Exercises and Solutions Manual

Springer
New York
Berlin
Heidelberg
Barcelona
Budapest
Hong Kong
London
Milan
Paris
Santa Clara
Singapore
Tokyo
Gerard Letac
Exercises and Solutions
Manual for

Integration
and Probability
by Paul Malliavin
Translated by Leslie Kay

, Springer
Gerard Letac
Laboratoire de Statistique
Universite Paul Sabatier
118 Route de Narbonne
F-31062 Toulouse
France

Translator:
Leslie Kay
Department of Mathematics
Virginia Polytechnic Institute
and State University
Blacksburg, VA 24061
USA

Mathematics Subject Classification (1991): 28-01, 43A25, 6OH07

Printed on acid-free paper.

© 1995 Springer-Verlag New York, Inc.

All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New
York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly
analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereaf-
ter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even
if the former are not especially identified, is not to be taken as a sign that such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely
by anyone.

Production managed by Frank Ganz; manufacturing supervised by Jeffrey Taub.

Camera-ready copy prepared by the translator.

98765432

ISBN-13 :978-0-387-94421-0 e-ISBN-13 :978-1-4612-4212-3

DOl: 10.1007/978-1-4612-4212-3
Contents

I Measurable Spaces and Integrable Functions 1

1 u-algebras and partitions . . . . . . 1
2 r-families . . . . . . . . . . . . . . . 3
3 Monotone classes and independence 3
4 Banach limits . . . . . . . . . . . . 4
5 A strange probability measure .. . 5
6 Integration and distribution functions 6
7 Evaluating 2:::='=1(-sn 8
8 Monotone convergence . . . . . . . . 9
9 Vector integration . . . . . . . . . . . 10
10 Convergence in measure and composition of functions 11
11 Principle of separation of variables. . 12
12 The Cauchy-Schwarz inequality . . . 13
13 Test that X ~ Y almost everywhere. 14
14 Image of a measure . . . . . . . . . . 14
15 Primitives of square integrable functions 15

II Borel Measures and Radon Measures 17

1 Positive measures on an open interval 17
2 Distribution functions . 20
3 Convexity and growth . 21
4 Convexity and measure 22
vi Contents

5 Integral representation of positive convex functions

on (0,00) . . . . . . . . . . . . . . . . . . . 23
6 Integral representations of Askey functions 24
7 Gauss's inequality . . . . . . . . . . . . . 26
8 Integral of a decreasing function . . . . . 27
9 Second mean value theorem for integrals 27
10 Variance of a distribution on [0, 1] . . . . 28
11 Variance of the distribution of a convex function on [0, 1] 28
12 Rational functions which preserve Lebesgue measure 29
13 A measure on the half-plane . . . . . . . . 33
14 Weak convergence and moments . . . . . . . . . . . . 34
15 Improper integrals and Lebesgue measure. . . . . . .. 35
16 1000 si~xdx, fooo(cosax - cos bx)d: ' 1000 (cos ax - cosbx); 36
17 Comparisons between different LP spaces . . 36
18 Differentiation under the integral sign . . . . . . . . . 37
19 Laplace transform of a measure on [0, +00) . . . . . . 40
20 Comparison of vague, weak, and narrow convergence 41
21 Weak compactness of measures. . . . . . . . . . . 41
22 Vague convergence and limit of JLn(O) . . . . . . . 41
23 Vague convergence and restriction to a closed set 43
24 Change of variables in an integral . . 45
25 Image of a measure and the Jacobian . . . . . . . 45

III Fourier Analysis 49

1 Characterizations of radial measures . 49
2 Radial measures and independence 50
3 Area of the sphere. . . . . . . . . . . 51
4 Fourier transform of the Poisson kernel of R~+1 52
5 Askey-Polya functions. . . . . . . . 53
6 Symmetric convex sets in the plane
and measures on [0,1T) .• 55
7 T. Ferguson's theorem .. 59
8 A counterexample of Herz 59
9 Riesz kernels. . . . . . . . 61
10 Measures on the circle and holomorphic functions 62
11 Harmonic polynomials and the Fourier transform 63
12 Bernstein's inequality. . . . . 64
13 Cauchy's functional equation . . . . . . 66
14 Poisson's formula . . . . . . . . . . . . 66
15 A list of Fourier-Plancheral transforms 68
16 Fourier-Plancheral transform of a rational function 69
17 Computing some Fourier-Plancheral transforms .. 70
18 Expressing the Fourier-Plancheral transform as a limit 71
19 An identity for the Fourier-Plancheral transform. 71
20 The Hilbert transform on L2(R) . . . . . . . . . . . .. 71
 Contents vii

