Professional Documents
Culture Documents
Principles of Real Analysis-Solutions-Aliprantis PDF
Principles of Real Analysis-Solutions-Aliprantis PDF
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 -- - -
ACADEMIC PRESS
a division of Harcourt Brace & Company
525 B Street, Suite 1900, San Diego, CA 92101-4495, USA
http://www.apnet.com
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
Foreword vii
v
vi CONTENTS
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.
vii
I
I
I
I
I
I
CHAPTER 1 ______________
FUNDA:MENTALSOF
REAL ANALYSIS
1. ELEMENTARY SETTIIEORY
~ [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).
(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
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
(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
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
and
Thus,
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.
For the converse, note that the relation l(f-\({b))) = {b} implies I-\({b}) =1=
o for each bEY so that I is onto.
(g 0 f)-leA) = I-\g-\(A)).
Solution. Note that
Problem 1.9. Show that the composition ollunctions satisfies the associative
law. That is, show that if X ~ Y ~ Z ~ V, then (h 0 g) ~ I = h 0 (g 0 f).
Therefore, (h 0 g) 0 I = h 0 (g 0 f).
Solution. We must show that the relation R is reflexive, symmetric, and trans i-
tive.
Reflexivity: Note that I(x) = I(x) implies xRx for each x E X.
Symmetry: Let X\RX2. Then, I(x\) = 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.
(b) Clearly,
If X and Y are two nonempty disjoint sets, then X U Y ¢. P(X) U P(y), and so
equality is seldom valid.
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.
X, if x ¢. C;
f(x) =( cn+I, if x = C2n+l (n = 0, 1,2, ... );
bn , if x = C2n (n = 1,2, ... ) .
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.
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.
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).
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.
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
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.
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
Problem 2.14. Show that the Well-Ordering Principle implies the Principle of
Mathematical Induction.
Problem 2.15. Show that the Principle of Mathematical Induction implies the
Well-Ordering Principle.
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
Problem 3.3. Show that the real numbers J2 and .,fi + .J3 are irrational num-
bers.
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,
(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
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.
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
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,
(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
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.
16 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
f(~) = ;;
and extend it to all of]R. via the formula
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
=
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 < sup A, or a < sup A + E,
holds for all 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,
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.
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-
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.
20 Chapter 1: FUNDAMENTAlS OF REAL ANALYSIS
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.
Problem 4.4. Find the lim sup and lim inf for the sequence {(-1)").
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
(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
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,
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).
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
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
24 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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
Problem 4.9. Show that lim..::lii = 1 (and conclude from this that lim:fO. = 1
for each a > 0).
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
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.
Section 4: SEQUENCES OF REAL NUMBERS 25
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
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
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}?
26 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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
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
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
Problem 4.12. For a sequence of real numbers {xn} establish the following:
a. If Xn+l - Xn -+ x in JR, then xnln -+ x.
Section 4: SEQUENCES OF REAL NUMBERS 27
* 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.
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 )
Xn+1 1 - .JI=X;;" 1 1
-- = = ---=== ~ -
Xn XII 1 + -v'1-xlI 2'
and the solution is complete.
is a convergent sequence.
n n n
Xn = 1 + L fi(1-~) ... (1- ;~I) ~ 2+ L h~ 2+ L ~ ~ 3.
;=1 ;=2 ;=2
for n = 2,3, ... and some fixed 0 < a < 1. Show that {xn} is a convergent
sequence.
Section 4: SEQUENCES OF REAL NUMBERS 29
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.
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
for each n = 2,3, .... By Problem 4.15, the sequence {xn} converges. Let
30 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
x = 11lim . ( 1) 1
Xn+1 = hm 1 + - - = 1 + - - .
....
00 n .... oo 1 + Xn 1+x
x2 I (x - -
-2=- 2 )2 >0
n+1 4 n x" -
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.)
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
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
and
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)
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
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, ... }.
Section 4: SEQUENCES OF REAL NUMBERS 33
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,
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
34 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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.)
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]
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
(n-k)-terms
°
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
36 Chapter 1: FUNDAMENTAlS OF REAL ANALYSIS
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
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 ....
(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
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
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
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
40 Chapter 1: FUNDAMENTAlS OF REAL ANALYSIS
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.
