Principles of Real Analysis-Solutions-Aliprantis PDF

PROBLEMS IN
REAL ANALYSIS
Second Edition
A Workbook with Solutions
CHARALAMBOS D. ALIPRANTIS
Departments of Economics and Mathematics
Purdue University
and
OWEN BURKINSHAW
Departments of Mathematical Sciences
Indiana University, Purdue University, Indianapolis
ACADEMIC PRESS -- - -
San Diego London Boston

.-
, i
._ .'
--::.
....•--. .. _- --
New York Sydney Tokyo Toronto

This book is printed on acid-free paper. e
Copyright © 1999, 1990 by Academic Press
All Rights Reserved.

No part of this publication may be reproduced or
transmitted in any form or by any means, electronic
or mechanical, including photocopy, recording, or
any information storage and retrieval system, without
permission in writing from the publisher.
ACADEMIC PRESS
a division of Harcourt Brace & Company
525 B Street, Suite 1900, San Diego, CA 92101-4495, USA
http://www.apnet.com
ACADEMIC PRESS LIMITED

24-28 Oval Road, London NWI 7DX, UK
http://www.hbuk.co.uk/ap/
International Standard Book Number 0-12-050253-4
This book accompanies the following title, catalogued by the Library of Congress:
Library of Congress Cataloging-in-Publication Data
Aliprantis, Charalambos D.
Problems in real analysis ICharalambos D. Aliprantis and Owen Burkinshaw.
p. cm.
Includes bibliographical references and index.
ISBN 0-12-050257-7 (acid-free paper)
1. Mathematical analysis. 2. Functions of real variables. 1. Burkinshaw, Owen. TI. Title.
QA300.A48 1998
515-dc21 98-3955
CIP
Printed in the United States of America

98 99 00 01 02 DS 9 8 7 6 5 4 3 2 1
CONTENTS
Foreword vii
CHAPTER 1. FUNDAMENTALS OF REAL ANALYSIS

1. Elementary Set Theory 1
2. Countable and Uncountable Sets 6
3. The Real Numbers 11
4. Sequences of Real Numbers 20
5. The Extended Real Numbers 34
6. Metric Spaces 45
7. Compactness in Metric Spaces 54
CHAPTER 2. IDPOLOGY AND CONTINUITY 65

8. Topological Spaces 65
9. Continuous Real-Valued Functions 73
10. Separation Properties of Continuous Functions 92
11. The Stone-Weierstrass Approximation Theorem 98
CHAPTER 3. THE THEORY OF MEASURE 107

12. Semirings and Algebras of Sets 107
13. Measures on Semirings 112
14. Outer Measures and Measurable Sets 116
15. The Outer Measure Generated by a Measure 122
16. Measurable Functions 133
17. Simple and Step Functions 137
18. The Lebesgue Measure 146
19. Convergence in Measure 157
20. Abstract Measurability 160
v
vi CONTENTS
CHAPTER 4. THE LEBESGUE INTEGRAL 171

21. Upper Functions 171
22. Integrable Functions 174
23. The Riemann Integral as a Lebesgue Integral 190
24. Applications of the Lebesgue Integral 206
25. Approximating Integrable Functions 220
26. Product Measures and Iterated Integrals 224
CHAPTER 5. NORMED SPACES AND Lp-SPACES 239

27. Normed Spaces and Banach Spaces 239
28. Operators between Banach Spaces 245
29. Linear Functionals 251
30. Banach Lattices 259
31. Lp-Spaces 271
CHAPTER 6. HILBERT SPACES 297

32. Inner Product Spaces 297
33. Hilbert Spaces 310
34. Orthonormal Bases 325
35. Fourier Analysis 333
CHAPTER 7. SPECIAL IDPICS IN INTEGRATION 345

36. Signed Measures 345
37. Comparing Measures and the
Radon-Nikodym Theorem 353
38. The Riesz Representation Theorem 365
39. Differentiation and Integration 379
40. The Change of Variables Formula 395
FOREWORD
This book contains complete solutions to the 609 problems in the third edition
of Principles of Real Analysis, Academic Press, 1998. The problems have been
spread over forty sections which follow the format of the book.
AH solutions are based on the material covered in the text with frequent refer-
ences to the results in the text. For instance, a reference to Theorem 7.3 refers to
Theorem 7.3 and a reference to Example 28.4 refers to Example 28.4, both in the
third edition of Principles of Real Analysis.
This problem book will be beneficial to students only if they use it "properly."
That is to say, if students look at a solution of a problem only after trying very
hard to solve the problem. Students will do themselves great injustice by reading
a solution without any prior attempt on the problem. It should be a real challenge
to students to produce solutions which are different from the ones presented here.
We would like to express our most sincere thanks to all the people who 'made
constructive recommendations and corrections regarding the text and the problems.
Special thanks are due to Professor Yuri Abramovich for his contributions and
suggestions during the writing of this problem book.
C. D. ALIPRANTIS AND O. BURKINSHAW

West Lafayette, Indiana
July, 1998
vii
I
I
I
I
I
I
CHAPTER 1 ______________
FUNDA:MENTALSOF
REAL ANALYSIS
1. ELEMENTARY SETTIIEORY
Probiem 1.1. Establish the following set theoretic relations:

1. (A U B) n C = (A n C) U (B n C) and
(A n B) U C = (A U C) n (B U C);
2. (A U B) \ C = (A \ C) U (B \ C) and
(A n B) \ C = (A \ C) n (B \ C);
3. A \ B = A nBc;
4. A S;;; B ¢::::::> BC S;;; N; and
5. (A U B)C = N n B C and (A n B)C = ACU B C.
AlsoJor an arbitrary function f: X ~ Y, establish the following claims:
6. f(UEI Ai) = UEI f(Ai);
7. f(nEI Ai) S;;; nEI f(A i);
8. f-I(UiEI Bi) = UEI f-I(Bi);
9. f-l(nEI Bi) = nEI f-I(Bi);and
10. f- I (Bc) = [J-I(B) r
Solution. (1) We establish the first formula only. We have
x E (A U B) n C ¢::::::> x E A U B and x E C
¢::::::> [x E A or x E B] and x E C
¢::::::> [x E A and x E C] or [x E B and x E C]
¢::::::> xEAnC orxEBnC
¢::::::> x E (A n C) lJ (B n C).
(2) Again, we establish the first formula only. Observe that
x E (A U B) \ C ¢::::::> x E A U B and x 1. C
1
2 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
~ [X E A or x E B] and x rf. C
~ [x E A and x rf. C] or [x E B and x rf. C ]
~ x E A \ C or x E B \ C
~ XE (A \ C) U (B \ C).
(3) Note that
xEA\B ~ x E A and x rf. B

~ X E A and x E B C ~ X E AnB c.
(4) Let .:1 S;; B. The~, x E BC implies x rf. B and so x rf. A (i.e., x E A C) so that
B C S;; A C: On the other hand, if B C S;; A C holds, then (by the preceding case) we
have A = =
(AC)C S;; (BC)C B.
(5) Note that
X E (A n B)C ~ xrf.AnB ~ xrf.A orxrf.B

~ x E AC or x E B C ~ X E ACU B C.
Moreover,
x E (A U B)C ~ x rf. A U B
~ x rf. A and x rf. B
~ X E A and x E B C ~ X E ACnBc.
C
(6) We have
Y E J(UAi) ~ :3 x E UAi with Y = f(x)

iEI iEI
~ :3 i E I with x E Ai and y = f(x)
~ :3 i E I with Y E f(Ai) ~ Y E U f(Ai)'
iEI
(7) From the inclusion f ( ni EI Ai) S;; f (A j) for each j, we see that
f(n Ai) S;;
iEI iEI
n
f(Ai).
(8) We have
x E f-1(UB i ) ~ f(X)'E UBi ~ :3 i E I with f(x) E Bi

iEI iEI
~ :3 i E I with x E f-1(Bi) ~ x E Uf-1(B i ).
iEI
Section 1: ELEMENTARY SET THEORY 3
(9) Note that
x E f-l(nBi) ..;=> f(x) E nBi ..;=> f(x) E Bi for each i E I

ieI ieI .
..;=> x E f-I(Bi) foreach i E I ..;=> x E nf-I(Bi ).

ieI
(10) Observe that
x E f- I (BC) ..;=> f(x) E BC ..;=> f(x) rj B

..;=> x rj f-I(B) ..;=> x E [I-I (B»)".
Problem 1.2. For two sets A and B show that the following statements are
equivalent:
a. A ~ B;
b. AU B = B;
c. A nB = A.
Solution. (a)=> (b)Clearly,B ~ AUB holds. On the other hand, if x E AUB,
then x E A or x E B, and so in either case, x E B. This means A U B ~ B, and
hence, A U B = B.
(b) => (c) By part (1) of the preceding problem, we have
An B = A n (A U B) = (A n A) U (A n B) = A U (A n B) = A.
(c) => (a) Clearly, A = A nB ~ B.
Problem 1.3. Show that (AflB)flC = Afl(8 flC) holds for every triplet of
sets A, B, and C.
Solution. Note first that for any three sets X, Y, and Z we have
X flY \ Z = [X \ (Y U Z)] U [Y \ (X U Z)]
and
Z \ (X fly) = [Z \ (X U Y)] U [X n Y n Z].
For instance, to verify the first identity, note that
x E X flY \ Z ..;=> [x E X \ Y or x E Y \ X] and x rj Z

..;=> [XEX,xrjY, andxrjZ] or [XEY, xrjX, andxrjZ]
..;=> x E [X \ (Y U Z)] U [Y \ (X U z)].
Thus,
(AIlB)IlC = [(AIlB) \ C] u [C \ (AIlB)]

= [A \ (B U C)] U [B \ (A U C)] U [C \ (A U B)] U [A n B n C]
= ([A\(B U C)] U (A n B nC)} U ([B\(C U A)] U [C\(B U A)]}
= [A \ (B IlC) ] U [ (B IlC) \ A]
= AIl(BIlC).
Problem 1.4. Give an example of a function f: X -+ Y and two subs~ts A and

B of X such that f(A n B) i= f(A) n f(B).
Solution. Define f: {O, I} -+ {O, I} by f(O) = f(l) = 0. If A = {OJ and

B = {I}, then f(A n B) = 0 i= {OJ = f(A) n f(B).
Problem 1.5. For a function f: X -+ Y, show that the following three state-
ments are equivalent:
a. f is one-to-one.
b. f(A n B) = f(A) n f(B) holdsforall A, B E P(X).
c. For every pair ofdisjoint subsets A and B of X, we have f(A)n f(B) = 0.
Solution. (a) ==> (b) If Y E f(A) n f(B), then there exist a E A and b E B
with Y = f(a) = f(b). Since f is one-to-one, a = b E A n B, and so
Y E f(A n B). Thus, f(A) n f(B) £; f(A n B) £; f(A) n f(B).
(b) ==> (c) Obvious.
(c) ==> (a) Let f(a) = f(b). If a i= b, then the two sets A = {a} and B = {b}
satisfy An B = 0, while f(A) n f(B) = {f(a)} i= 0.
Problem 1.6. Let f; X -+ Y be afunction. Show that fU- 1(A» £; A for all
A£; Y,and B £; f- 1U(B»for all B £; X.
Solution. Clearly, x E f- 1(A) if and only if f(x) E A. Thus, fU- 1(A» £; A.

Similarly, x E f- 1U(B» if and only if f(x) E f(B), and so B £; f- 1U(B»
holds.
Problem 1.7. Show thatafunction f; X -+ Y is onto ifand only if fU- 1(B»=B

holds for all B £; Y.
Solution. Assume that f is onto and B £; Y. If b E B, then there exists some

a E X with f(a) =
b; clearly, a E f- 1(B). Thus, b =
f(a) E fU- 1(B», and
1
so B £; fU- (B» £; B holds.
Section 1: ELEMENTARY SET THEORY 5
For the converse, note that the relation l(f-\({b))) = {b} implies I-\({b}) =1=
o for each bEY so that I is onto.
Problem 1.8. Let X ~Y .....!!..:,. Z. II A: S; Z, show that
(g 0 f)-leA) = I-\g-\(A)).
Solution. Note that
Problem 1.9. Show that the composition ollunctions satisfies the associative
law. That is, show that if X ~ Y ~ Z ~ V, then (h 0 g) ~ I = h 0 (g 0 f).
Solution. Observe that for each x E X we have
[(h 0 g) 0 I lex) = h 0 g(f(x)) = h(g(f(x))) = h«g 0 f)(x)) = [h 0 (g 0 f)](x).
Therefore, (h 0 g) 0 I = h 0 (g 0 f).
Problem 1.10. Let I: X -+ Y. Show that the relation R on X, defined by

X\RX2 whenever I(xd = l(x2), is an equivalence relation.
Solution. We must show that the relation R is reflexive, symmetric, and trans i-
tive.
Reflexivity: Note that I(x) = I(x) implies xRx for each x E X.
Symmetry: Let X\RX2. Then, I(x\) = l(x2)
= I(x\), so that X2Rx\.
or l(x2)
Transitivity: If x\ RX2 and X2Rx3, then I(x\) = l(x2) and l(x2) = l(x3) both
hold. It follows that I(x\) = l(x3), and so X\RX3.
Problem 1.11. II X and Yare sets, then show that
P(X) n P(Y) = P(X n Y) and P(X) U P(Y) S; P(X U Y).
Solution. (a) Note that
A E P(X) n P(Y) {:=::> A S; X and A S; Y

{:=::> A S; X nY {:=::> A E P(X n Y).
(b) Clearly,
A E P(X) U P(Y) :=:} A S; X or A S; Y ==> A S; X U Y :=:} A E P(X U Y).

If X and Y are two nonempty disjoint sets, then X U Y ¢. P(X) U P(y), and so
equality is seldom valid.
2. COUNTABLE AND UNCOUNTABLE SETS
Problem 2.1. Show that the set of all rational numbers is countable.
Solution. Let Q be the set of rational numbers and let Q+ = {r E Q: r > OJ.
Then the function f: 1N x 1N -+ Q+ defined by f(m, n) = r;;- is onto. The
conclusion now follows from Theorems 2.7 and 2.5.
Problem 2.2. Show that the set ofall fmite subsets ofa countable set is countable.
Solution. We can assume that A = {PI, P2, ... } is the set of all prime numbers.
Let :F denote the collection of all finite subsets of A. Define f::F -+ 1N by
f(F) = the product of the elements of F, for each F E :F. Then f is one-to-one,
and the conclusion follows from Theorem 2.5.
Problem 2.3. Show that a union of an at-most countable collection of sets, each
of which isfinite, is an at-most (:ountable set.
Solution. This follows immediately from Theorem 2.6.
Problem 2.4. Let A be an uncountable set and let B be a countable subset of

A. Show that A is equivalent to A \ B.
Solution. Let B = {b I , b 2 , ••• }. Since A is uncountable, the set A \ B is also

uncountable. Let C = {CI' C2, •.• } be a countable subset of A \ B. Now define
f: A \ B -+ A by
X, if x ¢. C;
f(x) =( cn+I, if x = C2n+l (n = 0, 1,2, ... );
bn , if x = C2n (n = 1,2, ... ) .
Then f is one-to-one and onto, proving that A ~ A \ B.
Problem 2.5. Assume that f: A -+ B is a sUljective (onto) function between

two sets. Establish the following:
a. card B ::: card A.
b. If A is countable, then B is at-most countable.
Section 2: COUNTABLE AND UNCOUNTABLE SETS 7
Solution. (a) Consider the family {f-I(b): bE B). Clearly, this is a family of
disjoint subsets of A. By the Axiom of Choice there exists a subset C of A such
that C n f- I (b) consists precisely of one element of A for each b E B. The
conclusion now follows by observing that f: C -+ B is one-to-one and onto.
(b) This follows immediately from part (a).
Problem 2.6. Show that two nonempty sets A and B are equivalent if and only
if there exists a function from A onto Band afunctionfrom B onto-A.
Solution. If A and B are equivalent, then there exists a function f: A -+ B

which is one-to-one and onto. Clearly, f- I : B -+ A is a surjective function.
For the converse, assume that there exists a function from A onto B and a
function from B onto A. A glance at the preceding problem guarantees that
card B ~ card A and card A ~ card B. Now, use the SchrOder-Bernstein theorem
to conclude that A and B are equivalent sets.
Problem 2.7. Show that if a finite set X has n elements, then its power set
P(X) has 211 elements.
Solution. We shall use induction on n. Assume that (I, ... , n) has 211 subsets.
Then the subsets of the set (I, ... , n, n + 1) consist of:
a. The subsets of (1, ... , n), which are 211 altogether; and
b. The subsets of the form AU (ll + 1), where A is a subset of {I, ... , nl,
again 211 altogether.
Thus, the number of subsets of (I, ... , n, II + 1) is 2" + 211 = 211 +1.
A direct proof goes as follows. Notice that the number of subsets of (I, 2, ... , h)
having k elements (where 0 ~ k ~ n) is precisely G). So, the total number of
subsets of {I, 2, ... , n) is
G) + G) + G) + ... + C) = (1 + 1)11 = 2n ,
where the last equality holds true by virtue of the binomial theorem.
Problem 2.8. Show that the set of all sequences with values 0 or 1 is uncount-
able.
Solution. For each subset A of IN define the sequence f(A) = {XII} by Xn = 1

if n E A and Xn = 0 if n fj; A. Then f defines a function from P(lN) onto
8 Chapter 1: FUNDAMENTAlS OF REAL ANALYSIS
the sequences with values 0 and 1. Since f is clearly one-to-one and onto, the
conclusion follows from Theorem 2.8.
Problem 2.9. If 2 = {O, I}, then show that 2 x ::::: P(X) for every set x.
Solution. Define f: P(X) ~ 2 x by A t-+ fA, where fA(X) = 1 if x E A
and fA(X) = 0 if x ¢. A. Note that f is one-to-one and onto. Therefore,
2x ::::: P(X).
Problem 2.10. Any complex number that is a root of a (nonzero) polynomial

with integer coeffiCients is called an algebraic number. Show that the set of all
algebraic numbers is countable.
Solution. Let Z = {... , -2, -1,0,1,2, ... }. Fix n :::: 1. Since every polyno-
mial p(x) = ao+alx+· . ·+anx n is determined uniquely by (ao, ai, ... , an), it is
easy to see that the nonzero polynomials of degree ::::;n with integer coefficients are
in one-to-one correspondence with the countable set zn+I\{(o, 0, ... , O)}. Let
{PI, P2, ... } be an enumeration of all these polynomials. By the Fundamental
Theorem of Algebra, the set Ak = {x E C: Pk(X) = O} is a finite set. Thus, the
set of all zeros of the polynomials {PI, P2, ... } of degree ::::; n is precisely the set
Rn = U%:I Ako which (by Theorem 2.6) is a countable set. Now, note that the set
of all algebraic numbers is U:I RI/' which-as a countable union of countable
sets-is itself countable; see Theorem 2.6.
Problem 2.11. For an arbitrary function f: 1R ~ 1R show that the set
A = {a E 1R: lim f(x) exists and lim f(x) =1= f(a)}

x-..Q x~a
is at-most countable.
Solution. Let I denote the set of all open subintervals of 1R with rational
endpoints and note that I is a countable set. Also, let Q denote the countable set
of all rational numbers of 1R.
For each rational real number r, let
AI" = {a E A: Either f(a) < r < lim f(x) or lim f(x) < r < f(a)}.
x-+a x-+a
Clearly, A = UreQA r holds. Thus, in order to establish that A is at most

countable, it suffices to show that each Ar is at-most countable.
So, fix some r E Q and a E A; and assume (without loss of generality) that
f(a) < r < lim x _ a f(x). Then there exists as 8> 0 such that a - 8 < y < a + 8
and y =1= a imply fey) > r. Next, pick an open interval Ia with rational
endpoints (i.e., Ia E I) such that a E Ia and Ia ~ (a - 8, a + 8). Since f (y) > r
Section 2: COUNTABLE AND UNCOUNTABLE SETS 9
holds for each y E Ia with y i= a, we see that y ¢. Ar for each y E Ia \{a}. In

particular, note that Ar n Ia = {a}.
Thus, we have established a mapping a 1---+ Ia from Ar into I (which in view
of Ar n Ia = {a} for each a EAr) is also one-to-one. This implies that Ar is
at-most countable, and hence, A is likewise at-most countable.
Problem 2.12. Show that the set of real numbers is uncountable by proving the
following:
a) (0, 1) ~ JR.; and
b) (0, 1) is uncountable.
Solution. (a) The function f: (0, 1) -+ JR. defined by the formula f(x) =
tan(1l'x - I) is one-to-one and onto.
(b) If. (0,1) is countable, then let {Xl, X2, ••• } be one enumeration of (0,1).
For each n write x" = 0.d,,\d,,2··· in its decimal expansion, where each dij
is 0,1, ... , or 9. Now, consider the real number y of (0,1) whose decimal
expansion y = 0.Y\Y2··· satisfies y" = 1 if d"" i= 1 and y" = 2 if d"" = 1. An
easy argument now shows (how?) that y i= x" for each n, which is a contradiction.
Hence, the interval (0, 1) is an uncountable set.
Problem 2.13. Using mathematical induction prove the following:

a. Ifa ~ -1,then(l+a)" ~ l+naforn = 1,2, ... (Bernoulli'sinequality).
b. If 0< a < 1, then 1 + 3"a > (1 + a)" for n = 1,2, ....
c. cos(n1l') = (-I)" for n = 1,2, ....
Solution. (a) Let a ~ -1. For n = 1 the inequality is trivially true; in fact, it is
an equality. For the induction step, assume that (1 + a)" ~ 1 + na holds true for
°
some n. Since 1 + a > is assumed to be true, it follows that
(1 + a)"+\ = (1+ a)(1 + a)" ~ (1 + a)(1 + na) = 1 + na + a + na2

= 1 + (n + l)a + na 2 ~ 1 + (n + l)a,
which is the desired inequality when n takes the value n + 1. This completes the
induction.
°
(b) Assume < a < 1. Since 1 + 3a > 1 + a, the desired inequality is true
for n = 1. For the inductive step assume 1 + 3" a > (1 + a)". Then, taking into
°
account that < a < 1, we see that
(1 + a)"+\ = (1+ a)(l + a)" < (1 + a)(1 + 3"a)

= 1 + 3"a + a + 3"a 2 = 1 + (3" + 3"a + l)a
< 1 + (3" + 3" + 3")a = 1 + 3· 3"a = 1 + 3"+\a,
which is the desired inequality valid when n is replaced by n + 1. By the Principle

of Mathematical Induction, the inequality is true for every natural number n.
(c) For n = 1, we have cos(1 .1l') = COS1l' = -1 = (_1)1. Now, assume
that cos(n1l') = (_1)n. Then, using the trigonometric formula cos(x + y) =
cos x cos y - sin x sin y, we see that
cos[(n + 1)1l'] = cos(n1l' + 1l') = cos(n1l') cos 1l' - sin(n1l')sin1l'

= (-1)n(-I) - sin(n1l')· 0 = (_1)n+1,
and the induction is complete.
Problem 2.14. Show that the Well-Ordering Principle implies the Principle of
Mathematical Induction.
Solution. Let S £; IN satisfy

a. 1 E S, and
b. n + 1 E§ whenever n E S.
We must show that S = IN, or equivalently that IN \ S = 0.

To this end, assume by way of contradiction that we have IN \ S '1= 0. Then,
by the Well Ordering Principle, n = min(lN \ S) exists. Clearly, 1 < n E IN \ S.
Thus, n - 1 E S, and consequently n = (n - 1) + 1 E S, a contradiction.
Therefore, IN \ S = 0 or S = IN.
Problem 2.15. Show that the Principle of Mathematical Induction implies the
Well-Ordering Principle.
Solution. Assume that the Principle of Mathematical Induction is true. Consider

the subset S of IN consisting of all natural numbers n with the property: whenever
a nonempty subset A of IN contains a natural number m ::: n, then A has a
least element. To establish the Well-Ordering Principle, we need to show that
S IN.=
To this end, note that 1 E S. Now assume that n E S. Also, assume that a
nonempty subset A of IN contains some natural number m ::: n + 1. If A contains
a natural number k < n + 1, then A also contains a natural number (namely k
itself) less than or equal to n, and so, in view of n E S, A must have a least element.
On the other hand, if A does not CO!1tain any natural number strictly less that n + 1,
it follows that n + 1 E A, in which case n + 1 is the least element of A. Therefore,
n + 1 E S, and so by the validity of the Principle of Mathematical Induction, we
infer that S = IN.
Section 3: THE REAL NUMBERS 11
3. THE REAL NUMBERS
Problem 3.1. If a V b = max{a, b} and a 1\ b = minta, b}, then show that

a vb = !(a + b + la - bl) and a 1\ b = 4(a + b - la - bl).
Solution. Since all expressions do not change their values if we interchange a

and b, we can assume a 2: b. Thus,
!(a + b + la - bi) = !(a +b +a - b) = a = a vb,
and
4(a +b - la - bl) = 4£a +b - (a - b)] = b = a 1\ b.
Problem 3.2. Show that Iial - Ibll ::: la + bl ::: lal + Ibl for all a, b E JR.
Solution. From -Ial ::: a ::: lal and -Ibl ::: b ::: Ibl, it follows that
-(Ial + Ibl) ::: a + b ::: lal + Ibl·

So, la + bl ::: lal + Ibl·
Substituting a - b in the place of a, we get lal :::: la - bl + Ibl so that
lal-Ibl ::: la - bl. Interchanging a and b yields -(Ial -Ibl) ::: la - bl, and so
Iial-Ibil ::: la - bl also holds.
Problem 3.3. Show that the real numbers J2 and .,fi + .J3 are irrational num-
bers.
Solution. Assume by way of contradiction that.,fi = ~ with m, n E .IN. We can

suppose that m and n have no common positive divisors other than 1. Squaring,
we get m 2 = 2n 2. This implies that m is even, i.e., m = 2k for some k E .IN
(otherwise m = 2k + 1 implies that m 2 is odd, a contradiction). It follows that
4k2 = 2n2, or n 2 = 2k 2, which in tum implies that 1l is even, i.e., n = 2e for some
e E .IN. But then, m and n have the common factor 2, which is a contradiction.
Hence, .,fi is not a rational number. (This simple proof is due to Eudoxus.)
With a different and more elegant proof one can establish the following general
result:
• The square root Jk of a natural number k is a rational number if and only

if k is a complete square, i.e., k = p2 for some p E .IN.
12 Chapter 1: F1JNDAMENTALS OF REAL ANALYSIS
If k = p2, then clearly -Ik = pEN. On the other hand, if -Ik is a rational
number, then -Ik is a rational root of the polynomial p(x) = x 2 - k. But the
positive rational roots of this polynomial are of the form ;-, where mEN is a
=
divisor of k and n EN is a divisor of 1. Thus, -Ik mEN, and so k m 2 . =
To see that ,Ji +,J3 is not a rational number, assume by way of contradiction
that ,Ji +,J3 = r > 0 is a rational number. Then,J3 = r - ,.fi and by squaring,
we get 3 = r2 - 2r,Ji + 2. This implies,Ji = r~;I, a rational number, contrary
to our previous conclusion. Hence, ,J2 + ,J3 is an irrational number.
Problem 3.4. Show that between any two distinct real numbers there is an irra-
tional numbel:
Solution. Let a < b. Choose a rational numberr with a < r < b, and then
select some n so that 0 < .;; < b-r. Note that the irrational number x = r+';;
satisfies a < x < b.
Alternatively: Note that the open interval (a, b) is· uncountable, while the set
of all rational numbers is countable.
Problem 3.5. This pj'oblem will introduce (by steps) the familiar process of sub-
traction in the framework of the axiomatic foundation of real numbers.
a. Show that the element 0 is uniquely determined, i.e., show that ifx + 0* = x
for all x E R and some 0* E R, then 0* = O.
b. Show that the cancellation law of addition is valid, i.e., show that x +a =
x + b implies a = b.
c. Use the cancellation law of addition to show that O· a = Ofor all a E R
d. Show that for each real number a the real number -a is the unique real
number that satisfies the equation a + x = O. (The real number -a is
called the negative of a.)
e. Show that for any two given real numbers a and b, the equation a + x = b
has a unique solution, namely x = b + (-a). The subtraction operation
- ofR is now defined by a - b = a + (-b),. the real number a - b is also
called the difference of b from a.
f. Foranyrealnumbersaandbshowthat-(-a) = aand-(a+b) = -a-b.
=
Solution. (a) Assume than another element 0* E R satisfies 0* + x x +0* x =
for all x E R Letting x = 0, we get 0* + 0 = O. Now, recalling that 0 + Y =
Y + 0 = y also holds for all y E R, letting y = 0* yields 0* = 0* + 0 = O.
(b) Let x + a = x + b. By Axiom 5 there exists some z E R such that z + x =
x +z = O. So,
a = O+a = (z+x)+a = z+(x+a) = z+(x+b) = (z+x)+b = O+b = b.

Section 3: TIlE REAL NUMBERS 13
(c) Clearly,
o. a + 0 = 0 . a = (0 + 0) . a = 0 . a + 0 . a,
and so by the cancellation law of addition, O· a = 0 for each a E JR.
(d) Assume a +x = = =
O. Since a + (-a) 0, we see that a +x a + (-a), and
so, by the cancellation law we have established in (b) above, x = -a; the negative
of a.
(e) If a + z = a + y = b, then by the cancellation law, we get z = y. Thus, given
a and b, the equation a + x = b has at-most one solution x E JR. Since
a+[b+(-a)] = (a+b)+(-a) = (-a)+(a+b) = [(-a)+a)]+b = O+b = b,_
we see that the only solution of the equation a + x = b is x = b + (-a). We

denot~ this number by b - a and call it the subtraction of a from b.
(f) A close look at the equation a + (-a) = (-a) + a = 0 guarantees immediately
that -(-a) = a. Moreover, from
a+b+(-a-b) = a+b+[ -a+(-b)] = [(a+b)+(-a)]+(-b) = b+(-b) = 0,

we easily infer that -(a + b) = -a - b.
Problem 3.6. This problem introduces (by steps) the familiar process of division
in the framework of the axiomatic foundation of real numbers.
a. Show that the element 1 is uniquely determined, i.e., show that if 1* . x = x
for all x E JR and some 1* E JR, then 1* = 1.
b. Show that the cancellation law of multiplication is valid, i.e., show that
x . a = x . b with x =f:. 0 implies a = b.
c. Show that for each real number a =f:. 0 the real number a-I is the unique
real nun:zber that satisfies the equation x· a = 1. The real number x = a-I
is called the inverse (or the reciprocal) of a.
d. Show that for any two given real numbers a and b with a =f:. 0, the equation
ax = b has a unique solution, namely x = a-lb. The division operation
-+- (or I) of JR is now defined by b -+- a = a-I b; as usual, the real number
b -+- a is also denoted by bla or~.
e. For any two nonzero a, bE JRshowthat(a-I)-1 = a and (ab)-I = a-Ib- I.
f. Show that T = a for each a, ~ = 0 for each b =f:. 0, and ~ = 1 for each
a =f:. O.
Solution. (a) Assume that some real number 1* satisfies 1* . x = x . 1* = x for

each x E JR.. In particular, letting x = 1, we get 1* . 1 = 1. Since y . 1 = Y for
all y E JR, letting y = 1* yields 1* = 1* . 1 = 1. So, 1 is the only real number r
which satisfies r . x = x for each x E JR.
(b) Assume x . a = x . b with x i= O. By Axiom 7 there exists a real number

y E JR such that y . x = 1. Now, observe that
a = 1 . a = (yx)a = y(xa) = y(xb) = (yx)b = 1 . b = b.

(c) If ax = ay = 1 with a i= 0, then by (b), we must have x = y. This shows
that the reciprocal a-I of a is uniquely determined.
(d) To see that the equation ax = b with a i= 0 has at-most one solution x,
notice that if ax = ay = b, then by the cancellation law of multiplication, we
have x = y. Moreover, notice that
The above show that the equation ax =

b with a i= 0 has the unique solution
x = a-lb.
(e) If a i= 0, then the equation a . a-I = 1 readily says that (a-I)-I = a. In
addition, from
(ab)· (b-Ia- I ) = a(b· b-I)a- I = a· 1· a-I = 1,

we easily obtain (ab)-I = b~la-l.
(t) Since 1 . a = a, we obtain if = a for each a E 1R. The equation b . 0 =0
also implies that ~ = 0 for each b i= O. From a . 1 = a, we get immediately ~ =1
for all a i= O.
Problem 3.7. Establish the following familiar properties of real numbers using
the axioms of the real numbers together with the properties established in the
previous two problems.
i. The zero product rule: ab = 0 if and only if either a = 0 or b = O.
ii. The multiplication rule of signs: (-a)b = a( -b) = -Cab) and (-a)
(-b) = ab for all a, bE JR.
iii. The multiplication rule for fractions: For b, d i= 0 and arbitrary real
numbers a, c we have
a c ac
_.-=-
b d bd
In particular, if~ i= 0, then (~)-I =~.

iv. The cancellation law of division: If a i= 0 and x i= 0, then !;= ~ for
each real number b.
v. The division rule for fractions: Division by a fraction is the same as
Section 3: TIlE REAL NUMBERS 15
multiplication by the reciprocal of the fraction, i.e., whenever the fraction

J
~ -=- is defined, we have
a cad ad
b -=- d = b . -;; = be
°. °
Solution. (i) We already know from the previous problem that b = for each
b E lR. On the other hand, if ab = 0(= a . 0) and a i= 0, the cancellation law of
mUltiplication shows that b = 0.
Oi) Clearly,
ab+(-a)b = [a+(-a)]b=O.b=O and ab+a(-b) = a[b+(-b)] =a·O=O,

and so -Cab) = (-a)b = a(-b). This implies
(-a)(-b) = -[a(-b)] = -[-Cab)] = abo
(iii) If b, d i= 0, then
an d th IS h'
· SOWS th at ac c s·mce Iia . a
bd = Iia . d' ab = 1,we see th at (a)-I
b = ab Ii = a'
b
(iv) If c = ~, then ac = b and so (ax)c = bx for each x i= 0, which shows that

c=£=bx.
a ax
(v) Notice that the identity
c ad adc a
d' be = dbc = b
guarantees ~ -=- J = ~~ = ~ . ~.
Problem 3.8. This problem establishes that there exists essentially one set of
real numbers that satisfies the eleven axioms stated in Section 3. To see this, let lR
be a set of real numbers (i.e., a collection of objects that satisfies all eleven axioms
stated in Section 3 of the text).
a. Show that 1 > 0.
b. A real number a satisfies a = -a if and only if a = 0,
C. If n = 1 + 1 + ... + 1 (where the sum has Un summands" all equal to I),
,then show that these elements are all distinct; as usual, we shall call the
collection IN of all these numbers the natural numbers oflR.
d. Let Z consist oflN together with its negative elements and zero; we shall
call Z, of course, the set of integers oflR. Show that Z consists of distinct
elements and that it is closed under addition and multiplication.
e. Define the set Q of rational numbers by Q = {~: m, n E Z and n =1= OJ.

Show that Q satisfies itself axioms 1 through 10 and that
a = sup{r E Q: r :::: a} = inf{s E Q: a :::: s}
holds for each a E ]R..
f. Now, let]R.' be another set of real numbers and let Q' denote its rational
numbers. If I' denotes the ullit element of]R.', then we write
Il' = I' + I' + ... + I'
for the sum having un-summands" all equal to I'. Now, define thefunction
f: Q -+ Q' by
f(~) = ;;
and extend it to all of]R. via the formula
f(a) = sup{f(r): r :::: a}.

Show that ]R. and]R.' essentially coincide by establishing the following:
1. a:::: b holds in]R. ifand only if f(a):::: feb) holds in ]R.'.
11. f is one-to-one and onto.
iii. f(a+b) = f(a)+ feb) and f(ab) = f(a)f(b) for all a, bE 1R.
Solution. (a) Since 1 =1= 0, we have two possibilities: either 1 > 0 or 0 > 1. If
0> 1, then (by Axiom 9) we have 0 + (-1) > 1 + (-1) = 0 or -1 > 0, i.e.,-I
is a positive number. Now, using Axiom 10, we infer that 0 . (-1) 2: 1 . (-1), or
o 2: -:-1, contrary to -1 > O. Hence, 1 > O.
(b) Since 0 + 0 = 0, we know that -0 = O. Conversely, assume that a real
= =
number a satisfies a = -a. This implies a + a (1 + 1)a O. However, since
1 > 0, we have 1 + 1 2: 1 + 0 = 1 > 0, and so 1 + 1 =1= O. Consequently, from
the zero product rule, (1 + I)a = 0 implies a = O.
(c) As shown in part (b) above, 1 + 1 =1= 0 and in fact 1 + 1 =1= 1; otherwise
1 + 1 = 1 = 1 + 0 implies (in view of the cancellation law) 1 = 0, which is
impossible. Now, by induction, assume that
0<1<1+1<1+1+1<···<1+1+···+1.
'----v--'
II-summands
We claim that the real number n + 1 = 1 + 1 + .. , + 1 + 1 (a sum of n + 1

summands) satisfies n + 1 > n =
1 + 1 + ... + 1 (where the last sum has n
summands). Indeed, if n + 1 :::: n, then (n + 1) + (-n) :::: n + (-n) or 1 :::: 0,
which is a contradiction. Hence, Ii + 1 > n and the induction is complete.
(d) By part (c) we know that the natural numbers together with zero are all
distinct real numbers. If -m = -n with m, n E IN, then m = n, which shows that
distinct natural numbers have distinct negatives. If m = -n with m, n E IN, then
Section 3: THE REAL NUMBERS 17
=
m + n 0 contradicting (c), and so no natural number can be equal to a negative
integer. It now follows that Z consists of distinct elements. .
(e) Observe that if ;- and ~ are two rational numbers, then
m p mq+np m p mp
- + - = --=----!- and ---=-,
n q nq n q nq
and if;- =I 0, then ( ;-) - I = !!i. That is, Q is closed under addition, mUltiplication
and inverses. Since all real numbers satisfy axioms 1 through 10, it follows that
Q itself satisfies axioms 1 through 10 in its own right.
For the second part, fix a E lR and let A = {r E Q: r ::: a}. Since there exists
a rational number between a-I and a (see Theorem 3.4), A is nonempty, and
clearly A is bounded from above by a. By the Completeness Axiom (Axiom 11),
sup A exists in lR and satisfies sup A ::: a.
No)\" let E > O. By Theorem 3.4, there exists some rational number r such
that a - E < r < a. Clearly, rEA, and so a - E O. This implies a ::: sup A, and hence a = sup A. The equality,
a = inf{s E Q: a ::: s} can be proven in a similar manner.
(f) Notice that the mapping is well defined. That is, if;- = ~ in Q, then
f(;-) = f(*). Indeed, since ;- *
= is equivalent to mq = np, we see that
m'q' = n'q' or ~
II
= t..
q
Now, let us verify properties (i), (ii), and (iii).
i. a ::: b holds in lR if and only if f(a) ::: feb) holds in lR'.
Note first that two rational numbers r, sEQ satisfy r ::: s if and only if r' ::: s'.
Indeed, to see this it suffices to assume that r and s are positive rational numbers
(why?). We have
p p'
r = -mn < s=-
- q
-¢=:} mq ::: np -¢=:} m'q'::: n'q' -¢=:}
m'
r'=- < - = s'.
n' - q'
Now, let a ::: ,b. Then {r E Q: r ::: a} S;; {s E Q: s ::: b}, and from this it
easily follows that f(a) ::: feb). For the converse, assume that f(a) ::: feb). If
a ::: b is not true, then we must have b < a. But then, by Theorem 3.4 there
exist two rational numbers r, sEQ such that b < r < s < a. This implies
feb) ::: r' < s' ::: f(a), a contradiction.
ii. f is one-to-one and onto.
To see that f is onto, let a' E lR'. Then by Theorem 3.4,
a' = sup{t E Q': t ::: a'}.

If, we iet S = {r E Q: r' ::: a'}, then this set is bounded above in lR (why?) and
so a = sup S exists in lR. Moreover, notice that
{t E Q': t ::: a'} = {r': r E Q and r ::: a}.
Now, it is easy to see that f(a) = a'.

To verify that f is one-to-one assume f(a) = feb). Then by part (i) we have
a :::: b and b :::: a, i.e., a = b.
iii. f(a + b) = f(a) + feb) and f(ab) = f(a)f(b) for all a, b E JR.
We verify the additivity property only and leave the multiplicative property
for the reader. Clearly, fer + s) = fer) + f(s) holds for all rational numbers
r, s. Now, fix a, b E JR and assume r, sEQ satisfy r :::: a and s :::: b. Then
fer) = r':::: f(a) and f(s) = s':::: feb). Sincer +s E Q andr +s:::: a +b, we
=
see that fer) + f(s), r' + s' =
fer + s) :::: f(a + b). This easily implies
f(a) + feb) :::: f(a + b).
For the reverse inequality, let E' > 0 in ]R.'. Then there exist rational numbers
r, sEQ with r :::: a and s :::: b such that f(a) - E' < fer) and feb) - E' < f(s).
Sincer +s :::: a+b, it follows that f(a)+ f(b)-2E' < f(r)+ f(s) = f(r+s) ::::
f(a + b) for each E' > O. This guarantees f(a) + feb) :::: fCa + b), and therefore
f(a + b) = f(a) + feb).
Problem 3.9. Consider a two-point set R = {O, I} equipped with the following
operations:
a. Addition (+) : 0 + 0 = 0, 0 + 1 = 1 + 0 = 1 and 1 + 1 = 0,
b. Multiplication (-) : 0·1 = 1 ·0= 0 and 1·1 = 1, and
c. Ordering: 0 2: 0, 1 2: 1 and 1 2: O.
Does R with the above operations satisfy all eleven axioms defining the real num-
bers? Explain your answer.
Solution. It satisfies all axioms except Axiom 9, which states that:
• ffx 2: y and z 2: 0, then x + z 2: y + z.

To see this, assume that Axiom 9 is valid. We distinguish two cases.
CASE I: 1> O.
In this case, we must have 0 = 1 + 1 2: 0 + 1 = 1, which contradicts 1 > O.
CASE II: 0 > 1.
This implies 1 = 0 + 1 2: 1 + 1 = 0, which again contradicts 0 > 1. Thus,
Axiom 9 does not hold in this case.
It should be noticed that Axiom 9 is the one that guarantees that 1 + 1 (i.e, the
number 2) is distinct from 0 and 1; and, of course, it is the axiom that establishes
(as we saw in part (b) of Problem 3.8) the existence of the set of integers.
Problem 3.10. Consider the set of rational numbers Q equipped with the usual
operations ofaddition, multiplication, and ordering. Why doesn't Q coincide with
the set of real numbers?
Section 3: TIIE REAL NUMBERS 19
Solution. The set of rational numbers satisfies all the axioms of real numbers
except the completeness axiom. This was proven in part (e) of Problem 3.8. To see
that Q does not satisfy the completeness axiom, assume by way of contradiction
that it does. Consider the set
s = {O :::; r E Q: r2 :::; 2}.
Then S is nonempty and bounded from above in Q(why?), and so b = sup Sexists
in Q. Now, repeat the proof of Theorem 3.5 to conclude that b2 = 2, i.e., that
b = -/2. However, we proved in Problem 3.3 that -/2 is not a rational number,
and we have reached a contradiction. Hence, Q does not satisfy the completeness
axiom and it cannot coincide with the set of real numbers.
Problem 3.11. This problem establishes the familiar rules of" exponents" based
on the axiomatic foundation of real numbers. To avoid unnecessary notation,
we shall assume that all real numbers encountered here are positive--and so by
Theorem 3.5, all non-negative real numbers have unique roots. As usual, the -
"integer" powers are defined by
= a . a ... a =1 1
all '-y---"
aD , a 1 = a, and a -Il = -
all
factors
Il-
Extending this to rational numbers,for each m, n lN we define

E
m nC;;; 1 1
an = ,yam and
nI
a- n = a -- :;am.
Establish the following properties:
a. a!ff = (::{ri)m for all m, n E IN.
b• IJrF m, n, p, q E I N
satls m=Eq'. tllen a !!!n
' if)' Ii =a I!.
q •
c. Ifr and s are rational numbers, then:

i. aras = ar+s and ~ r s
as = a - ,
r
ii. (aby = arb and (%Y = ~, and
iii. (ary = ars .
Solution. It should be noticed first that (all)m = (amy' = a mn for all a E lR and
all natural numbers m, n E IN.
(a) Notice that
[(,y'a)m)" = ~,yIat" (,yIa;m'" (,y'a)m, = ~
n- factors mn- factors
n'-)n . (n'-)II
= ( va va ... ( va
n'-)II =a·a···a=a.
m
, y '-y---',
m- factors m- factors
Since the nIh_roots are unique (Theorem 3.5), we infer that (::{ri)m =:;am = a!ff.
(b) Assume m, n, p, q E IN satisfy ;- = ~, or pn = mq. Using part (a), we

see that
and this shows that a ~ = a r; .

(c) The formulas can be established easily if rand s are integers. Now, let r and
s be rational numbers. We shall assume that r and s are also positive and leave
the "negative case" for th~ reader. By part (b), we can also suppose that r = ;-
and s = ;. Since (.::fo . ::/b)n = (.::fo)n . (::/b)n = ab, we see that::tab = .::fo::/b.
Now note that
i. ara S =,::((iiii.::[(iP = ::/ama P = ::/a m+p = a~ = ar+s.
ii. (a bY = ::.f((ili)"i = ::/am bm = (,::((iiii)(:fijiii) = arbr .
iii. (ary = ;ju:jarii)p =;j::/a pm = ~apm = a7;T = ars .
We leave the remaining cases for the reader.
4. SEQUENCES OF REAL NUMBERS
Problem 4.1. Show that iflxl < I, then lim x" = O.

Solution. Let Xn = Ixl" for each n. Then Xn+l = Ixlxn holds for each n, and
the assumption Ixl < 1 implies 0 ::: Xn+l ::: Xn' By Theorem 4.3, a = limxn
=
exists. It follows that a alxl (or (1 - IxDa =
0) must hold, and from this that
a =0.
A direct way of proving that lim xn = 0 goes as follows. Observe first that we
can suppose 0 < x < 1. Now, if E > 0 is given, then note that
InE
x" < E ¢=:> In(xn) = n lnx < In E ¢=:> n > -.
lnx
Problem 4.2. Show that lim XII = x holds if and only if evelY subsequence of
{XII} has a subsequence that converges to x.
Solution. If limxn = x, then every subsequence must converge to x. So, every

subsequence of a subsequence (as being itself a subsequence of {x n }) must converge
to x.
For the converse, assume that each subsequence of {x n } has a subsequence
that converges to x. Now, suppose by way of contradiction that {x n } does not
converge to x. Then for some £ > 0 we must have Ix - Xn I 2: £ for an infinite
number of n. So, there exists a subsequence {Yn} of {xn} such that Ix - Yn I 2: £
for each n. However, the latter contradicts the fact that {Yn} has a subsequence
that converges to x. Therefore, lim Xn = x.
Section 4: SEQUENCES OF REAL NUMBERS 21
Problem 4.3. Consider two sequences {k n} and {mil} of strictly increasing nat-
ural numbers such that for some e E IN we have
Ie, e + 1, e + 2, ... } £; {kl' k2, ... j U {ml, m2, .. .}.
Show that a sequence of real numbers {XII} converges in R if and only if both
subsequences {Xk n} and {xm.! of {xn} converge in R and they satisfy lim Xkn =
lim x m" (in which case the common limit is also the limit of the sequence).
In particulQ/; show that a sequence of real numbers {XII} converges in R if and
only if the "even" and "odd" subsequences {X2n} and {x211-d both converge in R
and they satisfy limx211 = limx2n-l.
Solution. If Xn ~ x, then clearly Xkn ~ x and x m" ~ x. For the converse,

assume that Xkn ~ x and xmn ~ x. Let E > O. Choose some no E IN such that
IXkn - xl < E and IXm" - xl < E for all n ::: no.
Put eo = maxIe, kilO' milo}' and we claim that
IXII - xl < E for all n ::: eo.
To see this, let 12 ::: eo. Then the assumption
guarantees the existence of some rEIN such that kr = n or mr = n. Since r < no

implies kr < kno :5 eo and mr < milo :5 eo, we see that r ::: no. Hence, either
Xn = Xkr of Xn = xmr (with r ::: no), and so from (*) it follows that IXII - xl < E.
This shows that XII ~ x.
The last part should be immediate from the above conclusion.
Problem 4.4. Find the lim sup and lim inf for the sequence {(-1)").
Solution. We have lim inf( _1)n = -1 and lim sup( _1)n = 1.

Problem 4.5. Find the lim sup and lim inf of the sequence {xn} defined by
XI = ~, X2n = ~X2n-I' and X2n+1 = ~ +X2n for n = 1,2, ....

Solution. We claim that
n-I II
I '\' I _I,\,I
X2n = 32 ~ 3k and X 2n+l -"3 ~ 3k
k=O k=O
hold for n = 1, 2, . ... The validity of the identities can be established by

induction. We shall establish the validity of the second identity and leave the
verification of the first to the reader. For n = 1, we have

I
x 3 -x
- 2·1+1 -- "3I + X 2 -- "3I + "3Ix I --"3I + "9I -- "31(1 + "3I) --"3I '~
\ ' y:.
I
k=O
Now, assume that X2n+1 = ~ I:Z=o:fr holds for some n. Then,
n n+1
I _I I _Il'\'l 1,\,1
X2(n+l)+1 = "3 + X2(n+I) - "3 + "3 X2n + 1 - "3 + "9 ~ y: = "3 ~ y:,
k=O k=O
and the induction is complete. Consequently,

co
11'm x 2n+1 -"3
- I '\' I -
~
I
y: - 2'
n-+co k=O
Now, we claim that andi 4

are the only limit points of {xn}' To see this, let
a be a real number different from and i 4.
Pick some e > 0 such that
(a - e, a + e) n (i - e, i + e) = 0 and (4 - e, 4+ e) = 0.
(a - e, a + e) n
Next, note that there exists some k such that X2n E (i - e, i + e) and X2n+1 E
(4 - 4e, + e) hold for all n ::: k. Therefore, IX n - al ::: e holds for all n ::: k,
and this shows that a cannot be a limit point of the sequence {xn}. Consequently,
· . f Xn = 6I
1ImlD and ·
1Imsupxn = 2'I
Problem 4.6. Let {xn} be a bounded sequence. Show that
limsup(-xn) = -liminfxn and liminf(-xn) = -limsupxlI'
Solution. We shall use the fact that lim sup XII and lim inf Xn are the largeSt and
smallest limit points of {x n }, respectively. We shall establish the first formula.
Choose two subsequences {Yn} and {zn} of {xn} such that lim Yn = lim inf Xn
and lim( -zn) = lim sup( -xn). Then
-liminfxn = lim(-Yn)
:::: limsup(-xn) = lim(-zn) = -limz n
:::: -lim inf Xn,
and so lim sup( - xn) = - lim inf X n •

Problem 4.7. If {xn} and {YII} are two bounded sequences, then show that
a. lim sup(xn + YIl) :::: lim sup Xn + lim sup YII' and
b. lim inf(xII + Yn) 2: lim inf Xn + lim inf YII'
Moreover, show that if one of the sequences converges, then equality holds in both
(a) and (b).
Solution. (a) By passing to a subsequence, we can assume that lim(xn + Yn) =

lim sup (XII + YIl)' Since {XII} is a bounded sequence, there exists a subsequence
{Xk,,} that converges. Let X = lim Xk n • By the same reasoning, there exists a
subsequence of {Yk n } that converges to some y. Thus, there exists a strictly
increasing sequence {mil} of natural numbers such that X = limxm" and Y =
lim Ym". Hence,
lim SUP(XII + YII) = x + Y = lim xm" + lim Ym" :::: lim sup XII + lim sup YII'
Finally, if X = lim XII holds, then pick a subsequence {Yk n } of {YII} such that
lim Yk" = lim sup YII, and note that
lim sup XII + lim sup YII = X + lim Yk"

= lim(xk + Yk,,) :::: lim sup(xll + YII)'
n
(b) It follows from (a) by using the preceding problem.
Problem 4.8. Prove that the lim sup and lim inf processes "preserve inequali-
ties:' That is, show that if two bounded sequences {XII} and {Yn} of real numbers
satisfy XII :::: Yn for all n 2: no, then
lim inf Xn :::: lim inf Yll and lim sup XII :::: lim sup Yn.
Solution. First, we shall show that if two sequences of real numbers {sn} and {til}
converge in lR (say Sn --+ s and tn --+ t) and Sll :::: til for each n 2: no, then s :::: t.
Indeed if, s > t is true, then let E = s"21 > 0 and note that for all n sufficiently
large, we must have
sn E (s - E, S + E) = et, 3s
2-1) and til E (t - E, t + E) = (312S, Sil).
That is, tn < S~I < Sll must hold for all n sufficiently large, which is impossible.
Hence, s :::: t. -
iSTANBUL BiLGi
UNIVERSITY LIBRARY
Now, assume that two bounded sequences of real numbers {xn} and {Yn} satisfy
Xn :s Yn for alln ::: no. Put
Sn = inf Xk
k~n
and tn = inf Yk.
k~n
If n ::: no, then notice that for each r ::: n we have Sn = infk~no Xk :s Xr :s Yr,
and so Sn :s infr~1I Yr = tn for each n ::: no. By the discussion of the first part, we
infer that
liminfx n = limsn :s limtn = liminfYn.

The lim sup case can be established in a similar manner, or by using the formula
lim supxn = -lim inf( -xn)'
Problem 4.9. Show that lim..::lii = 1 (and conclude from this that lim:fO. = 1
for each a > 0).
Solution. Note that ..::Iii

= (·:fl;n 2
• An easy inductive argument shows that
M> 1 holds for each n. Thus, we can write M

= 1 + Xn with XII O. >
Since (1 + a)n ::: 1 + na holds for each n and each a ::: 0 (see Problem 2.13),
we get
and so 0 < Xn :s In - ~. This implies limxn = O. Therefore,
An alternate proof goes as follows: By L'Hopital 's Rule, we have limx -+ oo = l:x
0, and so limn-+ oo l~n = O. Therefore, using that the exponential function is
continuous, we infer that
lim ..::Iii = lim el~n = eO = 1.

n-+oo n-+oo
For the parenthetical part, assume first a > 1. Then it is easy to see that 1 :s
:fO. :s ..::Iii holds true for all n > a. Consequently, by the "Sandwich Theorem,"
we see that lim:fO. = 1. If 0 < a < }, then 1 > 1, and so lim nil. = lim n~ = 1,
a Vii "a
from which it follows that lim :fO. = 1 holds true in this case, too.
Problem 4.10. If {xn} is a sequence of strictly positive real numbers, then show
that
1Imln Xn+1
· . f --::: . f '::;Xn:::
l'ImlD "Iv"" l'Imsup,::;xn:::
"Iv"" Xn+1
l'Imsup--.
n~oo...tll n-i-OO 11-+>00 n~oo Xn
Conclude from this that if lim:!!ttl

XII
"IX;; also exists and
exists in JR, then lim vA'n
"IX;; = lim
lim "'''''It .t",-!-I •
.t n
Solution. Let {XII} be a sequence of real numbers such that Xn > 0 holds for
each n. We shall establish lim sup ~ ::: lim sup:!!ttL
XII
and leave the similar proof
of the other inequality for the reader. Put
. xn+1 Xk+1
= /\ V-,
00 00
X = lim sup -
XII 11=1 k=n Xk
and note that if X = 00, then there is nothing to prove. So, we can assume x < 00.
Let s > 0 be fixed. Then there exists some k such that X,,+I < X + s holds
x"
for all n 2: k. Now, for n > k we have
where c = Xk(X + S)-k is a constant. Therefore, ~ ::: (x + s):jC holds for

each n 2: k and so, in view of lim:jC = 1 (see Problem 4.9) and Problem 4.8,
we infer that
lim sup -rx; ::: lim sup(x + s).::[C =

IJ-+OO
(x + s) lim .::[C = x
1I~00
+ S.
Since s > 0 is arbitrary, we infer that lim sup ~ ::: x = lim sup X,~:I •
Problem 4.11. The sequence of averages of a sequence of real numbers {xn} is
the sequence {an} defined by all = XI+X2;"+X". If {xn} is a bounded sequence of
real numbers, then show that
lim inf Xn ::: lim inf all ::: lim sup an ::: lim sup Xn.
In particular, if Xn ---+ x, then show that an ---+ x. Does the convergence of {an}
imply the convergence of {xn}?
Solution. The solution will be based upon the following properties of lim sup
and liminf:
• If {un} is a bounded sequence of real numbers, then for each E > 0 the
inequalities
Uk 2: lim sup un +E and Urn:::: lim inf Un - E
holdforfinitely many k and finitely many m.
To see this, assume by way of contradiction that Uk 2: lim sup Un + E holds true
for infinitely many k. Then there exists a subsequence {v n} of {un} satisfying
Vn 2: lim sup Un + E for each n. Since {v n} is a bounded sequence, there exists
a subsequence {wn} of {vn} (and hence of {un}) satisfying Wn -+ W E 1R. By
Problem 4.8, we know that W 2: lim sup Un + E, i.e., W is a limit point of {un}
which is greater than the largest limit point (lim sup un) of {un}, a contradiction.
Now, let {xn} be a bounded sequence of real numbers and fix E > O. Put
l = limsupxn and let K = {k E 1N: Xk 2: l + E}. By the above discussion, K is
a finite set. Put
Sn = {i E 1N: i E K and i :::: n} and Tn = {i E 1N: i rf. K and i :::: n},
and define the sequences {sn} and {tn} by
sn =: LXi and tn = LXi.

ieSII ieT"
Clearly, {sn} is an eventually constant sequence, tn :::: n(l + E) holds for each n
and an = ~ + ~. Since snln -+ 0 and tnln :::: l + E for each n, it follows from
Problems 4.7 and 4.8 that
lim supan = lim sup(~n + !u.)

n
= lim ~n + lim sup!u.n = lim sup!u.n <- l + E.
Since E > 0 is arbitrary, we get lim sup an :::: l = lim sup X n • Similarly, lim inf
Xn :::: liminfan. If Xn -+ x, then x = liminfxn = limsupxn, and so x =
liminfa n = lim supan. This implies an -+ x.
The convergence of the sequence {an} of averages does not imply the conver-
gence of {x n}. For instance, if Xn = (_1)n, then an -+ 0 while {xn} fails to
converge.
Problem 4.12. For a sequence of real numbers {xn} establish the following:
a. If Xn+l - Xn -+ x in JR, then xnln -+ x.
b. If {XII} is bounded and 2xII :s X II +l + Xn -l holds for all n = 2, 3, ... , then

X II +l - Xn to.
Solution. (a) Assume that Xn+l - XII -+ X in JR. Notice that I:7=; (Xi+l - Xi) =
Xn+l - Xl for each n. By Problem 4.11, we nave
n
* L(Xi+l -Xi)
i=l
= HXn+l -Xl) -+ X.
Since Xl / n -+ 0, it follows that Xn+l / n -+ x. Now note that
XII XII n-l
- = - - . - - --+ x·l=x.
n n- 1 n
(b) The condition 2xn :s Xn+l + Xn-l can be rewritten as Xn - Xn-l :s Xn+l - XII
for each n = 2, 3, ... , which implies that the bounded sequence {X II +l - xn} is an
increasing sequence, and hence convergent. Let X II +l - Xn t X in JR. By part (a),
we have Xn / n -+ x. But, since {xn} is a bounded sequence, Xn / n -+ O. Therefore,
=
x 0, and so X II +l - Xn t O.
Problem 4.13. Consider the sequence {xn} defined by 0 < Xl < 1 and x lI + 1 =
1 - .Jl - Xn for n = 1, 2, .... Show that XII .j.. O. Also, show that Xn+l
Xn
-+ 21.
Solution. We claim that

o< X II +l < XII < 1
holds for each n = 1, 2, . '" To verify this claim, we use induction. Since
0< Xl < 1, we have 0 < I-Xl < l,andsoO < I-Xl <.JI=XI < 1. Hence,
=
o < 1 - .JI=XI X2 < Xl < 1. That is, (*) is true for n 1. =
For the inductive argument, assume that (*) is true for some n. This implies
o < 1 - Xn < 1 - XII +l < 1, and so 0 < ~ < ,)1 - Xn+l < 1, from which
it follows that
0< Xn+2 =1- )1 - Xn+l < 1 -)1 - X;1 = Xn+l < 1,
which shows that (*) is true for n + 1. This completes the induction and guarantees
that (*) is true for each n.
Now, since {XII} is decreasing and bounded from below, it converges, say to
X E JR. Clearly, 0 :s X < 1. Moreover, we have
X = n-+co
lim Xn+l = lim (1- )1-x = 1-.Jf=X.
n-+oo
lI )
In other words, X is the non-negative solution of the equation X = 1 - .Jf="X.

Solving the equation yields x = 0 or X = 1. Hence, X = 0, and so XII .j.. O.
For the last part, notice that
Xn+1 1 - .JI=X;;" 1 1
-- = = ---=== ~ -
Xn XII 1 + -v'1-xlI 2'
and the solution is complete.
Problem 4.14. Show thqt the sequence {xn} defined by
is a convergent sequence.
Solution. From the binomial expansion:

n n
Xn = (1 + ~r = L G);!r = 1 + L GH-
;=0 ;=1
n
= 1 + ""
~
n(n-I) ,,:(n-;+I) .1,.
I! n'
;=1
n
= 1+ L fi(1 - ~)(1 - ~) ... (1 - ;~I)
;=1
11+1
< 1 + ""'+(1
L...., "
- _I
n+1
)(1 - .L)
n+1
... (1 _ cl)
n+1
;=1
= ( 1 + n+1I )n+1 = Xn+l·

Thus, Xn t holds. Also, note that for n ::: 2 we have
n n n
Xn = 1 + L fi(1-~) ... (1- ;~I) ~ 2+ L h~ 2+ L ~ ~ 3.
;=1 ;=2 ;=2
By Theorem 4.3, {xn} converges. (Of course, limxn = e = 2.718···.)
Problem 4.15. Assume that a sequence {xn} satisfies
IXn+1 - Xn I ~ alxn - xn-II
for n = 2,3, ... and some fixed 0 < a < 1. Show that {xn} is a convergent
sequence.
Solution. Let c = IX2 - xII. An easy inductive argument shows that for each n
we have IXII+I - xIII::: ca"-I. Thus,
P P
,x +
lI p - 1 "I
XII _ 1:< "
< L...; XII+I - XII+I-l :.... C L...; a 11+1-2 <
_ 1_,,0'.
C II-I
1=1 1=1
holds for all n and all p. Since lim a" = 0, it follows that {XII} is a Cauchy
sequence, and hence, a convergent sequence.
Problem 4.16. Show that the sequence {XII}, defined by

1
XI = 1 and Xn+1 = -3 +
- f o r n = 1,2, ... ,
XII
converges and determine its limit.
Solution. Clearly, XII > 0 holds for each n. Now, note that
.
IXII+I . I = I 3+xn
- XII I - 3+x,,_1
I 1 !.t,,-xn-d
= (3+x,,)(3+x < 9"I IXII - XII_I I
n_l) -
holds for n =2,3, .... By Problem 4. 15, the sequence {xn} converges. IflimxlI =X,
then X 2: 0 and
. 1 1
x = hrnx l1 +1 = . = --
3 +hmxn 3 +X
Solving the equation, we get X = -3-;JI3.
Problem 4.17. Consider the sequence {XII} of real numbers defined by XI = 1

and Xn+1 = 1 + 11xn for n = 1,2, .... Show that {x,d is a convergent sequence
and that lim XII = ../2.
Solution. An easy inductive argument shows that XII > 0 for each n. This implies
that, in fact, we have 1 ::: Xn ::: 2 for each n. Now, note that
IXn+1 -xnl = 1_1_ _ 1 1

1 +xn 1 +Xn-I
IXn -xn-II
=-------
(1 + xn)(1 + Xn-I)
IXn - xII-II I
< (1 + +
1)(1 1) = 41xn -xn-II
for each n = 2,3, .... By Problem 4.15, the sequence {xn} converges. Let
Xn -+ x. Since Xn 2: 1 for each n, we see that x 2: 1. Then
x = 11lim . ( 1) 1
Xn+1 = hm 1 + - - = 1 + - - .
....
00 n .... oo 1 + Xn 1+x
That is, x is the positive solution of the equation x = 1+ I !x ' or x 2

+x = 1+x + 1.
This implies X2 = 2, and so x = .J2.
" l
Problem 4.18. Define the sequence {xn} by XI = 1 and
XII+I = !(x n + ~), n = 1,2, ....

2 Xn
Show that {XII} converges and that lim Xn = .J2.

Solution. Clearly, Xn > 0 holds for each n. (Use induction to prove this!) Also,
x2 I (x - -
-2=- 2 )2 >0
n+1 4 n x" -
holds for each n. Thus, if n 2: 2, then
2-x~ 0
Xn+1 - XII = 2"I ( XII + X2 ) - Xn = -2x :::: ,
" "
and so 0 < Xn+1 < Xn holds for each n 2: 2. By Theorem 4.3, X = limx n exists.
Since x; 2: 2 holds for each n 2: 2, we see that x > O. From the recursive
formula, it follows that 2x = x + ~, or x2 = 2. (Note also that the limit is
independent of the initial choice XI > 0.)
Problem 4.19. Define the sequence Xn = L~=I for n i = 1, 2, .... Show that
{xn} does not converge in 1R. (See also Problem 5.10.)
Solution. The inequality
X2n - Xn = II~I + 1I~2 + ... + n~n

- -L+-L+···+-L=n·-L=l
> 211 2n 2n 2n 2
shows that {xn} is not a Cauchy sequence, and hence, is not convergent in JR.
Problem 4.20. Let -00 < a < b < 00 and 0 < A < 1. Define the sequence
Section 4: SEQUENCFS OF REAL NUMBERS 31
Xn+2 = AXn + (1 - A)Xn+1 for n = 1,2, ....

Show that {xn} converges in lR and find its limit.
Solution. Rewriting Xn+2 = AXn + (1 - A)xn+1 = AX n + Xn+1 - AXn+1 in the

form X n+2 - Xn+1 = A(xn - Xn+I), we see that
holds for each n. Now, a glance at Problem 4.15 guarantees that {xn} is a convergent
sequence. However, we cannot get the limit of the sequence {xn} by taking limits
in both sides of the recursive formula Xn+2 = AXn + (1 - A)xn+ I. We shall compute
the limit of the sequence {xn} using a different method.
For simplicity put fJ. = 1 - A. First, we shall verify that
XI < X3 < ... < X2n+1 < X2n < X2n-2 < ... < X2
holds for each n.

The proof is by induction. For n = 1, the inequalities reduce to XI < X2 which
is obviously true. So, for the inductive step, assume X2n-1 < X2n for some n. Then
and
Now, note that
X2n+1 < AX2n-l + (1 - A)XZ n = XZn+2 < XZ n •
Next, if we let d n = XZn - XZn-h then it is easily to verify (see Figure 1.1) that
d n+ 1 = AfJ.dn , (1)
X2n+l = XZn-1 + fJ.dn , and (2)
X2n+2 = X2n - AZdn . (3)
From (1) it follows that

dn .. I
I
-J.l.dn t...dn .\I
I
I
t... 2d n
r--t...J.l.dn
·11
X211-1 .X2 1I +1 X21l+2 XlII
FIGURE 1.1.
and so from (2) and (3), we obtain

n n
X2n+1 = XI + LCX 2i+1 - X2i-l) = XI +L J1.di
~I ~I
and
n n
X2n+2 = X2 - LC X 2i - X2i+2) = X2 -)...1 L0I.J1.i-ldi
i=1 i=1
Therefore,
·
an d consequent1y, 11m Xn =
)..2a+/-Lb
I-)../-L'
Problem 4.21. Let G be a nonempty subset of JR, which is a group under addition
(i.e., if x, Y E G, then X + y E G and -x E G). Show that between any two
distinct real numbers there exists an element of G or else there exists a E JR such
thatG = {na: n =0,±1,±2, ... }.
Solution. Assume G #- to}. Let a = infG nco, 00). We distinguish two cases.
(1) a > O. In this case, we shall show that G = {na: n = 0, ±l, ... }.
To see this, note first that a E G. Indeed, if a ¢. G, then there exist x, y E G

with a < x < y < ¥. Then, the element z = y - x E G satisfies < z < I'
contradicting the definition of a. Now, if x E G, then na ::: x < (n + l)a must
°
hold for some integer n. However, x = na. must also hold, since otherwise the
°
element x - na E G satisfies < x - na < a, which is again a contradiction.
(2) a = 0. In this case, we claim that between any two distinct real numbers there
is an element of G.
Choose some element Z E G with °°

To see this, we only need to consider < x < y. Let 8 = minIx, y - x} > 0.
< Z < 8. By the Archimedean property,
the set A = (n E :IN: nz 2: y} is nonempty, and by the Well Ordering Principle
the element k = min A exists. Now, note that the element b = (k - l)z E G
satisfies x < b < y.
Probl~m 4.22. Determine the limit points of the sequence (cosn}.
Solution. We claim that the set oflimit points ofthe sequence (cos n} is [-1, 1].
To prove this, we shall need two facts from elementary calculus.
a) The Intermediate Value Theorem; and
b) The inequality I cos x - cos y I ::: Ix - y I for all x, y E lR.
Let G= (n + 2mrr: n, m integers}. Clearly, G is a group under addition,
and since rr is an irrational number, it is easy to see that the group G is not of
the form (na: n = 0, ±1, ±2, ... }. Now, let x E [-1, 1] and let e > 0. By the
Intermediate Value Theorem, there exists some y E lR satisfying cos y = x. The
preceding Problem 4.21 shows that there exist two integers n and m satisfying
y < n + 2mrr < y + e. Thus,
Ix'- cosnl = Icos y - cos(n + 2mrr)1 ::: n + 2mrr' - y < e.
The above arguments show that given x E [-1,1] and e > 0, there exists
some non-negative integer n with Ix - cos nl < e. From this, it easily follows
(how?) that every point of [-1, 1] must be a limit point of {cos n}.
Problem 4.23. For each n define fll: [-1, 1] -+ lR by fll (x) = x". Determine
lim sup fll and lim inf 1".
Solution. We have
1, if x = -1
lim sup fll(x) = 0, if Ixl < 1
[
1, if x = 1
and
-1, if X = -1
= Ixi
liminf fn(x)
1 0,
1,
if
if X
< 1
= 1.
Problem 4.24. Show that every sequence of real numbers has a monotone sub-
sequence. Use this conclusion to provide an alternate proof of the Balzano-
Weierstrass property of the real numbers: Every bounded sequence has a con-
vergent subsequence. (See Corollary 4.7.)
Solution. Let {XII} be a sequence of real numbers. We consider the set of natural
numbers
S = {k E IN: Xk ::: Xm for all m 2: k},
and distinguish two cases.
1. S is infinite.
In this case, we can write S = {kl' k2' ... } with kl < k2 < .... Now, it should
be clear that the subsequence {Xk n } of {xn} is increasing.
2. S is finite (and possibly empty).
°
In this case, if we put kl = 1 + max S (let max S = if S = (5), then for each
k 2: kl there exists some m > ksuch that Xm < Xk. So, by induction, if k n has
been chosen, then we can select some natural number kn+1 with kn+1 > k n and
Xk n+ 1 < Xk". This implies that {Xk n } is a strictly decreasing subsequence of {x n },
and the claim is established.
For the Bolzano-Weierstrass property, notice that if {xn} is a bounded sequence,
then, by the above, {xn} has a monoto~e subsequence which (by Theorem 4.3) must
be convergent in lR. (We remark that this result shows that not only a bounded
sequence has a convergent subsequence but it also has a monotone convergent
subsequence.)
5. TIlE EXTENDED REAL NUMBERS
Problem 5.1. Let {xn} be a sequence oflR*. Define a limit point of {xn} in lR* to
be any element X oflR* for which there exists a subsequence of {xn} that converges
tax.
Show that
limsupxn = inf[sup xkj
n k?;11
and liminfxn = sup [inf
IIk?;n
Xk]
are the largest and smallest limit points of {xn} in lR*.

Section 5: TIIE EXTENDED REAL NUMBERS 35
Solution. The lim sup case is established. Let x = lim sup xn E JR *. Then
three cases arise:
a) x E lR.
In this case, repeat the proof of Theorem 4.6.
b) x = 00.
Inthis case, we have only to show that x is a limit point of {xn }. Note that
V:n Xi = 00 for each n. Choose some kl 2: 1 such that Xk l > 1. Now, by
induction: If k n has been selected so that Xkn > n, then use Vi>knXi = 00 to
choose some kn+1 2: kn + 1 > kn so that xkn+1 > n + 1. Clearly, {XkJ is a
subsequence of {xn} satisfying limxkn = 00.
x = -00.
c)
In this case, we shall show that limxlI = -00. Let < M < 00. From
V:nXi -I- -00, it follows that V:IIXi < -M for some n, and so Xi < -M for
°
all i 2: n. That is, lim XII = -00.
Problem 5.2. Let {xn} be a sequence of positive real numbers sllch that
e = lim Xn+1
Xn exists in JR. Show that:
a. if e < 1, then limxlI = 0, and
b. if e > 1, then limxn = 00.
Solution. (a) Assume e < 1 and fix some 0 such that e < 0 < 1; for instance,
let 0 = I~e.
_
Since lim Xn+1
Xn
= e, there exists some k > 1 such that Xn+1
Xn
< 0 holds
for all n 2: k. Now, if n > k, then note that
XII Xn-I Xk+1

Xn = -- . -- ... -- . Xk
Xn-I Xn-2 Xk
<
o 0
~
0
'Xk =u oll-k
Xk = (Xk)011
ok u ,
'
(n-k)-terms
an'd so 1'f C = 8k'

Xt th
en
°
holds for all n > k. Since (in view of < 0 < 1) on -+ 0, we easily infer that
Xn -+ 0.
(b) Assume now e > 1 and choose some 0 such that 1 < 0 < e. Since
lim X,,+I = e, there exists some k > 1 such that X,,+I > 0 holds for all n > k. Then,
~ ~-
as in the preceding case, there exists some constant C > satisfying Xn > con
for all n > k. From on -+ 00, it easily follows that XII -+ 00.
°
Problem 5.3. Let °: : all •m ::::: 00 for all m, n, and let a: 1N x 1N -+ 1N x 1N be
one-to-one and onto. Show that
n=1 m=1 n=1 m=1
Solution. It follows immediately from Theorem 5.4.
Problem 5.4. Show that
Solution. The convergence or divergence of the series is according to the con-

vergence or divergence of the double integral It'ItO :t:!;2'
Now, note that
00[1 00 dX] dy :::

00
I 1I -dyY = 00.
1I I x 2 + y2
An alternate solution goes as follows. Note first that the inequality
1 1
-::----::-
2 2 > > = -- - ----
n +m (n+m)2 (n+m)(n+m+l) n+m n+m+l
. l'
Imp Ies
,",00 I ,",00 (
L..m=1 n2+m2 ::: L..m=1
I
n+m -
I)
n+m+1 = n+I'
I Th f
ere ore,
Problem 5.5. This problem describes the p-adic representation of a real num-
ber in (0, 1). We assume that p is a natural number such that p > 2 and
x E (0, 1).
a. Divide the interval [0, 1) into the p closed-open intervals
[0, ~), [~, ~), ... , [7' 1),

°
and number them consecutively from to p - 1. Then x belongs precisely
to one of these intervals, say kl (0 :::: kl < p). Next, divide the interval
[;, kl;l) into p closed-open intervals (of the same length), number them
°
consecutively from to p - 1, and let k2 be the subinterval to which x
belongs. Proceeding this way, we construct a sequence {k n } ofnon-negative
Section 5: THE EX1ENDED REAL NUMBERS 37
integers such that °: : k n < p for each n. Show that
k
x=I:.l!...
00
n=1 pn
b. Apply the same process as in (a) by subdividing each interval now into p
open-closed intervals. For example. start with (0, 1] and subdivide it into
the open-closed intervals (0, *], (*, *], ... ,
(7,1].
As in (a), construct a sequence (m n ) of non-negative integers such that
°: : mil < p for each n. Show that
mn
x=I:-. 00
n=1 pn
c. Show by an example that the two sequences constructed in (a) and (b) may
be different.
In order to make the p-adic representation of a number unique, we shall agree to
take the one determined by (a) above. As usual, it will be written as x = O.kl k2 ....
Solution. For (a) and (b) note that Ix -

2:::;'=1 ?I : : ;"
holds for all n.
For part (c) take, for instance, p = 2 and note that for x = we have kl 4 =1
° °
and kn = if n > 1, while ml = and mn = 1 for n > 1.
Problem 5.6. Show that P(lN) ~ 1R by establishing the following:

i. If A is an infinite set. and f: A ~ B is one-to-one such that B \ f(A) is
at-most countable, then show that A ~ B.
ii. Show that the set of numbers of (0, 1) for which the dyadic (i.e., p == 2)
representation determined by (a) and (b) of the preceding exercise are·
different is a countable set.
iii. For each x E (0, 1), let x = 0.klk2··· be the dyadic representation
determined by part (a) of the preceding exercise; clearly, each k; is either
001' 1. Let f(x) = (n E IN: kn = 1). Show that f: (0, 1) ~ P(lN)
is one-to-one such that P(lN) \ f«O, 1)) is countable, and conclude from
part (i) that (0, 1) ~ P(lN).
Solution. (i) Let S = (aI, a2, ... ) be a countable subset of A.

(a) Assume B \ f(A) = (b l , ... , bn ) is a finite set. Then g: A ~ B defined
by g(x) = f(x) if x ¢:. S, g(a;) = b; for 1 :::: i :::: n, and g(a;+n) = f{a;) for
i = 1, 2, ... is one-to-one and onto.
(b) Assume B \ f(A) = (b l , b2, ... ) is countable. Then g: A ~ B defined
by g(x) = f(x) if x ¢:. S, g(a211+1) = f(a n) and g(a2n) = bll for each n is
one-to-one and onto.
(ii) Let D be the set of all numbers of (0, 1) for which the two sequences {kn}
and {mn} determined by the preceding problem are different. Assume x =
0.klk2· .. = 0.mlm2· .. E D and define the natural number r = min{n: kll i-
mil}' We can assume kr = 1 and mr = 0. Then the inequalities
r-l oc 00
x- L~~' = L~::::
n=l n=r n=r+l
L dn
-J..
- 2'
00 r-l
<l+"h-x_"h
- 2' L....t 2" - L....t 2"
n=r+l n=l
r-l
=X-L~:'
n=l
guarantee that x = ~ + ~ + ... + ~;:: + f,-. In particular, note that mn = 1 and

°
kn = hold for each n > r.
On the other hand, it is not difficult to see that every x of the above type belongs
to D. It is now a routine matter to verify that D is a countable set. (It is also
interesting to observe that D consists precisely of the endpoints of the subintervals
appearing during the construction of the expansions.)
(iii) Let A E P(lN). Define the sequence {mn} of {O, 1} by mn = 1 if n E A
°
and mn = if n ¢. A, and then set
00
X - " m"
- L....t 2"'
n=l
Note that A ¢. f(O,1)) if and only if XED. Thus, P(lN) \ f(O, 1)) is
countable, and so by part (i) and the fact that f is one-to-one, (0, 1) ~ P(lN)
holds.
Problem 5.7. For a sequence {xn} of real numbers show that the following con-
ditions are equivalent:
a. The series I::l Xn is rearrangement invariant in JR.
b. For every permutation a of IN the series I::l xu" converges in JR.
c. The series I::l IXnl converges in JR.
d. For evelY sequence {Sn} of {-1, 1}, the series I::l SnXn converges in JR.
e. For evelY subsequence {Xk"} of {x n}, the series I::l Xk" converges in IR.
f. For evelY E > 0, there exists an integer k (depending on E) such that for
every finite subset S of IN with min S ~ k, we have I I:nes Xn I < E.
(Any series I::l Xn satisfying anyone of the above conditions is also referred to
as an unconditionally convergent series.)
Section 5: TIlE EXTENDED REAL NUMBERS 39
Solution. (a)=>(b) Obvious.

(b)=>(c) Assume L:I IXIII = 00. From our hypothesis it follows that Xn > 0
and Xn < 0 both hold for infinitely many n. Split {XII} into two subsequences
{Yn} and {zn} such that Yn 2: 0 and Zn < O· hold for all n. We can assume that
L:I Yn = 00.
Now, use induction to construct a strictly increasing sequence of natural numbers
{kn } such that
1. kl = 1 and ZI + L1~1 Yi > 1; and

2. ZII + L~;:'+~n+1 Yi > 1 for n = 1,2, ....
Then note that
is a permutation of {xn} whose series is not convergent, contrary to our hypothesis.

(c)=>(d) Obvious.
(d)=>(e) Let {Xk n } be a subsequence of {xn}. Put Si = -1 if i =j:. kn for each
n, and Sk" = 1. Then
is a convergent series.
(e)=>(f) If (f) is false, then there exists some e > 0 and a sequence {Sn} of finite
subsets of natural numbers such that max Sn < min Sn+1 and ILiesnxd 2: e hold
for all n. Let
00
USn = {kl' k2,· .. }'

n=1
where kn t . Then, it is easy to see that the series L:::l Xkn does not converge in
JR., contradicting (e).
(f)=>(a) Let 0": IN -+ IN be a permutation. By our hypothesis, it is readily
. seen that the partial sums of both series L:::l Xn and L:I xUn form Cauchy
sequences, and hence, both series converge in JR.. Let X = L:I Xn and Y =
L:l xU".
Now, if e > 0 is given, then choose k so large such that
r r
Ix - LXnl < e, Iy - LXunl < e,
n=1 n=1
and ILXil
ieS
< e
hold for all r 2: k and all finite subsets S of IN with min S 2: k. Fix some r > k
such that for each 1 ::::: i ::::: k there exists 1 ::::: j ::::: r with Xi = x ui ' and note
that
k k r r
Ix - y I: : : Ix - L
n=1
Xn I+ IL Xn - L xUnI+ IL xUn - YI <
n=1 n=1 n=1
8 + 8 + 8 = 38
holds for all 8 > 0, and so x = y. In other words, the series L::I Xn is
rearrangement invariant.
Problem 5.8. A series of the form L::I(-l)n-I xn , where Xn > Ofor each n, is
called an alternating series. Assume that a sequence {xn} of strictly positive real
numbers satisfies Xn .r. o. Then establish the following:
a. The alternating series L::I(-l)n-I xn converges in R.
b. If L::I Xn = 00, then the alternating series L::I (_1)n-1 Xn is not rear-
rangement invariant.
Solution. (a) Let Sn = L:Z=I (-1/- 1Xk. We claim that
holds for each n. The proof is by induction. For n = 1, we have S2 = XI - X2 <

XI =
Sl. So, assume the inequalities to be true for some n. Then taking into
account that X2n - X2n+1 ::: 0 and X2n+1 - X2n+2 ::: 0, we see that
1. S2n::::: S2n + (X2n+1 - X211+2) = S2n+2 = S2(n+I),

2. S2(n+1) = S2n+2 = S2n+1 - X2n+2 ::::: S2n+1 = S2(n+I)-I,
3. S2(~+I)-1 = S2n+1 = S2n-1 - (X2n - X2n+l) ::::: S2n-l,
and our claim is established.

Now, if S2n t s and S2n-1 .r. t hold in R, then clearly s ::::: t. Moreover, from
S2n - S2n-1 = -X2n -+ 0, we obtain s = t. But then, this implies that {sn}
converges to s in R; see Exercise 4.3 of Section 4. Consequently, the alternating
series converges and L:~I(-I)k-Ixk limsn = s. =
(b) We must have either L:~I X2k-1 = 00 or L:~I X2k = 00. Assume
L:~I X2k-1 = 00; the other case can be treated in a similar manner.
Since L:~I X2k-1 = 00, there exist integers 0 = ko < kl < k2 < ... such
that [L:::~n+IX2i-d - X2n > 1 holds for each n = 0, 1, . ... Consider the
rearrangement {Yn} of the sequence {(-I )n-I xn} given by
and note that L::I Yn = 00 holds.

Problem 5.9. This problem describes the integral test for the convergence of
series. Assume that f: [1, 00) 4- [0,00) is a decreasing function. We define the
sequences {all} and {"ill} by
II
an = L f(k) and "ill = JI[" f(x) dx.

k=1
Establish the following:

a. O:s an - "ill :s f(1)for all n.
b. the sequence {an - "ill} is decreasing--and hence. convergent in JR.
c. Show that the series L~I f(k) converges in JR if and only if the improper
Riemann integral ItO
f(x)dx = lim r ...... oo I;
f(x)dx exists in 1R.
Solu~ion.Since f is decreasing, notice that for each k E 1N we have f(x) 2:

f(k+l) and f(x):s f(k) foreachk:s x:s k+1. So,integratingover[k, k+l],
we get:
(k+1
Jk f(x)dx 2: f(k + 1), and (1)
(k+1
Jk f(x) dx :s f(k) (2)
for each k = 1, 2, .... (We remark that as a decreasing function, f is Riemann

integrable over every closed subinterval of [1, 00); see Section 23 of the text.)
(a) Using (1), we see that
n n-I
all = f(l) + L f(k) = f(1) +L f(k + 1)
k=2 k=1
n-I r+1 ["

:s f(1) + (; i f(x)dx = f(l) + il f(x)dx
k
= f(1) + f n •
This implies all "i1l:S f(1) for each n. On the other hand, using (2), we see
that
II-I n-I {k+1 {"

all 2: ( ; f(k) 2: {; Jk f(x)dx = JI f(x)dx = "ill'
and so all - "in 2: 0 for each n.

(b) Using once more (1), we get

n +1
all+1 - TII+I = all + f(n + 1) -
"+ 1
Til -
1II
f(x)dx
=an-TII -
[1 n f(X)dx- f (n+l)]::::all -Tn .
This shows that {all - Til} is a decreasing sequence-and so lim(all - Tn) exists in
lR.
(c) Since {all} and {Til} are both increasing sequences of non-negative real num-
bers they both converge in lR*, and clearly
00
lim an =L f(k) and lim Til = foo f(x)dx.

I
n~oo k=1 n~oo
But from part (a), we have Tn :::: an :::: f(1) + Tn for each n, and therefore (by
letting n -+ 00), we have
0::::
fI
oo
f(x)dx :::: L f(k) :::: f(l) + foo f(x)dx.
00
k=1 I
This inequality shows that I:~I f(k) converges if and only the improper Ri~ann
oo
integral fl f (x) dx exists.
Problem 5.10. Use the preceding problem to show that the series I::I t!P does
not converge in lRfor 0 1. The following
are problems related to the harmonic series 2:::1 ~.
a. Prove with (at least) three different ways that 2:::1 ~ = 00.
b. If a computer starting at 12 midnight on December 31, 1939, adds one
million terms of the harmonic series every second, what was the value
(within an error of 1) of the sum at 12 midnight on December 31, 1997?
(Assume that each year has 365 days.)
00
c. Show that "\'(_1)"-1

~ 11
= n-+oo n n
1 ) = In2.
lim (_+11 + +12 + ... + -2
n
11=1
Solution. Notice that
oo dx . fr dx
= { lim
lim
f
rl-P-I if p =1= 1,
- = r~oo
hm - r~oo I-p
1 xp 1 Xp • _ lnr
r oo if p = 1.
This limit is finite if p > 1 and infinity if 0 < p :::: 1.

(a) We let 0"11 = I:Z=I t. Here are four proofs of the divergence of the harmonic
series.
1. Notice that 0"211 - O"n = n~1 + 1Z~2 +: .. + IZ~IZ 2: n = ~ for each n.'1n
This shows that {O"n} is not a Cauchy sequence, and hence, divergent.
2. As shown at the beginning of the solution of the problem, It" ~/ = 00, and
1
so £""'11=1 ii = 00.
,",00
3. We claim that 0"2n 2: 1 + ~ for each n. (If this inequality is established,

t
then clearly I:~I = lim0"2n = 00.) The proof of the inequality is by
induction. For n = 1, we have 0"21 = 0"2 = 1 + !.
Now, if we assume the
inequality true for some n, then
1 1 1
n 1
0"2 + = 0"2
n
+ 211 + 1 + 211 + 2 + ... + 211 + 2n
n n 1 n+l
> 1+-+2 · - -11= 1 + - - .
2 2.2 2
4. Note that
(b) From Problem 5.9, we know that the harmonic series is associated with
the function f(x) = ~ and that OSO"n-ln ns/(1) = 1 for eachn. So, Inn
approximates 0"n within an error of one. If the computer started adding the terms of
the harmonic series at 12 midnight on December 31,1939, then up to 12 midnight
on December 31,1997, there are
S7(years) x 365(days) x 24(hours) x 60(minutes) x 60(seconds)
= 1,797,552 x 103 seconds.
So, if the computer adds 1, 000, 000 = 106 terms per second of the harmonic
series, the last number N added a second before midnight on December 31 of
1997 is N = 1,797,552 x 109 • Therefore,
LN -1 = O"N ~ InN = 35.12520213 ... ,

n=1 n
which shows that the harmonic series is a "very slow" divergent series.
(c) From Problem 5.8, we know that the alternating series 2:~1 (-It
I is con-
vergent in 1R. Also, from Problem 5.9, we know that lim(an -Inn) y E 1R. So,=
if we let Xn = Y - (an -In n), then Xn -+ 0 and all = Y + Inn + Xli for each n.
Now, note that
~ (-lki-1 1 1 ... +---(-+-+

1 1 1 ... +-)
1
L..--'---- = 1+-+-+
k=1 3 5 2n - 1 2 4 2n
= 1+ ~ + ~ + ... + ..!... - 2(~ + ~ + ... + ..!...)

23 2n 24 2n
1 1 1
= n+l +n+2+···+2n
= a2n -an = [Y + In(2n) + X2n] - [Y +lnn +Xn]
= ln2+X2n -Xn,
for each n. This implies 2:~1 (-It I = In2.
Problem 5.11 (Toeplitz). Let {an} be a sequence of positive real numbers (i.e.,
an > Ofor each n) and put b n = 2:7=1 a;. Assume that bn t 2:;:1 a; = 00. If
{xn} is a sequence of real numbers such that Xn -+ X in JR, then show that
I n
lim - La;x; =x.
n-+oo bn ;=1
Solution. Let E > O. Choose some k such that IXn - xl < E for each n ::: k. Put
M = max{lx; -xl: i = 1, ... , k}, and then selectsome.e > k such that ~~k < E
for alln ::: e. Now, notice that if n ::: .e, then
1 n 1 n 1 n
Ibn t;a;x; -x I = Ibn t;a;x; - b t;a;x
n
I
1 kIn
:::: bLa;Jx;-x l + L a;lx;-xl
bII ;=k+1
n ;=1
Mbk
:::: !;; + E < E + E = 2E,
and the conclusion follows. (Note that this problem is a substantial generalization
of Problem 4.11.)
Problem 5.12 (Kronecker). Assume that a sequence of positive numbers {bll}

satisfies 0 < bl < b 2 < b3 < ... and bn t 00. If a series 2::lxn of real numbers
Section 6: MErRIe SPACES 45
converges in JR, then show that

1 II
lim - LbiXi = O.
II-+?C bll i=1 .
In particular, show that if {YII) is a sequence of real numbers such that the series
,,",00 Ia. converges in JR then YI+-·+Y. -+ O.
.L...,1l=1 IZ 'n
Solution. Let x = I::IXn. Put bo = 0, So = 0, and Sll = XI + ... + XII for

each n 2: 1. Now, notice that
Il n It
LbiXi = Lbi(Si - Si-I) = bllsll - LSi-l(bi - bi-I).

i=1 i=1 i=2
Ther~fore, *. I:;'=I biXi = SII - *. I:~=2 Si-I (bi - bi-I). Since bi - bi-I > 0 for
each i and I:;'= I(bi - bi-I) = bll t 00, it follows from the preceding problem that
*. I:;'=2Si-l(bi - bi-I) = x. Hence,
1 II 1 II
lim -b LbiXi = lim [SII - -b LSi-l(bi - bi-d ] = X - X = O.
n-+oo n ;=1 n-+oo IJ ;=2
For the second part, notice that if I::I ~ is convergent in JR, then let bn =n
for each n and notice that (by the above)
1 "II" . y; _ YI + .... + Yn
-b L....-b'T - ~O,
II i=1 n
as desired.
6. METRIC SPACES
Problem 6.1. For subsets A and B of a metric space (X, d) show that:
a. (A n B)O = AO n BO.
b. AOUB°S;(AUB)°.
c. AUB = AUB.
d. A nBS; An B.
e. If B is open, then A nBS; A n B.
Solution. (a) From (AnB)° S; AO and (AnB)O S; BO,itfollowsthat (AnB)O S;

A° n BO. On the other hand, since A° n BO S; A n B holds and A° n BO is open,
it easily follows that AO n BO S; (A n B)o.
(b) From A S; AUB,itfollowsthat AO S; (AUB)o. Similarly,BO S; (AUB)O,

and the desired inclusion follows.
(c) From S S; S and the fact that S is a closed set for any subset S, we see that
AU B S; AU B U AU B = AU B S; AU B = AU B.
(d) Since An B S; An B, we have An B S; An B = An B.
(e) If x E An B and r > 0, then choose some 0 < 8 < r with B(x,8) S; B,
and note that
B(x, r) n (A n B) ;2 B(x, 8) n B n A = B(x, 8) n A # 0.

That is, x E A n B, and so An B S; A n B holds.
Problem 6.2. Show that in a Euclidean space JRII with the Euclidean distance,
the closure ofany open ball B (a, r) is the closed ball {x E JRn: d(x, a) ::: r}. Give
an example of a complete metric space for which the corresponding statement is
false.
Solution. Let C(a, r) = {x E JRk: dCa, x) ::: r}. Since C(a, r) is a closed set,
it follows that B(a,r) S; C(a,r).
For the other inclusion, let x E C(a, r). For each n let Xn = ~a + (1 - ~)x.
The inequalities
imply that {xn} S; B(a, r) and XII ~ x. Consequently, x E B(a, r), and thus
C(a, r) S; B(a, r) also holds.
For a counterexample, consider X = {O, I} with the discrete distance, and
note that X is a complete metric space. Also, observe that B(O, 1) = {OJ, while
C(O, 1) = {O, I}.
Problem 6.3. If A is a nonempty subset of JR, then show that the set
B = {a E A: There exists some e > 0 with (a, a + e) n A = 0}

is at-most countable.
Solution. Foreacha E B pickarationalnumberro > a so that (a, ro) n A = 0.

We claim that if a, b E B satisfy a # b, then ro # rh. Indeed, if a < band
Section 6: METRIC SPACES 47
ra = rb hold, then-since b E (a, ra )-the open interval (a, ra) is a neighborhood

of b E A, and so (a, ra) n A =j:. 0, contrary to the choice of ra.
The above show that the mapping a J---+ ra , from B into the set of rational
numbers, is one-to-one. Consequently, the set B is at-most countable.
Problem 6.4. Let f: (X, d) -* (Y, p) be afunction. Show that f is continuous

if and only if f- I(B O) S;;; [I-I (B) t
for every subset B of Y.
Solution. Assume f continuous, and B S;;; Y. Since B O is open, the set

f-I(BO) is likewise open. Thus, in view of B O S;;; B, we have
In the opposite direction, assume that the condition is satisfied. If B S;;; Y is

=
open (i.e., if B BO holds), then
shows that f- I (B) is open. Therefore, f is continuous.
Problem 6.5. Show that the boundary of a closed or open set in a metric space
is nowhere dense. Is this statement true for an arbitrary subset?
Solution. Since aA = aAc = A n Ac holds, we can assume that A is closed.

Thus,
(aA)O = (AnAcr=(AnAcr=Aon(Acr
S;;; AD n AC = AD n (Ao)" = 0.
Since aA is closed, this shows that aA is nowhere dense.

