(Pure and Applied Mathematics) Donald L. Stancl, Mildred L. Stancl-Real Analysis With Point-Set Topology-Dekker (1987)

Real Analysis
with
Point-Set Topology
OONALD L. STANCL
MILDRED L. STANCL
ST. ANSELM COLLEGE
MANCHESTER, NEW HAMPSHIRE
MARCEL DEKKER, INC.
NEW YORK ANO BASEL
Library of Congress Cataloging-in-Publication Data

Stancl, Donald L.
Real analysis with point-set topology.
(Monographs and textbooks in pure and applied
mathematics; 113)
Bibliography: p.
Includes index.
1. Functions ofreal variables. 2. Mathematical
analysis. 3. Topology. 1. Stancl, Mildred L.
II. Point-set topology. III. Series: Monographs and
textbooks in pure and applied mathematics; v. 113.
QA331.5.S73 1987
515.8
87-3465
ISBN 0-8247-7790-5
COPYRIGHT 1987 MARCEL DEKKER, INC. ALL RIGHTS

RESERVED
Neither this book nor any part may be reproduced or transmi~ted in any
form or by any means, electronic or mechanical, including photocopying,
microfilming, and recording, or by any information storage and retrieval
system, without permission in writing from the publisher.
MARCEL DEKKER, INC.
270 Madison Avenue, New York, New York 10016
Current printing (last digit):

10 9 8 7 6 5 4 3 2
PRINTED IN THE UNITED STATES OF AMERICA
ln memory of
Bruce Amert Luzader
Minnie Farson Luzader
Frances W olff Stand
Preface
This book is designed as a text for a first course in real analysis. It is

specifically addressed to students who are unlikely to proceed to advanced
degrees in mathematics and for whom their first course in real analysis will
also be their last. It is our hope that the students who use this book will
develop
A solid understanding of the structure and properties ofthe real number
system and real-valued functions
A knowledge of the basic concepts and results of point-set topology
An appreciation for the interplay between these two areas ofmathematics
An appreciation for the role of examples in suggesting generalizations
An appreciation for the power of abstraction and its application to
particular cases.
This book presents the standard material of a first course in real
analysis: properties ofthe real numbers, functions on the reais, continuity,
sequences and series, sequences of functions, integration, and differentiation are ali covered. Mastery of this material will provide the student with
a ~ood understanding of the real numbers and real-valued functions. ln
V
Vl
Preface
addition we have also included material on the construction of the reais

and cardinal arithmetic, topics which are seldom found in a text at this
levei, on the ground that these are so important and interesting that they
should be made available to ali students of mathematics, even those who
will not proceed further in the discipline.
The major feature of this book lies in its presentation of the basic
concepts of point-set topology in tandem with real analysis. Our reasons
for including an introduction to point-set topology in a real analysis book
are as follows:
Point-set topology is a subject that the students to whom this book is
addressed are unlikely to encounter elsewhere.
lts ideas and results can easily be related to properties of the real numbers and real-valued functions, and often help in making these properties clearer.
lts proofs tend to be rigorous but relatively nontechnical, and proving
results in the general topological setting is seldom more difficult, and
often easier, than proving them on the real line.
The abstraction to the general topological setting helps students understand what a proof is and how proofs are constructed.
The interplay between topology and analysis allows the students to
observe and appreciate the interplay between abstraction and particularization that is so important to mathematics.
Students find point-set topology an easily accessible and interesting
subject.
We have attempted to interweave topology and analysis in such a way
that each illuminates the other. Thus we continually proceed in two directions: we use knowledge of the real numbers and functions on the reais to
suggest and illustrate topological generalizations, and we elucidate general topological concepts and results and immediately apply them to real
analysis. This approach not only presents the basic ideas of point-set
topology, but will also, we hope, make the power of generalization and
abstraction clear to the student.
As a supplement to the standard definition-theorem-proof format, we
have provided copious examples. There are also exercises at the end of
each section; these present a spectrum of difficulty, ranging from simple
applications of the results of the text to quite challenging problems. Some
exercises are designed to supplement the text by introducing additional
topics of interest. Hints are given for the more difficult exercises.
The organization of the text is straightforward. The sections concern-
Preface
Vil
ing construction of the reais (Section 2.2), completion of metric spaces

(Section 6.4), and function spaces and uniform approximation (Section
7.2) may be pmitted without affecting later material. Section 5.3, on infinite series, may be omitted at the cost of eliminating some later exercises,
while that on convergence ofsequences offunctions (Section 7.1) may be
omitted at the cost of eliminating discussion ofthe integrability and differentiability of sequences of functions in Chapter 8. The material on set
equivalence and cardinal numbers in the Appendix can be covered at any
time after Chapter 1.
We would like to thank the reviewers for their criticisms and suggestions. Our special thanks to Dr. Nicholas N. Greenbaun ofTrenton State
College, who class-tested the manuscript and provided us with many
helpful comments. Finally, our appreciation to the staff at Marcel
Dekker, Inc. for their cooperation and assistance.
Donald L. Stancl
Mildred L. Stancl
Contents
Preface
1 Sets and Functions

1.1 Sets
1.2 Functions
1.3 Finite and Infinite Sets
1
16
31
2 The Real Numbers

2.1 Properties of Real Numbers
2.2 Construction of the Real Numbers
41
3 Topology
3.1 Topological Spaces
3.2 Open Sets and Closed Sets
57
4 Continuous Functions
4.1 Continuity
4.2 Connectedness and Compactness
75
5 Sequences and Series

5.1 Sequences
5.2 Real Sequences
41
51
57
66
75
90
107
107
116
ix
Contents
5.3 Infinite Series

5.4 Functional Limits
126
Metric Spaces
139
139
6.1
6.2
6.3
6.4
7
The Metric Topology

Continuity in Metric Spaces
Sequences in Metric Spaces
Completion of Metric Spaces
Sequences of Functions
7.1 Pointwise and Uniform Convergence

7.2 Function Spaces and Uniform Approximation
8
134
149
156
165
177
177
190
Calculus
201
8.1 The Riemann Integral

8.2 Properties of the Integral
8.3 The Derivative
201
226
241
Appendix : Cardinal Numbers
265
Bibliography
281
Index
283
1
Sets and Functions
ln this book we shall study various sets and functions, paying particular
attention to the set of real numbers and to functions whose values lie in
the real numbers. This introductory chapter presents some basic definitions and results concerning sets and functions which will be used
throughout the remainder of the text.
1.1
SETS
We begin our study of sets with a definition.
Definition 1.1.1 A set is a collection of objects, called the elements of the

set. If object x is an element of set S, we say that x belongs to S, and write
xeS; if object x is not an element of set S, we say that x does not belong
to S, and write xrtS (see Figure 1.1). The empty set, or null set, is the set
which has no elements. The empty set is denoted by 0.
We specify a set S either by listing its elements between braces or by
writing a statement of the form
S = {x 1x has property P}
where P is some property which serves to define the elements of S unambiguously. The statement S = {x 1 x has property P} is read "S is the set
of ali elements x such that x has property P."
Sets and Functions
o
Figure 1.1
Element of a set.
1. The following notation will be standard throughout

the text: the set of natural numbers (positive integers) will be denoted by
N, the set of integers by Z, the set of rational numbers by Q, and the set
of real numbers by R. Thus we may write
Examples 1.1.2
N={l,2,3,4, ... }
Z={O, 1, -1,2, -2, ... }

and
Q
= {: m E Z, n E Z, n -:f
1
o}
We assume that the reader is familiar with the arithmetic operations

(addition, subtraction, multiplication, division, and exponentiation) and
the order relationships ( <, ~, >, ~) defined on these sets. The reader is
undoubtedly aware that the set of real numbers R may be represented as
a horizontal line called the real fine (see Figure 1.2). Each point on the real
line corresponds to a unique real number and each real number corresponds to a unique point on the line. We shall speak interchangeably of
the set of real numbers R and the real line R.
2. There may be severa! different ways of specifying the elements of
a given set. For example, if
s=
-3
-2
-1
{!, 2, 3}
Figure 1.2 The real line.
. 3
Sets
we may also write
S={xlxeN,x<4}
or
S = {xeZ 1x 3 - 6x 2 + I Ix - 6 =O}
3. Any statement of the form {x 1x has property P} where P is a

property which no object can satisfy serves to define the empty set f/J. For
instance,
f/J={x lxeR, x 2 <0}

since there is no real number whose square is negative.
4. Let aeR and beR, with a< b. The set
(a, b) = {xeR 1a< x < b}
is called an open interval in R (see Figure 1.3). The real numbers a and b
are called the endpoints of (a, b). Note that the open interval (a, b) consists
of ali real numbers which are between a and b on the real line, but that the
endpoints a and b do not belong to (a, b). The open interval (O, 1) is
referred to as the open unit interval.
a
Figure 1.3
The open interval (a, b).
5. Let aeR and beR, with
a~
b. The set
[a, b] = {xeR 1a~ x ~ b}
is called the closed interval in R with endpoints a and b (see Figure 1.4).
Note that the endpoints a and b belong to the closed interval [a, b]. Also,
if a = b, then [a, b] =[a, a] ={a}, so a closed interval may consist of a
single point. The interval [O, l] is the closed unit interval.
Sets and Functions
Figure 1.4 The closed interval [a, b].
6. If aeR and beR with a< b, then the sets
(a, b] = {xeR 1 a <x ~b}

and
[a,b)={xeRla~x<b}
are half-open intervals in R with endpoints a and b (see Figure 1.5).
7. If a e R, then the sets
(a, +oo) = {xeR 1a <x}

and
(-oo, a)= {xeR 1 x <a}

are called open rays in R. The sets
[a, +oo) = {xeR 1a ~x}

and
(-oo,a]={xeRlx~a}
are called closed rays in R (see Figure 1.6). The set Ris both an open ray
b
(a,b]
b
[a,b)
Figure 1.5 Half-open intervals.
Sets
a
(a,+:x:)
a
(-:x:,a)
a
(a,+:x:)
a
(-:x:,a]
Figure 1.6
Rays.
anda closed ray and may be written as
R=(-oo,+oo)
Next we define the notion of a subset of a given set.
Definition 1.1.3 A set S is a subset of a set T if and only if every element

of S is also an element of T. If S is a subset of T, we write S e T and say
that S is contained in T. If S is not a subset of T, we write S <t T and say
that S is not contained in T. If S e T and Te S, then S = T. If S e T but
S =F T, we say that S is a proper subset of T (see Figure 1.7).
Sets and Functions
S C T
cf_
Figure 1.7 Set containment.
Examples 1.1.4
1. The following are true for any set S:
</JcS
ScS
and
S=S
2. For the sets N, Z, Q, and R we have
NcZcQcR
Furthermore, N is a proper subset of Z, Z is a proper subset of Q, and Q
is a proper subset of R.
3. Let aeR and beR, with a< b; then
(a, b) is a proper subset of [a, b]
(a, b) is a proper subset of (a, b]

[a, b) is a proper subset of [a, b]
and
(b,
4.
subset
subset
5.
+ oo) is a proper subset of (a, + oo)
Every interval in R, whether open, half-open, or closed, is a proper

of R. Every ray in R, except for the ray ( - oo, + oo ), is a proper
of R.
Let
and
T=(O, 1)
We claim that S = T. To prove this, we must show that Se T and that
Sets
T e S. To show that S e T, let x be an arbitrary element of S. The

condition 1/x > 1 then implies thi:i.t x is positive and less than l; hence
xe(O, 1) = T. Therefore every element of S is also an element of T, and
thus Se T. To show that Te S, let ye T, so that O<y < 1. But then
1/y > 1 and hence y e S. Therefore every element of Tis also an element
of S, and thus Te S.
The technique used in the previous example is the standard method

by which set identities are proved. lt may be summarized as follows: to
prove that S = T, let x be an arbitrary element of S and show that xeT,
thus establishing the inclusion S e T; then let y be an arbitrary element of
T and show that y eS, thus establishing the inclusion Te S. The inclusions S e T and Te S then imply that S = T.
Now we are ready to consider how sets can be combined to form new
sets. There are four common methods of doing this: by forming unions,
intersections, complements, and Cartesian products of sets. We now
define these set operations.
Definition 1.1.5 Let S and T be sets.
1. The union of S and Tis the set Su T defined by
SuT= {x 1 xeS or xeT}
(see Figure l.8a).
2. The intersection of S and T is the set S n T defined by
S n T = {x 1 x eS and x e T}
(see Figure l.8b). If S n T = </J, we say that S and Tare disjoint sets.
3. The complement of T in S is the set S - T defined by
S - T = {x 1x e S and x ~ T}
(see Figure l.8c).

4. An ordered pair with first coordinate x from S and second coordinate y from Tis a pair (x, y) where x e S and y e T. Two such ordered pairs
(x, y) and (x', y') are equal if and only if x = x' in S and y = y' in T. The
Cartesian product of S with Tis the set S x T of ali such ordered pairs,
that is,
S x T={(x,y) lxeS;yeT}
Sets and Functions
SvT
union
(a)
SnT
intersection
(b)
S-T
complement
(e)
Figure 1.8 (a) Union, (b) intersection, (e) complement of sets.
Bxamples 1.1.6
1. Let S = {!, 2, 3} and T = {3, 4}. We have
SuT={l,2,3,4}
SnT={3}
S-T={l,2}
T-S={4}
T = {(l, 3), (1, 4), (2, 3), (2, 4), (3, 3), (3, 4)}
and
T x S = {(3, 1), (3, 2), (3, 3), (4, 1), (4, 2), (4, 3)}
Sets
9
2. Let
R 2 =R xR={(x,y) lxeR,yeR}
The set R 2 is called the Euclidean plane; it may be represented by the

familiar Cartesian coordinate system as depicted in Figure 1.9.
3. Consider the intervals (O, 1), [l, 2], and [2, 3] in R. We have
(O, 1) u[l, 2] =(O, 2]
(O, 1) n[l, 2] =
[l, 2] n[2, 3] = {2}
and
[l, 2] -[2, 3] = [l, 2)
[l, 2] - (O, l)
= [l, 2]
Note that (O, 1) and [l, 2] are disjoint, but that [l, 2] and [2, 3] are not
disjoint. The Cartesian product
[l, 2] x [2, 3] = {(x,y) 1xe[l,2], ye[2, 31}
is the subset of the Euclidean plane R 2 shown in Figure 1.10.
The algebraic rules which govern the set operations ofunion, intersection, and complementation are given in the following proposition.
y-axis
- - - - - - - , (x,y)
1
1
1
1
- - - - + - - - - - - - + - - - - - - - - x-axis
X
Figure 1.9
The Cartesian coordinate system for R 2
Sets and Functions
10
y
3
--~
--~
1
1
1
1
2
Figure 1.10 The Cartesian product [ l, 2] x (2, 3].
Proposition 1.1.7
Let S, T, and V be sets. The following set identities
hold:
l.
2.
3.
Idempotent laws
Commutative laws
Associative laws
4.
Distributive laws
S.
deMorgan's laws
SuS=S
SnS=S
SuT=TuS
SnT=TnS
Su(Tu V)= (Su T) u V
S n (T n V) = (S n T) n V
Su(Tn V)= (Su T) n(Su V)
S n (Tu V)= (S n T) u (S n V)
S - (Tu V) = (S - T) n (S - V)
S - (T n V) = (S - T) u (S - V)
Proof: We shall prove the identity S u (T n V) = (S u T) n (S u V) of

( 4). The remainder of the proof is similar and is left as an exercise.
Suppose that Su(Tn V) =F</J and let xeSu(Tn V), so that
either xeS or xeTnV. If xeS, then xeSuT and xeSuV, so
xe(S u T) n(Su V); if xeT n V, then xeTand xe V, so again xeSu T
and xeSuV, and hence xe(SuT)n(SuV). Therefore if
S u (T n V) =F </J then
[S u (T n V)] e [(S u T) n (S u V)]

But this inclusion is also true if S u (T n V) = </J, because the empty set is
a subset of every set; hence the inclusion always holds.
Now suppose that (S u T) n (S u V) =F </J and let y e(S u T) n
(S u V), so that yeSu T and yeSu V; if yeS, then surely
yeS u(T n V), while if y~S then yeT and ye V, so that yeT n V, and
11
Sets
again y eS u (T n V). Therefore if (S u T) n (S u V) =F </J then
[(Su T) n(Su V)] e [Su(Tn V)]
But, as above, this inclusion is also true if (S u T) n (S u V) = </J,

and hence it always holds. Thus we have shown that
Su(Tn V)= (Su T) n(Su V). D
It will sometimes be necessary for us to distinguish between finite and
infinite sets. We will postpone the formal definitions of these terms until
Section 1.3, but for now we will consider a set to be finite if its elements
can be counted and the counting process terminates; if the elements of a
set cannot be counted, or if they can but the counting process does not
termina te, then we consider the set to be infinite. Thus the set S = {a, b, e}
is finite; the set S = {1, ... , n} is finite for any n eN; and the empty set is
finite. On the other hand, the set N is infinite because if we count its
elements the counting process will not terminate. For the sarne reason, the
sets Z and Q are infinite. The set of real numbers R is infinite because its
elements cannot be counted. (We shall prove this fact in Chapter 2.)
We may form unions and intersections of any finite number of sets by
taking them two ata time. For instance, suppose we have sets S 1, S2 , S 3 ,
and S4 Then we may form
and
But the commutative and associative laws of Proposition 1.1.7 (together
with mathematical induction) imply that in such a union neither the order
of the sets nor the manner of their grouping matters. Thus we may write
Similar remarks apply to intersections of finitely many sets.

ln addition to thus being able to form unions and intersections of
finitely many sets, we would like to be able to form unions and intersections of infinitely many sets. An efficient way to do this is to use the
concept of an index set.
Let A be a nonempty set. (A may be finite or infinite.) If
to each element ex e A there corresponds a set X", then the elements of A
are sai d to index the family of sets {X" 1 ex E A}, and A is called an index
Definition 1.1.8
12
Sets and Functions
set. We define the union and intersection of the family of sets {XIX 1 tX EA}
as follows:
U XIX= {x 1xeXIX for some tXEA}
IXE
n XIX= {x
IXE
1 XEXIX
for ali tXEA}
The sets {XIX ltXEA} are mutually disjoint if and only if XIXnXp=</J
whenever tXEA, /JEA, and tX =F {3.
Examples 1.1.9 1. Let A = { 1, 2, 3}, and for each tX EA let XIX denote the
closed interval [tX, tX + 3] in R; then
U XIX= Xi uX2uX3 = [l, 4] u[2, 5] u[3, 6] .= [l, 6]
IXE
and
n XIX= Xi nX2nX3 = [l, 4] n[2, 5] n[3, 6] = [3, 4]
IXE
Here the index set A is finite and we have taken the union and intersection
of finitely many closed intervals in R.
2. Let A= N, and for each nEN let Xn = {l, ... , n}; then
U XIX= neN
U Xn=N
n XIX = n Xn = { 1}
and
IXE
czeA
neN
Here the index set is infinite, and we have taken the union and intersection
of infinitely many sets, each of which was a subset of N.
3. For each n EN, let Xn be the half-open interval [n - 1, n) in R; then
U Xn =[O, +oo)
neN
and
l Xn=</J
neN
(see Exercise 15 of Section 1.1). Note that the sets {Xn} are mutually
disjoint and that lneN Xn = 0.
4. The previous example suggests that if {XIX 1tX EA} is a family of
mutually disjoint sets, then
IXE
XIX=</J
This is indeed true (see Exercise 16 in Section 1.1). However, the intersec-
Sets
13
tion may be empty even if no two of the sets {X" 1tX E A} are mutually
disjoint. To see this, let
for each n eN. No two of the sets {Xn 1n eN} are- disjoint, for if m ~ n
then Xm e Xn and hence Xn n Xm =F f/J. However, it is easy to check that
Xn=</J
neN
( see Exerci se 15 in Section l. l ).

5. For each reR, let X, be the ordered pair (r, r
+ l) in R 2 ; then
U X,= {(r, r + l) 1 reR}

reR
Ifwe re)>resent R 2 by means ofthe Cartesian coordinate system ofExample 2 in Examples l. l .6, the above union is the line with slope l and
y-intercept l (see Figure l.l l).
Now that we can form unions and intersections of arbitrarily many
sets, we must generalize the distributive laws and deMorgan's laws of
Proposition l.l.7.
Proposition 1.1.10
Let Y be a set. If A is a nonempty set and
{X" 1 tXEA}
{(r,r+ 1) 1 rER}
Figure 1.11
A union of ordered pairs in R 2
Sets and Functions
14
is a family of sets indexed by A, then

1.
Yu ( (1 X"') =
ex e A
3.
Y-
U X"'=
ex e A
4.
Y - (1 X"' =
ex e A
(1 (
ex e A
Yu X"')
(Y-X"')
ex e A
U (Y -
X"')
ex e A
Proof: We will prove (1), leaving the remainder of the proof as an

exercise. For ease of notation, we write Ucx for UcxeA and lcx for lcxeA"
We wish to show that Yu ( lcx X"') = lcx ( Yu X"'). This is the generalization to arbitrarily many sts of the identity we proved in Proposition
1.1. 7. The reader should compare the proof we give here with that of
Proposition 1.1.7.
Suppose Y u ( lcx X"') # </J. If e Y u ( lcx X"'), then either x e Y or
x e lcx X"'. If x e Y, then x e Y u X"' for ali oc e A; if x e lcx X"', then x e X"' for
ali oceA and hence xe YuX"' for ali oceA. Thus, in either case, xe YuX"'
for ali oceA, so xe lcx (YuX"'). Therefore we have shown that if
Yu( lcx X"')# </J, then
and the inclusion also holds if Y u ( lcx X"') = </J.

Now suppose lcx (YuX"') # </J. If ye lcx (YuX"')' then ye YuX"' for
ali oceA. If ye Y, thenye Yu( lcx X"'); if y~ Y, thenyeX"' for ali oceA, and
hence y e lcx X"', which implies that y e Yu ( lcx X"'). Therefore
lcx (Y u X"') # </J, and since

lcx (Y u X"') = </J, we are done. D
if
the
inclusion
also
holds
when
15
Sets
To conclude this section, we offer a few remarks concerning "if and

only if" proofs. (Some of the exercises which follow request "if and only
if" proofs.) Suppose A and B are mathematical statements. Then the
phrase "A if and only if B" means that A implies B and also that B implies
A; that is, it means that if A is true, then B must be true and also that if
B is true, then A must be true. Therefore in order to prove that a statement of the form "A if and only if B" holds, it is necessary to do two
things: (a) assume that A is true and show that this implies the truth of B,
and (b) assume that Bis true and show that this implies the truth of A.
Thus every "if and only if" proof comes in two parts, which are called the
converses of one another.
EXERCISES
l.
Let S = {a, b, e, d}, T = {a, e}, and V= {e, d, e}. Find

SuT
SuTuV
Sn(TuV)
SnT
S-T
T-S
TxV
2.
S- T
4.
5.
6.
7.
8.
Su(TnV)
S-(TnV)
and
V-(S-T)
VxT
Let S = {n e N 1n = 3k for some k e N}, T = {n e N 1 n = 4k for

somekeN}, and V={l,2,3,5, 7, li}. Find
SuT
3.
SnTnV
SnT
T- S
SuTuV
and
SnTnV
V -(Su T)
Prove that any interval in R ( open, closed, or half-open) can be

written as the intersection of two properly chosen rays in R.
How many elements are there in the set O? ln the set {O}? ln the set
{O, {O}}?
Let S ={a, b, e}. Let V= {T 1 Tis a subset of S}. Write out the
elements of the set V.
Let n e N and let S = {1o ... , an}. Prove that S has 2n distinct subsets.
Let S and T be sets. Prove that S - T = 0 if and only if S e T.
Let S and T be sets. Prove that S = (S n T) u (S - T) and that
S n T and S - T are disjoint.
16
Sets and Functions
9.
Let S, T, T1, and T2 be sets, with T = T1 u T2 Prove that

s X T = (S X T1) u (S X T2).
Let S, T, and V be sets.
(a) Prove that if V e T, then (S - T) e (S - V).
(b) Suppose that both T and V are subsets of S. Prove that
V e T if and only if (S - T) e (S - V).
(e) Suppose that both T and V are subsets of S. Prove that
T = S - V if and only if V= S - T.
Prove (l) and (2) of Proposition l.l.7.
Prove (3) of Proposition l.l.7.
Prove the second distributive law in ( 4) of Proposition l. l. 7.
Prove (5) of Proposition l.l.7.
(a) For each neN, let Xn be the interval [n -1, n) in R. Show that
U.eN Xn =[O, + 00) and that lneN Xn = 0.
(b) For each neN, let Xn be the interval (O, l/2n) in R. Show that
U.eN Xn =(O, 1/2) and that lneN Xn = 0.
Let A be an index set. Prove that if the sets {X"' 1tX E A} are mutually
disjoint, then lixeA X"' = 0.
Prove Proposition l.l.10.
Let A be an index set, and for each tX EA, let X"' be an open interval
in R. Prove that if lixeA X"' # 0, then UixeA X"' is either an open
interval or an open ray.
10.
11.
12.
13.
14.
15.
16.
17.
18.
1.2 FUNCTIONS
ln arder to study sets more thoroughly, we need to be able to compare one
set with another. This means that we must have some method of "pairing" their elements and leads us to the concept of a function from one set
to another.
Let S and T be sets. A function from S to T is a subset
T such that
Definition 1.2.1
f of S x
l. For each x eS there is some y E T such that the ordered pair (x, y) Ef
2. If (x, y) and (x, y') are elements off, then y = y' in T.
The set S is called the domain of the function and the set T the range of
the function. If S = T, we refer to a function from S to S as a function
on S.
Note that according to this definition, a function from S to Tis a
17
Functions
subset of the Cartesian product S x T and hence is a set of ordered pairs,

with the first coordinate of each ordered pair being an element of the
domain S and the second an element of the range T. Condition ( l) of the
definition says that every element of S must appear as the first coordinate
of some ordered pair which belongs to the function; in other words, every
element of the domain must get "used" by the function. Condition (2)
says that no element of S can appear as the first coordinate in more than
one ordered pair which belongs to the function; in other words, no element ofthe domain can get "used" more than once by the function. There
are no restrictions on the use of the elements of the range T: the function
may "use" some or ali of the elements of T, and it may "use" them more
than once.
Examples 1.2.2 l. Let S = {l, 2, 3} and T ={ex,
subsets of S x T:
= {(l,
ex), (2, p), (3, P)}
h = {(2, p), (3, y)}
and
p, y }, and consider the
= {(I, y), (2, ex), (3, p)}
= {(l, ex), (l, p), (2, y), (3, y)}
Since each element of S appears as the first coordinate of some ordered pair belonging to f, and no element of S appears as the first coordinate of more than one ordered pair belonging to f, f is a function from S
to T. Note that not every element of the range T appears as the second
coordinate of an ordered pair inf; note also that the element pe T appears
as a second coordinate more than once.
The set g is also a function from S to T; in this case, every element of
T does appear as the second coordinate of some ordered pair belonging to
the function, and no element of T appears more than once.
The set h is not a function from S to T because it violates condition
(l) of the definition: 1 eS but there is no y e T such that ( l, y) eh. (However, h is a function from the set {2, 3} to T.)
The set k is not a function from S to T because the element l of
S appears as the first coordinate of more than one ordered pair belonging
to k.
2. Consider the subsetf of N x Q given by f= {(n, 1/n) 1neN}. We
claim thatf is a function from N to Q. This follows beca use if n eN, then
(n, l/n) e f, and if (n, y) and (n, y') both belong to f, then y = l/n = y'.
3. Let
f={(x,x 2 ) lxeR}
and
= {(x 2, x) 1 xeR}
Sets and Functions
18
The subsetf is a function on R (i.e., a function from R to R), because if

xeR, then (x,x 2)e/and if (x,y)e/and (x,y')ef, then y=x 2 =y'.
However, g is not a function on R, because, for instance, (l, 1) and
(l, -1) both belong to g.
4. Let
{(x, x) 1 xeR, x ~O} u{(x, -x) 1 xeR, x <O}
then f is a function on R, called the absolute value function. Figure l.12

depicts f as a subset of R x R: such a depiction is called the graph of the
function. Every function whose domain and range are subsets of R has a
graph.
We have defined a function from S to Tas a subset of S x T having
certain properties. It is often convenient to think of a function as a "rule
of correspondence" or "mapping" which assigns to each element of its
domain a unique element of its range. See Figure 1.13 for a depiction of
this viewpoint. ln line with this approach, we adopt the following notation: we will write /: S-+ T to signify that f is a function from S to T and
write f(x) = y if and only if the ordered pair (x, y) e f
l. Let S = {l, 2, 3} and T = {tX,
be the function defined by
Examples 1.2.3
f(l)
=(X
and
p, y }, and letf: S-+ T
/(2) = /(3) =
This is the functionf of the first example in Examples l.2.2. The function
y
{(x,x) 1 XER, x;;., O}
{(x,-x) 1 XER, x < O}
Figure 1.12 The graph of the absolute value function.
19
Functions
f:S ~ T,
f(x) = y
A function as a mapping.
Figure 1.13
g of the sarne example may be written as g: S-+ T, where

g(l) = y
g(2) =
(l(
g(3) =
2. Let/: R-+R be given by/(x) = x 2 for all xeR. This is the function
f of the third example in Examples 1.2.2. Its graph is depicted in Figure
1.14.
3. Let/: R -+R be given by /(x) = lxl for ali xeR, where by definition,
if x~O
if x<O
This is the absolute value function introduced in Example 4 of Examples
1.2.2.
y
f(x) = x2
Figure 1.14
The graph of f(x) = x 2
20
Sets and Functions
4. Let us show that the correspondence from S = [O, + oo) to

T=[O, +oo) given by f(x)=Jx for ali xe[O, +oo) is a function.
Here Jx denotes the nonnegative square root of x. We must show
that
a.
b.
For each x in the domain, f(x) is defined and belongs to the range.
Iff(x) =y andf(x) =y', then y =y'.
But if x E [O, + oo ), then x ~O, so f(x) = Jx ~O, and thus f(x) is

defined and belongs to [O, + oo ). Hence (a) is satisfied. Furthermore, if
f(x) = y and f(x) = y', then y = Jx = y', so (b) is satisfied. Therefore
f(x) = Jx is a function from [O, + oo) to [O, + oo ).
Now we are ready to consider two special types of functions: onto
functions and one-to-one functions.
Definition 1.2.4
Let S and T be sets and let /: S - T.
If for every y e T there is some x e S such that f(x) = y, then f is said

to be a function from S onto T.
2. If f(x) = f(z) implies that x = z, then f is said to be a one-to-one
function.
1.
Thus /: S - Tis onto if every element of its range gets "used," and it
is one-to-one if no two elements of its domain are assigned to the sarne
element of its range ( see Figures 1.15 and 1.16).
1. The function f of Example 1 in Examples 1.2.3 is not
a function of S onto T beca use y E T and there is no x E S such that
f(x) = y. Furthrmore,fis not one-to-one because/(2) = f3 and/(3) = f3.
However, the function g of the sarne example is both onto and oneto-one.
2. The function of Example 2 in Examples 1.2.3 is not a function
of R onto R beca use, for instance, - 1E R but there is no x E R
such that f(x) = - 1. It is not one-to-one because, for instance,
f ( - 1) = f (l) = 1.
3. The absolute value function of Example 3 in Examples 1.2.3
is neither onto .R nor one-to-one. (Why?) The square root function of
Example 4 in Examples 1.2.3 is onto [O, + oo) because if y e [O, + oo ),
then
Examples 1.2.5
y 2 e [O, +oo)
and
Functions
21
f is onto
f is not onto
Figure 1.15
The concept of an onto function.
lt is one-to-one because if xe [O, +oo), ze[O, +oo) and
Jx=Jz,
then
=z.
4. Let S and T be any nonempty sets and let e e T. Let /: S -+ T be
given by f(x) = e for all x e S. Such a function is called a constant function.
A constant function cannot be onto if its range contains more than one
element; it cannot be one-to-one if its domain contains more than one
element.
X
If /: S -+ T and U is a subset of S, we can consider the subset of T

which consists o"f all elements y e T such that y = f(x) for some x e U.
Similarly, if V is a subset of T, we can consider the subset of S which
consists of all elements x of S such that f(x) e V.
22
Sets and Functions
f is one-to-one
f is not one-to-one
Figure 1.16 The concept of a one-to-one function
Definition 1.2.6 Let /: S

l.
-+
T, and let U e S and V e T.
The image of U under f, denoted by f( U), is the subset of T defined by

f(U) = {f(x)eT 1xeU}
2. The preimage of V under f, denoted by 1- 1(V), is the subset of S

defined by
1- 1(V) = {xeS lf(x)eV}

( see Figure l.17).
Functions
23
f(U)CT
Figure 1.17 Image and preimage.
Examples 1.2.7
given,by ,
l. Let S ={a, b, e}, T = {l, 2, 3, 4}, and /: S-+ T be
f(a) = 3
f(b) = 4
f(c) = 3
If U ={a, b }, thenf(U) = {3, 4}. If V= {l, 2, 4}, thenf- 1(V) = {b }.

2. Let S and T be any sets and /: S -+ T any function from S to T. We
have/(0) = 0, 1- 1(0) = 0, 1- 1(T) = S, and /( U) e f(S) for ali U e S.
(Check these facts.)
3. Letf: R -+R be given by f(x) = x 2 for ali xeR. We will show that
the image of [O, 2] under f is [O, 4] and that the preimage of [O, 4] under
f is [ -2, 2].
24
Sets and Functions

Bydefinition,f([O, 2])
{x 2 eR 1xe[O,21}. Ifxe[O, 2], thencertainly
f(x) = x 2 e [O, 4]; hence/([O, 2]) e [O, 4]. On the other hand, if xe [O, 4],
then Jxe [O, 2] andf(Jx) = xe [O, 4]; hence [O, 4] c/([O, 2]). Therefore
/([O, 2]) =[O, 4].
Now consider 1- 1([0, 4]) = {xeR 1x 2 e [O, 41}. If xe [ -2, 2], then
~learly f(x) = x 2 e [O, 4]; hence [
2] c/- 1([0, 4]). On the other hand,
2
2
if f(x) = x e [O, 4], then x = y' x e [ -2, 2], so 1- 1([0, 4]) e [ -2, 2].
Therefore/- 1([0, 4]) = [ -2, 2].
-'fo
Our next proposition tells us how set operations are affected by functions. lt states that taking preimages preserves containments, unions,
intersections, and complements, while taking images preserves containments and unions. Intersections and complements are not in general preserved by taking images.
Let S and T be sets and /: S
subsets of S and V1 and V2 be subsets of T.
Proposition 1.2.8
1.
-+
T. Let U1 and U2 be
If U1 e U2 , then/(U1) c/(U2).
If V1 e V2 , then/- 1(V1) c/- 1(V2).
2. f(U 1 u U2 ) = f(U 1) u /(U2)

1- 1(V1 u V2 ) = / - 1(V1) u/- 1(V2)
3. f(U 1 n U2 ) e [f(U1) n/(U2)]
1- 1(V1 n V2 ) = 1- 1(V1) n/- 1(V2)
4. f(U 1 - U2) c/(U1)
J- 1(V1 - V2) =/- 1(V1) - / - 1(V2)
Proof: We will prove (3), leaving the remainder of the proof as an

exercise.
Suppose U1 n U2 =I= f/J and let x e U. n U2 ; then x e U1 and x e U2 , so
f(x)ef(U 1) and f(x)ef(U2). Thus f(x)ef(U 1)nf(U2 ). Hence
f(U 1 n U2) e [/(U1) n/(U2 )] when U1 n U2 =F f/J, and this inclusion also
holds when U. n U2 = f/J. (Why?)
Now assume that/- 1(V1 nV2)=Ff/J and let xe/- 1(V1 nV2). Then
/(x)eV1 n V2 , and therefore xe/- 1(V1) and also xe/- 1(V2). Hence
xe/- 1(V1) r.i/- 1(V2), and thus when 1- 1(V1 n V2 ) =F f/J, then
1- 1(V. n V2) e [/- 1(V1) n/- 1(V2)]. This inclusion also holds when
J- 1(V1 n V2) = f/J. (Why?)
Finally, if 1- 1(V.)n/- 1(V2 ) =Ff/J and xe/- 1(V1)n/- 1(V2), then
xe/- 1(V1) and xe/- 1(V2), so f(x)e V1 and /(x)e V 2 Therefore
/(x)eV1 n V 2 , which implies that xe/- 1(V1 n V 2 ). Hence when
25
Functions
1- 1(V1) nf- 1(V2)
=/d/J, we have
[f- 1(V1) nf- 1(V2)] cf- 1(V1 n V2 )
and this inclusion also holds when

J- 1(Vi n V2) = J- 1(V 1) n/- 1(V2). O
1- 1(V1) n/- 1(V2)
= </J. Therefore
Example 1.2.9 Let /: R -+R be given by f(x) = x 2 for ali xeR. Let
U 1 = [ - 2, O] and U2 =[O, 2]. We have U 1 n U2 = {O} and thus
whereas
f(U1) nf(U2) =[O, 4] n[O, 4] =[O, 4]
This shows that in general/(U1 n U2) # f(U 1) nf(U2 ). Of course Proposition 1.2.8 says that /( U 1 n U2) e [f( U1) n /( U2)], and this is certainly the
case here.
Now consider f(U 1 - U2) and/(U1) - /(U2)_. We have
f(U 1 - U2) = /([ -2, O))= (O, 4]
whereas
Thus in general f( U 1 - U2 ) # f( U 1) - f( U2). Note that Proposition 1.2.8

states that/(Ui - U2) c/(Ui), and that this isso here.
We now turn to the question of how to use known functions to create
new ones. Our first result in this regard says that we can create a new
function from an old one simply by restricting the domain of the ld one.
Prop9sition 1.2.10 Let /: S -+ T be a function from a set S to a set T and

let U be a subset of S. Define g by setting g(x) = f(x) for all x e U; then
g: U-+ Tis a function called the restriction off to U. Furthermore, if f is
one-to-one, so is g.
The proof of Proposition 1.2.10 will be left as an exercise.
Example 1.2.11 Let /: R -+ R be the absolute value function. If
U e [O, + oo ), the restriction of f to U is the function g: U-+ R given by
g(x) = x for ali x e U. If U e ( - oo, O), the restriction of f to U is the
function g: U-+R given by g(x) = -x for ali xeU.
26
Sets and Functions
(g f) (x) = g(f(x))
Figure 1.18 Composition of functions.
Next we show how to create new functions from old by means of a

process called composition offunctions. The idea of composition of functions is this: if /: S-+ T and g: T-+ U, then for each x e S, the element
f(x) e T, and we can apply g to f(x) to obtain a unique element of U. ln
this manner we can assign to each element of S a unique element of U,
thus constructing a function from S to U. Figure 1.18 depicts the process
of composition of functions.
Proposition 1.2.12 Let S, T, and U be sets, let f: S-+ T and g: T-+ U,

and define g f as follows:
(g f)(x) = g(f(x))
for ali xeS
Then (g f): S -+ U is a function called the composite of g with f ln both

f and g are onto, so is g f; if both f and g are one-to-one, so is g f
Proof: To show that g ofis a function from S to U, we must show that
for each xeS, (g of)(x) is a unique element of U. But since/is a function
from S to T, f(x) is a uni que element of T for each x e S and since g is a
function from T to U, g(f(x)) is a unique element of U. This establishes
that g f is a function from S to U.
Now we show that iff and g are both onto, so is g f If z e U, then
because g is onto, there is some y e T such that g(y) = z. But similarly,
since/is onto, there is some xeS such that/(x) = y. Therefore
(g of)(x) = g(f(x)) = g(y) = z
Functions
27
Finally we prove that if f and g are both one-to-one, so is g f If

(g of)(x) = (g of)(x'), for some xeS, x'eS, then g(f(x)) =g(f(x')). But
single g is one-to-one, we must have f(x) = f(x'), and then since f is
one-to-one, it follows that x = x'. D
1. Let S = {l, 2, 3}, T ={ex,
Let f: S -+ T and g: T-+ U be given by
Examples 1.2.13
f(l) =
f(2) =ex
f(3) =
and
p, y }, and
U = {u,
g(ex) = g(p) = u
v, w }.
g(y) = w
We then have g f: S-+ U given by

(g of)(l) = g(j(l)) = g(p) =
(g of)(2) =g(J(2)) =g(ex) =
(g of)(3) = g(j(3)) = g(p) =
and
Hence g of is a constant function from S to U.

2. Letf: R-+ R and g: R-+ R be given by f(x) = 2x and g(x) = x
for ali xeR. We then have (g of): R -+R given by
(g of)(x) = g(f(x)) = g(2x) = 2x
+ 1,
+l
Sincef and gare both one-to-one and onto, g ofis one-to-one and onto.
ln this case we can also form the composite functionf g. Thus we have
f g given by
(f g)(x) = f(g(x)) = f(x
+ 1) =
2(x
+ 1)
and f g is one-to-one and onto. Note that although both composite

functions g f and f g are defined, they are not equal.
The final method that we will consider for creating new functions
from old ones is that of forming the inverse function of a given one-to-one
function. We do this as follows: if f: S-+ T is one-to-one, then each
yef(S) corresponds to a unique xeS, namely, the unique element x such
thatf(x) = y. The inverse of f is a function fromf(S) to S, which assigns
the element x to the element y. See Figure 1.19.
28
Sets and Functions
f-1:f(S)-+S
Figure 1.19 Inverse function.

Proposition 1.2.14
Let S and T be sets, let f: S-+ T be a one-to-one

function, and for each yef(S), definef- 1(y) = x if and only if xeS and
f(x) = y; then f- 1:f(S)-+ S is a one-to-one onto function called the inverse function off Furthermore,
(f- 1 of)(x)
=X
for ali xeS
and
(j
f- 1)(y) = Y
for ali yef(S)
Proof: To show thatf- 1 is a function fromf(S) to S, we must show that

if yef(S), then f- 1(y) is a unique element of S. But by definition,
J- 1(y) =x, where xeS andf(x) =y. Since yef(S), such an element x
exists, and if x were not uni que, that is, if there were an elemen t z e S,
z # x, such thatf(z) = y, thenfwould not be one-to-one.
To show thatf- 1 is onto, note that if xeS, thenf(x) = yef(S), and
thus J- 1(y) = x. To show that J- 1 is one-to-one, suppose that y ef(S)
and also y'ef(S), withf- 1(y) =f- 1(y'); then there is some xeS such
thatf(x) = y andf(x) = y'. But sincefis a function, we must have y = y'.
To conclude the proof, let xeS and set y = f(x); then
Similarly, if yef(S) and x =f- 1(y), thenf(x) =y and

(f of- 1)(y) =f(J- 1(y)) =f(x) =y D
29
Functions
We emphasize here that the inverse functionf- 1 exists if and only if

the function f is one-to-one; if f is not one-to-one, we cannot form 1- 1
Also, the inverse function 1- 1 should not be confused with the preimage
1- 1( V) of a subset V of the range of /: this preimage is a subset of the
domain, and it always exists, whether or not f is one-to-one.
Examples 1.2.15 1. Let S = {l, 2}, T = {tX, p, y }, and let /: S-+ T be
given by /( 1) = tX, /(2) = p. Since f is one-to-one, 1- 1 exists, and
1- 1: {tX,P}-+Sis given by/- 1(tX) = l,J- 1(p) =2.
2. Letf: R -+R be given by f(x) = x + 1 for ali xeR. Sincefis oneto-one, 1- 1 exists. Furthermore, since f is onto (why?), f(R) = R and
hencef- 1: R -+R. Since (f of- 1)(x) = x for ali xeR, we have
for ali xeR. Thereforef- 1(x) =x-1 for ali xeR. Note also that
(f-I of)(x) = f- 1(/(x)) = f- 1(X + 1) =(X+ 1) - 1 =X
as required.
3. Let /:[O, +oo)-+[O, +oo) be given by f(x)=Jx for ali
xe[O, +oo). Sincefis one-to-one and onto,f- 1:[0, +oo)-+[O, +oo).
Thus for ali x e [O, + oo ),
X=
(f of- 1)(x) = f(f- 1(x)) = Jf- 1(x)
which implies thatf- 1(x) =x 2 for ali xe[O, +oo).

EXERCISES
1.
For each of the following, show that f either is or is not a function

from S to T. If it is a function and its domain and range are subsets
of R, draw its graph.
(a) S={l,2,3,4}, T={tX,p,y},
f = {(l, tX), (2, p), (3, p), (4, tX), (4, y)}.
(b) S={l,2,3,4}, T={tX,p,y},
f = {(1, tX), (2, tX), (3, p), (4, y)}.
(e) S = Q, T = Z,f = {(n/m, n) 1 neZ, meZ, m =F O}.
(d) S=Q, T=Z,f={(n/m,n+m) lneZ,meZ,m =FO}.
(e) S=T=R,f={(x,y) lxeR,yeR,x 2 +y 2 =1}.
30
Sets and Functions

(f) S = T = R,f = {(x, ax + b) 1xeR}. Here a and b are fixed real
numbers.
(g) S = T = R,f = {(x,
1x eR}.
(h) S = T = R,f given by f(x) = x 2 + x + 1 for ali xeR.
(i) S = T = R,f given by
Jx2+1)
f(x)
= {~
if X =/= 0
if X =0
U) S = T = R,f given by
if X ~0
if X< 0
f(x) = {: + 1
2.
3.
4.
For each of the functions of Exercise 1, state whether or not the

function is onto and whether or not it is one-to-one. Justify your
statements.
Let n eN, m eN, and let S be a set which has n distinct elements and
Ta set which has m distinct elements. Let/: S-+ T. Prove each ofthe
following:
(a) Iff is onto, then n ~ m.
(b) If f is one-to-one, then n ~ m.
Let S = {l, 2, 3}, T = {tX, p, y, }, and let/: S-+ T, g: S-+ Tbe given
by
/(1) = }'
/(2) =
/(3) =(X
and
g(l) = g(2) = }'
g(3) =
(a) State whether f and gare one-to-one and onto.

(b) If U = {l, 3} and V= {tX, y, }, find f(U), g(U),
1- 1(V),
g-i(V).
5.
Let/: R-+R, g: R-+R, and h: R-+R be given by

f(x) = 2x + 1 for ali xeR
g(x) = x 2 + 1 for ali xeR
and
2x
h(x) = { X -1
if X ~t
if X <t
and
31
Finite and Infinite Sets
(a) Findf([O, l]), g((O, l]), and h((O, 1)).

(b) Find/- 1([0, 1]), g- 1((0, 1]), and h - 1((0, 1)).
6. Prove ( l) of Proposition l.2.8.
7. Prove (2) of Proposition l.2.8.
8. Prove ( 4) of Proposition 1.2.8.
9. Prove Proposition l.2.10.
10. Let S = {l, 2, 3, 4}, T = {oc, {3, y }, and U = {u, v, w, z }. Letf: S-+ T
and g: T-+ U be given by
f(l) =
/3
/(2) = }'
/(3) =
(l(
/(4) =
/3
and
g(oc)
=V
g(/3) = z
g(y)
=u
Write out the defining equations for f g.

Let /: R -+R be given by f(x) = x 2 for all xeR. Let g: R-+ R be
given by g(x) = 2x + l for all xeR.
(a) Write out the defining equations for f g and g of Is
Jog=goj?
(b) Write out the defining equations for j2 = f of, g 2 = g g,
/ 3 =! o/ 2 , and g 3 =g og 2
12. Give an example of functions f: R -+ R and g: R -+ R such that f =F g
but/ og =g of
13. For each of the functions of Exercise l above, state whether or not
the inverse function/- 1 exists. lf/- 1 does exist, finda description or
defining equation for it. Check your answer by computing
(f of- 1)(y) and (/- 1 of)(x).
14. Let S, T, and V be sets, with V e T, and let f: S-+ T. Prove that
1- 1(V) = 0 if and only if V nf(S) = 0.
11.
1.3
FINITE AND INFINITE SETS
Suppose we have two sets and we wish to compare the number of elements
which they contain. One way to do this is to pair the elements of one set
with those ofthe other in a one-to-one manner: if every element ofthe first
set is paired with an element of the second set, and if each element of the
second set is paired with some element of the first one, then the sets must
have the sarne number of elements. We make this notion precise with the
following definition.
Sets and Functions
32
Definition 1.3.1 A set S is equivalent to a set T if there exists a function

f: S-+ T which is one-to-one and onto.
Examples 1.3.2 1. Any set S is equivalent to itself, since f: S-+ S defined

by f(x) = x for all xeS is a one-to-one onto function.
2. Let E= {2k' 1keN} = {2, 4, 6, ... }, and letf: N-+ E be given by
f(n) = 2n for all n eN. It is easy to check that f is one-to-one and onto.
Hence the set N of natural numbers is equivalent to the set E of even
natural numbers.
3. For ali zeZ, letf: Z-+N be given by
2z
f(z) = { -2z + 1
ifzeN
ifzeZ -N
Since f is a one-to-one, onto function ( check this), Z is equivalev,t to N.

4. Let [a, b] and [e, d] be closed intervals in R, with a < b ande< d.
Letf: [a, b]-+[c, d] be given by
f(x)
d-e
b _a (x - a)
+e
for all xe [a, b]. The functionfis one-to-one and onto (check this), and
thus any two closed intervals in R which contain more than one point are
equivalent. The sarne function may be used to show that any two open
intervals in R are equivalent.
Now we are ready to give a rigorous definition of finite and infinite
sets.
Definition 1.3.3 A set is infinite if it is equivalent to a proper subset of
itself. A set is finite if it is not infinite.
Examples 1.3.4 1. The set N of natural numbers is infinite because it is

equivalent to the set E of even natural numbers, and E is a proper subset
of N.
2. The empty set is finite, because if it were not, it would be equivalent to a proper subset of itself, and this is impossible since the empty set
has no proper subsets.
3. For any n eN, the set {l, ... , n} is finite. (See Exercise 5 at the end
of this section.)
33
4. Let [a, b] be a closed interval in R, with a < b. Choose real numbers e and d such that a < e < d < b. Since the interval [e, d] is a proper'
subset of [a, b] and since any two closed intervals having more than one
point are equivalent by Example 4 of Examples 1.3.2, it follows that [a, b]
is infinite. Thus any closed interval having more than one point is infinite.
Similar arguments show that open intervals, half-open intervals, and rays
in R are infinite.
Intuitively, we may think of equivalent sets as being of the sarne
"size." Thus if two sets are equivalent, then either both of them must be
finite or both must be infinite. Furthermore, if Se T and Tis finite, then
S must be at least as "small" as T, and hence S must itself be finite. On
the other hand, if S e T and S is infinite, then T must be at least as
"large" as S and hence must be infinite. The following proposition and its
corollary prove these facts.
Proposition 1.3.5
1.
2.
Any set equivalent to an infinite set is infinite.

Any set which contains an infinite set is infinite.
Proof: 1. If the set S is equivalent to the infinite set T, then there exists
a one-to-one function f from S onto T and also a one-to-one function g
from T onto some proper subset V of T. Consider
See Figure 1.20. The function (f- 1 g) f is one-to-one and onto ( why?),
r-1
Figure 1.20
(f- 1 g)
o/: S-+f- 1(V).
Sets and Functions
34
and since T - V # <b, we have
Thus (f- 1 g) f is a one-to-one function from S onto a proper subset of

S, so S is infinite.
2. Let T contain the infinite set S. Since S is infinite, there exists a
one-to-one function f from S onto some proper subset U of S. Define
g: T-+ U as follows:
g(x) = fx(x)
if XES
if xET-S
It is easy to check that g is a one-to-one function from Tonto its proper

subset(T- S) u U. Hence Tis infinite. D
Corollary 1.3.6
1.
2.
Any set equivalent to a finite set is finite.