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
(k+1
Jk f(x)dx 2: f(k + 1), and (1)
(k+1
Jk f(x) dx :s f(k) (2)
n n-I
all = f(l) + L f(k) = f(1) +L f(k + 1)
k=2 k=1
= f(1) + f n •
This implies all "i1l:S f(1) for each n. On the other hand, using (2), we see
that
=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
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 < p :::: 1 and converges in lRfor all p > 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
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.
(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
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,
(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
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.)
Il n It
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.
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
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
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?
(aA)O = (AnAcr=(AnAcr=Aon(Acr
S;;; AD n AC = AD n (Ao)" = 0.
Solution. Let I denote the set of all irrational numbers, and let {rJ, r2, ... } be
an enumeration of the rational numbers of JR.
48 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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.
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
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.
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.,
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
00
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.
f; X --+ JR defined by
= {~:
if x E A;
f(x) if ~ ¢. A,
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
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.
52 Chapter 1: FUNDAMENTAlS OF REAL ANALYSIS
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.
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.
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
Solution. Notice that if A is a nowhere dense set in a metric space X, then from
Lemma 6.8 we see that
54 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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
Problem 7.3. Show that if(X, d) is a separable metric space (see the preceding
exercise for the definition), then card X ~ c.
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
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).
(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,
~ 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)
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
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
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),
Problem 7.10. Let(X, d) be a metric space. Define the distance oftwo nonempty
subsets A and B of X by
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
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
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
Thus, 0 ::::: f(Zk) ::::: Cnd(Zb xo) holds for each k. In view of
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.
62 Chapter 1: FUNDAMENTALS OF REAL ANALYSIS
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.
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?
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
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.
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 ______________
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 V ~ A \ AO ~ A.
".
65
66 Chapter 2: TOPOLOGY AND CONTINUITY
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.
v n (A n 0) = (V n q) n A =1= 0,
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).
A
C
=X \ A = (U Oi) \ A = U(Oi \ A) =U Vi
iel iel iel
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.
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
If(x) - f(y)1 ::: If(x) - f(a)1 + If(a) - f(y)1 < E + E ::::: 2E,
and thus
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) = 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).
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.
(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
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
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}.
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 9.1. If u, v, and ware vectors in a vector lattice, then establish the
following identities:
a. uVv+U!\v=u+v;
74 Chapter 2: TOPOLOGY AND CONTINUITY
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
u +v + lu - vi = (u v v + u /\ v) + (u V v - U /\ v) = 2(u v v).
(g) As in (f), we get
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·
Solution. (a) Note that
Solution. If lui" Ivl = 0, then using parts (a) and (b) and part (e) of Problem 9.1,
we get
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
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.
Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 77
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
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
= °
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
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
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:
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
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),
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
"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
holds for all x E V, which shows that f is continuous at Xo. For the equality, note
that
while
Solution. Clearly,
Since f is not continuous, the convergence cannot be uniform; see Theorem 9.2.
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.
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:
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)
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
that if {fn} converges pointwise to a continuous lunction I on [a, b], then {f,,}
converges uniformly to I on [a, b].
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.
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
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 =
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].
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
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,
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
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
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.
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.
Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 87
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.
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
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.
I/(x) - l(y)1 ~ I/(x) - h(x)1 + Ih(x) - Ik(y)1 + I/k(y) - l(y)1 < 38.
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
By the compactness of X, there exist XI, ... ,Xk E X such that X = U~=I V Xi '
Section 9: CONTINUOUS REAL-VALUED FUNCTIONS 89
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
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
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)
D(f, fl) = IIglioo = e. On the other hand, if x [0,1) is fixed, then for all
sufficiently small h > we have ° E
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
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
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
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
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
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
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
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.
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
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
Problem 10.4. Show that a closed set of a normal space is itself 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.
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.
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 10.10. Show that a nonempty connected normal space is either a sin-
gleton or uncountable.
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
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.
98 Chapter 2: TOPOLOGY AND CONTINUITY
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.
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
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 ,
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].
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].
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
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]?
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
1 00
1 00
f(x)e-
nX
dx = ° forall n = k, k + 1, k + 2, ... ,
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
Jolx"g(x)dx
that g
°
= holds for each n = 0, 1,2, ....
= 0, and consequently, f = 0.
By Problem 11.4, it follows
°
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.
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.
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).
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
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
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
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!" 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 12.6. 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}.
S = {E S; X: E or E C is at-most countable}.
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
110 Chapter 3: TIlE TIlEORY OF MEASURE
{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.
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
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
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}.
A = {A n Y: A E B}.
c = {A E B: AnY E Br}.
112 Chapter 3: THE THEORY OF MEASURE
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.