An alternate proof goes as follows: If x E (aA)O, then there exists some r > 0
such that B(x, r) S;;; aA = An AC S;;; A. This implies B(x, r) n AC= 0, contrary
tox E AC.
The boundary of an arbitrary set need not be nowhere dense. An example: Let
=
X JR with the Euclidean distance, and let A =
Q (the set of rational numbers).
Note that aA lR.=
Problem 6.6. Show that the set of irrational numbers is not a countable union
of closed subsets oflR.
Solution. Let I denote the set of all irrational numbers, and let {rJ, r2, ... } be
an enumeration of the rational numbers of JR.
Assume by way of contradiction that there exists a sequence of closed sets {An}
of JR such that I = U:'I An. Then
and by the Baire Category Theorem (Theorem 6.18), we must have (An)O =1= (j;
for some n. Thus, some An contains an interval. However, since An S;;; I holds
and each interval contains rational numbers, this is impossible, and the conclusion
follows.
Problem 6.7. Let (X, d) be a metric space. Show that if{xn} and {Yn} are Cauchy
sequences of X, then {d(xn. Yn)} converges in JR.
Solution. Use the inequality
(Also, see the discussion before Theorem 6.19.)
Problem 6.S. Show that in a metric space a Cauchy sequence converges if and
only ifit has a convergent subsequence.
Solution. Let {XII} be a Cauchy sequence in a metric space (X, d). If XII -+ X
holds in X, then every subsequence of {xn} converges to x.
For the converse, assume that there exists a subsequence {Xkn } of {XII} such
that Xkn -+ X holds in X. Let E > O. Choose no such that d (Xkn ' x) < E and
d(x n , xm) < E for n, m > no. Now, if n > no, then k n 2: n > no, and so
This shows that, lim Xn = X holds in X.
Problem 6.9. Prove that the closed interval [0, 1] is an uncountable set:
a. by using Cantor's Theorem 6.14, and
b. by using Baire's Theorem 6.17.
Solution. (a) Assume by way of 'contradiction that [0, 1] is a countable set,

say [0,1] = {XI, X2, •• •}. We consider [0, 1] equipped with the usual distance
d(x, y) = Ix - yl so that [0, 1] is a complete metric space.
Subdivide [0, 1] into three closed subintervals (as in the construction of the
Cantor set) of equal length. Remove from [0, 1] the middle open subinterval and
consider the remaining two closed subintervals (here the subintervals [0, tJ
and
[~, 1]) and then select one of them, say II, such that Xl 1:- II. Next, repeat this
process with II in place of [0, 1] and select a closed subinterval h of II oflength
equal to one-third of II such that X2 1:- h Inductively, assume that we have chosen
n closed intervals II, ... , In such that:
1. In ~ Ill-I ~ ... ~ 12 ~ Ij,
2. Xk 1:- h for k = 1, ... , n, and
3. the length of each h is t..
As above, there exists a closed subinterval I n+ I of In of length equal to one-third
of In such that XII + 1 1:- IIl+l·
Thus, there exists a sequence {Ill} of closed subintevals of [0, 1] such that
111 +1' ~ In, XII 1:- III and dUll) = fn
for each n. By Theorem 6.14, we infer that
n:l III consists exactly of one point. But, since XIl 1:- III for each n, we see that
n::"=1 III = 0, a contradiction. Hence, [0, 1] must be uncountable.
(b) Again, assume by way of contradiction that [0, 1] = {XI, X2, ... } and again
we consider [0, 1] as a complete metric space. If All = {x ll }, then each All is closed
and has no interior points. However, Theorem 6.17 (or Theorem 6.18) applied to
the equality [0, 1] = U:l All implies that some All must have an interior point,
which is impossible. This shows that [0, 1] cannot be countable.
Problem 6.10. Let {rl, r2, ... } be an enumeration of all rational numbers in the
interval [0, 1] and for each X E [0, 1] let Ax = {n E IN: rn :::: x}. Define the
function f: [0,1] -+ lR by the formula
f(x) = L-in·
ilEA.,
Show that f restricted to the set of irrational numbers of[O, 1] is continuous.
Solution. Fix an irrational number a E [0, 1] and let 8 > O. Pick a natural
number k such that L:k -in < 8 and let
We claim that if x E [0, 1] is an irrational number, then la - xl < Ii implies

If(x) - f(a)1 < 8 (which tells us that f is continuous when restricted to the
irrational numbers).
To see this, let x E [0, 1] be an arbitrary irrational number satisfying Ix - a I < Ii.
Let Ix denote the half-open subinterval of [0, 1] which is open at left and closed
at right having endpoints a and x. If Bx = {n E IN: rn E Ix }, then note that

Bx ~ {k, k + 1, k + 2, ... } (why?), and so
00
If(x) - f(a)1 = Lt.:::: L t. < e

neB, n=k
holds, as claimed.
Problem 6.11. This problem concerns connected metric spaces. A metric space
(X, d) is said to be connected whenever 0 and X are the only subsets of X that
are simultaneously open and closed. A subset A of a metric space (X, d) is said
to be connected whenever (A, d) is itself a connected metric space. Establish the
following properties regarding connected metric spaces and connected sets.
a. A metric space (X, d) is connected if and only if every continuous function
f: X -+ to, I} is constant. where the two point set to, I} is considered to
be a metric space under the discrete metric.
b. If in a metric space (X, d) we have B ~ A ~ X. then the set B is a
connected subset of (A, d) if and only if B is a connected subset of (X, d).
c. If f: (X, d) -+ (Y, p) is a continuous function and A is a connected subset
ofX. then f(A) is a connected subset ofY.
d. If {Ai }iel is a family of connected subsets of a metric space such that
niel Ai #- 0. then UiE/ Ai is likewise a connected set.
e. If A is a subset of a metric space and a E A. then there exists a largest
(with respect to inclusion) connected subset C a of A that contains a. (The
connected set C a is called the component of a with respect to A.)
f. If G, b belong to a subset A of G metric space and Ca and Cb are the
components of G and bin A. then either Ca = Cb or else Ca nCb = 0.
Hence. the identity A = UaeA C a shows that A can be written as a disjoint
union of connected sets.
g. A non empty subset of R with at least two elements is a connected set if and
only ifit is an interval. Use this and the conclusion of(f) to infer that eve])
open subset of R can be written as an at-most countable union of disjoint
open intervals.
Solution. (a) Let (X, d) be a connected space and let f: X - + to, I} be a

continuous function. Then the set A = f-I(O) is an open and closed subset of
X. Since X is connected either A = 0 (in which case f (x) = 1 holds for each
x E X) or A = X (in which case f(x) = 0 holds for each x EX).
For the converse, assume that every continuous function from X into to, 1}
is constant and let A be a closed and open subset of X. Then the function
f; X --+ JR defined by
= {~:
if x E A;
f(x) if ~ ¢. A,
is continuous (why?). By our hypothesis, f must be a constant function, and this

implies that either A = 0 or A = X, Le., X is a connected metric space.
(b) It follows immediately from (a).
(c) Assume that f and A satisfy the stated properties and consider the contin-
uous functions (A, d) ...£,. (t(A), p) ~ {O, I}. By (a), the continuous function
g 0 f must be a constant function and from this, we see that g is also a constant
function. By (a), (t(A), p) is a connected metric space.
(d) Assume the family {Ai; i E I} satisfies the stated properties. Put A =
UielA i and let f;(A,d) ---+ {O, I} be a continuous function. Then the function
f; (Ai> d)---+{O, 1} is a continuous function and so f restricted to each Ai is
constant. Since nielAi i= 0, we see that f is constant on A, and so--by
(a)-the set A is connected.
(e) Fix a E A and let
A = {B ~ A; B is connected and a E B}.
Note that {a} E A and that nBeAB i= 0. By (d), the set Ca = UBeAB is a
connected subset of A that satisfies the desired properties.
(f) If Ca nCb i= 0, then by (d) we infer that Ca U Cb is a connected set
containing a. Hence, Cb ~ Ca U Cb ~ Ca. Similarly, Ca ~ Cb and so Ca = Cb.
(g) Let A be a connected subset of JR and let a, b E A satisfy a < b. If
a < x < b and x ¢. A, then the set A n (-00, x) is a proper and closed subset
of A (why?), a contradiction. Thus, (a, b) ~ A holds and this shOws that A is an
interval.
For the converse, assume that I is an interval of JR. Assume by way of
contradiction that there exists an onto continuous function f; I ---+ {O, I}. Pick
a, bEl such that f(a) = 0 and feb) = 1; we can suppose that a < b (and so
[a, b] ~ I). Now, let
= sup{c E [a, b); fCc) = OJ.

Co
By the continuity of f, we see that f(eo) = 0 and that Co < b. Then f(x) = 1
holds for all Co < x < b, and so (by the continuity of f again) f(co) = 1 must
also hold, which is impossible. Therefore, every continuous function from I into
{O, I} is constant, and so by (a) the interval I is a connected set.
Finally, note that if I is an open subset of JR, then by (f) we know that I =
Uael Ca, where each Ca is a connected set. It easily follows (how?) that each
C a is an open interval and that there are at-most countably many of them.
Problem 6.12. Show that JRn with the Euclidean distance is a connected metric
space. Use this conclusion to establish that, if the intersection o/two open subsets
olJRn is a proper closed set, then the two open sets must be disjoint.
Solution. Let d denote the Euclidean distance of JRn, i.e., let

n I
d(x,y) = [~)Xi - Yiir.

i=1
For each x E JRn, let Lx denote the line segment joining 0 and x, i.e., let
Lx = {tx: 0::: t ::: I}. We claim that Lx is a connected set.
To see this, note that the function I: [0, 1] -+ Lx, defined by I(t) = tx,
satisfies I/(t) - l(s)1 ::: d(x, O)ls - tl, and so I is (unifonnly) continuous. From
parts (g) and (c) of Problem 6.11, we see that Lx is a connected set.
Now, use part (d) of the preceding problem and the identity R n = UXERnLx to
infer that R n is itself a connected metric space.
For the last part of the problem, let U and V be two open subsets of R n such
that K = U n V is a closed set. Then K is both open and closed (and since K
is a proper subset of Rn) it must be the empty set.
Problem 6.13. Let C be a nonempty closed subset 01R. Show that a function
I: C -+ R is continuous if and only if it can be extended to a continuous real-
valued function on R.
Solution. Let C be a nonempty closed subset of R and let I: C -+ JR be a

function. If I can be extended to a continuous real-valued function on JR, then
I: C -+ R is obviously continuous.
For the converse, assume that I: C -+ R is a continuous function. Start
by observing that the complement CC of C is an open set and so (by part (g)
of Problem 6.11) Cc can be written as an at-most countable union of pairwise
disjoint open intervals; say CC = Uie/ (ai, bi ), where I is at-most countable.
Since the open intervals (ai, b i ): i E I} are pairwise disjoint, it follows that all
the endpoints ai and bi belong to C. Hence, I(aj) and I(b j ) are defined for
each i. Now, extend the domain of I by defining the graph of the function I on
the interval (aj, bj ) to be the straight line segment joining the points (aj, I(ai»
and (bj , I(bj». In other words, for each ai < x < b i we let I(x) = I(ai) +
f(b~:=~(ai) (x - aj )-in case (ai, b i ) = (-00, bj ) or (a;, bi ) = (aj, (0) let I (x) =
bj or I(x) = aj.
We claim that this extension of I to all of R is continuous. Clearly, I is
continuous at every point of Co and·at every point of CC (why?). We only need
to verify that I is continuous at the boundary points of C. So, let a E ac and
let Xn -+ a with {x n } S; CC-if (xn} S; C, then I(x n) -+ I(a) is trivially
true. Also, we shall assume that a is not one of the endpoints ai or bi . For each
n pick (the unique) in E I with ain ::::; Xn ::::; bin' Note that in this case, we must
have limain = lim bin = a (why?). From
= II(b;n)- I(a;n) (x
ain -bin n
-a·'II )1 -< \f(b.In ) - f(a·In )\
-+ If(a) - f(a)1 = 0,
we see that lim f(x lI ) = f(a). A similar conclusion holds true if a is one of the
endpoints ai or bi . This shows that f is continuous at a, as claimed.
For an alternate proof see Problem 10.11.
Problem 6.14. Show that a metric space is a Baire space if and only if the
comple"!ent of every meager set is dense.
Solution. Let X be a metric space. Assume first that X is a Baire space and
let A be a meager set. Pick a sequence {An} of nowhere dense sets such that
A = U:-;"=1 All' To show that N is dense, it suffices to show that V n AC :f= 0 for
each nonempty open set V. To see this, let V be a nonempty open set, and assume
by way of contradiction that V n A C = 0. This implies V S;; A, and so
00
V = UVnA n .
n=1
Hence, V is a nonempty open meager set, a contradiction. Hence, AC is a dense

set.
For the converse, assume that the complement of every meager set is dense,
and let V be an open meager set. Then V C is dense. So, if V is nonempty, then
V n v c :f= 0, which is impossible. Thus, the empty set is the only open meager
set, and hence, X is a Baire space.
Problem 6.15. A subset of a metric space is called co-meager if its complement

is a meager set. For a subset A of a Bail'e space show that:
a. A is co-meager if and only ifit contains a dense Go-set.
b. A is meager ifand only ifit is contained in an F,,-setwhose complement is
dense.
Solution. Notice that if A is a nowhere dense set in a metric space X, then from
Lemma 6.8 we see that
This implies that a subset A is nowhere dense if and only if the open set (A:)c is
dense.
Now, assume that X is a Baire space and let A be a subset of X.
(a) Suppose first that A is a co-meager set. Then there exists a sequence {All) of
nowhere dense sets such that A = (U~l Alit This implies
00 00
C
A = nAn ~ nAil.
n=1 n=1
By the above discussion, each set (An)C is an open dense set, and since X is a
Baire space, the G.-set E = n~1 (AII)C is also dense (see Theorem 6.16). Now,
a glance at (*) shows that E ~ A.
For the converse, assume that A contains a dense G.-set B, i.e., B ~ A. So,
there exists a sequence {Vn} of open sets such that B = n:1
Vn. From B ~ Vn,
we see that each VII is also dense. This implies
and so each (V,J is nowhere dense closed set. Now, use the inclusion
00
A ~ B = U(Vllt
C C
n=1
to conclude that A C is a meager set, i.e., A is a co-meager set.

(b) Assume first that A is a meager set, i.e., A C is a co-meager set. By part (a),
there exists a dense G • .:set E such that E ~ N. This implies A ~ £C, where now
E is an Fa-set whose complement (EC)C = E is dense.
For the converse, assume that A ~ F holds, where F is an Fa-set with dense
complement. It follows that F C ~ AC, where now F Cis a dense G.-set. By part
(a), N is co-meager set, which means that A is a meager set.
7. COMPACTNESS IN METRIC SPACES
Problem 7.1. Let I: (X, d) -+ (Y, p) be afunction. Show that I is continuous

if and only if I restricted to the compact subsets 01 X is continuous.
Solution. Assume that I is continuous on every compact set. Let Xn -+ x.
Then the set A = {x 1, X2, ••• } U {x} is compact (note that every open cover of A
can be reduced to a finite cover), and Xn -+ x holds in A. Since I restricted to
A is continuous, lim I(x n ) = I(x) holds, which shows that I is continuous.
Section 7: COMPACTNESS IN METRIC SPACES 55
Problem 7.2. A metric space is said to be separable if it contains a countable

subset that is dense in the space. Show that every compact space (X, d) is sepa-
rable.
Solution. Foreach n choose a finite subset Fn of X suchthatX=UxEFnB(x, ~).

Let F = U:t Fn, and note that F is at-most countable.
Now, let x E X and r > 0. Pick some n with ~ < r. Then there exists some
Y E Fn with d(x, y) <
X.
*
< r. Thus, B(x, r) n F "# 0, and so F is dense in
Problem 7.3. Show that if(X, d) is a separable metric space (see the preceding
exercise for the definition), then card X ~ c.
Solution. Let {XI, X2,"'} be a countable dense subset of X. Consider the

collection of open balls {B(Xi, ~): i, j = 1,2, ... }. Clearly, this collection is
countable; let {BI' B2, ... } be ~ne of its enumerations. Now, for each x E X
define the set Sx = {n E IN: x E Bn}. Thus, a mapping x 1--+ Sx from
X into P(lN) has been established that is clearly one-to-one. Consequently,
card X ~ card P(lN) = c. (See also Problem 5.6.)
Problem 7.4. Let (XI, d l ), ••• , (X n, dn ) be arbitrary metric spaces, and let
X = XI X .•• X Xn' ffx = (XI, ... ,xn) and y = (Yt, .. ·, Yn), define
n 11 I
DI(X, y) = I>m(Xm, Ym) and D2(X, y) = (L.:.:[dm(X m, Ym)fr.

m=1 m=1
a. Show that DI and D2 are distances on X.

b. Show that DI is equivalent to D2.
c. Show that (X, DI) is complete if and only if each (X/, di ) is complete.
d. Show that (X, Dt} is compact if and only if each (Xi, dd is compact.
Solution. (a) Routine.

(b) Use the inequalities
(c) Assume that each Xm (m = 1, ... , n) is a complete metric space. Let

{xd be a DI-Cauchy sequence of X, where Xk = (xt, ... , x~). Clearly, each
{x~} is a Cauchy sequence of X m, and thus there exists Xm E Xm such that
limk~oo dm(x!, xm) = 0. Hence, if X = (xt. ... , XI!) E X, then we have
limk~oo DI (Xb x) = 0, so that the metric space X is Dt-complete.
Now, let X be DI-complete. Fix an element (YI,"" Yn) E X. Let {x;/} be a

Cauchy sequence of X m. If Xk E X is the element whose jth component equals
Yj for j #- m and equals x~ if j = m, then {xd is a Cauchy sequence of X. If
x E X is its limit, then it is easy to see that limk->oo dm(x~, x m) = 0, so that each
Xm is complete.
(d) Assume first that each Xm is compact. Then following the proof of the
second part of Theorem 7.4, we can see that every sequence of X has a convergent
subsequence, and so X must be a compact metric space.
On the other hand, if X is a compact metric space, then the function 1m: X ---1-
X m, defined by 1m (x I, ... , Xn) = Xm, is continuous and onto for each 1 ::: m ::: n.
Hence, by Theorem 7.5, each Xm = Im(X) is compact.
Problem 7.5. Let {(X n , dn )} be a sequence 01 metric spaces, and let X =

n:1 X n. For each x = {xn} and Y = {Yn} in X, define
a. Show that d is a distance on X.

b. Show that (X, d) is a complete metric space if and only if each (X n , dn ) is
complete.
c. Show that (X, d) is a compact metric space if and only if each (X n , dn ) is
compact.
Solution. (a) Note first that if d is a distance on a set X, then p(x, y) = I!~(~:~)
is likewise a distance on X, which is equivalent to d. From this observation it
easily follows that
is a distance on X = n:IXn'
(b) Let {xk} be a sequence of X, where xk = (xf, x~, ...). The proof follows
from the following two properties (whose verifications are straightforward).
1. xk ---1- x holds in X if and only if xf ---1- Xi holds in Xi for each i; and

2. {xk} is a Cauchy sequence in X if and only if {xf} is a Cauchy sequence
in Xi for each i.
(c) Assume that (X, d) is a compa:ct metric space. Then the function /;: X ---1-
Xi, defined by I(x) = Xi for each x = (XI, X2, ... ) EX, is continuous and onto.
By Theorem 7.5, each Xi is a compact metric space.
Section 7: COMPACfNFSS IN MErRIC SPACES 57
For the converse, assume that each Xi is a compact metric space. By (b), X
is a complete metric space, and so by Theorem 7.8 it suffices to show that X is
totally bounded. To this end, let 8 > O. Choose n such that 2- n < 8, and note
that
~ 1 di(Xi, Yi)
Pn(x, y) = L..- 2i . 1 + d.(. .)
i=1 ,x"y,
defines a distance on f17=1 Xi. It should be clear that Pn is equivalent to the

distances of the preceding problem, and (I17=1 Xi, Pn) is a compact metric space.
Choose a finite subset F of I17=1 Xi such that the Pn-balls with centers at the
points of F and radii 8 cover f17=1 Xi. Next, extend each x E F to an element
of X (Le., add to each x E F arbitrary components Xn+1, Xn+2, ••• ). Now, if
Y = (Yl, Y2,"') E X, then pick some x E F with Pn(x, y) < 8, and note that
~ 1 di(Xi, yj) -n
d (x, y) = PIJ (.-c, y) + L..- --:-. d( < 8 +2 < 8 + 8 = 2e.
2
i=n+1 ' 1 + i Xi, Yi)
Thus, X = UrEF B (X, 28) holds, and therefore, X is totally bounded.
Problem 7.6. A family of set :F is said to have the finite intersection property
if every finite intersection of sets of:F is nonempty. Show that a metric space
is compact if and only if every family of closed sets with the finite intersection
property has a nonempty intersection.
Solution. Let X be compact, and let {Ai: i E I} be a family of closed sets with
the finite intersection property. If niEIAi = 0, then X = UiEIAT holds, and by
the compactness of X , there exist ii, ... , in E I such that X = Ui= 1AT}' Thus,
nj=1 Aij = 0, contrary to our hypothesis. Hence, niEI Ai =1= O.
For the converse, assume that every family of closed sets with the finite in-
tersectionproperty has a nonempty intersection. Let X =
UiEI Vj be an open
cover. Then niEI Vic = 0, and since {Vic: i E l} is a family of closed sets, our
hypothesis guarantees the existence of a finite number of indices ii, ... ,in such
that ni=l Vi~ = O. Thus, X = Ui=l Vij holds, so that X is a compact metric
space.
Problem 7.7. Let f: X --+ X be a function from a set X into itself. A point
a E X is called a fixed pointfor f if f(a) = a.
Assume that (X, d) is a compact metric space and f: X --+ X satisfies the
inequality d(f(x), f(y)) < d(x, y)for x =1= y. Show that f has a unique fixed
point.
S8 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
Solution. Note first that f has at most one fixed point. Indeed, if f(x) = x
and f(y) = y hold with x ::/= y, then
d(x, y) = d(J(x), f(y)) < d(x, y)
must hold, which is absurd.

Now, define the function g: X -+ 1R. by g(x) = d(x, f(x)). From the inequality
ig(x) - g(y)i 5 d(J(x), f(y)) + d(x, y) 5 2d(x, y)
(see the discussion preceding Theorem 6.19), it follows that g is continuous.

Since X is compact, g attains its minimum at some point a EX. If f(a) ::/= a,
then the inequality
g(J(a)) = d(J(a), f(f(a))) < d(a, f(a)) = g(a)
shows that g does not attain a minimum at a. Thus, f(a) = a must hold, and so
a is a (unique) fixed point for f.
Problem 7.S. Let (X, d) be a metric space. Afunction f: X -+ X is called a
contraction if there exists some 0 < a < 1 such that d(f(x), f(y)) 5 ad(x, y)
for all x, y EX; a is called a contraction constant.
Show that every contraction f on a complete metric space (X, d) has a unique
fixed point; that is, show that there exists a unique point x E X such that f(x) = x.
Solution. Note first that if f(x) = x and f(y) = y hold, then the inequality
d(x, y) = d(J(x), f(y)) 5 ad(x, y) easily implies that d(x, y) = 0, and so
x = y. That is, f has at-most one fixed point.
To see that f has a fixed point, choose some a EX, and then define the
sequence {x n } inductively by
Xl =a and xn+l = f(x n) for n = 1,2, ....

From our condition, it follows that
d(Xn+l, x n) = d(J(x n), f(xn-d) 5 ad(xn,xn-d
holds for n = 2,3, .... Thus, as in Problem 4.15, we can show that {x n } is
a Cauchy sequence. Since X is complete, {x n } is a convergent sequence. Let
x = lirnx n • Now, by observing that f is (uniformly) continuous, we obtain that
x = n-..oo
lim Xn+l = lim
. n~oo
f(x n) = f(x),
and so x is a (unique) fixed point for f.

Section 7: COMPACI'NFSS IN METRIC SPACES 59
Problem 7.9. A property of a metric space is called a topological property if it

is preserved in a homeomorphic metric space.
a. Show that compactness is a topological property.
b. Show that completeness, boundedness, and total boundedness are not topo-
logical properties.
Solution. (a) It follows from Theorem 7.5.

(b) Consider (0,1] and [1,00) as metric spaces under the usual Euclidean
distance d(x, y) = Ix - YI. Clearly, (0, 1] is not complete but it is bounded and
totally bounded. Also, [1, 00) is complete (because it is a closed subset of R), but
is neither bounded nor totally bounded. On the other hand, f: (0,1] -r [1,00),
defined by f(x) = 1,x is a homeomorphism, and the claims in (b) follow.
Problem 7.10. Let(X, d) be a metric space. Define the distance oftwo nonempty
subsets A and B of X by
dCA, B) = inf{d(x, y): x E A and y E B}.
a.
b.
°
Give an example of two closed sets A and B of some metric space with
An B = and such that dCA, B) = 0.
If An B = 0, A is closed, and B is compact (and, of course, both are
nonempty), then show that dCA, B) > 0.
Solution. (a) Let X = R2 with the Euclidean distance, and consider the closed
subsets of X
= {(x, ±): x 2: 1} and

A B = {(x, 0): x 2: I}.
Note that An B = 0, while dCA, B) = 0.

(b) Let A and B be as stated in the problem. If dCA, B) = 0, then pick two
sequences {XII} £; A and {Yn} £; B with d(xll' YIl) -+ 0. Since B is compact,
by passing to a subsequence (if necessary), we can assume that Y,t -+ Y holds
for some Y E B. The inequality
shows that d(xn' y) -+ 0. Since A is closed, YEA, and hence An B =P 0,

°
contrary to our hypothesis. Therefore, dCA, B) > must hold.
Problem 7.11. Let (X, d) be a compact metric space and f: X -r X an isome-

fly; that is, d(f(x), fey»~ = d(x, y) holds for all x, Y EX. Then show that f is
onto. Does the conclusion remain true if X is not assumed to be compact?
Solution. Let y EX. Define the sequence {xn} of I(X) by
Xl = I(y) and Xn+l = I(xn ) for n = 1,2, ....

Note that d(xn, x n+p) = dey, xp) holds for all n and all p. Since I(X) is
compact, {x,,} must have a limit point in I(X). Let a be a limit point of {xn}.
Now, let B > O. Pick n > 1 and p such that d(x n, a) 0, and so d(y, I(X)) = o. Thus, y E I(X) = I(X), so that
I(X) = X holds.
If X is not supposed to be compact, then the conclusion is no longer true. A
=
counterexample: Take X IN with den, m) =
In - m I and consider the function
I: IN -+ IN defined by I(n) = n +.1.
Problem 7.12. Show that a metric space X is compact if and only if every con-
tinuous real-valuedfunction on X attains its maximum value.
Solution. Let (X, d) be a metric space. Assume that X is compact and that
I: X -+ lR is a continuous function. By Theorem 7.5, we know that I(X) is a
compact subset oflR, and so (by Theorem 7.4) I(X) is closed and bounded. The
maximum of I(X) is the maximum value of Ion X.
For the converse, assume that every continuous real-valued function on X
attains a maximum value. Clearly, every continuous real-valued function on X
attains also a minimum value.
We shall establish first that X is a complete metric space. Let (X, d) denote
the completion of (X, d) and let X EX. The function I: X ~ lR, defined by
I(x) = d(i, x), satisfies inf{f(x): x EX} = O. So, there exists some Xo E X
satisfying I(xo) = d(i, xo) = O. It follows that i = Xo E X and so X = X.
This means that X is a complete metric space.
Next, we shall show that X is totally bounded. To establish this, assume
by way of contradiction that X is not totally bounded. Then an easy inductive
argument shows that there exist some B > 0 and a sequence {xn} of X such that
d(x", xm) 2: 3B holds for n =1= m. For each n consider the nonempty closed set
and then define the function In: X -:-+ lR by
In(x) = d(x, Cn) = inf{d(x, y): y E Cn}.

Section 7: COMPACfNFSS IN METRIC SPACES 61
So, fn is a bounded function, fn(x) = 0 holds for each x E C n and fn(xn) > O.
Multiplying by a constant cn , we can assume that sup{fn(x): x E X} > n holds
for each n. Now, define the function f: X ~ lR by
f(x) = { (/X), if x E B(x n, e)

if x ~ U~l B(xn, e),
and we claim that f is a continuous function. Clearly, f is continuous at the points

of the balls B(x lI , e). If Xo ~ U~l B(xn, e), note that B(xo, ~) n B(xn, e) =f:. 0
holds for at-most one n (why?). If B(xo, ~) n B(xlI , e) =0 for each n, then
f(x) = 0 for each x in B(xo, ~), and so f is continuous at Xo. Thus, we can
assume that B(xo, ~) n B(xlI , e) =f:. 0 for some n. We distinguish two cases.
CASE I: d(xo, XII) > e.
In this case, there exists some 0 < r < ~ such that B(xo, r) n B(xn, e) = 0.
Clearly, f(x) = 0 holds for each X E B(xo, r), and from this we see that f is
continuous at Xo.
CASE II: d(xo, xn) = e.
Let {Zk} be a sequence of X satisfying Zk --+ Xo; we can assume that Zk
belongs to B(xo, ~) for each k. Note that if Zk ~ B(x lI , e), then f(Zk) = O. On
the other hand, if Zk E B(x n , e), then
Thus, 0 ::::: f(Zk) ::::: Cnd(Zb xo) holds for each k. In view of
lim d(Zb xo)

k-+oo
= 0,
we see that lim f(Zk) = =
0 f(xo) and so f is continuous at Xo in this case too.
To contradict our hypothesis, note that f does not attain a maximum value.
Thus, X must also be totally bounded. By Theorem 7.8, we see that X is a compact
metric
space.
Problem 7.13. This exercise presents a converse of Theorem 7.7. Assume that
(X, d) is a metric space such that every real-valued continuous function on X is
uniformly continuous.
a. Show that X is a complete metric space.
b. Give an example of a noncompact metric space with the above property.
c. If X has a finite number of isolated points (an element a E X is said to
be an isolated point whenever there exists some positive r > 0 such that
B(a, r) n (X \ {a}) = 0), then show that X is a compact metric space.
Solution. Let (X, d) be a metric space such that every continuous real-valued
function on X is uniformly continuous.
x
(a) If E X(the completion of X) is an element that does not belong to X, then
the function f: X -+ R defined by f (x) = l(Lx) , x EX, is a continuous real-
valued function on X that fails to be uniformly continuous (why?), a contradiction.
Hence, X = X holds, which means that X is a complete metric space.
(b) Let X = {I, 2, ... } equipped with the discrete distance d. Then every set is
open and so every real-valued function f on X is continuous. Since d(x, y) < 1
implies x = y (and so f(x) - f(y) = 0), we see that every real-valued function
on X is uniformly continuous. Now, note that X is not a compact metric space.
(c) In view of (a), we need to establish that X is totally bounded. To this
end, assume that X is not totally bounded. Then, there exist some 8 > 0 and
a sequence of elements {xn} of X such that d(xn, xm) > 38 for n =1= m. From
our hypothesis, we can suppose that each Xn is an accumulation point of X. For
each n pick an element Yn such that 0 < d(x n, Yn) < ~ and let rn = d(xn' Yn).
Put
and define the functions fn and f as in the solution of Problem 7.12 (the open
ball B(xn , 8) is now replaced by B(xn, rn». Then f is a continuous function
and satisfies f(Yn) = 0 for each n. Pick Zn E B(xlI' rn) such that f(xn) > n,
and note that
This shows that the continuous function f is not uniformly continuous, contrary
to our hypothesis. Hence, X is totally bounded, as desired.
Problem 7.14. Consider a function f: (X, d) -+ (Y, p) between two metric

spaces. The graph G of f is the subset of X x Y defined by
G = {(x, y) E X X Y: Y = f (x)}.
If(Y, p) is a compact metric space, then show that f is continuous if and only if
G is a closed subset of X x Y, where X x Y is considered to be a metric space
under the distance D«x, Y), (u, v» = d(x, u) + p(y, v); see Problem 7.4. Does
the result hold true if (Y, p) is not assumed to be compact?
Solution. Observe that an arbitrary sequence {(xn, Yn)} of X x Y satisfies

(xn , Yn) -+ (x, y) in X x Y if and only if Xn -+ x and Yn -+ Y both hold.
Section 7: COMPACTNESS IN METRIC SPACES 63
Assume (Y, p) compact and G closed. If I is not continuous, then there exists
a sequence {xnl of X and some e > 0 such that XII -+ X and p(J(xn ), I(x)) ?:
e for all n (why?). Since (Y, p) is compact, by passing to a subsequence, we can
assume that I(Yn) -+ Y holds in Y. Now, observe that (xn,/(x n)) E G holds
for each n and (xn,/(x n)) -+ (x, y) holds in X x Y. Since G is closed, it
follows that (x, y) E G and so y = I(x). This implies
p(J(XII ), I(x)) -+ p(J(x), I(x)) = 0,
which contradicts p(J(xn),/(x)) ?: e for all n. Hence, I is a continuous

function.
If (Y, p) is not compact, then a function with closed graph need not be contin-
uous. For an example, consider the function I: lR -+ lR defined by
I(x) = {6 if x:;6 0;
if x = 0 .
Problem 7.15. A cover {Vi liEf 01 a set X is said to be a pointwise finite cover
whenever each x E X belongs at-most to afinite number olthe Vi.
Show that a metric space is compact if and only if every pointwise finite open
cover 01 the space contains a finite subcover.
Solution. Clearly, if X is compact, then every pointwise finite open cover of

X contains a finite subcover. For the converse, assume that every pointwise finite
open cover of X contains a finite subcover. To establish that the metric space X
is compact, it suffices to show that every sequence in X contains a convergence
subsequence.
Let {xnl be a sequence in X. We can suppose (why?) that the sequence
consists of distinct elements. Suppose by way of contradiction that {xn} has no
convergence subsequence. Then x I is not in the closure of the set {xn: n :;6 I} and
thus, there exists an open ball VI = B(XI, 01) about XI with radius 0 < 01 < 1
and satisfying Xn ¢ VI for all n :;6 1. Also, X2 is not in the closure of the set
{XII: n :;6 2} and thus, there exists an open ball V2 = B(X2,02) about X2 with
radius 0 < 02 < ~ and such that XII ¢ V2 for all n :;6 2. Proceeding inductively,
we see that for each k there exists an open ball Vk = B(Xk, Ok) with radius
ir
o < Ok < satisfying XII ¢ Vk for all n :;6 k.
Since the set F = {XI, X2,"'} contains no convergent subsequences, the set
F must contain all of its closure points. Thus, F is a closed set, and hence, the
set G = X \ F is an open set. Then, the collection C = {G, VI, V2, ... } is an
open cover of X. In fact, the collection C is a pointwise finite open cover of X
because if a point x belongs to an infinite number of sets in C, then X belongs to
an infinite number of the sets VII' However, this would imply that a subsequence
of {xn} converges to the point x. Since the sequence {xn} contains no convergent
subsequences, we infer that C is a pointwise finite open cover.
Therefore, C contains a finite subcover of X, say VI,"" Vm, G. Since G
ur=1
does not intersect {XI, X2, ... }, it follows that {x" X2, ... } s;; Vi. However,
this contradicts the fact Xn f/. Vk for n =1= k. Conclusion: The sequence {xn} must
have a convergent subsequence-and hence, the metric space X is compact.
CHAPTER 2 ______________
TOPOLOGY AND CONTINUITY
8. 1DPOLOGICAL SPACES
Problem 8.1. For any subset A of a topological space show the following:
a. AO = (N)c.
b. aA = A\Ao.
c. (A \ AO)O = 0.
x E A° ¢=> there exists a neighborhood V of x with V ~ A

¢=> there exists a neighborhood V of x with V nA C = 0
¢=> x rj;. A C ¢=> x E (N )c.
(b) Using (a), we see that aA = A n AC = A \ (N)C = A \ AO.

(c) If x E (A \ AO)O, then for some open set V we have
x E V ~ A \ AO ~ A.
This implies x E A \ A° and x E A0 , a contradiction. Hence, (A \ A 0)0 = 0.

Problem 8.2. If A and B are two arbitrary subsets of a topological space, then
show the following:
a. AUB =AUB.
b. (A U BY = A' UB'.
Solution. (a) See Problem 6.1.

(b) Clearly, A ~ B implies A' ~ B', and so A' U B' ~ (A U BY. For the reverse
inclusion, let x E (A U BY. If x rj;. A' UB', then there exist two neighbOrhOod~,
".
65
66 Chapter 2: TOPOLOGY AND CONTINUITY
V and W of x such that

A n (V \ {x}) = B n (W \ {x}) = (25.
Now, note that the neighborhood U = V nW of the point x satisfies

(A U B) n (U \ {x}) = (25,
proving that x rf. (A U B)" a contradiction.
Problem 8.3. If A is an arbitrary subset of a H ausdorjf topological space, then

show that its derived set A' is a closed set.
Solution. Let A be an arbitrary subset of a Hausdorff topological space X. We

shall establish that (A')C is an open set (and this will guarantee that A' is a closed
set). To this end, let x E (A')C, i.e., let x rf. A'. This means that there exists a
neighborhood V of x such that
V n (A \ {xl) = (25.
We claim that V S;; (A')C holds. To see this, let y E V with y '# x. Since X
is a Hausdorff topological space, there exist neighborhoods U and W of y and
x, respectively, such that Un W = (25. Now, note that V n U is a neighborhood
of y with x rf. V n U and so from (*), we see that (V n U) n A = (25. The latter
shows that y rf. A'. Hence, V S;; (A')C holds proving that every point of (A')C is
an interior point, as desired.
Problem 8.4. Let X = JR, and let r be the topology on X defined in Example 8.4.
In other words, A E r if and only iffor each x E A there exist E > 0 and an
at-most countable set B (both depending on x) such that (x -E, x +E) \ B S;; A.
a. Show that r is a topology on X.
b. Verify that 0 E (0, 1).
c. Show that there is no sequence {xn} of(O, 1) with limxn = O.
Solution. (a) Straightforward.

(b) Since for each e > 0 and each countable set B the set (-e, e) \ B is
uncountable, we must have (-e, e) \ B) n (0,1) '# (25. This easily implies that
o E (0, 1).
(c) If {XII} is a sequence of (0, 1), then V = (-1, 1) \ {XIoX2' ••• } is a neigh-
borhood of zero, and Xn rf. V for all n. This shows that no sequence of (0, 1) can
converge to O.
Problem 8.5. If A is a dense sub.set of a topological space, then show that

o S;; A n 0 holds for every open set O. Generalize this conclusion as follows: If
A is open, then A nBS;; A n B for each set B.
Section 8: lOPOLOGICAL SPACES 67
Solution. Let x E 0 and let V be a neighborhood of x. Since 0 is open,

V n 0 is a neighborhood of x, and so the denseness of A implies
v n (A n 0) = (V n q) n A =1= 0,
which means that x E A nO.

For the general case, assume A is an open set and let x E A n B. If V is a
neighborhood of x, then V n A is also a neighborhood of x. Since x E B, it
follows that V n (A n B) = (V n A) n B =1= 0. This shows that x E A n B, and
hence, A n B ~ A n B.
Problem 8.6. If {Oi liel is an open cover for a topological space X, then show
that a subset A of X is closed if and only if A n Oi is closed in Oi for each i E I
(where Oi is considered equipped with the relative topology).
Solution. If A is closed, then clearly A n Oi is closed in Oi for each i. For

the converse, assume that A n Oi is closed in Oi for each i. Put
Vi =Oi \ AnOi =Oi \ A,
and note that-by our hypothesis--each Vi is open in Oi. Since each 0i is an

open subset of X, it follows that each Vi is likewise an open subset of X. Now,
note that
A
C
=X \ A = (U Oi) \ A = U(Oi \ A) =U Vi
iel iel iel
is an open subset of X, and so A is a closed set.
Problem 8.7. If(X, r) is a Haus40rfftopological space, then show thefollow-

ing:
a. Every finite subset of X is closed.
b. Every sequence of X converges to at-most one point.
Solution. (a) Let A = {x} be a one-point set. If y ¢. A, then (since X is a

Hausdorff space) there exists a neighborhood V of y with x ¢ V, and so V ~ A C •
Thus, N is open, and hence A, is closed. Now, observe that every finite set is a
finite union of one-point sets.
(b) If x =1= y, then there exist neighborhoods v\ and Vy of x and y respec-
tively, such that V:t n Vy = 0. Now, a sequence of X cannot converge to x and
y at the same time simply because its terms cannot be eventually in both V:t and
Vy •
Problem 8.8. For afunction I: (X, T) ~ (Y, TI) show thelollowing:

a. II T is the discrete topoLogy, then I is continuous.
b. II T is the indiscrete topoLogy and TI is a Hausdorff topoLogy, then I is
continuous if and onLy if I is a constant function.
Solution. (a) Note that every subset of X is open. Thus, I-I (A) is an open set
for every subset A of Y, and so I is continuous.
(b) Recall that the indiscrete topology is the topology T = ((,25, X}. If I is a
constant function, then I-I(A) is either (,25 or X, and so I is continuous. For
the converse, let I be a continuous function. If for some x, y E X we have
I(x) =1= I(y), then there exists a neighborhood V of I(x) such that I(y) ¢. V.
Now note that I-I (V) is neither equal to (,25 nor equal to X, and so I-I (V) is
not open, a contradiction. Thus, I must be a constant function.
Problem 8.9. Let I and g be two continuous lunctions from (X, T) into a H aus-
dorff topoLogicaL space (Y, TI). Assume that there exists a dense subset A 01 X
such that I(x) = g(x)lor all x E A. Show that I(x) = g(x) hoLds lor all x EX.
Solution. Suppose that for some x E X we have I(x) =1= g(x). Pick a neigh-
borhood V of I(x) and another W of g(x) such that V n W = (,25. Since
I-I (V) n g-I (W) is a neighborhood of x and A is dense in X, there exists some
y E 1- 1(V) n g-I(W) n A. Now, note that I(y) = g(y) E V n W = (,25 must
hold, which is absurd. Thus, I (x) = g(x) holds for each x EX.
Problem 8.10. Let I: (X, T) ~ (Y, Tl) be a function. Show that I is continuous
if and onLy if 1-1 (BO) ~ [f-l (B)JD hoLds lor every subset B olY.
Solution. Repeat the solution of Problem 6.4.
Problem 8.11. II I: (X, T) ~ (Y, TI) and g: (Y, Tl) ~ (2, T2) are continuous
lunctions, show that their composition g 0 I: (X, T) ~ (2, T2) is aLso continuous.
Solution. Use the identity (g 0 f)-I (V) = 1-1 (g-I(V»). (See Problem 1.8.)
Problem 8.12. Let X be a topoLogicaL space, Let a E X, and Let Na denote the
collection 01 aLL neighborhoods at a. The oscillation 01 a function I: X ~ IR at
the point a is the extended non-negative real number
Wt(a) = inC { sup I/(x) - l(y)1 }.

VeNa x,yeV
Section 8: 1OPOLOGICAL SPACES 69
Establish the following properties regarding the oscillation:

a. The function f is continuous at a if and only if (V j(a) = O.
b. If X is an open interval oflR and f: X ~ lR is a monotone function, then
I
(Vj(a) = lim f(x) - lim f(x)
x~a+ x-+a-
I..
Solution. (a) Assume that f is continuous at a. Fix E > O. Then there exists
some W E Na (i.e., some neighborhood W of a) such that x E W implies
If(a) - f(x)1 < E. SO, if x, YEW, then
If(x) - f(y)1 ::: If(x) - f(a)1 + If(a) - f(y)1 < E + E ::::: 2E,
and thus
0::: (Vj(a)::: sup If(x) - f(y)1 ::: 2E

x,yeW
for each E > O. This implies (Vj(a) = O.

For the converse, assume (Vj(a) = O. Let E > O. Then from the definition
of the oscillation, we see that there exists some neighborhood V of a such that
SUPx,yeV If(x) - f(y)1 < E, In particular, we have If(x) - f(a)1 < E for all
x E V, and this shows that f is continuous at a.
(b) We can assume that f is an increasing function. Note that we can consider
neighborhoods of a of the form (c, d) with a E (c, d). Consider first a neigh-
borhood (c, d) of a and assume that x, y E (c, d) satisfy x < a < y. Since f is
increasing, it follows that 0 ::: limHa+ f(t) - limHa- f(t) ::: fey) - f(x), and
from this, we infer that
lim f(t) - lim f(t) :5 (Vj(a).

I-+a+ I-+a-
On the other hand, if E > 0 is given, then there exists some 8 > 0 such that the
open interval J = (a - 8, a + 8) satisfies J S;; X and
(Vj(a)::: sup If(x) - f(y)1 < [lim f(t) - lim f(t)]

x,yeJ t .... a+ I .... a-
+ E.
This implies (Vj(a) ::: limHa+ f(t) -limHa - f(t), and so
(Vj(a) = lim
t-+a+
f(t) - lim f(t)
I-+a-
holds true.
Problem 8.13. Show that a finite union of nowhere dense sets is again a nowhere
dense set. Is this statement true for a countable union of nowhere dense sets?
70 Chapter 2: 1OPOLOGY AND CONTINUTIY
Solution. Let A and B be two nowhere dense sets. Using the identity So = sc-c
(see Problem 8.1), we have
(A U B t = (A U B r-c= (A -c n B -c r S;; (A
c
-c- n B -c- r
= A- c- c U B- c- c = (At U (Bt = (2\U(2\ = (2\.
An easy induction argument can now complete the proof.
The countable union of nowhere dense sets need not be nowhere dense. An
example: Take X = JR, and let En = {r n}, where {rl, r2,"'} is an enumeration
of the rational numbers. Clearly, each En is nowhere dense, while U~I En =
{rl' r2, ... } is not nowhere dense.
Problem 8.14. Show that the boundary of an open or closed set is nowhere
dense.
Problem 8.15. Let f: (X, !") -+ JR, and let D be the set of all points of X where
f is discontinuous. If DC is dense in X, then show that D is a meager set.
r
Solution. From DC = X, it follows that DO = (Dc = (2\. Now, the proof can
be completed by observing that D is an Fa-set (Theorem 8.10).
Problem 8.16. Show that there is no function f: JR -+ JR having the irrational

numbers as the set of its discontinuities.
Solution. Let I denote the set of all irrational numbers of R If I is the set of
discontinuities ofa function f: JR --+ JR, then (by Theorem 8.10) I is an Fa-set.
However, this is impossible by Problem 6.6.
Problem 8.17. Show that every closed subset of a metric space is a G.-set and
every open set is an Fa-set.
Solution. Let A be a nonempty closed subset of a metric space X. Then the

function f: X --+ JR, defined by
f(x) = d(x, A) = inf{d(x, y): YEA},

is continuous (see the proof of Lemma lOA) and satisfies
'00 00
A = f-I({O}) = f-I(n(-~, ~)) = nf-I(-~, ~»).

n=1 n=1
Section 8: WPOLOGICAL SPACES 71
(See the discussion at the end of Section 6 of the text.) Thus, A is a G a-set. By
Theorem 8.9 every open set is an Fer-set.
Problem 8.18. Let B be a collection of ope,! sets in a topological space (X, 7:).
Iffor each x in an arbitrary open set V there exists some B E B with x E B £; V,
then B is called a basefor 7:. In general, a collection B of subsets of a nonempty
set X is said to be a base if
1. UBEB B = X, and
2. for every pair A, B E B and x E A n B, there exists some C E B with
xEC£;AnB.
Show that if B is a base for a set X, then the collection
7: = {V £; X: V x E V there exists B E B with x E B £; V}
is a topology on X having B as a base.
Solution. Obviously, B £; 7: holds. Clearly, (/J E r, and from condition (1)

it follows that X E r. Also, it should be clear that r is closed under arbitrary
unions.
Now, let V, W E r and x E V n W. Choose two sets A, B E B with
x E A £; V and x E B £; W. By condition (2), there exists some C E B with
x E C £; A n B £; V n W, that is, V n W E r. Thus, 7: is a topology.
The verification that B is a base for r is straightforward.
Problem 8.19. Let (X, r) be a topological space, and let B be a base for the
topology r (see the preceding exercise for the definition). Show that there exists a
dense subset A of X such that card A ::: card B.
Solution. If B E Band B =1= (/J, then fix some XB E B and consider the set
A = {XB: B E B \ {(/J)}. We claim that:
1. A is dense in X, and
2. card A ::: card B.
To see (1) let V be a nonempty open set. If x E V, then there exists some
B E B with x E B £; V. It follows that XB E V, and so V n A =1= (/J. This shows
that A is dense in X.
For (2) note that the function f: B \ {(/J} --7 A, defined by feB) = XB, is
e
onto. By the Axiom of Choice there exists a subset of B such that enf-I({xD
e
consists precisely of one point for each x E A. Then f: --7 A is one-to-one
and onto, proving that card A = card e ::: card B.
72 Chapter 2: IDPOLOGY AND CONTINUITY
Problem S.20. Let I: X -+ Y beajunction.II!' isatopologyonX, then the quo-

tienttopology!'I determined by I on Y isdefinedbY!'I = {O £;; Y: 1- 1(0) E !'}.
a. Show that!' I is indeed a topology on Y and that I: (X, !') -+ (Y,!' I) is
continuous.
b. II g: (y, !'I) -+ (Z, !'I) is ajunction, then show that the compositionjunction
go I: (X,!') -+ (Z, !'I) is continuous if and only if g is continuous.
c. Assume that I: X -+ Y is onto and that !'* is a topology on Y such that
I: (X, !') -+ (Y, !'*) is an open mapping (i.e., it carries open sets 01 X onto
open sets oly) and continuous. Show that!'* = !'I'
Solution. a. (1) Since I-I«(/;) = (/; E !' and I-I(y) = X E !', we see that
(/;, Y E !'I'
(2) If V, W E !'I, then the identity I-I(V n W) = I-I (V) n I-I(W) implies
that V n WE!' I'
(3) If {Vi: i E I} is a family of !'I,then in view of the identity 1-I(UVi ) =
U 1-I(Vi), we see that U Vi E !'I'
b. Assume gol is continuous. If V is an open subset of Z,then I-I (g-I(V») =
(g 0 f)-I(V) E !' shows that g-I(V) E !'I' That is, g is continuous.
c. Since I is continuous, it is easy to see that !'* £;; !'I holds. On the other hand,
let V E !'I' Then I-I (V) E !', and moreover, since I is an open mapping and
onto, we have V = I(J-I(V») E !'* (see Problem 1.7). That is, !'I £;; !'* also
holds, and so !'I = !'*.
Problem S.21. This exercise presents an example 01 a compact set whose closure
is not compact. Start by considering the interval [0, 1] with the topology!' gener-
=
ated by the metric d(x, y) Ix - YI. It should be clear that ([0, 1], !') is a compact
topological space. Next, put X = [0, 1] U IN = [0, 1] U {2, 3,4, ... }, and define
!'*=!'U{[O,I]UA: A£;;IN}.
a. Show that!'* is a non-Hausdorff topology on X and that!'* induces!' on

[0, 1].
b. Show that (X, !'*) is not a compact topological space.
c. Show that [0,1] is a compact subset ol(X, !'*).
d. Show that [0,1] is dense in X (and hence, its closure is not compact.
e. Why doesn't this contradict Theorem 8.12(1)?
Solution. a. (1) Clearly, (/;, X E !'*.

(2) Let V, W E !'*. Then we have the following cases:
CASE I. V, WE!'. In this case, V n WE!' £;; !'*.
CASE II. V E !' and W f/. !' (and vice versa). Note that V n W = V E !' £;; !'*.
Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 73
CASE ID. V rf; rand W rf; r. In this case, we have V n W = [0, 1] U A for
some A ~ N. That is, V n W E r*.
(3) Let {Vi: i E l} be a family of r*. If Vi E r holds for each i, then clearly
U Vi E r ~ r* holds. On the other hand, if some Vi is of the form [0, 1] U A,
then U Vi is of the same type, and hence, it' belongs to r*.
Thus, r* is a topology on X that induces r on [0, 1],
b. The cover X = U:2 ([0, 1] U {nl) cannot be reduced to a finite cover.
c. Since ([0, 1], r) is a compact topological space and r* induces r on [0, 1],
it follows that [0, 1] is a compact subset of X.
d. If x E X \ [0, 1], then every neighborhood V of x is of the form V =
[0, 1] U A for some A ~ N. Thus, V n [0, 1] = [0, 1] ::f. C/J holds for every
neighborhood V of x. Therefore, [0, 1]= X holds.
e. This does not contradict Theorem 8.12(1) because (X, r*) is not a Hausdorff
topological space.
Problem 8.22. A topological space (X, r) is said to be connected if a subset of

X that is simultaneously closed and open (called a e10pen set) is either empty or
else equal to X.
a. Show that (X, r) is connected if and only if the only continuous functions
from (X, r) into to, I} (with the discrete topology) are the constant ones.
b. Let f: (X, r) ~ (Y, r*) be onto and continuous. If (X, r) is connected,
then show that (Y, r*) is also connected.
Solution. (a) If f: X --+ to, I} is a nonconstant continuous function, then

f-1({0)) is a nonempty clopen set which is different from X, and so X is not
connected.
For the converse, assume that every continuous function from X into to, I} is
constant. If A is a clopen subset of X different from C/J and X, then the function
f: X --+ to, I}, defined by f(x) = 1 if x E A and f(x) = 0 if x rf; A,
is a nonconstant continuous function, a contradiction. Thus, X is a connected
topological space.
(b) Let A be a clopen subset of Y. By the continuity of f, the set f-1(A) is
a clopen subset of X. Since X is connected, f-1(A) = C/J or f-1(A) = X. Also,
since f is onto, f(J-I(A») = A holds (Problem 1.7). Thus, A = C/J or A = Y,
proving that Y is a connected topological space.
9. CONTINUOUS REAL-VALUED FUNCTIONS
Problem 9.1. If u, v, and ware vectors in a vector lattice, then establish the
following identities:
a. uVv+U!\v=u+v;
b. u- vV w =
(u - v) /\ (u - w);
c. u - v /\ w = (u - v) V (u - w);
d. a(u /\ v) =
(au) /\ (av) if a 2: 0;
e. lu - vi = u V v - U /\ v;
f. uV v = !(u + v + lu - vI);
g. U /\ V = !(u + v -Iu - vI).
Solution. We use the identities (a), (b), and (d) in Section 9 of the text.
(a) Replace w by -(u + v) in u /\ v + w = (u + w) /\ (v + w) to get
u /\ v - (u + v) = (-v) /\ (-u) = -u V v.
(b) u - v V w =
u + (-v) /\ (-w) = (u - v) /\ (u - w).
(c) u - V /\ W =
U + (-v) V (-w) = (u - v) V (u - w).
(d) If a 2: 0, then
a(u /\ v) = a[-(-u) V (-v)] = -a[(-u) V (-v)]

= -(-au) V (-av) = (au) /\ (av).
(e) Using (a), we see that
u vv- U /\ v = v v + [u V v - (u + v)] = 2(u v v) - (u + v)

U
= (2u) v (2v) - (u + v) = (u - v) v (v - u) = lu - vi.

(f) Using (e) and (a), we get
u +v + lu - vi = (u v v + u /\ v) + (u V v - U /\ v) = 2(u v v).
(g) As in (f), we get
u+v - lu - vi = u v v + u /\ V - (u V v - U /\ v) = 2(u /\ v).
Problem 9.2. Ifu and v are elements in a vector lattice, then show that:
a. lu + vi v lu - vi = lui + lvi, and
b. lu + vi /\ lu - vi = Ilul-Ivll·
lu + vi v lu - vi = [(u + v) V (-u - v)] v [(u - v) V (-u + v)]

= [(u + v) v (-u + v)] V [(-u - v) V (u - v)]
= [uv(-u)+v]v[(-u)vu-v]
= [u V (-u)] + [v V (-v)] = lui + Ivl.
Section 9: CONTINUOUS REAL-VALUED F1JNCI10NS 75
(b) Using the distributive law, we see that
lu + vi" lu - vi = [(u + v) v (-u - V)]" [(u - v) v (-u + v)]

= [(U+V) " (U-V)] V [(-U-V) " (U-V)] V [(U+V) " (-U+V)] V···
... V [(-U - V)" (-U + v)]
= [U + v" (-v)] V [(-U) ". U - v] V [u" (-U) + v] V [V" (-v) - U]
= (U -Ivl) V (-U -Ivl) V (v -luI) V (-v -luI)
= [U V (-U) -Ivi] V [v V (-v) -luJ] = (Iul-Ivl) V (lvi-luI)
= Ilul-Ivll·
Problem 9.3.
holds.
Show that lui" Ivl = ° holds if and only if lu + vi = lu - vi
Solution. If lui" Ivl = 0, then using parts (a) and (b) and part (e) of Problem 9.1,
we get
lu + vi" lu - vi = Ilul-Ivil = lui V lvi-lui" Ivl = lui V Ivl

= lui + Ivl -lui" Ivl = lui + Ivl = lu + vi V lu - vi.
This easily implies that III + vi = lu - vi holds.

For the converse, assume that lu + vi = lu - vi. Then by parts (a) and (b) of
Problem 9.2, we have
lui + Ivl = Ilul- Ivll = lui V lvi-lui" Ivl

= (lui + Ivl -lui" Ivl) -lui" Ivl = lui + Ivl- 2(lul" lvI),
from which it follows that lui" Ivl = 0.
Problem 9.4. Show that the vector space consisting of all polynomials (with
real coefficients) on JR is not a function space. Prove a similar result for the
vector space of all real-valued differentiable functions on JR.
Solution. If p is the polynomial defined by p(x) = x, then Ipl(x) = Ip(x)1 =

Ixl holds. Clearly, Ipl is not differentiable (and hence, it is not a polynomial
either).
Problem 9.5. Let X be a topological space. Consider the collection L of all

real-valued functions on X defined by
L = {f E JRx: 3 (fnl £:; C(X) such that limfll(x) = f(x) V x E Xl.
Show that Lis afunction space.
Solution. Clearly, L is a vector space. Now, let f, gEL. Choose two se-
quences {fn} and {gn} of C(X) with limfn(x) = f(x) and limgn(x) :;:: g(x)
for all x. Then fn v gn E C(X) for each n and
lim fn v gn(x) = lim ~[tn(X) + gn(x) + Ifn(x) - gll(x)l]

1
= 2[f(x) + g(x) + If(x) - g(x)l] =f v g(x),
so that f v gEL. Similarly, f /\ gEL, so that L is a function space.
Problem 9.6. Let L be a vector space of real-valued functions defined on a set

X. Iffor every function f E L the function If I (defined by Ifl(x) = If(x)1 for
each x E X) belongs to L, then show that Lis afunction space.
Solution. Use the identities

1 1
f vg = 2([ + g + If - gl) and f /\ g = 2([ + g -If - gl).
Problem 9.7. Consider each rational number written in the form ~,where n >
0, and m and n are integers without any commonfactors other than ±l. Clearly,
such a representation is unique. Now, define f: 1R -+ 1R by f(x) = 0 if x is
irrational and f(x) = ~ if x = ~ as above. Show that f is continuous at every
irrational number and discontinuous at every rational numbel:
Solution. The proof will be based upon the following property: Let {rnl be a
bounded sequence of distinct rational numbers. If rn = ;: (where kll > 0, and
mn and k n do not have common factors), then lim k n = 00 .
To see this, pick some number M > 0 such that Irnl ::: M for each n, and so
Imn I ::: M k n. Now, if for some C > 0 we have Ik nI ::: C for infinitely many
n, then Imnl ::: MC must also hold for the same infinitely many n. However,
this contradicts the fact that there is a finite number of rational numbers ~ with
Iml ::: MC and Inl < C.
Now, let x be an irrational number. If {x n } is a sequence of irrational numbers
with Xn --+ x, then 0 = f(x n) --+ 0 = f(x). Thus, if f is not continuous
at x, then there exists a sequence {rn} of rational numbers with r n --+ x and
lim fern) :f:: O. Since x is irrational, we can assume rn :f:: rm whenever n :f:: m.
Write rn = ~:, and note that fern) = t fr 0 implies kn fr 00, a contradiction.
Therefore, f is continuous at every irrational number.
Now, let r be a rational number. Choose a sequence {rn} of distinct rational
numbers with rn = t --+ r. Nc:>w, note that lim fern) = lim t= 0 :f:: fer)
holds, which shows that f is not continuous at r. That is, f is discontinuous at
every rational number.
Problem 9.8. Let f: [a, b] -+ JRbe increasing,i.e., x < y implies f(x) ,::: fey).
Show that the set of points where f is discontinuous is at-most countable.
Solution. Let f: [a, b] ---?- JR be increasing, and let D be the set of dis-
continuities of f. For each XED choose a rational number rx such that
limet.d(t) < rx < limltxf(t). Since x, y E D with x < y implies
rx < limf(t) < limf(t) < r y ,

It x Ity
it follows that rx '# ry whenever x '# y. Thus, x f-+ rx is a one-to-one function

from D into the set of rational numbers, and so D is at-most countable.
Problem 9.9. Give an example of a strictly increasing function f: [0, 1] -+ JR

which is continuous at every irrational number and discontinuous at every rational
number.
Solution. For each t E [0, 1], let fl: [0, 1] -+ [0, 1] be a strictly increasing
function which is continuous everywhere except at x = t. For instance, for
° < t ,::: 1 let
!,(x) = {0.5X ~f O,:::X <t,

I x Ift,:::x,:::l,
= °
and fo(x) 0.5 + O.5x if < x ,::: 1 and fo(O) 0. =
If {rt. r2, ... } is an enumeration of the rational numbers of [0, 1], then define
the function f: [0, 1] -+ [0, 1] by
1
=L
00
f(x) 2 n frn(x),
n=1
and note that f satisfies the desired properties.
Problem 9.10. Recall that a junction f: (X, i) -+ (Y, il) is called an open
mapping if f(V) is open whenever V is open. Prove that if f: JR -+ JR is a
continuous open mapping, then f is a strictly monotone junction-:-and hence, a
homeomorphism.
Solution. Let (a, b) be a finite open interval of JR. Since f attains a maximum
value on [a, b] and f(a, b») is an open set, it is easy to see that the extrema of
f on [a, b] take place at the end points. In particular, this implies f(a) '# feb).
(If f(a) = feb), then f(a, b») must be a one-point set, contradicting the fact
that f is an open mapping.) Next, we claim that f is strictly monotone on
(a, b). To see this, assume f(a) < feb), and a < x < y < b. Then note first that
78 Chapter 2: WPOLOGY AND CONTINUITY
f(a) < f(x) < feb) must hold. Indeed,if f(x):s f(a) holds, then f attainsits
minimum on [a, b1 at some interior point. Similarly, if f(x) 2: feb) holds, then
f attains its maximum value on [a, b1 at some interior point. However, (since f
is an open mapping) both cases are impossible, and so f (a) < f (x) < f (b) holds.
By the same arguments, f(x) < fey) < feb). Thus, f is strictly increasing on
(a, b). Similarly, if f(a) > feb) holds, then f is strictly decreasing on (a, b).
Now, assume that f is strictly increasing on (0,1), and let x < y. Choose
some n with (0, 1) f; (-n, n) and x, y E (-n, n). Since f is strictly monotone
on (-n, n), and strictly increasing on (0, 1), it is easy to see that f must be
strictly increasing on (-n, n). Thus, f(x) < fey) holds, and this shows that f
is strictly increasing on IR. (We remark that the function f need not be onto.
However, the mapping f: IR ---+ f(IR) is a homeomorphism.)
Problem 9.11. Let X be a non empty set, and for any two functions f, g E IRx
let
If(x) - g(x)1
d(f,g ) = xeX
sup
1 + If(x) - g(x)1

a. (IR x, d) is a metric space.
b. A sequence {fn} f; IRx satisfies d(jn, f) -+ 0 for some f E IR x if and
only if {fn} converges uniformly to f.
Solution. (a) Clearly, d(j, g) 2: 0 for all f, g E IRx and d(j, g) = 0 if and only
if f = g. Moreover, it should be clear that d (j, g) = d (g, f) for all f, g E IR x .
What needs verification is the triangle inequality. To do this, we need the following
two properties:
1. O:s x :s y impliesl~x :s I~Y , and

2. l~;';:Y:S l~x + It for all x, y 2: O.
Property (1) follows from the fact that the function f (t) = l~t (t 2: 0) is strictly
increasing on [0, (0); notice that f'(t) = (1 + t)-2 > 0 for each t > -1. For (2)
fix x, y 2: 0, and note that
(x+ y)(1 + x)(1 + y) = x(1 + x)(1 + y) + y(1 + x)(l + y)

:s [x(I +x)(1 + y) +xy(1 + y)] + [y(1 +x)(1 + y) +xy(1 +x)]
= x(I + y)(1 +x + y) + y(l +x)(1 +x + y).
Dividing across by (1 +x)(1 + y)(1 +x + y), the validity of (2) can be established.
Now, let I, g, h E R X and x EX. From
I/(x) - g(x)1 ::: I/(x) - h(x)1 + Ih(x) - g(x)1
and (1) and (2), we get
I/(x) - g(x)1
::: I/(x) - h(x)1
~~------~~~~~~
+ Ih(x) - g(x)1
1 + I/(x) - g(x)1 1 + I/(x) - h(x)1 + Ih(x) - g(x)1
::: I/(x) - h(x)1 + Ih(x) - g(x)1
1 + I/(x) - h(x)1 1 + Ih(x) - g(x)1
::: d(j, h) + d(h, g),
for all x EX. This implies
I/(x) - g(x)1
d(j, g) = ~~k 1 + I/(x) _ g(x)1 ::: d(j, h) + d(h, g).
(b) Let Un} s.;:; RX. Assume first that Un} converges uniformly to some function
X
I E R , and let E > O. So, there exists no such that I/,,(x) - l(x)1 < E holds for
all n 2: no and all x EX, and hence, I~·j~?;)!n~ll < Lfn(~)-=-LWJ_<-,--"E-=f=or=-a=l=-l_ _ _ _ __
n 2: no and all x EX. It follows that
lin (X) - l(x)1

d( I" I) = sup < E
In, xeX 1 + I/n(x) - l(x)1 -
for all n 2: no. This shows that d(jn, f) ~ O.

For the converse, assume d(jn, f) ~ 0, and let E > O. Then there exists some
no such that
I" I) _ lin (x) - l(x)1 E

d( In, -sup <--
.teX 1 + I/,,(x) - l(x)1 1+E
"11
lor a n 2: no, an d hence, I +l!.(xl-[(.tll
1[. (xl-[(xli <E" 11 d 11
\+E lor a n 2: no an a x E
X . Th'IS
implies I/,,(x) - l(x)1 < E for all n 2: no and all x E X, which means that Un}
converges uniformly to I.
Problem 9.12. Let I, 11, /Z, . " be real-valued lunctions defined on a com-
pact metric space (X, d) such that Xn ~ x in X implies In (x n) ~ I (x) in R II I is
continuous, then show that the sequence olfunctions {/,,} converges uniformly to I.
Solution. Assume that the functions I, II, /Z, '" satisfy the stated proper-
ties and that the function I: X ---7 R is continuous. Also, assume by way of
80 Chapter 2: IDPOLOGY AND CONTINUTIY
contradiction that the sequence Un} does not converge uniformly to I. Then an
easy argument shows (how?) that there exist e > 0, a subsequence {gn} of Un},
and a sequence {xn} of X such that
Since X is compact, the sequence {xn} has a convergent subsequence in X,

say Xkn ~ x. By the continuity of I, we see that I (Xk.) ~ I (x). Also, from
our hypothesis, it follows that gkn (Xk.) ~ I (x), and so
contrary to (*). Therefore, the sequence Un} converges uniformly to I.

Problem 9.13. For a sequence {In} olreal-valuedfunctions defined on a topo-
logical space X that converges uniformly to a real-valuedfunction I on X establish
the lollowing.
a. II Xn -+ x and I is continuous at x. then In(xn) -+ I(x).
b. II each In is continuous at some point Xo EX. then I is also continuous at
the point Xo and
lim lim In(x)

X-+Xo n-+oo
= n-+oo
lim lim In(x) = I(xo).
X-+Xo
Solution. (a) Assume I is continuous at x, Xn -+ x and let s > O. Choose

some k with Iln(Y) - l(y)1 < s for all n > k and all y EX. By the continuity
of I at x, there exists some m > k with I/(x n) - l(x)1 < s for all n > m.
Thus,
holds for all n > m, so that lim In (Xn) = I(x).

(b) Assume that each In is continuous at Xo E X and let E > O. Since Un}
converges uniformly to I on X, there exists some k satisfying Ilk (X) - I (X) I < E
for all x E X. Now, the continuity of !k at Xo guarantees the existence of a
neighborhood V of Xo such that Ilk(X) - Ik(Xo)1 < E for all x E V. Then
I/(x) - I(xo) I ::: I/(x) - Ik('~)1 + l!k(x) - Ik(Xo)1 + Ilk(XO) - l(xo)1

< E +E +E = 3E
Section 9: CONTINUOUS REAL-VALUED FUNCITONS 81
holds for all x E V, which shows that f is continuous at Xo. For the equality, note
that
lim lim fn(x)

x-rxo n-+-oo
= '\"-+Xo
lim f(x) = f(xo),
while
lim lim fn(x)

n-+OO X-+Xo
= lim fn(xo)
n-+OO
= f(xo).
Problem 9.14. Let fn: [0,1] -+ 1R be defined by fn(x) = xn for x E [0, 1].
Show that Un} converges pointwise and find its limit function. Is the convergence
uniform?
Solution. Clearly,
fnCx) -+ f(x) = {~: if 0 ::: x < 1;

if x = 1.
Since f is not continuous, the convergence cannot be uniform; see Theorem 9.2.
Problem 9.15. Let g: [0, 1] -+ 1R be a continuous function with g(1) = O. Show

that the sequence of functions Un} defined by fn (x) = x" g (x) for x E [0, 1],
converges uniformly to the constant zero function.
Solution. Let e > O. Choose some 0 < 8 < 1 with Ig(x)1 < e whenever
8 < x ::: 1. Now, pick some M > 0 with Ig(x)1 ::: M for all x E [0, 1], and
then select some k with M 8n < e whenever n > k. Thus, for each n > k we
have Ix" g(x)1 ::: M 8n < e for 0 ::: x ::: 8 and Ixn g(x)1 S Ig(x)1 < e for all
8 < x ::: 1. That is, the sequence Un} converges uniformly to the constant zero
function.
Problem 9.16. Let Un} be a sequence of continuous real-valued functions

defined on [a, b), and let {an} and {b n} be two sequences of [a, b) such that
lim an = a and lim bn = b. If U,,} converges uniformly to f on [a, b), then show
that
lim
ll-HXJ ian
bn
fn(x)dx =
lb
a
f(x)dx.
Solution. Let e > O. Pick some k such that for all n > k we have:
1. an - a < e and b - bn < e; and

2. Ifn(x) - f(x)1 < e for all x E [a, b].
82 Chapter 2: IDPOLOGY AND CONTINUI1Y
Also, since f is continuous (Theorem 9.2), there exists some M > 0 satisfying
If(x)1 :s M for all x E [a, b]. Thus,
n b
Il: f,,(X)dX - l f(X)dXI
rn ibn tn [b I
= IJan fn(x)dx - Jan f(x)dx - Ja f(x)dx - Jbnf(x)dx
b lan + lb If(x)1 dx
:s
la
Ifn(x) - f(x)1 dx +
a
If(x)1 dx
bn
:s s(b - a) + M(a n - a) + M(b - bn) < s(2M +b - a)
holds for all n > k, and our conclusion follows.
Problem 9.17. Let {f,,} be a sequence of continuous real-valued functions on a

metric space X such that {fn} converges uniformly to some function f on every
compact subset of X. Show that f is a continuous function.
Solution. Let Xn ---+ x in X. Put K = {Xl,X2, ... } U {x}, and note that K
is a compact set-every open cover of K can be reduced to a finite cover. Since
{fn} is a sequence of continuous functions that converges uniformly to f on K,
it follows from Theorem 9.2 that f is continuous on K. Since Xn ---+ x holds
in K, we get f(x ll ) ---+ f(x). That is, f is a continuous function.
Problem 9.18. Let {f,,} and {gn} be two uniformly bounded sequences of real-
valued functions on a set X. If both. {fn} and {gn} converge uniformly on X, then
show that {fngn} also converges uniformly on X.
Solution. Assume that {fn} and {gn} converge uniformly to f and g, respec-
tively. Let s > O. Choose some k with Ifn(x)- f(x)1 < s and Ign(x)-g(x)1 < s
for all n > k and all x EX. Also, pick some M > 0 so that Ifn(x)1 :s M and
Ign(x)1 :s M hold for all n and all x. Now, note that
Ifn (x)gn (x) - f(x)g(x)1

:s Ifn(X)I·lglI(x) - g(x)1 + Ig(x)I·l!n(x) - f(x)1 < 2Ms
holds for all n > k and all x EX.
Problem 9.19. Suppose that {fn} is a sequence of monotone real-valued func-

tions defined on [a. b] and not necessarily all increasing or decreasing. Show
that if {fn} converges pointwise to a continuous lunction I on [a, b], then {f,,}
converges uniformly to I on [a, b].
Solution. Let 8 > O. Since I is unifonnly continuous (Theorem 7.7), there

exists some 0 > 0 so that I/(x) - ICy)1 < e holds whenever Ix - yl < o. Fix
a finite number of points a = Xo < XI < ... < Xk = b with Xi - Xi-I < 0 for
1 ::: i ::: k, and then pick some m such that Iln(Xi) - I(Xi)1 < 8 holds for each
o ::: i ::: k and all n > m.
Now, let n > m. Assume that In is decreasing. If X E [a, b], then Xi-I :::
x ::: Xi holds for some 1 ::: i ::: k, and so
!In(X) - 1/I(Xi)! = In(x) - 1/I(Xi) ::: 1/I(Xi-l) - In(Xi)

= (fn(XI-I) - I(Xi-I)] + [f(Xi-l) - I(Xi)] + [/(x/) f,,(Xi)]
< 8 +8 +8 < 38.
A similar inequality holds true if In is increasing. Therefore,
!In(X) - I(x)! ::: !f,,(x) - f,,(Xi)! + !//I(Xi) - I(x/)! + !/(x/) - I(x)!

< 38 + 8 + 8 = 58
holds for all x E [a, b] and all n > m. That is, {fn} converges unifonnly to I.
Problem 9.20. Let X be a topological space and let {fn} be a sequence 01 real-
valued continuouslunctions defined on X. Suppose that there is afunction I: X -+
lR such that I (x) = lim In (x) holds lor all X EX. Show that I is continuous at
a point a if and only iflor each E > 0 and each m there exist a neighborhood V
ola and some k > m such that I/(x) - Ik(X)1 < E holds lor all x E V.
Solution. Assume that I is continuous at some point a. Let' 8 > 0 and an

integer m be given. Pick a neighborhood U of a such that I/(x) - l(a)1 < 8
holds for all X E U. Since lim In(a) = I(a) holds, there exists an integer r > m
such that I/(a) - In(a)1 < 8 holds for all n > r. Fix any integer k > r and
note that k > m. Since Ik is a continuous function, there exists a neighborhood
W of a such that I/k(a) - Ik(X)1 < 8 holds for all x E W. Now, note that if
x E V = U n W, then
For the converse, assume that I satisfies the stated condition at the point a
and let 8 > O. Since I(a) = lim I/I(a) holds, there exists an integer m such
84 Chapter 2: 1OPOLOGY AND CONTINUITY
that If(a) - fn(a)1 < 8 holds for all n > m. By the hypothesis, there exist a
neighborhood V of a and an integer k > m such that If(x) - fk(X)1 < 8 holds
for all x E V. By the continuity of fko there exists another neighborhood U of
a such that Ifk(a) - fk(X)1 < 8 holds for all x E U. Now, note that x E Un V
implies
which shows that f is continuous at the point a.
Problem 9.21. Let Un} be a uniformly bounded sequence of continuous real-

valued functions on a closed interval [a, b]. Show that the sequence offunctions
{¢n} defined by ¢n(x) = J:
fn(t)dt for each x E [a, b], contains a uniformly
convergent subsequence on [a, b].
Solution. Since the sequence Un} is uniformly bounded, there is some M > 0
such that Ifn(x)1 < M holds for all x E [a, b] and all n. Clearly,
holds for all x E [a, b] and all n. So, the sequence {¢n} is uniformly bounded
and we claim that it is an equicontinuous sequence.
To see this, let 8 > 0 and put 8 =
81M. Now, note that x, y E [a, b] and
Ix - yl < 8 imply
11 fn(t)dt -l fn(t)dt 1
x Y
l¢n(X) - ¢n(y)1 =
: : 11 lfn(t)1 dt 1: : 11 Mdt 1= Mix - yl <

Y Y
8.
Thus, the set A = {¢I, ¢2" ... } is equicontinuous. If A denotes the (uniform)
closure of A in C[a, b], then A is bounded, closed, and equicontinuous (why?).
By the Ascoli-Arzela theorem (Theorem 9.10), the set A is a compact set. Since
{¢n} is a sequence of A, it follows that {¢n} has a subsequence that converges
uniformly on [a, b].
Problem 9.22. For each n let fn: JR -+ JR be a monotone (either increasing or

decreasing) function. If there exists a dense subset A of R such that lim fn (x)
exists in JR for each x E A, then show that lim fn(x) exists in JR at-most for all
but countably many x.
Solution. Assume that the functions fn and the dense subset A of JR satisfy
the properties of the problem. Also, assume at the beginning that all but a finite
number of the ft, are increasing functions.
Define the function f: JR ---')0 JR by
f(x) = lim sup fn(x), x E JR.
Note that f(x) is a real number for each x E JR. Indeed, if x E JR, then there
exist a, b E A with a < x < b, and so fn(a) .::s fn(x) .::s ft,(b) holds for all
sufficiently large n. Consequently,
-00 < lim fn(a) = lim sup fn(a)

.::s lim sup fn(x) = f(x)
.::s lim sup ft,(b) = lim fn(b) < 00
Clearly, f (x) = lim ft, (x) holds for each x E A. Next, note that f is an increas-
ing function. Indeed, if x < y holds, then from fn(x) .::s f,,(y) for all sufficiently
large n, we see that f(x) = lim sup f,,(x) .::s lim sup fn(Y) = f(y). By Prob-
lem 9.8, we know that f has at-most countably many discontii1uities in every
closed subinterval of JR. Hence, f has at-most countably many discontinuities
(why?). Now, we claim that
lim fn(x) = f(x)
holds at every point of continuity of f. To see this, let xo be a point of continuity

of f and let 8 > O. ·Pick some 8 > 0 such that Xo - 8 < x < Xo + 8 implies
If(x) - f(xo)1 < 8, and then choose a, b E A with xo - 8 < a < Xo < b <
xo + 8. Also, pick some no such that for each n ::: no the function fn is increasing
and satisfies
\fn(b) - f(b)\ < 8 and \f,,(a) - f(a)\ < 8.
Now, note that for n ::: no, we have
f(xo) - f,,(xo) .::s f(xo) - fn(a)

= [f(xo) - f(a») + [f(a) - fn(a») < 8 + 8 = 28
and
fn(."(.o) - f(xo) .::s ft,(b) - f(xo)

= [f,,(b) - f(b») + [f(b) - f(xo») < 8 + 8 = 28.
Thus, Ifn(xo) - f(xo)1 < 28 holds for all n 2: no, proving that lim fn(xo) =
f(xo).
For the general case, assume that there are infinitely many increasing and in-
finitely many decreasing fn. Splitthe sequence {fn} into two subsequences {gn}
and {h n } such that each gn is increasing and each h n is decreasing. Put
g(x) = lim sup gn(x) and h(x) = liminfhn(x) = -limsup[-hn(x)],

and note that g(a) = h(a) holds for each a E A. By the above conclusion, g and
h are continuous except possibly at the points of an at-most countable subset C
of JR, and for each point x ¢. C we have
limgn(x) = g(x) and limhAx) = h(x).

Now, let c ¢. C and fix 8 > O. Pick some 8 > 0 such that Ix - cl < 8 implies
Ig(x) - g(c)1 < 8 and Ih(x) - h(c)1 < 8. Pick a E A with la - cl < 8 and note
=
that from g(a) h(a), it follows that
Ig(c) - h(c)1 ::: Ig(c) - g(a)1 + Ih(a) - h(c)1 < 8 +8 = 28.
Since 8 > 0 is arbitrary, we see that g(c) = h(c) holds for each c ¢. C. This
implies (how?) that lim fn(c) exists in JR for each c ¢. C.
Problem 9.23. Consider a continuous function f: [0, 00) -+ JR. For each n
define the continuous function fn: [0, 00) -+ JR by fn(x) =
f(x n ). Show that the
set of continuous functions {fl, /Z, ... }is equicontinuous at x =
1 if and only if
f is a constant function.
Solution. Let f E C[O,oo), let fn: [0, 00) ~ JR be defined by fn(x) =

f(x n), and let E = {fl, /Z, ... }. If f is a constant function, then it should be
clear that the set E is equicontinuous at x =
1.
For the converse, assume that the set E is equicontinuous at x = 1. Fix a > 0
and let 8 > O. The equicontinuity of E at x = 1 guarantees the existence of some
0<8 < 1 such that Ix -11 < 8 implies Ifn(x) - fn(1)1 < 8 for each n. From
lim.y/a = 1 (why?), we see that there exists some no such that l.y/a - 11 < 8
holds for each n 2: no. Thus, if n 2: no, then we have
Since 8 > 0 is arbitrary, it follows that f(a) = f(1) holds for each a > O. By
continuity, we see that f(a) = f(1) for each a 2: 0, and so f is a constant
function.
Problem 9.24. Let (X, d) be a compact metric space and let A be an equicon-
tinuous subset ofC(X). Show that A is uniformly equicontinuous, i.e., show that
for each E > 0 there exists some 0 > 0 such that x, y E X and d(x, y) < 0 imply
If(x) - f(y)1 < E for all f EA.
Solution. Let (X, d) be a compact metric space, let A be an equicontinuous

subset of C (X), and let 8 > O. For each x E X there exists (by the equicontinuity
of A) some Ox > 0 such that d(x, y) < Ox implies If(x) - f(y)1 < 8 for all
f E A. From X = UtEX B(x,~) and the compactness of X, we see that there
exist x\, ... ,Xn E X such that X = U7=1 B(Xi, 8:;,).
Let 0 = 1min{oxl' ... , OxJ > 0 and let x, y E X satisfy d(x, y) < O. Now,
pick some 1 ::: i ::: n with d(x, Xi) < ~ and observe that If(x) - f(Xi)1 < 8
for all f E A. In addition, from
ox; ox;
d(y,xi):5d(y,x)+d(x,xi) < T+T = ox;>
we see that If (y) - f (Xi) I < 8 holds for all f E A. Therefore, from the above,
if d(x, y) < 0, then
\f(x) - f(Y)\ ::: \f(x) - f(Xi)\ + \f(Xi) - f(y)\ < 8 + 8 = 28

holds for all f E A. That is, A is a uniformly equicontinuous subset of C(X).
Problem 9.25. Let X be a connected topological space (see Problem 8.22 of

Section 8for the definition) and let A be an equicontinuous subset ofC(X). lffor
some Xo E X, the set of real numbers (f(xo): f E A} is bounded, then show that
(f (x): f E A} is also bounded for each x EX.
Solution. Let X be a connected topological space, let A be an equicontinuous

subset of C (X), and let Xo E X be a point such that the collection of real numbers
(f(xo): f E A} is bounded. Consider the set
E = {x E X: The set (f(x): f E A} is bounded}.
Since Xo E E, we see that E is nonempty. We claim that E is both open and
closed. If this is the case, then by the connectedness of X we must have E = X,
and the desired conclusion follows.
We shall show first that E is a closed set. To this end, let y E E. By the
equicontinuity of A, there exists a neighborhood V of y such that
If(x) - f(y)1 < 1
holds for all X EV and all f E A. From y E E, we see that V n E =j:. C/J. Fix
some z E V n E, and then pick some M > 0 such that If(z)1 :5 M holds for
each lEA. In particular, we have
I!(y) I ~ I/(y) - l(z)1 + I/(z)1 < 1+M
for all lEA. This means that y E E, and so E = E, i.e., E is a closed set.
Next, we shall establish that E is an open set. To this end, let y E E. Pick
some C > 0 such that I/(y)1 ~ C holds for each lEA. By the equicontinuity
of A, there exists a neighborhood W of y such that I/(x) - l(y)1 < 1 holds
for each x E W and all I E A. In particular, if x E W, then
I/(x)1 ~ I!(x) - l(y)1 + I/(y)1 < 1+C
holds for all I E A, and so x E E. That is, W £; E holds, which shows that y
is an interior point of E. Therefore, E is also an open set.
Problem 9.26. Let {In} be an equicontinuous sequence in C(X), where X is not

necessarily compact. IIlor some function f: X -+ R we have lim In (x) = I (x)
lor each x E X, then show that f E C(X).
Solution. Let x E X and let 8 > O. Since {In} is an equicontinuous sequence,

there exists a neighborhood V of the point x such that IIn (y) - In (x) I < 8 holds
for all n and each y E V.
Now, let y E V. Pick some k with Ifk(X) - l(x)1 < 8 and I/k(y) - l(y)1 < 8,
and note that
I/(x) - l(y)1 ~ I/(x) - h(x)1 + Ih(x) - Ik(y)1 + I/k(y) - l(y)1 < 38.
That is, I is continuous at the arbitrary point x.
Problem 9.27. Let X be a compact topological space, and let {In} be an equicon-
tinuous sequence oIC(X). Assume that there exists some I E C(X) and some
dense subset A 01 X such that lim In (x) = I (x) holds lor each x E A. Then show
that {In} converges uniformly to I.
Solution. Let 8 > O. By the equicontinuity of {In} and the continuity of I, for
each x EX, there exists some neighborhood .Vx of x such that
1. I/n(Y) - In(x)1 < 8 holds for all y E Vx and all n; and

2. I/(y) - l(x)1 < 8 holds fot all y E Vx '
By the compactness of X, there exist XI, ... ,Xk E X such that X = U~=I V Xi '
Now, let Y E Yr;' Choose some x E A n Vx;, and then pick some mi with
Ifn(x) - f(x)1 < 8 for all n > mi. Clearly, If(x) - f(y)1 < 28. Thus,
\fn(Y) - f(Y)\
::s \fn(Y) - f,,(xi)\+\fn(Xi)- fn(x)\ + \fn(x) - f(x)\+\f(x) - f(y)\
< 58
holds for all Y E Yr; and all n > mi.

Finally, put m = max(mi: 1 ::s i ::s k}, and note that If,,(y) - f(y)1 < 58 for
all Y E X and all n > m.
Problem 9.28. Show that for any fixed integer n > 1 the set of functions f in
e[O, 1] such that there is some x E [0, 1 - ~ ] for which
If(x + h) - f(x)1 ::s nh whenever 0< h < ~,
is nowhere dense in e[O, 1] (with the uniform metric).

Use the above conclusion and Baire's theorem to prove that there exists a con-
tinuous real-valuedfunction defined on [0, 1] that is not differentiable at any point
of[O, 1].
Solution. Let D(j, g) = IIf - glloo = sup{lf(x) - g(x)l: x E [0,1]}. For

n 2: 2 define
An = {J E e[O, 1]: 3 x E [0, 1 -~] with \f(x + h) - f(x)\ ::s nh

whenever ° < h < ~ }.
We claim that:
1. Each An is closed; and
2. Each An is nowhere dense in e[O, 1] (i.e., (An)O = (25).
To see that each An is closed, let (fk} ~ An satisfy lim D(jk, f) = (Le., (fk}
converges uniformly to f on [0, 1]). For each k choose some Xk E [0, 1 - 1]
°
°
with Ifk(Xk + h) - fk(Xk)1 ::s nh for all < h < ~. Since [0, 1- ~] is compact,
there exists a subsequence of (xd that converges to some x E [0, 1 - 1].
n
We
n
can assume that limxk = x. By Problem 9.13, lim fk(Xk + h) = f(x + h) and
lim fk(Xk) = f(x), and so If(x + h) - f(x)1 ::s nh holds for all
Thus, f E An, and hence, An is a closed subset of e[O, 1].
< h < ~. °
Now, let f E An and let 8 > 0. Consider the function g E C[O, 1] whose
graph is shown in Figure 2.1. Note that for each x E [0, 1) we have Ig(x + h) -
g(x)1 = 3nh for all sufficiently small h > 0. Put fl = f + g, and note that
90 Chapter 2: IDPOLOGY AND CONTINUIIY
y=g(x)
FIGURE 2.1. The Construction of a Nowhere Differentiable Function
D(f, fl) = IIglioo = e. On the other hand, if x [0,1) is fixed, then for all
sufficiently small h > we have ° E
nh < 2nh = 3nh - nh ::: Ig(x + h) - g(x)I-lf(x + h) - f(x)1

::: Ig(x + h) - g(x) - [f(x) - f(x + h)]1 = Ifl(x + /z) - fl(x)l·
Thus, fl ¢. An, and so B(f, 2e) g; All for all e > 0. This shows that (An)O = 0.
Now, for each n ::: 2 let
Bn = {t E C[O, 1]: 3 x E [~, 1] with If(x - h) - f(x)1 ::: nh
whenever ° < h < ~ }.
By the same arguments, each Bn is closed and nowhere dense. Consequently,

from Baire's Theorem 6.17, we have
In particular, note that every f E C[O, 1] \ (U:2 An) U (U:2 Bn) does not
have anyone-sided derivative at any point of [0, 1].
Problem 9.29. Establish the following result regarding differentiability and uni-
form convergence. Let {fn} be a sequence of differentiable real-valuedfunctions
defined on a bounded open interval (a, b) such that:
a. for some Xo E (a, b) the sequence of real numbers (fn(XO)} converges in JR.
and
Section 9: CONTINUOUS REAl.rVALUED FUNCTIONS 91
b. the sequence of derivatives U~} converges uniformly to a function

g: (a, b) -+ IR..
Then the sequence Un} converges uniformly to a function f: (a, b) -+ lR which is
differentiable at Xo and satisfies 1'(xo) = g(xo).
Solution. First, we shall show that {fn} is a uniformly Cauchy sequence. To this
end, let E > 0 and pick some M > 0 such that Ix - xol ::: M for each x E (a, b).
Next, choose some k such that
\!~(x) - f~(x)1 < E for all m,11 ::: k and all x E (a, b) (*)
and
\!n(XO) - fm(xo)1 < E for all m, n ::: k.
Using the Mean Value Theorem, (*) and (**), we see that for each x E (a, b) and
each pair n, m ::: k there exists some t E (a, b) such that
If,,(x) - fm(x)1 ::: 1 [fn(X) fm(x)] - [1;,(xo) - fm(xo)] 1 + If,,(xo) - fm(Xo)1

= \!,;(t) - /';,(t)I· Ix - xol + \!n(Xo) - fm(Xo)1
::: ME+E=(l+M)E.
This shows that {f,,} is a uniformly Cauchy sequence, and hence, {f,,} converges
uniformly to a function f: (a, b) -+ IR..
Next, for each n we consider the continuous function ¢n: (a, b) -+ lR defined
by ¢,,(x) = Mt~=!n(.to)
.1: .to
if x i= Xo and ¢n(XO) = /'n'(XO). Using the Mean Value
Theorem and (*), we see that for each x E (a, b) there exists some Cx.E (a, b)
such that
l¢n(X) - ¢m(X)1 = ILf,,(x)-[m(x),~-=-~~,(xo)-[m(.to)ll = \!~(c.t) - f~(c.t)1 < E,
for all n, m ::: k. This shows that {¢n} is a uniformly Cauchy sequence, and hence,
it converges uniformly to the function ¢: (a, b) -+ lRdefined by ¢(x) = [(X)=[(x o)
x .to
if x i= Xo and ¢(xo) = g(xo).
Finally, from Problem 9.13, we obtain
g(xo) = lim f~(xo) = lim lim ¢n(x) = lim lim ¢n(x)

n-'1--CO tJ-+oo X~Xo X-+Xo 11-+00
' l' fn(x) - f(x) l' f(x) - f(xo)

I
=lm 1m =lm .
.<-+.to n-+oo X - Xo x-+.to X - Xo
This shows that f is differentiable at xo and that 1'(xo) = g(xo).

92 Chapter 2: roPOLOGY AND CONTINUITY
10. SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS
Problem 10.1. Let (X, d) be a metric space and let A be a nonempty subset of
X. The distance function of A is the function d(·, A): X -+ 1R defined by
d(x, A) = inf{d(x, a): a E A}.
Show that d(x, A) = 0 if and only if x EA.
Solution. Clearly,d(x, A) 2: Oforeachx EX. Assumethatx E AandletE > O.

Then B(x, E)nA i= 95, and so there exists some yEA such thatd(x, y) < E. From
the definition of the distance function, we see that 0 ::s d(x, A) ::s d(x, y) < E.
Since E > 0 is arbitrary, this implies d(x, A) = O.
For the converse, assume that d(x, A) = O. If E > 0, then it follows from
d(x, A) = inf{d(x, a): a E A} <, E that there exists some a E A with d(x, a) < E.
Hence, B(x, E) n A i= 95 for each E > 0, and this implies that x EA.
Problem 10.2. Let (X, d) be, a metric space, let A alld B be two non empty
disjoint closed sets and consider the junction f: X -+ [0, 1] defined by
f(x) = d(x,~\~:lx,B)' Show that:
a. f is a continuous junction,
b. f-I({O}) = A alld f- I({I}) = B, and
c. if d(A, B) = inf{d(a, b): a E A and b E B} > 0, then f is uniformly
continuous.
Solution. Let C be an arbitrary nonempty subset of X. We shall show first that

the function X 1-+ d (x, C) is uniformly continuous. To see this, fix x, Y EX.
Choosing some c E C, we see that
d(x, C) ::s d(x, c) ::s d(x, y) + d(y, c) ::s d(x, y) + d(y, C),
or d (x, C) - d (y, C) ::s d (x, y). Exchanging the roles of x and y in the last
inequality, we get d(y, C) - d(x, C) ::s d(x, y). Therefore,
Id(x, C) - d(y, C)I ::s d(x, y),
and the uniform continuity of x 1-+ d(x, C) follows.

(a) Observe that since A and B are disjoint closed sets, it follows from the
Problem 10.1 that d (x, A) + d (x, B) > 0 for each x 'E X. This, in connection
with the (uniform) continuity of the functions d(-, A) and d(·, B), guarantees that
f is a continuous function.
(b) Note that f(x) = 0 if and only if d(x, A) = o. Now, by Problem 10.1, we
have d(x, A) = 0 if and only if x E A = A. In other words, we have f(x) = 0 if
and only if x EA. This means f-I({O}) = A.
Section 10: SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS 93
Similarly, notice that f(x) = d(x.~~~~l.t.B) = 1 if and only if d(x, B) = 0. As

above, this shows that f- I ({1}) = B.
°
(c) Fix some E > such that d(u, v) ::: E for all u E A and v E B. If a E A
and b E B are arbitrary, then for each z E X 'we have
E ::= dCa, b) ::= d(z, a) + d(z, b) ::= d(z, A) + d(z, B).

Now, if x, y EX, then the inequalities
If(x) -
d(x, A)
I = Id(x, A) + d(x, B) -
dey, A) I'
fey) dey, A) + dey, B)
+ dey, B)]d(x, A) - [d(x, A) + d(x, B)]d(y, A)1
l[d(y, A)
= [d(x, A) + d(x, B)][d(y, A) + dey, B)]
l[d(x, A) - dey, A)]d(x, B) + [dey, B) - d(x, B)]d(x, A)1
= [d(x, A) + d(x, B)][d(y, A) + d(y, B)]
[d(x, B) + d(x, A)]d(x, y)
::= ---------------------------
[d(x, A) + d(x, B)][d(y, A) + dey, B)]
d(x, y)
::=
E
guarantee that f is uniformly continuous.
Problem 10.3. Let A and B be two nonempty subsets of a metric space X such
that A n B = =
A n B 0. Show that there exist two open disjoint set U and V
such that A S;; U and B S;; V.
Solution. From the solution of Problem 10.2, we know that for each nonempty
subset C of X the function x 1-+ d(x, C) is (uniformly) continuous. Now, consider
the function f: X -+ 1R defined by
f(x) = d(x, A) - d(x, B).
By the above, f is a continuous function. From An B = 0 and Problem 10.1, we

see that f(x) =
-d(x, B) < °
holds for each x E A. Similarly, f(x) > holds
for each x E B. Consequently, the two disjoint open sets U = f-I(-oo,O))
°
and V = f-I(O, 00)) satisfy AS;; U and B S;; V.
Problem 10.4. Show that a closed set of a normal space is itself a normal space.
Solution. Let C be a closed subset of a normal space X. We consider C equipped

with the topology induced by X. Now, assume that A and B are two disjoint closed
subsets of C. Since C is closed, it is easy to see that A and B are also closed subsets
of X. Pick two open subsets VI and WI of X satisfying A 5; VI, B 5; WI and

VI n WI = 0. Now, if V = C n VI and W = C n WI, then V and W are two
disjoint open subsets of C satisfying A 5; V and B f; W. This shows that C
equipped with the relative topology is a normal space.
Problem 10.5. Let X be a normal space and let A and B be two disjoint closed
subsets 01 X. Show that there exist open sets V and W such that A 5; V, B f; W
and V n W = 0.
Solution. Assume that A and B are two disjoint closed subsets of a normal space
X. Pick two disjoint open sets V and WI satisfying A 5; V and B 5; WI. We
claim that V n WI = 0. Indeed, if x E V n WI, then on one hand WI is a
neighborhood of x, and on the other hand, x belongs to the closure of V, which
imply WI n V i= 0, a contradiction.
Now, since V n B = 0 and X is normal, there exist two disjoint open sets VI
and W such that V 5; VI and B f; W. As before, VI n W = 0, and clearly the
open sets V and W satisfy the desired properties.
Alternatively: If a continuous function I: X -+ [0,1] satisfies A 5; I-I({O})
and B 5; I-IC{1}), then the open sets V = 1- 1 ([0,4)) and W = I-I((~, 1])
satisfy A 5; V, B 5; W, and V n W = 0.
Problem 10.6. Show that a topological space is normal if and only iflor each
closed set A and each open set V with A 5; V, there exists an open set W such
that A 5; W 5; W 5; V.
Solution. Let X be a topological space. Assume first that X is a normal space

and let a closed set A and an open set V satisfy A 5; V. Then A n V c = 0
and V C is a closed set. Pick two disjoint open sets Wand U such that A 5; W
and V C 5; U. In particular, W n U = 0. This implies W n v c = 0, and so
W 5; V.
For the converse, assume that the property is satisfied and let A and B be two
disjoint nonempty closed sets. If V = B C , then V is an open set such that A 5; V.
By our hypothesis, there exists an open set W such that· A 5; W 5; W 5; V = B C •
If U = we,then U is an open set disjoint from W and satisfies B 5; U. This
shows that X is a normal space.
Problem 10.7. For a closed subset A 01 a normal topological space X, establish

the lollowing:
a. There exists a continuous function I: X -+ [0,1] satisfying I-IC{O}) = A
if and only if A is a G s-set.
Section 10: SEPARATION PROPERTIFS OF CONTINUOUS FUNCTIONS 95
b. If A is a Go-set and B is another closed set satisfying A n B = 0, then

there exists a continuous function g: X -+ [0, 1] such that g-I ((O)) = A
and g(b) = 1 for each bE B.
Solution. Let A be a closed subset of a nonnal topological space X.