21 Action of Ll(R) on L2(R) . . . . . . . . . . . 72

22 Another expression for the Hilbert transform . 73
23 A table of properties of the Hilbert transform 73
24 Computing some Hilbert transforms . . . 74
25 The Hilbert transform and distributions 74
26 Sobolev spaces on R 75
27 H. Weyl's inequality. . . . . . . . . . . . 76

IV Hilbert Space Methods and Limit Theorems

in Probability Theory 79
1 Fancy dice . . . . . . . . . . . . . . . . 79
2 The geometric distribution . . . . . . . 80
3 The binomial and Poisson distributions 80
4 Construction of given distributions 83
5 Von Neumann's method 84
6 The laws of large numbers . . . . 85
7 Etemadi's method . . . . . . . . . 86
8 A lemma on the random walks Sn 89
9 o(s) = limn--+oo(P[Sn 2: s· n])l/n exists 89
10 Evaluating o(s) in some concrete cases 90
11 Algebra of the gamma and beta distributions . 92
12 The gamma distribution and the normal distribution 93
13 The Cauchy distribution and the normal distribution 94
14 A probabilistic proof of Stirling's formula . . . . 96
15 Maxwell's theorem. . . . . . . . . . . . . . . . . 97
16 If Xl and X 2 are independent, then ()f;x"\)l/2
is uniform ................1..2... 98
17 Isotropy of pairs and triplets of independent variables 100
18 The only invertible distributions are concentrated
at a point . . . . . . . . . . . . . . . . . . 102
19 Isotropic multiples of normal distributions 103
20 Poincare's lemma . . . . . . . . . 104
21 Schoenberg's theorem . . . . . . . . 104
22 A property of radial distributions 106
23 Brownian motion hits a hyperplane
in a Cauchy distribution 107
24 Pittinger's inequality . . 109
25 Cylindrical probabilities. 112
26 Minlos's lemma . . . . . 113
27 Condition that a cylindrical probability
be a probability measure 115
28 Lindeberg's theorem. . . 116
29 H. Chernoff's inequality. 117
30 Gebelein's inequality .. 120
31 Fourier transform of the Hermite polynomials 121
viii Contents

32 Another definition of conditional expectation. 122

33 Monotone continuity of conditional expectations. 123
34 Concrete computation of conditional expectations 123
35 Conditional expectations and independence . 124
36 E(XIY) = Y and E(YIX) = X . . . . . . . . . . 125
37 Warnings about conditional expectations . . . . . 126
38 Conditional expectations in the absolutely continuous case
and the Gaussian case. . 126
39 Examples of martingales . . . 127
40 A reversed martingale . . . . . 128
41 A probabilistic approximation
of an arithmetic conjecture . . 129
42 A criterion for uniform integrability 129
43 The Galton-Watson process and martingales 130

v Gaussian Sobolev Spaces

and Stochastic Calculus of Variations 133
1 d and 6 cannot both be continuous 133
2 Growth of the Hermite polynomials 134
3 Viskov's lemma .. . . . . . . 134
4 Cantelli's conjecture . . . . . . 136
5 Lancaster probabilities in R2 . 138
6 Sarmanov's theorem. . . . . . 139
I
Measurable Spaces
and Integrable Functions

Problem 1-1. If 9 is a family of subsets of a set X, we denote bya(Q) the

Boolean algebra generated by 9 and by a(Q) the a-algebra generated by 9.
A partition of X is a family P = {Pj}jEJ of nonempty subsets of X such
that Pi n Pj = 0 ifi =1= j and UjEJ = X.
(1) Let P = {Pj}jEJ be a partition of X. Characterize
a) a(P) if J is finite,
b) a(P) if J is infinite,
c) a(P) if J is finite or countable, and
d) a(P) if J is uncountably infinite.
(2) Show that the family A of subsets of X is a Boolean algebra generated
by a finite number of elements if and only if there exists a partition P =
{Pj }jEJ, with J finite, such that A = a(P).
(3) Let A be a a-algebra on a countable set X. Show that there exists a
partition P of X such that A = a(P).
(4) Show that a a-algebra never has a countable number of elements.

