homotopy equivalence does give a partition of the family of
topological spaces, just as homotopy gave a partition of the family of
continuous functions from one space into another.

Proposition 11.Let the phrase “

is of the same homotopy type as”be
denoted by and let T denote the family of all topological spaces.
Then ^ is an equivalence relation on T.

Proof. If X E T, then X ~ X. Let / = g — ix •Then / o g — g of = ix -}

therefore I ~ I .
11.4 The Fundamental Group and Continuous Functions 259

space without any holes. On the other hand, it should be noted that al­
though the sphere also has a trivial fundamental group, it could hardly
be said not to have any holes. The hole in a sphere, however, is a higher­
dimensional hole, and is not one which can be registered by the funda­
mental group. Note that a torus has two types of holes and that its
fundamental group has two generators. We should not, however, try to
push this point too far, since it is only approximately true.
The fact that there are “ higher-dimensional holes,”as well as the fact
that the fundamental group can give but rather limited information, leads
us to hope that there are other algebraic structures which can be asso­
ciated with a space to supplement the information given by the funda­
mental group. Such is indeed the case. There are higher homotopy groups
[previously implied by using the notation instead of merely
homology groups, cohomology groups, and a long list of others, but these
will not be explored in this text.

EXERCISES

1. Supply an argument to prove more fully that 0 is a deformation retract of

Q (Example 11).
2. Classify each of the diagrams in Fig. 11.24 (considered as subspaces of R2)
according to homotopy type.

A B C D E R T

8 1/08 Figure 11.24

3. In each of the following, decide whether or not the two spaces given are of
the same h om otopy type.
a) a circle in R2 and R2 — {(0, 0)}
b) the sphere in R3 and a circle in R2
c) a triangle and a circle in R2
d) R3 and R2
4. In which of the follow ing is the second space a deform ation retract of the
first space? An intuitive argument m ay be all the reader will be able to give.
a) {(*, y
)x|2 + y2 < 1} and {
(x
,y) \x2+ y2 = 1}
b) R2 and {(0, 0)}
c) {Or, y)| x2+ y2 < 1} and {(x,y) \+ < 1}
5. Suppose TE is a subspace of X, r such that %w — ix | IE can be extended to a
continuous function from X into IE. D iscuss the relation between tti(IE) and
260 Introduction to Homotopy Theory 11.4

7ri(X). Formulate a proposition which tells us when iw cannot be extended

to a continuous function from X into W. Let

W = {(x,y)\x2+ y 2 = 1 } C I = {(x, y) \x2+ y2 < 1}.

In this case can iw be extended to a continuous function from X into W?

How could we interpret this result geometrically?
6. Suppose X, t is a space with a contractible subspace A. Let X/A be the
identification space obtained by identifying all the points of A (and leaving
the other points of X as they are). What can be said about the relation of
tti(X) to tt\(X/A)? Give an argument for your assertion. Can you interpret
your statement geometrically?
7. We will call a continuous function/ from a space X, r into a space F, r' trivial
if /* takes all of m(X) onto the identity of 7ti(F). Examine each of the follow­
ing and decide whether or not there is a nontrivial function from the first
space into the second. If there is a nontrivial function, describe one.
a) a circle in R2 and the open interval (0, 1)
b) a circle in R2 and R2 — {(0, 0)}
c) a circle C in R2 and the torus C X C
d) a torus and a circle in R2
e) a space with fundamental group Z$ and a space with fundamental group
Z3, where Z 5 and Z3 represent the additive groups of integers modulo 5
and 3, respectively
8. If X and Y have the same homotopy type, which of the following situations
cannot occur?
a) X compact, but Y not compact
b) X connected, but Y disconnected
c) X arcwise connected, but Y not arcwise connected
APPENDIX ON INFINITE PRODUCTS

Let {X i, T{}, i E I, be any family of topological spaces indexed b y som e set I.