Any set which is contained in a finite set is finite.
The proof of the corollary is left as an exercise.
Example 1.3.7 The set Z of integers, the set Q of rational numbers, and
the set R of real numbers are ali infinite sets because each of them contains
the infini te set N.
There is more than one kind of infinity; that is, infinite sets differ in
the type or "intensity" of their infiniteness. As we shall see, the sets N, Z,
and Q ali possess the sarne type of infinity, that which is the least "intense." The set R of real numbers, on the other hand, has a different and
more "intense" type ofinfiniteness. Our next result is the first step toward
a classification of the different types of infinity.
Proposition 1.3.8 If Sisa nonempty set, then either there is some n EN
such that S is equivalent to {l, ... , n}, or else S conta.ins a subset equivalent to N.
Proof: Suppose that S is not equivalent to {1, ... , n}, for any n EN.
Since S is nonempty, we may choose an element x 1 eS. If S - {x 1} = </J,
35
then S = {x 1}; but since the set {x 1} is obviously equivalent to {l}, this
cannot occur. Hence S - {x1} =F </J.
Now suppose n ;;?; l and assume that we have selected n distinct elements x., ... , Xn of S. If S - {x., ... , xn} = </J, then S would be equivalent to {l, ... , n }. Therefore S - {x 1, , xn} =F </J, and we may select an
element Xn + 1 e S such that Xn + 1 ~ {xi> ... , Xn}. Thus by induction we
produce a subset {x 1, , Xm xn + 1, } of distinct elements of S, and
this subset is clearly equivalent to N by means of the functionf defined by
f(xn) = n for all n eN. D
Corollary 1.3.9 Every finite set is equivalent to {l, ... , n}, for some
neN. Every infinite set contains a subset equivalent to N.
Definition 1.3.10 An infinite set is denumerable if it is equivalent to the

set N of natural numbers. A set is countable if it is finite or denumerable.
A set which is not countable is said to be uncountable.
Examples 1.3.11 l. The sets N ofnatural numbers and Z ofintegers are

denumerable, hence countable.
2. For any neN, the set {l, ... , n} is finite, hence countable.
3. By Corollary l.3.9, every infinite set contains a denumerable subset. Therefore the denumerable sets are the "smallest" infinite sets.
A countable set is just that: a set whose elements can be counted. For
if the elements of a set can be counted, then the counting process establishes a correspondence between the set and N or between the set and a
subset {1, ... , n} of N, and such a correspondence shows that the set is
either denumerable or finite, as the case may be. Of course, in order to
countth elements of a set, we must be able to list them in such a way that
each element appears somewhere in the list, and when we count them we
must make sure that every element gets counted. Summarizing these notions, we have the countability criterion:
A nonempty set is countable if and only if its elements may
be listed according to some well-defined scheme and then counted
using some well-defined procedure.
Let us use the countability criterion to show that the set Q of rational
numbers is countable. We list the rationals in order of increasing
36
Sets and Functions
denominators, as follows:
o
l' l'
-1
1 '
-1
2'
2'
-2
-3
l'
1 '
l'
2 '
2'
2
-1
3' 3'
3 '
3'
-1 2
o
4' 4' 4' 4'
-2 3
2 ' 2'
-2 3
3 ' 3'
-2 3
4 '
4'
' ...
-3
2
-3
3 ' ...
-3
4 ' ...
Clearly, every element of Q appears in this list. (Some elements appear

more than once, but in fact this does not matter: counting some elements
of the set more than once will not affect its countability. This is a consequence of the Schroeder-Bernstein Theorem, which is provcd in the appendix. See also Proposition A.3 of the appendix.)
Now we count the elements of Q by counting along diagonais as
indicated by the arrows:
;1
o
-2
1,
2'
2
3'
3
-1
4
-2 3
,
2'
2
-2 3
3
-2
4'
. 3'
.
4'
-3
1 , ...
-3
.....
2
-3
...
3
-3
....
Thus we have demonstrated that the set Q is countable.

So far we have not produced any examples of uncountable sets. It is
a major result about the set R of real numbers that Ris uncountable. We
intend to prove this result, but in order to doso we will need the following
proposition.
37

Proposition 1.3.12 Every subset of a countable set is countable.
A formal proofrof this proposition is given in the appendix. Here we

will outline an argument based on the countability criterion. Thus, suppose that S is a countable set, so that there exists a list of the elements of
S and a procedure whereby the elements of S can be counted. But if Tis
contained in S, the elements of T will appear in the list for S and will be
counted as S is counted. Therefore T must be countable.
Now we are ready to prove that the reais are uncountable. The proof
proceeds by showing that the closed unit interval [O, l] is uncountable,
and this is accomplished by associating with each xe[O, l] a unique decimal expansion and then showing that no matter how we attempt to list ali
such expansions, we can always find an expansion which is not in the list.
The argument is called "Cantor's Diagonalization Procedure."
Theorem 1.3.13 (Cantor)
The set R of real numbers is uncountable.
Proof: If Ris countable, then sois [O, l], by Proposition 1.3.12. Therefore it suffices to show that [O, I] is uncountable.
Each xe [O, l] has a decimal expansion of the form
where a;e{O, 1, ... , 9} for each i. Some numbers have two such expansions: for instance
1
2=
0.5000 ... = 0.4999 ...
However, if a number does have two such expansions, ali the digits from
some p-J'lt on in one of them will be zeros, while ali the digits from some
point on in the other will be nines. Let us agree that if xe [O, l] has two
decimal expansions, we will always use the one which has a tail of nines.
With this agreement in force, we may consider that every xe [O, l] has
associated with it a unique decimal expansion.
Now assume that [O, l] is countable. Since [O, l] is infinite, it must be
denumerable, and hence we must be able to index the elements of [O, l] by
the natural numbers, thus:
38
Sets and Functions
Suppose that for each neNwe let O.an 1an2an 3 be the unique decimal
expansion of Xm so that
Xi
= O.a11a12a13
X2
= O.a21022023
Define ye [O, l] as follows:
where for each n eN,

if ann # 1
if ann = l
Obviously ye [O, l] and hence y = Xm for some neN, and therefore the
unique decimal expansions for y and xn must be the sarne. But this is
impossible, beca use for every n e N, the expansion for y differs in its n th
digit from the expansion for xn. Thus we have arrived at a contradiction,
and therefore [O, l] must be uncountable. D
Corollary 1.3.14
Every subset of R which contains an open interval is
uncountable.
The proof of the corollary will be left as an exercise.
There is much more to be said about set equivalence, finite and infinite sets, and the various types of infinity than we have been able to
present here. The reader interested in pursuing these topics may begin by
studying the appendix.
EXERCISES
1.
Prove:
(a) If set S is equivalent to set T, then set Tis equivalent to set S.
39
(b) If set S is equivalent to set T and set Tis equivalent to set U,

then set S is equivalent to set U.
2. (a) Prove that lV is equivalent to {1, 3, 5, ... } ( the set of odd natural numbers).
(b) Let k eZ. Prove that N is equivalent to {nk 1n eN}.
3. Prove that the function of Example 4 of Examples 1.3.2 is one-toone and onto, and then show that
(a) Any two open intervals (a, b) and (e, d) are equivalent.
(b) Any two half-open intervals [a, b) and [e, d) are equivalent.
(e) Any two closed rays [a, + oo) and [b, + oo) are equivalent.
4. Prove that Ris equivalent to the open interval ( -1, 1) by showing
that the function f defined by
X
f(x) = 1 +
5.
6.
7.
8.
9.
10.
lxl
for ali xeR
is one-to-one and onto.

Let n e N. Show that {1, ... , n} is finite.
Prove Corollary 1.3.6.
(a) Is the intersection of finite sets always a finite set? Justify your
answer.
(b) Is the union offinite sets always a finite set? Justify your answer.
Let S and T be countable sets. Prove that
(a) S n Tis countable.
(b) S u Tis countable.
(e) S x Tis countable.
Let I = {xeR 1 xQ}. The set /is the set of irrational numbers. Is I
countable or uncountable? Justify your answer.
2
The Real Numbers
The set R of real numbers will be of fundamental importance to us.

Throughout this text we will always apply our general results concerning
sets and functions to the set R and in tum use our knowledge of the
properties of R as a guide to further generalization. ln this chapter we
discuss some of the basic properties of the real number system and also
present a method for constructing the system of real numbers from the
system of rational numbers.
2.1
PROPERTIES OF REAL NUMBERS
As sta~ in Chapter 1, we assume that the reader is familiar with

the properties of the arithmetic operations (addition, subtraction, multiplication, division, and exponentiation) and the order relations
( <, ~, >, ;;?; ) on R. We shall not develop these here.
The first property of the real numbers which we wish to discuss is that
of distance between numbers. Between any two real numbers there is a
well-defined distance, the definition of which depends on the concept of
the absolute value of a real number.
Definition 2.1.1
Let
xe R.
The absolute value of x, denoted by
defined by
41
lxl, is
The Real Numbers
42
if X ~0
if X<
If x e R and y e R, we define the distance between x and y to be lx

Figure 2.1.
- y I See
Examples 2.1.2 We have
ll=O
141=4
l-41=4
l5fil=5J2
l4-Il=3
II-41=3
Our first proposition lists some of the properties of absolute value and
distance in R.
Proposition 2.1.3 If x e R and y e R, then

l.
2.
lxl ~O.
lxl =O if and only if x =O.
3. If a is a positive real number, then lxl ~a if and only if -a~ x ~a.

4. lxyl = lxllYI
5. (triangle inequality) lx + YI ~ lxl + IYI
Proof: We prove only (5), the triangle inequality, leaving the remainder
of the proof as an exercise.
To prove the triangle inequality, we proceed by cases.
Case 1: Both x and y are nonnegative: then x ~ O, y ~ O, and
X+ y
~0, SO
lx + YI = x + Y = lxl + IYI
lyl
IXI
~~
o
IX - yl
Figure 2.1 Absolute value and distance in R.
Properties of Real Numbers
43
Case 2: Both x and y are negative: then x < O, y < O, and x
+ y < O,
so
lx
+ YI = -(x + Y) = -x-y = lxl + IYI
Case 3: One of x or y negative and the other"is nonnegatiye: suppose

that y <O~x, so that either x + y ~O or x + y <O. If x + y ~O, then
lx + YI = x + y. But y <O implies that y < -y, so
lx + Y 1 = x
If x
+ y <O,
then lx
+ YI =
+Y < x -
-(x + y)
lx 1 + IY 1
-x-y. But x ~O implies that
-x~x,so
lx + YI
-x - Y ~ x - Y
lxl + IYI
This completes the proof of the triangle inequality. O
Corollary 2.1.4 If x, y, and z are real numbers, then

1. lx -yl ~O;
2. lx - y 1 = O if and only if x = y;
3. If a is a positive real number, then lx - YI ~a if and only if
y-a ~x~y +a;
4. lx - YI = IY - xi;
5. (triangle inequality) lx - YI ~ lx - zl + IY - zl.
The proof of the corollary is left as an exercise.
We now tum to another property of the real numher system,
the property known as boundedness. This is a property possessed by
certain subsets of R which depends on the order relationships among
real numbers.
Definion 2.1.5 Let S be a nonempty subset of R. If there exists a real

number tX such that x ~ tX for ali x e S, then S is bounded above by tX, and
tX is an upper bound for S. Similarly, if there exists a real number f3 such
that /3 ~ x for ali x e S, then S is bounded below by {3, and f3 is a lower
boundfor S. The set S is bounded if it is both bounded above and bounded
below; otherwise S is unbounded.
44
The Real Numbers
Examples 2.1.6
1. Let S be the open unit interval (O, 1) in R. Any real

number tX ~ 1 is an upper bound for (O, 1), and any real number f3 ~O is
a lowe'r bound for (O, 1). Thus (O, 1) is a bounded subset of R. Note that
the upper and lower bounds are not unique: clearly, if tX is an upper bound
for a set S, then sois any tX' > tX. Similarly, if f3 is a lower bound for S, then
so is /3' for any /3' < {3.
2. Let S be the open ray ( - oo, 1) in R. Any real number tX ~ 1 is an
upper bound for ( - oo, 1), but this ray has no lower bound. Thus
( -oo, l) is bounded above and unbom:1ded below, so is an unbounded
subset of R.
3. The set N of natural numbers is bounded below, but unbounded
above. The set Z of integers and the set Q of rational numbers are unbounded above and unbounded below.
Consider again the open unit interval (O, 1) in R. Although any real
number tX ~ 1 is an upper bound for (O, 1), it is clear that tX = 1 is the
smallest possible upper bound for the set. When we need to work with an
upper bound of a subset of R, it will be convenient to select the smallest
one (if there is a smallest one). Similar remarks apply to lower bounds.
This leads us to our next definition.
Definition 2.1.7 Let S be a nonempty subset of R. A real number

least upper bound, or supremum, for S if
l.
2.
is an upper bound for S and

For every tX' which is an upper bound for S,
is a
tX
Similarly, a real number

3.
4.
tX
tX ~ tX'.
f3 is a greatest lower bound,
or infimum, for S if
f3 is a
lower bound for S and

For every {3' which is a lower bound for S,
f3
{3'.
We shall prove immediately that if a subset of R has a least upper

bound, then that least upper bound must be unique, and similarly for
greatest lower bounds.
Proposition 2.1.8 Let S be a nonempty subset of R. If S has a least upper

bound, then that least upper bound is unique. If S has a greatest lower
bound, then that greatest lower bound is unique.
Proof: We prove the proposition only for the case of least upper
bounds.
45
Suppose ex 1 and ex 2 are both least upper bounds for the set S. Then by
(2) of Definition 2.1.7, ex 1 an upper bound and ex2 a least upper bound
implies that ex 2 ~ ex1> while ex2 an upper bound and ex 1 a least upper bound
implies that ex 1 ~ ex 2 Therefore ex 1 = ex2 D
Proposition 2.1.8 allows us to define the least upper bound and the
greatest lower bound of a set.
Definition 2.1.9 Let S be a nonempty subset of R. If they exist, the
unique least upper bound of S is denoted by sup S and the unique greatest
lower bound of S by inf S.
Examples 2.1.10 1. We claim that sup (O, 1) = 1 and inf(O, 1) =O.
Since we know that 1 is an upper bound for (O, 1), to show that 1 is the
least upper bound it suffices to show that if ex is any upper bound for (O, 1)
then 1 ~ ex. Suppose the contrary; that is, suppose that ex is an upper
bound for (O, 1) such that ex < 1. Since ex is an upper bound for (O, 1), we
must have O< ex. But then x =ex+ ( 1 - ex)/2 is in (O, 1) and ex < x, thus
contradicting the fact that ex is an upper bound for (O, 1). Therefore ex < 1
is impossible, so sup (O, 1) = 1. A similar argument shows that
inf (O, 1) =O. Note that in this example sup S and inf Sare not elements
of the set S.
2. The argument of the previous example shows that sup [O, l] = 1
and also that inf[O, l] = 1. Note that here both sup S and inf S are
elements of S.
3. Let S = {I/n 1 n e N}. We claim that sup S = 1: for 1 is certainly
an upper bound for S, and no real number ex < 1 can be an upper bound
for S because 1 e-S-:"Note that sup S e S.
4. If S is not bounded above, then of course sup S cannot exist; if S
is not bounded below, then inf S cannot exist. Hence sup N, sup Z, and
sup Q do not exist, nor do inf Z and inf Q. However, inf N = 1.
The preceding examples demonstrate that sup S, if it exists, may or

may not be an element of S, and similarly for inf S. ln nane of the
examples, however, was S bounded above but without a least upper
bound, or bounded below and without a greatest lower bound; this
is because such situations cannot occur in R. This is a very important
fact about the real numbers; it is known as the Completeness Property
for R.
46
The Real Numbers
Completeness Property for R Every nonempty subset of R which is

bounded above has a least upper bound. Every nonempty subset of R
which is bounded below has a greatest lower bound.
,
The Completeness Property for R is really a theorem about the real

number system. We will not prove it here, for it is not possible to prove
that R has the Completeness Property while working entirely within the
real number system: if we attempt to construct a proof of the Completeness Property using only facts about R, we will be forced to rely on some
other property of R which is equivalent to the Completeness Property,
and hence our proof will be circular. The proper way to show that R has
the Completeness Property is to construct the system R in such a way
that the Completeness Property becomes true in R by construction. We
will do this in Section 2.2 of this chapter, but for now let us accept the
Completeness Property as a fact and use it to obtain more properties of
the real number system. The first of these is the well-known
Archimedean Property for R.
(Archimedean Property for R) Given any two positive real numbers x and y, there exists meN such that mx > y.
Theorem 2.1.11
Proof: Let S = {nx 1neN}; then S is a nonempty subset of R. If

there is no me N such that mx > y, then y is an upper bound for S, and
hence by the Completeness Property, S has a least upper bound. Let
sup S = tX.
Since (n + l)x = nx + x is an element of S for ali n eN, we must
have nx +x ~tX for ali neN, and thus nx ~tX -x for ali neN. But this
implies that tX - x is an upper bound for S which is less than the least
upper bound tX, a contradiction. O
Theorem 2.1.12
Between any two distinct real numbers there is a
rational number.
Proof: We must show that if x and y are real numbers with x < y, then
there is some rational number r such that x < r < y. If x <O and y >O,
we may take r = O, so that it suffices to prove the theorem for the cases
O~ x < y and x < y ~O.
Assume first that O< x <J'. By the Archimedean Property for R,
there exist natural numbers m and n such that mx < 1 and n(y - x) > 1.
Let k be the larger of m and n; k is a natural number, and since kx > l
and k(y - x) > 1, we have I/k < x and l/k < (y - x). Again by the
47
Archimedean Property for R, there exists a natural number p such that

p( l/k) > x. Now consider the rational numbers
1 2
p
k' k" .. 'k
Since
1
p
-<x<k
k
there exists a rational number q / k, where 1 < q < p - l, such that

q
q+l
-<x<-k
k
But then
q+l q 1
1
x<--=-+-<x+-<x+y-x=y
Letting r = (q + 1)/k, we have shown the existence of a rational number

between x and y when O< x < y.
Now suppose O~ x < y. Since O~ x < (x + y)/2 < y, by what we
have just shown there is a rational number between (x + y)/2 and y, and
hence between x and y. Therefore the theorem is proved in the case that
O~x <y.
Finally, if x < y ~O, then O~ -y < -x, and by what we have just
proved'lhere is a rational number r such that -y < r < -x. But then
x < - r < y, and since r rational implies - r rational, we are done. D
Corollary 2.1.13 If x is any positive real number, then there is some n eN
such that O< 1/n < x.
Proof: There is some rational number m /n, where me N and n e N, such

that O< m/n < x. Hence O< 1/n < m/n < x. D
Example 2.1.14 1. Consider the set S = {l/n 1 neN}. Previously we
showed that sup S = 1. Now we claim that inf S =O: clearly O is a lower
bound for S, and if p > O, then by the corollary there is some element of
S between O and p, so p cannot be a lower bound for S.
2. Let S = {xeR 1 xeQ and x 2 < 2}. The set S is nonem~ and
bounded above, so sup S exists. Obviously, the real number ..j2 is an
then there is a
upper bound for S and sup S is positive. If sup S <
J2,
The Real Numbers
48
rational number r such that

O< sup S < r <
.J2
But then reQ and r 2 < 2, so reS, and thus sup S < r contradicts the fact
that sup S is an upper bound for S. Therefore sup S =
.J2.
The preceding example implies that the system Q of rational numbers

does not have the Completeness Property. For consider the subset S of Q
defined by
Then S is bounded above in Q (by the rational number 2, for instance),

but S has no least upper bound in Q; if it did, the least upper bound would
and -.;2 is nota rational number. (Exercise 12 at the end
have to be
of this section asks you to fill in the details of this argument.) Hence the
Completeness Property fails to hold in Q. This is the major defect of Q as
a number system.
We conclude this section by constructing a subset of the closed unit
interval [O, l] known as the Cantor ternary set. The Cantor ternary set
is more subtle in its form and properties than the intervals and rays
which we have studied so far, and it is a fruitful source of examples and
counterexamples in analysis.
.J2,
Example 2.1.15 (Cantor Ternary Set) We begin with a closed unit

interval [O, I]. We remove the open middle third (1/3, 2/3) from [O, l],
thus obtaining the set
Next we remove from C 1 its open middle thirds ( 1/9, 2/9) and (7/9, 8/9) to
obtain
-[O, !]9
C2 -
2] 1]
[~ !] [~3' 9 u [~9'
u 9' 3 u
Now we remove the open middle thirds from C2 to obtain
J[ J
J[ J
1 u 27'
2 9
l u [29' 27
7 u 27'
8 3
l u [23' 27
19] u [20
7]
C3 = [ O, 27
27' 9
u
[2627' l J
[9'825]
27
u
49
2
3
E J
_1_
9
E:3 B
E
_g_
9
_1_
2
3
BB
E
--9
B E3
o
C3
Figure 2.2
Construction of the Cantor set.
(See Figure 2.2.) Continuing in this manner, for each n eNwe define a set
Cn. The Cantor ternary set C is then given by
C=
Cn
neN
Thus the Cantor ternary set C consists of exactly those elements of [O, l]
which remain after ali open middle thirds have been removed. Another
way of writing the Cantor set is to let
C =[O, l] -
LJ
Sn
neN
where for each n e N,
(See Exercise 13 at the end of this section.)
, 'f}le Real Numbers
50
The Cantor set is nonempty, since Oe C and 1e . Jn fact, every real

number of the form l/3n, n eN is an element of C, ~o C is an infinite set.
Since we obtain C from [O, l] by removing most of the elements of [O, l],
we feel intuitively that C must be small compar~d to [O, l]. ln some ways
this isso: for instance, C does not contain any open interval. (See Exercise
16 at the end of this section.) Also, the length of C is zero, because for
each neN, C e Cn and Cn has length (2/3)n. Therefore in these senses C
is indeed a smaller set than [O, l]. On the other hand, C is uncountable
(see Exercise 15 at the end of this section), and hence in this sense it is just
as large as [O, l].
EXERCISES
1.
Find ali values of x which satisfy the given equality or inequality.
(a) lx - 21=3
(b) lx - 21<3
(e) l3x+5l~6
(d) 12-xl ~ 4
(e) IS+ 2xl ~ 6
(f) lx+ll<lx-31
2. (a) Prove 1, 2, 3, and 4 of Proposition 2.1.3.
(b) Prove Corollary 2.1.4.
3.
Let xeR, yeR. Prove:
(a) 1-xl = lxl

= lxl/IYI (y ::/=O)
(e) lx - Y 1~ lx 1+ IY 1
(d) llxl- IYll ~ lx - YI
(b) x/yl
4.
5.
6.
7.
8.
9.
Let S and Tbe bounded subsets of R. Prove that S u T and S n Tare

bounded.
Prove or give a counterexampli;: if X" is a bounded subset of R for
each ixeA, then UoceA X" is bounded.
Let [a, b] be a closed interval in li. Prove that inf[a, b] =a and
sup [a, b] = b.
Let I be a nonempty subset of R having the following property:
whenever xe/, yel, and x ~y, then [x,y] e/. Prov~ that I is an
interval or a ray.
Let xeR and let S = {qeQ 1 q < x}. Show that sup S = x.
Let S be a nonempty subset of R which is bounded above. Let
T = {xeR 1 x is an upper bound for S}.
Prove that inf T = sup S.
Construction of Real Numbers

10.
11.
12.
13.
14.
51
Let S be a nonempty subset of R. Prove that if x is an upper bound

for S such that xeS, then x = sup S.
Let S and T be nonempty subsets of R, with T bounded and S e T.
(a) Prove that sup S ~ sup T and that inf T ~ inf S.
(b) Give an example to show that even if Sisa proper subset of T,
it may be the case that sup S = sup T and inf S = inf T.
Fill in the details ofthe argument following Example 2.1.14 to show
that the system Q of rational numbers does not have the Completeness Property.
Show that the two expressions for the Cantor ternary set C given in
Example 2.1.15 are equivalent.
Just as every xe [O, l] has a decimal expansion, so does every
xe [O, l] have a ternary expansion
where\a;e{O, 1, 2}
(Such a ternary expansion for x may be obtained by writing x,=
atf3 + a 2 /3 2 + a 3 /3 3 + .) Prove that the Cantor set C consists of
exactly those elements of[O, l] whose ternary expansions have every

digit a;e{O, 2}.
15. If xe [O, l] has two ternary expansions, one ofthem ends in a tail of
zeros while the other ends in a tail of twos. Use this fact and the
result of Exercise 14 to prove that the Cantor set C is uncountable.
16. Prove that the Cantor set C does not contain an open interval.
2.2 CONSTRUCTION OF THE REAL NUMBERS*
ln this section we will construct the set R of real numbers from the set Qof
rational numbers by what is known as the Dedekind cut method. Once
this has been done, it will be a relatively simple matter to prove that R has
the Completeness Property.
We assume that the operations of arithmetic and the usual arder
relations have been defined on Q. As noted in the discussion following
Example 2.1.14 of this chapter, Q does not have the Completeness Property. The reason this is so is that, if we think of Q as consisting of the
rational points on a line, there are gaps between the g_oints which represent elements of Q. (For instance, the real number .J2 represents such a
gap. See Figure 2.3.) The Dedekind cut method constructs R from Q
*This section is optional.
52
The Real Numbers
(v12
Dots are rationals, gaps are irrationals
Figure 2.3
J2 as a gap in the rationals.
by filling in these gaps. We begin by defining the concept of a Dedekind

cut.
Definition 2.2.1
A nonempty proper subset S of Q is a Dedekind cut if
l. ses, qeQ, q < s imply that qeS and

2. seS implies that there is some teS such that s < t.
Note that a Dedekind cut is a subset of Q. Condition (1) of the

definition says that if a particular rational number s belongs to a
Dedekind cut, then so do ali rationals less than s; condition (2) says that
a Dedekind cut contains no 1argest element. Thus we see intuitively that
a Dedekind cut consists either of ali the rationals less than (but not equal
to) some specific rational, or of ali the rationals less than some particular
gap in Q (see Figure 2.4).
= {p e Q 1p < r}, then S, is a Dedekind
If r e Q and S,
Proposition 2.2.2
cut.
____
_______
,_ _,/
v
\...._
Rational cut: all rationals

less than a given rational

'-------.
V
..... ,..
lrrational cut: all rationals

less than a given irrational
Figure 2.4
Dedekind cuts.
Construction of the Real Numbers
53
Proof: The set S, is obviously a nonempty proper subset of Q, and if

s E S, and q E Q with q < s, then q < s < r, so q E S,. Furthermore, if s E S,
and we let t = (s + r) /2, then s < t < r and t E S,. Therefore ( 1) and ( 2) of
Definition 2.2.1 are satisfied, and hence S, is a Dedekind cut. D
Definition 2.2.3 A Dedekind cut which is of the form S, = {p E Q 1 p < r}
for some r E Q is called a rational cut. A Dedekind cut which is not a
rational cut is called an irrational cut.
Now we are ready to define the set of real numbers.

Definition 2.2.4 A real number is a Dedekind cut. The set R of real
numbers is the set of ali Dedekind cuts. A real number is rational if it is a
rational Dedekind cut; otherwise it is irrational.
Note that there is a one-to-one correspondence between rational

numbers and rational Dedekind cuts. (See Exercise 1 at the end of this
section.) Therefore ifwe identify each rational cut with the rational number which determines it, we may consider Q to be a subset of R.
We have now constructed R as the set of ali Dedekind cuts. Ali the
usual properties of R may be proved using Definitions 2.2.1 and 2.2.4 and
the known properties of Q. For instance, if x and y are real numbers
(Dedekind cuts), we may define their sum as follows:
x
+y
{p +q lpex and qey}
It is easy to check that x + y is a Dedekind cut, and hence a real number,

and thus addition is defined in R. Since our immediate objective here is to
show that R has the Completeness Property, we shall not pursue this
further. (But see the exercises at the end of this section).
Before we can prove that R has the Completeness Property, we must
define the order relation < on R.
Definition 2.2.5 Let x and y be real numbers (Dedekind cuts), and define
x < y if x is a proper subset of y. Define x ~ y if x < y or x = y.
Now we are ready to prove that R has the Completeness Property.

Actually, this is so by construction. The reason that Q does not have the
Completeness Property is that there are gaps between elements of Q; but
we have constructed R by filling in these gaps with Dedekind cuts. Hence
R must have the Completeness Property.
The Real Numbers
54
Theorem 2.2.6
The system R of real numbers has the Completeness
Property.
Proof: Let X be a nonempty subset of R which is bounded above. We
will show that X has a least upper bound.
Every real number xeX is a Dedekind cut and hence is a nonempty
proper subset of Q which satisfies properties ( 1) and (2) of Definition
2.2.1. Let
S = (qeQ 1 qex for some xeX}
Thus
S=
Ux
xeX
Since X is nonempty, S is a nonempty subset of Q. Furthermore,

since X is bounded above, there is some Dedekind cut which does not
belong to X, and thus there is some rational number which does not
belong to S. Therefore S is a proper subset of Q.
Let us show that S is a Dedekind cut. If ses, then sex for some
xeX. Because x is a Dedekind cut, we must have qex for ali rational q
such that q < s. But q ex implies that q e S. This shows that ( 1) of Definition 2.2.1 is satisfied. To prove that (2) is satisfied, we note that if s < S,
then, since x is a Dedekind cut, there is some t ex such that s < t. But
tex implies that teS. Therefore Sisa Dedekind cut, and hence a real
number.
If xeX then xeS and therefore by definition x ~ S. Thus S is an
upper bound for X. On the other hand, if T is any upper bound for X,
then for ali xeX we must have x ~ T, or x e T, so that
S=UxcT
xeX
and hence S ~ T. Thus S is a least upper bound for X, and we have

shown that every nonempty subset of R which is bounded above has a
least upper bound.
To complete the proof we must show that every nonempty subset Y
of R which is bounded below has a greatest lower bound. We outline an
argument which establishes this fact, leaving the details to the reader.
Constructioh df the Real Numbers
55
(See Exercise 7 at the end of this section.) Let

X = { x e R 1 x ~ y for ali y e Y}
then X is a nonempty subset of R which is bounded above, and hence by

what we have already proved, Xhas a least upper bound; but a least upper
bound for X is a greatest lower bound for Y. O
EXERCISES
1.
2.
Prove that two rational Dedekind cuts S, and Sq are equal if and only
if r = q.
If x and y are real numbers (Dedekind cuts), define
x
3.
4.
+ y = {p +q lpex and qey}
Prove that x + y is a Dedekind cut and hence a real number.

ln the notation of Proposition 2.2.2, S0 is the rational Dedekind cut
determined by the rational number O. Define the real number O by
setting O= S0 and then use the definition of addition given in Exercise
2 to show that for every real number x, x + O= x.
Let x, y be real numbers. If x is irl'ational, define
-x
{qeQ l -q~x}
If x is rational, say x = S,, where r e Q, then define
-x
5.
{qeQ 1 -q~x and -q =F r}
(a) Prove that -x is a real number (Dedekind cut).

(b) Show that x + y =O (see Exercise 3) if and only if y = -x. This
allows us to define subtraction in R. (How?)
(e) Prove that O< x if and only if -x <O.
Let x, y be real numbers. Define multiplication for nonnegative real
numbers as follows: if O~ x and O~ y, define
xy = Ou{qp 1 qe x, pe y, O~ q, O~p}
(a) Prove that xy is a Dedekind cut.
(b) Prove that xO =O.
56
The Real Numbers
(e) ln the notation of Proposition 2.2.2, S 1 is the rational Dedekind

cut determined by the rational number 1. Define the real number
1 by setting 1 = S 1 and the then use the definition of multiplication to show that x 1 = x.
6. If x is a real number, x > O, prove that there is a unique real number
y such that xy = 1. This allows us to define division in R. (How?)
7. Fill in the details for the argument outlined in the last paragraph of
the proof of Theorem 2.2.6.
3
Topology
Thus far we have been concerned mainly with combining sets by means of
set operations and comparing them using functions. ln so doing we have
been looking at sets primarily from an externai point of view, and have
not examined their internai structures in any detail. lt is important to
consider the internai structures ofsets, however. For instance, the set R of
real numbers is more than just an uncountable collection of elements:
certain subsets of R (such as the open intervals) have important properties
of their own, and we cannot understand the real numbers thoroughly
unless we examine these subsets and their properties. ln this chapter we
will study the internai structures of sets by considering certain collections
oftheir subsets known as topologies. We begin by introducing the concept
of a topological space and then explore such spaces with the aid of continuous functions. Ali our general results about topology can be applied to
the set R of real numbers, and beginning in this chapter and throughout
the rest of the text, the reader should always give thought to what our
general definitions and results say about R.
3.1
TOPOLOGICAL SPACES
We commence with the definition of a topological space.

57
58
Topology
Definition 3.1.1 A topological space is a pair (X, r), where X is a set and
r is a collection of subsets of X such that
l. Xer and 0er.
2. If A is an index set and Grxe 't" for each oieA, then UrxeA Grxe r.
3. If A is a finite index set and Grxe for each oieA, then lrxeA Grxe.
Such a collection r of subsets of X is called a topology on X, and X is then
referred to as the space, or underlying point set, of r. Also, the elements of
X are called the points of the topological space. A subset G of X is an open
set in the topology r if G e r; a subset F of X is a closed set in the topology
r if its complement X - Fe r.
Note that a collection r of subsets of a set X is a topology on X if and

only if both the empty set and X itself belong to r; any union of sets
belonging to ris a set belonging to r; and any intersection offinitely many
sets belonging to r is a set belonging to r.
Examples 3.1.2 l. Let X= {x, y, z} and let r = {0, {x }, {y }, {x, y }, X}.
The collection r satisfies the conditions of Definition 3.1, so (X, r) is a
topological space, that is, r is a topology on X. The open sets in r are the
sets
0, {x}, {y}, {x,y}, and X
The closed sets in r are the complements of the open sets, namely
X, {y, z}, {x, z}, {z}, and 0
Notice that 0 and X are subsets of X which are both open and closed in
the topology. Also, in this example every subset of the topological space
is either open, closed, or both open and closed.
2.Let X= {x,y,z} and let
1
{0, {x}, X}
2 = {0, {x}, {x,y}, {x, z}, X}

T3
{0, {x}, {y}, X}
't"4
{0, {x,y}, {x, z}, X}
The collections r 1 and r 2 are topologies on X(check this), but r 3 and r 4 are
Topological Spaces
59
not: -r 3 is not a topology on X because
{x}u{y} = {x,yH-r3
while -r4 is not a topology on X because
{x, y} n {x, z}
{x} ~'r4
Notice that there are subsets of Xwhich are neither open nor closed in the
topology -r 1
3. Let X be any set and let -r = {S 1 S is a subset of X}; then -ris a
topology on X, called the discrete topo/ogy on X. Every subset ofthe space
is both open and closed in the discrete topology.
4. Let X be any set and let -r = {0, X}; then -r is a topology on X,
called the indiscrete topology on X. ln the indiscrete topology, the empty
set and the entire space are the only open sets and also the only closed
sets.
Assigning a topology to a set is called topo/ogizing the set. There are
always at least two ways to topologize a set, for we may always assign it
either the discrete topology or the indiscrete topology.
Now consider the set R of real numbers. We can of course topologize
R by assigning it either the discrete or the indiscrete topology, but these
are not very interesting. Instead, we will use the open intervals as the basis
for a topology on R. This topology, called the natural topology on R,
will be of fundamental importance to us throughout the remainder of this
text.
Proposition 3.1.3 Let -r consist of ali subsets S of R which have the

property that if xeS then there is some open interval lx e S such that
xelx. The collection -r is a topology on R, called the natural topo/ogy
on R.
Proof: We must verify that the collection -r satisfies the conditions of
Definition 3.1.1. It is certainly the case that if xeR, then there is some
open interval lx e R such that xe lx. Therefore Re -r.
To show that the empty set belongs to -r, we must demonstrate that if
x e 0 then there is some open interval lx e 0 such that x e lx. But there is
no x which is an element of 0, so there is nothing to prove. (When a
situation like this occurs, we say that the condition is vacuously satisfied.)
Hence 0e -r.
Topology
60
Now let A be an index set and suppose that G~ E r for all r:t. e A. If
= </), then U~EA G~ E r by what we have just proved, so let us
assume that U~EA G~ # </). If x E U~EA G~, then there is some /3 e A such
that x E Gp. But since Gp E r, there is some open interval
lx e Gp e U~EA G~ such that x E lx, and therefore the set U~EA G~ E r.
Finally, let A be a finite index set and suppose that G~ E r for all rx e A.
Again, if n~EA ~ = </), we are done, so assume that n~EA ~ # </) and let
xen~EA~ Then XEG~ for each r:t.E, and since G~Er, for each r:t.Er
there is some open interval (a~, b~) e G~ such that x E (a~, b~). Because A
is a finite set, the set {~ 1 rx E A} is finite and therefore has a largest
element: let a = max {~ 1 rx E A} be this largest element. Similarly, let
b = min {b~ 1 rx e A}. Then xe (a, b) e (a~, b~) for all r:t. e A, and hence
(a, b) is an open interval contained in n~EA ~ such that XE (a, b). Therefore n~EA ~E r, and we are done. D
U~EA G~
Note that a set is open in the natural topology on R if and <?nly if we

can surround each of its points with an open interval which is itself
contained in the set. Thus open intervals are open sets in the natural
topology, as are open rays. On the other hand, closed intervals and closed
rays are not open sets: for example, if [a, b] is a closed interval, it is not
possible to surround the endpoint a with an open interval which is contained within [a, b]. Closed intervals and closed rays are closed sets in the
natural topology, however, because they are complements of open sets.
We collect these anda few other facts in a proposition, the proof ofwhich
is left to the reader.
Proposition 3.1.4
ln the natural topology on R,
1. Every open interval is an open set.

2. Every open ray is an open set.
3. Every closed ray is a closed set.
4. Every closed interval is a closed set.
5. Every one-element subset of R is a closed set.
6. Every finite subset of R is a closed set.
We have seen that a given set may be topologized in several different
ways. But since a topology is itself a set, we can compare topologies which
have the sarne underlying point set by means of set containment.
Definition 3.1.5 Let r 1 and r 2 be topologies on the set X. If r 1 is a proper
subset of r 2, we say that r 1 is smal/er than r 2, or that r 2 is /arger than r 1,
Topological Spaces
61
and write 1 e 2 If 1 e 2 and 2 e 1 then 1 = 2 If 1 <t:- 2 and 2 <t:- 1,

we say that 1 and 2 are not comparable.
Note that if and 2 are topologies on X then 1 e 2 if and only if

every set which is open in 1 is also open in 2 Thus if 1 is smaller than
2 , then 1 has fewer open sets than 2 , and if 1 = 2 , then 1 and 2 have
exactly the sarne open sets.
Examples 3.1.6
1. Let X= {x, y, z }, 1 = {0, {x }, X}, and 2 = {0,

't" 1 and ,;-are topologies on X, and 1 is a proper
{x}, {x, y}, X}. Both
subset of 2 Hence 1 is smaller than 't" 2

2. Let X= {x, y, z }, 1 = {0, {x }, X}, and 2 = {0, {y }, X}. Then 1
and 2 are not comparable.
3. Let 1 be the natural topology on R and let 2 be defined as follows:
2
= {G e
1G=0 or G is a union of open intervals}
We claim that = t"2 Thus 2 is merely another way of describing the

natural topology on R. Note that it is not necessary to prove directly that
2 is a topology on R: if we can show that 1 and 2 have the sarne
elements, then 2 must satisfy the conditions of Definition 3.1. l because
we have already shown that does. Therefore in order to prove the claim,
it will be sufficient to show 1 e 2 and 2 e 1
Let Ge t" 2 If G = 0, then Ge 1 If G =F 0, then G is a union of open
intervals, which are open sets in the natural topology, and therefore
Ge 1
Now let Ge 1 If G = 0, then Ge 2 If G =I= 0, then for each xeG
there is some open interval lx such that xelx e G. But this implies that
G = UxeG lx. Therefore G e 2 , and the claim is proved.
By virtue of the previous example, we can think of natural topology

on R as being generated by the open intervals, in the sense that every
nonempty open set in the natural topology is a union of opeti. intervals
and conversely every su.ch union is an open set. The following useful result
further characterizes these open sets as countable unions of mutually
disjoint subsets of R.
Proposition 3.1.7 If G is a nonempty open set in the natural topology on
R, then G is a union of countably many mutually disjoint subsets of R,
each of which is either an open interval or an open ray.
62
Topology
Proof: Let xeG. Since G is open in the natural topology, there is some
open interval lx such that x elx e G. Let Jx denote the union of all subsets
of G which contain x and are either open intervals or open rays. lt is easy
to check that Jx must itself be either an open interval or an open ray. (See
Exercise 18 of Section 1.1, Chapter 1.)
We claim thatif xeG andyeG, then either Jx nJy = </J or Jx = JY. For
if Jx nJy # </J then Jx uJY must be either an open interval or an open ray,
and since x eJx u JY and (Jx u Jy) e G, the definition of Jx tells us that
(Jx uJy) e Jx. Similarly, (Jx uJy) e JY. Therefore Jx = JY whenever
Jx n JY # </J. Clearly G = UxeG Jx, and thus we have established that G is
a union ofs'ets each ofwhich is either an open interval or an open ray, and
that the distinct sets in the union are mutually disjoint. But any union of
mutually disjoint subsets of R is countable, because each set in the union
must contain a rational number which does not belong to any of the other
sets in the union, and hence there is a one-to-one correspondence between
the sets in the union and a subset of the rational numbers. D
We have been examining the natural topology on R in some detail

and shall continue to do so throughout this text, for the natural topology
is, as its name implies, the topology on R that is most useful to us. lt is not
the only nontrivial topo!Ogy on R, however. For example, suppose we let
r={</J}u{R}u{(a, +oo) laeR}
Then r is a topology on R ca!led the open ray topology. It is neither
the discrete nor the indiscrete topology, and it is not the natural topology because it is smaller than the natural topology. We leave the
proof of these facts to the reader. (See Exercise 10 at the end of this
section.)
We conclude this section with a method for constructing new topological spaces from old ones. The next proposition says that, given a
topological space (X, r) we can topologize any subset Y of X by intersecting Y with the open sets of r.
Proposition 3.1.8 Let (X, r) be a topological space and let Y be a subset

of X. If
r' = {G' e Y 1 G' = Y nG for some Ge r}
then r' is a topology on Y called the subspace topology induced on Y by r.
Topological Spaces
Proof:
63
Since Xe and 0e <, it follows that Y = Y nXe ' and
0=Yn0e'.
Let A be an index set and suppose that G ~e ' for ali oc e A; then for
each oc e A we have G~ = Y n G,,, for some G,,, e Thus
u G~ u
=
a.e A
(YnG,,,) = Yn
a.e A
(u G,,,)
a.e A
ds
where G,,, e "t" for each oc e A. Since

a topology, Ua.eA G,,, e<, and therefore Ua.eA G~ = Y n ( Ua.eA G,,,) E'
Now let A be a finite index set and suppose that G~ e' for ali oc e A;
then for each oceA we have G~ = Y nG,,, for some G,,,e <, and hence
where G,,, e for each oc e A. Since is a topology,

fore la.eAG~e"t"'. D
la.eA
G,,, e<, and there-
Examples 3.1.9 1. Let X= {x, y, z} and "t" = {0, {x, y }, X}. Let
Y = {y, z }. The subspace topology induced on Y by is the topology
' =
{0, {y }, Y}.
2. Consider the closed unit interval [O, l] in R. Let us describe the

subspace topology induced on [O, l] by the natural topology on R. A
subset G of [O, l] is open in the subspace topology if and only if it is the
intersection of [O, l] with a set which is open in the natural topology on
R; that is, if and only if it is the intersection of [O, l] with a union of open
intervals. It is easy to check that the intersection [O, l] with any open
interval is either 0, [O, l], an open interval (a, b), ora half-open interval of
the form (a, l] or [O, b). Therefore a nonempty proper subset of [O, l] is
open in the subspace topology if and only if it is a union of intervals of the
form (a, b), (a, l], or [O, b), where O~ a < b ~ 1.
EXERCISES
1.
Let X= {x, y }. Find ali topologies on X. For each topology, find ali
the open sets, ali the closed sets, ali the sets which are both open and
closed, and ali the subsets of X which are neither open nor closed in
the topology.
64
Topology
2.
Repeat Exercise 1 for the set X= {x, y, z }. (There are 14 topologies

on this set.)
Let r ={Se N 1S = N or Sisa finite subset of N}. Is r a topology
on N? Justify your answer.
Let r ={Se N 1S=0 or N - S is a finite subset of N}. Is r a
topology on N? Justify your answer.
Prove Proposition 3.1.4.
Prove that in the natural topology on R half-open intervals are
neither open nor closed.
For each topology r of Exerci se 1 above, list ali the topologies which
are smaller than r, ai! those which are larger than r, and ali those
which are not comparable to r.
For each topology r of Exercise 2 above, list ali the other topologies
of Exercise 2 which are smaller than r, those which are larger than
r, and those which are not comparable to r.
Prove that the indiscrete topology is the smallest topology on any set
and that the discrete topology is the largest topology on any set.
Let r be the open ray topology on R defined by
3.
4.
5.
6.
7.
8.
9.
10.
{0}u{R}u{(a, +oo) 1 aeR}
Prove that ris a topology on R,. that it is neither the discrete nor the
indiscrete topology, and that it is smaller than the natural topology.
11. Considering the natural numbers N and the rational numbers Q as
subsets of R,
(a) Show that the natural topology on R induces the discrete topology on N;
(b) Describe the open sets in the subspace topology induced on Q
by the natural topology on R.
12. Let R have the open ray topology ofExercise 10 above. Describe the
open sets in the subspace topologies induced on N and on Q by the
open ray topology.
13. Let X be a set, Y a nonempty subset of X, and r 1 and r 2 topologies
on X with r 1 # r 2 Is it possible for r 1 and r 2 to induce the sarne
subspace topology on Y? Justify your answer.
14. Let r 1 and r 2 be topologies on a set X and define
r 1 rH 2 = {G e X 1G e r 1 and G e r 2 }
and
Topological Spaces
15.
(a) Prove that r 1 n r 2 is a topology on X which is smaller than or

equal to r 1 and also smaller than or equal to r 2
(b) Give an example to show that r 1 ur2 need not be a topology.
Let X be a set and ffe be a collection of subsets of X such that
(a) f/Jeffe and Xeffe;
(b) lf A is an index set and Fa.eff for ali oieA, then la.eA Fa.effe;
(e) lf A is a finite index set and Fa.eff for ali exeA, then
Ua.eA
16.
65
Fa. effe.
Let r = { G e X 1X - G effe}. Prove that r is a topology on X and

that ffe is the collection of closed sets for the topology r.
Consider the Euclidean plane R 2 = R x R. A subset S of R 2 is an
open rectangle if it is of the form
S = { (x, y) 1 a < x < b, e < y < d}

for some real numbers a, b, e, and d. A subset D of R 2 is an open disc
with center (h, k) and radius r if it is of the form
for some real numbers h and k and some positive real number r.
(a) Let r 1 = {G e R 2 G = f/J or G is a union of open rectangles}.
Prove that r 1 is a topology on R 2 . This topology is called the
product topology on R 2 .
(b) Let r 2 = {G e R 2 G = f/J or G is a union of open discs}. Prove
that r 2 is the product topology of part (a).
Let X be a set, r be a topology on X, and x e X. A subset V of X is
a neighborhood of x if there is some open set G e r such that G e V
and xeG.
(a) Let X= {x, y, z} and r = {f/J, {x }, {x, y }, X}. Find ali neighborhoods of each element of X.
(b) Suppose X= R and ris the natural topology. Show that Vis a
neighborhood of x if and only if there is some open interval I
such that xe/ and I e V.
(e) Suppose r is the discrete topology. For any xeX, describe ali
neighborhoods of x. Do the sarne if r is the indiscrete topology.
(d) Prove that a subset G of X is open in r if and only if G is a
neighborhood of each of its points.
17.
66
Topology
3.2 OPEN SETS ANO CLOSED SETS

An understanding of open sets and closed sets is fundamental to an
understanding of topological spaces. ln this section we characterize open
and closed sets in a general topological space. We remark here that if
(X, t) is a topological space and Tis the only topologyon Xwhich is under
consideration, we will speak of "the topological space X" rather than "the
topological space (X, -r)." We will also refer to "open sets in X" rather
than use the more precise language "sets open in the topology on X," and
similarly for closed sets. This should cause no confusion.
Our first proposition collects some elementary properties of open sets
and closed sets. lts proof follows direct\y from the definitions of these
concepts and will be left to the reader.
Proposition 3.2.1 Let X be a topological space and A be an index set. If

for ali a e A, Ga. is an open set in X and Fa. a closed set in X, then
Ga. is an open set in X and la.eA Fa. is a closed set in X.
la.eA Ga. is an open set in X and Ua.eA Fa. a closed set in
1.
Ua.eA
2.
If A is finite,
X.
Suppose that (X, -r) is a topological space. According to the definition

of open set, a subset G of X is open in X if and only if G E -r. This is what
is known as a global characterization of open sets: if we wish to use this
definition to test whether or nota given set is open, we mustcheck to see
whether or not the set as a whole satisfies the required condition. lt would
be convenient for us to have a local characterization of open sets, that
is, one which would enable us to test whether or not a given set is open
by checking to see if an arbitrary element of the set, rather than the set
as a whole, satisfies some condition. The next proposition affords us
just such a local characterization of open sets. Note that the characterization is similar to the property we used to define the natural topology
onR.
Proposition 3.2.2 Let X be a topological space. A subset G of X is open

in Xifand only iffor each xeG there is some open set Hx such that xeHx
and Hx e G.
Proof: Let G be open in X. If G = 0 then it is vacuously true that for
each x E G there is some open set H x such that x EHx e G. (Why?) If G # 0
and xeG, choose Hx = G. Thus, in either case, if G is open there is an
open set Hx such that xeHx e G.
Open Sets and Closed Sets
67
To prove the converse, let G be a subset of X such that for each xeG
there is some open set Hx such that xeHx e G. But then G = UxeG Hx is
a union of open sets and hence is itself open. D
We now have two characterizations for open sets: a global one which
says that a set G is open if and only if it belongs to the topology, and a
local one which says that a set G is open if and only if we can surround
each of its points with an open set which is itself contained in G.
Now we tum our attention to closed sets. We have a global characterization of closed sets: a subset F of the topological space X is closed in X
ifand only ifits complement X - Fis open in X. We would like to obtain
a local characterization of closed sets. ln order to do so, we must introduce the concept of a limit point of a set. This concept is important not
only for closed sets, but for ali sets in a topological space.
Definition 3.2.3 Let X be a topological space and let S be a subset of X.
A pointxeXis a limit point of Sifwhenever G is open in Xand xeG, then
(GnS)-{x}#0
The set of ali limit points of S, called the derived set of S, is denoted
by
s.
Note that x is a limit point of S if every open set containing x also

contains at least one point of S different from x. Note also that a limit
point of S need not be an element of S.
Examples 3.2.4 1. Let X= {x, y, z} and r = {0, {x }, {x, y }, X}, and
consider the set S = { x}. The element x is not a limit point of S beca use
{x} is an open set containing x which does not contain any point of S
different from x. However, y is a limit point of S because the only open
sets which contain y are {x, y} and X, and each of these contains a point
of S different from y. Similarly, z is a limit point of S. Therefore
S = {y, z}
Note that S is not a closed set and that S <t;. S.
Now consider the subset T = {y, z} of X. The element x is nota limit
point of T beca use {x} is an open set containing x which does not contain
a point of T different from x. Furthermore, y is not a limit point of T
because {x, y} is an open set containing y which does not contain a point
68
Topology
of T different from y. However, z is a limit point of T stnce the only open

set containing z is X, and X contains a point of T different from z.
Therefore
t= {z}
Note that Tis closed and that te T.
2. Let R have the natural topology and consider the set Z of integers
as a subset of R. For any xeR we can always find an open interval which
contains x and does not contain any element of Z, except possibly x if
xeZ. Therefore no element of R can be a limit point of Z, and thus
Z=</J
Now consider the set Q of rational numbers as a subset of R. If xeR
and G is any open set containing x, then there is some open interval (a, b)
such that xe(a, b) e G. But then there is a rational number q such that
a< q < x, and hence qe(Q nG) - {x }. Therefore every xeR is a limit
point of Q, and thus
Q=R
3. Let R have the natural topology and let S be a nonempty subset of
R which is bounded above. If sup S rtS and (a, b) is an open interval which
contains sup S, then (a, b) must contain a point of S, for otherwise the
endpoint b would be a least upper bound for S which is less than sup S.
Hence if sup S S then sup S is a limit point of S. A similar result holds
for greatest lower bounds.
4. If R has the natural topology and S =(a, b) is an open interval and
T = [a, b] a closed interval in R, then
S =[a, b]
and
=[a, b]
We leave the proof to the reader. (See Exercise 7 at the end of this
section.) Note that S is not closed and does not contain ali its limit points,
while Tis closed and does contain ali its limit points.
Now we use the concept of limit point to obtain a local characterization of closed sets. The characterization says that a set is closed if and only
if it contains ali its limit points; it will appear as a corollary to the next
proposition.
69
Proposition 3.2.5 If X is a topological space and S is a subset of X, then

S u S is closed in X.
Proof: We prove that S u S is closed by showing that its complement is

open. Since this is certainly the case if the complement is empty, we
suppose that the complement of S u S is nonempty.
Let xeX - (SuS). Since xrtS, x is nota limit point of S and hence
there must exist an open set G such that x e G and '(G n S) - {x} = </J. But
since also x S, it follows that in fact G n S = </J. Furthermore, G n S = </J,
for if this were not so, G would contain a limit point of S, and therefore
a point of S different frolll the limit point, thus contradicting the fact that
G nS = </J. We have now shown that G nS = </J and that G nS = </J, and
thus that G is contained in X - (S u S). Therefore every element x in the
complement of S u S is contained in an open set G, which is itself contained in the complement. Hence the complement is open, and we are
done. D
Corollary 3.2.6 Let X be a topological space. A subset F of X is .closed
in X if and only if F contains all its limit points.
Proof: If F contains ali its limit points, then F =Fui, which is closed
by the proposition. Conversely, suppose F is closed but does not contain
all its limit points. ln this case there is some x E X such that x F and x
is a limit point of F. But then xeX - F, which is an open set containing no point of F, and this contradicts the fact that x is a limit point
of F. D
Now that we have considered open sets and closed sets, what about
sets which are neither open nor closed? For example, consider the halfopen interval (a, b] in R. This set, which is neither open nor closed in the
natural topology, contains a "largest open set," namely (a, b), and is
contained in a "smallest closed set," namely [a, b]. Thus if we wished we
could study (a, b] by analyzing the open set (a, b), the closed set [a, b], and
the "boundary set" {a, b }. This sort of dissection of a set can be done in
any topological space, provided we define "largest open set," "smallest
closed set," and "boundary set" properly.
Definition 3.2.7
1.
Let X be a topological space and let S be a subset of X.
An open set G in X is the largest open set contained in S if G is

contained in S and whenever H is an open set contained in S then
HcG.
70
Topology
2. A closed set F in X is the smallest closed set containing S if F contains

S and whenever E is a closed set containing S then F e E.
Note that if a largest open set contained in S exists, it must be unique:
for if G and G' are both largest open sets contained in S then necessarily
G e G' and G' e G, so that G = G'. A similar remark applies to smallest
closed sets containing S. Therefore use of the phrases "the largest open
set" and "the smallest closed set" in the definition is justified. Furthermore, the largest open set G contained in S always exists: let G = UH,
where H ranges over ali open sets contained in S. Of course, if the only
open set contained in S is the empty set, then G = f/J. Similarly, the
smallest closed set F containing S always exists: let F = lE, where E
ranges over ali closed sets containing S. If the only closed set containing
S is the space X, then F = X.
Definition 3.2.8
The interior of S, denoted by S, is the largest open set contained

in S.
2. The closure of S, denoted by S, is the smallest closed set containing S.
3. The boundary of S, denoted by bd S, is defined by bd S = S - S.
1.
The remarks preceding the definition show that for any subset S of a
topological space X, the interior, closure, and boundary of S exist and are
unique.
Examples 3.2.9
1. Let X= {x, y, z} and r
= {0, {x }, {x, y }, X}. Let
T= {y}
and
V= {x, z}
S=S
S=X
and
bd S= {y, z}
T=f/J
f'= {y, z}
and
bd T= {y, z}
V= {x}
V=X
and
bd V= {y, z}
S= {x}
We have
2. Let R have the natural topology and let S be any of the intervals
(a, b), [a, b], (a, b], or [a, b); then
S =(a, b)
S= [a, b]
and
bd S= {a, b}
71

Also
Z=0
Z=Z
and
bdZ=Z
Q=0
Q=R
and
bdQ=R
and
The interior, closure, and boundary can be used to characterize open

and closed sets.
Prop~sition
1.
2.
3.
3.2.10 If X is a topological space and Sisa subset of X, then
ScS cS.
S is open in X if and only if S = S.
S is closed in X if and only if S = S.
Proof: Part ( l) is true by definition. To prove (2), note that if S is open,

then it is the largest open set contained in S; conversely, the interior S of
S is open by definition. The proof of ( 3) is similar and is left to the reader.
(See Exercise 11 at the end of this section.) O
Corollary 3.2.11
l.
2.
3.
If X is a topological space and S is a subset of X, then
S is open in X if and only if S n bd S = 0.