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.
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.
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.
00
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
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.
k k m
00 00
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
°= 00
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
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.
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
116 Chapter 3: TIlE TIlEORY OF MEASURE
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
Problem 14.1. Show that a countable Ullioll of Ilull sets is agaill a Ilull set.
Problem 14.2. If IL is all outer measure all a set X alld A is a null set, then show
that
!
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
Consequently, we have
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( An [U
n=1
En]) = /-L(U A n En) ::: L /-L(A n En).
n=1 n=1
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
Problem 14.8. Let /-L be an outer measure on a set X. Show that a subset E of
2: J-L(A n F) + J-L(A n E C
)
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.
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).
120 Chapter 3: THE THEORY OF MEASURE
(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) =
00 00
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
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.)
Section 14: OUTER MEASURES AND MEASURABLE SETS 121
oc
I>(A~) < fl.*(En) + rne = v(E Il ) + rile.
k=1
:GOO 00 00
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.
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
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.)
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
00 00
fL*(AnE)=fL*(U(AnnE)) = LfL*(AnE),
n=1 11=1
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
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
b. I = (a, b). Pick a < XII < b with Xn -I- a and observe that[xlI , b) t (a, b).
124 Chapter 3: TIlE TIlEORY OF MEASURE
d. 1= (-00, a]. Note that (-00, a) S;;; (-00, a] and so from the inequality
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 JR and since A. * ([an, bll ») = A. * (all, bll »), it follows that A. *(aA) = lalA. *(A)
for each a E JR. Now, the identities
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
Show by a counterexample that it is possible to have p, f:. p,* on 5. Why does this
not contradict Theorem 15.1?
Av = {F S; E: FE A/1-}'
126 Chapter 3: THE THEORY OF MEASURE
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
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
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
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
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.
128 Chapter 3: TIlE TIlEORY OF MEASURE
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.
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
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).
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.
130 Chapter 3: THE THEORY OF MEASURE
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.
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.
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
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
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. a. (i) The atoms of the counting measure on a set are precisely the
one-point sets.
132 Chapter 3: THE THEORY OF MEASURE
(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?
f..L*(B) =
. °
1001
if B = <;b;
if B =1= <;b and B ~ A;
if BS?;A.
(e) If a is any non-negative extended real number, then the set function
v: A -7 [0,00], defined by
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.
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
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.
134 Chapter 3: TIlE TIlEORY OF MEASURE
X if x E E;
f(x) =:= { _~, if x E [0, 1] \ E.
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. (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) }.
that g-I«()) is an open set, and the conclusion follows from the identity
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.
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)}.
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)·
138 Chapter 3: THE THEORY OF MEASURE
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.
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.
Section 17: SIMPLE AND STEP FUNCTIONS 139
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
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.
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
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
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.
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.
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.
Section 17: SIMPLE AND STEP FUNCTIONS 141
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
co kl/
f(x) = LL'>i'XEj',
11=1 i=1
142 Chapter 3: TIlE TIlEORY OF MEASURE
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
A = {x EX: ¢(x) 2: 4}
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
144 Chapter 3: THE THEORY OF MEASURE
(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
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
and
00 00
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.
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
= 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
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. 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
00 00 00
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
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:
y y y
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
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.
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.
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.
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
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,
Problem 18.17. Let f: lRn -7 lR be a continuous function. Show that the graph
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
Solution. (a) Since every a-set is a Borel set, Problem 15.2 shows that
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,
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
156 Chapter 3: TIlE TIlEORY OF MEASURE
f(t) = A(A n [-k, t]) ~ A(A n [-k, s]) + A(A n (s, t]) ~ f(s) +t - s.
implies
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
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.
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
Solution. The conclusion follows immediately from Theorems 19.4 and 19.5.
{x: Ifn(x) - fm(x)1 2: 28} ~ {x: lfn(x) - f(x)1 2: 8} U {x: Ifm(x) - fex)1 2: 8}
00
Fn = U Ek = {x: k
Igk+l(X) - gk(x)1 :.::: 2- holds for some k :.::: n }.
k=1l
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
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
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
is an algebra of sets.
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.
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.
'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.
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).
164 Chapter 3: TIlE TIlEORY OF MEASURE
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}.
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
Section 20: ABSTRACf MEASURABILITY 165
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 20.10. 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.
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
SE = {E n A: A E S} = {B E 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
00 00
= /.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.