If 1 is countable, then we already have a definition of the product space of this
family. But / need not always be countable; thus if we are to consider product
spaces in their full generality, we need to have a product of uncountably many
spaces also.
Let R be the set of real numbers. Then, as we know, R X R = R 2 is the
set of all ordered pairs (x, y) where x and y are elements of R. T o each ordered
pair (x, y), we can associate a function c from the set {1, 2} into R defined b y
c(l) = x and c(2) = y. For each (x, y) E R, there is a distinct function from
{1,2} into R, and for each function c: {1,2} —> R , there is a unique point
(c(1), c(2)) of R 2.
If X i and X 2 are any two sets, then

X\ X X 2 = {(xi, X2) \ x i E X i and X2 E X 2}.

But we are able to show that I i X I 2 can be associated in a natural way with
the set of all functions c from {1, 2} into T i U X 2 which have the property that
c(l) E X\ and c(2) E X 2 . N ote that {1, 2} is the index set for the fam ily of sets
{X\, X 2) of which we are taking the product. We therefore make the follow ing
definition.

Definition 1. Let {Xi}, i E I, be any fam ily of sets. Define the product of
the fam ily {Xi}, i E I, to be the collection of all functions c from I into
Uz X i such that
c(i) E X i for each i E I.

The product of {Xi}, i E I, is denoted by Xz Xi. If c E Xz Xi, then c(i),

usually denoted b y a, is called the ith coordinate of c. X i is called the ith
component of the set Xz Xi.

N ote that this definition of the product does not depend on the cardinality
of I. It should not cause the reader much work to show that where I is countable
between the old definition of Xz X i and the new, there is a natural equivalence.
We now use the considerations put forth in Chapter 4 concerning what prop­
erties the product top ology should have to define a topology on Xz X i if each X i
is also a topological space.
261
262 Appendix

Definition 2. Let {Xi, r;}, i E l, be a family of spaces, and let Xi Xi be the

product set of the family of Xi as defined in Definition 1. Let

S = {Xi Vi | Vi — Xi for all but at most one i E I

and each Vi is an open subset of Xi].

Then S is a subbasis for a topology r on Xi Xi called the product topology.

The space Xr Xi, t is called the product space of {Xi, rt}, i E I. (Compare
this to Definition 7, Chapter 4.)
There is a natural function
P i - . X Xi ^ X i
I
defined by pi(c) — Ci for each i E I. We call pi the projection into the ith
component.
The reader should promptly prove that the product topology is the coarsest
topology which makes each projection pi continuous.
The propositions and proofs in this text which deal with product spaces
have purposely been designed so that it would be easy to adapt them to the more
general notion of a product space (provided that they are valid when generalized).
We now give two examples of propositions and their proofs which generalize and
one example of a proposition which is not true when stated for the product of
uncountably many spaces.

Proposition 1. Let Y = Xr Xi be the product space of the family of non­

empty spaces {Xi, n], i E I. Then Y is T2 if and only if each Xi is T2.
(See Proposition 3, Chapter 5.)
Proof. Suppose each Xi is T2, and let x and y be distinct points of Y. We will
use Xi and yi to denote the ith coordinate of x and y, respectively. Since x 5* y,
Xi 7* yi for at least one i E I, say for i'. Therefore there are open sets Ui' and
Vi’in X^ such that

x^ E Ui’
, y^ E V i’
, and Ui’n V = <t>.
Set U = Xi Hi, where Hi = Xi, i i f, and Hi* = Ui>; and set V = Xj Ui,
where Gi = Xi, i ^ ir, and Gi> = Vi’ . Then U and V &re neighborhoods of x
and y, respectively, and since any point of U differs from any point of V at least
in the ith coordinate, U fl V = <£. Therefore Y is TV
By the generalization of Proposition 20, Chapter 4, each Xi, n is homeo-
morphic to a subspace of Y. Thus if Y is T2, then each Xi is also T2.