S is closed in X if and only if bd S e S.
S is both open and closed in X if and only if bd S = 0.
Proof: We prove ( l), leaving the restas an exercise. (See Exercise 17 at

the end of this section.)
If S is open, then S = S, and this implies that bd S = S - S. But then
Snbd S =Sn(S-S) =0
Conversely, if S n bd S = 0, then S n (S - S) = 0, and since S - S is
contained in S - S, it follows that S - S = 0. But this implies that
S = S and thus that S is open. O
Definition 3.2.12 Let X be a topological space and let S be a subset of X.
If S = X, we say that S is dense in X.
Examples 3.2.13 1. Let X= {x, y, z} and r = {0, {x }, {x, y }, X}. Then
{x} is dense in X but {y} is not dense in X.
Topology
72
2. If [O, l] has the subspace topology induced by the natural topology

on R, then (O, 1) is dense in [O, l].
3. Let R have the natural topology. The set Q of rational numbers is
dense in R.
We conclude this section with another characterization ofthe interior,

closure, and boundary of a set.
Proposition 3.2.14 If X is a topological space and Sisa subset of X, then
~ = {xe.s 1 xeG for some open set G e S}.

S=SuS.
bdS={xeXlif xeG, G open, then GnS=F</J and Gn(X-S)
=F <b}.
1.
2.
3.
Proof: We prove (1) and (2), leaving (3) as an exercise. (See Exercise 18
at the end of this section.)
Let us put T = {xeS 1xeG for some open set G e S}: then in arder
prove ( l) we must show that S = T. If .S = </J, then the only o pen set
contained in S is </J, and T must be empty also. On the other hand, if T = <b
then S cannot contain a nonempty open set and hence S = </J. Therefore
( 1) holds if either S or Tis empty, so we assume that these sets are not
empty.
If x e S then x e T, since S is an o pen set contained in S. Therefore
S e T. Conversely, if x e T, then there is some o pen set G such that x e G
and G e S. But then G e S, because S is the largest open set contained
in S. Therefore Te S, and we have proved (1).
Now we prove (2). Since Se S, every limit point of S is also a limit
point of S. (See Exercise 8 at the end of this section.) Since S is closed, it
contains ali its limit points. Therefore S contains ali the limit points of S,
and it follows that (S u S) e S. On the other hand, since S u S is a closed
set containing S, we must have Se (S u S), and we are done. O
Corollary 3.2.15 Let X be a topological space. A subset S of X is dense

in X if and only if every point of X is either a point of S ora limit point
of S.
EXERCISES
l.
2. Give anexample ofa topological space Xand a collection {G,,, ex e A}

of open sets in X such that
in X.
la.eA
G,,, is not an open set

3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
73
Give an example of a topological space X and a collection

{F,,_ 1tX E A} of closed sets in X such that U,,_eA F,,_ is not a closed set
in X.
Let [O, l] have the subspace topology induced by the natural topology on R. Prove that the Cantor ternary set is clo~ed but not open
in the subspace topology.
Let X be a discrete topological space and let S be any subset of X.
Show that S = </J.
Let X be an indiscrete topological space and let S be any nonempty
subset of X. Show that either S = X or S = X - S.
Let R have the natural topology and let S =(a, b) and T =[a, b].
Prove that S = t =[a, b].
Let X be a topological space, and let S and T be subsets of X such
that S e T. Prove that S e t.
Let R have the open ray topology, and let S =[O, l]. Find Q, Z, and
S in this topology.
Let [O, l] have the subspace topology induced by the natural topology on R. Let C denote the Cantor ternary set. Show that C = C.
Prove ( 3) of Proposition 3.2.1 O.
Let X= {x, y, z, w} and T = {<b, {x }, {x, y }, {x, y, z }, X}. Let
S = {x, y} and T = {x, z }. Find S, S, S, bd S, t, T, t, and bd T.
Let R have the open ray topology. Find Q'\ Q, bd Q, Z, Z, and
bd Z. If S =[O, l], find S, S and bd S.
Let the Euclidean plane R 2 have the product topology. (See Exercise
16 of Section 3.1.) Let D be an open disc in R 2 and La line in R 2 .
Find D, D, 15, bd D, l, L, L, and bd L. Is L open in R 2? closed in
R2?
Let X be a topological space and let S be a subset of X. Let S have
the subspace topology induced by the topology on X. Prove that S
is dense in S.
Give an example of topological space X which is
(a) Infinite and such that every nonempty subset of X is dense in X.
(b) Infinite and such that no proper subset of X is dense in X.
(e) Uncountable, not R, and such that it contains a countable subset which is dense in X and a countable subset which is not dense
in X.
Finish the proof of Corollary 3.2.11.
Finish the proof of Proposition 3.2.14.
Let X be a topological space and let S be a subset of X. Prove that
(S) = S and that S = S.
Let X be a topological space. Let S and T be subsets of X.
74
Topology
(a) Prove that (S n T) = Sn T.
(b) Prove that Su T = Su T.
(e) Give examples to show that in general (Su T) # Su T and
SnT#SnT.
21.
Let X be a topological space.

(a) Suppose X has the property that every subset of X which consists of exactly one point is closed. Prove that if S is any subset
of X then S is closed in X.
(b) Give an example oftopological space X anda subset S of X such
that S is not closed in X.
4
Continuous Functions
ln this chapter we define the notion of a continuous function from one

topological space to another, examine the properties of such functions,
and use them to compare topological spaces. We then'study the important
topological properties of connectedness and compactness and the preservation of these by continuous functions. As usual, ali of our general
topological results will be applied to the set R of real numbers, and
conversely we will use our knowledge of the system R to suggest topological generalizations.
4.1
CONTINUITY
We have stated that we intend to compare topological spaces by means of

continuous functions, but what is a continuous function from one topological space to another? The reader is undoubtedly familiar to some
extent with continuous functions from R to R. Let us examine such functions to see what it is they do in terms of the natural topology on R.
A common intuitive definition of a continuous function from R to R
that is sometimes used in elementary courses is this: a functionf from R
to R is continuous if one can draw its graph without lifting one's pencil
from the paper. What does this mean in terms of the natural topology on
R? To be more specific, what does it mean in terms of the basic open sets
75
76
y
Figure 4.1
f((a,b)) = [c,d)
The continuous image of an open interval need not be an open set.
r1((a,b)) is a union of open intervals
Figure 4.2 The preimage of an open interval under a continuous function is an

open set.
77
Continuity
in the natural topology, namely, the open intervals? As Figure 4.1 shows,
a function from R to R which is continuous according to our intuitive
definition need not carry open intervals onto open intervals, or even onto
open sets. Thus in Figure 4.1 the image under the continuous function f
of the open interval (a, b) is the half-open interval [e, d). Hence we
may conclude that the continuous image of an open set is not always an
open set.
Now suppose we consider preimages under continuous functions. If
we begin with an open set in the range of a continuous function and take
its preimage, we will obtain a set in the domain of the function. Figure 4.2
shows a continuous function f from R to R. Note that in the figure the
preimage under f of the open interval (a, b) in the range is a union ofopen
intervals, and hence an open set, in the domain. This is always so for a
continuous function from R to R: under such a function, the preimage of
an open set is an open set. On the other hand, if a function from R to R
is not continuous, then there will be some open set in its range whose
preimage under the function is not open in its domain. For example,
Figure 4.3 shows a function g from R to R which is not continuous: note
that the preimage under g of the open interval (a, b) is not an open set.
The above remarks suggest that if we want continuity in topological
spaces to be a generalization of our intuitive notion of continuity on R,
.............
--?:
1
1
X
g-1((a,b)) is a half-open interval
Figure 4.3 The preimage of some open interval under a discontinuous function
will not be an open set.
78
then we should consider a function from one topological space to another

to be continuous provided that the preimage of any open set in its range
is an open set in its domain.
Definition 4.1.1 Let (X, r) and (Y, u) be topological spaces. A function
/:X-+ Yis continuous ifwhenever G eu its preimagef- 1(G) er. If/ is not
continuous, it is said to be discontinuous.
Before we proceed to some examples, let us agree that from now on

the set R of real numbers will always have the natural topology unless we
explicitly state otherwise. This will save us constant repetition of the
phrase "let R have the natural topology."
Examples 4.1.2 1. Let X={x,y,z} and r={0,{x},{x,y},X}. Let
Y = {tX, p, y} and u = {0, {tX}, Y}. Let/: X-+ Yand g: X-+ Ybe given by
f(x) = tX
f(y) = /(z) =
and
g(x) =
g(z)
tX
g(y) =
We have
and
Therefore the preimage under f of every open set in Y is an open set in X,
so f is a continuous function from X to Y. For g, on the other hand,
and
g- 1({tX})={x,z}
and since {tX} is open in Y but its preimage under g is not open in X, g is
a discontinuous function from X to Y.
2. Continuity depends on the topology as well as on the function. A
function which is continuous for one topology may not be continuous for
another. For instance, suppose we define X, Y, u,f, and g as in Example
1 above, but change the topology on X by setting r = {0, {x, z }, X};
then f is a discontinuous function from X to Y and g a continuous one.
( Check this.)
3. Every constant function is continuous. To see this, suppose X and
Yare topological spaces and/: X-+ Yis given by f(x) =e for all xeX. If
Continuity
79
G is an open set in Y, then

ifc~G
if CEG
Thereforef- 1(G) is always open in X, sofis continuous.
4. Let /: R -+ R be given by f(x) = x + 1 for ali x E R. Let us show
that f is continuous. It will suffice to show that under f the preimage of
every open interval is an open set in the natural topology. (Why?) But this
isso, for if (a, b) is an open intervalf- 1((a, b)) =(a - l, b - 1).
5. Let XQ: R-+ R be defined as follows:
if XEQ
ifxeR-Q
The function XQ is called the characteristic function of the rationals. It is
discontinuous because every open interval (a, b) contains both rational
and irrational numbers and hence
and {O, 1} is not an open set in the natural topology on R.

6. A function /: X-+ Y may be discontinuous, but its restriction to a
subset of X supplied with the subspace topology may be continuous. For
instance, _suppose /: R -+ R is given by
f(x) =
if x E [0, l]
if x ~ [0, l]
Thenfis discontinuous, sincef- 1((!, ~)) =[O, l], which is not an open set.
However, if we restrict f to the subset [O, l] supplied with the subspace
topology induced by the natural topology, the restriction of f to [O, l] is
a constant function, which is continuous by virtue of Example 3 above.
The definition of a continuous function from one topological space to
another is given in terms of open sets, but of course there must be an
equivalent formulation in terms of closed sets.
Proposition 4.1.3 Let X and Y be topological spaces. A function
/: X-+ Y is continuous if and only if whenever F is a closed set in Y its
preimage 1- 1(F) is a closed set in X.
80
Proof: Supposefis a continuous function. If Fis a closed set in Y, then

Y - F is an open set in Y and therefore
must be an open set in X. But X - J- 1(F) open implies that J- 1(F) is

closed.
Conversely, suppose that J- 1(F) is closed in X whenever F is closed
in Y. If G is open in Y, then Y - G is closed in Y, and therefore
must be closed in X. But this implies thatf- 1(G) is open in X. Hencef is

continuous. D
The next proposition is sometimes useful when a function must
be tested for continuity. lt is really just a restatement of the definition
of continuity in terms of the domain of the function rather than its
range.
f: X-+ Y is continuous if and only if f(S) cf(S) for ali subsets S of X.
Proof: Suppose f is continuous. Then since f(S) is closed in Y, w~
havef- 1(f(S)) closed in X. But S cf- 1(j(S)), and thusf- 1(f(S)) closed,
implies that S cf- 1{f(S)). Applying f to both sides of this last inclusion~
we obtainf(S) cf(S).
To prove the converse, let us suppose that f is a function from X to
Ysuch thatf(S) cf(S) for every subset S of X. Let Fbe a closed set in Y
and let S = f- 1(F). We will be done if we can show that S is closed in X.
We have
f(S) cf(S) = f(f
1(F))
F= F
But f(S) e F implies that S cf- 1(F) = S. We have thus shown that
Se S, and since S e S always holds, we conclude that S = S. Therefore
S is a closed set in X. D
Thus far ali our characterizations of continuity have been global

in nature. The next proposition affords us a local characterization of
continuity.
Continuity
81

f: X-+ Y is continuous if and only if it satisfies the following property for
each xeX:
Whenever G is an open set in Y such that/(x) E G, there is some open

set H in X such that x E H and f(H) e G.
Proof: Let/be continuous and let xE X. If G is open in Yand/(x) E G,
thenf- 1(G) is open in X and xe/- 1(G), so we may take H = 1- 1(G).
Now suppose that f satisfies the property stated in the proposition.
Let G be any set open in Yand consider 1- 1(G). If1- 1(G) is empty, then
it is certainly an open set; if it is not empty, then for each x ef- 1(G) there
is some open set Hx such that XE Hx and f(Hx) e G. But f(Hx) e G
implies that Hx cf- 1(G), and it follows that
where x ranges over ali elements of1- 1(G). This shows thatf- 1(G) is a
union of open sets in X and hence is itself open in X, and therefore that
f is continuous. D
Corollary 4.1.6 Let S be a subset of R and let S have the subspace
topology induced by the natural topology on R. A function f: S-+ R is
continuous if and only if it satisfies the following property for ali x E S:
Whenever (e, d) is an open interval containing x, there is some open

interval (a, b) such that x E (a, b) and f(S n (a, b)) e (e, d).
The proof of the Corollary is left as an exercise. (See Exercise 6 at the
end of this section.)
Proposition 4.1.5 not only gives usa local characterization of continuity, but it also suggests how we may define continuity ata point: we will
say that a function f from X to Y is continuous at a point x E X if it
satisfies the property of the proposition at x.
Definition 4.1.7 Let X and Y be topological spaces and let xe X. A
function f: X-+ Y is continuous at x if it satisfies the following property:
Whenever G is an open set in Y such that/(x) E G, there is some open

set H in X that xE H andf(H) e G.
Iff is not continuous at x, it is discontinuous at x.
82
Note that in the light of the preceding definition and Proposition

4.1.5, we may conclude that a function is continuous if and only if it is
continuous at every point of its domain.
Examples 4.1.8
Y = {tX,
1. Let
X= {x, y, z}
r = {0, {x }, X}.
and
Let
p, y} and u = {0, {tX, p}, Y}. Let/: X-+ Y be given by

f(x) = tX
f(y) =
and
f(z) =y
Clearly,f is continuous at x: for any open set G which contains/(x) = tX

we always have xe {x} and /({x}) e G, and therefore we may always
choose the o pen set H of Defini tion 4.1. 7 to be the set {x}. Similarly, f is
continuous at z because we may always choose H = X. However, f is
discontinuous at y, beca use {tX, p} is an o pen set which contains f(y) = p,
and there is no open set H such that ye H and/(H) e {tX, P}. Thus the
function f is not a continuous function from X to Y.
2. Let/: R -+R be given by f(x) = 2x - 1 for all xe R. We will show
that/ is continuous at each xe R and hence that it is a continuous function from R to R. To show that/is continuous at xe R, it suffices to show
that whenever f(x) = 2x + 1e (e, d), there is some open interval (a, b)
such that xe (a, b) and f((a, b)) e (e, d). But if a= (e+ 1)/2 and
b =(d+ 1)/2, this is the case, so we are done.
3. Let /: R -+ R be given by
f(x) = { _:
ifx >O
if X
~0
Then f is discontinuous at x = O: for f( O) = - 1e ( - 2, O), but there is

no open interval (a, b) such that Oe (a, b) and f((a, b)) e ( -2, O). On
the other hand, f is continuous at x for all x =F O. For instance, if x > O
and f(x) = l e (e, d), then (x/2, 3x/2) is an open interval such that
xe (x/2, 3x/2) and
1((~. 3;))={l}c(c,d)
A similar argument holds if x < O.
4. Let XQ be the characteristic function of the rationals. Thus
if XE Q
ifxeR-Q
83
Continuity
Since every open interval (a, b) contains both rational and irrational numbers, it follows that
XQ((a, b)) = {O, 1}
for every open interval (a, b). Therefore if x is rational, so that xQ(x) = 1,
then there can be no open interval (a, b) such that
xQ((a, b)) e(!,~)
Hence the function is discontinuous at x if x is rational. Similarly, there

can be no open interval (a, b) such that
XQ((a, b)) e (
-!, !)
so the function is discontinuous at x if x is irrational. Therefore the

characteristic function of the rationals XQ is discontinuous at every real
number.
Now that we have various characterizations of continuity at our disposal, we tum to methods of combining known continuous functions to
obtain new ones. Our first result says that the composition of continuous
functions is a continuous function.
Proposion 4.1.9 Let X, Y, and Wbe topological spaces and letf: X-+ Y
and g: Y-+ W. Iff is continuous at xe X and g is continuous atf(x) e Y,
then g f is continuous at x.
Proof: Let G be an open set in W such that (g f)(x) is an element of G.
Since g is continuous at f(x), there is some open set H in Y such that
f(x) e H and g(H) e G. Since f is continuous at x, there is some open
set K in X such that x e K and f(K) e H. But then K is open, x e K, and
\Since
(g f)(K) = g(f(K)) e g(H) e G
0
we have g f continuous at x. O
0
Corollary 4.1.10 Let X, Y, and W be topological spaces. If/: X-+ Y and

g: Y-+ W are continuous functions, then the composite function
g /: X-+ W is also continuous.
84
Since we do not in general have the arithmetic operations of addition,

subtraction, multiplication, and division defined on topological spaces, we
cannot always arithmetically combine functions on such spaces. We can
doso for functions on R, however. The next definition shows how functions on R can be combined arithmetically, and the proposition which
follows demonstrates that arithmetic combinations of continuous functions are continuous.
Definition 4.1.11 Let S be a subset of R, with the subspace topology
induced by the natural topology on R. Let/and g be functions from S to
R. Define
1.
2.
3.
4.
5.
rf: S-+ R, for any r E R, by (rf)(x) = r f(x) for ali x E S;

f + g: S-+ R by (/ + g)(x) = f(x) + g(x) for ali x E S;
f-g: S-+R by (/- g)(x) = f(x) - g(x) for ali xe S;
fg: S-+ R by (/g)(x) = f(x) g(x) for ali x E S;
f/g:S-+R by (f/g)(x)=f(x)/g(x) for ali xeS, provided that
g(x) =F O for all x E S.
Let S be any subset of R and let S have the subspace

topology induced by the natural topology on R. lf/: S-+ R and g: S-+ R
are functions which are continuous at x E S, then
Proposition 4.1.12
1.
2.
3.
4.
The
The
The
The
function rf is continuous at x for every r E R;

functions f + g and f - g are continuous at x;
function fg is continuous at x;
function f/g is continuous at x if g is never zero on S.
Proof: We prove (1) and (2), leaving (3) and (4) as exercises. (See
Exercises 9 and 10 at the end of this section.)
To prove ( 1) it will suffice to show that if (rf)(x) = r f(x) is an
element of an open interval (e, d), then there is some open interval (a, b)
such that xe (a, b) and (rf)(S n(a, b)) e (e, d). (Why?) lf r =O, then
(rf)(x) =O, and if Oe (e, d), we may take (a, b) to be any open interval
containing x and obtain (rf)(S n (a, b)) ={O} e (e, d). Thus ( 1) is true if
r=O.
Now suppose that r >O. lf r -f(x) E (e, d), then/(x) E (c/r, d/r), and
since f is continuous at x, there is some open interval (a, b) such that
x E (a, b) and f(S n ((a, b)) e (e /r, d/r). But this in tum implies that
(rf)(S n (a, b)) e (e, d)
Continuity
85
Hence (1) holds if r >O, and a similar argument proves the case where
r <0.
Now we show thatf + g is continuous at x. Suppose that/(x) + g(x)

is an element of the open interval (e, d). Let us choose real numbers
such that
Ct
<f(x) <d1
and
Thus/(x) e (c1> d1) and g(x) e (c2 , d2 ). Sincefis continuous at x, there is an

open interval (a1' b1) such that
and
Since g is continuous at x, there is an open interval (a2 , b2) such that
and
Let
Then (a, b) is an open interval, x e (a, b), and

(f + g)(S n(a, b)) e (e, d)
Therefore f + g is continuous at x.
Now consider f - g: by part ( 1) of the proposition, ( - l)g is continuous at x, and therefore by what we have just proved, so is
f + ( - l)g = f - g. D
Let S be any subset of R and let S have the subspace
topology induced by the natural topology on R. Iff: S-+R and g: S-+R
are continuous functions, then
Corollary 4.1.13
1. The function rf: S-+ R is continuous, for every r e R;

2. The functions f + g: S-+ R and f - g: S-+ R are continuous;
86
3.
4.
The function fg: S-+ R is continuous;
The functionf/g: S-+R is continuous if g(x) #O for ali
xE
S.
We conclude this section with a brief discussion of homeomorphic

topological spaces. The following definition plays much the sarne role for
topology as the definition of set equivalence (Definition l.3.1, Section l.3
of Chapter 1) does for set theory.
Definition 4.1.14 A functionf from a topological space X to a topological space Y is a homeomorphism from X to Y if it is continuous and its
inverse function J- 1 exists and is a continuous function from Y to X. If
there is a homeomorphism from X to Y, we say that X is homeomorphic
to Y.
Implicit in the definition of a homeomorphism is the fact that both the
function and its inverse must be one-to-one and onto. Furthermore,
if a function is a homeomorphism from X to Y, then its inverse is a
homeomorphism from Y to X. Note also that if X and Y are homeomorphic topological spaces, then they are equivalent sets as defined in Chapter 1, Section 1.3. Therefore sets which are not equivalent cannot be
homeomorphic, no matter how they are topologized. For instance, R and
Q cannot be homeomorphic because they are not equivalent. However,
equivalent sets may not be homeomorphic even if they are topologized in
what appear to be similar ways. For instance, the results of the next
section will allow us to prove that R with the natural topology and [O, l]
with the subspace topology induced by the natural topology are not
homeomorphic even though they are equivalent sets. (See Exercise 15,
Section 4.2 of this chapter.)
Our final proposition in this section compares the topologies on
homeomorphic spaces.
Proposition 4.1.15 Let X and Y be topological spaces. If f: X-+ Y is a

homeomorphism from X to Y, then
l.
2.
A set G is open in X if and only if its image f(G) is open in Y;

A set F is closed in X if and only if its image f(F) is closed in Y.
If G is open in X then, since J- 1: Y-+ X is continuous, we must

have (f- 1) - 1(G) = f(G) open in Y. Conversely, ifj(G) is open in Y, then
the continuity ofjimplies thatf- 1(f(G)) = G must be open in X. This
proves (1). We leave (2) as an exercise. (See Exercise 17 at the end ofthis
section.) D
Proof:
Continuity
87
The import of Proposition 4.1.15 is that if X and Y are homeomorphic topological spaces, then there is a one-to-one correspondence between their open sets (and between their closed sets). Thus homeomorphic
spaces are topologically indistinguishable from one another, being topologically identical in much the sarne way that equivalent sets are set-theoretically identical.
EXERCISES
1.
2.
3.
Let X and Y be topological spaces.

(a) Prove that if X has the discrete topology, then any function
from X to Y is continuous.
(b) Prove that if Y has the indiscrete topology, then any function
from X to Y is continuous.
(e) Suppose X has the indiscrete topology, Y has the discrete topology, andfis a continuous function from X to Y. What can you
say aboutf?
Let X and Y be topological spaces and letfbe a continuous function
from X to Y. Let S be a subset of X, and let S have the subspace
topology induced by the topology on X andf(S) the subspace topology induced by the topology on Y. Prove that the restriction of/to
S is a continuous function from S to f(S).
Let X be a set and S a subset of X. Define the characteristic function
Xs of S as follows:
xs(x) =
4.
5.
if xe S
ifxeX-S
Thus the characteristic-function of S is a function from X to the set

{O, 1}. Suppose X is a topological space and let {O, 1} have the
discrete topology. Prove that a function from X to {O, 1} is continuous if and only if it is the characteristic function of a subset of X
which is both open and closed in X.
Let X and ..Y be topological spaces. Prove that a functionf: X-+ Y is
continuous if and only if it satisfies the following property: whenever
S is a subset of X and x is a limit point of S, then either f(x) ef(S)
or f(x) is a limit point of f(S).
A linear function on R is a function of the form f(x) = ax + b, for
ali x e R, where a and b are fixed real numbers. Using only Definition 4.1.1, prove that every linear function on R is continuous.
88
6.
7.

Use Corollary 4.1.6 to prove that the absolute value function
f(x) = lxl is continuous on R.
8. Let /: R -+ R be given by f(x) = [x], where [x] denotes the greatest
integer less than or equal to x. Prove that f is continuous at x if and
only if x~Z.
9. Let S be a subset of R with the subspace topology induced by the
natural topology on R. Let xe S and let/: S-+R and g: S-+R be
continuous at x.
(a) Prove that if g(x) =O, then fg is continuous at x.
(b) Define a function h from S to R be setting
h
10.
11.
12.
13.
14.
15.
16.
= f(g
- g(x))
+ g(x)(f -
/(x))
Prove that if h is continuous at x, then so is fg.

(e) Use (a) and (b) to prove (3) of Proposition 4.1.12.
Prove ( 4) of Proposition 4.1.12.
Let n e N. Let /: R-+ R be given by f(x) = xn, for ali x e R. Prove
that f is a continuous function on R.
A polynomial function on Ris a function whose defining equation is
of the form f(x) = a0 xn + a 1xn - 1 + +ano where n e N and
a0 , a 1, , an are fixed real numbers with a0 =F O. Prove that every
polynomial function on R is continuous.
Let/: R-+R.
(a) Suppose/is continuous when R has the natural topology. Is it
true that/must be continuous when R has the open ray topology? Justify your answer.
(b) Suppose that/is continuous when R has the open ray topology.
Is it true that f must be continuous when R has the natural
topology? Justify your answer.
Does Exercise 13 suggest to you a general result concerning continuity and comparable topologies? If so, state and prove it.
Let (O, 1) have the subspace topology induced by the natural topology on R and let /:(O, 1)-+ R be given as follows: f(x) =O if x is
irrational; if x is rational, write x = p /q, where p e N, q e N and p, q
have no common factors, and set/(x) = 1/q. Prove that/is continuous at x if and only if x is irrational.
Let f be a continuous function on R and suppose that f is not
identically zero on R, that is, that there is some y e R such that
f(y) =F O. Prove that there exists a closed interval [a, b], with a < b,
such that f(x) =F O for any x e [a, b].
89
Continuity
17.
18.
Prove ( 2) of Proposition 4. l.15.

Let X, Y, and W be topological spaces. Prove that
(a) X is homeomorphic to X;
(b) If X is homeomorphic to Y then Y is homeomorphic to X;
(e) If X is homeomorphic to Y and Y is homeomorphic to W then
X is homeomorphic to W.
19. Prove that a functionf from a topological space X to a topological
space Y is a homeomorphism if and only if f is continuous, one-toone, onto, and has the property that /( G) is open in Y whenever G
is open in X.
20. Give an example of topological spaces X and Y and a one-toone onto function f from X to Y such that f is continuous but
not a homeomorphism, and show that the inverse of f is not
continuous.
21. Letfbe a function on R and let xe R. We say thatfis continuous
from the right at x if whenever f(x) e (e, d) there is some b e R such
that f([x, b)) e (e, d). Similarly, f is continuous from the left at x if
whenever f(x) e (e, d) there is some a e R such thatf((a, x]) e (e, d).
(a) Letf: R-+R be given by
f(x) = {
22.
23.
-1
1
ifx <O
ifx ~o
Show thatfis continuous from the right at Obut not continuous

from the left at O.
(b) Letfbe the greatest integer function of Exercise 8 above. Show
thatfis continuous from the right at every xe R but continuous
from the left at x if and only if x~Z.
Let a, b be real numbers with a < b and let [a, b] have the subspace
topology induced by the natural topology on R. Show that a function from [a, b] to Ris continuous if and only if it is continuous on
(a, b), continuous from the right ata, and continuous from the left
at b.
Let the Euclidean plane R 2 have the product topology. Define
by px((x, y))
=X
for ali (x, y) e R 2
by py((x, y))
=y
for ali (x, y) e R 2
and
90
(The functions Px and Py are the projections of the plane onto the
x- and y-axes, respectively.) Prove that
(a) The projections Px and Py are continuous functions;

(b) If X is a topological space andf: X-+ R 2 , thenf is continuous if
and only if Px f and Py f are continuous functions from X
to R.
4.2 CONNECTEDNESS AND COMPACTNESS
ln this section we consider two important topological properties, connectedness and compactness, which are preserved by continuous functions.
We will make a particularly close study of these properties and their
relationship to the natural topology on the set R of real numbers.
Let us begin by raising again the question we asked at the start of this
chapter: what do continuous functions do to open intervals in R? As we
saw in Figure 4.1, if f is continuous on the open interval (a, b) then
f((a, b)) can be a half-open interval, which is neither open nor closed in
the natural topology. We might guess, then, that if the continuous image
of an open interval is not necessarily open, at least it must be an interval.
But this is not so: Figure 4.4 shows that the continuous image of an open
interval can be a ray. lt is true, however, that these are the only possibilities: the continuous image of an open interval must be either an interval
ora ray. As we shall see, the reason this isso is that continuous functions
preserve a topological property known as connectedness.
y
1
1
1
1
1
e
1
------,-1
X
a
f((a,b))
= [e,+ :xi)
Figure 4.4 The continuous image of an open interval need not be an interval.
Connectedness and Compactness
91
Definion 4.2.1 A subset S of a topological space X is disconnected in X

if there exist open sets G and H in X such that G n S and H n S are
nonempty disjoint sets and
S = (G nS) u(H nS)
(see Figure 4.5). The set S is connected in X if it is not disconnected in X.

Note that the subset S referred to in the definition need not be a
proper subset of the topological space X; if S =X, we simply refer to the
space X as being disconnected or connected.
Examples 4.2.2 1. Let X= {x, y, z} and let r = {0, {x }, {y, z }, X}. The
space X is disconnected because if G = {x} and H = {y, z }, then G and H
are open in X, G n X = G and H n X = H are nonempty disjoint sets, and
X= (G nX) u(H nX)
2. Let X= {x,y, z} and r = {0, {x}, {x,y}, {x, z}, X}. Let
S = {y, z}. The subset S is disconnected in X, since the sets G = {x, y} and
H = { x, z} are open sets in X such that G n S and H n S are nonempty
and disjoint and
S = (G nS) u(H nS)
(Note that the sets G and H are not disjoint in X: the definition only
requires that G n S and H n S be disjoint, not that G and H be disjoint.)
S is disconnected in X
Figure 4.5 A disconnected set.
92
Now let T = {x, y }; then Tis connected in X, for if
T= (G n T) u(H n T)
with G arid H open sets and G n T, H n T nonempty and disjoint, then

xe Timplies that x is an element ofjust one ofthe open sets G and H, and
this is impossible since every nonempty open set in X contains x. For the
sarne reason, the space X is a connected space.
3. The set Q of rational numbers is disconnected in the space R of real
numbers. To see this, let
G =(-oo,
.j2.)
and
H=(fi, +oo)
Then G and H are open sets, G n Q and H n Q are nonempty and disjoint,
and
Q = (GnQ)u(HnQ).
4. The empty set cannot be disconnected (why not?) and therefore is
a connected subset of every topological space.
We will soon prove that intervals and rays are connected subsets of R,
and that in fact these are the only nonempty connected subsets of R.
Our first proposition concerning the phenomenon of connectedness
provides us with an alternate definition of disconnectedness, one which is
often easier to use than that given in Definition 4.2.1.
Proposition 4.2.3 Let X be a topological space. A subset S of X is disconnected in X if and only if S contains a nonempty proper subset T such that
Tis both open and closed in the subspace topology induced on S by the
topology on X.
Proof:
Suppose S is disconnected in X, so that

S
= (G nS) u(H nS)
for some sets G and H such that G, H are open in X and G n S, H n Sare
nonempty and disjoint. Let T = G n S. Then T is a nonempty proper
subset of S, T is open in the subspace topology on S because it is the
intersection of S with the open set G, and Tis closed in the subspace
topology on S because its complement in S is H n S, which is also open
in the subspace topology on S.
93
Conversely, suppose that Tis a nonempty proper subset of S which

is both open and closed in the subspace topology on S. Since
S = Tu (S - T) and since T and S - T are both open in the subspace
topology on S, there exist sets G and H which are open in X such that
T = G n S and S - T = H n S. Therefore
S = (G nS) u(H nS)
where G, H are open in X and G n S and H n S are nonempty disjoint

sets. D
We next use Proposition 4.2.3 to show that continuous functions
preserve connectedness.
Proposition 4.2.4 Let X and Y be topological spaces. lf S is a connected
subset of X and f is a continuous function from X to Y, then f(S) is a
connected subset of Y.
Proof: Suppose S is connected in X. Since f is a continuous function
from X to Y, f restricted to S is a continuous function from S with the
subspace topology induced by X to f(S) with the subspace topology induced by Y. (See Exercise 2 ofExercises 4.1.) If/(S) is disconnected in Y,
then f(S) must contain a nonempty proper subset T which is both open
and closed in the subspace topology on f(S). But since /: S-+ f(S) is
continuous,J- 1(T) is a nonempty proper subset of S which is both open
and closed in the subspace topology on S; hence S must be disconnected,
a contradiction. Therefore f(S) is connected in Y. D
As a consequence of Proposition 4.2.4, we can prove the important

Intermediate Value Theorem, which states that if f is a continuous function from a topological space X into R, then on any connected subset of
X the function f will assume every value intermediate to any two of its
values. Figure 4.6 depicts the Intermediate Value Theorem when X= R
and the connected subset of X is the interval [O, 1].
Theorem 4.2.5 (Intermediate Value Theorem) Let X be a topological
space and let /: X-+ R be a continuous function. Let S be a connected
subset of X. lf xeS and yeS and/(x) # f(y), then for every real number
t between/(x) and/(y) there is some element zeS such that/(z) = t.
Proof: By Proposition 4.2.4,f(S) is connected in R. We must show that
if t is any real number between f(x) and /(y), then t ef(S). The rays
94
y
f(x)
1
1
- - i1- - - 1
f(y)
1
1
1
_ _ _ ,_ _
1
---r-----
1
1
f(y) .;; t .;; f(x) implies t = f(z)
Figure 4.6 The Intermediate Value Theorem.
(-oo, t) and (t, +oo) are open in R and (-oo, t)nf(S), (t, +oo)nf(S)
are disjoint sets which are nonempty because f(x) is an element of one of
them andf(y) is an element of the other. If t ~f(S), we have
f(S) = (( - oo, t) nf(S)) u ((t, + oo) nf(S))
and hencef(S) is disconnected in R, a contradiction. Therefore tef(S). D

Note that the conclusion ofthe Intermediate Value Theorem need not
hold if the subset S is not connected. For instance, letf: R-+ R be defined
by f(x) = x for ali x eR and consider the disconnected subset Q of R. The
is between /(1) = 1 and
function f is certainly continuous, and
/(2) = 2, but there is no zeQ such thatf(z) =
Now we tum to an examination of the connected subsets of R. As
previously mentioned, the only nonempty connected subsets of R are the
intervals and rays.
J2
J2.
Proposition 4.2.6 A nonempty subset I of R is connected if and only if it

is either an interval or a ray.
95
Proof: We prove first that if I is either an interval or a ray, then it

is connected in R. If I is a closed interval which contains only one
point, it is trivial that I is connected, so we will assume that I contains more than one point. Note that if xe/ and ye/ with x :s:;;y then
[x, y] e/.
Suppose that I is disconnected in R, so that I = J u (/ - J), where J
is some nonempty proper subset of I which is both open and closed in the
subspace topology induced on I by the natural topology on R. Let x eJ
and y e/ - J. Since x =F y, we may assume that x < y. Let
S={xeJlx:s:;;z<y}
Then x e S, so S is nonempty, and since S is bounded above by y we
conclude that s = sup S exists. Note that x :s:;; s :s:;; y, so se/, and hence
either s eJ or se/ - J. We shall show that neither of these cases can occur,
and therefore that I must be connected.
Suppose se/ - J: Se J implies Se J, and since s = sup Ses, we
must have s eJ. But J is closed in the subspace topology on /, so se/ and
s eJ together imply that s eJ. Therefore we have s ef:J and s eJ, which is
impossible.
Now suppose that s eJ: then s ef: I - J, and thus s =F y, so s < y. Consider the interval (s, y] e/. If J n (s, y] is not empty, then s =F sup S; hence
J n (s, y] = f/J and this implies that (s, y] e (/ - J). Since s is a limit point
of (s, y], it is therefore a limit point of I - J, and since I - J is a closed set,
it follows that se/ - J. Thus we have sef:/ -J and se/ - J, and we have
shown that I is connected.
To finish the proof we must demonstrate that if I is a nonempty
connected subset of R, then it must be an interval ora ray. Ifwe can find
xel, yel, and zeR - / such that x < z <y, then
/=((-oo,z)n/)u((z, +oo)n/)
is disconnected in R, a contradiction. Therefore we conclude that the
connected set I has the property that whenever x e/, y e/ and x <y, then
[x, y] e /. But any subset of R which has this property is either an interval
ora ray. (See Exercise 7 of Section 2.1, Chapter 2.) D
Corollary 4.2. 7 The set R of real numbers contains no nonempty proper
subset which is both open and closed in the natural topology on R.
96
Corollary 4.2.8 Let I be a subset of R and letfbe a continuous function

from I to R. If I is an interval, then f(/) is either an interval or a ray; if
I is a ray, thenf(/) is either an interval ora ray.
We now know what continuous functions on R do to open intervals:

they carry them onto either intervals or rays, and, in the light of Figures
4.l and 4.4, this is as specific as it is possible to be. However, we can be
more specific about what continuous functions do to closed intervals.
As we shall see, continuous functions carry closed intervals onto closed
intervals, and the reason this is so is that, in addition to connectedness,
continuous functions also preserve a topological property known as
compactness.
Let A be an index set and for each tX e A let G,,_ be an open set in X. The
collection of sets {Ga. 1tX e A} is an open covering for S if
Definition 4.2.9
Se
U G,,_
a.e A
An open covering {Ga. 1tX e A} for S has a finite subcovering if there is a

finite subset B of A such that
Se
U Gp
PeB
The subset S of X is compact in X if every open covering for S has a finite

subcovering.
Note that the subset S of the definition need not be a proper subset
of X. If S = X, then we refer to X as a compact topological space.
Examples 4.2.10 l. Let X be any topological space. If S is a finite subset
of X, then S is compact in X. To prove this, let {G,,_ tXEA} be an open
covering for S. If S = 0, then we may choose any {JeA and {Gp} will
certainly be a finite subcovering for S. If S is not empty, then
S = {x., ... , xn}, for some n eN. For each i eN, l ~ i ~ n, there is some
tXieA such that x;eG,,_;- But then {G,,_1 t ~ i ~ n} is a finite subcovering
for S. We have thus shown that every open covering of S has a finite
subcovering, so S is indeed compact in X.
97
2. It follows from the previous example that every subset of a finite

topological space is compact in the space. Thus, as a special case, every
finite topological space is a compact space.
3. The space R of real numbers is nota compact space. The collection
of intervals {( -n, n) 1n eN} is an open covering for R, and if this covering had a finite subcovering, say {( -n1o n 1), ... , ( -nk, nk)}, then we
would have
R=
U (-n,, n
1)
1 ~i~k
But if m = max {n1o ... , nd this would imply that R = ( -m, m), which is
impossible. Since we have produced an open covering of R which has no
finite subcovering, it follows that R is not compact.
4. Open intervals are not compact in R: an open covering of (a, b)
which has no finite subcovering is {(a+ (l/n), b) 1nen}. Similarly, rays
are not compact in R. We leave the proofs of these facts as an exercise.
We will soon show that every closed interval is compact in R. Before
doing so, however, let us prove the important fact that continuous functions preserve compactness.
Proposition 4.2.11 Let f be a continuous function from a topological
space X to a topological space Y. If S is a compact subset of X, then/(S)
is a compact subset of Y.
Proof: Let {G"' 1 tXEA} be an open covering for /(S) in Y. Since/ is

1(GIX) 1 tXEA} is an open covering for s in X,
continuous, the family
and therefore it must have a finite subcovering, say
u-
But then
Se
LJ 1- 1(Gix)
1 ~i~n
implies that
/(S) e
LJ
G"',
1 ~i~n
Therefore the open covering {G"' 1 tX e A} for /(S) has a finite subcovering,
namely {G"'' ... , G"'J. O
Now we show that closed intervals are compact subsets of R.
98
Proposition 4.2.12 Every closed interval in R is a compact subset of R.

Proof: Let {G" 1ex e A} be an open covering for the closed interval [a, b].
lt follows that for each x E [a, b], {G" 1 ex e A} is also an open covering for
the interval [a, x]. Let S be the set of ali elements x of [a, b] such that the
open covering {G" 1cxeA} for [a, x] has a finite subcovering.
Since there is some f3 e A such that a eGp, we have {Gp} a finite
subcovering for the interval [a, a]. Hence a eS, so S is nonempty, and
since S is clearly bounded above by b, this implies that s = sup S exists.
Note that a ~ s ~ b.
Since there is some yt= A such that s E Gy, and since Gy is an open set
in R, there exists an open interval (e, d) such that s E (e, d) e Gr Now
s = sup S is either an element of S or a limit point of S, and in either case,
there exists an element x of S such that e < x ~ s. But x E S implies that
there is a finite subcovering, say {G" 1, , G"J' for the interval [a, x].
Thus {G" 1, , G"n' Gy} must be a finite subcovering for [a, s]. If s < b,
then there is some y E (e, d) n (a, b) such that s < y. But then
is a finite subcovering for [a, y], and this implies that y eS, contradicting
the fact that s = sup S. Therefore s < b is impossible, so s = b, and
is a finite subcovering for [a, b]. D

We have now shown that closed intervals are compact subsets of R.
Of course, not ali closed subsets of R are compact, and an examination of
the proof of Proposition 4.2.12 shows that the boundedness of a closed
interval is crucial to its compactness. However, it is true that a compact
subset of R must be closed. (ln fact, it is true that a subset of Ris compact
if and only if it is closed and bounded: this is the important Heine-Borel
Theorem, which we will soon prove.) lt is important to realize, however,
that in a general topological space compact subsets need not be closed.
The following example exhibits a topological space and a compact subset
of that space which is not closed in the topology.
Example 4.2.13 Let R have the open ray topology. The interval [O, l] is
neither open nor closed in the open ray topology (why?), but it is compact. To see this, suppose that {G" 1cxeA} is an open covering for [O, l]
99
in the open ray topology. Since Oe Gp for some f3 e A, and since Gp is an

open ray, it follows that either Gp = R or Gp =(a, + oo) for some a <0.
ln either case, [O, l] e Gp, so the open covering {G"' 1 cxeA} has a finite
subcovering {Gp}.
Let us consider for a moment the difference between R with the
natural topology and R with the open ray topology. The open ray topology is much smaller than the natural topology, so small in fact that it is
not possible in the open ray topology to separate distinct points by surrounding them with open sets, although this can always be done in the
natural topology. Thus, in the natural topology if x # y we can always
find open sets G and H such that xeG, yeH, and G nH = f/J, whereas in
the open ray topology this is not the case, for if x < y, then any open ray
which contains x must also contain y. See Figure 4.7. This situation
occurs because the open ray topology has comparatively few open sets,
while the natural topology is rich in open sets. A topological space which
is like the reais with the natural topology in that it has enough open sets
to separate points is called a Hausdorff space.
Definition 4.2.14 A topological space is a Hausdorff space if whenever
x e X, y e X, and x # y, then there exist open sets G and H in X such that
xeG, yeH, and G nH = f/J.
( 1 )
( 1 )
A with the natural topology
r~~~~~J\-.~~~-..
'
A with the open ray topology
Figure 4.7 Separation of points depends on the topology.
100
Examples 4.2.15 l. Let X= {x, y, z} and r = {0, {x }, {x, y }, X}; then X

is not a Hausdorff space because there are no open sets G, H such that
xeG, yeH, and G nH = 0.
2. Every discrete topological space is a Hausdorff space. Every indiscrete topological space which has more than one point is not a Hausdorff
space.
3. The set R of real numbers with the natural topology is a Hausdorff
space. The set R with the open ray topology is not a Hausdorff space.
The distinction between Hausdorff and non-Hausdorff spaces is a
fundamental one. Many things which are true in Hausdorff spaces are not
true in non-Hausdorffspaces, and one ofthese is that in Hausdorffspaces,
compact subsets must be closed.
Proposition 4.2.16 A compact subset of a Hausdorff space is closed.

Proof: Let S be a compact subset of the Hausdorff space X. If S is empty
or if S = X, we are done, so we assume that S is a nonempty proper subset
of X. We will prove that S is Closed by showing that X - S is open.
Let y e X - S. Since X is a Hausdorff space, for each x e S there ex~st
open sets Gx and Hx such that xeGx, yeHx, and Gx nHx = 0. Clearly
Se
LJ
Gx
xeS
and because S is compact the open covering {Gx 1 xeS} has a finite
subcovering, say {Gx 1, , GxJ Therefore
and
Note that l 1 ,,.;,,.n Hx, is an open set.
For each i, l ~ i ~ n, Gx, n Hx, = 0 implies that
Hence
We have thus shown that for each y e X - S there exists an open set which
101
contains y and is contained in X - S, and it follows that X - S is an open

set in X. D
Now we can prove the Heine-Borel Theorem, which completely characterizes the compact subsets of R in the natural topology.
Theorem 4.2.17 (Heine-Borel)
if it is closed and bounded.
A subset S of Ris compact if and only
Proof: Suppose S is a compact subset of R. Since Ris a Hausdorff space,

S is closed. Furthermore, S compact implies that the open covering
{( -n, n) 1 n eN} has a finite subcovering, say {( -n;, n;) l 1 ~ i ~ k }. Let
m = max {n 1, , nk}. Then Se ( -m, m), which shows that S is
bounded.
Conversely, suppose that S is closed and bounded in R, and let
{ G"' 1 oc e A} be an open covering for S. Because S is bounded, there exists
a closed interval [a, b] such that S e [a, b,], and it follows that
{ G"' 1 oc e A} u (R - S) is an open covering for [a, b]. But since closed intervals are compact, this open covering must have a finite subcovering, say
{G"' 1, , G'n' R - S}. It follows that {G"' 1, , G"'J is a finite subcovering for S. Hence S is compact. D
The following consequence of the Heine-Borel Theorem, known as

the Extreme Value Theorem, is very important. It says that a continuous
function into R attains its maximum and its minimum on any nonempty
compact subset of its domain. Figure 4.8 illustrates the theorem for a
function from R into R and the compact subset [O, l].
Theorem 4.2.18 (Extreme Value Theorem) If /:X-+ R is continuous
and Sisa nonempty compact subset of X, thenf attains its maximum and
minimum on S; that is, there exist elements s and t in S such that
f(s) ~f(x) ~f(t) for ali xeS.
Proof:
Since f is continuous and S is compact in X, f(S) is compact in
R. Thus, by the Heine-Borel Theorem, f(S) is a closed and bounded

subset of R. Since S is nonempty so is f(S), and hence sup f(S) and
inf/(S) exist. Note that inf/(.sj ~f(x) ~ supf(S) for ali xeS. Sincef(S)
is closed in R, it contains ali its limit points, and from this it follows that
supf(S) and inf/(S) are elements of f(S). But this means there exist
elements s and t in S such thatf(s) = inf/(S),f(t) = supf(S), and we are
done. D
102
y
f(t)
1
1
1
1
1
1
- - ---1--- --1
f(s)
1
1
1
s
1
1
1
X
f(s).;; f(x).;; f(t) for all xe[0,1)
Figure 4.8
The Extreme Value Theorem.
Corollary 4.2.19 If S is a nonempty, closed, and bounded subset of

R andf: S-+R is continuous, thenf attains its maximum and minimum
on S.
Note that in the preceding corollary both closedness and boundedness are necessary. For example, letf(x) = x for ali xeR. This function is
certainly continuous on R, but it attains neither its maximum nor its
minimum on the closed (but unbounded) set R; similarly, it attains neither its maximum nor its minimum on the bounded (but not closed) set
(O, l).
We now bring this section to a close by proving that the continuous

image of a closed interval in Ris a closed interval.
If f is a continuous real-valued function defined on
the closed interval [a, b], thenf([a, b]) is a closed interval.
Proposition 4.2.20
Proof: Since [a, b] is connected and compact and f is continuous, it

follows thatf([a, b]) is connected and compact. Butf([a, b]) connected in
R implies that it is either an interval ora ray, andf([a, b)] compact in R
implies that it is closed and bounded. Since closed rays are unbounded,
f([a, b]) must therefore be a closed interval. D
103
EXERCISES
1.
2.
3.
4.
5.
6.
7.
8.
9.
Let X be a topological space and let S be a nonempty subset of X.

Prove:
(a) If X has the indiscrete topology, then S is connected;
(b) If X has the discrete topology, then S is connected if and only
if S has exactly one point.
Let X be a topological space and let {O, 1} have the discrete topology. Prove that X is a connected space if and only if there does not
exista continuous function from X onto {O, l}.
Prove that the set of all irrational real numbers is disconnected in R.
Let [O, l] have the subspace topology induced by the natural topology on R. Prove that the Cantor ternary set is disconnected in
[O, l].
Let X be a topological space and let S be connected in X. Prove that
(a) The closure S is connected in X;
(b) If S is dense in X, then X is a connected space.
(a) Letfbe a nonconstant function from R to R. Show thatfcannot
be continuous if either f(R) e Q or f(R) e (R - Q).
(b) Show that there is no continuous function /: R-+ R such that
f(x) is rational whenever x is irrational and f(x) is irrational
whenever x is rational.
A polynomial p(x) = a0 xn + + am where a0 #O, is said to be of
degree n. Prove that a polynomial of odd degree has at least one real
root; that is, prove that if n is odd then there is at least one real
number r such that p(r) = O.
Prove that if xe[O, +oo) and neN, then
exists in [O, +oo).
Let [O, l] have the subspace topology induced by the natural topology on R and let /: [O, l]-+ [O, l] be continuous. Prove that f has a
fixed point; that is, prove that there is some x e [O, l] such that
f(x)
10.
11.
12.
=X.
Let X be a topological space and let S be a subset of X. Prove the

following:
(a) If X has the indiscrete topology, then S is compact in X;
( b) If X has the discrete topology, then S is compact in X if and only
if S is a fini te set.
Prove from the definition of compactness that open intervals and
rays are not compact in R.
Let /: R-+ R be continuous. What can be said about f((a, + oo ))?
About f([a, + oo ))?
104
13.
Prove that if X is a compact topological space and S is a closed set

in X, then S is compact in X.
Let [O, l] have the subspace topology induced by the natural topology on R. Prove that the Cantor ternary set is compact in [O, l].
Let [O, l] have the subspace topology induced by the natural topology on R. Prove that R is not homeomorphic to [O, l].
Let X be a compact topological space and let Y be a Hausdorff
space. Letf: X-+ Ybe one-to-one, onto, and continuous. Prove that
f is a homeomorphism.
Let neN. Letf: [O, +oo)-+[O, +oo) be given by f(x)
for ali
x e [O, + oo ). Prove that f is a continuous function.
Prove the Cantor Intersection Theorem: let {Fn 1 n e N} be a collection of nonempty closed sets in R such that
(a) F 1 is bounded; and
(b) Fn+ 1 e Fn for ali neN.
Then lneN Fn =F 0.
(Hint: if the intersection is empty, then {R - Fn 1n eN} is an open
covering for F 1.)
Let X be a topological space. A collection {F.. 1 oc e A} of closed sets
in X has the finite intersection property if the intersection of any finite
number of the sets in the collection is nonempty. Prove that X is a
compact topological space if and only if whenever {F"' 1oceA} is a
collection of closed sets in X having the finite inters(-..;tion property,
then lixeA F"' =F 0.
Topological spaces can be classified according to the manner in
which their open sets can be used to separate points. The following
axioms are known as separation axioms.
Let X be a topological space; then
14.
15.
16.
17.
18.
19.
20.
=..::0
O.
X is a T0-space if whenever x e X and y e X with x =F y there is an

open set in X which contains one of the points x or y but not the
other.
1. X is a T 1-space if whenever x e X and y e X with x =F y there are
open sets G and H in X such that xeG, yeH, x~H, and y~G.
2. X is a T2 -space if whenever x e X and y e X with x =F y there are
open sets G and H in X such that x eG, y eH, and G n H = 0.
(Thus T2-space is merely another name for Hausdorff space.)
3. X is a regular space if whenever F is a closed set in X and
xeX - F there are open sets G and H in X such that xeG,
F e H, and G n H = 0. If X is both a regular space and a
T1-space, it is a T 3-space.