Section 20: ABSTRACf MEASURABILITY 167
f-I(C) = n00
n=1
f-ICVn)'
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.
168 Chapter 3: THE THEORY OF MEASURE
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.
is jointly 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.
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.
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
11 n 00
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
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.
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
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,
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. }.
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
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
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).
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).
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. LetFII = U~=I Ei. Clearly, IfXFn I:::: If I foreachn andfXFn --+ fXE
176 Chapter 4: TIlE LEBESGUE INTEGRAL
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
{x E X: I(x) =1= O}
lor t 2: 0 is continuous at t = O.
Section 22: INTEGRABLE FUNCTIONS 177
Then, we have
• 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.
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.
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.
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.
+
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.
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),
180 Chapter 4: TIlE LEBESGUE INTEGRAL
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
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.
then show that fg is integrable with respect to the measure space (X, S, fJ-),
and that
J =1gdv gf dfJ-.
°° °
(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
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
lal1g d)'"
J
= lal n-H;X)
lim 1Vrn d)... = lim j¢1I d)... = j f d)"'.
J n .... oo I 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.
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
Section 22: INTEGRABLE FUNCTIONS 183
2->
l+x -
fI
If - fll dA. = 2 [ U -
JEI
it) dA..
184 Chapter 4: THE LEBESGUE INTEGRAL
lEI
r(J - II) d'A + ! I\E,
(J - II) d'A = !(J -
I
II) d'A
= 11 d'A -1 I d'A = O.
Consequently,
= r (J -
lEI
II) d'A + !I\E,
(JI - I) d'A
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.
Section 22: INTEGRABLE FUNCTIONS 185
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
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
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 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
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.
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").
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 < b ::s 1 and let s > 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
~
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,
= (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.
Section 22: INTEGRABLE FUNCTIONS 189
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.
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
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.
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
l a
f(x) dx = lim
II~OO
b-aI::n (
--
n 1=1
.
f a + i(b-a))
n
.
192 Chapter 4: THE LEBESGUE INTEGRAL
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).
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
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
=
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.)
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
194 Chapter 4: THE LEBESGUE INTEGRAL
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].
lim
n--).oo
lb
a
fn(x)dx = lb
a
f(x)dx.
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.
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
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
196 Chapter 4: THE LEBFSGUE INTEGRAL
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)
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
as desired.
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
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,
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
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
=1 1 1
[/(t)
1g
(X)dX)]dt
= (1 (1 1 1
I(t)dt) . g(x)dx).
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
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.
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).
(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, ....
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
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.
202 Chapter 4: THE LEBESGUE INTEGRAL
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
Consequently, we have
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
= S*(f, P x Q).
Section 23: THE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 203
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
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
b fg(X)
=
l
a
(
J(x)
h(x,y)dy)dx.
1 [a,b)
I'd). = I(b) - I(a).
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
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.
Section 23: TIIE RIEMANN INTEGRAL AS A LEBESGUE INTEGRAL 205
[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) = ~ 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
206 Chapter 4: TIlE LEBESGUE INTEGRAL
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)
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).
Problem 24.3. Show that f(x) = I~it is Lebesgue integrable over [1,00) and
J
that f d)'" = 1.
Therefore,
. 1/!( + -X)II
hm
n..... oo 0 n
e- 1 2x
dx = 1.
= = 1.
Jo[00 e- lim (1 - e- r )
x dx = lim r e-x dx
r ..... oo Jo r->OO
lim - -
1
1
x..... oo j(x) x
00
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
Consequently, we have
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
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
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
[
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.
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
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
1
ox.
----dx=lnt
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) = +
holds for each t > O.
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 '
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
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
2C =2F(1) = 2faTln(COSX)dX
= faT In(sin22.t) dx
= faTln(Sin2X)dX - Iln2
= ~ Lrln(SinX)dX - Iln2
=C- I ln2
Thus, C = -I In 2, and so
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 '
214 Chapter 4: TIlE LEBESGUE INTEGRAL
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
F'(t) =
1o
00
-ata [ x(1Sinxt]
+x 2 )
dx = 1 0
00
cosxt
- -dx
1 +x 2
00 •
F " (t) = -
1 x SlDXt d x.
o l+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)
-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
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,
.,
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.
r'(t) = foooxl-le-Xlnxdx
holds.
e. Show that r has derivatives of all order and that
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
and so there exists some M > 0 such that xIJe-~(1nx)n .:5 M for all x > 1.