(a) If there exists a continuous function f: X -+ [0, 1] such that f- I ((O)) = A,
then the identity
00 00
A = f-I((O}) = f-l(n[O, ~)) = nf-I([O, ~))

n=1 ,,=1
shows that A is a Go-set.

For the converse, assume that A is aGo-set. Pick a sequence (VII} of open sets
such that A = n:1 VII' Since A n V,~ = 0, it follows from Uryson's lemmathat
there exists a continuous function f,,: X -+ [0,1] satisfying fll(a) = 0 for each
a E A and f,,(x) = 1 for all x E V,;, Now, consider the function f: X -+ [0, 1]
defined by
00
f(x) = L t. f,,(x).
,,=1
From the Weierstrass' M-test (Theorem 9.5) and Theorem 9.2, it is easy to see that
f is a continuous function, and we claim that f- I ((O}) = A. Clearly, f(x) = 0
for each x E A. Now, assume f(x) = O. Then f,,(x) = 0 for all n, and so (in view
E E
of f,,(v) = 1 for each v V,~) we have x V" for each n, i.e., x En::"=1
V" = A.
Therefore, f-I({Ol) = A.
(b) Assume now that A is a closed Go-set and B is another closed set such that
A n B = O. SO, there exist two disjoint open set V and W such that A ~ V and
B ~ W. This implies that the sequence {V,,} introduced in part (a) can be assumed
to satisfy V" ~ V for each n. In particular, each f" satisfies f,,(b) = 1 for each
b E B. Now, it is easy to see that the continuous function f constructed in the
preceding part satisfies the desired property.
Problem 10.8. Show that a compact subset A of a Hausdorff locally compact

topological space is a Go-set if and only if there exists a continuous junction
f: X -+ [0, 1] sllch that A = f-I({O}).
Solution. If A = f-I({O}), then-as in the solution of part (a) of the preceding

problem-the set A is aGo-set. For the converse, assume that A = n:1
VII' where
each V" is an open set. By Theorem 10.8, for each n there exists a continuous
function 1,,: X -7 [0, 1] such that f,,(x) = 1 for each x E A and fn(x) = 0 for
each x ¢. Vn' Now, as in the the solution of part (a) of the preceding problem,
notice that the function f: X -7 [0, 1] defined by f(x) = 2:::1 :In fn(x) satisfies
the desired properties. -
Problem 10.9. A topological space X is said to be perfectly normal iffor evelY

pair of disjoint closed sets A and B, there is a continuous function f: X -7 [0, 1]
such that A = f- I ({O}) and B = f- I({l}). (Part (b) of Problem 10.2 shows that
evelY metric space is pelfectly normal.)
Show that a Hausdorff normal topological space is pelfectly normal if and only
if evelY closed set is a Go-set.
Solution. Let X be a Hausdorff normal topological space. Assume first that X

is perfectly normal and let A be a proper closed subset of X. If a E X satisfies
a ¢. A, then A n {a} = 0 and {a} is a closed set. So, there exists a continuous
function f: X -7 [0, 1] with f- I ({O}) = A. This implies (as in the solution of
Problem 10.7), that A is a Go-set.
For the converse, assume that every closed set is a G s-set. Let A and B be
two closed disjoint sets. By Problem 10.7 there exist two continuous functions
g, h: X -7 [0,1] such that:
i. g-I ({O}) = A and g(b) = 1 for each bE B, and

ii. 11- 1 ({O})= B and h(a) = 1 for each a E A.
Now, let f = ig + i(1 - h), and note that f: X -7 [0,1], A = f-I({O}) and
B = f- I ({l)).
Problem 10.10. Show that a nonempty connected normal space is either a sin-
gleton or uncountable.
Solution. Let X be a (nonempty) Hausdorff connected normal space. If X is

not a singleton, then there exist a, b E X with a i= b. Since X is Hausdorff,
singletons are closed sets, and we have {a} n {b} = 0. Now, pick a continuous
function f: X -7 [0, 1] such that f(a) = 0 and feb) = 1. The assumption that
X is connected guarantees (according to Problem 6.11(g» that f(X) is an interval
and so f(X) = [0, 1]. This easily implies that X is uncountable-in fact, it has
cardinality greater than or equal to the cardinality of the continuum.
Problem 10.11. Let X be a normal space, let C be a closed subset of X, and let
I be a nonempty interval-with the possibility I = (-00, 00). If f: C -7 I is a
continuous function, then show that f has a continuous extension to all of X with
values in I.
Section 10: SEPARATION PROPERTIES OF CONTINUOUS FUNCTIONS 97
Solution. Assume that C is a closed subset of a normal space X and that f: X -+

I is a continuous function, where I is an interval. The interval I must be one of the
following type: (a, b), [a, b], [a, b), (a, b]. So, we shall establish the continuous
extension of f by steps.
STEP I: I is either of the form [a, b) or (b, a].
In this case, there exists a homeomorphism h: I -+ [0, 1). For instance, if
-00 < a < b < 00, then hex) = ~::::~ is a homeomorphism between (a, b] and
[0, 1). Likewise, if a E JR, then hex) = I~:~x defines a homeomorphism between
(-00, a] and [0, 1).
Fix a homeomorphism h: I -+ [0, 1) and consider the continuous (composition)
function h 0 f: C -+ [0, 1) S; [0, 1]. By Tietze's extension theorem, there exists a
continuous function g: X -+ [0,1] satisfying g(x) = h(j(x») for all x E C. The
continuity of g guarantees that the set A = g -I ({ 1}) is a closed subset of X. Also,
since for each x E C, we have g(x) = h(j(x») E [0, 1), we see that en A = 0.
By Uryson's lemma, there exists a continuous function 8: X -+ [0, 1] such that
°
8(a) = for each a E A and 8(e) = 1 for each e E C.
Now, consider the function ¢: X -+ [0, 1] defined by ¢(x) = 8(x)g(x). We
claimthat¢(X) S; [0, 1). To see this, let x E X. Ifx E A, then¢(x) = 8(x)g(x) =
0·1 = 0, and if x ¢. A, then 0.::: g(x) < 1 and so ¢(x) = 8(x)g(x) < 1 is also true.
Next, define the function j: X -+ I by
= (h- I ¢)(x) = h- I (8(x)g(x»).

j(x) 0
If x E C, then 8(x)g(x) = g(x) = h(j(x»), and hence,
j(x) = h-I(h(j(x»)) = f(x).

This shows that j: X -+ I is a continuous extension of f to all of X.
STEP II: I = [a, b] with -00 < a < b < 00.
The function h: [a, b] -+ [0, 1], defined by hex) = ~::::~, is a homeomorphism.

By Tietze's extension theorem, there exists a continuous function g: X -+ [0, 1]
satisfying g(x) = (h 0 f)(x) for each x E C. Then the continuous function
j = h- I 0 g: X -+ [a, b] satisfies j(e) = fee) for each e E C.
STEP ill: Assume I = (a, b) with -00 ::: a < b ::: 00.
In this case, there exists a homeomorphism h: (a, b) -+ (-1, 1). (For instance,
for -00 < a < b < 00 let hex) = 2~':::) - 1 and if (a, b) = (-00, (0) take
hex) = ~ arctan x.) Now, consider the continuous function h 0 f: C -+ (-1, 1) S;
[ -1, 1] and note that by STEP II there exists a continuous function g: X -+ [- 1, 1]
satisfying gee) = (h 0 f)(e) for each e E C.
Next, let B = =
g-I({-I, 1l). Then B is closed and B n C 0. By Uryson's
lemma, there exists a continuous function e: X -7 [0, 1] satisfying e(b) =°
for
e =
each b E B and (c) 1 for each c E C. As before, define the continuous function
¢: X -7 [-1,1] by ¢(x) = e(x)g(x). Then it is easy to see that ¢(X) f; (-1, 1)
and thefunction j: X -7 (a, b), defined by j = h- I o¢, is a continuous extension
of f.
11. TIlE STONE-WEIERSTRASS APPROXIMATION THEOREM
Problem 11.1. Let X be a compact topological space. For a subset L ofC(X),

let L denote the uniform closure of Lin C(X). Show the following:
a. If L is afunction space, then so is r.
b. If L is an algebra, then so is L.
Solution. Let f, gEL. Pick two sequences {[,,} and {gn} of L that converge
°
uniformly to f and g, respectively. Also, pick some M > so that IIfn II 00 ~ M
and IIgn 1100 ~ M hold for all n.
(a) The inequality II["I-Ifll ~ I[,,-fl shows that {Ifni} converges uniformly
to If I. Since Ifn I E L for each n, it follows that I fiE r.
This implies that L
is a function space.
(b) From the inequalities
IIfngn - fglloo ~ IIglioo 'lIfn -!IIoo + IIfnlloo ·lIgn - glloo

~ M(lIfn - !lloo + IIgII - glloo),
it follows that the sequence {fngn} of L converges uniformly to fg. Thus,

f gEL, and so L is an algebra.
Problem 11.2. Let L be the collection of all continuous piecewise linearfunc-

tions defined on [0, 1]. That is, f E L if and only iff E C[O, 1] and there exists a
°
finite number of points = Xo < XI < ... < Xn = 1 (depending on f) sl!ch that
f is linear on each interval [Xm-I, xm]. Show that Lis afunction space but not
an algebra. Moreover, show that L is dense in C[O, 1] with respect to the uniform
metric.
Solution. The verification that L is a function space is routine. Since the func-
tion f(x) = X satisfies f, ELand f2 ¢. L, it follows that L is not an algebra
of functions.
To see that L is dense, let f E C [0, 1] and let B > 0. By the uniform continuity
°
°
of f, there exists some 8> such that Ix -'YI < 8 implies I/(x) - f(y)1 < B.
Let = Xo < XI < ... < Xn = 1 be a finite collection of points with Xi-Xi-I < 8
Section 11: THE STONE-WEIERSTRASS APPROXIMATION THEOREM 99
for each 1 ::s i ::s n. The function g, defined on each subinterval [Xi-I, x;] by
get) = f(xo-
I 1
) + !(x;)-!(X;-I\t
X;-Xi_1
- X"_ )
I 1 ,
belongs to L and satisfies II f - g II < 8. (Xl
An alternate way of proving the denseness of L is the following: Note that 1 E

Land L separates the points of [0, 1] (why?). Thus, by the Stone-Weierstrass
theorem, L is dense in C[O, 1].
Problem 11.3. Show that a continuous function f: (0, 1) --+ 1R is the uniform
limit of a sequence of polynomials on (0, 1) if and only if it admits a continuous
extension to [0, 1].
Solution. Let f: (0, 1) -+ 1R be a continuous function. Assume first that f

has a continuous extension to [0, I]-which we denote by /. Then, by Cor-
ollary 11.6, the function !
is the uniform limit of a sequence of polynomials
on [0, 1], and consequently f: (0, 1) -+ 1R is likewise the uniform limit of a
sequence of polynomials on (0, 1).
For the converse, assume that there exists a sequence of polynomials {PII 1 that
converges uniformly to f on (0, 1). Let 8 > 0 and then pick some no such that
IplI(x) - f(x)1 < 8 holds for all x E (0, 1) and all n :::: no. From the triangle
inequality, we see that
\PII(X) - Pm(X)\ ::s \PII(X) - f(x)\ + \Pili (x) - f(x)\ < 8 +8 = 28

for all x E (0, 1) and all n :::: no. By continuity, we infer that
IplI(x) - Pm(x)1 ::s 28

holds for all x E [0, 1] and all n :::: no. The above show that {PII 1 is a Cauchy
sequence of qo, 1], and so (by Theorem 9.3) the sequence {Pill converges in
qo, 1], say to g E qo, 1]. It follows that f(x) = g(x) for all x E (0, 1), and so
g is a continuous extension of f: (0, 1) -+ 1R to [0, 1].
Problem 11.4. Iff isacontinuollsfimctionon [0, 1] such thatfol xllf(x)dx = 0

for n = 0, 1, ... , then show that f(x) = 0 for all x E [0, 1].
Solution. By Corollary 11.6, there exists a sequence of polynomials {PII} that

converges uniformly to f. It easily follows that {PII f} also converges uniformly
to f2, and by our hypothesis we see that fol
PIl(x)f(x) dx = 0 holds for each n.
Now, invoke Problem 9.16 to infer that fol
f2(x) dx = lim Pn(x)f(x) dx = O. fol
The latter easily implies that f(x) = 0 holds for each x E [0, 1].
100 Chapter 2: lOPOLOGY AND CONTINUITY
Problem 11.S. Show that the algebra generated by the set {l, x 2 } is dense in
C[O, 1] but fails to be dense in C[-I, 1].
Solution. Since the function f(x) = x 2 separates the points of [0, 1], the
algebra generated by {l, x 2 } also separates the points of [0,1]. Thus, by the
Stone-Weierstrass, this algebra must be dense in C[O, 1].
To see that the algebra generated by {l, x 2 } is not dense in C[-I, 1], note that
for every f in the closure of this algebra, we have f ( -1) = f (1). Thus, this
algebra is not dense in C [ -1, 1].
Problem 11.6. Let us say that a polynomial is odd (resp. even) whenever it
does not contain any monomial of even (resp. odd) degree.
Showthatacontinuousfunction f: [0,1] -7 1R vanishes at zero {i.e., f(O) = 0)
if and only if it is the uniform limit of a sequence of odd polynomials on [0, 1].
Solution. If f is the unifonn limit of a sequence of odd polynomials, then it

should be clear that f vanishes at zero. For the converse, assume that fEe [0, 1]
°
satisfies f(O) = and let e > 0. Define the function g: [-1, 1] -7 1R by
g(x) = { f(x), ifO<x<l·

- .- ,
- f(-x), if -1 ~ x < 0,
and note that g E C[-I, 1]. By the Stone-Weierstrass theorem there exists a
polynomial p such that Ig(x) - p(x)1 < e for each x E [-1, 1].
Next, write p = q + r , where q is the odd polynomial consisting of the sum of
all odd tenns of p and r is the even polynomial consisting of the sum of all even
tenns (including the constant tenn) of p. In particular, note that q( -x) = -q(x)
°
and r(-x) = r(x) hold for each x. Thus, if ~ x ~ 1, then
If(x) - q(x) - r(x)1 = Ig(x) - p(x)1 < e,
and g( -x) = - f(x) implies
If(x) - q(x) + r(x)1 = Ip( -x) - g( -x)1 < e.
from which it follows that V(x) - q(x)1 < e. (Here we use the elementary
b ab
property: ifla+bl < e and la-bl < e,then lal = lai + 2 l ~ lai b l+la 2 l <
b
e and Ibl < e.) In other words, the "Odd polynomial q is e-unifonnly close to f
on [0, 1], and the desired conclusion follows.
Section 11: TIlE STONE-WEIERSTRASS APPROXIMATION TIlEOREM 101
Problem 11.7. Iff: [0, 1] -+ JR. is a continuous function such that fd f( 2,,:JX)
dx = 0 for n = 0, 1,2, ... , then show that f(x) = 0 for all x E [0, 1]. Does the
same conclusion hold true if the interval [0, 1] is replaced by the interval [-1, I]?
Solution. Assume that a continuous function f E qo, 1] satisfies fol fe":JX)

dx = 0 for each n = 0, 1, 2, .. " Then the change of variable u = 2,,:JX
(or x = u211 + I) yields
and so fol X 211 f (x) dx = 0 holds for all n = 0, 1, 2, .... The conclusion now
follows immediately from Problem 11.5.
The conclusion is not valid if we replace the interval [0, 1] by the inter-
val [-1,1]. For instance, if f(x) = x for all x E [-1,1], then note that
f~lfe":JX)dx 0 holds for all n 0,1,2, ....
= =
Problem 11.8. Assume that a function f: [0, 00) -+ JR. is either a polynomial or
else a continuolls bounded function. Then show that f is identically equal to zero
(i.e., show that f = 0) if and only if fo f(x)e- IIX dx = 0 for all n = 1,2,3, ....
oo
Solution. Let f: [0, 00) --+ JR. be a continuous bounded function. If f = 0,

then clearly fo f(x)e- IIX dx = 0 holds for all n = 1,2,3, ....
oo
For the converse; assume that
1 00
f(x)e- IIX dx =0 holds for all n = 1,2,3, . . . . (*)
Using the change of variable u = e- x , it follows from (*) that
roo f(x)e-IIX dx =. t f(-lnll)1I"- 1du = 0, n = 1,2,.... (**)

10 10+
In particular, fo~g(u)ull du = 0 holds for each n =
0,1, ... , where g(u) =
uf(-lnu). Since f is bounded, note that limu-+o+ g(u) = 0 holds, and so g
defines a continuous function on [0, 1]. From (**), we see that fol g(x)x" dx 0 =
holds for all n = 0, 1,2, .... Problem 11.4 implies that g = 0, and consequently
f =0.
A closer look at the above arguments reveals that we have actually proven the
following result.
102 Chapter 2: IDPOLOGY AND CONTINUTIY
• Assume that f: [0, 00) -+ 1R is a continuousfunctioll such that
1 00
f(x)e-
nX
dx = ° forall n = k, k + 1, k + 2, ... ,
where k is a positive integel: If liml/ .... o+ urn f (-In u) =°

for some natural
number m, then the function f is identically equal to zero.
Indeed, replacing 11 by n + k + m + 1 in (**), we get
10+
t urn + f(-lnu)u
k ll
du = 0, n = 0,1,2, ... ,
which implies (as above) that f = 0. The reader can verify easily that any function
°
°
f that satisfies If(x)1 :s Ce ax for some C > and CY. > 0 and all x ::: Xo also
satisfies liml/ .... o+ un! f (- In u) = for some natural number m. In particular, the
reader should notice that every polynomial p satisfies an estimate of the form
Ip(x)1 :s Ce ax .
One more comment regarding the above discussion is in order. Recall that if
f: [0, 00) -+ 1R is a "nice" function, then the formula
.c(f)(s) = 1 00
e-S / f(t)dt
is called the Laplace transform of f. The Laplace transform is a linear operator

and plays an important role in a wide range of applications. The reader should
notice that in actuality property (.) asserts that the Laplace transform is a one-to-
one operator when defined on an appropriate linear space of functions. (See also
Example 30 of Chapter 5 in the text.)
Problem 11.9. Show that a continuous bounded function f: [1, 00) -+ 1R is

identically equal to zero if and only if fl X-II f(x) dx = for each n = 8,9,
10, ....
oo
°
Solution. The "if" part only needs verification. Therefore, assume that the func-
tion f: [I, 00) ---7 1R satisfies ftx-n f(x)dx = 0 for each n = 8,9,10, ....
Using the change ofvarlable u = x-I, we see that
where g(u) = u 6 f( 1).

11
Since f is bounded, we see that limu_o+ g(u) = 0, and
.
so
g defines a continuous function on [0, 1]. In addition, from (* * *), we see that
Jolx"g(x)dx
that g
°
= holds for each n = 0, 1,2, ....
= 0, and consequently, f = 0.
By Problem 11.4, it follows
Problem 11.10. Let A be an algebra of cofltinuous real-valuedfunctions defined

on a compact topological space X and separating the points of X. Show that the
closure A of A il1 C (X) with respect to the uniform metric is either all of C (X) or
else that there exists a E X such that A = (f E C(X): f(a) = o}.
Solution. Let A £ C(X) be an algebra, where X is compact. Now, consider

the sequence of polynomials (P,,(x)} on [0, 1] defined by
PI(x) = ° and Pn+l(x) = P,,(x) + Hx - (p,,(x))2] for n = 1,2, ....

An easy inductive argument shows that each polynomial P,,(x) has a constant
term equal to zero. This guarantees that if f E A, then P,,(f) E A for each 11.
Also, by Lemma 11.4, we know that the sequence {P,,(x)} converges uniformly
to ,JX on [0, 1]. Thus, if f E A is non-zero, then put c = IIflloo, and note that:
1. The sequence (PII ( :,:)} £ A converges uniformly to 1:'1. Hence, If I E A.

2. Since (p,,( I[.I)} £ A converges uniformly to flij, we see thatflij E A.
Thus, if f E A, then both If I and .Ji7i belong to A. In particular, A is an

algebra and a function space.
Now, suppose that A is not of the form (f E C(X): f(a) = o} for some
a EX. This implies that for each x EX, there exists some f E A with
f(x) =1= 0. Thus, for each x EX, there eX'ists some fr E A and a neighborhood
°
Vr of x with frey) =1= for all Y E Yr. By the compactness of X, there exist
XI, .•• ,x" E X with X = U'=I Vrj. Note that the function g = I~ + ... + I;',
°
of A satisfies g(x) > for each x EX. Multiplying by an appropriate constant,
we can assume that g(x) > 1 holds for all x. Put h" = ;;g, and note that
h" E A and that h,,(x) t 1 for each x EX. By Dini's theorem, {hIll converges
uniformly to the constant function 1, and so 1 E A. Theorem 11.5 now guarantees
that A = C(X) must hold.
Problem 11.11. Let A be the vector space generated by the functions
1, sinx, sin2 x, sin3 x, ...
defined on [0, 1]. That is, f E A if and only if there is a non-negative integer k and
real numbers ao, aI, ... , ak (all depending on f) such that f(x) = L~=o a" sin" x
104 Chapter 2: WPOLOGY AND CONTINUITY
for each x E [0,1]. Show that A is an algebra and that A is dense in C[O, 1] with
respect to the uniform metric.
Solution. Clearly, A is an algebra of functions that contains the constant function

1. Also, since the function f(x) = sinx separates the points of [0, 1], the algebra
A likewise separates the points of [0, 1]. By the Stone-Weierstrass theorem, A
is dense in C[O, 1].
Problem 11.12. Let X be a compact subset oflR. Show that C (X) is a separable
metric space (with respect to the uniform metric).
Solution. The polynomials with rational coefficients form a countable set (why?).
By Corollary 11.6, this set is dense in C(X).
Problem 11.13. Generalize the previous exercise as follows: Show that if

(X, d) is a compact metric space, then C(X) is a separable metric space.
Solution. By Problem 7.2, we know that X is a separable metric space. Fix a

countable dense subset {XI, X2, ... } of X and for each n let fll: X ---+ lR be
the function defined by f,,(t) = d(t, xn) for each t EX.
Now, let x, Y E X satisfy X =1= y. Put d(x, y) = 28 > O. Choose some n
with d(x, XII) < 8, and note that
fll(Y) = d(y, XII) 2: d(x, y) - d(x, XII) 2: 28 - 8 = 8> d(x, xn) = fn(x),
so that f,,(x) =1= fn(Y). This implies that the algebra-generated by {1, /J, 12, ... }
separates the points of X. By the Stone-Weierstrass theorem (Theorem 11.5), this
algebra must be dense in C(X).
Next, consider the collection C of all finite products of the countable collec-
tion {1, /J, h ... } and note that C is a countable set, say C = {gl, g2, ... }. To
complete the proof note that the finite linear combinations of {1, gl, g2, ... } with
rational coefficients form a countable dense subset of C(X).
Problem 11.14. Let X and Y be two compact metric spaces. Consider the
Cartesian product X x Y equipped with the distance D J given in Problem 7.4, so
that X x Y is a compact metric space. Show that iff E C(X x Y) and E > 0, then
there exist functions {fl, ... , fn} f; C (X) and {gl, ... , gil} f; C (Y) such that
11
If(x, y)":' L f;(x)g;(y)1 < E

;=1
holds for all (x, y) E X X Y.

Solution. Consider the set
A = {h E C(X x Y): :I {fI, ... , f,,) ~ C(X), {gl,"" gn) ~ C(y)

/I
with hex, y) =L !i(X)gi(Y) V (x, Y) E X x Y }.

i=l
Then, A is an algebra offunctions of C (X x Y) and 1 E A. On the other hand, if

(Xl, Yl) =1= (X2, Y2), then either Xl =1= X2 or Yl =1= Y2. If Xl =1= X2, then select some
! E C(X) with !(Xl) =1= !(X2), and let F(x, y) = lex) for all (x, y) E X X Y. If
Yl =1= Y2, then pick some g E C(Y) with g(Yl) =1= g(Y2), and put F(x, y) = g(y).
In either case, F E A and F(Xl, Yl) =1= F(X2' Y2) holds, so that A separates the
points of X x Y. Now, by the Stone-Weierstrass theorem (Theorem 11.5), we
have A = C(X x y), and the desired conclusion follows.
C RAPTER 3 ______________
THE THEORY OF MEASURE
12. SEMIRlNGS AND ALGEBRAS OF SETS
Problem 12.1. If X is a topological space, then show that the collection
5 = {C n 0: C closed and 0 open} = {C[ \ C2: C[, C2 closed sets}
is a semiring of subsets of x.
Solution. From 0 = ono and X = X n X, we see that 0, X E 5. Next,
notice that C [ n 0[, C2 n 02 E 5 imply
Now, ifC[ n 01, C2 n 02 E 5, then
C I n 0 1 \ C2 n 02 = (C[ n O2) n (C2 n 02)C

= (C I n 0 1) n (C~ U OD
= (CI n 0 1) n [q U (O~ n C2 )]
= [Cln(0[nC2)]U[(ClnC2n02)nOI]
= AUB,
where A = CI n (01 n q) E 5 and B = (C I n C2 n 0i) n 0 1 E 5 satisfy
AnB =0.
Problem 12.2. Let 5 be a semiring of subsets of a set X, and let Y s:;; X. Show
that Sy = {Y n A: A E 5} is a semiring ofY (called the restriction semi ring of
5 to Y).
107
108 Chapter 3: TIlE TIlEORY OF MEASURE
Solution. The conclusion follows from the identities:

a. Y n (Z5 =
(Z5 ;
b. (Y n A) n (Y n B) =
Y n (A n B); and
c. YnA \ YnB =
Y n (A \ B).
Problem 12.3. Let S be the collection of all subsets of [0, 1) that can be written
asfinite unions of subsets of[O, 1) of the form [a, b). Show that S is an algebra of
sets but not a a-algebra.
Solution. Let A = U;'=1 [ai, bi ) and B = Uj=1 [c j, d j). Then, we have

a. A UB E S;
b. An B = U'=1 Uj=l[ai, bi ) n [Cj, dj ) E S; and
c. [0, 1) \ A = n;'=1 ([0, 1) \ [ai, bi )) E S, where the last membership holds
since each [0, 1) \ [ai, bi ) can be written as a finite union of sets of the
form [a, b).
To see that S is not a a-algebra note that n:l [0, ~) = {OJ rf. S.
Problem 12.4. Prove that the a -sets of the semiring
S = ([a, b): a, bE 1R}
form a topology for the real numbers.
Solution. Let!" be the collection of all a-sets of S. Clearly, (Z5 E !" and 1R =
U:l[-n, n) E !". It should be clear that!" is closed under finite intersections.
Thus, in order to establish that !" is a topology, we need to show that !" is closed
under arbitrary unions. That is, if {[ai, bi ): iE/} is a collection of nonempty
members of S, then we must show that A = UiE/[ai, bj) belongs to !" (i.e., that
A is a a-set).
To see this, let V = U iE1 (ai, bi ). Then, V is an open set, and thus, there exists
an at-most countable collection of pairwise disjoint open interval {(c j, d j ): j E J}
(see part (g) of Problem 6.11) such that V = UjEiC j, d j). For each j E J, let
Aj = [Cj, d j ) if Cj = ai for some iE/ and let Aj = (Cj, d j ) if Cj =1= ai for all
iE/. Clearly, each A j is a a -set. Moreover, it is easy to see that A = UJEJ A j
holds, which shows that A is a a-set.
Problem 12.5. Let S be a semiring of subsets of a nonempty set X. What addi-

tional requirements must be satisfieafor S to be a base for a topology on X? (For
the definition of a base see Problem 8./8.) Prove that if such is the case, then each
member of S is both open and closed in this topology.
Section 12: SEMIRINGS AND ALGEBRAS OF SETS 109
Solution. Since S is already closed under finite intersections, it follows from

the definition of a base that S will be a base if and only if UAeSA = X.
Now, assume that UAesA = X holds. Note first that if A, B E S, then (since
S is a semiring) A \ B can be written as.a finite union of (disjoint) members
of S. It follows that A \ B belongs to the topology generated by S. Thus, if
B E S, then the relation
Be =X \ B = (UA) \ B = U(A \ B),

AeS AeS
shows that Be belongs to the topology generated by S. That is, in this case, every
B E S is a closed and open set.
Problem 12.6. Let A be afixed subset of a set X. Determine the two a-algebras
of subsets of X generated by
a. {A}, and
b. {B: A S; B S; X}.
Solution. (a) {0, A, N, X} and (b) {B: AS; B or AS; BC}.
Problem 12.7. Let X be an uncountable set, and let
S = {E S; X: E or E C is at-most countable}.
Show that S is the a -algebra generated by the one-point subsets of X.
Solution. Clearly, S contains the one-point subsets of X, and every member of

S must be a member of the a-algebra generated by the one-point sets. Thus, it
remains to be shown that S is a a-algebra.
Clearly, 0, XES and S is closed under complementation. Then let
{An} S; S. If each All is at-most countable, then U:;:1 All is at-most countable,
and consequently U:;:1 All E S. qn the other hand, if some Ak is uncountable,
then (Ad is at-most countable and the inclusion (U:;:1 An)" S; (Ad shows
that U:;:l All E S.
Problem 12.8. Characterize the metric spaces whose open sets form a a -algebra.
Solution. We shall show that the open sets of a metric space X form a a -algebra
if and only if X is a discrete metric space (i.e., if and only if every subset of X
is open).
Let r be the collection of all open sets. If r = P(X), then clearly r is
a a-algebra. On the other hand, if r is a a-algebra and x EX, then
{x} = n~l B(x, ~) shows that {x} is an open set. This easily implies that every
subset of X is open (Le., 1: = P(X) holds).
Problem 12.9. Determine the a-algebra generated by the nowhere dense subsets
of a topological space.
Solution. Let X be a topological space. Define
A = {A S;; X: A is meager or A C is meager}.

Recall that a set is called meager if it can be written as a countable union of nowhere
dense sets-a set A is nowhere dense if (A)O = C/J. We claim that A is the a-
algebra generated by the nowhere dense sets of X. Clearly, every nowhere dense
set belongs to A, and every member of A belongs to the a-algebra generated by
the nowhere dense sets. So, it suffices to establish that A is a a-algebra of sets.
Clearly, C/J, X E A. Also, it should be obvious that A is closed under comple-
mentation. Now, let {An} S;; A. If each An is meager, then clearly U~l An E A.
On the other hand, if (Ak)C is a meager set for some k, then the set inclusion
(U~l Ant S;; (Ad implies U~l An E A. Therefore, A is a a-algebra.
Problem 12.10. Let X be a nonempty set, and let F be an uncountable collection

of subsets of X. Show that any element of the a-algebra generated by F belongs
to the a-algebra generated by some countable subcol~ection of F.
Solution. Assume F to be uncountable. Let A be the a-algebra generated by

F. Denoteby {A;: i E I} the family of all a-algebras each of which is generated
=
by a countable subset of F. It suffices to show that B UiEI A; is a a-algebra
(because if this is the case, then A = B must hold, and the conclusion follows).
Clearly, C/J E B. Also, if A E B, then it is easy to see that AC E B likewise
holds. Now, let {An} S;; B. Since each An belongs to a a-algebra generated
by a countable subset of F, it easily follows that there exists some i E I with
{An} S;; Ai. Thus, U~l An E Ai S;; B. That is, B is a a-algebra, as required.
Problem 12.11. Show that every Fa- and every Go-subset ofa topological space
is a Borel set.
Solution. The Borel sets are the members of the a -algebra generated by the open
sets. So, a countable intersection of open sets (or a countable union of closed sets)
is always a Borel set.
Problem 12.12. Show that every infinite a -algebra ofsets has uncountably many
sets.
Section 12: SEMIRINGS AND ALGEBRAS OF SETS 111
Solution. Let A be an infinite a -algebra of subsets of a set X. If A contains a

sequence {An} of nonempty pairwise disjoint sets, then A has uncountably many
members. Indeed, if this is the case, then for each subset s of natural numbers
let As = UnEs An E A, and note that As =ft AI if s =ft t. By Problem 5.6, the
collection {As: s E PCJN)} has uncountably many members, and so A must
likewise have uncountably many members.
Next, we shall show that there exists a sequence {B II} S; A with Bn+ 1 S; Bn
and Bn+ 1 =ft Bn for all n. If this is done, then put An = Bn \ Bn+l, and use the
above arguments to see that A is an uncountable set.
Using induction, we shall establish the existence of a sequence {B II } such that:
1. BII + 1 S; Bn and BII+l =ft Bn for all n, and

2. {Bn n A: A E A} is an infinite set.
The basic step of the induction is the following: Assume that Bn E A has been
chosen so that {Bn n A: A E A} has infinitely many members. Choose C E A
so that ¢ S; BII n C S; Bn is a proper inclusion at both ends. In view of
we see that either {CBnnC)nA: A E A} or {CBII \ C)nA: A E A} is infinite. If

{CBnnC)nA: A E A} is infinite, put BII + 1 = BllnC. If {CBllnC)nA: A E A}
is finite, put BII+l = Bn \ c.
Start the induction with Bl = X.
Problem 12.13. Let CX, i) be a topological space, let B be the a-algebra of its
Borel sets, and let Y be an arbitrary subset of X. If Y is considered equipped with
the induced topology and By denotes the a-algebra of Borel sets of CY, i), then
show that
By = {A n Y: A E B}.
Solution. Let CX, L), Y. and By be as in the problem, and let
A = {A n Y: A E B}.
We have to show that By S; A and A S; By both hold.

Clearly, A is a a-algebra of subsets of Y and 0 n YEA holds for each
o E i. Thus, A contains the open sets of Y, and so By S; A. Now, consider the
collection of sets
c = {A E B: AnY E Br}.
112 Chapter 3: THE THEORY OF MEASURE
It is easy to see that C is a a -algebra of subsets of X satisfying r S;; C. Hence,

C = B, and this implies that A S;; By. Therefore, By = A, as claimed.
Problem 12.14. Let AI, ... , An be sets in some semiring S. Show that there
exists afinite Ilumber ofpaiJwise disjoint sets B I , ... , Bm of S such that each Ai
call be written as a union of sets fi'om the B I , ... , Bm.
Solution. We use induction on n. For n = 1 the result is trivial. Thus, assume

that the result is true for some n, and let AI, ... , An, An+1 be members of S.
Pick a finite number of pairwise disjoint members B I, ... ,Bm of S such that
each Ai, 1 :s i :s n, can be written as a union of sets from B I , ... , Bm. Clearly,
U;'=I Ai S;; Uj'=1 B j • The sets BI n A n+I, ... , Bm n An+1 are pairwise disjoint
members of S. On the other hand, for each 1 :s i :s m there exists a finite pairwise
disjoint collection :F; S;; S with Bi \ An+1 = UCE:F; C (by the definition of the
semiring). Thus, the collection
is finite and pairwise disjoint. Moreover, each Ai (1 :s i :s n) can be written as a

union of members of F. Now, observe that
m •
A II +I = (A II +I \ UB j) U (B I n A II +I) U··· U (Bm n AII+d.
j=1
By Theorem 12.2(1) there exist pairwise disjoint sets DI, ... ,Dk in S such that
AII+I \ Uj~1 B j = U;=I Dr. Finally, the collection F U {DJ, ... , Dd S;; S is
finite and pairwise disjoint, and each set Ai (1 :s i :s Il + 1) can be written as a
union from these sets.
13. MEASURES ON SEMIRINGS
Problem 13.1. Let {all} be a sequence ofnon-negative rea/numbers. Let JL(¢) =

0, and for evelY nonempty subset A of IN put JL(A) = LnEA an. Show that
JL: P(lN) -+ [0,00] is a measure.
Solution. If {An} is a sequence of pairwise disjoint subsets of IN and A =

U:I An, then note that
Section 13: MEASURES ON SEMIRINGS 113
Problem 13.2. Let S be a semiring, and let fL: S -+ [0,00] be a set function
such that fL(A) < 00 for some A E S. If fL is a-additive, then show that fL is a
measure.
Solution. Write A = A U C/J U C/J U .... Then,

fL(A) = fL(A) + fL(C/J) + fL(C/J) + ....
If fL(C/J) > 0, then fL(A) = 00, contrary to our hypothesis. Thus, fL(C/J) = 0, and
so fL is a measure.
Problem 13.3. Let X be an uncountable set, and let the a-algebra
S = {E ~ X: E or E C is at-most countable};
see also Problem 12.7. Show that fL: S -+ [0, (0), defined by fL(E) = 0 if E is
at-most countable and fL(E) = 1 if E C is at-most countable, is a measure on S.
Solution. Clearly, fL(C/J) = O. For the a-additivity of fL let {Ell} ~ S be a

pairwise disjoint sequence. Let E = U:1 Ell' If each Ell is at-most countable,
then E itself is at-most countable, and so fL(E) = I::1 fL(E II ) = 0 holds. On the
other hand, if E'k is at-most countable for some k, then (in view of Ell n Ek = C/J
for n i= k) we must have Ell ~ E'k for n i= k, and so Ell is at-most countable
for each n i= k. Thus,
00
1 = fL(E) = fL(Ed = L fL(EII)'

11=1
= [0, 1], then S is a a-subalgebra of the

It is interesting to observe that if X
Lebesgue measurable subsets of [0, 1], and fL is the restriction of the Lebesgue
measure to S.
Problem 13.4. Let X be a non empty set, and let f: X -+ [0,00] be afunction.
Define fL: P(X) -+ [0,00] by fL(A) = I:xEA f(x) if A i= C/J and is at-most
cOllntable, fL(A) = 00 if A is uncountable, and fL(C/J) = O. Show that fL is a
measure.
Solution. For the a-additivity of fL, let {All} be a pairwise disjoint sequence of
subsets of X. Let A = U:1 All' If some All is uncountable, then A is likewise
uncountable, and hence, in this case fL(A) = I::1fL(An) = 00 holds. On the
114 Chapter 3: TIlE TIIEORY OF MEASURE
other hand, if each All is at-most countable, then A is also at-most countable,
and so
00 00
JL(A) = Lf(x) = L[Lf(x)] = LJL(An)

xeA n=1 xeA n n=1
also holds.
Problem 13.5. Let 5 be a semiring, and let JL: 5 -+ [0,00] be afinitely additive
measure. Show that if JL is a-subadditive, then JL is a measure.
Solution. Let {An} S; 5 be a pairwise disjoint such that A = U:I An E5. By

hypothesis, JL(A) :5 L:I JL(An) holds. On the other hand, if k is fixed, then there
exist pairwise disjoint sets B I , ••• , Bm E 5 such that A \ U;'=I An = U~=I Bi
(see Theorem 12.2). Since A = [U;'=I An] U [U~=I Bi] is a finite union of
pairwise disjoint members of 5, the finite additivity of JL implies
k k m
L JL(An) :5 L JL(An) + L JL(B i ) = JL(A).

n=1 n=1 i=1
Since k is arbitrary, L:I JL(An) :5 JL(A) also holds, and so JL is a measure .
. Problem 13.6. Let {JLn} be an increasing sequence of measures on a semiring

5; that is, JLn(A) :5 JLn+1 (A) holdsfor all A E 5 and all 11. Define JL: 5 -+ [0,00]
by JL(A) = sUP{JLn(A)} for each A E 5. Show that JL is a measure.
Solution. ClearlY,JL(<;ZJ) = 0. Now,let {An} S; 5 be a pairwise disjoint sequence

such that U:I An = A E 5. Since each JLi is a measure,
00 00
JLi(A) = L JLi(An) :5 L JL{An)

n=1 n=1
holds, and so JL{A) :5 L:I JL(An). On the other hand, for each k we have
Thus, L:J JL{An) :5 JL(A) also holds, which shows that the measure JL is a-
additive.
Section 13: MEASURES ON SEMIRINGS 115
Problem 13.7. Consider the semiring S = (A S; lR.: A is at-most countable).

°
and define p,: S -+ [0, co] by p,(A) = if A is finite and p,(A) = co if A is
countable. Show that p, is a finitely additive measure that is not a measure.
Solution. Let A I, ... , An be pairwise disjoint members of S. Put A = U;'= I Ai.

If each Ai is a finite set, then A is likewise a finite set, and I:7=1 P,(Ai) = p,(A) =
° holds. On the other hand, if one of the Ai is countable, then A itself is also
countable, and I:7=1 P,(Ai) = p,(A) = co holds. Thus, p, is a finitely additive
measure.
To see that p, is not a-additive, note that IN = U:I (n), while
°= 00
LP,«(nj) < p,(lN)

11=1
= co.
Problem 13.8. Show that every finitely additive measure is monotone.
Solution. Assume that p.: S -+ [0, co] is a finitely additive measure. Let
A, B E S satisfy A S; B . Choose a finite collection of disjoint sets C I, ... , CII
of S such that B \ A = U;'=I C i . Then,
B = A U CI U··· U CII
is a finite union of pairwise disjoint sets of S. Thus, by the finite additivity of p"
we have
p,(A) ::: p,(A) + p,(C I) + ... + p,(C II ) = p,(B).
Problem 13.9. Consider the set function p, defined in Example 13.6. That is,
consider a nondecreasing and left-continuous function f: lR. -+ lR. and then define
the set function p.: S -+ [0, co) by p,([a, b)) = f(b) - f(a), where S is the
semiring S = ([a, b): -co < a ::: b < co}. Prove alternately the fact that p, is a
measure.
Solution. An alternate way of proving the a-additivity of p, is as follows. Let

a < b and let [a, b) = U;:O=I[all , bll ) with the sequence ([an, b n)} pairwise
disjoint. For each a < x ::: b let
Sx = :l]f(bi ) f(ai)],
i
where the sum (possibly a series) extends over all i for which [ai, bi) S; [a, x)
holds; we let Sx = °
if there is no such interval. Since f is nondecreasing, we
have Sx :s f(x) - f(a). Next, note that the set
A = {x E (a, b]: Sx = f(x) - f(a)}
is nonempty. Let t = sup A, and note that a < t :s b. Now, for x E A, we have
f(x) - f(a) = Sx :s Sr :s f(t) - f(a),
and so, by the left-continuity of f, we get Sr = f(t) - f(a). That is, tEA.
Our objective is to establish that t = b holds. Assume by way of contradiction
that a < t < b. Then ak :s t < bk must hold for some k. Since the sequence
([a", b,,)} is pairwise disjoint, observe that [ai, bi) f; [a, t) holds if and only if
[a;, bi ) £; [a, ak). Thus, Sr = sa, holds. In particular, the relation
f(t) - f(a) = Sr = sa, :s f(ak) - f(a) :s f(t) - f(a)
guarantees that ak EA. However, this implies bk E A, which is impossible. There-

fore, t = b holds, which guarantees that
00
1L([a, b)) =L lL([a n , b,,)).

n=1
14. OUTER MEASURES AND MEASURABLE SETS
Problem 14.1. Show that a countable Ullioll of Ilull sets is agaill a Ilull set.
Solution. The conclusion follows from the inequality

00 00·
IL(U An) :s L IL(An)·

n=1 n=1
Problem 14.2. If IL is all outer measure all a set X alld A is a null set, then show
that
IL(B) = IL(A U B) = IL(B \ A)

holds for every subset B of X.
Section 14: OUTER MEASURES AND MEASURABLE SETS 117
Solution. The conclusion follows from the inequalities:
fL(B) ::: fL(B U A) = fL((B \ A) U A)

::: fL(B \ A) + fL(A) = fL(B \ A) ::: fL(B).
Problem 14.3. Let fL be an outer measure on a set X. If a sequence {An} of

subsets of X satisfies I::I fL(A Il ) < 00, then show that the set
!
E = x EX: x belongs to An for infinitely many n }
is a null set.
Solution. Assume that a sequence {All} satisfies I::lfL(A Il ) < 00. For each n
let Ell = U~1l Ai, and note that E ~ Ell holds for each n. Therefore,
00
0::: fL(E)::: fL(En)::: LfL(A i ) --+ 0,

i=n
and hence fL(E) = O.

Problem 14.4. If E is a measurable subset of X, then show that for every subset
A of X the following equality holds:
fL(E n A) + fL(E U A) = fL(E) + fL(A).

Solution. The measurability of E gives
Consequently, we have
fL(E U A) + fL(E n A) = fL(E) + fL(A n E + fL(A n E) = fL(E) + fL(A).

C
)
Problem 14.5. Let fL be an outer measure on a set X. If A is a nonmeasur-

able subset of X and E is a measurable set such that A ~ E, then show that
fL(E \ A) > O.
Solution. If fL(E \ A) = 0 holds, then E \ A E A. Thus, A = E \ (E \ A) E

A, which is a contradiction. Therefore, fL(E \ A) > O.
118 Chapter 3: TIm TImORY OF MEASURE
Problem 14.6. Let A be a subset of X, and let {Ell 1be a disjoint sequence of
measurable sets. Show that
00 00
/-L( U A n En) = L/-L(A nEil)'

11=1 n=1
Solution. From the a-subadditivity of /-L, we see that

00' 00 00
/-L( An [U
n=1
En]) = /-L(U A n En) ::: L /-L(A n En).
n=1 n=1
On the other hand, Lemma 14.5 implies
Problem 14.7. Let {An 1be a sequence of subsets of X. Assume that there exists
a disjoint sequence {Bill of measurable sets such that An £; Bn holds for each n.
Show that
Solution. Put A = U:::I All and note that An Bn = An holds for each n.
Thus, using the preceding problem, we see that
00 00 00 00
/-L(U An) = /-L( An [U Bn]) = L

n=1 n=1 n=1
/-L(A n Bn) = L
n=1
/-L(An).
Problem 14.8. Let /-L be an outer measure on a set X. Show that a subset E of
that F £; E, and /-L(E \ F) < E.

°
X is measurable if and only iffor each E > there exists a measurable set F such
Solution. If E is measurable, then F = E satisfies the condition for each s > 0.

For the converse, assume that the condition is satisfied.
Start by choosing for each n a measurable set Fn with FII £; E and/-L(E \ Fn) < ~.
Put F = U:::I Fn £; E, and note that F is measurable. Consequently,
/-L(E \ F) ::: /-L(E \ Fn) < ~ for each n implies /-L(E \ F) = 0, and so
E \ F is measurable. The measurability of E now follows from the identity
E = F U(E \ F).
An alternate proof of the preceding part goes as follows. Let A be a subset of

X with J-L(A) < 00. If E > 0 is given, pick a measurable subset F with F ~ E
and J-L(E \ F) < E. Then
J-L(A n E) = J-L(A n [F U (E \ F)])

::: J-L(A n F) + J-L(A n (E \ F») ::: J-L(A n F) + E
implies J-L(A n F) - J-L(A n E) > -E, and so
J-L(A) = J-L(A n F) + J-L(A n F C

)
2: J-L(A n F) + J-L(A n E C
)
= J-L(A n E) + +J-L(A n E + [J-L(A n F) -

C
) J-L(A n E)]
> J-L(A n E) + J-L(A n E E
C
) -
for all E > O. This implies J-L(A) 2: J-L(A n E) + J-L(A n E C

), which shows that E
is a measurable set.
Problem 14.9. Let J-L be an outer measure on a set X. Assume that a subset E
of X has the property that for each E > 0, there exists a measurable set F such
that J-L(E llF) < E. Show that E is a measurable set.
Solution. Let 8 > O. According to the preceding problem, it suffices to show

that J-L(E \ G) < 8 holds for some measurable set G with G ~ E.
For each n choose Fn E A with J-L(E llFn) < 2- n8. Put F = n:1 FIl E A.
Since F \ E ~ Fn \ E holds, we have
J-L(F \ E) ::: J-L(Fn \ E) < T n8
for each n, and so J-L(F \ E) =

O. Thus, F \ E E A, and hence F n E =
F \ (F \ E) is also a measurable set. Now, note that F n E ~ E holds and
co co
J-L(E \ En F) = J-L(E \ F) = J-L(U(E \ Fn») ::: LJ-L(E \ Fn) < 8.
n=1 n=1
Problem 14.10. Let X = {I, 2, 3}, :F = {0, {l}, {l, 2} } and consider the set
function J-L::F -+ [0,00] defined by J-L(0) = 0, J-L({l}) = 2 and J-L({l, 2}) = l.
a. Describe the outer measure J-L* generated by the set function J-L.
b. Describe the a-algebra of all J-L*-measurable subsets of X (and conclude
that the set {I} E :F is not a measurable set).
Solution. (a) The outer measure J1-*: P(X) -+ [0,00] is given by
J1-*(0) = 0, J1-*({1}) = J1-*({2}) = 1, J1-*({3}) = 00,

J1-*((1, 2}) = 1, J1-*({1, 3}) = J1-*({2, 3J) = J1-*({1, 2, 3}) = 00.
(b) The o--algebra of all measurable sets is A = {0, (3), (I, 2), x}.
Problem 14.11. Let v: P(X) -+ [0,00] be a setfunctioll. Show that v is an outer
measure if alld only if there exist a collection:F ofsubsets ofX containing the empty
set and a set function J1-::F -+ [0,00] with J1-(0)
for all A E P(X).
= °
satisfying v(A) J1-*(A) =
Solution. Assume first that v: P(X) -+ [0,00] is an outer measure. Let:F =

P(X) and J1- = v. We claim that v(A) = J1-*(A) holds for each A E P(X), where
00 00
J1-*(A) = inf{LJ1-(A n): {All} s;;:F and AS;; An}, U

n=1 n=)
and inf 0 = 00. To see this, let A E P(X). From A = A U 0 U 0 U 0· . " we see
that J1-*(A) ::s J1-(A) = v(A). On the other hand, if A s;; U:1 An holds true, then
from the o--subadditivity of v, we see that
00 00
v(A) ::s L v(AII) =L J1-(A n),

11=1 11=1
and so V(A) ::s J1-*(A) is also true. Hence, v(A) = J1-*(A) for each subset A of X.
For the converse, assume that the outer measure J1- * generated by a set function
°
f.1-::F -+ [0,00] satisfies J1-(0) = and v(A) = J1-*(A) for each A E P(X). We
shall show that v is an outer measure by verifying the three properties required to
be satisfied by v in order to be a measure.
°
(I) From ::s v(0)
v(0) =0.
=
J1-*(0) ::s J1-(0) + J1-(0) + J1-(0) + ... 0, we see that =
(2) (Monotonicity) Let A s;; B and let {An} be a sequence of :F with B C
U:1 An· Then, AS;; U:) All, and so J1-*(A) ::s 2::::1 J1-(An). Therefore,
00 00
v(A) = J1-*(A)::s inf{LJ1-(A II ): {An} s;;:F and B s;; U All} = J1-*(B) = v(B).
n=1 n=1
(If there is no sequence {All} of:F with B s;; U:1 All, then J1-*(B) = 00, and
v(A) ::s v(B) = J1-*(B) is trivially true.)
(3) (a-Subadditivity) Let {En} be a sequence of subsets of X and let E =

U~1 Ell' If 'L~1 fl.*(EIl) =
00, then veE) = fl.*(E) :::: 'L:l fl.*(EIl) =
'L:l v(En) is trivially true. So, assume 'L:l fl.*(En) < 00 and let e > O.
For each n pick a sequence {A~} of F wit~ Ell ~ U~1 A~ and
oc
I>(A~) < fl.*(En) + rne = v(E Il ) + rile.
k=1
Clearly, E ~ U:l UZ:l A~ holds, and so
:GOO 00 00
veE) = fl.*(E) :::: L I>(A~) < IJv(E Il ) + rneJ =L v(EIl) + e.

11=1 k=l n=1 n=1
Since e > 0 is arbitrary, veE) :::: 'L:l v(E Il ), and we are done.
Problem 14.12. Consider an outer measure fL on a set X and let A be the
collection of all measllrable sllbsets of X offinite measure. That is, consider the
family A = (A E A: fl.(A) < oo}.
a. Show that A is a semiring.
b. Define a relation::::::: on A by A ::::::: B If fL(A6.B) = O. Show that::::::: is an
equivalence relation on A.
c. Let D denote the set of all equivalence classes of A. For A E A let
A denote the eqllivalence class of A in D. Now, for A, BED define
dCA, B) = fl.(A6.B). Show that d is well defined and that (D, d) is a
complete metric space.
Solution. Note that if A, B, and C are three arbitrary sets, then
A6.C ~ (A6.B) U (B 6.C).
(a) Straightforward. (Note that in actuality A is a ring of sets.)

(b) If A, B, and C in A satisfy A ::::::: Band B ::::::: C, then the relation
fL(A6.C) :::: fl. (A6.B) U (B 6.C)) :::: fL(A6.B) + fL(B 6.C) = 0
shows that A ::::::: C.

(c) If A ::::::: Al and B ::::::: B\. then
fL(A6.B) :::: fL(A6.A 1) U (AI6.B 1) U (BI6.B))

:::: fl.(A6.Al) + fL(A I 6.B I ) + fL(B I 6.B) = fL(A j 6.B I ).
Similarly, /L(A1LlB 1) :::: /L(ALlB), and so /L(ALlB) = /L(A1LlB 1). This shows
that dCA, 13) = /L(ALlB) is well defined.
For the triangle inequality, note that
dCA, 13) = /L(ALlB) :::: /L(ALlC) + /L(C LlB) = dCA, C) + d(C, 13).
Thus, (D, d) is a metric space. What remains to be shown is that (D, d) is a
complete metric space.
To this end, let {An} be a Cauchy sequence of D. By passing to a subsequence,
we can assume that
holds for each n. Set A = n~1 U~k Ai EA. Now, let n be fixed and note that
A ~ U~n Ai = All U (U~n(Ai+1 \ Ai)) holds. Thus,
00
/L(A) :::: /L(An) +L /L(A;+I \ A;) < /L(An) + Tn < 00,

;=IJ
and so A ED. Moreover, we have
00 00
11
/L(A \ All):::: /L(U(A i+1 \ Ai)):::: L/L(A;+I \ Ai) < 2- •
;=11 ;=n
On the other hand, if x E All \ A, then x E All and x ¢. A = n~1 U~k Ai.
Consequently, there exists some k 2: 11 with x ¢. A; for each i 2: k. This implies
A" \ A ~ U~,,(Ai \ Ai+I ), and so /L(A" \ A) :::: L~" f.1(A; \ Ai+l) < 2-"
also holds. Therefore,
holds for each n. This shows that limd(A", A) = 0, and so (D, d) isa complete
metric space. (For an alternate proof of this part, see Problem 31.3.)
15. THE OUTER MEASURE GENERATED BY A MEASURE
Problem 15.1. Let (X, S, /L) be a measure space, and let E be a measurable
subset of x. Put SE = {E n A: A E S), the restriction of S to E. Show that
(E, SE, /L*) is a measure space.
Section 15: TIlE OUTER MEASURE GENERA TED BY A MEASURE 123
Solution. Let E be a measurable subset of X and let {An} be a sequence of

A such that
a. {An n E} is a pairwise disjoint sequence; and
b. there exists some A E S such that A n E = U:I An n E.
Using the fact that fL*: A ---+ [0,00] is a measure, we see that
00 00
fL*(AnE)=fL*(U(AnnE)) = LfL*(AnE),
n=1 11=1
and so fL * is a measure when restricted to the semiring S E.
Problem 15.2. Let (X, S, fL) be a measure space. Show that
fL*(A) = inf{fL*(B): B is a a-set such that A ~ B}
holds for every subset A of X.
Solution. Let {An} ~ S and let B = U:I An. By Theorem 12.2(3), there
exists a pairwise disjoint sequence {B II } of S such that B = U:1 B II • Thus,
for A ~ X, there exists a sequence (An} of S with A ~ U:1 An if and only if
there exists a a-set B with A ~ B. The desired equality now follows from the
relation
00 00
fL*(A) ~ L fL(B II ) = fL*(B) ~ L fL(An).

n=1 11=1
Problem 15.3. Show that eVel) interval I of 1R is Lebesgue measurable and

"A.*(I) = III (=the length of I).
Solution. In Example 15.5, we established thatthe intervals I of the form [a, b]
and [a,oo) are Lebesgue measurable and that "A. *(1) = III holds for these cases.
We shall consider the other cases separately. Assume -00 < a < b < 00.
a. I = (a, b]. Choose a sequence {XII} of real numbers with Xn -I- a and
a < XII < b for each n. Thus, by Example 15.5, we have
"A.*(a, b1) = lim "A.*([xn' b1) = lim (b - xn) = b - a =

Il~OO n-+oo
Ill.
b. I = (a, b). Pick a < XII < b with Xn -I- a and observe that[xlI , b) t (a, b).
c. I = (-00, a). Note that [a - n, a) t (-00, a) and so
A*(-oo, a») = n-+oo

lim A([a - n, a») = n-+oo
lim n = 00 = III.
d. 1= (-00, a]. Note that (-00, a) S;;; (-00, a] and so from the inequality
00 = 1(-:-00, a)1 = A*(-oo, a») :::: A*(-oo, al),

we see that III = A*(1) = 00.
e. I = (a, (0). The conclusion follows immediately from the obvious inclu-
sion [a + 1, (0) S;;; (a, (0).
f. I = (-00, (0). Note that [0, (0) S;;; (-00, (0).
Problem 15.4. Show that evelY countable subset of lR has Lebesgue measure
zero.
Solution. Let a E 1R. Then, {a} S;;; [a - e, a + e) holds for each' e > 0 and so
A*({a}) :::: A*([a - e, a + e») = 2e for all e > O. Therefore, A*({a)) = 0 holds
for all a E 1R. If A = {at, a2, ... } = U:t {an} is a countable set, then note that
A*(A) :::: L:t A*({a ll )) = 0 so that A*(A) = O.
Problem 15.5. For a subset A of lR and real numbers a alld b, define the set
aA + b = {ax + b: x E A}. Show that
a. A*(aA+b)=laIA*(A),and
b. if A is Lebesgue measurable, then so is aA + b.
Solution. LetA S;;; lRandfix two real numbers a andb. Since A S;;; U:J [all, bll )
holds if and only if A + b S;;; U:t [all + b, bll + b) holds, it is easy to see that
A*(A + b) = A*(A). The identities
En (b + A) = b + (E - b) n A and En (b + A/ = b + (E - b) n AC
imply
A*(E n (b + A») + A*(E n (b + A)C) = A*(E - b) n A) + A*(E - b) n A C),
which shows that A is measurable if and only if b + A is measurable for each

bE lR.
Next, note that A*(C(S, 1)) = IC~A*(s, 1)) = IcICt-s) holds. On the other hand,
since A S;;; U:t(an , bn ) holds if and only if aA S;;; U~=t a(a n , bn ) holds for each
Section 15: TIlE OUTER MEASURE GENERATED BY A MEASURE 125
a E JR and since A. * ([an, bll ») = A. * (all, bll »), it follows that A. *(aA) = lalA. *(A)
for each a E JR. Now, the identities
En aA = a(a- I E) n A) and En (aAt = a(a- I E) n N) (a f:. 0),

imply
which shows that A is measurable if and only if aA is measurable for each

a E JR.
Now, (a) and (b) follow from the preceding discussion.
Problem 15.6. Let 5 be a semiring of subsets of a set X, and let p,: 5 ~ [0, 00]
be a finitely additive measure that is not a measure. For each A S; X define (as
usual)
00 00
p,*(A) = inf{ L p,(AIl): {An} S; 5 and A S; U All }.

11=1 11=1
Show by a counterexample that it is possible to have p, f:. p,* on 5. Why does this
not contradict Theorem 15.1?
Solution. Consider the finitely additive measure p, of Problem 13.7. Clearly,

p,(JN) = 00. Since IN = U:I {n} E 5, we have p,*(JN) S r::1 p,({n}) 0, and =
=
so 0 p, *(IN) < p,(JN) 00. =
This conclusion does not contradict Theorem 15.1, since the cr -additivity of the
measure was essential for its proof.
Problem 15.7. Let E be an arbitral) measurable subset of a measure space

(X, 5, p,)andcollsiderthemeasurespace(E, 5 E, v), where5E {EnA: A E 5} =
=
and veE n A) p,*(E n A); see Problem 15.1. Establish the following properties
regarding the measure space (E, 5 E, v);
a. The outer measure v* is the restriction of p,* on E, i.e., v*(B) = p,*(B)for
each B S; E.
b. The v-measurable sets of the measure space (E, 5E, v) are precisely the
sets of the form E n A where A is a p,-measurable subset of X, i.e.,
Av = {F S; E: FE A/1-}'
Solution. Let (X, S, J.t), E, and v be as defined in the problem.

(a) Let B be an arbitrary subset of E. If {An} is a sequence of S satisfying
B 5; U:1 An, then note that B 5; U:1 E n An and so
00 00 ex;
v*(B) ::: L v(E nAil) = L J.t*(E nAil) ::: L J.t(A II )·

n=1 11=1 n=1
This implies v*(B) ::: J.t*(B). On the other hand, if {An} is a sequence of S
satisfying B 5; U:1 E n An, then we have
00 00
J.t*(B) ::: L J.t*(E nAn) = L v(E n A,J.

11=1 n=1
Thus, J.t*(B) ::: v*(B) also holds, and so v*(B) = J.t*(B) for each subset B
of E.
(b) Let F be a subset of E. Assume first that F is v-measurable. If A E S,
then note that
J.t(A) = J.t*(A n E) + J.t*(A n (X \ E))

= v*(A n E) + J.t*(A n (X \ E))
= v*(A n E) n F)+v*(A n E) n (E \ F))+I1*(A n (X \ E))
2:: J.t*(A n F) + J.t*([A n (E \ F)] U [A n (X \ En)
= J.t*(A n F) + J.t*(A n (X \ F)),
which shows that F is J.t-measurable.

For the converse, assume that F is l1-measurable. If A is an arbitrary subset
of E, then note that
v*(A) = J.t*(A) = J.t*(A n F) + J.t*(A n (X \ F))

= J.t*(AnF)+J.t*(An(E \ F))
= v*(A n F) + v*(A n (E \ F)),
which means that F is also v-measurable.
Problem 15.8. Show that a subset E ala measure space (X, S, J.t) is measurable
if and only iflor each E > 0 there exists a measurable set AE and two subsets BE
and CE satisfying
Section 15: TIlE OU1ER MEASURE GENERA TED BY A MEASURE 127
Solution. Let E be a subset of a measure space (X, S, fJ.-). If E is a measurable

set and 8 > 0 is given, then let AE = E and BE = Cs = cj;, and note that these
sets satisfy the desired properties.
For the converse, assume that for each 8 >: 0 there exist a measurable set AE
and subsets BE and CE satisfying
Replacing CEby (AE U BE) n C E, we can assume that C E is a subset of AE U BE'

From (*), we see that
Now, by Theorem 15.11, there exists a measurable set DE such that BE ~ DE

and fJ.-*(DE) = fJ.-*(BE)' Using (**), we get
E U CEU (DE \ BE) = AE U BE U (Ds \ Bs) = As U Ds.

Clearly, As U DE is a measurable set and
In other words, the preceding show that for each 8 > 0 there exist a measurable
set Fs and a subset G s suc!) that
Now, for each n pick a measurable set FII and a subset G n with fJ.-*(G n) < ~
and E U Gil = Fn. Clearly, the set F = n~l Fn is measurable. Also, the set
G = n~l Gil is a null set-and hence G \ E is also measurable. In view of
00 00
E UG = nCE UG II ) = rTFII = F,
n=l 11=1
we see that E U G is a measurable set. Finally, the measurability of E follows

immediately from the identity
E = (E U G) \ (G \ E) =F \ CG \ E).
Problem 15.9. Let (X, S, fJ.-) be a measure space, and let A be a subset of X.
Show that if there exists a measurable subset E of X such that A ~ E, fJ.-*(E) < 00,
and fJ.-*(E) = fJ.-*(A) + fJ.-*(E \ A), then A is measurable.
Solution. By Problem 15.7, we know that the outer measure generated by the
measure space (E, SE, j.L*) coincides with j.L* and the a-algebra of all measurable
sets ofthemeasure space (E,SE,j.L*) is {A E AIL: AS;; E}.
Now, to complete the proof, assume E E AIL and that a subset A of E satisfies
j.L*(A) + j.L*(E \ A) = j.L*(E). If p,*(E) < 00 holds, then it follows from
Theorem 15.8 that A is a measurable set for (E, SE, j.L*). Thus, by the preceding
discussion, A E AIL"
Problem 15.10. Let A be a subset ofR with)"*(A) > 0. Show that there exists
a nonmeasurable subset B ofR such that B S;; A.
Solution. If A is nonmeasurable, then there is nothing to prove. So, assume that

A is measurable. Since some [n, n + 1] n A must have nonzero measure (why?),
by translating appropriately (and using Problem 15.5), we can also assume that
A S;; [0, 1].
As in Example 15.13 define an equivalence relation :::: on A by saying that
x :::: y whenever x - y is a rational number. By the Axiom of Choice, there exists
a subset B of A containing precisely one member from each equivalence class.
Let {rl' r2, ... } be an enumeration of the rationals of [-1, 1] and let Bn = rn+B.
Then:
a. The sequence {Bn} is pairwise disjoint;
b. )..*(Bn) = )..*(B) holds (by Problem 15.5) for each n; and
c. AS;;U:IBnS;;[-1,2].
Now, note that if B is a measurable set, then each Bn is likewise a measurable
set (see Problem 15.5 again). Thus, from (c), it follows that
00 00
0< )"*(A) :::: )..*(UB;) = I>*(B;)

;=1 ;=1
n
= lim'"' )..*(B;)
lJ-+OO ~
= lim n)"*(B):::: 3,
Il-+OO
;=1
which is impossible. Therefore, B is a nonmeasurable subset of A.
Problem 15.11. Give an example of a disjoint sequence {En 1of subsets of some
measure space (X, S, j.L) such that
00 00
j.L*( U"En ) < Lj.L*(E n).

11=1 n=1
Section 15: TIlE OUTER MEASURE GENERATED BY A MEASURE 129
Solution. Let Ell be the disjoint sequence of nonmeasurable sets described in

Example 15.13, where Ell = I'll + E. Since E is anonmeasurable set, A*(E) > 0
holds, and so A*(E II ) = A*(E) > O. In particular, 2:::1 A*(En) = 00. On the
other hand, U:1 Ell S; [-1,2] implies A*(U:1 Ell) :::: 3 < 00.
Problem 15.12. Let (X, S, f.L) be a measure space, and let {An} be a sequence
of subsets of X such that All S; An+1 holds for all n. If A =
U:1 All, then show
t
that f.L *(AIl) f.L *(A).
Solution. Choose some E E A with A S; E and f.L*(A) =

f.L*(E). (This is
possible by Theorem 15.11.) By the same theorem, for each n there exists some
Ell E A with All S; Ell S; E and f.L*(AII) = f.L*(En). Now, for each n put
FII= n~1I Ek E A, and then let F =
U:1 FII EA. Then, we have:
a. All S; FII and f.L*(AII) =
f.L*(FII) for each n; and
b. FII t F and f.L*(A) = f.L*(F).
By Theorem 15.4, it follows that
Problem 15.13. For subsets of a measure space (X, S, f.L) let us define the fol-
lowing almost everywhere (a.e.) relations:
a. A S; B a.e. if f.L*(A \ B) = 0;
b. A = B a.e. if f.L*(At:.B) = 0;
c. Allt A a.e. if All S; An+1 a.e. for all n and A = U~=1 An a.e. (The
meaning of All t A a.e. is similar.)
Generalize Theorem 15.4 by establishing the following properties for a sequence
{Ell} of measurable sets:
i. If Ell t E a.e., thell f.L *(EII) t f.L *(E).
ii. If Ell t E a.e. and f.L*(Ek) < 00 for some k, thell f.L*(EII) t f.L*(E).
Is (i) true without assuming measurability for the sets Ell?
I
Solution. (i) Assume that {Ell} is a sequence of measurable sets such that Ell t
E a.e. holds. Let
and note that f.L*(A) = O. Now, define F = E U A and Fn = Ell U A for each n.
Clearly, JL*(E) = JL*(F) and JL*(En) = JL*(Fn) for each n (see Problem 14.2)
and Fn t F.
Now, apply Theorem 15.4(1) to get
(ii) Assume that {Ell} is a sequence of measurable sets such that En .J, E
a.e. and JL*(Ek) < 00 holds for some k. Define B [(n:l En)~E] U=
[U:l(E II + 1 \ En)]. Clearly, JL*(B) =
O. Now, apply Theorem 15.4(2) to
En U E U B .J, E U B.
Statement (i) is also true without assuming measurability for the En. This
follows from the arguments of (i) previously and Problem 15.12.
Problem 15.14. Give an example 0/ a sequence {En} a/measurable sets 0/ some

measure space (X, 5, JL) such that En+l .£ En holds/or alln and
Solution. Consider 1R with the Lebesgue measure, and let Ell = (n, 00) for each
n. Then, Ell .J, C/J holds, while A*(En) = 00 for each n.
Problem 15.15. For a sequence {An} 0/ subsets 0/ a set X define
liminf An = un
00 00
n=1 i=n
Ai and lim sup An = nu
00
n=1
00
i=1I
Ai.
Now, let (X, 5, JL) be a measure space and let {En} be the sequence a/measurable
sets. Show the/allowing:
a. JL*(liminfE n):::: liminfJL*(En).
b. 1/ JL*(U:l En) < 00, then JL*(lim sup En) 2: lim sup JL*(En)·
Solution. (a) Note that n:::n Ei t lim inf En and n:::n Ei .£ Ell holds for each
n. By Theorem 15.4(1), we get
00
JL*(lim inf En) = lim JL*

n-...oo
(n
Ei) :::: lim inf JL*(EII)'
. i=n
(b) Use similar arguments and Theorem 15.4(2).

Section 15: THE OUTER MEASURE GENERATED BY A MEASURE 131
Problem 15.16. Give an example ofa sequence {An} ofsubsets of some measure
space (X, S, M) such that A n+ 1 ~ An holds for each n, M*(AI) < 00, and
00
n~~M*(An) > M*(n All)'

11=1
Solution. Let An = U~n Ei, where {En} is the sequence of Example 15.13.
Note that An t cj; holds. Indeed, if x E An, then x E Ek for some k ~ n. Since
Ei n E j = cj; whenever i =1= j, it follows that x ¢. A k+1 so that An t cj; holds.
Now, observe that A*(An) ~ A*(En) = A*(E) > 0 holds for all n.
Problem 15.17. Let (X, SI, MI) and (X, S2, M2) be two measure spaces. Show
that MI and M2 generate the same outer measure on Xif and only if MI = Mi on
SI and M2 = Mi on S2 both hold.
Solution. If MI and M2 generate the same outer measure, then clearly MI = Mz

on SI and M2 = Mi on S2 both hold.
For the converse, assume that MI = Mz on SI and M2 = M; on S2 both hold.
Let A ~ X. If Mi(A) = 00, then Mi(A) S Mi(A) holds. If Mi(A) < 00, then
given e > 0 there exists a sequence {An} of S2 such that A ~ U:I An and
2:::1 Mz(A II ) < Mi(A) + e. Thus,
00 00
Mi(A) sL MTCAn) =L M2(An) < Mi(A) +e

11=1 n=1
holds for all e > 0, and so Mi(A) S Mi(A).

Similarly, Mi(A) ~ Mi(A) holds, and therefore MiCA) = Mi(A) holds for all
A~X.
Problem 15.18. Let (X, S, M) be a measure space. A measurable set A is called

an atom if M*(A) > 0 and for eve,) measurable subset E of A we have either
M*(E) = 0 or M*(A \ E) = O. If (X, S, M) does not have any atoms, then it is
called a nonatomic measure space.
a. Find the atoms of'
i. the counting measure, and
ii. the Dirac measure based at a point a.
b. Show that the real line with the Lebesgue measure is a nonatomic measure
space.
Solution. a. (i) The atoms of the counting measure on a set are precisely the
one-point sets.
(ii) The atoms of the Dirac measure based at a point a are precisely the sets
containing the point a.
b. Let A ~ lR be measurable with)... *(A) > 0. Pick some integer n such
that )...*([n, n + 1] n A) = 0 > 0. Subdivide [n, n + 1] into a finite number of
subintervals all of the same length less than o. For one of them, say I, we must
have
)..."'([n,n+l]nAnI) >0.
Now, note that the set E = [n, n + 1] nAn I ~ A is measurable and satisfies
° < )...
*(E) < 0 :::: )... *(A). This shows that A is not an atom, and hence lR
with the Lebesgue measure is nonatomic. (For more about this problem, see
Problem 18.19.)
Problem 15.19. This exercise presents an example of a measure that has in-
finitely many extensions to a measure on the a-algebra generated by S. Fix a
proper nonempty subset A of a set X (i.e., A =1= X) and consider the collection of
subsets S = {<;b, A}.
a. Show that S is a semiring.
b. Show thatthe setfunction p.,: S-7[O, 00] defined by f..L(<;b) = Oand f..L(A) = 1
is a measure.
c. Describe the CaratheodOlY extension f..L* of f..L.
d. Determine the a-algebra of measurable sets AI-"
e. Show that f..L has uncountably many extensions to a measure on the a-
algebra generated by S. Why doesn't this contradict Theorem 15.IO?
Solution. The validity of (a) and (b) should be obvious.

(c) The Caratheodory extension of f..L is given by
f..L*(B) =
. °
1001
if B = <;b;
if B =1= <;b and B ~ A;
if BS?;A.
(d) The a-algebra generated by S is
(e) If a is any non-negative extended real number, then the set function
v: A -7 [0,00], defined by
v(<;b) = 0, v(A) = 1, v(A C ) = a, and veX) = 1 + a,

Section 16: MEASURABLE FUNCTIONS 133
is a measure which is an extension of J.L to all of A. This shows that there are
uncountably many extensions of J.L to the a-algebra generated by S.
The latter conclusion does not contradict Theorem 15.10 because J.L is not a
a-finite measure.
16. MEASURABLE FUNCTIONS
Problem 16.1. Let (X, S, J.L) be a measure space. For a function f: X -')- JR
show that the following statements are equivalent:
a. f is a measllrablefunction.
b. f-'((-oo,a» is measurable for each a EJR.
c. f-I((a,(0» is measurable for each a E lR.
Solution. (a)==} (b) Note that f-I (( -00, a») is a measurable set simply because
the interval (-00, a) is an open set.
tb)==}(c) Observe that the identity
f-I((-oo, aJ) = n
00
11=1
f-I((-oo, a + ~»)
implies that f-I (( -00, a J) is a measurable set for each a E JR. Consequently,
the set f-I ((a, (0») = X \ f-I (( -00, aJ) is also measurable for each a E JR.
(c)==}(a) Clearly, f-I((-oo, aJ) = X \ f-I((a, (0») is measurable for
each a E JR. Thus, by condition (5) of Theorem 16.2, the function f is measur-
able.
Problem 16.2. Let (X, S, J.L) be a measure space, and let A be a dense subset
of lR. Show that a function f: X -')- JR is measurable if and only if the set
{x EX: f(x):::: a} is measurable for each a EA.
Solution. Only the "if" part needs proof. Let a E JR. Since A is dense in JR,
there exists a sequence {all} of A with all < a for each n and all t a. Now,
note that. the identity
f-I ([a, ~») = n 00
11=1
f-I ([all, (0»)
shows that the set f-I ([a, (0») is measurable. Therefore, by Theorem 16.2, the
function f is measurable.
Problem 16.3. Give an example of a nonmeasurable function f such that If I is

a measurable function and f-I ({a}) is a measurable set for each a E 1R.
Solution. Take a non-Lebesgue measurable subset E of [0, 1] and consider the

function f: [0, 1] ~ 1R defined by
X if x E E;
f(x) =:= { _~, if x E [0, 1] \ E.
It is straightforward to verify that the function f satisfies the desired properties.
Problem 16.4. Show that iff: 1R -+ 1R is continuous a.e., then f is a Lebesgue

measurable function.
Solution. Let f: 1R ~ 1R be a function that is continuous almost everywhere.

Put E = {x E 1R: f is continuous at x} and note that A*(1R \ E) = 0. Hence,
1R \ E and E are both measurable sets.
Now, let 0 be an arbitrary open subset of 1R. Clearly, the set f- I (O)n(1R \ E)
(as a null set) is measurable. Since f restricted to E is continuous, f-I (0) n E
is an open set in E, and consequently there exists an open subset V of 1R such
that f-I(O) n E = V n E. In particular, note that f-I(O) n E is a measurable
set. Therefore,
is likewise measurable, so that f is a measurable function.
Problem 16.5. Let f: 1R -+ lR be a differentiable function. Show that f' is

Lebesgue measurable.
Solution. For each n define
gn(x) = n[J(x + ~) - f(x)] = f(x+~tf(X),

n
and note that each gn is measurable (since it is continuous). In view of gil (x) -+
f' (x) for each x E 1R, it follows from Theorem 16.6 that f' is a measurable
function.
Problem 16.6. Let (X, S, fL) be a measure space and let f: X -+ 1R be a mea-
surable function. Show that:
a. Ifl P is a measurable function for all p 2:: 0, and
b. if f(x) =1= 0for each x E X, then l/f is a measurable function.
Section 16: MEASURABLE FUNCI10NS 135
Solution. Let I: X --+ lR be a measurable function.

(a) Assume p > O. By Theorem 16.5, III is measurable. The conclusion
now follows from the identities {x E X: I/IP(x) ::: a} = X if a :s 0, and
• I
{x EX: I/IP(x)::: a} = {x EX: I/(x)1 ::: a p } if a> O.
(b) Assume that I(x) =1= 0 holds for each x EX. Note that
{x EX: j(x) > O} = {x EX: I(x) > o},

{x EX: j(x) > a} = {x EX: I(x) < ~} if a> 0, and
{x EX: j(x) > a} = {x EX: I(x) < ~} U {x EX: I(x) > O} if a < O.
The preceding identities guarantee that j is measurable.
Problem 16.7. Let {f,,} be a sequence 01 real-valued measurable lunctions on a

measure space (X, S, f.L). Then show that the sets
a. A = {x EX: IIl(x) -+ ooJ,
b. B = {x EX: In(x) -+ -ooJ, and
c. C = {x E X: lim In (x) exists in lR}
are all measurable.
Solution. (a) For each m and k let A m.k = {x EX: In(x) ::: k for all n ::: m}.
From Am.k = n:m {x E X: f,,(x) ::: k}, we see that A m •k E 11./1. for each m, k.
Now, note that A = n:l U:::'=l
Am.k.
(b) Put Bm.k = {x E X: f,,(x) :s -k for each n 2: m} and note that B =
n:l U:::'=l Bm•k •
(c) Let Y = X \ (A U B) and consider the measure space (Y, Sy, f.L*). Also,
consider all functions restricted to Y. In view of Problem 15.7, all functions are
measurable with respect to this space. By Theorem 16.6, both functions lim inf f"
and lim sup In are measurable. The conclusion now follo.ws from Theorem 16.4(c)
by observing that
C = {x EX: lim f,,(x) exists in lR}
= {x E Y: lim sup f,,(x) = liminf In (X) }.
Problem 16.8. Let (X, S, f.L) be a measure space. Assume that I: X -+ lR is a

measurablelunction and g: lR -+ lR is a continuousftmction. Show that go I is
a measurable ftmction.
Solution. Consider the functions X lR ~ lR with I measurable and g

continuous, and let V be an open subset of lR. Since g is continuous, we know
that g-I«()) is an open set, and the conclusion follows from the identity
Problem 16.9. Let F be a nonempty family of continuous real-valued functions

defined onlR.. Assume that there exists a junction g: lR. -+ lR. such that f (x) ::: g(x)
for each x E lR. and all f E :F. Show that the supremum fUllction h: lR. -+ lR.,
defined by h(x) = sup{f(x):. f E F}, is (Lebesgue) measurable.
Solution. We shall show that h- I (a, (0)) is an open set for each a E lR. (and
hence, a Lebesgue measurable set).
To see this, let a E lR. and fix Xo E Jz -I (a, (0)), i.e., h(xo) > a. So, there exists
some f E F such that f(xo) > a. Since f is a continuous function, there exists
some neighborhood V of Xo such that f (x) > a for each x E V. This implies
h(x) ::: f(x) > a for each x E V, and so V S;; h- I (a, (0)). This shows that Xo
is an interior point of h -I (a, (0)) and consequently, h -I (a, (0)) is an open set.
Note: A real-valued function f: X -+ lR. defined on a topological space X is
said to be lower semicontinuous if f- I (a, (0)) is an open set for each a E lR..
The preceding arguments show that we have proven the following result: The
pointJ.1lise supremum of a family of lower semicontinuous junctions is likewise
lower semicontinuous.
Problem 16.10. Show that iff: X -+ lR. is a measurable function, then either f
is constant almost everywhere or else (exclusively) there exists a constant c such
that
J.L*({x EX: f(x) > cD > 0 and J.L*({x EX: f(x) < cD > O.
Solution. Let f: X -+ lR. be a measurable function which is not a constant almost

everywhere. Assume first that f(x) ::: 0 holds for each x E X and let
Co = sup{c E lR.: J.L*({x EX: f(x) ::: c) = OJ.

Clearly, 0 ::: Co < 00 and J.L*({x EX: f(x) < col) = O. Since f is not constant
almost everywhere, there exists some c > Co such that J.L*({x E X: f(x) >
cl) > o. Now, if k satisfies Co < k < c, then by the definition of Co we have
J.L*({x E X: f(x) < cl) ::: J.L*({x E X: f(x) ::: k}) > 0, and the desired
conclusion is established in this case.
In the general case, either f+ or f- is not equal to a constant almost everywhere.
We consider the case where f+ is not equal to a constant almost everywhere (the
other case can be treated in a similar fa~hion). By the preceding case, there exists
c > 0 with J.L*({x E X: f+(x) > c)) > 0 and J.L*({x E X: f+(x) < c)) > O.
Section 17: SIMPLE AND STEP FUNCTIONS 137
To finish the proof, notice that
(x E X: f+(x) > c) = (x EX: f(x) > c)
and (x EX: f+(x) < c) = (x E X: f(x) < C}.
17. SIMPLE AND STEP FUNCTIONS
Problem 17.1. For subsets A and B of a set X, establish the following state-
ments:
1. XcjJ = 0 and Xx = l.
2. A S; B if and only if
XA ::: XB·
3. XAnB = XA . XB = XA 1\ XB·
4. XAUB = XA + XB - XAnB = XA V XB·
5. XA \ B =
XA - XAnB·
6. If A = U~= I All and (All) is a pairwise disjoint sequence of subsets of
X, then XA = :L:I XA,,·
7. XAx B = XA • XB· (Here the set B can be considered to be a subset of some
other set Y.)
Solution. The proofs of the statements are straightforward. To indicate how one
can prove them, we shall establish the validity of statements (3) and (7).
(3) We have
I if x E An B
XAnB(X) = { 0', if x ~ A n B
1, if x E A and x E B
= [ 0, if x ~ A
0, if x ~ B
= XA(X)' XB(X)
= XA . XB(X)
= min(XA(x), XB(X)}.
(7) Note that
1 if (x, y) E A x B
XAxB(X, y) = { 0: if (x, y) ~ A x B
1, if x E A and y E B
= 0, if x ~ A
[
0, if y ~ B
= XA(X)' XB(Y)·
Problem 17.2. Let ¢ be a step function and 1/1 a simple function such that
o ::: 1/1 ::: ¢ a.e. Show that 1/1 is a step function.
Solution. Let E = {x: 0 ::: 1/I(x) ::: ¢(x)}, and observe that p..*(X \ E) = o.
If F = {x E X: ¢(x) > O}, then the measurable set A = (X \ E) U F satisfies
p..*(A) < 00, and 1/I(x) = 0 for each x EX \ A.
Problem 17.3. Show that If(X, S, p..) isafinite measure space, then evelysimple
function is a step function.
Solution. If ¢ is a simple function and E = {x EX: ¢(x)::j:. O}, then note that
p..*(E) ::: p..*(X) < 00 holds.
Problem 17.4. Give an alternate proof of the linearity of the integral (Theo-
rem 17.2) based on Problem 12.14.
Solution. The linearity follows immediately from the following property.
• 1f ¢ is a step function and ¢ = :Lj~l b j XBj is an arbitral}' representation

of¢, then
m
1(¢) = Lbjp..*(B j ).
j=J
We shall establish the preceding property below.

To this end, let ¢ = :L~=J ai XAj be the standard representation of ¢. Assume
first that the B j are pairwise disjoint. Since neither the function ¢ nor the sum
:Lj~Jbjp..*(Bj) changes by deleting the terms with b j = 0, we can assume that
bj ::j:. 0 for each}. In such a case, we have U~=J Ai = U~J B j. Moreover, note
that aip..*(Ai n Bj) = bjp..*(A i n Bj) for all i and}. Indeed, if Ai n Bj = 0 the
equality is obvious and if x E Ai n B j, then ai = b j = ¢(x). Therefore,
n m n
/(¢) = Laip..*(A i ) = L Laip..*(A i n Bj)
i=l i=j i=J
m 11 m
= L Lbjp..*(A i n Bj ) = Lbjp..*(Bj}.
j=J i=l j=J
Now, consider the general case. 'By Problem 12.14, there exist pairwise disjoint
measurable sets C J, ... , C k such that each C i is included in some B j and B j =
UfC i : C i S; Bj }. For each i and} let8! = 1 ifC i S; Bj and8! = OifCi ~ Bj.
So, by the preceding case, we have
k m m k m
1(4)) = L[2:)jo! ]fJ-*(C;) = L bj [ LO! fJ-*(C;)] = LbjfJ-*(Bj ).
;=1 j=1 j=1 ;=1 j=1
Problem 17.5. Show that 11(4))1 .:::: 1(14)1) holds for every step function 4>.
Solution. From -14>1 .:::: 4> .:::: 14>1 and the monotonicity of the integral (Theorem
17.3), it follows that
-1(14)1) = 1(-14>1)':::: 1(4)).:::: 1(14)1),
and so 11(4))1.:::: 1(14)1) holds.
Problem 17.6. Let 4> be a step function such that 1(14) I) = O. Show that 4> =0
a.e. holds.
Solution. Let 4> = L::=I a; XAj be the standard representation of 4>. Then, note
that 14>1 = L:;'=I la;\XA is a representation of 14>1, imd therefore
j
n
0= 1(14)1) = L la;\fJ-*(A;).
;=1
Since lad > 0 holds for each 1 .:::: i .:::: n, it follows that fJ- *(A;) = 0 for each
1 .:::: i .:::: n, and so 4> = 0 a.e. holds.
Problem 17.7. Let 4> be a step function. Let A = {x E X: 4>(x) i= O} and

M = max{I4>(x)l: x E X}. Show that 11(4))1.:::: MfJ-*(A).
Solution. Apply the monotonicity of the integral (Theorem 17.3) to the inequality
-MXA .:::: 4>.:::: MXA'
Problem 17.8. Let {4>II} be a sequence of step functions. Show that if4> is a step
function and 4>11 ,!.. 4> a.e. holds, then 1(4)11) ,!.. 1(4)) also holds.
140 Chapter 3: TIIE TIIEORY OF MEASURE
Solution. If ¢n t ¢ a.e. holds, then ¢n - ¢ t 0 a.e. likewise holds. Thus, by

the order continuity of the integral (Theorem 17.4),1 (¢n) - 1 (¢) = l(¢n -¢) to
so that l(¢n) t 1(¢).
Problem 17.9. Let {¢n} be a sequence of step junctions and ¢ a simple junction
such that 0 ~ ¢n t ¢ a.e. holds. Show that if lim 1 (¢n) < 00, then ¢ is a step
junction.
Solution. Assume ¢(x) 2: 0 for each x and let ¢ = 2:7=1 aiXAj be the standard
representation of ¢. Now, let i be fixed. Then, for each n the function Vrn =
¢n A ai XAj is a step function, Vrn ~ ¢n holds, and Vrn t n ¢ A ai XAj = ai XAj a.e.
By Theorem 17.6, we see that
o ~ ailL*(Ai) = lim l(Vrn) ~ lim l(¢n) <

n-",CX) n-+oo
00,
and so 1L*(Ai) < 00 holds for each 1 ~ i ~ k. That is, ¢ is a step function.
Problem 17.10. Let (X, S, IL) be a measure space, and let f: X --+ 1R be a
function. Show that f is a measurable function if and only if there exists a sequence
{¢n} of simple functions such that lim ¢n(x) = f(x) holds for all x E X.
Solution. Assume f to be measurable. Then both f+ and f- are measurable

functions. By Theorem 17.7 there exist two sequences of simple functions {sn}
and {tn} with 0 ~ sn(x) t f+(x) and 0 ~ tn(x) t f-(x) for each x EX. Now,
note that ¢// = Sn - tn satisfies ¢//(x) -+ f(x) for all x.
For the converse, note that (by Theorem 16.6) the pointwise limit of a sequence
of measurable functions is always a measurable function.
Problem 17.11. Let (X, S, IL) be a CJ -finite measure space, and let f: X --+ 1R
be a measurable function such that f(x) 2: Ofor all x EX. Show thatthere exists
a sequence {¢n} of step functions such that 0 ~ ¢n t f(x) holdsforal! x EX.
Solution. By Theorem 17.7 there exists a sequence {Vrn} of simple functions

satisfying 0 ~ Vrn(x) t f(x) for all x E X. Now, pick a sequence {En} of
measurable sets with IL *(En) < 00 for each n, and En t X. Let ¢n = Vr// A XEn'
Then, {¢n} is a sequence of step functions satisfying 0 ~ ¢n (x) t f (x) for each
x E X.
Problem 17.12. Give a proof of the order continuity of the integral, i.e., ¢// t 0
a.e. implies I(¢n) t 0, based on Egorov's Theorem 16.7.
Solution. Assume that {<PII} is a sequence of step functions of some measure

space (X, 5, f-L) satisfying <PII to a.e. Without loss of generality, we can suppose
that <PII(X) ,j, 0 for each x E X. Let E = (x E X: <PI(X) > O} and note that
f-L*(E) < 00. Also, let M = max!<PI(x): x EX}.
Now, let E >0. By Egorov's Theorem 16.7 there exists a measurable set F ~ E
such that f-L*(F) < E and {<PII} converges uniformly to zero on E \ F. So, there
exists some k such that 0 ~ <PII(X) < E for all x E E \ F and all n ::: k. Thus, for
n ::: k, we have
o ~ <PII ~ EXE\F + MXF ~ EXE + MXF,

and consequently, by the monotonicity of the integral
for all n ::: k. This shows that 1(<PII) ,j, O.
Problem 17.13. Let (X, 5, f-L) be a measure space, and let f: X -+ [0, (0) be
a function. Show that f is measurable if and only if there exist non-negative
constants CI, C2, ... and measurable sets E I, E2, ... such that
00
f(x) = I:>IIXE,,(X)
11=1
holds for each x EX.
Solution. Consider a measure space (X, 5, f-L) and a non-negative real-valued

function f: X -+ [0, (0). Assume first that there exist non-negative constants
CI, C2,.·· and measurable sets E I , E2, ... such that f(x) = L::I cllxdx)
holds for each x EX. If we let <PII = L:;'=I Ci XE" then <Pn is a measurable
function (in fact, it is a simple function) fmd <Pn(x) ---7 f(x) holds for each
x E X. Now, by Theorem 16.6(1), the function f is necessarily a measurable
function.
For the converse, assume that f is a measurable function. By Theorem 17.7
there exists a sequence of simple functions {<Pn} such that 0 ~ <Pn (x) t f (x) for
each x EX. If we let <Po = 0, then f(x) = L::I [ <Pn(x) - <PII-I (x)] holds for
each x EX. For each n write <PII - <Pn-I = L:7:::1 <' XEf with cj' ::: 0 for each i
and n. Thus,
co kl/
f(x) = LL'>i'XEj',
11=1 i=1
and by rearranging the terms of the preceding series in a single series, our conclu-
sion follows.
Problem 17.14. Let (X, S, JL) be afinite measure space satisfying JL*(X) = 1,
t
and let E\. E 2 , ••• , EIO be ten measurable sets such that f.J.*(E;) = holds for
each i. Show that four of these sets have an intersection of positive measure. Is
the conclusion true for nine measurable sets instead often?
Solution. Consider the step function ¢ = L:~I XE;. Clearly, the function ¢
assumes only integer values and
¢(x) = the cardinality of the set {i E {1, ... , lO}: x E E;}.
If ¢(x) :s: 3 = 3 Xx (x) for almost all x, then

10 10
3< ~ = LJL*(E;) = LI(XE;) = I(¢):s: I(3Xx) = 3
;=1 ;=1
a contradiction. Hence, the measurable set
A = {x EX: ¢(x) 2: 4}
must have positive measure.