SOLUTION. (1) If T C J, let A(T) = UjETPj , with the convention that

A(0) = 0.
a) T f-> A(T) is a bijection between the set P(J) of all subsets of J and
the set a(P) .

• A = {A(T) : T E P( J)} is a Boolean algebra which contains P; hence

it contains a(P) .

• Conversely, it is trivial that a(P) :J A.

2 I. Measurable Spaces and Integrable Functions

b) Let :F(J) be the set of subsets of J which are either finite or cofinite
(complements of finite sets). Then F(J) is a Boolean algebra. Moreover,
T ....... A(T) is a bijection between F(J) and a(P).

• A = {A(T) : T E F( J)} is a Boolean algebra which contains P and

hence a(P).

• Conversely, a(P) J A trivially.

c) T ....... A(T) is a bijection between P(J) and a(P).

• A = {A(T) : T E P(J)} is a a-algebra which contains P and hence

a(P).

• Conversely, a(P) J A trivially.

d) Let V(J) be the set of subsets of J which are either finite, countable,
or cocountable (complements of finite or countable sets). Then V(J) is a
a-algebra. Moreover, T ....... A(T) is a bijection between V(J) and a(P).

• A = {A(T) : T E V(J)} is a a-algebra which contains P and hence

a(P).

• a(P) J A trivially.

(2) Sufficiency is trivial. If A is generated by Q = {G 1 ,G2, ... ,Gn }, set

G} = Gj and ail = Gj. For every f = (fll f2, ... , fn) with fj = ±1, let
G. = nj=l G;;. Let E = {f : G. 1=- 0}. Then P = {G.hEE is a partition of
X, and it is easy to see that a(Q) C a(P) and a(P) C a(Q).
(3) For every x in X, let A(x) = n{A : A E A and x E A} and let
an equivalence relation,..., be defined on X by x ,..., y if A(x) = A(y). Let
P = {Pj hEJ be the partition of X into equivalence classes. (This is just
the image of the mapping x ....... A(x) from X to P(X).) We show that
A(x) E A for every x.

(A(xW = {A C : A E A and x E A} = {B : B E A and x ¢ B}.

Hence, for every y E (A(xW, there exists B(y) such that y E B(y), B(y) E
A, and x ¢ B(y). Thus

(A(xnc = U{B(y) : y E (A(xnc}.

This union is countable since X is. Hence AC(x) E A and therefore a(P) C
A.
Conversely, A E a(P) for every A in A. For, if not, there would exist
j E J and x, y E X such that x E Pj n A and y E Pj n A c. Since x ,..., y and
y ¢ A(x), this gives a contradiction. Hence A C a(P).
 I. Measurable Spaces and Integrable Functions 3

(4) Let A be a a-algebra on X with a countable number of elements. Let

A(x), "', and P = {Pj}jEJ be defined as in (3). Then A(x) E A. But either
J is finite and (1) implies that IAI = 21J1 , or J is infinite and (lc) and (ld)
imply that a(P) is uncountable.

Problem 1-2. Let 9 be a family of subsets of a set X such that X E 9 and

9 is closed under finite intersections. An r-family is a family R of subsets
of X which is closed under finite intersections of pairwise disjoint sets and
such that if Bl and B2 E 'R with Bl C B2, then B2 \ Bl E 'R. Let r(g)
be the smallest r-family containing g. Show that r(9) equals the Boolean
algebra a(9) generated by 9.

METHOD. Consider the families

'Rl = {B : B E r{Q) and An BE r(Q) VA E 9}

R2 = {B : B E r(9) and An BE r(9) VA E r{Q)}
and show that they are r-families.