Proposition 2. Suppose / is a function from a space X, r into the product

space Xi Xi, r'. Define /t: X, r —> Xi, Ti by
fi(x) = pi°f(x) for each x EX,
where pi is the projection into the ith component. Then / is continuous if
and only if is continuous for each i E I. (Cf. Proposition 21, Chapter 4.)
 Infinite Products 263

Proof. If/ is continuous, then/» •= fop{ is the composition of two continuous

functions and therefore is continuous.
Suppose now that /*•is continuous for each i G I. We first note that the
rth coordinate of f(x) is fi(x) for each x G X. A basis (B for r' consists of all sets
of the form X/ Ft-, where F»is open in X»for each i G I and Ft = Xi for all but
finitely many i. Suppose Xr Vi is any member of (B, and Vi = Xi for each i E /
except t‘i, . . . , £m. Now/'^Xr F») is the set of all points x of X such that

fix) e X Vi.
I
But this is easily seen to be flr/-1(Fi). For every i G I, except i\,... , im,
f t 1(Vi]) = X (because F* = X»). Since /t is continuous for each i G /, /7/(F»y)
is open in X, j = 1, . . . , m. Therefore

/-1(X f.) = ijn •••n/,ra(F,m),

V
M

which is open in X since it is the intersection of finitely many open sets. Hence,
by Proposition 7, Chapter 4, / is continuous.
Proposition 3. Suppose {s»}, i G /, is a net in the product space X/Xy.
Then —*■t/ if and only if s{ —> yj, where 2/y is the jth coordinate of y and
is defined to be {py(s»)}, i E I.
The proof of Proposition 12 of Chapter 6 may be used verbatim.
Proposition 4 (Tychonoff). Let XiXi be the product space of the nonempty
family of nonempty spaces {Xi, n}, i G I. Then X/X» is compact if and
only if each component space is compact.
The proof of Proposition 13 of Chapter 7 may be used verbatim; alternatively,
one may use the proof of Proposition 16.
The propositions we would expect not to generalize are those which deal
with the cardinality of a basis for the product topology. For example, Prop­
osition 4(d), Chapter 7, does not generalize, as the following example shows.
Example 1. Let I be an uncountable set, and let Xi = {0,1} for each i G I.
Give each X* the discrete topology. Then certainly each X* is second countable.
But the product space X/ X* is not second countable. This is true since the sub­
basis S, as described in Definition 2 for the product topology on X/ X<, contains
uncountably many distinct members of the form X/ F», where Vi = X», except
for precisely one i G /. The family of such members of S can be shown to be a
minimal subbasis for the product topology; thus the product topology could not
be second countable. We could also prove that the product space Xr X t is not
second countable as follows: If Xr X* is second countable, then Xr Xi is Lindelof,
and hence every open cover of Xr Xi has a countable subcover. But the collection
of Xr Vi, where V% = Xi, except for precisely one i, is an open cover of Xr Xi
which has no countable subcover.
The generalization of the notion of a product space also expands our horizons
as to the possible uses of product spaces.
264 Appendix

Example 2. Given spaces X , r and Y, r', then Yx was used to denote the family of
all continuous functions from X into Y (cf. Proposition 1, Chapter 11). Actually,
Yx more appropriately stands for the family of all functions from X into F, that
is, the product set Yx . Since F is a space, Yx can be given a topology as a product
space; the family of continuous functions from X into F is a subspace of this
space. This in turn leads to the possibility of putting topologies on the set of
homotopy equivalence classes of functions from X into F (using the identifica­
tion topology) and of even giving a topology to fundamental groups. In point
of fact, more important topologies than the product topology are used on Yx,
but having a generalized notion of product has awakened us to this possibility.