Next, let AI, A2, ... , Ak denote the collection of all (nonempty) intersections
of the sets E; taken four at a time; clearly, k :s: (~) = 210. Now, an easy argument
guarantees that A ~ U~=1 A j, and from this it easily follows that at least one of
the A j must have positive measure.
For nine sets the conclusion is false. For a counterexample take X = [0, 1],
JL = A, EI = E2 = E3 = (0, t), E4 = E5 = E6 = G, D,and E7 = E8 = E9 =
(~, 1).
Problem 17.15. Iff: X -+ [0, 1] is a measurablefunction, then show that either

f = XA a.e. for some measurable set A or else (exclusively) there exists a constant
° < c < ~ such that
JL*({x EX: c < f(x) < l-e}) > 0.
Solution. For each nlet An = {x EX: 1

2n < f(X) < 1-:lid- °
If JL*(An) > for
some n, then the constant e = in satisfies JL*({x EX: c < f(x) < 1 - c}) > 0.
Now, assume that p,*(AII) = 0 for each n. Then from

All t (x EX: 0 < f(x) < 1},
we see that p,*({x E X: 0 < f(x) < 1}) = O. This easily implies that f = XA
a.e. for the measurable set A = f- l ({ I}).
Problem 17.16. Let (X, S, p,) be a measure space, and let ¢: X -+ lR. be a
simple function having the standard representation ¢ = L~=l ai XAi' If ¢ 2: 0 a.e.,
then the sum L~=laip,*(Ai) makes sense as an extended real number (it may be
infinite). Call this extended real number the Lebesgue integral of ¢, and write
I(¢) = L~=laip,*(Ai)'
a. If ¢ and 1/1 are simple functions such that ¢ 2: 0 a.e., then 1/1 2: 0 a.e., then
show that I(¢ + 1/1) = I(¢) + 1(1/1).
b. If ¢ and 1/1 are simple functions such that 0 ::S ¢ ::S 1/1 a.e., then show that
I(¢) ::s 1(1/1).
c. Show that if {¢II} and {1/I1I} are two sequences of simple functions and
f: X -+ lR.* such that 0 ::S ¢II t f a.e. and 0 ::S 1/111 t f a.e., then
lim I (¢II) = lim I (1/111) holds (with the limits possibly being infinite).
d. Assume that {¢II} is a sequence of simple functions such that 0 ::S ¢II t XA
a.e. holds. Show that lim I (¢n) = p, *(A).
e. Give an example of a sequence {¢II} of simple functions on some measure
space sllch that ¢n to (everywhere) and lim I(¢n) =1= O.
Solution. Clearly, a simple function ¢ is a step function if and only if I(¢) < 00.
(a) Note that ¢ + 1/1 is a step function if and only if both ¢ and 1/1 are step
functions. In this case, the equality I(¢ + 1/1) = I(¢) + 1(1/1) follows from
Theorem 17.2. On the other hand, if ¢ + 1/1 is not a step function, then either ¢ or
1/1 fails to be a step function and hence, in this case, I (¢ + 1/1) = I (¢) + I (1/1) = 00
holds.
(b) If 1(1/1) = 00, then I(¢) ::S 1(1/1) holds trivially. On the other hand, if
1(1/1) < 00, then 1/1 is a step function. It follows (from Problem 17.2) that ¢ is a
step function, and the desired inequality follows from Theorem 17.3.
(c) If both [¢n} and {1/In} are sequences of step functions, then the conclusion
follows from Theorem 17.5. Thus, we only need to consider the case when {¢n}
is a sequence of step functions and I (1/Ik) = 00 holds for some k.
In view of ¢n /\ 1/Ik tn f /\ 1/Ik = 1/Ik a.e., it follows from Problem 17.9 that
limn->oo I(¢II /\ 1/Id = 00. From ¢n /\ 1/Ik ::S ¢n, we obtain that liinI(¢II) = 00.
Hence, lim l( 1/111) = lim I (¢n) = 00 holds in this case.
(d) We can suppose 0 ::S ¢II(X) t XA(X) holds for each x. If for each n we let
All = [x EX: ¢n (x) > OJ, then each An is measurable and An t A holds. Since
XAn t XA, part (c) coupled with Theorem 15.4 gives
(e) Consider IR with the Lebesgue measure, and let </111 = X(II,oo)'
Problem 17.17. Let (X, :E, JL) be a measure space with :E being a a-algebra.
Let us say that a function f:X -+ IR. is :E-measurable if f-I (A) E :E for each
open subset A of IR. Also, let ML denote the collection of all :E-measurable
functions. Establish the following:
a. ML is afunction space and an algebra offunctions.
b. ML is closed under sequential pointwise limits.
c. If JL is a-finite and f: X -+ IR is a measurable function, then there exists
a :E-measurablefunction g: X -+ IR such that f = g a.e.
Solution. (a) In order to show that ML is closed under addition and multiplica-
tion, we need the following properties among :E-measurable functions f and g:
The sets
1. {x EX: f(x) > g(x)},
2. {x EX: f(x) 2: g(x)}, and
3. {x EX: f(x) = g(x)}
all belong to :E. To see (1), let 1'1,1'2 • ••• be an enumeration of the rational numbers
of 1R, and note that
J.
00
{x EX: f(x) > g(x)} = U[ {x EX: f(x) > rn} n {x EX: g(x) < I'll}
n=1
which belongs to :E, since it is a countable union of sets from the a -algebra :E. For
(2), note that {x EX: f(x) 2: g(x)} = (x EX: g(x) > f(xW, which belongs to
:E by (1). Finally, for (3), observe that
{x EX: f(x) = g(x)} = {x EX: f(x) 2: g(x)} n {x E X: g(x) 2: f(x)},
which belongs to :E by (2).

To complete the proof of part (a), we shall establish that for :E-measurable
functions f and g, the following statements hold:
i. f + g is a :E-measurable function.
ii. f g is a :E-measurable function.
lll. If I, f+, and f- are :E-measurable functions.

iv. f V g and f /\ g are :E-measurable functions.
The proofs of these claims are given below. .

(i) Note first that if e is a constant number, then e - g is a :E-measurable function.
[Reason: If a E JR, then (x EX: e - g(x) 2: a} = (x EX: g(x) ::: e - a} E :E.]
Now, if a E JR, then the set
(f + g)-I ([a, 00)) = {x E X: f(x) + g(x) 2: a} = {x E X: f(x) 2: a - g(x)}
belongs to :E by the preceding observation and (2). This implies (how?) that f + g
is a :E-measurable function.
(ii) Note first that f2 is a :E- measurable function. To see this, let a E R
Then {x E X: f2(x) ::: a} = 0 if a < 0 and {x E X: f2(x) ::: a} =
f- I ([ -.j(i, .j(i]) if a 2: O. This implies that f2 is a :E-measurable function.
Also, if e is a constant, then ef is measurable. [Reason: If A = (x EX: ef(x) 2:
a}, then A = {x EX: f (x) 2: a/ e} for e > 0 and A = {x EX: f (x) ::: a / e} for
e < 0.] The result now follows from the preceding observations combined with
(i) and the relation
(iii) The :E-measurability of If I follows from the relation
{XEX: If(x)l:::a}=0 ifa<O,
and
{x EX: If(x)l::: a} = {x EX: f(x)::: a} n (x EX: f(x) 2: -a} if a 2: O.
For the :E-measurability of f+ and f- use the identities
f+ = Hlfl + f) and f- = Hlfl - f).
(iv) The identities
fvg=Hf+g+lf-gl) and f/\g=4(J+g-lf-gl)
show that f V g and f /\ g are :E-measurable functions.

(b) Assume that Un} is a sequence of :E-measurable functions such that

fn(x) -+ f(x) holds for each x EX. Observe that the equality
00 00
f-I(a, 00)) = U nfi-I((a +~, 00))

11=1 i=n
and the :E-measurability of each fi show that f- I (a, 00)) belongs to:E. This
implies that f is a :E-measunible function.
(c) We can assume f (x) :::. 0 for each x E X (otherwise, we apply the arguments
below to f+ and f- separately). Assume first that f = XA for some A E :E.
Since fL is a-finite, it follows from Theorem 15.11 that there exists a fL-null set C
such that B = AU C E :E. So, if g = XB, then g is :E-measurable and f = g
fL-a.e. It follows that if <P is a fL-simple function, then there exists a :E-simple
function 0/ such that 0/ = <P fL-a.e.
Now, by Theorem 17.7, there exists a sequence {<PII }of simple functions such that
<PII(X) t f(x) for each x EX. Replacing each <PII by a :E-simple function 0/11 (as
above) we have Vln(X) t f(x) for fL-almost all x. So, there exists a fL-measurable
set E such that o/n(x) t f(x) for each x rt E. Now, use Theorem 15.11 to select
a set F E :E with E f; F and fL*(F) = O. Clearly, o/lI(X)XFc(X) t f(X)XFc(X) =
g(x) for each x E X. By part (b), g is :E-measurable and satisfies g = f fL-a.e.
18. THE LEBESGUE MEASURE
Problem 18.1. Let I = n;~1 Ii be an inten1al ofJRII. Show that I is Lebesgue

measurable and that AU) = n;'=1
IIi I. where IIi I denotes the length of the interval
Ii'
Solution. The verification of the formula can be done by cases as in Problem 15.3.
To show this, we establish the formula for two cases, and leave the rest for the
reader.
The first case is when Ii = [ai, bi], where -00 < ai < bi < 00 holds for each
-1 ::: i ::: n. Then, n;~I[ai' bi + t)
tk ni'=IIi = I. Thus, from Theorem 15.4, it
follows that
n n
AU) lim A(TI[ai , bi + ~))= lim TI(bi -
= k->oo ai +~)
k k->oo k
i=1 i=1
n n
= TI (bi - ai) ='TI 11;1.
i=1 i=1
The second case is when I = [a, 00) x [a2, b2] x '" x [all, bn]' Then, note
that [a, a + k] x [a2' b2] x ... x [an, bn] tk l. Taking into account the preceding
Section 18: TIlE LEBESGUE MEASURE 147
case, it follows from Theorem 15.4 that
= k_oo
lim k . (b2 - a2) ... (b ll - an) = 00 = nlId·
00
i=1
Problem 18.2. Let 0 be an open subset oflR.. Show that there exists an at-most
countable collection {fa: Q! E A} of pairwise disjoint open intervals such that
0= UaEA la. Also, show that A(O) = LaEA Ilal·
Solution. Let 0 be an open subset of JR.. By part (g) of Problem 6.11, we know
that there exists an at-most countable collection {fa: Q! E A} of pairwise disjoint
open intervals such that 0 = UaEA la.
Now, using the fact that the length of each la coincides with its Lebesgue
measure (Problem 15.3), we see that
A(O) = A(U la) = I>Ua) = L Ilal·

aEA aEA aEA
Problem 18.3. Show that the Borel sets of JR." are precisely the members of the
a-algebra generated by the compact sets.
Solution. Let C denote the a -algebra generated by the compact sets. Since every
compact set is closed (which is the complement of an open set), it follows that
C ~ B. On the other hand, if C is a closed set and CII = (x E C: dCO, x) ::: n},
then {CII} is a sequence of compact sets satisfying CII t C. This implies that C
contains all the closed sets (and hence, all the open sets). Thus, B ~C also holds,
and so B = C.
Problem 18.4. Show that a subset E of JR." is Lebesgue measurable if and only
if for
each E > 0 there exists a closed subset F of lR." such that F ~ E and
A(E \F) < E.
Solution. Assume that E is Lebesgue measurable and let s > O. Since E C

is also Lebesgue measurable, there exists an open set V such that E C ~ V and
A(V \ E C ) = A(E n V) < s. Then, the closed set C = V C satisfies C ~ E
and A(E \ C) = A(E n V) < s. For the converse, either reverse the preceding
arguments and use Theorem 18.2, or else use Problem 14.8.
Problem 18.5. Show that if a subset E of [0,1] satisfies ACE) = 1, then E is

dense in [0, 1].
Solution. Let I be a (nonempty) subinterval of [0, 1]. If I n E = 0, then

we have A(E) + A(l) = A(E U 1) ::: A([O, 11) = 1, and hence, in this case,
A(E) ::: 1 - A(l) < 1 holds, which is a contradiction. Thus, I n E =j:. 0 holds for
each subinterval I of [0, 1], and so the set E is dense in [0, 1].
Problem 18.6. If E £; ]Rn satisfies A(E) = 0, then show that EO = 0.

Solution. If V is a nonempty open set with VeE, then note that 0 < A(V) :::
A(E) holds. Therefore, the open set EO must be empty.
Problem 18.7. Show that if E is a Lebesgue measurable subsetof]Rn, then there

exist an Fa-set A and a Gs-set B such that A £; E £; Band A(B \ A) = O.
Solution. By Problem 18.4, for each k there exists a closed set C k with C k £; E
and A(E \ Cd < i.
Similarly, by Theorem 18.2,forevery k there exists an open
set Vk with E £; Vk and A(Vk \ E) < i.
Put A = U~1 Ck (an Fa-set) and
B = n~1 Vk (a Gs-set). Clearly, A£; E £; B holds, and in view of
A(B \ A) ::: A(Vk \ Cd = A(Vk \ E) U (E \ Cd)

::: A(Vk \ E) + A(E \ Ck) < f
for each k, we see that A(B \ A) = O.
Problem 18.8. Let {En} be a sequence of non empty (Lebesgue) measurable sub-
sets of[O, 1] satisfying limA(EII ) = 1.
a. Show that for each 0 < E < 1 there exists a subsequence {Ek n } of {Ell}
such that A( n:1
Ek,,) > E.
b. Show that n~n Ek = 0 is possible for each n = 1,2, ....
Solution. Let {En} be a sequence of nonempty Lebesgue measurable subsets of

[0,1] satisfying limA(E n) = 1.
(a) Fix 0 1 - 1;£. Now, consider the measurable sets
E= n:1
Ekn and F = [0, 1] \ E. Then, we have
00
A(F) = A([O, 1] \ E) = A(U([O, 1] \ EkJ)

n=1
00 00 00
= 2::>([0,1] \ EkJ = :L]1 - A(EkJ] < L 1;£ = 1- B.

n=1 n=1 n=1
Hence, A(E) =1- A(F) > 1 - (1 - B) = B.

(b) Let AZ = [0, 1] \ [k~1 , ~], 1 :s k :s n; n ?: 2. Clearly, A(Ak) = 1- holds f.

for each 1 :s k :s nand nX=1 Ak = 0 holds for each n ?: 2. Let Ell denote the
sequence
A 2l' A22' A 3I' A 2'

3 A3
3" .. , A"l' An2" .. , A"II' A"+ 1
1 , ....
Now, note that A(En) ~ 1 and n~1I Ek =0 holds for each n ?: 1.
Problem 18.9. Assume that afunction f: I -7 lR defined on a subinterval oflR

satisfies a Lipschitz condition. That is, assume that there exists a constant C > 0
sllch that If(x) - f(y)1 :s Clx - yl holds for all x, y E I. Show that f carries
(Lebesgue) null sets to null sets;
In particular, if a function f: I -7 lR defined on a subinterval of lR has a
continuous derivative, then show that f carries null sets to null sets.
Solution. Assume that a function f: I ~ lR satisfies the condition of the

problem. Clearly, f is a (unifonnly) continuous function. In particular, note that
if J is a subinterval of I, then f(1) is also a subinterval of lR (see part (g) of
Problem 6.11), and our condition implies (how?) that the length of f(1) is less
than or equal to C times the length of J, i.e., A* (j (1)) :s C A*(1) holds.
Now, let A be a null subset of I and let 8 > O. Pick a sequence ([all, bn ) of I
half-open intervals such that
00 00 00
A ~ U[a" , b,,) and I>*([a ll , bll )) = L(b" all) < 8.

11=1 ,,=1 ,,=1
Hence, f(A) C f(U:I[a ll , bn) n I) = U~=I f([a n , bn) n I), and so by the
preceding
00
A*(j(A)):S A*(Uf([a",bn)nl))
n=1
00 00
:s LA*(j([an , bn) n I)) :s L C(b n - an) < C8.

n=1 n=1
Since 8 > 0 is arbitrary, we see that A*(j (A)) = 0 holds, as desired.

For the second part notice that if [a, b] is a closed subinterval of I, then there
exists some constant M > 0 satisfying If'Ct)1 :s M for all t E [a, b]. Now, if
x, y E [a, b], then there exists (by the Mean Value Theorem) some z between x and
y satisfying f(x) - fey) = l'(z)(x - y). This implies If(x) - f(y)1 :s Mix - yl
for all x, y E [a, b]. So, by the firs t part, f carries null sets of [a, b] to null sets.
Now, fix a sequence {[an, bll ]} of closed subintervals of I such that I =

U:I [all, bnl and let A be a null subset of I. Then A n [all, bnl is a null sub-
set of [all, bnl, and so f(A n [all, bill) is a null subset of JR. Now, notice that the
identity
00 00
f(A) = f(U A n [an, bill) = U f(A n [all, bnl)

n=1 n=1
guarantees that f(A) is a null subset of lR.
Problem 18.10. Show that the Lebesgue measure of a triangle inJR2 equals its
area. Also. determine the Lebesgue measure of a disk in JR2 .
Solution. Start by observing that every line segment has Lebesgue measure zero
(why?). Thus, the Lebesgue measure of a triangle is the same with or without
some of its edges. Also, every triangle is Lebesgue measurable (since without its
edges.it is an open set). Since A is translation invariant, we can assume that all
triangles have one of their vertices at zero. Let T be such a triangle, and let A(T)
denote its area. Following the graphs in Figure 3.1 (from left to right) we see that:
2A(T) = A(T) + A( - T) = A(T) + A(T2) = A(T) + A(TI)

= A(TI U T) = A(P) = A(Q) = A(P) = 2A(T).
y y y
FIGURE 3.1. The Lebesgue Measure of a Triangle

Area OAD = ~r2 sin (~) cos (~)
Ar.ea OAB = ~ r2 tan (~)
8
8
FIGURE 3.2. Thy Computation of the Lebesgue Measure of a Disk
That is, )...(T) = A(T). In particular, this implies that the Lebesgue measure of
any polygon equals its area.
Now, let D be a closed disk of radius r; see Figure 3.2. To compute its Lebesgue
measure, we use the Eudoxus-Archimedes Method of Exhaustion. For each n, let
Pn and Qn be the inscribed and circumscribed regular n-polygons, respectively.
Clearly, PI! £;; D £;; Qn holds. Now, note that
)"'(Pn) = TC/.2 [ sin(Z!:.)]

~n cos(~) ~)"'(D) ~ )...(Qn) = rrr2 [tan(Z!:.)]
~" ,
and so, by letting n -+ 00, we see that
)"'(D) = rrr2 = A(D).

Problem 18.11. If fL is a translation invariant Borel measure on lR", then show
that there exists some c ::: 0 such that fL*(A) = c)...*(A) for all sub~et A oflR".
Solution. By Theorem 18.8, fL = c)... holds on B for some constant c ::: O. Now,
by Theorem 14.10, (cA)* = c)...* holds, and consequently fL*(A) = (cA)*(A) =
cA*(A) for each subset A of lRn.
Problem 18.12. Show that an arbitrary collection of pairwise disjoint measur-

able subsets oflR, each of which has positive measure, is at-most countable.
Solution. Let C be a collection of pairwise disjoint measurable subsets of lR

such that )"'(C) > 0 holds for each C E C. For each n let
Cn = {C E C; )"'(C n [-n, n]) ::: ~},

and note that C = U~I Cn. Now if C I , ... , C k E CIl, then we have
k k
~ :::: I),(Ci n [-n, n]) = A( (UCi ) n [-n, 11]) : : A([-n, 11]) = 2n,
i=1 i=1
and so k :::: 2112 holds. This shows that each Cn is a finite set, and consequently
C is at-most countable.
Problem 18.13. Let G be a proper additive subgroup of JRn • If G is a measur-

able set, then show that A(G) = 0.
Solution. If A(G) > 0, then, by Theorem 18.13, the element zero is an interior
point of G - G. Since G is an additive group, G - G =G holds, and from this
it follows that G = JRn , which is a contradiction.
Problem 18.14. Let f: 1R -+ 1R be additive (i.e., f(x + y) = f(x) + f(y)for

all x, y E JR) and Lebesgue measurable. Show that f is continuous-alld hence,
of the form f(x) = cx.
Solution. Assume f ;:j:. ° and let e > 0. Since f is an additive function,

I1f- 1([0, e1) = f- I ([0, ne1) holds (why?). Thus, if A(J-I ([0, e1)) = 0, then
A(J-I ([-ne, 01)) = A(J-I ([0, ne1)) = I1A(J-I ([0, e1)) = °

holds for each 11, and so A(J-I ([ -ne, l1e1)) = °for all 11. From
f-I([-ne, ne1) t JR,

it follows that A(JR) = 0, which is impossible. Thus, A(J-J ([0, e1)) > 0. Since
f is also measurable, there exists (by Theorem 18.13) some 8 > with °
That is, -8 < x < 8 implies -e :::: f(x) :::: e so that f is continuous at zero.
Now apply Lemma 18.7.
Problem 18.15. Show that an arbitrary union of proper intervals of 1R is a

Lebesgue measurable set.
Solution. Let {fa: a E A} be a family of "proper" intervals (an interval is

proper whenever its endpoints a and b satisfy a < b) and let E = UaeAIa.
Write E = UxeE Cx, where C x . denotes the component of x in E. Since each
x belongs to a proper subinterval of E, we see that each Cx is a proper interval;
see part (g) Problem 6.11. Since the distinct components C x are pairwise disjoint,
Section 18: THE LEBESGUE MEASURE 153
we see that there are at-most countably many C x and so E is the union of at-
most countably many intervals. Now, use the fact that each interval is a Lebesgue
measurable set to infer that E itself is a Lebesgue measurable set.
Problem 18.16. Let C be a closed nowhere dense subset of lRn such that
}"(C) > O. Show that the characteristic function Xc cannot be continuous on
the complement of any Lebesgue null set of lRn. Also, show that Xc will be contin-
uous on the complement of a properly chosen open set whose Lebesgue measure
can be made arbitrarily small.
Solution. Let A ~ lR" be a Lebesgue null set. Since }"(C) > 0 and }"(A) = 0,
it follows that N n C =1= 0. Fix some a E AC n C. We claim that Xc: N -+ lR
is not continuous at x = a.
Indeed, if Xc: AC -+ lR is continuous at a, then there exists some open ball
B(a, r) with XcCx) = 1 for all x E B(a, r) n A C ; i.e., B(a, r) n N ~ C holds.
Since }"(A) = 0, it follows that B(a, r) n A C is dense in B(a, r), and therefore,
B(a, r) ~ C (since B(a, r) n A C ~ C and C is closed), contradicting the fact
that C is nowhere dense.
Now, let e > O. By Theorem 18.2, there exists an open set V with C ~ V and
}"(V \ C) < e. Note that the set 0 = V \ C = V n Cc is open, and }"(O) < e.
We claim that Xc: Dc -+ lR is continuous.
To see this, let a¢.O = V n CC. We have two cases.
1) a E C. Since C ~ V, there exists some open ball B(a, r) with B(a, r) ~ V.
Now, note that
B(a, r) n oc = B(a, r) n [V u C] = B(a, r) n C

C
~ C.
Thus, if x E B(a, r) n oe, then XcCx) = 1. This shows that the function
Xc: OC -+ lR is continuous at x = a.
2) a E C e . Choose an open ball B(a, r) such that B(a, r) ~ CC. Then,
B(a, r) n oc = B(a, r) n [V e u C] = B(a, r) n V e ~ V e ~ CC.

Thus, x E B(a, r) n oe implies XcCx) = 0, which shows that in this case
Xc: oe -+ lR is continuous at x = a.
Problem 18.17. Let f: lRn -7 lR be a continuous function. Show that the graph
of f has (n + I)-dimensional Lebesgue measure zero.

Solution. Denote by An+l and An the (n + I)-dimensional and n-dimensional

Lebesgue measures, respectively. Fix some k and let A = [-k, k] x· .. x [-k, k].
Now, let E > O. By the uniform continuity of f on A, there exists some a > 0
such that x, YEA and IXi-yd < a for I:::: i :::: n imply If(x)- f(y)1 < E. Fix
a partition P of [-k,k] with mesh IPI < a,andlet Q = P x ... x P. Then,
Q subdivides A into a finite number of distinct closed cells, say Ai, ... , Ap.
(Note that the open cells corresponding to AI, ... , A p are pairwise disjoint). For
each I :::: i :::: p fix some. ai E Ai, and let Ii = [f(ai) - E, f(ai) + E]. Then,
Gk s:;; Uf=I(A i x Ii) holds, and so
p p
An+l(Gk) :::: I>n+l(A i x h) = I>n(Ai)' 2B = (2kt ·2B
i=1 i=1
holds for all B > O. This shows that An+I(G k ) =0 for each k. To complete the
proof, now apply Theorem 15.4 to Gk t G.
Problem 18.18. Let X be a Hausdorff topological space, and let f.L be a regular
Borel measure on X. Show the following:
a. If A is an arbitral) subset of X, then
f.L*(A) = inf{f.L(O): 0 open and A s:;; OJ.
b. If A is a measurabLe subset of X with f.L*(A) < 00, then
f.L*(A) = sup{f.L(K): K compact and K s:;; A}.
c. If f.L is a -finite and A is a measurabLe subset of X, then
f.L*(A) = sup{f.L(K): K compact and K s:;; A}.
Solution. (a) Since every a-set is a Borel set, Problem 15.2 shows that
f.L*(A) = inf{f.L(B): B is a Borel set satisfying AS:;; B}.
Now, use property (2) of Definition 18.4.

(b) Let A be a measurable set with f.L*(A) < 00 and let E > O. Pick an open set
V with A s:;; V and f.L*(V) < f.L*(A) + B. Similarly, choose an open set W such
that V \ A s:;; W s:;; V and
f.L*(W) < f.L*(V \ A) +B = f.L*(V) - f.L*(A) +B < 2E.

Section 18: THE LEBESGUE MEASURE 155
Next, pick a compact set C such that C ~ V and jJ-*(V) < jJ-*(C) + 8. Set
K = C n We, and note that K is a compact subset of A. Moreover,
o :::: jJ-*(A) - jJ-*(K) = jJ-*(A \ K) :::: jJ-*(V \ K)

= jJ-*(V \ C) u W) :::: [jJ-*(V) - jJ-*(C)] + jJ-*(W) < 38
holds, and the desired conclusion follows.

(c) Straightforward using (b).
Problem 18.19. If A is a (Lebesgue) measurable subset oflR ofpositive measure

and 0 < 0 < A(A), then show that there exists a measurable subset B of A
satisfying A(B) = o.
Solution. We shall present two solutions. The first one will employ the Axiom
of Choice (via Zorn's Lemma); the second one will establish the validity of the
conclusion without using the Axiom of Choice and without assuming that A is a
measurable set.
(a) Consider a measurable subset A of lR and some 0 > 0 satisfying 0 < 0 <
A(A). Since A(A n [-n, n]) t A(A) holds, replacing A by some An [-11,11],
we can assume that A(A) < 00 also holds.
Next, we shall denote by A the set of all collections C of pairwise disjoint
measurable subsets of A such that:
a) A(C) > 0 holds for each C E C (and so C is at most countable); and
b) The Lebesgue measurable set UCEC C satisfies A(UCEC C) :::: o.
From Problem 15.18, it is easy to see that A =1= 0. Under the inclusion relation
~ the set A is a partially ordered set. We claim that the partially ordered set
(A,~) satisfies the hypothesis of Zorn's Lemma. To see this, let {Ci : i E I} be
a chain of A (i.e., for each pair i, j E I either Ci ~ C j or C j ~ Ci holds
true). Our claim, will be established, if we can show that C = UiEI Ci E A. Note
first that if B, C E C, then B, C E Ci must hold for at least one i E I, and
so B n C = 0. In particular, it follows that C is at most countable. Now, if
BI, ... , Bk E C, then BI,"" Bk E Ci also must hold for some i (why?), and
so A(U;=I Br) :::: A(UBEC; B) :::: o. Since C is at most countable, it follows that
A(U BEC B) :::: o. That is, C E A.
Now, by Zorn's Lemma, the collection A has a maximal element, say C. If
B = UCEC C, then we claim that the measurable set B satisfies A(B) = 0 (and
this will complete the proof). To see the latter, assume by way of contradiction
that A(B) < o. Then, we have 0 < 1) = 0 - A(B) .:::: A(A) - A(B) = A(A \ B)
holds, and so by Problem 15.18 there exists a measurable subset D of A \ B
satisfying 0 < A(D) < rJ (clearly, D ¢. C). In view of B nD = C/J and
A(B U D) = A(B) + A(D) < A(B) + 0 - A(B) = 0,

we see that C1 = C U {D} E A. However, this contradicts the maximality property
of C, and so A(B) = 0 must hold, as desired.
(b) For this solution the set A is an arbitrary subset of R satisfying A(A) > O.
As in the preceding, we can assume that A ~ [-k, k] holds for some k. Now,
consider the function f: [ -k, k] --+ R defined by
f(t) = A(A n [-k, rD, t E [-k, k].
Clearly, f( -k) = 0 and f(k) = A(A). We claim that f is a continuous function.

Indeed, if -k ~ s < t ~ k, then
f(t) = A(A n [-k, t]) ~ A(A n [-k, s]) + A(A n (s, t]) ~ f(s) +t - s.
Therefore, If(s) - f(t)1 ~ It - sl holds for all s, t E [-k, k] ~d so f is a

continuous function.
Finally, by the Intermediate Value Theorem, there exists some -k ~ x ~ k such
that the subset B = A n [-k, x] of A (which is measurable if A is measurable)
satisfies f(x) = A(B) = o.
Problem 18.20. Let E be a Lebesgue measurable subset of R offinite Lebesgue
measure. Show that the function fE: R -+ R. defined by
lE(x) = A(E ll(x + E»,

is uniformly continuous.
Solution. The solution goes by steps.

(1) Assume first that E = (a, b) is a bounded open subinterval ofR. In this case,
an easy calculation shows that
21xl, if Ixl < b - a

lE(x) = { 2(b - a), if Ixl 2: b - a.
This guarantees that IE is uniformly continuous in this case.

(2) Assume that E and F are twd' Lebesgue measurable subsets of R of finite
measure such that IE and fF are both uniformly continuous. Put G = E U F. We
shall show that !G is also uniformly continuous.
Section 19: CONVERGENCE IN MEASURE 157
To see this, notice first that
implies
>-(Gb..(x + G)) ::: >-(E b..(x + E)) + >-(F b..(x + F)).

Hence,
I!a(x) - fa(Y)\ = \>-(Gb..(x + G)) - >-(Gb..(y + G))\

::: >-([Gb..(x + G)] b..[G b..(y + G)J)
= A(X + G)b..(y + G)) = A(Gb..(y - x + G))
::: A(E b..(y - x + E)) + A(F b..(y - x + F))
= fe(y - x) + !F(y - x).
Since IE and !F are uniformly continuous, it follows that fa is likewise uniformly

continuous. (Actually, the continuity of IE and fF at zero is what is needed
here.)
(3) By induction, we can show that if E = U;'=I Ei with each Ei Lebesgue
measurable having finite measure and Ei n E j = (/; if i =1= j, then IE is uniformly
continuous.
(4) Now, let E > O. Pick a finite collection of pairwise disjoint bounded open
intervals II, ... , In such that the set G = U7=1 Ii satisfies A(E b..G) < E. Then,
as previously, we have
\1E(x) - IE(Y)\ = \A(E b..(x + E)) - >-(E b..(y + E))\

::: A(Eb..(y x+E))
::: A(E b..G) + A(G b..(y-x + G)) + A«y-X + G)b..(y-x + E))
< 2E + A(Gb..(y - x + G))
= 2E + fa(Y - x).
This easily implies that IE must be a uniformly continuous function.
19. CONVERGENCE 1N MEASURE
Problem 19.1. Let {f,,} be a sequence of measurable functions and let f: X ~

JR. Assume that limf.L*({x EX: Ifn(x) - f(x)1 ::: En = 0 holdsfor every E > o.
Show that f is a measurable function.
Solution. Pick a sequence {kn } of strictly increasing positive integers such that
J.L*({x EX: I/k(X) - l(x)l2: ~})< 2- n for all k > kn. Set
Ell = {x EX: I/kn(X) - l(x)1 2: *}

for each n and let E = n:.'=1 U:m En. Then,
00 00
J.L*(E) :::- J.L*(U En) ::: L J.L*(E II ) : : : 21- m

n=m n=m
holds for all m, so that J.L*(E) = O. Also, if x ¢ E, then there exists some m
such that x ¢ U:m En, and so I/kn(X) - l(x)1 < *
holds for each n 2: m.
Therefore, lim !kn (x) = I (x) for each x ¢ E, and so Ik n -+ I a.e. holds. The
latter (by Theorem 16.6) easily implies that I is a measurable function.
. ~
Problem 19.2. Assume that {f,,} S; M satisfies In t and In -+ I. Show that
In t I a.e. holds.
Solution. By Theorem 19.4, there exists a subsequence {fkJ of the sequence

{fn} with Ikn -+ I a.e. Since 1" t, it easily follows that 1" t I a.e. holds.
Problem 19.3. II {fn} S; M satisfies In ~ I and In 2: 0 a.e. lor each n, then

show that I 2: 0 a.e. holds.
Solution. Since, by Theorem 19.4, some subsequence of {fn} converges almost

everywhere to I, we must have I 2: 0 a.e.
Problem 19.4. Let {fn} S; M and {gn} S; M satisfy In ~ I, gn ~ g, and

In = gn a.e. lor each n. Show that I = g a.e. holds.
Solution. Since 1" ~ I implies Ik n ~ I for each subsequence {fkn }

of {fn}, by passing to two subsequences (if necessary), we can choose a strictly
increasing sequence {kn} of positive integers such that Ik n -+ I a.e. and
gkn -+ g a.e. This easily implies I = g a.e.
Problem 19.5. Let (X, S, J.L) be a finite measure space. Assume that two sequ-
enCp.5 {-f,,} and {gil} aiM satisfy In ~ I and gn ~ g. Show that f"gn ~ Ig·
ISJhis !!!atement trUf~ if J.L*(X) = oo?
Section 19: CONVERGENCE IN MEASURE 159
Solution. By Theorem 19.4, the only possible limit of {fngn} is fg. Conse-
quently, if fngn ~ fg does not hold, then there exist 8 > 0 and 0 > 0 and
some subsequence of {fngn} (which we shall denote by {fngn} again) such that
p,*({x EX: Ifn(x)gn(x) - f(x)g(x)1 2: 8l) 2: 0
holds for all n. In view of fn ~ f and gn ~ g, Theorem 19.4 shows that

for some subsequence {ik"gk"l of {fngn} J.Lwe must have ik"gkn -+ fg a.e.
Now, n:;te that (by Theorem 19.5) ik"gk" -+ fg holds, contrary to (*). Thus,
fngn -+ fg holds.
If p,*(X) = 00, then the conclusion is no longer true. An example: Take X =
(0, (0) with the Lebesgue measure. Consider the functions fn(x) = Jx 4 + :I and
A • 2 A n
f(x) = x-. Then, /" -+ f, whIle fn -F+ f-·
ry ry
Problem 19.6. Show that a sequence of measurable functions {fn} on a finite

measure space converges to f in measure if and only if every subsequence of {fn}
has in turn a subsequence which converges to f a.e.
Solution. The conclusion follows immediately from Theorems 19.4 and 19.5.
Problem 19.7. Define a sequence {fn} of M to be p,-Cauchy whenever for

each E > 0 and 0 > 0 there exists some k (depending on E and 0) such that
p,*({x EX: Ifn(x) - fm(x)1 2: ED < 0 holds for all n, m 2: k.
Show that a sequence {fn} of M is a p,-Cauchy sequence if and only if there
exists a measurable fimction f such that fn ~ f·
Solution. If /" ~ f, then the inclusion
{x: Ifn(x) - fm(x)1 2: 28} ~ {x: lfn(x) - f(x)1 2: 8} U {x: Ifm(x) - fex)1 2: 8}
easily implies that {fn} is a p,-Cauchy sequence.

For the converse, assume that {fn} is a p,-Cauchy sequence. It suffices to show
that {fn} has a subsequence that converges in measure (why?). To this end, start
by selecting a subsequence {gn} of {fn} satisfying
160 Chapter 3: TIIE TIIEORY OF MEASURE
00
Fn = U Ek = {x: k
Igk+l(X) - gk(x)1 :.::: 2- holds for some k :.::: n }.
k=1l
Clearly,jL*(Fn)::: L~lljL*(Ek)::: 2 1- n holds for all n,andhencethemeasurable

set F = n:1Fn satisfies jL*(F) = O. Now, note for each fixed x rf. F there
exists some positive integer" kx such that x rf. Fn holds for all n :.::: kx . Thus, for
n :.::: kx, we have
00
1n
Ign+p(x) - gn(X)1 ::: ~]g;+1 (x) - g;(x)1 ::: 2 - .
i=n
Therefore, {gn(x)} is a Cauchy sequence of real numbers for each x rf. F. Thus,
there exists a function gEM such that gn(x) -4 g(x) holds for each x rf. F.
Now, if 1l > k and x rf. FIl , then
00
Ign+1(X) - gn+p(X)I::: L Ig;+1 - g;(x)1 ::: Tn

;=11+1
implies that Igll+1(X) - g(x)1 ::: 2- n < 2- k for all n > k. Thus,
holds for all n > k. Finally, to see that gn ~ g holds, note that for n > k, we
have
{x EX: Ign(x) - g(x)l:.::: 2 1- k }

C {x EX: Ign(x) - gn+1(.x)l:.::: Tk}U{x E X: Ign+1(X) - g(x)l:.::: Tk}
s;:; En U Fn = Fn·
20. ABSTRACT MEASURABll.ITY
Problem 20.1. Let R be a nonempty collection of subsets of a set X. Show

that R is a ring if and only if R is dosed under symmetric differences and finite
intersections.
Section 20: ABSTRACf MEASURABILITY 161
Solution. Assume first that the nonempty collection n is a ring. That is, assume
that A, BEn imply A U BEn and A \ BEn. Then the identities
A6.B = (A \ B) U (B \ A) and An B = A \ (A \ B)
easily imply that n is closed under symmetric differences and finite intersections.
n
For the converse assume that is closed under symmetric differences and finite
intersections. Then, the identities
A \ B = A6.(A n B) and AU B = (A6.B)6.(A n B)

guarantee that n is a ring.
Problem 20.2. If n is a ring of subsets of a set X, then show that the collection
A = {A 5;; X: Either A or N belongs to n}
is an algebra of sets.
Solution. From the definition of A, it easily follows that if A E A, then ACE A,

i.e., that A is closed under complementation.
Now, assume that A, B E A. If A, BEn, then since n (as being a ring) is
closed under finite unions, we have AU BEn and so AU B E A. If A c, B C E n,
then N \ (N \ B C ) En, and so
AU B = (A Cn BC)C = [N \ (A C\ BC)r E A.
Now, assume that A E n and B C E n. Then, B C\ A = B C nNE n, and
consequently (from the definition of A), AU B = (A C n BC)C E A. The preceding
show that A is an algebra.
Problem 20.3. In the implication scheme of Figure 3.3 show that no other im-
plication is true by verifying the following regarding an uncountable set X.
~a-ring~
a-algebra ring ===:;::;> scmiring
~algebra~
FIGURE 3.3.
162 Chapter 3: THE TIIEORY OF MEASURE
a. The collection of all singleton subsets of X together with the empty set is a
semiring but not a ring.
b. The collection of all finite subsets of X is a ring but is neither an algebra
nor a a-ring.
c. The collection of all subsets of X that are either finite or have finite com-
plement is an algebra but is neither a a-algebra nor a a-ring.
d. The collection of all at-most countable subsets of X is a a-ring but not an
algebra.
e. The collection of all subsets of X that are either at-most countable or have
at-most a countable complement is a a-algebra (which is, infact, the a-
algebra generated by the singletons).
Solution. (a) If A and B are singletons, then An B and A \ B are either empty or
singletons. This shows that the collection of all singletons together with the empty
set is a semiring. However, it should be obvious that finite unions of singletons
need not be a singleton, and so the collection of all singletons is not an algebra.
(b) Let n denote the collection of all finite subsets of (the infinite) set X. If
A, BEn, then A U B and A \ B are finite sets and so A U B and A \ B belong to
n. n
This shows that is a ring. Since the complement of a finite set is infinite, it
n n
follows that is not closed under complementation, and so is not an algebra.
n
To see that is not a a-ring, let A = raJ,a2,."} be a countable subset of X,
and for each n let An = {all} En. Then, U:] An = A rf. n,and this shows that
n is not a a-ring.
(c) tfn is the ring of all finite subsets, then by part (b) the collection
A = {A ~ X: Either A or A C belongs to n}
is an algebra of sets. To see that A is not a a-ring (and hence neither a a-algebra),
let A = {a], a2,"'} be a countable subset of X such that X \ A is an infinite
set. Clearly, A rf. A. On the other hand, if An = {an}, then An E A and
U:] An = A rf. A. This shows that A is not a a-ring.
(d) Let C denote the collection of all at-most countable subsets of X. Clearly,
A, B E C imply A \ B E C. Also, C is closed under countable unions (recall that
an at-most countable union of sets each of which is at-most countable is at-most
countable; see Theorem 2.6). Therefore, C is a a-ring. However, when X is an
uncountable set, C is not closed under complementation, and hence it cannot be
an algebra.
(e) This is Problem 12.7.
Problem 20.4. Show that a Dynkin system is a a-algebra if and only if it is

closed under finite intersections.
Section 20: ABSTRACT MEASURABILITY 163
Solution. Let 'D be a Dynkin system that is closed under finite intersections.
Since 'D is also closed under complementation, it is easy to see (by using the
identity A U B = (A e n Ben that'D is in fact an algebra. So, if A = U::l All
with {All} s:;; 'D, then by letting Bn = UZ=IAk E 'D, and noting that BIl t A, we
see that A E 'D. In other words, 'D is a (J"-algebra.
Problem 20.5. Give an example of a Dynkin system which is not an algebra.
Solution. Consider the set X = {I, 2, 3, 4}, and let
'D = {O, {l, 2}, {3, 4}, {l, 3}, {2, 4}, X}.
Then'D is a Dynkin system (why?), which (since {I, 2} U {I, 3} = {I, 2, 3} does
not belong to 'D) fails to be an algebra.
Problem 20.6. A monotone class of sets is a family M of subsets of a set X

such that if a sequence {An} of M satisfies All t A or An .j, A, then A E M.
Establish the following properties regarding monotone classes:
a. We have the following implications:
(J"-algebra ==} Dynkin system ==} monotone class
Give examples to show that no other implication in the preceding scheme

is true.
b. An algebra is a monotone class if and only if it is a (J" -algebra.
c. The (J"-algebra (J"(A) generated by an algebra A is the smallest monotone
class containing A.
Solution. (a) The implication scheme follows immediately from the definitions
of the three classes of sets involved. An example of a Dynkin system which is not
an algebra was exhibited in the preceding problem. Now, if X = {I, 2}, then the
collection M = {X, {I}} is a monotone class but not a Dynkin system.
(b) Let A be an algebra of sets. If A is a (J"-algebra, then it is clearly a monotone
class. For the converse assume that the algebra A is a monotone class.
Assume {All} s:;; A and put A = U::l All' Let BIl = UZ=1 Ak E A and note
that BIl t A. Since A is a monotone class, it follows that A E A, and so A is a
(J" -algebra.
(c) Let A be an algebra of sets and let M be the smallest monotone class that
contains A, i.e., j\lt is the intersection of the collection of all monotone classes
that include A. Clearly, A s:;; M s:;; (J"(A).
Let C = {B EM: B \ A E M for each A E A}. An easy verification shows

that C is a monotone class that includes A, and so M = C. Now, let
1) = {B EM: M \ B E M for each M EM}.
Again, 1) is a monotone class which (in view of M = C) satisfies A 5; 1).

Thus,1) = M. This shows that M is, in fact, a Dynkin system. By Dynkin's
Lemma 20.8, a(A) 5; M, and so M = a(A).
Problem 20.7. Show that if X and Yare two separable metric spaces, then
BxxY = Bx ® By.
Solution. Assume that X and Y are two arbitrary topological spaces. For each
subset A of X, let
:EA = {B 5; Y: A x B E Bxxy}.
From the identities A x (B \ C) = (A x B) \ (A x C), we see that if B, C E :EA,

then B \ C E :EA. From A x (n:::\ Bn) = n:::\(A x Bn), it follows that :EA is
closed under countable intersections. Observing that C/J E :E A, we infer that :EA is
a a-ring. Clearly, :EA is a a-algebra if and only if Y E :EA.
Next, note that for any open subset 0 of X, V E :Eo for every open subset V
of Y. Since Y is itself open, if 0 is open, then :Eo is a a-algebra of subsets of Y
that includes the open subsets of Y. Thus, By 5; :Eo for each open subset 0 of X.
Now, let
As we have just noticed, V E A holds for each open subset V of X. Since

(as easily checked) :EA = :EAc for each A E A, we see that A is closed under
complementation. Moreover, if {An} 5; A, then for any Borel subset B of Y, we
haveAnxB E Bxxyforeachn. Thus,inviewofn:::\(AnxB) = (n:::\ An)xB,
we obtain B E :En:;';,) An' In other words, A is closed under countable intersections,
and so A is a a-algebra including the open subsets of X. This implies Bx 5; A.
We have just established the following: If A is a Borel subset of X and B is a
Borel subset of Y, then A x B E B xxY . Therefore,
Bx ® By 5; BxxY·
For the reverse inclusion, assume that X and Yare two separable metric spaces.
Then every open subset of X x Y is an at-most countable union of sets of the form
v xU, where V is an open subset of X and U an open subset of Y. Consequently,

BxxY ~ Bx ® By, from which it follows that Bx ® By = B xxY .
Problem 20.8. Show that the composition function of two measurablefunctions

is measurable.
Solution. Assume that (X, 2: 1).L (Y,2:2) (Z, 2:3) are measurable func-
tions. If A E 2:3, then g-I(A) E 2: 2 , and so (g 0 f)-I(A) = f- I (g-I(A)) E 2: 1,
This shows that g 0 f is measurable.
Problem 20.9. ff(X, 2:) is a measurable space, then show that

a. the collection of all real-valued measurable functions defined on X is a
function space and an algebra offtll1ctions, and
b. any real-valuedfunction on X which is the pointwise limit of a sequence of
(2:, B)-measurable real-valued functions is itself (2:, B)-measurable.
Problem 20.10. Let (X, 2:) be a measurable space. A 2:-simple function is any
measurableftll1ction ¢: X -+ lR having afinite range, i.e, if¢ hasfinite range and
its standard representation ¢ = L:;'=I ai XA/ satisfies Ai E 2: for each i.
Show that a ftll1ction f: X -+ [0, 00) is measurable if and only if there exists a
sequence {¢II) of2:-simplefunctions such that ¢n(x) t f(x) holdsfor each x EX.
Solution. The proof is identical to the proof of Theorem 17.7. Here it is.
Foreachn letA~ = {x EX: (i -1)2- 11 ~ f(x) < i2- n } fori = 1,2, ... , n2n ,
and note that A:, n A~ = 0 if i i= j. Since f is measurable, all the A:, belong to
2:.
Now, for each n define ¢II = L:~,:nl 2- n(i - l)XA~' and note that {¢n) is a
¢II+I (x) ~ f(x) holds for all x and all n. Moreover, if x is fixed, then
° °
sequence of 2:-simple functions. Also, an easy verification shows that ~ ¢n (x) ~
~
f(x) - ¢II(X) ~ 2- 11 holds for all sufficiently large n. This implies ¢II(X) t f(x)
for all x E X.
Problem 20.11. Use Corollary 20.10 to show that if a measure JL is a-finite,

then JL* is the one and only extension of JL to a measure Oil a(S).
Solution. Let v: a(S) -+ [0,00] be a measure satisfying yeA) = JL(A) for each
A E S. We shall establish that yeA) = JL*(A) for each A E a(S).
166 Chapter 3: TIlE THEORY OF MEASURE
Fix E E S with /.L(E) < 00 and let
SE = {E n A: A E S} = {B E S: B S; E}.
Clearly, SE is a semiring of subsets of E and /.L restricted to E is a measure.

Moreover, we know (see Problem 15.7) that the outer measure generated by the
measure space (E, SE, /.L) is simply the restriction of /.L* to peE). In addition, we
claim that if a(SE) denotes the a-algebra generated by SE in peE), then
a(SE) = {A nE: A E a(S)} = {B E a(S): B S; E}. (*)
To see this, note first that since {B E a(S): B S; E} is a a-algebra containing SE,
we have
On the other hand, the collection
is a a-algebra of subsets of X satisfying S S; A. Hence, a(S) S; A. In particular,

if B S; E satisfies B E a(S), then B E A and so B = B nEE a(SE).
Consequently, {B E a(S): B S; E} S; a(SE), and the validity of (*) follows.
Next, note that since SE is closed under finite intersections, /.L*(E) = /.L(E) =
veE) < 00, and v(F) = fJ,(F) = /.L*(F) for all F ESE, it follows from Corol-
lary 20.10 that v(F) = fJ,*(F) for all F E a(S) with F S; E.
Now, let {En} be a pairwise disjoint sequence of S satisfying X = U:l En and
fJ,(En) < 00 for each n. If A E a(S), then by the preceding discussion we have
v(A n Xn) = /.L*(A n Xn) for each n, and so
00 00
v(A) = v(A n X) = v(U An Xn) =

n=l
L v(A n Xn)
n=l
00 00
= L fJ,*(A n Xn) = /.L*(U A n Xn)

n=l n=l
= /.L*(A n X) = /.L*(A),
and we are finished. (For more about the extension of /.L, see Problem 15.19.)
Problem 20.12. Show that the uniform limit of a sequence of measurable func-
tions from a measurable space into a metric space is measurable.
Solution. Let Un} be a sequence of measurable functions from a measurable

space (X, 2;) into a metric space (Y, d). Suppose f is the uniform limit of U;,}.
Thatis,assumethatforeachE > othere exists some no such that d(jn(x), f(x») <
E for all x E X and all n 2: no. By passing to a subsequence, we can assume that
d(jn(x), f(x») < ~ holds for each n and all x EX.
Since the family of closed sets generates the Borel sets of Y, in order to establish
the measurability of f, it suffices to prove that f- I (C) E 2; for each closed set C.
To this end, let C be a closed subset of Y.
Let Vn = /y E Y: dey, C) < ~}. We claim that
f-I(C) = n00
n=1
f-ICVn)'
To see this, assume x E f-I(C), i.e., let f(x) E C. From d(jn(x), C) :S

d(jn(x), f(x») < ~,we get fn(x) E Vn or x E h;I(Vn) for each n. Conversely,
if fn(x) E Vn for each n, then d(f,,(x), C) < ~ for each n, and so if we pick some
Cn E C with d(f,,(x), cn) < ~, then we have
d(j(x), C) :S d(j(x), en) :S d(j(x), f,,(x») + d(f,,(x), en) < *+ *= ~

for each n. This implies d(j(x), C) = O. Since C is a closed set, it follows that
f(x) E C, or x E f-I(C).
Next, use the measurability of each fn and the fact that each Vn is open to obtain
that h-;I(Vn) E 2; for each n. Now, invoke (**) to conclude that f-I(C) E 2;.
Problem 20.13. Let f, g: X --* lR be two functions and let E denote the a-
algebra of all Borel sets oflR. Show that there exists a Borel measurable function
h: lR --* lR satisfying f = hog if and only if f- I(E) S; g-I (E) holds.
Solution. Assume f = hog holds for some Borel measurable function h: lR --*
lR. Fix BEE and note that h-I(B) E E. Therefore, f-I(B) = g-I(h-I(B») E
g-I(E), and so f-I(E) S; g-I(E) holds.
For the converse, assume f-I(E) S; g-I(E). The existence of the Borel mea-
surable function h will be established by steps.
Step I: Assume f = XA for some A E f- I(E).
Since f-I(E) S; g-I(E) is true, there exists some BEE such that A = g-I(B).
Let h = XB, and note that hog = f.
Step II: Let f = I:~=I ai XAj with the Ai pairwise disjoint and Ai E f- I(E)
for each i. For each i choose some Bi E g-I(E) such that Ai = g-I(B i ). If
we consider the Borel step function h = I:7=lai XBp then it is easy to see that
hog = f.
Step III: The general case.

The preceding problem applied with I; = f-I(B) guarantees the existence of
a sequence {<Pnl of f-I(B)-simple functions satisfying <PI/(x) t f(x) for each
x EX. Now, by Step II, for each n there exists a Borel measurable function
111/: lR -7 lR such that hn 0 g = <Pn. Next, let
B = {x E lR: lim hl/(x)

n-->oo
= hex) eXists}.
It follows (as in Problem 16.7) that B E Band hn(x)Xn(x) -7 hex) for each
x E lR. If we let h (x) = 0 for x fj. B, then h: lR -7 lR is Borel measurable and
satisfies hog = f.
Problem 20.14. Let (X, I;) be a measurable space, Y, Z I, and Z2 be separable

metric spaces and IJI a topological space. Now, assume also that the functions
Ii: X x Y -7 Zi, (i = 1, 2), are Caratheodory functions and g: Z I x Z2 -7 IJI is
Borel measurable. Show that the function h: X x Y -7 W, defined by
hex, y) = g(fI(X, y), hex, y)),
is jointly measurable.
Solution. By Theorem 20.15, each fi: X x Y -7 Zi is jointly measurable. This

implies that the function F: X x Y -7 ZI X Z2, defined by
F(x, y) = (f1(X, y), hex, y))

is measurable (why?). Since g: ZI x Z2 -7 IJI is (B Z , XZ 2' BIjI)-measurable and
(by Problem 20.7) B z , ® BZ2 = B Z , XZ 2' it follows that II = go F is likewise
measurable.
Problem 20.15. Let (X, I;) be a measurable space and (Y, d) a separable metric
space. Show that afunction f: X -7 Y is measurable if and only iffor each fixed
y E Y the function x f-+ dey, f(x)),from X into lR, is measurable.
Solution. Let f: (X, I;) -7 Y be a function from a measurable space to a

separable metric space. For each y E Y define the function gy: X -7 lR by
gy(x) = d(y, f(x)). Note that for each r > 0 and each y E Y, we have
f- 1 (B(y,r)) = {x E X: f(X).E B(y,r)} = {x EX: d(y, f(x)) < r}

= {x E X: gy(x) < r} = g;I(-oo,r)).
Section 20: ABSTRACT MEASURABILTIY 169
Assume that each gy is measurable. Then, by the preceding identity, we have

I-I (B(y, r») = g;1 (-00, r») E I; for each y E Y and all r > O. Since Y is a
separable metric space, every open set can be written as an at-most countable union
of open balls, and so I-I (0) E I; holds for-each open set 0 _ By Theorem 20.6,
I is a measurable function.
For the converse, suppose that I is a measurable function and let y E Y. From
=
gyl(-oo,r») 1-I(B(y,r») ifr > 0 and gyl(-oo,r») cj; ifr ~ 0, we =
easily infer that g}: is a measurable function for each y E Y.
Problem 20.16. Let (X, S, fL) be a a-finite measure space, where S is a a-

algebra. III: X -+ lR is a AIL -measurable lunction, then show that there exists a
S-measurable lunction g: X -+ lR such that I = g a.e.
Solution. We can assume I(x) 2: 0 for each x E X (otherwise, we apply the

arguments below to 1+ and 1- separately). If I = XA for some A E AIL' then
an easy argument (using Theorem 15.11) shows that there exists a null set C such
that B = A U C E S. So, if g = XB, then g is S -measurable and I = g a.e. holds.
It follows that if I is a AIL-simple function, then there exists a S-simple function
g such that I = g a.e.
Now, we consider the general case. By Problem 20.10 there exists a sequence
{<PII} of A IL -simplefunctions satisfying <Pn Ct') t I (x) for each x EX. For each n fix
as-simple function 1/111 such that 1/In = <Pn a.e. By Theorem 15.11, for each n there
exists a null set All E S with 1/111 (x) = <PII(X) for all x ¢. All' PutA = U:l All E S,
and note that A is a null set. Moreover, we have 1/1Ii XAC (x) t I XAC (x) for each x.
If g = IXAc, then (by Problem 20.9(b» g is a S-measurable function satisfying
1= g a.e.
CHAPTER 4 ______________
THE LEBESGUE INTEGRAL
21. UPPER FUNCI10NS
Problem 21.1. Let L be the collection ofall step functions ¢ such that there exist
afinite number of members A I, ... , An ofS all offinite measure and real numbers
a I, ..• , an such that ¢ = 'L,;'= 1ai XA;. Show that L is a function space. Is L an
algebra offunctions?
Solution. Let ¢ = 'L,~=1 aiXA; and 1/1 = 'L,:;~I bjXBj' where the Ai and B j
belong to S and they all have finite measure. By Problem 12.14, there exist
pairwise disjoint sets C I, ... , Ck of S such that each Ai and each B j can be
written as a union from the C i . We can assume that U;=l C r = [U;'=l Ai] U
J.
[Uj= 1 B j It is easy to see that ¢ and 1/1 can be written in the form ¢ =
'L,;=1 Cr Xc, and 1/1 = 'L,;=1 dr Xc" Now, everything follows from the equali-
ties:
1.
2.
3.
Problem 21.2. Consider the function f: lR ~ lR defined by f(x) = 0 if x ~

(0, I], and f(x) = ..jii if x E (n~l' ~] for some n. Show that f is an upper
function and that - f is not an upper function.
Solution. Put Ak = (k~l ' t] and note that A(Ak) = k(k~1)' Thus, if we let
n
¢n = L .JkXAk'
k=l
then {¢n} is a sequence of step functions satisfying ¢n (x) t f (x) for each x. On
171
172 Chapter 4: THE LEBFSGUE INTEGRAL
the other hand, the relations
11 n 00
f ¢11 d)" = L ./k . k(k~1) = L ~(~+I)

k=1 k=1
::: L r~ <
k=1
00
guarantee that f is an upper function.

Since - f is not bounded from below, there is no step function ¢ satisfying
¢ ::: - f. This implies that -- f cannot be an upper function.
Problem 21.3. Compute Jf d)" for the upperfunction f of the preceding exer-
cise.
00
Solution. We have Jf d)" = L(n+:)..;n'
n=1
Problem 21.4. Verify that every continuous function f: [a, b] -+ R is an upper

function-with respect to the Lebesgue measure on [a, b].
Solution. For each n let Pn = {xo, XI, ... , X2n} be the partition that divides
[a, b] into 211 subintervals all of the same length (b - a)2- n ; that is, Xi =
a+i(b-a)2- 11 • Let mi =min{f(x): X E [Xi_l,xi]},andthendefine
211
¢11 = LmiX[Xi-J,Xi)'
i=1
Clearly, each ¢11 is a step function. Using the uniform continuity of f, it is not
difficult to see that ¢n(x) t f(x) holds for all X E [a, b). On the other hand, if
J
f(x) ::: M holds for each x, then ¢n d).. ::: M(b - a) holds for all n, implying
that f is an upper function.
Problem 21.5. Let A be a measurable set, and let f be an upper function. If

XA ::: f a.e., then show that JL*(A) < 00.
Solution. Choose a sequence {¢n} of step functions with ¢11 t f a.e. Then,
¢11 /\ XA t f /\ XA
= XA a.e., and so, by Theorem 17.6,
Section 21: UPPER FUNCTIONS 173
Problem 21.6. Let f be an upper function, and let A be a measurable set of

finite measure such that a :::: f(x) :::: b holds for each x E A. Then, show that
a. f XA is an upper function, and
b. J
a,u*(A):::: fXA d,u :::: b,u*(A).
Solution. (a) Pick a sequence {<Pn} of step functions with <PII t f a.e. For each
n define the step function 1{!1I = (<Pn XA) /\ b XA. Then,
Since 1{!1I t f XA a.e. holds, f XA is an upper function.

(b) Apply the monotone property of the integral (Theorem 2l.5) to the inequality
aXA :::: fXA :::: bXA.
Problem 21.7. Let (X, S,,u) be afinite measure space, and let f be a positive
measurable function. Show that f is an upperfitnction if and only if there exists a
real number M such that J <P d,u :::: M holds for every step function <P with <P :::: f
a.e. Also, show that if this is the case, then
f f d,u = sup {f <P d,u: <P is a step function with <P :::: f a.e. }.
Solution. If f is an upper function, then by Theorem 21.5 every step function

J J
<P with <P :::: f a.e. satisfies <P d,u :::: f d,u < 00.
Conversely, by Theorem 17.7 there exists a sequence {<PI!} of simple functions
with <Pn t f a.e. Since (X, S,,u) is a finite measure space, we know that each
<Pn is a step function. If J<PI! d,u :::: M holds for all n, then this readily implies
that f is an upper function.
The last formula is immediate from Theorem 2l.5.
Problem 21.8. Show that every monotone function f: [a, b] -+ lR is an upper

function--with respect to the Lebesgue measure on [a, b].
Solution. We assume that f: [a, b] -+ lR is an increasing function. The "de-

creasing case" can be proven in a similar fashion and is left for the reader.
By Problem 9.8, we know that the set D of all discontinuities of f is an at-most
=
countable set. In particular, )"'(D) O. Now, for each n, let Pn be the partition that
subdivides [a, b] into 211 equal subintervals. That is, let P II = {as, ai, ... ,a2n },
where a? = a + b:;,.a i for i = 0, 1, ... , 2n. Next, for each 1 ::: i ::: 2n let
and put 4>11 = I:~:I mj X[aF_I,aj)' Clearly, each 4>n is a step function and, in view
of the monotonicity of f, we have
for all x E [a, b). Put E = D U PI U P2 U P3 U··· and note that )"(E) = 0. We
shall establish that 4>n(x) t f(x) for each x E [a, b] \ E.
To this end, fix some t E [a, b] \ E and let E > 0. Since f is continuous at t,
there exists some 8 > such that °
X E [a, b] and t - 8 < x < t +8 imply f(t) - E < f(x) < f(t) + E. (*)
Next, pick some k such that ~ < 8 for all n 2: k, and then choose the subinterval
[Xi-I, x;] of Pk such that t E (Xi-I, Xi). From (*), it easily follows that
4>k(t) = inf{j(x): x E [xi-I> x;]} 2: f(t) - E.
Therefore, f(t) - E ::: 4>k(t) ::: 4>n(t) ::: f(t) holds for all n 2: k, and this shows
that 4>n(t) t f(t), as claimed.
Finally, note that the monotonicity of f guarantees that mj ::: f(b) holds for
aliI::: i ::: 2n. This implies
f 4>n d)"
2" 2"
= t;mi(a~ - a~_I) ::: f(b) t;(a~ - a~_I) = f(b)(b - a) < 00
for each n, and this establishes that f is an upper function.
22. INTEGRABLE FUNCTIONS
Problem 22.1. Show by a counterexample that the integrable functions do not

form an algebra.
Solution. Consider the function -j: 1R. -+ 1R. defined by f(x) = if x rf. (0, 1],
and f(x) = .[ii if x E (n~I' ~] for some n. From Problem 21.2, we know that f
°
is an integrable function.' Now, note that f2 is not an integrable function.
Section 22: INTEGRABLE FUNCTIONS 175
Problem 22.2. Let X be a nonempty set, and let 0 be the Dirac measure on X with
respect to the point a (see Example 13.4). Show that every junction f: X -+ JR is
integrable and that J f do = f(a).
Solution. Note that f = f(a)X{a) a.e. holds. Consequently, the function f is

J
integrable and f do = f(a)o({aJ) = f(a).
Problem 22.3. Let f.1, be the counting measure on IN (see Example 13.3). Show
that a function f: IN -+ JR is integrable if and only ifL:', If(n)1 < 00. Also,
show that in this case J f df.1, = L:'I fen).
Solution. Let f: IN --+ JR. Since every function is measurable, f is integrable

if and only if both f+ and f- are integrable. So, we can assume that f(k) 2:
holds for each k.
°
If ¢n = L~=I f(k)X{k), then (¢n} is a sequence of step functions such that
¢n(k) tn f(k) for each k, and
f ¢n df.1, = t
k=1
f(k) til :t
k=1
f(k).
This shows that f is integrable if and only if L::I f(k) < 00, and in this case
Jf df.1, = L::I f(k) holds.
Problem 22.4. Show that a measurable function f is integrable if and only ifl fl
is integrable. Give an example of a nonintegrable junction whose absolute value
is integrable.
Solution. Apply Theorems 22.2 and 22.6. For a counterexample: Let E

be a non-Lebesgue measurable subset of [0, 1] and consider the function
f = XE - X[O.I]\E·
Problem 22.5. Let f be an integrable function, and let {En} be a sequence of

disjoint measurable subsets of X. If E = U:'I Ell, then show that
Solution. LetFII = U~=I Ei. Clearly, IfXFn I:::: If I foreachn andfXFn --+ fXE
176 Chapter 4: TIlE LEBESGUE INTEGRAL
a.e. Thus, by the Lebesgue Dominated Convergence Theorem, we have
JrE I df.L = !IXE df.L = lim !IXFn df.L

n-+CXl
= }i~(t! IXEi df.L) = ~ Lnl df.L.
Problem 22.6. Let I be .an integrable function. Show that lor each E > 0
there exists some 0 > 0 (depending on E) such that I lEI df.L I < E holds lor all
measurable sets with f.L*(E) < o.
Solution. Consider an integrable function I and let e > O. From 0::: IliA n t
II I and the Lebesgue Dominated Convergence Theorem we get II IA n d f.L t 1
1 III df.L. So, there exists some no such that 1(1/1 - IliA no)df.L < ~ for all
n 2: no. Now, put 0 = 2~o and note that if a measurable set E satisfies f.L*(E) < 0,
then
ILldf.L1 = Ll/ldf.L= L(I/I-I/IAno)df.L+ LI/IAnodf.L
::: !(I/I-I/IAno)diL+ L nodf.L
< ~ + nof.L*(E) < ~ + ~ = e,

as desired.
Problem 22.7. Show thatlor every integrablelunction I the set
{x E X: I(x) =1= O}
can be written as a countable union olmeasurable sets ollmite measure-orelerred

to as a a-finite set.
Solution. Each En = {x E X: I/(x)1 2: *} is a measurable set and, by
Theorem 22.5, f.L*(En) < 00 holds. Now, observe that
CXl
{x E X: I(x) =1= o} = UE".

n=1
Problem 22.8. Let I: lR -+ lR be integrable with respect to the Lebesgue mea-

sure. Show that the function g: [0, 00) -+ lR defined by
g(t) = sUP{! I/(x + y) - I(x) I d)"(x): Iyl ::: t}
lor t 2: 0 is continuous at t = O.
Solution. Let f: lR -+ lR be an upper function and let {4>II} be a sequence of

step functions with 4>11 t f a.e. Fix some y, and note that 4>n(x + y) t f(x + y)
holds for almost all x. Since XA(X + y) = XA-ix) and A(A) = A(A - y), it
follows that f(x + y) as a function of x i~ integrable and J f(x + y) dA(X) =
J fdA. Thus, if f is integrable, then f(x + y) is integrable with respect to x for
each fixed y and J f(x + y) dA(X) = J fdA holds. In particular, for each fixed
y we have J If(x + y) - f(x)1 dA(X) ::: J If(x + y)1 dA(X) + J If(x)1 dA(X) =
2JlfidA < 00.
Now, for each integrable function f: lR -+ lR and each t 2: 0, define
gf(t) = sUP{jlf(X + y) - f(x)1 dA(X): IYI::: t } 2: O.
Then, we have
These relations show that the set
L = {f: f is integrable and gf is continuous at zero}
is a vector space. Moreover, L has the following approximation property:
• If f is an integrable function such that for each e > 0 there exists some
h E L with Jlf - hi dA < e, then f E L.
Indeed, if f is such a function and e > 0 is given, then choose h E L with

Jlf - hi dA < e. Pick some 8 > 0 with g,,(t) < e whenever 0 < t < 8, and
note that for Iyl ::: t we have
jlJ(X + y) - f(x)1 dA(X)
::: jlJ(X + y) - hex + y)1 dA(X) + jlh(X + y) - h(x)1 dA(X)
+ jlh - fldA < 3e.
Thus, gf(t) ::: 38 holds for all 0 < t < 8 so that f E L.

Now, assume that f == X[a.b). If 0 < t < b - a, then for Iyl ::: t we have
jlf(X + y) - f(x)1 dA(X) = jIX[a-y.b-y)(X) - X[a.b)(x)1 dA(X)
= A([a - y, b - y)L\[a, b)) = 21yl ::: 2t,

178 Chapter 4: THE LEBESGUE IN1EGRAL
and so gf(t) ::: 2t holds for all 0 < t < b - a, i.e., f E L. By the approximation
property, we have XA E L for every (1' -set A of finite measure, and hence, by the
same property, XA E L for every A E A with )'(A) < 00 (see Problem 15.2). It
follows that L contains the step functions. Since for every integrable function f
and each 8 > 0 there exists a step function ¢ with Ilf - ¢I d)' < 8, we infer
that L consists of all the integrable functions.
Note. We basically verified here that the collection L satisfies properties (1), (2),
and (3) of Theorem 22.12. This guarantees that L coincides with the vector space
of all integrable functions.
Problem 22.9. Let g be an integrable function and let Un} be a sequence of

integrable functions such that Ifni::: g a.e. holds for all n. Show that if fn ~ f,
then f is an integrable function and lim I Ifn - fl dJL = 0 holds.
Solution. By Theorem 19.4, the sequence Un} has a subsequence that converges
to f a.e., and so If I ::: g a.e. Thus, by Theorem 22.6, the function f is integrable.
Assume that for some 8 > 0 there exists a subsequence {gn} of Un} such
that Ilgn - fldJL 2: 8. By Theorem 19.4, there exists a subsequence {h n } of
{gn} with h n ---+ f a.e. Now, note that-the Lebesgue Dominated Convergence
Theorem implies 0 < 8 ::: Ilh n - fl dJL ---+ 0, which is impossible. Hence,
limIlfn - fl dJL = O.
Problem 22.10. Establish the following generalization of Theorem 22.9: If

Un} is a sequence of integrable functions such that 2::, I Ifn I dJL < 00, then
2::, fn defines an integrable function and
Solution. By Theorem 22.9, the series g = 2::, Ifni defines an integrable

function and 12:~=, /',1 ::: g a.e. holds for each k. Since 2::, fn is conver-
gent for almost all points, it follows from the Lebesgue Dominated Convergence
Theorem that 2::, fn defines an integrable function and that
Problem 22.11. Let f be a positive (a.e.) measurable function, and let
ei = JL*({x EX: i-i < f(x) ::: il)

for each integer i. Show that f is integrable if and only if 2::_002i ei < 00_
Solution. Let Ei = (x E X: i - I < I(x) ::: i}, i = 0, ±1, ±2, .... For
each n let <Pn = L~=-n 2i XE;. Then, there exists some function g with <Pn t g
a.e. Clearly, g is a measurable function and 0 ::: I ::: g a.e. holds.
Assume that I is integrable. Then, each <Pn is a step function, and in view of
<Pn ::: 21 (why?), it follows that
~
~
i ei = limIl-+-OO
f<Pn dJ.L ::: 2fl dJ.L < 00.
i=-oo
On the other hand, if L~-oo 2i ei < 00, then each <Pn is a step function, and
so g is integrable. Since 0 ::: I ::: g, Theorem 22.6 shows that I is also
integrable.
Problem 22.12. Let Un} be a sequence 01 integrable functions satisfying 0 :::

In+1 ::: In a.e. lor each n. Then, show that In +
0 a.e. holds if and only if
+
fin dJ.L o.
+
Solution. Assume f IndJ.L O. Let 1" +
I a.e.; clearly, I 2: 0 a.e. It follows
that f I dJ.L = 0, and thus (by Theorem 22.7) 1=0 a.e.
Problem 22.13. Let I be an integrable function such that I(x) > 0 holds lor
almost all x. II A is a measurable set such that fA I dJ.L = 0, then show that
J.L *(A) = O.
Solution. Let A be a measurable set satisfying fAI dJ.L = O. Next, consider

the set B = (x E A: I(x) ::: O}, and note that, by our hypothesis, J.L*(B) = O.
Also, for each n put
An = (x E A: I(x) 2: *}.
Then, A = (U:I An) U B, and IXA" ::: IXA a.e. holds for each n. Thus,
which shows that J.L*(An) = 0 for each n. This easily implies J.L*(A) = O.
Problem 22.14. Let (X, S, J.L) be aftnite measure space and let I: X -+ 1R be
an integrable ftmction satisfying I(x) > 0 lor almost all x. 110 < 8 ::: J.L*(X),
then show that
inf{L f dp,: E E A/J. and p,*(E) 2: B} > O.
Solution. We can assume that f(x) > 0 holds for each x E X. If for some
0< B ~ p,*(X) we have
then there exists a sequence {En} of A/J. satisfying P,*(En) 2: E and fEJ dp, < f.
for each n. Put Fn = U~n Ek and note that:
• each Fn is measurable;
• Fn+l ~ F/I holds for each n; and
• P,*(Fn) 2: P,*(En) 2: B holds for each n.
If F = n~1 Fn, then F is a measurable set and (by Theorem 15.4(2» we have
p,*(F) = lim p,*(FlI) 2: B > O.

IJ-+OO
From f XF" ~ L~/1 f XEp we infer that
and so f fXFn -l- O. The latter implies fXFn -l- 0 a.e. (see Problem 22.12), and
since f XF" -l- f XF, we infer that f XF = 0 a.e. In view of f (x) > 0 for each x,
the latter (in view of Problem 22.13) implies p,*(F) ~ 0, contrary to (*), and the
desired conclusion follows.
Problem 22.15. Let f be a positive integrable function. Define v: A -+ [0, (0)

by v(A) = fAf dp, for each A E A. Show that
a. (X, A, v) is a measure space.
b. If Av denotes the a-algebra of all v-measurable subsets of X, then show
that A ~ Av. Give an example for which A '# Av.
c. If p,*({x EX: f(x) = OJ) = 0, then show that A = Av.
d. If g is an integrable function with respect to the measure space (X, A, v),
then show that fg is integrable with respect to the measure space (X, S, fJ-),
and that
J =1gdv gf dfJ-.
Solution. (a) This part follows immediately from Problem 22.5.

(b) The measure v has initial domain A. Hence, by Theorem 15.3, A ~ Av
holds.
Consider 1R with the Lebesgue measure, and let f = XO.2). Since, in this
case, v([O, 11) = 0, it follows that every subset of [0, 1] is a v-null set (and
hence v-measurable). On the other hand, not every subset of [0, 1] is Lebesgue
measurable. Thus, A i= Av holds in this case.
(c) First, observe that v is a finite measure. Now, let A E Av with v*(A) = 0.
By Theorem 15.11, there exists some B E A such that A ~ Band v(B) = 0.
°
The relation IBf dfJ- = v(B) = combined with Problem 22.13, shows that
fJ-*(A) = 0. Thus, A E A. Now, if V E A v , then pick some W E A with
V ~ W and v(W) = v*(V) (Theorem 15.11). Note that v*(W \ V) = 0, and so,
by the preceding discussion, W \ V E A. Finally, V = W \ (W \ V) E A holds,
which shows that A = Av.
°° °
(d) We shall assume g(x) 2: and f(x) 2: for all x. Pick a sequence {<Pill of
v-step functions such that :s <PII t g v-a.e. Let
G = {x EX: f(x) > OJ.
Clearly, G E A. Since G C = {x E X: f(x) = OJ, it follows that v(G C ) =

IGJ dfJ- = 0. Since f is strictly positive on G, the arguments of part (c) show
that whenever A ~ G, we have:
1) If A E A v , then A E A, and
° °
2) By Problem 22.13, v*(A) = if and only if fJ-*(A) = (and in this case
A E A).
In particular, it follows that <pllf t fg fJ--a.e. holds. Now, if A E Av satisfies
v*(A) < 00, then
J XA dv = v(A n G) = inG f dfJ- = J

XAf dfJ-.
This implies that if <P is a v-step function, then <pf is fJ--integrable, and that
°
I <P dv = I <pf dfJ- holds. Now, note that :s <pllf t fg fJ--a.e. and I <Pn dv =
I <pllf dfJ-, show that fg is fJ--integrable and that I g dv = I gf dfJ- holds.
182 Chapter 4: TIlE LEBFSGUE IN'IEGRAL
Problem 22.16. Let I be an interval of1R, and let f: I -+ 1R be an integrable

function with respect to the Lebesgue measure. For a pair of real numbers a and
b with a =f:. 0, let J = {(x - b)/a: x E l}. Show that the function g: J -+ 1R
defined by g(x) = f(ax+b)forx E lis integrable and that flf d)... = lal fJg d)...
holds.
Solution. Assume first that f = XA for some measurable set A f: I. Clearly,

HA - b) f: J. Thus, in vi~w of the identity, XA(ax + b) = X~(A-b)(X), it follows
from Problem 15.5 that
i gd )'" = I!I)"'(A) = 1!ljf d)"'.

Thus, the formula is true for the characteristic function of a measurable set. It
follows that it is also true for step functions.
Now, let f be an upper function. Chooseasequence {¢n} of step functions with
¢n t f a.e. on I. If VrIl(X) = ¢n(ax + b) for x E J, then Vrn is a step function
on J and Vrn t g a.e. holds on J. (Note that if B f: I satisfies )"'(B) 0, then =
by Problem 15.5, we have )"'U(B - b») = I!I)"'(B) = 0.) Therefore,
lal1g d)'"
J
= lal n-H;X)
lim 1Vrn d)... = lim j¢1I d)... = j f d)"'.
J n .... oo I I
Thus, the formula holds true for every integrable function f on I.
Problem 22.17. Let (X, S, f,L) be a finite measure space. For every pair of
measurable functions f and g let
If-gl
dU, g) = f 1 + If
_ gl df,L.
a. Show that (M, d) is a metric space.

b. Show that a sequence {fn} of measurable functions {i.e., {fn} f: M)satis-
fies fn ~ f if and only if lim dUn, f) = 0.
c. Show that (M, d) is a complete metric space. That is, show that if a
sequence {fn} of measurable functions satisfies dUn, fm) -+ as n,
m -+ 00, then there exists a measurable function f such that lim dUn, f)
°
=0.
Solution. (a) We assume that functions equal f,L-a.e. are considered identical.
Only the triangle inequality needs verification. To this end, let f, g, hEM. The
triangle inequality follows immediately from the inequality
If(x) g(x)1 If(x) - h(x)1 Ih(x) - g(x)1

--'-=-:----=--- < + ,.
1 + If(x) - g(x)1 - 1 + If(x) - h(x)1 1 + Ih(x) - g(x)1
For details see the solution of Problem 9.11.

(b) Start by observing that for x 2: 0 and s > 0 we have:
2->
l+x -
Now, assume that limd(f", f) = 0 holds. Then, the inequality
easily implies that 1" ~ f·

For the converse, assume fn ~ f. If lim d(f" , f) =1= 0, then there exists
some s > 0 and some subsequence {gil} of {fll} with d(gn, f) 2: s for all n.
By passing to a subsequence, we can assume that gil --+ f a.e. (Theorem 19.4).
In view of l~I~~!jl ::: 1 and the finiteness of the measure space, the Lebesgue
Dominated Convergence Theorem yields 0 < s ::: limd(gll' f) = 0, which is
absurd. Hence, lim dUn, f) = 0 holds.
(c) Assume dUn, fm) --+ 0 as n, m --+ co. The inequality
shows that {f,,} is a JL-Cauchy sequence. Thus, by Problem 19.7, 1" ~ f

holds for some f, and by (b) above, lim d(f" , f) = 0 also holds.
Conversely, if limd(f", f) = 0 holds, then by part (b) above fn ~ f, which
implies that {f,,} is a JL-Cauchy sequence.
Problem 22.18. Let f: lR -7 lR be a Lebesgue integrable function. For each fi-

nite interval I let fl=J..tI)JlfdA. and EI={XEI: f(x»fl}. Show
that
fI
If - fll dA. = 2 [ U -
JEI
it) dA..
184 Chapter 4: THE LEBESGUE INTEGRAL
Solution. We follow the notation of the problem. Start by observing that
lEI
r(J - II) d'A + ! I\E,
(J - II) d'A = !(J -
I
II) d'A
= l ld'A-1 IId 'A
= 11 d'A -1 I d'A = O.
Consequently,
r(J - II) d'A = !

lEI I\E,
(II - I) d'A.
Now, note that
= r (J -
lEI
II) d'A + !I\E,
(JI - I) d'A
= r(J - II) d'A + r(J - II) d'A 2 r(J - II) d'A.

lEI lEI
=
lEI
Problem 22.19. Let I: [0, 00) -+ JR be a Lebesgue integrable lunction such

that f~ I(x) d'A(x) = 0 lor each t ::: O. Show that I(x) = 0 holds lor almost
all x.
Solution. Start by observing that
1 [a,b)
I d'A = r
l[o,b)
I d'A - r
l[O,a)
I d'A =0
holds for each interval [a, b). By Problem 22.5, we see that fAI d'A = 0 holds
for each a-set A. From Problem 15.2 (and the Lebesgue Dominated Convergence
Theorem), we see that fAI d'A = 0 holds for each Lebesgue measurable subset
A of JR.
Now, let X = {x E JR.: f(x) > O} and Y = {x E X: f(x) < O}. Clearly, X
and Yare both Lebesgue measurable sets, and
Ixf d)" = !rf d ).. = O.
Now, invoke Problem 22.13 to obtain )"(X) = )..(y) = O. Therefore, f(x) = 0

holds for almost all x.
Problem 22.20. Let (X, S, f.L) be a measure space and let f, fl' 12, ... be non-
negative integrable functions such that fn -+ f a.e. and lim Jfn df.L = Jf d f.L.
If E is a measurable set, then show that
lim r
n->oo lE
fn df.L =
lE
r
f df.L.
Solution. Assume that the integrable functions f, fl' 12, ... are non-negative
satisfying the hypotheses of the problem and let E be a measurable set. Then the
functions f XE, fl XE, 12XE, . .. are non-negative and integrable (because 0 ::::
fXE :::: f and 0 :::: fnXE :::: f,,) and f"XE -+ fXE holds. Using Fatou's
Lemma, we see that
Similarly, we have
r f df.L :::: liminf lEcr fn df.L.

lEc
Therefore,
I
f df.L = rf df.L + l~r f df.L :::: liminf hrfn df.L + liminf l~r f" df.L
h
:::: lim inf(L f" df.L + L/n df.L)
= liminf I fndf.L
= I f df.L,
186 Chapter 4: TIm LEBESGUE INTEGRAL
where the second inequality holds by virtue of Problem 4.7(b). It follows that
r f dlL + J£cr f dlL = liminf JrEfn dlL + liminf JEer /" dIL,
JE
and from (*) and (**), we see that
Now, let {gn} be a subsequence of {fn}. Then,
By the preceding conclusion, we infer that
and so there exists a subsequence {gk,,} of the sequence {gn} such that
limIEgk"dIL = IEf dIL·
Thus, we have demonstrated that every subsequence of the bounded sequence
of real numbers UE fn d IL) has a convergent subsequence to IE f d IL. This means
that
lim r/" dlL = JrEf dlL

11->00 JE
holds; see Problem 4.2.
Problem 22.21. If a Lebesgue integrable function f: [0, 1] ~ 1R satisfies

101 211 f(x) d)'(x) =
X ° °
for each n = 0, 1,2, ... , then show that f = a.e.
Solution. Let an integrable function f: [0, 1] ---+ 1R satisfy
l1x2n f(x)d)'(x) = ° for n = 0,1,2, ....

Since the algebra offunctions generated by {I, x 2 ) is uniformly dense in C[O, 1]
(see Problem 1l.5), it follows that Id g(x)f(x)d)'(x) = ° holds for all g in
Section 22: INTEGRABLE ruNCTIONS 187
e[o, 1]. Consider the two measurable sets
E = (x E [0, 1]: I(x) > O} and F = (x E [0, 1]: I(x) < O}.
We have to show that )"'(E) = )"'(F) = O. We shall establish that )"'(E) = 0 holds
and leave the identical arguments for F to the reader.
Pick a sequence {K II } of compact sets and a sequence {all} of open sets
of [0, 1] satisfying KII ~ E ~ On for each n, KII t, all .j" and )"'(E) =
lim )...(Kn) = lim)...( On). (Here we use the regularity of the Lebesgue measure.) For
each n there exists (by Theorem 10.8) a continuous function gn: [0, 1] ~ [0, 1]
satisfying gll(x) = 1 for each x E KII and I(x) = 0 for each x ¢. all' Clearly,
Ig,,/1 ::s III and gill ~ IXe a.e. By the Lebesgue Dominated Convergence
Theorem, we get
lim
11_00
r
io
gll(x)/(x) d)"'(x) = r
io
l(x)Xe(x) d)"'(x) = rid)....
ie
Taking into account that fOI gIlCr)/(x) d)"'(x) = 0 holds for all n, we infer that
fe l d)... = O. Now, invoke Problem 22.13 to infer that )"'(E) = 0, as claimed.
Problem 22.22. For each n consider the partition
{O, Til, 2 . Til, 3 . Til, ... , (2" - 1) . Til, 1}
01 the interval [0, 1] and define the function r,,: [0, 1] -i>- JR. by rll (1) = -1 and
rll(x) = (-1/- 1 lor (k - I)T" ::s x < kT" (k = 1,2, ... ,2").
a. Draw the graphs 011'1 and r2.

b. Show that if I: [0,1] -i>- JR. is a Lebesgue integrable function, then
lim trll(x)/(x)d)"'(x)
11-+00 io
= O.
Solution. (a) The graphs of rl and r2 are shown in Figure 4.1.
(b) By Theorem 22.12, it suffices (how?) to establish the claim for the case
I = X[a,b), where [a, b) is a subinterval of [0, 1]. Clearly, foirll(x)x[a,b) d)"'(x) =
f:rll(x) dx. Therefore, it suffices to show that lim !:rll(x) dx =0 holds for each
O::sa<b::s 1.
To this end, fix 0 ::s a O. Fix no such that 2-110 <
min {s, b;a }. Pick n ::: no and consider the partition {O, f.;, .t" ... ,
2'~:-1 ,I}; for
188 Chapter 4: TIIE LEBFSGUE INTEGRAL
y y
Y='1 (x) y = r2(x)
!"-l
: :
5
:
!:
5 i
x x
i!
. i2lJ: !:::
-1 -1 ---
1 i
~
FIGURE 4.1. The Graphs of rl and r2
simplicity, let Xi = :in and note that the points a and b are related to the Xi as
shown in Figure 4.2.
Since for any three
b
consecutivex points Xi-I, Xi,
b
.Ci
Xi+1 we have A,_I rll(x)dx =
0, we see that fa rn(x)dx = fa krn(x)dx + fe rn(x)dx, where c = Xm-I or
c = Xm; see Figure 4.2. Hence,
11brn(X)dxl ::::: lxklrn(X)ldX+ l\'n(X)ldX
= (Xk - a) + (b - c) < e + 2e = 3e
for each n ::: no. This means that lim f:rll(x) dx = 0, as desired.
Problem 22.23. Let {En} be a sequence of real numbers such that < Ell < 1 °
for each n. Also, let us say that a sequence {An} of Lebesgue measurable subsets
of[O, 1] is consistent with the sequence {En} if A(An) = En for each n. Establish
the following properties of[ En}:
a. The sequence {En} converges to zero if and only if there exists a consistent
sequence {All} of measurable subsets of[O, 1] such that L:::IXAn(X) < 00
for almost all x.
b. The series L::: I En converges in 1R if and only if for each consistent se-
quence {All} ofmeasurable subsets of[O, 1] we have L:::IXA,,(X) < oofor
almost all x.
Solution. (a) If ell --7 0, then let All = (0, en) (n = 1,2, ... ), and note that
A(An) = en holds for each n and that L:::I XAn(X) < 00 for each X E [0,1].
• II • .. .. I ....
FIGURE 4.2.
For the converse, assume that there exists a consistent sequence of measurable
subsets {All} of [0,1] satisfying L~l XAn(X) < 00 for almost all x. For each
n let BII = U~1l Ak and note that BII ,). B = n~l Bk. If A(B) > holds,
then note that L~l XAn(X) = 00 for each ,~ E B (why?), which contradicts our
°
hypothesis. Thus, A(B) = 0. From the continuity of the measure (Theorem 15.4),
°
we see that )"(B Il ) ,). 0. In view of An ~ B n, we have < 8n = A(An) ::: )"(B II )
for each n, and so lim 8 n = 0.
(b) Assume L~l 8 11 < 00 and that {All} is a consistent sequence of measurable
subsets of [0, 1]. Then,
~1
.i.....J XA"d)" =~ II ) ~ 8n <
.i.....J )"(A = .i.....J 00,
n=l [0,1] 11=1 n=l
and so, by the series version of Levi's Theorem 22.9, we have L~l XAn(X) < 00
for almost all x.
For the converse, assume that for every consistent sequence {An} of measurable
subsets of [0,1], we have L::;"=l XA,,(X) < .00 for almost all x. Suppose by
way of contradiction that L~l 8 11 = 00. Using an inductive argument (how?);
we see that there is a sequence {k ll } of strictly increasing natural numbers such
that L7:::kn+18; > 1 holds for each n. Next, for each n we can choose (how?)
subintervals Ak,,+l, Akn+2' ... ,Akn+ , of [0, 1] such that )"(A;) = 8; for kll + 1 :::
k
i ::: k ll +1 and U;:::kn+1A; = [0, 1]. Now,notethatthesequenceofmeasurablesets
{All} is consistent with {8 11 } and L~l XAn(X) = 00 holds for each x E [0,1],
contrary to our hypothesis. So, L::;"=l 8 11 < 00 must hold.
Problem 22.24. Let (X, S, JL) be afmite measure space and let f: X -+ JR be a
measurable function.
a. Show that if r r
is integrable for each n and that lim J d JL exists in lR,
then If (x) I ::: 1 holds for almost all x.
b. If ris integrable for each n, then show that Jr
dJL = c (a constant)
for n = 1, 2, ... if and only if f = XA for some measurable subset A
ofX.
Solution. Keep in mind that r

denotes the function f": X ----* JR defined by
rex) = [f(x))" for each x EX.
(a) Assume that r
is Lebesgue integrable for each n and that lim J r dJL exists
in JR. Assume by way of contradiction that the measurable set
E = {x EX: If(x)1 > I}
satisfies JL*(E) > 0. From the identity E = U~l Ek, where
Ek = {x EX: If(x)1 ~ 1 + t},

we see that there exists some 8 > 1 such that the measurable set
190 Chapter 4: TIIE LEBFSGUE INTEGRAL
F = {x EX: I/(x)1 > o} satisfies p.*(F) > O. Now, note 1211 ::: 0211 XF holds for
each n, and so from
we infer that lim 11211 df-L = 00, contradicting the existence in 1R of the limit
lim II" df-L. Hence, 1/(.x)1 ::::: 1 must hold for almost all x.
(b) Assume II" df-L = c holds for each n = 1,2, .... By part (a), we know
that I/(x)1 ::::: 1 holds for almost all x E X. Now, define the sets A = {x E
X: I(x) = I}, B = {x EX: f(x) = -I}, and C =
{x EX: I/(x)1 < I}. Then,
for each n we have
f 1" df-L = i 1" dp. + i l " df-L + i 1" dp.
= i Idf-L + i(-1)1I df-L + i l " dp.
= f-L*(A) + (_1)" p.*(B) + i 1" dp. = c.
Since I"(x) -+ 0 holds for each x E C,itfollowsfromtheLebesgueDominated

Convergence Theorem that lim Ie 1" d p. = O. Hence,
lim [f-L*(A)
11 .... 00
+ (-1)"f-L*(B)] = c.
Since lim(-I)" does not exist, we infer that p.*(B) = 0, and therefore, c =
p.*(A) = f-L*(A) + leI" dp. for each n. In particular, we have Ie 12 dp. = 0, and
so I (x) = 0 must hold for almost all x E C. The latter implies that I = XA a.e.
holds.
23. TIlE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL
Problem 23.1. Let I: [a, b] -+ 1R be Riemann integrable. Show that I is

Riemann integrable on every closed subinterval of[a, b]. Also, show that
Id I(x)dx ~ Ie I(x)dx + Id I(x)dx

holds lor evelY three points c, d, and e of[a, b].
Section 23: TIIE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 191
Solution. Let f: [a, b] -+ 1R. be a Riemann integrable function and let [u, v]
be a closed subinterval of [a, b]. If f: [u, v] -+ 1R. is discontinuous at some
point x E [u, v], then f: [a, b] -+ 1R. is also discontinuous at the point x-note
that, in this case, there exists a sequence {-~n} of [u, v] (and hence of [a, b])
such that {f(x n)} does not converge to f(x). Thus, the set D of all points of
discontinuity of f: [u, v] -+ 1R. is a subset of the set of all points of discontinuity
of f: [a, b] -+ 1R.. Since f: [a, b] -+ 1R. is Riemann integrable, we know that
)"'(D) = 0, and so (by Theorem 23.7) the function f: [u, v] -+ 1R. is Riemann
integrable.
Now, assume that a ::: c < e < d ::: b. Since f: [c, e] -+ 1R. is Riemann in-
tegrable (and hence Lebesgue integrable), there exists a sequence of step functions
°
{<Pn} over [c, e] (i.e., <p,,(x) = holds for x rt. [c, e]) with <PII(X) t f(x) for al-
most all x E [c, e]. Similarly, there exists a sequence of step functions {¥rn} over
[e, d] such that ¥rn(x) t f(x) holds for almost all x E [e, d]. Then, {<Pn + ¥rll}
is a sequence of step functions over [c, d] satisfying <PnCx) + ¥r,,(x) t f(x) for
almost all x E [c, d]. Therefore,
I c
d
f(x)dx = 1[c,d]
f d)'" = lim
II~OO
1[c,d]
(<p" + ¥rn)d)'"
= lim
,,~oo
1[c,d]
<p" d)... + lim
n~oo
1
[c,d]
¥rn d)...
= 1[c,e]
f d )...+l fd)'"
[e,d].
= Ie f(x)dx + [d f(x)dx.
Now, the equality t f(x)dx = f: f(x)dx+ 1: f(x)dx for arbitrary elements
c, d, and e of [a, b] can be obtained by considering all possible cases. We prove
it for one such case and leave the rest for the reader. Assume that a ::: e < c <
d ::: b. Then, by the preceding case, we have
[d f(x)dx = [C f(x)dx + ld f(x)dx = -l e

f(x)dx + ld f(x)dx,
from which it follows that fee f (x) dx + 1: f (x) dx = t f (x) dx.

Problem 23.2. Let f: [a, b] -+ 1R. be Riemann integrable. Then, show that
l a
f(x) dx = lim
II~OO
b-aI::n (
--
n 1=1
.
f a + i(b-a))
n
.
Solution. The conclusion follows from Theorem 23.5 by observing that
n
b~a L I(a + ib~a) = Sj(P n, Tn),
;=1
where the partition Pn = {Xo, Xl, ... ,Xn } and T = {tl, ... , t n } satisfy Xi =
a+ i b~a (0 :::: i :::: n) and t; = X; (1 :::: i :::: n).
Problem 23.3. Let Un} be a sequence olRiemann integrable functions on [a, b]

such that Un} converges uniformly to a function I. Show that I is Riemann
integrable and that
lim
11-+00
lb
a
In(x)dx = lb
a
I(x)dx.
Solution. Choose some k such that Ilk(X) - In(x)1 < 1 holds for all n > k
and all X E [a, b]. Thus; Ilk(X) - l(x)1 :::: 1 holds for all x E [a, b]. Since Ik
is bounded, it is easy to see that there exists s?me M > 0 such that I/(x)1 :::: M
holds for all x E [a, b].
If D n S;; [a, b] denotes the set of discontinuities of In, then (by Theorem 23.7)
D = U:::I Dn satisfies 'A.(D) = O. Since each In is continuous on [a, b] \ D, it
follows from Theorem 9.2 that I is continuous on [a, b] \ D, i.e., I is continuous
almost everywhere. By Theorem 23.7, I is Riemann integrable.
For the last part, let E > O. Pick some k such that Iln(x) - l(x)1 < E for all
n ::: k and all x E [a, b]. So, for n ::: k we have
and this shows that limn-+oo 1: I,,(x)dx = J: I(x)dx.