SOLUTION. It is clear that a Boolean algebra is an r-family and hence that

a(9) ~ r(9). Conversely, it must be shown that r(9) is a Boolean algebra,
which will imply that r{Q) ~ a(9). Since X E 9 C r(9), clearly r(9) is
closed under complements. It remains to show that r(9) is closed under
finite intersections.
Rl and 'R2 are obviously closed under disjoint unions. If B} E 'Ri and
B2 E Ri with i = 1 or 2 and Bl C B 2, then, for every A,

which implies that 'Rl and'R2 are r-families.

Finally, since 9 is closed under intersections, 9 C 'Rl C r(Q) and thus
Rl = r(9). It follows that 9 C R2 C r(9), which implies that r{Q) is closed
under intersections.

Problem 1-3. Let 91 and 92 be two nonempty families of subsets of a set

X which are closed under finite intersections. Let AI, A2, and A denote
the a-algebras generated by 91, 92, and 91 U 92, respectively. Let P be a
measure of total mass 1 on (X, A). Show that if

P(A I n A 2 ) = P(A l )P(A2) for all Al E 91 and A2 E 92,

then the same equality holds for all Al E Al and A2 E A2.

METHOD. Consider the families

4 I. Measurable Spaces and Integrable Functions

M2 = {A ; A E A and P(A I n A) = P(Al)P(A) VAl E AI}

and apply the theorem on monotone classes, using Problem 1-2.

SOLUTION. It is clear that Ml and M2 are monotone classes. By hypoth-

esis, 91 C MI. Evidently X E MI. Applying the result of Problem 2 to
the family Fl = 91 U {X}, which is closed under finite intersections, we
see that Ml is an r-family; hence

By the theorem on monotone classes, we can write

Similarly, it is clear that X E M 2 • We now apply the result of Problem

1-2 to the family F2 = 92 U {X}, which is closed under finite intersections.
Clearly M2 is an r-family, and hence

By the theorem on monotone classes,

which completes the proof.

REMARKS.
1. This result is especially useful in probability theory. Thus, if X = R2,
Al(X) = {(X},X2); Xl < X}, and A 2(y) = {(Xl,X2); X2 < y}, then
91 = {Al(X); X E R} and 92 = {A 2(y); Y E R} are closed under finite
intersections and A is the set of Borel subsets of R2. If P is a probability
measure on (R2, A), it is the distribution of a pair (Xl, X 2 ) of real ran-
dom variables. By Problem II-3, (Xl, X 2 ) is a pair of independent random
variables if and only if

P[XI < X; X2 < y] = P[XI < x] . P[X2 < y]

for all (x,y) E R2.
2. The result can be extended from two factors to n factors by constructing
monotone classes Mk for k = 1,2, ... , n and using induction on k. One
proves first that Mk :::> 9k, then that Mk :::> Ak.
Problem 1-4. Let x = {xn}~=o and let

£00 = {X; Xn E R Vn E Nand Ilxttoo = s~p tXnt < oo}.

Define T ; £00 -+ £00 by (Tx)o = Xo and (Tx)n = Xn - Xn-l if n > o.
 I. Measurable Spaces and Integrable Functions 5

(1) If e = (1,1, ... ,1, ... ), show that the equation Tx = e has no solution
x in loo.
(2) Let F = TR,00 be the image ofT. Assume without proof that there exists
a continuous linear functional f on loo such that f(x) = 0 for every x in
F, f(e) = 1, and sup{lf(x)1 : IIxll oo ~ I} < +00 (Hahn-Banach Theorem).
Show that if x = {xn}~o is such that Xn 2: 0 for every n, then f(x) 2: O.
(3) Let S : loo -+ lOO be defined by (Sx)n = Xn+l if n 2: o. Show that
f(x) = f(Sx) for every x in lOO.
(4) Show that liminfn...... +oo Xn 2: 0 implies that f(x) 2: o. Conclude that
liminf n...... +oo Xn ~ f(x) ~ limsupn ...... +oo Xn for every x E £00.
(5) Let A c N and let lA E lOO be defined by lA(n) = 0 if n =1= A and
lA(n) = 1 ifn E A. If P(A) = f(lA), show that P(AUB) = P(A) +P(B)
if A n B = 0 and that P does not satisfy the countable additivity axiom.