Therefore,
Both can be proven by induction. For this limit use induction and L'Hopital's rule
by observing that
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
. _ {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
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
and
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
:: 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
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.
220 Chapter 4: TIlE LEBESGUE INTEGRAL
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 < b < 00. In this case, for each
t > 0 we have
I
I
I I
xme:;:T.:I
(xz_l)k(x_1) _<
I
e-:m=x;
(I_X)'+1
I= tHI e-1 , 1
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 < b - e, then Supp gx = [x - e, x + e] and so
h(x) =
l
a
b
gAt)dt = lx+e pC~X)dt = e
x-e
11
-1
p(u)du =c > O.
holds for all x. Thus, f = hi c satisfies 0 ::: f (x) ::: 1 for all x.
Finally, observe that
y=f(x)
FIGURE 4.3.
:::: 8 + t
i=1
ICil! IX[a;,b;) - gil d'A
n
< 8 + L IC;/n'~;1
;=1
= 8 +8 =28.
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
(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.
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
224 Chapter 4: TIlE LEBESGUE INTEGRAL
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.
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,
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
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
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,*(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
226 Chapter 4: THE LEBESGUE INTEGRAL
00 00
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.
and
00 00
0000 0000
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
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
and
and
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
CASE 3. The general case. Here we shall use the hypothesis that fL is CY-finite.
(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
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
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
Section 26: PRODUcr MEASURES AND ITERATED INTEGRALS 231
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
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
= J[00 [_ ysinx+;osx
f e
r sinx
x
dx
o 1+y-
e- xy \x=r] dy
x=e
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
232 Chapter 4: TIlE LEBESGUE INTEGRAL
°
Now, if < e < 8, then (by taking into account that 0 :::; ~~;~ :::; 1 for y :::: 0)
we infer that
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
= l°00dY = o.
Section 26: PRODUCT MEASURES AND ITERATED INTEGRALS 233
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
Problem 26.11. Show that if f(x, y) = ye- o+x2 )y2 for each x and y, then
~
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
Solution. Since
+ 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
we get
[00£
10 l+y
_ [00 y2 sinr+cosr e- ry2 dy
10 l+y4
_../ii
- 2
r
10 .fi
sinx dx
.
1o
00 .!!L _
I+y· -
11-
0
.!!L
Hy· +
lex>.!!L
1+
_
l+y4 -
rr./2
4 .
[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
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
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?
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.
Section 26: PRODUcr MEASURES AND ITERATED INTEGRALS 237
. g(x,y) =0
g(x,y) = 0
FlGURE 4.5.
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.
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}.
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
Problem 27.5. Assume that two vectors x and Y in a normed space satisfy
IIx + yll = IIxll + lIylI. Then show that
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
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.
is countable.
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.
AX + (1 - A)Y = ax + by E (a + b)A = A.
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
and so for proving PA(ax) = lalpA(x), we can suppose without loss of generality
that a > O. Now, note that
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
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
246 Chapter 5: NORMED SPACES AND Lp.SPACES
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
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.
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
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].
248 Chapter 5: NORMED SPACES AND Lp-SPACES
FIGURE 5.1.
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?
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.
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
and so II Til::: 1. On the other hand, for the constant function 1, we have
(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
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/·
Section 29: LINEAR FUNCTIONALS 253
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
It is easy to verify that V is a vector subspace of JR.". Now, define the linear
functional g: V -+ JR. via the formula
=
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.
lim TIl(e)
n-+oo
= n-+co
lim Yn = TCe).
254 Chapter 5: NORMED SPACES AND Lp-SPACES
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
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
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
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
Section 29: LINEAR FUNCTIONALS 255
by
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.
256 Chapter 5: NORMED SPACES AND Lp-SPACES
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
L(x) = nlim
...... oo
Xn = p(x).
Xl - X3 Xl - X4 Xl - Xn+l )
A(x - y) = ( Xl - X2, - - ,
2 3
, ... ,
n
, ....
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.
(Tx, f) = (x, T* f)
258 Chapter 5: NORMED SPACES AND Lp·SPACES
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,
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
we see that S is a bounded operator and that IISII :s IIgll . lIuli. Now, if x E X
Section 30: BANACH LATTICES 259
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.
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+.
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
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.
Section 30: BANACH LATTICES 261
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] •
262 Chapter 5: NORMED SPACES AND Lp-SPACES
Problem 30.5. Show that in a normed vector lattice X its positive cone X+ is a
closed set.
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.
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.
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.
·
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.