4.
105
X is a normal space if whenever E and F are disjoint closed sets

in X there are open sets G and H in X such that E e G, F e H,
and G n H = f/J. If X is both a normal space and a T1-space it is
a T4 -space.
(a) Give an example of a T4 -space.

(b) Give an example of a space which is nota T0-space.
(e) The five separation axioms are progressively stronger: prove that
if X is a Ti-space, 1 ~ i ~ 4, then X is a Ti_ 1-space.
For each i, O~ i ~ 3, there exist spaces which are Ti-spaces but not
Ti+ 1-spaces.
(d) Give an example of an infinite topological space which is a
T0 -space but not a T1-space.
(e) Prove that if X is a T1-space, then every subset of X which consists of a single point is closed in X.
(f) Give an example of a topological space which is a T1-space but
not a T2-space.
It is not so easy to find examples of spaces which are T2 but not T 3
and T 3 but not T4 The interested reader should consult Kelley's
General Topology or another textbook in topology. It is not too
difficult, however, to show that Ris a T 4 space.
(g) Let X be a topological space. Suppose that whenever E and F are
disjoint closed sets in X there is a continuous function /: X-+ R
such thatf(E) e ( -oo, O) andf(F) e (O,+ oo). Prove that Xis a
normal space.
(h) Use (g) to show that R is a normal space, and hence a T4 -space.
(Hint: define functions f, g from R to R by
f(x) = inf {lx -
YI 1 y e E}
for ali
XER
g(x) = inf {lx -
YI 1yeF}
for ali
XER
and
and consider f
- g.)
5
Sequences and Series
ln the previous chapter we utilized continuous functions to study topological spaces, paying particular attention to the space R of real numbers. ln
this chapter we will continue this process, but concentrating our attention
now on the continuous functions known as sequences. The first section of
the chapter examines the properties of sequences in general topological
spaces, while the second considers sequences in the space of real numbers.
The third section is devoted to the special type of real sequence known as
an infinite series, and the fourth examines the relationship between real
sequences and functional limits.
5.1
SEQUENCES
A sequence in a topological space X is a function from the set N of natural

numbers with the discrete topology into X.
Definition 5.1.1 Let X be a topological space and let N be the set of
natural numbers provided with the discrete topology. A functionf: N-+ X
is called a sequence in X. If n eN, the elementf(n) e X is called the nth term
of the sequence and is denoted by Xn-
Note that every sequence is a continuous function, since any function

whose domain has the discrete topology must be continuous.
107
108
Sequences are important for what they tell us about the range space
X, and hence it is common practice to suppress ali mention ofthe function
f and focus attention only on the image (i.e., the terms) of the sequence.
We shall adopt this convention and in addition shall use the notation {xn}
for a sequence. Therefore rather than speaking of
"the sequence f: N-+ X given by f(n) = Xn for ali n eN"
we will refer to
"the sequence {xn} in X"
Our next definition states what convergence means for a sequence in
a general topological space.
Definition 5.1.2 Let {xn} be a sequence in the topological space X and let
xeX. If for every open set G containing x there is some meN such that
Xn eG for ali n ;;?; m, then we say that the sequence converges to x, and that
x is a limit of the sequence. If the sequence has a limit in X, it is a
convergent sequence; if it has no such limit, it is a divergent sequence.
Thus a sequence converges to x if for every open set G containing x

there is some term of the sequence such that this term and ali succeeding
terms are in G. Another way to say this is that the sequence converges to
x if every open set which contains x also contains ali but finitely many
terms of the sequence (see Figure 5.1).
Examples 5.1.3 1. Let X= {x, y, z} and r = {0, {x }, {x, y }, X}. Let
{xn} be the sequence in X defined by Xn = x for ali n eN. This sequence
converges to x, since if G is an open set containing x then we have either
G = {x}, G = {x, y }, or G =X; hence Xn = xeG for ali n;;?; 1. This sequence also converges to y, for if G is an open set containing y, then either
G = {x, y} or G =X; in either case xn = x is an element of G. Similarly,
the sequence converges to z. This example shows that a convergent
sequence need not have a unique limit.
2. Consider the sequence {xn} = {( - l)n} in R. This sequence is divergent because for any xeR there exists an open interval (a, b) containing x such that either 1 or -1 does not belong to (a, b); therefore for any
m eN there is some n ;;?; m such that xn (a, b). This proves that the sequence cannot converge to x for any xeR, and hence that it is divergent.
109
Sequences
G
{x 0 } converges to x
{x 0 } does not converge to x
Figure 5.1
Convergence of sequences.
3. Let {xn} = {1/n} in R. We shall demonstrate that zero is the unique

limit of this sequence. Suppose that G is an open set in R such that Oe G.
Let (a, b) be an open interval such that Oe (a, b) e G. From Corollary
2.1.13, there is somem eN such that O< 1/m < b. But then 1/n ~ 1/m for
ali n ~ m, and thus xn e G for ali n ~ m. Therefore the sequence converges
to O.
Now we will show that if x =F O, then the sequence cannot converge to
x. First, if x <O, then G = (3x/2, x/2) is an open set containing x and
Xn = 1/n is not an element of G for any n eN because x/2 <O. Thus the
sequence cannot converge to x if x <O. If x >O, then there is some keN
such that O< 1/k < x, and then G = ( 1/k, x + l) is an open set which
contains x and l/n is not an element of G for ali n ~ k, so the sequence
cannot converge to x.
4. Let x e R and consider the sequence {x + 1/n} in R. This sequence
converges to the unique limit x in R. The proof is similar to that of the
previous example. (See Exercise 4 at the end of this section.)
5. ln this example we show that every real number is the limit of a
sequence of rational numbers. Let xeR. By Theorem 2.1.12, for each
n eN there is some rational number qn such that x < qn < x + 1/n. We
110
claim that the sequence {qn} converges to x. To show this, it suffices to

show that if (a, b) is an open interval containing x, then there is some
meN such that qne(a, b) for all n ~m. Rut by the previous example,
the sequence (x + l/n) converges to x, so there is some m eN such that
x + l/n is an element of (a, b) for all n ~ m, and thus qn must be an
element of (a, b) for all n ~ m. A similar argument shows that every real
number is the limit ofa sequence ofirrational numbers. (See Exercise 5 at
6. Let R have the open ray topology; then the sequence {1/n} converges to x for ali x ~O. (See Exercise 6 at the end of this section.)
Examples 1 and 6 ofExamples 5.1.3 show that in general a convergent
sequence need not have a unique limit. Note that the topological spaces
involved in these examples are not Hausdorff spaces. On the other hand,
all the sequences we have seen so far which do have unique limits have
been sequences in Hausdorff spaces. This is another reason why Hausdorff spaces are important: a convergent sequence in a Hausdorff space
must have a unique limit.
Proposition 5.1.4 Every convergent sequence in a Hausdorff space has a

unique limit.
Proof: Let {xn} be a convergent sequence in a Hausdorff space X, and
assume that this sequence converges to the elements x and y of X, where
x # y. Since X is Hausdorff, there exists open sets G and H such that x e G,
yeH, and G nH = f/J. Since x is a limit of the sequence, there is some
me N such that xn e G for all n ~ m, and since y is also a limit of
the sequence, there is some keN such that xneH for all n ~ k. But if
p = max {m, k}, then xP e G n H, a contradiction. D
Corollary 5.1.5' Every convergent sequence in R (with the natural topology) has a unique limit.
Proposition 5.1.4 allows us to use the limit notation in Hausdorff
spaces.
Definition 5.1.6 Let X be a Hausdorff space. lf the sequence {xn} converges to xeX, we write Limn--+co Xn =x.
Next we show that continuous functions preserve the limits of sequences. This does not depend on whether or not the spaces involved are
Hausdorff.
Sequences
111
Proposition 5.1.7 Let f be a continuous function from the topological

space X to the topological space Y. If {xn} is a sequence which converges
to x in X, then the sequence {f(xn)} converges to f(x) in Y.
Proof: Let G be an open set in Y such thatf(x)eG; thenf-i(G) is an
open set in X. Since x ef- i( G) and {xn} converges to x, there is some
me N such that Xn ef- i( G) for ali n ~ m. But then f(xn) e G for ali n ~ m,
so {f(xn)} converges to f(x). O
The converse of the preceding proposition is not true. That is, it is in

general not the case that if f: X-+ Y is such that the sequence {f(xn)}
converges to f(x) whenever {xn} converges to x, thenf must be a continuous function. ln fact, this is not even true if X and Y are both Hausdorff
spaces. It is true for functions from R to R, however, and this is an
important fact about continuity on the reais.
Proposition 5.1.8 A function f: R-+ R is continuous if and only if for

every convergent sequence {xn} in R,
Lim xn =x implies that
n--+oo
n--+oo
Proof: Iffis continuous, the conclusion follows from Proposition 5.1.7.

We must prove the converse: ifjis a function on R which has the property
that
Lim f(xn) = f(x) whenever
Lim Xn =X
n--+
oo
thenf is continuous. We establish this by showing thatf(S) cJ{S'j for ali

subsets S of R ( see Proposition 4.1.4).
For any subset S of R, we have S = S u S, and since f(S) cf(S),
we need only show thatf($) cf(S). This is trivial if S = 0, so we assume
that S is nonempty. Let xeS; then x is a limit point of S and we must
prove that f(x)ef(S). ln other words, we must demonstrate that if x
is a limit point of S, then f(x) is either a point of f(S) or a limit point
of f(S).
Suppose x is a limit point of S. Iff(x) is a point ofj(S), we are done,
so assume thatf(x) does not belong tof(S) and consider the open interval
(x - 1, x + 1). Since x is a limit point of S, there exists Xi e (x - 1, x + 1)
such that Xi eS and Xi =F x. Now suppose that we have found points
112
such that
Xk E (X -
~' X + ~)
and
for l
~k ~n
Since x is a limit point of S, there exists
such that Xn + 1 e S and Xn + 1 =F x. Thus we define inductively a sequence

{xn} such that for each n eN,
and
lt is easy to check that this sequence converges to x, and therefore by
hypothesis the sequence {f(xn)} converges to f(x).
Now we are ready to show that/(x) is a limit point of/(S). Let G be
an open set containing/(x). Since the sequence {f(xn)} converges to f(x),
there is some me N such that f(xn) e G for ali n ~ m. Hence f(xm) is an
element of G. But Xm eS implies that f(xm) e/(S), and f(xm) =F f(x) be-
cause/(x) ~/(S) by assumption. We have thus proved that every open set
which contains f(x) must contain a point of /(S) distinct from f(x) and
this establishes that/(x) is a limit point of /(S). D
Corollary 5.1.9 Let S be a subset of R and x e S. A function /: S-+ R is

continuous at x if and only if for every sequence {xn} in S which converges to x, the sequence {f(xn)} converges to f(x) in R.
Notice that the proof of Proposition 5.1.8 depended heavily on the
fact that for an arbitrary xeR it is possible to define a family of open sets
{Gn 1 neN} such that Gn+ 1 e Gn and xeGn for ali n eN. (ln the proof, we
took Gn = (x - l/n, x + l/n).) The reason that Proposition 5.1.8 does not
hold in general is that it is not always possible to define such a family of
open sets for every element in a topological space.
Examples 5.1.10 Let /: R -+ R be given by

f(x)
= {_
if
if
>0
X ~0
113
Sequences
We see thatf is discontinuous at O, since the sequence {l/n} converges to

Owhile the sequence {f(l/n)} = {l} converges to 1. However,fis continuous at x for ali x =F O. For instance, suppose x >O; then if {xn} converges
to x, there must be some me N such that xn > O for ali n ~ m. But then
f(xn) = 1 for ali n ~ m, so the sequence {J(xn)} converges to f(x) = 1, and
hencefis continuous at x. A similar argument shows thatfis continuous
at
if X< 0.
The existence of the arithmetic operations in the system R allows us

to form new sequences from old ones. Thus, if {xn} and {Yn} are sequences
in R, we may form the sequences {xn + Yn}, {xn - Yn}, {XnYn}, and {rxn},
for any r e R. Furthermore, if Yn =F O for ali n e N, we may also form the
sequence {xn/Yn} Our next proposition describes the manner in which the
limits of these newly constructed sequences are related to the limits of the
original ones.
Proposition 5.1.11
1.
If {xn} and {Yn} are convergent sequences in R, then
For ali reR, the sequence {rxn} converges and

Lim rxn =r. Lim Xn
n-+oo
2.
The sequences {xn
+ Yn}
n-+oo
and {xn - Yn} converge and
Lim [xn +Yn] = Lim Xn

n-+ oo
n-+ oo
+n-+OO
Lim Yn
Lim [xn - Yn] = Lim Xn - Lim Yn

n-+ oo
n-+ oo
n-+ oo
3.
The sequence {xnYn} converges and
n-+OO
4.
n-+oo
n-+OO
If Yn =I= O for ali n eN and Limn-+ 00 Yn =FO, the sequence {xn/Yn} converges and
00 Xn
. [Xn]
Limn-+
L im
- =
--Limn-+oo Yn
n-+ oo Yn
Proof:
We shall prove ( 1) and part of ( 2), lea ving the rest as an exercise.
114
Let reR. We know that the function f defined by f(t)

teR, is continuous on R, and hence by Proposition 5.1.7,
Lim Xn =X
implies that
n--+oo
= rt, for all
Lim f(xn) = f(x)

n--+OO
But f(xn) = rxn and f(x) = rx, so

Lim Xn=X
implies that
n--+oo
Lim rxn =rx

n--+
oo
and we have established that (1) holds.

Now suppose that {xn} converges to x and {Yn} converges to y. We
wish to show that {xn + Yn} converges to x + y. lt will suffice to show that
if (a, b) is an open interval containing x + y, then there is somem eN such
that (a, b) contains Xn + Yn for ali n ;;:;: m.
Now, it is always possible to choose real numbers e., d., c2 , d2 , such
that xe(c 1,d1), ye(c2 , d2), and
Since {xn} converges to x, there is some k 1 eN such that

for ali n;;:;: k 1
Similarly, there is some k2 eN such that
If m = max {k., k2 }, then n eN, n;;:;: m implies that
But since (c 1 + c2 , d 1 + d2) e (a, b), this shows that Xn

of (a, b) for ali n;;:;: m, and we are done. O
+ Yn is an element
ln a practical sense, the definition of a convergent sequence is not very

useful for deciding whether or not a particular sequence converges, for in
order to apply the definition, we must know ( or guess) beforehand what
its limit is. Thus, in using the definition to show that a sequence {xn}
converges, we must somehow decide that x is a limit of the sequence and
then verify that this is so by showing that for every open set G which
Sequences
115
contains,x there is some meNsuch that xneG for ali n ?;;m. It would be
convenient to have a characterization of convergent sequences which
would allow us to examine a particular sequence and determine whether
or not it is convergent without having to predetermine its limit, if any.
Such a characterization exists for sequences of real numbers, and we will
develop it in the next section.
EXERCISES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Let X be an indiscrete topological space. Prove that any sequence in

X converges to every point of X.
Let X be a discrete topological space. Prove that a sequence {xn} in
X converges if and only if there is somem eN such that Xn = Xm for
ali n ?;:; m.
Is the converse of Proposition 5.1.4 true? That is, if X is a topological
space which is not Hausdorff, does it follow that there must be a
sequence in X which converges to more than one point? Justify your
answer.
Let xeR and let Xn = x + 1/n, for ali neN. Prove that {xn} con-.
verges to x.
Prove that every real number is the limit of a sequence of irrational
numbers.
Let R have the open ray topology. Show that the sequence {l/n}
converges to x for ali x ~ O.
Let X be a topological space. Show that if {xn} is a sequence in X
which converges to x and xn =F x for any n e N, then x is a limi t point
of the set of points {xn 1 n eN}.
Is the converse of Exercise 7 true? That is, if a sequence in a topological space has a limit point which is not a point of the sequence,
then does the sequence converge to the limit point? Justify your
answer.
Let Sbe a nonempty subset of R. Prove that xeSif and only ifthere
is a sequence of points of S which converges to x.
Letf: (O, 1) -+R be given by f(x) =O if x is irrational,f(x) = 1/q if
x = p /q where p, q are integers having no common factor. Show that
f is discontinuous at every rational point and continuous at every
irrational point of (O, 1).
Prove that if {xn} and {Yn} are convergent sequences in R with
Xn ~ Yn for ali n EN, then Limn .... 00 Xn ~ Limn .... 00 Yn
116
12.

Let {xn} be a convergent sequence in R. Prove that
Lim
oo
Xn =X
n-+
if and only if Lim [xn -x] =O

n-+ oo
13.
14.
Finish the proof of Proposition 5.1.11.

Let the Euclidean plane R 2 have the product topology. Prove that a
sequence {(xm Yn)} in R 2 converges to (x, y) in R 2 if and only if
Limn-+oo Xn =X and Limn-+OO Yn = y in R.
5.2
REAL SEQUENCES
ln this section we characterize the convergent sequences in the system of

real numbers. Our procedure will be as follows: we will show how to
construct from every real sequence two associated sequences of a special
kind, and will then prove that the original sequence converges if and only
if both of these associated sequences converge to the sarne limit.
A sequence {xn} of real numbers is a function from N to R whose
image is the set ofreal numbers {xn 1 neN}. This subset of R may or may
not be bounded.
Let {xn} be a sequence of real numbers. If its image
{xn 1 neN} is bounded above in R, we say that the sequence is bounded
above, and refer to the least upper bound of the image as the least upper
bound of the sequence. Similarly, the sequence is bounded below if its image
is bounded below, and the greatest lower bound of the sequence is the
greatest lower bound of the image. A sequence of real numbers whose
image is a bounded subset of R is said to be a bounded sequence.
Definition 5.2.1
If a sequence is convergent, we know that its terms must get closer

and closer to its limit. But for a sequence of real numbers, this means that
the absolute value of the terms cannot grow without bound; in other
words, a convergent real sequence must be a bounded sequence.
Proposition 5.2.2
Every convergent sequence of real numbers is a
bounded sequence.
Proof: Let {xn} be a sequence of real numbers which converges to the

real number x; then there exists me N such that xn e (x - 1, x + 1) for ali
n ~m. Let
a
= min {X 1, ... , Xm _ ., X
1}
117
Real Sequences
and
b
= max {x., ... , Xm-1> x + l}
Since a ~ Xn ~ b for ali n EN, the image {Xn 1n EN} of the sequence is a
bounded subset of R. D
Proposition 5.2.2 tells us that a convergent sequence in R must be
bounded, but of course the converse is not true: a bounded sequence need
not be convergent. (For example, the sequence {( - l)n} is bounded but
not convergent in R.) However, the converse is true for the class of
sequences known as monotonic sequences.
Definition 5.2.3 A sequence {xn} in R is monotonically increasing if
Xn ~ Xn + 1 for ali n EN, and monotonically decreasing if Xn ~ Xn + 1 for ali
n EN. A sequence is referred to as monotonic if it is either monotonically
increasing or monotonically decreasing.
Example 5.2.4 The sequence {2n} is monotonically increasing, the sequence {l/n} is monotonically decreasing, and the constant sequence {l}
is both monotonically increasing and monotonically decreasing; therefore
these are ali monotonic sequences. On the other hand, the sequence
{( - l)n} is neither monotonically increasing nor monotonically decreasing, and hence is not monotonic.
Proposition 5.2.5 Every monotonically increasing sequence in R which is
bounded above converges to its least upper bound and every monotonically decreasing sequence in R which is bounded below converges to its
greatest lower bound.
Proof: Let {xn} be a monotonically increasing sequence which is
bounded above, and let x = sup {xn 1n EN}; then x is either an element or
a limit point of the set {xn 1n EN}. We will show that in either case the
sequence converges to x.
Suppose x is an element of {xn 1 n EN}, so that x = xm for somem EN.
Since x is an upper bound for the set and the sequence is monotonically
increasing, it follows that xn = x for ali n ~ m. But this surely implies that
the sequence converges to x.
Now suppose that x is a limit poiilt of {xn 1 n EN}, and let (a, b) be an
open interval which contains x. Because x is a limit point, (a, b) contains
some element xm of the set, and since x is an upper bound for the
118
monotonically increasing sequence, it follows that Xm ~ Xn ~ x for all

n ~ m, and hence that Xn E (a, b) for all n ~ m. Therefore the sequence
converges to x in this case also. We have thus shown that every monotonically increasing sequence which is bounded above converges to its least
upper bound, and the proof for monotonically decreasing sequences is
similar. O
Corollary 5.2.6 A monotonic sequence in R converges if and only if it is

bounded.
The preceding corollary completely characterizes convergent
monotonic sequences in R. Our goal is a similar characterization for ali
convergent real sequences. We shall accomplish this by making use of
our characterization of convergent monotonic sequences. We will first
show that from any bounded sequence we can construct two bounded
monotonic sequences, then prove that the original sequence converges
if and only if the two monotonic sequences both converge to the sarne
limit.
Before beginning the program outlined above, we introduce some
convenient notation for divergent monotonic sequences. If a monotonically increasing sequence is not bounded above, it diverges because its
terms grow without bound, in which case we say that the sequence diverges to plus infinity. Similarly, a monotonically decreasing sequence
which is not bounded below is said to diverge to minus infinity.
Definition 5.2.7 If {xn} is a monotonically increasing sequence which is

unbounded above, we write
Lim
n-+
oo
Xn =
+oo
and say that the sequence diverges to plus infinity. If {xn} is a monotonically decreasing sequence which is unbounded below, we write
Lim
Xn = -00
n-+oo
and say that the sequence diverges to minus infinity.

Next we show that it is possible to construct from every bounded real
sequence a monotonically increasing sequence and a monotonically decreasing sequence.
Real Sequences
Proposition 5.2.8
1.
119
Let {xn} be a sequence in R.
If { xn} is bounded above, for each k e N let

sk
= sup {xn
1n
~ k}
The sequence {sd is monotonically decreasing and it converges if

{xn} is bounded below.
2. If { Xn} is bounded below, for each k e N let
The sequence {td is monotonically increasing and it converges if {xn}

is bounded above.
Proof: We prove (1), leaving (2) as an exercise. (See Exercise 5 at the
If the sequence {xn} is bounded above, then certainly for each k eN
the set {xn 1n ~ k} is bounded above, and hence sk exists for each k e N.
Since
we have
for ali k e N, and thus {sd is a monotonically decreasing sequence. If { xn}

is bounded below, then sois {sk}, and hence being a bounded monotonically decreasing sequence, it must converge to its greatest lower bound. O
Examples 5.2.9 1. Let {xn} = { l, 4, t.... }. This sequence is bounded
above and below, so we may form both of the sequences {sk} and {td:
si= sup {l, 4, !, , ... } = 1

s2 = sup {4,
S3
!, ... } = 4
1 1
}
1
sup { 3,
4, = 3
4, !, ... } = O
t2 = inf {4, !, ... } = O
t 1 = inf {1,
t3 =
inf {t, !, ... } = O
and so on. Therefore {sd = {l/k}, which is a monotonically decreasing

sequence which converges to O, and {tk} ={O}, which is a monotonically
increasing sequence which converges to O. Note that both {sd and {tk}
120
converge to the sarne limit and that the original sequence {xn} also converges to this limit.
2. Let {xn} = {l, -1, 4, -2, -3, ... }. This sequence is bounded
above, so the sequence {sd exists, and we have
t.
{ sk} = { l,
4, 4, !, !, }
Note that Limk--.oo sk =O. Since {xn} is not bounded below, the sequence
{ tk} does not exist.
Now we are ready to define the concepts of the limit superior and
limit inferior of a real sequence. These concepts will be useful in our
characterization of convergent sequences in R.
Definition 5.2.10 Let {xn} be a sequence in R. Define the limit superior of

{xn}, denoted by Lim sup Xm and the limit inferior of {xn}, denoted by
Lim inf Xm as follows:
1. If {xn} is bounded above, let {sd be the monotonically decreasing
sequence defined in Proposition 5.2.8 (1) and set
Lim sup Xn = Lim sk
k--+oo
2.
If {xn} is not bounded above, set Lim sup Xn = + oo.

If {xn} is bounded below, let {td be the monotonically increasing
sequence of Proposition 5.2.8 (2) and set
Lim inf Xn = Lim
tk
k--+00
If {xn} is not bounded below, set Lim inf xn = - oo.
Examples 5.2.11 1. Let {xn} = {l/n}. This sequence is bounded above

and below, and by virtue of Example 1 of Examples 5.2.9,
Lim sup Xn = Lim sk = Lim
k-+OO
Lim inf Xn = Lim
k-+OO
1
-k
=O
k-+OO
tk
= Lim O= O
k-+oo
Note that Lim sup Xn = Lim inf Xn =O and that {xn} also converges to O.
Real Sequences
121
2. Let {xn} = {( - l)n}. This sequence is bounded and

sk
= sup {( - 1)n 1 n ~ k} = 1
tk
= inf {( - l)n 1n ~ k} = -1
Thus
= Lim 1 = 1
Lim sup Xn = Lim sk

k-+OO
k-+OO
and
Lim inf Xn = Lim
tk
= Lim - 1 = - 1
k-+oo
k-+oo
Note that Lim sup Xn =F Lim inf Xn and that the sequence {xn} diverges.
3. Let {xn} = {l, -1, !, -2, t, -3, ... }. By virtue of Example 2 of
Examples 5.2.9,
Lim sup Xn = Lim sk = O
k-+00
while, since {xn} is not bounded below,

Lim inf Xn
=-
00
4. Consider the sequence {xn} = {( - l)nn }. This sequence is unbounded above and unbounded below, so
Lim sup Xn =
+ oo
and
Lim inf Xn = -
00
5. Consider the sequence {xn} = {n }. This sequence is unbounded

above, hence
Lim sup Xn =
+ oo
It is bounded below, however, and since tk = k, for ali keN, the monotonically increasing sequence {td diverges to + oo. Therefore we have
Lim inf Xn = Lim

k-+
tk
+ oo
00
Similarly, the sequence {Yn} = { -n} has

Limsupyn = -oo
and
Lim inf Yn = -oo
122
We will soon prove that a seque.nce of real numbers converges if and

only if its limit superior and limit inferior are finite and equal. ln general,
the limit superior and limit inferior of a sequence may be thought of
as measuring the "degree of divergence" of a sequence. For instance, if
the limit superior and limit inferior are both infinite, as in Examples 4
and 5 of Examples 5.2.11, then no part of the sequence converges, and it
is very badly divergent. If one of the limit superior or limit inferior is
finite while the other is infinite, as in Example 3 above, then the sequence
is less badly divergent, for at least "part" of it converges. If the limit
superior and limit inferior are finite but unequal, as in Example 2, then
the sequence is "almost" convergent, for its "parts" converge even
though it does not converge as a whole. Finally, if the limit superior and
limit inferior are finite and equal, as in Example 1, then the sequence
converges.
lt is easy to show that Lim inf Xn ~ Lim sup Xn for any sequence
{xn} in R. We outline a proof of this fact, leaving the details as an
exercise. (See Exercise 6 at the end of this section.)
Proposition 5.2.12 If {xn} is a sequence in R, then

Lim inf Xn
Lim SUp Xn
Proof: Suppose first that {xn} is bounded above and l:...:low. It is easy
to check that
for ali ieN,jeN, and from this it follows that

Lim inf Xn = Lim
t; =
sup {t;} ~ inf {sJ = Lim = Lim sup xn
k-+CO
j-+CO
On the other hand, if {xn} is not bounded above, then

Limsupxn = +oo
so surely
Lim inf Xn
Lim sup Xn
in this case, and similarly if {xn} is not bounded below. O
Real Sequences
123
Now we are ready to prove the theorem which characterizes convergent sequences in R. The theorem not only tells us exactly which real
sequences are convergent but also what the limit of a convergent sequence
must be.
Theorem 5.2.13
If { Xn} is a sequence in R, then {Xn} converges to x E R
if and only if
Lim sup Xn
= Lim inf Xn = X
Proof: If {xn} converges to x, then {xn} must be a bounded sequence,

and therefore the monotonic sequences {sd and {td of Proposition 5.2.8
exist and are convergent. Hence
Lim sup Xn = Lim sk = inf {sk} = s

k--+oo
and
Lim inf Xn = Lim
k--+
tk =
sup {tk} =
00
both existas finite real numbers. Note that for ali keN,
and
t ;;:;: t1r
If Limn .... 00 Xn = x < s, then there exists an open interval (a, b) such that
xe(a, b) and b <s. But then there is some meNsuch that xne(a, b) for
ali n ~ m, and therefore
sm
sup {Xn 1 n ;;:;: m} ~ b < s
which contradicts the fact that s is a lower bound for the sequence {sd.
Thus we must have x ;;:;: s. A similar argument shows that x ~ t, and hence
But t ~ s by Proposition 5.2.12. Therefore x = s = t.

Now we prove the converse. Suppose that {xn} is a real sequence such
that Lim sup Xn = Lim inf Xn = x, where x is a finite real number. We
must show that {xn} converges to x. Let (a, b) be an open interval which
124
contains x. Since
x
= Lim sup Xn = Lim
sk
k~oo
there exists m 1 eN such that sk E (a, b) for ali k ~ m 1 Similarly, because

x = Lim inf xm there exists m2 eN such that tk E (a, b) for ali k ~ m2 If
m = max {m\' m2} then for ali k ~ m, we have sk E (a, b) and tk E (a, b). But
sk = sup {Xn 1 n ~ k} implies that sk ~ xk for all k EN, and similarly
tk ~ xk for all k eN. Hence tk ~ xk ~ sk for all k ~ m, and it follows
that xk E (a, b) for all k ~ m. Therefore {xn} converges to x, and we are
done. D
Theorem 5.2.13 shows that the question of convergence for an arbitrary real sequence can be reduced to a question of convergence for two
associated monotonic sequences: the original sequence converges if and
only if the associated monotonic sequences both converge to the sarne
limit. Since convergence for monotonic sequences is merely a question of
the boundedness of their images as subsets of the reais, this is indeed a
result of considerable power.
EXERCISES
1.
For each of the following real sequences, prove convergence or divergence. If the sequence converges, find its limit.
{n: 1}
(b) {n:1}
(d)
(a)
(e)
3.
{2n-:~2}
Let reR, r ~O. Under what circumstances does the sequence {rn}
converge, and what is its limit? Prove your answers.
Let {Xn} be a real sequence. Show that
Lim
k~oo
4.
(f) {r/n }, any reR
(c) {n 2 -n}
2.
{2nn ++ l}
Xn
=O
if and only if Lim
k~oo
lxn 1 = O
Let {xn} be a real sequence which converges to the real number x.

Prove that for any m eN,
(a) The sequence {x:;'} converges to xm;
(b) If it is defined, the sequence {zy'Xn} converges to zy'X.
Real Sequences
5.
6.
7.
125
Prove (2) of Proposition 5.2.8.

Prove Proposition ,5.2.12.
For each of the following sequences, find Lim sup xn and
Lim inf Xm state whether the sequence converges or diverges, and
find its limit if it converges.
(a) {n(n - 1)}

(b)
t;: 1}
(e) {l, 1,2, 1, 2, 3, 1, 2, 3, 4, ... }

(f) {O, -1,0, -2,0, -3,. .. }
(c)
t 2n~ 1}
(g) {2, O, -2, O, 2, O, -2, O, ... }
(d) {l, 2, 3, 4, 1, 2, 3, 4, ... }

8.
9.
10.
11.
Let {xn} be a real sequence. A sequence {Yn} is said to be a subsequence of {xn} if both of the following conditions hold:
(i) For every m eN, there is some n eN such that Ym = xn;
(ii) If Ym = Xn and Ym+ 1 = xk, then n < k.
Prove that if {xn} has a subsequence which is bounded below by b
then Lim sup Xn ~ b, while if {xn} has no subsequence which is
bounded below, then Lim sup xn = - oo. State and prove a similar
result for Lim inf Xn.
Let {xn} be a bounded real sequence. Let Lim sup Xn = s and
Lim inf xn = t, and let e be any positive real number. Prove:
(a) There are infinitely many values of n for which xn > s - e;
(b) There are infinitely many values of n for which xn < t +e;
(c) Xn < s +e for all but finitely many values of n;
(d) xn > t - e for all but finitely many values of n.
Prove that every bounded real sequence has a convergent subsequence by showing that if {xn} is such a sequence, then it has a
subsequence which converges to Lim sup xn. State and prove a similar result for Lim inf Xn.
Let {xn} and {Yn} be bounded real sequences, with Xn ~Yn for ali
n eN. Show that
Lim sup Xn
Lim sup Yn
Lim inf Xn
Lim inf Yn
and
12.
Let {xn} and {Yn} be bounded real sequences, and consider the
sequence {xn + Yn}
126
(a) Prove that

Lim sup [xn
+ Ynl
Lim sup Xn
+ Lim sup Yn
Lim inf[xn
+ Ynl
Lim inf Xn
+ Lim inf Yn
and
(b) Give examples to show that equality need not hold in (a).
5.3
INFINITE SERIES
ln this section we consider the special kind of real sequence known as an

infinite series. We begin with the definition of an infinite series.
Definition 5.3.1
Let {xn} be a sequence in R. For each m eN let

m
sm =Xi
+ X2 + ... + Xm = L
xk
k-1
The real number Sm is the mth partia/ sum of the sequence {xn} The
sequence
of m th partia! sums of {Xn} is called the infinite series with n th term Xn and
is denoted by
Note that an infinite series ~:= 1 xn is not an "infinite sum." The

notation ~:= 1 Xn merely stands for a special sequence, namely the sequence of m th partial sums of the sequence {xn}.
Examples 5.3.2
where
l. The infinite series~:= 1 (
l)n is the sequence {Sm},
Clearly, Sm = - l if m is an odd natural number and Sm = O if m is
Infinite Series
127
an even natural number. Therefore

- 1, o, ... }.
2. The infinite series
00
1
{ - 1, o,
~::"= 1 ( -
l)n is the sequence
I-
n= 1 2n
is the sequence of partia! sums {~;;' = 1 ( 1/2k)}. But

m
L - = 1 -2mk= 1 2k
so ~::"= 1 (1 /2n) is the sequence {1 - (1 /2m)}
a. ....}.
~. ~
Since an infinite series is justa special kind of sequence, we can speak

of its convergence or divergence.
Definition 5.3.3 The infinite series ~::"= 1 xn is said to converge if its sequence of partia! sums {~;;' = 1 xk} converges, and to diverge if its sequence
ofpartial sums diverges. Ifthe sequence ofpartial sums converges to xeR,
we say that the infinite series converges to x. Although the series ~::"= 1 xn
is a sequence, not a number, it is traditional to write
00
L Xn=X
n= 1
to indicate that the series converges to x.

Examples 5.3.4 1. The infinite series~::"= 1 ( - l)n of Example 1, Examples 5.3.2, diverges because its sequence of partia! sums is
{-1, O, -1, O, ... }, which is clearly divergent.
2. The series~::"= 1 ( l/2n) of Example 2, Examples 5.3.2, converges to
1 because its sequence of partial sums is {l -(l/2m)}, which converges
to 1.
3. The infinite series~::"= 1 ( l/n) is known as the harmonic series. We
will show that the harmonic series diverges. If {Sm} is the sequence of
partia! sums of the harmonic series, we have
S2 =I+t>I
S4 = S2 + t + ~ > S2 + 2(~) = S2 + t
S8 = S4 + ! + ~ + t + i > S4 + 4(!) = S4 + t
S 16 = S8 + ~ + + i<; > S8 + 8(1<;) = S 8 + t
128
and so on. Thus the sequence of partia! sums is clearly unbounded, and
hence divergent, and therefore the harmonic series diverges.
Now we prove some results about infinite series. Our first result will
completely characterize the type of series known as a geometric series.
Definition 5.3.5
A series of the form

00
L arn = a + ar + ar + ar +
n=O
2
is called a geometric series with ratio r.
lrl
lrl
Proposition 5.3.6 If
< 1, the geometric series
a/( 1 - r). If a =I O and
~ 1, the series diverges.
Proof:
:r.;:o= 0 arn
converges to
It is easy to show that if r =I 1 the m th partia! sum of the geomet-
ric series is
S = a(l - rm)
m
1-r
and the proposition follows from this fact and the examination of a few
special cases. The details are left to the reader. (See Exercise 1 at the end
of this section.) D
The geometric series are virtually the only important series for which
it is possible to determine at a glance whether or not a particular series is
convergent. Other types of series generally require more sophisticated
tests to determine whether or not they converge, and there are many such
tests, some of them quite specialized. ln the remainder of this section, we
presenta few ofthe more general convergence tests; others are given in the
exercises at the end of the section.
Our first test is one which is freque'ltly useful in establishing the
divergence of a series.
Proposition 5.3.7
If the series
:r.;:o= 1 xn converges,
then Limn_, 00 xn =O.
Proof: If the series :r.;:o= 1 xn converges, then its sequence of partia! sums
{Sm} = {S 1, S2 , S3 , } converges to some real number x. But then the
sequence {Tm} ={O, S 1, S 2 , } also converges to x. (Why?) Since
129
Infinite Series
Xm = sm - Tm for ali meN, we have
Lim Xm = Lim [Sm - Tm] = Lim Sm - Lim Tm = x - x = 0. D
m-+oo
m-+oo
m-+OO
m-+OO
Proposition 5.3.7 provides a divergence test for series because it implies that if Limn .... oo Xn #O, then the series ~:'= 1 Xn must diverge. Note
that the converse of the proposition is not true: a series ~n = 1 Xn may
diverge even though Limn .... oo Xn =O, as the example of the harmonic
series shows.
Example 5.3.8 The
series~:'= 1 (n
+ l)/n diverges, since
Lim n + 1 =1
n-+ 00
The next result may also be thought of as providing a test for divergence, though not a particularly easy one to apply. It is included here
because it is needed to prove the more useful comparison test.
Proposion 5.3.9 (Boundedness Test) If ~:'= 1 Xn is an infinite series

with xn ~O for ali neN, then the series converges if and only if its sequence of partia! sums is bounded above.
Proof: The infinite series converges if and only if its sequence of partia!
sums converges. But since xn ~ Ofor ali n e N, the sequence of partia! sums
is monotonically increasing and therefore converges if and only if it is
bounded above. D
The next result, known as the comparison test, is perhaps the most
useful of ali convergence tests.
Proposition 5.3.10 (Comparison Test) If ~:'= 1 Xn and

finite series with O~ Xn ~ Yn for ali n eN, then
~:'= 1 Yn
are in-
1. If ~:'= 1 y n converges, so does ~:'= 1 Xn;

2. If ~:'= 1 Xn diverges, so does ~:'= 1 Yn
Proof: If ~:'= 1 Yn converges, then by Proposition 5.3.9, the sequence of
partial sums {~k' = 1 Yk} is bounded above. But O~ Xn ~ Yn for ali n e N
then implies that the sequence of partia! sums {~k' = 1 xk} is bounded
above, and hence the series~:'= 1 xn must converge. This proves ( 1), and
the proof of (2) is similar. (See Exercise 2 at the end of this section.) D
130
Examples 5.3.11
1. Let p be a real number and consider the p-series

~ l/nP when p < l and since I::'= 1 (l/n) diverges, we see that when p < l, the p-series diverges by comparison with
the harmonic series.
2. Since O< I/nn ~ I/2n for all n ~ 2 and since I::'= 2 ( l/2n) converges
(why?), the series I::'= 2 ( l/nn) converges by comparison. Therefore the
series I::'= 1 ( l/nn) = l + I::'= 2 ( I/nn) also converges. (Why?)
I::'= 1 (l/nP). Since O< I/n
The final convergence test we shall present is the well-known ratio

test.
Let I::'= 1 xn be a series ofpositive real
Proposition 5.3.12 (Ratio Test)

numbers.
l.
n-+co
2.
I
> l, then I
lf Lim Xn + 1 < l, then

Xn
If Lim Xn +
n-+oo
Xn
xn converges;
n=l
Xn diverges.
n=1
Proof: 1. lf Limn-+co (xn+ ifxn) = L < l, let r be a positive real number

such that L < r < 1. Since there is somem EN such that (xn + ifxn) < r for
ali n ~ m, we have
and so on. Therefore, by induction

for ali keN
But the infinite series
is a geometric series with ratio

series
r, and since lrl < l, it converges. Thus the

co
k=I
Xm+k
Infinite Series
131
converges (by comparison wth the geometric series). But if ~k= 1 Xm+k
converges, so does ~;::'= 1 xn. (See Exercise 3 at the end of this section.)
2. If Limn--+oo (xn+ ifxn) > l, then there is some meN such that
(xn + if xn) > l for ali n ~ m, and thus
Xm < Xm + 1 < Xm + 2 <
Since ~;::'= 1 xn is a series of positive real numbers, this shows that

Limn--+oo Xn #O, and hence the series diverges by Proposition 5.3.7. D
Note that if Limn .... 00 (xn + ifxn) = l, then the ratio test says nothing
about convergence or divergence: in such a case the series may either
converge or diverge.
Examples 5.3.13
beca use
l. The series
~;::'= 1 (
l/n!) converges by the ratio test
Lim Xn + 1 = Lim 1/(n + 1) ! = Lim - 1- = O

n--+OO Xn
n--+OO
l/n!
n--+OO n + l
2. The series
~;::'= 1
(2n/n) diverges by the ratio test because
Xn+t
L" 2n+t/(n + l) L" 2 n
2
Li. m
-- =
im
= im - - =
n--+OO Xn
n--+OO
2n;n
n--+OO n + l
EXERCISES
1.
2.
3.
4.

Prove (2) of Proposition 5.3.10.
Prove that a series ~;::'= 1 xn converges if there is some me N such that
~k= 1 Xm + k converges.
For each of the following series, prove convergence or divergence.
(a)
00
(-Itn+l
n=I
(b)
n~I ]n
(c)
I 1
n= 1 n.
<d)
3( 5) 1 -
n=I
oo n3
2n
(e)
n~t
(f)
n~I l +~n-1
oo n3
132
5.
~::"= 1
(a) Show that every series of the form
(an/10n), where
for all neN
ane{O, 1, 2, ... , 9}
converges to some xe [O, l].

(b) Show that if x e [O, l] has the decimal representation
then
6.
(a) Prove that if ~::"= 1 Xn = x and
~::"=
Yn = y, then
00
(Xn
n=I
+ Yn) = X + Y
(b) Prove that if ~::"= 1 xn = x and ris any real number, then
00
n=I
7.
rxn = rx
(a)
Generalize Exercise 3 above by showing that if one infinite

series is created by eliminating finitely many terms from another one, then either both series converge or both diverge.
(b) Show that the hypotheses ofthe comparison test may be weakened by replacing the condition
for all neN
with
for ali but finitely many indices n
8. Show that if ~::"= 1 lxnl converges, then ~::"= 1 Xn converges.

9. The ratio test may be generalized as follows:
Let ~::"= 1 xn be a series of nonzero real numbers.
( i) If Lim sup lxn +
Xn
( ii) If Lim inf lxn +
Xn
Prove this.
I Xn converges;
> 1, then I Xn diverges.
n=
11
11
< 1, then
n= 1
1
lnfinite Series
133
10. The root test for convergence of a nonnegative series may be stated
as follows:
Let I::'= 1 xn be a series of nonnegative real numbers.
00
( i) If Lim
z(i;. < 1, then L
Xn
converges;
Xn
diverges.
n= t
n-+oo
00
(ii) If Lim
z(i;. > 1, then L
n= t
n-+oo
(a) Apply the root test to the following series:

1
00
L-;;
n~I
oo
n=ln
11.
(n+l)n
2n+1
(b) Prove the root test.

Considera series of the form I::'= 1 ( - l)n+ 1xm where Xn is positive
for ali n eN. Such a series is called an alternating series. The following is a convergence test for alternating series:
The alternating series I::'= 1 ( - 1Y + 1xn converges if
(i) Xn ~ Xn+ 1 for ali n eN, and
(ii) Lim
Xn
=O.
R-+00
(a) Apply the test to the following series:
I. (- l)n +
n= 1
1n
+1
n
(-l)n+I lnn
n+1
n=I
(ln n is the natural logarithm, or logarithm to the base e, of n.)

(b) Prove the test.
12. The integral test for the convergence of positive series may be stated
as follows: Let I::'= 1 xn be a series of positive real numbers, and let
/: [l, +oo)-+[O, +oo) be a continuous function such thatf(n)=xn
for ali n eN andf(x) ~f(y) if x ~ y. The series I::'= 1 xn converges if
and only if the improper integral Jt 00 f(x) dx exists.
(a) Apply the integral test to the following series:
00
L -nP (p any real number)
n= 1
(b) Prove the integral test.
I.-1n(ln n)
n=2
134
5.4
FUNCTIONAL LIMITS
Letfbe a function from R to R. The reader has undoubtedly encountered

expressions of the form Limx-+c f(x) = x 0 An expression such as this,
which is known as afunctional limit, is meant to convey the information
that the functional values f(x) approach x 0 as x approaches e. ln this
section we present a brief discussion of functional limits. We begin with
the definition of such a limit, show how questions about the limit of a
functionf can be reduced to questions about the limits of sequences of the
form {f(xn) }, and conclude with a characterization of continuous functions on R which utilizes functional limits.
Our definition of functional limit is given in terms of the natural
topology on R.
Definition 5.4.1 Let S be a subset of R, f: S-+ R, and let e be a limit point

of S. The real number x 0 is the limit of f as x approaches e, written
Pmf(x)
x-+c
=X0
if for every open set G containing x 0 there exists an open set H containing
e such thatf(x) eG for ali xeS n H - {e}.
Suppose that Limx-+cf(x) = x 0 The definition says that if we surround the point x 0 with an open set G, then there must be an open set H
about e such thatf carries the elements of S n H (with the possible exception of e, if e eS) into the set G. Therefore by choosing G to be smaller and
smaller, we can force the functional values ofjto approach x 0 more and
more closely. Note that the value of f at e is immaterial, for we never
considerf(c); indeed,fneed not even be defined ate, for we do not require
that e be an element of the domain S off, but only that it be a limit point
of S. (It is necessary to require that e be a limit point of S in order to
ensure that the open set H contains some point ofthe domain S.) We also
remark that use of the phrase "the limit of f as x approaches e" in the
definition is justified, for the limit of f as x approaches e is unique if it
exists. As usual, this depends on the fact that R is a Hausdorff space, and
we leave the proofto the reader. (See Exercise l at the end ofthis section.)
Examples 5.4.2
l. Let f: R
-+
f(x) =
R be given by
{~
if x = l
if X # l
135
Functional Limits
We claim that Limx .... t f(x) = 1. To prove this, we must show that if G is
an open set which contains 1, then there exists an open set H which
contains 1 and such thatf(x) eG for all xeH - { 1}. But sincef(x) = x for
all x =F 1, we may take H = G. Note that Limx .... tf(x) = 1 even though
f(l) =I= 1.
2. Let f: (O, 1) -+R be given by f(x) = 2x + 1 for all xe(O, 1). We
claim that Limx_,J(x) = 2c + 1 for all ce[O, l]. To see this, let G be an
open set containing 2c + 1. There is some open interval (a, b) contained in
G such that 2c + 1 e(a, b). If H =((a - 1)/2, (b - 1)/2), then H is open,
ceH, andf(x)e (a, b) e G for all xe(O, 1) nH - {e}. Note that if e= O
ore= 1, thenf(c) is not defined, but Limx_,J(x) exists nonetheless.
3. Let f: R -+ R be given by
f(x) = { _
ifx >O
if X ~0
If e> O, then Limx_,J(x) = 1, for if G is an open set containing l, we

may choose H =(a, b), where O< a< e< b. A similar argument shows
that if e< O, then Limx_,J(x) = -1. We claim, however, ' that
Limx .... of(x) does not exist. To see this, suppose Limx .... of(x) = x 0 does
exist. If G is an open set containing x 0 , there must exist an open set H
containing O such thatf(x)eG for ali xeH - {O}. Choose G to be any
open set which does not contain both 1 and - 1. Since an open set H
which contains O must contain both a positive number x 1 and a negative
number x 2 , and sincef(x 1) = 1 andf(x2 ) = -1, it is not possible to have
f(x)eG for all xeH - {O}, and this establishes the claim.
Now, as promised, we demonstrate the connection between the limit
of a functionf as x approaches e and the limits of certain sequences of the
form {f(xn)}.
Proposition 5.4.3 Let S be a subset of R, f: S-+ R, and let e be a limit
point of S; then Limx_,J(x) = x 0 if and only if Limn_, 00 f(xn) = x 0 for
every sequence {Xn} in S which converges to e and for which xn =F e for ali
neN.
Proof: Suppose Limx_,J(x) = x 0 and let {xn} be a sequence ofpoints of
S which converges to e and such that xn =F e for all n eN. If G is an open
set containing x 0 , there is an open set H containing e such thatf(x) eG for
ali x e S n H - {e}. Since Limn .... 00 Xn = e, there is some m e N such that
Xn eH for all n ~ m. Since Xn =F e for all n eN and since {xn} is a sequence
136
of points of S, it follows that xn e S n H - {e} for ali n ;;:;: m. But then

f(xn) eG for ali n ;;:;: m, and this shows that the sequence {J(xn)} converges to x 0
To prove the converse, we show that if Limx .... cf(x) # x 0 , then there
is some sequence {xn} in S such that Xn #e for ali n eN and
Limn .... oo Xn =e, but Limn_, 00 f(xn) # x 0 We argue as follows: if
Limx .... cf(x) # x 0 , there exists and open set G containing x 0 such that
for every o pen set H containing e there is some point x e S n H - {e}
for which f(x) ~G. ln particular, this is so if we choose
H = Hn =(e - I/n, e+ l/n), n eN. Thus for every n eN there is some
point xneSnHn -{e} such thatf(xnHG. Clearly, {xn} is a sequence of
points of S which converges to e, Xn # e for ali n e N, and {J(xn)} cannot
converge to x 0 because f(xnH G for any n e N. D
Proposition 5.4.3 says that the study of functional limits in R reduces to the study of certain convergent sequences in the domain of the
function. Thus the concept of the limit of a function on R does not really
yield anything new, for fi1s merely sequences "in disguise": whatever can
be accomplished by using limi(s of functions can also be accomplished
using limits of sequences. Many of our earlier results concerning functions and sequences can be used to derive corresponding results concerning functional limits. As an example, we present a result which is
sometimes used as the definition of continuity on the real line.
Proposition 5.4.4 Let S be a subset of R and let ceS be a limit point of
S. A function f: S-+ R is continuous at e if and only if
Limx_,J(x) = f(c).
Proof: If f is continuous at e, then for every sequence {xn} in S which
converges to e, we must have {J(xn)} converging to f(c) by Proposition
5.1.7. But then Proposition 5.4.3 shows that Limx_,J(x) = f(c). Conversely, if Limx_,J(x) = f(c), then for every open set G containing f(c)
there is an open set H containing e such that f(x) eG for ali
xeSnH-{c}. But sincef(c)eG, we conclude that for every open set
G containing f(c) there is an open set H containing e such that f(x)
eG for ali xeSnH, and this is the definition of continuity of fat the
point e. D
We conclude this section with a statement of the familiar properties

of limits. The proof is left to the reader. (See Exercise 3 at the end of this
section.)
Functional Limits
137
Proposition 5.4.5 Let S be a subset of R, e be a limit point of S, and

/: S-+R, g: S-+R. IfLimx_,J(x) and Limx_,cg(x) exist, then
1.
2.
3.
4.
For all reR, Lim rf(x) exists and Lim rf(x) = r Limf(x);
X-+C
X-+C
X-+C
Lim (/ + g)(x) exists and Lim (/ + g)(x) = Limf(x) + Lim g(x);
X-+C
X-+C
X-+C
X-+C
Lim (fg)(x) exists and Lim (fg)(x) = [Limf(x)][Lim g(x)];
x-+c
x-+c
X-+C
X-+C
If Lim g(x) =F O, then Lim (f/g)(x) exists and
X-+C
X-+C
(L)
Lim
(x) = L~mx .... cf(x)
x-+c g
L1mx .... c g(x)
EXERCI SES
1.
2.
Let S be a subset of R, e be a limit point of S, andf: S--+ R. Prove that

if Limx_,J(x) exists it is unique.
For each of the following, find the indicated limit or show that it does
not exist. Justify your answers.
(a) Lim lxl, where e is any real number;
X-+C
(b) Lim xn, for neN ande any real number;
X-+C
(e) Lim [x], where meZ and [x] is the greatest integer less than or
X-+m
e9ual to x;
. x 2 +x + 1
(d) L 1m
;
X-+2
X- 1
(e) Lim x
X-+I
3.
2+
X
+1
x1
-
Prove Proposition 5.4.5
6
Metric Spaces
Throughout this text we have been using the real number system R as
both an example of a topological space and as a guide toward further
topological generalization. ln this latter connection, there are several important properties of R which we have not yet considered from a topological point of view, and foremost among these is the property that
between any two points of R there is a well-defined distance. (Recall that
the distance between the real numbers x and y is defined to be lx - yl,
the absolute value of their difference.) ln this chapter we will extend
the concept of distance between points to sets other than R by means
of functions known as metrics. We will see that if it is possible to define
a metric function on a set, then it is possible to topologize the set in
a very interesting and fruitful manner, thus creating a topological
space called a metric space. Metric spaces share many of the properties
of the real number system R and are of considerable importance in
analysis.
6.1
THE METRIC TOPOLOGY
Suppose x and y are real numbers. Then (see Corollary 2.1.4, Section 2.1
of Chapter 2)
139
140
1.
Metric Spaces
The distance between x and y is a nonnegative real number:
lx:-YI ~o
2.
The distance between x and y is O if and only if x = y:
lx-yl =0
3.
if and only if x
=y
The distance between x and y is the sarne as the distance between y

and x:
lx -yl = IY-xl
4.
Furthermore, if z is also a real number, then

The triangle inequality holds for x, y, and z:
lx - YI ~ lx - zl + IY - zl
These properties of distance in R motivate the following definition of a
distance function, or metric, on an arbitrary set M.
Definition 6.1.1
Mif
1.
2.
3.
4.
Let M be a set. A function p: M x M-+ R is a metric on
p((x,y)) ~O for ali xeM, yeM;

p((x, y)) =O if and only if x = y in M;
p((x,y)) =p((y, x)) for ali xeM, yeM;
(triangle inequality) p((x, y)) ~ p((x, z))
yeM, and zeM.
+ p((y, z))
for ali xeM,
ln order to simplify the notation, from now on we will write p(x, y)

rather than p((x, y)).
It should be clear that if pisa metric on M, then it defines a distance
between any two points of M: the distance between xeM and yeM, as
measured by p, is just the nonnegative real number p(x, y). Thus, the
smaller the real number p(x, y), the closer together the points x and y are
in M, as measured by p.
Let us consider some examples of metrics on sets.
Examples 6.1.2 1. Let p: R x R -+ R be given by p(x, y) = lx - y 1, for
ali x e R, y e R. Then p is a metric on R, called the absolute value metric or
natural metric on R.
The Metric Topology
141
2. Let M be any set, and define p: M x M

for ali xeM, yeM
p(x,y) =
-+ R
as follows:
if X# y
ifx =y
It is easy to check that the function p satisfies the conditions of Definition

6.1. l and is therefore a metric on M. This metric is called the discrete
metric on M.
3. Let M be any set, S a subset of M, and p: M x M-+ R a metric on
M; then the restriction of the function p to S x S is a metric on S. For
instance, the restriction of the absolute value metric on R to the closed
unit interval [O, l] is a metric on [O, l].
4. Define p: R 2 x R 2 -+ R as follows: for ali points (x1o y 1) and (x2 , yi)
in R 2 , let
The function pisa metric on the Euclidean plane R 2 . We leave the proof
of this fct as an exercise. (See Exercise 2 at the end of this section.) This
metric is known as the product metric on R 2 .
5. Let M = {bounded sequences in R}, and define p: M x M -+R as
follows: if {xn}eM and {Yn}eM, then
p( {xn}, {yn})
= sup
{lxn - Ynl 1 n eN}
Let us show that p is a metric on M.