SOLUTION. (1) Tx = e implies that Xo = 1 and, by induction on n, that

Xn = n + 1. Hence e =1= Tl oo •
(2) Assume that f(x) < o. Then IIxli oo > OJ if Y = e - IIxlloo' with Yn =
1 - IIx:~oo' then 0 ~ Yn ~ 1. Hence If(y)1 ~ 1, but f(y) = 1 - ~ > 1, a
contradiction.
(3) In order to see that f(x) = f(Sx), it suffices to show that x - Sx E F,
and hence that for every x the equation x - Sx = Ty has a solution y.
Indeed, we find that Yn = Xo - Xn+1 for every n, and this defines an
element y of loo .
(4) If liminfn......+oo Xn 2: 0, then for every e > 0 there exists an integer
N(e) such that Xk 2: -e if k 2: N(e). Thus SN(f)X + ee)n 2: 0 for every
nj by (2), f(SN(E)X + ee) 2: 0 and hence f(SN(E)X) 2: -e. By (3), we have
f(x) = f(SN(f)X), whence f(x) 2: -e for every e > 0, or f(x) 2: O.
Moreover, if x is an arbitrary element of loo, let m = lim infn...... +oo Xn and
let M = lim suPn ...... +oo x n . Then f( x - me) and f( Me - x) are nonnegative
by the first paragraph of (4), and hence m ~ f(x) ~ M.
(5) That P(A U B) = P(A) + P(B) if A n B = 0 follows immediately from
the linearity of f. To see that P does not satisfy the countable additivity
axiom, consider P({k}) = f(l{k}). Since limn-++ool{k}(n) = 0, it follows
from (4) that P({kll) = 0, and hence that

1 = f(e) = P(N) = P(UkEN{k}) =1= L P({k}) = O.

kEN

REMARKS. The linear functional f above is called a Banach limitj it cannot

be written down explicitly since it is constructed by means of the Hahn-
Banach theorem and the axiom of choice. Similarly, it is impossible to
give an explicit example of an additive but not a-additive measure on a
a-algebra.
6 I. Measurable Spaces and Integrable Functions

Problem 1-5. Let X be an uncountable set and let A be the {}'-algebm

genemted by the family of I-element subsets of X. (See Problem 1, question
(Id}.) Let P : A -+ [0,1] be defined by
peA) = 0 if A is finite or countable
P( A} = 1 if A is cocountable.
(A is cocountable if A C is finite or countable.) Show that P is a probability
measure on (X, A).

SOLUTION. Let {An}~=o be a sequence of pairwise disjoint elements of A.

Then

• either, for every n, An is finite or countable, in which case so is

u~=oAn and hence

= 0 = L 0 = L P(An};
00 00

P(U~=oAn}
n=O n=O

• or there exists no E N such that Ana is cocountable. Since An is

disjoint from Ano if n -=1= no, un#oAn is finite or countable. Hence
00

P(U~=oAn} = 1 = P(Ano} + P(Un#noAn) = L P(An).

n=O

Problem 1-6. Let (X, A, JL) be a measure space and let f be a nonnegative
measumble function on X. For every t ~ 0, set
F(t) = JL{x : f(x) > t} and G(t) = JL{x : f(x} ~ t}.
(I) Assume that f(X) S;;; N and that f is integmble. Prove that

JJ
r f(x}dJL(x) = L F(n) 00
= L G(n}.
00

x n=O n=l

METHOD. Set JLn = JL{x : f(x} = n} and show that Ix f(x}JL(dx)

r:::'=onJLn'
(2) Assume that r is integrable for a > O. Prove that

Ix r(x)dJL(x) = a 10+ 00
t a- 1F(t)dt = a 10+ 00
ta-1G(t)dt.

METHOD. Show that (2) holds for a = 1 by considering the functions

fn(x) = [2" !Jz:») , where [a] means "the greatest integer :s; a", and using the
 I. Measurable Spaces and Integrable Functions 7

monotone convergence theorem. The general case can then be reduced to

the case Q = 1.