EXERCISES

1. If {Xi, Ti}, i E /, is a countable family of spaces, show that there is a natural

correspondence between the product space of this family as defined in this
appendix and the product space as defined previously. In other words, prove
that the product spaces formed in both ways are actually homeomorphic.
2. Formulate and prove a generalization of Proposition 20, Chapter 4.
3. Prove or disprove: The product of any family of nonempty first countable
spaces is first countable.
4. Show that the product of a family of discrete spaces may not have the discrete
topology. Does the product of any family of spaces with the trivial topology
necessarily have the trival topology? Is it possible for an infinite product of
infinite spaces to have the discrete topology?
5. Prove that the product topology on Yx (Example 2) is equivalent to the
topology generated by saying that any net {/»}, i E I, in Yx converges to /
if and only if fi(x) —>f(x) for all x E X.
 INDEX OF SYMBOLS

e, 1 D(x, A), 34
U, 1 D(A, B), 34
n, l Cl, 37, 55
c, l Fr, 38, 55
{ I }, i A°, 43, 55
<*>, i Ext, 56
S - T, A', 56
S XT, 2 XjSi,84, 261
r \ 3 a - > X, 1
33
f°g, 3 {iSj}, id , 3
/ 1W, 3 {s„ }, neN, 3
<, 5 {Si},id, 117
gib, 6 d{A), 218
lub, 6 f ~ 9, 236
f:S T, 3 Yx , 236
5 © T, 14 L(Y, y0), 241
X, D, 16 /:[0, 1], {0, l} - + 7 , y0, 241
X, t , 40 M, 241
max(xi, . . . , == maximum of t t i (Y,y0),#, 242, 246
the numbers , ••* , 17 a x # a2, 241
min^!, . . . , minimum of l«iI # M , 242
the numbers •••, xn, 21 a~\ 13, 246
\x - y\, 16 /*, 253
N(x, p), 21 /, 253
S i y , 26, 122 X ~ Y, 257
D(x, y), 16
265
IND EX

Absolute retract, 112 Compactification, 175

Absolute value metric, 16 Alexandroff, or one-point, 175
Accumulation point, 179 Complement, 1
Alexandroff compactification, 175 Complete metric, 218
Arc, 236 Completely normal, 112
Arc-connected, 239 Completely regular, 230
Associative operation, 13 Completion, 221
Axiom of choice, 7 Component, 197
Component space, 85, 261
Baire category theorem, 223 Composition of functions, 3
Base point, 240 Condensation point, 151
Basis, 44 Connected, 183
of a filter, 134 locally, 198
Bolzano-Weierstrass theorem, 180 polygonally, 189
Bounded function, 167 simply, 251
Bounded subset, 162, 164 Connected subset, 183
Connected, path, 189
Cardinal number, 12 Continuity, 29, 70
Cardinality, 9 and convergence, 130
Cartesian product, 2, 84 uniform, 166
Category, first or second, 226 Continuum, 202
Cauchy sequence, 214 irreducible, 202
Chain, 7 Contractible, 233, 239
Class, 12 Convergence, in a metric space, 26,
Closed subset, 24, 41 113
Closure, 37, 55 of filters, 133
Cofinal, 117 of nets, 122
Cohomology groups, 259 Convex, 190
Compact, countably, 179 Coordinate, 85, 261
locally, 169 Countable, 9, 145
sequentially, 179 Covering property, 142
Compact space, 152 Cube, 158
Compact subset, 152 Cut point, 196, 202
267
268 Index