264 Chapter 5: NORMED SPACES AND Lp-SPACES
(1, -1, 1, 1, -1, -1,1,1,1,1,1, 1, -1, -1, -1, -1, -1, -1, ... ).
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.
Problem 30.12. Determine the norm completion of the normed vector lattice of
the preceding problem.
Solution. Let
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).
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).
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).
266 Chapter 5: NORMED SPACES AND Lp-SPACES
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.
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.
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.
T(P(rp)) = hP(rp)
268 Chapter 5: NORMED SPACES AND Lp-SPACES
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].
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
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
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
Section 30: BANACH LATTICES 269
II
II
L G)~ t k (1 - t)n-k = t.
k=O
II
II
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
Now, note that these identities readily imply that lim II Til I
I = 1, x, and x2.
- !lloo = °
holds for
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
-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
Consequently,
and so
Section 31: Lp-SPACES 271
31. Lp·SPACES
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,
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.
Problem 31.3. Let (X, S, p,) be a measure space and consider the set
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.
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 - -.
Section 31: Lp-SPACES 273
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
,
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 < p < q < 00. Let f E Lq(f.L). From
Theorem 31.14, we know that Lq(f.L) <;; Lp(f.L), and so f E Lp(f.L).
Put r = 'l..p > 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
(b) Let {Pn} be a sequence of positive real numbers satisfying Pn > 1 for each
n and lim PII = 00. From the inequality
it follows that
Solution. Assume I E, L2[0, 1] satisfies the stated properties and let 0 < f3 < a.
Then, note 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·
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.
0 if Ix~ I 2: 1
Y/I = {2 if Ix~ I < 1 .
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.
276 Chapter 5: NORMED SPACES AND Lp.SPACES
°
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.
lim
,,-;00 1" a
g(x)fll(x)dx = 1"
a
g(x)f(x)d"A(x).
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
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.
278 Chapter 5: NORMED SPACES AND Lp-SPACES
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
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
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
That is,
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).
280 Chapter 5: NORMED SPACES AND Lp-SPACES
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
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.
Consequently,
n n
Fg(f) = Fg(L I(Xi)xEi) = L I(Xi)Fg(XEi)
i=1 i=1
n fl
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 < p < 1 and consider the space Lp([O, 1]). Take I =
X(o.~) and g = X(!.I)' and note that
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
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.
In the other direction, note that if for each n we consider the open set
Problem 31.21. If {fn} is a norm bounded sequence of L 2(Jk), then show that
fnln -+ Oa.e.
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
284 Chapter 5: NORMED SPACES AND Lp.SPACES
see that
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
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.
2
(x) = {If(X)1 if x E A
g If(x)1 if x ¢. A ,
2
If(x)I P < {1/(X)1 if x E A= g(x),
- - If(x)1 if x ¢. A
(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
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
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
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.).
Show that for each 0 < E < 1 there exists some 8p > 0 sLich 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
"
Section 31: Lp-SPACES 289
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.
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·
290 Chapter 5: NORMED SPACES AND Lp-SPACES
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
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
! 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
In addition, we have
r
Jx
h(t,x)dfL(X) = r
JE,
pt p- 1 df.L(x) = ptP-1f.L*(Er).
:: 1 00
as desired.
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 < P < 00, e > 0, h > 0,
and a positive integer n are given, then there exists a COO-function ¢: lR --7 1R
such that
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 < p < 00, we see that limk-+oo (kt2)P . k P-1 = 0, and so a
sufficiently large k will yield a function ¢ with the desired properties.
To complete the proof, let I E LP(lR) and let 8 > 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:
Problem 31.34. Let (0, (0) be equipped with the Lebesgue measure, and let
1 < P < 00. For each IE Lp().) let
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
it follows that
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 .
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
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
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
Consequently, we have
and so
This implies
. 1
I TII > -
~.
p_I+~
[1 + _1_]-;;
lI(p-l)
--+....L
p-I'
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
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
is an inner product on Xc' Also, show that the norm induced by the inner product
(', .) on Xc is given by
297
298 Chapter 6: HD...BERT SPACES
1. (Additivity)
(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).
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.
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
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. Observe that (XII' x) -+ IIxI\2 implies (x, XII) = (XII' x.) -+ IIxll2 =
IIx1I2. So, from
IIx n _x1I 2 = (XII -X,XII -x)
Problem 32.9. Show that the norms of the following Banach spaces cannot be
induced by inner products.