SOLUTION. (1) If Ik(X) = x for x ~ k and h(x) = 0 for x > k, Ik is a

simple function (one which assumes only finitely many values) and

1 x
h(x)dtt(x) = 0 x tt( {x: x > k}) + L
n=O
k
nttn·

Since Ik approaches I as k --+ +00, I is integrable, and 0 ~ h ~ I, the

dominated convergence theorem (1-7.6) implies that
k 00

( h(x)tt(dx) = lim ( Ik(X)tt(dx) = lim L nttn = L nttn·

}x k.... +oo } X k.... +oo n=O n=O

This implies that F(n) < +00. Observe next that

N N-l N-l 00 00

L nttn = L(F(n) - F(N)) ~ L F(n) = Lmin(n,N)ttn ~ L nttn·

n=O n=O n=O n=O n=O
Hence
00 00

L nttn =L F(n).
n=O n=O
Since F(n) = 1::+1 F(t)dt, the first equality has been proved. The second
is clear, since G(n + 1) = F(n) for every n ~ O.
(2) We begin by proving the second equality. Let A(t) = {x : I(x) = t}. The
A(t) are disjoint and, since tt is u-finite, D = {t : tt(A(t)) > O} cannot be
uncountable. The functions t",-l F(t) and t",-lG(t) coincide on [0, +00) \D
and their integrals are equal.
Assume now that Q = 1.
Since 2n In has integer values, it follows from (1) that

i In (x)dtt(x) = 21n ~ G ( ; ) .
Since G is decreasing, we have the following bounds

"
00
~~
1 ( -k ) <
-G
~ -
1+ 00
G(x)dx < "
-~~
00 1 ( -k ) .
-G
~
k=l 0 k=O

Moreover, In(x) increases to I(x), and hence In{x)dtt(x)Ix I{x)dtt(x) Ix

as n --+ +00 by the monotone convergence theorem. This implies the result
for Q = 1.
8 I. Measurable Spaces and Integrable Functions

Finally, if a> 0, we can write

Ix r(x)dp,(x) = 1 00
p,{x: r(x) > u}du = a 1 00
t a - 1 F(t)dt,

where the last equality is obtained by the change of variable 11. = t a and
the first by applying the result for a = 1 to the function rather than f. r
Problem 1-7. For 0 < r < 1, we write the Poisson kernel as
00 1- r2
Pr ( 8) =1+ 2 ' " rn cos n8 = 1
L.-;
2 8
- r cos + r
2.
n=1

(1) Show that r2 + cos 8(1 - 2r) ~ 0 if 0 ~ 8 ~ and ~ ~ r ~ 1. Deduce 7r

2 (1_r 2 '8 2 rIC 2
that 0 Pr(O) ~ ~ and evaluate limr-+l Jo 8 Pr(O)dO.
(2) Show that Jo"lf 02Pr(8)d8 = 33 + 47r 2::'1 and use this to derive (-.;r
another expression for liIDr-+l 82Pr(O)dO. J:
(3) Use (1) and (2) to compute the sums of the series 2::=1 (-,.V n ,
,",00 1 d ,",00 1
Lm=1 (2n-l)2, an Lm=1 ~.

(4) Express J~(log(1 - X))2~ as the sum of a double series and show that
rl(log(1 _ x 2))2dx = 2 ,",00 1.
Jo ? Lm=1 ~

SOLUTION.
(1) 0 ~ (l-r)2 = r2 + 1- 2r ~ r2 +cos8(1-2r). The inequality 02Pr(8) ~
(~=:~~2 is immediate. Since Pr(O) ~ 0 and 1-~:e8 is integrable, it follows
from the dominated convergence theorem that

l "1f 02dO
-=---:-----;:---;;-->
o 1-2rcosO+r2
11"1f
2 0
02d8
l-cos8
as r->1.

Hence limr-+l Jo"lf 82 Pr(O) = O.

(2)
{"If {"If
in
o
02Pr(O)d8 = in
0
02dO + 2
n=1
L rn in("If 02cos nO d8
00

= "lf3 + 22::=1 rn
3
[~ sin nO + ~Ocos nO - ~ sin n8]: .