First we must show that p( {xn}, {yn}) exists for ali {xn}eM and
{Yn}eM. However, since {xn} and {Yn} are bounded sequences in R, it
follows that the set {lxn - Yn 11 n e N} is a bounded subset of R ( why?) and
hence p({xn}, {Yn}) = sup {lxn -Ynl 1 neN} exists.
If {x }eM and {Yn}eM, then lxn - Ynl ~O for ali n eN, so surely
p( {xn}, {yn}) = sup {lxn - Ynl 1 neN} ~O

Furthermore, sup {lxn - Yn 11 n e N} = O if and only if Xn = Yn for ali n e N,
that is, if and only if {xn} = {Yn} in M. Thus the function p satisfies
conditions ( 1) and (2) of Definition 6.1.1. It satisfies condition (3) because
sup {lxn - Ynl 1 neN}
and thus
= sup {IYn - xnl 1 neN}
Metric Spaces
142
Now suppose that {zn} is also a bounded sequence of real numbers.

By the triangle inequality for absolute value, we have
for ali n e N, and it is easy to check that this in tum implies that
sup {lxn - Ynl 1 n eN} ~ sup {lxn - znl 1 n eN} + sup {IYn - znl 1 n eN}
Thus
Therefore p is indeed a metric on M. The set M of ali bounded sequences
in R is commonly denoted by / 00 , and the metric of this example is called
the supremum metric on / 00
If M is a set, p is a metric on M, and x is a point of M, it is natural
to consider the subset of M consisting of ali points which lie within a
specified distance of x.
Definition 6.1.3 Let M be a set and let p be a metric on M. If xeM and

e is a positive real number, then
1. The open bali of radius e about x is the set
P(x, e)
= {y e M 1 p(x, y)
< e}
2. The closed bali of radius e about x is the set
if(x, e)= {yeM 1p(x, y) ~e}

Thus the open bali p(x, e) consists of ali points of M which are at a
distance less than e from x, while the closed bali if(x, e) consists of ali
points of M which are at a distance of e or less from x. ln Figure 6.1, the
,,,,_- .....
/
f. ' \
I
\
...--,
I
_,,
'
......
~(X,f.)
Figure 6.1
G
~(X,f.)
Open and closed balls.
The Metric Topology
143
open bali consists of ali points of M which lie within the interior of the
circle, while the closed bali consists of the interior points together with
those which lie on the circumference of the circle.
Examples 6.1.4 1. Let p be the absolute value metric on R, let x e R, and
let e be a positive real number. We have
f3(x, e)= {yeR lx -yl <e}= (x - e, x +e)
Hence for any xeR and any positive real number e, the open bali f3(x, e)
is just the open interval (x - e, x +e). Similarly, the closed bali p(x, e) is
just the closed interval [x - e, x +e].
2. Let M be a set and let p be the discrete metric on M, so that
p(x, y) = 1 if x ~ y in M, and p(x, y) = Oif x = y in M. If e e R, e > 1, we
have
f3(x, e)= {yeM 1p(x,y) <e}= M
while if O< e~ 1, we have

f3(x, e)= {yeM 1 p(x,y) <e}= {x}
Similarly, if e~ 1, then p(x, e)= M, while if O< e< 1, then P(x, e)= {x }.
3. Let p be the product metric on R 2 of Example 4 ofExamples 6.1.2.
Let (h, k) be a point in R 2 , and let e be a positive real number; then
f3((h, k), e)= {(x, y) eR 2 (x - h) 2 + (y - k) 2 < e2 }
which is the set of ali points in the plane which lie in the interior of the
circle with center at (h, k) and radius e; that is, f3((h, k), e) is the open disc
with center (h, k) and radius e. ~ilarly, the closed bali P((h, k), e) is the
closed disc with center (h, k) and radius e.
The preceding examples suggest a connection between a metric on a
set and a certain topology on the set. For instance, consider the absolute
value metric on R. It is clear that a subset G of R is open in the natural
topology on R if and only if for each xeG there is some positive real
number e such that f3(x, e) = (x - e, x +e) e G. (Why?) Thus we could
use the natural metric on R to define the natural topology on R: we could
define a subset G of R to be open in the natural topology if for every xeG
there is some positive real number e such that f3(x, e) e G. ln this manner,
given any set M with a metric p, we may use p to define a topology on M.
144
Metric Spaces
p
p
Proposition 6.1.5 Let M be a set and let p be a metric on M. If is the

collection of ali subsets G of M having the property that for each xeG,
there is some positive real number e such that P(x, e) e G, then
is a
topology on M.
We leave the proof of Proposition 6.1.5 as an exercise. (See Exercise
6 at the end of this section.)
Definition 6.1.6 Let M be a set and let p be a metric on M. Let be the

topology on M defined in the statement of Proposition 6.1.5. The topological space (M, rp) is called a metric space, and the topology p is the metric
topology induced on M by p.
We remark that it is common practice to suppress mention of the
metric topology rP and thus refer to "the metric space (M, p)" rather than
"the metric space (M, p)."
We have seen that if pisa metric on a set M, then p induces a metric
topology on M. Of course, this topology may be identical to a topology
defined on M without reference to p.
Examples 6.1.7 1. Let p be the absolute value metric on R. We will show
that the metric topology induced on R by p is the natural topology, thus
justifying the remarks of the paragraph preceding Proposition 6.1.5.
lf G is a subset of R which is open in the metric topology, then for
each xeG, there is some positive real number e such that P(x, e) e G. But
since P(x, e) = (x - e, x + e) is an open interval, this shows that G is open
in the natural topology on R. Conversely, if G is open in the natural
topology on R, then for each x eG there is some open interval (a, b) such
that xe(a, b) and (a, b) e G. Thus if e= min {x -a, b -x}, then e is a
positive real number such that p(x, e) = (x - e, x +e) e (a, b) e G, so G
is open in the metric topology on R.
We have thus shown that the metric topology induced on R by
the absolute value metric is identical to the natural topology on R. This
result has an important consequence, for it implies that ali the results we
have previously obtained concerning R with the natural topology remain
valid when we consider R to be a metric space with the absolute value
metric.
2. Let M be a set and let p be the discrete metric on M. For ali x e M,
the one-point subset {x} is open in the metric topology because
The Metric Topology
145
But if every one-point subset of a topological space is open, then the

topology is the discrete topology. Hence the metric topology induced on
a set by the discrete metric is the discrete topology.
Metric spaces are a special type of topological space, and in the next
two sections of this chapter we shall study them as such. Before we
conclude this section, however, we must clear up two more points concerning the metric topology by showing that open balis are open sets and
closed balis are closed sets in the metric topology. The proofs of these
facts are typical examples of metric space arguments which use the triangle inequality.
Proposition 6.1.8 Let (M, p) be a metric space. ln the metric topology on

M, every open bali is an open set and every closed bali is a closed set.
Proof: Let p(x, e) be an open bali in M. If p(x, e) = 0, then surely it is

open, so assume that p(x, e) =F 0. ln order to show that P(x, e) is an open
set in the metric space (M, p), we must show that if z is any element of
P(x, e), there is some positive real number '1 such that P(z, '7) e P(x, e). If
z = x, we choose '1 = e. If z =F x, then O < p(x, z) < e, and we choose
'1 =e - p(x, z). Clearly, '1 is a positive real number, and we claim that
P(z, '7) e p(x, e). To see this, note that if y ep(z, 'J), then p(z, y) < '1 and by
the triangle inequality we have
p(x, y)
p(x, z)
+ p(z, y) < p(x, z) + '1 =e
But p(x, y) <e implies that y ep(x, e). Therefore P(z, '7) e P(x, e), and we
have proved that every open bali is an open set in the metric topology.
Now we prove that every closed bali if(x, e) is closed in (M, p). If
if(x, e) = M, then surely it is closed, so assume that if(x, e) =F M and consider M - if(x, e). lt will suffice to show that M - if(x, e) is open. If
zeM - if(x, e), then p(x, z) >e. Choose '1 = p(x, z) - e. Again, '1 is a positive real number, and we claim that P(z, '7) e M - if(x, e). This follows
because if yep(z, 'J), then p(y, z) < 'J, and by the triangle inequality,
p(x, z) ~ p(x, y) + p(y, z). Therefore
p(x, y)
p(x, z) - p(y, z) > p(x, z) - '1 =e
Hence y if(x, e) and therefore P(z, '7) is contained in M - if(x, e). We

have thus shown that for every z eM - if(x, e) there is an open bali about
z which is contained in M - if(x, e), and hence M - if(x, e) is open in the
metric topology, and we are done. D
146
Metric Spaces
We should note that although 'iJ(x, e) is a closed set containing the

open set /J(x, e), it is not true in general that p(x, e) is the closure of {J(x, e).
For example, suppose M = {x, y} with the discrete metric. Then
/J(x, 1) = {X}
and
p(x, 1) =M
while the closure of {J(x, 1) is the set

/J(x, 1) = {X}
EXERCISES
1.
Verify that the discrete metric of Example 2 of Examples 6.1.2

satisfies the conditions of Definition 6.1.1 and thus is indeed a
metric.
2. Verify that the product metric of Example 4 of Examples 6.1.2 is a
metric on R 2 .
3. Let M be the set of ali .sequences in R which converge to O. If {xn}
and {Yn} are elements of M, set
4.
5.
Prove that p is a metric on M.

Let u: R 2 x R 2 -+ R be given by
for ali (x 1, y 1), (x 2 , y.2) in R 2 . Prove that u is a metric on R 2 .

For each qeQ, q #O, we may write q = (s/t)2m, where s, tare integers not divisible by 2 and m eZ, and define
Also define IOl 2=O. (This is known as the 2-adic valuation on Q; if

we had used another prime number p in place of 2, we would have
the p-adic valuation on Q.) Prove that
(a) lql2~0 for ali qeQ;
(b) lql 2=O if and only if q =O;
(e) lqrji = lqjilrl 2 for ali qeQ, reQ;
(d) lq + rji ~ max {iqji, lr'2} for ali qeQ, reQ.
The Metric Topology
6.
7.
147
Let p: Q x Q-R be given by p(q, r) = lq-rl 2 for ali qeQ, reQ,

and show that p is a metric on Q.
Let M = [ -1, 1] u{2} and define <p: M x M-R by
ifx =y
if x =I y and neither x nor y is 2
if one (but not both) of x or y is 2
8.
9.
10.
11.
Prove that <p is a metric on M.

Let u be the metric of Exercise 4 above. Let (x 1, y 1) eR 2 and let e be
a positive real number. Give a geometric description of the open bali
/3((x 1, y 1), e) and the closed bali i]((x 1, y 1), e) in the metric space
(R 2 , u).
Let r be the product topology on R 2 , let p be the product metric of
Example 4 of Examples 6.1.2, and let u be the metric of Exercise 4
above.
(a) Are the metric topologies induced on R 2 by p and u identical?
Justify your answer.
(b) Are either of the metric topologies induced on R 2 by p or a
identical with the product topology r on R 2? Justify your
answer.
(a) Let p be the 2-adic metric of Exercise 5 above. Let q e Q and let
e be a positive real number. Describe the open bali f3(q, e) and
the closed bali if(q, e) in the metric space (Q, p).
(b) The absolute value metric on R induces an absolute value metric
on Q by restriction. Do the absolute value metric on Q and the
2-adic metric induce the sarne metric topology on Q?
Define 1/1: R x R _,. R by
lx-yl
ifl(x, y) = l + lx - y
12.
for all x E R, y E R. Is 1/1 a metric on R? If so, does it induce the sarne

metric topology on R as does the absolute value metric?
Let / 1 denote the set of all sequences {xn} in R such that the series
1 lxnl converges. Define p: / 1 x / 1 _,.R by
:r.:=
00
p( {xn}, {Yn})
L lxn - Ynl
n=I
for all {xn}e/ 1, {Yn}e/ 1 Prove that (/ 1, p) is a metric space.
Metric Spaces
148
13. We shall use this exercise to construct the metric space (Rn, Pn)
known as Euclidean n-space. The readeris asked to fill in the details.
Let n eN, and set Rn = {(x1o ... , xn) 1xkeR, 1 ~ k ~ n }. Thus, Rn is
the set of ali n-tuples of real numbers.
(a) Prove the Cauchy-Schwartz inequality for Rn: if (x1o ... , xn)
and (y 1, , Yn) are elements of Rn, then
Hint: Let teR and consider the nonnegative real number

n
L (txk + Yk) 2= t 2k=I

L xk 2+ 2tk=I
L XkYk + k=I
L Yk 2
k=I
Set
rearrange, and take square roots.

(b) Prove the Minkowski inequality for Rn: if (x 1,
(y 1, , Yn) are elements of Rn, then
n
[ k~1 (xk
+ Yk)2
]1/2 [
~ k~1xk2
]1/2 [
+ k~1Yk2
xn) and
]1/2
Hint: Expand
and use (a).

(e) If x = (x., ... , Xn) and y = (y 1,
Yn) in Rn, define
Prove that Pn is a metric on Rn.

(d) Show that p 1 is the absolute value metric on R and that p2 is the
product metric on R 2 Describe open balis and closed balis in
the metric space (R 3 , p 3).

14.
149
We shall use this exercise to construct an important metric space

known as Hilbert space, or 12. The reader is asked to fill in the
details. Let /2 denote the set of all sequences {xn} in R such that the
associated series ~:'= 1 xn 2 converges.
(a) Prove that if {xn}e/2 and {Yn}e/2, then {xnYn}e/2. Hint: Use
Exercise 13(a).
(b) Prove the Cauchy-Schwartz inequality for 12: if {xn}e/2 and
{Yn} E/2, then
1
n~l XnYn ~ n~l x/

00
00
]1/2. [n~l y/ ]1/2

00
(e) Prove that if {xn} and {Yn} are elements of 12, then {xn + Yn} is
an element of 12. Next show that the Minkowski inequality
00
n~l (Xn
+ Yn)2
]1/2 [
00
~ n~l Xn 2
]1/2 [
00
+ n~l Yn 2
]1/2
holds in /2.
(d) If {xn}e/2, define the norm of {xn}, denoted by llxn li. as follows:
Prove that
(i) llxnll ~O for all {xn}e/2;
(ii) llxn li =O if and only if Xn =O for all n eN;
(iii) llcxnll =lei llxnll for all {xn}e/2 and ali ceR;
(iv) llxn + Ynll ~ llxnll + llYnll for all {xn}e/2 and {Yn}e/2.
Show that the function p defined on /2 x / 2 by
15.
is a metric on 12.
Let (M, p) be a compact metric space. Prove that M contains a
countable dense subset.
6.2.
CONTINUITY IN METRIC SPACES
ln this section we study continuous functions from one metric space to

another. We begin with a characterization of continuity in terms of the
150
Metric Spaces
metrics on the spaces, then define the concept of a uniformly continuous

function from one metric space to another, and finally show that a function continuous on a compact metric space is uniformly continuous there.
Let (M 1, p 1) and (M2, p 2) be metric spaces andf: M 1 -+ M 2. According
to Definition 4.1. 7,f is continuous at x eM1 if and only if, for every open
set G in M 2 which contains/(x), there is some open set H in M 1 such that
xeH andf(H) e G. Our first result in this section reformulates this characterization of continuity at a point in terms of the metrics p 1 and p2
Proposition 6.2.1 Let (M1o p 1) and (M2 , p2 ) be metric spaces and let
/: M 1 -+ M 2 If x eM1o then f is continuous at x if and only if for every
positive real number 6 there is some positive number such that for ali
yeM1,
p 1(x,y) <
implies
P2(f(x),f(y)) < 6.
Proof: Suppose/is continuous at xeM. If G is open in M 2 andf(x) is

contained in G, then there must be some open set H in M 1 such that xeH
and f(H) e G. ln particular this is so if G = P(f(x), 6), the open bali of
radius 6 about f(x). Thus there is some open set H in M 1 such that x e H
and/(H) e P(f(x), 6). But since H is open, there must be some positive
real number such that P(x, ) e H. Now if y eM1 with p 1(x, y) <, then
y ep(x, ) and hence f(y) eP(f(x), 8), so that p 2(f(x),f(y)) < 6.
Conversely, let x eM1 and suppose that for every positive real number
6 there is some positive real number such that whenever y eM1 and
p 1(x, y) < then p 2(f(x),f(y)) < 6. If G is an open set in M 2 which contains/(x), then there is some positive real number 6 such that the open bali
P(f(x), 6) e G. For this 8, there is some positive real number such that
yeM1, p 1(x,y)< implies that p2(f(x),f(y))<6. But then
f(P(x, )) e P(f(x), 6), and since p(x, ) is an open set in M 1 which contains x, this shows that f is continuous at x. O
Corollary 6.2.2 Let S be a subset of R and/: S-+R. If xeS, thenf is

continuous at x if and only if for every real number 6 > O there is some
real number > Osuch that if yeS and lx - YI <, then lf(x) - f(y)I < 6.
Proposition 6.2.1 is often used as the definition of continuity at a
point for functions from one metric space to another; similarly, Corollary
6.2.2 is often used as the definition of continuity at a point for functions
on R. When this is dane, these definitions are known as " - 6 defini-
151
tions," and when we use Proposition 6.2.l or Corollary 6.2.2 to show that
a function is continuous ata point, we will be doing " - e proofs." The
following examples illustrate the technique of - e proofs.
Examples 6.2.3 1. Let /: R-+ R be given by f(x) = 3x + 2 for ali x eR.
We use Cotollary 6.2.2 to show thatfis continuous at x for ali xeR.
Given any real number e > O, we must find a real number > Osuch
that lx - YI < implies lf(x) - f(y)I <e. But if lx - YI <, then
lf(x) - f(y)I = l(3x
+ 2) -(3y + 2)1=3lx -
YI < 3
Thus if we take = e/3, then lx - YI < = e/3 implies
lf(x) -f(y)I = 3lx - YI < 3
= 3
(i)
=e
Therefore f is continuous at x for ali xeR. Note that our choice of

depended only on e.
2. Let /: R -+ R be defined by
if X> 0
if X ~0
The functionf is not continuous at x =O. To see this, choose e = 1; then
there is no > Osuch that IO- YI < implies lf(O) - f(y)I <e, because no
matter how is chosen, there is some y such that O< y < , and hence
such that
IO- YI = y < while lf(O) -f(y)I = 1- I - II= 2 ., e= 1
3. Letf: (O, l]-+R be given by f(x) = 1/x for ali xe(O, l]. To show
thatfis continuous at each xe(O, l], we must show that for every e> O
there is some > O such that if ye(O, l] and lx - YI <, then
lf(x) -f(y)I = lx - YI <e
xy
But lx - YI < implies that
lx-yl
--<xy
xy
Metric Spaces
152
From this last inequality we see that we must choose in such a way that
the quotient /xy is less than e. lt will not help us to choose to be a small
positive real number if at the sarne time we force the product xy to be
small, for then the quotient /xy may be very large. To keep this from
happening, we will select a preliminary which forces xy to be bounded
below and then adjust this preliminary to obtain /xy <e.
Suppose we choose ~ x/2. If x and y are in the interval (O, l] and
lx - YI <~ x/2, then x/2 < y < 3x/2 and therefore xy > x(x/2) =
x 2/2. Choosing ~ x/2 thus forces xy to be bounded below by x 2/2. Now
let us see what effect this has on lf(x) - f(y)i: since xy > x 2/2, we have
lx -yl
2
lf(x) -f(y)I = ~ < xy < x2;2 = x2
Clearly, if we now adjust so that it is less than or equal to ex 2 /2 as well
as less than or equal to x/2, then we will have lf(x) - f(y)I <e. Therefore
we let = min {x/2, ex 2/2}, and conclude that if x and y are in (O, l] with
lx - YI <, then xy > x 2/2 and hence
lf(x) - f(y)I = lx - YI < !..__ < 2~ < 22 ex22 =e

xy
xy x
x
Note that in this example our choice of depended on both e and on x.
ln Example 1 of Examples 6.2.3 we were able to find a >O such that
lx - YI < implied lf(x) - f(y)I <e, and depended only on e, and not
on the point x under consideration. ln Example 3 of Examples 6.2.3, on
the other hand, the we obtained depended on both e and on the point
x. Continuous functions for which it is possible to find a >O which
depends only on e and not on x are said to be uniformly continuous.
Definition 6.2.4 Let (M1> p 1) and (M2 , p 2) be metric spaces. A function

f: M 1 -+ M 2 is uniformly contlnuous if for every positive real number e
there is some positive real number such that whenever xeM1 and yeM 1
with p 1(x, y) <, then p2(f(x),f(y)) <e.
Let us contrast Proposition 6.2. l with Definition 6.2.4: the proposition says thatfis continuous at x if for every positive real number there
is some positive real number e such that if y E M 1 and p 1(x, y) < , then
p2 (f(x),f(y)) <e; this is a local condition at the point x, and thus in
general the number will depend on both the choice of e and on the point
153
x being considered. The definition says that/ is uniformly continuous on

the space M 1 if for every positive real number e there is some positive real
number such that whenever xeM1 and yeM1 with p 1(x,y) <, then
p2(/(x),f(y)) <e; this is a global condition on MI> and thus can depend
only on the choice of e, and not on the points x and y.
Uniform continuity is clearly a stronger condition than continuity: if

f is a function which is uniformly continuous on a metric space MI> then
Proposition 6.2. l shows that f is continuous at every point of MI> hence
continuous on M 1 Therefore uniform continuity implies continuity. The
converse is not true, however, for (as Example 2 of Examples 6.2.5 below
demonstrates), a function can be continuous on a metric space without
being uniformly continuous there.
Ex:amples 6.2.5 l. Let/: R -+R be given by f(x) = 3x + 2 for ali xeR.

As Example l of Examples 6.2.3 shows, for any e :>O, we may choose
= e/3 and have lf(x) - /(x)I <e whenever lx - YI <. Therefore the
function f is uniformly continuous on R.
2. Let/: (O, l]-+ R be given by f(x) = 1/x for ali x e (O, l]. Example 3
ofExamples 6.2.3 shows that this function is continuous on (O, l], but we
claim that it is not uniformly continuous there. To see this, suppose the
contrary; then for any e > O there is some > O such that
lx-yl<<>
implies
lx-yl
--<e
xy
If x is any point of (O, l) such that O< x <, and if y = x/2, then
X
lx-yl=-<
2
Therefore
lx-yl l
--=-<e
xy
X
But this is impossible, for no matter how is chosen, l/x will be arbitrarily
large for x sufficiently near zero. Thus f cannot be uniformly continuous
on (O, l].
3. Let aeR, O< a< l, and let/: [a, l]-+ R be given by f(x) = l/x for
ali x e[a, l]. We claim that f is uniformly continuous on [a, l]. For any
e> O, choose = ea 2 If x and y are points in [a, l], then xy > a 2 ; if also
Metric Spaces
154
lx - YI < = ea 2, then
lf(x) - f(y)I = lx - YI < (ea2) < (ea2) =e
xy
xy
a2
and therefore f is uniformly continuous on [a, l].

We have just shown that the function f defined by f(x) = 1/x is
uniformly continuous on the metric space [a, l] if O< a < l, but is not
uniformly continuous on (O, l]. The reason this is so is that [a, l] is
compact while (O, l] is not. The following important result tells us th.at a
function which is continuous on a compact metric space is uniformly
continuous there.
Theorem 6.2.6 Let (Mh p 1) and (M2, p2) be metric spaces and let
f: M 1 -+M2 be continuous. If M 1 is compact, thenfis uniformly continuous on M 1
Proof: Let e be a positive real number. Since f is continuous at each
point t of M 1, there is some = (t) >O such that if z eM1 and
p 1(t, z) < (t) then p2(f(t),f(z)) < e/2. (We write = (t) to emphasize
that depends on the point tas well as on the number e.) The collection
of open balis {/J(t, (t)/2) 1 teM1} is an open covering for M 1 and since
M 1 is compact, it must have a finite subcovering, say
Let = min {(t 1)/2, ... , (tn)/2}. Clearly is a positive real number, and
we claim that if x and y are points of M 1 such that p 1(x, y) <, then
P2(f(x),f(y)) < e.
If xeM1o then xe{J(tk, (tk)/2), for some k, 1 ~ k ~ n, and hence
But this implies that
155
If yeM1 and p 1(x, y) <, then p 1(x, y) < (tk)/2. By the triangle inequality, we have
and therefore p 2(f(tk),f(y)) < e/2 also. Thus
and we are done. O

Corollary 6.2.7 Let /:[a, b]-+ R. If f is continuous on [a, b], then it is
uniformly continuous on [a, b].
EXERCI SES
1.
2.
Let a, b be fixed real numbers. If /: R -+ R is given by f(x) = ax + b

for all xeR, use Corollary 6.2.2 to show thatf is continuous on R.
For each of the following functions, use Corollary 6.2.2 to show that
the function is continuous at the given values of x.
(a) /(x) = x 2, for all xeR;
(b) f(x) =
for xeR, x ;;;i:: O;
(e) f(x) = l/(x 2 - 1), for xeR, x > 1.
Let /: R -+ R be given by
Jx,
3.
f(x)
4.
{~
ifx ;60
if X =0
Use Corollary 6.2.2 to show that f is not continuous at x = O.

Let a be any real number and let /: R -+ R be given by
ifx ;60
ifx =O
Use Corollary 6.2.2 td show thatf is continuous at x if and only if
x#O.
Metric Spaces
156
5.
6.
Consider the identity function f(x) = x.

(a) Using only Definition 6.2.4, show that this function is uniformly
continuous on [O, l].
(b) Is this function uniformly continuous on R? Justify your answer
using Definition 6.2.4.
For each of the following, either prove that the given function is
uniformly continuous on the given set or prove that it is not.
(a) f(x) = x 2, on [ -n, n], neN;
(b) f(x)
on [O, +oo);
(c) f(x) = l/(x 2 - l), on (1, +oo);
(d) f(x) = l/(x 2 - 1), on [a,+ oo), where a> 1.
Let S be a subset of R and let f: S-+ R. We say that f satisfies a
Lipschitz condition on S if there exists ceR, e> O, such that
=JX,
7.
lf(x) -f(y)I <
clx - YI
for ali elements x and y of S. Prove that if f satisfies a Lipschitz

condition on S then f is uniformly contit?-uous on S.
8. Give an example to show that the converse of the result stated in
Exercise 7 above is not true.
6.3
SEQUENCES IN METRIC SPACES
ln this section we study sequences in metric spaces. We will characterize

convergent sequences in a metric space in terms of the metric, then define
the concepts of Cauchy sequence and complete metric space, show that R
is a complete metric space, prove that the intersection of nested closed
balis in a complete metric space is nonempty, and finally prove the
important Bolzano-Weierstrass Theorem. Before we can do any of this,
however, we must show that metric spaces are Hausdorff spaces, and thus
that convergent sequences in them have unique limits.
Proposition 6.3.1
Every metric space is a Hausdorff space.
Proof: If x and y are distinct points in a metric space (M, p) and we let
e= p(x, y), then p(x, e/2) and p(y, e/2) are nonempty disjoint open sets
which separate the points x and y. D
Corollary 6.3.2
unique limit.
Every convergent sequence m a metric space has a
157
Now we are ready to characterize convergent sequences in a metric

space in terms of the metric.
Let (M, p) be a metric space. If {xn} is a sequence in
xn
= x if and only if for every positive real number e
00
there is some m EN such that p(x, xn) < e for ali n :;;i: m.
Proposition 6.3.3
M, then
Limn~
Proof: Suppose Limn ~ 00 Xn = x in M; then for every open set G in M

which contains x there is somem EN such that xn EG for ali n :;;i: m. If e is
a positive real number, then f3(x, e) is an open set which contains x, and
hence there is some mEN such that XnE/3(x,e) for ali n :;;i:m. But this
implies that p(x, xn) < e for ali n :;;i: m.
Conversely, let xEM, let {xn} be a sequence in M, and suppose that
for every positive real number e there is somem EN such that p(x, xn) <e
for ali n :;;i: m. Let G be any open set which contains x. Since G is open,
there is some e >O such that f3(x, e) e G, and therefore for this e, there is
some m EN such that p(x, xn) <e for ali n :;;i: m. But this implies that
Xn E f3(x, e) e G for ali n :;;i: m, and thus that Limn ~ 00 xn = x. D
Corollary 6.3.4 If { Xn} is a sequence in R, then Limn~ 00 xn = x if and

only if for every positive real number e there is some m EN such that
lx - Xn 1 < e for ali n :;;i: m.
Examples 6.3.5 1. Let us use Corollary 6.3.4 to show that {(n + I)/n}
converges to 1 in R. Given e >O we must find m EN such that l1 - (n + 1) /
n 1 = 1/n < e for ali n :;;i: m. However, since e > O, there is some m EN
such that O< I/m <e; hence for ali n :;;i: m, we have
(n + l)/nl =
1/n < I/m <E.
2. Consider the sequence {( - 1Y} in R. If this sequence converged to
x, then we could take e = 1 in Corollary 6.3.4, and therefore there would
be some m EN such that lx - ( - 1)nl < 1 for ali n :;;i: m. But this is impossible, for if x :;;i: O and n is any odd natural number, then
lx - ( - l)nl = x + 1 :;;i: 1, while if x <O and n is any even natural number,
then lx - ( - l)nl = 1 - x > 1.
11-
Proposition 6.3.3 is often used as the definition of a convergent sequence in a metric space. Similarly, Corollary 6.3.4 is often used as the
definition of a convergent sequence in R.
lt is possible to do more with sequences in metric spaces that it is
in more general topological spaces because in metric spaces there is
an explicit measure of the distance between points. Thus, in a manner of
Metric Spaces
158
speaking, Proposition 6.3.3 says that if a sequence {xn} converges to x in

a metric space, then the terms ofthe sequence are getting closer and closer
to x, and it follows as a consequence ofthe triangle inequality that the terms
of the sequence must be getting closer and closer to each other. Sequences
whose terms get closer and closer to each other are called Cauchy sequences.
Definition 6.3.6 Let (M, p) be a metric space. A sequence {xn} in M is a
Cauchy sequence if for every positive real number e there is somem e N such
that p(xm xk) < e for all n ;;:;: m and k ;;:;: m.
Proposition 6.3.7
Every convergent sequence in a metric space is a
Cauchy sequence.
Proof: Suppose {xn} converges to x in the metric space (M, p). By Proposition 6.3.3, for every positive real number e there is somem eN such that
p(x, xn) < e/2 for all n ;;:;: m. But then for ali n ;;:;: m and k ;;:;: m we have
p(xm xk) ~ p(xm x) + p(xk, x) < e/2 + e/2 =e, which shows that {xn} is a
Cauchy sequence. D
If { xn} is a convergent sequence in R, then for every e > O
there is some me N such that lxn - xk 1 < e for ali n ;;:;: m and k ;;:;: m.
Corollary 6.3.8
Unfortunately, the converse of Proposition 6.3.7 is not true in general:

in an arbitrary metric sp!'lce, Cauchy sequences need not converge.
Example 6.3.9 Let R have the absolute value metric, as usual, and consider the set Q of rational numbers as a subspace of R. ln the metric space
Q, Cauchy sequences need not converge. To see this, let {qn} be a sequence
of rational numbers which converges to
in R. (We know that such a
sequence exists. Why?) Since this sequence converges in R, it must be a
Cauchy sequence in R, and hence it must also be a Cauchy sequence when
considered as a sequence in the subspace Q. But the sequence {qn} cannot
converge in Q, for if it did it would necessarily converge to a rational
number, and therefore could not converge to the irrational number
in R.
J2.
J2.
A metric space is complete if every Cauchy sequence in

the space converges to a point of the space.
Definition 6.3.10
We have just seen that Q with the absolute value metric is not a
complete metric space, so not all metric spaces are complete. It is a very
159
important fact that R with the absolute value metric is a complete metric
space, and we will prove this. Before we can doso, however, we need the
following. result.
Proposition 6.3.11
Every Cauchy sequence in R is bounded.
Proof If {xn} is a Cauchy sequence in R, then there is somem eN such

that lxn - xk 1 < 1 for ali n ~ m and k ~ m. Therefore, for ali n ~ m,
Let a= max {lxd, ... , lxml}: for ali neN, lxnl < 1 +a, and hence {xn} is
bounded. D
Theorem 6.3.12 R is a complete metric space.

Proof: Let {xn} be a Cauchy sequence in R. We must show that {xn}
converges to some real number. By Proposition 6.3.11, {xn} is bounded;
therefore s = Lim sup xn exists as a finite real number, and if {xn} does
converge, it must converge to s. We will prove that this is so.
We claim that for every positive real number e and every me N there
is some peN, p > m, such that is - xPI <e. To see this, recall that
s = Limk-.oo sk, where sk = sup {xn 1 n ~k} for each keN. Since the sequence {sd converges tos, there is some teN such that
for ali k
Let meN and choose reN such that r > max {t, m}. Then r > m and
If is, - xPI ~ e/2 for ali p ~ r, then s, cannot be the supremum for the set
{xn 1n ~ r }; but this is the definition of s,. Therefore there is some
p ~ r > m such that Is, - xPI < e/2. Thus we have p > m and
This proves the claim made at the beginning of this paragraph.
Metric Spaces
160
Now we are ready to show that the Cauchy sequence {xn} converges
tos. Since {xn} is a Cauchy sequence, there is some meN such that
for ali n
m and k
By the claim of the preceding paragraph, however, there is some p > m

such that is - xPI < e/2. Thus for ali n ~ p,
Hence {xn} converges to s and we are done. D

Corollary 6.3.13
A sequence in R converges if and only if it is a Cauchy
sequence.
Complete metric spaces have many properties which incomplete ones
do not. For instance, suppose that {if(xm en) 1n eN} is a collection of
closed balis in a metric space such that the real sequence {en} converges to
O and if(xn+ 1, Bn+ 1) e if(xm en) for ali neN. If the metric space is complete, then the intersection of these nested closed balis must consist of
exactly one point, while if it is not complete the intersection may be
empty.
Theorem 6.3.14
Let (M, p) be a complete metric space. If
is a collection of closed balis in M such that Limn .... 00 en = O in R and
for ali n eN, then
contains exactly one point.
Proof: We will show that the sequence {xn} of center points of the closed
balis is a Cauchy sequence in M; since M is complete, this will imply that
161
the sequence {xn} converges to a point x of M, and the point x will be the
unique element of lneN P(xm en).
Let e be a positive real number. Since Limn .... 00 en =O in R, there is
some m eN such that O< en < e/2 for ali n ~ m. But if n ~ m and k ~ m
then P(xm en) e P(xm, em) and P(xk, ek) e P(xm, em), and thus
Hence {xn} is a Cauchy sequence in M and therefore converges to some

xeM.
We claim that xe lneN P(xm en). If x is not an element ofthis intersection, then there is some me N such that x e M - p(xm, em), which is an
open set. Therefore there is some e > O such that
But then p(x, e) n P(xm en) = 0 for ali n ~ m, and this in tum implies that
p(x, xn) ~e for ali n ~ m, which contradicts the fact that {xn} converges
to x. Hence X E lneN P(xm en).
Finally we show that x is the only element in lneN P(xm en). To see
this, suppose that y is also a point in this intersection; then
for ali n e N. But O ~ p(x, y) < 2en for all n e N and Limn .... 00 en = O together imply that p(x, y) =O and hence that y = x. O
Corollary 6.3.15
Let (M, p) be a complete metric space. If
is a collection of closed balls in M such that
for all neN, then lneNP(xm en) is nonempty.

(Nested Intervals Theorem) If {Jn 1n eN} is a collection of closed intervals in R such that Jn + 1 e ln for ali n e N, then lneN Jn
is nonempty.
Corollary 6.3.16
162
Metric Spaces
The proofs of Corollaries 6.3.15 and 6.3.16 will be left to the reader.
Example 6.3.17 As was previously noted, the conclusion of Theorem
6.3.14 need not hold if (M, p) is not a complete metric space. As an
example, consider the metric space (O, 1) with the absolute value metric.
The sequence {l/(n + 1)} is a Cauchy sequence in (O, 1) which does not
converge to a point of the space, so the space is not complete. Note that
the collection of closed balis {if(O, l/(n + 1)) 1n eN} satisfies the hypotheses of Theorem 6.3.14, but that the intersection of these closed balls is
empty.
Theorem 6.3.14 allows us to prove the important Bolzano-Weierstrass theorem.

Theorem 6.3.18 (Bolzano-Weierstrass)
set of R then S has a limit point.
If S is a bounded infinite sub-
Proof: Since S is bounded, there exists a closed interval [a, b] such that
Se [a, b]. Let x 1 be the midpoint of [a, b] and let 6 1 = (b - a)/2; then
Se [a, b] = if(x 1, 6 1). Because S is infinite, at least one of the intervals
[a, xi], [x 1, b] must contain an infinite subset of S. Choose one of the
intervals [a, x 1], [x 1, b] which does contain an infinite subset of S; if x 2 is
the midpoint of the chosen interval and
then the interval may be written as if(x2 ,
6 2).
Note that
Now bisect the closed interval if(x2 , 6 2) and argue as before. ln this manner we obtain for each n eN a closed interval if(xno 6n) which contains an
infinite subset of S and such that 6n = (b - a) /2n and
if(xn + 1o 6n + 1) e if(xno 6n). Therefore by Theorem 6.3.14 there is a unique
point x e lneN if(xno 6n). The point x is a limit point of S, because if G is
any open set containing x then there is some n e N such that if(xno 6n) e G
and if(xno 6n) contains an infinite subset of S. D
Note that the conclusion of the Bolzano-Weierstrass theorem fails to
hold if either of the hypotheses is contravened: thus, a finite subset of R
163
will have no limit point, while N provides an example of an infinite

unbounded subset of R which has no limit point.
EXERCISES
1.
For each of the following sequences in R use Corollary 6.3.4 to

establish convergence or divergence.
(a)
{~}
(b)
{n
(c)
2.
3.
4.
5.
6.
7.
8.
2:
1}
t2: l}
Let M be a set and p the discrete metric on M. Prove that (M, p) is

a complete metric space.
Prove that if (M, p) is complete metric space and Sisa closed subset
of M, then (S, p) is a complete metric space. Give an example to
show that the conclusion fails to hold if S is not closed in M.
Let p be the absolute value metric on R. Give an example of an
uncountable subset S of R such that (S, p) is a complete metric space
which contains no open interval.
Let p be the product metric on R 2 Prove that (R 2 , p) is a complete
metric space.
Use Corollary 6.3.13 to establish convergence or divergence for the
following sequences in R.
(a) {xn}, where x 1 =1, x 2 = 2, and Xn = (l/2)(xn-J + Xn_ 2) for ali
n >2;
(b) {xn}, where Xn =~Z= 1 (l/k) for ali nEN.
Prove Corollaries 6.3.15 and 6.3.16.
Let (M, p) be a metric space and let S be a subset of M. Define S to
be bounded in M if there is some x EM and some positive real number e such that Se if(x, e). Define S to be totally bounded in M if for
every positive real number e there is a finite set of points xi> ... , xn
in M such that
S e
if(xk, e)
l~k~n
(a) Prove that if S is totally bounded in M, then S is bounded in M.
164
Metric Spaces
(b) Give an example of a metric space (M, p) anda subset S of M

such that S is bounded in M but not totally bounded in M.
9. Prove that in R, boundedness is equivalent to total boundedness;
that is, prove that a subset S of R is bounded if and only if it is
totally bounded.
10. Prove the generalized Bolzano-Weierstrass Theorem: if Sisa totally
bounded infinite subset of a complete metric space, then S has a limit
point in the space.
11. Give an example of a metric space (M, p) and a subset S of M such
that
(a) S is infinite and totally bounded in M but has no limit point;
(b) (M, p) is complete, S is totally bounded in M, but S has no limit
point;
( c) (M, p) is complete, S is infinite and bounded in M, but S has no
limit point.
Thus if any one of the three hypotheses of the generalized BolzanoWeierstrass Theorem is removed, the theorem fails.
12. Prove that the property fTheorem 6.3.14 is equivalent to completeness; that is, prove that a metric space (M, p) is complete if and only
if
contains exactly one point
whenever
Lim
Bn
=O
n--+<X>
13.
Let M be the set of all sequences in R which have a finite number of

nonzero terms. Thus
M = {{xn} in R 1 for some m eN,
14.
Xn
=O for all n ~ m}
If p( {xn}, {Yn}) = sup {lxn - Ynl 1 n eN} for ali {xn} and {Yn} in M,
then (M, p) is a subspace of the metric space / 00 , and hence is itself
a metric space. Show that (M, p) is not complete. (Hint: consider
the sequence {sn} in M, where for each n e N, sn is the sequence
{l, 1/2, ... , I/n, O, O, ... }.)
Let S be a subset of R and let f: S -+ R. A point se S is a fixed point
of fif f(s) = s. Supposef satisfies a Lipschitz condition with positive
165
constant e < l. (See Exercise 7 of Section 6.2.) Show that f has a

unique fixed point. (Hint: let s 1 eS, define sn + 1 = f(sn) for ali n eN,
and consider the limit of {sn}.)
6.4 COMPLETION OF METRIC SPACES*
ln this section we will show that every metric space can be identified with
a dense subset of a complete metric space. Our proof of this fact will be
constructive: that is, we will start with an arbitrary metric space (M, p)
and demonstrate how to construct from it a complete metric space
(M*, p *) which contains as a dense subset a "copy" of (M, p ). Our proof
will also suggest an alternative method of constructing the real numbers
R from the rational numbers Q.
lt will be helpful to us to have available the notion of an equivalence
relation on a set.
Definition 6.4.1 Let S be a set. A subset R of S x S is called a relation on
S. If (x, y) E R, we write xRy; if (x, yH R, we write xRy. A relation R on
S is an equivalence relation if it is
l.
2.
3.
Reflexive: xRx for ali xeS;

Symmetric: if xeS, yeS, and xRy, then yRx; and
Transitive: if xeS, yeS, and zeS, and xRy and yRz, then xRz.
If R is an equivalence relation on S and x e S, then [x] = {y E S 1xRy} is

called the equivalence class of x under R, and x is referred to as the
representative of the equivalence class.
Examples 6.4.2 l. Let S={a,b,c} and R={(a,a),(b,b),(a,b)}, so
that aRa, bRb, and aRb. The relation R is not an equivalence relation on
S because, for instance, ceS but cRc. (Also, aRb but bRa.)
2. Let S ={a, b, e} and R ={(a, a), (b, b), (e, e), (a, b), (b, a)}. The
relation R is an equivalence relation on S. Note that [a] = {a, b },
[b] ={a, b }, and [e]= {e}; thus the equivalence class {a, b} has two different representatives, namely a and b. Note also that distinct equivalence
classes are mutually disjoint.

3. If x and y are real numbers, define xRy if x = y. This relation is an
equivalence relation on R. Note that [x] = {x }, for ali x eR.
*This section is optional.
166
Metric Spaces
4. If x and y are real numbers, define xRy if x ~ y. This relation is

reflexive and transitive, but not symmetric; hence it is not an equivalence
relation on R.
5. If {xn} and {Yn} are Cauchy sequences in Q, define {xn}R{yn} ifthe
sequence {lxn - Yn I} converges to zero. Let us show that this is an equivalence relation on the set of ali Cauchy sequences in Q:
Ref/exivity: If {xn} is a Cauchy sequence in Q, then surely the sequence

{lxn - xnl} ={O} converges to zero; hence {xn}R{xn}
Symmetry: If {xn}, {Yn} are Cauchy sequences in Q with {xn}R{yn}, then
{lxn - Ynl} converges to zero; but lxn - Ynl = IYn - xnl so {lyn - xnl}
converges to zero, and hence {Yn}R{xn}
Transitivity: If {xn}, {Yn}, and {zn} are Cauchy sequences in Q with
{lxn - Ynl} and {lyn - znl} converging to zero, then
implies that {lxn - znl} converges to zero (why?) and hence that
{xn}R{zn}
Note that [{O}]= [{l/n}l = [{an}], where {an} is any Cauchy sequence in
Q which converges to zero.
As the preceding examples suggest, if R is an equivalence relation on
a set S, then either xRy, in which case [x] = [y] and the elements x and
y are representatives of the sarne class, or x.Ry, in which case [x] n [y]
= f/J. Thus an equivalence relation on a set S partitions the set; that is, it
defines a collection of mutually disjoint subsets of S whose union is S.
(The mutually disjoint subsets are the distinct equivalence classes of S
under R.)
Proposition 6.4.3 Let S be a set and let R be an equivalence relation on

S. If xeS and yeS, then
1.
2.
xRy implies that [x] = [y];

xRy implies that [x] n [y] = f/J.
Hence the distinct equivalence classes of S under R partition the set S.

The proof of Proposition 6.4.3 is left to the reader. (See Exercise l at
At the beginning of this section we spoke of a "copy" of a metric
space (M, p); let us formalize this notion.
167
Definition 6.4.4 Let (M1> p 1) and (M2 , p 2 ) be metric spaces and letfbe a
function from M 1 onto M 2 lf p 1(x, y) = p(f(x),f(y)) for ali xEM1,
yEM1, thenfis called an isometry and M 1 and M 2 are said to be isometric
spaces.
It is easy to show that every isometry is a homeomorphism. (See
Exercise 2 at the end of this section.) Indeed, since an isometry preserves
distances, isometric spaces are identical as metric spaces; hence, if
/: M 1 -+M2 is an isometry, then we may think of M 2 as a "copy" of M 1
Now we are ready to begin the process of completing a metric space.
We wish to prove that every metric space is isometric to adense subset of
a complete metric space. Thus if (M, p) is any metric space, we must
construct from it a complete metric space (M*, p*) and then produce a
functionf: M -+M* such thatfis an isometry from (M, p) to (f(M), p*)
andf(M) is dense in M*. The basic idea of the proof is to construct M*
as a set of equivalence classes of Cauchy sequences in M.
Definition 6.4.5 Let (M, p) be a metric space and let C be the set of ali
Cauchy sequences in M. Define a relation R on C by setting {xn}R{yn} if
the sequence of nonnegative real numbers {p(xm Yn)} converges to zero.
Proposition 6.4.6 The relation R of Definition 6.4.5 is an equivalence
relation on C.
The proof of the Proposition is similar to that of Example 5 of Examples 6.4.2, and is left to the reader. (See Exercise 3 at the end of this
section.)
Definition 6.4.7 Let (M, p) be a metric space and let R be the equivalence
relation of Definition 6.4.5. If C is the set of ali Cauchy sequences in M
and {xn}EC, denote the equivalence class of {xn} under R by [xnl Let M*
denote the set of distinct equivalence classes of C under R, and define
p*: M* x M*-+R as follows: if[xn]EM* and [Yn]EM*, then
P*([xn], [yn]) = Lim p(Xm Yn)
n--+
oo
in R.
Our first task is to show that p*([xn], [yn]) exists and is well-defined.
Proposition 6.4.8
For ali [xn] EM* and [Ynl EM*, p*([xn], [YnD exists and
Metric Spaces
168
its value does not depend on the Cauchy sequences {xn}, {Yn} which are
chosen as representatives for the equivalence classes [xnl, [Ynl
Proof: To show that p *([xn], [yn]) exists for ali [xnl e M* and [Ynl e M*,
it will suffice to show that {p(xm Yn)} is a Cauchy sequence in R. (Why?)
By the triangle inequality in (M, p), we have
and
for ali n e N and me N. Therefore

(1)
and
(2)
for ali neN and meN. But because {xn} and {Yn} are Cauchy sequences
in (M, p ), for every positive real number e there is some k e N such that
n ~ k and m ~ k imply that
and
Hence the inequalities (1) and (2) together show that
and therefore {p(xm Yn)} is a Cauchy sequence in R.