302 Chapter 6: HILBERT SPACES
Solution. (a) Consider the vectors x = (1, 0, ... , 0) and y = (0, 1, 0, ... , 0).
Clearly,
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
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
This shows that the norm II . lip does not satisfy the Parallelogram Law and so it
is not induced by an inner product.
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
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
304 Chapter 6: HILBERT SPACES
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
1
(IX, y) = 4(1I1x + yll2 - IIlx - yll2 + IlIlx + lyll2 - IlIlx _ lyll2)
Solution. Assume that (T x, x) = 0 holds for all x EX. From the identity
(Tx, y) + (Ty, x) = 0
-(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
00
Solution. (a) If x, YEA l. and a, {3 are arbitrary scalars, then for each Z E A we
have
(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
= l b
[p(X)f'(X)],g(X)dX + 1b q(x)f(x)g(x)dx
-1b
a a .
= 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:
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
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
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
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.
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
<
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.
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
This implies ~ Ian 12 < (1 - EC 2)la n 12 ::: E2(b - a) for alln ::: m, or
for alln ::: m. Since 0 < E < 2~2 is arbitrary, we have established that an -+ O.
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).
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,
314 Chapter 6: HILBERT SPACES
it follows that lim(xll , Yn) = lim(x~, y~). In other words, the formula
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. 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..
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.
316 Chapter 6: HILBERT SPACES
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
Vf(x) = l x
f(t)dt.
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.
::: (b - a)211f112.
lim (XII' y)
11""""00
=0
holds for each vector y E £2.
Section 33: HILBERT SPACES 317
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
< 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
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
318 Chapter 6: HILBERT SPACES
II "L,7:1 Xkj r = "L,;'=I (Xkl' xd + "L,7=1 (Xk2~:k) + ... + "L,7=1 (Xkn , Xk)
This implies
as required.
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).
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
320 Chapter 6: HILBERT SPACES
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
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
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)
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.
322 Chapter 6: HILBERT SPACES
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
00
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
Section 33: HILBERT SPACES 323
I: ai(x, ¢i)<!>i f = I:
00 ? 00
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.
for all x, Y E H. Show that T and T* are both bounded linear operators satisfying
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
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
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
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.
326 Chapter 6: HILBERT SPACES
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
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
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.
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
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
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
n=1 n=1
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
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.
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
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.
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
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
~ 2
L If -snI 2d J.L. L 1 dJ.L
~ IIf - snll 2 J.L*(E) --+ o.
L
j=l+1
lIekj - fk j UZ = 0 < 1,
00
00
00
IIxII2 = L
iei
ICx, ei)1 2
= L
j=l+1
I(x, ekj - fk)1 2
00
{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.
elX _ e- 1X
sinx=----
21
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
334 Chapter 6: HILBERT SPACES
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
1 /I
Integrating gives
1
-
j1f D/I(t)dt = 1 + -1 L j1f cosktdt = 1. /1
1r -1f 1r k= 1 -1f
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
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
1 121f
IIS"II ::: - ID,tCx - t)1 dt.
rr 0
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.
10r n: g(x)cosnxdx
2
(g,cosnx) =
= 1n:
10r f(x)cosnxdx+
On:.
2
f(2n-x)cosnxdx.
l~-, (;r)
(;r) 2*·
2~- cos x, (;r) 2~- cos 3x,
- cos 2x, (;r) ~ ...
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].
{(;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.
{( i?)~-smx,
. (?)!; .
i -sm2x, (2)!; .
;r -sm3x, (?)!; .
i -sm4x, '" }
f (x ) if 0:::: x :::: rr
g(x) = { -f(2rr - x) l'f rr<x_ < 2 rr.
338 Chapter 6: HILBERT SPACES
2rr
(g,cosnx) = Jor g(x)cosnxdx
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.
Section 35: FOURIER ANALYSIS 339
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.
I ifO::::x<I
fex) ={0 if I:::: x < 2n.
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
= ~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
00 sinnx
x =:rc - 2 2 : - - .
n=1 n
Solution. We consider the periodic function /: [0, 2:rc] -+ R defined by
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.