Problem 23.4. For each n. let I,,(x): [0, 1] -+ 1R be defined by In(x) = nt;~1
lor all x E [0, 1]. Then. show that lim Jo1In (x) dx = 4.
Solution. Integrating by parts, we get
11 In(x)dx
0 0 . 0
xn
= I+x I+ 111 XII
(I+X)2 dx = "21 + 11
0
xn.
(I+x)2 dx.
Since 0 :::: (I~:)2 :::: 1 holds for all x E [0, 1] and lim (I~:)2 = 0 for each x in
Section 23: TIm RIEMANN INTEGRAL AS A LEBFSGUE INTEGRAL 193
[0, 1), the Lebesgue Dominated Convergence Theorem yields lim fd (1~~t)2 dx = O.
Thus, from (*), we see that lim fol fll (x) dx = t.
Problem 23.5. Let f: [a, b] -+ lR be an increasing function. Show that f is
Riemann integrable.
Solution. Let PII = {xo, XI, ••• , XII} be the partition that subdivides [a, b] into
n subintervals ofequallength b~a. Since f is increasing, note that mi = f(Xi-l)
and Mi = f(Xi) hold for each lSi S n. Next, observe that
II
= L[f(Xi) - f(Xi-\)](Xi -Xi-\)

i=1
= [feb) - f(a)] . b~a,
=
holds for each n. Thus, I*(f) - I*(f) 0 and so f is Riemann integrable.
An alternate proof goes as follows: According to Problem 9.8 the set of disconti-
nuities of f is at-most countable-and hence, it has Lebesgue measure zero. Now,
Theorem 23.7 guarantees that f is Riemann integrable. (See also Problem 21.8.)
Problem 23.6 (The Fundamental Theorem of Calculus). If f: [a, b] -+ lR is

a Riemann integrable fimction, then define its area function A: [a, b] -+ lR by
A(x) = f~t f(t)dt for each X E [a, b]. Show that
a. A is a uniformly continuolls function.
b. Iff is continuous at some point c of[a, b], then A is differentiable at c and
A'(C) = fCc) holds.
c. Give an example of a Riemann integrable function f whose area function
A is differentiable and satisfies A' i= f.
Solution. (a) Choose some M > 0 with If(x)1 S M for each x in [a, b]. The
uniform continuity of A follows from the inequalities
IA(x) - A(y)1 = 11 Y
f(t)dt IS 11 If(t)1 dt ISMix - YI·
Y
(b) Let f be continuous at some point c E [a, b] and let s > O. Choose some
8> 0 such that If(x)- f(c)1 < s holds whenever x E [a, b] satisfies Ix-cl < 8.
Then, for x E [a, b] with 0 < Ix cl < 8 we have If(t) - f(c)1 < E for all tin
the interval with endpoints x and c, and so
IA(x) - A(c) - f(C)1 =

x- c
-l-Iic
Ix - cl x
f(t)dt - f(c)(x - C)I
= 'x~c,lic[f(t) - f(c)] dt I
< _1_.
- Ix-cl
six - cl = s.
This shows that A'(c) exists and that A'(c) = f(c) holds.
(c) We consider the function f: 1R -+ 1R of Problem 9.7 defined by f(x) = 0 if
x is irrational and f(x) = ~ if x = 7,- with n > 0 and with the integers m and n
without having any common factors other than ±l. It was proven in Problem 9.7
that f is continuous at every irrational and discontinuous at every rational. This
implies that f restricted on an arbitrary closed interval [c, d] is continuous almost
everywhere and f = 0 a.e. From Theorems 23.6 and 23.7, we infer that f is
Riemann integrable over [c, d] and t
f(x)dx = f d)'" = O. J
J:
In particular, if[a, b] is any closed interval, then A(x) = f(t) dt = ofor each
x E [a, b]. Thus, A'(x) = 0 for each x E [a, b], and consequently A'(x) =1= f(x)
at each rational number x in [a, b].
Problem 23.7 (Arzela). Let Un} be a sequence ofRiemann integrable functions

on [a, b] such that lim fn(x) = f(x) holdsfor each x E [a, b] and f is Riemann
integrable. Also, assume that there exists a constant M such that 1f,,(x)1 ::5 M
holds for all x E [a, b] and all n. Show that
lim
n--).oo
lb
a
fn(x)dx = lb
a
f(x)dx.
Solution. Using Theorem 23.6 and the Lebesgue Dominated Convergence

Theorem, we see that
b 1 1 lb
la
f(x)dx =
[a.b]
f d)'" = lim
n-+oo [a.b]
fn d)... = lim
n-.oo a
fn(x)dx.
Problem 23.8. Determine the lower and upper Riemann integrals for the func-
tion f: [0,1] ---7 1R definedby f(x) =
0 if x isarationalnumberand f(x) I =
if x is an irrational number.
Section 23: THE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 195
Solution. Let P be a partition of [0, 1]. Since each interval contains rational and
irrational numbers, we have mi = 0 and Mi = 1 for all i. Thus, S*(f, P) = 1
and S*(f, P) = 0 for all partitions P. Therefore, I*(f) = 1 and I*(f) = O.
Problem 23.9. Let C be the Cantor set (see Example 6.15). Show that Xc is
l
Riemann integrable over [0, 1]. and that Jo Xc dx = O.
Solution. Note that Xc is continuous at every point of [0, 1] \ C and discon-

tinuous at every point of C. Since A(C) = 0, it follows from Theorem 23.7 that
Xc is Riemann integrable over [0, 1]. Since Xc = 0 a.e. holds, we see that
1 o
1 xc(x) dx = r
i[o.l]
Xc dA = O.
Problem 23.10. Let 0 < E < 1, and consider the E-Cantor set C€ of [0, 1].
Show that Xc, is not Riemann integrable over [0, 1]. Also. determine I*(Xc,) and
I*(Xc,).
Solution. Consider the e-Cantor set for some 0 < e < 1. Since Ce is nowhere
dense in [0, 1], it is easy to see that Xc, is discontinuous at every point of C e and
continuous at every point of [0, 1] \ Ceo Since A(Ce) = e > 0, it follows from
Theorem 23.7 that Xc, is not Riemann integrable ..
Now, let P = {xo, XI, ... , xn} be a partition of [0, 1]. Since Ce is nowhere
dense, it follows that mi = 0 for each 1 ::: i ::: n. Thus, S*(Xc" P) = 0 for
each partition P, and so I*(Xc) = O. Clearly, Mi = 1 if [Xi-I, X;] n Ce =1= 0,
and Mi = 0 if [Xi-I, X;] n C e = 0. Since Ce = U7=1 [Xi-I, X;] n Ce, it follows
that
n n
e = A(Ce) ::: I:>([Xi-l, x;] n Ce) ::: L Mi(Xi - Xi_I) = S*(Xc" P),
i=1 i=1
and so I*(Xc) ::: e.

On the other hand, if 0 < 8 < 1 - e, then there exist pairwise disjoint open
subintervals (al. b l ), ... , (an, bn) such that
n
[a,., b;] £;; [0,1] \ Ce (1 ::: i ::: n) and - L(bi - ai) > 1 - e - 8.
i=1
The endpoints of all these subintervals together with 0 and 1 form a partition P
of [0, 1] such that
e ::: J*(XC.) ::: S*(XC" P) ::: 1 - (1 - e - 8) = e + 8.

Since 0 < 8 < 1 - e is arbitrary, it easily follows that 1*(Xc.) = e.
Problem 23.11. Give a proof of the Riemann integrability of a continuous func-

tion based upon its uniform continuity (Theorem 7.7).
Solution. Let e > 0. Since (by Theorem 7.7) f is uniformly continuous, there
is some 8 > 0 such that x, y E [a, b] and Ix - yl < 8 imply If(x) - f(y)1 < e.
Let P be a partition of [a,b] withmesh IPI < 8. Then,M; -m; < e holds for
each 1 ::: i ::: n (why?), and so
n
°: : J*(f) - 1*(f) ::: L(M; ;:- m; )(Xi - Xi-I) < e(b - a).
;=1
Since e > 0 is arbitrary, we see that 1*(/) = 1*(/) holds, and therefore f is
Riemann integrable.
Problem 23.12. Establish the following change of variable formula for the Rie-
mann integral of continuous functions: If [a, b] -4 [c, d] -4. lR are con-
tinuous functions with g continuously differentiable (i.e., g has a continuous
derivative), then
l
19(b)
f(g(x))g'(x)dx = f(u)du.
a g(a)
Solution. We shall apply the Fundamental Theorem of Calculus in connection

with the Chain Rule. We consider the two functions F, G: [a, b] --+ lR defined by
F(x) =
1
8 (X)
g(a)
f(u)du and G(x) = l x
f(g(x))g'(x)dx
for all x E [a, b]. Next, we shall compute the derivatives of F and G separately.
For the derivative of F we use the Fundamental Theorem of Calculus and the
Chain Rule to get F'(x) = f(g(x))g'(x) for each x E [a, b]. For the derivative of
G, the Fundamental Theorem of Calculus yields G'(x) = f(g(x))g'(x) for each
x E [a, b]. So, F'(x) = G'(x) for boll x E [a, b].
The latter implies that there exists a constant c such that F(x) = G(x) + c for
all x E [a, b]. Letting x = a and taking into account that F(a) = G(a) = 0,
Section 23: TIIE RlEMANN INTEGRAL AS A LEBESGUE INTEGRAL 197
we get c = 0. Thus, F(x) = G(x) for all x E [a, b]. Finally, letting x = b, we
obtain
hf(g(x))g'(x)dx ='jRW f(u)dll,

la R(a)
as desired.
Problem 23.13. Let f: [0, (0) -r JR be a continuous function such that

lim.r-oo f(x) = D. Show that limll -+ oo
a
Jo
f(nx) dx = aD for each a > 0.
by f,,(x)
°
Solution. Fix a > and then define the sequence of continuous functions (f,,}
=
f(nx). Clearly, limll -+ oo f,,(x) =
0 holds for all x E (0, a]. We
claim that the sequence of functions (f,,} is uniformly bounded on the interval
[0, a]. Indeed, since limx-+oo f(x) =
0 holds, there exists a number M > Osuch
that If(x)1 < 101 + 1 whenever x > M:
Also, since f is a continuous function,
it is bounded on the interval [0, M]. Thus, there exists a constant C such that
If(x)1 ~ C holds for all x, and hence If,,(x)1 = If(nx)1 ~ C holds for all x.
Now, an application of the Lebesgue Dominated Convergence yields
II_X
lim
Jor f(nx)dx = 11-+00
lim r fll(x)dx = r odx = aD.
Jo Jo
Problem 23.14. Let f: [0, (0) -r JR be a real-valued continuous junction such

that f(x + 1) = f(x) for all x 2: 0. If g: [0, 1] -r JR is an arbitrary continuous
function, then show that
Solution. Let the functions f and g satisfy the hypotheses of the problem.
Observe first that an easy inductive argument establishes that f(x + k) = f(x)
holds for all x 2: 0 and all non-negative integers k.
The change of variable II = nx yields
198 Chapter 4: THE LEBESGUE IN1EGRAL
Letting t = u - i + 1, we get
1
;
;-1
g(~)!(u)du = rg(t+~-I)!(t +
Jo
1
i - l)dt = r
Jo
1
gC+:1-1)!(t)dt.
Consequently,
11
o
g(x)!(nx)dx = 11 [2:::
0
II
;=1
~gC+~-1 )]!(t)dt = 11
0
hn(t)dt,
where hn(t) = [L~=l ~g(t+~-I )]!(t). Clearly, hn is a continuous fu~ction

defined on [0, 1]. In addition, note that if Ig(x)1 :::: K and 1!(x)1 :::: K hold for
each x E [0, 1], then
Ihn(t)1 :::: K2 for all t E [0, 1],
i.e., the sequence {h n } is unifonnly bounded on [0, 1]. Next, note that
implies ;~I :::: '+~-I :::: ~. Thus, if mj and Mi denote the minimum and
t :::: 1 °: :
maximum values of g, respectively, on the closed interval [;~I, ~], then
m'! < g(,+;-I) < M~

1- n - I
holds for each °: : t :::: 1. Next, put
n n
Rn = 2:::~m7 and Sn = 2::: ~Mi,
;=1 ;=1
and note that Rn and SII are two Riemann sums-the smallest and largest ones,
respectively-for the function g corresponding to the partition {o,~,~, ... ,
n;l, I}. Hence, limn .... ooR n =limn .... ooSII = Jo1g(x)dx. From
n
Ihn(t)-Rn·!(t)1 = 1[?:~gC+~-I)]!(t)-Rn.!(t)1
1=1
n
= ([2::: ~gC+~-I)] -
;=1
Rn) ·1!(t)1
Section 23: TIlE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 199
Jo1
we see that limn-oo hnCt) = ICt) g(x) dx-in fact, the sequence {h n} con-
verges uniformly (why?).
Now, use (*) and the Lebesgue Dominated Convergence Theorem to obtain
lim tg(x)/(nx)dx = lim thnCt)dt

n-..OO 10 n->oo 10
=
10t[n-..oo
lim hnCt)] dt
=1 1 1
[/(t)
1g
(X)dX)]dt
= (1 (1 1 1
I(t)dt) . g(x)dx).
Problem 23.15. Let I: [0, 1] -+ [0,00) be Riemann integrable on every closed

subinterval 01 (0, 1]. Show that I is Lebesgue integrable over [0, 1] if and only
iflime.t.o JEl I(x) dx exists in R. Also, show that if this is the case, then we have
JI dJ... = lime.t.o J:l I(x)dx.
Solution. Assume that I is Lebesgue integrable. Let {en} be an arbitrary
sequence of (0, 1] with en +
O. For each n, we consider the upper function
gil = IX[8n ,l]' Then, gn t I a.e. holds and so, by Theorem 21.6, we have
1/
ldJ...= lim JglldJ...= lim 1 (X)dx.
J n-+oo n-+oo Elf
1
This shows that lim8 to Je I (x) dx exists and that
lim 11/(X)dX
8tO 8
= JI dJ....
Conversely, assume that the limit exists. Let 8 n = ~ and consider the sequence
of upper functions {gn} as previously (Le., let gn = I X[8 n .lj). Then, gil t I a.e.
and
lim Jgn dJ... = lim

n->OO n->oo
11
8n
I(x)dx = lim
8 to
118
I(x)dx < 00.
By Theorem 21.6, I is an upper function, and hence, Lebesgue integrable.

Problem 23.16. As an application of the preceding problem, show that the func-
tion f: [0, 1] -+ 1R defined by f(x) = x P if x E (0, 1] and f(O) = 0 is Lebesgue
integrable if and only if p > -1. Also, show that if f is Lebesgue integrable,
then
Jf dA =_I_.
l+p
- rl I p+1
Solution. If 0 < e < 1, then note that Je x P dx = ~~I for p =1= -1 and
Jelx-Idx = -lne. Thus,lime~oJelxPdx exists if and only if p > -1, and,
in this ~ase, the limit is P~I. The conclusion now follows immediately from the
precedmg problem.
Problem 23.17. Let f: [0,1] -+ 1R be a function and define g: [0,1] -+ 1R by

g(x) = ef(x).
a. Show that iff is measurable (or Borel measurable), then so is g.
b. Iff is Lebesgue integrable, is then g necessarily Lebesgue integrable?
c. Give an example ofan essentially unboundedfunction f which is continuous
on (0, 1] such that ris Lebesgue integrable for each n = 1,2, .... (A
function f is "essentially unbounded," if for each positive real number
M > 0 the set {x E [0, 1]: If(x)1 > M} has positive measure.)
Solution. (a) Let h(x) = eX and note that g = h 0 f. The conclusion follows
from the identity (h 0 f)-I (V) = f- I (g-I(V)) and the fact that h is a continuous
function.
(b) The measurable function g need not be necessarily Lebesgue integrable.
Here is an example. Consider the function f: [0, 1] -+ 1R defined by f(x) = Jx;
at x = 0 we let f (0) = O. If 0 < e < 1, then the change of variable t = Jx
yields
11e
f(x)dx = 11
e
dx
r:: = 2
~x
/1
vre
dt = 2(1 - v's).
Therefore, from Problem 23.16, we see that f is Lebesgue integrable and J fdA =
2. On the other hand, for each 0 < E < 1 the change of variable u = yields Jx
11
e g(X)dA(X) =
11 I
e e 7f dx = 2
/;7;
I ~ du ::: 2
u /;7;
I ell du = 2(e7< -
I
1).
This implies lime~oJ: e 7x dx = 00, and so by Problem 23.15 the function g is

not Lebesgue integrable over [0, 1].
(c) The function f: [0, 1] -+ lR defined by f(x) = lnx for 0 < x < 1 and
f(O) = 0 satisfies fol r(x)d"A(x) = (-1)lI n ! for each n = 1,2, ....
Problem 23.18. Let f: [0, 1] -+ lR be Lebesgue integrable (with respect to the

Lebesgue measure). Assume that f is differentiable at x = 0 and f(O) = O. Show
that the function g: [0, 1] -+ lR defined by g(x) = x-~ f(x) for x E (0, 1] and
g(O) = 0 is Lebesgue integrable.
Solution. Start by observing that (by Problem 23.16) the function hex) = x-~
for x E (0, 1] is Lebesgue integrable over [0, 1]. Since f(O) = 0 and 1'(0)
exists, there exist 0 < 8 < 1 and M > 0 such that If(x)1 ::: Mx for all
o ::: x :::3 8. Since for 8 ::: x ::: 1 we have _c~ ::: 8-~, we can assume that
M > 8-"2. Now, note that for 0 < x ::: 1, we have
Ig(x)1 = Ix-% f(x)1 ::: M {x-~ if 0 < x ::: 8

If(x)1 if 8 < x ::: 1
::: M(h + Ifl)(x).
Since h + If I is integrable and (obviously) g is measurable, Theorem 22.6 guar-

antees that g is also Lebesgue integrable.
Problem 23.19. Let f: [a, b] x [c, d] -+ lR be a continuousfunction. Show that

the Riemann integral of f can be computed with two iterated integrations. That
is, show that
bfd f(x,y)dxdy= lb [ Id f(x,y)dy]dx= fd [ lb f(x,y)dx]dy.

1 a c a e c a
Generalize this to a continuous function of n variables.
Solution. Note first that the functions
x I--+ lcr
d
f(x, y)dy and y I--+ lba f(x, y)dx
are both continuous-and so both iterated integrals are well defined. Indeed, since
the function f is uniformly continuous, given e > 0 there exists some 8 > 0
such that IXI - x21 < 8 and IYI - Y21 < 8 imply If(xl, YI) - f(X2, Y2)1 < e.
Thus, if IXI - x21 < 8 and IYI - Y21 < 8 both hold, then
and
Let P = {xo, XI. ••• , xn} be a partition of [a, b] and Q = {Yo, YI,···, yd
be a partition of [e, d]. Put Rij = [Xi-I, xil X [Yj_l, Yj], and then define
mij = inf{/(x, y): (x, y) E Rij} and Mij = sup{J(x, y): (x, y) E Rij}.
From the inequality mij :::: I(x, y) :::: Mij for (x, y) E Rij, it follows that
for all Xi-I:::: x :::: Xi, and so
n k
S*(f, P x Q) = L Lmij(Xi - Xi-I)(Yj - Yj-I)
i=1 j=1
: : t t [tl (jYj
i=1 j=1 .11-1 Yj-I
I(x, y)dy) dx
= l b([d I(x, y)dy) dx

/I k
:::: LLMij(Xi -xi-d(Yj - Yj-I)
i=1 j=1
= S*(f, P x Q).
Section 23: THE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 203
Since P and Q are arbitrary and f is Riemann integrable, it follows that
The other equality can be proven in a similar manner.
Problem 23.20. Assume that f: [a, b] -+ JR and g: [a, b] -+ JR are two contin-
uousfunctions such that f(x) ::: g(x)for each x E [a, b]. Let
A = «x, y) E JR2 : x E [a, b] and f(x) ::: y ::: g(x)}.
a. Show that A is a closed set-and hence, a measurable subset ofJR2 .

b. If h: A -+ JR is a continuous function. then show that h is Lebesgue inte-
grable over A and that
b f~(X)
1 l A
hdA=
a
[.
J(x)
h(x,y)dy Jdx.
Solution. (a) Let «XII' YII)} be a sequence of A such that XII -+ x and YII -+ y.
From the inequality f(x lI ) : : : YII ::: g(x lI ) and the continuity of f and g,itfollows
that f(x) ::: y ::: g(x), i.e., (x, y) E A. Thus, A is a closed set.
(b) Let c < inf{f(x): X E [a, b]} and d > sup{g(x): X E [a, b]}. Consequently,
A 5:; [a, b] x [c, d] = E. Extend h to E by hex, y) = 0 if (x, y) ¢; A, and note
that the set of all discontinuities of h is a subset of
= {(x, y) E JR2 : a::: x::: band y = f(x) or y = g(x)}.

D
By Problem 18.17, A(D) = 0, and so h is Riemann integrable on E (and hence,

Lebesgue integrable). Now, by modifying the arguments of Problem 23.19, we
easily see that
b fg(X)
=
l
a
(
J(x)
h(x,y)dy)dx.
Problem 23.21. Let f: [a, b] -+ JR be a differentiable function-with one-sided

derivatives at the end points. If the derivative f' is uniformly bounded on [a, b].
then show that !' is Lebesgue integrable and that
1 [a,b)
I'd). = I(b) - I(a).
Solution. Let I: JR - - 7 JR be a differentiable function such that for some M > 0

we have 1!,(x)1 ::: M for all x E [a, b]. By letting I(x) = I(a) + !'(a)(x - a)
for x < a and I(x) = I(b) + i'(b)(x - b) for x > b, we can assume that I is
defined (and is differentiable) on JR.
Next, consider the sequence of differentiable functions Un) defined by
In(x) = n[J(x +~) - I(x)]= I(x + ~1- I(x) , x E JR,

n
and note that In(x) --7 !'(x) holds for each x E 1R. Also, by the Mean Value
Theorem, it is easy to see that I/n(x)1 ::: M holds for each x. Consequently, by
the Lebesgue Dominated Convergence Theorem, I' is Lebesgue integrable over
[a, b] and
b
1[a,b]
I'd). = lim
11-+00
r II1(x)dx.
Ja
Now, using the change of variable u = x + ~, we see that
rh+f.
= n [ Ja+ I(u)du - J
r I(x)dx ]
1 a
"
b+ 1
= n[l "/(x)dx ~ 1 a+ 1
"/(X)dx]
ra + 1
= Ja "/(x)dx --7 I(b) - I(a),
1
n
where the last limit is justified by "V'irtue of the Fundamental Theorem of Cal-
culus. A glance at (*) guarantees that ha,b)!' d). = I(b) - I(a), and we are
finished.
Problem 23.22. Let f, g: [a, b] -+ JR be two Lebesgue integrable junctions

satisfying
[t f(t)d)...(t):s [~g(t)d)...(t)
for all x E [a, b]. If rP: [a, b] --+ JR is a non-negative decreasing function, then
show that the functions rPf and rPg are both Lebesgue integrable over [a, b] and
that they satisfy
[t rP(t)f(t) d)...(t) :s [t rP(t)g(t) d)...(t)
for all x E [a, b].
Solution. Since rP is decreasing there exists some M > 0 satisfying IrP(t)1 :s M

for each t E [a, b]. Since f and g are Lebesgue integrable, it follows from the
inequalities IrPU)fCt)1 :s Mlf(t)1 and IrP(t)g(t)1 :s Mlg(t)1 for each t E [a, b]
that rP f and rPg are both Lebesgue integrable functions over [a, b].
To obtain the desired inequality, fix x E [a, b]. Assume first that rP is a non-
negative decreasing function of the form rP = 2:::7=1 Ci X[ai_I,ui)' where {a = aD <
al < '" < ak = b) is a partition of [a, b]. Since f is decreasing, we know that
CI 2:: C2 2:: ... 2:: Ck 2:: O. Clearly,
rP = (CI - C2)X[a.atl + (C2 - c3)X[a.a2) + ... + (Ck-I - cdX[a,ak_,) + CkX[a,b)

k
= LYi X[a.ai)'
i=1
with Yi 2:: 0 for each i. Pick 1 :s m :s k such that am-I :s X < am, and note that
[t rP(t)f(t) d)...(t) = ~ Yi la i
f(t) d)...(t) + Ym [t f(t) d)...(t)
:s ~ Yi la i
g(t)d)...(t) + Ym [t g(t)d)...(t)
= lX rP(t)g(t)d)...(t).
Now, we consider the general case. Fix x E [a, b). As in the solution of
Problem 21.8, we see that there exists a sequence {rPn} of non-negative decreasing
step functions (as above) satisfying ¢n(t) t ¢(t) for almost all t E [a, b]. Since
l¢n(t)!(t)1 :::; MI!(t)l, l¢n(t)g(t)1 :::; Mlg(t)l, ¢n(t)!(t) -+ ¢(t)!(t), and since
¢n(t)g(t) -+ ¢(t)g(t) for almost all t E [a, b], it follows from the inequality
l X
¢n(t)!(t) d'A(t):::; l x
¢n(t)g(t) d'A(t)
and the Lebesgue Dominated Convergence Theorem that
l a
x
¢(t)!(t)d'A(t) = lim
n-+OO
la
x
¢n(t)!(t)d'A(t)
:::; lim
n-+OO
la
x
¢n(t)g(t)d'A(t) = la
x
¢(t)g(t)d'A(t).
24. APPUCATIONS OF THE LEBESGUE INTEGRAL
holds!or n = 0,1,2, ....

Solution. We shall establish the formula by induction on n. For Il = the
formula is true by virtue of Theorem 24.6. If the formula is true for some n 2: 0,
°
then an integration by parts yields
for x 2(n+l)e- x2 dx = -~ for x2n+1 d(e- X2 )

= _~r2n+le-r2 + 2n;-1 for x 2n e- x2 dx
for each r > 0. This implies
['Xl x2(n+l)e-X2 dx = lim {r x2(n+l)e-x2 dx = 2n;-1 (!Xl x 2n e-x2 dx

h r-ooh h
(2n+I)(2n+2) (211)! ..fii
= 22(n+l) . 22"n! . Z
_ [2(n+ 1)i! ..fii
- 22("+I)(n+I)! . z'
Section 24: APPLICATIONS OF TIlE LEBESGUE INTEGRAL 207
Problem 24.2. Show that J: e- tx2

dx = ~.jif for each t > 0.
Solution. Let u = x.ji. Then, J; e- tx2 d~ = Jr J;0 e- 1I2

du holds for each
r > 0. Therefore,
Problem 24.3. Show that f(x) = I~it is Lebesgue integrable over [1,00) and
J
that f d)'" = 1.
Solution. Since InxXT > - °

holds for each x > 1, it suffices (in view of
Theorem 24.3) to show that Jl I~il' dx exists.
oo
If r > 1, then an integration by parts yields
r I~il' dx = _ rlnxd(l) = _In:t\r + JIr -\dx = 1_1- Inr.

JI .r JI x x I x r r
Therefore,
. 1/!( + -X)II
hm
n..... oo 0 n
e- 1 2x
dx = 1.
Solution. Note that
= = 1.
Jo[00 e- lim (1 - e- r )
x dx = lim r e-x dx
r ..... oo Jo r->OO
Therefore, the function e- x is Lebesgue integrable over [0,00). Now, let

gn(x) = (1 + ~)"e-2tx[O.nl(X), and note that each gn is Lebesgue integrable
over [0,00).
From elementary calculus, we know that (1 + ;; rt eX for each x ::: 0, and

so g/l(x) t e-x holds for each x ::: O. Thus,
10t(1 + :rye-x dx = nlim jg/l d'A = (JO e- = 1.

lim x dx
/1 ..... 00 /I ..... oo 10
Problem 24.5. Let j: [0,00) -+ (0,00) be a continuous, decreasing, and

Lebesgue integrable function. Show that
lim - -
1
1
x..... oo j(x) x
00
j(s) ds = 0 if and only if •

hm
x..... oo
j(x+t)
j(x)
= 0 jor each t > O.
oo
Solution. Assume that limx..... oo flx) Jx j(s)ds = 0 and let t > 0 be fixed.
Since j is decreasing, we see that j (x + t) :::: j (s) for all x :::: s :::: x + t, and so
x+' 1x+,
tj(x+t)= x j(x+t)ds:::: x j(s)ds.
1
o< f(HI) < 1 . 1;'+1 f(s)ds < 1 . 1,00 f(s)ds

f(x) - I f(x) - I f(x)'
from which it follows that limx..... oo fj~;;) = O. For the converse, assume
that limx ..... oo ft~l) = 0 holds for each fixed t, and, for simplicity, let us write
oo
F(x) = flx) Jx j(s)ds for each x E [0, (0). Fix B > 0 and then choose
some 0 < 8 < 1 such that ~8 < B. (Since lim8 ..... 0+ 1~8 = 0 such a 8 always
exists.) From limx-+oo fi<;:J
= 0, we infer that there exists some M > 0
such that fj~;:) < 8 holds for all x > M. That is, j(x + 8) :::: 8j(x) holds for
each x > M. Now, note that for x > M, we have
x
+.j(s)ds + flx) 100
F(x) = flx)
1x x+. j(s)ds
+. +x
100 + 8)du
= Ax) x
1 j(s)ds Ax) x j(u
+.x
100
1 j(~)ds +
:::: flx) x flx) x 8j(u)du
+. +
x
= flx) x
1 j(s)ds 8F(x).
Section 24: APPLICATIONS OF THE LEBESGUE INTEGRAL 209
Consequently, if x > M, then
(1 - 8)F(x) :s flx) 1 X+O

f(s)ds =? At) lx+of(x)ds = 8,
x
and so 0< F(x) :s I~O < t: holds for all x > M. Thus, limx->oo F(x) = 0.
Problem 24.6. Show that the improper Riemann integrals
(which are known as the Fresnel integrals) both e.xist. Also, show that COS(X2)
and sin(x2) are not Lebesgue integrable over [0,00).
Solution. We shall work with fr~o sin(x2) dx. Similar arguments will establish
oo
the corresponding result for fo cos(x2) dx.
°
Let < s < t. The substitution u = x 2 followed by an integration by parts
gives
This inequality, combined with Theorem 24.1, guarantees the existence of the
oo
improper Riemann integral fo sin(x 2 ) dx. The inequality
j ..([ii Isin(x 2)ldx = 1 lkrr Isinlll du > _1- l krr Isinxldx = _1_
../krr-rr 2 krr-rr...;u - 2...(jik krr-rr .fifk
implies
which shows that f:'

Isin (x2) I dx does not exist in JR.-and hence, that sin(x 2)
is not Lebesgue integrable over [0,00).
Problem 24.7. Show that f;si~~x dx =~.

Solution. Consider the function
f( X ) --
{
I
x2 x
sin"
ifx=O
°
if < X <_ 1
:& ifx>l,
and note that f is Lebesgue integrable over [0,00). In view of the inequality
°:s Si~~
[0,00).
x :s f (x), we see tJ:1at the function Si~~ x is Lebesgue integrable over
Now, note that for each r, 8 > 0, we have
[
r sin~
x-
x dx = _ [ r sin2 x d(l) =
x
_sin" x
x
I+
r
[ r 2sinxcosx
X
dx
£ £ £ £
= sin"e _ sin2 r
£ r
+ 12r
2£
sinx
x
dx.
Thus, by Theorem 24.8, we see that
kr'"
o
sin: x
r
dx lim [ r ~ dx =
= r_oo r
£_0+ £
10
00
sinx
x
dx = zr..
2
Problem 24.8. Let (X, s, f.L) be an arbitrary measure space, T a metric space,
and f: X x T -r JR afunction. Assume that f(', t) is a measurable function for
each t E T and f (x, .) is a continuous function for each x EX. Assume also that
there exists an integrable function g such thatfor each t E T we have If(x, t)1 :s
g(x) for almost all x EX. Show that the function F: T -r JR, defined by
F(t) = Lf(X, t)df.L(x),

is a continuous function.
Solution. Let tn --+ t in T. Define the function gn: X --+ JR by the formula
gn(x) = f(x, tn). By our assumptions each gn is integrable, Ignl :s g a.e., and
gn(x) --+ f(x, t) holds for each x E X. Thus, by the Lebesgue Dominated
Convergence Theorem, we have
This shows that F is a continuous function.


00 e-X - e- XI
1
ox.
----dx=lnt
holds for each t > O.
Solution. Consider the function f(x, t) = e-X_C" x

for x > 0 and t > O.
Observe that the value f(O, t) = t -I extends f continuously to the point (0, t),
t > O. Next, note that the function g(x, t), defined by
g(x, t) = { eI~\x, t)~

x + e XI
if 0 ::: x ::: 1 and t > 0
if x > 1 and t > 0 ,
is Lebesgue integrable for each t > O. Moreover, If (x, t) I ::: g(x, t) holds. This
implies that
F(t) = 1 00
f(x, t) dx = 1 e-"~e-"
00
dx
exists both as an improper Riemann integral and as a Lebesgue integral; see also
Theorem 24.3.
Next, note that ~ (x, t) = e-xI holds for all x > 0 and all t > O. The
inequality 0 ::: e-xI ::: e-xa for all t > a > 0 and all x 2: 0, coupled with
Theorem 24.5, shows that
F' (t) = 1 00
¥C(x, t) dx = 1 00
XI
e- dx =+
holds for all t > O. Thus, F(t) = In t +C. Since F(I) = 0, it follows that C = 0
and so
F(t) = 1 e-'~e-'"
00
dx = Int.
Problem 24.10. For each t > 0, let F(t) = Jooo f~:~2 dx.
a. Show that the integral exists as an improper Riemann integral and as a
Lebesgue integral.
b. Show that F has a second-order derivative and that F"(t) + F(t) = +
212 Chapter 4: TIlE LEBESGUE IN1EGRAL
Solution. (a) The integrability of F follows from Theorem 24.3 and the inequa-
lity
1.c:...1 <_I
I+x" - l+x 2 '
(b) If f(x, t) = t~;2' then

af ( ) _ -xe-Xl
at x, t - I+x2
Since l¥r(x, t)1 ::: e-xt and I~(x, t)1 ::: e-xt both hold, by applying
Theorem 24.5 twice we get
Consequently,
+ F(t) =
10roo xI+x2 +
10roo I+X2
2
e-·f/ dx = t>Oe- dx = 1
x1
F"(t) e- dx xt
10 t'
Problem 24.11. Show that the improper Riemann integral Jo! In(t cos x) dx ex-
ists for each t > 0 and that it is also a Lebesgue integral. Also, show that
holds for all t > O.
Solution. Let f(x, t) = In(t cosx) for 0 ::: x < I'

t > 0, and let g(x) =
r:1
I
(I - x for 0 ::: x < I'

An easy argument shows that the improper
Riemann integral (and hence, the Lebesgue integral) of g exists over [0, I)'
Also, L'Hopital's Rule shows that limxg [f;~;;)] = O. Thus, for each t > 0 there
exists some 0 < Xo < I
such that If(x, t)1 ::: g(x) holds for all Xo < x < I'
Since f(x, t) is continuous for 0 ::: x < I'
an easy application of Theorem 22.6
guarantees that
F(t) = l\n(t cosx)dx

Section 24: APPLICATIONS OF TIm LEBESGUE INTEGRAL 213
exists both as a Lebesgue and as an improper Riemann integral. Next, note that
~ (x, t) = f, and that for 0 < a < t we have ~{ (x, :s ~. Thus, byI t)1
Theorem 24.5, we have
F'(t) = faT ~ (x, t)dx = fr

for each t > 0, and therefore F(t) = I In t + C.
Since JoTln(cosx) dx = JoTln(sinx) dx (why?), it follows that
2C =2F(1) = 2faTln(COSX)dX
= fa TIn (cos x) dx + fa TIn (sin x) dx
= faT In(sin22.t) dx
= faTln(Sin2X)dX - Iln2
= ~ Lrln(SinX)dX - Iln2
=C- I ln2
Thus, C = -I In 2, and so
fat In(t cosx)dx =I lnt - I In2 = IlnU)

Problem 24.12. Show that for each t 2: 0 the improper Riemann integral
J:'X~;~~:2) dx exists as a Lebesgue integral and that
OO sinxt IT -/
.,,0 - e ).
fao x(x-
?
+ 1) dx = -
oo
Solution. For each t E lR let F (t) = Jo X~;~~:2) dx. From
sinxt I Ixtl It I
Ix(l + x 2) :s Ix(l + x 2 )1 = 1 + x 2 '
we see that F is indeed a well defined real-valued function on JR and that the
integral defining F exists both as a Lebesgue integral and as an improper Riemann
integral. Moreover, the relations
a [ sinxt] cosxt cosxtl 1

at x(1 +x 2 ) = 1 +x 2 and
I1 +x 2 :5 1 +x 2
in connection with Theorem 24.5 guarantee that F is a differentiable function and
F'(t) =
1o
00
-ata [ x(1Sinxt]
+x 2 )
dx = 1 0
00
cosxt
- -dx
1 +x 2
holds for each t E lR.

Since .!1..[f2W.]
at l+x2 = _xsinxt
l+x2
and the natural dominating function in the inequal-
I
ity I_·\~;;t :5 l~l2 is not Lebesgue integrable over [0,00), we cannot use
Theorem 24.5 to conclude that
00 •
F " (t) = -
1 x SlDXt d x.
o l+x
--2
As a matter of fact, the identity
x sinxt x 2 sinxt sil}xt sinxt

(**)
1 +x 2 = x(1 +x 2 ) = -x- - x(1 +x 2 )
shows that, on one hand, the function x I--?- xl~~~t is not Lebesgue integrable over
°
[0,00) for each t > and, on the other hand, that
1
00
x sinxt
-
o 1+x
- 2
d
x = 1 --
0
00
sinxt d x = -
X
1 0
00
sinxt dx
x(1 + x 2 )
= -TC2 - F
(t)
for each t > O.

We shall establish the validity of (*) for each t > 0 using another method. For
each n, let
-In
G n(t ) -
• 0
cosxt2 d X •
1 +x
Clearly, Gn(t) -+ 10 00
~~:~ dx = F'(t) for each t E JR. Now, from Theorem 24.5
Section 24: APPLICATIONS OF 11IE LEBFSGUE INTEGRAL 215
and (**), we see that
G ' (t )
n
= - in0 ---
X sinxt d
I +x2
X =-
in -
sinxt
0 . X
- dx + in sinxt d X
0 xO + x 2 )
nt sinx
= -
l o
--dx+
X
in sinxt
0 x(1 + x 2 )
dx,
and consequently for each t > 0, we have
.,
hmGn(t)=-
'1-+00
1 0
00
sinx
--dx+
X
1 0
00
sinxt 2 dx=-1-+ F(t)=g(t).
x(1 +X )
We claim that for each a > 0 the sequence of derivatives {G~) converges uniformly
to the function g(t) = -1-
+ F(t) on the open interval (a, (0). To see this, fix
a > 0 and let E > O. Choose some Xo > 1 such that
11 s
00
sin x dx
X
I < E and 11°O_s_in_,_xt-::-dx
s x(l + x 2 )
1< 1°O_d,_X_ <
- s 1+ x2
E
for all s > Xo. Now, if we fix some natural number k satisfying k ::: Xo and
ka > Xo, then for each n ::: k and all t > a, we have
,
jGn(t) - g(t)j = 11
nt
00
-sinx dx -
X
100
I
sinxt 2 dx < 2E.
n x(1 + X )
This shows that {G~) converges uniformly to the function get) = + F(t). -1-
Finally, using Problem 9.29, we get F"(t) = [limn.... oo Gn(t)J' = + F(t), or -1-
F"(t) - F(t) = -1-
for each t > O. Solving the differential equation, we obtain
Since F and F' are continuous at zero (why?), it follows from F(O) = 0 and
F'(O) = Jo
oo
= 1!:2 1-
and the preceding formula of F(t) that = CI -1- and
C2 = O. Hence, F(t) = 1-(1-
e- t ) for each t ::: O.
Problem 24.13. The Gamma function for t > 0 is defined by an integral as

follows:
a. Show that the integral
exists as an improper Riemann integral (and hence. as a Lebesgue integral).

b. Show that f(~) = .jir.
c. Show that r(t + 1) = t f(t) holds for all t > O. alld use this conclusion
to establish f(n + 1) = nl for n = 1,2, ....
d. Show that r is differentiable at evelY t > 0 and that
r'(t) = foooxl-le-Xlnxdx
holds.
e. Show that r has derivatives of all order and that
r(II)(t) = fooo xl-1e-X(lnx)" dx

holds for n = 1,2, ... and all t > O.
Solution. (a) Since xl-1e-x :::; X I - 1 holds for 0 < x :::; 1, it follows from
Problem 23.l6 that folxl-le-X dx exists both as an improper Riemann integral
and as a Lebesgue integral. .
Now, for each fixed t > 0 we have limx-+ oo X I - 1e-! = O. Thus, there exists
some M > 0 (depending upon t) such that 0 :::; xr-le-~ :::; M holds for all
x ::: 1. Hence, xl-1e-x :::; Me-~ holds for each x ::: 1. This shows that
flooxl-le-X dx exists both as an improper Riemann integral and as a Lebesgue
integral for each t > O.
oo
The preceding show that fo x l - 1e-x dx exists both as an improper Riemann
integral and as a Lebesgue integral.
(b) Substitute u = x! to get
(c) Integrating by parts, we get

Consequently, we see that
r(n + 1) = nlro) = nl rOe-x dx = nl.

.fo
(d) and (e). Note that ;1: (xl-Ie-X) = xl-Ie-x(lnx)" holds for all t > 0 and all
x> o.
Now, let a < t O. We claim that there exists a
Lebesgue integrable function g: (0,00) -+ (0,00) such that Ih(x, t)1 .:5 g(x)
holds for all x > 0 and all a < t < b. If this is the case, then Theorem 24.5
allows us to "differentiate under the integral sign," and since 0 < a < b are
arbitrary this shows that r must have derivatives of all orders and that the desired
formulas hold. So, we must construct a positive Lebesgue integrable function g
over (0,00) such that Ih(x, t)1 .:5 g(x) holds for each a < t O.
I
Note that for x 2: 1, we have 0.:5 X
l
- .:5 x IJ • Using L'Hopital's rule, we see
that
lim x IJ e- 1(lnx)" = lim

x~cx:: .'C-+oo e4
x:. lim (In.~)'' =0·0=0,
x-+co e~
and so there exists some M > 0 such that xIJe-~(1nx)n .:5 M for all x > 1.
Therefore,
holds for all x 2: 1 and all a < t < b.

For the rest of our discussion, we shall need two facts from calculus.
lim xa (In x)" = 0 and t X a-I (1nx )" dx -_

10+ (-1)"n!
an+1 •
X-+O+
Both can be proven by induction. For this limit use induction and L'Hopital's rule
by observing that
lim x a lnx = lim

.<-+0+
(lnx):
a
x-+o+ (.c )
= x-+o+
lim -:£: =
a
0
and
218 Chapter 4: 1HE LEBESGUE INTEGRAL
For the integral, use induction and take into account that
11
0+
xa-Ilnxdx=~ 11
0+
InXd(Xa)=~Xalnxl -~ 1 11
0+ 0+
x a- 1 dx=--1r
and
txa-I(lnx)"+1 dx =~ t(lnx)"+1 d(x a)

Jo+ Jo+
= ~xa(lnxf+t+ _n!1 1>a-I(lnx)" dx
= _n!1 txa-I(lnxf dx.
Jo+
°
Since either (lnx)n ::: holds for all x E (0,1] or (lnx)" ::: holds for all °
x E (0, 1], it follows that the function </lex) = xa-I(lnxl, x E (0, 1], is Lebesgue
integrable over (0, 1]. Now, let
. _ {xa-lllnxln
g(x) - Me-~
°
if < x::: 1
if x::: 1 '
and note that g is Lebesgue integrable over (0, (0). To finish the proof, notice
that
Ih(x, t)1 ::: g(x)

holds for all x > ° and all a < t < b.
Problem 24.14. Let f: [0, 1] --+ 1R be a Lebesgue integrable function and define
the function F: [0, 1] --+ 1R by F(t) = Jd
f(x) sin(xt) d)"(x).
a. Show that the integral defining F exists and that F is a uniformly continuous
function.
b. Show that F has derivatives of all orders- and that
F(2n)(t) = (-It 11 x2n f(x)sin(xt)d)"(x)
and
F(2n-I)(t) = (~I)n 11 x 2n - 1f(x) cos(xt) d)"(x)
for n = 1,2, ... and each t E [0, 1].

Section 24: APPUCATIONS OF THE LEBESGUE INTEGRAL 219
c. Show that F =
a.e.
° (i.e., F(t) = °for all t E [0, 1]) if and only if f = °
Solution. (a) Note that for each fixed t € [0, 1] the function x J-7 sin(xt)
is continuous and hence, measurable. The inequality If(x) sin(xt)I ::: If(x)1
guarantees that x J-7 f(x) sin(xt) is integrable for each t E [0, 1]. So, F is a
well-defined function.
For the uniform continuity of F note that
IF(t) - F(s)1 = III f(x)sin(xt)dA(x) -11 f(X)Sin(xs)dA(X)1
= 111 f(x)[sin(xt) - sin(xs)] dA(X)\
:: l l
lt(x)llsin(xt) - sincu)1 dA(X)
:: l l
lt(x)llxt -xsldA(X)
= [l
l
lf(x)1 dA(X)]It - sl
holds for all s, t E [0, 1].

(b) Consider the function of two variables hex, t) = f(x) sin(xt). Then an easy
inductive argument shows that for each n = 1, 2, ... we have
foreacht E [0, 1] and almost allx. This implies lan~;.~.t)l::: Ix n f(x)1 = gn(x) for
all t E [0, 1] and almost all x. Since gn is Lebesgue integrable for each n, it easily
follows from Theorem 24.5 that we can "differentiate under the integral sign" and
get the desired formulas.
°
(c) Assume F(t) = for each t E [0, 1]. Then F(2n)(t) = for all n, and so
° °
from (b) we get fol x2n f(x) sin(xt) dA(X) = for each n and all t E [0, 1]. Letting
t = 1, we get
for each n. Now, invoke Problem 22.21 to conclude that f(x) sin x = for almost
° °
all x. Since sinx > for each < x ::: 1, we easily infer that f(x) = for
° °
almost all x.
25. APPROXIMATING INTEGRABLE FUNCITONS
Problem 25.1. Let f: JR -+ JR be a Lebesgue integrable function. Show that
lim If(X) cos(xt) dJc(x) = lim If(X) sin(xt)dJc(x) = O.

1-+00 1-+00
Solution. By Theorem 25.2, it suffices to establish the result for the special case
f = X[a.b)· So, let f = X[;.b), where -00 < a 0 we have
II f(x) cos(xt) dJc(x) I = lib COS(xt)dXI

= I I x=a I= I
sin(XI) IX=b sin(bl)-sin(al)
I
I-< 6-I'
J
and so lim(-+oo f(x) cos(xt)dJc(x) = 0 holds. In a similar fashion, we can show
that lim(-+oo Jf(x) sin(xt) dJc(x) = O.
Problem 25.2. Afunction f: 0 -+ JR (where 0 is a nonempty open subset of

JRn) is said to be a COO-function if f has continuous partial derivatives of all
orders.
a. Consider the function p: JR -+ JR defined by p(x) = exp[xL I ] if /x/ < 1
and p(x) = 0 if /x/ 2: 1. Then show that p is a COO-function such that
=
Suppp [-1, 1].
b. For f > 0 and a E JR show that the function f(x) =
p(x;a) is also a
COO-function with Supp f = [a - f, a + f].
Solution. (a) We shall establish that p(n)(I) exists for each n.

Start by observing that, by L'Hopital's Rule, liml ...... ::x; t k e-4 1 = 0 holds for
k = 0, 1,2, .... Notice that if for 0 < x < 1 we let t = I~X' then we have the
inequality
I
I
I I
xme:;:T.:I
(xz_l)k(x_1) _<
I
e-:m=x;
(I_X)'+1
I= tHI e-1 , 1
from which it follows that
• xme*' • k+1 _11

hm(Z_l)k(._I)=hmt e 2 =0 fork,m=0,1,2,.... (*)
xtl x x Itoo
N ow, by a simple induction argument, we see that for -1 < x < 1 the deriva-
...,L
tive p(II)(X) is a finite sum of terms of the form ~:i~-I~! . Using (*) and another
Section 25: APPROXIMATING INTEGRABLE FUNCTIONS 221
simple inductive argument, we can also see that p(n)(l) =0 holds for n =
1,2, ....
(b) Note that: f(x) i= 0 if and only if -1 < x~a < 1 if and only if a - e < x <
a + e. Therefore, Supp f = [a - e, a + e],
Problem 25.3. Let [a, b] be an interval, E > 0 such that a +E < b- E. and
p as in the previous exercise. Define h: JR -+ JR by h(x) = 1: pC:X) dt for all
x E JR. Then show that
a. Supph S; [a - E, b + E].
b. h(x) = c (a constant function) for all x E [a + E, b - E].
c. h isa COO-function and h(Il)(x) = J: :';"pe:X)dt holdsforall x E 1R. and
d. the COO-function f = hlc satisfies 0::: f(x) ::: 1 for all x E JR. f(x) = 1
for all x E [a + E, b - E]. and J IX[a.b) - fl d'A < 4E.
Solution. For simplicity, let gx: JR ---+ JR be the function defined by gAt) =
pC~X), and so h(x) = J: gAt)dt.
(a) By part (b) of the preceding problem, we know that Supp gx = [x - e, x + e].
Thus, if a < t < b and x if. [a - e, b + e], then gx(t) = pC~x) = 0 (since
I I
t~.t 2: 1). This implies that h(x) = 0 holds for all x if. [a - e, b + e], so that
Supp h S; [a - e, b + e].
(b) If a + e < x O.
(c) Since every partial derivative ::n

p( t~X) is continuous, it must be bounded on
[a, b] (and hence, on JR). Now, the desired conclusion follows from Theorem 24.5.
(d) Since Supp gr =
[x - e, x + e] and gx is a positive function, it follows that
holds for all x. Thus, f = hi c satisfies 0 ::: f (x) ::: 1 for all x.
Finally, observe that
IX(a.b) - fl ::: X(a-e.a+e) + X(b-e.b+e)
holds, and so JIX[a.b) - fl d'A = JIX(a.b) - fl d'A < 4e.

The graph of f is shown in Figure 4.3.
y=f(x)
a-e a a+e b-e b b+e x
FIGURE 4.3.
Problem 25.4. Let f: IR -+ IR be an integrable function with respect to the

Lebesgue measure, and let E > O. Show that there exists a COO-function g such
that Jlf - gl d'A < E.
Solution. Let f: IR ~ IR be a Lebesgue integrable function and let 8 > O. By

Theorem 25.2, there is a step function 41 = 2:7=1 C; X[a;,b;) (with Ci #- 0 for each
i) such that Jlf - 411 d'A < 8. Now, by the preceding problem, for each i there
exists a COO-function gi with compact support such that JIXla;,b;) - gil d'A < n'~;I'
Now, note that the COO-function g = 2:7=lcigi has compact support and satisfies
!If-gld'A:::: !If-41ld'A+ !I41-gld'A
< 8+ !1'tciXla;,b;)- 'tc;gild'A
:::: 8 + t
i=1
ICil! IX[a;,b;) - gil d'A
n
< 8 + L IC;/n'~;1
;=1
= 8 +8 =28.
Problem 25.5. The purpose of this problem is to establish the following

general result. If f: IRn -+ IR is an integrable function (with respect to the
Section 25: APPROXIMATING INTEGRABLE FUNCTIONS 223
Lebesgue measure) and E > 0, then there exists a C oo -ftll1ction g such that
Ilf gldJ.. < E.
a. Let a,. < b,. for i = 1, ... , n, and let I = TI7=1 (a,., b,.). Choose E > 0 sllch
that a,. + E < b,. - E for each i. Use Problem 25.3 to select for each i a
COO-function Ii: lR -+ 1R such that 0 ::: f,.(x) ::: 1 for all x, f,.(x) = 1 if
x E [a +E, bi - E], and Supp J; f; [ai - E, bi +E]. Now, define h: lRn -+ lR
by h(xl, ... , xn) = TI7=1 f,.(x,.). Then show that h is a COO-function onlRn
and that
b. Let f: lRn -+ lR be Lebesgue integrable, and let E > O. Then use part (a)
to show that there exists a C oo 1unction g with compact support such that
flf - gldJ.. < E.
Solution. (a) Clearly, h is a COO-function. Let A = TI7=1 (a,. 8, b,. + 8),

B = TI;'=I(a,. + 8, b,. + 8), and C = TI7=I(a,. - 8, bi - 8). Now, the desired
conclusion follows from the inequality
(b) Let f be an integrable function and let8 > O. Pick a step function of the
form 4> = L~=I C,. XI;
(where each I,. is a finite open interval of lRn) such that
Ilf - 4>1 dJ.. < 8. From part (a) it follows that there exists a COO-function g
with compact support such that 114> - gl dJ.. < 8. Consequently, Ilf - gl dJ..
< 28.
Problem 25.6. Let fJ., be a regular Borel measure on1R", fa fJ.,-integrableftll1c-

tion,andE > o. Show that there exists a Coo-function g such that Ilf-gldfJ., < E.
Solution. Let I = TI7=1 [a,., b;] be a finite closed interval. Given 0 > 0, pick
8 > 0 such that the closed interval J = TI;'=I [a,. -28, b,. +28] satisfies fJ.,(J \ 1) =
fJ.,(J)- fJ.,(/) < o. (This is always possible since m'=I[ai - t, bi +t] +k I.) As in
Problem 25.5, there exists a COO-function 11: lR" --+ lR such that 0 ::: hex) ::: 1
for all x E lRn, hex) = 1 for x E I, and Supph f; J. Therefore, if f = XI,
then
Thus, the desired result holds true for the characteristic functions of the finite
closed intervals.
Now, let I = n;'=l[ai, bi) be finite. Since n7=1[ai, bi - tJ
tk I, it follows
that the approximation result is also true for the characteristic functions of sets
of the form n7=1 [ai, bi ). Since these sets form a semiring and every open set
is a a-set (for this semiring), it is not difficult to see that the result is true for
the characteristic functions of open sets of finite measure. The regularity of f.L
guarantees the validity of the approximation result for characteristic functions of
f.L-measurable sets of finite f.L-measure. This in tum implies that the result holds
true for f.L-step functions. Finally, since for each f.L-integrable function f and each
8 > 0, there exists some f.L-step function <P with f l f - <PI df.L < 8, it follows
that the COO-functions with compact support satisfy the desired approximation
property.
Problem 25.7. Let f: [a, bJ ~ JR be a Lebesgue integrable function, and let

E > 0. Show that there exists a polynomial p such that flf - pi d)" < E, where
the integral is considered, of course, over [a, bJ.
Solution. Let f: [a, b] ---7 JR be integrable (over [a, b]), and let 8 > 0.
By Theorem 25.3 there exists a continuous function g: [a, b] ---7 JR such that
f If - gl d)" < 8. Now, by Corollary 11.6, there exists a polynomial p such that
Ig(x) - p(x)1 < 8 holds for all x E [a, b]. Thus,
! If - pi d)" :s ! If - gl d)" + ! Ig - pi d)" < 8 + 8(b - a) = 8(1 +b - a),
and our conclusion follows.
26. PRODUCT MEASURES AND ITERATED INTEGRALS
Problem 26.1. Let (X, S, f.L) and (Y, 1;, v) be two measure spaces, and assume
that A x BEAlL ® Av.
a. Show that f.L*(A) . v*(B) :s
(f.L x v)*(A x B).
b. Show that if f.L*(A)· v*(B) =f:. 0, then (f.L x v)*(A x B) = f.L*(A)· v*(B).
c. Give an example for which (J1. x v)*(A x B) =f:. J1.*(A)· v*(B).
Solution. (a) We have S ® 1; 5;; .A IL ® Av. Let A x BEAlL ® Av. Also, let
{An BII}be a sequence of S ® 1; such that A x B 5;; U:I An X Bn. Since (by
X
Theorem 26.1) f.L* x v* is a measure on the semiring AIL ® A v, it follows from
Section 26: PRODUcr MEASURES AND ITERATED INTEGRALS 225
Theorem 13.8 that
00 00
p,* x v*(A x B) .:::: l : p,* x v*(AII .x B II ) = l : p, X v(AII x BII)'

11=1 11=1
Consequently, we see that
p,*(A)· v*(B) = p,* x v*(A x B)

00 00
.:::: inf{l:p, x v(AII X Bn): {An X Bn} ~ S®:E and A x B ~ UAII X Bnl
n=1 n=1
= (p, x v)*(A x B).
(b) If 0 < p,*(A) < 00 and 0 < v*(B) < 00, then
(p, x v)*(A x B) = p,*(A)· v*(B)
holds true by virtue of Theorem 26.2. On the other hand, if either p,*(A) = 00 or
v*(B) = 00, then-by (a)-the equality holds with both sides equal to 00.
(c) Let X = Y =(OJ, S =
{(Zj}, and :E = P(Y). Also, let p, = 0 on S (the only
choice!) and v = 0 on :E. Now, note that p,*(X) . v*(Y) = 00 . 0 = 0, while
(p, x v)*(X x Y) = 00.
Problem 26.2. Let (X, S, p,) and (Y, :E, v) be two a-finite measure spaces.
Then show that (p, x v)*(A x B) =
p,*(A) . v*(B) holds for each A x B in
AI-'®Av.
Solution. Let {Xn} ~ AI-' and {YII } ~ Av satisfy Xn t X, YII t Y, P,*(Xn) <
00, and v*(Yn) < 00 for each n. Using Theorems 15.4 and 26.2, we see that
(p, x v)*(A x B) = n-+oo

lim (p, x v)*(A n XII) x (B n Yn»)
= lim [P,*(A n XII) . v*(B n Y

IZ~OO
II ) ]
= p,*(A) . v*(B)
for each A x BE AI-' ® Av.
Problem 26.3. Let (X, S, p,) and (Y, :E, v) be two measure spaces. Assume that
A and B are subsets of X and Y, respectively, such that 0 < p,*(A) < 00, and
0< v*(B) < 00. Then show that A x B is p, x v-measurable ifand only ifboth A
and B are measurable in their corresponding spaces. Is the preceding conclusion

true if either A or B has measure zero?
Solution. If A E AJ1. and B E A v, then by Theorem 26.3, A x B E AJ1.xv, For

theconverse,assumethat AxB E AJ1.xv, We claim first that (p.xv)*(AxB) < 00.
To see this, pick two sequences {An} £; 5 and {Bm} £; :E with A £; U:'l An,
L::'l p.*(An) < p.*(A) + 1, B £; U:::'=l Bm, and L::::'=l v*(Bm) < v*(B) + 1.
Now, from A x B £; U:'I.U:::'=1 An X Bm, we see that
00 00
(p. x v)*(A x B) :::: L L(P. x v)*(An x Bm)

n=l m=!
00 00 00 00
= LLP.(An)· v(Bm) = [LP.(A n)]. [LV(B m)]

n=l m=! n=l m=l
< [p.*(A) + 1] . [v*(B) + 1] < 00.
Therefore, by Theorem 26.4, (A x B)Y is p.-measurable for v-almost all y. Since

(A x B)Y = A holds for all y E B and v*(B) > 0, it follows that A is p.-
measurable. Similarly, B is v-measurable.
Finally, if p.*(A) = 0, A "# 0, and A x B E AJ1.xv, then B need not be
necessarily v-measurable. An example: Take X = Y = 1R with p. = v = A. If
E £; [0, 1] is nonmeasurable, then {OJ x E is a p. x v-null set (since {OJ x E £;
{OJ x [0, 1]), and so {OJ x E is a p. x v-measurable set.
Problem 26.4. Let (X, 5, p.) and (Y, :E, v) be two a-finite measure spaces, and
let f: X x Y ~ 1R be a p. x v-measurable function. Show that for p.-almost all x
the function fx is a v-measurable function. Similarly, show that for v-almost all
y the function fY is p.-measurable.
Solution. We can assume f (x, y) ::: ° for each (x, y) E X x Y. Since (in
this case) the product measure is a-finite, there exists a sequence {An} of p. x v-
measurable sets with An t X x Y and (p. x v)*(An) < 00 for each n.
Now, by Fubini's Theorem, the function (f /\ XAn)X is v-integrable for p.-almost
all x. Since (f /\ XAn h t fx, it follows that fx is v-measurable for p.-almost all
x.
Problem 26.5. Show that if f(x, y) = (x 2 - y2)/(x 2 + y2)2, with f(O, 0) = 0,

then
Section 26: PRODUCT MEASURES AND ITERATED INTEGRALS 227
Solution. Note that
and
11[11 f(x, y)dy ] dx = 1t2r 21~:~] dx

y = 11 ILl dx = %.
Problem 26.6. Let X = Y = IN, S = l: = the collection of all subsets oflN,

and p., = v = the counting measure. Give an interpretation of Fubini's Theorem
in this case.
Solution. Let f: IN x IN ---+ 1R be a non-negative p., x v-integrable function.

Then, by Problem 22.3 and Fubini's Theorem, we see that
f fd(p., x v)= f f f dp.,dv= f['fr fen, m)]dv(n)= ~ 'frf(n, m).

On the other hand, if f: IN x IN ---+ 1R is a non-negative function such that
00 00
LLf(n,m) < 00,

11=1 m=1
then it follows from Tonelli's Theorem that f is p., x v-integrable. Conclusion: A

function f: lNxlN ---+ 1R is p., x v-integrable if and only if L:;:I L~=I If(n, m)1
< 00, and in this case
0000 0000
f fd(p., x v)= LLf(n,m)= LLf(n,m).

n=lm=1 m=III=1
Problem 26.7. Establish the following result, known as Cavalieri's Principle.

Let (X, S, p.,) and (Y, l:, v) be two measure spaces, and let E and F be two p., x v-
measurable subsets of X x Y of fmite measure. If v*(Ex) = p., *(F.t) holds for
p.,-almost all x, then
(p., x v)*(E) = (p., x v)*(F).

228 Chapter 4: TIlE LEBFSGUE INTEGRAL
Solution. By Theorem 26.4, we have
Problem 26.8. For this problem A denotes the Lebesgue measure on lR.. Let
(X, S, JL) be a a-finite measure space, and let f: X --+ lR.be a measurablefunction
°
such that f (x) ::: holds for all x EX. Then show that
°
a. The set A = {(x, Y) E X x lR.: ~ Y ~ f(x)}, called the ordinate set of
f, is a JL x A-measurable subset of X x 1R.
°
b. The set B = {(x, y) E X x lR.: :s Y < f(x)} is a JL x A-measurable
subset of X x lR. and (JL x A)*(A) = (JL x A)*(B) holds.
c. The graph of f, i.e., the set G = {(x, f(x)): x E X}. is a JL x A-measurable
subset of X x lR..
d. Iff is JL-integrable, then (JL x A)*(A) = f f dJL holds.
e. Iff is JL-integrable, then (JL x A)*(G) = holds. °
Solution. If g: X -+ lR. is an arbitrary positive measurable function, then we
shall write Ag = {(x, y) E X x lR.: o:s Y :s g(x)} and
Bg = {(x,y) E X x lR.: O:s y < g(x)}.
Observe that if fn(x) t f(x) and hn(x) ~ f(x) hold for each x E X, then
BIn t B I and A"n ~ A I'
. Assume first that g is a positive simple function. Let g = :L;'=I ai Xc; be the
°
standard representation of g, where ai > for each 1 :s i :s n. Then, it is easy
to see that
Ag = (X x (OJ) U (C 1 x [0, ad) U (C 2 ?< [0,a2J) U ... U (C II x [O,anJ) (*)
and
Bg = (C 1 X [0, ad) U (C 2 x [0, a2J) u··· U (C II X [0, an]). (**)
By Theorem 26.3, both Ag and Bg are JL x A-measurable subsets of X x 1R.

(a) First assume that f is a bounded measurable function. That is, assume that
there exists some M > ° °
such that :s f(x) :s M holds for all x EX. By
Theorem 17.7 there exists a sequence {1/rn} of simple functions with 1/rn(x) t
M - f (x) for each x EX. Thus, the sequence {rPlI} of simple functions, defined
by rPn(x) = M - 1/rn(x), satisfies 4n(x) ~ f(x) for each x E X. This implies
A¢n ~ A I' Since (by the preceding discussion) each A¢n is JL x A-measurable,
we see that in this case A I is likewise a JL x A-measurable set.
Section 26: PRODUcr MEASURES AND ITERATED INTEGRAlS 229
Now, let f be an arbitrary positive measurable function. For each n, let

1" = f 1\ nl,and note that (by the preceding case) each AI" is fL x A-measurable.
To infer that A I is a fL x A-measurable set, observe that A I" tAl holds.
(b) By Theorem 17.7 there exists a sequence {Sll} of simple functions such that
o ::: SIl(X) t f(x) holds for all x EX. Clearly, BSn t B I holds. Since each BSn
is fL x A-measurable, it follows that B I is likewise fL x A-measurable.
Next, we shall establish the equality (fL x A)*(A I) = (fL x A)*(B I) by cases.
CASE 1. Assume fL*(X) < 00.
Clearly, (fL x A)*(X x {O}) = fL*(X) . A({O}) =

O. Also, assume that 0 :::
f (x) ::: M < 00 holds for all x. Then, there exist two sequences {<PII} and {o/II}
of step functions with 0 ::: <PII (x) t f (x) and 0/11 (x) t f (x) for all x EX.
Clearly, B",,, t Bland A",,, tAl' Now, use (*) and (**) in connection with
Theorem 26.3 and the Lebesgue Dominated Convergence Theorem to see that
and
Thus, in this case (fL x A)*(A I) = (fL x A)*(B I) = Jf dfL holds.
CASE 2. Assume fL *(X) < 00 and that f is a positive fL-measurable function.
For each n let fll =f 1\ nl. Note that BI" t BI and AI" t AI' By the
preceding case, we have (fL x A)* (A I,,) =
(fL x A)* (B I,,) for each n. Thus, from
Theorem 15.4, it follows that
(fL x A)*(A I) = (fL x A)*(BI) = lim !flldfL.

11->00
CASE 3. The general case. Here we shall use the hypothesis that fL is CY-finite.
Choose a sequence {Ell} of measurable subsets of X with Ell t X and

fL*(EII) < 00 for each n. Let gil = fXE", and observe that Bgn t BI and
AI" tAl' Using the preceding case and Theorem 15.4, we see that
Also, it should be noted here that if f is integrable, then
(c) From the identity G = A f \ Bf' it follows that the graph G of f is p., x A-
measurable.
(d) The equality follows from the discussion in part (b).
(e) From G = A f \ B f and part (d), we see that
Problem 26.9. Let g: X -+ 1R be a p.,-integrablefunction, and let h: Y -+ 1R be

a v-integrable function. Define f: X x Y -+ 1R by f(x, y) = g(x)h(y)for each
x and y. Show that f is p., x v-integrable and that
Solution. We can assume g ::: 0 and h ::: O. Choose a sequence {¢n} of p.,-step
functions and a sequence {o/n} of v-step functions with ¢n t g and 0/nth.
Then, {¢n o/n} is a sequence of p., x v-step functions such that ¢n o/n t gh. The
conclusion now follows from the relation
Problem 26.10. Use Tonelli's Theorem to verify that
1€
r
sin x
7 ('"
dx = 10
(1 €
r
e-
XY
sinx dx
)
dy
holds for each 0 < E < r. By letting E -+ 0+ and r -+ 00 (and justifying your
steps) give another proof of the formula
r'o
10
.sinx dx
x
= ~.
2
Solution. Fix 0 < 8 < r and consider the function
if (x, y) E [8, r] x [0,1]

if (x, y) E [8, r] x [1,00).
Clearly, the continuous (and hence, measurable) function I(x, y) = e-xy sinx
satisfies \/(x, y)\ ::: ge,I'(.'(' y) for all (x, y) E [8, r] x [0, (0). From
fe J1roo ge,I'(X' y)dy ]dx::: fl'e -;;1 dx = r

r [ 8
-8- < 00
and Tonelli's Theorem, we see that the function ge,1' is Lebesgue integrable over
[8, r] x [0, (0). So, thefunction I(x, y) is integrable over [8, r] x [0, (0). Now
Fubini's Theorem guarantees that
Using the elementary integral
and performing the innermost integrations in (*), we get
= J[00 [_ ysinx+;osx
f e
r sinx
x
dx
o 1+y-
e- xy \x=r] dy
x=e
= [00 ysine+cose e-ey dy _ [00 ysinr+cosr e-I'Y dy

Jo 1+y2 Jo 1+),2 ,
and consequently,
fe
r sinx
x
dX=Sin8 [OOye_ EY
Jo 1+),2
dY+COS8 {OOe_ EY
Jo 1+y2
dy- {ooysinr+cosr
Jo 1+),2
e- ry dy. (**)
We shall compute the limits of the three terms in the right-hand side of (**) as
r -+ 00 and 8 -+ 0+.
OO
We start by computing lime-->-o+ sin 8 fo l~;~ dy. To this end, let 7] > O. Since
limy->oo l:y1 = 0, there exists some Yo > 0 such that 0 < 1:y2 < 7] holds for
all y 2: Yo. Now, from lime-->-o+ Yo sin8 = 0 and 1ime-->-0+ si~e = 1, we see that
there exists some ° < 8 < 1 such that
° < E < 8 implies ° < Yo sin e < 1] and
°
Now, if < e < 8, then (by taking into account that 0 :::; ~~;~ :::; 1 for y :::: 0)
we infer that
Isine (Xl ~~;~ dyl":::; Isine [Yo 1~):~ dy I + ISinel°O ~~;~ dy I

10 10 )'0
:::; Yo sin e + 1] sin e roo e- EY dy

1yo
:::; Yo sin e + 1] sin el°Oe-EY dy
= Yo sine + 1]Si~E < 1] + 21] = 31].
.. . roo ye- C)'

That IS, hm 8 ....,.0+ sm e Jo 1+y2 dy = 0.
For the second limit, note that :~~~ I I : :;
1~)'2 holds for each y E [0, (0). Thus,
in view of the Lebesgue integrability of the function h(y) = l~y2 over [0, (0),
Theorem 24.4 yields
lim [cose roor~C)~ dyJ = lim [cose]. lim [roo t~C~~.\ dyJ
8-+0+ 10 y E-->O+ 8-+0+ 10
= 1· 1o008-->0+
. e- 100-f.!L
CY 1f
hm [1+),2J dy = 1+),2 = z'
0
For the third limit, note that for each r :::: 1 and each y :::: 0, we have
and so by the Lebesgue integrability of g(y) = 2e-Y over [0, (0), it follows from
Theorem 24.4 that
lim roo vsinr+;osr e- ry dy = roo lim [vsinr~~osr e-ryJ dy

r->oo 10 l+y 10 r->oo 1+.\
= l°00dY = o.
Section 26: PRODUCT MEASURES AND ITERATED INTEGRALS 233
Finally, from (**), we see that
10
roo sinx
X
dx = lim
0-+0+ 10
r sinx
X
dx = lim
0-+0+ .
[Sin e10roo ..:!-re-
l+y
oy dy] +
r-+oo
+ lim [cos e roo e-'Y2 dy] _ lim roo ysinr+~osr e- ry dy

0-+0+ 10 l+y r-+oo 10 l+y
=O+I+O=I'
Problem 26.11. Show that if f(x, y) = ye- o+x2 )y2 for each x and y, then
Use the preceding equality to give an alternate proof of the formula
~
1o
00
e
-x2d
X=-.
2
Solution. Note that
and
1 [1 00 00
x2
ye- O+ )y2 dX] dy = (1 00
x2
e- dX) . (1 00
y2
e- dy)
= (1 00
x2
e- dx r.
2 2
Since ye-(1+x )y ::: 0 holds for all x ::: 0 and y ::: 0, Tonelli's Theorem shows
that
roo , 2
(10 X
e- - dX) =~.
1 (1 00 r
xy2
e- sinx dX) dy = l (1r 00
e-
xy2
sinx dy) dx
234 Chapter 4: TIlE LEBESGUE 1N'lEGRAL
holds for all r > 0. By letting r -+ 00 show that
[00 sin x dx = ../iii.

10.JX 2
In a similar manner show that 10 c~ dx = ../iii/2.

00
Solution. Since
1OO(lre-Xy2 dx)dy = 11(lre-Xy2dx)dY+ 1(ir OO

1ry2e-tdt)dY
+ 1 (? 1
1 r OO 00
t
:s 1 (l Idx) dy e- dt) dy
:s r + 1?OO
dy = r + 1,
it follows from Tonelli's Theorem that e-xy2 is integrable over [0, r] x [0, 00).
I
In view of le- Xy2 sin x :s e-xy2 , we see that e-xy2 sin x is also integrable over
[0, r] x [0,00), and the stated identity follows from Fubini's Theorem.
Performing the innermost integrations and using the elementary integral
f e-at sint dt = asint+cost

l+a1
e- ar
,
we get
[00£
10 l+y
_ [00 y2 sinr+cosr e- ry2 dy
10 l+y4
_../ii
- 2
r
10 .fi
sinx dx
.
ry2 1+y2 = f(y) holds , and f is Lebesgue integrable

l+y4 e- 1 -< l+y4
Since 1)'2 sinr+cosr
over [0,00), it follows from Theorem 24.4 that
Now, using the elementary integral
f .!!L = _1_ [In ( 1+y./2+y2)

I+y· 4./2 l-y./2+y~
+ 2 arctan I-y-
yv'2, ]
and an easy computation, we see that
1o
00 .!!L _
I+y· -
11-
0
.!!L
Hy· +
lex>.!!L
1+
_
l+y4 -
rr./2
4 .
Thus, from (*), we see that
[e:::o sinor dx
JO..fi
= lim
r-+oo
r
Jo
sinor
..fi
dx = -L . rr./2
.fii 4
= ..tiii.
2
Problem 26.13. Using the conclusions of the preceding problem (and an ap-
propriate change of variable), show that the values of the Fresnel integrals (see
Problem 24;6) are
1 00.
0 0
sm(x 2 )dx = 1 00
cos(x 2 )dx ../2ii

= --.
4
Solution. Using the change of variable x = -/U, we get
1 o
r
sin{x 2) dx = 120 1.fi sinll
.[ii
du and
1o
r
cos{x 2 ) dx =~ 1.fi "II
0
CO~' duo
Now, let r ~ 00 and use the preceding problem.
Problem 26.14. Let X = Y = [0, 1], fJ- = the Lebesgue measure on [0, 1], and
v = thecounang measure on [0, 1]. Consider the "diagonal" t::.. = {(x, x); x E X}
of X x Y. Then show that
a. t::.. is a fJ- x v-measurable subset of X x Y, and hence, Xl!. is a non-negative
fJ- x v-measurable function.
b. Both iterated integrals JJ Xl!. dfJ-dv and JJ Xl!. dvdfJ- exist.
C. The function Xl!. is not fJ- x v-integrable. Why doesn't this contradict
Tonelli's Theorem?
Solution. (a) Consider the two sets
A={(X,Y)EXxY;X>y} and B={(X,Y)EXxY;X<Y}.

a b x
FIGURE 4.4.
Note that A = U[a, b] x [c, d], where the union extends over all rectangles
[a, b] x [c, d] with rational end points and a > d; see Figure 4.4.
Clearly, the collection of all such rectangles is countable. Since each rectangle
is IL x v-measurable, it follows that A is IL x v-measurable. Similarly, the set B
is IL x v-measurable. Hence, fl = X x Y \ A U B is a IL x v-measurable set.
(b) Note that
ff Xl:>.dlL dv = 101[1o1Xl:>.(X,Y)dlL(X)]dV(Y)
1 1
= 10 [10 X{y}dlL(X)] dv(y) = 1010. dv(y) = 0,
and
ff 1 1
Xl:>. dvdlL = 10 [10 Xl:>. (x, Y)dV(Y)] dlL(X)
1 1
= 10 [10 X{x} dV(Y)] dlL(X) = 10\' dlL(X) = 1.
(c) Fubini's Theorem combined with part (b) shows that Xl:>. is not integrable over
X x Y (i.e., (IL x v)*(fl) = 00 must hold). This does not contradict Tonelli's
Theorem because v is not a a-finite measure.
Problem 26.15. Let f: 1R -+ 1R be Borel measurable. Then show that the func-
tions f(x + y) and f(x - y) are both A x A-measurable.
. g(x,y) =0
g(x,y) = 0
FlGURE 4.5.
Solution. Since f is the limit of a sequence of step functions, it suffices to

establish the result for characteristic functions of measurable sets of finite measure.
The regularity of the Lebesgue measure allows us to reduce it to the characteristic
functions of open sets of finite measure. Finally, this can be reduced to the case
when f = X(a,b) for some finite open interval (a, b).
The A x A-measurability of the function g(x, y) = X(a,b)(X + y) follows easily
from the graph shown in Figure 4.5.
The A x A-measurability of f(x - y) can be proven in a similar manner.
CHAPTER 5 ______________
NORMED SPACES AND

Lp·SPACES
27. NORMED SPACES AND BANACH SPACES
Problem 27.1. Let X be a normed space. Then show that X is a Banach space
If and only if its unit sphere {x
EX; IIx II = I} is a complete metric space (under
the induced metric d(x, y) = IIx - yll).
Solution. Let S = {x EX; IIx II = I}, and note that S is a closed set. Clearly,
if X is a Banach space, then S is a complete metric space.
Conversely, assume that S is complete. Let {xn} be a Cauchy sequence of X.
In view of the inequality
we see that {lIx n II} is a Cauchy sequence of real numbers. If lim IIxn II = 0, then
lim Xn = O. So, we can assume that 0 = lim IIxnll > O. In this case, we can also
assume that there exists some M > 0 such that lI';n II ::: M and IIxn II ::: M both
hold for each n. The inequalities
t. Xm I - II (lIxmllx.-lIxnllxml II
I lI:tnll - IIXmll - IIx.II·llx.. 1I
::: M211l1 xmll(xn -Xm) - (lIx"II-lIxmll)xmII
::: 2M 3 11x" - xmll,
show that the sequence {II:~: II} is a Cauchy sequence of S. If x is its limit in S,
then Xn = IIx nII . ~ ---+ ox holds in X, and so X is a Banach space.
Problem 27.2. Let X be a normed vector space. Fix a E X and a nonzero

scalar ct.
239
240 Chapter 5: NORMED SPACES AND Lp.SPACES
a. Show that the mappings x H- a + x and x H- ax are both homeomor-

phisms.
b. If A and B are two sets with either A or B open and a and f3 are nonzero
scalars, then show that aA + f3B is an open set.
Solution. (a) Observe that lI(a + x) - (a + y)/I = IIx - y/l holds for all x and
y. This shows that x H- a +
x is, in fact, an isometry.
Also notice that for all x, y E X we have /lax - ay/l = lal . /Ix - YII. This
easily implies that x H- ax .is a homeomorphism.
(b) Assume first that B is an open set. Since the mapping x H- a + x is a
homeomorphism, we know that a + B is an open set for each a EX. This implies
that the set A + B = UoeA (a + B) is an open set for each subset A of X.
Now, assume that B is an open set and that a and f3 are nonzero scalars. Since
the mapping x H- f3x is a homeomorphism, the set f3B is an open set. So, by the
preceding discussion, aA + f3B must be an open set.
Problem 27.3. Let X bea normed vector space, and let B = {x EX: IIxll < I}
be its open unit ball. Show that Ii = {x EX: IIxll :::: I}.
Problem 27.4. Let X be a normed space, and let {x n} be a sequence of X such

that limxn =x holds. If Yn = n-1(xl + ... + Xn) for each n, then show that
limYn = x.
Solution. Let 8 > O. Choose some k with IIx" - xII < 8 for all n > k. Fix
some m > k so that 1I~[(xl - x) + .. , + (Xk - x)]11 < 8 holds for all n > m.
Thus, if n > m, then
llYn-xII = IIH(xl -xH"'+(Xk -X)+(Xk+l -x)+",+(xn -x)]11

:::: IIH(xl - x) + ... + (Xk -x)]11 + Hllxk+l -xII + ... + IIxn - xII]
< 8 +8 = 28.
That is, lim Yn = X holds. (See also Problem 4.11.)
Problem 27.5. Assume that two vectors x and Y in a normed space satisfy
IIx + yll = IIxll + lIylI. Then show that
lIax + f3ylI = allxll + f3l1yll

holds for all scalars a 2: 0 and f3 2:' O.
Solution. Assume IIx + yll = IIxll + lIylI holds for two vectors x and y in a
normed space, and let a 2: 0 and f3 2: O. Without loss of generality, we can
Section 27: NORMED SPACES AND BANACH SPACES 241
suppose that a 2: ,B 2: O. From the triangle inequality, it follows that
lIax + ,8yll ::: allxll + ,Bllyll·

Next, notice that
lIax + ,By II = \\a(x + y) + (,8 - a)y\\

2: \\a(x + y)\\-\\(,B - a)y\\
= allx + yll - (a - ,B)lIyll = a(lIxll + lIyll) - (a - ,B) lIy II
= allxll + ,Bllyll.
Hence, lIax + ,By II = allxll + ,Bllyll, as desired.
Problem 27.6. Let X be the vector space of all real-valued functions defined on
[0, 1] having continuousfirst-order derivatives. Show that II!II = If(O)1 + 111'1100
is a norm on X that is equivalent to the norm II !II 00 + IIf'lIoo.
Solution. The verification of the norm properties of II . II are straightforward.

From
f(x) = f(O) + foX I'(t) dt,

we see that If(x)l::: If(O)1 + 111'1100 holds for each x E [0, 1], and consequently,
1If1l00 ::: If(O)1 + 111'1100 holds.
The equivalence of the two norms follows from the inequalities
1If1l00 + 111'1100 ::: If(O)1 +2111'1100 ::: 2(lf(0)1 + IIf'lIoo)

= 211f1l ::: 2(II!11oo + IIf'lIoo).
Problem 27.7. A series 2::1X/l in a normed space is said to converge to x If

lim IIx - 2:;'=lx;1I = O. As usual, we write x = 2::1xn. A series 2::1X/I is
said to be absolutely summable if 2::1I1x/l1l < 00 holds.
Show that a normed space X is a Banach space if and only if every absolutely
summable series is convergent.
Solution. Let X be a Banach space, and let 2::1 xn be an absolutely summable

series. For each n let Sll = 2:7=1 Xi. The inequality
II+P II+P
\\Sn+p - Sn \\= L
1\ X; II ::: L IIx; \I
;=n+l ;=11+1
implies that {Sl!} is a Cauchy sequence, and hence, convergent in X.
For the converse, let {xn} be a Cauchy sequence in a nonned space X whose
absolutely summable series are convergent. By passing to a subsequence (if nec-
essary), we can assume that IIxn+! - Xn II < 2-n holds for each n. Put Xo = 0, and
note that 2::0 IIxn+!-xn ll < 00. Thus, limn .... ~ 2:~,:d(Xi+!-Xi) = limn .... ooxn
exists in X so that X is a Banach space.
Problem 27.8. Show that a closed proper vector subspace of a normed vector
space is nowhere dense.
Solution. Let E be a proper closed subspace of a nonned space X. Assume that

°
E has an interior point a. Then, there exists some r > such that B (a, r) S;; E.
Now, if y is an arbitrary nonzero element of X, then a + 211Yll Y E B(a, r) S;; E,
and so y = 21~" [(a + 211YIlY) - a]E E. That is, E = X holds, a contradiction.
Thus, EO = C/J.
Problem 27.9. Assume that f: [0, 1] -+ R is a continuous function which is

not a polynomial. By Corollary 11.6 we know that there exists a sequence of
polynomials {Pn} that converges uniformly to f. Show that the set of natural
numbers
{k E 1N: k = degree of Pn for some n}
is countable.
Solution. Let f: [0, 1] -+ R be a continuous function which is not a polynomial,

and let {PII} be a sequence of polynomials that converges unifonnly to f on [0,1].
Assume by way of contradiction that the set of natural numbers
K = {k E 1N: k = degree of Pn for some n}
is bounded. This means that there exists some mEN such that every Pn has
degree at-most m. So, if V is the finite dimensional vector subspace generated in
C[O, 1] by the functions {I, x, x2, ... , x m }, then {Pn} S;; V holds. Now, a glance
at Theorem 27.7 guarantees that V is a closed subspace of C[O, 1], and thus the
(sup) nonn limit f of {PII} must lie in V. That is, f must be a polynomial of degree
at-most m, contrary to our hypothesis. Hence, K is not bounded, and therefore it
must be a countable set; see Theorem 2.4.
Problem 27.10. This problem describes some important classes of subsets of a

vector space. A nonempty subset A of a vector space X is said to be:
a. symmetric, ifx E A implies -x E A, i.e., if A = -A;
Section 27: NORMED SPACES AND BANACH SPACES 243
b. convex, if x, YEA implies AX + (1 - A)Y E A for all 0 :s A :s 1, i.e.,

for every two vectors x, YEA the line segment joining x and Y lies in
A; and
c. circled (or balanced) if x E A impli~s AX E A for each IAI :s 1.
i. A circled set is symmetric.
ii. A convex and symmetric set containing zero is circled.
iii. A nonempty subset A of a vector space is convex if and only if aA + bA =
(a + b)A holds for all scalars a 2: 0 and b 2: O.
iv. If A is a convex subset of a normed space, then the closure A and the
interior AD of A are also convex sets.
Solution. (i) Let A be a circled set. Since I - 11 = 1 :s 1, it follows that

- x = (-I)x E A for each x E A. Thus, A is a symmetric set.
(ii) Let A be a convex symmetric set containing zero. Fix x E A and IAI :s 1.
If 0 :s A :s 1, then Ax = AX + (1 - A)O E A and if -1 :s A < 0, then
AX = (-A)( -x) + (1 + A)O E A. So, A is a circled set.
(iii) Let A be a subset of a vector space. Assume first that A is a convex set, and
let a > 0 and b > O. If x E (a + b)A = {(a + b)u: u E A}, then for some u E A,
we have x = (a + b)u = au + bu E aA + bA, and so (a + b)A ~ aA + bA
is always true. Now, let x E aA + bA. Then, there exist £I, v E A such that
x = au + bv. Since A is convex, we have z = a~bu + a~b V E A, and so
x = (a+b)[a~bu+ a~b v] = (a+b)z E (a+b)A. Therefore, aA+bA ~ (a+b)A
is also true, and consequently aA + bA = (a + b)A.
Next, suppose that aA + bA = (a + b)A holds true for all a 2: 0 and b 2: O.
Let x, yEA and 0 :s A :s 1. Letting a = A and b = 1 - A, we see that
AX + (1 - A)Y = ax + by E (a + b)A = A.
This shows that A is a convex set.

(iv) Let A be a convex subset of a normed space. We show first that A is a convex
set. To this end, let x, yEA and fix 0 :s A :s 1. Pick two sequences {x n} and {Yn}
of A such thatxn -+ x andy" -+ y. Putz" = AXn +(l-A)Y" and note that {z,,} is
a sequence of A. Since the function f: X -+ X, defined by feu) = All +(I-A)u,
is continuous (see Problem 27.2), it follows that z" -+ AX +(1- A)y. This implies
AX + (1 - A)y E A, so that A is a convex set.
Next, we shall show that AO is a convex set. Fix 0 :s A :s 1. Since AO is an
open set, it follows (from Problem 27.2) that the set AAo + (1 - A)AD is also an
open set, which (since A is convex) is contained in A. Since AD is the largest open
set contained in A, we infer that AAD + (1 - A)AD ~ AD. This shows that AD is
a convex set.
244 Chapter 5: NORMED SPACES AND Lp-SPACES
Problem 27.11. This problem describes all norms Oil a vector space X that are
equivalent to a given norm. So, let (X, II . II) be a normed vector space. Let
A be a norm bounded convex symmetric subset of X having zero as an interior
point (relative to the topology generated by the norm II . II). Define the function
PA:X -+ lRby
PA(X) = inf{A > 0: x E AA}.

a. Thefunction PA is a well-defined norm on X.
b. The norm PA is equivalent to II . II, i.e., there exist two constants C > 0 and
K > 0 such that C IIx II :::: P A(x) :::: K IIx II holds for all x EX.
c. The closed unit ball ofpA is the closure of A, i.e., {x EX: PA(X) :::: I} = A.
d. Let II I· II I be a norm on X which is equivalent to II . II, and consider the norm
bounded nonempty symmetric convex set B = {x EX: II Ix II I :::: I}. Then
=
zero is an interior point of B and II Ix II I P B (x) holds for each x EX.
Solution. Assume that A is a norm bounded convex symmetric subset of a

normed space (X, II . II) such that zero is an interior point of A.
(a) Pick some r > 0 such that B(O, 2r) = {x EX: IIxil < 2r} S;;; A. If x E X is
a nonzero vector, then 1I:lIx E B(O, 2r) S;;; A, and sox E 11;11 A. This shows thatthe
set {A > 0: x E AA} is nonempty and so the formula PA (x) = inf{A > 0: x E AA}
is well defined and satisfies
for all x EX. Next, we shall show that PAis a norm on X.

Clearly, PA(X) ?: 0 and PA(O) = O. Now, if PA(X) = 0, then there exist
a sequence {an} S;;; A and a sequence {An} of positive real numbers satisfying
An -+ 0 and x = Anan for each n. Since A is a norm bounded set, it easily follows
that x = lif!1 Anan = O. Thus, PA(X) = 0 if and only if x = O.
Next, we shall show that PA (ax) = la IpA (x) holds for all a E lR and all x EX.
Since A is symmetric, we have
{A> 0: AX E A) = {A > 0: A(-x) E A},
and so for proving PA(ax) = lalpA(x), we can suppose without loss of generality
that a > O. Now, note that
PA(ax) = inf{A > 0: ax E AA) = infrA > 0: x E ~A}

= a inf{ ~: x E ~ A} = a inf{1l > 0: x E IlA)
= apA(x).
Section 28: OPERATORS BETWEEN BANACH SPACES 245
For the triangle inequality, let x, Y E X and fix E > O. Choose A > 0 and x E AA
such that A < PA (x) + E. Likewise, pick some fJ, > 0 such that Y E fJ,A and
fJ, < PA(Y)+E. FromProblem27.lOweknowthatx+y E AA+fJ,A = (A+fJ,)A,
and so
Since E > 0 is arbitrary, we infer that PA(X + y) ::: PA(Y) + PA(Y)'

(b) Let x E X and fix some M > 0 such that lIall ::: M holds for all a E A.
Now, if A > 0 satisfies x E AA, then there exists some YEA such that x = Ay.
Hence, IIxll = AlIylI ::: AM, or A 2: IIxIlIM. This implies PA(X) 2: -ttllxll. Now,
combine this inequality with (*) to establish that PAis a norm equivalent to II . II.
(c) Assume first that PA(X) ::: 1 and x ¥= O. Then for each n there exist
o < All ::: 1 + ~ and all E A such that x = Anan. By passing to a subsequence, we
can assume All -+ A. Since x ¥= 0 and A is a norm bounded set, it easily follows
that 0 < A ::: 1. Now, note that the sequence {an} ~ A satisfies all = t;x -+ ix,
and so ix E A. Since A is also a convex set (see Problem 27.10), we see that
x = A(tX) + (1- A)O E A.
Now,letx EA. Then, there exists a sequence {XII} ~ A such that IIxll - x II -+ O.
Since II . II is equivalent to PA, we also have PA(Xn - x) -+ O. In particular,
PA (x,,) -+ PA (x). Now, notice that since XII E A, we have PA (x n) ::: 1 for each n.
This implies PA (x) ::: 1. Therefore, the closed unit ball of PA is A.
(d) Let III . III be a norm on X which is equivalent to II . II. It is easy to check that
the closed unit ball B of III . III is a bounded convex and symmetric set containing
zero as an interior point. We shall show next that II Ix III = PB(x) holds for each
x EX.
To see this, let x E X be a nonzero vector. Since xl II Ix III E B, we see that
PB(x)/lIlxlll = PB (xIII Ix II I) ::: 1, and so PB(X) ::: IIlxlll. On the other hand, there
exist a sequence {An} of positive real numbers and a sequence {b ll } of B such that
An -+ PB(X), b" E B and x = A"bll for each n. Since IIlxlll = An IIlbll III ::: An, we
easily infer that II Ix III ::: PB(x). Hence, PB(x) = II Ix III for each x EX.
28. OPERATORS BETWEEN BANACH SPACES
Problem 28.1. Let X and Y be two Banach spaces and let T: X -+ Y be a

bounded linear operator. Show that either T is onto or else T(X) is a meager set.
Solution. Assume that T(X) is not a meager set. Then, we have to show that
T (X) = Y holds.
Let V = {x E X: IIxll ::: I}. Since (by assumption) T(X) is not a meager
set, some nT(V) = nT(V) has an interior point. This implies that T(V) has
an interior point. So, there exists some Yo E T(V) and some r > 0 such that
B(yo,2r) S;; T(V) = -T(V). Note that if y E Y satisfies lIyll < 2r, then
y - Yo = -(Yo - y) E T(V) and so y = (y - Yo) Yo E T(V) + T(V) S;; +
2T(V). (The last inclusion follows, of course, from the identity V V 2V.) + =
. Consequently, we have established that {y E Y: lIyll < r} S;; T(V). From the
linearity of T, we infer that
holds for each n.

Next, let y E Y be fixed such that lIyll < 2- l r = ~. From (**), we know
that y E T(2- 1V). So, for some vector XI E 2- 1V we have lIy - T(xI)1I <
2- 2r. Now, proceed inductively. Assume that Xn E 2- nV has been selec-
ted such that Ily - I
L7=1 T(x;) < 2-n- I r. From (**) it follows that y -
L7=1 T(x;) E T(2- n- 1V), and so there exists some Xn+1 E 2-n- 1V such that
I y - L7:;;11 T (x; ) I < 2-n- 2,.. Thus, there exists a sequence {xn} of X such that
IIxn II ::::: 2- n and
Ily - tT(x;)11
1=1
= Ily - T ( t x ;)
1=1
I < Tn-Ir
hold for each n. Now, for each n let Sn = XI + ... + Xn and note that the relation
n+p n+p 00
IISn+p - Sn I =I L X; I : : : ;=n+1
;=n+1
L Ilx; II::::: ;=n+1
L T; = Tn
shows that {sn} is a Cauchy sequence of X. Since X is a Banach space, the
sequence {sn} converges; let s = limsn • Clearly, IIsll ::::: L:I IIxnll ::::: 1 (Le.,
S E V), and by the continuity and linearity of T, we see that
n
T(s) = lim T(sn) = lim "T(x;) = y.
.i....J
n-'oo n-+oo
;=1
That is, y E T(V), and so {y E Y: lIyll < ~} S;; T(V) S;; T(X). Since T(X) is a
=
vector subspace of Y, the latter inclusion implies that T (X) Y must hold.
Problem 28.2. Let X be a Banach space, T: X -+ X a bounded operato/~ and

I the identity operator on X. If II Til < 1, then show that I - T is invertible.
Solution. If A, B E L(X, X), then the inequalities
easily imply that IIABII ::: IIAII . IIBII. In particular, if a sequence {All} of
operators of L(X, X) satisfies lim An = A in L(X, X) and B E L(X, X), then
the inequality
shows that lim BAn = BA. Similarly, limAnB = AB.

Now, assume T E L(X, X) satisfies IITII < 1. In view of the inequality
IIT"II ::: IITll n , it follows that
00 00
LIIT"II ::: LllTf = I-~ITII < 00.
n=O 11=0
Thus, L:o yn is an absolutely summable series. Since L(X, X) is a Banach

L:o yn converges in L(X, X); see Problem 27.7. Moreover,
space, S =
(1- T)S = lim (I - T)("\' Ti) = lim

" (I - yn+l) = I,
Il~OO ~ IJ-+OO
i=O
and similarly S(I - T) = I. Therefore, S = (I - T)-I .
Problem 28.3. On C[O, 1] consider the two norms
11/1100 = sup{l/(x)l: x E [0, I]} and 11/111 = foll/(X)1 dx.
Then show that the identity operator I: (C[O, 1], II . 1100) -+ (C[O, 1], II . III) is
continuous, onto, but not open. Why doesn't this contradict the Open Mapping
Theorem?
Solution. Clearly, I is onto, and in view of the inequality II fill ::: II 11100, we
see that I is also continuous.
For the rest of the proof, we need to show that (C[O, 1], II· III) is not a Banach
space. To establish this, consider the sequence {f,l} of continuous functions whose
graphs are shown in Figure 5.1.
The inequality II In+p - In III ::: fz shows that {f,l} is a Cauchy sequence for
the norm II . III. Assume by way of contradiction that lim II In - 1111 = holds °
for some I E C[O, 1].
FIGURE 5.1.
Let a E (0, t).

If I(a) #- 1, then there exist some 8 > and some < 8 <° °
t
minta, -a} such that I/(x) -112: 8 holds whenever Ix -al < 8. Now, note
that 288 ~ !oll/n(x)-/(x)ldx = II/n-/11I for all sufficiently large n,contrary
to lim II In - Ilh =
0. Thus, I(a) = 1 holds for all a E (0, t).
Similarly,
I(a) =°
for all a E (t,
1). Now, it is readily seen that I cannot be a continuous
function, contrary to IE C[O, 1]. Thus, Un} does not converge in C[O, 1] with
respect to the II . III norm.
Finally, I: (C[O, 1], II . 11(0) ---7 (C[O, 1], II . III) cannot be an open mapping.
Since otherwise, II· III and II . 1100 would be equivalent norms, and therefore
(C[O, 1], II . III) would be a Banach space.
Problem 28.4. Let X be the vector space 01 all real-valued functions on [0, 1]
that have continuous derivatives with the sup norm. Also, let Y = C[O, 1] with
the sup norm. Define D: X -+ Y by D(!) = /'.
a. Show that D is an unbounded linear operato/:
b. Show that D has a closed graph.
c. Why doesn't the conclusion in (b) contradict the Closed Graph Theorem?
Solution. (a) The standard properties of differentiation guarantee that D is a

linear operator. Now, for each n let In(x) = xn. Then, In E X and II In 1100 =
sup{If,/(x)l: °
~ X ~ I} = 1 for each n. Now, notice that D(fn)(x) = nx IJ - I
holds for each n, and from this it follows that
Therefore, II D II = 00, and so D is an unbounded operator.