The first equality is justified by the fact that if SN(O) = 82 + E:=l r n 82

cos nO, then ISN(O)I ~ 02(1 + 2rI2;) for fixed r in (0,1). Hence sN(8) ~
02 ~. Thus the dominated convergence theorem can be applied to the
sequence {SN }~=1' Furthermore, since I( _r)nl ~ 1, applying dominated
convergence again shows that

lim
r-+l
l0
"1f
82 Pr(8)d8 = -3 + 47r L -2-'
7r3

n=1
00 (_I)n
n
 1. Measurable Spaces and Integrable Functions 9

(3) From (1) and (2) it follows that "3 + 47rL:n=1 7

3 00 !_1)n
= 0 and hence
that L:::1 <-sn
= - ~;. Let I = L:%"=1 (2k~1)2 and let P = L:%"=1 ~.
It is easy to see that I + P = 4P and, by the result above, P - I = - ~2 .
2

Hence I = 8' ,,2 P ,,2 d""'oo 1 ,,2

= 24' an L.m=1 ~ ="6.
(4) ~ 10g(1-x) = L:~=o :;1
for 0 < X < 1. Since this is a convergent series
with positive terms, the double series L:~=o L::=o (n+x1)(':+1) is convergent
with positive terms and has sum (10g(1 - x))2~. Thus, by monotone con-
vergence,
d
10r (10g (1-x)?~
1 00 00

_"" 1
x ~~o(n+m+1)(n+1)(m+1)"
2 -

If n of 0, we can use partial fractions to re-write the summands as

1 1 [1 1]
(n+m+1)(n+1)(m+1) = (n+1)n m+1 - n+m+1 '
and hence

The n = 0 term in the double series also gives L::=o (m!1)2 . Thus

1
o
1 dx
(10g(1 - x))22"
X
= 2 L 2" = -.
00

n=1 n
1 7r 2
3

00 (_1\n+l 1 d
Problem 1-8. EvaluateL:n=l ~ by using the integral fo H~x and the
monotone convergence theorem.
10 I. Measurable Spaces and Integrable Functions

SOLUTION.
1 00 00

- - = ~(_x)n = ~x2n(1_ x) for 0 < x < l.

1+x n=O n=O

Since the second series has positive terms, the monotone convergence the-
orem can be applied to give

r1 dx 00 1 1 00 (_1)n+1
log 2 = Jo 1+x=~(2n+1-2n+2)=~ n .

REMARK. Even an example as simple as this shows how much easier it is

to work with the Lebesgue integral than the Riemann integral. Justifying
the preceding calculation is much more delicate when 101la.;X is viewed as
a Riemann integral.

Problem 1-9. Let (X, A, J-t) be a measure space and let x f-t f(x) =
(h(x), h(x), . .. ,fn(x» be a measurable mapping from X to Rn. Suppose
that R n is equipped with a norm II I such that x f-t Ilf(x)1I is integrable.
(1) Show that Ii is integrable for every j = 1,2, ... , n.
(2) Defining Ix f(x)dJ-t(x) in R n by

(Ix h(x)J-t(dx), ... , Ix h(X)J-t(dX») ,

show that I Ix f(x)J-t(dx) II :S Ix Ilf(x)IIJ-t(dx).

METHOD. On the dual space (Rn)* consisting of linear functionals a : v f-t

(a,v) on R n , introduce the dual norm Ilall* = suPv¥o I(O~")I and use the
fact that Ilvll = sUPa¥O lil;111 1.

SOLUTION.
(1) If v = (Vl,V2,'" ,vn ), the mapping v f-t Vj from R n to R is linear
and hence continuous, since it is a map between finite-dimensional spaces.
Hence there exists a number k > 0 such that IVjl :S kilvil for every v in Rn.
Thus
Ix Ifj(x)IJ-t(dx) :S k Ix Ilf(x)IIJ-t(dx).

(2) If a E (Rn)*, we can write l(a,J(x»)1 :S Ilf(x)llllall*, and hence

(a, Ix f(x)J-t(dx») = Ix (a, f(x»J-t(dx) :S Ilall* Ix IIf(x)IIJ-t(dx).

 1. Measurable Spaces and Integrable Functions 11

Taking the sup over a and using the fact mentioned above, we find that

II Ix f(x)JL(dx) II :s; Ix Ilf(x)IIJL(dx).