Section 35: FOURIER ANALYSIS 341
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 •
2rr
ao -
-1rr
1o
-, - 3rr 12rr -_8.".2
-,v2dv_x3
0
3" ,
342 Chapter 6: IDLBERT SPACES
+4 L (COS
OO
2
x =-x
4 2
-n- nx- x sin nx) ,
3 2 n
n=1
is an orthonormal basis for L2[0, x]. Also, as usual, the norm of the operator is
given by
/ = Lc (:)2sinnx
ll
n=1
Section 35: FOURIER ANALYSIS 343
in its Fourier expansion relative to the above basis. By Parseval's Identity, we have
00
00 00
SPECIAL lOPICS IN
INTEGRAnON
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·
Problem 36.3. If JL is a signed measure, then show that for each A E I: we have
00 00
345
346 Chapter 7: SPECIAL IDPICS IN INTEGRATION
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
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=!'
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
Solution. The proof parallels the one of Theorem 36.1. We shall verify that if
f.L, v E M(2:), then the formula
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: 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
00 00
= L[f.L(Bn) + v(An \ B Il )]
n=1
00 00
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
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 A z) v (y A z) = u 2: x v Y + z - (x v y) v z = (x V y) A z.
and
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 n E) + 1tJ,I(A n E
C
),
,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
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
there exists some B E ~ with. B S;;; E and /LI (E \ B) + /L2(B) < e. Thus,
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. 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
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.
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 37.3. 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
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
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
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.
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.
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
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
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
AnY = B n Y \ (B \ A) n Y E AJl."
(b) It follows immediately from Problem 22.15.
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
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.
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
/~ /·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).
1/ dv = Ix / 0 T d/L.
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.
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:
362 Chapter 7: SPECIAL TOPICS IN INTEGRATION
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,
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
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
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).
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
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
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 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)
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.
[ fd)'" =
JB
1aB
fd)....
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)
f(a) = f~l)
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,
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.
"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,
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,
(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).
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,
368 Chapter 7: SPECIAL TOPICS IN INTEGRATION
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.
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
)
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
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
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.
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).
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
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
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
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.
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) ::: 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.
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
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.
374 Chapter 7: SPECIAL TOPICS IN INTEGRAnON
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
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
Thus,
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}.
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}.
00 00
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
From An C n t A, it follows that A(A n Cn) t A(A), and an easy argument shows
that
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
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
we see that
00
Hence, )"(A) = inf{)"( 0): A S;; 0 and 0 open} also holds, and so the Lebesgue
measure ).. is a regular Borel measure.
f I ol/Jdf-L = f I df-L
holdsloreach IE C(X).
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
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}.
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
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
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
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
- .- - ,
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
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
JL(E) = r
lEnBn
I
Vex) - a dA(X).
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
: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,
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
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
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
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.
384 Chapter 7: SPECIAL IDPICS IN INTEGRATION
o ifx=O
I(x) = {x 2 cos(~) if 0 < x ::: 1
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
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
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.
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)
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,
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.
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).
holds.
J.L[(E) = Lf'(X)dX
Start by observing that if a = to < tl < ... < tn = b is a partition of [a, b],
then
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)]
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
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.
f gdIL/ = lb g(x)f'(x)dx.
~
d)" = D IL/ = f' a.e.
/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
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
.
Problem 39.16. Let I: JR. --+ JR. be a (uniformly) bounded function and let
00 00
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
°
Since e > is arbitrary, we infer that 'A(J(En») = 0, as desired. (Compare this
problem with Problem 18.9.)
00
[h(y)-h(XO)-/L(V-XO)l+[h(Xo)-h(X)-/L(XO-X)ll
I y-x
::::
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:
00
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
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.
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
ffi(x)
ax,
T'(x) = :
[
?Tk
ax, (x)
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,
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.
A = [ co~ e sin e ]
-sme cose '
.......... ..., 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
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.
r r t/'I x
FIGURE 7.4.
4(1
7r - e-r 2 ) ::=: ( l0
r
e-·t.2 dx )
2
::=: 4(1
7r - e- 2r 2 ),
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
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 ).
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 •
. [ 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
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
Section 40: TIlE CHANGE OF VARIABLES FORMULA 401
corresponding spaces and their corresponding Lebesgue measures are zero. Thus,
if
Z Z
(x, y, z)
(I', e, z)
•
! 21Z"
e y
r
JT(G)
fdA=jj
.
rf(rcose,r sine, z)rdrdedz
JG
holds.
z
(x,y, z)
e y
x
FIGURE 7.6. The Spherical Coordinate Transformation
Section 40: TIIE CHANGE OF VARIABLES FORMULA 403
rid)"
JT(O)
=// r I(r cose sin ¢, r sine sin¢, r cos¢)r2 sin¢ dr de d¢
Jo
holds.