(b) To see that D has a closed graph, assume In -7 0 in X and Din = I~ -7 g

in Y. That is, Un} converges uniformly to zero, and U~} converges uniformly
to g. We have to show that g = O.
From 10" I,;(t)dt = In(x) - In(O) (and P:oblem 9.16), it follows that
10r'g(t)dt = n-.oo
lim r'f,;(t)dt = lim [In(X) -
10 n ..... oo
j;1(0)] =0
holds for all x E [0, 1]. Differentiating, we get g(x) = 0 for each x E [0, 1], as
required. (See also Problem 9.29.)
(c) The conclusion in (b) does not contradict the Closed Graph Theorem since
X is not a Banach space. For instance, we know (from Corollary 11.6) that every
function I E C[O, 1] is the uniform limit of a sequence of polynomials. So, if
I E C [0, 1] is a nondifferentiable continuous function and {Pn} is a sequence of
polynomials that converges uniformly to I, then {Pn} is a Cauchy sequence of X
which cannot converge in X.
Problem 28.5. Consider the mapping T: C[O, 1] -7 C[O, 1] defined by T I(x) =

x 2 l(x)lor all IE C[O, 1] and each x E [0, 1].
a. Show that T is a bounded linear operator.
b. II I:C[O, 1] -7 C[O, 1] denotes the identity operator (i.e., I(f) = I lor
each IE C[O, 1]), then show that III + TIl = 1 + IIT11.
Solution. (a) From the identities
T(f + g)(x) = x 2 (f + g)(x) = x 2 /(x) + x 2 g(x) = (T 1+ T g)(x)

and
T(af)(x) = ax 2 /(x) = (aTI)(x),
we easily infer that T is a linear operator. For the norm of T, note that for each
I E C[O, 1] we have
IITllloo = sup ITI(x)1 = sup x 2 1/(x)l::: sup I/(x)1 = 1111100,

.rE[O, I] .tE[O,I] .tE[O,I]
and so II Til::: 1. On the other hand, for the constant function 1, we have
liT II :::: IIT11100 = sup x 2 = 1.

.tE[O,!]
Thus, II Til = 1, and so T is a bounded operator.

(b) Clearly, II I + Til::: II I II + II Til = 1 + 1 = 2. Moreover, we have
III + TIl 2: 11(1 + T)ll1oo = sup (1 + x2) = 2,

xe[O,l]
and so III + Til = 1 + liT II = 2 holds true.

Problem 28.6. Let X be a vector space which is complete in each of the two
norms II· lit and 11·112. If there exists a real number M > Osuchthatllxllt ::: Mllxll2
holds for all x EX, then show that the two norms must be equivalent.
Solution. The identity operator I: (X, II . 112) -+ (X, II . lit) is one-to-one,

continuous, and onto. By the Open Mapping Theorem it is a homeomorphism,
and the conclusion follows.
Problem 28.7. Let X, Y, and Z be three Banach spaces. Assume thatT: X -+ Y

is a linear operator and S: Y -+ Z is a bounded, one-to-one linear operator. Show
that T is a bounded operator if and only if the composite linear operator SoT
(from X into Z) is bounded.
Solution. Assume that SoT is a bounded operator. Let Xn -+ 0 in X and

T(xn) -+ y in Y. Using that SoT and S are both continuous, we get
S(y) = lim S(T(x lI )) = lim So T(x n) = O.

n~oo n-+oo
Since S is one-to-one, we infer that y = 0, and hence-by the Closed Graph

Theorem-the operator T is continuous.
Problem 28.8. An operator P: V -+ V on a vector space is said to be a pro-

jection if p2 = P holds. Also, a closed vector subspace Y of a Banach space is
said to be complemented if there exists another closed subspace Z of X such that
Y EB Z = X.
Show that a closed subspace of a Banach space is complemented if and only if
it is the range of a continuous projection.
Solution. Let Y be a closed subspace of a Banach space X. Assume first that

there exists a continuous projection P: X -+ X whose range is Y, i.e., P (X) = Y.
From p2 = P, it follows that Y = {y E X: y = Py}.
If I: X -+ X denotes the identity operator, let Z = (I - P)(X), the range of
the continuous operator I - P. Clearly, Z is a vector subspace of X and in view
of x = Px + (I - P)(x), we see that Y + Z = X. Now, if u E Y n Z, then
Section 29: LINEAR FUNCTIONALS 251
u = Z - pz for some Z E Z, and so u =

P(u) =
P(z - pz) =
P(z) - p2(z) o.=
This means that Y EB Z = X. Finally, to see that Z is also closed, assume that a
sequence {zn} of X satisfies (1- P)(ZIl) -r z. Then, the continuity of P implies
0= P(I - P)(zn) -r Pz, and so pz = o. =
Bence, z (l - P)(z) E Z, proving
that Z is also closed. Thus, Y is a complemented closed subspace.
For the cQnverse, assume that Z is a closed subspace such that Y EB Z =
X. So,
for each x E X there exist Y E Y and z E Z (both uniquely determined) such
that x = Y + z. Define an operator P: X -r X via the formula P (x) =
Y, where Y
satisfies x - y E Z. We claim that P is a continuous projection whose range is Y.
Notice first that P is a linear operator. Also, p2(X) = P(y) = y =
P(x) holds
for each x EX, so that P is a projection. Clearly, the range of P is Y. To finish
the proof, we must show that P is also continuous. For this, it suffices to show (in
view of the Closed Graph Theorem) that P has a closed graph.
To this end, assume that a sequence {xn} of X satisfies Xn -r x and P(xn) ~ y
in X. For each n letxn = Yn + Zn, where Yn E Y and ZII E Z. Clearly, Yn = P(xn)
for each n. Since Y is a closed subspace, it follows that Y E Y. Now, from
Zn =XII - Yn -r x - Y and the closedness of Z, we infer that Z =
x - Y E Z.
=
Thus, x = Y + z, and so Y P(x). This shows that P has a closed graph, and we
are done.
29. LINEAR FUNCTIONALS
Problem 29.1. Let I: X ~ lR be a linear functional defined on a vector space

X. The kernel 01 I is the vector subspace
Ker 1= I-I({O}) = {x EX: I(x) = O}.
II X is a normed space and I: X -r lR is nonzero linear functional, establish the

lollowing:
a. I is continuous if and only if its kernel is a closed subspace 01 X.
b. I is discontinllolls if and only if its kernel is dense in X.
Solution. (a) Clearly, if I is continuous, then its kernel I-I({O}) is a closed

set. For the converse, assume that I =1= 0 and that I-I ({O}) is a closed set. Pick
some e E X with I(e) = 1.
Suppose by way of contradiction that II I II =
00. Then, there exists a sequence
{xn} of X with IIxnll =
1 and I/(xn)1 2: n for each n. Note that the sequence
{Yn},definedby Yn =
e- f~tn),satisfies Yn E l-t({O}) foreach nand YII -:-+ e.
Since the set I-I({O}) is closed, it follows that e E l-t({O}), and so I(e) = 0,
which is a contradiction. Thus, I is a continuous linear functional.
(b) IfKer I is dense in X, then Ker I is not closed and hence, by part (a), I is not
continuous. For the converse, assume that I is a discontinuous linear functional,
i.e., 11/11 = 00. This implies (as in the previous part) that there exists a sequence
(x n} of X satisfying IIxnll = 1 and I/(xn)1 2: n for each ll. Now, if x E X and
Yn = X - fg;)x n, then (yn} is a sequence in Ker I and satisfies Yn -+ x. This
shows that Ker I is dense in X.
Problem 29.2. Show that.a linear functional Ion a normed space X is discoll-
tinuous if and only iflor each a E X and each r > 0, we have
I(B(a, r» = (f(x): lIa - xII < r} = JR.
Solution. Let I be a linear functional on a normed space X and let B =

B(O, 1) = (x EX: IIxll < I}. Assume that I is discontinuous. Fix a E X and
r > O. From the relation B(a, r) = a + r B(O, 1) = a + r B and the linearity of
I, it follows that I(B(a, r») = JR holds if and only if I(B) = m.
We claim first that I(B) is unbounded from above in JR. To see this, assume
by way of contradiction that there exists some M > 0 such that I (x) ::: M holds
for each x E B. Note that if x E X satisfies IIxll ::: 1, then ±~x E B, and so
from
±~/(x) = I(±~x) ::: !!f,

we see that II (x) I ::: M holds for all x E X with IIx II ::: 1. That is, II! II =
sup{l/(x)l: IIxll ::: I} ::: M < 00, and so I is a continuous linear functional, a
contradiction. Thus, I(B) is unbounded from above in JR. Now,let a > 0 be an
arbitrary positive real number. By the above, there exists some x E B satisfying
I(x) > a. Now, note that the element y = /(x)x E B satisfies I(y) = a (and,
of course, -y E B satisfies I( -y) = -a). Consequently, I(B) = JR.
Forthe converse, assume that I (B (a, r») = JR holds for each a E X and each
r > O. In particular, from
00 = sup{l/(x)l: IIxll < !} ::: sup{l/(x)l: IIxll ::: I} = 1If1l,

we see that 11/11 = 00. Thus, I is unbounded and so (by Theorem 28.6) I is a
discontinuous linear functional.
Problem 29.3. Let I, II, /Z, ... ,In be linear lunctionals defined on a com-
mon vector space X. Show that there exist constants AI, ... , An satisfying I =
2.:::7:1 A;/i (i.e., I lies in the linear span 01II, ... , In) ifand only if n?=1 Ker Ii S;
Ker/·
Solution. If f = L;'=I Ai fi holds, then clearly n;'=1 Ker fi ~ Ker f. For the
converse, assume n;'=1 Ker fi ~ Ker f. Let
v = {Y E JR.": 3 x E X.such that y::;: (f1(X), h(x), ... , fll(x»)},
It is easy to verify that V is a vector subspace of JR.". Now, define the linear
functional g: V -+ JR. via the formula
g(fI(X), h(x), ... , /,,(x») = f(x).
Notice that g is well defined. To see this, assume
(f1(X), h(x), ... , fll(x») = (f1(Y), h(y),···, fll(Y»)'
=
Then, J;(x - y) 0 for each i, and sox - Y E n7=1 Ker J;. From our hypothesis,
it follows that x Y E Ker f, which means that f(x) =
fey). Now, it is a routine
matter to verify that g is linear.
Denote by g again a linear extension of g to all of JR.". This implies that
there exist scalars AI,"" AI! such that g(ZI, ... , ZIl) =
L;'=l AiZi holds for all
(Zl, ... , ZIl) E JR.". In particular, we have
n
f(x) = g(fI(X), h(x), ... , /,,(x») = I>;J;(x)
i=1
for all x EX, as desired.
Problem 29.4. Prove the converse of Theorem 28.7. That is, show that if X and
Yare (nontrivial) normed spaces and L(X, Y) is a Banach space, then Y is a
Banach space.
Solution. Let (YIl) be a Cauchy sequence of Y. Pick some f in X* with

f i= 0, and then consider the sequence of operators {Til} of L(X, Y) defined by
Tn(x) = f(x)YIl' The inequality
IIT,,(x) Tm(x) II = II!(X)(Yn - Ym)11 :s IIfII·IIYn - Ymll'lIxll,
shows that II Til - Tm II :s II f II . II Yll - Ym II, and so (T,,) is a Cauchy sequence of

L(X, Y). By the completeness of L(X, Y), there exists some T E L(X, Y) with
lim Til = T. Now, pick some e E X with fee) = 1, and note that
lim TIl(e)
n-+oo
= n-+co
lim Yn = TCe).
Problem 29.5. The Banach space B(1N) is denoted by loo. That is, loo is the
Banach space consisting of all bounded sequences with the sup norm. Consider
the collections of vectors
Co = {x = (XI, X2, X3, ... ) E loo: XII --+ O}, and

c = {x = (XI, X2, X3, ... ) E loo: limxII exists in 1R}.
Show that Co and c are both closed vector subspaces of loo.
Solution. It should be obvious that Co and c are vector subspaces of loo (and
that Co is a vector subspace of c). What needs verification is their closedness. To
see that Co is closed, assume that a sequence {XII} of co, where XII = (xf, x!J., ... ),
=
satisfies IIxlI - X 1100 --+ 0. If X (XI, X2, ... ), we must show that lim XII O. =
To this end, let € > O. Fix some k such that IIxn - X II < € for all n 2: k; clearly
Ixi - Xi I < € also holds for all n 2: k and all i. Since lirni-+oo xf = 0, there exists
some m 2: k such that Ixfl < € holds for all i 2: m. Now, notice that if i 2: m,
then
Ixil ::::: IXi - xf I + Ixf I < € + E = 2E.

This shows that limxn = 0, as desired.
Next, we shall establish that c is closed. For simplicity, for a sequence X =
(XI, X2, ... ) E C we shall write Xoo = limxll • Now, assume that a sequence {XII}
in c satisfies xn --+ x = (XI, X2, ... ) E loo. We must show that lim XII exists in 1R.
Start by fixing some E > 0. Then, there exists some k such that
IIx" - x 1100 < € holds for all n 2: k.
This implies IIx n - xm 1100 < 2E for all n, m 2: k, and so Ixi - xf'1 < 2E for all
n, m 2: k and each i. Consequently, Ix~ - x~ I ::::: 2E for all n, m 2: k. This shows
that {x~} is a Cauchy sequence of real numbers. Let Xoo = lim x~ and note that
Ix~ - xool ::::: 2E holds for all n 2: k.
We claim that Xn --+ Xoo. To see this, let again € > 0 and choose k so that (*)
is true. Next, fix some r 2: k such that Ix~ - x~ I < E holds for all n 2: r. Now,
note that if n 2: r, then
IXn - Xoo I ::::: IXII - x! I+ Ix! - x~ I+ Ix~ - I

Xoo < E + € + 2E = 4€.
This shows thatxn --+ Xoo , and sox· E c. Therefore, cis a closed subspaceofloo.
Problem 29.6. Let c denote the vector subspace of loo consisting of all con-
vergent sequences (see Problem 29.5). Define the limit functional L: c --+ 1R
by
L(x) = L(xt. X2, ... ) = lim x n,

IJ~OO
and p:.e oo ~ lR by p(x) = P(XI, X2, ... ) = lim supxn.

a. Show that L is a continuous linearfunctional, where c is assumed equipped
with the sup norm.
b. Show that pis sub linear and that L(x) = p(x) holds for each x E c.
c. By the Hahn-Banach Theorem 29.2 there exists a linear extension of L to
all of.e oo (which we shall denote by L again) satisfying L (x) ~ p(x) for
all x E .eoo . Establish the following properties of the extension L:
i. For each x E .e oo , we have
liminfxlI ~ L(x) ~ limsupxn.

n-+oo IJ-+oo
ii. L is a positive linear functional, i.e., x ?: 0 implies L(x) ?: O.

iii. L is a continuous linear functional (and in fact II L II = 1).
Solution. (a) Clearly, L is a linear functional. Moreover, if x = (XI, X2, ... ) E c,

then IXn I ~ IIxlioo = supm IXm I for each n, and so IL(x)1 = lim IX nI ~ IIx 1100. This
shows that L is a continuous linear functional. (Since L(l, I, I, ... ) = I, it is easy
to see that ilL II = 1.)
(b) The sublinearity of p follows immediately from Problem 4.7. The equality
p(x) = L(x) = limxn for each x E c should be also obvious.
(c) We shall establish the stated properties.
(i) If x E .e oo , then notice that
-L(x) = L(-x) ~ limsup(-xn) = -liminfxn,

n->oo
n-+oo
and so lim inf Xn ~ L (x) ~ lim sup Xn holds true.

(ii) If x = (XI, X2, ... ) ?: 0 (i.e., if Xn ?: 0 for each n), then it follows from
Problem 4.8 and the preceding conclusion that L(x) ?: lim inf Xn ?: O. That is, L
is a positive linear functional.
(iii) If IIxlioo ~ 1 (i.e., if IXnl ~ 1 for each n), then it follows from part (i) and
Problem 4.8 that
-1 ~ liminfxn ~ L(x) ~ limsupxn ~ I,

ll-HX) n-+oo
and so IL(x)1 ~ 1. This implies IILII = sup{lL(x)l: IIxlioo < I} < 1. Since
L(1,I,I, ... ) = I, we easily infer that ilL II = 1.
Problem 29.7. Generalize Problem 29.6 as/allows. Show that there exists a lin-
earjunctional £im:.e oo -+ 1R (called a Banach-Mazur limit) with the/allowing
properties.
a. £im is a positive linear/unctional a/norm one.
b. For each x = (Xl, X2, ••• ) E .e oo , we have
· . f
1Imm
Xl + X2 + ... + X/I t'.
:::: J..-zm
()
X ::::
l'
Imsup
Xl + X2 + ... + X/I .
n...... oo n. /1 ...... 00 n
In particulal~ £im is an extension 0/ the limit junctional L.

c. For each X = (Xl, X2, ••• ) E .e oo , we have
Solution. For each X = (Xl, X2, ••. ) E .e oo , let

_ (
A(X ) - Xl,
Xl
2
X2 +
' ... ,
Xl + X2 +n ... + Xn •... )
be the sequence of averages of x. If we define p:.e oo -+ 1R via the formula
p(X) = lim sup A(x) = lim sup Xl +X2 + ... +X n

. ,
n...... oo n
then a glance at Problem 4.7 guarantees that p is a sublinear functional. Moreover,

it is easy to see that if X E c, then
L(x) = nlim
...... oo
Xn = p(x).
Now, by the Hahn-Banach Theorem 29.2, L has an extension £im:.e oo -+ 1R sat-

isfying £im(x) :::: p(x) for each X E .e oo • Properties (a) and (b) can be established
exactly as in the solution of Problem 29.6.
To verify (c), let X = (Xl, X2, ... ) E .e oo and put y = (X2, X3, ... ). Then, an easy
computation shows that
Xl - X3 Xl - X4 Xl - Xn+l )
A(x - y) = ( Xl - X2, - - ,
2 3
, ... ,
n
, ....
Since X = (Xl, X2, ••. ) is a bounded sequence, the latter implies

. . Xl -Xn+l
p(X - y) = hmsup n
= O.
Hence, Lim(x - y) .::: p(x - y) = 0. Similarly, L(y - x) .::: 0, and so Lim(x) -

Lim(y) = Lim(x - y) = 0. Thus, .cim(x) = Lim(y), as desired.
Problem 29.S. Let X be a normed vector space. Show that if X* is separable

(in the sense that it contains a countable dense subset), then X is also separable.
Solution. Let {fl, fz, ... } be a countable dense subset of X*. For each n
choose some x" E X with IIx,,1I = 1 and If,,(xn)1 2: ~ II /',11, and let Y be the
closed subspace generated by {Xl, X2, ..• }. We claim that Y = X.
To see this, assume by way of contradiction that Y =1= X. Fix some a ¢. Y with
lIall = 1. By Theorem 29.5, there exists some f E X* with fey) = for all
°
y E Y and f(a) =1= 0. Given £ > choose some n with IIf - f,,11 < £, and
°
note that
Thus, If(a)1 .::: If(a) - fn(a)1 + Ifn(a)1 < 3£ holds for all £ > 0, and so f(a) =
0, a contradiction. Therefore, Y = X holds.
Now, note that the collection of all finite linear combinations of the countable
set {Xl, X2, •.• } with rational coefficients is a countable dense subset of X.
Problem 29.9. Show that a Banach space X is reflexive if and only if X* is

reflexive.
Solution. Assume that X* is reflexive. If X =1= X**, then by Theorem 29.5

there exists some nonzero F E X*** with F(x) = °
for each X E X. Since X*
is reflexive, there exists a nonzero x* E X* so that FU) = f(x*) holds for all
f E X**. In particular,
x*(x) = _t(x*) = F(x) = °

holds for all X EX, and so x* = 0, a contradiction. Therefore, X must be a
reflexive Banach space.
Problem 29.10. This problem describes the adjoint of a bounded operator. If

T: X -+ Y is a bounded operator between two normed spaces, then the adjoint
ofT is the operator T*: y* -+ X* defined by (T* j)(x) = f(Tx) for all f E y*
and all X EX. (Writing h(x) = (x, h), the definition of the adjoint operator is
written in "duality" notation as
(Tx, f) = (x, T* f)
258 Chapter 5: NORMED SPACES AND Lp·SPACES
for all f E Y* and all x EX.)

a. Show that T*: Y* -+ X* is a well-defined bounded linear operator whose
norm coincides with that ofT, i.e., IIT*II = II TIl·
b. Fix some g E X* and some u E Y and define S: X -+ Y by Sex) = g(x)u.
Show that S is a bounded linear operator satisfying liS II = IIgil . IIuli. (Any
such operator S is called a rank-one operat01:)
c. Describe the adjoint of the operator S defined in part (b).
d. Let A = [aU] be an m x n matrix with real entries. As usual, we consider
the adjoint A * as a (bounded) linear operator from lRn to lRm. Describe
A*.
Solution. As usual, we shall denote from simplicity T (x) by T x.

(a) Fix f E Y*. Then, for x, y E X and a, f3 E lR, we have
(T* f)(ax + f3y) = f(T(ax + f3y)) = f(aTx + f3Ty)

= af(Tx) + f3f(Ty) = a(T* f)(x) + f3(T* f)(y),
so that T* f is a linear functional on X. To see that T* f is also continuous, notice
that
I(T* f)(x) I = If(Tx)1 :s IIfil . IIT(x)II :s IIf11 . IITII ·lIxll
holds for all x E X. This shows that T* f is a bounded (and hence, continuous)
linear functional and that IIT* fll :s 1IT11 . IIfII holds true for each f E Y*.
The last inequality also shows that T*: y* -+ X* is a bounded operator and
that II T* II :s II TIl. For the reverse inequality, let x E X satisfy IIx II :s 1. By
Theorem 29.4 there exists some h E y* satisfying IIhll = 1 and h(Tx) = IITxll.
So,
IIT*II ::: IIT*hll ::: IIT*h(x)1I = IIh(Tx)II = IITxll,
for each x E X with IIxll :s 1. This implies 1IT11 = sup{IITxll: IIxli :s 1}:s IIT*II·
Hence, IIT*II = IITII·
(b) It is a routine matter to verify that S is linear. From
IIS(x)1I = IIg(x)ull = Ig(x)l· lIuli

:s IIgli . 1I:x1I·lIull = (lIglI· lIull)lIxll,
we see that S is a bounded operator and that IISII :s IIgll . lIuli. Now, if x E X
Section 30: BANACH LATTICES 259
satisfies IIx II :::: 1, then we have
IISII 2: IIS(x)1I = IIg(x)ull 2: Ig(x)l· lIull,
and so IISII 2: sup{lg(x)I·lIull: x E X and IIxll :::: I} = IIgll·lIuli. The preceding

show that IISII = IIgll ·llull·
(c) Note that for each f E y* and each x E X we have
(S* f)(x) = f(Sx) = f(g(x)u) = f(u)g(x) = [J(u)g](x).

So, S* f = f(u)g holds for all f E Y*.
(d) Let A = [aij] be an m x n real matrix. Note that the norm dual of JR./I is
again JR./J, where every y = (Yl, ... , Y/I) E JR./I defines a linear functional on JR."
via the formula
/I
y(x) = (x, y) = X· Y= I:>iYi.
;=1
This easily implies that the adjoint A * of A is an n x m matrix B = [bij] that

satisfies the duality identity (Ax, y) = (x, A*y), or Ax· y = x· By. That is, the
elements of B satisfy the equation
for all x . E JR." and all y E JR.m • This easily implies bij = a j; for all i and j.
Therefore, A * is the transpose of A, i.e. A * = AI.
30. BANACH LATTICES
Problem 30.1. Let X be a vector lattice, and let f: X+ -+ [0, (0) be an additive
function (that is, f(x + y) = f(x) + fey) holds for all x, y E X+). Then show
that there exists a unique linear functional g on X such that g(x) = f(x) holds
for all x E x+.
Solution. Note first that if x 2: y 2: 0 holds, then
f(x) = f(y + (x - y)) = fey) + f(x - y) 2: fey)·
Also, the arguments of the proof of Lemma 18.7 show that f(rx) = rf(x) holds
for all x E X+ and all rational numbers r 2: o.
260 Chapter 5: NORMED SPACES AND Lp·SPACES
Now, let Cl > 0 and x 2: O. Pick two sequences {r ll } and {til} of rational
numbers with 0 ::: rn t Cl and til -J, Cl. Then, the inequality rnx ::: ClX ::: tllX
implies
from which it follows that Clf(x) = f(ClX) holds.

Now, define g: X ~ JR- by
Note that if x = y - z holds with y, z E X+, then the relation x+ + z = y + x-,

coupled with the additivity of f on X+, shows that f(x+)+ fez) = f(y)+ f(x-).
That is,
g(x) = f(x+) - f(x-) = fey) - fez).
In particular, for x, Y E X we have
g(x + y) = g(x+ + y+ - (X- + y-») = f(x+ + y+) - f(x- + y-)

= f(x+) + f(y+) - f(x-) - f(y-)
= [J(x+) - f(x-)] + [f(y+) - f(y-)]
= g(x) + g(y).
Moreover, for Cl > 0 we have
and if Cl < 0, then
g(ClX) = -Clg(-X) = -Cl[g(X- - x+)] = -Cl[J(X-) - f(x+)] = Clg(X).
Thus, g is a linear functional on X, which is clearly a unique extension of f.
Problem 30.2. A vector lattice is called order complete ifevelY nonempty subset
that is boundedfrom above has a least upper bound (also called the supremum of
the set).
Show that if X is a vector lattice, then its order dual X- is an order complete
vector lattice.
Solution. Let A be a nonempty subset of X- that is bounded from above by

some g E X~ . By replacing A with the set (g - f: f E A}, we can assume that
A S;; X:;:. Let B denote the collection of all finite suprema of A, i.e., fEB if
and only if there exist fl, ... , fn E A with .f = V;'=I fj. Clearly, f ::: g also
holds for all fEB. Next, define h: X+ -+ R+ by
h(x) = sup{f(x): fEB}
for each x E X+. Clearly, 0 ::: h(x) ::: g(x) holds.

Let x, y E X+. Since f(x + y) = f(x) + f(y) ::: h(x) + h(y) holds for all
fEB, we see that
h(x + y) ::: h(x) + h(y).

On the other hand, given £ > 0 choose fl, h E B such that h(x) - £ < fl (x)
and h(y) - £ < h(y). Taking into account that fl V 12 E B, we see that
h(x) + h(y) - 2£ ::: fl (x) + h(y) ::: fl v h(x) + fl v 12(y)

= fl v h(x + y) ::: h(x + y)
holds, for all £ > O. Thus,
h(x) + h(y) ::: h(x + y)

also holds, and so h(x + y) = h(x) + h(y).
By the preceding problem, h extends uniquely to a positive linear functional.
Clearly, f ::: h holds for all f E A. On the other hand, if f ::: 41 holds for all
f E A, then f ::: 41 also holds for all fEB. This easily implies h ::: 41. That
is, h = sup A holds in X~.
Problem 30.3. Show that the collection of all boulldedfunctions on [0, 1] is an

ideal ofR[D,l]. Also, show that C[O, 1] is a vector sublattice ofR[D,I] but not an
ideal.
Solution. Let f: [0, 1] -+ R be a bounded function. If If(x)1 ::: M holds for

all x E [0, 1] and g E R[D,I] satisfies Igl ::: If I, then Ig(x)1 ::: M also holds for
all x E [0, 1]. This implies that the space of all bounded functions is an ideal of
R[D,1l.
The function X(D, t) is not a continuous function and satisfies 0 ::: X(D, ~) ::: 1,
where 1 denotes the constant function one on [0, 1]. Hence, C[O, 1] is not an
ideal of R [0,1] •
Problem 30.4. Let X be a vector lattice. Show that a norm II . II on X is a lattice

norm if and only if it satisfies the following two properties:
a. If 0::::: x ::::: y, then IIxll ::::: lIyll, and
b. IIx II = II Ix III holds for all x EX.
Solution. Assume that II . II is a lattice nonn. Clearly, 0 ::::: x ::::: y implies

IIxli ::::: lIyll. Also, Ixl = Ilxll holds, and so IIxli ::::: IIlxlll ::::: IIxll·
Conversely, assume (a) and (b) to be true. If Ixl ::::: Iyl, then
IIxli = IIlxlll ::::: Illylll = lIyll
so that II . II is a lattice nonn.
Problem 30.5. Show that in a normed vector lattice X its positive cone X+ is a
closed set.
Solution. From Theorem 30.1(3) we see that
This implies that the function x 1--+ x- from X into X is (uniformly) continu-
ous. Thus, X+ = {x E X: x- = O} is a closed set.
Problem 30.6. Let X be a normed vector lattice. Assume that {x n} is a sequence

of X such that Xn ::::: Xn+1 holds for all n. Show that if lim Xn = x holds in X,
then the vector x is the least upper bound of the sequence {x n} in X. In symbols,
Xn t x holds.
Solution. Assume that {x n} satisfies Xn ::::: Xn+1 for each 11 and limxn = x.
Then, x n+p - Xn ?: 0 holds for all n and all p and limp-Hlo(xn+p - xn) = x - Xn.
Since (by Problem 30.5) the positive cone X+ is closed, we see that x - Xn ?: 0,
or x ?: Xn for each n. This shows that x is an upper bound for the sequence {x n}.
To see that x is the least upper bound for the sequence {xn }, assume that y ?: Xn
holds for each n. So, y - Xn ?: 0 holds for all nand lim(y - xn) = Y - x. Using
once more that X+ is closed, we get y - x E X+. That is, y - x ?: 0, or y ?: x.
Therefore, x = sup{xn}, or Xn t x holds true, as desired.
Problem 30.7. Assume that Xn -+ x holds in a Banach lattice and let {En} be a
sequence of strictly positive real numbers. Show that there exists a subsequence
{Xk n} of{xn} and some positive vector u such that IXkn -xl::::: EnU holdsfor each n.
Solution. An easy inductive argument guarantees the existence of a subsequence

{Xk,,} of {xn} satisfying IIxk" - x II ::::: En2-n for each n. Now, notice that the series of
positive vectors I::l (EIl)-llxkn -xl is absolutely summable. Since X is a Banach

space, u = I::I (En)-IIXkn -xl exists inX. Now, a glance at Problem 30.6 shows
that (En)-llxkn - xl :5 u for each n. Thus, IXkn - xl :5 E/ZU holds for each n, as
desired.
Problem 30.S. Let T: X -+ Y be a positive operator between two normed vector

lattices., i.e, x 2: 0 in X implies T x 2: 0 in Y. If X is a Banach lattice, then show
that T is continuous.
Solution. Let T: X ---+ Y be a positive operator, where X is a Banach lattice

and Y is a normed vector lattice. Assume by way of contradiction that T is not
continuous. Then, there exist a sequence {xn} of X and some E > 0 satisfying
Xn -+ 0 and IITxnll 2: E for each n. By Problem 30.7 there exists a subsequence
{YIl}of{xn}andsomeu E X+ satisfying IYnl ::: ~uforeachn. Now,noticethatthe
positivityofTimplieslTYnl :5TIYnl:5 ~Tuforeachn,andsoIlTYnll:5 ~IITull
for each n. Since ~ II T u II -+ 0, it follows that II TYn II -+ 0 contrary to II T Yn II > E
for each n. Thus, T is a continuous operator.
Problem 30.9. Show that any two complete lattice norms on a vector lattice
must be equivalent.
Solution. If II . III and II . 112 are two complete lattice norms on a vector lattice
X, then, by Problem 30.8, the identity operator I: (X, II . II d ---+ (X, II . 112) is a
homeomorphism. That is, II . III and II . 112 are two equivalent norms.
Problem 30.10. The averaging operator A: £00 -+ £00 is defined by
A(x) _- ( Xl" XI + X2 XI + X2 + X3 , ... , XI + X2 + ... + Xn , .... )

2 3 n
for each x = (XI, X2, ... ) E £00' Establish thefollowing:

a. A is a positive operat01:
b. A is a continuous operator.
c. The vector space
is a closed subspace of £00' Is V = too?
·
S oIu t Ion. ( a) If X - (XI, X2, . .. ) >
_ 0 , th en Xi >
_ 0 lor
C each I. and so x, +X'+'+X
-n n
2:
o for each n. This implies A(x) 2: 0, and so A is a positive operator.
(b) By Problem 30.8 every positive operator on a Banach lattice is continuous.

Therefore, A (as a positive operator) is continuous.
(C) We know from Problem 29.5 that the vector space of all convergent sequences
c = {x = (XI, xi, ... ) E too: lim Xn exists in 1R}

IJ-+-OO
is a closed subspace oU oo . Clearly, V = A -I (c). Since A is continuous, the latter

guarantees that V is a closed: subspace of too.
There are bounded sequences having divergent sequences of averages. Here is
an example:
(1, -1, 1, 1, -1, -1,1,1,1,1,1, 1, -1, -1, -1, -1, -1, -1, ... ).
Hence, V is a proper closed subspace of too.
Problem 30.11. This problem shows that for a normed vector lattice X its norm
dual X* may be a proper ideal of its order dual X-. Let X be the collection of all
sequences {xn} such that Xn = 0 for all but a finite number of terms (depending
on the sequence). Show that:
a. X is afunction space.
b. X equipped with the sup norm is a normed vector lattice, but not a Banach
lattice.
c. Iff: X -+ 1R is defined by f(x) = L:lnxn for each x = {Xn} E X, then
f is a positive linear functional on X that is not continuous.
Solution. (a) Routine.

(b) If Xn = (1,4, t, ... , ...),
~, then {xn} is Cauchy sequence of X that does
not converge in X.
(c) Clearly, f is a positive linear functional. If en denotes the sequence whose nth
component is one and every other zero, then lien 1100 = 1 and n = f(e n ) ::: II til·
That is, II til = 00.
Problem 30.12. Determine the norm completion of the normed vector lattice of
the preceding problem.
Solution. Let
Co = {x = (XI,X2, ... ) E too: lim Xn = OJ.

n->oo
Clearly, Co is a vector sublattice of too. Also, it is not difficult to see that Co is
a closed subspace, and so Co is a Banach lattice (with the sup norm). We claim
that Co is the norm completion of the normed vector lattice X of the preceding
problem.
To see this, note first that X is a vector sublattice of Co. Now, let x =
(Xl,X2, ... ) E Co and let 8 > O. Choose some 12 with IXk I < 8 for all k :::: 12,
and note that the element y = (Xl, ... , x n , 0, 0, ... ) E X satisfies IIx - y 1100 :::: 8.
Thus, X is dense in Co, and our claim follow.s.
Problem 30.13. Let C c(X) be the normed vector lattice-with the sup norm-of
all continuous real-valued functions on a Hausdorff locally compact topological
space X. Determine the norm completion of Cc(X).
Solution. Consider the vector space of functions
co(X) = {J E C(X): V 8> 03 K compact with If(x)1 < 8 for X fj K}.
Clearly, co(X) is a vector sublattice of B(X). We claim that co(X) is a closed

subspace. To see this, let [fn} s;:; co(X) satisfy ~, ----+ f in B(X), and let 8 > O.
By Theorem 9.2, f E C(X). Pick some n with IIf - fnlloo < 8, and then select
a compact set K with Ifn(x)1 < 8 for x fj K. Thus,
If(X)1 :::: Ii(x) - fn(X)1 + Iin(X)1 < 28
holds for each X fj K, and so f E co(X). Therefore, co(X) (with the sup norm)
is a Banach lattice.
Clearly, Cc(X) is a vector sublattice of co(X), and we claim that Cc(X) is
dense in co(X). To see this, let f E co(X) and let 8 > O. Choose some compact
set K with If(x)1 < 8 for all x fj K, and then use Theorem 10.8 to pick some
g E Cc(X) with g(x) =
1 for all X E K and 0:::: g(x) :::: 1 for x fj K. Then,
fg E Cc(X) and IIfg - flloo :::: 8 holds, proving that Cc(X) = co(X). Thus,
co(X) is the norm completion of Cc(X).
Problem 30.14. Let X and Y be two vector lattices, and let T: X -+ Y be a

linear operator. Show that the following statements are equivalent:
a. T(x V y) = T(x) v T(y) holds for all x, y EX.
b. T(x 1\ y) = T(x) 1\ T(y) holds for all x, y EX.
c. T(x) 1\ T(y) = 0 holds in Y whenever x 1\ y = 0 holds in X.
d. IT(x)1 = T(lxl) holds for all x EX.
(A linear operator T that satisfies the preceding equivalent statements is referred
to as a lattice homomorphism.)
Solution. (1) =:::} (2) From the identity (a) of Problem 9.1, we get
T (x 1\ y) = T (x + y - x V y) = T (x) + T (y) - T (x V y)
= T(x) + T(y) - T(x) V T(y) = T(x) 1\ T(y).
(2) ~ (3) If X 1\ Y = 0, then

T(x) 1\ T(y) = T(x 1\ y) = T(O) = o.
(3) ~ (4) Using the identity (e) of Problem 9.1, we see that
IT(x)1 = IT(x+) - T(x-)I = T(x+) v T(x-) - T(x+) 1\ T([)

= T(x+) v T(x-) = T(x+) + T(x-) - T(x+) 1\ T(x-)
= T(x+) + T(x-) = T(x+ + x-) = T(lxJ).
(4) ~ (l) From the identity (f) of Problem 9.1, we get
T(x v y) = T(Hx + y + Ix - yin = HT(x) + T(y) + T(lx - yJ)]

= HT(x) + T(y) + IT(x) - T(Y)I] = T(x) v T(y).
Problem 30.15. Let loo be the Banach lattice of all bounded real sequences;
that is, loo = B(JN), and let {r}, 1'2, ... } be an enumeration of the rational num-
bers of [0,1]. Show that the mapping T: C[O, 1] -+ .too defined by T(f) =
(f(r}), f(r2), ... ) is a lattice isometry that is not onto.
Solution. Clearly, T is a linear operator. Let f E C[O, 1]. Since f is a

continuous function and the set of all rational numbers of [0, 1] is a dense set, it
easily follows that
11T(f)1100 = sup{lf(r/l)I: n = 1,2, ... }

= sup{lf(x)l: x E [0, I]} = Ilfll x '
In addition, note that
which shows that T is a lattice isometry.

To see that T is not onto, note that
T(f) f= (0, 1,0,1, ... )
holds for each f E C[O, 1].

Problem 30.16. Let X be a normed vector lattice. Then show that an element
x E X satisfies x 2: 0 if and only if f(x) 2: 0 holds for each continuous positive
linear functional f on X.
Solution. If x 2: 0 holds, then clearly f(x) 2: 0 also holds for each 0 :5 f E

x-.
For the converse, assume that x is fixed and satisfies f (x ) 2: 0 for each
f E X.:;.. Let 0 :5 f E X- be fixed. Since -g(x) :5 0 holds for all 0:5 g :5 f,
it follows from Theorem 30.3 that
0:5 f(x-) = sup{-g(x): g EX-and 0:5 g:5 f} :5 O.
That is, f(x-) = 0 holds for all 0 :5 f E X~ and consequently fex-) = 0 for all
f E X*. From Theorem 29.4, we see that x- = O. Thus,x = x+-x- = x+ 2: 0,
as required.
Problem 30.17. Let X be a Banach lattice. If 0 :5 x EX, then show that
IIxll = sup{f(x): 0:5 f E X* and 1If11 = 1}.

Solution. Let x 2: O. In view of the inequality If(x)1 :5 Ifl(x) :5 1If11 . IIxll,
we have
Ilxll = sup{lf(x)l: f IItIl = I}

E X* and
:5 sup{lfl(x): f E X* and 1If11 = I}
= sup{f(x): 0:5 f E X* and 1If11 = I} :5 IIxll,
and the conclusion follows.
Problem 30.18. Assume that rp: [0,1] -T 1R is a strictly monotone continu-

ous function and that T: C[O, 1] -T C[O, 1] is a continuous linear operator. If
T(rpf) = rpT(f) holds for each f E C[O, 1] (where rpf denotes the pointwise
product ofrp and f). Show that there exists a uniquefunction h E C [0, 1] satisfYing
T(f) = hf for all f E C[O, 1].
Solution. Taking f = 1, the constant one function, and letting h = T1, we

obtain T(rp) = hrp, and by induction T(rpn) = hrpn for each n 2: O. Hence, by the
linearity of T, we see that
T(P(rp)) = hP(rp)
for each polynomial P of one variable. Since the function cp is strictly increasing,
the algebra A = (P(cp): P polynomial} separates the points and contains the
constant function 1. Consequently, by the Stone-Weierstrass Theorem 11.5, A is
dense in C[O, 1]. From (*), it easily follows that T(f) = hi for each I E C[O, 1].
Problem 30.19. II I E C[O, 1], then the polynomials
n
BII(x) =L G)/(~)xk(1 - x)"-k,
k=O
where G)
is the binomial coefficient defined by G)
= k!(I;'~k)!' are knm'l'n as the
Bernstein polynomials 01 I.
Show that if I E C[O, 1], then the sequence {Bn} 01 Bernstein polynomials 01
I converges uniformly to I·
Solution. Let {Til} be the sequence of positive operators from C[O, 1] into
C[O, 1] defined by
n
Tnl(t) = L G)/(~)tk(1 - t)"-k
k=O
for all I E C[O, 1] and each t E [0,1]. We must show that
lim IITnl - 11100 = 0
holds for each I E C[O, 1]. By Korovkin's Theorem 30.13, it suffices to establish
that lim IITnl - 11100 = 0 holds for 1= 1, x, and x 2 •
To do this, we need some elementary identities. First note that by the binomial
theorem
11
L (Z)t k(1 - t)"-k = [t + (1 - t)f =1

k=O
holds for all t. Differentiating (*), we get
n .
L (~)[ktk-l(1 - t)n-k - (n - k)t k(1 - t)n-k-1J
k=O
n
= L G)t k- 1(1 - t)"-k-l(k - nt) = O.
k=O
Multiplication by t(1 - t) yields
II
L G)tk(l - t)"-k(k - nt) = 0,

k=O .
and by using (*), we see that
II
L G)~ t k (1 - t)n-k = t.
k=O
Differentiating (**) yields
II
L (%H t k- 1(1 - tt-k-1(k - nt) = 1,

k=O
and multiplying by t(1 - t), we get
II
L GH t k(1 - t/- II (k - nt) = t(1 - t).

k=O
That is,
n n
L G)(~)2l(1 - ti- II
- t L GH t k(1 - t)k-II = 1(1,;1),
k=O k=O
and by taking into account (**), we see that
The identities (*), (**), and (* * *) can be rewritten as follows:
Til 1 =1, TIIX = x,. and [ 2 - x 2]( t ) = -

Tllx 1(1-1)
n- .
Now, note that these identities readily imply that lim II Til I
I = 1, x, and x2.
- !lloo = °
holds for
Problem 30.20. Let T: C[O, 1] -+ C[O, 1] be a positive operator. Show that if

T I = I holds true when I equals 1, x, and X2, then T is the identity operator
(that is, T I = I holds lor each I E C[O, 1]).
Solution. Foreach n,let Tn = T. Clearly, lim Till = I holds in C[O, 1] when

I = 1, x, and x 2 • By Korovkin's Theorem 30.13, we have T I = lim Til I = I
for each I E C[O, 1].
Problem 30.21 (Korovkin). Let {Til} be a sequence 01 positive operators Irom

C[O, 1] into C[O, 1] satisfying TnI = 1. Iithere exists some c E [0,1] such that
°
lim Tn g = holds lor the function g(t) = (t - c?, then show that lim Tn I =
l(c)·1 holds lor all IE C[O, 1].
Solution. Let I E C[O, 1] and let 8 > 0. It suffices to show that there exist
constants C 1 and C 2 such that
holds for all 11.

Set M = 11/1100' By the continuity of I at the point c there exists some 0 > 0
such that -8 < l(t) - I(c) < 8 holds whenever t E [0, 1] satisfies It - cl < O.
Next, observe that
-8 - 2M ( t
"]2 - c)2 :5 I(t) - I(c) :5 8 2M ( t - c )2
+ 12 (a)
holds for all t E [0, 1]. (To see this, repeat the arguments in the proof of Theo-
rem 30.13.) Since each Tn is positive and linear, it follows from (a) that
Put C = 28"! , and note that
Consequently,
jTnl - I(c) . Ij :5 jT;J - I(c)·Tnlj + j/(c)j.jTnI-Ij

:5 81 + (8 + I/(c)l)jT I- Ij + CTng,
ll
and so
Section 31: Lp-SPACES 271
31. Lp·SPACES
Problem 31.1. Let f E L p(p,), and let E > O. Show that
p,*({x EX: If(x)1 2: E)) s.. E- P ! Ifl Pdp,.
Solution. Consider the measurable set E = (x EX: If (x) I 2: s}, and note that
E = {x E X: If(x)iP 2: sP}. Thus,
!IfI Pd P,2: !XElfiPdP,2:! sPXEdp, = sPp,*(E).
Problem 31.2. Let {ft,} be a sequence of some L p(p,)-space with 1 S. P < 00.
Show that if lim IIfn - flip = 0 holds in Lp(p,), then {ft,} converges in measure
to f.
Solution. From the preceding problem, we see that
holds. Clearly, this inequality shows that fn ~ f holds whenever lim II fn -

flip =0.
Problem 31.3. Let (X, S, p,) be a measure space and consider the set
E = {XA: A E AIL with p,*(A) < oo}.

Show that E is a closed subset of L 1(p,) (and hence, a complete metric space in
its own right). Use this conclusion and the identity
p,(AflB) = !IXA - XBI dp, = IIxA - XBIII
to provide an alternate solution to Problem 14.12(c).
Solution. Assume that {XA,,} is a sequence of E such that J XAn - f d p, -+ 0 I I

holds for some f ELI (p,). By Lemma 31.6 there exists a subsequence {XAkn} of
{XA n} such that XAkn -+ f a.e. This implies (how?) that f = XA a.e. for some
A E AIL with p,*(A) < 00. Thus, fEE and so E is a closed subset of LI (p,).
Problem 31.4. Show that equality holds in the inequality
albl-t ::; ta + (1 - t)b, 0 < t < 1; a ~ 0; b ~ 0
if and ollly if a = b. Use this to show that if I E L p(Jl) and g E Lq(Jl), where
~ < p, q .< 00 and ~ + ~ = I, then J Ilgl dJl = IIIIIp' IIgll q holds ifand only
ifthere eXist two constants C 1 and C 2 (not both zero) such that Cd/l P = C21gl q
holds.
Solution. Clearly, if a = b ~ 0, then a lbl- 1 = ta + (1 - t)b = a holds.

For the converse, let alb 1- 1 = ta + (1 - t)b hold for some a, b > O. Put
y = ~, and rewrite the given equality as 1 - t + ty - i = O. Since the function
I(x) = I-t +tx _Xl for x ~ 0 (and some fixed 0 < t < 1) attains its minimum
when x = 1 (see the proof of Lemma 31.2), it follows that y = ~ = 1, and so
=
a = b. Thus, db 1- 1 ta + (1- t)b holds if and only if a b. =
For the second part, assume first that there exist two constants C 1 and C 2
(which are not both zero) such that Cd/l P C2lgl q. We can assume C 1 > 0
=
and C 2 ~ O. Then, we have
jllgldJl = j (~)~lgl~lgldJl = (g~)~ jlglq dJl

= [j(g~)lglqdJl]~ ·[jlglqdJl]~
= (ji/IP dJl)* . (jlglq dJl)~ = II/IIp' IIgllq·
For the converse, assume J Ilgl dJl = IIfllp . IIgl/ q. If either I or g is zero,
then the conclusion is trivial. (If I = 0, then put C 1 = 1 and C 2 = 0.) So, we
..J. 0 'T'~1."
..J. 0
can assume I Tang d T 1
. HU\.mg t = Ii' a = (If(X)I)P
IIfllp , and b = (1.iL<.:Ul)q
IIgllq ,
the inequality alb - ::; ta + (1 - t)b gives
1 1
Integrating (and using our hypothesis), we get
0<
-
j[l(l.&ll)P
llfilp
P
+ 1(18(X)I)q
IIgllqq
_ IIIf(X)8(X)I]dJl(X)
IIg IIqflip'
- 1 +1 _ J If(x)g(xlldJ1.(x) - 1 - 1 - 0
- P q IIfllp'lIgll q - -.
Consequently,
If(x)g(x)1 _ .!.(If(X)I)P + 1 (lg(XlI)q
IIfllp·lIglI" - P IIfllp q IIglI"
holds for almost all x, and by the first part of the problem, we see that (I{;-~~I) P =
( lg(X)I)q
111111'1
holds so that
,
holds for almost all x, as required.
Problem 31.5. Assume that f.L*(X) = I and 0 < p < q :::: 00. If f is in Lq(f.L),
then show that II fII P :::: II fIIq holds.
Solution. Assume f.L*(X) = I and 0 1, and then choose s > I so that .!.r + .!.S = 1 holds. Since
Ifl P E Lr(f.L) and 1 E Ls(f.L), it follows from Holder's inequality that
(lIfllp)P =j \f\P df.L =j \f\P ·1 df.L :::: (j \f\pr df.L): . (j IS df.L) ~
= (j\f\pr df.L): = (j\f\q df.L)~ = (lIfllq)P.

Consequently, IIfllp :::: IIfllq holds.
If q = 00, then
also holds in this case.
Problem 31.6. Let f ELI (f.L) n Loo(f.L). Then show that

a. f E Lp(f.L)for each I < p < 00.
b. If f.L*(X) < 00, then lim p.... oo II flip = IIflloo holds.
Solution. (a) If M = II f 1100, then the inequality
shows that f E L p(f.L) for each 1 1 for each
n and lim PII = 00. From the inequality
it follows that
Let 0 < e < M. Then, the measurable set
E = {x EX: I/(x)1 :::: 1111100 - e}

satisfies f*(E) > O. From (1111100 - e)P"XE :::: I/IP", we see that (1111100 - e)
[JL*(E)]Pn :::: II/lI pn , and so 1111100 - e :::: lim inf IIIIIPn holds for all 0 < e < M.
That is,
1111100 :::: lim inf 11111 Pn'

Thus, lim sup II I II p" :::: II I II 00 :::: lim inf II I II p" holds. This shows that lim II I II Pn =
11/1100, and from this it follows that limp-+oo II/lIp = 11/1100'
Problem 31.7. Let IE L2[0, 1] satisfy 111112 1 and = Jd

I(x)d)"(x) 2: a > O.
AlsoJor each f3 E lR let EfJ =
{x E [0,1]: I(x) 2: f3}. 110 < f3 < a, show that
'A(EfJ) 2: (f3 - a)2.
(This inequality is known in the literature as the Paley-Zygmund Lemma.)
Solution. Assume I E, L2[0, 1] satisfies the stated properties and let 0 < f3 < a.
Then, note that
and so, from Holder's inequality, it follows that
O<a-f3:::: l i/
(X)d'A(X)-f3 = l i
[/(X)-f3]d'A(X):::: foi/(X)XEP(X)d)"(X)
1 1
:::: 11/112' ['A(EfJ)J2 = ['A(E fJ )]2·
This implies 'A(EfJ) 2: (a - f3)2.

Problem 31.8. Show thatior 1 :s p < 00 each.e p is a separable Banach lattice.
Solution. Let en denote the sequence whose nth component is one and every
other is zero. Also, denote by E the set of all finite linear combinations of
{eI. e2, ... } with rational coefficients. Clearly, E is a countable set, which we
claim is also dense in .e p whenever 1 :s p < 00.
To see this, let x = (XI. X2, ... ) E .e p (1 :s p < 00), and let 8 > O. Fix
some natural number n with 2::::/1+1 Ixd P < Ef.
Then, pick rational num-
bers rl, r2, ... , rn with 2::7=1 Ix; - r;Jp < Ef'
and note that the element a =
(rl, 1"2,.·., r/l, 0, 0, .. .)=rlel + r2e2+" ·+r/le/l E E satisfies
That is, the countable set E is dense in .e p, and so each .e p (1 :s p < 00) is a
separable Banach lattice.
Problem 31.9. Show that .eeo is not separable.
Solution. Let E = {Xl, X2, ... } be a countable subset of loa. Write Xn =

(x~, x 2, ... ) for each n. Now, define
0 if Ix~ I 2: 1
Y/I = {2 if Ix~ I < 1 .
Clearly, Y = (YI, Y2, ... ) E .e oo and
holds for each n. Thus, B(y, 1) n E = 0, and this shows that no countable subset
of .eoo can be dense.
An alternate way of proving that .e oo is not separable is as follows: Consider
the set F of all sequences whose coordinates are zero or one. By Problem 2.8,
the set F is uncountable, and it is not difficult to see that IIx - Ylloo = 1 holds
for each pair x, Y E F with x :f:. y. It follows that {B(x,1): x E F} is an
uncountable collection of pairwise disjoint open balls. This easily implies that
every dense subset of .eoo must be uncountable.
Problem 31.10. Show that Loo([O, 1]) (with the Lebesgue measure) is not sepa-
rable.
°
Solution. Write fx = X[O,x), < x < 1. Since IIfx - 11'1100 = 1 holds when-
ever x =f. y, it follows that (BUx, 1): x E (0, I)} is an uncountable collection of
pairwise disjoint open balls of Loo([O, 1]). This easily implies that every dense
subset of Loo([O, 1]) must be uncountable, and so Loo([O, 1]) is not a separable
Banach lattice.
Problem 31.11. Let X be a Hausdorfflocally compact topological space, andfix

a point a EX. Let f.i be the 11!easure on X defined on all subsets of X by f.i(A) 1=
if a E A alld f.i(A) = °
if a ¢ A. III other words, f.i is the Dirac measure (see
Example 13.4). Show that f.i is a regular Borel measure and that Supp f.i =
{a}.
Solution. The regularity of f.i will be established first.

1) Clearly, f.i(A) ::: 1 holds for each A ~ X.
2) Let B ~ X. If a E B, then
1 = f.i(B) ::: inf{f.i(O): 0 open and B ~ O} ::: f.i(X) = 1.

On the other hand, if a ¢ B, then use the open set X \ {a} to see that
°= f.i(B) ::: inf{f.i(O): 0 open and B ~ O} ::: f.1.(X \ (a}) = 0.

3) Let B ~ X. If a ¢ B, then each subset C of B satisfies f.i(C) = 0, and so
0= sup{f.i(K): K compact and K ~ B} ::: f.1.(B) = 0.
Now, if a E B, then using that {a} is a compact subset of B we see that
1 = f.i({a}) ::: sup{f.i(K): K compact and K ~ B} ::: f.i(B) = 1.

Thus, f.i is a regular Borel measure. Since f.i(X \ (a}) = 0, it is easy to see that
Supp f.i = {a} holds.
Problem 31.12. Ifg E C 1 [a, b] and f E L 1 [a, b], then

a. show that the function F: [a, b] -+ 1R defined by F(x) = J~t f(t)d)...(t) is
uniformly continuous, and
b. establish the following "Integration by Parts" formula:
1ar g(x)f(x)d)"'(x) = g(X)F(X)Lb - 1ar g'(x)F(x)dx.

Section 31: Lp·SPACES 277
Solution. (a) The uniform continuity of F follows immediately from Prob-

lem 22.6.
(b) Start by choosing some constant C > 0 such that Ig(x)1 ::: C and Ig'(x)1 ::: C
hold for each x E [a, b]. Now, by Theorem 25.3 there exists a sequence of
continuous functions {fll} satisfying lim Ja" If - fn I d"A = O. From Lemma 31.6,
we can suppose (by passing to a subsequence if necessary) that there exists some
function 0 ::: h E L, [a, b] satisfying Ifn I ::: h a.e. for each n and fll -+ f a.e. Let
FIl (x) = J~t fll (t) d t , and note that by the "standard" Integration by Parts Formula
we have
r g(x)fll(x)d"A(x) iar g(x)fn(x)dx

1a
=
b
= g(X)FIl(X)\" _ r" g'(x)FIl(x)dx.
a 1a
(*)
From Igflll ::: Ch E L,[a,b], gj" -+ gf a.e., and the Lebesgue Dominated
Convergence Theorem, it follows that
lim
,,-;00 1" a
g(x)fll(x)dx = 1"
a
g(x)f(x)d"A(x).
Likewise, the Lebesgue Dominated Convergence Theorem implies
FII(x) = 1 x
j,,(t)dt - + L' f(t)d"A(t) = F(x).
for each x E [a, b]. Observing that Ig' Fill::: C 1:h d"A and g' FII -+ g' F, the
Lebesgue Dominated Convergence Theorem once more yields
1ar g'(x)Fn(x)dx = Jar g'(x)F(x)dx.

b b
lim
Il~OO
Finally, letting n -+ 00 in (*), we obtain
1ar" g(x)f(x)d"A(x) = g(X)F(x)L-

" 1ar" g'(x)F(x)dx,
as desired.
Problem 31.13. Let f.,L be a regular Borel measure on JR.". Then show that
the collection of all real-valued functions on JR.II that are infinitely many times
differentiable is norm dense in L p(f.,L) for each 1 ::: p < 00.
Solution. Let S be the semiring consisting of the sets of the fonn n7=I[ai, hi).
By Theorem 15.10, the outer measure generated by (]R",S,J.l) agrees with I'- on
thea-algebra B of all Borel sets of ]R.n. Thus, what needs to be shown is that given
I = n7=I[ai, hi) and s > 0, there exists some Coo-function I with compact
support such that IIxl - flip < s.
To this end, let I = n;'=l [ai, hi) and let s > 0. The arguments of the first part
of the solution of Problem 25.6 show that there is a COO-function I:]R.II ---+ [0, 1]
satisfying J IXI - II dl'- < 2- ps p. Since IXI - II :::: 2 holds, it follows that
Ilxl - Illp = (/ixi - liP dl'-) ~= (/ixi - II P - 1 ·Ixi - II dl'-) *

:::: 2(/IXI-/ldl'-)~ 1
<2·T s=s,
Problem 31.14. Let (X, S, 1'-) he a measure space with I'-*(X) = 1. Assume that
In(!)
°
a function I ELI (I'-) satisfies I (x) 2: M > lor almost all x. Then show that
E L1(1'-) and that Jln(f)dl'- :::: In{j I dl'-) holds.
Solution. The function get) = t - 1 -lnt, t > 0, attains its minimum value
°
at t = 1. Thus, = g(1) :::: g(t) = t - 1 - Int holds for all t > 0, and so
In t :::: t - 1. Replacing t by +, the last inequality yields 1- +:::: In t. Therefore,
1 - 11<- In t -< t - 1
holds for each t > 0.

Since the function In x is continuous on (0, (0) and I is a measurable function,
it follows that In(f) is a measurable function. (See the solution of Problem 16.8.)
Replacing t by (J'~~ in (*), we see that
1 - .!!Llli.
f(x) -
< In(/(x)) -In(11111 1) -< IIflli
f(x) - 1
holds for almost all x. From our assumptions, it is easy to see that both functions
1 - %~') and (j~~ are integrable. Thus, from (**) and Theorem 22.6, it follows
that In(f) ELI (1'-).
Finally, integrating the right inequality of (**) (and taking into account that
I'-*(X) = 1), we see that
/ In(f)dl'- -In(lIlll1) :::: / dill dl'- - 1 = 0.

That is,
f In(f)dp, :s In(lIfIIl).= In(J f dp,)
holds, as required.
Problem 31.15. Theorem 31.7 states that: If 1 :s p < 00, f in Lp(p,), {!t,} S;
Lp(p,), fn ~ f a.e., and lim IIfnllp = II flip, then lim Ilfn - flip = O.
Show with an example that this theorem is false when p = 00.
Solution. Consider the sequence {fn} of Loo([O, 1]) defined by fll = X(~.ll'
Then fn ~ 1 a.e., and IIfnlloo = 1 ~ 1 = 1111100' However, IIfn -11100 = 1
holds for each n.
Problem 31.16. This exercise presents a necessary and sufficient condition for
the mapping g!---+ Fg from Loo(p,) into Li(p,) (defined by Fg(f) = J fgdp,)
to be an isometry.
a. Show that for each g E Loo(p,) the linear functional Fg(f) = J fg dp"
for fELl (p,), is a bounded linearfunctional on Ll (p,) such that IIFg II :s
IIglioo holds.
b. Consider a nonempty set X and p, the measure defined on every subset of
X by p,(0) = 0 and p,(A) = 00 if A '# O. Then show that Ll (p,) = {OJ
and Loo(p,) = B(X) [the boundedfunctions on Xl and conclude from this
that g E Loo(p,) satisfies IIFglI = IIglioo ifand only if-g = o.
c. Let us say that a measure space (X, S, p,) has the finite subset property
whenever every measurable set of infinite measure has a measurable subset
offinite positive measure.
Show that the linear mapping g!---+ Fg from Loo(p,) into Li(p,) is a
lattice isometry if and only if (X, S, p,) has the finite subset property.
Solution. (a) Let g E Loo(p,). Then, for each f E L1(p,), we have Ifgl :s
IIglioo' If I, and so
That is, Fg is a bounded linear functional on Ll (p,), and IIFg II :s IIglioo holds.
(b) Since every nonempty set has infinite measure, it is easy to see that there is
only one step function. Namely, the constant function zero. That is, L I (p,) = {OJ
holds. On the other hand, since every one-point set has infinite measure, each
equivalence class of Loo(p,) consists precisely of one function. This implies that
Loo(p,) = B(X).
Finally, note that in view of L!(JL) = {OJ, we must have Fg = 0 for each
g E Loo(JL). Thus, IIFg II= IIglioo holds if and only if IIglioo = 0 (i.e., if and only
if g = 0).
(c) Assume that a measure space (X, S, JL) has the finite subset property. Let
o < g E Loo(JL) and let 0 < e < IIglioo. The set
E =:= {x EX: Ig(x)l2: IIglioo - e}
is measurable and JL*(E) > 0 holds. By the finite subset property, there exists a
measurable set F with F £;:; E and 0 < JL*(F) < 00. Put 1= St~MF E LI(JL),
and note that II fill = 1. Therefore,
Since 0 < e < IIglioo is arbitrary, IIFgII 2: IIglioo holds. Now, using part (a), we
seethat IIFglI = IIglioo holds for all g E Loo(JL). Therefore,g 1----7 Fg is a lattice
isometry.
For the converse, assume that g 1----7 Fg is a lattice isometry, and let E be
a measurable set with JL*(E) = 00. Then g = XE E Loo(JL), and so IIFgII =
IIglioo = 1. Pick some 0::: I E LI(JL) with FgU) = I Ig dJL = IE I dJL > 4·
It is easy to see that there exists a step function 0 ::: ¢ ::: I XE with I ¢ d JL > 4.
From this, it easily follows that there exists a measurable set F £;:; E with 0 <
JL*(F) < 00.
Problem 31.17. Let (X, S, JL) be a measure space. Assume that there exist
measurable sets E I, ... , En such that 0 < JL(E i ) < 00 lor 1 ::: i ::: n, X =
U7=IEi, and each Ei does not contain any proper nonempty measurable set.
Then show that L~(JL) = LI(JL); that is, show that g 1-+ Fg Irom LI(JL) to
L~(JL) is onto.
Solution. From our assumptions, we see that Ei n E j = 0 holds whenever

i :j:: j. For each 1 ::: i ::: n fix some Xi E Ei and note that JL*({Xi}) > o. If I is
a measurable function and ai = I(Xi), then the set I-I ({ad) n Ei is nonempty
and measurable. Thus, by our hypothesis, I-I ({ad) n Ei = Ei holds, and
therefore, I must be constant on each E i . In other words, I = :L7=1 I(Xi)XE;
holds for each measurable function I.
To see that g 1----7 Fg from L I (JL) to L~(JL) is onto, let F be an arbi-
trary functional in L~(JL). Put Ci = F(XE) for 1 ::: i ::: n, and then let g =
Section 3i: Lp-SPACES 281
Consequently,
n n
Fg(f) = Fg(L I(Xi)xEi) = L I(Xi)Fg(XEi)
i=1 i=1
n fl
= LI(Xi)F(XEi) = F(LI(Xi)xE,) = F(f)

i=1 i=1
holds for all I ELI (f.J,), and so F = Fg. That is, g f---* Fg is onto.
Problem 31.18. Let (X, S, f.J,) be a measure space, and let 0 < p < 1.
a. Show by a counterexample that II . lip is no longer a norm on L p(f.J,).
b. For each I, g E Lp(f.J,) let d(f, g) = III - glP df.J, = (III - gllp)P. Show
that d is a metric on L P(f.J,) and that L P(f.J,) equipped with d is a complete
metric space.
Solution. (a) Let 0 2 1- ~ = (4) *+ (4) ~ = I I I p + I g I p'

That is, II . II p does not satisfy the triangle inequality.
(b) If a > 0 and b > 0 (and 0 < p < 1), then
(a+b)P = (a+b)(a+by-l =a(a+b)P-l +b(a+b)P-l

:s a . a P- 1 + b· b P- 1 = a P + bP.
Thus, (a + b)P :s a P+ b P holds for each a 2: 0 and each b 2: O. This inequality

easily implies that d(f, g) = I II - glP df.J, is a metric on Lp(f.J,).
For the completeness, let {f,,} be a Cauchy sequence in the metric space
(L p(f.J,), d), where 0 < p < 1. By passing to a subsequence, we can assume
that I I/n+l - f"IP df.J, < 2- n holds for each n. We shall establish the existence
of some I E L P(f.J,) such that lim III" - !II P = O.
Set gl = 0 and gn = II1I + liz - III + ... + II,I - 1,1-11 for n 2: 2. Clearly,
0::: gil t and
holds for each n. By Levi's Theorem 22.8, there exists some g E Lp(f.J.) such
that 0 ::: gn t g a.e. From .
n+k n+k
Iln+k - In I = I;=n+1
L (/; - Ii-!) I::: ;=n+1
L II; - 1;-1 I = gn+k - gn,
it follows that Un} converges pointwise (a,e.) to some function I. Since IIn I =
IiI + 1:.~=2(f; - I;-dl ::: gn ::: g hold a.e., we see that III::: g a.e. also holds.
Therefore, IE Lp(f.J.). Now, note that II,I - II ::: 2g and lin - liP -+ 0 hold,
and so by the Lebesgue Dominated Convergence Theorem, we see that
Therefore, (L p(f.J.), d) is a complete metric space.
Problem 31.19. II (X, S, f.J.) is afinite measure space, then show that the vector
space 01 all step junctions is norm dense in Loo(f.J.).
Solution. Let I E Loo(f.J.) and let e > O. Choose some C > 0 such that
I/(x)1 < C holds for almost all x, and then pick a partition -C = aD < al <
... < an = C of [-C, C] with a; - a;_1 < e for each 1 ::: i ::: n. Let
E; = 1-I([a;_I,a;)), and note that (since f.J.*(X) < 00) the simple function
¢ = 1:.7=1 aiXE; is a step function satisfying II! - ¢1I00 ::: e.
Problem 31.20. II K is a compact subset 01 a metric space X, then show that

there exists a regular Borel measure f.J. on X such that Supp f.J. = K.
Solution. Let K be a compact subset of a metric space X. Pick a countable

dense subset {XI,X2,"'} of K (see Problem 7.2) and then for each n consider the
Dirac measure OXn supported at the point Xn (see Example 13.4). Now, consider
the measure f.J.: P(X) -+ [0, 1] defined by
00
f.J.(A) = L.Tnoxn(A) = L n
2- ,
n=1 neA
where A = {n E IN: XII E A}.

Clearly, Jk(X \ K) = O. On the other hand, if 0 is an open subset of X

satisfying 0 n K ::j:. 0, then for some n we have Xn E 0, and so Jk(O n K) ::::
2- n oxn (0 n K) = 2- n > O.
It remains to be shown that Jk is a regular Borel measure. To this end, let
A £; X be fixed. Note first that if Cn = {XI, ... , xn} n A £; A, then Cn is a finite
set (and hence, a compact set) and, moreover, Jk(C n) t Jk(A) holds. Therefore,
Jk(A) = sup{Jk(C): C compact and C £; A}.
In the other direction, note that if for each n we consider the open set
On = X \ {Xi: 1::::: i ::::: n and Xi ~ A},
then A £; On and Jk(On) -l- Jk(A) (why?). Therefore,
Jk(A) = inf{Jk(O): 0 open and A £; O}
also holds, proving that Jk is a regular Borel measure.
Problem 31.21. If {fn} is a norm bounded sequence of L 2(Jk), then show that
fnln -+ Oa.e.
Solution. Assume that a sequence {!,,} £; L2(Jk) satisfiesJCfn)2dJk::::: C for all

n, where C > 0 is a constant. Then,
f= f (~)2
n=1
dJk ::::: C f=,!r
n=1
< 00
holds. By the series version of Levi's Theorem 22.9, we know that the series
L::I (~)2 defines an integrable function. Therefore, ~ -+ 0 a.e. must hold.
Problem 31.22. Let (X, S, Jk) be a measure space such that Jk*(X) = 1. If
f, gEL I (Jk) are two positive functions satisfying f (x )g(x) :::: 1 for almost all X.
then show that
Solution. Note that the functions ,J1 and ..;g both belong to L2(Jk) and
satisfy .j f(x).jg(x) :::: 1 for almost all x. Applying Holder's inequality, we
see that
Squaring, we get (J f d fL) . (J g d fL) ::: 1.
Problem 31.23. a
Consider measure space (X, S, fL) with 1l*(X) = 1, and let
f, g E L2(1l). If Jf dfL = 0, then show that
Solution. Put ex = Jg dfL. Then, using Holder's inequality, we get
If fgdlll = If (Jg - exf) dfLl = If f(g - ex) dill
~ flfllg-exldfL~ (f f 2dfL)4 . (f(g-ex)2)4
4
= (f f2 dfL) (f g2dfL - 2ex f gdfL +ex 2 )t
= (f f 2dfL)
4
[fg2dll -2(fgdfL) (fgdll ) + (fgdfL rJ!
= (f f2 dfL) t [f g2 dfL - (f g dfL r J4,
and our inequality follows.
Problem 31.24. If two functions f, g E L 3(fL) satisfy
then show that g = If I a.e.
Solution. Let p = ~ and q = 3, and note that ~ +~ = 1. Clearly, f2 E

Lp(fL) = L~(fL), and since g E L3(fL), we see that f2g E L,(fL).
Section 31: Lp.SPACES 285
Now, using Holder's inequality, we obtain
I = If f2g dfL I :5 f f21g1 dfL :5 lit lip ·llgll q

= [f (2)~ dfL J~ '11g113 = (11/113)2 '11g113 = 1,
and so Jf21g I d fL = I f211 p . II g IIq = 1. By Problem 31.4, there exists a constant
C> 0 such that Clf 21P = Iglq, or Clfl 3 = Ig1 3. From IIfII3 = IIgll3 = 1, we
infer that C = 1, and so Ifl 3= Igl 3 holds. Therefore,
bigllfl = Igl a.e.
From the relation
and f2(lgl - g) 2: 0 a.e., we conclude that f2(lgl - g) = 0 a.e. Taking into

account (*), the latter easily implies that g = Igl = If I a.e. holds.
Problem 31.25. For afunction f ELI (fL)nL 2 (fL) establish the following prop·
erties:
a. f E Lp(fL)foreach 1:5 p:5 2, and
b. limp_l+ IIfllp = IIflll.
Solution. Let f E LI(fL) n L2(fL); we can assume that f(x) E lR for each
x E X. Consider the measurable set A = {x E X: If(x)1 2: I} and then define
the function g: X -+ lR by
2
(x) = {If(X)1 if x E A
g If(x)1 if x ¢. A ,
Le., g = f2XA + fXAc. From our hypothesis, we see that g E LI(fL).

(a) Let 1 :5 p :5 2. Then, the inequality
2
If(x)I P < {1/(X)1 if x E A= g(x),
- - If(x)1 if x ¢. A
implies f E L p(fL) for each 1 :5 P :5 2.

(b) Let a sequence {Pn} of the interval [1,2] satisfy PII -+ 1. From (*), we
see that Ifl Pn :s g holds for each n. Now, from Ifl Pn -+ If I a.e. and the
Lebesgue Dominated Convergence Theorem, we infer that
The preceding easily implies that lim IIfllp

p .... 1+
= IIfliI holds.
Problem 31.26. Assume that the positive real numbers aI, ... , an satisfy < °
aj < 1for each i and ",£7=laj= 1. If fl' ... , fn are positive integrable functions
on some measure space, then show that
a. Ifl f;2 ... J:n E LI(fL), and
b. Jffl f;2 ... J:n dfL :s (lIfllll)"1 (1112111),'2 ... (lIfn liI)"n .
Solution. We shall establish the result by using induction on n. For n = 1
the result is trivial. For n = 2, note that ffl E L -L (fL) and f;2 E L -L (fL).
Since (t r l (t;r l
+ = al + a2 = 1, it follows ff(;~ HOlder's inequalitY that
ffl f;2 ELI (fL) and that
f ffl f;2 dfL :s (f(Jfl) *r dfL (f(g2) * f2 dfL
= (II fl III) (1112 III)

UI U2 •
For the inductive argument, assume that the result is true for some n. Let
fl' ... , fn, fn+1 be n + 1 integrable positive functions and let aI, ... ,an, an+1
be positive constants such that ",£7:!:i aj = 1. Put a = ",£7=1 aj > 0, and note that
~ 2LL
",£7=1 ~ = 1. Now, by our induction hypothesis, we have f l" ... fn" E LI(fL).
Also, applying the case n = 2 for a and 1 - a = a n + I, we see that
and the induction is complete.

Problem 31.27. Let (X, S, {J..) be a measure space and let {An} be a sequence
of measurable sets satisfying 0 < J.l.*(An) < 00 for each n and lim J.l.*(AII) = O.
Fix 1 < p < 00 and let gil = [J.l.*(AII)r~ XA" (n = 1,2, ... ), where + ~ = 1. *
Prove that lim Jfgn dJ.l. = 0 for each f E ·Lp(J.l.).
Solution. Pick 1 < q < 00 such

.
that 1p + 1q = 1 and let f EL p(J.l.). Then, by
Holder's inequality, we have
If fgll dJ.l.1 = If (JXAn)gll dJ.l.1
:s (f ItXAn IP dJ.l.) *(f Iglll q

~
dJ.l.) = (inlflP dJ.l.) *.
From Problem 22.6, we know that lim h, Ifl PdJ.l. = 0, and therefore lim
Jfgll dJ.l. = 0 likewise holds.
Problem 31.28. Let (X, S, J.l.) be a measure space such that J.l.*(X) = 1. For
each 1 0 sLich that
J.l.*({x EX: If(x)1 > En 2: 8p

for each f E Ep.
Solution. Fix 0 < e < 1. For each f E Ep put
Ef = {x EX: If(x)1 > e} and Ff = X\Ef = {x EX: If(x)1 :s e}.
{ Itl dJ.l. = lltl

iEt x
dJ.l. - { If IdJ.l. 2: 1 -
iFt e.
Now, if 1 < q < 00 satisfies *+ ~ = 1, then Holder's inequality implies
r Ifld/L ::: (rlEI Ifl

lEI
P
d/L)* . (rlEI lq d/L)~
= (rlEI Ifl d/L)*[/L*(Ef)J~ ::: 2*[/L*(Ef)J~.
P
holds for each f E cP' and the desired conclusion follows.

Problem 31.29. Let (X, S, /L) be a measure space and let 1 ::: p < 00 and 0 <
rJ < p.
a. Show that the nonlinear function 1/1: L p(/L) -7 L r.(/L), where 1/I(f) =
Ifll), is norm continuous. . "
b. If 1" -7 f and gn -7 g hold in Lp(/L), then show that
Solution. (a) It should be clear that 1/1 maps indeed L p(fJ..) into L I!. (/L), and that
1/1 is nonlinear. Let 1" -7 f in Lp(/L) (i.e., let IIfn - flip -7 0)" and assume
by way of contradiction that 1/1(1,,) fr 1/I(j) in LI!.(fJ..). So, by passing to a
subsequence, we can assume that there exists some e "> 0 such that
Now, by passing to a subsequence again, we can assume that there exists some
function 0 ::: gEL p(/L) such that Ifn I ::: g /L-a.e. holds for each n and
I!.
fn ---+ f a.e.; see Lemma 31.6. Therefore, the relations 111,,11) - Ifll)l" :::
{igllJ+lfll))* L\(/L) and Ilfnll)-lfll)l* ---+ 0 a.e., coupled with the Lebesgue
E
e.
Dominated Convergence Theorem, imply 1111,,11) - Ifll)l" d/L ---+ 0, which
contradicts (**). Consequently, the nonlinear mapping 1/1 is norm continuous.
(b) Notice that the two nonlinear ,functions 1/1\: L p(/L) -7 L ,!!://L) and 1/12:
Lp(/L) -7 Le.(/L), defined by
"
are-by part (a)-both norm continuous. Therefore,
Illf"IP-ry -lflp-ryll.L. ~ 0 and Il lg" lry -Igirylll! ~ O.

p-q '1
Now, observe that (L f-:;; (J.L))* = L ~ (J.L) holds. Consequently, from the duality
(L.L (J.L), L e (J.L)), we see that·
P-q "
as claimed.
Problem 31.30. Let T: Lp(J.L) ~ Lp(J.L) be a continuous operator, where 1 <

p < 00, and let 0 ::: 1] ::: p. Show that:
a. If f E Lp(J.L), then Iflp-rylTflry E L1(J.L) and
b. Iffor some f E Lp(J.L) with IIfllp ::: 1 we have J Iflp-rylTflry dJ.L =

IITllry, then ITfl = IITIIlfl·
Solution. Assume T, 7], and f are as stated in the problem.

(a) If 1] =
0 or 1] = p, then the desired inequality is obvious. So, assume
o< 1] < p and consider the conjugate exponents
r =.-E...-
p-ry and s = (1 _ 1)-1
r = E..
ry
Since IflP-ry E Lr(J.L) and ITfiry E Ls(J.L), we see that Iflp-rylTflry belongs to
LI(J.L). Also, applying Holder's inequality with exponents r and s, we obtain
.
f Iflp-rylT flry dJ.L ::: [f (lfIP-ry) f-:;; dJ.L r~ [f (IT flry) ~ dJ.L] *
= (f Ifl PdJ.L f~ .(f ITfl PdJ.L) *
= (lIfllpy-ry(IITfllp)ry ::: (lIfllpy-ryIlTIIIJ(lIfllp)ry
= IITIIIJ(lIfllpY·
(b) Assume that some IE Lp(JL) with II/l1p S 1 satisfies
From Holder's inequality, we see that
IITIII) = ! IIIP-I)ITIII) dJL S 11111P-1)1Ir . II ITIII) lis
= (1I/I1 p)P-I)(II Tfl/p)1) S IITIII)·
Thus,J IIIP-I)ITIII) dJL = IIIIIP-I) ·IIITIII) t lis·

From Problem 31.4, there exists
°
a constant c :::: such that (ITIll)r = c(IIIP-l)f, or
Therefore, ITI I = A. II I holds for some A. :::: 0. This implies A. II! II P = II T fI/ P S
II T 1111 I II P and so A. S II T/I. Also, from
liT III) = ! IIIP-I)ITIII) dJL = ! IIIP-I)A.I) 1111) dJL S A.I),
we see that liT II s A.. Hence, A. = liT II, and so IT II = liT 11111.
Problem 31.31. Let (X, S, JL) be a measure space and let I E L p(JL) lor some
1 S p.< 00. Show that the junction g: [0, 00) -+ [0,00] defined by
g(t) = pt P- 1JL*(fx EX: I/(x)1 :::: tJ)

is Lebesgue integrable over [0,00) and that
! IIIPdJL= {
1[0,00)
g(t)dA.(t)=p ('''tP-1JL*({XEX: I/(x)l::::t})dt.
10
Solution. Let I E L p(JL); we shall assume that I (x) E 1R holds for each
x EX. Let T = [0,00) and consider the product measure space T x X. Also,
let
A = {(l,X) E T x X: ° S l S I/(x)I},
and note that (by Problem 26.8) the set A is JL x A.-measurable. Now, consider
Section 31: Lp.SPACES 291
the function h: T x X -+ [0,00) defined by
h(t,x)= { gtp-l ifOSfSlf(x)1 =ptp-1XA(t,X).

if t > 1f(x.) 1
In addition, we have
By Tonneli's Theorem (Theorem 26.7), the function h is integrable over T x X

and
Put Er = {x EX: If(x)1 2: t} and note that
r
Jx
h(t,x)dfL(X) = r
JE,
pt p- 1 df.L(x) = ptP-1f.L*(Er).
By Fubini's Theorem (Theorem 26.6), we know that the function
is integrable over [0,00) and from (*), we see that
That the Lebesgue integral frtp-lf.L*(Er)d)...(t) is also an improper Riemann

integral follows from the fact that the function t 1---+ f.L*(Er) is decreasing-and
hence continuous for all but at-most countably many t.
Problem 31.32. Let (X, S, f.L) be a measure space and let f: X -+ lR be a

measurable function. If f.L*({x EX: If(x)l2: tl) S e- r foral! t 2: 0, then show
that f E Lp(f.L) holds for each 1 S p < 00.
Solution. Let g(t) = f.L*({x E X: I/(x)1 ~ tl), t ~ 0. In view of Prob-

lem 31.31, we must show that fo""t P- 1g(t)d'A.(t) < 00. Since for each t ~ we
have 0::: g(t) ::: e- 1 , it suffices to establish that fo""t p - 1e- 1 dt < 00.
°
To see this, start by observing that by L'H6pital's Rule we have
lim tP-le-~ = lim lP~1 = 0.

1-+00 1-+00 e~
So, there exists some M >

Hence,
° satisfying °: : tP-le-~ < M for all t > 0.
0::: lOOtP-le-r dt = lOOtP-le-~e-~ dt
:: 1 00
M e-~ dt = 2M < 00,
as desired.
Problem 31.33. Consider the vector space olftmctions
E = {I: lR" -+ lRl I is a Coo -function with compact support and r I d'A. = O}.
JR"
Show thatlor each 1 < P < 00 the vector space E is dense in Lp(1Rn). Is E dense
in L 1(1R")?
Solution. We shall prove the result for the special case n = 1. The general case
(whose details can be completed as in Problem 25.3) is left for the reader. The
proof will be based upon the following property: If 1 0, h > 0,
and a positive integer n are given, then there exists a COO-function ¢: lR --7 1R
such that
1. Supp ¢ is compact and Supp ¢ ~ [12, 00);

2. 0::: ¢(x) ::: h for all x E 1R;
3. fR¢ d'A. = 1; and I
4. 1I¢lI p = (fR¢P d'A.)Ii < e.
To see this, assume 1 0, h > 0, and the positive integer n are
given. If k is an arbitrary positive integer, then (by Problem 25.3) there exists a
COO-function I: 1R --7 1R such that:
a. Supp I ~ [11, 11 + k + 2];

b. 0::: I(x) ::: 1 for each x E lR and I(x) = 1 for each x E [n + 1, n +

k + 1].
If c = JRI d)' > 0, then the COO-function .¢ = ~I satisfies Supp¢ £;; [n, n +
k + 2] £;; [n, 00), JR¢ d)' = 1, and (in view of c = JRI d)' 2: J,~:;k+l1 dx = k)
°: : ¢(x) ::: t for each x E lR. In addition, we have
I I
In view of 1 O. As in Problem 2S.S(b)
(how?), there exists a COO-function g with compact support such that III - g lip <
8. If m =
JRg d)' =
0, then gEE, and we are done. So, assume that m i= O.
Pick a positive integer n such that Supp g n [n, 00) =
0, and then (by the prior
discussion) pick a COO-function ¢ with compact support such that:
i. Supp ¢ £;; [n, 00);

ii.
iii.
°
¢(x) 2: for each x E lR;
JR¢ d)' = 1; and I
IV. 1I¢lI p = (JR¢P d). ) P < I!I .
Now, consider the function 1/1 = g - m¢, and note that 1/1 E E and
I I -1/I11 p= Ilu - g) + m¢llp

::: III - gllp + Imlll¢llp
< 8 + Imlll¢llp < 8 +8 = 28.
Therefore, E is dense in L p(lR).

The vector space E is not dense in L 1 (lR). For instance, consider the function
I = X[O,I] E LI (JR). If ¢ E LI (lR) satisfies JII - ¢I d). < then from i,
1 -1 ¢d)' = 1u - ¢)d)'::: 111 - ¢I d). < i,
it follows that JR¢d)' > 1 - i = i, and so ¢ ¢. E. This shows that E is not
dense in L 1(lR).
Problem 31.34. Let (0, (0) be equipped with the Lebesgue measure, and let
1 0.
Then show that T defines a one-to-one bounded linear operatorjrom Lp«O, (0))
into itself such that" T" = J!::i.
Solution. For simplicity, we shall write T I instead of T (f). Consider an
arbitrary function °: :I E Cc(O, (0»). Choose some M > so that
I(x) :::: M holds for all x > 0.
° °: :
If I = 1000
I(t)dt, then the function
M ifO<x::::1
g(x) = { ~ if x > I
belongs to L p (0, (0»). Since °: :

T I :::: g holds, we see that T I belongs to
Lp(O, (0»). Also, in view of the inequalities
it follows that
Now, integrating by parts and using Holder's inequality, we get
(IITllipY = foooU foX l(t)dtY dx = I~P fooo(foX l(t)dtY d(x l- P)

= l~p[xUfox/(t)dtYI: -p fooo/(x>(~foxI(t)dtr-ldX]
= -/:::r fooo/(X)[TI(x)y-1 dx
:::: -/:::r II/IIp(fooo[I:I(X)]q(P-I) dX) ~
= -/:::r II/IIp' (1ITlllp)~.

This easily implies that
holds for all IE Cc(O, (0»). In other words,
T: Cc(O, (0») -+ Lp(O, (0»)

defines a continuous operator such that II Til :s p~ I holds.
Since Cc(O, (0») is norm dense in Lp(O, (0») (Theorem 31.11), T has a
unique continuous (linear) extension T* to all of L p(0, (0») such that II T* II :s
-!=r holds. Our next objective is to show that T*/(x) = ~ Jot/(t)d'A(t) = TI(x)
holds for all IE Lp(O,oo») and all x > O.
To this end, let 0 :s ¢ be a step function. Choose some C > 0 satisfying
o :s ¢(x) :s C for all x > O. By Theorem 31.11, there exists a sequence (f,,} of
Cc(O, (0») with lim J If" - ¢IP d'A = O. We can assume that lim III (x) = ¢(x)
holds for almost all x (see Lemma 31.6). In view of
I Itll /\ C -
1/11 /\ C - ¢ = ¢ /\ C I :s Itll - C I,
replacing UII} by (I" /\ C}, we can assume that 0 :s
III (x) C holds for all:s
x > 0 and all n. Since lim IITf" - T*¢lIp =
0, we can also assume (by passing
to a subsequence) that T III (x) -+ T*¢(x) holds for almost all x. Next, observe
that for each fixed x > 0 we have ¢ E LI (0, x») and so, by the Lebesgue
Dominated Convergence Theorem, we see that .
T*¢(x) lim TIII(x) = lim ~ tf,,(t)d'A(t) = ~ t¢(t)d'A(t)

= 11-+00 Il-+OO x io x io
holds for almost all x. Now, let 0 :s I E L p (0, (0»). Choose a sequence (¢Il}
of step functions with 0 :s ¢Il t I. In view of
lim IIT*¢1l - T* flip = 0,
we can assume that T*¢II(x) -+ T* I(x) holds for almost all x. Taking into
account that for each fixed x > 0, we have I E Lp(O, x») S; LI (0, x»), the
Lebesgue Dominated Convergence Theorem implies
T* I(x) lim T*¢II(x) = lim 1 t

= n-+oo Il-+OO x io
¢1I(t) d'A(t) = ~ iot
.1
l(t) d'A(t)
holds for almost all x. Thus, T* =T holds.

Next, we shall show that II TIl = -/:r holds. We already know that II TIl :::: -/:r
holds. So, it must be established that II Til::: -/:r. To this end, let
_ lip ( -I - I)-I
XII P ifO<x<l
TlnCx) - I+n(p-l) { X-I
if x ::: 1
and so
This implies
. 1
I TII > -
~.
p_I+~
[1 + _1_]-;;
lI(p-l)
--+....L
p-I'
from which it follows that II Til 2! --s

also holds ..
Finally, we establish that T is 6ne-to-one. Assume that T I = 0 holds for
some I E Lp(CO, 00)). Then, Jo'/(t)d>..(t) = 0 holds for all x > O. Now,
by Problem 22.19, we infer that I = 0 a.e. holds, and so the operator T is
one-to-one.
CHAPTER 6 ______________
HILBERT SPACES
'32. INNERPRODUcrSPACES
Problem 32.1. Let CI, C2, ... ,CII be n (strictly) positive real numbers. Show
that the function of two variables (', .):JR.n x JR.II --+ JR., defined'by (x, Y) =
,\,11 .. .d
L..,i=1 ciXiYi, IS an lI1ner plO uct on
JR.II .
Solution. Notice that for all vectors x = (XI, ... ,XII)' Y = (YI,"" Yn) and
Z = (ZI, ••• , Z,.) in JR.n we have
n II II
(ax + {3y, z) = L Ci( axi+{3Yi)zi=a L Ci Xi Zi+{3 L CiYizi=a(x, zH{3(y, z),

i=1 i=1 i=1
IJ II
(x,y) = LC;X;Yi = LCiYiXi =(y,x), and

i=1 i=1
n
(x, x) =L cixl2: O.
i=1
Moreover, (x, x) = 2:7=1 cixf = 0 implies c;Xf = 0 for each i, and so (since
Ci > 0 for each i) Xi = 0 for each i, i.e., x = O. The above show the function (', .)
is an inner product on JR.n .
.»
Problem 32.2. Let (X, (', be a real inner product vector space with complex-
ification Xc' Show that the function (', .): Xc x Xc --+ C defined via the formula
(x + IY, XI + IYI) = (x, xd + (y, YI) + l[ (y, XI) - (x, YI)].
is an inner product on Xc' Also, show that the norm induced by the inner product
(', .) on Xc is given by
IIx + lyll = J(x, x) + (y, y) = (lIxll 2 + IIYII2)~.
297
298 Chapter 6: HD...BERT SPACES
Solution. Let XI + I YI, X2 + I Y2, X3 + I Y3 E Xc. We check below the properties

of the inner product. .
1. (Additivity)
+ Iyd + (X2 + IY2), X3 + IY3}

(XI
= (XI + X2 + I(YI + Y2), x3 + IY3) .
= (XI + X2, X3) + (YI + Y2, Y3) + I [ (YI + Y2, X3) - (XI + x2, Y3) ]
= (XI, X3) + (YI, Y3) + I [(YI, X3) - (XI, Y3)] + «X2, X3) + (Y2, Y3)
+ I [(Y2, X3) - (X2, Y3)]}
= (XI + IYI, X3 + IY3) + (X2 + IY2, x3 + IY3).
2. (Homogeneity)
+ IfJ)(XI + IYI), X2 + IY2}

(a
= (axI - fJYI + I (fJXI + aYI), X2 + IY2}
= (axI - fJYI, X2) + (fJXI + aYI, Y2) + I [(fJXI + aYI, X2)
-(axI - fJYI, Y2)]
= (a + IfJ)[ (XI, X2) + (YI, Y2) + I [(YI, X2) - (XI, Y2)] ]
= (a + IfJ)(XI + lYI, X2 + IY2).
3. (Conjugate Linearity)
(XI + lYI, x2 + lY2) = (XI, X2) + (YI, Y2) + I [(YI, X2) - (XI, Y2)]
= (XI, X2)+ (YI, Y2) + I [(XI, Y2) - (YI, X2)]
= (X2, xd + (Y2, YI) + I [(Y2, xI> - (X2, yI>]
= (X2 + lY2, XI + lYI).
4. (Positivity)
Moreover,_
Section 32: INNER PRODUcr SPACES 299
Problem 32.3. Let Q be a Hausdorff compact topological space and let f1. be
a regular Borel measure on Q such that Supp f1. =
Q. Show that the function
(., .): C(Q) x C(Q) -+ JR, defined by
is an inner product. Also, describe the complexiftcation oIC(Q) and the extension
olthe inner product to the complexification oIC(Q).
Solution. If I, g, h E C(Q) and a, f3 E JR, then note that
(al+f3g,h) = l(al+f3g)hdf1.=a 1Ihdf1.+f31ghdf1.=a(f,hHf3(g,h),

(f,g) = 1Igdf1.= 19ldf1.= (g, f), and
(f, f) = 112 df1. 2: O.
Moreover, observe that (since Supp f1. = Q) a function 1 E C(Q) satisfies
The complexification Cc(Q) of C(Q) consists of all complex-valued functions

1 + 19, where I, g E C(Q). The complex inner product is given by
for all I, g E Cc(Q)·
Problem 32.4. Show that equality holds in the Cauchy-Schwarz inequality (i.e.,
I(x, y)1 = IIx lilly II) if and only if x and yare linearly dependent vectors.
Solution. Assume I(x, y)1 = IIxIiIlYIi. If x = 0, then the conclusion is obvious.

So, assume x -:f. O. Let (x, y) = rete. Replacing x by e-,e x , we can assume
without loss of generality that (x, y) = r 2: 0, and so (x, y) = IIxlillyli. Now,
300 Chapter 6: HILBERT SPACES
notice that for each real ).. we have
O:s ()..x + y, Ax + y) = )..2 IIx 112 +)..[ (x, y) + (x, y)] + lIyll2
= )..2 IIx 112 + 2)"(x, y) + lIyll2
= )..2 IIx 112 + 2)..lIxlillyll + lIyll2
= ()..lI x ll + lIyllt
So, if)" =- ::.~::, then (Ax + y, Ax y) + =0 or Ax + Y =

O. This implies
lIyllx - IIxlly = 0, which means that the vectors x and y are linearly dependent.
If x and yare linearly dependent, then the equation I(x, y)1 =
IIxlillyll should
be obvious.
Problem 32.5. If x is a vector in an inner product space. then show that
IIxll = sup I(x, Y)I.

11)'11=1
Solution. If x = 0, then the conclusion is obvious. So, we consider the case

x =1= o. If lIyll = 1, then the Cauchy-Schwarz inequality implies I(x, y)1 :s
IIxlillyll :s IIxll· Therefore, we have
sup I(x, y)1 :s IIxll·

11)'11=1
For the reverse inequality, let z = xlllxll. Then, IIzll = 1, and so
sup I(x, y)1 2: I(x, z)1 = l(x,xlllxll)1 = (x,x)/lIxll = IIxli.

lIyll=1
Therefore, IIxll = sUPII)'II=1 I(x, Y)I, with the supremum being in actuality the
maximum.
Problem 32.6. Show that in a real inner product space x .l y holds if and only
ifllx + yll2 =
IIxll2 + lIyll2. Does IIx + yll2 =
IIxll2 + lIyf in a complex inner
product space imply x .l y?
Solution. Let x and y be two vectors in a real inner product space. If x .l y,

then the Pythagorean Theorem gives IIx + yll2 = IIxll2 + lIyll2. Conversely, if
IIx + yll2 = IIxll2 + lIyll2, then from

IIx + y 112 = (x + y, x + y) = (x, x) + (x, y) + (y, x) + (y, y)
= IIxll2 + (x, y) + (y, x}+ lIyll2
= IIxl\2 + 2(x, y) + lIyll2,
it follows that 2(x, y) = 0, and so x 1- y.
In complex inner product space the Pythagorean identity I\x + y 1\2 = I\x 112 + 1\ y UZ
does not imply x 1- y. To see this, consider a non-zero vector x and let y = IX.
Clearly, lIyll2 = (u:, IX) = I\x1l2. Now, note that
IIx + yl\2 = IIx + Ixll2 = 1\(1 + l)xlI 2 = 11 + 11211xl\2

= 21\xl\2 = IIxI\2 + l\yl\2,
while (x, y) = (x, IX) = -llIxlI 2 =1= 0.

Problem 32.7. Assume that a sequence {XII} in an inner product space satisfies
(x n , x) -+ IIxll2 and IIxlll\ -+ IIxli. Show that Xn -+ x.
Solution. Observe that (XII' x) -+ IIxI\2 implies (x, XII) = (XII' x.) -+ IIxll2 =
IIx1I2. So, from
IIx n _x1I 2 = (XII -X,XII -x)
it follows that I\xlI - x 1\ -+ 0, i.e., XII -+ x.
Problem 32.8. Let S be an orthogonal subset of an inner product space. Show

that there exists a complete orthogonal subset C such that S ~ C.
Solution. Assume that S is an orthogonal subset of an inner product space X.