To show that the value of p*([xn], [yn]) does not depend on the choice
of the representatives used for the equivalence classes, suppose that {xn},
{Yn}, {sn}, and {tn} are Cauchy sequences in (M, p) with {xn}R{sn} and
{Yn}R{tn} It will suffice to show that the sequences {p(XmYn)} and
{p(sm tn)} converge to the sarne limit in R. (Why?)
Since {xn}R{sn} and {Yn}R{tn}, we have
and
n-+oo
n-+OO
Also, since p *([xn], [yn]) and p *([sn], [tn]) exist by the first part of this
169
proof, the sequences {p(xm Yn)} and {p(sm tn)} converge in R. By the
triangle inequality,
for ali n eN, and therefore
= Lim p(Xm Yn)

n--+oo
Thus
n--+oo
n--+oo
and a similar argument shows that

n--+oo
n--+
oo
Hence
n--+
oo
n--+oo
and we are done. O

Now we know that p* is a well-defined function from M* x M* to R,
it is easy to show that it is a metric on M*.
Proposition 6.4.9
The function p* of Definition 6.4.7 is a metric on M*.
Proof: We will show that the triangle inequality holds, leaving the remainder of the proof to the reader. (See Exercise 4 at the end of this
section.)
Let [xn], [yn], and [zn] be elements of M*. By the triangle inequality for
(M,p),
for ali n eN, and hence

n--+
Therefore
oo
n--+
oo
n--+oo
Metric Spaces
170
We have now constructed from the metric space (M, p) a new metric
space (M*, p *) whose points are equivalence classes of Cauchy sequences
in M. We still must show that (M, p) is isometric to a dense subset of
(M*, p*) and that (M*, p*) is a complete metric space. Note that if xeM
then {x} is the constant sequence ali of whose terms are equal to x, and
[x] EM* is the equivalence class of this sequence.
Proposition 6.4.10 The functionf: M --+M* defined by f(x) = [x] for ali
xeM is an isometry from (M, p) onto (f(M), p*), and f(M) is a dense
subset of (M*, p *).
Proof:
For ali xeM, yeM, we have [x]eM*, [y]eM*, and

p*([x], [y]) = Lim p(x, y) = p(x, y)
n~oo
thusfis an isometry. To show thatf(M) is dense in (M*, p*) it will suffice

to show that for any [xn] eM* and any positive real number e, there is
some [x] ef(M) such that p*([xn], [x]) ~e. (Why?)
Since {xn} is a Cauchy sequence, there is some k EN such that
p(xm xm) < e for all n ~ k, m ~ k. Let x = xk> and consider
p*([xn], [x]) = Limn~ 00 p(xm x). Since p(xm x) = p(xm xk) <e for ali
n ~ k, it follows that Limn~ 00 p(xm x) ~e, and therefore that
p*([xn], [x])
~e
Now we are ready to begin the proof that the metric space (M*, p*)
is complete.
Proposition 6.4.11 If {xm *} is a sequence in (M*, p*), then for each
meN there is a Cauchy sequence {xnm 1 neN} in (M, p) such that
Xm * = [xnml and p(xnm' Xkm) < 1/m for ali n EN and k EN.
Proof: For each meN, xm *is a point of M*, and hence is an equivalence class of Cauchy sequences in (M, p); let {xnm} denote a representative of the class which determines xm *, so that Xm * = [xnml Since {Xnm} is
a Cauchy sequence, there is some tm eN such that p(xnm' xkm) < I/m for ali
n ~ t and k ~ tm. Consider the Cauchy sequences {Xnm 1 n EN} and
{xnm n ~ tm}: it is easy to show that these two sequences in (M, p) are
equivalent under the equivalence relation R of Definition 6.4.5 ( see Exercise 5 at the end of this section), and hence they define the sarne point,
namely xm * , in M*. If we now renumber the terms of the sequence
171
{xnm 1n ~ tm} so that its first term is Xin rather than x,mm we obtain a
sequence which satisfies the requirements of the proposition. D
Proposition 6.4.12 Let {xm *} be a Cauchy sequence in (M*, p*), and for
each m eNlet {xnm} be a Cauchy sequence in (M, p) such that Xm * = [xnm1
and p(xnmXkm) < 1/m for ali neN and keN. Then the sequence
{xim 1 meN} is a Cauchy sequence in (M, p).
Proof: The existence ofthe sequences {xnm} is guaranteed by the previous proposition; the sequence {xim 1m eN} consists of the first terms of
ali these sequences. We must prove that {xim} is a Cauchy sequence in
(M,p).
If e is a positive real number, then, since {xm *}is a Cauchy sequence,
there is some Pi eN such that p*(xm *, x, *) < e/3 for ali m ~Pi and t ~Pi
Choose p eN such that p > max {p., 3/e }: then 1/p < e/3 and
p*(xm *, x, *) < e/3 for ali m ~ p and t ~ p. Now,
p *(xm *, X1 *) = p *([XnmJ, [xn,]) = Lim p(Xnm Xn1)

n--+OO
in R. Since p*(xm, x,) < e/3 for m ~ p and t ~ p, it follows that there must
be q eN such that p(Xnm Xn1) < e/3 for ali n ~ q.
Suppose m ~ p and t ~ p and consider p(xim Xi 1). By the triangle
inequality,
But
1 1 6
p(Xim Xqm) <- <- <3
p(xqt Xi,) < - < - < -3

t p
and
Therefore p(xim Xi 1) <e for ali m ~ p and t ~ p, and thus {xim} is a

Cauchy sequence in (M, p). D
Metric Spaces
172
Theorem 6.4.13 The metric space (M*, p *) is complete.
Proof: By virtue of the previous two propositions we may assume that

if {xm *} is a Cauchy sequence in (M*, p *), then for each me N there is a
Cauchy sequence {xnm 1neN} in (M, p) such that Xm * = [Xnml and
p(xnl' xkm) < l/m. Furthermore, we may also assume that the sequence
(x 1m m eN} is a Cauchy sequence in (M, p).
Let e be a positive real number. Since {x 1m} is a Cauchy sequence,
there is some k 1eN such that p(x 1m, x 1,) < e/2 for ali m ~ k 1 and t ~ k 1.
Choose keN such that k >max {kt> 2/e}: then l/k <e/2 and
p(x 1m, x 1,) < e/2 for ali m ~ k and t ~ k. But then
We have thus shown that for any positive real number e there is some
k eN such that m ~ k and t ~ k imply p(x 1m, xm,) <e.
The Cauchy sequence {x 1m} in (M, p) defines a point x* = [x 1ml in
M*, and we claim that the Cauchy sequence {xm *} in (M*, p*) converges
to x*. ln arder to prove this, we must show that if e is a positive real
number, then there is some keN such that p*(x, *, x*) <e for t ~ k.
However, by the result proved in the preceding paragraph, there is some
keN such that p(x 1m, xm,) < e/2 for m ~ k and t ~ k. Therefore for t ~ k
we have
p*(x, *, x*) = p*([xm,l, [X1mD = Lim p(xmt X1m) <-28 <
n-+OO
and we are finished.
[J
We have thus shown that every metric space (M, p) is isometric to a

dense subset of a complete metric space (M*, p*). The metric space
(M*, p*), which was explicitly constructed from (M, p), is called the completion of (M, p). The completion (M*, p*) is essentially unique, in the
sense that if (M, p) is isometric to adense subset of another metric space
(MI> p 1), then (MI> p 1) must be isometric to (M*, p*). (See Exercise 7 at
Since the metric space Q of rational numbers is not complete and is
a dense subset of the complete metric space R, it is reasonable to ask if in
fact R is the completion of Q. The answer is yes: the procedure we have
173
used to construct the completion of a metric space can, with some modification, be used to construct R from Q. (Modification is required because
several of our definitions and proofs made use of real numbers, and of
course if we intend to construct R from Q, we cannot use real numbers in
so doing.) Thus, we begin with the number system Q, and assume that the
operations of arithmetic, the order relation <, and absolute value have
been defined in Q. We then define the concept of an open interval in Q and
allow the open intervals to generate the natural topology on Q in the usual
manner. Imitating the proof of Proposition 6.3.3, we show that a sequence
{xn} in the topological space Q converges to x e Q if and only if for every
qeQ, q >O, there is somem eN such that lxn - xi< q for ali n ~ m. Next
we define {xn} to be a Cauchy sequence in Q if for every q e Q, q > O, there
issomemeNsuch that lxn - xkl < qforall n ~ m, k ~ m. Nowitispossible
to define a relation R on the set of ali Cauchy sequences in Q by setting
{xn}R{yn} if the sequence {lxn - Ynl} converges to O in Q; as Example 5
of Examples 6.4.2 shows, this relation is an equivalence relation. Finally,
we define the set of real numbers to be the set of ali equivalence classes of
rational Cauchy sequences under the relation R.
Once the construction of R outlined above has been accomplished, it
is possible to define the usual operations of arithmetic, the order relation
<, and the concepts of convergent sequence and Cauchy sequence in R.
lt then requires only minor modifications of the relevant proofs of this
section to show that Q is isometric to a dense subset of R and that every
Cauchy sequence in R converges. We do not intend to pursue this topic any
further, but the interested reader may consult any one of a number of texts
on advanced real analysis. Incidentally, if R has been constructed both by
means of Dedekind cuts and as equivalence classes of rational Cauchy
sequences, then it becomes necessary to show that the two constructions
yield the sarne number system. This can be done by defining a one-to-one
function from the set of all equivalence classes of rational Cauchy sequences onto the set of ali Dedekind cuts, which preserves the arithmetic
operations and ordering. (An excellent source for the complete details of
the construction ofthe reais by both methods is C. Goffman, Real Analysis,
Holt, Rinehart and Winston.)
EXERCISES
1.
2.

Prove that every isometry from one metric space to another is a
homeomorphism.
174
3.
4.
5.
6.
Metric Spaces
Prove that the Cauchy sequences {xnm 1neN} and {xnm 1n ~ tm} in
the proof of Proposition 6.4.11 are equivalent.
Let Mbe adense subset ofthe metric space (M1, p 1) and M' be adense
subset of the metric space (M2 , p2 ). Prove that if (M, p 1) is isometric
to (M', p2 ), then (M1, p 1) is isometric to (M2 , p2 ). (Hint: if f is an
isometry from M to M', extendf to M 1 by defining
f(x) = Lim f(xn)

n--+OO
whenever {xn} is a sequence in M which converges to xeM 1.)

Prove that if the metric space (M, p) is isometric to dense subsets of
both the metric space (MI> p 1) and the metric space (M2 , p2 ), then
(M1> p 1) is isometric to (M2 , p2 ).
8. Suppose that R has been constructed from Q in the manner outlined
in Section 6.4, so that a real number is an equivalence class of Cauchy
sequences in Q, where by definition, {xn} is a Cauchy sequence in Q
if for every q e Q, q >O, there is some m eN such that n ~ m, k ~ m
imply lxn - xkl < q.
(a) Show that if {xn} and {Yn} are Cauchy sequences in Q, then sois
{xn + Yn}
(b) Define addition of real numbers (equivalence classes of rational
Cauchy sequences) as follows: if {xn} and {Yn} are Cauchy sequences in Q, with equivalence classes [xn] and [Yn], then
7.
Prove that addition of real numbers is well-defined by showing

that if [xn] = [rn] and [Ynl = [sn], then
(c) Show that if {xn} and {Yn} are Cauchy sequences in Q, then sois
{XnYn}
(d) Define multiplication of real numbers ( equivalence classes of
rational Cauchy sequences) as follows: if {xn} and {Yn} are
Cauchy sequences in Q, with equivalence classes [xn] and [Yn],
175
then
Prove that multiplication of real numbers is well-defined by

showing that if [xn] = [rn] and [Ynl = [sn], then
This page intentionally left blank
7
ln this chapter we examine the behavior of sequences of functions in

metric spaces. We define and illustrate the concepts of pointwise and
uniform convergence of a sequence of functions, show that certain properties of functions ( e.g., continuity) are preserved by uniform convergence, introduce and briefly study several metric spaces whose points are
functions, and finally prove two important approximation theorems for
functions. For simplicity, we will only consider functions from a metric
space to the real numbers, although many of our definitions and results
can be extended without difficulty to the more general setting offunctions
from one metric space to another.
7.1
POINTWISE AND UNIFORM CONVERGENCE
ln this section we define two types of convergence for sequences of functions and show how these preserve or fail to preserve certain properties.
We begin with the definition of pointwise convergence for a sequence of
functions.
Definition 7.1.1 Let (M, p) be a metric space, letfbe a function from M
to R, and for each n e N let f,,: M--+ R. We say that the sequence offunc-
tions {fn} converges pointwise to the function f if for each x e M the sequence of real numbers {f,,(x)} converges to the real number f(x).
177
178
Thus {fn} converges pointwise to f if Limn .... ,xJ(xn) = f(x) for all
xeM.
Examples 7.1.2 1. For each n eN, definefn: R-+ R by f,,(x) = x/n for all
xeR (see Figure 7.1). Since Limn--+cxJ(xn) = Limn--+co x/n =O for ali
xeR, the sequence offunctions {x/n} converges pointwise to the function
f defined by f(x) = O for ali x E R.
2. For each neN, definefn: R-+R by f,,(x) =xn for ali xeR (see
Figure 7.2). If x > 1, Limn--+cofix) = Limn--+co xn does not exist, so the
sequence {xn} does not converge pointwise.
3. Pointwise convergence of a. sequence of functions depends on the
metric space as well as on the functions. To see this, letf,,: [O, l] -+R be
given by f,,(x) = xn for ali n eN and x e [O, l] (see Figure 7.3). These are
the sarne functions as in the previous example, but here
Lim f,,(x) = Lim xn =
n-+oo
n-+oo
ifxe[O, 1)
if X= 1
hence the sequence {xn} converges pointwise to the functionf: [O, l] -+R
Figure 7.1
{f,.(x)}
= {x/n}
on R.
Pointwise and Uniform Convergence
179
Figure 7.2 {f,,(x)} = {xn} on R.
Figure 7.3 {f,,(x)} = {xn} on [O, 1).
180
defined by
f(x) =
{~
ifxe[O,l)
if X= 1
4. Let f,,: (O, + oo)-+ R be defined by f,,(x) = 1+x/(1 + nx) for all
n e N and all x > O ( see Figure 7.4). Since
X
l<l+--<l+-=l+l+nx
nx
n
we have
1
~ Limfn(x) ~ Lim (1+~)=1

n
H-+00
H-+00
Therefore the sequence {/,,}converges pointwise to the functionf defined

byf(x) = 1 for ali xe(O, +oo).
y
Figure 7.4 {f,,(x)} = { 1 + [x /(1
+ nx)]}
on (O,
+ oo ).
181
Figure 7.5 {f,,(x)} = {1/(1 +nx 2 )}.
5. Letfn: R-+ R be defined by f,,(x) = 1/( 1 + nx 2 ) for ali n eN and ali

xeR (see Figure 7.5). It is easy to check that if x =O, then
Limn-+ 00 fn(x) = 1, while if x =F O, then Limn-+ 00 f,,(x) = O. Thus the sequence ifn} converges pointwise to the functionf: R-+R given by
f(x)
= {~
if X =/= 0
ifx =0
Our first result in this section characterizes pointwise convergence in

terms of the absolute value metric on R.
Proposition 7.1.3 Let (M, p) be a metric space,f be a function from M
to R, and f,,: M-+ R for ali n e N. The sequence of functions {/,,} converges pointwise to f if and only if for each x e M and for each positive real
number e there is some me N such that lfnCx) - f(x) 1 < e whenever n ~ m.
Proof: The sequence {/,,} converges pointwise to f if and only if for each
xeM, Limn-+oof,,(x) =f(x). But Limn-+oofn(x) =f(x) if and only if for
every positive real number e there is some me N such that n ~ m implies
lfn(x) - f(x)I <e. O
Note that the proposition states that {/,,} converges pointwise to f if

for each xeM and eeR, e> O, there is some meN such that n ~m
implies IJ,,(x) - f(x)I <e; in general, then, m will depend on x as well as
182
on e. ln other words, for a particular e > O it may be necessary to choose

different values ofm for each xeM; on the other hand, it may be possible
to select a single value of m which will serve for ali x eM. To illustrate, let
us examine two of our previous examples using Proposition 7.1.3.
Examples 7.1.4 l. Let ifn} be the sequence {x/n} of Example l, Examples 7.1.2. Ifwe wish to use Proposition 7.1.3 to show that this sequence
converges pointwise to the zero function, then we must show that given
any e> O we can find m eN such that lx/n - OI= lxl/n <e whenever
n ';;:; m. Clearly, it suffices to choose m > lxl/e. Note that m depends on
both e and x.
2. Let ifn} be the sequence {1 + x /( 1 + nx)} of Example 4, Examples
7.1.2. If xeM ande> O, we seek m eN such that n ';;:; m implies
lfn(x)-f(x)I =11 + -x1 - li = -x1 <e
+nx
+nx
But since
X
l
n
---<1 +nx
for ali xe(O, +oo), if we choose meN such that m > 1/e, then n ';i:;m
implies
X
lfn(x) - f(x) 1 = -1 <- ~- <e

+nx n m
Note that here m does not depend on x but only on e: for a given e> O,
the sarne value of m will work for all xe(O, + oo).
Suppose the sequence {f,,} converges pointwise to the functionf As
we have just seen, the value of m guaranteed by Proposition 7.1.3 may
depend on e alone or on both e and x. This is reminiscent of the situation
encountered in Section 6.2 of Chapter 6, where we discussed continuity in
metric spaces. Justas we there defined a stronger form of continuity called
uniform continuity, so here we may define a stronger form of convergence
known as uniform convergence.
Definition 7.1.5 Let (M, p) be a metric space, letfbe a function from M
to R, and for each n e N let f,,: M-+ R. The sequence of functions {f,,}
183
converges uniformly to f on M if for every positive real number e there is

some meN such that n ~ m implies lfn(x) -f(x)I <e for ali xeM.
Thus {f,,} converges uniformly to f if, given e > O, there is some me N,
m depending only on e, such that lfn(x) - f(x) 1 < e whenever n ~ m, for ali
xeM.
Examples 7.1.6 1. Example 1 ofExamples 7.1.2 shows that the sequence
{x /n} converges pointwise to the function defined by f(x) = Ofor ali x e R,
while Example 1 of Examples 7.1.4 suggests ( but does not prove) that this
sequence does not converge uniformly to this function. Here we will prove
that the convergence is not uniform. To see this, suppose the sequence
{x/n} does converge uniformly to the zero function on R; then there is
some m e N, m depending only on e = 1, such that n ~ m implies
lfn(x) - f(x)I
= lxl/n <
for ali x e R. But this cannot be true for ali x e R, for if n = m and x = m,
then lxl/m = 1.
2. Example 2 of Examples 7.1.4 shows that the sequence
{1 + x /( 1 + nx)} converges uniformly on (O, + oo) to the function f
defined by f(x) = 1 for ali xe(O, +oo).
3. Uniform convergence depends on the metric space as well as on the
functions. For instance, let fn: [O, l]--+ R be given by fn(x) = x/n for ali
n e N and ali x e [O, 1]. These are the sarne functions as in Example 1
above, but here the sequence {x/n} converges uniformly to the zero function on [O, l]. To see this, suppose e >O is given and choose m > 1/e. Since
O~ x ~ 1, n ~ m implies that
X
1=-~-~-<e
X
1 1
lfix)-f(x)I= I--0
n
n n m
for ali xe[O, l], and hence {x/n} converges uniformly on [O, l].
4. The sequence of functions {1/(1 + nx 2)} of Example 5, Examples
7.1.2, does not converge uniformly on R to the function f defined by
f(x) =
{~
if X =/=
ifx =O
To see this, suppose the convergence is uniform; then there is somem eN

such that n ~ m implies lfn(x) - f(x)I <!for ali xeR. But it is easy to
check that this is impossible when n = m and O~ x < l/fo.
184
5. Let a eR, a> O, and for each n eN definef,.(x) = 1/( 1 + nx 2) for ali
+ oo ). The sequence {ln} converges uniformly to the zero function
+ oo ), because if me N, m > ( 1 - e)/a 2 , then n ~ m implies
x e [a,
on [a,
lfn(x) -
for ali x e [a,
OI = 1 + nx2 ~ 1 + mx2 ~ 1 + ma2 <e
+ oo ).
lt is clear that if a sequence of functions Un} converges uniformly to

f on a metric space, then the sequence also converges pointwise to f; on
the other hand, Examples 1 and 4 of Examples 7.1.6 show that pointwise
convergence does not imply uniform convergence. Thus uniform convergence is a stronger condition than pointwise convergence.
Application of the.definition of uniform convergence requires knowledge ofthe function/which is the limit of the sequence Un} lt would be
convenient to have a characterization of uniformly convergent sequences
which does not utilize the limit functionf The next result, known as the
Cauchy criterion for uniform convergence, provides such a characterization.
Proposition 7.1.7 (Cauchy Criterion) Let (M, p) be a metric space and
let f,.: M--+ R for ali n e N. The sequence Un} converges uniformly to a
function from M to R if and only if for every positive real number e there
is somem eN such that n ~ m and k ~ m imply lf.(x) - ,h(x)I <e for all
xeM.
If the sequence of functions Un} converges uniformly to f on M,

then for every positive real number e there is somem eN such that n ~ m
implies lfn(x) - /(x)I < e/2 for all xeM. Therefore n ~ m and k ~ m imply that
Proof:
lfn(x) -.h(x)I ~ lfn(x) -/(x)I + lfk(x) -/(x)I <
2 + 2 =e
for ali xeM.

Conversely, suppose that Un} is a sequence of functions from M to R
such that for every positive real number e there is some m eN such that
n ~ m and k ~ m imply lf.(x) -.h(x)I <e for ali xeM. For fixed xeM,
this implies that the sequence of real numbers Un(x)} is a Cauchy
185
sequence, and hence that Limn-+cofn(x) exists. Define/: M -+R by

f(x) = Lim fn(x)
n-+CO
for ali xeM. We claim that Un} converges uniformly to this functionf
If e eR, e >O, then there is some m eN such that n:;:::: m and k:;:::: m
imply lfn(x) - .fk(x)I < e/2 for ali xeM. For fixed k, k:;:::: m, and fixed
xeM, consider the sequence {IJ,,(x) - .fk(x)l neN}: since
Lim fn(x) = f(x)

n-+CO
and
IJ,,(x) - fk(x)I <
for n:;:::: m
we have
Thus if k:;:::: m, then lf(x) - .fk(x)I <e for ali xeM, which shows that ifn}
converges uniformly to f on M. D
Suppose {fn} is a sequence of functions which converges to f and f,,
has property P for each n e N. Will the limit function f necessarily have
property P? The answer is often no if the convergence is pointwise, but yes
if it is uniform. Thus uniformly convergent sequences of functions are
important because properties shared by the functions ofthe sequence tend
to be preserved in the limit, whereas this is frequently not the case for
pointwise convergence. We will illustrate by showing that the properties
ofboundedness aIJ.d continuity are preserved by uniform convergence, but
not by pointwise convergence.
First we must define what is meant by a bounded function from a
metric space to R.
Definition 7.1.8 If (M, p) is a metric space andfis a function from M
to R, thenf is bounded on M if f(M) is a bounded subset of R.
Now we prove that uniform convergence preserves boundedness.

Proposition 7.1.9 Let (M, p) be a metric space and let {/,,} be a sequence
of bounded functions from M to R. If {fn} converges uniformly on M to
the function f, then f is bounded on M.
Proof:
Since {/,,} converges uniformly to f on M, there is some m eN
186
such that n ~ m implies IJ(x) - fn(x)I < 1 for all x eM. Since fm is
bounded on M, there is some reR such that lfm(x)I < r for all xeM.
Therefore
IJ(x)I ~ IJ(x) - fm(x)I
+ lfm(x)I < 1 + r
for all xeM, and hencefis bounded on M, O

Example 7..10 This example shows that the conclusion of Proposition
7.1.9 does not hold if the convergence is pointwise but not uniform.
For each neN, definefn: [O, +oo)-+R as follows:
{~
fn(x) =
ifx ~n
if k e N and k - 1 ~ x < k
+ oo ), then there is some meN such that m - 1 ~ x < m, and it

follows that fn(x) = m for all n ~ m. Therefore if m - 1 ~ x < m, me N,
then Limn .... 00 ,f,,(x) = m, and hence {ln} converges pointwise to the functionf defined by f(x) = m if m - 1 ~ x < m, m eN. Note that each of the
functions f,, is bounded on [O, + oo) because O ~ fn(x) ~ n for all
x e [O, + oo ), but that the limit functionfis not bounded on [O, + oo ). The
reader should check that {f,,} does not converge uniformly to f
If x e [O,
We conclude this section by showing that uniform convergence preserves continuity while pointwise convergence need not e!~ so.
Proposition 7.1.11 Let (M, p) be a metric space and let {f,,} be a sequence of continuou~ functions from M to R. If {f,,} converges uniformly
on M to the functionf, thenfis continuous on M.
Proof: Since Un} converges uniformly to f on M, for every e e R, e > O,

there is somem eN such that n ~ m implies lfiy) - f(y)I < e/3 for all
yeM. If xeM, the continuity offm at x implies that there is a real number
>O such that
p(x,y) <
implies
Thus if p(x, y) < , then

IJ(x) - f(y)I ~ IJ(x) -fm(x)I + lfm(X) - m(Y)I + lfm(y) - f(y)I
6
<3+3+3=e
which proves that the limit function f is continuous at x. O
187
Example 7.1.12 This example shows that the conclusion of Proposition

7.1.l l does not hold if the convergence is pointwise but not uniform.
The sequence {f,,} of Example 5, Examples 7.1.2 and Example 4,
Examples 7.1.6 converges pointwise but not uniformly on R to the function f defined by
f(x)
= {~
if X =/=
ifx =0
It is clear thatf,, is continuous on R for all n eN but that/ is not continuous on R.

EXERCISES
l.
For each of the following, consider the sequence of functions Un}

defined on the given subset of R. If the sequence converges uniformly to a function f, find f and show that the convergence is
uniform; if the convergence is pointwise but not uniform, show this;
if the sequence does not converge, show this.
X
(a) fn(x) = 2n, on R

X
(b) fn(x) = 2n,
OI
[O, l]
(c) f,,(x) = nx, on [a, b]
nx
(d) f,,(x) = -1 - , on (O,

+nx
+ oo)
nx
(e) fn(x) = - , on [a, +oo), a>O

1+nx
xn
(f) fn(x) =1-- , on [O, l]
+xn
xn
(g) f,,(x) = -1 - , on [O, b], O< b < l
+xn
Let /: R -+R be uniformly continuous. For each neN, define
f,,(x) =f(x + l/n), for all xeR. Prove that the sequence {f,,} converges uniformly to f on R.
3. Let (M, p) be a metric space and for each n eN letfn be a function
from M to R. Prove that if Un} converges uniformly on a dense
subset of M, then it converges uniformly on M.
2.
188
4.
Let
x be the characteristic function of the rationals, defined by

x(x) =
5.
6.
7.
8.
9.
if xeQ
if x~Q
Prove that there is no sequence of continuous functions on R which

converges uniformly to X
Give an example of a sequence of functions {ln} which converges
pointwise to a bounded functionlbut does not converge uniformly
to f This shows that the converse of Proposition 7.1.9 does not
hold.
Give an example of a sequence of functions Un} which converges
pointwise to a continuous function I but does not converge uniformly to f This shows that the converse of Proposition 7.1.11 does
not hold.
Let (M, p) be a metric space and for ali n eNletln be a function from
M to R. Show that if Un} converges uniformly on each of the
subsets S 1, S2 , , Sk of M, then it converges uniformly on
U. ";" k S;. Give an example to show that the conclusion does not
hold for denumerable unions.
Let (M, p) be a metric space and for each n eN letln be a function
from M to R. The sequence {f,.} is uniformly bounded on M if there
is some r e R such that lfn(x) 1 ~ r for ali n e N and ali x e M.
(a) Prove that if {f,.} is uniformly bounded on M, then ln is
bounded on M for ali n e N.
(b) Give an example of a metric space (M, p) and a sequence Un}
which is not uniformly bounded on M even though f,. is
bounded on M for ali n e N.
Let (M, p) be a metric space and for each n eN let f,. be a function
from M to R. The sequence Un} is equicontinuous on M if for every
positive real number e there is a positive real number such that if
x e M and y e M with p(x, y) < , then lfn(x) - ln(Y) 1 < e for ali
neN.
(a) Prove that if {f,.} is equicontinuous on M, thenf,. is a continuous function from M to R for ali n eN.
(b) For each neN, define ln: [O, l] -+R by ln(x) = x/n for ali
xe[O, l]. Show that Un} is equicontinuous on [O, l].
(c) For each n eN, definef,.: [O, l]-+ R by f,.(x) = 1/( 1 + nx 2 ) for ali
xe[O, l]. Show that the sequence of continuous functions {f,.} is
uniformly bounded on [O, l] but not equicontinuous there.

10.
11.
189
Let (M, p) be a metric space and let Un} be a sequence ofcontinuous

function from M to R which converges uniformly on M. Prove that
if M is compact, then
(a) {f,,} is uniformly bounded on M;
(b) {f,,} is equicontinuous on M.
We utilize this exercise to prove the Arzela-Ascoli Theorem, which
is a partia! converse to the result of Exercise 10. The reader is asked
to fill in the details.
Arzela-Ascoli Theorem. Let (M, p) be a compact metric space. If

Un} is a sequence of functions from M to R which is uniformly
bounded and equicontinuous on M, then some subsequence of Un}
converges uniformly on M.
(a) Since (M, p) is compact, it contains a coup.table dense subset S.
(See Exercise 15 of Section 6.1, Chapter 6.)
(b) Let S = {x1> x 2, }. Since Un} is uniformly bounded, the sequence Un(x 1)} is bounded in R and hence has a convergent
subsequence
{fim(X1)} = {fi1(X1).fdx1), .}
(See Exercise 10, Section 5.2, Chapter 5.) Similarly, the sequence
{!1m(x2 )} is boqnded and therefore has a convergent subsequence
U2m(X2)} = {/21(X2),J;2(X2), .}
Thus we construct a family of subsequences of {f,,}, namely
each of which is a subsequence of the preceding one, and such

that Unm(xn)} converges in R, for ali n e N.
(c) Considerthesubsequence{fmm} of {f,,}. IfxneS, thenform >n
we have {fmm(xn)} a subsequence of the convergent sequence
Unm(xn) }, and therefore Umm(xn)} converges for m > n. But this
implies that the subsequence Umm} converges at every point
of S.
(d) We now use the Cauchy Criterion for uniform convergence, the
equicontinuity of Un}, and the fact that S is dense in M to show
that Umm} converges uniformly on S. But Exercise 3 above then
shows that Umm} converges uniformly on M.
190
7.2 FUNCTION SPACES ANO UNIFORM APPROXIMATIONt

ln this section we introduce several important metric spaces whose elements are functions. We also consider the approximation of a given function f by a sequence of functions which converges uniformly to f; our
major result in this regard will be the famous Weierstrass Approximation
Theorem.
It is possible to make the set of ali bounded functions from a metric
space to the reais into a metric space. ln order to do this, we need the
concept of the supremum norm of a bounded real-valued function.
Definition 7.2.1
Let (M, p) be a metric space and let

B(M) = {j: M--+ R 1f is bounded on M}
If f e B(M), define the supremum norm of f, denoted by li f li, as follows:
li! li= sup {lf(x)l xeM}
Proposition 7.2.2 Let (M, p) be a metric space. IffeB(M) and geB(M),

then
l.
2.
3.
4.
llf
11
11
>O;
f li = O if and only if f is the zero function from
M to R;
llr/ li= lrl li! li for ali reR;

li!+ gll ~li! li+ llgll
The proof of Proposition 7.2.2 will be left to the reader. ( See Exercise
l at the end of this section.)
Now suppose that for all/eB(M) and geB(M), we define
p*(f,g) = llJ-gll
then p* will be a metric on B(M). ln fact, not only is (B(M), p*) a metric
space, it is a complete metric space.
Proposition 7.2.3 Let (M, p) be a metric space. If p*: B(M) x B(M)--+ R

is defined by p*(f, g) =li! - gil for ali feB(M) and geB(M), then
(B(M), p*) is a complete metric space.
fThis section is optional.
191
Function Spaces and Uniform Approximation
Proof: As indicated above, the fact that (B(M), p *) is a metric space

follows from Proposition 7.2.2. We leave the details to the reader. (See
Exercise 2 at the end of this section.) To prove that (B(M), p*) is complete, we must show that every Cauchy sequence in (B(M), p*) converges
to an element of B(M).
If Un} is a Cauchy sequence in B(M), then for every positive real
number e there is some me N such that n ~ m and k ~ m imply
But
llfn - fk li= sup {lf,,(x) - fk(x)l xeM}

and thus llJ,, - fk li <e surely implies that lfn(x) - fk(x) 1 <e for ali x eM.
Therefore if {f,,} is a Cauchy sequence in (B(M), p *), then for every e e R,
e > O, there is some m e N such that n ~ m and k ~ m imply
lfn(x) - fk(x)I <e for ali xeM. But this is just the Cauchy Criterion for
uniform convergence, and hence we conclude that if Un} is a Cauchy
sequence in (B(M), p*), then it converges uniformly to some functionf
from M to R. Sincefn is bounded on M for ali n eN, so isf, by Proposition
7.1.9. ThereforefeB(M), and we are done. D
Corollary 7.2.4 A sequence Un} in the metric space B(M) converges
if and only if it converges uniformly as a sequence of functions from
MtoR.
The proof of the corollary is left as an exercise. (See Exercise 3 at the

Since uniform convergence preserves continuity in the limit, it follows
from Corollary 7.2.4 that the set consisting of ali continuous functions in
B(M) is also a complete metric space under p*.
Proposition 7.2.5 Let (M, p) be a metric space. If C(M) denotes the set
of ali continuous bounded functions from M to R, then (C(M), p*) is a
complete metric space.
The proof of the proposition is left to the reader. (See Exercise 4 at

The metric spaces (B(M), p*) and (C(M), p*) are referred to asfunction spaces, and the metric p* is the supremum metric. Study of these
spaces yields many interesting results concerning real-valued functions on
192
metric spaces. We will not pursue this topic any further, but merely
remark that function spaces are of considerable importance in higher
analysis.
Now we tum our attention to the uniform approximation of functions. Suppose Un} is a sequence of functions which converges uniformly to f on M. Then for any e e R, e > O, there is some m e N such that
n ~ m implies lf(x) - fn(x) 1 <e for ali x eM. Therefore by making e
small enough we can find a function f m which is arbi trarily elo se to f on
M. For this reason we say that the limit function f is uniformly approximated by the functions Un}. There are many approximation theorems of
the following form: let f be a function ha ving property P; then f can be
uniformly approximated by functions having property P'. This is the
sarne as saying that if f has property P, then there is a sequence of
functions having property P' which converges uniformly to f We will
prove two important approximation theorems of this type for continuous functions on R.
Definition 7.2.6 Let S be a subset of R and letfbe a function from S to
R. We say that f is a step function on S if S is a finite union of mutually
disjoint intervals or rays and f is constant on each of the intervals or

rays.
Example 7.2.7 Let /: R -+ R be defined by
if X~ -1
if-l<x<!
1
2
l 2~X ~ 1
if X> 1
Thenf is a step function on R. See Figure 7.6.
Our first approximation theorem says that a function which is continuous on a closed interval can be uniformly approximated there by
step functions.
Proposition 7.2.8 If/:[a, b]-+ R is continuous, then f can be uniformly
approximated on [a, b] by step functions.
Proof: We must produce a sequence {f,.} of step functions which converges uniformly to f on [a, b].
193
-1
-1
-2
-3
Figure 7.6 A step function.
For each neN, set

In1
b-a)
[ a, a +-n-
b-a
2(b-a))
l n2= [ a +
--,a+--n
n
l nn
=[
a+
(n - I)(b-a)
'
b]
Then for each n e N, [a, b] = Ui . _ k .._ n Ink and the intervals Ink are mutually
disjoint. For each k, 1 ~ k ~ n, let Xnk denote the midpoint of the interval
Ink For each neN, define/,,: [a, b]-+R by
if X E [nk> 1 ~ k
194
y
Figure 7.7 f,,(x) = f(xnk) on lnk
See Figure 7.7. Clearly each f,, is a step function on [a, b]. We will show
that the sequence {/,,} converges uniformly to f on [a, b].
Suppose e is a positive real number. Since f is continuous on the
compact set [a, b] it is uniformly continuous there, and hence there exists
a positive real number such that whenever x and y are in [a, b] with
lx -yl <
<>
then
lf(x) -f(y)I <e
Choose meNsuch that (b - a)/m <. For n;;:;: m, xElnk and yelnk imply
b-a
n
b-a
m
lx - YI < - - ~ - - <
<>
ln particular, this isso if y = Xnk the midpoint of Ink Therefore if n ;;:;: m,

then
IJ(x) - fn(x)I = IJ(x) - f(xnk)I <e
for ali xe[a, b], and we are done. D

Our final result of this chapter is the important Weierstrass Approximation Theorem, which says that a function continuous on a closed
195
interval can be uniformly approximated there by polynomials. Before we

can prove this theorem, we need several definitions and a proposition.
Recall the n ! (n factorial) is defined by
for ali neN
n ! = n(n - 1)(n - 2) 1
O!= 1
Also, the binomial coefficient
(n) =
k
for O ~ k
n!
k!(n -k)!
n, and the Binomial Theorem states that
for ali n eN.

We will need the following inequality in the proof of the Weierstrass
Approximation Theorem.
Proposition 7.2.9
For ali xe[O, l] and ali neN,
L
n
k=O
(n)
k
( k)2
xk(I -x)n-k X - -
1
~4n
The proof is left to the reader. (See Exercise 9 at the end of this
section.)
Next we define the polynomial functions which we will use to approximate a given continuous function.
Definition 7.2.10
Let /: [O, l]
-+ R.
For ali n eN, the polynomial
is the nth Bernstein polynomial for f

Example 7.2.11 The nth Bernstein polynomial for f is calculated from
the values of/ at the rational points O, I/n, 2/n, ... , 1. Let us find the first
-196
two Bernstein polynomials for f(x) = x 2 :
n = 1: Bn
n = 2: Bn
G)
(1-x)O +xi =x
(~) x(I -x) /(0) + G) x (1- x) /G)
x 0 (1 - x) 1/(0) +
C)
x 1(1 - x)f(l)
+G)x 2(1-x)f(l)
= (1 -
x) 20 + 2x(1 - x)! + x 2 I = 4x + !x 2
ln fact, it is easy to check that
1 (n - 1)
Bn =;;x + -n- x2
for ali n e N. Note that the sequence of Bernstein polynomials converges

uniformly to f(x) = x 2 on [O, l].
Now we are ready to prove the Weierstrass Approximation Theorem.
Theorem 7.2.12 (Weierstrass Approximation Theorem) Iff: [a, b] -+R

is continuous, then f can be uniformly approximated on [a, b] by polynomials.
Proof: We first prove the theorem for the case where [a, b] =[O, l].
Let /: [O, l]-+ R be continuous. We will show that the sequence of
Bernstein polynomials for f converges uniformly to f on [O, l]. Since
it will suffice to show that if e is a positive real number, then there is some
me N such that n ;;:;: m implies
for ali x e [O, l].
197
Since I is continuous on the compact set [O, l], it is both uniformly

continuous and bounded there. Hence for every e eR, e >O, there is some
eR, > O, such that lx - YI < implies ll(x) -l(Y)I < e/2, and also
there is some reR such that ll(x)I < r for all xe [O, l]. Choose neN such
that I/n <, and write
kto
(~)xk(I-x)n-kk(x)-1(~)1
=
~ G)xP(I -x)n-pk(x) -1(~)1

+ ~ (:)xq(I
-x)n-qk(x)
-1(~)1
where p ranges over ali values of k such that lx - (k/n)I <, and q ranges
over all values of k such that lx - (k/n)I ~. Because lx - (p/n)I < for
each p, we have ll(x) - l(p/n)I < e/2 for each p, and thus
But by the Binomial Theorem,
(n) xk(I -x)n-k = (x
k~O k
+ (1-x))n = 1
Therefore we have shown that if I/n <, then
for ali xe[O, l].

Now consider
Since ll(x)I < r for all xe[O, l], we have ll(x) -l(q/n)I < 2r, and hence
L (n) xq( 1 - x)n - qll(x) - j fj_)I < 2r L (n) xq(I - x)n-q (x - q/n) 2
q q
J\.n
q q
(x-q/n) 2
198
But lx -q/nl ~, so (x - q/n) 2 ~ 2 , and thus
But by the inequality of Proposition 7.2.9,
(2r) 7~ (n)
2
(2') _r
xq(l _ x)n-q(x _ Cf._) 2 ~

I_ =
n
2 4n 2n 2
and r/2n 2 < e/2 if l/n < 2e/r. Thus if we choose meN such that
1/m < min {, 2e/r}, what we have done shows that neN, n ~ m implies
Therefore the theorem is proved for the case where [a, b] =[O, l].
Now suppose /: [a, b] -+ R is continuous. Define
h(x) = f(a
+ (b -
a)x)
for ali x e [O, l]. Then h is a continuous function on [O, l], since it is the
composite ofthe continuous function g(x) =a+ (b - a)x with the continuous function f By what we have already shown, for every positive real
number e there is some polynomial P(x) such that lh(x) - P(x) 1 < e for ali
xe[O, l]. But then
p
(X -a)
b-a
is a polynomial and
for ali x e [a, b], so we are done. D

EXERCISES
1.
2.
3.

Prove that, as claimed in Proposition 7.2.3, (B(M), p*) is a metric
space.

4.
5.
199

Let (M, p) be a metric space and let a, b be real numbers with a< b.
Consider the subset C(a, b) of C(M) defined by
C(a, b) = {feC(M) 1a :;;;;,.f(x):;;;;,. b for ali xeM}
6.
7.
Prove that (C(a, b), p*) is a complete metric space. (Hint: use the
result of Exercise 3 of Section 6.3.)
Find a sequence of step functions which converges uniformly to f on
[O, l] if
(a) f(x) = x for ali xe[O, l];
(b) f(x) =x 2 for ali xe[O, l].
A function g: [a, b]-+ Ris piecewise linear on [a, b] if there exist points
t0 , t1o . .. , tn and real numbers r0 , r1o ... , rno s0 , s1o . .. , sn such that
and
8.
9.
Prove that if /: [a, b] -+ R is continuous, then f can be uniformly

approximated on [a, b] by piecewise linear functions.
Find a sequence of polynomials which converges uniformly to f on
the indicated interval:
(a) f(x) = lxl for ali xe[-1, l];
(b) f(x) = l/x for ali xe[l, 2].
Prove Proposition 7.2.9. (Hint: write
2
I (n)k xk(l -x)n-k(x - ~)
n
k=O
and show that the three terms on the right-hand side reduce to x 2,
-2x 2 , and x/n + [(n - l)x 2]/n, respectively.)
8
Calculus
ln this chapter we will study the basic concepts of calculus, the definite
integral and the derivative. We will not be concerned with techniques of
integration and differentiation, but rather with the theory of integration
and differentiation. We begin with a section devoted to the definition of
the Riemann integral and integrable functions, proceed to consider the
properties of the Riemann integral, and conclude with an examination of
the derivative and its properties.
8.1
THE RIEMANN INTEGRAL
ln this section we will define the integral of calculus, which is known as the
Riemann integral, and discuss briefly the question of the Riemann integrability of certain functions. Before we can define the integral, we need
some preliminary results about partitions of intervals and Riemann sums.
Definition 8.1.1 Let [a, b] be a closed interval in R. A partition of [a, b] is

a finite set of real numbers
such that
a = x0 < x 1 < <
201
Xn
= b
Calculus
202
The closed intervals

l~k~n
are called subintervals of the partition P. If P and P' are both partitions
of [a, b] and P is a proper subset of P', we say that P' is a refinement
of P.
Example 8.1.2
Let [a, b] =[O, l] and let

. { l 1 }
P= '4'2' 1
and
, {o'4'2'
I I I i}
J2'
p =
The sets P, P', and P" are ali partitions of [O, l]. The partition P' is a
refinement of P and P" is also a refinement of P; however, P' is not a
refinement of P", nor is P" a refinement of P'.
Now we define the upper and lower Riemann sums of a bounded
function over a partition.
Definition 8.1.3
Let a, b be real numbers, with a < b, and let

P
={a= x0 , X1o
, Xn
= b}
be a partition of [a, b]. If /: [a, b] -+ R is bounded, define the upper

Riemann sum U(f, P) of f over P as follows:
n
U(f,P)=
sup{f(x) lxe[xk_ 1,xk]}(xk-xk_ 1)
k=I
Similarly, define the lower Riemann sum L(f, P) of f over P as follows:

n
L( f, P) =
L
k=l
inf {f(x) 1 x e [xk _ 1o xk]}(xk - xk _ 1)
The Riemann Integral
203
U(f,P) is the sum of the areas of the rectangles
Figure 8.1
An upper Riemann sum for f
Note that sincef is bounded on [a, b] both U(f, P) and L(f, P) exist
as finite real numbers for any partition P of [a, b]. Furthermore, it is clear
from the definition that L(f, P) ~ U(f, P) for any partition P of [a, b].
Iff is a continuous function on [a, b], then U(f, P) is the sum of the
areas of the rectangles whose bases are the subintervals [xk _ 1, xk] and
whose heights are equal to the maximum values which f attains on the
subintervals (see Figure 8.1). Similarly, if f is continuous on [a, b], then
L(f, P) is the sum of the areas of the rectangles whose bases are the
subintervals [xk _ 1, xk] and whose heights are equal to the minimum values whichf attains on the subintervals (see Figure 8.2). Notice, however,
that Definition "8.1.3 does not require thatf be continuous on [a, b], but
only that it be bounded there.
Examples 8.1.4 1. Let e eR and letfbe the constant function defined by
f(x) =e for ali xe[a, b]. If P ={a= x0 , x., ... , Xn = b} is any partition
of [a, b], then
n
U(f, P) =
c(xk -xk_ 1)
= c(xn -x0) = c(b - a)
c(xk...- xk_ 1)
= c(xn - x0) = c(b - a)
k=I
and
n
L(f, P) =
L
k=I
204
Calculus
y
b
L(f,P) is the sum of the areas of the rectangles
Figure 8.2 A lower Riemann sum for f
2. Let f: [a, b] -+R be given by f(x) = x for all xe[a, b]. For each
n eN let Pn be the partition

{
b- a
2(b - a)
(n - l)(b - a) b}
Pn a, a +
,a+
, ... , a +
,
Then
U(f, Pn) =
f
k=I
(a+ k(b - a))(b n

n
b- a
=--na+
n
(b - a) 2
= b- a
k=I
k=I
Using the well-known identity

1 + 2 + ... + n = _n(_n_+_I)
2
for the sum of the first n natural numbers, we see that
U(f, Pn) = a(b - a)+ (b
~2a)2 n(n; 1)
1
1
=-(b 2 -a 2) +-(b-a) 2
2n
a+ (b -2a)2
k
n
k=I
205
Similarly,
3. Letfbe defined on [O, l] by

f(x) =
ifO ~x < !
{~
if!~x ~ 1
If P = {O= x 0 , x., ... , Xn = 1} is any partition of [O, l], there is some
ieN, 1 ~ i ~ n, such that X;_ 1 ~!<X;. Thus

i-1
U(f, P) =
O(xk -xk_ 1)
+L
l(xk -xk-1)
k=i
k=l
=Xn -X;-1 =
-X;-1
Similarly, there is somejeN, 1 ~j ~ n, such that xi_ 1 <!~xi, and thus

j
L(f, P)
O(xk - xk_ 1)
+ L
l(xk - xk_ 1)
k=j+l
k=l
= Xn - xi = 1 - xi
4. Letf be the function defined on [O, l] by

if x # ! for any n e N
n
if x =!for some neN
n
For any partition P of [O, l] we will have L(f, P) =O, because every
subinterval of P must contain a point x such that f(x) =O. Now let us
consider upper Riemann sums. We claim that for every positive real
number e we can find a partition P. of [O, l] such that O< U(f, PJ <e.
We may assume that e< 1, so that there is some meN such that
1
e 1
0<--<-<m+ 1 2 m
Calculus
206
It is possible to choose eR, >O, such that <e/( 4m) and also such
that
for l
m unless n = k
(Why?) Let
el
l
l
l
}
p = { o, -2 -m - , -m + , - . - , - . + , ... ' 1 - , l
mmWe have
sup {/(x) 1x e [
~ - , ~ + ]} = ~
~k ~m
sup {J(x) 1xe[l -, 1] = l

and sup {J(x)} = O on the remaining subintervals of P. Therefore
o< U(f, P.)
=--.1
m
e
-2+
(ml)
L -k 2 + l
k=2
<-+
L - =-+-=e
2 k= 1 2m 2 2
5. Let x denote the characteristic function of the rationals restricted
to [O, l], so that
x(x) =
{~
if x e [O, l] is rational
if x e [O, l] is irrational
If P is any partition of [O, l], then every subinterval of P contains both
rational and irrational numbers, and hence

sup {x(x)} = l
and
inf {x(x)} = O
207
on every subinterval of P. Therefore U(x, P) = 1 and L(x, P) =O for every

partition P of [O, l].
We need to show that for a given bounded function, every lower
Riemann sum is less than or equal to every upper Riemann sum. This
result will appear as a corollary to the following proposition.
Proposition 8.1.5 Let f be a bounded real-valued function on [a, b],
where a < b. If P and P' are partitions of [a, b] with P' a refinement of P,
then
L(f, P)
L(f, P')
U(f, P')
U(f, P)
Proof: We have already observed that, as a consequence of the definition of upper and lower Riemann sums, the lower Riemann sum over a
given partition is less than or equal to the upper Riemann sum over the
sarne partition. Therefore
L(f, P')
U(f, P')
and thus it will suffice to show that

L(f, P)
and
L(f, P')
U(f, P')
U(f, P)
Let us first consider the special case in which P' has been obtained by
adding one point to P; say
P ={a= x0 ,
X1o , Xn
= b}
and
P'
={a= x0 , x 1,
, X;-1'
y, X;,
. . . , Xn
= b}
lt is clear that
and
sup{f(x) lxe[y,x;]}~sup{f(x) lxe[x;_ 1,xJ}
Calculus
208
Hence
n
U( f, P') =
sup {f(x) 1 x e [xk _ 1> xkl}(xk - xk _ 1)
k=I
k-Fi
+ sup {f(x)
+ sup {f(x)
1x e [xi- 1> yl}(y - xi- 1)

1x e [y, x;l}(xi - y)
sup {f(x) 1 x e [xk _ 1> xkl}(xk - xk _ 1)
k=I
k"" i
+ sup {f(x)
+ sup {f(x)
1xe [x;-1> xJ}(y - X;_ 1)

1x e [x; _ 1> xJ }(x; - y)
sup{f(x) lxe[xk_ 1,xkl}(xk-xk-1)
k=I
= U(f, P)
We have thus shown that if P' is a refinement of P obtained by

adding one point to P, then U(f, P') ~ U(f, P), and a similar calculation establishes that L(f, P) ~ L(f, P'). (See Exercise 2 at the end of this
section.) But every refinement of P can be obtained by adding finitely
many points to P, so it follows by induction that if P' is any refinement
of P, then
U(f, P')
U(f, P)
and
L(f, P)
L(f, P')
We leave the details of the induction argument to the reader. (See Exercise 3 at the end of this section.) D
Corollary 8.1.6 Letfbe a bounded real-valued function on [a, b], where

a < b. lf P and P' are any partitions of [a, b], then L(f, P) ~ U(f, P').
Proof: If P and P' are the sarne partition, the conclusion is clearly true.
lf P and P' are distinct partitions, then P u P' is a refinement of P and
also a refinement of P', and hence by the proposition
L(f, P)
L(f, P u P')
U(f, P u P')
U(f, P') D
Thus for a given bounded functionf, every lower Riemann sum is less
209
than or equal to every upper Riemann sum, and it follows that

{U(f, P) 1 P is a partition of [a, b]}
is bounded below and therefore has a greatest lower bound, and that
{L(f, P) 1 Pisa partition of [a, b]}
is bounded above and therefore has a least upper bound.