Cylinder, 158 Group, 13

fundamental, 246
Deformation retract, 256
De Morgan formulas, 2 Hahn-Mazurkiewicz theorem, 206
Dense, 62 Half-plane, 48
nowhere, 62 Hausdorff space, 95
somewhere, 62 Higher homotopy groups, 259
Derived set, 56 Hilbert cube, 209
Diagonal, 89 Homeomorphism, 75
Diameter of a set, 218 Homology groups, 259
Direct sum, 14 Homomorphism, 14
Directed set, 116 Homotopic, 111
Disconnected, 183 functions, 235
totally, 187 Homotopy, 235
Disconnecting subset, 196 relative, 240
Domain, 3 Homotopy class, 238
Homotopy type, 256
Element, 1
Embedding, 75, 88 Ideal in a ring, 51
Empty set, 1 Ideal point, 175
Equivalence class, 8 Identification mapping, 80
Equivalence relation, 7 Identification space, 80
Equivalent metrics, 23 Identity, with respect to an operation,
Euclidean n-space, 163 13
Existence proof, 226 Identity function, 32
Extension of a function, 106 Image, 3
Exterior, 56 Inclusion map, 74
Indices, 3
Filter, 133 Infinite, 9
basis for, 134 Interior, 43, 55
finer, 136 Intersection, 1
generated by a net, 134 Inverse, with respect to an operation,
ultra-, 136 13
Finite, 9 Inverse relation, 3
First countable, 115, 145 Irreducibly connected, 196
Frontier, 38, 55 Isometry, 223
Function, 2 Isomorphism, 14
bounded, 167
closed, 207 Least upper bound, 6
continuous, 29, 70 Lebesque number, 165
open, 89, 172 Limit, 26, 122
order-preserving, 119 of a filter, 133
selection, 118 Limit point, 122, 133
Fundamental group, 246 Lindelof, 144
Locally compact, 169
Geometry, 77 Locally connected, 198
Greatest lower bound, 6 Locally metrizable, 230
 Index 269

Loop, 240 Product set, 85, 261

Lower bound, 5 Product space, 85, 261
greatest, 6 Projection, 85, 262
Pseudocompactness, 182
Mappings, 3 Pseudometric, 92
Maximal, 6 Pythagorean metric, 16
Metacompactness, 182
Metric, 16 Quasi-component, 206
complete, 218 Quotient space, 80
Metric space, 16 Quotient group, 82
Metrizable, 208
Metrizable, locally, 230 Range, 3
Minimal, 6 Refinement, 143
Mutually separated, 193 Refinement, locally finite, 228
Regular, 97
Neighborhood, 19, 60 Regular, completely, 230
Net, 117 Relation, 2
based on a filter, 135 Residual, 117
Noncut point, 202 Restriction, 3
Normal, 102 Retraction, 256
Riemann integral, 124
One-one, 3 Ring, 51
One-point compactification, 175 Rotation, 71
Onto, 3
Open cover, 142 Same number of elements, 9
Open function, 89, 172 Second countable, 145
Open neighborhood system, 49 Separable, 145
Open set, 21, 40 Separate points, 232
Open subcover, 143 Sequence, 3, 26, 113
Open in a subspace, 64 Set, 1
Operation, 13 partially ordered, 5
Ordering, induced, 5 Split, 202
partial, 5 Subbasis, 46
total, 5 Subcontinuum, 205
Subcover, 143
Paracompactness, 228 Subgroup, 15
Partial ordering, 5 Subnet, 119
Partition, 8, 123 Subsequence, 119
finer, 124 Subset, 1
induced by equivalence relation, 8 bounded, 162, 164
mesh of, 124 Subspace, 17, 65
Path, 111, 206
Path connected, 189 To, 91
Permutation, 89 T i,93
Point, 1 T2, 95
cut, 196, 202 Ts, 97
Polygonally connected, 189 T4, 102
270 Index

Tietze’s extension theorem, 110 Totally bounded, 223

Topological space, 41 Tychonoff space, 230
metrizable, 208 Tychonoff theorem, 158, 263
Topology, 40
coarser, 52 Ultrafilter, 136
discrete, 40 Uncountable, 9
finer, 52 Union, 1
identification, 80 Upper bound, 5
metric, 40 least, 6
open interval, 45 Urysohn’ s lemma, 107
order, 95 Urysohn’ s metrization theorem,
product, 85, 262 209
subspace, 65
torus, 101 Weak derived set, 56
trivial, 40
Total ordering, 5 Zorn’
s lemma, 7, 149