Let C denote the collection of all orthogonal sets that contain S. That is, an
orthogonal set A of vectors of X belongs to C if and only if S ~ A. If we consider
C partially ordered by the inclusion relation ~; then it is easy to see that C satisfies
the hypotheses of Zorn's Lemma. Now, notice that any maximal element C of C
is a complete orthogonal set satisfying S ~ C.
Problem 32.9. Show that the norms of the following Banach spaces cannot be
induced by inner products.
a. The norm IIxll = max{lxIi, IX21, ... , IXnll on JR.".

b. The sup norm on C[a, b].
c. The Lp-norm on any Lp(f.J..)-spacelor each 1 ::: p ::: 00 with p =1= 2.
Solution. (a) Consider the vectors x = (1, 0, ... , 0) and y = (0, 1, 0, ... , 0).
Clearly,
IIxll = lIy/i = /Ix + y/l = /Ix - yll = 1,

and so
Therefore, /Ix + y/l2+ /Ix - y/l2 =1= 2/1x/l2+2/1y/l2 and consequently the norm /1./1
does not satisfy the Parallelogram Law. This implies that the norm /I . /I cannot be
induced by an inner product.
(b) Again, we shall show that the sup norm 11·/100 does not satisfy the Parallelogram
Law-and this will guarantee that the sup norm is not induced by an inner product.
To see this, consider the two functions 1 (the constant function one) and I: [a, b] -+
JR. defined by I(x) = ~:::~. Now, note that
111/100 = 111/100 = 1, 111 + 1/100 = 2, and 111- 11100 = 1.
Therefore,
so that the norm II . 1100 does not satisfy the Parallelogram Law.
(c) Assume that there are two disjoint measurable sets E and F such that < °
°
f.J..*(E) < 00 and < f.J..*(F) < 00. First, we consider the case p = 00. Then,
note that
and consequently,
This shows that the norm /I . /100 does not satisfy the Parallelogram Law and so is
not induced by an inner product.
Section 32: INNER PRODUCT SPACES 303
Now, consider the case 1 ::: p < 00 with p =1= 2. The functions f
1 1
=
= [p.*(F)r"p XF satisfy IIfllp = IIglip = 1, and hence,
[p.*(E)r"p XE and g
2(lIfllp)2 + 2(lIg II pi = 4 = 22.

Also, from If + glP = If - glP = Ifl P + IgIP, we see that
Since p =1= 2, we have 21+% =1= 22, and so
This shows that the norm II . lip does not satisfy the Parallelogram Law and so it
is not induced by an inner product.
Problem 32.10. Show that a norm II . II in a complex vector space is induced by

an inner product if and only if it satisfies the Parallelogram Law, i.e., if and only if
holds for all vectors x and y . Moreover, show that if 11·11 satisfies the Parallelogram
Law, then the inner product (', .) that induces II . II is given by
(x, y) = 41 ( IIx + y 112 - IIx Y 112 + llix + l y 112 - llix - l y UZ ).
Solution. If II . II is induced by the inner product (', .), then for all vectors x and
y, we have
IIx+yIl2+lIx-yIl2 = (x+y,x+y)+(x-y,x-y)
= [(x, x) + (y, x) + (x, y) + (y, y)]
+ [(x, x) - (y, x) - (x, y) + (y, y)]
= 211xll2 + 211y1l2.
For the converse, assume that the norm II . II satisfies the Parallelogram Law.
Consider, the complex-valued function (., .) defined by
(x, y) = 41( IIx + yll 2 - IIx - yll-? + llix + lyll-? -llix - lyll-?) .
Clearly, (x, x) = IIxll2 holds for all vectors x. To finish the solution, we shall
verify that (', .) is a complex inner product. Start by observing that

1 .
(y,x) = 4( lIy +x1I2 -lly _x1I 2 + lily + IX 112 -Illy -lxII2)
= ~(IIX + yll2 -lix - yll2 -lll(-I)(X + ly)1I 2 + I III (X _ly)1I 2 )
= ~(IIX+YIl2_IIX_YIl2_IIIX+lYIl2+1IlX_IYIl2)
= (X,y).
Next, note that for all vectors u, v and w, we have
4(u + v, w) + 4(u - v, w)
= [liu + v + wll2 -liu + v - + diu + v + Iwll2 -Iliu + v -lwIl2]
wll2
+ [liu - v + wll2 - lIu - v - wll2 + Iliu - v + Iwll2 -iliu - v _lwIl 2]
= [liu + w + vll2 + lIu + w - v1l2] - [liu - w + vll2 + lIu - w - v1l2]
+ 1[lIu + IW + vll2 + lIu + IW - v1l2] -1[lIu -IW + vll2 + lIu -IW - v1l2]
= 211u + wII2 + 211vll2 - 211u - wII2 - 211vll2
+ I [211u + iwII2 + 211vll2 - 211u - IwII2 - 211v1l2]
= 2[lIu + wII2 - lIu - wII2 + Iliu + lwII2 - Iliu -lwII2]
= 8(u, w).
Thus, for all vectors u, v, and W we have
(u + v, w) + (u - v, w) = 2(u, w).
When v =
u, (*) yields (2u, w) = 2(u, w). Now, letting u = t(x + y), v =
~(x - y) and w =
z in (*), we get
(x, z) + (y, z) = (u + v, z) + (u - v, z) = 2(u, z) = (2u, z) = (x + y, z),
which is the additivity of (., .) in the first variable.

For the homogeneity, note first that
1
(IX, y) = 4(1I1x + yll2 - IIlx - yll2 + IlIlx + lyll2 - IlIlx _ lyll2)
= ~(1I1X + yll2 - 1I1~ _ ylf + Ilix + yll2 -Ilix _ Y1l2)

= I[~(IIX + yll2 -lix - yll2 -llIlx + yll2 + IlIlx - YIl2)J
= z[~(IIX + yll2 - IIx - yll2 - zllz(x - zy)1I 2 + I liz (X + ZY)1I 2)]
= ZU(IIX+ yll2 -lix - yll2 +.zllx + zyll2 -zllx _ZYIl2)]

= z(x, y).
Now, as in the proof of Lemma 18.7, we can establish that (rx, y) = rex, y) holds
for each "real" rational number I' and all x, y E X. Since (', .), as defined above,
is a jointly contil!uous function (relative to the norm II . 11), it easily follows that
(ax, y) = a(x, y) holds for all a E lR and all x, y E X. Finally, for an arbitrary
complex number a + z{3 and arbitrary vectors x and y, note that
(a + z{3)x, y) = (ax + z{3x, y) = (ax, y) + ({3(zx), y)

= a(x, y) + {3(zx, y) = a(x, y) + {3z(x, y)
= (a + z{3)(x, y).
This establishes that (', .) is an inner product that induces the norm II . II.
Problem 32.11. Let X be a complex inner product space and let T: X -+ X be

a linear operato1: Show that T = =
0 if and only if (T x, x) 0 for each x EX.
Is this result true for real inner product spaces?
Solution. Assume that (T x, x) = 0 holds for all x EX. From the identity
(T(x + y), x + y) = (Tx, x) + (Tx, y) + (Ty, x) + (Ty, y)

and our hypothesis, it follows that
(Tx, y) + (Ty, x) = 0
for all x,y E X. Replacing y by zy in (**) yields (Tx,zy) + (T(zy),x) =

z [-(Tx, y) + (Ty, x)] = O. So,
-(Tx, y) + (Ty, x) = 0
holds for all x, y E X. Adding (**) and (***), weget2(Ty, x) = 0 or(Ty, x) = 0
for all x, y E X. Letting x = Ty, we get (Ty, Ty) = 0 and so Ty = 0 for all
y EX, i.e. T = O.
For real inner product spaces the preceding conclusion is false. Here is an
example. Consider the Euclidean space lR2 equipped with its standard inner prod-
uct and define the linear operator T: lR2 -+ lR2 by T (x) = (- X2, x d for all
306 Chapter 6: Hll..BERT SPACES
x= (Xl, X2) E ]R2. Clearly, T #- 0, and
holds for all X E ]R2.
Problem 32.12. If {xn} is an orthonormal sequence in an inner product space,

then show that lim(xn, y) =·0 for each vector y .
. Solution. Let {x n } be an orthonormal sequence in an inner product space, and

let y be an arbitrary vector. Then, from Bessel's fuequality, we have
00
L I(x n, y)1 2 :s lIyll2 < 00.

n=l
This implies I(xn , y)1 2 -+ 0, and so (x n , y) -+ O.
Problem 32.13. The orthogonal complement of a nonempty subset A of an

inner product space X is defined by
A l. = {x EX: x ..L y for all YEA}.
We shall denote (Al.)l. by AH. Establish the following properties regarding

orthogonal complements:
a. a
A l. is closed subspace of X, A s;:; A H and A n A l. {O}. =
b. If AS;:; B, then Bl. s;:; Al..
c. Al. = ifJ. = [L(A)]l. = [L(A)] l., where L(A) denotes the vector
subspace generated by A in X.
d. IfM andN aretwovectorsubspacesofX, then MH+NH s;:; (M+N)H.
e. If Mis afinite dimensional subspace, then X = M EB Ml..
Solution. (a) If x, YEA l. and a, {3 are arbitrary scalars, then for each Z E A we
have
(ax + (3y, z) = a(x, z) + (3(y, z) = a 0 + {3 0 = 0,

and so ax + {3y E Al.. Therefore, Al. is a vector subspace of X. Since x E A
implies x ..L y for all y E Al., it follows that x E Al.l., i.e., A s;:; AH. Now if
x E AnAl., then (x, x) =
0 or x = 0, and thus AnAl. {OJ. =
(b) Assume A s;:; B and x E Bl.. If YEA, then y E B, and so y ..L x. This
implies x E Al., and so Bl. s;:; Al..
(c) From A S;; A and Part (b), it follows that A.L S;; A.L. Now, let x E A.L and
let YEA. Pick a sequence {YIl} S;; A satisfying Yll -+ Y and note that
(y, x) = n-+oc
lim (YIl' x) = O.
.
Therefore, x E A .L. Hence, A.L S;; A .L, and thus A.L A .L. =
For the other equalities, note first that A S;; .c(A) implies [.c(A)].L S;; A.L. Now,
fix x E A.L, and let Y E .c(A). Pick Yl, ... , Yk E A and scalars AI, ... , Ak such
= ",k
that Y L...i=1 AiYi· Then,
k k
(y, x) = (I>iYi, X) = I>;(Yi, x) = O.
i=1 i=l
This shows that x E [.c(A)].L. Thus, A.L S;; [.c(A)].L, and so A.L = [.c(A)].L.
(d) From M S;; M + N, it follows that MH S;; (M + N)H holds. Likewise,
N S;; M + N implies N H S;; (M + N)H. Therefore, M H + N H S;; (M + N)H.
(e) Let M be a finite dimensional subspace of dimension 11. In order to establish
that X = M EB M.L, we must show that every vector can be written in the form
Y + z with Y E M and Y E M.L. (The uniqueness of the decomposition should be
obvious.)
Start by fixing a Hamel basis {XI, X2, •.• , XII} of M. Replacing (if necessary)
{Xl, X2, ••• , XII} by the normalized set of vectors that can be obtained by applying
the Gram-Schmidt orthogonalization process (Theorem 32.11) to {XI, X2, •.. , XII},
we can assume that the set {XI, X2, .•. , XII} is also an orthonormal set.
Now, fix X E X and consider the vectors
n /I
Z = L(x, Xk)Xk and y=x-L(X,X})Xk.

k=1 k=l
Clearly, z E M and since (y, Xk) =

0 for each k, it easily follows that Y E M.L.
Now, note that X =
Y + Z EM EB M.L.
Problem 32.14. Let V be a vector subspace of a real inner product space X. A

linear operator L: V -+ X is said to be symmetric if (Lx, y) = (x, Ly) holds
for all x, Y E V.
a. Considerthe realinnerproduct space C[a, b] and let V = {f E C 2 [a, b]:
f(a) = f(b) = O}. Also, let p E Cl[a, b] and q E C[a, b] be two fixed
functions. Show that the linear operator L: V -+ C[a, b], defined by
L(f) = (p!'Y + qf,

is a symmetric operatOl:
b. ConsiderJR." equipped with its standard innerproductand let A: JR." -+ JR."

be a linear operator. As usual, we identify the operator with the matrix
A = [aij] representing it, where the jth column of the matrix A is the
column vector Ae j. Show that A is a symmetric operator if and only if A
is a symmetric matrix. (Recall that an n x n matrix B =
[bij] is said to be
symmetric if bij = b ji holds for all i and j.)
c. Let L: V -+ X be a symmetric operat01: Then L extends naturally to
a linear operator L: Vc =
{x + IY: x, y E V} -+ Xc via the formula
L(x + lY) = Lx + Ity. Show that L also satisfies (Lu, v) = (u, Lv) for
all u, v E Vc and that the eigenvalues of L are all real numbers.
d. Show that eigenvectors of a symmetric operator corresponding to distinct
eigenvalues are orthogonal.
Solution. (a) If f, g E V, then note that
(Lf, g) = 1b ([p(x)f'(x)]' + q(x)f(x»)g(x)dx
= l b
[p(X)f'(X)],g(X)dX + 1b q(x)f(x)g(x)dx
-1b
a a .
= P(X)f'(X)g(X)I: p(x)f'(x)g'(x)dx + 1b q(x)f(x)g(x)dx
= 1b q(x)f(x)g(x)dx -1 b
p(x)f'(x)g'(x)dx
= (f, Lg).
(b) Recall that the transpose of a matrix B = [bij] is the matrix Bt = [bjiJ. In
tenns of the transpose, a matrix A is symmetric if and only if At = A. Now, our
conclusion follows immediately from the following two identities:
(Ax, y) = (x, Aty) for all x, y E JR.", and

aij = (ei, Aej).
(c) If u = x + Iy and v = Xl + IYl are vectors of Vc , then note that
(Lu, v) = (L(x + IY), Xl + IYl) = (Lx + ILy, Xl + IYI)
= (Lx,xd+(LY,yd+I[(Ly,xd-(LX,YI)]
= (X, LXI) + (y, LYI) + I[(Y, Lxd - (X, LYI)]
= (x + IY, LXI + ILyd = (u, Lv).
Now, assume that A E C is an eigenvalue of L: Vc -7 Xc' Fix a unit vector

u EX c satisfying L u = AU, and note that
A = A(U, u) = (AU, u) = (Lu, u) = (u., Lu) = (u, AU) = I(u, u) = I.

This shows that A is a real number.
(d) Assume that L: V -7 X is a symmetric operator and let two nonzero vectors
u, v E Vc satisfy Lu = AU and Lv = /1-V with A t= /1-. By part (c), we know that
A and /1- are real numbers. Therefore,
(A -/1-)(u, v) = A(U, v) -/1-(u, v) = (AU, v) - (u, /1-v) = (Lu, v) - (u, Lv) = 0,
and so (u, v) = o.
Problem 32.15. Let (', .) denote the standard inner product on JR. n, i.e., (x, y) =
L;'=l Xi Yi for all x, y E JR.". Recall that an n x n matrix A is said to be positive
definite if (x, Ax) > 0 holds for all nonzero vectors x E JR.".
Show that a function of two variables (', .): JR.n x JR." -7 JR. is an inner product
on JR." if and only if there exists a unique real symmetric positive definite matrix
A such that
(x, y) = (x, Ay)
holds for all x, y E JR.". (It is known that a symmetric matrix is positive definite if
and only if its eigenvalues are all positive.)
Solution. Let (', .): JR." x JR." -7 JR. be a function of two variables. Assume first
that there exists a real symmetric positive definite matrix A such that
(x, y) = (x, Ay)

holds for all x, y, z E JR. n • Then, for all x, y E JR." and all a, {3 E JR., we have
(x, y) = (x, Ay) = (Ax, y) = (y, Ax) = (y, x),

(ax + {3y, z} = (ax + (3y, Az) = a(x, Az) + (3(y, Az) = a(x, z} + {3(y, z}, and
(x,x) = (x, Ax) 2: 0 for all x E JR." and (x,x) = 0 ~ x = O.
This shows that (', .) is an inner product.

For the converse assume that the function of two variables (', .): JR." x JR.n -7 JR.
is a real inner product. Let el, e2, ... , en denote the standard unit vectors, and so
310 Chapter 6: HILBERT SPACFS
each vector x E JRn is written as x = 2:7=1 Xiei. It follows that

n n n n
(x, y) = (:L:'>ieh I:>jej) = LLxiYj(ei, ej} = (x, Ay)
i=1 j=1 ;=1 j=1
for all x, Y E JRII, where A is the n x n matrix A = J.

[(ei' e j) Clearly, A is a real
symmetric matrix and in view of (x, Ax) = (x, x), we see that A is also a positive
definite matrix. The uniqueness of A should be obvious.
33. IllLBERTSPACES
Problem 33.1. Let (X, S, /L) be a measure space and let p: X -+ (0,00) be
a measurable function-called a weight function. Show that the collection of
measurable functions
under the inner product (', .): L2(P) x L 2 (p) -+ JR, defined by
(j, g) = ! pfg d/L,
is a real Hilbert space.
Solution. It should be clear that L 2(p) is a vector space. Moreover, since f E

L2(P) is equivalent to ..fPf E L2(/L), it follows from Holder's inequality that
and so (', .) is well-defined. We leave it as an exercise for the reader to verify that
(', .) is indeed a real inner product. We shall prove that L 2(p) is a Hilbert space
by establishing that it is complete.
To this end, let Un} ~ L 2 (p) be a Cauchy sequence. That is, for each E >
there exists some no such that
°
Section 33: HILBERT SPACES 311
holds for all n, m ~ no. This means that the sequence of functions {-/PIn} S;
L 2(/L) is a norm Cauchy sequence of the Hilbert space L2(/L). Since L 2(/L) is a
Banach space (Theorem 31.5), it follows that there exists some function g E L2(/L)
such that
f I.../Pln - g 12 d/L -+ O.
Now, note that if I = g / -/p, then I E L p(/L) and
This shows that L2(P) is norm complete andhence, it is a Hilbert space.

The reader should also notice that L2(p) is exactly the Hilbert space L2(V) for
the measure v: All -+ [0,00] defined by
v(A) = iP(X)d/L(X)
for each A E All'
Problem 33.2. Show that the Hilbert space L2[0, (0) is separable.
Solution. Consider the countable set offunctions {fk,lI: k, n = 1,2, ... }, where
if O::sx::Sk
An(x) = { X"
0
if k < x.
We know that the continuous functions with compact support are dense in L 2 [0, (0)
(Theorem 31.11) and so, we need only prove that the linear span of {All} is dense
in the vector space of continuous functions with compact support. Observe that if
this is established, then the linear span of {fk,lI} with rational coefficients would
be a countable dense set.
Let I E L2[0, (0) be a continuous function with compact support, and letE > O.
Fix an integer k such that I(x) = 0 for all x ~ k. By the Stone-Weierstrass
approximation theorem, there exists a polynomial P(x) = r:::"o
cllx n satisfying
II(x) - P(x)1 < E/,JJ( for each x E [0, k]. Now, notice that if we consider the
function g E L2[0, (0) defined by g(x) = r:::"o
CII Ik,II(X), then
III - gll2 = (1 00
II(x) - g(x)1 dX)

2 1
= (10
k
2
I/(x) - P(x)1 dX)
1
<
r sf:
(}o dx r
1
=E
312 Chapter 6: Im-BERT SPACES
holds. This shows that the linear span of the countable set {!k.n: k, n = 1,2, ... }
is dense in L2[O, 00), and so L 2[O, 00) is separable.
Problem 33.3. Let {1/rn} be an orthonormal sequence offunctions in the Hilbert

space L2[a, b] which is also uniformly bounded. If {an} is a sequence of scalars
such that an 1/rn -+ 0 a.e., then show that lim an = O.
Solution. Fix some constant C such that I1/rn(x) I ::: C hold for all n and for all
x E[a, b]. Also, let{a n} be a sequence of scalars such that a n1/rn(x) -+ 0 holds
for almost all x.
Next, fix E > 0 so that EC 2 < ~. Now, by Egorov's Theorem 16.7, there exists
a measurable set E f; [a, b] with A(E C ) < E such that the sequence of functions
{an 1/rn} converges uniformly to zero on E. So, there exists an integer m such that
lan1/rn(x)1 ::: E for all n ::: m and all x E E. Then, we have
Ian 12 = lb Ian 1/rn (t) 12 dt = { Ian 1/rn(t)12 dt + { Ian 1/rn(t)12 dt

a lE lEe
::: {E 2 dt + lan l2 { I1/rn(t)1 2 dt
lE lEe
2(b-a)+la f ( C 2 dt
::: E n
lEe
::: E2(b - a) + Ela nl2C 2.
This implies ~ Ian 12 < (1 - EC 2)la n 12 ::: E2(b - a) for alln ::: m, or
lanl ::: EJ2(b - a)
for alln ::: m. Since 0 < E < 2~2 is arbitrary, we have established that an -+ O.
Problem 33.4. Let {¢n} be an orthonormal sequence offunctions in the Hilbert

space L2[-I, 1]. Show that the sequence offunctions {1/In}, where
I
1/rn(x) =(b=a)2¢n(b=a (x - ~ )),

is an orthonormal sequence in the Hilbert space L2[a, b].
Solution. Observe that the inner product satisfies
(1/rn, 1/Im) = lb 1/rn(X)1/r~(x)dx
= b=a lb ¢n(b=a(x - b!u))¢m(b=a(x - b!a))dx.

Making the substitution t = b~a (x b!u), we have dt = b~adx, and so
That is, {o/II} is an orthonormal sequence in the Hilbert space L2[a, b].
Problem 33.5. Show that the norm completion X of an inner product space X
is a Hilbert space. Moreover. if x, Y E X and two sequences {XII} and {YII} of X
satisfy XII -+ x and YII· -+ Y in X. then establish that the inner product of X is
given by
(x. y) = lim (x
Il~OO
ll • Yn).
Solution. Assume that X is the norm completion of an inner product space X

and let x, Y EX. Pick two sequences {XII} and {YIl) of X such that IIx" x II -+ 0
and lIy" - yll -+ 0, where II . II is the norm of X (which is the unique continuous
extension of the norm of X to X). Fix some constant M > 0 such that IIx n II :::: M
and lIy,,1I :::: M hold for each 11. Then, using the Cauchy-Schwarz inequality, we
have
I(XII' y,J - (xm• Ym) I = I(XII' YII) - (x", Ym) + (x", Ym) - (xm, Ym) I
= I(x"' y" - Ym) + (XII -Xm, YIII) I
< l(xn,YII-Ym)I+I(x,,-Xm,Ym)1
< IIx"lIlIy" - Ymll + IIx" -xmllllYmll
:::: M{lIxlI-xmll+IIY,,-Ymll).
This shows that the sequence of scalars {(x"' y,,)} is a Cauchy sequence and hence,
convergent.
Next, assume two other sequences {x:,} and {Y:,} of X satisfy II x:, - xII -+ 0
and lIy:, - Y II -+ O. We can assume without loss of generality that IIx~ II :::: M and
lIy;'1I :::: M holds for each n. By the preceding lim(x;', Y;') exists, and since
I(XII' y,,) - c<, Y;,) I = I(x"' y,,) - (x", Y;,) + (x"' Y;) - (;.<, Y;,) I
= I(xn, y" - Y;,) + (xn - x;" Y;,) I
< I(XII' YII - Y;) I + I(x" - x~, y~) I
< II XII II llYn - Y~II + IIxn -x~lIlIy;,1I
:::: M( IIx" X;, II + llYn - y;,II) ~ 0,
it follows that lim(xll , Yn) = lim(x~, y~). In other words, the formula
gives rise to well-defined scalar-valued function on X x X.

Now, it should be clear that the properties of the inner product are transferred
via (*) from the inner product of X to the function (., .): X x X ~ C. In other
words, the formula of (*) is an inner product on X. Moreover, from
IIxll2 = lim IIXn 112

n-+-oo
= lim (X n , XII) =
11-+00
(x, X),
we see that the inner product given by (*) induces the norm of X.
Problem 33.6. Show that the closed unit ball of £2 is not a norm compact set.
Solution. Let U = {x E £2: IIx II ::: I} be the closed unit ball of £2. Now, for
each n let en = (0, 0, ... , 0, I, 0, 0, ... ), the sequence with I in its nth coordinate
and zero elsewhere. Note that lien II = I for each n and thus {en} is a sequence of
the unit ball of £2. (In fact {ell} is an orthonormal sequence of £2.) Now, notice
that for n =1= m we have lien - em II = ../2. This implies that {ell} does not have any
Cauchy subsequences-and hence, it does not have any convergent subsequences
either. Now a glance at Theorem 7.3 guarantees that U is not a norm compact
subset of £2.
Problem 33.7. Show that the Hilbert cube (the set of all x = (XI, X2, ... ) E £2
such that IXn I ::: ~ holds for all n) is a compact subset of £2.
Solution. Let C = {(XI,X2,.") E £2: IXIII ::: ~ foreach n = 1,2, ... }.

Clearly, C is a closed subset of £2. Thus, in order to establish the compactness of
C, it suffices to prove (by Theorem 7.8) that C is totally bounded.
To this end, let 8 > O. Fix some n such that I:f:n+1 b< 8. Since the set
t
A = {(x I, ... , xn) E ]R.n: Ix;! ::: for I ::: i ::: n} is closed and bounded, it must
be a compact subset of ]R.II. Pick X I , ... , xm E A (where Xi = (xl, ... , xII») so
that A S;;; U7~1 B(xi, 8) holds. (We consider, of course,]R.n equipped with the
Euclidean distance.) Now, for each I ::: i ::: m let Yi = (xl, ... , x~, 0, 0, ... ).
Then, it is easy to see that C S;;; U::I B(Yi, 28) holds in £2. This shows that C
is totally bounded, as required.
Problem 33.8. Show that every subspace M of a Hilbert space satisfies M =

Ml.l..
Solution. It should be clear that (M)l. = Ml.; see Problem 32.13. Therefore, by
Theorem 33.7, H = M Ef) MJ... Also, it should be noticed that M s:;; M s:;; MH.
Now, let x E M l.l.. From H = M Ef) M l., it follows that we can write x = u + v
with U E M and v E Ml.. This implies x-u ::= v E Ml. and since u E M s:;; MH,
we have x - u E Ml.l.. Hence, x - u E Ml. n Ml.l. = {O} or x = u E M.
Therefore, M l.l. s:;; M also holds true, and so M l.l. = M, as desired.
Problem 33.9. For two arbitrmy vector subspaces M and N of a Hilbert space
establish the following.'
a. (M + N)l. = Ml. n Nl., and
b. if M and N are both closed, then (M n N)l. = M l. + N l..
Solution. (a) Let x 1. M + N. Then, x 1. y holds for all y E M and x 1. z holds

for all zEN. That is, x E M l. and x E N l.. Therefore, (M + N)l. s:;; M l. n N l..
For the reverse inclusion, let x E M l. n N l. and let y E M + N. Write
y = u + v with u E M and v E N and note that (x, y) = (x, u) + (x, v) = 0
holds. Hence, x E (M + N)l., and therefore Ml. n Nl. s:;; (M + N)l.. Thus,
(M + N)l. = Ml. n Nl..
(b) Now, suppose that M and N are closed subspaces. By the preceding problem
we know that M = Ml.l. and N = Nl.l.. Now, use part (a) to get
Therefore,
Problem 33.10. Let X be an inner product space such that M = M l.l. holds for
evelY closed subspace M. Show that X is a Hilbert space.
Solution. We need to show that X is complete in the induced norm. For this, it
suffices to establish that X = X, where Xdenotes the norm completion of X. (We
already know that X is a Hilbert space; see Problem 33.5.) To this end, let fi EX
be a nonzero vector.
The linear functional f: X -+ C defined by f (x) = (x, fi) is nonzero and
continuous. So, f restricted to X is also continuous and since X is norm dense
in X, it follows that f: X -+ C is a nonzero continuous linear functional. In
particular, its kernel M = (x E X: f(x) = O} is a proper closed subspace of
X. We claim that M l. =1= {O}. Indeed, if M l. = {O}, then it follows from our
hypothesis that M = Ml.l. = {O}l. = X, which is a contradiction.
Next, fix a vector u E M.l. with lIuli = 1 and let v = f(u)u EX. Now, taking
into account that f (x)u - f (u)x E M holds for each x EX, it follows that
f(x) = f(x)(u, u) = f(u)(x, u) = (x, v).

for each x EX. That is, (x, u) = (x, v) for all x EX. Since X is dense in X, we
get (x, v) = (x, u) for all x EX. That is, (x, v - u) = 0 for all x EX, and from
this we conclude that ii = v. EX. So, X = X, and thus X is a Hilbert space.
Problem 33.11. Consider the linear operator V: L2[a, b] -+ L2[a, b] defined

by
Vf(x) = l x
f(t)dt.
Show that the norm of the operator satisfies II V II ::: b - a.
Solution. By HOlder's Inequality, we get
IVf(x) I ::: l x
If(t)1 dt ::: lb If(t)1 dt
l
::: [ a
b
2
If(t)1 dt
I
r[lb r
1
a
2
1 dt
I
::: (b - a)2I1fll.
Therefore, the norm of V satisfies
IIV f112 = lb IV f(t)12 dt ::: (b - a) lb IIfll2 dt
::: (b - a)211f112.
This implies II V II ::: b - a.

Problem 33.12. Let {XII} be a norm bounded sequence of vectors in the Hilbert
space £2, where XII = (xf, xi, x 3' ... ). If for each fixed coordinate k we have
limn-+oo xl: = 0, then show that
lim (XII' y)
11""""00
=0
holds for each vector y E £2.
Solution. Choose some A > 0 such that IIxnll :::: A holds for all n. Now,
fix a vector Y, = (YI, Y2, ... ) E l2 and let E > O. Pick some m satisfying
O:=;:m IYkl p <
2
E.
Since for each fixed k we have limn-+co xl: = 0, there exists an integer no
I I
satisfying "LZ'::I xl: Yk < E for all n::: no. Now, using the Cauchy-Schwarz
inequality, we see that for each n ::: no we have
00 m 00
l(xn,y)1 = I Lxl:Yk I:::: 1Lxl:Yk 1+ L IXI:Ykl

k=1 k=1 k=m+1
< E+( L
00
k=m+1
Ix1: 12 r( L 2r
'00
k=m+1
IYk!
,
:::: E + AE = (1 + A)E .
Since E > 0 is arbitrary, we have shown that limn-+oo(xn, y) = O.
Problem 33.13. Let H be a Hilbert space and let {xn} be a sequence satisfying
lim (x n , Y) = (x, y)
n-oo
for each Y E H. Show that there exists a subsequence {Xk,,} of {xn} such that
lim Iloq , +Xk, +Xk, +·+Xkn - X II = O.

Il~OO Il
Solution. Start by noticing that we can assume without loss of generality that
x = O. Therefore, suppose
lim (xn, Y)
Il-+OO
=0
for each Y EX. We claim that the sequence {x n } is norm bounded. To see
this, for each 11 consider the continuous linear functional fn: H -+ C defined by
fn(Y) = (y, xn) for each Y E H. By our condition, the sequence of bounded
linear functionals Un} is pointwise bounded. So, by the Principle of Uniform
Boundedness (Theorem 28.8), there exists some C > 0 such that II 1,,11 :::: C for
each 11. Now, notice that (by Theorem 33.9) IIxn II = II 1,,11 holds for each n.
Now, let kl = 1 and then choose k2 > kl with I(Xk, ,Xk2)1 < 1. Next, an
inductive argument shows that there exist integers kl < k2 < k3 < ... < k n <
kn+ I < . .. satisfying
To finish the solution, we shall show that
holds for each n. To see this, take inner products to get
II "L,7:1 Xkj r = "L,;'=I (Xkl' xd + "L,7=1 (Xk2~:k) + ... + "L,7=1 (Xkn , Xk)
n[C 2 +(1 + 2- 1 + 2-2 + 2- 3 + ... + 2-n )] 2 + C2

-< n2 < -----
- n .
This implies
as required.
Problem 33.14. Let p: [a, b] -T (0, (0) be a measurable essentially bounded

function and for each n =
0, 1, 2, ... let Pn be a nonzero polynomial of degree n.
Assume that
lb p(x)P,,(x)Pm(x)dx = °for 11 i= m.
Show that each P" has n distinct real roots all lying in the open interval (a, b).
Solution. By Theorem 33.12, we know that the sequence of orthogonal poly-

nomials Po, PI, P2, ... is complete and coincides (aside of scalar factors) with
the sequence of orthogonal functions of L 2 (p) that is obtained by applying the
Gram-Schmidt orthogonalization process to the sequence of linearly independent
functions {1,x,x 2,x3 , •• • }. In particular, we have I:
p(x)xmpn(x)dx = for
all m = 0, 1, ... ,11 - 1. Also, by mUltiplying each Pn by an appropriate scalar,
°
we can assume that each Pn has real coefficients and leading coefficient 1.
Now, fix one of these polynomials Pn , where n ?: 1. First, we shall show that Pn
cannot have any complex roots. If Ph has a complex root, then Pn has a factorization
of the form Pn(x) = [(x +a? + ,82]Q(X), where a and,8 are real numbers and Q
is a polynomial of degree 12 - 2. Hence Q(x)P,,(x) = [(x +a)2 + ,82][Q(X)]2 ?: 0,
and from J: p(x)xm Pn(x) dx = 0 for all m = 0, 1, ... , n - 1, it follows that
o< lb p(x)Q(x)Pn{x) dx = 0,
which is impossible. Hence, each Pn has only real roots. This means that Pn has
a factorization of the form
where 1'1,1'2, ... , rk are real number and m I, m2, ... ,mk are natural numbers
such that m I + m2 + ... + mk = n.
Next, we claim that Pn does not have any root outside of the open interval
(a, b). To see this, assume that one root lies outside of (a, b), say 1'1 :s a. Then the
polynomial Q(x) = (x - r2)m 2 ••• (x - rk)m k has degree less than n and satisfies
Q(x)Pn(x) ::: 0 for each a :s x :s b. But then, we have
o< lb p(x)Q(x)P,,(x) dx = 0,
which is a contradiction.
Finally, to see that each root appears with multiplicity one, assume by way of
contradiction that one root has multiplicity more than one, say m I > 1. If, again
if ml is even
if ml is odd,
then Q is a polynomial of degree strictly less than n and satisfies Q(x)Pn(x) ::: 0
for all a :s x :s b. But then, as previously,
0= lb p(x)Q(x)P,,(x)dx > 0,
which is absurd. Hence, each polynomial P" has n distinct real roots all lying in
the open interval (a, b).
Problem 33.15. In Example 33.13 we defined the sequence Po, PI, P2, ... of
Legendre polynomials by the formulas
1 d"
P,,(x) = -nn. 2
-n (x -It·
2,dx
We also proved that these are (aside of scalarfactors) the polynomials obtained by
applying the Gram-Schmidt orthogonalization process to the sequence of linearly
independent functions {I, x, x 2 , ••• } in the Hilbert space L 2([ -1, 1]). Show that
for each n we have
Solution. The proof of the formula PIl (I) = 1 is by induction. Notice that for
n = 0 and n = 1 the formula is trivially true. So, for the induction argument,
assume that Pn (1) =
1 holds true for some n. To complete the proof, we must
show that Pn+I(1) = 1. To see this, note that
1 d n+1 2 n+1
Pn+1 (x) = 2n+1(n + I)! dX"+ 1 (x - 1)
n
1 d [( 2 1)11+1)']
= 21l+I(n + I)! dx" x-
n
= 2n+1 (n1+ 1.)1 ddxn [2(n + l)x(x 2 - 1)n]
n n
= -1 -d [ (x -1)(x 2 _1)11] + _
1 d
_ (x 2 _1)11
211n! dx" 211n! dx ll
= (x - I)Q(x) + PII(x),
where the term (x - I)Q(x) designates the form of the expression
1 dn ll
- - ( x -1)(x 2 _1)n = -1- (d x _1)II+I(x + 1)".

2"n! dx" .2"n! dxl1
So, PI1 + I (1) = PII (1) = l.

Next, we shall compute the norm of PI1 • Clearly,
IIPII 1I 2 = i: PII(x) PI1 (x)dx
II n l1
= -1- d
_(x 2 d 2
- 1)" _(x - 1)" dx.
(2l1n!)2 -I dx n dx"
Next, observe that the function (x 2 _1)11 = (x -l)l1(x+ 1)" and all of its derivatives
of order less than or equal to n - 1 vanish at the points ±l. So, integrating by
Section 33: lDLBERT SPACES 321
parts IHimes and using the previous observation, we obtain
( -1)// 11 2 d 2// (-1)// 11 2

IIP//1I 2 =- - (x -
(2//n!? _I
1)//--(X2
dX 211
- 1)// dx
.
=- - (x -
(2//n!p_1
1tc2n)! dx
= (_1)11 (2n)! 11 (x 2 _1)" dx = (2n)! 11

(1 X2)" dx.
(2"n!)2 -1 (2"n!)2 -1
Using integration by parts to evaluate this last integral gives
11(1 - x t dx = 11 (1 - X)" + xt dx
2 (l
-I -1
- / +1 1+11 n(1- x
- (l-X)"(I+X),,+III
- -1
)"-1
/ +1
(l+.t),,+1 d
X
l
= _n-l (l-X)II-1(1+X)II+1 d X = ...
n+1 -1
= ---'
nl --:--
(n + 1)· .. (2n) -1
11 (1 + x)2// dx
(n!)22 211 +1
=-----
(2n)! (2n + 1)
Consequently, the norm of PII is given by
as claimed.
Problem 33.16. Let {T" }"EA be a family of linear continuous operators from a
complex Hilbert space X into another complex Hilbert space Y. Assume that for
each x E X and each y E Y the set of complex numbers {(T,,(x), y): a E A} is
bounded. Show that the family of operators {T" }"EA is uniformly norm bounded,
i.e., show that there exists some constant M > 0 satisfying IIT"II ::: M for all
a E A.
Solution. Observe that if Z is a Banach space over the field of complex numbers,
then we may also consider Z as a Banach space over the field of real numbers.
Therefore, the Principle of Uniform Boundedness (Theorem 28.8) can be applied
to any Banach space over the field of complex numbers.
Fix a vector x EX. For each a E A define the complex valued continuous
linear operator Ba: Y -+ C by
Thus, {BalaeA is family of bounded linear operators from the Banach space Y to
the Banach space of complex numbers C. From Theorem 33.9, we know that
Next, notice that for each fixed y E Y, it follows from
IBa(y)1 = I(y, Ta(x» I = I(Ta(x), y)1

and our hypothesis that the family of continuous linear operators {Ba laeA is point-
wise bounded. Hence, by the Principle of Uniform Boundedness (Theorem 28.8),
the family {BalaeA is norm bounded. This means that there exists a constant
Mx > 0 (that depends upon x) such that IIBall :s Mx for all a E A. Thus, we
have IITa(x)1I :s Mx for all a.
Therefore, the family {TalaeA of continuous linear operators Ta: X -+ Y is
pointwise bounded. Invoking the Principle of Uniform Boundedness once more,
we conclude that there exists a constant M > 0 satisfying II Ta II :s M for all
a E A.
Problem 33.17. Let {4>lIl be an orthonormal sequence in a Hilbert space Hand

consider the operator T: H -+ H defined by
00
T(x) = Lan(x, 4>1I)4>n,

n=1
where {an 1 is a sequence of scalars satisfying lim an = O. Show that T is a

compact operat01:
Solution. Let B be the open unit ball of H. We need to show that T (B) is a
compact set. For this, it suffices to show that T (B) is totally bounded. To this end,
fix E > 0 and observe that there exists an integer m such that Ian I < E holds for
all n > m.
Next, define the operator Tm: H -+ H by
m
Tm(x) = Lai(X, 4>i)4>i.
i=1
Clearly, the range of Tm is a finite dimensional subspace of H and thus, Tm is a

compact operator. Therefore, Tm(B) is a totally bounded set. Thus, there exists a
finite set {YI , Y2, ... , Yn} such that for each x E B there exists some 1 ::: k ::: n such
that IITm(x) - Ykll < E. Now, Parseval's IneR.uality 'L:I I(x, ¢i)1 2 ::: IIxll2 < 1
implies
I: ai(x, ¢i)<!>i f = I:
00 ? 00
\\T(x) - Tm(x) \\2 = II lail 2 1(x, ¢i)1 2

i=m+1 i=m+1
00
::: E2 I: I(x, ¢i)1 2 ::: E211xll2 < E2.

i=m+1
Therefore, for each x E B there exists some 1 ::: k ::: n such that
This shows that T(B) is totally bounded, and hence T is a compact operator.
Problem 33.18. Assume that T, T*: H -7 H are two functions on a Hilbert

space satisfying
(Tx, y) = (x, T*y)
for all x, Y E H. Show that T and T* are both bounded linear operators satisfying
IlTii = IIT*II and IITT*II = liT 112.

Solution.. By the symmetry of the situation, it suffices to show that T is a bounded
linear operator. We shall show first that T is a linear operator. To this end, fix
x, Y E H and two scalars a and f3. Then, for each z E H we have
(T(ax + f3y), z) = (ax+ f3y, T*z) = a(x, T*z) + f3(y, T*z)

= a(Tx, z) + f3(Ty, z) = (aTx + f3Ty, z),
or (T(ax + f3y) - aTx - f3Ty, z) = 0 for all z E H. This implies (by letting
z = T(ax + f3y) - aTx - f3Ty) that
T(ax + f3y) = aT(x) + f3T(y),

i.e., that T is a linear operator.
Next, for each y E H with lIyll :::: 1 consider the linear functional fy: H -+ H
defined by fy(x) = (x, T*y). Then, using the Cauchy-Schwarz Inequality, we
see that
Ify(x)1 = I(x, T*(Y)I = I(T(x), y)1 :::: IIT(x)lIl1yll :::: IIT(x)1I
for all y E H with lIy II :::: 1. This implies that the set of linear functionals
{fy: lIyll :::: I} is pointwise .bounded and therefore, by the Principle of Uniform
Boundedness, the set of linear functionals {fy: II y II :::: I} is norm bounded. Thus,
there exists some constant C > 0 such that II fy II :::: C for all y E H with II y II :::: 1.
Now, assume that y E H satisfies lIyll :::: 1 and T*(y) =f: O. Letting x =
T*(Y)/IIT*(y)lI, we obtain
1
IIT*(Y)II = IIT*(Y)III(T*(y), T*(y))1 = I(x, T*(y))1 = Ify(x)1 :::: C.
This implies
IIT*II = sup{IIT*(Y)II: lIyll :::: 1} :::: C,

and thus, T* is a bounded linear operator. By the symmetry of the situation, T is
likewise a bounded linear operator.
Now, note that for all x, y E H with IIxll = lIyll = 1, we have
I(Tx, y)1 = I(x, T*Y)I :::: IIxIlIlT*YII :::: IIxllllT*lIl1ylI = IIT*II,

and hence, IITxll = sup{I(Tx, y)l: lIyll :::: 1} :::: IIT*II for all unit vectors x E H.
This implies
1IT11 = sup IIT(x)lI:::: IIT*II·

IIxll=l
Using the symmetry once more, we get II T *II :::: II TIl, and so II TIl = II T * II·
Finally, for each x E H with IIx II = 1, we have
IITT*xll :::: II TIl IIT*II IIxll = II TIl 2, and

IIT*xIl2 = (T*x, T*x) = (TT*x, x) :::: IITT*lIlIxllllxll = IITT*II,
and so by taking suprema, we get II T T* II = II T 112.
Problem 33.19. Show that if T: H -+ H is a bounded linear operator on a

Hilbert space, then there exists a unique bounded operator T*: H -+ H (called
Section 34: ORTIiONORMAL BASES 325
the adjoint operator ofT) satisfying
(Tx, y) = (x, T*y)
forallx,y E H. Moreover, show that liT II = IIT*II.
Solution. Assume that T: H -+ H is a bounded linear operator on a Hilbert

space. For each fixed y E H, the formula fy (x) = (T x, y) defines a bounded
linear functional on H. By Theorem 33.9, there exists a unique vector T*y E H
satisfying
fy(x) = (Tx, y) = (x, T*y)

for all x E H. Now, use the preceding problem to conclude that the unique
function T *: H -+ H defined above is, in fact, a bounded linear operator satisfying
IIT*II = IITJ1·
34. ORTHONORMAL BASES
Problem 34.1. Let lei lief and {fJ 1jel be two orthonormal bases of a Hilbert
space. Show that 1 and J have the same cardinality.
Solution. Assume that {edief and {fJ bel are two orthonormal bases of a
Hilbert space. First, suppose that 1 is a finite set. Then, from Theorem 34.2, it
follows that lei lief is also a Hamel basis and so H is finite dimensional. Since
the /j (as being mutually orthogonal vectors) are also linearly independent, we
conclude that J must also be a finite set. This implies that {fJljel is itself a
Hamel basis for H, and so I and J must have the same number of elements.
Now, suppose that 1 and J are infinite sets. For each i E I, we define the set
of indices Ji = {j E J: (ei, fJ) =f: Ol. By Theorem 34.2, we know that each Ji is
nonempty and at-most countable. Next, we claim that
J=UJi .
ie!
To see this, let j E J. Since lei lie! is an orthonormal basis, it follows from
Parseval's Identity that Lie! I(h, ei)1 2 = IIfjll2 = 1, and so (ei, fJ) =f: 0 holds
true for some i E I. Thus, j E Ji holds true for at least one i E I, and thus
J = Uie ! 1;.
Finally, to see that I and J have the same cardinality use (*) together with
the standard "cardinality" arithmetic; see, for instance, P. R. Halmos, Naive Set
Theory, Springer-Verlag, 1974, pp. 94-98.
Problem 34.2. Let {ei liEf be an orthonormal basis in a llilbert space H. If D

is a dense subset of H • then show that the cardinality of D is at least as large as
that of I . Use this conclusion to provide an alternate pro(!t" of Theorem 34.4 by
proving that for an infinite dimensional Hilbert space H tile following statements
are equivalent:
1. H has a countable orthonormal basis.
2. H is separable.
3. H is linearly isometric to .e2.
Solution. Let {ei liEf be an orthononnal basis in a Hilbert space H and let D be
a dense subset of H. Consider the family of open balls {B (ei, ~) LeI'
where
B(ei'~) = {x E H: lIei -xII < ~l·
Since lIei - ejll = -Ii for i =1= j, it follows that {B(ei, ~)}iel is a pairwise
disjoint family of open sets. Since D is dense in H for each i E I. there exists
some di ED n B(ei, ~). Clearly, the mapping i 1-+ di , from I into D, is one-to-
one and this shows that D has cardinality greater than or equal of the cardinality
of/.
Next, we shall prove the equivalent statements. To this end, assume that H is
an infinite dimensional Hilbert space.
(1) ¢=:::> (2) Let {el, e2,"'} be a countable orthononnal basis for H. Then,
the finite linear combinations of the en with "rational" coefficients is a countable
dense set.
Now, assume that H is separable, and let D be a countable dense subset of H. If
{ei liE! is an orthononnal basis, it follows fonn the first part that I has cardinality at
most that of D, and hence, I is at-most countable. Since H is infinite dimensional,
I must be countable, and so H has a countable orthonomlul basis.
(2) ===? (3) If H has a countable orthononnal basis, then H is linearly isometric
(by Theorem 34.9) to .e 2 (lN) = .e2.
(3) ==? (1) Obvious.
Problem 34.3. Let I be an arbitrary nonempty set alld for each i E I let
ei = Xli}' Show that the family offunctions {edie! is till orthonormal basis for
the Hilbert space .e2(1).
Solution. For each i, let ei = XiiI and note that the family of functions lei liE!
is an orthononnal family. Now, notice that if x = {Xi }iel E .e2(1), then
" ' Ix;! 2 = ~

IIxll 2 = '~ ~ I(x, ei)1 2 .
ie! ie!
This shows that {edie! is an orthononnal basis for the Hilbert space ('.2(1).
Section 34: ORTIlONORMAL BASES 327
Problem 34.4. Let lei }iEi be an orthonormal basis in a Hilbert space and let x be
a unit vector, i.e, IIx II = 1. Show thatfor each k E IN the set {i E I: I(x, ei)1 2:t}
has at most k 2 elements.
Solution. From Parseval's Identity we know that
1 = IIxll2 = L I(x, ei)12.

iEI
Let A = {i E I; 1(x, ei ) 1 2: t }.
If A has more than k2 elements, then by choosing
e + 1 indices from A, we see that
1 = IIxll 2 = 'L...,1(x,
" ei)1 2 2: (k2 + 1)k2"1 > 1,
iEI
which is impossible. Therefore, A has at most k 2 elements.
Problem 34.5. Let M be a closed vector subspace of a Hilbert space H and let
{e;}iEI be an orthonormal basis of M; where M is now considered as a Hilbert
space in its own right under the induced operations. If x E H, then show that
the unique vector of M closest to x (which is guaranteed by Theorem 33.6) is the
vector y = LiE/(X, ei)ei·
Solution. Assume that lei }iEI is an orthonormal basis for a closed subspace M
of a Hilbert space H and let x E H be a fixed vector. Note first that Parseval's
Inequality guarantees that LiEI I(x, ei )1 2 :::: IIx1l 2, and so y = LiE/(X, ei )ei is a
well-defined vector of M.
We claim that x - y ..L M. To see this, let Z be an arbitrary vector of M, and let
Z = LjE/(Z, ej)ej be its Fourier series expansion as a vector of M. Then, we have
(Z,x-y) = (L(z,ej)ej,x- L(x,ei)ei)

JEI iEI
= L(Z, ej)(ej, x) - L L(z, ej)ej, (x, ei)ei)
JEI JEI iEI
= L(Z, ej)(ej, x) - L(z, ej)(ej, x) = O.
JEI JEI
Now, if z is an arbitrary vector of M, then y - Z E M and so x - Y ..L Y - z.

Hence, by the Pythagorean Theorem,
This shows that y is the vector in M closest to x.

Problem 34.6. Let {ell} be an orthonormal basis of a separable Hilbert space.

For each n let 1" = ell+1 - en. Show that the vector subspace generated by the
sequence Un} is dense.
Solution. We need to show that if x -L fll for all n, then x = 0. So, let x be a
vector satisfying x -L (en+1 - en) for each n. That is,
This implies (x, ell+l) = (x, en) for each n. If we let 8 = (x, el), then 8 = (x, ell)
for all n, and by Parseval's Identity
00 00
IIxll2 = l)(x, ell )1 2 = L8 2
n=1 n=1
Therefore, 8 = 0, and hence, IIx II = 0, or x = 0.

Problem 34.7. Show that a linear operator L: HI --+ H2 between two Hilbert
spaces is norm preserving if and only if it is inner product preserving.
Solution. Let L: HI --+ H2 be a linear operator between two Hilbert spaces. If

L is inner product preserving, then
IILxll2 = (Lx, Lx) = (x, x) = IIxll2
holds or IILx II = IIx II for each x E HI, i.e., Lis norm preserving. For the converse,
assume that L is norm preserving. Then, from Theorem 32.6, it follows that
(Lx, Ly) = HIILx + Lyll2 - IILx - Lyll2 + lllLx + ILyll2 -IIlLx - ILyIl2)
= HIIL(x + y)1I 2 - IIL(x - y)1I 2 + I ilL (x + ly)1I 2 - IIlL(x - ly)1I 2)
= Hllx + yll2 -lix - yll2 + llix + lyll2 -llix _ lyll2)
= (x, y).
That is, L is inner product preserving.
Problem 34.8. Show that the vector space .e2(Q) ofall square summable complex-
valued functions defined on a nonempty set Q under the inner product
(x, y) ~ Lx(q)y(q)
qEQ
is a Hilbert space.
Section 34: ORTHONORMAL BASES 329
Solution. The verification of the inner product properties of the function (., .)
are straightforward. We shall show that £2(Q) is a Hilbert space, i.e., complete
under its induced normed.
To see this, assume that Q is an infinite set, and let {XII} S; £2(Q) be a Cauchy
sequence. Since for each n we have xll(q) =1= 0 for at-most countably many q E Q,
there exists an at-most countable subset C of Q such that xll(q) = 0 for all
q E Q \ C and alln. We consider only the case when C is a countable set, say
C = {ql, q2, ... j. For each n, let
Then, it is easy to see that we can consider {Yn j as a Cauchy sequence in £z. The
completeness of £2 implies that {YII j converges to some sequence Y = (YI, Y2, ... )
in £2. If X: Q -+ C is defined by X(qi) = Yi and x(q) = 0 whenever q E Q \ C,
then X E £2(Q) and IIxlI - xII = IIYII - yll -+ 0 holds in £2(Q). This shows that
£z(Q) is a Hilbert space.
Problem 34.9. Let lei liEf be an orthonormal basis ofa Hilbert space H. Show
that the linear operator L: H -+ £z(1), defined by
L(x) = {(x, ei)}iEf,
is a surjective linear isometry.
Solution. Clearly, L is linear and by Parseval's Identity (Theorem 34.2(5», it

is also an isometry. We shall verify next that L is also surjective. To this end,
let P. diEf E £2(1). From 'LiEf IAd 2 < 00, it follows that Ai =1= 0 for at-most
countably many indices i. Assume that {i E I: Ai =1= O} = {Aii' Ai2' ... }. (We
consider only the countable case; the finite case is trivial.) Clearly, 'L:'I IAi" 12 <
00.
From the Pythagorean Theorem, we have
and from this it follows that the series 'L:'I Ai" ei" is norm convergent in H. Let
x = 'L:'I Ai"ei" = 'LiEf Aiei· Then, (x, ei) = Ai for each, i and so L(x) =
{Ai LEf. This shows that L is also surjective, as required.
Problem 34.10. Let {ell} be an orthonormal sequence of vectors in the Hilbert

space L 2[0, 2Jr]. Sl/ppose that for each continuous function f in L2[0, 2Jr] we
have f = 'L:'I (f, ell)e ll · Show that {en} is an orthonormal basis.
330 Chapter 6: Hll..BERT SPACES
Solution. We need only show that the linear span of the set {en} is dense. Let
E> °
andletg E L2[0, 21r]. Since the continuous functions are dense inL 2[0, 21r]
(see Theorem 31.10), there exists a continuous function f E L 2[0,21r] with
IIf - gil ::: E. By our assumption, we have f = L~!(f, en)en and, by Bessel's
Inequality, we know th~t L~! l(f, en )12 < 00. Next, choose an integer m such
p:
that [L~m l(f, ek)1 2 < E, and then let h =
L~~! (f, edek. Then, h is in the
linear span of the sequence {en}, and moreover
m 00
IIg - hll = II g - L(f, ek)ek II ::: IIg - fll +I L (f, edek II

k=! k=m+!
::: IIg - fll +[

00
L l(f, ek)1 2
k=m+!
r
I
< E +E = 2E.
Therefore, the linear span of {en} is dense and hence, {en} is a complete orthonormal
set, i.e., it is an orthonormal basis.
Problem 34.11. Let {¢n} be an orthonormal sequence of vectors in the Hilbert

space L2[0, 21r]. Suppose that for each continuous junction f in L2[0, 21r] we
have IIff = L~! l(f, ¢n)12. Show that {¢n} is an orthonormal basis.
Solution.
andletg E
It suffices to show that the linear span of the set {¢n} is dense. Let E >
L2[0, 21r]. Since the continuous functions are dense in L2[0, 21r], there
°
exists a continuous function f E L 2[0, 21r] with IIf - gil < E.
Now, by our hypothesis, we have IIfII2 = L~! l(f, ¢n)12. Choose an integer
1-
m such that [L~m l(f, ¢k)fp < E, and note that
II g - t(f, ¢d¢k II ::: IIg - fII + II f - t(f, ¢k)¢k II·

k=! k=!
Using once more our hypothesis, we see that
::: IIg - fll +[

00
L l(f, ¢k)1
k=m+J
2
rI
< E +E = 2E.
Therefore, the linear span of {¢n} is dense and hence, {¢n} is an orthonormal
basis.
Section 34: ORTIIONORMAL BASES 331
Problem 34.12. Let {if>I, if>2, ... } be an orthonormal basis of the Hilbert space
L 2(J.L), where J.L is afinite measure. Fix afunction f E L 2 (J.L) and let {ai, a2, ... }
be its sequence of Fourier coefficients relative to {<Pnl. i.e., an = J f if>n dJ.L.
Show that (although the series L:I anif>n ,!eed not converge pointwise almost
everywhere to f) the Fourier series L:I anif>n can be integrated term-by-term in
the sense that for every measurable set E we have
Solution. Let Sn = L~=J anif>n, and note that IIf - snll -+ O. Now, using the
Cauchy-Schwarz inequality, we see that
= ILU- Sn)dJ.L1 ~(Llf-SnldJ.L/

2
ILfdJ.L- L Sn d J.L12
~ 2
L If -snI 2d J.L. L 1 dJ.L
~ IIf - snll 2 J.L*(E) --+ o.
Problem 34.13. Establish the following "perturbation" property of orthonor-

mal bases. If lei liE! is an orthonormal basis and {fi liE! is another orthonormal
family satisfying
L lI ei - fill2 < 00,

iE!
then {fi liE! is also an orthonormal basis.
Solution. Let {ediE! be an orthononnal basis in a Hilbert space H, and let

{fi liE! be another orthononnal family satisfying LiE! lIei - fi 112 < 00. To
establish that the orthononnal family {fi liE! is an orthononnal basis, it suffices
to show that if a vector u satisfies u J.. fi for each i E I, then u = O. So, fix a
vector U E H such that II J.. fi for all i E I.
From LiE! lIei - fill2 < 00, we know that the set J = {i E I: fi =1= e;} is
at-most countable. We distinguish two cases.
CASE I: J isfinite, say J = {kJ, k2 , •.. , ke}.
Let M = {y E H: y J.. fi for all i ¢. J}. Then, M is a closed vector subspace
of H satisfying {fkl' ... , fke} ~ M and {ekl , ... , eke} ~ M. Moreover, we claim
that {ekl ' •.• , eke} must be an orthononnal basis for M. Indeed, if x E M satisfies
z..L ekr forr = e,

1, ... , then (in view x ..L fi =
ei for each i ¢ J) we have x ..L ei
for each i E I. Since {ei liel is an orthononnal basis of H, it follows that x = 0.
Thus, {ekl' ... , ek l 1is (as being an orthononnal basis) also a Hamel basis for M,
and so M is i-dimensional. This implies that {fk l ' ••• , fkl 1is also a Hamel basis.
The latter implies that every ekr is a linear combination of the vectors fk l , ••• , ikc'
e,
Consequently, u ..L ekr for each r = 1, ... , and hence u ..L ei for all i E I. This
implies u = 0, and thus, in this case, {fi }iel is an orthononnal basis.
CASE IT: J is countable, say J = {k l , k2' k 3 , ••• ,}.
In this case, choose a natural number e such that

00
L
j=l+1
lIekj - fk j UZ = 0 < 1,
and let J I = {k l , k2, ... , ke}. Next, define the vectors
00
gr = ekr - L (ek r , fk)ik j , r = 1,2, ... , e.

j=l+1
We claim the following:
• Ifavectorx E H satisfies x ..L grforr = 1,2, ... , iandx..L /jfor j ¢ JI,

then x = 0.
To see this, assume that vector x E H is orthogonal to gr for r = 1, 2, ... , e

and to each fj for j ¢ JI. Then, for j ¢ J lo we have (x, ej) = Cx, ej - fj) and
for each 1 ::s r ::s e, we have
00
(x, ekJ = Cx, gr) + L (ek

j=l+1 .
r, fk)(X, fk) = 0.
Now, from Parseval's Identity, we have
00
IIxII2 = L
iei
ICx, ei)1 2
= L
j=l+1
I(x, ekj - fk)1 2
00
::s [ L lIekj - fkj 112 ] IIxll2 = 01lx1l2.

j=l+1 .
This implies °::s (1 - o)lIx 112 ::s 0, or x = 0, as claimed.

Section 35: FOURlER ANALYSIS 333
Next, we consider the closed vector subspace
M = {y E H: y.l fi for all i ¢; fd.
Clearly, {gl, g2, ... , ge} ~ M. Moreover, if some vector x E M is orthogonal

to g I , g2, ... , g e, then by property (e) we have x = O. This means that the
vector space generated by gl, g2, ... , ge coincides with M. Since the orthogonal
vectors h" fk2' ... , h, belong to M, M is e-dimensional and so Uk, ' fk2' ... , fke}
is a Hamel basis of M. In particular, for each 1 ~ r ~ e the vector gr is a
linear combination of the vectors fkl ' fk2' ... , fk e. This implies u .1 g,. for each
e
1 ~ r ~ and u .1 fJ for j ¢; fl. Using (e) once more, we conclude that u = O.
Therefore, the orthonormal family Ui liEf is an orthonormal basis.
35. FOURIER ANALYSIS
Problem 35.1. Show that sinll x is a linear combination of
{1, sin x, cos x, sin 2t, cos 2t, sin 3x, cos 3x, ... , sin nx, cos nx }.
Furthermore, show that the coefficients of the cosine terms are zero when n is an
odd integel~ and the coefficients of the sine terms are zero when n is an even integer.
Solution. Observe that
elX _ e- 1X
sinx=----
21
Then, using the binomial theorem, we get
1 I:
= (e e-l. )1I = --
lX t II
• II - [ n!( _l)k 1(II-k)r -lkrJ
sm x e -e -
21 (2/)11 k=O k!(n - k)!
1 "[ I( l)k
= __ II
11. - el (II-2k)XJ
(2/)11 6J k!(n - k)!
1
= -(?II
_I)
I:[ knl(-ll
II
k=O .(n
k (cos(n -
l
).
' _ I 2k)x + I sin(n - 2k)x) ]
[n!(-l)k ]
= -"
1
(2/)11
II
6J k!(n-k)!
cos(n-2k)x + I
- " [n!(-l)k
II
(2/)n 6J k!(n-k)!
]
sin(n-2k)x .
Now, observe that if n is odd, then I" = ±I and if n is even, then I" = ±l. Since
sinn x is equal to the real part of the preceding expression, we have the following
two cases:
1 n n'(-l)k
sinn x = -(2)
I
"[ k .
f;o !(n -
/I k)!
cos(n - 2k)X] for n even, and
sinn x
I
= -(2 Ln [ n'(-l/
k .
l)n k=O. '!(n - k)!
Sin(11 - 2k)x
]
for 11 odd.
Problem 35.2. Show that the Dirichlet kernel Dn and the Fejb' kernel Kn
satisfy
-
1 j1f D/I(t)dt = -1 j1f K/I(t)dt = 1.
1r -1f ,1r -1f
Solution. The Dirichlet kernel is given by
1 /I
D/1(t) ="2 + Lcoskt.

k=l
Integrating gives
1
-
j1f D/I(t)dt = 1 + -1 L j1f cosktdt = 1. /1
1r -1f 1r k= 1 -1f
Likewise, the Fejer kernel is defined by
So, integrating and using the previous result on the Dirichlet kernel, we get
-
1 j1f K/I(t)dt = - 1
L
n
-
1 j1f Dn(t)dt = - 1
L
n
1 = 1.
1r -1f n +1 k=O 1r -1f n+1 k=O
Problem 35.3. Let X denote the Banach space of all continuous periodic real-
valuedful1ctions defined on [0, 21r]. Fix some x E [0, 21r] and define the linear
Section 35: FOURIER ANALYSIS 335
functional S,,: X -+ 1R by the formula
S,,(!) = -1 121f fU)Q,,(x - t)dt.

rr 0
Show that the norm of the linear functional S" satisfies
IIS"II = -rr1 121f

0
ID,,(x - t)1 dt.
Solution. The norm of Sn is defined by
IIS"II= sup
IIflloo~l
1- 1 121f
rr 0
fU)D,,(x-t)dtl·
Since IIflloo ::: 1 implies If(t)D(x - t)1 ::: ID(x - t)1 for each t, we see that
1 121f
IISn II ::: - ID,,(x - t)1 dt.
rr 0
Next, we shall establish the reverse inequality. Since the Dirichlet kernel Dn
has period 2rr, it follows that the continuous function
Dn(.x - t)
f( E,t)=~----
ID,,(x - t)1 +E
also has period 2rr with respect to t, and clearly If (E, t) I ::: 1 for each t and all
E > O. This implies
1 121f ID (x t)12
IISnll ::: Sn(!) = - ,,- dt.
rr 0 ID lI (x - t)1 + E
Taking into account Theorem 24.4 and letting E -+ 0+ yields
1 121f
IIS"II ::: - ID,tCx - t)1 dt.
rr 0
Therefore, IISnll = ~ J~1fID,,(x - t)1 dt holds true.

Problem 35.4. Show that the sequence offunctions

I I I I ? I
{(~P, (~pcosx, (~pcos2x, (~pcos3x, (~pcos4x, ... }
is an orthonormal basis in L2[0, n]. Also show that the preceding sequence is Gil
orthogonal sequence offunctions in L2[0, 2n] which is not complete.
Solution. It is easy to verify that the functions

I I I ') J
(~ p, (~P cos x, (~P cos 2x, (~p cos 3x, ...
are mutually orthogonal and of nonn one in L 2 [0, n].

To show that the preceding orthononnal sequence is complete (i.e., that it is an
orthononnal basis), we need to show that if f E L2[0, n] is perpendicular to the
functions 1, cos x, cos 2x, cos 3x, ... , then f = 0. To this end, suppose that a
function f E L2[0, n] satisfies
Ion: f(x) cos nx dx = °

for all n = 0,1,2, .... Define the function g: [0, 2n] -+ 1R by
f(x) if 0:::: x :::: n

g(x) = { f(2n - x) if n < x :::: 2n.
Then, in L 2 [0, 2n], for each n we have
10r n: g(x)cosnxdx
2
(g,cosnx) =
= 1n:
10r f(x)cosnxdx+
On:.
2
f(2n-x)cosnxdx.
The change of variable t = 2n - x gives
(g,cosnx) = Ion: f(t)cosntdt+ Ion: f(t)cosntdt=O.

Next, observe that
= 10rn: g(x) sin nx dx =.10r f(x) sin nx dx + 1n:r n: f(2n -

2 2
(g, sin nx) x) sin nx dx
= Ion: f(t) sin nt dt - Ion: f(t) sin nt dt = 0.

The preceding show that g is perpendicular to every vector of a complete orthogonal

°
sequence of L 2 [0, 2rr]. Therefore, g = and hence, f = 0. Thus, the sequence
l~-, (;r)
(;r) 2*·
2~- cos x, (;r) 2~- cos 3x,
- cos 2x, (;r) ~ ...
is an orthonormal basis of L2[0, rr].

To see that the orthogonal set
is not complete in L2[0, 2rr] notice that the nonzero function sin x is perpendicular
to each of these functions in L 2 [0, 2rr].
Problem 35.5. Show that the sequence offunctions
{(;r2)~'sm.\,
. (2)~'
;r sm 2x, (2;r) ~ sm
. 3x, (2;r) 1·sm 4x,. '" }
is an orthonormal basis of L 2 [0, rr]. Also prove that this set offunctions is an
orthogonal set offunctions in L2 [0, 2rr] which is not complete.
Solution. It is easy to verify that the collection of functions
{( i?)~-smx,
. (?)!; .
i -sm2x, (2)!; .
;r -sm3x, (?)!; .
i -sm4x, '" }
is an orthonormal set of functions of L2[0, rr].

Thus, in order to show that it is an orthonormal basis, we need to show that
if a function f E L 2 [0, rr] is perpendicular to each function of the preceding
set in L2[0, rr], then f = O. To this end, assume that a function f E L2[0, rr]
satisfies
folt f(x) sin nx dx = 0
for alln = 1,2,3, .... Define the function g: [0, 2rr] -+ 1R by
f (x ) if 0:::: x :::: rr
g(x) = { -f(2rr - x) l'f rr<x_ < 2 rr.
Then, in L 2[O, 21l'], for each n we have
(g, sinnx) = !o2rr g(x) sinnx dx

= !orr f(x) sinnx dx -12rr f(21l' - x) sinnx dx
= !orr f(t) sinnt dt + !orr f(t) sinnt dt = 0 + 0 = O.

Next, observe that
2rr
(g,cosnx) = Jor g(x)cosnxdx
=J ro f(x)cosnxdx - Jrrr 2rr

f(21l' -x)cosnxdx
= !orr f(t) cos nt dt - !orr f(t) cos nt dt = O.

Therefore, g is perpendicular to every vector of a complete orthogonal set of
functions inL 2[O, 21l']. Therefore, g = 0 and hence, f = O. Thus, the orthonormal
set of functions
I I I I
{( l)i:sinx
rr (1)i:sin2x ' 1(1)i:sin3x
'rr r
(1)i:sin4x ,
'1l' •••
}
is an orthonormal basis of L2[O, 1l'].

To see that the orthogonal set of functions
I I I I
{( l)i:sinx
:rr (1)'isin2x ' t(1)'isin3x
'rr r ''!(
(1)'isin4x , •••
}
is not complete in L2[O, 21l'], observe that cos x is perpendicular to each of these
functions.
Problem 35.6. The original Weierstrass approximation theorem showed that ev-
elY continuous function of period 21l' can be uniformly approximated by trigono-
metric polynomials. Establish this result.
Solution. Weierstrass originally gave a direct proof, however, the result can be
derived directly from Fejer's Theorem 35.8. Let f be a continuous function of
period 21l' defined on the entire real line. Let E > 0 be fixed. Let {sn} be the
sequence of partial sums of the Fourier series of f and let {ern) be the sequence
arithmetic means.
From Theorem 35.8 we know that the sequence {all} converges uniformly to f
on [0, 2n]. Now, notice that each an is a trigonometric polynomial, and the claim
is established.
Problem 35.7. Find the Fourier coefficients of the function
I ifO::::x<I
fex) ={0 if I:::: x < 2n.
Solution. The Fourier coefficients are given by the formulas
1 1 if 1
ao = - r
n 10
fex) dx = - r dx =-,
2rr
n 10 2
1 ~ 1 if 1
an = - r fex)cosnxdx = - r cosnxdx = -sin(n!f), and
n 10 n 10 nn-
bll = -1
non
l 2rr
fex)sinnxdx = -1 lif 0
1
sinnxdx = --[cos(nI)
nn
-1].
Simplifying yields
ao = '21>
= 2, 4, 6, .. .
all={~/nn
if n
if n= 1,5,9, .. .
-linn if n= 3, 7, 11, ... ,
linn if n = 1,3,5, .. .
bll = { ~/nn if n = 2, 6, 10, .. .
if n = 4, 8, 12, ... .
Problem 35.8. Find the Fourier series of the function
fex) = { sin:, ~ 0:::: x < n

- sm x if n :::: x < 2n.
Solution. The function f is continuous, even, and periodic. Its Fourier coeffi-
cients are given by
rr
ao = 1i
l 2
0
•
sm x dx - 1i'
_ 4
= ~l
rr
an sin x COS nx dx = { ~ rr(II~-l) if n is even
if n is odd, and
bll = O.
340 Chapter 6: IDLBERTSPACFS
So, the Fourier series of / is given by ~ - ~ L::I ~~2~ir. Since this series
converges at every x and the periodic function / is continuous everywhere, it
follows from Corollary 35.9 that the series converges to lex) for each x. That is,
we have
lex) = ~ _ ~ ~ cos2nx
:rc :rc L..
11=1
4n 2 - 1
for each real number x.
Problem 35.9. Show that/or each 0 < x < 2:rc we have
00 sinnx
x =:rc - 2 2 : - - .
n=1 n
Solution. We consider the periodic function /: [0, 2:rc] -+ R defined by
/ (x) = {X ~f 0 :::: x < 2:rc

o If x = 2:rc.
Computing the Fourier coefficients of /, we obtain
211"
ao = IT1 1 0 x dx
2
= 21fx 211"
10 = 2:rc,
all rx
= 1.11" 10
21f
cos nx dx =..L
1111" 10r x d(sin nx)
21f
= n~ [ X sinnx 1
0211" - 10r211" sinnx dx ] = 0, . and
2 2
bn =~ r 11" x sin nx dx = - II~ r 11" x d(cos nx)
~ ~.
2
= _..L[xcosnx
n1l"
1211" _
0 10 r 11" cosnxdx] = -..L[2:rc
n1l"
_1 sinnx 1211"] = _.6.
non
So, the Fourier series of the function / is :rc - 2 L:::O=I Si"nnx. Given that the
function / is continuous at every 0 < x < 2:rc and that the preceding Fourier series
converges for each 0 < x < 2:rc (see Example 9.7), it follows from Corollary 35.9
that
~ sinnx
x=:rc-2L..--
n=1 n
holds for each 0 < x < 2:rc.
x2 n: 2 00 cos nx
=n:x--+2"'- -
2 3 /1
2
f:t
holds for all 0 .:s x .:s 2n:. Letting x = 0 we obtain the formula r:: 1~ = ~2 •
Solution. Consider the periodic function f: [0, 2n:] -7 lR defined by f(x) =

f- n:x. Computing its Fourier coefficients, we get
{2rr , 3 ' 12rr ,

ao = ~ 10 (T-n:x)dx=~(~-rr;-) 0 =_2~-,
all = 1:r 10(2rr(f- -n:x)cosnxdx = ly,

n
and
Therefore, the Fourier series of the function f is ~2 + 2 r::1

co~i'x. Since
this series converges for each x and f is a c<?ntinuous function, it follows from
Corollary 35.9 that
and the desired identity follows.

00 •
x2 = :!n: 2 + 4 ",(cosnx _ n: smnx)
3 ~ n2 n
11=1
holds for each 0 < x < 2n:.
Solution. Consider the periodic function f: [0, 2n:] -7 1R defined by
x2 if 0 .:s x < 2n:

f(x) = { 0
if x = 2n:.
Computing the Fourier coefficients of f, we obtain
2rr
ao -
-1rr
1o
-, - 3rr 12rr -_8.".2
-,v2dv_x3
0
3" ,
342 Chapter 6: IDLBERT SPACES
bll = ~ Jrrr x2sinnxdx = _4,~.

o
Therefore , the Fourier series of the function / is i3 X 2 + 4 L..,1l=1

,"",00 (COSIlX _ rr sinnx).
--;:r- n
Since this series converges for each x (see Example 9.7) and / is continuous at
°
each < x < 2x, it follows from CoroIIary 35.9 that
+4 L (COS
OO
2
x =-x
4 2
-n- nx- x sin nx) ,
3 2 n
n=1
for each °< x < 2x.
Problem 35.12. Consider the "integral" operator T: L 2[O, x] -+ L2[O, x] de-

fined by
T/(x) = 1rr K(x, t)/(t)dt,
where the kernel K: [0, x] x [0, x] -+ IR is given by
~ [sin(n + l)x] sinnt

K(x, t) = L..- 2 •
n=1 n
Show that the norm a/the operator T satisfies·"T" = x /2.

Solution. By Problem 35.5, we know that the sequence of functions
I
{(:)2sinnx: n = 1,2, ... }
is an orthonormal basis for L2[0, x]. Also, as usual, the norm of the operator is
given by
IITII = sup{IIT(f)II: / E L2[0, x] and Ilfll = 1 }.
Now, fix a function / E L2[O, x] with II/II = I, and write

OC I
/ = Lc (:)2sinnx
ll
n=1
in its Fourier expansion relative to the above basis. By Parseval's Identity, we have
00
1If112 = L Icn l2.

n=1
Next, notice that the operator satisfies
Now, notice that the latter expression is the Fourier expansion of T (n

with respect
to the orthonormal basis described at the beginning of the solution. Thus, by
Parseval's Identity, we have
00 00
IIT(f)1I 2 = ~2 L I~;,t S rr; L Ic l2 = rr4211/112,

n
n=1 11=1
from which it follows that II Til S n: /2 holds.

Finally, if 10 (x ) = (1.) ~ sin x, then II 10 II = 1 and by Parseval's Identity we have
rr
Cl = 1 and CII = 0 for n =1= 1, and so IIT(fo)lI- = ~ . Therefore, liT!! 2: n: /2, and
? '
hence, II Til = n: /2.

CHAPTER 7 ______________
SPECIAL lOPICS IN
INTEGRAnON
36. SIGNED MEASURES
Problem 36.1. Give an example of a signed measure and two Hahn decompo-
sitions (A, B) and (AI, B I ) of X with respect to the signed measure sllch that
A i= A I and B i= B I·
Solution. Let X = lR and let I: be the a-algebra of all Lebesgue measurable

sets. Consider the measures JLI, JL2 E M(I:) defined by JLI(E) = A(E n [0,1])
and JL2(E) = A(E n [1,2]) for each E E I: (where A denotes the Lebesgue
measure on lR). Now, consider the signed measure JL = JLI - JL2, and note that
( -00, 1), [1,00») and ([0, 1), (-00,0) U [1,00») are two Hahn decompositions
of X with respect to the signed measure JL.
Problem 36.2. If JL is a signed meaSllre, then show that JL+ /\ JL- = O.

Solution. Let (A, B) be a Hahn decomposition of X with respect to JL. If
E E I:, then note that
0:::: JL+ /\ JL-(E) = JL+ /\ JL-(E n B) + JL+ /\ JL-(E n A)

n B) + JL-(E n A)
:::: JL+(E
= JL(E n B n A) - JL(E nAn B) = o.
Problem 36.3. If JL is a signed measure, then show that for each A E I: we have
00 00
IJLI(A) = sup { L IJL(An)l: {An} is a disjoint sequence of I: with U An = A}.

n=1 n=1
345
346 Chapter 7: SPECIAL IDPICS IN INTEGRATION
Solution. Fix A E :E. From Theorem 36.9, we know that
k k
IJlI(A) = SUP{.I)L(AII)I: {A!, ... , Ad 5; :E is disjoint and UAn 5; A }.
II=! II=!
Also, let
00 . 00
s=suP{LIJl(An)l: {An} 5; :E is disjoint and A=U An }.

n=! 1I=!
Now, let {An} be a pairwise disjoint sequence of :E such that U:! An = A.

"k
Clearly, .L..-1I=! IJl(AII)1 ::s: IJlI(A) holds for each k, and so
00 k
L IJl(An)1 = k~~ L IJl(An)1 ::s: IJlI(A).
n=! n=!
Therefore, s ::s: IJlI(A). On the other hand, if {A!, ... , Ad is a finite pairwise
disjoint collection of :E satisfying U~=! An 5; A, then
k
A = A! U ... U Ak U (A \ U Ak) U (2S U (2S ••• ,
n=!
and so
k k k
LIJl(An)1
n=!
::s: LIJl(An)1
n=!
+ IJl( A \ UAn) 1+ Jl«2S) + Jl«2S) + ... ::s: s.
n=!'
Consequently, IJlI(A) ::s: s also holds. Thus, IJlI(A) = s, as claimed.
Problem 36.4. Verify that if Jl and v are two finite signed measures, then the
least upper bound Jl V v and the greatest lower bound Jl/\ v holds in M(:E) are
given by
Jl v v(A) = sup{Jl(B) + v(A \ B): B E :E and B 5; A}, and

Jl/\ v(A) = inf{Jl(B) + ~(A \ B): B E :E and B 5; A}
for each A E :E.

Section 36: SIGNED MEASURES 347
Solution. The proof parallels the one of Theorem 36.1. We shall verify that if
f.L, v E M(2:), then the formula
w(A) = inf{f.L(B) + v(A \ B): B E:£ and B ~ A}, A E 2:,
defines a finite signed measure (Le., w E M(2:», and that w is the greatest lower
bound of f.L and v in M(2:).
Since f.L and v are both bounded from below (and also both bounded from
above), it follows that w(A) E 1R for each A E 2:. Clearly, w(¢) = 0 holds.
Next, we shall establish that w is a-additive. To this end, let (All) be a pairwise
disjoint sequence of 2: and let A = U:I An. If B E 2: satisfies B ~ A, then
00 00
f.L(B)+v(A\B) = f.L(UAllnB)+v(U(An\AllnB))
11=1 n=1
00
= '2]f.L(A Il n B) + v(AIl \ All n B)]

n=1
00
2: Lw(A n),
11=1
and so w(A) 2: L~=I w(AIl) holds. For the reverse inequality, let 8 > O. Then,
for each n pick some Bn E 2: with BII ~ All and
Obviously, (Bnl is a pairwise disjoint sequence of 2:. Put B = U:I BII ~ A,

and note that U:I(A n \ BIl ) = A \ B holds. Moreover,
00 00
w(A) ::: f.L(B) + v(A \ B) = f.L(U Bn) + v(U(A Il \ Bn))

n=1 11=1
00
= L[f.L(Bn) + v(An \ B Il )]
n=1
00 00
::: L[w(An) + f.]= Lw(An) + 8.

n=1 11=1
Since 8 > 0 is arbitrary, we infer that w(A) ::: L:1 w(A II ) also holds, and so
WE M(2:).
348 Chapter 7: SPECIAL IDPICS IN INTEGRAnON
. Finally, we shall establish that w is the greatest lower bound of f.L and v in
M(~). Note first that w is a lower bound for both f.L and v. Indeed, if A e ~,
then (by letting B = A), we see that
w(A) ~ f.L(A) + v«(25) = f.L(A) and w(A) ~ f.L«(25) + v(A) = v(A).
On the other hand, if n e M(~) satisfies n ~ f.L and n ~ v and A e ~, then

for each B e ~ with B S;. A we have
n(A) = nCB) + n(A \ B) ~ f.L(B) + v(A \ B),

from which it follows that n(A) ~ w(A), i.e., n ~ w. This shows that w = f.L /\ v
holds in M(~).
Problem 36.5. Let A be the Lebesgue measure on the Lebesgue measurable

subsets of JR.. If f.L is the Dirac measure, defined by f.L(A) 0 if 0 ¢. A and =
JL(A) = 1 if 0 e A, describe A V JL and A /\ JL.
Solution. If A = 1R \ to}, B = to}, and E is an arbitrary Lebesgue measurable

set, then
o ~ A /\ JL(E) ~ A(E n B) + f.L(E n A) = 0
holds. That is, A /\ JL = O. Moreover, we have

A v JL = A v f.L + A /\ f.L = A + JL.
Problem 36.6. Show that the collection of all a-finite measures forms a dis-
tributive lattice. That is, show that if JL, v, and ware three a-finite measures,
then
(JL v v) /\ W = (JL /\ w) v (v /\ w) and (JL /\ v) V W = (JL v w) /\ (v V w).
Solution. We shall show first that every vector lattice satisfies the distributive
law. To do this, we shall use the identity (a) of Problem 9.1.
Let x, y, and z be elements in a vector lattice. Since x v Y 2:: x, it follows
that (x v y) /\ z 2:: x /\ z, and similarly (x v y) /\ z 2:: y /\ z. Thus,
(x v y) /\·z 2:: (x /\ z) V (y /\ z).
On the other hand, if u = (x /\ z) V (y /\ z), then u 2:: x /\ Z = X + Z -x v z holds.

Hence, x ::: u-z+xvz ::: u-z+(xvy)vz, and similarly y ::: u-z+(xvy)vz.

It follows that x v y ::: u - z + (x v y) v z, and so
(x A z) v (y A z) = u 2: x v Y + z - (x v y) v z = (x V y) A z.
Therefore, (x V y) A Z = (x A z) V (y A z) holds. The other identity can be

established in a similar manner.
Now, let {XII} S; 2: satisfy tJ,(X n) < 00, v(Xn) < 00, w(X II ) < 00 for all n,
and XII t X. If 2: n = {A n X n: A E 2: }, then clearly tJ" v, and w (restricted
to 2: 11 ) belong to the vector lattice M(2: n ). Thus, if E E 2:, then
(tJ, V v) A w(E n Xn) = (tJ, A w) V (v A w)(E n Xn)
and
(tJ, A v) V w(E n Xn) = (tJ, V w) A (v V w)(E n XII)

hold. To finish the proof note that En Xn t E, and then use the "order continuity"
of the measure (Theorem 15.4).
Problem 36.7. If 2: is a a-algebra of subsets of a set X and tJ,: 2: -+ 1R* is a

signed measure, then show that
Solution. Assume that tJ,: 2: --+ 1R* is an arbitrary signed measure. Let E be
in AJ1.+ n AJ1.- and let A E 2: be an arbitrary set. Then,
1tJ,I(A) = tJ,+(A) + tJ,-(A)

= [tJ,+(A n E) + tJ,+(A n E + [tJ,-(A n E) + tJ,-(A n E
C
)]
C
)]
= [tJ,+(A n E) + tJ,-(A n E)] + [tJ,+(A n E + tJ,-(A n E

C
)
C
)]
= 1tJ,I(A n E) + 1tJ,I(A n E
C
),
and so E E AIJ1.I, i.e., AJ1.+ n Aw S; AIJ1.I·

For the reverse inclusion, let E E AIJ1.I' If A E 2: is arbitrary, then note that
tJ,+(A) + tJ,-(A) = 1tJ,I(A) = 1tJ,I(A n E) + 1tJ,I(A n E C )

= [tJ,+(A n E) + tJ,+(A n E C
)] + [tJ,-(A n E) + tJ,-(A n E C
)].
350 Chapter 7: SPECIAL TOPICS IN INTEGRAnON
,C(A) :s J.L"""(A n E) + /-C(A nEe) both hold, it follows from the preceding
equality that
which shows that E E Ap.+ n Ap.-. Thus, A1p.1 S; Ap.+ n Ap.-, and consequently,
AIp.1 = Ap.+ n Ap.- holds, as desired.
Problem 36.8. Let J.L and v be two measures on a a-algebra :E with at least one
01 them finite. Assume also that S is a semiring such that S S; :E, XES, and that
the a-algebra generated by S equals :E. Then show that J.L = von :E if and only
ifJ.L = von S.
Solution. Assume that J.L is finite and that J.L = v on S. If we consider the
measure space (X, S, J.L), then it is easy to see that S S; :E S; Ap. holds. Now,
apply Theorem 15.10 to get that J.L = J.L* = v holds on :E.
Problem 36.9. Let (X, S, J.L) be a measure space, and let I ELI (J.L). Then show
that
v(A) i
= I dJ.L
lor each A E AJl. defines afinite signed measure on Aw Also, show that
holds lor each A E Aw
Solution. If {An} is a pairwise disjoint sequence of Ap. satisfying A = U:'I An,

then lim 2:::7=1 IXAi = IXA and 12:::;1=1 IXAil :s III holds for each n. Thus,
from the Lebesgue Dominated Convergence Theorem, it follows that
Therefore, v is a finite signed measure.

Now, note that if
A = {x EX: I(x) ::: O} and B = {x EX: I(x) < OJ,

then it is easy to see that (A, B) is a Hahn decomposition of X with respect to

v. Since / XEnA = /+ XE holds, we see that
foreach E E Aw Theprooffor v- is similar. The absolute value formula follows

from the identity Ivl = v+ + v-.
Problem 36.10. Let v be a signed measure on 2:. A/unction /: X -+ lR is said

to be v-integrable if / is simultaneously v+ - and v- -integrable (in this case, we
write J/ dv = J/ dv+ - J/ dv-). Show that afunction / is v-integrable ifand
only if/ E L1(lvi).
Solution. Assume that / is simultaneously v+ - and v- -integrable. We can

assume that lex) 2: 0 holds for all x EX. Since each set ofthe form {x E X: a .:s
lex) < b} belongs to AI1-+ n AI1-- = 1\111-1 (for this identity see Problem 36.7), we
see that there exists a sequence {4>n} of simultaneously v+ - and v- -step functions
such that 4>n(;r) t lex) holds for all x E X; see the proof of Theorem 17.7.
Clearly, each 4>11 is a 11,L1-step function and from
we see that / is IJLI-integrable and that J/ dlJLI = J/ dJL+ + J/ dJL- holds.

For the converse, assume that / belongs to LI (Ivi). We can assume that
lex) 2: 0 holds for each x. Note first that if / = XA for a lvi-measurable set
A with Ivl*(A) < 00, then there exists (by Theorem 15.11) some B E 2: with
A S; Band Ivl*(A) = Ivl*(B). It follows that Ivl*(B \ A) = 0, and in view of
o .:s v+ .:s lvi, we have (v+)*(B \ A) = O. Thus, B \ A is a v+ -measurable set,
and consequently, A = B \ (B \ A) is also v+ -measurable. This shows that XA
is v+ -integrable.
Now, choose a sequence {4>II} of lvi-step functions with 0 .:s 4>1I(X) t lex)
for each x. By the previous discussion, {4>II} is a sequence of v+ -step functions.
Moreover,
f 4>n dV+ .:s f 4>11 dlvl .:s f/

dlvl < 00
holds for all n. Thus, / is v+ -integrable.

The v- -integrability of / can be established in a similar manner.
352 Chapter 7: SPECIAL WPICS IN INTEGRATION
Problem 36.11. Show that the Jordan decomposition is unique in the following
sense. If v is a signed measure, and /LI and /L2 are two measures such that
v = /LI - /L2 and /LI 1\ /L2 = 0, then /LI = v+ and /L2 = V-.
Solution. First, we shall establish that v+ = /LI holds. Start by observing that
v ::: /LI implies v+ ::: /LI.
Now, let E E :E. If v+(E) = 00, then v+(E) = /LI(E) = 00 holds trivially.
Thus, we can suppose v+(E) < 00. Since vee) = /LI (E) - /L2(E) ::: v+(E) <
00, it follow~ that /LI(E) < 00. Let e > 0. Then, in view of
°= /LI 1\ /L2(E) = inf{/LI(E \ B) + /L2(B): B E ~ and B S;;; E },
there exists some B E ~ with. B S;;; E and /LI (E \ B) + /L2(B) < e. Thus,
v+(E) = sup{ v(F): F E ~ and F S;;; E } 2: v(B) = /LI (B) - /L2(B)

2: /LI(B) - e = /LI(E) - /LI(E \ B) - e 2: /LI(E) - 2e
holds for all e > 0. That is, v+(E) 2: /LI (E) for each E E ~, and therefore
v+ = /LI holds. For the other identity note that
Problem 36.12. In a vector lattice Xn t x means that XII +I ::: XII for each nand
that X is the greatest lower bound of the sequence {x n}. A normed vector lattice is
said to have a-order continuous norm if Xn to implies lim IIxnll =0.
a. Show that every Lp(/L) with 1 ::: p < 00 hps a-order continuous norm.
b. Show that Loo([O, 1]) does not have a-order continuous norm.
c. Let ~ be a a -algebra of sets, and le't {/Ln} be a sequence of M (~) such that
/Ln t /L. Show that lim /Ln(A) = /L(A) holds for all A E ~.
d. Show that the Banach lattice M(~) has a-order continuous norm.
Solution. (a) Note first that fn t f in L p(/L) is equivalent to fn t f a.e.

(why?). If for some 1 ::: p < 00 a sequence {fn} of Lp(/L) satisfies fn t
a.e., then
°
holds by virtue of the Lebesgue Dominated Convergence Theorem.

°
(b) If fn = X(O,l), then fn,J.. holds in Loo([O, 1]). However, note that IIfnlloo =
1 holds for each n.
Section 37: COMPARING MEASURFS AND THE RADON-NIKODYM THEOREM 353
(c) Let ILII +

IL in M(L;). Then 0 ::: ILl - ILII t ILl - IL in M(L;). By
Theorem 36.2, it follows that ILI(A) - ILn(A) t ILI(A) - IL(A) holds for each
+
A E L;. Thus, ILn(A) IL(A) holds for each A E L;.
+
(d) If ILn 0 in M(L;), then from part (c) it follows that lilLI/II = ILn(X) O. +
Problem 36.13. Prove the following additivity property of the Banach lattice
M(L;): If IL, v E M(L;) are disjoint (i.e., IILI t\ Ivl = 0), then IIIL+vll = IIILIl + II v II
holds.
Solution. If IILI t\ Ivl = 0 holds in M(L;), then IlL + vi = IILI + Ivl holds
(see Problems 9.2 and 9.3). Thus, IIIL + vii = IlL + vl(X) = IILI(X) + Ivl(X) =
II IL II + II v II·
Problem 36.14. Let L; be a a-algebra of subsets of a set X and let {ILn} be a
disjoint sequence of M(L;). If the sequence of signed measures {ILII} is order
bounded, then show that lim IIILIIII = O.
Solution. Let {ILn} be a disjoint sequence of the Banach lattice M(L;) such that
for some 0 ::: IL E M(L;) we have IILIII ::: IL for each n. From IILn I t\ IILm I = 0 for
n :j:. m, we see that
k k
L IILn I = V IILn I ::: IL
1/=1 n=1
holds for each k. In particular, we have
k k k
L IIILnll = L IILnl(X) ::: [V IILIII](X)::: IL(X) < 00
11=1 1/=1 11=1
holdsforeach n,andso I::I IIILnll < 00. The latter easily implies lim IIILnll = O.
37. COMPARING :MEASURES AND THE RADON-NIKODYM

THEOREM
Problem 37.1. Verify the following properties of signed measures:

a. IL« IL·
b. v« IL and IL «
w imply v « w.
c. If 0 ::: v ::: IL, then v IL· «
d. «
If IL 0, then IL O. =
354 Chapter 7: SPECIAL 10PICS IN INTEGRATION
Solution. (a) From Theorem 36.9, we have IJ.L(A)I :::: IJ.LI(A), and so if IJ.LI(A) =
o holds, then J.L(A) = 0 likewise holds. That is, J.L J.L. «
(b) Assume v « «
J.L and J.L w and Iwl(A) = O. Theorem 37.2 applied twice
shows that IJ.LI(A) = 0 and Ivl(A) = O. Hence, v « w holds.
«
(c) Let 0 :::: v :::: J.L. If IJ.LI(A) = J.L(A) = 0, then clearly v(A) = 0, and so v J.L
holds.
(d) Let J.L « O. Since the zero measure assumes the zero value at every A E I:,
it follows that J.L(A) = 0 hol~s for every A E I:. This means that J.L = O.
Problem 37.2. Verify the following statements about signed measures on a a-

algebra I: of sets:
1. If J.L « w and v « w, then IJ.LI + Ivl « w.
2. If J.L 1. wand v 1. w, then IJ.LI + Ivl 1. w.
3. If J.L « wand Ivl :::: IJ.LI, then v « w.
4. If J.L 1. wand Ivl :::: IJ.LI, then v 1. w.
5. If v « J.L and v 1. J.L, then v = O.
Solution. (1) This follows immediately from Theorem 37.2.

(2) Since J.L 1. w, there exists (by Theorem 37.5) some AlE I: with Iwl(A 1) =
11l1(AD = O. Similarly, there exists some A2 E I: with Iwl(A 2) = Ivl(A~) = O.
Put A = Al UA2 and B = (Al UA2r = A~ nA~. Then A, B E I:, AUB = X,
An B = 0, Iwl(A) = 0, and CiJ.L1 + Ivl)(B) = O. By Theorem 37.5 we infer that
w 1. IJ.LI + Ivl holds.
(3) This follows easily from Theorem 37.2.
(4) This follows immediately from Theorem 37.5.
(5) Since v 1. J.L, there exists some A E I: such that Ivl(A) = IJ.LI(A C ) = O. By
v « 11 and Theorem 37.2, Ivl(A C ) = 0, and so Ivl(X) = Ivl(A) + Ivl(N) = O.
That is, Ivl = 0, so that v = O.
Problem 37.3. Let J.L and v be two measures on a a-algebra I:. Ifv is a finite
measure, then show that the following statements are equivalent.
a. v« J.L holds.
b. For each sequence {An} of I: with lim J.L(An) = 0, we have lim v(An) = O.
c. For each E > 0 there exists some 0 > 0 (depending on E) such that whenever
A E I: satisfies J.L(A) < 0, then v(A) < E holds.
Solution. (a) ==> (b) If (b) is not true, then there exists some e > 0 and some
sequence {All} of I: such that J.L(An) < 2- n and v(An) > e for each n. Set
Section 37: COMPARING MEASURES AND TIlE RADON-NIKODYM TIlEOREM 355
OC 00
jJ,(A)::s I:jJ,(Ai)::S L t. = 2'-"

i=n ;=n
holds for each n, and so jJ,(A) = O. However, from Theorem 15.4(2), we see that
v(A) ::: 8, contrary to v « jJ,. Hence, (a) implies (b).
(b) =? (c) If (c) is not true, then there exist some 8 > 0 and a sequence (A,,}
of :E such that jJ,(A,,) <
tradi cts (b).
* and v(A,,) ::: 8 hold for all n. Clearly, this con-
(c) =? (a) Let A E :E satisfy jJ,(A) = O. Given 8 > 0, choose some 0 > 0 so
that (c) is satisfied. In view of jJ,(A) < 0, it follows that v(A) < 8. Since 8 > 0
is arbitrary, v(A) = 0, and so v « jJ, holds.
Problem 37.4. Let jJ, be a finite measure, and let (v n } be a sequence oj finite
measures (all on :E) such that Vn « jJ, holds Jar each n. Furthermore, assume
that lim v,,(A) exists in IRJor each A E :E. Then, show that;
a. For each E > 0 there exists some 0 > 0 such that whenever A E :E satisfies
jJ,(A) < 0, then vn(A) < E holds Jar each n.
b. The set Junction v::E -7 [0,00], defined by v(A) = lim v,,(A) Jar each
A E :E, is a measure such that v « jJ,.
Solution. (a) From Problem 31.3, we know that :E under the distance dCA, B) =
jJ,(AIlB) is a complete metric space. From Vk « jJ, and the inequality
iteasilyfollowsthatthefunction Vk::E ~ IR is well defined (Le., vk(A) = vk(B)

holds whenever jJ,(AIlB) = 0) and is continuous.
Now, let 8 > O. Define
Ck = {A E :E: Iv,,(A) - vm(A)1 ::s 8 for all n, m ::: k }.
Note that each Ck is closed and that :E = U~, Ck holds. By Baire's Category
Theorem 6.18), we have Ck =1= 0 for some k. Thus, there exist Ao E Ck and
0, > 0 such that A E :E and jJ,(AIlAo) < 0, imply A E Ck.
From Vi « jJ, (1 ::s i ::s k) and the preceding problem, there exists some
o < 0 < 0, such that A E :E and jJ,(A) < 0 imply vi(A) < 8 for all 1 ::s i ::s k.
356 Chapter 7: SPECIAL IDPICS IN INTEGRAnON
Now, if A E 2: satisfies IL(A) < 8, then A U (Ao \ A) = A U Ao satisfies

IL( (A U Ao).6.Ao) ::: IL(A) < 81, and so
IVII (A) - vk(A)1 = I(vn - vk)(A U Ao) - (vn - vk)(Ao \ A)I

::: I(v n - vk)(A U Ao)1 + I(v n - vk)(Ao \ A)I ::: 28
holds for all n > k. Thus, A E 2: and IL(A) < 8 imply
for all n > k (and aliI::: n ::: k).