REMARKS.
1. The shortest path between two points is a straight line. Consider R n
1
with the Euclidean norm Ilvll = [vf + v~ + ... v~l'. Let X = [0,1] with
Lebesgue measure. (See Chapter II.) Let F be a function from [0,1] to
R n such that the derivative f = F' exists everywhere and is continuous.
Then fol Ilf(x)lIdx can be interpreted as the Euclidean length of the curve
described by F, and II fo1 f(x)dxll = IIF(l) - F(O)II is the length of the line
segment with endpoints F(O) and F(l).
2. Case of equality. It can be shown that, when the unit ball B is strictly
convex (that is, when IIv111 = IIV211 = IIAv1 + (1 - A)v211 = 1 for A E [0,1]
holds only for A = 0 or 1), the inequality is strict unless there exist vERn
and a function g(x) ~ 0 such that f(x) = g(x)v JL-almost everywhere. The
application to the Euclidean length of a curve is immediate.

Problem 1-10. LetX,X1 , ••• , X n , ... be measurable functions from a space

(E, E, JL) to an open set n of Euclidean space R d such that

Vf > 0 JL({IIXn - XII ~ f}) ---> 0 as n ---> 00.

(1) Show that Vf > 0 there exists a compact set Ken such that JL({X ~
K}) :s; f and, for every n, JL( {Xn ~ K}) :s; €.
(2) If f : n ---> R m is continuous, then Vf > 0

JL({Ilf(Xn ) - f(X) II ~ f}) ---> 0 as n ---> 00.

SOLUTION.
(1) For any fixed f > 0, there exists a compact subset C 1 of n such that
JL({X ¢. Cd):S; ~. Let a = inf{llx-yll : x E nand y E C1 } and let 0 < b <
a be fixed. Then there exists an integer N such that JL( {II Xn - X II ~ b}) :s; ~
if n ~ N. Let C 2 be the compact subset of n defined by

C2 = {x En: inf{llx - yll : y E Cd :s; b}.

Then, if n ~ N,

JL({Xn ~ C 2 }) JL({X n ~ C 2 } n {IIXn - XII :s; b})

+JL({Xn ~ C 2 } n {IIXn - XII ~ b})
:s; JL({X ~ Cd) + JL({IIXn - XII ~ b}) :s; f.
12 I. Measurable Spaces and Integrable Functions

Moreover, if n < N, there exists a compact subset Kn of n such that

1L({Xn ¢. Kn}) ~ f. Set K = C2 uKo u ... UKN-l; the compact set K has
the desired property.
(2) We may assume without loss of generality that m = 1 since, if !j(x) is
the jth component of !,
m

1L({I1!(Xn ) - !(X) II ~ f}) ~ r>({IIfJ(Xn) - !(X) II ~ f}).

j=1

Now let 8> O. By (1), there exists a compact set Kin n such that IL( {Xn ¢.
K}) ~ 8 for every nand IL( {X ¢. K}) ~ 8. The function ! restricted to
K is uniformly continuous; that is, for every f > 0 there exists 17 > 0 such
that I!(x) - !(y)1 ~ f if IIx - yll ~ 17 with x and y in K. Hence, for every
n,
1L({I!(Xn) - !(X)I ~ f})
~ 1L({Xn E K, X E K, If(Xn) - !(X)I ~ E})
+IL( {Xn f- K}) + IL( {X f- K})
~ 1L({IIXn - XII ~ 17}) + 28,
and therefore
lim sup 1L({I!(Xn ) - !(X)I ~ f}) ~ 28.

Since 8 is arbitrary, limn_co 1L({I!(Xn ) - !(X) I ~ f}) = O.

Problem 1-11. Let (X, A, IL) and (Y, B, v) be measure spaces such that
IL(X) and v(Y) > O. Let a : X --+ C and b : Y --+ C be /unctions, respec-
tively A and B measurable, such that
a(x) = b(y) IL ® v-almost everywhere on X x Y.
Show that there exists a constant oX such that a(x) = oX IL-a.e. and b(y) = oX
v-a.e.

SOLUTION. If Y E Y, let All = {x : a(x) i- b(y)} and let C = UIIEyAIi .

Clearly All E A and C E A®B (since (x,y) f-+ a(x) -b(y) is A®B measur-
able). Then b(ytl = b(Y2) for any Yl, Y2 ¢. B. For otherwise we would have
a(x) = b(Yl) on A~, and a(x) = b(Y2) on A~2' with A~, and A~2 disjoint
and AliI U AII2 = X. But this is impossible since IL(A II1 ) = IL(A II2 ) = 0 and