Definition 8.1.7 Let f be a bounded real-valued function defined on
[a, b], where a < b. Set
sup L(f, P)
p
sup {L(f, P) 1 P is a partition of [a, b]}
and
inf U(f, P) = inf {U(f, P) 1P is a partition of [a, b]}
p
lt is a very important fact about Riemann sums that
sup L(f, P)
p
inf U(f, P)
p
Proposition 8.1.8 Iff is a bounded real-valued function defined on [a, b],

where a < b, then
sup L(f, P)
p
Proof:
inf U(f, P)
p
If
sup L(f, P) > inf U(f, P)

p
then by virtue of Definition 8.1. 7 there is some partition P of [a, b] such

that
L(f, P) > inf U(f, P)
p
But then, again by virtue of Definition 8.1. 7, there is some partition P'
such that
L(f, P) > U(f, P')
However, this last inequality contradicts Corollary 8.1.6. O
Calculus
210
Since
sup L(f, P)
p
inf U(f, P)
p
always holds, we may single out those bounded functions f for which
sup L(f, P) = inf U(f, P)
p
Such functions are said to be Riemann integrable.
Definition 8.1.9 Letfbe a bounded real-valued function defined on the

closed interval [a, b], where a < b. If
p
on [a, b], thenf is Riemann integrable on [a, b], and the Riemann integral
of f on [a, b], denoted by
is defined by
b f dx =
Ja

P
The Riemann integral is the definite integral studied in calculus. There

are more general definite integrais, one of which, the Riemann-Stieltjes
integral, is introduced in the exercises at the end of this section. We also
remark here that as far as we are concerned the dx which appears in
does not stand for any specific quantity, but is merely part of the notation
for the Riemann integral. When dealing with a more general integral than
the Riemann integral, the dx is replaced by another, more appropriate,
symbol. Thus the function of the dx is to tell us that we are dealing with
the Riemann integral of f rather than one of the more general integrais
off
211
The reader should note that the Riemann integral as defined above
has no connection with derivatives. As we shall see in Section 8.3, it is
indeed true that for some Riemann integrable functions f, it is possible to
calculate the value of
by finding a function whose deriva tive is/ and evaluating this function at
a and b. However, this is nota definition of the Riemann integral, but a
theorem about it. Furthermore, there are many Riemann integrable functions for which this theorem does not hold.
We have not yet demonstrated that there are any Riemann integrable
functions. The first example below shows that every constant function
from a closed interval to R is Riemann integrable.
Examples 8.1.10 1. Let ceR and/be defined on [a, b] by f(x) =e for ali
x e [a, b]. From Example 1 of Examples 8.1.4,
U(f, P) = L(f, P) = c(b - a)
for every partition P of [a, b], and therefore

sup L(f, P) = inf U(f, P) = c(b - a)
p
Hence/is Riemann integrable on [a, b] and
r r
f dx =
e dx = c(b - a)
2. Let/be defined on [a, b] by f(x) = x for ali x e [a, b]. From Example 2 of Examples 8.1.4, we know that for each n e N we can find a
partition Pn such that
and
212
Calculus
Therefore
for all neN and
for all n e N. But these inequalities imply that

inf U(f, P)
p
~ -21 (b 2 -
a 2)
~ sup
L(f, P)
p
and since
sup L(f, P)
p
inf U(f, P)
p
always holds by Proposition 8.1.8, we conclude that

inf U(f, P) = sup U(f, P) = -21 (b 2 - a 2)
p
p
Hence f is Riemann integrable on [a, b] and
ibf ib
dx =
x dx = -1 (b 2 - a 2)
3. Letf be defined on [O, l] by

f(x) =
{~
ifO~x <!
if!~x ~ 1
Let P be any partition of [O, l] such that xk =!for some k. Example 3 of

Examples 8.1.4 shows that
and
L(f, P) = 1 - xk =
213
Therefore
and
Again, since
sup L(f, P)
p
inf U(f, P)
p
always holds, it follows that

sup L(f, P) = inf U(f, P) = -21
p
p
Hence f is Riemann integrable on [O, l] and
l
fdx=-
4. Let/be the function defined on [O, l] by

if x :; ! for any n e N
n
if x =!for some neN

n
From Example 4 of Examples 8.1.4, L(f, P) = O for every partition P of

[O, l], and also for every positive real number e there is some partition P.
such that
O< U(f, P.) <e
lt follows that
sup L(f, P) =O= inf U(f, P)

p
Hence f is Riemann integrable on [O, l] and
5. Let
x denote the characteristic function of the rationals, restricted
214
Calculus
to [O, l]. From Example 5 of Examples 8.1.4, for every partition P of [O, l]
we have U(x, P) = 1 and L(x, P) = O. Thus
inf U(x, P) = 1
p
so the function
sup L(x, P) = O
and
x is not Riemann integrable on [O, l].
Let us consider the question of the integrability of a real-valued function f defined on a closed interval [a, b]. What must we do to show thatf
is Riemann integrable on [a, b]? We must first prove that it is bounded on
[a, b], for if it is not, we cannot form its Riemann sums and it certainly is
not Riemann integrable. Having established thatf is bounded on [a, b], it
then suffices to show that
inf U(f, P)
p
sup L(f, P)
p
since this, together with the result of Proposition 8.1.8, implies that
inf U(f, P)
p
= sup
L(f, P)
p
and hence thatfis Riemann integrable on [a, b]. The task ofshowing that
inf U(f, P)
p
sup L(f, P)
p
can be simplified somewhat, as our next result demonstrates.
'Proposition 8.1.11 If f is a bounded real-valued function defined on

[a, b], where a < b, then/ is Riemann integrable on [a, b] if and only if for
every positive real number e there is some partition P. of [a, b] such that
U(f, P.) - L(f, P.) <e
Proof:
Let e be a positive real number. Because

inf U(f, P) = inf {U(f, P) 1 P a partition of [a, b]}
p
there must be a partition P' of [a, b] such that
U(f, P') <
i~f U(f, P) + ~
215
Similarly, there must be a partition P" of [a, b] such that

L(f, P") >
s~p
L(f, P) -
Let P. = P'uP". Since P. is a refinement of both P' and P", it follows

that
~
U(f, P')
and
L(f, P.)
i~f U(f, P) + ~
and
L(f, P.) > s~p L(f, P)
U(f, P.)
L(f, P")
Hence
U(f, P.) <
-2
But if/is Riemann integrable on [a, b] then infp U(f, P) = suppL(f, P),
so
U(f, P) -L(f, P) <
2 +2 =e
To prove the converse, suppose that for every e e R, e > O, there is

some partition P. such that
U(f, P) - L(f, P) <e
then
inf U(f, P)
p
U(f, P.) < L(f, P.) +e
sup L(f, P)
p
+e
Thus
inf U(f, P) < sup L(f, P)
p
+e
for every positive real number e, and it follows that

inf U(f, P)
p
sup L(f, P)
p
and hence thatf is Riemann integrable on [a, b]. D

Every function continuous on a closed interval [a, b] should be
Riemann integrable ther~: its upper Riemann sums are approximations
Calculus
216
from above the graph of the function to the area between the graph and
the x-axis from x =a to x = b (see Figure 8.1), its lower Riemann sums
are approximations from below the graph to this sarne area (see Figure
8.2), and these approximations can be made to come arbitrarily close to
the true value ofthe area, and hence to each other, by taking the partition
fine enough. Therefore for a continuous functionf, it should be the case
that
inf U(f, P) = sup L(f, P)
p
and thus thatf is Riemann integrable on [a, b]. We now prove this.
Proposition 8.1.12 Every continuous real-valued function defined on a
closed interval [a, b], where a< b, is Riemann integrable on [a, b].
Proof: Letfbe a continuous real-valued function defined on [a, b]. Since
/is continuous on the co~pact set [a, b], it is both bounded and uniformly
continuous on [a, b]. Since f is uniformly continuous on [a, b], for every
positive real number e there is some positive real number such that
xe[a, b], ye[a, b], lx -
YI <
implies lf(x) - f(y) 1 < b
~a
Let P. = {a = x 0 , x 1, , Xn = b} be a partition of [a, b] such that

xk - xk _ 1 < for ali k, 1 ~ k ~ n. Since f is continuous on [xk _ 1, xk], it
attains its maximum and minimum there, and hence for each k, 1 ~ k ~ n,
there are points a.k, Pk in [xk _ 1, xk] such that
and
Thus
n
U(f, P.) - L(f, P.) =
L
k=I
(f(Pk) - f(a.k))(xk - xk-1)

for 1 ~ k
217
n, and therefore
n
U(f, P.) - L(f, P.) < k~ 1 b _a (xk - xk _ 1) = b _ a (b - a) = e
Hence f is Riemann integrable by Proposition 8.1.11. O

We have thus established that continuity of a function on a closed
interval is a sufficient condition for its Riemann integrability on the interval. Continuity is not a necessary condition for Riemann integrability,
however; Examples 3 and 4 ofExamples 8.1.10 show that a function may
be discontinuous at a finite or even a denumerable number of points and
still be Riemann integrable. On the other hand, the mere fact of boundedness is not sufficient for Riemann integrability, as Example 5 of Examples
8.1.1 O shows. These examples and Proposition 8. l.12 suggest that necessary and sufficient conditions for the Riemann integrability of a function
will involve its boundedness together with some other condition concerning the set of its discontinuities. ln fact, it turns out that a function is
Riemann integrable on a closed interval if and only if it is bounded there
and its set of discontinuities in the interval has measure zero. Hre the
concept of the measure of a set is a generalization of the idea of length: the
measure of an interval is its length, and ali finite and denumerable sets
have measure zero, as do some uncountable sets (such as the Cantor set,
for instance). We will not prove this characterization of Riemann integrable functions here, but instead refer the reader to Exercises 13-16 at
the end ofthis section, where the relevant definitions are given anda proof
is outlined.
EXERCISES
l.
Letf, g, and h be defined on [O, l] as follows:
f(x) = x 2 for ali x e [O, l]

g(x) = {
h(x)
-1
1
= {~
ifO~x ~t
ift<x~l
if x is irrational, x e [O, l]
if x is rational, x e [O, 1]
Calculus
218
2.
3.
4.
Let neN and let Pn ={O, l/n, 2/n, ... , (n -1)/n, l}. Write out the
upper and lower Riemann sums U(f, Pn), L(f, Pn), U(g, Pn),
L(g, Pn), U(h, Pn), and L(h, Pn).
ln Proposition 8.1.5, show that L(f, P) ~ L(f, P').
Carry out the induction argument which completes the proof of
Proposition 8.1.5.
Using only the definition of the Riemann integral and the identity
2
m(m + 1)(2m + 1)
+ 22 ++m 2 =
--6- - -
show thatf(x) = x 2 is Riemann integrable on [a, b] and evaluate
5.
Letfbe defined on [O, l] by
f(x) = {
-1
ifO ~x <
r 1
3
1 3~X<5
iq~x ~ 1
Show that f is Riemann integrable on [O, l] and evaluate

Ifdx
6.
Let Qn[O, l] = {q 1, q2 ,
}.
For each neN, let
if xe{q1> q2, ... , qn}

if xe[O, l] - {q1> q2,. , qn}
(Thus Xn is the characteristic function ofthe set {q 1, , qn}.) Show
that Xn is Riemann integrable on [O, l] and evaluate
Xndx
Now show that the sequence of functions {Xn 1 n eN} converges

pointwise on [O, l] to x, the characteristic function of the rationals.
7.
Since x is not Riemann integrable (Example 5 of Examples 8.1.10),

this demonstrates that the pointwise limit of a sequence of
Riemann integrable functions need not be Riemann integrable.
Letfbe defined on [a, b] by
f(x) =
8.
219
{~
if x is irrational, x E [a, b]
if x is rational, x E [a, b]
Prove thatf is not Riemann integrable on [a, b].

Letf be defined on [O, 1] by /(O) = l,f(x) =O if x is irrational, and
f(x) = 1/n if x = m/n, where m, n are natural numbers which have
no common factor. Prove thatf is Riemann integrable on [O, l] and
evaluate
ffdx
9.
Prove that if f is Riemann integrable on [a, b], and g is obtained

from f by changing the values of f at finitely many points of [a, b],
then g is Riemann integrable on [a, b] and
rfdx=
gdx
(Hint: prove for the case where g differs from f at only one point,
then use induction.)
10. Let f be a step function defined on [a, b], where a < b. (Thus
[a, b] is a union of mutually disjoint intervals / 1, , Im and
f(x) = ck for ali xelk.) Prove thatf is Riemann integrable on [a, b]
and evaluate
11.
Prove that a real-valued functionf defined on [a, b], where a< b, is

Riemann integrable on [a, b] if and only if for every e >O there are
step functions g. and h. defined on [a, b] such that
g.(x)
~f(x) ~
h.(x)
for ali x E [a, b]
220
Calculus
and
(h. - g.) dx < e
12.
This result is sometimes used as the definition of Riemann integrability for a function f
Let P ={a = x0 , xt> . .. , Xn = b} be a partition of [a, b], where
a < b, and let f: [a, b]--+ R be bounded. Define
n
R(f, P) =
L f(yk)(xk -
xk _ 1)
k=l
where for each k, l ~ k ~ n, Yk may be any point in the subinterval

[xk _ 1, xk]. Such a sum is called a Riemann sum, and its value depends on the choice ofboth P and the points Yk (see Figure 8.3, and
contrast it with Figures 8.1 and 8.2). Show that f is Riemann integrable on [a, b] with Riemann integral equal to oc if and only if for
every positive real number e there is a partition P. of [a, b] such that
if P is any refinement of P., then
IR(f,P)-oci<e
(Hint: P = P. u {y 1,
Yn} is a refinement of P. )
a
!I
Xo
Xn-2
li
Yn-1
R(f,P) is the sum of the areas of the rectangles
Figure 8.3 A Riemann sum for f
Yn b
li
Xn
221
This result is sometimes used as the definition of Riemann integrability for a function f
We utilize Exercises 13-16 to prove that a function is Riemann integrable on a closed interval if and only if it is bounded there and its set of
discontinuities within the interval is a set of measure zero.
13. If I is an interval in R, then I =(a, b), I =[a, b], I =(a, b], or
I = [a, b), and the length of I is defined to bem(/) = b - a. A subset
S of R is said to be a set of measure zero if for every positive real
number e there exists a collection of open intervals {/"' 1 tXeA} such
that (i) A is a countable set; (ii) Se UoceA /"'; and (iii)
I:oceA m(l") <e. Note that the collection of open intervals {!"' 1 tX e A}
can be either finite or denumerable. Prove:
(a) If S is a finite subset of R, then S is a set of measure zero;
(b) If S is the union of countably many sets of measure zero, then
sisa set ofmeasure zero. (Hint: if s = Un Sm then Sn e
Ink
where I:k m(lnk) < e/2n.)
(e) If S is a countable subset of R, then S is a set of measure zero;
(d) If T is a subset of S and S is a set of measure zero, then T is a
set of measure zero;
(e) If S contains an open interval, then S is not a set of measure
zero.
14. Letfbe a bounded function on [a, b], where a < b. If I is any interval
contained in [a, b], define w1 (I), the oscillation of f over /, by
uk
wJ(l)
sup {f(x) 1xe/} - inf {f(x) 1xe/}
If te [a, b], define w1 (t), the oscillation off at t, by
w1 (t) = inf {wJ{/) 1I an open interval containing t}
15.
Prove that
(a) lffis continuous at t, then w1 (t) =O;
(b) Iff is discontinuous at t, then w1 (t) >O;
(e) If Sn = {te(a, b] 1W1(t);;;.:: l/n}, then Sn is ciosed. (Hint: if ris a
limit point of Sm any open interval I containing r must contain
a point s of Sm and then w1 (1) ;;;.:: w1 (s) ;;;.:: l/n.)
Letfbe a real-valued function defined on [a, b], where a < b. Prove
that if f is Riemann integrable on [a, b], then the set of discontinuities of f within [a, b] is a set of measure zero.
222
Calculus
(a) Let Sn ={te [a, b] 1 mj(t) ~ l/n }. Show that Sn is a set of
measure zero:
(i) There is a partition P of [a, b] such that
U( f, P) - L( f, P) < 2n
(Why?) Show that

n
m1([xk- i xk])(xk - xk _ i) = U( f, P) - L( f, P)
k=i
(ii) Show that Sn =Ti u T2 , where
Ti= {xeSn 1 xeP}

and
T2 = {xeSn 1 x is in the interior of a subinterval of P}

(iii) Show that Ti can be covered by a union of open intervals
the sum of whose lengths is less than e/2.

(iv) Now consider T 2 : let Jki , Jk, be the subintervals of P
whose interiors cover T 2 If te T2 , show that
and then use this and the result of (i) above to show that
and thus that
(v) Use the results of (iii) and (iv) above to prove that Sn is a
set of measure zero.
223
16.
(b) Show that the set of discontinuities offwithin [a, b] is the union
of the sets Sm n e N, and then conclude that the set of discontinuities of f is a set of measure zero.
Letfbe a real-valued function defined on [a, b], where a <b. Prove
that iff is bounded on [a, b] and its set of discontinuities on [a, b] is
a set of measure zero, then f is Riemann integrable on [a, b].
(a) Show that we may assume that wf([a, b]) >O.
(b) Choose n eN such that (b - a)/n < e/2, and define
sn = {re[a, b] 1 wf(t) ~~}

Show that there exists a countable collection of open intervals
{J"' 1 tXeA} such that
e
L
m(J"')
<
2
([
ixeA
wf a, b])
(e) Show that a finite number of the open intervals {J"'} of (b)
above must cover Sn. Let Jk 1, , Jk, be these open intervals,
and show that
where J.i is a closed interval for 1 ~j ~ s.

(d) Show that for each x e J.i there is an open interval Kx which
contains x and such that wf(Kx) < 1/n, and then show that a
finite number of the open intervals Kx must cover J.i.
(e) Let Pi be a partition of J.i whose points are the endpoints of the
finite number of open intervals Kx which cover J.i. Show that
m(J.i)
U( f, P) - L( f, P) < -
(f) Define a partition P of [a, b] by setting
Show that the subintervals of P are the subintervals of Pi, for
Calculus
224
1 ~j
that
17.
s, together with the intervals Jk 1,
Jk, and then show
We utilize this exercise to introduce the Riemann-Stieltjes integral.

Let f, g be bounded real-valued functions defined on [a, b], where
a< b. Let
P ={a= x 0, xh ... , Xn = b}
be a partition of [a, b]. Define the Riemann-Stieltjes sum for f with

respect to g over P, denoted R(f, g, P), as follows:
n
R(f, g, P) =
f(yk)(g(xk) - g(xk - 1))
k~I
where for each k, 1 ~ k ~ n, Yk may be any point of the subinterval

[xk-1' xk] (compare Exercise 12 above). Define f to be RiemannStieltjes integrable with respect to g on [a, b], with Riemann-Stieltjes
integral tX, if there is a real number tX such that for every e e R, e > O,
there is some partition P. of [a, b] such that if P is any refinement of
P., then IR(f, g, P) - tXI <e. If/is Riemann-Stieltjes integrable with
respect to g on [a, b], then the Riemann-Stieltjes integral of f with
respect to g over [a, b] is
'
The Riemann-Stieltjes integral is a generalization of the Riemann

integral which reduces to the Riemann integral when g(x) is the
identity function on [a, b] (compare Exercise 12 above).
(a) Let
f(x) = g(x) =
{~
if ~X< t
ift~x~ 1
Prove thatf is not Riemann-Stieltjes integrable with respect to
g on [O, l].
225

(b) Let/ be continuous on [O, l] and let
g(x) =
ifO~x<!
if!~x~ 1
{~
Prove that/ is Riemann-Stieltjes integrable with respect to g on

[O, l] and show that
(e) Let g be a constant function on [a, b]. Prove that if f is any

bounded function on [a, b], then f is Riemann-Stieltjes integrable with respect to g on [a, b] and
(d) Let/(x) = 1 for all xe[a, b] and let g(x) be bounded on [a, b].
Prove that/ is Riemann-Stieltjes integrable with respect to g on
[a, b] and that
r
f
dg = g(b) - g(a)
(e) Evaluate the integrais

and
18.
if they exist.
(a) If f and g are bounded functions on [a, b], prove that f is
Riemann-Stieltjes integrable with respect to g on [a, b] if and
only if for every positive real number e there is a partition P. of
[a, b] such that whenever P', P" are refinements of P., then
IR(f, g, P') - R(f, g, P")I <e
(Hint: if the property holds, let Pn be the partition such that

R(f, g, P') - R(f, g, P") < I/n for ali refinements P', P" of Pn.
Calculus
226
Show that the sequence of real numbers {R(f, g, Pn)} converges

and that its limit is the Riemann-Stieltjes integral of f with
respect to g.)
(b) A function g defined on [a, b] is monotonically increasing if
xe[a, b], ye[a, b], x
~y
imply g(x)
g(y)
Prove that if f is bounded and continuous on [a, b] and g is

monotonically increasing on [a, b], then f is Riemann-Stieltjes
integrable with respect to g on [a, b].
(Hint: use the uniform continuity of f to find > O such that
6
IJ(x) - f(y)I < 2(g(b) - g(a))
whenever
lx - y 1 <
Let P be a partition such that max {xk - xk _ 1} < , and show

that if P' is any refinement of P, then IR(f, g, P) - R(f, g, P')I
<e/2. Finally, show that if P' and P" are refinements of P, then
IR(f, g, P') - R(f, g, P")I <e.)
8.2 PROPERTIES OF THE INTEGRAL
ln this section we prove some ofthe elementary properties ofthe Riemann
integral and also discuss the question of whether integrability is or is not
preserved under pointwise and uniform convergence.
Our first result is a technical one which will be useful to us as we
develop the properties of the integral.
Proposition 8.2.1 Letfbe a real-valued function defined on [a, b], where

a < b. For every partition P of [a, b],
1.
If reR, r ~O, then

U(rf, P) = r U(f, P)
2.
and
L(rf, P) = r L(f, P)
and
L(rf, P) = r U(f, P)
If reR, r <O, then

U(rf, P) = r L(f, P)
Proof: We prove the first part of ( 1), lea ving the remainder of the proof
as an exercise (See Exercise_ l at the end of this section.)
Properties of the Integral
227
If reR, r;;:;: O, and P ={a= x 0 , xi> . .. , Xn = b} is a partition of[a, b],
then
n
U(rf, P) =
sup {(rf)(x) 1x e [xk _ 1> xk]}(xk - xk- 1)
k=I
n
.=
sup{rf(x) lxe[xk_.,xk]}(xk-xk-1)
k=I
n
rsup{f(x) lxe[xk-lxk]}(xk-xk-1)
k=I
n
= r
sup {f(x) 1xe[xk-., xk]}(xk -xk_ 1)
k=I
= r U(f, P) D
Corollary 8.2.2
If f is a real-valued function defined on [a, b], where
a< b, then
reR, r;;:;: O implies
1.
inf U(rf, P) = r inf U(f, P)

p
2.
and
sup L(rf, P) = r sup L(f, P)
and
sup L(rf, P) = r inf U(f, P)
reR, r <O implies
inf U(rf, P) = r sup L(f, P)

p
Proof: If reR, r;;:;: O, then U(rf, P) = r U(f, P) for every partition P of

[a, b], so
inf U(rf, P) = inf {r U(f, P)} = r inf U(f, P)

We leave the remainder of the proof as an exercise. (See Exercise 2 at the
end of this section.) D
Now we are ready to prove the familiar properties of the Riemann
integral. We begin by showing that the integral of a constant times a
function is the constant times the integral of the function.
228
Calculus
Proposition 8.2.3 If/is a function which is Riemann integrable on [a, b],

then rf is Riemann integrable on [a, b] for ali reR, and
rfdx =r rfdx
Proof: Let reR. Since f is Riemann integrable on [a, b] it is bounded

there, and this surely implies that rf is bounded on [a, b].
If r ~O, then
and
sup L(rf, P) = r sup L(f, P)

p
inf Utj; P) = r inf U~f, P)

p
by Corollary 8.2.2. But since f is Riemann integrable on [a, b],

sup L(f, P) = inf U( f, P) =
p
b f dx
Ja
Thus we have
sup L(rf, P) = r
p
b f dx
Ja
and
inf U(rf, P) = r
p
b f
Ja
dx
and the proposition is proved when r ~ O. The proof for the case r < O is
similar. D
Next we prove that the integral of a sum (difference) is the sum
( difference) of the integrais.
Proposition 8.2.4 Iff and g are functions which are Riemann integrable
on [a, b], then f + g and f - g are Riemann integrable on [a, b], and
r
r
r(/- r r
(f+g)dx= rfdx+
gdx
g dx
g) dx
f dx -
Proof: It will suffice to prove the proposition for f + g, since the conclusion for f - g will then follow from Proposition 8.2.3 and the fact that
f-g = f + ( - l)g.
229
Since f and g are Riemann integrable on [a, b], they are bounded
there, and thus f + g is also bounded there. Because f is Riemann integrable on [a, b], there is for every positive real number e a partition P. of
[a, b] such that U(f, P.) - L(f, P.) < e/2, and it follows that
Similarly, there is for every e >O a partition
U(g, P;)
+ g(x)
b
6
<]a gdx +2
p] it is clear that
For any closed interval [tX,

sup {f(x)
P; such that
1xe[tX,
P]} ~ sup {f(x) 1xe[tX, P]}
+ sup {g(x)
1 xe[tX,
P]}
and this implies that for every partition P of [a, b],
U(f + g, P)
U(f, P)
+ U(g, P)
ln particular, then,
U(f + g, P. u P;)
U(f, P. u P;)
+ U(g, P. u P;)
But
U(f, P. uP;) ~ U(f, P.) <

and
U(g, P.uP;)
Jb f dx +26
r
r r
~ U(g, PJ<
Therefore
U(f + g, P. u P;) <
f dx
gdx
+~
g dx +e
which implies that for every e > O,

inf U(f + g, P) <
P
b f dx + b g dx +e
]a
Ja
Calculus
230
Hence
inf U(f +g, P)
P
~ b fdx + b gdx
Ja
Ja
Since the preceding inequality holds for every pair of functions which
are Riemann integrable on [a, b], it must hold for -fand -g, which are
Riemann integrable on [a, b] by Proposition 8.2.3. Therefore
infU(-(f +g),P) =infU(-f-g,P)
P
-f
~ b
Ja
-fdx+
b
Ja
-gdx
Jdx-f gdx
But by Proposition 8.2.l with r = -1,

inf U( -(f + g), P)
p
-sup L(f + g, P)
p
Thus
- sup L(f + g, P)
P
~ - b f
Ja
dx -
b g dx
Ja
or
sup L(f + g, P)
(**)
~ b fdx + b g dx
Ja
Ja
Now, supp L(f + g, P) ~ infp U(f + g, P) by Proposition 8.1.8, and this

together with the inequalities ( ) and ( **) implies that
inf U(f + g, P)
p
= sup (f + g, P) = b f
p
Ja
dx
+ b g dx
Ja
and we are done. D

Now we are ready to show that the integral is additive on intervals.
Properties of the Integral.
231
Proposition 8.2.5 Let f be a function which is Riemann integrable on

[a, b]. If a <e < b, thenf is Riemann integrable on [a, e] and on [e, b], and
Proof: It is clear thatf is bounded on [a, e] and on [e, b]. Furthermore,

for every positive real number e there is a partition P. of [a, b] such that
U(f, P.) - L(f, P.) <e
We may assume that ceP. (If crtP., then P; = P. u {e} is a partition of

[a, b] such that ceP; and U(f, P;) - L(f, P;) <e.)
Let P consist of those points of P. which belong to [a, e] and let Pb
consist of those points of P. which belong to [e, b]. Clearly,
Since U(f, Pa) "'."'" L(f, P) and U(f, Pb) - L(f, Pb) are nonnegative, it follows that
and
Hencefis Riemann integrable on [a, e] and on [e, b] by virtue of Proposition 8.1.11.
Now we show that
If P is any partition of [a, e] and P' any partition of [e, b], then
f dx
~ U(f, P u P') =
U(f, P)
+ U(f, P')
But this implies that
b f
]a
dx
~ inf U(f, P) + inf U(f, P')

P
232
Calculus
or
The preceding inequality is valid for ali functions which are Riemann
integrable on [a, b], hence is valid for -f, and thus
b
]a
rc
-fdx ~]a -fdx +
-fdx
or
(**)
The inequalities ( *) and ( **) now imply that
and we are done. D

It is the usual practice to define the Riemann integral on a closed
interval of the form [a, a] to be zero; thus, by definition,
(This makes sense, for the only "partition" of [a, a] is {a = x 0 , x 1 =a},

and certainly every upper and lower Riemann sum over this "partition"
will be zero.) Also, if a< b, we define
provided the integral on the right-hand side of the equation exists. With
these definitions established, the equality
233
of Proposition 8.2.5 holds for ali real numbers a, b, and e, as long as ali
three of the integrais exist. (See Exercise 2 at the end of this section.)
We return to our study of the properties ofthe Riemann integral. Our
next result is often useful when it is necessary to bound an integral.
Proposition 8.2.6 Letland g be functions which are Riemann integrable

on [a, b]. Ifl(x) ~ g(x) for ali x e [a, b], then
rldx~r gdx
Prool:
Let h = g -
Then h is Riemann integrable on [a, b] and
r r
hdx=
gdx-f ldx
But h(x) ~O for ali x E [a, b], so L(h, P)

which implies that
supL(h,p) =
P
~O
for every partition P of [a, b],
b hdx ~O
]a
Corollary 8.2.7 Iflis a function which is Riemann integrable on [a, b],

then Ili is also Riemann integrable on [a, b], and
Prool: If e eR, e >O, there is a partition P. = {x0 , xi> . .. , xn} of [a, b]

such that U(f, P.) - L(f, P.) <e. We may write
U(f, P.) - L(f, PJ= (sup {l(x) 1 xe [xk-1> xk]}
- inf {J(x) 1 xe[xk-1> xk]})(xk - xk-1)
The integrability of Ili now follows from this identity and the inequality
lll(x)l- ll(Y)ll ~ ll(x) -l(Y)I
which holds for ali x and y in [a, b]. We leave the details to the reader. (See
Exercise 5 at the end of this section.)
234
Calculus
Since -lf(x)I ~f(x) ~ lf(x)I for ali xe[a, b], it follows from Proposition 8.2.6 that
and thus
Next we state the First Mean Value Theorem for Integrais, which we
will need in the next section to prove the Fundamental Theorem of Calculus. The proof of the First Mean Value Theorem is left to the reader.
Proposition 8.2.8 ( First Mean Value Theorem for Integrais) If f is a
real-valued function defined and continuous on [a, b], there is some
e E (a, b) such that
f dx =f(c)(b - a)
This concludes our development of the basic properties of the

Riemann integral. The reader is asked to prove additional properties of
the integral in the exercises.
We end this section with a discussion of the realtionship between
Riemann integrability and the convergence of a sequence of functions.
The questions are these: If {ln} is a sequence of Riemann integrable
functions which converges to a functionf, isfRiemann integrable? Iffis
Riemann integrable, is it true that
b f
Ja
b fn dx
n-+ oo Ja
dx = Lim
The answer to both these questions is no if the convergence is merely

pointwise, but yes if it is uniform.
Examples 8.2.9 l. This example shows that if Un} is a sequence of
Riemann integrable functions which converges pointwise tof, thenfneed
235
not be Riemann integrable. Let Q n[O, l] = {q1> q2 ,

let
and foreach neN
if xe{ql> q2, ... , qn}'

if xe[O, l] - {q1> q2, ... , qn}
Then (see Exercise 6 of Exercises 8.1) Xn is Riemann integrable on [O, l]
for each n eN, with
Xndx =0
and the sequence {Xn} converges pointwise (but not uniformly) to x, the
characteristic function ofthe rationals restricted to [O, l]. But as we showed
in Example 5 of Examples 8.1.10, x is not Riemann integrable on (O, l].
2. This example will show that even if {f,,} is a sequence of Riemann
integrable functions which converges pointwise to a Riemann integrable
function f, it need not be the case that
b f
Ja
b f,, dx
n-+ oo ]a
dx = Lim
For each n eN, letfn be the function whose graph over ( 1/n, l] is zero
and whose graph over [O, l/n] is the isosceles triangle with altitude n (see
Figure 8.4). The defining equations for fn are
ifO ~ x ~ 1/2n
if 1/2n < x ~ 1/n
if 1/n <X~ l
Each functionf,, is continuous on [O, l], hence Riemann integrable there,
and using the results of Propositions 8.2.3, 8.2.4, and 8.2.5 and ofExamples
1 and 2 of Examples 8.1.10, we have
f,, dx = 2n 2
ll/2n
x dx
+ 2n il/n
1 dx - 2n 2
l/2n
21( 1 ) (1 1)
1) +o (1--n1)
-2n 21(1
- --2 n
4n
=2n - - - 0 +2n - - 2 4n 2
n 2n
2
=-
il/n x dx +
l/2n
il
l/n
Odx
Calculus
236
y
Figure 8.4 The sequence {f,,} of Example 2, Examples 8.2.9.
The sequence of functions {f,,} converges pointwise (but not uniformly)

to the zero function on [O, l]. (Check this.) Thus the limit function is
certainly Riemann integrable, with integral equal to O, whereas
Lim
n-+oo
Jo f, dx =
1
Lim ! = !
2 2
n-+oo
Now let us show that uniform convergence does preserve integrability, and that under uniform convergence the limit of the integrais is the
integral of the limit.
237
Proposition 8.2.10 For each n EN let fn be Riemann integrable on [a, b].

If {fn} converges uniformly tof on [a, b], thenfis Riemann integrable on
[a, b] and
Proof: Since fn is Riemann integrable on [a, b], it is bounded there, and

hence by Proposition 7.1.9, the limit functionfis bounded on [a, b].
Let e be a positive real number. Since {fn} converges uniformly to f,
there is some m EN such that n ~ m implies
8
IJ(x) - f,,(x)I < 3(b - a)
for ali x E [a, b]. -Because fm is Riemann integrable on [a, b], there is a
partition P = {x0 , . . . , xn} of [a, b] such that
U(fm, P) - L(fm, P) <
Now,
8
lf(x) - m(x)I < 3(b _a)
for all xe[a, b] implies that

8
fm(x) - 3(b - a) < f(x) < fm(x)
+ 3(b -
a)
for all x E [a, b]. Since this last inequality must also hold on any subinterval of [a, b], it follows that
= U(fm, P)
+ 3(b _a) k~l (xk

8
= U(fm, P) +3
-xk_ 1)
Calculus
238
Similarly,
L(f, P)
3e
L(fm, P) -
Therefore
U(f, P) - L(f, P) ~ U(fm, P) - L(fm, P)
+ 3+ 3 < 3+ 3+ 3 =
Hencefis Riemann integrable over [a, b].

To complete the proof we must show that
b f
Ja
b f,, dx
n-+ oo Ja
dx = Lim
Since Un} converges uniformly tof on [a, b] there is for every positive real
number e some me N such that n ~ m implies
lf(x) -fn(x)I < b _a

for ali n e [a, b]. By virtue of Corollary 8.2. 7, Proposition 8.2.6, and Example 1 of Examples 8.1.10, we have that
r r r(/- r
f dx -
fn dxl
=1
f,,) dxl
li - fnl dx
8 -dx
rb-
J
0
b -a
This proves that the sequence of real numbers
converges to the real number
rfdx
and we are done. O
8 -(b-a) =e
=-
b-a
239
EXERCISES
1.
(a) Complete the proof of Proposition 8.2.1.

(b) Complete the proof of Corollary 8.2.2.
2. If f(a) exists, define
If a< b and
exists, define
Prove that for any real numbers a, b, and e,
3.
4.
5.
6.
7.
provided ali three of these integrais exist.

Letfbe a real-valued function defined on [a, b] and let a <e< b. If
fis Riemann integrable on [a, e] and on [e, b], show thatfis Riemann
integrable on [a, b] and that
Letfbe a bounded real-valued function defined on [a, b]. If the set

of discontinuities of f is a finite subset of [a, b], prove that f is
Riemann integrable on [a, b].
Fill in the details in the proof of Corollary 8.2.7.
Give an example which shows that the converse of Corollary 8.2. 7
does not hold.
(a) Letfbe Riemann integrable on [a, b] with
ex = inf {f(x) 1 x e [a, b]}
and
p=
sup {f(x) 1 x e [a, b]}
240
Calculus
Prove that
a.(b - a)
8.
dx
~ /J(b -
a)
(b) Prove the First Mean Value Theorem for Riemann integrais.
Let f be continuous on [a, b] and let
F(t) =
9.
fdx
for ali t E [a, b]. (Note that F(a) =O by definition.) Prove that F is a
continuous function on [a, b].
(a) Letfbe continuous on [a, b] withf(x) ~O for ali n E [a, b]. Prove
that if f(x 0 ) > O for some x 0 e [a, b], then
.
rfdx>O
(b) Letf, g be continuous on [a, b] withf(x) ~ g(x) for ali xe [a, b].
Prove that if f(x 0 ) < g(x0 ) for some x 0 E [a, b], then
rfdx <
10.
11.
gdx
Give an example to show that the converse of Proposition 8.2.10 is

false. That is, find a sequence of functions Un} which converges
pointwise but not uniformly to a function f, and such that
(Hint: modify Example 2 ofExamples 8.2.9 by changing the heights

of the triangles.)
Iffn is continuous on [a, b] for ali n eN, the sequence {f,,} converges
uniformly on [a, b] to f, and g is continuous on [a, b], prove that fg
is Riemann integrable on [a, b] and that
Lim jb fng dx = jb fg dx
n-+ oo Ja
Ja
241
The Derivative
12. This exercise continues the study of the Riemann-Stieltjes integral

begun in Exercise 17 of Exercises 8.1. Most of the properties of the
Riemann integral have Riemann-Stieltjes analogs. Prove that the
following properties hold for Riemann-Stieltjes integrais, provided
the integrais on the right-hand side of each equation exist.
r r
r r
(a)
(b)
rf dg = r
f dg
for ali reR
f d(rg) = r
f dg
for ali reR
r(/
(e)
(d)
+g)dh = rfdh
rfd(g +h) = rfdg
(e) If a< e< b,
gdh
+ rfdh
r r r
r r
f dg =
(f)
f dg
f(b)g(b) - f(a)g(a) =
f dg
f dg
g df
8.3 THE DERIVATIVE

ln this section we define the deriva tive of a function on the reais, study the
properties of the derivative, and examine the connection between the
derivative and the Riemann integral. The derivative will be defined by
means of a functional limit, and thus we will make repeated use of the
results concerning functional limits developed in Section 5.4, Chapter 5; the
reader is urged to review that section now. ln particular, we will find it
useful to have the definition of a functional limit (Definition 5.4.1) restated
in -e terms; we include this restatement as our first result of this section.
Proposition 8.3.1
Letfbe a real-valued function defined on a subset S of
R. If e is a limit point of S and x 0 eR, then
Limf(x) = x 0
X-+C
242
Calculus
if and only if for every e e R, e > O, there is some e R,
x eS, O<
lx - e 1<
imply
{>
> O, such that
IJ(x) - x0 I < e
The proof of the proposition will be left to the reader. (See Exercise
1 at the end of this section.)
Now let us define the derivative of a function at a point.
Definition 8.3.2 Let/be a real-valued function whose domain is a subset
S of R, and let e e S be a limit point of S. The deriva tive off at e, denoted
by /'(e), is defined by
f'(c) = Lim /(x) - f(c)
x--+c
-e
Note that we consider the derivative off at e only for those points e
of the domain S which are also limit points of S. (If S is an interval or ray,
as will frequently be the case, then every point of S is a limit point of S,
and we may consider the derivative off ate for ali ceS.) However, even
if e e S is a limit point of S, the deriva tive off at e need not exist, for the
limit as x approaches e of the function
f(x) -f(c)
x-c
may not exist. Ifj'(c) does exist for ali ceS, we say that/is dijferentiable
on S. The function
f(x) -f(c)
x-c
called the dijference quotient off at e, is a real-valued function of x whose

domain is S - {e}. Thus when working with the derivative we need only
consider the values of the difference quotient when x eS, x #e. Also, if
S' = {ceS lf'(c) exists}
then the deriva tive function f' is the function from S' to R which assigns
to each ceS' the real number f'(c).
The Derivative
243
Examples 8.3.3 1. Let S be any subset of R, let reR, and let f be the
constant function defined by f(x) = r for ali xeS. If ceS is a limit point
of S, then
O
. f(x) - f(c) L' r - r L'
Lim
= im--= im-x-+c
X - e
X-+C X e X-+C X - e
But since x # e in the limit
--=0
x-c
so
f'(c) = Lim O
X-+C
provided this limit exists. But the constant function g(x) = O is certainly
continuous at e, so by Proposition 5.4.4 of Chapter 5,
Lim O= Lim g(x) = g(c) =O
x-+c
X-+c
Thus we have shown that ifjis a constant function on S and ceSis a limit
point of S, then f'(c) exists and f'(c) =O. Therefore we may say that if
f(x) = r on S, thenf'(x) =O for ali xeS such that x is a limit point of S.
2. Let S be any subset of R, and letf be the identity function on S,
defined by f(x) = x for ali xeS. If ceS-is a limit point of S, then
f'(c) = Lim f(x) - f(c) = Lim x - e
x-+c
X -
X-+C
X -
Again, since x #e in the limit,
x-c
--=l
x-c
and therefore the continuity of the constant function g(x) = 1 implies that
f'(c) = Lim 1 = 1
X-+C
Thus iff(x) = x on S, thenf'(x) = 1 for ali x e S such that x is a limit point

of S.
244
Calculus
3. Let/(x) = x 2 for ali xeR. If ceR, then
. f(x) - f(c) L' x 2 - c 2 L' (x + c)(x - e)
Llffi
= lffi
= lffi - - - - X-+C
X-C
X-C
X-+C
X-C
X-+C
Since x =F e in the limit,

. (x
L lffi
+ c)(x -
x-+c
e)
x - e
L'
lffi X
+e)
x-+c
and the continuity of the linear function g(x) = x
+ e then implies that
f' (e) = Lim (x +e) = Lim g(x) = g(c) = 2c

X-+C
X-+C
Thus we have shown that if f(x) = x 2 on R, thenf'(x) = 2x on R.

4. Let/be the absolute value function on R. Then/(x) = x on the ray
(O, + oo), so if ceR, e> O, then by Example 2 above,f'(c) = 1. Similarly,
since f(x) = -x on the ray ( - oo, O), if e< O we have
f'(c) = Lim f(x) - f(c) = Lim -x +e= Lim ( -1) = -1
x-+c
x- e
x-+c
x- e
x-+c
However,f'(O) does not exist. To see this, supposef'(O) does exist; then
f'(O) = Lim f(x) - /(O)
x-+0
X -
and by Proposition 8.3.1, there is a eR, >0, such that O<lxl<<>

implies
But no matter what the value of , there are real numbers y and z such
that - < y < O< z < , and hence such that
11~1_f(o)I=1-1-f'(O)I<1
The Derivative
245
and
11:1- f'(O)I =II - /'()I < 1

Since there is no real number r such that 1- I - ri < 1 and ll - ri < 1, we
conclude that /'(O) cannot exist. Thus the absolute value function does
not have a derivative at O.
5. Letf(x) =
for ali xe[O, +co). If c>O, then
Jx,
Lim f(x) - /(e) = Lim

x~c
X -
x~c
Jx - Jc = Lim Jx - Jc Jx + Jc
x~c
.jX + Jc
X -
= L1m
X -
-e
x~c (X - c)(.jX
+ jC)
.
1
= L1m - - - x~c
.jX + Jc
Lim ( .jX + jC)

x~c
Jc> 2Jc
Limx~cCJx +
=
by continuity of the function
g(x) =
+
at e. Therefore if f(x) =
and x >O, then f' =
l/2Jx. However,f'(O) does not exist, because
But
Jx Jc
Jx
Lim f(x) - /(O)= Lim

x~o
X -
Jx = Lim-1x~o Jx
x~o X
does not exist. (Check this.)

The reader no doubt recalls from earlier studies that the derivative
f'(c) of f at e is the slope of the line tangent to the graph off at the point
(c,f(c)). (See Exercise 6 at the end of this section.) Therefore if the graph
does not have a unique nonvertical tangent line at (c,f(c)), then it stands
to reason that it cannot have a derivative ate. (Thus, the absolute value
function does not have a unique tangent line at (O, O) and, as we have
seen, it has no derivative at O; similarly, the square root function has a
vertical tangent line at (O, O), and it too has no deriva tive at O.) Since it is
intuitively clear that ifj is discontinuous ate, then its graph cannot have
a unique tangent line at (c,f(c)), this line of reasoning suggests that if
f is discontinuous at e, then f'(c) does not exist. This is indeed true,
and we begin our study of the properties of the derivative by proving this
important fact.
Calculus
246
Proposition 8.3.4 Letfbe a real-valued function defined on a subset S of

R, and let ceS be a limit point of S. If the derivativef'(c) exists, thenf is
continuous at e.
Proof:
If xeS, x #e, then
f(x) - f(c)
f(x) - f(c) (x - e)
x-c
Therefore
Lim (f(x) - /(e)) = Lim f(x) - f(c) Lim (x - e)
x-+c
x- e
x-+c
x-+c
provided that
f(x) - f(c)
L i. m
---x-+c
-e
and
Lim (x -e)
X-+C
both exist. However

Lim f(x) - f(c) = f (e)
x-+c
x- e
exists by hypothesis, and
Lim (x - e) = e - e = O
X-+C
by virtue of the continuity ofthe linear function g(x) = x - e and Proposition 5.4.4 of Chapter 5.-Thus
Lim (f(x) - /(e)) = f (e) O =O
X-+C
and hence
Limf(x) = f(c)
x-+c
which by Proposition 5.4.4 implies that f is continuous at e. D

Proposition 8.3.4 tells us that if f is not continuous at e, then /'(e)
cannot exist. lt is important to realize that the converse ofthe proposition
is not true: iff is continuous ate, it does not follow thatf(c) exists. Thus
the absolute value function is continuous at e = O, but as we have seen, it
has no derivative at O. Indeed, it is not terribly difficult to show that there
247
The Derivative
are functions which are continuous at every real number but which do not
have a derivative at any real number. (See Exercise 24 at the end of this
section.)
We next state some of the well-known rules of differentiation. The
proofs will be left to the reader. (See Exercises 7 and 8 at the end of this
section.)
Proposition 8.3.5 Let ceR. Iffand gare real-valued functions defined at

e, and f (e) and g'(c) exist, then
1.
2.
(rf)'(c) exists and (rf)'(c) = r f(c), for ali reR;

(f + g)'(c) and (f - g)'(c) exist and
(f + g)'(c) = f'(c)
3.
4.
+ g'(c),
(f- g)'(c) = f (e) - g'(c)
(fg)'(c) exists and (fg)'(c) = f(c)g'(c)

If g(c) =F O, (f/g)'(c) exists and
( [_)'(e) =
g
+ f(c)g(c);
f (c)g(c) - g'(c)f(c)
[g(c)]2
Now we are ready to prove the chain rule for derivatives.
Proposition 8.3.6 (Chain Rule) Let ceR. Iffis a real-valued function

defined at e, g is a real-valued function defined at f(c), and f(c) and
g'(f(c)) both exist, then (g f)'(c) exists ~nd
(g 0 f)'(c) = g'(f(c))f'(c)
Proof:
Define a function u by setting

u(x) = f(x) - f(c) - f (e)
x-c
for ali x in the domain of f such that x =F e. The following identity holds
for ali x in the domain of f, including x =e:
[u(x)
+f
(c)](x - e)
f(x) - f(c)
Similarly, we define a function r by setting