(b) Let A = U:1 All with the sequence {An} of 2: pairwise disjoint, and let
8 > O. Choose some 8 > 0 so that statement (a) is satisfied. Next, choose some
m so that IL(A \ U7=1 Ai) < 8 holds for all n 2: m. Then,
n n
IVk(A) - ?=
1=1
vk(Ai) WAi) <
1= Vk( A \ 1=1 8
holds for all k and all n 2: m. Thus, Iv(A) - 2::7=1 v(Ai)1 ::: 8 holds for all
n 2: m, and so Iv(A) - 2::~1 V(Ai) I :::
8. Since 8 > 0 is arbitrary, we see that
v(A) = 2:::1 v(An). Thus, v is a measure, and from part (a) and the preceding
problem it follows immediately that v « IL holds.
Problem 37.S. Let {VII} be a sequence of nonzero finite measures such that
lim vlI(A) exists in1Rfor each A E 2:. Show that v(A) = lim VII (A) for A E :E is
a finite measure.
Solution. Consider the set function IL: 2: ~ [0, (0) defined by
00
IL(A) = '"
L...t v,,(A)
v,,(X)
Tn,
n=1
and note that IL is in fact a measure. In addition, note that Vn « IL holds for each
IZ.Now, invoke part (b) of the preceding problem to conclude that the set function
v is also a measure.
Problem 37.6. Verify the uniqueness of the Radon-Nikodym derivative by prov-

ing the following statement: If ("X, S, IL) is a measure space and f E L 1(1L)
satisfies fA f dlL = Ofor all A E S, then f = 0 a.e.
Section 37: COMPARING MEASURES AND TIlE RADON-NIKODYM TIIEOREM 357
Solution. From the given condition, it is easy to see that fAf dfJ- = 0 must hold
for each a-set A. Now, consider the measurable sets
A = Ix EX: f (x) > 0} and' B = Ix EX: f (x) < 0 }.
By Problem 22.7 we know that A and B are both a-finite sets. Now, in view of
fAf dfJ- = fBf dfJ- = 0, it follows from Problem 22.13 that fJ-*(A) = fJ-*(B) = O.
Therefore, f = 0 a.e. holds.
Problem 37.7. This problem shows that the hypothesis of a-finiteness of fJ- in
the Radon-Nikodym Theorem cannot be omitted. Consider X = [0, 1], :E the
a -algebra of all Lebesgue measurable subsets of [0, 1], v the Lebesgue measure
on :E and fJ- the measure defined by fJ-(C/J) = 0 and fJ-(A) = 00 if A =1= C/J.
(Incidentally, fJ- is the largest measure on :E.) Show that:
a. v is afinite measure, fJ- is not a-finite, and v « fJ-.
b. There is no function f ELI (fJ-) such that v(A) = fA f dfJ- holds for all
A E :E.
Solution. (a) Note that fJ-(A) = 0 means A = C/J, and so v « fJ- holds.
(b) Observe that LI(fJ-) = (OJ.
Problem 37.S. Let fJ- be afinite signed measure on :E. Show that there exists a
unique function fELl (1fJ-1) such that
fJ-(A) = i f dlfJ-1
holds for all A E :E.
Solution. The conclusion follows from the Radon-Nikodym Theorem by ob-

serving that fJ- « 1fJ-1 holds.
Problem 37.9., Assume that v is a finite measure and fJ- is a a-finite measure
such that v « fJ-. Let g = dv/dfJ- E LI(fJ-) be the Radon-Nikodym derivative of
v with respect to fJ-. Then show that:
a. If Y = (x E X: g(x) > OJ, then Y n A is a fJ--measurable set for each
v-measurable set A.
b. If f E LI(v), then fg E LI(fJ-) and f f dv = f fg dfJ- holds.
Solution. (a) Note first that by Theorem 37.3, :E ~ Ap. ~ Av holds, and that
Y E Ap..
358 Chapter 7: SPECIAL TOPICS IN INTEGRATION
First consider the case when A E Av satisfies A S;; Y and v*(A) = 0. By

Theorem 15.11 there exists some B E 2: with A S;; B and v*(B) = 0. Now, if
f,L*(B n Y) > 0, then we have the contradiction 0= v*(B n Y) = fBnr g df,L >
°
(see Problem 22.13). Consequently, f,L*(B n Y) = f,L*(A) = holds, and so
°
A EAJl."
Now, let A E Av. Choose some B E 2: with A S;; B and v*(A) = v*(B).
Thus, v*(B \ A) = 0, and so (B \ A) n Y E AJl." Now, note that
AnY = B n Y \ (B \ A) n Y E AJl."
(b) It follows immediately from Problem 22.15.
Problem 37.10. Establish the chain rulefor Radon-Nikodym derivatives: If w

is a a-finite measure and v and f,L are two finite measures (all on 2:) such that
v « f,L and f,L « w, then v « wand
dv dv df,L
-dw = -df,L . -dw (w-a.e.)
holds.
Solution. Clearly, v « f,L and 2: S;; Aw S;; AIL S;; Av. Put f = ~~ E LI(f,L)
and g = ~~ ELI (w). If A E 2:, then by part (b) of the preceding problem, we
infer that
This combined with the Radon-Nikodym Theorem shows that
dv
-df,L = fg, w-a.e.
Problem 37.11. All measures considered here will be assumed defined on afixed
a-algebra 2:.
a. Call two measures f,L and v equivalent (in symbols, f,L == v) if f,L « v
and v « f,L both hold. Show that == is an equivalence relation among the
measures on 2:.
b. If f,L and v are two equivalent a-finite measures, then show that AIL = Av.
c. Show that if f,L and v are two equivalent finite measures, then
-df,L .
dv df,L
dv
- = 1 a.e. holds.
Section 37: COMPARING MEASURES AND THE RADON-NIKODYM THEOREM 359
d. If J1. and v are two equivalent finite measures, then show that
dJ1.
f 1-+ f· dv'
from L I (J1.) to L I (v), is an onto lattice isometry. Thus, under this identifi-
cation L I(J1.) = LI(v) holds.
e. Generalize (d) to equivalent a-finite measures. That is, if J1. and v are two
equivalent a-finite measures, then show that the Banach lattices L I (J1.) and
LI (v) are lattice isometric.
f. Show that if J1. and v are two equivalent a-finite measures, then the Banach
lattices L p(J1.) and L p(v) are lattice isometric for each I :s p :s 00.

(b) It follows immediately from Theorem 37.3.
(c) Use the relation v « J1. « v and the preceding problem.
(d) Let f 1-+ f·~~ = T(f). Since ~~ E L I (v),itfollowsfromProolem37.9(b)
J J
that T(f) = f· ~~ E LI(v) and f dJ1. = f· ~~ dv hold for each f E L I(J1.).
Thus, T defines a mapping from L I(J1.) to LI(v) which is clearly linear. Since
~~ 2: 0 holds, it follows that
T(lfD = If I. ~~ = If· ~~ 1= IT(nl

and
IIT(nll l = fit· ~~ Idv = flfl. ~~ dv = flfl dJ1. = Ilflll

hold for each f ELI (J1.). Thus, T: L 1(J1.) -+ L I (v) is a lattice isometry. To see
that T is also onto, note that if g E LI(V), then g. ~~ E L 1(J1.) and by part (c),
we see that
T (g . ~~) = g . ~~ . ~~ = g.
(e) Let {Ell} be a pairwise disjoint sequence of:E such that U:I E,,=X, J1.(E,,) <
00, and v(En) < 00 for each n. Let
be the onto lattice isometry determined by part (d) previously. Now, it is a routine
matter to verify that T: L I (J1.) -+ L 1(v) defined by
00
T(f) = LT,,(JXE,J
,,=1
for each f ELI (J1.) is an onto lattice isometry.

(f) Suppose first that /L and v are finite. Then,
/~ /·e~)~
is a lattice isometry from Lp(/L) onto Lp(v) for each 1 ::: p < 00. Now, if /L
and v are u-finite, then use the arguments of part (e) to establish that L p(/L) and
L p( v) are lattice isometric.
If p = 00, then from part (b) it follows that Loo(/L) = Loo(v) holds, and so in
this case the identity operator is a lattice isometry.
Problem 37.12. Let /L be a u-finite measure, and let AC(/L) be the collection 0/
allfinite signed measures that are absolutely continuous with respect to /L; that is,
AC(/L) = {v E M(2:): v «/L}.
a. Show that AC(/L) is a norm closed ideal 0/ M(2:) (and hence AC(/L), with
the norm IIvll = Ivl(X), is a Banach lattice in its own right).
b. For each / E Lt (/L),let /L I be thefinite signed measure defined by /L I(A) =
fA / d/L/or each A E 2:. Then show that / H- /LI is a lattice isometry
from Lt(/L) onto AC(/L).
Solution. (a) Clearly, AC(/L) is an ideal of M(2:). If {vn} is a sequence of

AC(/L) satisfying Vn -+ v in M(2:), then vn(A) -+ v(A) holds for each
A E 2:. Problem 37.4 shows that v E AC(/L). Thus, AC(/L) is a closed vector
sublattice of M(2:), and hence, a Banach lattice in its own right.
(b) Clearly, / 1---+ /L I is a linear operator. By Problem 37.6 this operator is
one-to-one. From Problem 36.9, it follows that / ~ /L I is a lattice isometry,
and the Radon-Nikodym Theorem implies that it is also onto.
Problem 37.13. Let 2: be a u -algebra 0/ subsets 0/ a set X and /L a measure on

2:. Assume also that 2:* is au-algebra 0/ subsets 0/ a set Y and that T: X -+ Y
has the property that T- t (A) E 2: for each A E 2:*.
a. Show that v(A) = /L(T-t(A))/or each A E 2:* is a measure on 2:*.
b. 1// E Lj(v), then show that / 0 T E Lt(/L) and
1/ dv = Ix / 0 T d/L.
c. 1/ /L is finite and w is a u-finite measure on 2:* such that v « w, then show

that there exists a/unction g E L1(w) such that
holds/or each / E Lt(v).

Section 37: COMPARING l\1EASURES AND THE RADON-NIKODYM THEOREM 361

(b) Note first that if A is a v-null set, then T-l(A) is a /L-null set. Indeed,
if v*(A) = 0 holds, then there exists (by Theorem 15.11) some B E 2;* with
A S;; B and v(B) = O. Therefore,
Now, let A bea v-measurable set with v*(A) < 00. Choose some B E 2;* with
A S;; B and v*(B) = v*(A). Since v*(B \ A) = 0, it follows from the preceding
discussion that /L*(T-l(B \ A») = O. Thus, T-l(A) = T-1(B) \ T-1(B \ A) is
/L-measurable, and moreover,
It follows that for every v-step function ¢ we have ¢ 0 TEL 1(/L), and Jy¢ dv =
Jx ¢ 0 T d/L. An easy continuity argument can complete the proof.
(c) It follows immediately from part (b) and Problem 37.9.
Problem 37.14. Let (X, S, /L) be a a-finite measure space, and let g be a mea-
surable function. Show that if for some 1 :::: p < 00 we have f gEL 1(/L) for all
t
f E Lp(/L), then g E Lq(/L), where ~ + = 1.
Also, show by a counterexample that for 1 < p < 00 the a-finiteness of /L
cannot be dropped.
Solution. We can assume that g ::: 0 holds (why?). Then the formula F(f) =
J f g d/L for f E L p(/L) defines a positive linear functional on L p(/L). By
Theorem 40.10, F is continuous. Now, by Theorems 37.9 and 37.10 there exists
some h E Lq(/L) such that Jfgd/L = Jfhd/L for each f E Lp(/L). This
implies (how?) JA (g - h) d/L = 0 for each measurable subset A. Now, a glance
at Problem 22.13 guarantees that g = h a.e. holds.
The a-finiteness of /L cannot be dropped. Consider X = (0,00) with the
measure /L defined on the a-algebra P(X) by /L(A) = 00 if A =1= (/J and
/L«(/J) = O. Then for 1 < P < 00 we have L1(/L) = Lp(/L) = Lq(/L) = {O}, and
Loo(/L) = B(X), the bounded real-valued functions on X. On the other hand, if
g(x) = x, then fg = 0 E Ll(/L) holds for all f E Lp(/L) (1 :::: p < 00), while
g ~ Lq(/L).
Problem 37.15. Let (X, S, /L) be a a-finite measure space, g a measurable func-
tion, and 1 :::: p < 00. Assume that there exists some real number M > 0 such
that¢g E Ll(/L) and J¢gd/L:::: MII¢lI p holds for every step function ¢. Then,
show that:
a. g E Lq(f.L), where ~ +~ = 1, and

b. Ilg df.L ::: Mllfllq holds lor all IE Lp(f.L).
Solution. Let L denote the vector space of all step functions. The given con-
ditions show that the function F: L -+ JR, defined by F (¢) = I ¢ g d f.L, is a
continuous linear functional. Since L is dense in Lp(f.L) (Theorem 31.10), it
follows that F has a continuous extension (which we shall denote by F again)
to all of Lp(f.L). By Theorems 37.9 and 37.10 there exists some h E Lq(f.L) such
that FU) = Ilh df.L holds for all IE Lp(f.L). Clearly,
IFU) = I If Ih df.L I::: Mll/lip

holds for all IE Lp(f.L).
To complete the proof, it suffices to show that g = h a.e. holds. To see
this, let E E All satisfy f.L*(E) < 00. Then, consider the step function ¢ =
xESgn (g - h) E L, and note that I ¢(g - h) df.L = 0 implies IE Ig - hi df.L = O.
That is, g = h a.e. holds on E; see Problem 22.13. Since f.L is a-finite, we see
that g = h a.e. holds on X.
Problem 37.16. Let f.L be a Borel measure on JRk and suppose that there exists
a constant c > 0 such that whenever a Borel set E satisfies A(E) c, then =
f.L(E) =c. Show that f.L coincides with A, i.e., show that f.L A. =
Solution. Assume that the Borel measure f.L. and the constant c > 0 sat-
isfy the properties of the problem. Clearly, f.L is a a-finite Borel measure. By
Theorem 37.7, we can write
f.L = f.L1 + f.L2, where f.L1 «A and f.L2 .1. A.
First, we shall establish that f.L2 = O. From f.L2 .1. A, there exist two disjoint
Borel sets A and B with A U B = JRk and f.L2(A) = A(B) = O. Since A(A) =
00, there exists (by Problem 18;19) a Borel subset C of A with A(C) = c.
From A(C U B) =A(C) + A(B) =
A(C) = c and our hypothesis, we see that
f.L(C U B) = c. Now, note that
c ::: c + f.L2(B) ::: c + f.L(B) = f.L(C) + f.L(B) = f.L(C U B) = c,

and so f.L2(B) = O. This shows that f.L2 = 0, and consequently f.L = f.L1 is
absolutely continuous with respect to A.
Section 37: COMPARING MEASURES AND THE RADON-NIKODYM TIIEOREM 363
Next, fix a compact set K with 'A(K) 2: c and consider both fJ.. and 'A restricted
to K. By the Radon-Nikodym Theorem, there exists a non-negative function
IE LI(K, S, 'A) such that
fJ..(E) = L I d'A
holds for each Borel subset E. of K (see Problem 12.13). We claim that I = 1 a.e.
To establish this, assume by way of contradiction that the Lebesgue measurable
set D = {x E K: I(x) < l} satisfies 'A(D) > 0; we can assume (why?)
that D is a Borel set. If 'A(D) 2: c holds, then pick a Borel subset DI of D
with 'A(DI) = c; if 'A(D) < c, then pick a Borel set DI with D s;; DI s;; K
and 'A(D I) = c; (see Problem 18.19). Now, note that in either case, we have
fJ..(DI) < c, which contradicts our hypothesis. Hence, 'A(D) = 0. Similarly,
'A({x E K: I(x) > 1} = 0, and so I = 1 a.e. Therefore, fJ..(E) = 'A(E) holds
for each Borel subset E of K. Now, pick a sequence {Kn} of compact subsets
of JRk with 'A(Kn) 2: c and KII t JRk . If E is an arbitrary Borel subset of JRk ,
then note that
fJ..(E) = n-+-oo
lim fJ..(E n K II ) = lim 'A(E n Kn) =
11-+00
'A(E).
Problem 37.17. Let fJ.. and v be two a-finite measures on a a-algebra :E 01

subsets 01 a set X sllch that v «
fJ.. and v =1= 0. Show that there exist a set E E :E
and an integer n slIch that
a. v(E) > 0; and
b. A E :E and A s;; E imply ~ fJ..(A) ::: v(A) ::: nfJ..(A).
Solution. Pick a sequence {Xn} of :E with X = U~l XII' v(X II ) < 00, and
fJ..(X n) < 00 for each n. Since v =1= 0, there exists some n such that v(X II ) > 0.
From v « fJ.., it follows that fJ..(X n) > °
also holds. Thus, replacing X by Xn,
we can assume from the outset that both v and fJ.. are finite measures.
Now, by the Radon-Nikodym Theorem, there exists a function °: :
I ELI (fJ..)
such that
v(A) = i l dfJ..
holds for each A E :E. From v =1= 0, we see that I =1= 0, and so the fJ..-measurable
set F = {x EX: I(x) > OJ satisfies fJ..*(F) > 0. Next, put
Ell = {x E X: ~ ::: I(x)::: nl

364 Chapter 7: SPECIAL WPICS IN INTEGRATION
and note that E" t E a.e. Thus, for some n, we have IL*(E,,) > O. By
Theorem 15.11, there exists some E E 1: with En 5; E and IL(E) = IL*(E,,).
We claim that the set E satisfies the desired properties.
To see this, note first that
Now, if A E 1: satisfies A 5; E, then note that ~ XA :::: f :::: n XA IL - a.e., and

consequently
Problem 37.18. Let IL be afinite Borel measure on [1,00) such that

a. IL« A, and
b. IL(B) = alL(aB)for each a ::: 1 and each Borel subset B of[I, 00), where
aB = tab: b E B}.
If the Radon-Nikodym derivative dlL/dA is a continuous function, then show that
there exists a constant c ::: 0 such that [dlL/dA](x) = fr for each x ::: 1.
Solution. For simplicity, let us write 0/1: = f. Then, the given identity IL(B) =
alL(aB) can be written in the fonn
r fdA = al
1B aB
fdA.
For B = [1, x], we get

t
11
f(t) dt = a lax f(t) dt
. a
for each a ::: 1 and each x ::: 1. Differentiating with respect to x (and taking
into account the Fundamental Theorem of Calculus), we see that
f(x) = a 2 f(ax)
holds for each x ::: 1 and each a ::: 1. Letting x = 1, we obtain

/(a) =~
for all a ::: 1, and our conclusion follows.
Section 38: TIlE RIFSZ REPRESENTATION TIlEOREM 365
Problem 37.19. Let f..I- be afinite Borel measure on (0,00) such that
a. f..I-«)..., and
b. f..I-(aB) = f..I-(B) for each a > 0 and each Borel subset B of(O, 00).
If the Radon-Nikodym derivative is a contfnuous function, then show that there
exists a constant c 2: 0 such that [df..l-Id)"'](x) = f for each x > O.
Solution. Let ~i = f. Then, (by The Radon-Nikodym Theorem) the given

identity f..I-(B) = f..I-(aB) can be written in the form
[ fd)'" =
JB
1aB
fd)....
For B = [1,x] (put B = [x, 1] if 0 < x < 1), we get
fX f(t) dt = lax f(t) dt
for each a > 0 and each x > O. Differentiating with respect to x (and taking
into account the Fundamental Theorem of Calculus), we see that
f(x) = af(ax)
holds for each x > 0 and each a > O. Letting x = 1, we obtain
f(a) = f~l)
for all a > 0, as desired.
38. THE RIFSZ REPRESENTATION THEOREM
Problem 38.1. If X is a compact topological space, then show that a continuous

linear functional F on C(X) is positive ifand only if F(l) = IIFII holds.
Solution. Let F be a continuous linear functional on C (X), where X is com-

pact. Note first that
{f E C(X): 1If1l00:S 1} = (J E C(X): If I :s I}.

Thus, if F is also positive, then
IIFII = sup{F(f): f E C(X) and IIflloo :s l}

= sup{F(f): f E C(X) and If I :s I} = F(I).
On the other hand, assume F(l) = I!PII. Let O:s I E C(X) be nonzero, and
put g = ~. Clearly, 111- glloo :s 1. Thus,
F(l) - F(g) = F(l- g) :s I!PII = F(l)

holds, which implies F(g) 2:: O. Therefore, F(f) = IlI1100F(g) 2:: 0 holds, and
so F is a positive linear functional.
Problem 38.2. Let X be a compact topological space, and let F and G be two
positive linearfunctionals on C(X). II F(l) + G(l) :s IIF - Gil, then show that
F 1\ G = O.
Solution. Since F, G 2:: 0, it follows that F - G :s F v G and G - F :s F v G,

and so IF - GI :s F v G. Thus, by the preceding problem
"F - Gil :s I F
v Gil = F v G(l) :s I F + Gil :s I F I + I Gil
= F(l) + G(l) :s IIF - Gil,
and hence, F v G(l) = F(l) + G(l) holds. From F +G= F vG+F 1\ G, it

follows that
I F 1\ Gil = F 1\ G(l) = F(l) + G(l) - F v G(l) = 0,

and so F 1\ G = O.
Problem 38.3. Let X be a Hausdorff locally compact topological space and let
co(X) = {f E C(X): 'V E > 03 K compact with I/(x)1 < E 'V x ¢ K}.
Show that:
a. co(X) equipped with the sup norm is a Banach lattice.
b. The norm completion oICc(X) is the Banach lattice co(X).
Solution. (a) Clearly, co(X) with the sup norm is a normed vector lattice. For
the completeness, let {In} be a Cauchy sequence of co(X). Then {In} converges
uniformly on X to some function I. By Theorem 9.2 we infer that I E C(X).
Now, if 8 > 0 is given, pick some n with IIfn - 11100 < 8, and then choose some
compact set K with I/n(x)1 < 8 for x ¢ K. Thus,
I/(x)1 :s Ifn(X) - l(x)1 + Ifn(X) I < 8 + 8 = 28

holds for all x ¢ K, so that I E co(X).
Section 38: TIlE RIFSZ REPRFSENTATION TIlEOREM 367
(b) Obviously, Cc(X) is a vector sublattice of co(X). We have to show that Cc(X)
is dense in co(X).
To this end, let f E co(X) and let 8 > O. Pick some compact set K with
1f(x)1 < 8 for x ¢. K, and then choose some open set V with compact closure
such that K S; V. By Theorem 10.8 there exists a function g: X --'? JR with
K -< g -< V. Then, fg E Cc(X) and I1f - fglloo ::5 28 holds, proving that
Cc(X) is dense in the Banach lattice co(X).
Problem 38.4. Let F be a positive linearfunctional on Cc(X), where X is Haus-

dorff and locally compact, and let f.L be the outer measure induced by F on X.
Show that if f.L* is the outer measure generated by the measure space (X, B, f.L),
then f.L*(A) = f.L(A) holds for every subset A of x.
Solution. Let A S; X. We know that
f.L(A) = inf{f.L(V): V open and A S; V }.
So, if A S; V holds with V open, then f.L*(A) ::5 f.L*(V) = f.L(V) also holds,
and thus f.L*(A) ::5 f.L(A). On the other hand, by Theorem 15.11 there exists some
B E B with A S; B and f.L*(A) = f.L(B). Thus, f.L(A) ::5 f.L(B) = f.L*(A), proving
that f.L*(A)= f.L(A) holds.
Problem 38.5. Let f.L and v be two regular Borel measures on a Hausdorfflocally
compact topological space X. Then show that f.L ::: v holds if and only if f d f.L ::: J
Jf dv for each f E Cc(X)+.
Solution. Let f.L and v be two regular Borel measures on a Hausdorff locally
compact topological space X.
Assume first that f.L ::: v holds (i.e., assume that f.L(A) ::: v(A) holds for each
=
A E B). Clearly, if 11 is a f.L-step function of the form 11 'L,;'= I ai XA; with each
J
ai ::: 0 and each Ai E B, then 11 is a v-step function and 11 df.L ::: 11 dv holds. J
Now, let 0::5 f E Cc(X). Since
it follows from Theorem 17.7 that there exists a sequence {1111} of f.L-step functions
of the preceding type satisfying 1111 (x) t f (x) for each x EX. This implies
I
f df.L = lim
11-+-00
11111 df.L::: n-+oo
lim l11n dv = I f dv.
Forthe converse, assume that Jf df.L ::: Jf dv holds for each 0 ::5 f E Cc(X).
In view of the regularity of the measures, in order to establish that f.L ::: v holds
it suffices to show that f.L(K) ::: v(K) holds for each compact set K. To this end,
let K be a compact set. Given 8 > 0, choose an open set V such that K S; V
and IL(V) < IL(K) + 8. By Theorem 10.8 there exists a function / E Cc(X) such
that K -< / -< V. Now note that from XK :s / :s Xv, it follows that
v(K) = f XK dv :s f / :s f / :s f
dv dIL Xv dIL = IL(V) < IL"(K) + 8
for all 8 > 0. That is, v(K) :s IL(K) holds, as desired.
Problem 38.6. Fix a point x in a Hausdorff locally compact topological space

X, and define F(f) = /(x)/or each / E Cc(X). Show that F is a positive linear
/unctionalon Cc(X) and then describe the unique regular Borel measure IL that
satisfies F(f) = J/ dIL/or each / E Cc(X). What is the support 0/ IL?
Solution. Clearly, F is a positive linear functional. The regular Borel measure

representing F is the Dirac measure with "base point" at x. Its support is, of
course, the set {x}.
Problem 38.7. Let X be a compact Hausdorff topological space. 1/ IL and v

are regular Borel measures, then show that the regular Borel measures IL v v and
IL 1\ v satisfy
a. SUPP(IL v v) = Supp IL U Supp v, and
b. SUPP(IL 1\ v) S; Supp IL n Supp v.
Use (b) to show that ifSuPPIL n Suppv = C/J, then IL .1. v holds. Also, give an
example/or which SUPP(IL 1\ v) =1= SUPP IL n SUPP v.
Solution. (a) Let A = SUPP(IL v v), B = SUPP IL, and C = SUPP v. From
IL :s IL v v, v :s IL v v, and IL v v(N) = 0, it follows that = v(A = 0,
IL(A C ) C
)
and so B S; A and C S; A. That is, B U C S; A. On the other hand, the inequality

IL v v :s IL + v implies
and so AS; (BC ncct = B uC.

(b) The inclusion follows easily from the inequalities
IL 1\ V :s IL and IL 1\ V :s v.
If SUPP IL n SUPP v = C/J, then by part (b) SUPP(IL 1\ v) = C/J holds, and so
IL 1\ v = 0. For an example showing that equality need not hold in (b), let X = R,
IL = the Lebesgue measure, and v = the Dirac measure with "base point" at 0.
Section 38: THE RlESZ REPRESENTATION THEOREM 369
By Problem 36.5, we have p., 1\ V == O. Therefore, Supp(p., 1\ v) = C/J holds, while

=
Supp p., n Supp v 1R n {O} ={O}.
Problem 38.8. Let X be a Hausdorff loeally compact topological space X.

Characterize the positive linear functionals F on Cc(X) that are also lattice
homomorphisms; that is, F(f v g) = max{F(f), F(g)} holds for each pair
f, g E Cc(X).
Solution. Let F be a positive linear functional on Cc(X). Then we shall show

that F is a lattice homomorphism if and only if there exist some c 2: 0 and some
=
a E X such that F(f) cf(a) holds for all f E Cc(X).
Clearly, iffor some c 2: 0 and some a E X we have F(f) = cf(a) for each
f E Cc(X), then F is a lattice homomorphism. For the converse, assume that
F is a non-zero lattice homomorphism. Let p., be the regular Borel measure that
represents F. If x, y E Supp p., satisfy x =1= y, then it is not difficult to see that
=
there exist f, g in Cc(X) with f 1\ g 0 and f(x) = g(y) =
1. Therefore,
F(J v g) = F(J + g) = F(J) + F(g) > maxI F(f), F(g)}
must hold, which is a contradiction. Thus, Supp p., consists precisely of one point;
let Supp p.,= {a}. Set c = p.,({a}) > 0, and note that for every f E Cc(X), we
have
F(f) = f f dp., = f(a)· p.,({a}) = cf(a).
Problem 38.9. Let X be a Hausdorff locally compact topological space such

that X is an uncountable set. Then show that
a. C~(X) is not separable, and
b. C[O, 1] (with the sup norm) is not a reflexive Banach space.
Solution. (a) For each x EX define the positive linear functional Fx: Cc(X) ---+
1R by FAf) = f (x) and note that IlFx - Fy 1\ = 2 holds for x =1= y. Clearly, the set
{Fx: x E X} is an uncountable subset of C~(X). Therefore, {B(Fx , 1): x EX}
is an uncountable collection of pairwise disjoint open balls. From this, it easily
follows that no countable subset of C~(X) can be dense in C~(X).
(b) By Problem 11.12, we know that C[O, 1] is separable. If C[O, 1] is reflexive,
then its second dual is likewise separable. But then (by Problem 29.8) its first dual
must be separable, contradicting part (a). Thus, C[O, 1] is not a reflexive Banach
lattice.
Problem 38.10. Let X be a Hausdorff locally compact topological space. For a

finite signed measure p., on B show that the following statements are equivalent:
a. f-L belongs to Mb(X).

b. f-L+ and f-L- are both finite regular Borel measures.
c. For each A E J3 and E > 0, there exist a compact set K and an open
set V with K S;; A S;; V such that If-L(B)I < E holds for all B E J3 with
BS;;V\K.
Solution. (a) ==> (b) Pick two finite regular Borel measures f-Ll and f-L2 such
that f-L = f-Ll - f-L2. Then,. f-L+ = (f-Ll - f-L2)+ = f-Ll V f-L2 - f-L2 holds. By
Theorem 38.5, f-Ll v f-L2 is a finite regular Borel measure, and from this it follows
that f-L + is a finite regular Borel measure. Similarly, f-L - is a finite regular Borel
measure.
(b) ==> (c) Note that If-LI = f-L+ + f-L- is a finite regular Borel measure. Now, let
A E J3 and let e > 0 be given. Then, there exists a compact set K and an open
set V with K S;; A S;; V and If-LI(V \ K) < e. Therefore, if B E J3 satisfies
B S;; V \ K, then If-L(B)I ::: If-LI(B) ::: If-LI(V \ K) < e holds.
(c) ==> (a) Let A E J3 and let e > O. Choose a compact set K and an open set
V so that (c) is satisfied. Then, by Theorem 36.9, we have
and
Thus, f-L+(A) - f-L+(K) ::: e and f-L+(V) - f-L+(A) ::: e both hold. Hence, f-L+ is a
finite regular Borel measure. Similarly, f-L - is a finite regular Borel measure, and
so f-L = f-L+ - f-L- E Mb(X).
Problem 38.11. A sequence {xn} in a normed space is said to converge weakly

to some vector x if lim f(xn) = f(x) holds for every continuous linearfunctional
f·
a. Show that a sequence in a normed space can have at most one weak limit.
b. Let X be a Hausdorff compact topological space. Then show that a se-
quence Un} of C (X) converges weakly to some function f if and only if
{f,l} is norm bounded and lim fn (x) = f (x) holds for each x EX.
Solution. (a) Assume that a sequence {xn} in a normed vector space Y satisfies
limf(xn) = f(x) and limf(xn) = f(y) forevery f E Y*. Then, f(x - y) = 0
holds for all f E Y*. By Theorem 29.4, we see that x - y = 0, and so {xn} can
have at most one weak limit.
(b) Assume first that the sequence Un} of C(X) converges weakly to some
Section 38: THE RIESZ REPRFSENTA'fION THEOREM 371
function f E C(X). By Theorem 29.8, {fIll is norm bounded. If x E X, let f.1,x

denote the Dirac measure with support {x}, and note that
fn(x) = f fn df.1,x -'r f f df.1,x = f(x).
Conversely, if {fn} is norm bounded and lim f,,(x) = f(x) holds for each
x X (where, of course, f E C(X», then the Lebesgue Dominated Convergence
E
J J
Theorem implies that lim f" df.1, = f df.1, holds for every Borel measure f.1,.
This, coupled with the Riesz Representation Theorem, shows that {f,,} converges
weakly to f.
Problem 38.12. Let f.1, be a regular Borel measure on a Hausdorff locally com-
pact topological space X, and let f ELI (f.1,). Show that the finite signed measure
v, defined by
for each Borel set E, is a (finite) regular Borel signed measure. In other words,
show that v E Mb(X).
Solution. We can assume that f(x) 2: 0 holds for all x. By Problem 22.7, the
set A = {x EX: f(x) > O} is a a-finite set with respectto f.1,. Choose a sequence
{Xn} of f.1,-measurable sets with f.1,(X n) < 00 for each n and X" t A.
Now, let E be a Borel set and let 8 > 0; clearly, vee) = veE n A). Select some
n with vee) - v(X" n E) < 8. Also, using the regularity of f.1, and Problem 22.6,
we see that there exists a compact set K s:;; X nnE with
v(X n n E) - v(K) = [ f dfJ- - [ f dfJ- < 8.

Jx"nE JK
Thus, the compact set K s:;; E satisfies
0::: vee) - v(K) = [vee) - v(X" n E)] + [v(X" n E) - v(K)] < 28.
Next, use Problem 22.6 and the regularity of f.1, to see that for each n there
exists an open set Vn satisfying X" n E s:;; Vn and v(V,,) - v(X" n E) < in.
Then, the open set V = U:I V" satisfies E s:;; V, and in view of V \ E s:;;
U:I(V" \ Xn n E), we see that
00
0::: v(V) - vee) = v(V \ E) ::: L v(V" \ X" n E) < 8.

n=1
Altogether, the preceding show that v is a regular Borel measure.
Problem 38.13. Generalize part (3) o/Theorem 38.5 as/ollows: 1/ /L and v are
two regular Borel measures on a Hausdorff locally compact topological space and
one o/then is a-finite, then show that /L /\ v is also a regular Borel measure.
Solution. Let /L and v be two regular Borel measures on a locally compact

Hausdorff topological space. X and assume that /L is a-finite. Also, let w = /L/\ v
and note (in view of w ::: /L) that w is a a -finite Borel measure which is absolutely
continuous with respect to /L.
Now, let E be a Borel subset of X satisfying /L(E) < 00 and let e > O.
Consider w and /L restricted to the Borel sets BE of E (from Problem 12.13 we
know that BE = {B n E: B E B}). Now, by the Radon-Nikodym Theorem there
exists a (unique) non-negative function / E L 1(E, BE, /L) satisfying
weB n E) = ( / d/L, for each B E B.

JBnE
Since /L is a regular Borel measure, it follows from Problem 22.6 that there exists
a compact subset K of E such that
0::: wee) - w(K) = { / d/L - { / d/L = ( / d/L < e.

JE JK JE\K
Therefore, we infer that
wee) = sup{w(K): K compact and K £;; E}.
Now, use the a-finiteness of w to show that (*) holds true for each Borel subset of
E.
It remains to be shown that the measure of every Borel set can be approximated
from above by the measures of the open sets. To this end, let E be an arbitrary
Borel set, and recall that
wee) = /L /\ v(E) = inf{/L(B) + veE \ B): B E B and B £;; E}.
Let c = inf{w(O): 0 open and E £;; 0) and let e > O. Given B E B

with B £;; E, choose open sets V and W such that B £;; V, E \ B £;; W,
/L(V) ::: /L(B) + e, and v(W) ::: v~E \ B) + e. Then, we have
wee) ::: c ::: w(V u W) ::: w(V) + w(W) ::: /L(V) + v(W)
::: /L(B) + e + veE \ B) + e = /L(B) + veE \ B) + 2e.
Section 38: 11IE RIFSZ REPRESENTATION 11IEOREM 373
Thus, weE) :::: c :::: weE) + 28 holds for each 8 > 0, and so weE) = c, and we
are finished.
Problem 38.14. Show that every finite Borel measure on a complete separable
metric space is a regular Borel measure. Use this conclusion to present an alternate
proof of the fact that the Lebesgue measure is a regular Borel measure.
Solution. Let X be a complete separable metric space, let B be the a-algebra

of all Borel sets of X, let {XI, X2, ••• } be a dense countable subset of X, and let
J.L: B --+ [0,00) be a measure.
Consider the collection A of subsets of X defined by
A = {A E B: J.L(A) = inf{J.L( 0): A£;O and 0 open}

= sup{J.L(K): K £; A and K compact}).
The collection A has the following properties:

1. A contains the open and closed sets.
To see this, assume first that V is an open set, and let 8 > 0. For each n let Fn
be the collection of all open balls of the form B (Xi, r) with r a rational number
less than or equal to f.
and B (Xi, r) £; V. Clearly, each Fn is at most countable
and V = UBeFl holds. For each n pick B;', ... , Bk" E Fn such that
J.L ( V \ U Bf
k"
i=1
)
< -;.
8
2
Next, put C = n:1 U7;:'1 Bj', and note that C is a totally bounded set. Hence,
its closure C is also a totally bounded set (why?). Since (by Theorem 6.13) C is
a complete metric space in its own right, it follows from Theorem 7:8 that C is a
compact set. Now, note that C £; V holds, and that
°: : J.L(V) - J.L(C) :::: J.L(V) - J.L(C) = J.L(V \ C)

00 kn 00 kn
= J.L(U(V\ UB!')):::: LJ.L(V\ UBi')

n=1 i=1 /1=1 i=1
00 8
< L
n=1
2
n =8.
Therefore,
J.L(V) = inf{J.L(O): V£;O and 0 open}
= sup{J.L(K): K £; V and K compact}
holds, and so V E A.
Now, let C be a closed set, and let 8 > O. By the preceding, there exists a
compact subset K with fJ,(X) - fJ,(K) < 8. Then, the compact subset C n K of
C satisfies
/L(C) - /L(C n K) = /L(C \ C n K)

= /L(C \ K) ::: /L(X \ K) = /L(X) - /L(K) < 8.
Also, by the previous part, there exists a compact set K 1 with K 1 S;;; X \ C and
/L(X \ C) - /L(K 1) < 8. Now, the open set 0 = X \ K 1 satisfies C S;;; 0 and
/L(O) - /L(C) = /L(X \ Kd - fJ,(C) = /L(X) - /L(Kl) - /L(C)

= /L(X \ C) - /L(K 1) < 8.
Thus,
/L(C) = inf{/L(O): C S;;; 0 and 0 open}

= sup{/L(K): K S;;; C and K compact}
also holds, and so C E A.

2. If A E A, then N E A.
From /L(A) = sup{/L(K): K S;;; A and K compact}, it follows that
C
/L(A ) = /L(X) - /L(A)
= inf{/L(X) - /L(K): K S;;; A and K compact}
= inf{/L(K C ): K S;;; A and K compact}
= inf{/L(O): A C S;;; 0 and 0 open}.
Similarly, fJ,(A) = inf{/L(O): A S;;; 0 and 0 open} implies
Since, by part (1), /L(C) = sup{/L(K): K S;;; C and K compact} holds for each
closed set C, we see that
C
/L(A ) = sup{/L(K): K S;;; A C and K compact}.
3. If {All} is a sequence of A, then· 1 An EA. U:

Let {An} S;;; A, let A = U:
1 An, and let 8 > O. For each n pick some open set
On with An S;;; On and /L(On \ All) < 82- n. Then, the open set 0 = U:l On
Section 38: THE RIESZ REPRESENTATION THEOREM 375
satisfies A ~ 0 and from 0 \ A = U~I 011 \ U~I All ~ U~I(On \ An), we

get
00 00
J1.(0) - J1.(A) = J1.(0 \ A) ~ J1.(U(O/~ \ All)) ~ L J1.(01l \ All) < s.

11=1 11=1
On the other hand, fix some k with J1.(A \ U~=I Ai) < s, and then for each
1 ~ i ~ k pick a compact set Ki ~ Ai with J1.(Ai \ K i ) < s2- i . Then, the
compact set K = Uki=1 Ki satisfies K ~ Uki=1 Ai ~ A and
k k
J1.(A) - J1.(K) = J1.(A \ K) = J1.(A \ U Ai) + J1.( (U Ai) \ K)
i=1 i=1
k k k
< s+J1.((UAi)\K) =S+J1.(UAi \ UKi)

i=1 i=1 i=1
k
~ S + LJ1.(Ai \Ki ) < s +s = 2s.
i=1
The validity of statement (3) has been established.

Now, from the preceding statements, we see that A is a a-algebra that contains
the open sets. Consequently, every Borel set belongs to A (Le., A = B), and so
J1. is a regular Borel measure.
Now, let us use the previous conclusion to establish that the Lebesgue measure
A on lRlI is a regular Borel measure. To this end, let A be an arbitrary Borel set.
Also, let VII (resp. Cn) denote the open (resp. the closed) ball of JRII with center
at zero and radius n. Since each C n is a complete separable metric space in its
own right, it follows from the previous result that A restricted to each C n is a
regular Borel measure. Therefore, we have
A(A n Cn) = sup{A(K): K ~ An Cn and K compact}.
From An C n t A, it follows that A(A n Cn) t A(A), and an easy argument shows
that
A(A) = sup{A(K): K ~ A and K compact}.
Next, note that if A(A) = 00, then

A(A) = inf{A(O): A ~ 0 and 0 open}
376 Chapter 7: SPEClAL'IDPICS IN INTEGRATION
is trivially true. So, assume that )"(A) < 00, and let e > O. By the regularity of
).. on Cn (and the fact that Vn is an open set), we see that
)"(A n Vn) = inf{)"(On Cn): Vn n AS;; 0 n Cn and 0 open}

= inf{)"(O n Vn): VII n A S;; 0 n Vn and 0 open}
= inf{)"(O): An Vn S;; 0 and 0 open}.
Therefore, for each n there exists an open set On with A n Vn S;; On and
)"(On \A n VII) < e2- n. Now, the set 0 = U:
1 0n is open and satisfies
AS;; O. From
00 00 00
0\ A = U 011 \ UAn Vn S;; U(On \ An Vn),

n=1 11=1 n=1
we see that
00
0::: ),,(0) - )"(A) = ),,(0 \ A) ::: I>(On \ An Vn) < e.

n=1
Hence, )"(A) = inf{)"( 0): A S;; 0 and 0 open} also holds, and so the Lebesgue
measure ).. is a regular Borel measure.
Problem 38.15. Let X be a Hausdorff compact topological space. II l/J: X -+ X

is a continuous junction, then show that there exists a regular Borel measure on
X such that
f I ol/Jdf-L = f I df-L
holdsloreach IE C(X).
Solution. Let X be a Hausdorff compact topological space and let l/J: X -+ X

be a continuous function. Fix some wE X and let Cim: too -+ too is a Banach-
Mazur limit (see Problem 29.7). Now, consider the positive linear functional
F: C(X) -+ 1R defined by
F(f) = Cim(f(l/J(w)), I(l/J\w)), 1(¢3(w)), ... ),
and let f-L be the regular Borel measure on X representing F, i.e., F (f) = JI d f-L
Section 38: THE RlFSZ REPRESENTATION THEOREM 377
holds for each f E C(X). The identity
Cim(x" X2, ... ) = Cim(x2, X3, ... )
for all (X"X2, ... ) E loa easily implies F(f) = F(f 0 ¢) for each f E C(X).
Consequently, the regular Borel measure J.1, satisfies J
f dJ.1, = f 0 ¢ dJ.1, for J
each f E C(X).
Problem 38.16. This exercise gives an identification of the order dual C;:-(X) of
Cc(X), Consider the collection M(X) of all formal expressions J.1" - J.1,2 with J.1"
and J.1,2 regular Borel measures. That is,
M(X) = {J.1" - J.1,2: J.1" and J.1,2 are regular Borel measures on X}.
a. Define J.1" -J.1,2 == V'-V2 inM(X) tomeanJ.1,,(A)+V2(A) = v, (A)+J.1,2(A)

for all A E B. Show that == is an equivalence relation.
b. Denote the collection of all equivalence classes by M(X) again. That is,
J.1" - J.1,2 and v, - V2 are considered to be identical if J.1" + V2 = V, + J.1,2
holds. In J\.It(X) define the algebraic operations
V2) = (J.1" + v,) - (J.1,2 + V2),

if a:::O
if a < O.
Show that these operations are well defined (i.e., show that they depend only
upon the equivalence classes) and that they make M(X) a vector space.
c. Define an ordering in M(X) by J.1" - J.1,2 ::: VI - V2 whenever
holds for each A E B. Show that::: is well defined and that it is an order
relation on M(X) under which M(X) is a vector lattice.
d. Consider the mapping J.1, = J.1" - J.1,2 l-7 FJ1. fi'om M(X) to C;:-(X) defined
by FJ1.(f) = Jf dJ.1" - Jf dJ.1,z/or each f E Cc(X). Show that FJ1. is well
defined and that J.1, l-7 FJ1. is a lattice isomorphism (Lemma 38.6 may be
helpfitl here) from M(X) onto C;:-(X). That is, show that C;:-(X) = M(X)
holds.
Solution. (a) Clearly, J.1" - J.1,2 == J.1" - J.1,2 and J.1" - J.1,2 == VI - V2 implies
V'-V2 For the transitivity, let J.1,,-J.1,2 == VI-V2 and VI-V2 = WI-W2.
== J.1,,-J.1,2'
That is, assume that J.1,1 + V2 = VI + J.1,2 and VI + W2 = WI + V2. Adding the last
two equalities, we see that
378 - Chapter 7: SPECIAL 'IDPICS IN INTEGRATION
Since all measures involved are regular Borel measures, it follows from (*) that
f.1.1 (K) + w2(K) = WI (K) + f.1.2(K) holds for each compact subset K of X. The
regularity of the measures implies
for each A E B, and so f.1.1 - f.1.2 = WI - W2 holds.

(b) To see that the addition is .well defined, assume that f.1.1 - f.1.2 == VI - V2 and
WI - W2 == 7l"1 - 7l"2. That is, f.1.1 + V2 = VI + f.1.2 and WI + 7l"2 = 7l"1 + W2, and so
(f.1.1 + wd + (V2 + 7l"2) = (VI + 7l"d + (f.1.2 + (2). That is,
(f.1.1 - f.1.2) + (WI - (2) + wd - (f.1.2 +(2)

= (f.1.1
== (VI + 7l"1) - (V2 + 7l"2)=(VI - V2) + (7l"1 - 7l"2).
Similarly, the mUltiplication is well defined. Now, it is a routine matter to verify

that under these algebraic operations M(X) is a vector space.
(c) To verify that::: is well defined, proceed as in part (b) above. It is a routine
matter to check that::: makes M(X) a partially ordered vector space.
Next, we shall show that M(X) is a vector lattice. It suffices to verify that
(f.1.1 - f.1.2)+ exists in M (X) for each JL I - f.1.2 E M (X). To this end, let f.1.1 - f.1.2 in
M(X). By Theorem 38.5, f.1.1 v f.1.2 is a regular Borel measure, and we claim that
(f.1.1 - f.1.2)+ = f.1.1 V f.1.2 - f.1.2 holds in M(X). Clearly,
f.1.1 - f.1.2 ::: f.1.1 v f.1.2 - f.1.2 and 0::: f.1.1 v f.1.2 - f.1.2
both hold. To see that f.1.1 v f.1.2 - f.1.2 is the least upper bound of f.1.1 - f.1.2 and 0,
assume f.1.1 - f.1.2 ::: VI - V2 = V and V ::: O. Then, V + f.1.2 is a regular Borel
measure such that V + f.1.2 ::: f.1.1 and V + f.1.2 ::: f.1.2 both hold. By Theorem 38.5,
V + f.1.2 ::: f.1.1 v f.1.2, and hence v ::: f.1.1 v f.1.2 - f.1.2 holds in M(X). This shows
that f.1.1 v f.1.2 - f.1.2 is the least upper bound of f.1.1 - f.1.2 and O.
(d) It is a routine matter to verify that f.1. 1---+ F/l from M(X) into C;:-(X) is
well defined and linear. Moreover, since every F E C;:-(X) can be written as a
difference of two positive linear functionals, the Riesz Representation Theorem
guarantees that f.1. 1---+ FJ.L is onto. To see that f.1. 1---+ FJ.L is one-to-one, assume
that F/l = O. Then, Jf df.1.1 = Jf df.1.2 holds for each f E Cc(X), and so by
(the Riesz Representation Theorem) f.1.1 = f.1.2. Therefore, f.1. = f.1.1 - f.1.2 = 0 and
so f.1. 1---+ F/l is one-to-one.
Finally, observe that f.1. ::: 0 holds in M(X) if and only if F/l ::: 0 holds in
C;:- (X), and then invoke Lemma 38.6 to see that f.1. 1---+ FJ.L is a lattice isomorphism
from M(X) onto C;:-(X). Thus, under this lattice isomorphism, we can say that
C;:-(X) = M(X).
Section 39: DIFFERENTIATION AND INTEGRATION 379
Problem 38.17. This problem shows that for a noncompact space X, in gen-
eral C~(X) is a proper ideal of C;;(X). Let X be a Hqusdorff locally compact
topological space having a sequence {On} of open sets such that On ~ 0 11 +I and
On =1= 011 +1 for each n, and with X = U:I 911.
a. Show that if X is a-compact but not a compact space, then X admits a
sequence {Oil} of open sets with the preceding properties.
b. Choose XI E 0 1 and XII E On \ On-I for n 2: 2. Then, show that
00
F(f) = L f(x lI ) for f E Cc(X)

11=1
defines a positive linear functional on Cc(X) that is not continuous.
c. Determine the (unique) regular Borel measure fL on X that represents F.
What is the support of fL?
Solution. (a) Let {K II } be a sequence of compact sets with KII t X. For each n
pick an open set VII with compact closure such that KII ~ Vn. Put On = U7=1 Vi
and note that all t X. Since each On has compact closure and X is not compact,
On =1= X holds for each n. By passing to a subsequence of {On}, we can assume
that On =1= On+1 also holds for each n.
(b) Let f E Cc(X). Since Supp f .s; U:I all holds and Supp f is compact,
there exists some k with Supp f ~ Ok. Thus, f(x lI ) = 0 for n ::> k, and so F
clearly defines a positive linear functional on Cc(X).
Next, we shall show that F is not continuous. By Theorem 10.8 there exists
some gn E Cc(X) with {XI, ... ,xn} -< gil -< a". Therefore, IIFII2: F(gn) = n
holds for each n, and so II F II = 00.
(c) The regular Borel measure fL that represents F is defined on the Borel set B
by
fL(B) = The number of elements of {XI, X2, ... } n B.

(If {Xl, X2, ... } n B is countable, then fL(B) = 00, and if (x), X2, ... } n B = 0,
then fL(B) = 0.) Also, note that
Supp fL = {Xl, X2, ... }.
39. DIFFERENTIAnON AND INTEGRAnON
Problem 39.1. If fL is a Borel measure on lRk, then show that fL .l A holds if

and only if D fL(X) = 0 for almost all x.
Solution. Assume fL 1.. A. Choose two disjoint Borel sets A and B with
AU B = IRk and fL(A) = A(B) = O. By Lemma 39.3, DfL(X) = 0 holds for
almost all x in A, and so D fL(x) = 0 holds for almost all x in IRk.
Now, suppose that D fL(x) =
0 holds for almost all x E IRk. Use the Lebesgue
Decomposition Theorem 37.7 to writefL = fLI + fL2 withfLI « A andfL2 1.. A.
By the preceding, DfL2(X) = 0 holds for almost all x in IRk. Thus, from
Theorem 39.4, it follows that
d/J.l -DIl
([f" - .-1 -D"-O
- .- - ,
and so fLI = O. Therefore, fL = fL2 1.. A holds.

Problem 39.2. Show that if E is a Lebesgue measurable subset of IRk, then
almost all points of E are density points.
Solution. For each x =

(XI, ... ,Xk) E IRk and each e > 0, consider the open
interval Ie = n~=I(Xi - e, Xi + e). If E is a Lebesgue measurable set, then
lim >..(En[,)
e->O+ (2e)
= XE(X)
holds for almost all x. To see this, note first that we can assume without loss of
generality that A(E) < 00 holds (why?). Now, consider the finite Borel measure
fL on IRk defined by
fL(A) = A(E n A) = i XE dA.
Clearly, fL « A and *
= XE. By Theorem 39.4, we have DfL = XE a.e., and
the validity of (*) follows.
Problem 39.3. Write Br(a)for the open ball with center at a E IRk and radius
r. If f is a Lebesgue integrable function on IRk, then a point a E IRk is called a
Lebesgue point for f if
I
lim (B »
r->O+ A r(a JfBr(a) If(x) - f(a)1 dA(x) = O.
Show that if f is a Lebesgue integrable function on IRk, then almost all points
of IRk are Lebesgue points.
Section 39: DIFFERENTIATION AND IN1EGRA TION 381
Solution. Denote by Q the set of all rational numbers of R Fix some a E Q,

and let Bn = B(O, n). Now, define the finite Borel measure JL by
JL(E) = r
lEnBn
I
Vex) - a dA(X).
Since JL « A, it follows from Theorem 39.4 that

DJL = ~~ = If - alxBn a.e.
Consequently,
lim )"(B\X»
r-.O+ r 1rBr(x) If(t) - a IdA(t) = If(x) - a I
holds for almost all x in B Il , and therefore (since n is arbitrary) (*) holds for
almost all x in JRk • Let Ea be a Lebesgue null set for which (*) holds for all
x ¢. Ea. Set E = UaeQ Ea, and note that A(E) = 0.
Now, let y ¢. E and let 8 > 0. Choose some rational number SEQ with
Is - f(y)1 < 8 (we shall assume that f is real-valued everywhere). In view of
If(x) - f(y)1 :s If(x) - sl + Is - fey)!. we see that
lim sup A(B~(Y» r I

If(x) - fey) dA(X)
r ..... O+ lBr(Y)
:s r lim
..... O+
)"(B\ »
r Y 1Br(Y)
r
Vex) - s IdA(X)
+ rlim
..... O+
)"(B\)) r
r Y 1Br(y)
Is - I
fey) dA(x)
= If(y) - sI+ Is - I
fey) < 28,
and from this the desired conclusion follows.
Problem 39.4. Let f: JR -+ JR be an increasing, left continuous function. Show

directly (i.e., without using Theorem 38.4) that the Lebesgue-Stieltjes measure JL I
is a regular Borel measure.
Solution. Let (a, b) be an open interval. Then, there exists a sequence {[an, bn]}
of closed intervals with [an, bnl t (a, b). Itfollowsthat JLI([a n, bnD t JLI«a, b)).
Since every open subset of JR can be written as an at most countable union of
pairwise disjoint open intervals, it follows that
ILI(O) = SUP{JLI(K): K ~ 0 and K compact}
holds for all open sets O.
382 ~hapter 7: SPECIAL WPICS IN INTEGRAnON
Now, let [a, b) be a finite interval. Then, for each point c < a of continuity of
f, we have [a, b) S; (c, b) and JLf(C, b»)-JLf([a, b») = f(a)- f(c). By the left
continuity of f, we see that JLf([a, b» = inf{J..lA(c, b»): c < a}. Next, consider
aa-set A with JLf(A) < 00. Choose a pairwise disjoint sequence ([an, bnH with
A = U:I [an, bn). Given B > 0, for each n choose some real number Cn < an
with JLf(cl!' bn) \ [an, bn») < {;;, and then set V = U:I (c n, bn). Clearly, V is
an open set, A S; V, and
00
JL f(V) - JL f(A) = JL f(V \ A) ~ JL f (U [(c n, bn) \ [an, bl!) J)

n=1
00 00
~ LJLf(Cn,bn)\ [an, bn») < L{;; =B.

n=1 n=1
Thus, JLf(A) = inf{JLf(V): A S; V and V open}.

Now, to complete the proof, use Problem 15.2. (For a general result about regular
Borel measures, see also Problem 38.14.)
Problem 39.5 (Fubini). Let Un} be a sequence of increasing functions defined

on [a, b] such that 2::1 fn(x) = f(x) converges in IRfor each x E [a, b]. Then,
show that f is differentiable almost everywhere and that f'(x) = 2::1 t,;(x)
Solution. Replacing each fn by fn - fn(a), we can assume that fn ?: 0

holds for each n. Set Sn = fl + ... + fll, and note that each Sn is increasing
and sn(x) t f(x) holds for each x. Clearly, f is also an increasing function.
By Theorem 39.9, f and all the fn are differentiable almost everywhere. Since
fn+1 = Sn+1 -SIl is an increasing function, we see that s~+] (x) ?: s~(x) must hold
for almost all x. Similarly, since f(x) - Sn(X) = 2:~n+] Si(X) is an increasing
function, it follows that f' (x) ?: s~ (x) holds for almost all x. Thus,
00
lim s~(x) = " f~(x)

ll~OO ~
n=]
exists for almost all x.

Now, for each n let
00
tn(x) = f(x) - Sn(X) = L fi(X)?: O.

i=n+]
Section 39: DIFFERENTIAnON AND INTEGRAnON 383
Clearly, each tn is an increasing function. Pick a subsequence (Sk"} of (sn} such

that
00 00
I:>k,,(X) :s I]f(b) - sk,,(b)] < 00.

n=1 n=1
The same arguments applied to (tk"} instead of (S/l} show that
00 00
I>U
n=1
x ) = L[J'(X) - s~"Cx) ]
n=1
converges for almost all x. In particular, sUx) -+ f' (x) holds for almost all x,
and so
00
L f~(x) = j'(x)
/l=1
Problem 39.6. Suppose (f,I} is a sequence of increasing functions on [a, b] and

that f is an increasing function on [a, b] such that fl, J" t fl, f. Establish that
f'(x) = lim f~(x) holds for almost all x.
Solution. We shall present a solution of this problem based upon the following
general continuity property of the Differential Operator D: If (fl,/l} is a sequence
of Borel measures in JRk and fl,/l t fl, holds for some Borel measure fl" then
D fl,n t D fl, a.e. also holds.
If this property is established, then using Theorem 39.8, we see that
f~(x) = Dfl,f"(x) t Dfl,f(x) = f'ex)
must hold for almost all x.

To establish the validity of the continuity property start by observing that if two
Borel measures fl, and v satisfy fl, :s v, then Dfl, :s Dv a.e. holds. Indeed, by
Theorem 39.6 both fl, and v are differentiable almost everywhere. If x E JRk is
a point for which Dfl,(x) and Dv(x) exist and Bn = B(x, then *),
Dfl,(x) = n-+oo
lim cl!hl < lim v(B,,) = Dv(x).
/l-+OO
)"(B") - )"(B")
Now, let fl,n t fl,. Restricting ourselves to the open balls (x E ]R.k: IIx II < n},
we can assume without loss of generality that all measures are finite.
By the Lebesgue Decomposition Theorem 37.7, we can write f.Ln = vn + WII

with Vn « A and Wn 1. A. It follows from the proof of Theorem 37.7 that
f.Ln 1\ mA tm Vn' Clearly, this implies VII ::: Vn+1 for each n. From formula (c) of
Problem 9.1, we get
f.Ln - f.L1I 1\ mA = 0 v (f.Ln - mA) ::: 0 v (f.Ln+1 - mA) = f.Ln+1 - f.L1I+1 1\ mA
for each m. Letting m ~ 00, we obtain
WIl = f.Ln - Vn ::: f.Ln+1 - VIl+1 = Wn+1

for each n. Let Vn t V and WIl t w. Since f.Ln = Vn + Wn t f.L, it follows that
=
f.L v + w. The relation Vn « A for each n easily implies v « A. In view of
W Il 1. A = 0 for each n, it follows from Lemma 37.6 that W 1. A = O. That is,
v « A and W 1. A both hold, and so f.L = v + W is the Lebesgue decomposition
of f.L with respect to A.
From Problem 39.1 (or by repeating the proof of Theorem 39.6), we see that
Df.Ln(x) = Dvn(x) and Df.L(x) = Dv(x) both hold for almost all x. Let Df.Ln =
d//).n =
In for each n. In view of f.LnORk) =I
In dA ::: f.L(JRk) < 00, Levi's
Theorem 22.8 shows that there exists some I E L1(JRk) with In t I. Now
note that vn(E) = IE In dA implies v(E) = IE I dA for each Borel set E. This
implies 1= Dv a.e., and so
holds for almost all x, as desired.
Problem 39.7. This problem reveals some basic properties ollunctions 01

bounded variation on an interval [a, b].
a. II I is differentiable at evelY point and 1/'(x)1 ::: M < 00 holds lor
all x E [a, b], then show that I is absolutely continuous (and hence, 01
bounded variation).
b. Show that the function I: [0, 1] ~ JR defined by
o ifx=O
I(x) = {x 2 cos(~) if 0 < x ::: 1
is differentiable at each x, but is not 01 bounded variation (and hence, I

is continuous but not absolutely continuous).
c. II I is a function 01 bounded variation and I/(x)1 2: M > 0 holds lor
each x E [a, b], then show that g(x) = t:X) is a function 01 bounded
variation.
d. II a function I: [a, b] ~ 1R satisfies a Lipschitz condition (i.e., if there

exists a real number M such that I/(x) - l(y)1 :::: Mix - yl holds lor all
x, y E [a, bD. then show that I is absolutely continuous.
Solution. (a) If (ai, bl) •... , (a", bn ) are pairwise disjoint open subintervals of
[a, b], then by the Mean Value Theorem we have
n n
LI!(b;) - I(a;) I:::: M L(b; - a;).
;=1 ;=1
The preceding easily implies that I is an absolutely continuous function.

(b) Only the differentiability of I at zero needs verification. The inequality
If(.~:=b(O) I= Ix cos(f,:) I : : x
for 0 < x :::: 1 yields /,(0) = O. Now if
P" = {o, p,;, J(2"~I)1T' ... , ii, If, I},
then an easy computation shows that the variation of I with respect to the partition
P" is
n
cos 1 + 1.
1T . "" 1 < Vf.
L..,..k-
k=1
This implies Vf = 00 .
(c) Note that for each a :::: x < y :::: b, we have
Ig() y I-
x - g () - If(x)- {(y)1 I
If(x)f(y)1 :::: M2
I/()
x - I()
Y
I•
Therefore, Vg :::: ,~~2 Vf < 00 holds.
(d) Let a function I: [a. b] ~ 1R satisfy a Lipschitz condition as stated in the
problem and let e > O. Put 8 = 11
> 0 and note that if (aJ, b l ), ... , (an, bn) are
pairwise disjoint open subintervals of [a, b] satisfying 'L7=I(b; - a;) < 8, then
n n
LI/(b;) - l(a;)1 :::: LM(b; - a;)
;=1 ;=1
n
= ML(b; -a;) < M8 = e.
;=1
386 Chapter 7: SPECIAL 10PICS IN INTEGRAnON
Problem 39.8. This p'roblem presents an example of a continuous increasing

function (and hence, of bounded variation) that is not absolutely continuous.
Consider the Cantor set C as constructed in Example 6.15 of the text. Recall
that C was obtained from [0, 1] by removing certain open intervals by steps. 1n
the first step we removed the open middle third interval. At the nth step there were
2n- 1 closed intervals, all of the same length, and we removed the open middle
third interval from each one of them. Let us denote by 1), ... , 1;._1 (countedfrom
left to right) the removed open.intervals at the nth step. Now, define the function
f: [0, 1] -+ [0, 1] asfollows:
i. f(O) = 0;
ii. if x E It for some 1 ::: i ::: 2n- 1, then f(x) = (2i - 1)/2n ; and
iii. ifx E C with x :j:. 0, then f(x) = sup{f(t): t < x and t E [0, 1] \ C}.
Part of the graph of f is shown in Figure 7.1.
a. Show that f is an increasing continuous function from [0, 1] to [0, 1].
b. Show that f'(x) = 0for almost all x.
c. Show that f is not absolutely continuous.
d. Show that JL f 1.. A holds.
Solution. Notice again that parts of the graph of the function f are shown in
Figure 7.1.
(a) Straightforward.
(b) Observe that f is constant on each lin. This implies that f' (x) holds =°
...'1'
3
4" r r
.1..
2 l-----<i,j
r
1
ri
4" !~
~.
!
1 1. .1.. 1. 1. 8 x
"9 9 3 .3 9 "9
FIGURE 7.1.
Section 39: DIFFERENTIAnON AND INTEGRA nON 387
°
for all x E [0, 1] \ C. Since A(C) = 0, it follows that f'(x) = holds for almost
all x.
(c) If f is absolutely continuous, then by Theorem 39.15 we should have
1
1 = f(1) - f(O) = 10 f'(x)dA(x) = 0,
which is impossible.
(d) Note that if B = [0, 1] \ C, then B U C = [0, 1] and J.L feB) = A(C) = 0.
Hence, J.L f ..L A holds.
Let f: [a, b] --+ lR be an absolutely continuous function. Then

Problem 39.9.
°
show that f is a constant function if and only if f' (x) = holds for almost all x.
Solution. Assume that f: [a, b] -+ lR is an absolutely continuous function

°
such that f'(x) = holds for almost all x. By Theorem 39.15, we have
f(x) - f(a) = lX f'(t)dA(t) = °

for each x E [a, b]. Hence, f(x) = f(a) holds for each x E [a, b], so that f is
a constant function.
Problem 39.10. Let f and g be two left continuous functions (on lR). Show that
J.L f = J.Lg holds if and only if f - g is a constant function.
Solution. If f - g is a constant, then it is easy to see that J.L f = J.Lg holds. For
the converse, assume J.L f = J.Lg. If x > 0, then
f(x) - f(O) = J.L f ([0, x») = J.Lg ([0, x») = g(x) - g(O)
implies f(x) - g(x) = f(O) - g(O). Similarly, if x < 0, then
f(O) - f(x) = J.Lf([X, 0») = J.Lg([x, 0») = g(O) - g(x),
and so f(x) - g(x) = f(O) - g(O) holds in this case too.
Problem 39.11. This problem presents another characterization of the norm

dual ofC[a, b]. Start by letting L denote the collection of all functions of bounded
variation on [a, b] that are left continuous and vanish at a.
a. Show that L under the usual algebraic operations is a vector space, and
that f 1-* J.Lf,from L to Mb([a, b]), is lineG1~ one-to-one, and onto.
b. Define 1 ?:: g to'mean that 1 - g is an increasingfunction. (Note that 1 2: g

does not imply 1 ?:: g.) Show that Lunder?:: is a partially ordered vector
space such that 1 ?:: g holds in L if and only if /1-1 2: /1-g in Mb([a, bD.
c. Establish that L with the norm" 1 II = VIII is a Banach lattice.
d. Show, with an appropriate interpretation, that C*[a, b] = L.
Solution. (a) Clearly, L under the usual algebraic operations is a vector space.
Also, it should be clear that I,'t---+ /1-1 from L to Mb([a,b]) is a linear mapping.
To see that 1 't---+ /1-1 is one-to-one, assume /1-1 = O. Then,
I(x) = I(x) - I(a) = /1-1 ([a, x») = 0

holds for all a < x ~ b, and so 1 O. =
Next, we shall show that /1- 't---+ /1-1 is onto. Assume at the beginning that
o ~ /1- E Mb([a, bJ). Define the function
0 ifx<a
I(x)= /1-([a,x») ifa<x<b
1 /1-([a, b]) if x 2: b ,
and note that 1 is increasing, left continuous, and satisfies I(a) =
o. Thus,
1 E L. Now, an easy argument shows that /1- = /1-1' Finally, if/1- E Mb([a, bJ),
then pick two increasing functions I, gEL with /1-+ = =
/1-1 and /1-- /1-g. Note
thatthefunction h = I-g E L satisfies /1- =
/1-+-/1-- = = =
/1-1-/1-g /1-1-g /1-11·
(b) Straightforward.
(c) Since 1 ?:: g holds in L if and only if /1-1 2: /1-g holds in Mb([a, bJ)
and Mb([a, bJ) is a vector lattice, it is easy to see that L must likewise be a
vector lattice. Moreover, by Lemma 38.6 the mapping 1 't---+ /1-1 is a lattice
isomorphism from L onto Mb([a, bJ).
Now, note that if 1 ?:: 0 holds in L (i.e., if 1 is an increasing function), then
holds. Thus, for each 1E L we have
11/1-/11 = 111/1-/111 = 11/1-l/ill = VIII'

This implies that "I" = VIII defines a lattice norm on L, and that 1 't---+ /1- I
from L onto Mb([a, bJ) is a lattice isometry. In particular, L with the norm
"I" = VIII is a Banach lattice.
(d) Using the notation of Theorem 38.7, we see that the composition of the two
operators
is a lattice isometry from L onto C*[a, bJ.

Problem 39.12. Iff: [a, b] -+ 1R is an increasing function, then show that l' is
J:
Lebesgue integrable and that 1'(x)dx :s
f(b) - f(a) holds. Give an example
of an increasing function f for which J: f:(x)dx < f(b) - f(a) holds.
Solution .. Let gn(x) = n[f(x + ~) - f(x)] for each x E [a, b] (where, of

course, f(x) = f(b) for x > b.) Clearly, gn(x) ---+ 1'(x) holds for almost all
x; see Theorem 39.9. On the other hand, the relation gn(x) 2: 0 for each x, and
coupled with Fatou's Lemma show that l' ELI ([a, b1) and that
b
= lb
l a
f'(x)dx lim gn(x)dx :s liminflb gn(x)dx :s f(b) -
a n-+OO n-+oo a
f(a).
Finally, an example of a function that yields strict inequality is provided by the

function described in Problem 39.S.
Problem 39.13. If f: [a, b] -+ 1R is an absolutely continuous junction, then

show that
holds.
Solution. By Theorem 39.15 we have l' E Ll ([a, bl) and
J.L[(E) = Lf'(X)dX
holds for each Borel subset E of [a, b].

Start by observing that if a = to < tl < ... < tn = b is a partition of [a, b],
then
tlfCti) - f(ti-I) = I tlf~/'(X)dX I

_ :::: t f~llf'(X)1 dx = lb If'(x)1 dx.
Therefore, Vf :::: I:
1f'(x)1 dx holds. Now, since the continuous functions are
dense (in the LI-norm) in LI ([a, b1) (Theorem 25.3) and the functions of the
form
n
4> = LaiX(li-IoI;),
i=1
where a = 10 < II < ... < tn = b is a partition of [a, b], are dense (in the
LI-norm) in C[a, b], these functions are also dense in LI ([a, b1). Thus, given
s > 0, there exist a partition a = to < II < ... < tn = b and real numbers
aI, .. , ,an so that 4> = 2:7=1 ai X(I;-IoI;) satisfies 114> - Sgn f' IiI < s. In view of
1(-1 v 4» 1\ I - Sgnf'1 :::: 14> - Sgnf'l, we can assume that 14>(x)1 :::: 1 holds
for all x E [a, b]. Moreover, we have
b
nIl; n
1 a 4>(x)j'(x)dx = t;:ai
/I
1;-1 f'(x)dx = t;:ai[fCti) - f(ti-I)]
:::: Llf(ti) - f(ti-I) I:::: Vf'

i=1
Next, choose a sequence {4>n} of step functions of the previous type satisfying
4>n ---* Sgnf' a.e. (see Lemma 31.6 of the text). In view of l4>nf'l:::: If'l,the
Lebesgue Dominated Convergence Theorem implies
lb If'(x)1 dx = lb f'(x) . Sgn f'(x) dx
= lim {b 4>n(x)f'(x)dx :::: Vf'

n-4QO la
Thus, Vf = I: 1f'(x)1 dx holds.

It is interesting to observe that Vf = IILfl([a, bJ) also holds. To see this, let
a = to < tl < ... < tn = b be an arbitrary partition of [a, b]. Then, we have
n n
:Llf(t;) - f(t;-I)I = :LIIL/([t;-I, t;)) 1
;=1 ;=1
n
::: :L IlL/I ([ti-l , t;)) = IIL/I([a, bJ),
;=1
and so V/ ::: IlL/ I([a, bJ). On the other hand, if E I, ... , En are pairwise disjoint
Borel subsets of [a, b], then
tIIL/(E;)1 =
;=1
til f'(x)dx I::: t llf'(,,)1 dx
;=1 Ei ;=1 Ei
holds, which (by Theorem 36.9) implies that IIL/I([a, bJ) ::: Vf' Consequently,
IIL/I([a, bJ) = V/ holds.
Problem 39.14. For a continuously differentiable function f: [a, b) -+ JR es-

tablish the following properties:
a. The signed measure IL f is absolutely continuous with respect to the Lebesgue
measure and dILt/dA = f' a.e.
b. If g: [a, b) -+ JR is Riemann integrable, then gf' is also Riemann integrable
and
f gdIL/ = lb g(x)f'(x)dx.
Solution. (a) By Problem 39.7, we know that f is absolutely continuous and

so (by Theorem 39.12) IL / is absolutely continuous with respect to the Lebesgue
measure. Now, combining Theorems 39.14(2), 39.8, and 39.4, we see that
~
d)" = D IL/ = f' a.e.
(b) Since f' is a continuous function and g is Riemann integrable, it follows

that gf' is also Riemann (and hence Lebesgue) integrable over [a, b). From
/L I(A) = fA I'd)" for every Borel subset A of [a, b] and Problem 22.15, we see
that
!.
[a,b]
gd/LI= { gl'd)..= {bg(x)/'(X)dX.
J[a,b] Ja
Problem 39.15. For each n consider the increasing continuous function In: JR.
--+ JR. defined by
I
if x> 0,
In(x) = n(x - 1) + 1 if 1 - ~ < x < I,
{ o zif x<I_1.
- n
II I: JR. --+ JR. is a continuous function, then show that

a. I is /LIn -integrable lor each n, and
b. limfld/Lln=/(l).
Solution. Note that Supp /LIn = [1 - ~,1]. This easily implies that I is
/LIn-integrable for each n. In addition, note that I~(x) = n holds for each
1 - ~ :s x :s 1. By the preceding problem, we see that
t t f/ ,/(x)dx
f I d/Lln = JI_1 l(x)/~(x)dx = JI_/I(x)dx = -;;
n n
1
n
.
Therefore, by the Fundamental Theorem of Calculus, we infer that lim f I d /L In =

1(1).
Problem 39.16. Let I: JR. --+ JR. be a (uniformly) bounded function and let
E = {x E JR.: I'(x) exists in JR.}.
If )"(E) = 0, then show that )..(I(E)) = o.

Solution. For each natural number n , let
En = {a E E: /I(x) - I(a)/ :s nix - al for all x E JR.}.
Since I is bounded, it is easy to see that E = U:I En, and so I(E) =

U:;O=I I(En) (see Problem 1.1(6)). Thus, in order to establish that )..(I(E)) = 0,
it suffices to show that )..(I(En)) ='0 holds for each n. To this end, fix n and
e > O.
From 'A(E) = 0, we obtain 'A(En) = 0, and so there exists a sequence of open

I
intervals (b k - rb bk + rk)} satisfying
00 00
En ~ U(bk - rk, bk + rk) and 2 l:rk < e.

k=l k=l
Now, note that if a E En, then there exists some m with Ibm - al < rm, and
hence I!(b m) - f(a)1 ::: nlbm - al < nrm holds. It follows that f(En) ~
U~l (J(bk) - nrk, f(bd + nrk)' Therefore,
00 00
2: 'A((f(bd - nrb f(bd + nrk») = 2n l:rk < 2ne.

k=l k=l
°
Since e > is arbitrary, we infer that 'A(J(En») = 0, as desired. (Compare this
problem with Problem 18.9.)
Problem 39.17. This problem presents an example of a continuous function

f: lR ~ lR which is nowhere differentiable; this example should be compared
with Problem 9.28. Consider the function ¢: [0, 2] ~ lR defined by ¢(x) = x if
°: : x ::: 1 and ¢(x) = 2 - x if 1 < x ::: 2. Extend ¢ to all oflR (periodically)
so that ¢(x) = ¢(x + 2) holds for all x E lR. Now, define the function f:
lR ~ lRby
00
f(x) = l:(ir¢(4'1 x).

11=0
Show that f is a continuous nowhere differentiable function.
Solution. Since the series Z:::our

converges and 0 .::: ¢(x) ::: 1 holds for
all x, it is easy to see that the sequence of partial sums of the series f(x) =
Z:::oUf¢(4 x) converges uniformly to f on lR. So, by Theorem 9.2, f is a
I1
well-defined continuous function. .

Now, fix Xo E lR. The proof of the nondifferentiability of f at Xo will be based
upon the following property of differentiable functions.
• If h: (a, b) ~ lR is differentiable at some Xo E (a, b) and f..L = h'(xo), then

°
for each E > there exists some 8 > 0 such that whenever x, y E (a, b)
satisfy x < Xo < y and y - x < 8, then I h(Y;-~(X) f..L1 < E.
394 Chapter 7: SPECIAL roPICS IN INTEGRA nON
This conclusion follows easily from the inequalities
[h(y)-h(XO)-/L(V-XO)l+[h(Xo)-h(X)-/L(XO-X)ll
I y-x
< I h(y)-h(xo)-/L(Y-xo) \·I.I'-XO 1 + \h(xo)-h(x)-/L(Xo-x» \.\ ~I

y-~ Y~ ~~ y~
::::
h(y)-h(xo) -
y-Xo
II 1
I""
+ 1h(xn)-h(x)
Xo-X
- II
I""
I·
I
Now, for each natural number m, then there exists a unique integer km such that
km :::: 4mxo < k m + 1. Let
and note that Sm :::: .to < tm holds for each m. From tm - Sm = 4-m , we see that
limtm = limsm = Xo.
The reader should keep in mind that if p and q are two integers, then ¢(p) -
¢(q) = 0 if p - q is an even integer and J¢(p) - ¢(q)J = 1 if p - q is an odd
integer. Next, observe that if 12 is a non-negative integer, then 4" tm - 4" Sm = 4"- m •
So, from the definition of ¢, we have:
a. if 12 > m, then ¢(4"tnJ - ¢(4"sm ) = 0,

b. if 12 = m, then ¢(4"tm ) - ¢(4"sm) = I, and
c. if 0 :::: 12 < m, then ¢(4I1 tm ) - ¢(4"sm) = 4"- m •
Therefore, for each m we have
00
IfCtm) - f(sm)1 = II)~r[¢(4I1tm) - IJ

¢(4 sm )JI
11=0
m
= 12)~r[¢(4IJtm) - IJ
¢(4 sm )JI
11=0
m-l
> (~)'" - I)~r 4"-111 > H~)",·
11=0
I
This implies f(t;~~={.~Sm) > I 3;'
for each m. Now, a glance at (.) shows that f
cannot be differentiable at Xo. Since Xo is arbitrary, f is differentiable at no point
ofR.
Section 40: 11IE CHANGE OF VARIABLES FORMULA 395
40. THE CHANGE OF VARIABLES FORMULA
Problem 40.1. Show that an open ball in a Banach space is a connected set.
That is, show that if B is an open ball in a Banach space such that B = 0 1 U O2
holds with both 0 1 and O2 open and disjoint, then either 0 1 = 0 or
O2 =0.
Solution. Let B be an open ball in a Banach space. Assume by way of contradic-

tion that there exist two nonempty open sets 01 and O2 such that B = 01 U O2
and 0 1 n 02 = 0. Fix two elements a E 0 1 and b E O2 , and then define
the function f: [0, 1] --:> B by f(t) = ta + (1 - t)b. Clearly, f(O) = band
f (1) = a. Moreover, in view of the inequality
Ilf(t) - f(s)11 = 11(t - s)a + (s - t)bll ::: (li all + II b ll)lt - sl,
we see that f is a (uniformly) continuous function.
Let a = inf{t E [0,1]: f(t) E Od. Choose a sequence {a/l} of [0,1]
with a/l -* a and f(a/l) E 0 1 for each n. By the continuity of f we have
f(a/l) -* f(a). Since 02 is open and disjoint from Ol,itfollowsthat f(a) ~ 02.
with °
°
In particular, a > must hold. Thus, there exists a sequence {,8/1} of real numbers
< ,8/1 < a for each n and ,8/1 --:> a. By the definition of a, we see
that f(,8/1) E O2 holds for each n, and hence, as above f(a) ~ 0 1. Now, note
that
holds, which is impossible.
Problem 40.2. Let T: V -* JRk be CI-differentiable. Show that the mapping

x ~ T' (x) from V into L(JRk , JRk) is a continuous function.
Solution. We know that
ffi(x)
ax,
T'(x) = :
[
?Tk
ax, (x)
So, if a = (al, ... , ad E JR'k satisfies lIall2 = ( Li=1

k
ai?) 1 = 1, then using the
2
Cauchy-Schwarz inequality, we see that

k k I
/I [T'(x) - T'(y)]a/l 2 = (L[~)~(x) - :~(y)). ajYr

;=] j=]
k k
:s [2:(2:[~(x) - ~(y)]2) . (2:
i=] j=]
k
j=]
an r
I
- k k I
= (2:2:[~(x) - ~(y)]2r .
i=1 j=]
Consequently,
IIT'(x) - T'(y)/I = sup{ II [T'(x) - T'(y)]all2: lIall2 = I}
for each pair x, y E V. This inequality, coupled with the fact that T: V ~ ]Rk
is C I-differentiable, implies that x I---T T' (x ) from V into L (]Rk , ]Rk) is a
continuous function.
Problem 40.3. Show that the Lebesgue measure on ]R2 is "rotation" invariant.
Solution. A "rotation" of the plane is a linear operator T:]R2 ~ ]R2 whose

representing matrix A is orthogonal (i.e., it satisfies AAI = N A = I). Any such
orthogonal matrix is of the form
A = [ co~ e sin e ]
-sme cose '
where e represents the angle of rotation; see Figure 7.2.

In particular, note that det A = 1. Thus, by Lemma 40.4, we see that
A(A(E)) = /detA/A(E) = A(E)
holds for each Lebesgue measurable subset E of ]R2.
Problem 40.4 (Polar Coordinates). Let
E = {(r. e) E ]R2: r 2: 0 and 0 :s e :s 27r}.

Section 40: TIlE CHANGE OF VARIABLES FORMULA 397
.......... ..., x
.•..
,
FIGURE 7.2. Rotation by an Angle e
The transformation T: E -7 JR2 defined by T (r, e) = (r cos e, r sin e), or as it is
usually written
x = r cos e and y = r sin e,
is called the polar coordinate transformation on JR2, shown graphically in
Figure 7.3.
a. Show that A(E \ EO) = O.
b. If A = {(x, 0): x 2: O}, then show that A is a closed subset ofJR 2 whose
(2-dimensional) Lebesgue measure is zero.
c. Show that T: EO -7 JR2 \ A is a diffeomorphism whose Jacobian determi-
nant satisfies her, e) = r for each (r, e) E EO.
d. Show that if G is a Lebesgue measurable subset of E with A( G \ GO) 0, =
then T(G) is a Lebesgue measurable subset ofJR2 . Moreovel~ show that if
fELl (T(G)), then
[
JT(a)
fdA = 11 a
fer cose, r sine)r dr de
holds.
e y
T y .......................... (x,y)
----....
r r x
FIGURE 7.3. The Polar Coordinate Transformation

398 Chapter 7: SPECIAL 10PICS IN INTEGRAnON
Solution. (a) If we consider the sets X = {(r, 0): r 2: O}, Y = {(r, 2:lr): r 2: O},
and Z = {(O,B): 0 :::: B :::: 2rr}, then E\EO = XU Y U Z. To show that
=
)"(E \ EO) 0, it suffices to establish that )"(X) )..(Y) = )"(Z) = O. =
Let Xn = {(r,O): 0 :::: r :::: n} and Yn = {(r,2rr): 0 :::: r :::: n}. In view
of Xn ~ [0, n] x [-e, e] and Yn ~ [0, n] x [2rr - e, 2rr + e], we see that
)..(Xn) = )..(Yn ) = 0 holds for each n. Since Xn t X and Y n t Y, it follows that
)"(X) = )..(Y) = O.
Also, the inclusion Z ~ [-8., e] x [0, 2rr] implies )"(Z) :::: 4rr e for each e > 0,
and thus )"(Z) = O.
(b) This is proven in part (a) previously.
(c) Clearly, T: EO --+ JR2\A is one-to-one, onto, and Cl-differentiable. The
Jacobian determinant is
ax
her, B) = det [ g~ ilYJ
ar - d
et [ cosB . ]-
smB r
al] :~ - -r sin B r cos 8 - ,
which implies that her, 8) = r =1= 0 holds for each (r, B) E £0. The preceding
are enough to guarantee that T: EO --+ JR2 \ A is a diffeomorphism.
(d) Clearly, GO ~ EO. Thus, by part (c), T(GO) is an open subset of JR2 \ A and
T: GO --+ T( GO) is a diffeomorphism.
Since T: JR2 --+ JR2 (defined by T(r,8) = (r cos8, r sinB» is a CI-diffeo-
morphism, it follows from Lemma 40.1 that )"(T(G \ GO») = O. Now, if we
consider the sets A = G, B = T(G), V = GO, and W = T(GO), then
)"(A \ V) = )"(G \ GO) = 0 and )"(T(G) \ T(GO») :::: )"(T(G \ GO») = 0 both
hold. Thus, Theorem 40.8 applies and gives us the desired formula.
Problem 40.5. This problem uses polar coordinates (introduced in the preceding
problem) to present an alternate proofofEuler's formula e-x2 dx = ,.;;r/2. It
a. Foreachr>O,letC,. = ((x,y) E JR 2:x2+y2::::r2, x2:0, Y2:0} and
Sr = [0, rJ x [0, rJ. Show that Cr ~ S,. ~ Cr..f'i:
If f(x, y) = e-(x +y lJ, then show that
2
b.
where).. is the two-dimensional Lebesgue measure.

c. Use the change of variables to polar coordinates and Fubini's Theorem to
show that
!!. r
rr '
1c,
f, d)" =
1 2
[e- I t dt dB
0 10
2
= -(1
4
- e-r-).
r r t/'I x
FIGURE 7.4.
d. Use (b) to establish that
4(1
7r - e-r 2 ) ::=: ( l0
r
e-·t.2 dx )
2
::=: 4(1
7r - e- 2r 2 ),
and then let r -+ 00 to obtain the desired formula.
Solution. (a) Geometrically the three sets are as shown in Figure 7.4.
(b) Since f(x, y) = e-(x +y2) ~ 0 holds for all (x, y), we see that
2
fXc, ::=: fXs, ::=: fXc,.fJ.'

and the desired inequality follows.
(c) Consider the polar coordinates transformation described in the preceding prob-
lem. For the set G = {(t, e): 0 ::=: t ::=: rand 0 ::=: e ::=: I}' we have
1 c,
f d).. = r
JT(G)
f d)" = 11 G
f(t cose, t sine)t dt de
ll
Y r
= 0 0 e-1-tdtde
7
= Hl-e- r 2 ).
(d) Note that

Thus, using (b) and (c), we see that
and by letting r -+ 00 we get Uo e-

oo x2
dX) 2 = %.
Problem 40.6. In R 4 , "double" polar coordinates are defined by
x = r cos e, y = r sin e , Z = P cos ifJ, W = P sin ifJ.
State the change of variables formula for this transformation, and use it to show
that the "volume" of the open ball in R4 with center at zero and radius a is
4:n- 2a4 •
Solution. The transfonnation T: 1R4 -+ R4 is given by
T (r, p, e, ifJ) = (r cos e, r sin e, p cos ifJ, p sin ifJ )

for each (r, p, e, ifJ) E R4. Its Jacobian detenninant is
. [ cose sine
°
h(r, p, e, ifJ) = det 0. e ° cosifJ
.
-rsm
°
rcose
°
-psinifJ
°
Write R4 = 1R2 X R2, and consider the Lebesgue measure on 1R4 as the product
measure of the corresponding Lebesgue measures on the two factors. Fix a > 0,
and let
and
F = [0, 2:n-] x [0, 2:n-] S; 1R2.
Put G = E x F S; 1R2 X 1R2 = 1R4, and note that T(G) = B, the open ball
of R4 with center at zero and radius a. Now, if C = {(r, p): rp = OJ S; R2
and D =
{(r, p, 0, 0): r ::: 0, p ::: OJ S; R 4 , then both sets are closed in their
corresponding spaces and their corresponding Lebesgue measures are zero. Thus,
if
V=(E\C)x[(0,2rr)x(O,2rr)] and W=B\D,
then both V and Ware open subsets of JR4 and T: V -+ W is a diffeomor-

phism (onto). Since A(G \ V) = A(B \ W) = 0, Theorem 40.8 combined with
Fubini's Theorem shows that
Problem 40.7 (Cylindrical Coordinates). Let
E = {(I', e, z) E JR3: I' 2: 0, 0::: e ::: 2rr, Z E JR).
The transformation T: E -+ JR3 defined by T(r, e, z) = (I' cose, I' sine, z) or as

it is usually written
x = I' cos e, y= I' sin e, z = z,
is called the cylindrical coordinate transformation, shown graphically in

Figure 7.5.
a. Show that A(E \ £0) = 0.
b. If A = {(x, 0, z) E JR3: x 2: 0, Z E JR}, then show that A is a closed subset
ofJR3 whose (three-dimensional) Lebesgue measure is zero.
Z Z
(x, y, z)
(I', e, z)
•
! 21Z"
e y
FIGURE 7.5. The Cylindrical Coordinate Transformation

402 Chapter 7: SPECIAL lDPICS IN INTEGRATION
c. Show that T: EO -+ R3 \ A is a diffeomorphism whose Jacobian determi-

nant satisfies her, e, z) = I" for each (I", e, z) E EO.
d. Show that ifG is a Lebesgue measurable subset of E with A(G \ GO) = 0,
then T(G) is a Lebesgue measurable subset ofR3. Moreovel~ show that if
f E LJ(T(G», then
r
JT(G)
fdA=jj
.
rf(rcose,r sine, z)rdrdedz
JG
holds.
Problem 40.8 (Spherical Coordinates). Let
E = {(I", e, ¢) E R3: I" :::: 0, 0 :s e :s 2.7r, 0 :s ¢ :s .7r}.

The transformation T: E -+ R3 defined by
T(r, e, ¢) = (r cos e sin ¢, r sine sin ¢, r cos ¢),

or as it is usually written
x = r cos esin ¢, y = r sin esin ¢, z = r cos ¢,

is called the spherical coordinate transformation, shown graphically in
Figure 7.6.
a. Show that A(E \ EO) == o.
b. If A = {(x, 0, z): x :::: 0 and z E R}, then show that A is a closed subset
qfR3 whose (3-dimensional) Lebesgue measure is zero.
z
(x,y, z)
e y
x
FIGURE 7.6. The Spherical Coordinate Transformation
Section 40: TIIE CHANGE OF VARIABLES FORMULA 403
c. Show that T: EO -+ lR3 \ A is a diffeomorphism whose Jacobian determi-

nant satisfies ly(r, e, ¢) = _r2 sin ¢.
d. Show that ifG is a Lebesgue measurable subset 01 E with )"(G \ GO) = 0,
then T(G) is a measurable subset ·01JR.3 . In addition, show that if I E
L](T(G», then
rid)"
JT(O)
=// r I(r cose sin ¢, r sine sin¢, r cos¢)r2 sin¢ dr de d¢
Jo
holds.

Principles of Real Analysis-Solutions-Aliprantis PDF

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

Principles of Real Analysis-Solutions-Aliprantis PDF

Uploaded by

Copyright:

Available Formats

PROBLEMS IN

San Diego London Boston

New York Sydney Tokyo Toronto

All Rights Reserved.

ACADEMIC PRESS LIMITED

International Standard Book Number 0-12-050253-4

Printed in the United States of America

CHAPTER 1. FUNDAMENTALS OF REAL ANALYSIS

CHAPTER 2. IDPOLOGY AND CONTINUITY 65

CHAPTER 3. THE THEORY OF MEASURE 107

CHAPTER 4. THE LEBESGUE INTEGRAL 171

CHAPTER 5. NORMED SPACES AND Lp-SPACES 239

CHAPTER 6. HILBERT SPACES 297

CHAPTER 7. SPECIAL IDPICS IN INTEGRATION 345

C. D. ALIPRANTIS AND O. BURKINSHAW

Probiem 1.1. Establish the following set theoretic relations:

(3) Note that

xEA\B ~ x E A and x rf. B

X E (A n B)C ~ xrf.AnB ~ xrf.A orxrf.B

Y E J(UAi) ~ :3 x E UAi with Y = f(x)

x E f-1(UB i ) ~ f(X)'E UBi ~ :3 i E I with f(x) E Bi

(9) Note that

x E f-l(nBi) ..;=> f(x) E nBi ..;=> f(x) E Bi for each i E I

..;=> x E f-I(Bi) foreach i E I ..;=> x E nf-I(Bi ).

(10) Observe that

x E f- I (BC) ..;=> f(x) E BC ..;=> f(x) rj B

X flY \ Z = [X \ (Y U Z)] U [Y \ (X U Z)]

Z \ (X fly) = [Z \ (X U Y)] U [X n Y n Z].

For instance, to verify the first identity, note that

x E X flY \ Z ..;=> [x E X \ Y or x E Y \ X] and x rj Z

(AIlB)IlC = [(AIlB) \ C] u [C \ (AIlB)]

Problem 1.4. Give an example of a function f: X -+ Y and two subs~ts A and

Solution. Define f: {O, I} -+ {O, I} by f(O) = f(l) = 0. If A = {OJ and

Solution. Clearly, x E f- 1(A) if and only if f(x) E A. Thus, fU- 1(A» £; A.

Problem 1.7. Show thatafunction f; X -+ Y is onto ifand only if fU- 1(B»=B

Solution. Assume that f is onto and B £; Y. If b E B, then there exists some

Problem 1.8. Let X ~Y .....!!..:,. Z. II A: S; Z, show that

Solution. Observe that for each x E X we have

[(h 0 g) 0 I lex) = h 0 g(f(x)) = h(g(f(x))) = h«g 0 f)(x)) = [h 0 (g 0 f)](x).

Problem 1.10. Let I: X -+ Y. Show that the relation R on X, defined by

Problem 1.11. II X and Yare sets, then show that

P(X) n P(Y) = P(X n Y) and P(X) U P(Y) S; P(X U Y).

Solution. (a) Note that

A E P(X) n P(Y) {:=::> A S; X and A S; Y

A E P(X) U P(Y) :=:} A S; X or A S; Y ==> A S; X U Y :=:} A E P(X U Y).

2. COUNTABLE AND UNCOUNTABLE SETS

Solution. This follows immediately from Theorem 2.6.

Problem 2.4. Let A be an uncountable set and let B be a countable subset of

Solution. Let B = {b I , b 2 , ••• }. Since A is uncountable, the set A \ B is also

Then f is one-to-one and onto, proving that A ~ A \ B.

Problem 2.5. Assume that f: A -+ B is a sUljective (onto) function between

Solution. If A and B are equivalent, then there exists a function f: A -+ B

Solution. For each subset A of IN define the sequence f(A) = {XII} by Xn = 1

Problem 2.10. Any complex number that is a root of a (nonzero) polynomial

Problem 2.11. For an arbitrary function f: 1R ~ 1R show that the set

A = {a E 1R: lim f(x) exists and lim f(x) =1= f(a)}

Clearly, A = UreQA r holds. Thus, in order to establish that A is at most

holds for each y E Ia with y i= a, we see that y ¢. Ar for each y E Ia \{a}. In

Problem 2.13. Using mathematical induction prove the following:

(1 + a)"+\ = (1+ a)(1 + a)" ~ (1 + a)(1 + na) = 1 + na + a + na2

(1 + a)"+\ = (1+ a)(l + a)" < (1 + a)(1 + 3"a)

which is the desired inequality valid when n is replaced by n + 1. By the Principle

cos[(n + 1)1l'] = cos(n1l' + 1l') = cos(n1l') cos 1l' - sin(n1l')sin1l'