r(y) =
g(y~ =~~~(e)) -
g'(f(c))
248
Calculus
for ali y in the domain of g such that y =F f(c). The following identity holds
for ali y in the domain of g, including y =/(e):
( **)
[-r(y)
+ g'(f(c))](y -
f(c)) = g(y) - g(f(c))
Note that
Lim cr(x) = Lim f(x) - f(c) - Limf'(c) = f'(c) - f'(c) = O
x-+c
X-+C
X -
X-+C
and
Lim -r(y) = Lim g(y) y-+f(c)
y -
y-+f(c)
~~~(e)) e
Lim g'(f(c))
y-+f(c)
= g'(f(c)) - g'(f(c)) =O
Using the identities ( *) and ( ** ), we have
(g f)(x) - (g f)(c) = g(f(x)) - g(f(c))
0
= [-r(/(x))
+ g'(f(c))](f(x) -
= [-r(/(x))
+ g'(f(c))][cr(x) + f'(c)](x
f(c))
- e)
Thus if x =F e, then
(g 0 f)(x) - (g 0 f)(c)
x-c
= [-r(/(x)) + g'(f(c))][cr(x) +f'(c)]
Hence
Lim (g f)(x) - (g 0 f)(c) = Lim [-r(/(x)) + g'(f(c))] Lim [cr(x) +/'(e)]
x-+c
x- e
x-+c
x-+c
provided the limits on the right-hand side of the equation exist. However,
Limx-+c cr(x) =O implies that Limx-+c [cr(x) +f'(c)] = f'(c). Furthermore,
since /'(e) exists, f is continuous at e, and therefore Limx_...J(x) = f(c);
that is,f(x) approaches/(c) as x approaches e. Thus
Lim (f(x)) =
X-+C
Lim -r(/(x))
f(x)-+f(<)
249
The Derivative
(See Exercise 2 at the end ofthis section.) But since Limy .....f(c) r(y) =O, we
have Limf(x)-+f(c) r(f(x)) =O. Therefore
Lim [r(.[(x))
+ g'(f(c))] = g'(f(c))
X-+C
and we are done. O

The technique used in the previous proof is one that is often useful
when wo1king with derivatives. lt consists of writing the difference quotient as the derivative plus an error function which goes to zero as x
approaches c, then manipulating the error function to establish the
desired result.
Next we are going to consider very briefly the well-known connection
between the derivative and relative maxima and minima of a function.
Definition 8.3.7 Let f be a real-valued function defined on the open
interval (a, b). The functionfhas a relative maximum at cE (a, b) if there
is some positive real number such that
f(x)
~f(c)
for ali
xE(c -, c +)
Similarly,fhas a relative minimum at cE (a, b) ifthere is some positive real

number such that
f(x) ;;;i.f(c)
for ali
xE(c - , c +)
.Proposition 8.3.8 Let f be a real-valued function defined on the open

interval (a, b). If f has a relative maximum or a relative minimum at
cE(a, b) andf'(c) exists, thenf'(c) =O.
Proof: We prove this for the case in whichfhas a relative maximum at

e. The proof for a relative minimum is similar and is left to the reader.
Sincefhas a relative maximum at cE(a, b) there is some positive real
number y such that f(x) ~ f(c) for ali x E (e - y, e + y). If f'(c) =F O there
is some positive real number 17 such that
Calculus
250
Let = min {y, 17 }. Then
( )
xe (e - , e+ ), x =F e
implies
1f(x) - f(c) -
x-c
f'(c)I < lf'(c)I
Now supposef'(c) >O. If xe(c, e+ ), then

f(x) -f(c) 0
----~
x-c
so that
1
f(x) - f(c) - f'(c)l

x-c
';J::; f'(c) =
lf'(c)I
which contradicts ( *) above. Hence f'(c) >O is impossible. But if

f'(c) <O, then x E (e - , e) implies
f(x) -f(c)
- - - - ";::::0
x-c
so that
1
f(x) - f(c) - f'(c)l

x-c
';J::;
-f'(c) = lf'(c)I
and thusf'(c) <O is also impossible, and we are done. D

Theorem 8.3.9 (Rolle's Theorem) Let f be a continuous real-valued
function defined on [a, b] and such that /(a) = f(b) =O. If f is differentiable on (a, b), then there is some ce(a, b) such thatf'(c) =O.
Proof: If f(x) =O for ali x E [a, b] then surely f'(c) =O for all e E (a, b)
and we are done, so we assume thatfis not identically zero on [a, b]. Since
f is continuous on [a, b], it attains its maximum at some point e E [a, b] and
its minimum at some point d E [a, b]. Since f is not identically zero on [a, b]
at least one of f(c), f(d) must be nonzero. For definiteness, suppose
f(c) =F O; then e =F a and e =F b, so e E (a, b), and beca use
/(e)= max {/(x) 1 xe[a, b]}, the functionfcertainly has a relative maximum at e. But f is differentiable on (a, b), so f'(c) exists, and therefore
f'(c) =O by Proposition 8.3.8. D
251
The Derivative
Corollary 8.3.10 (The Law of the Mean) Iffis a real-valued function

defined and continuous on [a, b] and differentiable on (a, b), then there is
some e E (a, b) such that f'(c)(b - a) = f(b) - f(a).
Proof:
Define
g(x) = f(x) -
f(b) -f(a)
b _ a (x - a) - f(a)
for all x e [a, b] and apply Rolle's Theorem to g. D

Corollary 8.3.11 Iffis a real-valued function defined on [a, b] and differentiable on (a, b) and if f'(x) =O for all x e (a, b), then f is a constant
function on [a, b].
Proof:
Apply the Law of the Mean on [a, x], for all x E (a, b]. D
Corollary 8.3.12 Letf and g be continuous real-valued functions defined

on [a, b] and differentiable on (a, b). If f'(x) = g'(x) for ali x E (a, b), then
there is some r E R such that f(x) = g(x) + r for all x E [a, b].
We leave it to the reader to finish the proofs of the preceding corollaries. (See Exercise 17 at the end of this section.)
Now we tum to an examination of the relationship between the
derivative and the Riemann integral. The key theorem in this regard is
the following one, which says that the derivative of the integral is the
integrand.
Theorem 8.3.13 Letfbe a real-valued function defined and continuous
on [a, b]. If F is defined on [a, b] by
F(x) =
lx
f dx
for all x E [a, b], then F is differentiable on [a, b] and F'(x)

XE[a, b].
= f(x) for ali
Proof: Since/is continuous on [a, b], it is Riemann integrable there, and

thus the function F is defined for all x e [a, b]. F is defined at x = a beca use, by definition,
Calculus
252
Let e e [a, b], and consider the difference quotient
F(x) -F(c)
x-c
for xe[a, b], x #e.
If x > e, we have
F(x) -F(c)
X - e
=-1-[ Ja
x fdx - cfdx]
X -
Ja
=-1- J
x fdx
X -
by Proposition 8.2.5. By Proposition 8.2.8, there is some y e (e, x) such

that
f dx =f(y)(x -e)
Therefore if x > e, then

F(x) - F(c) = f(y)
x-c
for some ye (e, x). Similarly, if x <e, then

F(x) - F(c) = _ l [ - c f dx] = _ l c f dx
X - e
X - e
Jx
e - X Jx
1
c-x
= -f(y)(c - x) = f(y)
for some y e (x, e). Hence for ali x e [a, b], x #e, there is some point y
between x and e such that
F(x) - F(c) = f(y)
x-c
Since y is between x and e, y approaches e as x approaches e. Therefore

. F(x) - F(c) =1my=1my
L ji( ) L ji( )
Lim
X-+C
X - e
X-+C
y-+c
But f is continuous on [a, b], so
Limf(y) = f(c)
y-+c
which proves the theorem. O
The Derivative
253
Corollary 8.3.14 (The Fundamental Theorem of Calculus) Let/be a

real-valued function defined and continuous on [a, b]. If G is any function
defined on [a, b] such that G'(x) = f(x) for all x e [a, b], then
f dx = G(b) - G(a)
Proof: Let Fbe as in Theorem 8.3.13. Since F'(x) =f(x) = G'(x) for ali
xe [a, b], by Corollary 8.3.12 there is some reR such that F(x) = G(x) + r
for all xe[a, b]. But then G(a) + r = F(a) =O, so r = -G(a), and hence
f dx = F(b)
= G(b)
+ r = G(b) -
G(a). D
The Fundamental Theorem of Calculus is the result which allows us

to evaluate certain definite integrais by the familiar process of finding an
antiderivative of the integrand and evaluating it at the endpoints of the
interval of integration. Thus it is the Fundamental Theorem of Calculus
which justifies calculations such as
i2
x412 =-(2
l
x 3 dx = 4
15
-1 4 ) = 4
However, it is important to realize that not every Riemann integral can be

evaluated in this manner. For e:J$:ample, the function/ defined by
is Riemann integrable on [O, l], with
l
fdx=-
but the Fundamental Theorem of Calculus does not apply here, because
f is not continuous on [O, l] and hence there cannot be any function

such that G' = f on [O, l].
The Fundamental Theorem of Calculus has several consequences

which yield useful techniques of integration. As an example, we state and
Calculus
254
prove the Integration by Substitution Theorem. For another example,

namely the Integration by Parts Theorem, see Exercise 21 at the end of
this section.
Proposition 8.3.15 Let g be a real-valued function which is differentiable

on [a, b] and suppose tht g' is continuous on [a, b]. Iff is a real-valued
function which is continuous on g([a, b]), then
g(b)
f dx =
ib
g(a)
( / 0 g)g' dx
Proof: The continuity of f, g, and g' implies the existence of the integrais. Let F be any function such that F' = f on g([a, b]). ( Such a function
F exists by Theorem 8.3.13.) Let G = F g. The function G is defined on
[a, b] and by the Chain Rule
0
G' = (F' g)g' = (f g)g'

Hence by the Fundamental Theorem of Calculus,
g(b)
dx = F(g(b)) - F(g(a)) = (F 0 g)(b) - (F 0 g)(a)
g(a)
= G(b) - G(a) =
r(/
g)g' dx D
Proposition 8.3. l 5 is known as the Integration by Substitution

Theorem because if it is written in the form
g(b)
f(x) dx =
g(a)
ib
f(g(x))g'(x) dx
it is evident that the right-hand side involves the substitution of g(x) for
x in the expression for f

Example 8.3.16 Proposition 8.3. l 5 is a useful tool for evaluating integrais, and it may be used in both directions. For instance, suppose we
wish to evaluate
x(x 2 + 1) 5 dx
255
The Derivative
Letting/(x) = x 5 and g(x) = x 2 + 1 for ali xe[O, l], we have
li
o
li
x(x 2 +1) 5 dx=-
2o
1ig(I) fdx=-112x
(f 0 g)g'dx=-
2g(O)
21
dx
But if F(x) = x 6 /6, then F'(x) = x 5 (see Exercise 9 at the end of this
section) and hence
112 x
x(x 2 +1) 5 dx=o

2
1x612
dx=-2 6
21
4
On the other hand, suppose we wish to evaluate
i3 X dx
Jo JXTI
If we let/(x) = x/jXTI and g(x) = x 2 - 1, then
1
3
~dx
o V .. '
ig-1(3)
g - l(O)
fdx =
12 (f
g)g' dx =
= 2 Jir2 (x 2 - 1) dx = 2
12x2l
--2xdx
1
(x33 _ x ) 12 = '83
1
We conclude this section with a discussion of the behavior of sequences of differentiable functions. Thus suppose {f,,} is a sequence of
functions which converges to the function f on [a, b]. As we did for
integrais, we can ask two questions: if fn is differentiable on [a, b] for ali
n eN, is the limit functionf differentiable on [a, b]? Iffis differentiable on
[a, b], isf = Limn .... cxJ~? We might expect the answers to these questions
will be no if the convergence is only pointwise, but yes if it is uniform.
Unfprtunately, this is not the case: even uniform convergence is not
enough to ensure that the limit function is differentiable, or, if it is, that
the derivative of the limit is the limit of the derivatives.
Examples 8.3.17 1. This example will show that a sequence of differentiable functions may converge uniformly to a nondifferentiable function.
Let f be the absolute value function, restricted to the interval [ -1, l].
According to the Weierstrass Approximation Theorem of Section 7.2,
Chapter 7, there is a sequence of polynomials which converges uniformly
Calculus
256
to f Each polynomial is differentiable on [ -1, l] (see Exercise 7 at the

end of this section), butf is not differentiable on [ -1, l] sincef(O) does
not exist.
2. This example will show that even if a sequence of differentiable
functions converges uniformly to a differentiable function, the derivative
of the limit need not be the limit of the derivatives. For each n eN, let
xn
f,(x)
=n
n
for ali xe[O, l]; then for each neN,

fn(x) = xn- t
for ali x e [O, l] ( see Exercise 7 at the end of this section). But the sequence
{f,,} converges uniformly to the zero function on [O, l], whereas the sequence of deriva tives {!~} does not converge to the deriva tive of the zero
function (which is again the zero function), because for x = 1,
Lim f,,(l) =F f' (1) = O
n-+
oo
Clearly we will need a stronger condition than uniform convergence

of the sequence to ensure that a sequence of differentiable functions converges to a differentiable function. As the following proposition shows,
what is needed is the uniform convergence of the sequence of deriva tives.
Proposition 8.3.18 Let Un} be a sequence offunctions defined and differentiable on [a, b]. If the sequence of deriva tives {f,,} converges uniformly
on [a, b], if the sequence Un(c)} converges for some e e [a, b], and if f,, is
continuous on [a, b] for ali n eN, then {f,,} converges to a differentiable
functionf on [a, b], and
Limf,, =f
n-+OO
on [a, b].
Proof: Suppose that {f,,} converges uniformly to the function g on
[a, b]. By Proposition 8.2.10,
Lim
n~co
jx f~ dx = jx g dx
Jc
Jc
The Derivative
257
for ali x E [a, b] and the Fundamental Theorem of Calculus implies that
lx f~ dx
= fn(x) - fn(c)
for ali x E [a, b]. Hence

Limfn(x) = ix g dx
n-+
oo
+ Limfn(c)
oo
n-+
for.-all x E [a, b]. Thus Limn-. 00 fn(x) exists for ali x E [a, b], so the sequence
{!~ l_'converges pointwise on [a, b]. Let Limn-. 00 fn = f; then
f(x) =
!-_!~fn(x) =
g dx
+f(c)
for ali x E [a, b]. Since {f,,} is a sequence of continuous functions which
converges uniformly to g, g must be continuous, and it follows from
Theorem 8.3.13 and ( *) above that f is differentiable and f' = g on the
closed interval with endpoints e and x, for ali x E [a, b]. Hencefis differentiable on [a, b] and
f =g =
Limf,,
n-+oo
on [a, b]. O
We should remark that uniform convergence of the sequence of
derivatives {f,,} is indeed stronger than uniform convergence of the sequence {fn}, for it can be shown that the former implies the latter,
whereas we know from Examples 8.3.16 that the latter does not imply the
former. Furthermore, Proposition 8.3.17 remains true if we remove the
hypothesis that each derivative function f,, be continuous. We assumed
the continuity of the derivatives in order to produce a relatively straightforward proof based on the Fundamental Theorem of Calculus. A proof
which does not require the continuity of the derivatives is outlined in
Exerci se 23 below.
EXERCISES
1.
2.

Prove that if Limx-.cf(x) = x 0 , then
Limz-.xo g(z), provided Limz-.x g(z) exists.
Limx-.c g(f(x)) =
258
Calculus
3.
Letfbe a real-valued function defined on an interval I and let ce/.

Prove thatf'(c) exists if and only if
Lim f(c
4.
5.
+ h) -
f(c) = f' (e)
h--+0
This result is often used as the definition of the deriva tive off at e.
For each of the following, find the derivative without using any
differentiation formulas.
(a) f(x) = 3x + 2 for ali xeR;
(b) f(x) = x 3 for ali x eR;
(e) f(x) = 1/x for ali xe(O, +oo);
(d) f(x) = 1/x 2 for ali x e(O, + oo ).
(a) Let
f(x) = {
.
-1
O
1
if X <0
if X= 0
if X> 0
Show thatf'(O) does not exist.

(b) Letf(x) = x 2' 3 for ali xeR. Without using any differentiation
formulas, prove that
f'(x) = jx -
1/3
for ali xeR, x ~O, and show thatf'(O) does not exist.
Letf be a real-valued function defined at ceR. Show that if f'(c)
exists, it is the slope of the tine tangent to the graph off at the point
(c,f(c)) in the Euclidean plane.

7. Prove (1) and (2) of Proposition 8.3.5.
8. Prove (3) and ( 4) of Proposition 8.3.5. (Hint: for (3),
6.
(fg)(x) - (fg)(c) =f(x) - f(c) g(x)

x-c
x-c
9.
+ g(x) -
g(c) f(c))
x-c
Let neN and letf(x) = xn for ali xeR. Prove that

f'(x) = nxn- 1
for ali x e R and then show that every polynomial function on R is

differentiable on R.
259
The Derivative
1O.
Let n be a negative integer and let f(x)

Prove that
f'(x) = nxn 11.
= xn for ali n E R, x #O.
for ali n eR, x #O.

Let q E Q, q # O, and let f(x) = xq for ali x E( O,
+ oo ). Prove that
f'(x) = qxq- I
for ali xe(O, +oo). (Hint: if q=m/n, then [f(xW=xm.)
12. Letfbe a real-valued function which is differentiable on [a, b]. Prove
that even though f' need not be continuous on [a, b], it is still true
thatf' has the intermediate value property; that is, show thatf' takes
on every value betweenf'(a) andf'(b). (Hint: if e is betweenf'(a) and
f'(b), consider g(x) = f(x) - ex.)
13. Letfbe a real-valued one-to-one function defined on [a, b] and let g
be its inverse function. Prove that iff is continuous at e E [a, b] and
g'(f(c)) exists and is nonzero, thenf'(c) exists and
1
f'(c) = g'(f(c))
14.
15.
Prove Proposition 8.3.8 for the case of a relative minimum.

Let f be a real-valued function which is continuous on [a, b] and
differentiable on (a, b).
(a) Prove that if f'(x) ~ O for ali x e (a, b), then
xe[a, b], ye[a, b], x
~y
imply
f(x)
~f(y)
(b) Prove that if f'(x) >O for ali x E (a, b), then
xe[a, b], ye[a, b], x <y
16.
imply
f(x) <f(y)
(e) State and prove results analogous to (a) and (b) if/'(x) ~O and
f'(x) <O on (a, b)
(a) Prove the First Derivative Test for relative maxima:
Letfbe a real-valued function which is continuous at cER and
suppose there exists a positive real number such that f is
differentiable on (e - , e +), except possibly at e. If f'(x) >O
for x E (e - , e) andf'(x) <O for x e(c, e +), thenfhas a relative maximum at e.
(b) State and prove a First Derivative Test for relative minima.
260
Calculus
17.
Fill in the details of the proofs for Corollaries 8.3.l O, 8.3. l l, and
8.3.12.
Prove the Cauchy form of the Law of the Mean:
18.
If f and g are continuous real-valued functions on [a, b] which

are differentiable on (a, b), then there is some e e (a, b) such that
f'(c)[g(b) - g(a)] = g'(c)[f(b) - f(a)]
19. Prove the following version of L'Hospital's Rule:

Let f and g be continuous real-valued functions on [a, b] which are
differentiable on (a, b). Iff(a) = g(a) =O, g'(x) =F Ofor x e (a, b), and
Limx .... a f(x) = Limx .... a g(x) =O, then
f'
. f- = L.im---;
L im
X-+Q
20.
X-+a
provided the limit on the right-hand side of the equation exists.

Prove the following generalization of Theorem 8.3. l 3:
Letf be Riemann integrable on [a, b], and define
F(x)
lx
f dx
for ali xe[a, b]. Iffis continuous at ce[a, b], then F'(c) exists and
F'(c) = f(c).
21.
Prove the Integration by Parts Theorem:

Iff and gare real-valued functions differentiable on [a, b] andf' and
g' are continuous on [a, b], then
fg' dx
22.
= f(b)g(b) - f(a)g(a) -
f' g dx
Use Proposition 8.3.15 to evaluate the following integrais.

(a)
(b)
I
l
x 2Jx3+} dx
X+ l
------dx
(2x 2 +4x+l) 3
The Derivative
23.
261
(e)
(d)
J2
xJx+}dx
13
(x - 1)3 dx
This exercise shows that Proposition 8.3.18 remains true if we drop

the hypothesis that the derivatives f,, be continuous. The reader is
asked to fill in the details. (Note:
li! - g li = sup {IJ(x) -
g(x) 11 x
E [a,
b]})
(a) If x E [a, b], apply the Law of the Mean to f,, - fm on [e, x] ( or
[x, e]) and take least upper bounds to conclude that
llfn - m li ~ (b -
a) llJ~
- fm li + lfn(c) -
fm(c)I
Use this inequality, the uniform convergence of {J',,}, and the

convergence of Un(c)} to show that Un} converges uniformly to
a function f on [a, b]. Note that f is continuous on [a, b].
(b) Let XE [a, b], ye [a, b]. Apply the Law of the Mean to/,, - fm on
[y, x] ( or [x, y]) and take least upper bounds to conclude that
fn(X) -fn(Y) _fm(X) -fm(Y)I <

x-y
x-y.
llJ~ -J',,, li
Let e >O be given. Because of the uniform convergence of {f,,},

there is some k 1 eN such that n ~ k1o m ~ k 1 imply
- f,,(y) .:...fm(x) -fm(y)I < ~

lf,,(x)x-y
x-y
3
from which it follows that if m
l
k1o then
f(x) -f(y) _fm(x) - fm(Y)I ~ ~

x-y
x-y
3
(e) Let g = Limn .... co~ on [a, b]. There is some k 2 EN such that
implies
IJ'm(y) - g(y)I < 3
262
Calculus
Let k = max {k1, k 2 }. Since/~(y) exists, there is some >O such
that
0< lx-yl <
implies
fk(x) - fk(y)
x-y
-!:' ( )1 < ~
k
Now show that O< lx - YI < implies that

l
24.
f(x) - f(y) _ g(y)I <

x-y
and hence thatf(y) =g(y).

There are many examples of functions which are continuous everywhere yet differentiable nowhere. The particular one we will consider in this exercise can also be found in Chapter 2 of Gelbaum and
Olmsted, Counterexamples in Analysis, Holden-Day, Inc., San Francisco, 1964. Before we can define the function, we need some preliminary results on series of functions.
If {f,,} is a sequence of functions d.efined on a set Se R, let
for all keN. (Thus sk is the kth partia! sum of the sequence of
functions Un}.) If the sequence {sd converges pointwise (uniformly) to a function f on S, we say that the series offunctions
00
L. ln
n=I
converges pointwise ( uniformly) to the function f on S, and write
00
L. ln=!
n=I
(a) Prove the Weierstrass M-Test:
if ~::'= 1 Mn is a convergent series of nonnegative constants
and lfn(x)I ~ Mn for ali neN and ali xeS, then
00
L. ln
n=I
converges uniformly on S.
263
The Derivative
Now we define a functionfwhich is continuous everywhere and

differentiable nowhere. Let
/1(x) = lxl
for
xe[ -!, !]
and extend the domain of / 1 to R by requiring that
f1(X
+ 1) = f1(X)
for ali xeR. For neN, n > 1, define
for ali xeR.

(b) Show that lfn(x)I ~ !{ 4 1- n) for ali x eR and ali n eN. Then show
that the series
converges uniformly on R, and hence that it converges to a

functionfwhich is continuous on R.
Finally, we prove that the functionf = ~.:'= tfn is nowhere differentiable.
(c) Let ceR. Show that if
Xn =
+4-1-n
lfm(Xn) - fm(c)I =
or
Xn =
{~-l -n
_4-1-n
ifm ~n
ifm >n
(d) Show that
is an even integer if n is even and is an odd integer if n is odd.

(e) Prove that
f'(c) = Lim/(x) - f(c)

x--+c
x- e
cannot exist.
Appendix
Cardinal Numbers
ln this appendix we continue the study of sets which we initiated in

Chapter 1. We begin by proving the important Schroeder-Bernstein
Theorem and showing that subsets and countable unions of countable sets
are themselves countable sets. We then tum to our main topic in this
appendix, cardinal numbers: we define the qoncept of the cardinal number
of a set, examine the properties and arithmetic of cardinal numbers, prove
Cantor's Theorem, and conclude with some remarks about the Continuum Hypothesis. The definitions, results, and examples of Section 1.3 of
Chapter 1 will form the basis of our work here, and the reader is urged to
review the material of Section 1.3 before proceeding.
Recall that we defined a set S to be equivalent to a set T if there is a
one-to-one function from S onto T. Since the existence of such a function
means that the elements of S and Tare paired off in a one-to-one manner,
it is clear that equivalent sets are the sarne "size," in the sense that they
have the sarne number of elements.
Suppose we wish to show that a given set S is equivalent to another
set T. ln arder to do this, we would need to find a one-to-one function
from S onto T, and this is not always easy to accomplish. The difficulty
usually lies in finding a function which is onto. However, it is often the
case that we can rather easily find a one-to-one function from S to T
which is not onto, and also a one-to-one function from Tto Swhich is not
265
266
Appendix
onto. ln such a situation S is equivalent to a subset of T and Tis equivalent to a subset of S, and intuition tells us that then S and T should be
equivalent. This important fact is indeed true; it is known as the
Schroeder-Bernstein Theorem.
(Schroeder-Bernstein) Let S and T be sets. If S is equivalent to a subset of T and T is equivalent to a subset of S, then S is
equivalent to T.
Theorem A.1
Proof: Let f: S ~ T and g: T ~ S be one-to-one functions, so that S is

equivalent to f(S) e T and Tis equivalent to g(T) e S. See Figure A. l.
The strategy of the proof is to find a subset X of T such that g - 1 is defined
on g(X) e S and Tis the disjoint union of X andf(S - g(X)) (see Figure
A.2), for then we can define a one-to-one function h from S onto T by
setting h equal to g - 1 on g(X) and h equal to f on S - g(X).
Figure A.l
S is equivalent to f(S), Tis equivalent to g(T).
S - g(X)
Figure A.2
Tis the disjoint union of X and f(S - g(X)).
267
Cardinal Numbers
Let "//'={V e T 1 V e T- f(S -g(V))}. Thus "//'is a set of subsets

of T. Since <bis a subset of every set, <b e T - f(S - g(f/J)), and hence f/Je"f/'.
Therefore "//' is a nonempty set of subsets of T.
Note that
()
AcBcT
implies
T-f(S -g(A)) e T- f(S -g(B)):
if A e B e T, then g(A) e g(B) e S, whence S - g(B) e S - g(A), which

implies f(S - g(B)) e f(S - g(A)), which in turn implies
T - f(S - g(A)) e T - f(S - g(B))
Let X= Uve-r V. For ease of notation, let us set
X'= T - f(S - g(X)). We will show that X= X'. If V e"f/', then
V e X e T and thus by virtue of ( *) above,
T-f(S -g(V)) e T-f(S-g(X)) =X'
However, Ve"f/' implies that V e T-f(S -g(V)), and therefore V e X'
for ali Ve"f/'. But then
X=
U VcX'
Ve"Y
and hence by ( *),

XcX'cT
implies
X'= T - f(S - g(X)) e T - f(S - g(X'))
which shows that X'e"f/'. Therefore X' e Uve-r V= X, and X e X',

X' e X imply that X = X' = T - f(S - g(X)).
Since X= T - f(S - g(X)), it follows that T - X = f(S - g(X)).
Define h: S-+ T by
if seg(X)
if seS-g(X)
The remainder of the proof consists of showing that h is one-to-one and
onto.
Suppose s 1 and s2 are elements of S. If
and
Appendix
268
then
and
h(s 1) e/(S - g(X)) = T - X
and hence h(s 1) = h(s2) is impossible. Thus if h(s 1) = h(s2), then either s 1
and s2 are both elements of g(X), in which case the fact that g - 1 is
one-to-one implies that s 1 = s2 , or s 1 and s2 are both elements of S - g(X),
in which case the fact that f is one-to-one implies that s 1 = s2 Therefore
h is one-to-one.
Finally, suppose te T. If te X, then g(t) eg(X) and thus
t = g - 1(g(t)) = h(g(t))
If te T - X= f(S - g(X)), then t

t
f(s) for some se S - g(X), and thus
= f(s) = h(s)
Therefore h is onto, and we are done. D
Examples A.2
1. The function f: R
-+ ( -
1, 1) defined by
f(x)
= 1 + lxl
is clearly one-to-one and thus Ris equivalent to a subset of ( - 1, 1). Since

( -1, 1) is equivalent to itself, it is equivalent to a subset of R. Thus by the
Schroeder-Bernstein Theorem, Ris equivalent to ( -1, 1).
2. By Example 4 of Examples 1.3.2 of Chapter l, any two closed

intervals having more than one point are equivalent; but every open
interval contains a closed interval with more than one point, so every
closed interval with more than one point is equivalent to a subset of every
open interval. Similarly, any two open intervals are equivalent, and every
closed interval with more than one point contains an open interval; thus
every open interval is equivalent to a subset of every closed interval with
more than one point. Therefore by the Schroeder-Bernstein Theorem,
every closed interval with more than one point is equivalent to every open
interval. Combining this result with that of Example 1 above, we see that
R is equivalent to every interval which contains more than one point.
We will use the Schroeder-Bernstein Theorem to give a rigorous
proof of the fact that every subset of a countable set is countable.
269
Cardinal Numbers
Proposition A.3
Every subset of a countable set is countable.
Proof: Let S be a countable set. If S is a finite, then by Corollary 1.3.6

every subset of S is finite, hence countable. If S is infinite, then it is
denumerable. Let T be a subset of the denumerable set S. If Tis finite, it
is countable. If Tis infinite, then by Corollary 1.3.9 it contains a denumerable subset T', and we have T' e Te S, with S and T' denumerable,
hence equivalent. Therefore S is equivalent to a subset of T. But T is
equivalent to a subset of S, namely itself, so by the Schroeder-Bernstein
Theorem, T is equivalent to S, and thus Tis denumerable. D
Next we prove that a countable union of countable sets is countable.

The proof is a slight generalization of the argument we gave in Section 1.3
when showing that the set Q of rational numbers is countable.
Proposition A.4 The countable union of countable sets is a countable set.
Proof: Let A be a countable index set and for each ixeA let X" be a
countable set. We must show that UoceA X" is a countable set.
Suppose first that A= N, Xn is denumerable for each neN, and the
sets Xn are mutually disjoint. For each n eNthere is a one-to-one function
f,, from N onto Xn. For each meN letfn(m) = Xnm in Xn. Thus for each
neN we have Xn = {xnh Xn 2 , , Xnm .. .} and it follows that we may
write the elements of UneN Xn in the following array:
Xi 1
X12
X13
X1.4
X21
X22
X23
X24
X31
X32
X33
X34
X41
X42
X43
X44
Since the sets Xn are mutually disjoint, the elements in the array are ali
distinct. Furthermore, every element of U,.eN Xn appears in the array. We
now define a function g: N-+ Xn by counting along diagonais ofthe array:
14
270
Appendix
Thus
g(l)
=X11,
g(2)
=X21>
g(3)
=X12
g(4)
=X31' ...
Since the function g is clearly one-to-one and onto, UneN Xn is equivalent to N and hence is countable.
Now we prove the proposition in its full generality. Suppose that A
is any countable index set and that Y"' is a countable set for every a e A.
We wish to show that UixeA Y"' is countable. Since A is countable, and
nonempty because it is an index set, A is equivalent to some subset of N.
This subset may be a proper subset of N, but in any case each a e A
corresponds to a unique neN and we may write Y"' = Yn. But ym being
countable, is equivalent to a subset (which may be proper) of the set Xn
used above, and thus UixeA Y"' is equivalent to a subset of UneN Xn. Since
we have shown that UneN Xn is countable, UixeA Y"' is countable by
Proposition A.3. D
Corollary A.5 The countable union of denumerable sets is a denumerable set.
The proof of the corollary is left to the reader.

Corollary A.6 The set Q of rational numbers is denumerable and hence
countable.
Proof: Writing Q = UneN {m/n 1m eZ} shows that it is the countable
union of the denumerable sets {m /n 1me Z}. D
Now we are ready to refine our techniques for comparing sets. We

will attach to each set a symbol called its cardinal number, then compare
sets by manipulating their cardinal numbers according to the rules of
cardinal arithmetic.
Definition A.7 Every set has attached to it a symbol called its cardinal
number. If S is a set, the cardinal number of S is denoted by !SI. Order
and equality among cardinal numbers are defined as follows: if S and T
are sets, then !SI~ ITI if S is equivalent to a subset of T, and !SI= ITI
if S is equivalent to T. If !SI~ ITI and !SI =F ITI, we write !SI< ITI. Furthermore, we say that a cardinal number is infinite if it is the cardinal
number of an infinite set, andfinite if it is the cardinal number of a finite
set.
Cardinal Numbers
271
Definition A. 7 is rather nebulous: it says that a cardinal number is a

symbol attached to a set, but it does not tell us which symbols to attach
to which sets. We remedy this in the following examples.
Examples A.8 1. By definition, the cardinal number of the empty set is

zero: 101 = O.
2. Let S be a nonempty finite set. By Corollary 1.3.9, there is some
n EN such that S is equivalent to {l, 2, ... , n }. We define ISI = n.
3. We define INI = ~ 0 (The symbol ~is the Hebrew letter aleph; the
cardinal number of the set N is thus "aleph-null. ") Note that
~o= INI = IZI = IQI
and that in fact, ISI = ~o for any denumerable set S. Furthermore, since
every infinite set contains a denumerable subset, ~o is the smallest infinite
cardinal number. Note also that ITI <~o for any finite set T.
4. We define IRI =e. (The letter e stands for continuum; the real
numbers form a continuum of points.) Note that e is an infinite cardinal
number and that ~o < e. (Why?)
Our first result concerning cardinal numbers shows that the arder
relation < behaves as expected.
ex, p, and y be cardinal numbers.

If ex ~ p and p ~ ex, then ex = p.
If ex ~ p and p ~ y, then ex ~ y.
Proposition A.9
1.
2.
Let
Proof: We prove (1), leaving (2) as an exercise. (See Exercise 2 at the

end of this appendix.)
Let S and Tbe sets such that ex= ISI and P= ITI. The inequality ex~ P
implies that S is equivalent to a subset of T, while p ~ ex implies that T
is equivalent to a subset of S. Therefore by the Schroeder-Bernstein
Theorem, S is equivalent to T, and hence ex = p. O
We begin our consideration of the arithmetic of cardinal numbers by

defining addition and multiplication for cardinais.
ex and p be cardinal numbers.

Let S and T be disjoint sets with ISI = ex and 1TI = p. Define
ex +P = ISuTI
Definition A.10
1.
Let
272
2.
Appendix
Let S and
Tbe sets with ISI = oc and ITI = p. Define

ocp
=IS x
TI
Examples A.11
1. ln this example we show that addition and multiplication of finite cardinal numbers is the sarne as addition and multiplication
of nonnegative integers. For instance, for any cardinal number oc, we have
oc + O= oc and oc O= O (here O is the cardinal number of the empty set),
because if S is a set with cardinal number oc, then
oc +o =
is u 01 = 1s1 = oc
and
(l(
o = IS
01 = 101 = o
Now suppose that oc and p are nonzero finite cardinais; then there exist
natural numbers n and m such that
oc
= 1{ 1,
2, ... , n} 1
and
p=
1{
1, 2, ... , m} 1
However, since {l, 2, ... , m} is clearly equivalent to {n + 1, . .. , n + m },

we have
oc +
P = 1{1, ... , n} u{n +
l, ... , n + m}I = l{l, ... , n + m}I = n + m
A similar argument shows that ocp = nm. (See Exerci se 3 at the end of this
appendix.)
2. The arithmetic of infinite cardinais is quite different from that of
finite cardinais. For instance, ~o+ ~o = ~0 To see this, let S and T be
disjoint denumerable sets; then
ISl=~o.
and
~o+~o= ISuTI
But since S, Tare denumerable, so is S u T, and hence IS u TI = ~ 0

3. Let us show that ~o ~o= ~ 0 Since ~o= INI, we have
~o ~o = IN x NI. But IN x NI is obviously denumerable, so
IN x Nl=~0 .
4. We claim that e+ e= e ande e= e. To prove the first of these
equalities, recall that every interval in R which has more than one point
Cardinal Numbers
273
is equivalent to R. Hence
e+ e= l[O, 1) u[l, 2ll = l[O, 2ll =e

To prove the second equality, note that e = l[O, ll 1, and hence
e e = l[O, ll x [O, ll 1
But since [O, ll is clearly equivalent to the subset {(x, O) 1x e[O, 11} of
[O, ll x [O, ll, we conclude that e ~e e. On the other hand, we may
represent each x e[O, ll by a unique decimal expansion, as in the proof of
Theorem 1.3.13, and use this to define a functionf: [O, ll x [O, ll-+[O, ll
as follows:
if x = O.a 1a 2
and y = O.b 1b2 are in [O, ll
let
f(x, y) = O.a 1b 1a 2b2
It is easy to check thatf is one-to-one (doso), and this establishes that

[O, ll x [O, ll is equivalent to a subset of [O, ll and thus that e e~ e.
Our next definition shows how to raise a cardinal number to a cardinal number power.
Definition A.12 Let S and T be sets. The power set T 8 is the set of ali
functions from S to T. If ex and f3 are cardinal numbers and S and T are
sets such that ISI =ex, ITI = /3, then 1311. = IT8 I.
Example A.13 Let S = {a, b} and T = {O, 1}. There are exactly four
functions from S to T, namely f, g, h, and k, defined by
f(a) = f(b) =O
g(a) = 1, g(b) =O
h(a) =O, h(b) = 1
k(a) = k(b) = 1
Therefore
T8 = {f, g, h, k}, and since 2 =

22 =
as expected.
ISI and 2 = ITI, we have
IT8 I = l{J.g, h, k}I = 4
Appendix
274
Notice that in the preceding example the set S has four distinct subsets, and each of these consists of the elements of S which one and only
one of the functions from S to {O, 1} carries onto 1. Thus
f/J= {xeS lf(x) = l}

{a}= {xeS 1 g(x) = l}
{b}={xeSlh(x)=l}
and
S= {xeS 1k(x) = l}
Clearly, there is a one-to-one correspondence between the power set
{O, 1}8 and the set of ali subsets of S, and therefore the power set {O, 1}8
is equivalent to the set of ali subsets of S. This is so for every set S: the
power set {O, 1} 8 is equivalent to the set of ali subsets of S. Before we
prove this it will be convenient. to define the characteristic function of a
set.
Let S be a set and let T be a subset of S. The characteristic function of Tis the function XT: S-+ {O, 1} defined by
Definition A.14
if xeT
if xeS-T
Let S be a set. The power set {O, 1}8 is equivalent to the
set of ali subsets of S.
Proposition A.15
Proof: Let Y' denote the set of ali subsets of S. For each TeY', the
characteristic function XT is an element of {O, 1} 8 , and
T{xeS 1XT(X) = l}
Define a function /: Y' -+ {O, 1} 8 as follows:
f(T) = XT
for ali TeY'
We will show thatfis one-to-one and onto and hence that Y' is equivalent
to {O, 1}8 .
Suppose that/(T) = f(V) for subsets Tand Vof S; then XT = Xv. and
Cardinal Numbers
275
hence Xr(x) = xv(x) for ali xeS. But then
T= {xeS 1Xr(x) = l} = {xeS 1xv(x) = l} =V

Therefore f is one-to-one.
Now suppose that g is an element of the power
a function from S to {O, l}. Let T = {xeS 1g(x)
g(x) = l and also XT(x) = l, while if xeS - T,
XT(x) =O. Thus f(T) = Xr = g, which shows that f
done. D
Corollary A.16
number 21s1.
set {O, l }s; then g is

= l}. If xeT, then
then g(x) =O and
is onto, and we are
If S is a set, the set of ali subsets of S has cardinal
Example A.17 We claim that 21{ =e. Let .A denote the set of ali subsets
of N. We know that
so it will suffice to show that l.Al =e. We will do this by demonstrating

that .A is equivalent to the closed interval [O, I].
Every xe[O, l] has a binary expansion of the form x = O.a 1a 2 ,
where an e {O, 1} for every n e N. If x has two such binary expansions, we
agree to use the one in which ali terms after some kth term are zero. With
this agreement in force, we may consider that every x e [O, l] has a unique
binary expansion.
If xe[O, l] has the binary expansion x = O.a 1a 2 , define
f(x) = {neN 1an "#O}
Clearly f is a function from [O, l] to .A, and if f(x) = f(y), then x and y
have the sarne nonzero terms in their binary expansions and hence have
the sarne binary expansions, so x = y. This shows thatfis one-to-one and
thus that [O, l] is equivalent to a subset of .A.
Now we show that .A is equivalent to a subset of [O, l]. Each xe [O, l]
has a ternary expansion x = O.b 1b2 , where bne{O, 1, 2} for ali neN. If
x has two such expansions, then in one of them every term after some kth
term is a 2. Define a function g: .A -+ [O, 1] as follows: for every subset M
of N, g(M) = O.b 1b2 , where O.b 1b2 is the ternary expansion with
ifneM
ifneN-M
276
Appendix
If M and M' are subsets of .;V such that M # M', then the ternary
expansions g(M) and g(M') are different in the sense that their terms are
not identical. If g(M) and g(M') represent the sarne real number in [O, l],
then every term in one of them past some kth term must be a 2. But this
is impossible, for by the way g is defined, neither g(M) nor g(M') can have
a 2 in its ternary expansion. Therefore g is one-to-one, so .;V is equivalent
to a subset of [O, l], and the Schroeder-Bernstein Theorem now implies
that .;V is equivalent to [O, l]. O
If tX is a finite cardinal number then of course tX < 2"'. As the preceding
example suggests, this inequality remains true for infinite cardinais. This
result is known as Cantor's Theorem.
Theorem A.18 (Cantor)
For any cardinal number
tX, tX
< 2"'.
P;oof: Let S be a set such that tX = ISI; then 2"' is the cardinal of !/, the
set of ali subsets of S. Therefore it will suffice to show that for every set
S, S is equivalent to a subset of !/, but that S cannot be equivalent to!/.
Let f: S-+ !/ be given by f(x) = { x} for ali x E S. Since f is obviously
one-to-one, S is equivalent to a subset of !/.
We show that S cannot be equivalent to!/ by proving that ifjis any
function from S to!/, thenf cannot be onto. Letfbe a function from S
to!/, so thatf(x) is a subset of S for ali xeS. Let T be the subset of S
defined by T = { x E S 1 x f(x)}. The set Te!/, but there is no x E S such
thatf(x) = T. To see this, supposef(x) = Tand consider the cases xET,
x T: if x E T, then by the definition of T, x f(x) = T, while x T = f(x),
then by the definition of T, x E T. Since each possibility leads to a contradiction, f cannot be onto, and we are done. O
Corollary A.19 There is no largest cardinal number.

According to Cantor's Theorem, we can construct a chain of strictly
increasing cardinal numbers, thus:
The first three of the cardinais in this chain are cardinais of what might be
called "naturally-occurring" sets: ~o is the cardinal of the natural numbers, e the cardinal of the reais, and 2c the cardinal of the set of ali graphs
which can be drawn in the Euclidean plane. (See Exercise 15 at the end of
this appendix.) To our knowledge, no one has ever given an example of a
Cardinal Numbers
277
naturally-occurring set which has a cardinal greater than 2c. For this
reason it is sometimes said that as far as infinite cardinal numbers are
concerned, we can only count up to three.
Let us again consider the chain
Are there cardinal numbers between the ones in this chain? To be more
specific, is there a cardinal number between ~o and e? This leads us to a
statement of the famous Continuum Hypothesis:
Continuum Hypothesis.
There is no cardinal number oc such that

~o<
oc <e
The Continuum Hypothesis was a famous unsolved problem in mathematics for over 50 years. ln order to disprove it, one would have to
exhibit a set with cardinal greater than ~o but less than e; in other words,
a set "larger" than Q but "smaller" than R. Since no such set could be
found, the Continuum Hypothesis seemed reasonable, but ali efforts to
prove it failed. P. Cohen finally settled the matter by showing that in fact
the Continuum Hypothesis is independent of the other axioms of set
theory. This means that it cannot be proved as a theorem of set theory,
nor can it be disproved. If we wish it to. be true, we must take it as an
axiom of set theory; if we wish it to be false, we must take its denial as an
axiom.
We conclude this appendix with a paradox. Let A be the set of ali
cardinal numbers. For each oceA let S"- be a set such that oc = IS"-1' and
consider Y' = urJ.EA s(J.. Each set s(J. is obviously equivalent to a subset of
Y' and therefore
for ali oc e A. Therefore jY' 1 is the largest cardinal number. But this contradicts Corollary A.19, so we have a paradox in the theory of cardinal
numbers. What is wrong? Notice that at the beginning of our construction
of the largest cardinal we said, "let A be the set of ali cardinal numbers."
ln fact, the collection of ali cardinal numbers is nota set; it is too large to
be a set, and it cannot be treated as if it were one. Thus our construction
Appendix
278
of the largest cardinal is invalid. ln order to avoid paradoxes such as this,

it is necessary to define the concept of a set quite precisely. We have not
done this, for it would have taken us much too far afield. The interested
reader can find a good introduction to axiomatic set theory in J. L.
Kelley's General Topology.
EXERCISES
1.
2.
3.
Prove Corollary A.5.

Prove part (2) of Proposition A.9.
Let tX and f3 be finite cardinais, with
tX = l{l, ... , n}I
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
and
/3 = l{l, ... , m}I
Prove that tX/3 = nm.

Prove that if tX is any cardinal number, then tX 1 = tX.
Find the cardinal numbers of the following sets:
(a) The set of ali open intervals in R;
(b) The set of ali open intervals in R which have rational endpoints;
( c) The Cantor ternary set;
(d) The set of ali lines in the Euclidean plane R 2 .
Let tX, /3, y, and be cardinal numbers. Prove that if tX ~ f3 and y ~,
then tX + y ~ /3 +.
Prove that if tX and f3 are cardinal numbers with tX ~~o and f3 ~ ~ 0 ,
then tX + /3 ~ ~ 0
Let tX, {3, y, and be cardinal numbers. Prove that if tX ~ f3 and y ~,
then tX}' ~ {J.
Prove that if tXllnd f3 are cardinal numbers with tX ~~o and f3 ~ ~ 0 ,
then tX/3 ~ ~0
Prove that if tX is any cardinal number, then tX = 1.
Let tX, {3, and y be cardinal numbers. Prove that tXPtXY = tXP+Y.
Let tX, {3, and y be cardinal numbers. Prove that (tXP)Y = tXP1
Prove that
(a) ct{0 =c;
(b) ~ 01{0 =e.
Prove that cc = 2".
Let S be the set of ali graphs which can be drawn in the Euclidean
plane R 2 . Prove that ISI = 2".
A polynomial of degree n is a function p: R -+ R whose defining
equation is of the form
p(x)
= OnXn + an-1Xn- I +'' '+ 01X +ao
Cardinal Numbers
279
where ak e R for O~ k ~ n and an # O. The real numbers ak are called

the coefficients of the polynomial. A real number r is an algebraic
number if for some n eN there exists a polynomial p of degree n
having rational coefficients such that p(r) = O. If there is no such
polynomial, the real number r is said to be a transcendental number.
Prove that the set of ali algebraic numbers is countable and that the
set of ali transcendentaLnumbers is uncountable. You may use the
fact that for a polynomial p of degree n there are at most n real
numbers y such that p(y) =O. {Hint: begin the proof by showing
that the set of ali polynomials of degree n which have integer coefficients is countable.)
Bibliography
ANALYSIS
Bartle, Robert G. ( 1964). The Elements of Real Analysis. John Wiley and
Sons: New York.
Gelbaum, Bernard R. and Olmsted, John M. H. (1964). Counterexamples
in Analysis. Holden-Day: San Francisco.
Goffman, Casper. (1961). Real Functions. Holt, Rinehart and Winston:
New York.
Randolph, John F. ( 1968). Basic Real and Abstract Analysis. Academic
Press: New York.
Royden, H. L. ( 1963). Real Analysis. The Macmillan Company: New
York.
Rudin, Walter. ( 1964). Principies of Mathematical Analysis. McGrawHill: New York.
TOPOLOGY
Dugundji, James. ( 1970). Topology. Allyn and Bacon: Boston.
Kelley, John L. (1964). General Topology. D. Van Nostrand Company:
Princeton, New Jersey.
Moore, Theral O. (1964). Elementary General Topology. Prentice-Hall:
Englewood Cliffs, New Jersey.
281
282
Bibliography
Wilansky, Albert. ( 1970). Topvlogy for Analysis. Ginn: Waltham, Massachusetts.

SET THEORY AND CARDINAL NUMBERS
Fraenkel, A. A. ( 1966). Set Theory and Logic. Addison-Wesley: Reading,

Massachusetts.
Halmos, Paul R. (1960). Naive Set Theory. D. Van Nostrand Company:
Princeton, New Jersey.
lndex
Absolute value, 41
Absolute value function, 18
Absolute value metric, 140
Algebraic number, 279
Alternating series, 133
Approximation theorems, 192-199
Archimedean Property, 46
Arzela-Ascoli Theorem, 189
Bernstein polynomials, 195
Binomial Theorem, 195
Bolzano-Weierstrass Theorem, 162
Boundary of a set, 70
Bounded function, 185
Bounded sequence, 116
Bounded set, 43
Boundedness Test, 129
Cantor ternary set, 48-50
Cantor's Diagonalization Procedure, 37-38
Cantor's Theorem, 276
Cardinal numbers, 270-278
arithmetic of, 271-276, 278
finite and infinite, 270
set of ali, 277
Cartesian product of sets, 7
Cauchy Criterion, 160, 184
Cauchy Sequence, 158
Cauchy-Schwartz inequality, 148, 149
Chain Rute, 247
Characteristic function, 79, 87, 274
Closed bali, 142
Closed set, 58, 66-74
Closure of a set, 70
Compactness, 96-102
and uniform continuity, 154
in Hausdorff space, 100

preservation of, 97
Comparison Test, 129
Complement of a set, 7
Complete metric space, 158
Completeness Property, 45-46, 54-55
Completion of a metric space, 165-175
Composition of functions, 26
Connectedness, 90-96
and intervals, 94
preservation of, 93
Constant function, 21
Construction of the real numbers, 51-56,
172-173
Continuity, 75-90, 149-156
and compactness, 97
and connectedness, 93
and differentiability, 246
and functional limits, 136
at a point, 81
delta-epsilon definition of, 150
from the left and from the right, 89
in metric spaces, 149-156
uniform, 152-155
Continuous function, 78
Continuum Hypothesis, 277
Convergence, of a sequence, 108
of a series, 127
pointwise, 177
uniform, 182-183
Countability Criterion, 35
Countable set, 35-37, 269-270
Dedekind cut, 52
deMorgan's laws, 10
283
lndex
284
Dense subset, 71
Denumerable set, 35, 270
Derivative, 241-263
of a function at a point, 242
Derived set, 67
Difference quotient, 242
Differentiability, 242
and continuity, 246
of a sequence of functions, 255-256, 261
Differentiation, rules of, 247-248
Disconnected set, 91
Discontinuous function, 78
Discrete metric, 141
Discrete topology, 59
Disjoint sets, 7
Divergence, of a sequence, 108, 118
of a series, 127
Domain of a function, 16
Empty set, 1
Equicontinuity, 188
Equivalence class, 165
Equivalence relation, 165
Equivalent sets, 32
Euclidean n-space, 148
Euclidean plane, 9
Extreme Value Theorem, 101
Finite Intersection Property, 104
Finite set, 11, 31-39
Finite subcovering, 96
First Derivative Test, 259
First Mean Value Theorem for Integrais,
234
Fixed point, 164
Function, 16-31
absolute value, 18
and preservation of set operations, 24
bounded, 185
composite, 26
constant, 21
continuous, 78
continuous everywhere, differentiable
nowhere, 263
derivative of, 242
difference quotient of, 242
differentiable, 242
discontinuous, 78
domain of, 16
graph of, 18
image under, 22
inverse, 27-28
one-to-one, 20
onto, 20
oscillation of, 221
piecewise linear, 199
preimage under; 22
range of, 16
restriction of, 25
Riemann integrable, 210
Riemann-Stieltjes integrable, 224
sequences of, 177-199
space, 191
step, 192
Function spaces, 190-192, 198-199
Functional limits, 134-137
and continuity, 136
and sequences, 135-136
properties of, '!37
Fundamental Theorem of Calculus, 253
Geometric series, 128
Graph of a function, 18
Greatest lower bound, 44
Greenbaun, N. N., vii
Harmonic series, 127
Hausdorff space, 99
Heine-Borel Theorem, 101
Hilbert space, 149
Homeomorphism, 86
Image, 22
Index set, 11-12
lndiscrete topology, 59
Infimum, 44
lnfinite series, 126-133, 262
alternating, 133
boundedness test, 129
comparison test, 129
convergence tests for, 129-131, 132-133,
262
convergent, 127
divergence test, 128
divergent, 127
geometric, 128
285
Index
harmonic, 127
integral test, 133
nth partia! sum, 126
of functions, 262
p-series, 130
partia! sum of, 126
ratio test, 130
root test, 133
Infinite set, 11, 31-39
lntegers, 2
Integrability theorems, 214-217, 219, 224,
228-233, 239
Integral Test, 133
lntegration by Parts Theorem, 260
lntegration by Substitution Theorem, 254
Interior of a set, 70
Intermediate Value Theorem, 93
Intersection of sets, 7
Interval, 3-4
partition of, 201
continuous image of, 96, 102
Inverse function, 27-28
lrrational numbers, 39
lsometry, 167
L'Hospital's Rute, 260
100 ' 142
/ 1, 147
Law of the Mean, 251
Cauchy form, 260
Least upper bound, 44
Limit inferior, 120
and convergence, 123
Limit of a function, 134
and sequences, 135
properties of, 137
Limit of a sequence, 108
Limit point, 67
Limit superior, 120
and convergence, 123
Lipschitz condition, 156
Local continuity, 81
Lower Riemann sum, 202
properties of, 207-210
Lower bound, 43
Measure zero, 221
Metric, 140
2-adic, 146--147
absolute value, 140
discrete, 141
natural, 140
product, 141
supremum, 142, 191
Metric space, 144-175
complete, 158
completion of, 165-175
continuity in, 149-156
Euclidean n-space, 148
Hilbert space, 149
/ 00 ' 142
/ 1, 147
sequences in, 156--165
uniform continuity in, 152-155
Metric topology, 144
Minkowski inequality, 148, 149
Monotonic sequence, 117
Monotonically increasing function, 226
Mutually disjoint sets, 12
Natural metric, 140
Natural numbers, 2
Neighborhood, 65
Nested Intervals Theorem, 161
Norm, 149, 190
Normal space, 105
Nul! set, 1
One-to-one function, 20
Onto function, 20
Open bali, 142
Open covering, 96
Open ray topology, 62, 98-99
Open set, 58, 66--74
in the natural topology, 59-60, 61
Ordered pair, 7
Oscillation of a function, 221
p-adic valuation on Q, 146--147
p-series, 130
Partia! sum, 126
Partition, of an interval, 201
of a set, 166
Piecewise linear function, 199
Pointwise convergence, 177-182
Power set, 273
286
Preimage, 22
Product metric, 141
Product topology, 65
Range of a function, 16
Ratio Test, 130
Rational numbers, 2
completion of, 172-173
countability of, 35-36, 270
dense in reais, 46, 72
incompleteness of, 48, 158
Rays, 4-5
Real tine, 2
Real numbers, 2, 41-56
Archimedean Property, 46
as a complete metric space, 159-160
as Dedekind cuts, 53-56
as equivalence classes of Cauchy sequences, 173, 174-175
completeness of, 45-46, 54, 159
construction of, 51-56, 172-173
distance between, 42
natural topology on, 59
sequences of, 116-126
series of, 126-133
triangle inequality for, 42
uncountability of, 37
Refinement of a partition, 202
Regular space, 104
Relation on a set, 165
Relative maximum, minimum, 249
Restriction of a function, 25
Riemann integral, 210-226
and continuity, 216
and differentiation, 251
and sequences of functions, 234-237, 239
First Mean Value Theorem, 234
integrability theorems, 214-217, 219-224,
228-233, 239
integration by parts, 260
integration by substitution, 254
Riemann sum, 220
upper, lower, 202
Riemann-Stieltjes integral, 224
integrability theorems, 225-226
properties of, 241
Riemann-Stieltjes sum, 224
Index
Rolle's Theorem, 250
Root Test, 133
Schroeder-Bernstein Theorem, 266
Separation axioms, 104-105
Sequence, 107-126, 156-165, 177-199
and continuity, 111-112
bounded, 116
Cauchy, 158
Cauchy Criterion, 160
convergent, 108
divergent, 108
in Hausdorff space, 110
in a metric space, 156-165
limit inferior of, 120
limit of, 108
limit superior of, 120
monotonic, 117
of functions, 177-199
of real numbers, 116-126
Sequence of functions, 177-199, 234-238,
255-257
and differentiability, 255-257, 261
and integrability, 234-237, 239
Cauchy Criterion, 184
equicontinuous, 188
pointwise convergence of, 177
uniform convergence of, 182-183
uniformly bounded, 188
Series, of functions, 262
of real numbers, 126-133
Set, 1-16
boundary of, 70
bounded, 43, 163
Cantor ternary, 48-50
Cartesian product, 7
characteristic function of, 274
closure of, 70
compact, 96
complement of, 7
connected, 91
containment, 5
countable, 35-37, 269-270
denumerable, 35, 270
derived, 67
disconnected, 91
disjoint, 7
element of, 1
287
lndex
empty, 1
equivalence, 32
finite, 11, 31-39
identities, 10, 13-14
index, 11-12
infinite, 11, 31-39
interior of, 70
intersection, 7
limit point of, 67
of measure zero, 221
operations, 7
partition of, 166
power, 273
proper subset of, 5
relation on, 165
subset of, 5totally bounded, 163
unbounded,43
uncountable, 35
union, 7
Step function, 192
Subsequence, 125 '
Subset, 5
Subspace topology, 62
Supremum, 44
Supremum metric, 142, 191
Supremum norm, 190
Topological space, 57-65
closed s'et in, 58
Hausdorff space, 99
neighborhood in, 65
point of, 58
subspace of, 62
normal space, 105
regular space, 104

T1-space, 104-105
Topologies, comparison of, 60-61
Topology, 58
closed set of, 58
discrete, 59
indiscrete, 59
metric, 144
natural, 59
open ray, 62
open set, 58
product, 65
subspace, 62
Totally bounded set, 163
Transcendental number, 279
Triangle inequality, 42, 140
Unbounded set, 43
Uncountable set, 35
Uniform approximation, 192
Uniform boundedness, 188
Uniform continuity, 152-155
and compactness, 154
Uniform convergence, 182-185
and preservation of continuity, 186
and preservation of differentiability, 255257, 261
and preservation of integrability, 234237, 239
Union of sets, 7
Upper Riemann sum, 202
Upper bound, 43
Weierstrass Approximation Theorem, 196
Weierstrass M-Test, 262

(Pure and Applied Mathematics) Donald L. Stancl, Mildred L. Stancl-Real Analysis With Point-Set Topology-Dekker (1987)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Pure and Applied Mathematics) Donald L. Stancl, Mildred L. Stancl-Real Analysis With Point-Set Topology-Dekker (1987)

Uploaded by

Copyright:

Available Formats

Real Analysis

MARCEL DEKKER, INC.

NEW YORK ANO BASEL

Library of Congress Cataloging-in-Publication Data

COPYRIGHT 1987 MARCEL DEKKER, INC. ALL RIGHTS

Current printing (last digit):

This book is designed as a text for a first course in real analysis. It is

addition we have also included material on the construction of the reais

ing construction of the reais (Section 2.2), completion of metric spaces

1 Sets and Functions

2 The Real Numbers

5 Sequences and Series

5.3 Infinite Series

The Metric Topology

7.1 Pointwise and Uniform Convergence

8.1 The Riemann Integral

Appendix : Cardinal Numbers

We begin our study of sets with a definition.

Definition 1.1.1 A set is a collection of objects, called the elements of the

Sets and Functions

1. The following notation will be standard throughout

Z={O, 1, -1,2, -2, ... }

We assume that the reader is familiar with the arithmetic operations

Figure 1.2 The real line.

we may also write

3. Any statement of the form {x 1x has property P} where P is a

f/J={x lxeR, x 2 <0}

The open interval (a, b).

5. Let aeR and beR, with

[a, b] = {xeR 1a~ x ~ b}

Sets and Functions

Figure 1.4 The closed interval [a, b].

6. If aeR and beR with a< b, then the sets

(a, b] = {xeR 1 a <x ~b}

(a, +oo) = {xeR 1a <x}

(-oo, a)= {xeR 1 x <a}

[a, +oo) = {xeR 1a ~x}

Figure 1.5 Half-open intervals.

anda closed ray and may be written as

Definition 1.1.3 A set S is a subset of a set T if and only if every element

Sets and Functions

Figure 1.7 Set containment.

1. The following are true for any set S:

2. For the sets N, Z, Q, and R we have

(a, b) is a proper subset of (a, b]

+ oo) is a proper subset of (a, + oo)

Every interval in R, whether open, half-open, or closed, is a proper

We claim that S = T. To prove this, we must show that Se T and that

T e S. To show that S e T, let x be an arbitrary element of S. The

The technique used in the previous example is the standard method

(see Figure l.8c).

Sets and Functions

Figure 1.8 (a) Union, (b) intersection, (e) complement of sets.

1. Let S = {!, 2, 3} and T = {3, 4}. We have

The set R 2 is called the Euclidean plane; it may be represented by the

[l, 2] n[2, 3] = {2}

The Cartesian coordinate system for R 2

Sets and Functions

Figure 1.10 The Cartesian product [ l, 2] x (2, 3].

Let S, T, and V be sets. The following set identities

Proof: We shall prove the identity S u (T n V) = (S u T) n (S u V) of

[S u (T n V)] e [(S u T) n (S u V)]

But, as above, this inclusion is also true if (S u T) n (S u V) = </J,